* [PATCH 0/7] validate-bpf: add standalone validation debug tool
@ 2026-09-11 10:39 Marat Khalili
2026-09-11 10:39 ` [PATCH 1/7] bpf/validate: fix finished status on restart Marat Khalili
` (7 more replies)
0 siblings, 8 replies; 38+ messages in thread
From: Marat Khalili @ 2026-09-11 10:39 UTC (permalink / raw)
Cc: dev
This patchset introduces a new standalone tool for pre-validating eBPF
programs and debugging validation problems. Its should allow developers
to trace validator state changes per instruction and understand verifier
decisions, simplifying the debugging of rejected programs before loading
them into a real application context.
Marat Khalili (7):
bpf/validate: fix finished status on restart
bpf/validate: refactor internal step function
bpf/validate: formalize call back requirements
bpf/validate: add jump notification events
bpf/validate: add get current event API
app/test: add test for bpf validate debug events
app/validate-bpf: add BPF validation application
MAINTAINERS | 2 +
app/meson.build | 1 +
app/test/test_bpf_validate.c | 341 +++++++-
app/validate-bpf/alloc_list.c | 53 ++
app/validate-bpf/args.c | 263 ++++++
app/validate-bpf/debug.c | 1099 ++++++++++++++++++++++++
app/validate-bpf/debug_command.c | 383 +++++++++
app/validate-bpf/debug_command.h | 65 ++
app/validate-bpf/eal_init_args.c | 57 ++
app/validate-bpf/internal.h | 151 ++++
app/validate-bpf/main.c | 77 ++
app/validate-bpf/meson.build | 13 +
app/validate-bpf/parse_decl.c | 611 +++++++++++++
doc/guides/rel_notes/release_26_11.rst | 5 +
doc/guides/tools/index.rst | 1 +
doc/guides/tools/validate_bpf.rst | 97 +++
lib/bpf/bpf_validate.c | 49 +-
lib/bpf/bpf_validate_debug.c | 92 +-
lib/bpf/bpf_validate_debug.h | 10 +-
lib/bpf/rte_bpf_validate_debug.h | 42 +-
20 files changed, 3347 insertions(+), 65 deletions(-)
create mode 100644 app/validate-bpf/alloc_list.c
create mode 100644 app/validate-bpf/args.c
create mode 100644 app/validate-bpf/debug.c
create mode 100644 app/validate-bpf/debug_command.c
create mode 100644 app/validate-bpf/debug_command.h
create mode 100644 app/validate-bpf/eal_init_args.c
create mode 100644 app/validate-bpf/internal.h
create mode 100644 app/validate-bpf/main.c
create mode 100644 app/validate-bpf/meson.build
create mode 100644 app/validate-bpf/parse_decl.c
create mode 100644 doc/guides/tools/validate_bpf.rst
--
2.43.0
^ permalink raw reply [flat|nested] 38+ messages in thread
* [PATCH 1/7] bpf/validate: fix finished status on restart
2026-09-11 10:39 [PATCH 0/7] validate-bpf: add standalone validation debug tool Marat Khalili
@ 2026-09-11 10:39 ` Marat Khalili
2026-09-11 10:40 ` [PATCH 2/7] bpf/validate: refactor internal step function Marat Khalili
` (6 subsequent siblings)
7 siblings, 0 replies; 38+ messages in thread
From: Marat Khalili @ 2026-09-11 10:39 UTC (permalink / raw)
To: Konstantin Ananyev; +Cc: dev
If evaluation was restarted, requesting evaluation result would return
result from the previous evaluation instead of -EAGAIN.
Make evaluation start clear the finished status.
Fixes: 6b0da0602cc7 ("bpf/validate: introduce debugging interface")
Signed-off-by: Marat Khalili <marat.khalili@huawei.com>
---
lib/bpf/bpf_validate_debug.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/lib/bpf/bpf_validate_debug.c b/lib/bpf/bpf_validate_debug.c
index 5d18804a74bc..9a039766322c 100644
--- a/lib/bpf/bpf_validate_debug.c
+++ b/lib/bpf/bpf_validate_debug.c
@@ -568,6 +568,7 @@ __rte_bpf_validate_debug_evaluate_start(struct rte_bpf_validate_debug *debug,
debug->last_point = NULL;
debug->pc = 0;
debug->evaluate_started = true;
+ debug->evaluate_finished = false;
const int rc = debug_send_event(debug,
RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_START);
--
2.43.0
^ permalink raw reply related [flat|nested] 38+ messages in thread
* [PATCH 2/7] bpf/validate: refactor internal step function
2026-09-11 10:39 [PATCH 0/7] validate-bpf: add standalone validation debug tool Marat Khalili
2026-09-11 10:39 ` [PATCH 1/7] bpf/validate: fix finished status on restart Marat Khalili
@ 2026-09-11 10:40 ` Marat Khalili
2026-09-11 10:40 ` [PATCH 3/7] bpf/validate: formalize call back requirements Marat Khalili
` (5 subsequent siblings)
7 siblings, 0 replies; 38+ messages in thread
From: Marat Khalili @ 2026-09-11 10:40 UTC (permalink / raw)
To: Konstantin Ananyev; +Cc: dev
Rename `__rte_bpf_validate_debug_evaluate_step` to
`__rte_bpf_validate_debug_evaluate_update` to better reflect planned
usage, and make it accept a bitmask of events instead of a single event.
Step event is still implicitly triggered for now.
No functional changes.
Signed-off-by: Marat Khalili <marat.khalili@huawei.com>
---
lib/bpf/bpf_validate.c | 17 ++++++++---------
lib/bpf/bpf_validate_debug.c | 20 +++++++++++---------
lib/bpf/bpf_validate_debug.h | 11 ++++++-----
3 files changed, 25 insertions(+), 23 deletions(-)
diff --git a/lib/bpf/bpf_validate.c b/lib/bpf/bpf_validate.c
index f9960088a285..b317abb81172 100644
--- a/lib/bpf/bpf_validate.c
+++ b/lib/bpf/bpf_validate.c
@@ -2989,10 +2989,9 @@ evaluate(struct bpf_verifier *bvf)
break;
}
- rc = __rte_bpf_validate_debug_evaluate_step(debug, idx,
+ rc = __rte_bpf_validate_debug_evaluate_update(debug, idx,
prev_nb_edge > 1 ?
- RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER :
- RTE_BPF_VALIDATE_DEBUG_EVENT_STEP);
+ RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER) : 0);
if (rc < 0)
break;
@@ -3024,9 +3023,9 @@ evaluate(struct bpf_verifier *bvf)
}
if (bvf->evst->unreachable) {
- rc = __rte_bpf_validate_debug_evaluate_step(
+ rc = __rte_bpf_validate_debug_evaluate_update(
debug, get_node_idx(bvf, next),
- RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE);
+ RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE));
if (rc < 0)
break;
@@ -3038,9 +3037,9 @@ evaluate(struct bpf_verifier *bvf)
*/
} else if (node->nb_edge > 1 &&
prune_eval_state(bvf, node, next) == 0) {
- rc = __rte_bpf_validate_debug_evaluate_step(
+ rc = __rte_bpf_validate_debug_evaluate_update(
debug, get_node_idx(bvf, next),
- RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE);
+ RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE));
if (rc < 0)
break;
@@ -3059,9 +3058,9 @@ evaluate(struct bpf_verifier *bvf)
*/
if (prev_nb_edge != 0) {
- rc = __rte_bpf_validate_debug_evaluate_step(
+ rc = __rte_bpf_validate_debug_evaluate_update(
debug, get_node_idx(bvf, node) + 1,
- RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN);
+ RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN));
if (rc < 0)
break;
}
diff --git a/lib/bpf/bpf_validate_debug.c b/lib/bpf/bpf_validate_debug.c
index 9a039766322c..f76303a7fd2a 100644
--- a/lib/bpf/bpf_validate_debug.c
+++ b/lib/bpf/bpf_validate_debug.c
@@ -582,8 +582,8 @@ __rte_bpf_validate_debug_evaluate_start(struct rte_bpf_validate_debug *debug,
}
int
-__rte_bpf_validate_debug_evaluate_step(struct rte_bpf_validate_debug *debug,
- uint32_t pc, debug_event_t event)
+__rte_bpf_validate_debug_evaluate_update(struct rte_bpf_validate_debug *debug,
+ uint32_t pc, uint64_t events)
{
int rc;
@@ -595,8 +595,7 @@ __rte_bpf_validate_debug_evaluate_step(struct rte_bpf_validate_debug *debug,
return -ECHILD;
}
- if (pc > debug->bpf_prm->raw.nb_ins || event < 0 ||
- event >= RTE_BPF_VALIDATE_DEBUG_EVENT_END)
+ if (pc > debug->bpf_prm->raw.nb_ins)
return -EINVAL;
debug->pc = pc;
@@ -606,11 +605,13 @@ __rte_bpf_validate_debug_evaluate_step(struct rte_bpf_validate_debug *debug,
rc = debug_send_event(debug,
RTE_BPF_VALIDATE_DEBUG_EVENT_INVALID_STATE);
- if (event != RTE_BPF_VALIDATE_DEBUG_EVENT_STEP)
- rc = rc < 0 ? rc : debug_send_event(debug, event);
+ for (enum rte_bpf_validate_debug_event event = 0;
+ event != RTE_BPF_VALIDATE_DEBUG_EVENT_END; ++event)
+ if (events & RTE_BIT64(event))
+ rc = rc < 0 ? rc : debug_send_event(debug, event);
- if (event == RTE_BPF_VALIDATE_DEBUG_EVENT_STEP ||
- event == RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER)
+ if (events == 0 || events == RTE_BIT64(
+ RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER))
/* Stepping into a real instruction to execute. */
rc = rc < 0 ? rc : debug_points_call_back(debug,
&debug->breakpoint_lists[pc]);
@@ -651,7 +652,8 @@ __rte_bpf_validate_debug_evaluate_finish(struct rte_bpf_validate_debug *debug,
event = RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS;
}
- rc = __rte_bpf_validate_debug_evaluate_step(debug, pc, event);
+ rc = __rte_bpf_validate_debug_evaluate_update(debug, pc,
+ RTE_BIT64(event));
}
debug_evaluate_close(debug);
diff --git a/lib/bpf/bpf_validate_debug.h b/lib/bpf/bpf_validate_debug.h
index a91f3e9c48b2..2b0c3bb646a7 100644
--- a/lib/bpf/bpf_validate_debug.h
+++ b/lib/bpf/bpf_validate_debug.h
@@ -46,7 +46,7 @@ __rte_bpf_validate_debug_evaluate_start(struct rte_bpf_validate_debug *debug,
const struct bpf_verifier *verifier, const struct rte_bpf_prm_ex *bpf_prm);
/*
- * Signal each instruction, branch end, or evaluation end.
+ * Signal evaluation events.
*
* Immediately return 0 if debug is NULL.
*
@@ -54,16 +54,17 @@ __rte_bpf_validate_debug_evaluate_start(struct rte_bpf_validate_debug *debug,
* Validate debug instance configured by user, can be NULL.
* @param pc
* Current value of the program counter, or next after last instruction.
- * @param event
- * Specific evaluation event if any, or RTE_BPF_VALIDATE_DEBUG_EVENT_STEP.
+ * @param events
+ * Bitmask of events.
+ * Step bit is always implied and should not be set.
* @return
* Non-negative value: evaluation should continue;
* -ECANCELED: evaluation should fail without calling this API again;
* Other negative value: evaluation should fail signalling failure;
*/
int
-__rte_bpf_validate_debug_evaluate_step(struct rte_bpf_validate_debug *debug,
- uint32_t pc, debug_event_t event);
+__rte_bpf_validate_debug_evaluate_update(struct rte_bpf_validate_debug *debug,
+ uint32_t pc, uint64_t events);
/*
* Signal end of evaluation process.
--
2.43.0
^ permalink raw reply related [flat|nested] 38+ messages in thread
* [PATCH 3/7] bpf/validate: formalize call back requirements
2026-09-11 10:39 [PATCH 0/7] validate-bpf: add standalone validation debug tool Marat Khalili
2026-09-11 10:39 ` [PATCH 1/7] bpf/validate: fix finished status on restart Marat Khalili
2026-09-11 10:40 ` [PATCH 2/7] bpf/validate: refactor internal step function Marat Khalili
@ 2026-09-11 10:40 ` Marat Khalili
2026-09-11 10:40 ` [PATCH 4/7] bpf/validate: add jump notification events Marat Khalili
` (4 subsequent siblings)
7 siblings, 0 replies; 38+ messages in thread
From: Marat Khalili @ 2026-09-11 10:40 UTC (permalink / raw)
To: Konstantin Ananyev; +Cc: dev
Formalize requirements to evaluation events in documentation and adjust
their handling in the actual code to match.
Particularly, specify more precisely:
* conditions for each catchpoint callback;
* relative order of callbacks within the step;
* expected value of the program counter;
Behaviour changes:
* one branch-return event is now generated for each branch-enter event
(previously multiple branch-return events could be coalesced);
* branch-return event now points to the corresponding conditional jump
(previously past the end of the last branch);
* step event is now only generated before an actual instruction
evaluation (previously also for branching and finishing events);
* program counter is no longer allowed to point past the program end;
These changes should simplify branch tracking of the validator by tests
and interactive applications. E.g. branch-enter and branch-return events
can now be used to push and pop branching points to/from some stack.
Signed-off-by: Marat Khalili <marat.khalili@huawei.com>
---
lib/bpf/bpf_validate.c | 28 +++++++++----
lib/bpf/bpf_validate_debug.c | 72 +++++++++++++++++++-------------
lib/bpf/bpf_validate_debug.h | 1 -
lib/bpf/rte_bpf_validate_debug.h | 24 +++++++----
4 files changed, 77 insertions(+), 48 deletions(-)
diff --git a/lib/bpf/bpf_validate.c b/lib/bpf/bpf_validate.c
index b317abb81172..f4258c2fb54f 100644
--- a/lib/bpf/bpf_validate.c
+++ b/lib/bpf/bpf_validate.c
@@ -2892,6 +2892,23 @@ prune_eval_state(struct bpf_verifier *bvf, const struct inst_node *node,
return rc;
}
+static bool
+is_branch_start(const struct inst_node *node)
+{
+ return node->prev_node != NULL && node->prev_node->nb_edge > 1;
+}
+
+static uint64_t
+step_events(const struct inst_node *node)
+{
+ uint64_t events = RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_STEP);
+
+ if (is_branch_start(node))
+ events |= RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER);
+
+ return events;
+}
+
/* Do second pass through CFG and try to evaluate instructions
* via each possible path. The verifier will try all paths, tracking types of
* registers used as input to instructions, and updating resulting type via
@@ -2918,7 +2935,6 @@ evaluate(struct bpf_verifier *bvf)
const char *err;
const struct ebpf_insn *ins;
struct inst_node *next, *node;
- int prev_nb_edge; /* branching number of the previous instruction */
int rc, debug_rc;
struct rte_bpf_validate_debug *const debug = bvf->prm->debug;
@@ -2954,7 +2970,6 @@ evaluate(struct bpf_verifier *bvf)
ins = bvf->prm->raw.ins;
node = bvf->in;
next = node;
- prev_nb_edge = 1;
memset(&stats, 0, sizeof(stats));
@@ -2990,8 +3005,7 @@ evaluate(struct bpf_verifier *bvf)
}
rc = __rte_bpf_validate_debug_evaluate_update(debug, idx,
- prev_nb_edge > 1 ?
- RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER) : 0);
+ step_events(node));
if (rc < 0)
break;
@@ -3047,7 +3061,6 @@ evaluate(struct bpf_verifier *bvf)
stats.nb_prune++;
} else {
next->prev_node = node;
- prev_nb_edge = node->nb_edge;
node = next;
}
} else {
@@ -3057,9 +3070,9 @@ evaluate(struct bpf_verifier *bvf)
* and proceed with parent.
*/
- if (prev_nb_edge != 0) {
+ if (is_branch_start(node)) {
rc = __rte_bpf_validate_debug_evaluate_update(
- debug, get_node_idx(bvf, node) + 1,
+ debug, get_node_idx(bvf, node->prev_node),
RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN));
if (rc < 0)
break;
@@ -3067,7 +3080,6 @@ evaluate(struct bpf_verifier *bvf)
node->cur_edge = 0;
save_safe_eval_state(bvf, node);
- prev_nb_edge = 0;
node = node->prev_node;
/* first node will not have prev, signalling finish */
diff --git a/lib/bpf/bpf_validate_debug.c b/lib/bpf/bpf_validate_debug.c
index f76303a7fd2a..db6d13f9cb22 100644
--- a/lib/bpf/bpf_validate_debug.c
+++ b/lib/bpf/bpf_validate_debug.c
@@ -225,6 +225,13 @@ debug_points_call_back(struct rte_bpf_validate_debug *debug,
return rc;
}
+/* Call back all breakpoints for the specified program counter. */
+static int
+debug_trigger_breakpoints(struct rte_bpf_validate_debug *debug, uint32_t pc)
+{
+ return debug_points_call_back(debug, &debug->breakpoint_lists[pc]);
+}
+
/* Call back all catchpoints for the specified event. */
static int
debug_send_event(struct rte_bpf_validate_debug *debug, debug_event_t event)
@@ -585,6 +592,21 @@ int
__rte_bpf_validate_debug_evaluate_update(struct rte_bpf_validate_debug *debug,
uint32_t pc, uint64_t events)
{
+ /* Required order of sent events according to the documentation. */
+ static const enum rte_bpf_validate_debug_event ordered_events[] = {
+ RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_START,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_INVALID_STATE,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_STEP,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_FAILURE,
+ };
+ RTE_BUILD_BUG_ON(
+ RTE_DIM(ordered_events) != RTE_BPF_VALIDATE_DEBUG_EVENT_END);
+
int rc;
if (debug == NULL)
@@ -595,29 +617,31 @@ __rte_bpf_validate_debug_evaluate_update(struct rte_bpf_validate_debug *debug,
return -ECHILD;
}
- if (pc > debug->bpf_prm->raw.nb_ins)
+ if (pc >= debug->bpf_prm->raw.nb_ins)
return -EINVAL;
debug->pc = pc;
rc = __rte_bpf_validate_state_is_valid(debug->verifier);
if (rc == 0)
- rc = debug_send_event(debug,
- RTE_BPF_VALIDATE_DEBUG_EVENT_INVALID_STATE);
+ events |= RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_INVALID_STATE);
+
+ for (uint32_t index = 0; index < RTE_DIM(ordered_events); index++) {
+ const enum rte_bpf_validate_debug_event event =
+ ordered_events[index];
+ if ((events & RTE_BIT64(event)) == 0)
+ continue;
- for (enum rte_bpf_validate_debug_event event = 0;
- event != RTE_BPF_VALIDATE_DEBUG_EVENT_END; ++event)
- if (events & RTE_BIT64(event))
- rc = rc < 0 ? rc : debug_send_event(debug, event);
+ if (event == RTE_BPF_VALIDATE_DEBUG_EVENT_STEP)
+ rc = rc < 0 ? rc : debug_trigger_breakpoints(debug, pc);
- if (events == 0 || events == RTE_BIT64(
- RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER))
- /* Stepping into a real instruction to execute. */
- rc = rc < 0 ? rc : debug_points_call_back(debug,
- &debug->breakpoint_lists[pc]);
+ rc = rc < 0 ? rc : debug_send_event(debug, event);
+ events -= RTE_BIT64(event);
+ }
- rc = rc < 0 ? rc : debug_send_event(debug,
- RTE_BPF_VALIDATE_DEBUG_EVENT_STEP);
+ if (events != 0)
+ /* Received unsupported events. */
+ rc = rc < 0 ? rc : -EINVAL;
return rc;
}
@@ -627,8 +651,6 @@ __rte_bpf_validate_debug_evaluate_finish(struct rte_bpf_validate_debug *debug,
int result)
{
int rc = 0;
- uint32_t pc;
- debug_event_t event;
if (debug == NULL)
return 0;
@@ -641,20 +663,10 @@ __rte_bpf_validate_debug_evaluate_finish(struct rte_bpf_validate_debug *debug,
debug->evaluate_finished = true;
debug->evaluate_result = result;
- if (result != -ECANCELED) {
- if (result < 0) {
- /* Last known pc is the place we failed. */
- pc = debug->pc;
- event = RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_FAILURE;
- } else {
- /* Show program end, not particular instruction. */
- pc = debug->bpf_prm->raw.nb_ins;
- event = RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS;
- }
-
- rc = __rte_bpf_validate_debug_evaluate_update(debug, pc,
- RTE_BIT64(event));
- }
+ if (result != -ECANCELED)
+ rc = debug_send_event(debug, result < 0 ?
+ RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_FAILURE :
+ RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS);
debug_evaluate_close(debug);
diff --git a/lib/bpf/bpf_validate_debug.h b/lib/bpf/bpf_validate_debug.h
index 2b0c3bb646a7..6d1f1ac99e3e 100644
--- a/lib/bpf/bpf_validate_debug.h
+++ b/lib/bpf/bpf_validate_debug.h
@@ -56,7 +56,6 @@ __rte_bpf_validate_debug_evaluate_start(struct rte_bpf_validate_debug *debug,
* Current value of the program counter, or next after last instruction.
* @param events
* Bitmask of events.
- * Step bit is always implied and should not be set.
* @return
* Non-negative value: evaluation should continue;
* -ECANCELED: evaluation should fail without calling this API again;
diff --git a/lib/bpf/rte_bpf_validate_debug.h b/lib/bpf/rte_bpf_validate_debug.h
index f30fa926f10a..b3fedd5428bc 100644
--- a/lib/bpf/rte_bpf_validate_debug.h
+++ b/lib/bpf/rte_bpf_validate_debug.h
@@ -31,25 +31,31 @@ extern "C" {
* Supported validate events.
*
* Valid events begin from 0 and end before `RTE_BPF_VALIDATE_DEBUG_EVENT_END`.
+ *
+ * At any given evaluation step, callbacks are fired in the following order:
+ * - Validation start event;
+ * - Branching and invalid-state events (can be interleaved);
+ * - Instruction breakpoints (before evaluating instruction);
+ * - Step (before evaluating instruction) or validation result (if done) event;
*/
enum rte_bpf_validate_debug_event {
- /* Just before every instruction, at branch or validation end. */
+ /* Just before every evaluated instruction. */
RTE_BPF_VALIDATE_DEBUG_EVENT_STEP,
/* Validator has failed its internal self-checks. */
RTE_BPF_VALIDATE_DEBUG_EVENT_INVALID_STATE,
- /* Start of validation. */
+ /* Start of validation (pc points to first instruction). */
RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_START,
- /* Successful finish of validation. */
+ /* Successful finish (pc undefined). */
RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS,
- /* Finish of validation with error. */
+ /* Finish of validation with error (pc points at error). */
RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_FAILURE,
- /* Beginning of a branch just after the jump. */
+ /* Beginning of a branch evaluation (pc points to branch start). */
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER,
- /* Pruning branch as verified earlier. */
+ /* Pruning branch as verified earlier (pc points to branch start). */
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE,
- /* End of branch verification, after the last verified instruction. */
+ /* End of branch verification (pc points to jump instruction). */
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN,
- /* Pruning branch as dynamically unreachable. */
+ /* Pruning branch as dynamically unreachable (pc points to branch start). */
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE,
/* Number of valid event values. */
RTE_BPF_VALIDATE_DEBUG_EVENT_END,
@@ -208,7 +214,7 @@ rte_bpf_validate_debug_get_last_point(const struct rte_bpf_validate_debug *debug
* @param debug
* Debug instance.
* @return
- * Current program counter being validated, or one after last.
+ * Current program counter being validated.
* UINT32_MAX if no program is being validated.
*/
__rte_experimental
--
2.43.0
^ permalink raw reply related [flat|nested] 38+ messages in thread
* [PATCH 4/7] bpf/validate: add jump notification events
2026-09-11 10:39 [PATCH 0/7] validate-bpf: add standalone validation debug tool Marat Khalili
` (2 preceding siblings ...)
2026-09-11 10:40 ` [PATCH 3/7] bpf/validate: formalize call back requirements Marat Khalili
@ 2026-09-11 10:40 ` Marat Khalili
2026-09-11 10:40 ` [PATCH 5/7] bpf/validate: add get current event API Marat Khalili
` (3 subsequent siblings)
7 siblings, 0 replies; 38+ messages in thread
From: Marat Khalili @ 2026-09-11 10:40 UTC (permalink / raw)
To: Konstantin Ananyev; +Cc: dev
Add events notifying validate debug client of instructions that may
change program counter. This is particularly useful to an interactive
debugger to track conditional and unconditional jumps without trying to
parse the program and interpret instructions independently.
Signed-off-by: Marat Khalili <marat.khalili@huawei.com>
---
lib/bpf/bpf_validate.c | 10 ++++++++--
lib/bpf/bpf_validate_debug.c | 2 ++
lib/bpf/rte_bpf_validate_debug.h | 5 +++++
3 files changed, 15 insertions(+), 2 deletions(-)
diff --git a/lib/bpf/bpf_validate.c b/lib/bpf/bpf_validate.c
index f4258c2fb54f..d503db836105 100644
--- a/lib/bpf/bpf_validate.c
+++ b/lib/bpf/bpf_validate.c
@@ -2899,13 +2899,19 @@ is_branch_start(const struct inst_node *node)
}
static uint64_t
-step_events(const struct inst_node *node)
+step_events(const struct inst_node *node, uint8_t op)
{
uint64_t events = RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_STEP);
if (is_branch_start(node))
events |= RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER);
+ if (op == (BPF_JMP | BPF_JA))
+ events |= RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS);
+
+ if (node->nb_edge > 1)
+ events |= RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL);
+
return events;
}
@@ -3005,7 +3011,7 @@ evaluate(struct bpf_verifier *bvf)
}
rc = __rte_bpf_validate_debug_evaluate_update(debug, idx,
- step_events(node));
+ step_events(node, op));
if (rc < 0)
break;
diff --git a/lib/bpf/bpf_validate_debug.c b/lib/bpf/bpf_validate_debug.c
index db6d13f9cb22..098a88026d6d 100644
--- a/lib/bpf/bpf_validate_debug.c
+++ b/lib/bpf/bpf_validate_debug.c
@@ -600,6 +600,8 @@ __rte_bpf_validate_debug_evaluate_update(struct rte_bpf_validate_debug *debug,
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE,
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN,
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL,
RTE_BPF_VALIDATE_DEBUG_EVENT_STEP,
RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS,
RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_FAILURE,
diff --git a/lib/bpf/rte_bpf_validate_debug.h b/lib/bpf/rte_bpf_validate_debug.h
index b3fedd5428bc..f1ee8769d06a 100644
--- a/lib/bpf/rte_bpf_validate_debug.h
+++ b/lib/bpf/rte_bpf_validate_debug.h
@@ -35,6 +35,7 @@ extern "C" {
* At any given evaluation step, callbacks are fired in the following order:
* - Validation start event;
* - Branching and invalid-state events (can be interleaved);
+ * - Jump-instruction events (before evaluating jump instruction);
* - Instruction breakpoints (before evaluating instruction);
* - Step (before evaluating instruction) or validation result (if done) event;
*/
@@ -57,6 +58,10 @@ enum rte_bpf_validate_debug_event {
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN,
/* Pruning branch as dynamically unreachable (pc points to branch start). */
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE,
+ /* Program counter points to a jump-always instruction. */
+ RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS,
+ /* Program counter points to a conditional jump instruction. */
+ RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL,
/* Number of valid event values. */
RTE_BPF_VALIDATE_DEBUG_EVENT_END,
};
--
2.43.0
^ permalink raw reply related [flat|nested] 38+ messages in thread
* [PATCH 5/7] bpf/validate: add get current event API
2026-09-11 10:39 [PATCH 0/7] validate-bpf: add standalone validation debug tool Marat Khalili
` (3 preceding siblings ...)
2026-09-11 10:40 ` [PATCH 4/7] bpf/validate: add jump notification events Marat Khalili
@ 2026-09-11 10:40 ` Marat Khalili
2026-09-11 10:40 ` [PATCH 6/7] app/test: add test for bpf validate debug events Marat Khalili
` (2 subsequent siblings)
7 siblings, 0 replies; 38+ messages in thread
From: Marat Khalili @ 2026-09-11 10:40 UTC (permalink / raw)
To: Konstantin Ananyev; +Cc: dev
Add API for retrieving the event currently processed by a callback.
Useful when single callback is registered for multiple events to put
less pressure on the context argument.
Signed-off-by: Marat Khalili <marat.khalili@huawei.com>
---
lib/bpf/bpf_validate_debug.c | 13 +++++++++++++
lib/bpf/rte_bpf_validate_debug.h | 13 +++++++++++++
2 files changed, 26 insertions(+)
diff --git a/lib/bpf/bpf_validate_debug.c b/lib/bpf/bpf_validate_debug.c
index 098a88026d6d..c8e9e5dd9227 100644
--- a/lib/bpf/bpf_validate_debug.c
+++ b/lib/bpf/bpf_validate_debug.c
@@ -43,6 +43,7 @@ struct rte_bpf_validate_debug {
const struct bpf_verifier *verifier;
const struct rte_bpf_prm_ex *bpf_prm;
struct point_list *breakpoint_lists;
+ enum rte_bpf_validate_debug_event current_event;
struct rte_bpf_validate_debug_point *last_point;
uint32_t pc;
/* Evaluate stage (only tracking `evaluate` part at the moment). */
@@ -236,6 +237,7 @@ debug_trigger_breakpoints(struct rte_bpf_validate_debug *debug, uint32_t pc)
static int
debug_send_event(struct rte_bpf_validate_debug *debug, debug_event_t event)
{
+ debug->current_event = event;
return debug_points_call_back(debug, &debug->catchpoint_lists[event]);
}
@@ -342,6 +344,17 @@ rte_bpf_validate_debug_get_ins(const struct rte_bpf_validate_debug *debug,
return 0;
}
+RTE_EXPORT_EXPERIMENTAL_SYMBOL(rte_bpf_validate_debug_get_event, 26.11)
+enum rte_bpf_validate_debug_event
+rte_bpf_validate_debug_get_event(const struct rte_bpf_validate_debug *debug)
+{
+ if (debug == NULL)
+ /* Just to be fool-proof, not really required by API. */
+ return -EINVAL;
+
+ return debug->current_event;
+}
+
RTE_EXPORT_EXPERIMENTAL_SYMBOL(rte_bpf_validate_debug_get_last_point, 26.07)
struct rte_bpf_validate_debug_point *
rte_bpf_validate_debug_get_last_point(const struct rte_bpf_validate_debug *debug)
diff --git a/lib/bpf/rte_bpf_validate_debug.h b/lib/bpf/rte_bpf_validate_debug.h
index f1ee8769d06a..6d4bf53589e0 100644
--- a/lib/bpf/rte_bpf_validate_debug.h
+++ b/lib/bpf/rte_bpf_validate_debug.h
@@ -194,6 +194,19 @@ int
rte_bpf_validate_debug_get_ins(const struct rte_bpf_validate_debug *debug,
const struct ebpf_insn **ins, uint32_t *nb_ins);
+/**
+ * Get event currently processed by a catchpoint callback.
+ *
+ * @param debug
+ * Debug instance.
+ * @return
+ * Event currently processed by a catchpoint callback.
+ * Undefined if no event is currently being processed.
+ */
+__rte_experimental
+enum rte_bpf_validate_debug_event
+rte_bpf_validate_debug_get_event(const struct rte_bpf_validate_debug *debug);
+
/**
* Get last triggered breakpoint or catchpoint.
*
--
2.43.0
^ permalink raw reply related [flat|nested] 38+ messages in thread
* [PATCH 6/7] app/test: add test for bpf validate debug events
2026-09-11 10:39 [PATCH 0/7] validate-bpf: add standalone validation debug tool Marat Khalili
` (4 preceding siblings ...)
2026-09-11 10:40 ` [PATCH 5/7] bpf/validate: add get current event API Marat Khalili
@ 2026-09-11 10:40 ` Marat Khalili
2026-09-11 10:40 ` [PATCH 7/7] app/validate-bpf: add BPF validation application Marat Khalili
2026-09-14 14:54 ` [PATCH v2 0/7] validate-bpf: add standalone validation debug tool Marat Khalili
7 siblings, 0 replies; 38+ messages in thread
From: Marat Khalili @ 2026-09-11 10:40 UTC (permalink / raw)
To: Konstantin Ananyev; +Cc: dev
Add a test that verifies that eBPF validator generates callbacks with
the number, order and program counter specified in the documentation.
Signed-off-by: Marat Khalili <marat.khalili@huawei.com>
---
app/test/test_bpf_validate.c | 341 ++++++++++++++++++++++++++++++++++-
1 file changed, 337 insertions(+), 4 deletions(-)
diff --git a/app/test/test_bpf_validate.c b/app/test/test_bpf_validate.c
index 066f1fa156ae..ea4621b257f4 100644
--- a/app/test/test_bpf_validate.c
+++ b/app/test/test_bpf_validate.c
@@ -9,10 +9,6 @@
#include <rte_bpf_validate_debug.h>
#include <rte_errno.h>
-/*
- * Tests of BPF validation.
- */
-
extern int test_bpf_validate_logtype;
#define RTE_LOGTYPE_TEST_BPF_VALIDATE test_bpf_validate_logtype
#define TEST_LOG_LINE(level, ...) \
@@ -26,6 +22,343 @@ RTE_LOG_REGISTER(test_bpf_validate_logtype, test.bpf_validate, NOTICE);
/* Special value indicating that register variable is not being used. */
#define NO_REGISTER UINT8_MAX
+static const char *const event_names[RTE_BPF_VALIDATE_DEBUG_EVENT_END] = {
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_STEP] = "step",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_INVALID_STATE] = "invalid-state",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_START] = "validation-start",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS] = "validation-success",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_FAILURE] = "validation-failure",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER] = "branch-enter",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE] = "branch-prune",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN] = "branch-return",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE] = "branch-unreachable",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS] = "jump-always",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL] = "jump-conditional",
+};
+
+/*
+ * Self-test of BPF validation debug events.
+ */
+
+#define EVENT_BIT(name) RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_ ## name)
+#define EVENT_BITS_ALL RTE_LEN2MASK(RTE_BPF_VALIDATE_DEBUG_EVENT_END, uint64_t)
+
+struct test_events_context {
+ /* Tested program. */
+ const struct ebpf_insn *ins;
+ uint32_t nb_ins;
+ /* Number of times each of the events was encountered. */
+ int event_counts[RTE_BPF_VALIDATE_DEBUG_EVENT_END];
+ /* Events active within the current step. */
+ uint64_t active_events;
+ /* Breakpoint was called within the current step. */
+ bool breakpoint_called;
+ /* Currently active conditional jump pc. */
+ uint32_t branch_pc;
+ /* Program counters of all conditional jumps in the current code path. */
+ uint32_t branch_pc_stack[4];
+ size_t branch_pc_stack_length;
+};
+
+/* Push current branch_pc to stack and clear it. */
+static int
+test_events_branch_pc_push(struct test_events_context *ctx)
+{
+ TEST_ASSERT_NOT_EQUAL(ctx->branch_pc, NO_PROGRAM_COUNTER,
+ "no branch pc to push");
+ TEST_ASSERT(ctx->branch_pc_stack_length < RTE_DIM(ctx->branch_pc_stack),
+ "branch stack overflow");
+ ctx->branch_pc_stack[ctx->branch_pc_stack_length++] = ctx->branch_pc;
+ ctx->branch_pc = NO_PROGRAM_COUNTER;
+ return TEST_SUCCESS;
+}
+
+/* Pop branch_pc from stack. */
+static int
+test_events_branch_pc_pop(struct test_events_context *ctx)
+{
+ TEST_ASSERT(ctx->branch_pc_stack_length > 0, "branch stack underflow");
+ ctx->branch_pc = ctx->branch_pc_stack[--ctx->branch_pc_stack_length];
+ return TEST_SUCCESS;
+}
+
+/* Verify breakpoint, checking that it is only called once and remembering it. */
+static int
+test_events_break_cb(struct rte_bpf_validate_debug *debug, void *void_ctx)
+{
+ struct test_events_context *const ctx = void_ctx;
+ const uint32_t pc = rte_bpf_validate_debug_get_pc(debug);
+
+ TEST_LOG_LINE(DEBUG, "%2u: break", pc);
+
+ TEST_ASSERT(!ctx->breakpoint_called,
+ "breakpoint called again at pc %u", pc);
+ ctx->breakpoint_called = true;
+
+ return TEST_SUCCESS;
+}
+
+/*
+ * Verify event, checking that:
+ * - validation-start is called at pc 0;
+ * - branch-return is called at a corresponding conditional jump pc;
+ * - branch-enter and branch-return are correctly paired;
+ * - jump-conditional and jump-always are called for corresponding opcodes;
+ * - jump-conditional and only it is followed by branches events;
+ * - events and breakpoints are correctly ordered;
+ *
+ * Increment event counter to compare against expected counts later.
+ */
+static int
+test_events_catch_cb(struct rte_bpf_validate_debug *debug, void *void_ctx)
+{
+ /* Bitmask of events allowed to precede each given event within the step. */
+ static const uint64_t allowed_preceding_events[] = {
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_START] = 0,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE] =
+ EVENT_BIT(BRANCH_PRUNE) |
+ EVENT_BIT(BRANCH_UNREACHABLE) |
+ EVENT_BIT(BRANCH_RETURN),
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE] =
+ EVENT_BIT(BRANCH_PRUNE) |
+ EVENT_BIT(BRANCH_UNREACHABLE) |
+ EVENT_BIT(BRANCH_RETURN),
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER] =
+ EVENT_BIT(BRANCH_PRUNE) |
+ EVENT_BIT(BRANCH_UNREACHABLE) |
+ EVENT_BIT(BRANCH_RETURN),
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN] =
+ EVENT_BIT(BRANCH_PRUNE) |
+ EVENT_BIT(BRANCH_UNREACHABLE) |
+ EVENT_BIT(BRANCH_RETURN),
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL] =
+ EVENT_BIT(BRANCH_ENTER) |
+ EVENT_BIT(BRANCH_PRUNE) |
+ EVENT_BIT(BRANCH_UNREACHABLE) |
+ EVENT_BIT(BRANCH_RETURN),
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS] =
+ EVENT_BIT(BRANCH_ENTER) |
+ EVENT_BIT(BRANCH_PRUNE) |
+ EVENT_BIT(BRANCH_UNREACHABLE) |
+ EVENT_BIT(BRANCH_RETURN),
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_STEP] =
+ EVENT_BITS_ALL - EVENT_BIT(VALIDATION_SUCCESS),
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS] =
+ EVENT_BIT(BRANCH_PRUNE) |
+ EVENT_BIT(BRANCH_UNREACHABLE) |
+ EVENT_BIT(BRANCH_RETURN),
+ };
+
+ struct test_events_context *const ctx = void_ctx;
+ const enum rte_bpf_validate_debug_event event =
+ rte_bpf_validate_debug_get_event(debug);
+ const uint32_t pc = rte_bpf_validate_debug_get_pc(debug);
+
+ TEST_LOG_LINE(DEBUG, "%2u: %s", pc, event_names[event]);
+
+ /* Verify event program counter and underlying instruction. */
+ switch (event) {
+ case RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_START:
+ TEST_ASSERT_EQUAL(pc, 0,
+ "event %s called at pc %u",
+ event_names[event], pc);
+ break;
+ case RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER:
+ TEST_ASSERT_SUCCESS(test_events_branch_pc_push(ctx),
+ "error pushing branch for event %s at pc %u",
+ event_names[event], pc);
+ break;
+ case RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN:
+ TEST_ASSERT_SUCCESS(test_events_branch_pc_pop(ctx),
+ "error popping branch for event %s at pc %u",
+ event_names[event], pc);
+ TEST_ASSERT_EQUAL(pc, ctx->branch_pc,
+ "event %s called at pc %u after jump pc %u",
+ event_names[event], pc, ctx->branch_pc);
+ break;
+ case RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS:
+ TEST_ASSERT_EQUAL(BPF_CLASS(ctx->ins[pc].code), BPF_JMP,
+ "event %s called at non-jump pc %u",
+ event_names[event], pc);
+ TEST_ASSERT_EQUAL(BPF_OP(ctx->ins[pc].code), BPF_JA,
+ "event %s called at non-jump-always pc %u",
+ event_names[event], pc);
+ break;
+ case RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL:
+ TEST_ASSERT_EQUAL(BPF_CLASS(ctx->ins[pc].code), BPF_JMP,
+ "event %s called at non-jump pc %u",
+ event_names[event], pc);
+ TEST_ASSERT_NOT_EQUAL(BPF_OP(ctx->ins[pc].code), BPF_JA,
+ "event %s called at jump-always pc %u",
+ event_names[event], pc);
+ break;
+ default:
+ break;
+ }
+
+ /* Verify events order. */
+ TEST_ASSERT_EQUAL(
+ (ctx->active_events & ~allowed_preceding_events[event]), 0,
+ "event %s at pc %u preceded by %#jx",
+ event_names[event], pc, (uintmax_t)ctx->active_events);
+
+ /* Update events for the current step as well as branch program counter. */
+ if (event != RTE_BPF_VALIDATE_DEBUG_EVENT_STEP) {
+ TEST_ASSERT(!ctx->breakpoint_called,
+ "event %s called after a breakpoint at pc %u",
+ event_names[event], pc);
+ ctx->active_events |= RTE_BIT64(event);
+ } else {
+ TEST_ASSERT_EQUAL(ctx->branch_pc, NO_PROGRAM_COUNTER,
+ "no branch-enter between branch- and next instruction");
+
+ if (ctx->active_events & EVENT_BIT(JUMP_CONDITIONAL))
+ /* Remember branch instruction pointer, branches to follow. */
+ ctx->branch_pc = pc;
+
+ TEST_ASSERT(ctx->breakpoint_called,
+ "step called before or without a breakpoint at pc %u", pc);
+
+ /* Prepare for the next step. */
+ ctx->active_events = 0;
+ ctx->breakpoint_called = false;
+ }
+
+ /* Update total event counts. */
+ ++ctx->event_counts[event];
+
+ return TEST_SUCCESS;
+}
+
+/*
+ * This test verifies that eBPF validator generates expected numbers of events
+ * of each kind in the expected order (e.g. step after jump after branch).
+ */
+static int
+test_events(void)
+{
+ /*
+ * Program demonstrating the following properties:
+ * - prunnable code paths;
+ * - dynamically unreachable code paths;
+ * - conditional jumps landing on conditional jumps;
+ * - conditional jumps landing on unconditional jumps;
+ */
+ static const struct ebpf_insn ins[] = {
+ /* 0: r0 = 0 */
+ { .code = (EBPF_ALU64 | EBPF_MOV | BPF_K), .dst_reg = EBPF_REG_0, .imm = 0 },
+ /* 1: if r1 > 0 goto 4 */
+ { .code = (BPF_JMP | BPF_JGT | BPF_K), .dst_reg = EBPF_REG_1, .imm = 0, .off = 2 },
+ /* 2: r1 = 0 */
+ { .code = (EBPF_ALU64 | EBPF_MOV | BPF_K), .dst_reg = EBPF_REG_1, .imm = 0 },
+ /* 3: if r0 < 1 goto 8 */
+ { .code = (BPF_JMP | EBPF_JLT | BPF_K), .dst_reg = EBPF_REG_0, .imm = 1, .off = 4 },
+ /* 4: if r1 > 0 goto 6 */
+ { .code = (BPF_JMP | BPF_JGT | BPF_K), .dst_reg = EBPF_REG_1, .imm = 0, .off = 1 },
+ /* 5: goto 9 */
+ { .code = (BPF_JMP | BPF_JA), .off = 3 },
+ /* 6: r1 = 0 */
+ { .code = (EBPF_ALU64 | EBPF_MOV | BPF_K), .dst_reg = EBPF_REG_1, .imm = 0 },
+ /* 7: if r0 > 0 goto 9 */
+ { .code = (BPF_JMP | BPF_JGT | BPF_K), .dst_reg = EBPF_REG_0, .imm = 0, .off = 1 },
+ /* 8: goto 10 */
+ { .code = (BPF_JMP | BPF_JA), .off = 1 },
+ /* 9: r0 = 1 */
+ { .code = (EBPF_ALU64 | EBPF_MOV | BPF_K), .dst_reg = EBPF_REG_0, .imm = 1 },
+ /* 10: exit */
+ { .code = (BPF_JMP | EBPF_EXIT) }
+ };
+
+ /*
+ * Verified code paths (order may vary by library implementation):
+ * 0 -- 1 -- 4 -- 6 -- 7 -- unreachable 9
+ * | | `-- 8 -- 10 -- return 7 -- return 4
+ * | `-- unreachable 5 -- return 1
+ * `-- 2 -- 3 -- prune 8
+ * `-- unreachable 4 -- return 1 -- success 11
+ *
+ * Note that 4 encountered conditional jump instructions generate:
+ * - 4 pairs of branch-enter/branch-return events;
+ * - 3 branch-unreachable events;
+ * - 1 branch-prune event;
+ * for a total count of 8 (4 instructions * 2 branches/instruction).
+ */
+ static const int expected_counts[RTE_BPF_VALIDATE_DEBUG_EVENT_END] = {
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_STEP] = 9,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_START] = 1,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS] = 1,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER] = 4,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE] = 1,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN] = 4,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE] = 3,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS] = 1,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL] = 4,
+ };
+
+ struct test_events_context ctx = {
+ .ins = ins,
+ .nb_ins = RTE_DIM(ins),
+ .branch_pc = NO_PROGRAM_COUNTER,
+ };
+
+ struct rte_bpf_validate_debug *debug = rte_bpf_validate_debug_create();
+ TEST_ASSERT_NOT_NULL(debug, "validate debug create error");
+
+ /* Set breakpoints for all instructions. */
+ for (uint32_t pc = 0; pc != RTE_DIM(ins); ++pc)
+ rte_bpf_validate_debug_break(debug, pc,
+ &(struct rte_bpf_validate_debug_callback){
+ .fn = test_events_break_cb,
+ .ctx = &ctx,
+ });
+
+ /* Set catchpoints for all events except invalid-state. */
+ for (enum rte_bpf_validate_debug_event event = 0;
+ event != RTE_BPF_VALIDATE_DEBUG_EVENT_END; ++event)
+ if (event != RTE_BPF_VALIDATE_DEBUG_EVENT_INVALID_STATE)
+ rte_bpf_validate_debug_catch(debug, event,
+ &(struct rte_bpf_validate_debug_callback){
+ .fn = test_events_catch_cb,
+ .ctx = &ctx,
+ });
+
+ struct rte_bpf *const bpf = rte_bpf_load_ex(&(struct rte_bpf_prm_ex){
+ .sz = sizeof(struct rte_bpf_prm_ex),
+ .origin = RTE_BPF_ORIGIN_RAW,
+ .raw.ins = ins,
+ .raw.nb_ins = RTE_DIM(ins),
+ .prog_arg[0] = { .type = RTE_BPF_ARG_RAW },
+ .nb_prog_arg = 1,
+ .debug = debug,
+ });
+
+ rte_bpf_destroy(bpf);
+ rte_bpf_validate_debug_destroy(debug);
+
+ TEST_ASSERT_NOT_NULL(bpf, "validation failed");
+
+ /* Verify event counts. */
+ for (enum rte_bpf_validate_debug_event event = 0;
+ event != RTE_BPF_VALIDATE_DEBUG_EVENT_END; ++event)
+ TEST_ASSERT_EQUAL(
+ ctx.event_counts[event], expected_counts[event],
+ "expected %d %s events, but got %d",
+ expected_counts[event], event_names[event],
+ ctx.event_counts[event]);
+
+ /* Additionally double-check final stack state. */
+ TEST_ASSERT_EQUAL(ctx.branch_pc_stack_length, 0,
+ "unbalanced branch enters and returns");
+
+ return TEST_SUCCESS;
+}
+
+REGISTER_FAST_TEST(bpf_validate_events_autotest, NOHUGE_OK, ASAN_OK, test_events);
+
+/*
+ * Tests of BPF validation.
+ */
+
/* Sizes of text buffers used for formatting various debug outputs. */
#define VALUE_FORMAT_BUFFER_SIZE 32
#define INTERVAL_FORMAT_BUFFER_SIZE 64
--
2.43.0
^ permalink raw reply related [flat|nested] 38+ messages in thread
* [PATCH 7/7] app/validate-bpf: add BPF validation application
2026-09-11 10:39 [PATCH 0/7] validate-bpf: add standalone validation debug tool Marat Khalili
` (5 preceding siblings ...)
2026-09-11 10:40 ` [PATCH 6/7] app/test: add test for bpf validate debug events Marat Khalili
@ 2026-09-11 10:40 ` Marat Khalili
2026-09-14 14:54 ` [PATCH v2 0/7] validate-bpf: add standalone validation debug tool Marat Khalili
7 siblings, 0 replies; 38+ messages in thread
From: Marat Khalili @ 2026-09-11 10:40 UTC (permalink / raw)
To: Thomas Monjalon, Konstantin Ananyev; +Cc: dev
Add a new DPDK application, dpdk-validate-bpf, for pre-validating eBPF
programs for compatibility with the lib/bpf execution context.
The application allows evaluating eBPF programs against the DPDK verifier
before loading them into a real application. It includes an interactive
debugging mode to trace state changes per instruction and understand the
validator's decisions.
Signed-off-by: Marat Khalili <marat.khalili@huawei.com>
---
MAINTAINERS | 2 +
app/meson.build | 1 +
app/validate-bpf/alloc_list.c | 53 ++
app/validate-bpf/args.c | 263 ++++++
app/validate-bpf/debug.c | 1099 ++++++++++++++++++++++++
app/validate-bpf/debug_command.c | 383 +++++++++
app/validate-bpf/debug_command.h | 65 ++
app/validate-bpf/eal_init_args.c | 57 ++
app/validate-bpf/internal.h | 151 ++++
app/validate-bpf/main.c | 77 ++
app/validate-bpf/meson.build | 13 +
app/validate-bpf/parse_decl.c | 611 +++++++++++++
doc/guides/rel_notes/release_26_11.rst | 5 +
doc/guides/tools/index.rst | 1 +
doc/guides/tools/validate_bpf.rst | 97 +++
15 files changed, 2878 insertions(+)
create mode 100644 app/validate-bpf/alloc_list.c
create mode 100644 app/validate-bpf/args.c
create mode 100644 app/validate-bpf/debug.c
create mode 100644 app/validate-bpf/debug_command.c
create mode 100644 app/validate-bpf/debug_command.h
create mode 100644 app/validate-bpf/eal_init_args.c
create mode 100644 app/validate-bpf/internal.h
create mode 100644 app/validate-bpf/main.c
create mode 100644 app/validate-bpf/meson.build
create mode 100644 app/validate-bpf/parse_decl.c
create mode 100644 doc/guides/tools/validate_bpf.rst
diff --git a/MAINTAINERS b/MAINTAINERS
index e99a65d1974d..39cd0a3de6fa 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -1923,8 +1923,10 @@ F: lib/bpf/
F: examples/bpf/
F: app/test/test_bpf*
F: app/test/bpf/
+F: app/validate-bpf/
F: app/test-pmd/bpf_cmd.*
F: doc/guides/prog_guide/bpf_lib.rst
+F: doc/guides/tools/validate_bpf.rst
Graph
M: Jerin Jacob <jerinj@marvell.com>
diff --git a/app/meson.build b/app/meson.build
index 1798db3ae43f..9b59ae483056 100644
--- a/app/meson.build
+++ b/app/meson.build
@@ -33,6 +33,7 @@ apps = [
'test-regex',
'test-sad',
'test-security-perf',
+ 'validate-bpf',
]
if get_option('tests')
diff --git a/app/validate-bpf/alloc_list.c b/app/validate-bpf/alloc_list.c
new file mode 100644
index 000000000000..2f4d18ad3df6
--- /dev/null
+++ b/app/validate-bpf/alloc_list.c
@@ -0,0 +1,53 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include "internal.h"
+#include <stdlib.h>
+
+
+/* Needs to be a power of two. */
+#define START_CAPACITY (1u << 3)
+
+size_t
+alloc_list_append(struct alloc_list *alloc_list, void *ptr)
+{
+ if (alloc_list->count == 0) {
+ RTE_ASSERT(alloc_list->ptrs == NULL);
+ alloc_list->ptrs = malloc(
+ sizeof(alloc_list->ptrs[0]) * START_CAPACITY);
+ RTE_VERIFY(alloc_list->ptrs != NULL);
+ } else if (alloc_list->count >= START_CAPACITY &&
+ /* Power of two detection */
+ (alloc_list->count & (alloc_list->count - 1)) == 0) {
+ /*
+ * We allocate in powers of two, and current count is one of
+ * these powers, so need to reallocate to larger capacity.
+ */
+ RTE_ASSERT(alloc_list->ptrs != NULL);
+ const size_t new_capacity = alloc_list->count * 2;
+ alloc_list->ptrs = realloc(alloc_list->ptrs,
+ sizeof(alloc_list->ptrs[0]) * new_capacity);
+ RTE_VERIFY(alloc_list->ptrs != NULL);
+ }
+ alloc_list->ptrs[alloc_list->count] = ptr;
+ return alloc_list->count++;
+}
+
+void alloc_list_replace(struct alloc_list *alloc_list, size_t index, void *ptr)
+{
+ RTE_ASSERT(index < alloc_list->count);
+ alloc_list->ptrs[index] = ptr;
+}
+
+void alloc_list_free_all(struct alloc_list *alloc_list)
+{
+ /* Copy and clear fields first in case alloc_list itself gets freed. */
+ size_t count = alloc_list->count;
+ void ** const ptrs = alloc_list->ptrs;
+ *alloc_list = (struct alloc_list){};
+
+ while (count != 0)
+ free(ptrs[--count]);
+ free(ptrs);
+}
diff --git a/app/validate-bpf/args.c b/app/validate-bpf/args.c
new file mode 100644
index 000000000000..a23b9bfe152a
--- /dev/null
+++ b/app/validate-bpf/args.c
@@ -0,0 +1,263 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include "internal.h"
+
+#include <ctype.h>
+#include <getopt.h>
+#include <stdlib.h>
+
+#include <rte_errno.h>
+
+
+/* Values to be used in getopt_long option.val. */
+enum app_args {
+ ARG_UNRECOGNIZED = '?',
+ ARG_AUTO = 0, /* Value set by getopt_long when flag is non-NULL. */
+ ARG_HELP, /* Keep this one in the beginning for UX reasons. */
+ ARG_DEBUG,
+ ARG_MBUF_BUF_SIZE,
+ ARG_NO_PROG_ARGS,
+ ARG_PROG_ARG,
+ ARG_SECTION,
+ ARG_XSYM,
+};
+
+/* Program options. */
+static struct option OPTIONS[] = {
+ {
+ .name = "help",
+ .val = ARG_HELP,
+ },
+ {
+ .name = "debug",
+ .val = ARG_DEBUG,
+ },
+ {
+ .name = "mbuf-buf-size",
+ .has_arg = required_argument,
+ .val = ARG_MBUF_BUF_SIZE,
+ },
+ {
+ .name = "no-prog-arg",
+ .val = ARG_NO_PROG_ARGS,
+ },
+ {
+ .name = "no-prog-args",
+ .val = ARG_NO_PROG_ARGS,
+ },
+ {
+ .name = "prog-arg",
+ .has_arg = required_argument,
+ .val = ARG_PROG_ARG,
+ },
+ {
+ .name = "section",
+ .has_arg = required_argument,
+ .val = ARG_SECTION,
+ },
+ {
+ .name = "xsym",
+ .has_arg = required_argument,
+ .val = ARG_XSYM,
+ },
+ { /* terminating zero record */ }
+};
+
+/* Default args value. */
+static const struct args ARGS_DEFAULT = {
+ .bpf_prm = {
+ .sz = sizeof(struct rte_bpf_prm_ex),
+ .origin = RTE_BPF_ORIGIN_ELF_FILE,
+ .elf_file.section = ".text",
+ },
+ .mbuf_buf_size = RTE_MBUF_DEFAULT_BUF_SIZE,
+};
+
+/* Default --prog-arg argument value. */
+static const char * const PROG_ARG_DEFAULT = "struct rte_mbuf *";
+
+
+void
+print_usage(const char *program_name)
+{
+ static const char *const options[][2] = {
+ { "--help", "Display help and exit." },
+ { "--debug", "Enable interactive debug mode." },
+ { "--mbuf-buf-size=MBUF_BUF_SIZE", "Size of the mbuf data buffer (in bytes)." },
+ { "--prog-arg=TYPE", "Expected type of the next BPF program argument (up to 5)." },
+ { "--no-prog-args", "BPF program does not take any arguments." },
+ { "--section=SECTION", "ELF section name in the BPF object file to load." },
+ { "--xsym='TYPE NAME[(TYPE, ...)]'", "External symbol BPF program can access." },
+ };
+
+ printf("USAGE: %s [OPTIONS]... BPF_PATH\n", program_name);
+ printf("OPTIONS:\n");
+ for (int oi = 0; oi != RTE_DIM(options); ++oi)
+ printf("\t%-31s %s\n", options[oi][0], options[oi][1]);
+}
+
+void
+print_defaults(void)
+{
+ printf("DEFAULTS:"
+ " --mbuf-buf-size=%#zx"
+ " --prog-arg='%s'"
+ " --section='%s'\n",
+ (size_t)RTE_MBUF_DEFAULT_BUF_SIZE,
+ PROG_ARG_DEFAULT,
+ ARGS_DEFAULT.bpf_prm.elf_file.section);
+}
+
+static int
+parse_size(size_t *result, const char *text)
+{
+ char *parse_end = NULL;
+
+ errno = 0;
+ const uintmax_t strtoumax_result = strtoumax(text, &parse_end, 0);
+ if (errno != 0 || *parse_end != '\0' || strtoumax_result == 0 ||
+ strtoumax_result > SIZE_MAX)
+ return -1;
+
+ *result = strtoumax_result;
+ return 0;
+}
+
+struct args *
+args_parse(int argc, char *argv[])
+{
+ int val;
+ size_t xsym_alloc_index;
+ struct rte_bpf_xsym *xsym = NULL;
+ const char * const program_name = argv[0];
+
+ /* Allocate args and set aliases for some of its members. */
+ struct args * const args = malloc(sizeof(*args));
+ RTE_VERIFY(args != NULL);
+ struct alloc_list * const alloc_list = &args->_alloc_list;
+ struct rte_bpf_prm_ex * const bpf_prm = &args->bpf_prm;
+
+ /* Set default values. */
+ *args = ARGS_DEFAULT;
+ RTE_VERIFY(parse_arg(&bpf_prm->prog_arg[bpf_prm->nb_prog_arg++],
+ PROG_ARG_DEFAULT) >= 0);
+ bool default_prog_args = true;
+
+ /* Reserve space for xsym in alloc_list. */
+ xsym_alloc_index = alloc_list_append(alloc_list, xsym);
+
+ while ((val = getopt_long(argc, argv, "", OPTIONS, NULL)) != EOF) {
+ int rc = 0;
+ switch (val) {
+ case ARG_AUTO:
+ /* getopt_long made the assignment, nothing to do */
+ break;
+ case ARG_HELP:
+ args->show_help = true;
+ break;
+ case ARG_DEBUG:
+ if (bpf_prm->debug == NULL)
+ bpf_prm->debug = debug_create();
+ if (bpf_prm->debug == NULL) {
+ rc = -rte_errno;
+ fprintf(stderr,
+ "%s: error %d creating debug session\n",
+ program_name, -rc);
+ }
+ break;
+ case ARG_MBUF_BUF_SIZE:
+ rc = parse_size(&args->mbuf_buf_size, optarg);
+ if (rc < 0)
+ fprintf(stderr,
+ "%s: invalid mbuf buf size '%s'\n",
+ program_name, optarg);
+ break;
+ case ARG_NO_PROG_ARGS:
+ bpf_prm->nb_prog_arg = 0;
+ default_prog_args = false;
+ break;
+ case ARG_PROG_ARG:
+ if (default_prog_args) {
+ bpf_prm->nb_prog_arg = 0;
+ default_prog_args = false;
+ }
+
+ if (bpf_prm->nb_prog_arg == RTE_DIM(bpf_prm->prog_arg)) {
+ fprintf(stderr,
+ "%s: at most %d program arguments allowed\n",
+ program_name,
+ (int)RTE_DIM(bpf_prm->prog_arg));
+ rc = -EINVAL;
+ break;
+ }
+
+ rc = parse_arg(&bpf_prm->prog_arg[bpf_prm->nb_prog_arg++],
+ optarg);
+ if (rc < 0)
+ fprintf(stderr,
+ "%s: unrecognized prog arg '%s'\n",
+ program_name, optarg);
+
+ break;
+ case ARG_SECTION:
+ bpf_prm->elf_file.section = optarg;
+ break;
+ case ARG_XSYM:
+ bpf_prm->xsym = xsym = realloc(xsym,
+ sizeof(xsym[0]) * (bpf_prm->nb_xsym + 1));
+ RTE_VERIFY(xsym != NULL);
+ alloc_list_replace(alloc_list, xsym_alloc_index, xsym);
+ rc = parse_xsym(&xsym[bpf_prm->nb_xsym++], optarg,
+ alloc_list);
+ if (rc < 0)
+ fprintf(stderr, "%s: unrecognized xsym '%s'\n",
+ program_name, optarg);
+ break;
+ case ARG_UNRECOGNIZED:
+ args_destroy(args);
+ return NULL;
+ default:
+ rte_panic("Unexpected getopt_long return value %d\n",
+ val);
+ }
+ if (rc < 0) {
+ args_destroy(args);
+ return NULL;
+ }
+ }
+
+ /* Set buf_size to the value specified in command line arguments. */
+ adjust_arg_buf_size(bpf_prm->prog_arg, args->mbuf_buf_size);
+ for (size_t xsymi = 0; xsymi != bpf_prm->nb_xsym; ++xsymi)
+ adjust_xsym_buf_size(&xsym[xsymi], args->mbuf_buf_size);
+
+ /* getopt_long moves all non-options to the end, starting at optind. */
+ const int nb_bpf_path = argc - optind;
+ switch (nb_bpf_path) {
+ case 0:
+ break;
+ case 1:
+ bpf_prm->elf_file.path = argv[optind];
+ break;
+ default:
+ fprintf(stderr, "%s: too many positional arguments\n",
+ program_name);
+ args_destroy(args);
+ return NULL;
+ }
+
+ return args;
+}
+
+void
+args_destroy(struct args *args)
+{
+ if (args == NULL)
+ return;
+ if (args->bpf_prm.debug != NULL)
+ debug_destroy(args->bpf_prm.debug);
+ alloc_list_free_all(&args->_alloc_list);
+ free(args);
+}
diff --git a/app/validate-bpf/debug.c b/app/validate-bpf/debug.c
new file mode 100644
index 000000000000..a2d6a44868c8
--- /dev/null
+++ b/app/validate-bpf/debug.c
@@ -0,0 +1,1099 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include "debug_command.h"
+#include "internal.h"
+
+#include <rte_bpf_validate_debug.h>
+#include <rte_errno.h>
+
+#include <stdlib.h>
+
+
+/* Write single line to the user, currently just to stdout. */
+#define PRINTLN(fmt, ...) do { \
+ RTE_LOG_CHECK_NO_NEWLINE(fmt); \
+ printf(fmt "\n", ##__VA_ARGS__); \
+} while (0)
+
+#define PROMPT "(validate) "
+
+#define INITIAL_CAPACITY 8u
+
+static const char *const event_names[] = {
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_INVALID_STATE] = "invalid-state",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER] = "branch-enter",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE] = "branch-prune",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN] = "branch-return",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE] = "branch-unreachable",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS] = "jump-always",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL] = "jump-conditional",
+};
+
+static const char *const register_names[] = {
+ [EBPF_REG_0] = "r0",
+ [EBPF_REG_1] = "r1",
+ [EBPF_REG_2] = "r2",
+ [EBPF_REG_3] = "r3",
+ [EBPF_REG_4] = "r4",
+ [EBPF_REG_5] = "r5",
+ [EBPF_REG_6] = "r6",
+ [EBPF_REG_7] = "r7",
+ [EBPF_REG_8] = "r8",
+ [EBPF_REG_9] = "r9",
+ [EBPF_REG_10] = "r10",
+};
+
+static const char *const comparison_operator_names[] = {
+ [BPF_JEQ] = "==",
+ [BPF_JGT] = ">",
+ [BPF_JGE] = ">=",
+ [EBPF_JNE] = "!=",
+ [EBPF_JSGT] = "s>",
+ [EBPF_JSGE] = "s>=",
+ [EBPF_JLT] = "<",
+ [EBPF_JLE] = "<=",
+ [EBPF_JSLT] = "s<",
+ [EBPF_JSLE] = "s<=",
+};
+
+enum point_type {
+ POINT_TYPE_BREAK,
+ POINT_TYPE_CATCH,
+};
+
+/* Local representation of points with additional info for UI. */
+struct point_info {
+ struct rte_bpf_validate_debug_point *point;
+ enum point_type type;
+ union {
+ uint32_t pc;
+ enum rte_bpf_validate_debug_event event;
+ };
+};
+
+/* Dynamically growing list of point infos. */
+struct point_infos {
+ struct point_info *elements;
+ uint32_t length;
+ uint32_t capacity;
+};
+
+/* Information about a single conditional or unconditional jump in code path. */
+struct branch_info {
+ uint32_t jump_pc;
+ uint32_t target_pc;
+ bool is_conditional;
+};
+
+/* Dynamically growing stack to track code path. */
+struct branch_stack {
+ struct branch_info *branches;
+ uint32_t length;
+ uint32_t capacity;
+};
+
+/* Flags telling if we should validate again, stopping or not at start. */
+static bool validate_again;
+
+/* List of point infos; element index is its ID for UI. */
+static struct point_infos point_infos;
+
+/* Catchpoint invisible to the user used for step-by-step validation. */
+static struct rte_bpf_validate_debug_point *step_point;
+
+/* Tracking entered branches. */
+static struct branch_stack branch_stack;
+static uint32_t pending_jump_pc = UINT32_MAX;
+static struct rte_bpf_validate_debug_point *jump_always_step_point;
+
+static int
+step_cb(struct rte_bpf_validate_debug *debug, void *ctx);
+static int
+point_cb(struct rte_bpf_validate_debug *debug, void *ctx);
+
+/* Find index of the string in the list. */
+static int
+find_name(const char *name, const char *const *names, int nb_names)
+{
+ if (nb_names < 0)
+ return -EINVAL;
+
+ for (int index = 0; index != nb_names; ++index)
+ if (names[index] != NULL && strcmp(names[index], name) == 0)
+ return index;
+
+ return -ENOENT;
+}
+
+/* Convert event name to its enum value. */
+static enum rte_bpf_validate_debug_event
+parse_event_name(const char *name)
+{
+ const int rc = find_name(name, event_names, RTE_DIM(event_names));
+ if (rc < 0)
+ PRINTLN("Error: invalid event name.");
+
+ return rc;
+}
+
+/* Convert comparison operator name to opcode. */
+static int
+parse_comparison_operator(const char *name)
+{
+ const int rc = find_name(name, comparison_operator_names,
+ RTE_DIM(comparison_operator_names));
+ if (rc < 0)
+ PRINTLN("Error: invalid comparison operator name.");
+
+ return rc;
+}
+
+/* Convert register name to its number. */
+static int
+parse_register(const char *name)
+{
+ const int rc = find_name(name, register_names, RTE_DIM(register_names));
+ if (rc < 0)
+ PRINTLN("Error: invalid register name.");
+
+ return rc;
+}
+
+/* Free local list of point infos (do not destroy points). */
+static void
+point_infos_free(void)
+{
+ free(point_infos.elements);
+ point_infos = (struct point_infos){};
+}
+
+/* Free local branch stack. */
+static void
+branch_stack_free(void)
+{
+ free(branch_stack.branches);
+ branch_stack = (struct branch_stack){};
+}
+
+/* Return existing element from the list of point infos. */
+static struct point_info *
+point_infos_at(uint32_t point_number)
+{
+ RTE_ASSERT(point_number < point_infos.length);
+ RTE_ASSERT(point_infos.elements[point_number].point != NULL);
+ return &point_infos.elements[point_number];
+}
+
+/* Print existing element from the list of point infos. */
+static void
+point_infos_print_at(uint32_t point_number)
+{
+ const struct point_info *const point_info = point_infos_at(point_number);
+
+ switch (point_info->type) {
+ case POINT_TYPE_BREAK:
+ PRINTLN("Breakpoint %d at %d.", point_number,
+ point_info->pc);
+ break;
+ case POINT_TYPE_CATCH:
+ PRINTLN("Catchpoint %d on %s.", point_number,
+ event_names[point_info->event]);
+ break;
+ default:
+ PRINTLN("Point %d of unknown type", point_number);
+ break;
+ }
+}
+
+/* Print all point infos. */
+static int
+point_infos_print_all(void)
+{
+ uint32_t nb_printed = 0;
+ for (uint32_t pn = 0; pn != point_infos.length; ++pn) {
+ const struct point_info *const point_info =
+ &point_infos.elements[pn];
+ if (point_info->point != NULL) {
+ point_infos_print_at(pn);
+ ++nb_printed;
+ }
+ }
+
+ if (nb_printed == 0) {
+ PRINTLN("No breakpoints or catchpoints set.");
+ return -ENOENT;
+ }
+
+ return nb_printed;
+}
+
+/* Allocate space for a new element in the list of point infos. */
+static uint32_t
+point_infos_append(void)
+{
+ if (point_infos.length == point_infos.capacity) {
+ /* Set to initial capacity or double previous one. */
+ point_infos.capacity = RTE_MAX(INITIAL_CAPACITY,
+ point_infos.capacity * 2);
+ point_infos.elements = realloc(point_infos.elements,
+ point_infos.capacity * sizeof(point_infos.elements[0]));
+ RTE_VERIFY(point_infos.elements != NULL);
+ }
+ return point_infos.length++;
+}
+
+/* Allocate space for a new element in the branch stack. */
+static void
+branch_stack_append(const struct branch_info *branch)
+{
+ if (branch_stack.length == branch_stack.capacity) {
+ /* Set to initial capacity or double previous one. */
+ branch_stack.capacity = RTE_MAX(INITIAL_CAPACITY,
+ branch_stack.capacity * 2);
+ branch_stack.branches = realloc(branch_stack.branches,
+ branch_stack.capacity * sizeof(branch_stack.branches[0]));
+ RTE_VERIFY(branch_stack.branches != NULL);
+ }
+ branch_stack.branches[branch_stack.length++] = *branch;
+}
+
+/* Destroy existing element from the list of point infos, printing it first. */
+static void
+point_infos_destroy_existing(uint32_t point_number)
+{
+ point_infos_print_at(point_number);
+
+ struct point_info *const point_info = point_infos_at(point_number);
+ rte_bpf_validate_debug_point_destroy(point_info->point);
+ *point_info = (struct point_info){};
+}
+
+/* Destroy point with specified number if it exists. */
+static int
+point_infos_destroy_at(uint32_t point_number)
+{
+ if (point_number >= point_infos.length ||
+ point_infos.elements[point_number].point == NULL) {
+ PRINTLN("No breakpoint number %d.", point_number);
+ return -ENOENT;
+ }
+
+ point_infos_destroy_existing(point_number);
+ return 1;
+}
+
+/* Destroy all point infos. */
+static int64_t
+point_infos_destroy_all(void)
+{
+ uint32_t nb_destroyed = 0;
+ for (uint32_t pn = 0; pn < point_infos.length; ++pn) {
+ const struct point_info *const point_info =
+ &point_infos.elements[pn];
+ if (point_info->point != NULL) {
+ point_infos_destroy_existing(pn);
+ ++nb_destroyed;
+ }
+ }
+
+ if (nb_destroyed == 0) {
+ PRINTLN("No breakpoints or catchpoints set.");
+ return -ENOENT;
+ }
+
+ return nb_destroyed;
+}
+
+/* Destroy all breakpoints at specified location. */
+static int64_t
+point_infos_destroy_breakpoints(uint32_t pc)
+{
+ uint32_t nb_destroyed = 0;
+ for (uint32_t pn = 0; pn < point_infos.length; ++pn) {
+ const struct point_info *const point_info =
+ &point_infos.elements[pn];
+ if (point_info->point != NULL &&
+ point_info->type == POINT_TYPE_BREAK &&
+ point_info->pc == pc) {
+ point_infos_destroy_existing(pn);
+ ++nb_destroyed;
+ }
+ }
+
+ if (nb_destroyed == 0) {
+ PRINTLN("No breakpoint at %u.", pc);
+ return -ENOENT;
+ }
+
+ return nb_destroyed;
+}
+
+/* Destroy all catchpoints for specified event. */
+static int64_t
+point_infos_destroy_catchpoints(int event)
+{
+ if (event < 0)
+ /* Error was already printed by parse_event_name. */
+ return event;
+
+ uint32_t nb_destroyed = 0;
+ for (uint32_t pn = 0; pn < point_infos.length; ++pn) {
+ const struct point_info *const point_info =
+ &point_infos.elements[pn];
+ if (point_info->point != NULL &&
+ point_info->type == POINT_TYPE_CATCH &&
+ (int)point_info->event == event) {
+ point_infos_destroy_existing(pn);
+ ++nb_destroyed;
+ }
+ }
+
+ if (nb_destroyed == 0) {
+ PRINTLN("No catchpoint on %s.", event_names[event]);
+ return -ENOENT;
+ }
+
+ return nb_destroyed;
+}
+
+/* Create new breakpoint at specified location. */
+static int
+add_breakpoint(struct rte_bpf_validate_debug *debug, uint32_t nb_ins, uint32_t pc)
+{
+ const uint32_t point_number = point_infos_append();
+
+ if (pc >= nb_ins) {
+ PRINTLN("Error: program only has %u instructions.", nb_ins);
+ return -ENOENT;
+ }
+
+ struct rte_bpf_validate_debug_point *const point =
+ rte_bpf_validate_debug_break(debug, pc,
+ &(struct rte_bpf_validate_debug_callback){
+ .fn = point_cb,
+ .ctx = (void *)(uintptr_t)point_number,
+ });
+ if (point == NULL) {
+ PRINTLN("Library error %d.", rte_errno);
+ return -rte_errno;
+ }
+
+ point_infos.elements[point_number] = (struct point_info){
+ .point = point,
+ .type = POINT_TYPE_BREAK,
+ .pc = pc,
+ };
+ point_infos_print_at(point_number);
+ return 0;
+}
+
+/* Create new catchpoint at specified location. */
+static int
+add_catchpoint(struct rte_bpf_validate_debug *debug, int event)
+{
+ if (event < 0)
+ /* Error was already printed by parse_event_name. */
+ return event;
+
+ const uint32_t point_number = point_infos_append();
+ struct rte_bpf_validate_debug_point *const point =
+ rte_bpf_validate_debug_catch(debug, event,
+ &(struct rte_bpf_validate_debug_callback){
+ .fn = point_cb,
+ .ctx = (void *)(uintptr_t)point_number,
+ });
+ if (point == NULL) {
+ PRINTLN("Library error %d.", rte_errno);
+ return -rte_errno;
+ }
+
+ point_infos.elements[point_number] = (struct point_info){
+ .point = point,
+ .type = POINT_TYPE_CATCH,
+ .event = event,
+ };
+ point_infos_print_at(point_number);
+ return 0;
+}
+
+static bool
+is_step_enabled(void)
+{
+ return step_point != NULL;
+}
+
+/* Enable step-by-step validation: make sure catchpoint is set on step event. */
+static int
+enable_step(struct rte_bpf_validate_debug *debug)
+{
+ if (is_step_enabled())
+ return 0;
+
+ step_point = rte_bpf_validate_debug_catch(debug,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_STEP,
+ &(struct rte_bpf_validate_debug_callback){ step_cb });
+ if (step_point == NULL)
+ return -rte_errno;
+
+ return 0;
+}
+
+/* Disable step-by-step validation: destroy catchpoint on step event if any. */
+static void
+disable_step(void)
+{
+ rte_bpf_validate_debug_point_destroy(step_point);
+ step_point = NULL;
+}
+
+/* Format and print information about specified frame offset. */
+static int
+print_frame_offset(struct rte_bpf_validate_debug *debug, int32_t offset)
+{
+ char *info;
+ int info_size, rc;
+
+ if (offset >= 0 || offset % sizeof(uint64_t) != 0) {
+ PRINTLN("Invalid frame offset, must be a negative multiple of %zu.",
+ sizeof(uint64_t));
+ return -EINVAL;
+ }
+
+ rc = rte_bpf_validate_debug_format_frame_info(debug, NULL, 0, offset);
+ if (rc == -ERANGE) {
+ PRINTLN("Offset is out of frame range.");
+ return rc;
+ }
+ if (rc < 0) {
+ PRINTLN("Error %d printing information.", -rc);
+ return rc;
+ }
+
+ info_size = rc + 1;
+ info = malloc(info_size);
+ if (info == NULL)
+ return -ENOMEM;
+
+ rc = rte_bpf_validate_debug_format_frame_info(debug, info, info_size,
+ offset);
+ if (rc + 1 != info_size) {
+ if (rc >= 0) {
+ PRINTLN("Expect format return value %d, got %d.",
+ info_size, rc);
+ rc = -EINVAL;
+ } else
+ PRINTLN("Error %d printing information.", -rc);
+ free(info);
+ return rc;
+ }
+
+ printf("%5jd: \t%s\n", (intmax_t)offset, info);
+ free(info);
+ return 0;
+}
+
+/* Format and print informatiion about the frame. */
+static int
+print_frame(struct rte_bpf_validate_debug *debug)
+{
+ int32_t frame_size;
+ int rc;
+
+ frame_size = rte_bpf_validate_debug_get_frame_size(debug);
+ if (frame_size < 0) {
+ PRINTLN("Error %d getting frame size.", -frame_size);
+ return frame_size;
+ }
+
+ for (int32_t frame_offset = 0;;) {
+ frame_offset -= sizeof(uint64_t);
+ if (frame_offset < -frame_size)
+ break;
+
+ rc = print_frame_offset(debug, frame_offset);
+ if (rc < 0)
+ return rc;
+ }
+
+ return 0;
+}
+
+/* Format and print informatiion about specified register. */
+static int
+print_register(struct rte_bpf_validate_debug *debug, int reg)
+{
+ char *info;
+ int info_size, rc;
+
+ if (reg < 0)
+ /* Error was already printed by parse_register. */
+ return reg;
+
+ rc = rte_bpf_validate_debug_format_register_info(debug, NULL, 0, reg);
+ if (rc < 0) {
+ PRINTLN("Error %d printing information.", -rc);
+ return rc;
+ }
+
+ info_size = rc + 1;
+ info = malloc(info_size);
+ if (info == NULL)
+ return -ENOMEM;
+
+ rc = rte_bpf_validate_debug_format_register_info(debug, info, info_size,
+ reg);
+ if (rc + 1 != info_size) {
+ if (rc >= 0) {
+ PRINTLN("Expect format return value %d, got %d.",
+ info_size, rc);
+ rc = -EINVAL;
+ } else
+ PRINTLN("Error %d printing information.", -rc);
+ free(info);
+ return rc;
+ }
+
+ printf("%5s: \t%s\n", register_names[reg], info);
+ free(info);
+ return 0;
+}
+
+/* Format and print informatiion about all registers. */
+static int
+print_registers(struct rte_bpf_validate_debug *debug)
+{
+ int rc = 0;
+
+ for (int reg = 0; reg != EBPF_REG_NUM; ++reg)
+ rc = rc < 0 ? rc : print_register(debug, reg);
+
+ return rc;
+}
+
+/* List one eBPF program instruction. */
+static int
+list_one(const struct ebpf_insn *ins, uint32_t nb_ins, uint32_t offset,
+ uint32_t pc, const char *comment)
+{
+ char hexadecimal[256], disassembly[256];
+
+ if (offset >= nb_ins) {
+ PRINTLN("Error: program only has %u instructions.", nb_ins);
+ return -EINVAL;
+ }
+
+ ins += offset;
+
+ if (offset == nb_ins - 1 && rte_bpf_insn_is_wide(ins)) {
+ PRINTLN("Error: truncated last instruction.");
+ return -EINVAL;
+ }
+
+ rte_bpf_format(hexadecimal, sizeof(hexadecimal), ins, 0,
+ RTE_BPF_FORMAT_FLAG_HEXADECIMAL |
+ RTE_BPF_FORMAT_FLAG_NEVER_WIDE);
+ rte_bpf_format(disassembly, sizeof(disassembly), ins, offset,
+ RTE_BPF_FORMAT_FLAG_DISASSEMBLY |
+ RTE_BPF_FORMAT_FLAG_ABSOLUTE_JUMPS);
+
+ if (comment == NULL)
+ comment = "";
+ PRINTLN("%2s %10u: \t%s \t%s%s%s",
+ offset == pc ? "=>" : "", offset, hexadecimal, disassembly,
+ comment[0] != '\0' ? " \t; " : "", comment);
+
+ if (rte_bpf_insn_is_wide(ins)) {
+ rte_bpf_format(hexadecimal, sizeof(hexadecimal), ins + 1, 0,
+ RTE_BPF_FORMAT_FLAG_HEXADECIMAL |
+ RTE_BPF_FORMAT_FLAG_NEVER_WIDE);
+ PRINTLN("%15s\t%s", "", hexadecimal);
+ }
+
+ return 0;
+}
+
+/* List specified range of eBPF program instructions, updating start offset. */
+static int
+list(const struct ebpf_insn *ins, uint32_t nb_ins, uint32_t *offset,
+ uint32_t count, uint32_t pc)
+{
+ uint32_t local_offset = 0;
+ if (offset == NULL)
+ offset = &local_offset;
+
+ if (*offset > nb_ins) {
+ PRINTLN("Error: program only has %u instructions.", nb_ins);
+ return -EINVAL;
+ }
+
+ const uint32_t end =
+ /* Calculate end in a way preventing overflow: */
+ *offset + RTE_MIN(nb_ins - *offset, count);
+ while (*offset < end) {
+ const int rc = list_one(ins, nb_ins, *offset, pc, NULL);
+ if (rc < 0)
+ return rc;
+
+ *offset += 1 + rte_bpf_insn_is_wide(&ins[*offset]);
+ }
+
+ return 0;
+}
+
+/* Print if specified conditional jump _may_ be executed. */
+static int
+print_if_may(struct rte_bpf_validate_debug *debug, const struct ebpf_insn *jump,
+ uint64_t imm64)
+{
+ const int result = rte_bpf_validate_debug_may_jump(debug, jump, imm64);
+
+ switch (result) {
+ case 0:
+ case RTE_BPF_VALIDATE_DEBUG_MAY_BE_FALSE:
+ PRINTLN("NO");
+ break;
+ case RTE_BPF_VALIDATE_DEBUG_MAY_BE_TRUE:
+ case RTE_BPF_VALIDATE_DEBUG_MAY_BE_FALSE | RTE_BPF_VALIDATE_DEBUG_MAY_BE_TRUE:
+ PRINTLN("YES");
+ break;
+ default:
+ PRINTLN("Error %d getting result.", -result);
+ break;
+ }
+
+ return result;
+}
+
+/* Print if specified condition with literal right hand side _may_ be true. */
+static int
+print_if_may_literal_rhs(struct rte_bpf_validate_debug *debug,
+ const struct debug_command_comparison *comparison)
+{
+ const int lhs = parse_register(comparison->lhs);
+ const int op = parse_comparison_operator(comparison->op);
+ const int64_t rhs = comparison->literal_rhs;
+
+ if (lhs < 0 || op < 0)
+ /* Error was already printed by parse function. */
+ return lhs < 0 ? lhs : op;
+
+ return print_if_may(debug, &(struct ebpf_insn){
+ .code = BPF_JMP | op | BPF_K,
+ .dst_reg = lhs,
+ }, /* imm64 = */ rhs);
+}
+
+/* Print if specified condition with register right hand side _may_ be true. */
+static int
+print_if_may_register_rhs(struct rte_bpf_validate_debug *debug,
+ const struct debug_command_comparison *comparison)
+{
+ const int lhs = parse_register(comparison->lhs);
+ const int op = parse_comparison_operator(comparison->op);
+ const int rhs = parse_register(comparison->register_rhs);
+
+ if (lhs < 0 || op < 0 || rhs < 0)
+ /* Error was already printed by parse function. */
+ return lhs < 0 ? lhs : op < 0 ? op : rhs;
+
+ return print_if_may(debug, &(struct ebpf_insn){
+ .code = BPF_JMP | op | BPF_X,
+ .dst_reg = lhs,
+ .src_reg = rhs,
+ }, /* imm64 = */ 0);
+}
+
+/* Return 1 on validation success, 0 on failure, -EAGAIN if still running. */
+static int
+get_validation_success(struct rte_bpf_validate_debug *debug)
+{
+ int validation_result, rc;
+
+ rc = rte_bpf_validate_debug_get_validation_result(debug,
+ &validation_result);
+ return rc < 0 ? rc : (validation_result >= 0);
+}
+
+static void
+print_status(const struct ebpf_insn *ins, uint32_t nb_ins,
+ int validation_success, uint32_t pc)
+{
+ if (validation_success == 1) {
+ PRINTLN("Validation succeeded.");
+ return;
+ }
+
+ list_one(ins, nb_ins, pc, pc, NULL);
+
+ if (validation_success == 0)
+ PRINTLN("Validation failed.");
+}
+
+static void
+debug_command_where(const struct ebpf_insn *ins, uint32_t nb_ins,
+ int validation_success, uint32_t pc)
+{
+ for (uint32_t bi = 0; bi != branch_stack.length; ++bi) {
+ const uint32_t jump_pc = branch_stack.branches[bi].jump_pc;
+ const uint32_t target_pc = branch_stack.branches[bi].target_pc;
+ const char *const comment =
+ !branch_stack.branches[bi].is_conditional ? NULL :
+ target_pc == jump_pc + 1 ? "fallen-through" : "taken";
+ list_one(ins, nb_ins, jump_pc, UINT32_MAX, comment);
+ }
+ print_status(ins, nb_ins, validation_success, pc);
+}
+
+/* Return true if pc is defined, otherwise print an error message. */
+static bool
+ensure_pc_defined(uint32_t pc)
+{
+ if (pc != UINT32_MAX)
+ return true;
+
+ PRINTLN("No current instruction.");
+ return false;
+}
+
+/* Return true if still validating, otherwise print an error message. */
+static bool
+ensure_still_validating(int validation_success)
+{
+ if (validation_success == -EAGAIN)
+ return true;
+
+ PRINTLN("Finished, use `start` or `run` to restart, `quit` to quit.");
+ return false;
+}
+
+/* Step-by-step validation callback: read and process user commands. */
+static int
+step_cb(struct rte_bpf_validate_debug *debug, __rte_unused void *ctx)
+{
+ int rc;
+ int validation_success;
+ const struct ebpf_insn *ins;
+ uint32_t nb_ins, pc, list_offset;
+
+ validate_again = false;
+
+ rc = rte_bpf_validate_debug_get_ins(debug, &ins, &nb_ins);
+ if (rc < 0) {
+ PRINTLN("Error %d getting program instructions.", -rc);
+ return rc;
+ }
+
+ validation_success = get_validation_success(debug);
+ if (validation_success < 0 && validation_success != -EAGAIN) {
+ PRINTLN("Error %d getting validation result.",
+ -validation_success);
+ return validation_success;
+ }
+
+ pc = validation_success == 1 ? UINT32_MAX :
+ rte_bpf_validate_debug_get_pc(debug);
+
+ print_status(ins, nb_ins, validation_success, pc);
+
+ list_offset = pc;
+
+ while (true) {
+ switch (debug_command_get(PROMPT)) {
+ case DEBUG_COMMAND_BREAK:
+ if (ensure_pc_defined(pc))
+ add_breakpoint(debug, nb_ins, pc);
+ continue;
+ case DEBUG_COMMAND_BREAK_PC:
+ add_breakpoint(debug, nb_ins, debug_command_parsed.pc);
+ continue;
+ case DEBUG_COMMAND_CATCH:
+ add_catchpoint(debug,
+ parse_event_name(debug_command_parsed.event));
+ continue;
+ case DEBUG_COMMAND_CLEAR:
+ if (ensure_pc_defined(pc))
+ point_infos_destroy_breakpoints(pc);
+ continue;
+ case DEBUG_COMMAND_CLEAR_EVENT:
+ point_infos_destroy_catchpoints(
+ parse_event_name(debug_command_parsed.event));
+ continue;
+ case DEBUG_COMMAND_CLEAR_PC:
+ point_infos_destroy_breakpoints(
+ debug_command_parsed.pc);
+ continue;
+ case DEBUG_COMMAND_CONTINUE:
+ if (ensure_still_validating(validation_success)) {
+ disable_step();
+ return 0;
+ }
+ continue;
+ case DEBUG_COMMAND_DELETE:
+ point_infos_destroy_all();
+ continue;
+ case DEBUG_COMMAND_DELETE_NUMBER:
+ point_infos_destroy_at(
+ debug_command_parsed.point_number);
+ continue;
+ case DEBUG_COMMAND_INFO_FRAME:
+ print_frame(debug);
+ continue;
+ case DEBUG_COMMAND_INFO_FRAME_OFFSET:
+ print_frame_offset(debug,
+ debug_command_parsed.frame_offset);
+ continue;
+ case DEBUG_COMMAND_INFO_POINTS:
+ point_infos_print_all();
+ continue;
+ case DEBUG_COMMAND_INFO_REGISTER:
+ print_register(debug,
+ parse_register(debug_command_parsed.register_));
+ continue;
+ case DEBUG_COMMAND_INFO_REGISTERS:
+ print_registers(debug);
+ continue;
+ case DEBUG_COMMAND_LIST:
+ if (ensure_pc_defined(pc))
+ list(ins, nb_ins, &list_offset, 10, pc);
+ continue;
+ case DEBUG_COMMAND_LIST_COUNT:
+ if (ensure_pc_defined(pc))
+ list(ins, nb_ins, &list_offset,
+ debug_command_parsed.instruction_count, pc);
+ continue;
+ case DEBUG_COMMAND_LIST_PROGRAM:
+ if (ensure_pc_defined(pc))
+ list(ins, nb_ins, NULL, UINT32_MAX, pc);
+ continue;
+ case DEBUG_COMMAND_MAY_LITERAL_RHS:
+ print_if_may_literal_rhs(debug,
+ &debug_command_parsed.comparison);
+ continue;
+ case DEBUG_COMMAND_MAY_REGISTER_RHS:
+ print_if_may_register_rhs(debug,
+ &debug_command_parsed.comparison);
+ continue;
+ case DEBUG_COMMAND_EOF:
+ case DEBUG_COMMAND_QUIT:
+ PRINTLN("Quitting...");
+ return validation_success == -EAGAIN ? -ECANCELED : 0;
+ case DEBUG_COMMAND_RUN:
+ PRINTLN("Re-running...");
+ disable_step();
+ validate_again = true;
+ return -ECANCELED;
+ case DEBUG_COMMAND_START:
+ PRINTLN("Re-starting...");
+ enable_step(debug);
+ validate_again = true;
+ return -ECANCELED;
+ case DEBUG_COMMAND_STEP:
+ if (ensure_still_validating(validation_success))
+ return 0;
+ continue;
+ case DEBUG_COMMAND_WHERE:
+ debug_command_where(ins, nb_ins, validation_success, pc);
+ continue;
+ default:
+ PRINTLN("INTERNAL ERROR");
+ return -ENOTSUP;
+ }
+ }
+}
+
+/* Any point callback: print it and enable step-by-step validation. */
+static int
+point_cb(struct rte_bpf_validate_debug *debug, void *ctx)
+{
+ const uint32_t point_number = (uintptr_t)ctx;
+ point_infos_print_at(point_number);
+ return enable_step(debug);
+}
+
+/* Branch stack machinery. */
+
+static void
+clear_jump_always_step_point(void)
+{
+ rte_bpf_validate_debug_point_destroy(jump_always_step_point);
+ jump_always_step_point = NULL;
+}
+
+static int
+reset_branch_tracking(struct rte_bpf_validate_debug *debug __rte_unused,
+ void *ctx __rte_unused)
+{
+ branch_stack.length = 0;
+ pending_jump_pc = UINT32_MAX;
+ clear_jump_always_step_point();
+ return 0;
+}
+
+/*
+ * Handling the jump-always instructions:
+ * - upon a jump-always event save the pc and set a custom step callback;
+ * - ignore the first custom step callback call (still on the same jump);
+ * - on the second call push the jump pc into stack and delete the callback;
+ */
+
+static int
+jump_always_step_cb(struct rte_bpf_validate_debug *debug, void *ctx __rte_unused)
+{
+ /* Step event is also emitted at the end of the jump instruction itself. */
+ if (rte_bpf_validate_debug_get_pc(debug) == pending_jump_pc)
+ return 0;
+
+ branch_stack_append(&(struct branch_info){
+ .jump_pc = pending_jump_pc,
+ .target_pc = rte_bpf_validate_debug_get_pc(debug),
+ .is_conditional = false,
+ });
+ clear_jump_always_step_point();
+ return 0;
+}
+
+static int
+jump_always_cb(struct rte_bpf_validate_debug *debug, void *ctx __rte_unused)
+{
+ pending_jump_pc = rte_bpf_validate_debug_get_pc(debug);
+ clear_jump_always_step_point();
+ jump_always_step_point = rte_bpf_validate_debug_catch(debug,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_STEP,
+ &(struct rte_bpf_validate_debug_callback){ jump_always_step_cb });
+ return 0;
+}
+
+/*
+ * Handling the jump-conditional instructions:
+ * - upon a jump-conditional event save the pc;
+ * - upon a branch-enter event push the conditional jump pc into stack;
+ * - upon a branch-return event pop conditional jump and all unconditional jumps
+ * preceding it from the stack;
+ * - during step execution notify the user about the step events;
+ */
+
+static int
+jump_conditional_cb(struct rte_bpf_validate_debug *debug, void *ctx __rte_unused)
+{
+ pending_jump_pc = rte_bpf_validate_debug_get_pc(debug);
+ return 0;
+}
+
+static int
+branch_enter_cb(struct rte_bpf_validate_debug *debug, void *ctx __rte_unused)
+{
+ branch_stack_append(&(struct branch_info){
+ .jump_pc = pending_jump_pc,
+ .target_pc = rte_bpf_validate_debug_get_pc(debug),
+ .is_conditional = true,
+ });
+ if (is_step_enabled())
+ PRINTLN("Entered new branch at pc %u.", pending_jump_pc);
+ return 0;
+}
+
+static int
+branch_return_cb(struct rte_bpf_validate_debug *debug __rte_unused,
+ void *ctx __rte_unused)
+{
+ clear_jump_always_step_point();
+
+ while (branch_stack.length > 0) {
+ const struct branch_info branch_info =
+ branch_stack.branches[--branch_stack.length];
+ pending_jump_pc = branch_info.jump_pc;
+ if (branch_info.is_conditional)
+ break;
+ }
+ if (is_step_enabled())
+ PRINTLN("Returned from branch at pc %u.", pending_jump_pc);
+ return 0;
+}
+
+/* Notify user when skipping branches in step mode. */
+
+static int
+branch_prune_cb(struct rte_bpf_validate_debug *debug __rte_unused,
+ void *ctx __rte_unused)
+{
+ if (is_step_enabled())
+ PRINTLN("Prunned branch at pc %u.", pending_jump_pc);
+ return 0;
+}
+
+static int
+branch_unreachable_cb(struct rte_bpf_validate_debug *debug __rte_unused,
+ void *ctx __rte_unused)
+{
+ if (is_step_enabled())
+ PRINTLN("Unreachable branch at pc %u.", pending_jump_pc);
+ return 0;
+}
+
+/* Global initialization and cleanup functions. */
+
+static int
+set_callbacks(struct rte_bpf_validate_debug *debug)
+{
+ static const struct rte_bpf_validate_debug_callback events_callback[
+ RTE_BPF_VALIDATE_DEBUG_EVENT_END] = {
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_START] = { reset_branch_tracking },
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS] = { step_cb },
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_FAILURE] = { step_cb },
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL] = { jump_conditional_cb },
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS] = { jump_always_cb },
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER] = { branch_enter_cb },
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN] = { branch_return_cb },
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE] = { branch_prune_cb },
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE] = { branch_unreachable_cb },
+ };
+
+ for (enum rte_bpf_validate_debug_event event = 0;
+ event < RTE_BPF_VALIDATE_DEBUG_EVENT_END; ++event)
+ if (events_callback[event].fn != NULL &&
+ rte_bpf_validate_debug_catch(debug, event,
+ &events_callback[event]) == NULL)
+ return -rte_errno;
+
+ return 0;
+}
+
+struct rte_bpf_validate_debug *
+debug_create(void)
+{
+
+ int rc = 0;
+
+ struct rte_bpf_validate_debug *const debug = rte_bpf_validate_debug_create();
+ if (debug == NULL)
+ rc = -rte_errno;
+
+ rc = rc < 0 ? rc : set_callbacks(debug);
+
+ rc = rc < 0 ? rc : enable_step(debug);
+
+ if (rc < 0) {
+ debug_destroy(debug);
+ rte_errno = -rc;
+ return NULL;
+ }
+
+ return debug;
+}
+
+void
+debug_destroy(struct rte_bpf_validate_debug *debug)
+{
+ /* No need to destroy created points, destroying debug will do it. */
+ point_infos_free();
+ branch_stack_free();
+ rte_bpf_validate_debug_destroy(debug);
+}
+
+bool
+debug_validate_again(void)
+{
+ return validate_again;
+}
diff --git a/app/validate-bpf/debug_command.c b/app/validate-bpf/debug_command.c
new file mode 100644
index 000000000000..505ae389c94f
--- /dev/null
+++ b/app/validate-bpf/debug_command.c
@@ -0,0 +1,383 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include "debug_command.h"
+
+#include <rte_debug.h>
+#include <cmdline.h>
+#include <cmdline_socket.h>
+#include <cmdline_parse_num.h>
+
+#include <string.h>
+
+
+#define EVENT_PATTERN \
+ "invalid-state#" \
+ "branch-enter#branch-prune#branch-return#branch-unreachable#" \
+ "jump-always#jump-conditional"
+
+#define REGISTER_PATTERN "r0#r1#r2#r3#r4#r5#r6#r7#r8#r9#r10"
+
+#define COMPARISON_OP_PATTERN "==#!=#<#<=#>#>=#s<#s<=#s>#s>="
+
+static void
+handle_command(void *parsed_result, struct cmdline *cl, void *data);
+
+/* Keywords */
+static cmdline_parse_token_string_t break_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "b#break");
+static cmdline_parse_token_string_t may_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "may");
+static cmdline_parse_token_string_t catch_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "catch");
+static cmdline_parse_token_string_t clear_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "clear");
+static cmdline_parse_token_string_t continue_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "c#continue");
+static cmdline_parse_token_string_t delete_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "delete");
+static cmdline_parse_token_string_t info_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "i#info");
+static cmdline_parse_token_string_t info_points_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword,
+ "b#break#breakpoints#points");
+static cmdline_parse_token_string_t info_frame_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "f#frame");
+static cmdline_parse_token_string_t info_registers_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "r#registers");
+static cmdline_parse_token_string_t list_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "l#list");
+static cmdline_parse_token_string_t program_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "program");
+static cmdline_parse_token_string_t quit_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "q#quit");
+static cmdline_parse_token_string_t run_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "run");
+static cmdline_parse_token_string_t start_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "start");
+static cmdline_parse_token_string_t step_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "s#step");
+static cmdline_parse_token_string_t where_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "where");
+
+/* Variable tokens */
+static cmdline_parse_token_string_t comparison_lhs_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, comparison.lhs, REGISTER_PATTERN);
+static cmdline_parse_token_string_t comparison_op_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, comparison.op, COMPARISON_OP_PATTERN);
+static cmdline_parse_token_num_t comparison_rhs_literal_tok =
+ TOKEN_NUM_INITIALIZER(struct debug_command_parsed, comparison.literal_rhs, RTE_INT64);
+static cmdline_parse_token_string_t comparison_rhs_register_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, comparison.register_rhs,
+ REGISTER_PATTERN);
+static cmdline_parse_token_string_t event_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, event, EVENT_PATTERN);
+static cmdline_parse_token_num_t frame_offset_tok =
+ TOKEN_NUM_INITIALIZER(struct debug_command_parsed, frame_offset, RTE_INT64);
+static cmdline_parse_token_num_t instruction_count_tok =
+ TOKEN_NUM_INITIALIZER(struct debug_command_parsed, instruction_count, RTE_UINT32);
+static cmdline_parse_token_num_t pc_tok =
+ TOKEN_NUM_INITIALIZER(struct debug_command_parsed, instruction_offset, RTE_UINT32);
+static cmdline_parse_token_num_t point_number_tok =
+ TOKEN_NUM_INITIALIZER(struct debug_command_parsed, point_number, RTE_UINT32);
+static cmdline_parse_token_string_t register_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, register_, REGISTER_PATTERN);
+
+
+/* Commands */
+static cmdline_parse_inst_t cmd_break = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_BREAK,
+ .help_str = "b|break: break at current instruction",
+ .tokens = {
+ (void *)&break_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_break_pc = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_BREAK_PC,
+ .help_str = "b|break <pc>: break at specified instruction",
+ .tokens = {
+ (void *)&break_tok,
+ (void *)&pc_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_catch = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_CATCH,
+ .help_str = "catch <event>: catch specified event",
+ .tokens = {
+ (void *)&catch_tok,
+ (void *)&event_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_clear = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_CLEAR,
+ .help_str = "clear: delete all breakpoints at current instruction",
+ .tokens = {
+ (void *)&clear_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_clear_event = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_CLEAR_EVENT,
+ .help_str = "clear <event>: delete all catchpoints for specified event",
+ .tokens = {
+ (void *)&clear_tok,
+ (void *)&event_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_clear_pc = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_CLEAR_PC,
+ .help_str = "clear <pc>: delete all breakpoints at specified instruction",
+ .tokens = {
+ (void *)&clear_tok,
+ (void *)&pc_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_continue = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_CONTINUE,
+ .help_str = "c|continue: continue validation",
+ .tokens = {
+ (void *)&continue_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_delete = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_DELETE,
+ .help_str = "delete: delete all breakpoints and catchpoints",
+ .tokens = {
+ (void *)&delete_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_delete_number = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_DELETE_NUMBER,
+ .help_str = "delete <point>: delete specified breakpoint or catchpoint",
+ .tokens = {
+ (void *)&delete_tok,
+ (void *)&point_number_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_info_frame = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_INFO_FRAME,
+ .help_str = "i|info f|frame: show information about all frame locations",
+ .tokens = {
+ (void *)&info_tok,
+ (void *)&info_frame_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_info_frame_offset = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_INFO_FRAME_OFFSET,
+ .help_str = "i|info f|frame -<offset>: show information about specified frame location",
+ .tokens = {
+ (void *)&info_tok,
+ (void *)&info_frame_tok,
+ (void *)&frame_offset_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_info_points = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_INFO_POINTS,
+ .help_str = "i|info b|break|breakpoints|points: "
+ "show information about all breakpoints and catchpoints",
+ .tokens = {
+ (void *)&info_tok,
+ (void *)&info_points_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_info_register = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_INFO_REGISTER,
+ .help_str = "i|info <register>: show information about specified register",
+ .tokens = {
+ (void *)&info_tok,
+ (void *)®ister_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_info_registers = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_INFO_REGISTERS,
+ .help_str = "i|info r|registers: show information about all registers",
+ .tokens = {
+ (void *)&info_tok,
+ (void *)&info_registers_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_list = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_LIST,
+ .help_str = "l|list: list ten instructions",
+ .tokens = {
+ (void *)&list_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_list_count = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_LIST_COUNT,
+ .help_str = "l|list <number>: list specified number of instructions",
+ .tokens = {
+ (void *)&list_tok,
+ (void *)&instruction_count_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_list_program = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_LIST_PROGRAM,
+ .help_str = "l|list program: list whole program",
+ .tokens = {
+ (void *)&list_tok,
+ (void *)&program_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_may_literal_rhs = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_MAY_LITERAL_RHS,
+ .help_str = "may <register> <comparison> <number>: "
+ "check if specified condition _may_ be true",
+ .tokens = {
+ (void *)&may_tok,
+ (void *)&comparison_lhs_tok,
+ (void *)&comparison_op_tok,
+ (void *)&comparison_rhs_literal_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_may_register_rhs = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_MAY_REGISTER_RHS,
+ .help_str = "may <register> <comparison> <register>: "
+ "check if specified condition _may_ be true",
+ .tokens = {
+ (void *)&may_tok,
+ (void *)&comparison_lhs_tok,
+ (void *)&comparison_op_tok,
+ (void *)&comparison_rhs_register_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_quit = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_QUIT,
+ .help_str = "quit: q|quit debugger",
+ .tokens = {
+ (void *)&quit_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_run = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_RUN,
+ .help_str = "run: re-run validation from the start",
+ .tokens = {
+ (void *)&run_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_start = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_START,
+ .help_str = "start: re-start validation and stop at start",
+ .tokens = {
+ (void *)&start_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_step = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_STEP,
+ .help_str = "s|step: validate one instruction",
+ .tokens = {
+ (void *)&step_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_where = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_WHERE,
+ .help_str = "where: show current branch stack",
+ .tokens = {
+ (void *)&where_tok,
+ NULL,
+ }
+};
+
+static cmdline_parse_ctx_t debug_ctx[] = {
+ &cmd_break,
+ &cmd_break_pc,
+ &cmd_catch,
+ &cmd_clear,
+ &cmd_clear_event,
+ &cmd_clear_pc,
+ &cmd_continue,
+ &cmd_delete,
+ &cmd_delete_number,
+ &cmd_info_frame,
+ &cmd_info_frame_offset,
+ &cmd_info_points,
+ &cmd_info_register,
+ &cmd_info_registers,
+ &cmd_list,
+ &cmd_list_count,
+ &cmd_list_program,
+ &cmd_may_literal_rhs,
+ &cmd_may_register_rhs,
+ &cmd_quit,
+ &cmd_run,
+ &cmd_start,
+ &cmd_step,
+ &cmd_where,
+ NULL
+};
+
+/* Receive, fill and return one command. */
+
+static enum debug_command debug_command;
+
+struct debug_command_parsed debug_command_parsed;
+
+static void
+handle_command(void *parsed_result, struct cmdline *cl, void *data)
+{
+ RTE_BUILD_BUG_ON(sizeof(debug_command_parsed) > CMDLINE_PARSE_RESULT_BUFSIZE);
+ memcpy(&debug_command_parsed, parsed_result, sizeof(debug_command_parsed));
+ debug_command = (uintptr_t)data;
+ cmdline_quit(cl);
+}
+
+enum debug_command
+debug_command_get(const char *prompt)
+{
+ debug_command = DEBUG_COMMAND_EOF;
+ struct cmdline *const cmdline = cmdline_stdin_new(debug_ctx, prompt);
+ RTE_VERIFY(cmdline != NULL);
+ cmdline_interact(cmdline);
+ cmdline_stdin_exit(cmdline);
+ /* Clear prompt, or it would prepend first message that follows. */
+ printf("\r%*s\r", (int)strlen(prompt), "");
+ fflush(stdout);
+ return debug_command;
+}
diff --git a/app/validate-bpf/debug_command.h b/app/validate-bpf/debug_command.h
new file mode 100644
index 000000000000..1a59ab251ebb
--- /dev/null
+++ b/app/validate-bpf/debug_command.h
@@ -0,0 +1,65 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include <cmdline_parse_string.h>
+
+#include <stdint.h>
+
+
+struct debug_command_comparison {
+ cmdline_fixed_string_t lhs;
+ cmdline_fixed_string_t op;
+ union {
+ cmdline_fixed_string_t register_rhs;
+ int64_t literal_rhs;
+ };
+};
+
+struct debug_command_parsed {
+ cmdline_fixed_string_t keyword; /* Any keyword we don't need */
+ union {
+ struct debug_command_comparison comparison;
+ cmdline_fixed_string_t event;
+ int32_t frame_offset;
+ uint32_t instruction_count;
+ uint32_t instruction_offset;
+ uint32_t pc;
+ uint32_t point_number;
+ cmdline_fixed_string_t register_;
+ };
+};
+
+enum debug_command {
+ DEBUG_COMMAND_EOF,
+ DEBUG_COMMAND_BREAK,
+ DEBUG_COMMAND_BREAK_PC,
+ DEBUG_COMMAND_CATCH,
+ DEBUG_COMMAND_CLEAR,
+ DEBUG_COMMAND_CLEAR_EVENT,
+ DEBUG_COMMAND_CLEAR_PC,
+ DEBUG_COMMAND_CONTINUE,
+ DEBUG_COMMAND_DELETE,
+ DEBUG_COMMAND_DELETE_NUMBER,
+ DEBUG_COMMAND_INFO_FRAME,
+ DEBUG_COMMAND_INFO_FRAME_OFFSET,
+ DEBUG_COMMAND_INFO_POINTS,
+ DEBUG_COMMAND_INFO_REGISTER,
+ DEBUG_COMMAND_INFO_REGISTERS,
+ DEBUG_COMMAND_LIST,
+ DEBUG_COMMAND_LIST_COUNT,
+ DEBUG_COMMAND_LIST_PROGRAM,
+ DEBUG_COMMAND_MAY_LITERAL_RHS,
+ DEBUG_COMMAND_MAY_REGISTER_RHS,
+ DEBUG_COMMAND_QUIT,
+ DEBUG_COMMAND_RUN,
+ DEBUG_COMMAND_START,
+ DEBUG_COMMAND_STEP,
+ DEBUG_COMMAND_WHERE,
+};
+
+extern struct debug_command_parsed debug_command_parsed;
+
+/** Get and return one command line command, storing its data in the struct above. */
+enum debug_command
+debug_command_get(const char *prompt);
diff --git a/app/validate-bpf/eal_init_args.c b/app/validate-bpf/eal_init_args.c
new file mode 100644
index 000000000000..ba85b7c58807
--- /dev/null
+++ b/app/validate-bpf/eal_init_args.c
@@ -0,0 +1,57 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include "internal.h"
+
+
+#define RTE_EAL_INIT_ARG_SIZE_MAX sizeof("--log-level=lib.eal:warning")
+
+static const char RTE_EAL_INIT_ARGS[][RTE_EAL_INIT_ARG_SIZE_MAX] = {
+ "--log-level=lib.eal:warning",
+ "--no-huge",
+ "--no-pci",
+ "--no-hpet",
+ "--no-shconf",
+};
+
+#define RTE_EAL_INIT_ARGC (/* program name */ 1 + RTE_DIM(RTE_EAL_INIT_ARGS))
+
+int
+get_eal_init_argc(void)
+{
+ return RTE_EAL_INIT_ARGC;
+}
+
+/*
+ * We cannot just return literal strings here, because rte_eal_init accepts
+ * an array of mutable pointers to mutable strings, so literals won't work.
+ * Instead we build a copy of RTE_EAL_INIT_ARGS in a static mutable area.
+ */
+char**
+get_eal_init_argv(char *prog_name)
+{
+ /* Static arrays for mutable copies of actual args */
+ static char mutable_args[RTE_DIM(RTE_EAL_INIT_ARGS)]
+ [RTE_EAL_INIT_ARG_SIZE_MAX];
+ /*
+ * Static array for pointers to args. First element will hold pointer
+ * to the prog_name, last will be set to NULL, the rest will point to
+ * elements of mutable_args with index one smaller.
+ */
+ static char *mutable_ptrs[RTE_EAL_INIT_ARGC + /* terminating NULL */ 1];
+
+ mutable_ptrs[0] = prog_name;
+ for (int argi = 0; argi != RTE_DIM(RTE_EAL_INIT_ARGS); ++argi) {
+ const char * const const_arg = RTE_EAL_INIT_ARGS[argi];
+ char * const mutable_arg = mutable_args[argi];
+ const int snprintf_rc = snprintf(mutable_arg,
+ RTE_EAL_INIT_ARG_SIZE_MAX, "%s", const_arg);
+ RTE_VERIFY(snprintf_rc >= 0 &&
+ (size_t)snprintf_rc < RTE_EAL_INIT_ARG_SIZE_MAX);
+ mutable_ptrs[1 + argi] = mutable_arg;
+ }
+ mutable_ptrs[RTE_DIM(mutable_ptrs) - 1] = NULL;
+
+ return mutable_ptrs;
+}
diff --git a/app/validate-bpf/internal.h b/app/validate-bpf/internal.h
new file mode 100644
index 000000000000..730a604d88bc
--- /dev/null
+++ b/app/validate-bpf/internal.h
@@ -0,0 +1,151 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include <rte_bpf.h>
+#include <rte_log.h>
+
+
+extern int rte_validate_bpf_logtype;
+#define RTE_LOGTYPE_VALIDATE_BPF rte_validate_bpf_logtype
+#define VALIDATE_BPF_LOG(level, ...) \
+ RTE_LOG_LINE(level, VALIDATE_BPF, "" __VA_ARGS__)
+
+struct rte_bpf_validate_debug;
+
+/**
+ * List of pointers to allocations, to keep track of things to free.
+ *
+ * May contain NULLs. May contain itself directly or via an outer struct.
+ */
+struct alloc_list {
+ size_t count;
+ void **ptrs;
+};
+
+/** Add new pointer to the alloc list, return its index. */
+size_t alloc_list_append(struct alloc_list *alloc_list, void *ptr);
+
+/** Replace pointer at the specified index in the alloc list. */
+void alloc_list_replace(struct alloc_list *alloc_list, size_t index, void *ptr);
+
+/** Free all allocations in the alloc list */
+void alloc_list_free_all(struct alloc_list *alloc_list);
+
+
+/** Parsed program arguments */
+struct args {
+ /* Allocations list, args.c internal use only. */
+ struct alloc_list _alloc_list;
+
+ /* Set if command line contains --help */
+ bool show_help;
+
+ /* BPF load parameters. */
+ struct rte_bpf_prm_ex bpf_prm;
+
+ /* Size of the mbuf data buffer. */
+ size_t mbuf_buf_size;
+};
+
+/** Print program usage information */
+void
+print_usage(const char *program_name);
+
+/** Print program defaults. */
+void
+print_defaults(void);
+
+/**
+ * Parse command-line arguments
+ *
+ * @param argc
+ * Command-line arguments count, as received by main.
+ * @param argv
+ * Command-line arguments array, as received by main.
+ * Modified during the call due to the use of getopt_long.
+ * Modifying it after the call invalidates the return value.
+ * @return
+ * - parse results in case of success;
+ * - NULL in case of an error (diagnostic will be printed to stderr)
+ */
+struct args *
+args_parse(int argc, char *argv[]);
+
+/** Destroy args struct returned by args_parse */
+void
+args_destroy(struct args *args);
+
+
+/** Value for the rte_eal_init argc argument */
+int
+get_eal_init_argc(void);
+
+/**
+ * Value for the rte_eal_init argv argument
+ *
+ * @param prog_name
+ * Program name, can be obtained from argv[0]
+ */
+char **
+get_eal_init_argv(char *prog_name);
+
+
+/** Print types supported in command-line. */
+void
+print_supported_types(void);
+
+/**
+ * Parse text into arg.
+ * @param arg
+ * Pointer to a variable to put parse result to.
+ * Member `buf_size` of types related to `struct rte_mbuf` is set to
+ * `RTE_MBUF_DEFAULT_BUF_SIZE`, adjust using `adjust_arg_buf_size` if needed.
+ * @param text
+ * Text representation of the type, e.g. "struct rte_mbuf *".
+ * @param alloc_list
+ * Alloc list to use for new allocations.
+ * @return
+ * 0 on success
+ * -1 on failure
+ */
+int
+parse_arg(struct rte_bpf_arg *arg, const char *text);
+
+/**
+ * Parse text into xsym.
+ * @param arg
+ * Pointer to a variable to put parse result to.
+ * Member `buf_size` of types related to `struct rte_mbuf` is set to
+ * `RTE_MBUF_DEFAULT_BUF_SIZE`, adjust using `adjust_xsym_buf_size` if needed.
+ * @param text
+ * Text representation of the external symbol, e.g. "void exit(uint32_t)".
+ * @param alloc_list
+ * Alloc list to use for new allocations.
+ * @return
+ * 0 on success
+ * -1 on failure
+ */
+int
+parse_xsym(struct rte_bpf_xsym *xsym, const char *text,
+ struct alloc_list *alloc_list);
+
+/** If arg->buf_size is non-zero, set it to mbuf_buf_size */
+void
+adjust_arg_buf_size(struct rte_bpf_arg *arg, size_t mbuf_buf_size);
+
+/** If buf_size members are non-zero, set them to mbuf_buf_size */
+void
+adjust_xsym_buf_size(struct rte_bpf_xsym *xsym, size_t mbuf_buf_size);
+
+/** Create and set up global debugging session. */
+struct rte_bpf_validate_debug *
+debug_create(void);
+
+/** Clear and destroy global debugging session. */
+void
+debug_destroy(struct rte_bpf_validate_debug *debug);
+
+/** Tells caller if validation should be re-tried. */
+bool
+debug_validate_again(void);
diff --git a/app/validate-bpf/main.c b/app/validate-bpf/main.c
new file mode 100644
index 000000000000..f2047001c045
--- /dev/null
+++ b/app/validate-bpf/main.c
@@ -0,0 +1,77 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include "internal.h"
+
+#include <stdio.h>
+#include <stdlib.h>
+
+#include <rte_bpf.h>
+#include <rte_debug.h>
+#include <rte_eal.h>
+#include <rte_errno.h>
+
+RTE_LOG_REGISTER(rte_validate_bpf_logtype, validate-bpf, NOTICE);
+
+static int
+test_bpf_load(struct rte_bpf_prm_ex *prm)
+{
+ struct rte_bpf * const bpf = rte_bpf_load_ex(prm);
+
+ const int rc = -rte_errno;
+
+ rte_bpf_destroy(bpf);
+
+ return bpf == NULL ? rc : 0;
+}
+
+/* Re-starts validation of asked from interactive debugger. */
+static int
+test_bpf_load_with_restarts(struct rte_bpf_prm_ex *prm)
+{
+ for (;;) {
+ const int rc = test_bpf_load(prm);
+
+ if (rc == -ECANCELED && debug_validate_again())
+ continue;
+
+ if (rc != 0)
+ fprintf(stderr, "Error %d loading BPF: %s\n",
+ -rc, strerror(-rc));
+ else
+ fprintf(stderr, "Validation succeeded.\n");
+
+ return rc;
+ }
+}
+
+int
+main(int argc, char *argv[])
+{
+ struct args * const args = args_parse(argc, argv);
+ if (args == NULL || args->show_help
+ || args->bpf_prm.elf_file.path == NULL) {
+ args_destroy(args);
+ print_usage(argv[0]);
+ if (args == NULL)
+ /* Could not parse arguments. */
+ return 2;
+ print_supported_types();
+ print_defaults();
+ return 0;
+ }
+
+ const int eal_init_argc = get_eal_init_argc();
+ char ** const eal_init_argv = get_eal_init_argv(argv[0]);
+ RTE_VERIFY(rte_eal_init(eal_init_argc, eal_init_argv) ==
+ eal_init_argc - 1);
+
+ const int ret = test_bpf_load_with_restarts(&args->bpf_prm);
+
+ args_destroy(args);
+
+ RTE_VERIFY(rte_eal_cleanup() == 0);
+
+ return ret < 0 ? EXIT_FAILURE : EXIT_SUCCESS;
+}
diff --git a/app/validate-bpf/meson.build b/app/validate-bpf/meson.build
new file mode 100644
index 000000000000..03cfc329efb5
--- /dev/null
+++ b/app/validate-bpf/meson.build
@@ -0,0 +1,13 @@
+# SPDX-License-Identifier: BSD-3-Clause
+# Copyright(c) 2025 Huawei Technologies Co., Ltd
+
+sources = files(
+ 'alloc_list.c',
+ 'args.c',
+ 'debug.c',
+ 'debug_command.c',
+ 'eal_init_args.c',
+ 'main.c',
+ 'parse_decl.c',
+)
+deps = ['bpf', 'cmdline']
diff --git a/app/validate-bpf/parse_decl.c b/app/validate-bpf/parse_decl.c
new file mode 100644
index 000000000000..5c05684ea51b
--- /dev/null
+++ b/app/validate-bpf/parse_decl.c
@@ -0,0 +1,611 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include "internal.h"
+
+#include <ctype.h>
+#include <stdlib.h>
+
+#include <net/ethernet.h>
+#include <netinet/ip.h>
+#include <netinet/tcp.h>
+#include <netinet/udp.h>
+
+#include <rte_mbuf_core.h>
+
+#define RETURN_TEXT_ERROR(text, text_start, message, ...) do { \
+ const int _offset = (text) - (text_start); \
+ VALIDATE_BPF_LOG(ERR, "at offset %d: " message, \
+ _offset, ## __VA_ARGS__); \
+ VALIDATE_BPF_LOG(NOTICE, "%s", text_start); \
+ VALIDATE_BPF_LOG(NOTICE, "%*c", _offset + 1, '^'); \
+ return -1; \
+} while (0)
+
+/* Used in place of any signature, this is not really being checked. */
+static uint64_t
+dummy_function(uint64_t arg1, uint64_t arg2, uint64_t arg3, uint64_t arg4,
+ uint64_t arg5)
+{
+ RTE_SET_USED(arg1);
+ RTE_SET_USED(arg2);
+ RTE_SET_USED(arg3);
+ RTE_SET_USED(arg4);
+ RTE_SET_USED(arg5);
+ return 0;
+}
+
+
+/* TOKENS AND TYPES */
+
+enum token {
+ TOKEN_UNRECOGNIZED = -1,
+ TOKEN_END = 0,
+
+ TOKEN_ASTERISK,
+ TOKEN_BRACKET_CLOSE,
+ TOKEN_BRACKET_OPEN,
+ TOKEN_COMMA,
+ TOKEN_PARENTHESIS_CLOSE,
+ TOKEN_PARENTHESIS_OPEN,
+ TOKEN_STRUCT,
+
+ TOKEN_TYPES_BEGIN,
+
+ TOKEN_TYPE_CHAR,
+ TOKEN_TYPE_ETHER_HEADER,
+ TOKEN_TYPE_INT32_T,
+ TOKEN_TYPE_IP_HEADER,
+ TOKEN_TYPE_IP_HEADERS,
+ TOKEN_TYPE_RTE_MBUF,
+ TOKEN_TYPE_TCP_HEADER,
+ TOKEN_TYPE_TCP_HEADERS,
+ TOKEN_TYPE_UDP_HEADER,
+ TOKEN_TYPE_UDP_HEADERS,
+ TOKEN_TYPE_UINT32_T,
+ TOKEN_TYPE_UINT64_T,
+ TOKEN_TYPE_UINTPTR_T,
+ TOKEN_TYPE_VOID,
+
+ TOKEN_TYPES_END,
+};
+
+/* Return true if the token is a type token */
+static bool
+is_type_token(enum token token)
+{
+ return token > TOKEN_TYPES_BEGIN && token < TOKEN_TYPES_END;
+}
+
+struct text_token {
+ const char *text;
+ unsigned int length; /* Not size_t to fit this struct in 2 regs. */
+ enum token token;
+};
+
+#define TEXT_TOKEN_DEF(text_, token_) { \
+ .text = text_, \
+ .length = sizeof(text_) - 1, \
+ .token = token_, \
+}
+
+/* Token search table, MUST BE SORTED! */
+const struct text_token TEXT_TOKENS[] = {
+ TEXT_TOKEN_DEF("(", TOKEN_PARENTHESIS_OPEN),
+ TEXT_TOKEN_DEF(")", TOKEN_PARENTHESIS_CLOSE),
+ TEXT_TOKEN_DEF("*", TOKEN_ASTERISK),
+ TEXT_TOKEN_DEF(",", TOKEN_COMMA),
+ TEXT_TOKEN_DEF("[", TOKEN_BRACKET_OPEN),
+ TEXT_TOKEN_DEF("]", TOKEN_BRACKET_CLOSE),
+ TEXT_TOKEN_DEF("char", TOKEN_TYPE_CHAR),
+ TEXT_TOKEN_DEF("ether_header", TOKEN_TYPE_ETHER_HEADER),
+ TEXT_TOKEN_DEF("int32_t", TOKEN_TYPE_INT32_T),
+ TEXT_TOKEN_DEF("ip_header", TOKEN_TYPE_IP_HEADER),
+ TEXT_TOKEN_DEF("ip_headers", TOKEN_TYPE_IP_HEADERS),
+ TEXT_TOKEN_DEF("iphdr", TOKEN_TYPE_IP_HEADER),
+ TEXT_TOKEN_DEF("rte_mbuf", TOKEN_TYPE_RTE_MBUF),
+ TEXT_TOKEN_DEF("struct", TOKEN_STRUCT),
+ TEXT_TOKEN_DEF("tcp_header", TOKEN_TYPE_TCP_HEADER),
+ TEXT_TOKEN_DEF("tcp_headers", TOKEN_TYPE_TCP_HEADERS),
+ TEXT_TOKEN_DEF("tcphdr", TOKEN_TYPE_TCP_HEADER),
+ TEXT_TOKEN_DEF("udp_header", TOKEN_TYPE_UDP_HEADER),
+ TEXT_TOKEN_DEF("udp_headers", TOKEN_TYPE_UDP_HEADERS),
+ TEXT_TOKEN_DEF("udphdr", TOKEN_TYPE_UDP_HEADER),
+ TEXT_TOKEN_DEF("uint32_t", TOKEN_TYPE_UINT32_T),
+ TEXT_TOKEN_DEF("uint64_t", TOKEN_TYPE_UINT64_T),
+ TEXT_TOKEN_DEF("uintptr_t", TOKEN_TYPE_UINTPTR_T),
+ TEXT_TOKEN_DEF("void", TOKEN_TYPE_VOID),
+};
+
+const struct text_token TEXT_TOKEN_UNRECOGNIZED = { "", 0, TOKEN_UNRECOGNIZED };
+const struct text_token TEXT_TOKEN_END = { "", 0, TOKEN_END };
+
+/* IP header maximum possible size with options. */
+#define IP_HEADER_MAX_SIZE 60
+#define IP_HEADERS_MAX_SIZE (sizeof(struct ether_header) + IP_HEADER_MAX_SIZE)
+#define TCP_HEADERS_MAX_SIZE (IP_HEADERS_MAX_SIZE + sizeof(struct tcphdr))
+#define UDP_HEADERS_MAX_SIZE (IP_HEADERS_MAX_SIZE + sizeof(struct udphdr))
+
+#define POINTER_SIZE (sizeof(char *))
+
+const size_t TYPE_TOKEN_SIZE[] = {
+ [TOKEN_TYPE_CHAR] = sizeof(char),
+ [TOKEN_TYPE_ETHER_HEADER] = sizeof(struct ether_header),
+ [TOKEN_TYPE_INT32_T] = sizeof(int32_t),
+ [TOKEN_TYPE_IP_HEADERS] = IP_HEADERS_MAX_SIZE,
+ [TOKEN_TYPE_IP_HEADER] = sizeof(struct iphdr),
+ [TOKEN_TYPE_TCP_HEADERS] = TCP_HEADERS_MAX_SIZE,
+ [TOKEN_TYPE_TCP_HEADER] = sizeof(struct tcphdr),
+ [TOKEN_TYPE_UDP_HEADERS] = UDP_HEADERS_MAX_SIZE,
+ [TOKEN_TYPE_UDP_HEADER] = sizeof(struct udphdr),
+ [TOKEN_TYPE_UINT32_T] = sizeof(uint32_t),
+ [TOKEN_TYPE_UINT64_T] = sizeof(uint64_t),
+ [TOKEN_TYPE_UINTPTR_T] = sizeof(uintptr_t),
+};
+
+/*
+ * Struct rte_bpf_arg augmented with a type of pointer to it, since rte_bpf_arg
+ * by itself may not contain enough information to determine its pointer type.
+ */
+struct arg_info {
+ struct rte_bpf_arg value;
+ enum rte_bpf_arg_type ptr_type;
+ bool is_array;
+};
+
+/* Return arg_info struct described by the specified type token. */
+static struct arg_info
+get_type_token_arg(enum token type)
+{
+ RTE_ASSERT(is_type_token(type));
+ switch (type) {
+ case TOKEN_TYPE_VOID:
+ return (struct arg_info){
+ .value = { .type = RTE_BPF_ARG_UNDEF },
+ .ptr_type = RTE_BPF_ARG_PTR,
+ };
+ case TOKEN_TYPE_RTE_MBUF:
+ return (struct arg_info){
+ .value = {
+ .type = RTE_BPF_ARG_RAW,
+ .size = sizeof(struct rte_mbuf),
+ .buf_size = RTE_MBUF_DEFAULT_BUF_SIZE,
+ },
+ .ptr_type = RTE_BPF_ARG_PTR_MBUF,
+ };
+ default:
+ /* Should only reach here for normal sized types. */
+ RTE_ASSERT((size_t)type < RTE_DIM(TYPE_TOKEN_SIZE) &&
+ TYPE_TOKEN_SIZE[type] != 0);
+ return (struct arg_info){
+ .value = {
+ .type = RTE_BPF_ARG_RAW,
+ .size = TYPE_TOKEN_SIZE[type],
+ },
+ .ptr_type = RTE_BPF_ARG_PTR,
+ };
+ }
+}
+
+
+/* PARSING TEXT INTO TOKENS */
+
+/* Return true if character matches [a-zA-Z0-9_] in regex */
+static bool
+iswordchar(char character)
+{
+ return isalnum(character) || character == '_';
+}
+
+/*
+ * Compare pointer to text with pointer to struct text_token to determine
+ * if text starts with the specified token.
+ */
+static int
+text_and_text_token_cmp(const void *text_void_ptr, const void *text_token_void_ptr)
+{
+ int result;
+ const char * const text = text_void_ptr;
+ const struct text_token * const text_token = text_token_void_ptr;
+
+ if (memchr(text, 0, text_token->length) != NULL)
+ /* Text is shorter than the token. */
+ return strcmp(text, text_token->text);
+
+ /* Cannot use strcmp because we are looking for a prefix. */
+ result = memcmp(text, text_token->text, text_token->length);
+
+ /* Checking the case of a partial word match. */
+ if (result == 0 && iswordchar(text[text_token->length - 1]) &&
+ iswordchar(text[text_token->length]))
+ /* Text word is longer than the token. */
+ result = 1;
+
+ return result;
+}
+
+/* Advance pointed character pointer to the first non-space character. */
+static void
+skip_space(const char **text_ptr)
+{
+ while (isspace(**text_ptr))
+ ++*text_ptr;
+}
+
+/*
+ * Recognize and return a token starting text.
+ * Return TEXT_TOKEN_END if text is empty.
+ * Return TEXT_TOKEN_UNRECOGNIZED if text starts with unknown token.
+ */
+static struct text_token
+peek_text_token(const char *text)
+{
+ if (*text == '\0')
+ return TEXT_TOKEN_END;
+ const struct text_token * const text_token = bsearch(text, TEXT_TOKENS,
+ RTE_DIM(TEXT_TOKENS), sizeof(TEXT_TOKENS[0]),
+ text_and_text_token_cmp);
+ if (text_token == NULL)
+ return TEXT_TOKEN_UNRECOGNIZED;
+ return *text_token;
+}
+
+/*
+ * Advance pointed text starting with the specified token to the first
+ * non-space character after it.
+ * Do nothing if called for TEXT_TOKEN_END or TEXT_TOKEN_UNRECOGNIZED.
+ */
+static void
+consume_text_token(const char **text_ptr, struct text_token text_token)
+{
+ RTE_ASSERT(memcmp(*text_ptr, text_token.text, text_token.length) == 0);
+ *text_ptr += text_token.length;
+ skip_space(text_ptr);
+}
+
+/* Return length of the word starting at `text`. */
+static size_t
+find_word_length(const char *text)
+{
+ size_t word_length = 0;
+ while (iswordchar(text[word_length]))
+ ++word_length;
+ return word_length;
+}
+
+/* Create a copy of the word starting text, advance to next token. */
+static const char *
+take_name(const char **text_ptr, struct alloc_list *alloc_list)
+{
+ const size_t word_length = find_word_length(*text_ptr);
+ if (word_length == 0)
+ /* Text does not start with a word. */
+ return NULL;
+
+ /* Allocate memory for the word and add it to the alloc_list */
+ char *word = malloc(word_length + 1);
+ RTE_VERIFY(word != NULL);
+ alloc_list_append(alloc_list, word);
+
+ /* Copy and terminate word contents. */
+ memcpy(word, *text_ptr, word_length);
+ word[word_length] = '\0';
+
+ /* Advance text pointer. */
+ *text_ptr += word_length;
+ skip_space(text_ptr);
+
+ return word;
+}
+
+/* Read a number starting text, advance to next token. */
+static int
+take_number(size_t *number, const char **text_ptr)
+{
+ /* Read a word and let strtoull to decide if it's a number. */
+ const size_t word_length = find_word_length(*text_ptr);
+ if (word_length == 0)
+ /* Text does not start with a word. */
+ return -ENOENT;
+
+ errno = 0;
+ char *number_end;
+ unsigned long long long_number = strtoull(*text_ptr, &number_end, 0);
+ if (errno > 0)
+ return -errno;
+ if (number_end != *text_ptr + word_length)
+ /* Could not parse whole word. */
+ return -EINVAL;
+ if (long_number > SIZE_MAX)
+ return -ERANGE;
+
+ *number = long_number;
+
+ /* Advance text pointer. */
+ *text_ptr += word_length;
+ skip_space(text_ptr);
+
+ return 0;
+}
+
+
+/* PARSING DECLARATION PARTS */
+
+/* Change arg into a reference. */
+static void
+change_into_reference(struct arg_info *arg)
+{
+ if (RTE_BPF_ARG_PTR_TYPE(arg->value.type) != 0) {
+ VALIDATE_BPF_LOG(WARNING,
+ "After taking reference to a pointer the latter "
+ "will be described as an opaque pointer-size blob.");
+ RTE_ASSERT(arg->ptr_type == RTE_BPF_ARG_PTR);
+ arg->value.size = POINTER_SIZE;
+ }
+ arg->value.type = arg->ptr_type;
+ arg->ptr_type = RTE_BPF_ARG_PTR;
+ arg->is_array = false;
+}
+
+/*
+ * Recognize and consume arg starting text, advance to next token.
+ */
+static int
+take_arg(struct arg_info *arg, const char **text_ptr, const char *text_start)
+{
+ struct text_token next = peek_text_token(*text_ptr);
+
+ if (next.token == TOKEN_STRUCT) {
+ consume_text_token(text_ptr, next);
+ next = peek_text_token(*text_ptr);
+ }
+
+ if (!is_type_token(next.token))
+ RETURN_TEXT_ERROR(*text_ptr, text_start, "expect type");
+ const enum token type = next.token;
+ consume_text_token(text_ptr, next);
+ next = peek_text_token(*text_ptr);
+
+ *arg = get_type_token_arg(type);
+
+ while (next.token == TOKEN_ASTERISK) {
+ consume_text_token(text_ptr, next);
+ next = peek_text_token(*text_ptr);
+ change_into_reference(arg);
+ }
+
+ while (next.token == TOKEN_BRACKET_OPEN) {
+ consume_text_token(text_ptr, next);
+
+ size_t array_length;
+ if (take_number(&array_length, text_ptr) < 0)
+ RETURN_TEXT_ERROR(*text_ptr, text_start, "expect length");
+ next = peek_text_token(*text_ptr);
+
+ if (next.token != TOKEN_BRACKET_CLOSE)
+ RETURN_TEXT_ERROR(*text_ptr, text_start, "expect ']'");
+ consume_text_token(text_ptr, next);
+ next = peek_text_token(*text_ptr);
+
+ change_into_reference(arg);
+ arg->value.size *= array_length;
+ arg->is_array = true;
+ }
+
+ return 0;
+}
+
+/* Fill struct rte_bpf_arg within xsym trying not to unzero the padding. */
+static void
+fill_xsym_arg(struct rte_bpf_arg *target, struct rte_bpf_arg source)
+{
+ /* Copy fields individually to try and prevent copying the padding. */
+ target->type = source.type;
+ target->size = source.size;
+ target->buf_size = source.buf_size;
+}
+
+/* Build and return struct rte_bpf_arg of type RTE_BPF_XTYPE_VAR */
+static void
+fill_var_xsym(struct rte_bpf_xsym *xsym, struct arg_info arg, const char *name,
+ struct alloc_list *alloc_list)
+{
+ /* Variables are passed by reference, except for arrays. */
+ if (!arg.is_array) {
+ if (RTE_BPF_ARG_PTR_TYPE(arg.value.type) != 0)
+ VALIDATE_BPF_LOG(WARNING,
+ "External pointers may not work as expected "
+ "because all external variables are passed by "
+ "reference but there is currently no way to "
+ "describe double pointer to the validator.");
+ change_into_reference(&arg);
+ }
+
+ /* Allocate something to assign to a val pointer. */
+ void * const val = calloc(1, RTE_MAX(1u, arg.value.size));
+ RTE_VERIFY(val != 0);
+ alloc_list_append(alloc_list, val);
+
+ /* Need all padding and unused fields to be zero-filled. */
+ memset(xsym, 0, sizeof(*xsym));
+ xsym->name = name;
+ xsym->type = RTE_BPF_XTYPE_VAR;
+ xsym->var.val = val;
+ fill_xsym_arg(&xsym->var.desc, arg.value);
+}
+
+/* Build and return struct rte_bpf_arg of type RTE_BPF_XTYPE_FUNC */
+static void
+fill_func_xsym(struct rte_bpf_xsym *xsym, struct arg_info arg, const char *name)
+{
+ /* Need all padding and unused fields to be zero-filled. */
+ memset(xsym, 0, sizeof(*xsym));
+ xsym->name = name;
+ xsym->type = RTE_BPF_XTYPE_FUNC;
+ xsym->func.val = &dummy_function;
+ fill_xsym_arg(&xsym->func.ret, arg.value);
+}
+
+/*
+ * Parse and store function arguments, advance to next token after ')'.
+ * Value of *text_ptr should point to the next token after '('.
+ */
+static int
+take_func_xsym_args(struct rte_bpf_xsym *xsym, const char **text_ptr,
+ const char *text_start)
+{
+ struct arg_info arg;
+ struct text_token delimiter = peek_text_token(*text_ptr);
+
+ while (delimiter.token != TOKEN_PARENTHESIS_CLOSE) {
+ if (xsym->func.nb_args == EBPF_FUNC_MAX_ARGS)
+ RETURN_TEXT_ERROR(*text_ptr, text_start,
+ "too many arguments, maximum %d allowed",
+ EBPF_FUNC_MAX_ARGS);
+
+ if (take_arg(&arg, text_ptr, text_start) < 0)
+ return -1;
+ if (arg.value.type == RTE_BPF_ARG_UNDEF &&
+ xsym->func.nb_args != 0)
+ RETURN_TEXT_ERROR(*text_ptr, text_start,
+ "arguments of type void are not allowed");
+ fill_xsym_arg(&xsym->func.args[xsym->func.nb_args++],
+ arg.value);
+
+ delimiter = peek_text_token(*text_ptr);
+ switch (delimiter.token) {
+ case TOKEN_COMMA:
+ consume_text_token(text_ptr, delimiter);
+ continue;
+ case TOKEN_PARENTHESIS_CLOSE:
+ break;
+ default:
+ RETURN_TEXT_ERROR(*text_ptr, text_start,
+ "expect ')' or ','");
+ }
+ }
+ consume_text_token(text_ptr, delimiter);
+
+ /* Special case of single void argument. */
+ if (xsym->func.nb_args == 1 &&
+ xsym->func.args[0].type == RTE_BPF_ARG_UNDEF) {
+ xsym->func.nb_args = 0;
+ /* Need all padding and unused fields to be zero-filled. */
+ fill_xsym_arg(&xsym->func.args[0], (struct rte_bpf_arg){});
+ }
+
+ return 0;
+}
+
+/* Make sure text has ended. */
+static int
+ensure_end(const char *text, const char *text_start)
+{
+ if (peek_text_token(text).token != TOKEN_END)
+ RETURN_TEXT_ERROR(text, text_start, "trailing garbage");
+ return 0;
+}
+
+/* Parse and store xsym, advance to next token. */
+static int
+take_xsym(struct rte_bpf_xsym *xsym, const char **text_ptr,
+ const char *text_start, struct alloc_list *alloc_list)
+{
+ struct arg_info arg;
+
+ if (take_arg(&arg, text_ptr, text_start) < 0)
+ return -1;
+
+ const char * const name = take_name(text_ptr, alloc_list);
+ if (name == NULL)
+ RETURN_TEXT_ERROR(*text_ptr, text_start, "expect name");
+
+ const struct text_token next = peek_text_token(*text_ptr);
+ switch (next.token) {
+ case TOKEN_END:
+ fill_var_xsym(xsym, arg, name, alloc_list);
+ break;
+ case TOKEN_PARENTHESIS_OPEN:
+ consume_text_token(text_ptr, next);
+ fill_func_xsym(xsym, arg, name);
+ if (take_func_xsym_args(xsym, text_ptr, text_start) < 0)
+ return -1;
+ break;
+ default:
+ RETURN_TEXT_ERROR(*text_ptr, text_start,
+ "expect '(' or text end");
+ }
+ return 0;
+}
+
+
+/* PUBLIC FUNCTIONS */
+
+void
+print_supported_types(void)
+{
+ printf("TYPE: [struct] BASIC_TYPE [*]... [[N]]...\n");
+ printf("BASIC_TYPE: one of\n");
+ for (int tti = 0; tti != RTE_DIM(TEXT_TOKENS); ++tti) {
+ if (is_type_token(TEXT_TOKENS[tti].token))
+ printf("\t%s\n", TEXT_TOKENS[tti].text);
+ }
+}
+
+int
+parse_arg(struct rte_bpf_arg *arg, const char *text)
+{
+ struct arg_info arg_info;
+ const char * const text_start = text;
+ skip_space(&text);
+ if (take_arg(&arg_info, &text, text_start) < 0)
+ return -1;
+ if (ensure_end(text, text_start) < 0)
+ return -1;
+ if (arg_info.ptr_type != RTE_BPF_ARG_PTR)
+ VALIDATE_BPF_LOG(WARNING,
+ "`%s` has a special pointer type which was left unused; "
+ "argument was set to an opaque blob.",
+ text_start);
+ *arg = arg_info.value;
+ return 0;
+}
+
+int
+parse_xsym(struct rte_bpf_xsym *xsym, const char *text,
+ struct alloc_list *alloc_list)
+{
+ const char * const text_start = text;
+ skip_space(&text);
+ if (take_xsym(xsym, &text, text_start, alloc_list) < 0)
+ return -1;
+ if (ensure_end(text, text_start) < 0)
+ return -1;
+ return 0;
+}
+
+void
+adjust_arg_buf_size(struct rte_bpf_arg *arg, size_t mbuf_buf_size)
+{
+ if (arg->buf_size != 0)
+ arg->buf_size = mbuf_buf_size;
+}
+
+void
+adjust_xsym_buf_size(struct rte_bpf_xsym *xsym, size_t mbuf_buf_size)
+{
+ switch (xsym->type) {
+ case RTE_BPF_XTYPE_FUNC:
+ for (uint32_t argi = 0; argi != xsym->func.nb_args; ++argi)
+ adjust_arg_buf_size(&xsym->func.args[argi],
+ mbuf_buf_size);
+ adjust_arg_buf_size(&xsym->func.ret, mbuf_buf_size);
+ break;
+ case RTE_BPF_XTYPE_VAR:
+ adjust_arg_buf_size(&xsym->var.desc, mbuf_buf_size);
+ break;
+ default:
+ rte_panic("Unexpected xsym type %d\n", xsym->type);
+ }
+}
diff --git a/doc/guides/rel_notes/release_26_11.rst b/doc/guides/rel_notes/release_26_11.rst
index 87c7e81bdebd..5c0caba50d07 100644
--- a/doc/guides/rel_notes/release_26_11.rst
+++ b/doc/guides/rel_notes/release_26_11.rst
@@ -55,6 +55,11 @@ New Features
Also, make sure to start the actual text at the margin.
=======================================================
+* **Added BPF validation application.**
+
+ Added ``dpdk-validate-bpf`` tool to pre-validate eBPF programs for
+ compatibility with the ``lib/bpf`` execution context.
+
Removed Items
-------------
diff --git a/doc/guides/tools/index.rst b/doc/guides/tools/index.rst
index 13f75a5bc652..1c43060e15cd 100644
--- a/doc/guides/tools/index.rst
+++ b/doc/guides/tools/index.rst
@@ -26,3 +26,4 @@ DPDK Tools User Guides
testmldev
graph
dts
+ validate_bpf
diff --git a/doc/guides/tools/validate_bpf.rst b/doc/guides/tools/validate_bpf.rst
new file mode 100644
index 000000000000..9e2d58cfbe5f
--- /dev/null
+++ b/doc/guides/tools/validate_bpf.rst
@@ -0,0 +1,97 @@
+.. SPDX-License-Identifier: BSD-3-Clause
+ Copyright(c) 2026 The DPDK contributors
+
+dpdk-validate-bpf Application
+=============================
+
+The ``dpdk-validate-bpf`` tool is an application that allows evaluating BPF
+programs against the ``lib/bpf`` library. It can be used to pre-validate BPF
+programs before loading them in a real application to ensure that they pass the
+internal verifier. This makes it a useful tool for integrating into build
+systems and continuous integration (CI) pipelines to verify the correctness
+of BPF programs during compilation, as well as for debugging validation issues.
+
+Running the Application
+-----------------------
+
+The tool requires a path to a compiled BPF object file. It also provides options
+to configure the execution environment and external symbol definitions.
+
+For a comprehensive list of options and their meanings, refer to the
+built-in help by running ``dpdk-validate-bpf --help``.
+
+The most important options for a minimal example are:
+
+* ``--prog-arg <type>``
+
+ Define program arguments, if different from ``struct rte_mbuf *``.
+ This argument may be repeated up to 5 times.
+
+* ``--xsym '<type> <name> | <type> <name>(<type>, ...)'``
+
+ Define an external symbol (variable or function) that the BPF program uses.
+ This allows the validator to recognize external calls and memory accesses.
+ Multiple external symbols can be defined by repeating this option.
+
+* ``--section <name>``
+
+ Specify the ELF section name in the BPF object file to load and validate.
+ The default section name is ``.text``, which is typically fine in cases
+ where the object file contains only one program.
+
+Interactive Debugging Mode
+--------------------------
+
+The interactive mode (enabled with the ``--debug`` flag) allows the user to
+debug the BPF validation process itself. This is particularly useful when
+BPF program fails verification, allowing you to trace the state changes per
+instruction and understand the validator's decisions.
+
+Design
+~~~~~~
+
+The interactive debugger provides a command-line interface similar to GDB.
+It allows setting breakpoints at specific instructions, or catchpoints on
+validator events (such as ``branch-enter`` or ``branch-prune``).
+Users can step through the validation process,
+inspect the state of registers, and query if certain conditional jumps may
+be taken based on the validator's knowledge of the program state.
+
+Usage Example
+~~~~~~~~~~~~~
+
+Launch the tool in debug mode with a compiled BPF object file:
+
+.. code-block:: console
+
+ $ dpdk-validate-bpf bpf_prog.o --prog-arg={'void *',uint64_t} --debug
+ => 0: b7 00 00 00 01 00 00 00 mov r0, #0x1
+ (validate) list 3
+ => 0: b7 00 00 00 01 00 00 00 mov r0, #0x1
+ 1: 25 02 01 00 0e 00 00 00 jgt r2, #0xe, L3
+ 2: b7 00 00 00 00 00 00 00 mov r0, #0x0
+ (validate) break 2
+ Breakpoint 0 at 2.
+ (validate) catch branch-enter
+ Catchpoint 1 on branch-enter.
+ (validate) continue
+ Catchpoint 1 on branch-enter.
+ Entered new branch at pc 1.
+ => 3: 95 00 00 00 00 00 00 00 exit
+ (validate) where
+ 1: 25 02 01 00 0e 00 00 00 jgt r2, #0xe, L3 ; taken
+ => 3: 95 00 00 00 00 00 00 00 exit
+ (validate) info r1
+ r1: %buffer<0> + 0
+ (validate) info r2
+ r2: 0xf..UINT64_MAX
+ (validate) may r2 <= 14
+ NO
+ (validate) continue
+ Catchpoint 1 on branch-enter.
+ Entered new branch at pc 1.
+ Breakpoint 0 at 2.
+ => 2: b7 00 00 00 00 00 00 00 mov r0, #0x0
+ (validate) continue
+ Validation succeeded.
+ (validate) quit
--
2.43.0
^ permalink raw reply related [flat|nested] 38+ messages in thread
* [PATCH v2 0/7] validate-bpf: add standalone validation debug tool
2026-09-11 10:39 [PATCH 0/7] validate-bpf: add standalone validation debug tool Marat Khalili
` (6 preceding siblings ...)
2026-09-11 10:40 ` [PATCH 7/7] app/validate-bpf: add BPF validation application Marat Khalili
@ 2026-09-14 14:54 ` Marat Khalili
2026-09-14 14:54 ` [PATCH v2 1/7] bpf/validate: fix finished status on restart Marat Khalili
` (8 more replies)
7 siblings, 9 replies; 38+ messages in thread
From: Marat Khalili @ 2026-09-14 14:54 UTC (permalink / raw)
Cc: dev
This patchset introduces a new standalone tool for pre-validating eBPF
programs and debugging validation problems. Its should allow developers
to trace validator state changes per instruction and understand verifier
decisions, simplifying the debugging of rejected programs before loading
them into a real application context.
v2:
* for FreeBSD compatibility switched from Linux to DPDK network structs;
* addressed small AI comments:
* clarified patch 1 commit message;
* renamed `rte_validate_bpf_logtype` to `validate_bpf_logtype`;
* corrected comment to an internal-use-only function to reflect that
next-after-last value of the program counter is no longer allowed;
Marat Khalili (7):
bpf/validate: fix finished status on restart
bpf/validate: refactor internal step function
bpf/validate: formalize call back requirements
bpf/validate: add jump notification events
bpf/validate: add get current event API
app/test: add test for bpf validate debug events
app/validate-bpf: add BPF validation application
MAINTAINERS | 2 +
app/meson.build | 1 +
app/test/test_bpf_validate.c | 341 +++++++-
app/validate-bpf/alloc_list.c | 53 ++
app/validate-bpf/args.c | 263 ++++++
app/validate-bpf/debug.c | 1099 ++++++++++++++++++++++++
app/validate-bpf/debug_command.c | 383 +++++++++
app/validate-bpf/debug_command.h | 65 ++
app/validate-bpf/eal_init_args.c | 57 ++
app/validate-bpf/internal.h | 151 ++++
app/validate-bpf/main.c | 77 ++
app/validate-bpf/meson.build | 13 +
app/validate-bpf/parse_decl.c | 611 +++++++++++++
doc/guides/rel_notes/release_26_11.rst | 5 +
doc/guides/tools/index.rst | 1 +
doc/guides/tools/validate_bpf.rst | 97 +++
lib/bpf/bpf_validate.c | 49 +-
lib/bpf/bpf_validate_debug.c | 92 +-
lib/bpf/bpf_validate_debug.h | 12 +-
lib/bpf/rte_bpf_validate_debug.h | 42 +-
20 files changed, 3348 insertions(+), 66 deletions(-)
create mode 100644 app/validate-bpf/alloc_list.c
create mode 100644 app/validate-bpf/args.c
create mode 100644 app/validate-bpf/debug.c
create mode 100644 app/validate-bpf/debug_command.c
create mode 100644 app/validate-bpf/debug_command.h
create mode 100644 app/validate-bpf/eal_init_args.c
create mode 100644 app/validate-bpf/internal.h
create mode 100644 app/validate-bpf/main.c
create mode 100644 app/validate-bpf/meson.build
create mode 100644 app/validate-bpf/parse_decl.c
create mode 100644 doc/guides/tools/validate_bpf.rst
--
2.43.0
^ permalink raw reply [flat|nested] 38+ messages in thread
* [PATCH v2 1/7] bpf/validate: fix finished status on restart
2026-09-14 14:54 ` [PATCH v2 0/7] validate-bpf: add standalone validation debug tool Marat Khalili
@ 2026-09-14 14:54 ` Marat Khalili
2026-09-14 14:54 ` [PATCH v2 2/7] bpf/validate: refactor internal step function Marat Khalili
` (7 subsequent siblings)
8 siblings, 0 replies; 38+ messages in thread
From: Marat Khalili @ 2026-09-14 14:54 UTC (permalink / raw)
To: Konstantin Ananyev; +Cc: dev
If debug session is reused, requesting evaluation result while
evaluation is still in progress would successfully return result from
the previous evaluation instead of expected -EAGAIN error.
Make evaluation start clear the finished status.
Fixes: 6b0da0602cc7 ("bpf/validate: introduce debugging interface")
Signed-off-by: Marat Khalili <marat.khalili@huawei.com>
---
lib/bpf/bpf_validate_debug.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/lib/bpf/bpf_validate_debug.c b/lib/bpf/bpf_validate_debug.c
index 5d18804a74bc..9a039766322c 100644
--- a/lib/bpf/bpf_validate_debug.c
+++ b/lib/bpf/bpf_validate_debug.c
@@ -568,6 +568,7 @@ __rte_bpf_validate_debug_evaluate_start(struct rte_bpf_validate_debug *debug,
debug->last_point = NULL;
debug->pc = 0;
debug->evaluate_started = true;
+ debug->evaluate_finished = false;
const int rc = debug_send_event(debug,
RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_START);
--
2.43.0
^ permalink raw reply related [flat|nested] 38+ messages in thread
* [PATCH v2 2/7] bpf/validate: refactor internal step function
2026-09-14 14:54 ` [PATCH v2 0/7] validate-bpf: add standalone validation debug tool Marat Khalili
2026-09-14 14:54 ` [PATCH v2 1/7] bpf/validate: fix finished status on restart Marat Khalili
@ 2026-09-14 14:54 ` Marat Khalili
2026-09-14 14:54 ` [PATCH v2 3/7] bpf/validate: formalize call back requirements Marat Khalili
` (6 subsequent siblings)
8 siblings, 0 replies; 38+ messages in thread
From: Marat Khalili @ 2026-09-14 14:54 UTC (permalink / raw)
To: Konstantin Ananyev; +Cc: dev
Rename `__rte_bpf_validate_debug_evaluate_step` to
`__rte_bpf_validate_debug_evaluate_update` to better reflect planned
usage, and make it accept a bitmask of events instead of a single event.
Step event is still implicitly triggered for now.
No functional changes.
Signed-off-by: Marat Khalili <marat.khalili@huawei.com>
---
lib/bpf/bpf_validate.c | 17 ++++++++---------
lib/bpf/bpf_validate_debug.c | 20 +++++++++++---------
lib/bpf/bpf_validate_debug.h | 11 ++++++-----
3 files changed, 25 insertions(+), 23 deletions(-)
diff --git a/lib/bpf/bpf_validate.c b/lib/bpf/bpf_validate.c
index f9960088a285..b317abb81172 100644
--- a/lib/bpf/bpf_validate.c
+++ b/lib/bpf/bpf_validate.c
@@ -2989,10 +2989,9 @@ evaluate(struct bpf_verifier *bvf)
break;
}
- rc = __rte_bpf_validate_debug_evaluate_step(debug, idx,
+ rc = __rte_bpf_validate_debug_evaluate_update(debug, idx,
prev_nb_edge > 1 ?
- RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER :
- RTE_BPF_VALIDATE_DEBUG_EVENT_STEP);
+ RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER) : 0);
if (rc < 0)
break;
@@ -3024,9 +3023,9 @@ evaluate(struct bpf_verifier *bvf)
}
if (bvf->evst->unreachable) {
- rc = __rte_bpf_validate_debug_evaluate_step(
+ rc = __rte_bpf_validate_debug_evaluate_update(
debug, get_node_idx(bvf, next),
- RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE);
+ RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE));
if (rc < 0)
break;
@@ -3038,9 +3037,9 @@ evaluate(struct bpf_verifier *bvf)
*/
} else if (node->nb_edge > 1 &&
prune_eval_state(bvf, node, next) == 0) {
- rc = __rte_bpf_validate_debug_evaluate_step(
+ rc = __rte_bpf_validate_debug_evaluate_update(
debug, get_node_idx(bvf, next),
- RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE);
+ RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE));
if (rc < 0)
break;
@@ -3059,9 +3058,9 @@ evaluate(struct bpf_verifier *bvf)
*/
if (prev_nb_edge != 0) {
- rc = __rte_bpf_validate_debug_evaluate_step(
+ rc = __rte_bpf_validate_debug_evaluate_update(
debug, get_node_idx(bvf, node) + 1,
- RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN);
+ RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN));
if (rc < 0)
break;
}
diff --git a/lib/bpf/bpf_validate_debug.c b/lib/bpf/bpf_validate_debug.c
index 9a039766322c..f76303a7fd2a 100644
--- a/lib/bpf/bpf_validate_debug.c
+++ b/lib/bpf/bpf_validate_debug.c
@@ -582,8 +582,8 @@ __rte_bpf_validate_debug_evaluate_start(struct rte_bpf_validate_debug *debug,
}
int
-__rte_bpf_validate_debug_evaluate_step(struct rte_bpf_validate_debug *debug,
- uint32_t pc, debug_event_t event)
+__rte_bpf_validate_debug_evaluate_update(struct rte_bpf_validate_debug *debug,
+ uint32_t pc, uint64_t events)
{
int rc;
@@ -595,8 +595,7 @@ __rte_bpf_validate_debug_evaluate_step(struct rte_bpf_validate_debug *debug,
return -ECHILD;
}
- if (pc > debug->bpf_prm->raw.nb_ins || event < 0 ||
- event >= RTE_BPF_VALIDATE_DEBUG_EVENT_END)
+ if (pc > debug->bpf_prm->raw.nb_ins)
return -EINVAL;
debug->pc = pc;
@@ -606,11 +605,13 @@ __rte_bpf_validate_debug_evaluate_step(struct rte_bpf_validate_debug *debug,
rc = debug_send_event(debug,
RTE_BPF_VALIDATE_DEBUG_EVENT_INVALID_STATE);
- if (event != RTE_BPF_VALIDATE_DEBUG_EVENT_STEP)
- rc = rc < 0 ? rc : debug_send_event(debug, event);
+ for (enum rte_bpf_validate_debug_event event = 0;
+ event != RTE_BPF_VALIDATE_DEBUG_EVENT_END; ++event)
+ if (events & RTE_BIT64(event))
+ rc = rc < 0 ? rc : debug_send_event(debug, event);
- if (event == RTE_BPF_VALIDATE_DEBUG_EVENT_STEP ||
- event == RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER)
+ if (events == 0 || events == RTE_BIT64(
+ RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER))
/* Stepping into a real instruction to execute. */
rc = rc < 0 ? rc : debug_points_call_back(debug,
&debug->breakpoint_lists[pc]);
@@ -651,7 +652,8 @@ __rte_bpf_validate_debug_evaluate_finish(struct rte_bpf_validate_debug *debug,
event = RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS;
}
- rc = __rte_bpf_validate_debug_evaluate_step(debug, pc, event);
+ rc = __rte_bpf_validate_debug_evaluate_update(debug, pc,
+ RTE_BIT64(event));
}
debug_evaluate_close(debug);
diff --git a/lib/bpf/bpf_validate_debug.h b/lib/bpf/bpf_validate_debug.h
index a91f3e9c48b2..2b0c3bb646a7 100644
--- a/lib/bpf/bpf_validate_debug.h
+++ b/lib/bpf/bpf_validate_debug.h
@@ -46,7 +46,7 @@ __rte_bpf_validate_debug_evaluate_start(struct rte_bpf_validate_debug *debug,
const struct bpf_verifier *verifier, const struct rte_bpf_prm_ex *bpf_prm);
/*
- * Signal each instruction, branch end, or evaluation end.
+ * Signal evaluation events.
*
* Immediately return 0 if debug is NULL.
*
@@ -54,16 +54,17 @@ __rte_bpf_validate_debug_evaluate_start(struct rte_bpf_validate_debug *debug,
* Validate debug instance configured by user, can be NULL.
* @param pc
* Current value of the program counter, or next after last instruction.
- * @param event
- * Specific evaluation event if any, or RTE_BPF_VALIDATE_DEBUG_EVENT_STEP.
+ * @param events
+ * Bitmask of events.
+ * Step bit is always implied and should not be set.
* @return
* Non-negative value: evaluation should continue;
* -ECANCELED: evaluation should fail without calling this API again;
* Other negative value: evaluation should fail signalling failure;
*/
int
-__rte_bpf_validate_debug_evaluate_step(struct rte_bpf_validate_debug *debug,
- uint32_t pc, debug_event_t event);
+__rte_bpf_validate_debug_evaluate_update(struct rte_bpf_validate_debug *debug,
+ uint32_t pc, uint64_t events);
/*
* Signal end of evaluation process.
--
2.43.0
^ permalink raw reply related [flat|nested] 38+ messages in thread
* [PATCH v2 3/7] bpf/validate: formalize call back requirements
2026-09-14 14:54 ` [PATCH v2 0/7] validate-bpf: add standalone validation debug tool Marat Khalili
2026-09-14 14:54 ` [PATCH v2 1/7] bpf/validate: fix finished status on restart Marat Khalili
2026-09-14 14:54 ` [PATCH v2 2/7] bpf/validate: refactor internal step function Marat Khalili
@ 2026-09-14 14:54 ` Marat Khalili
2026-09-14 14:54 ` [PATCH v2 4/7] bpf/validate: add jump notification events Marat Khalili
` (5 subsequent siblings)
8 siblings, 0 replies; 38+ messages in thread
From: Marat Khalili @ 2026-09-14 14:54 UTC (permalink / raw)
To: Konstantin Ananyev; +Cc: dev
Formalize requirements to evaluation events in documentation and adjust
their handling in the actual code to match.
Particularly, specify more precisely:
* conditions for each catchpoint callback;
* relative order of callbacks within the step;
* expected value of the program counter;
Behaviour changes:
* one branch-return event is now generated for each branch-enter event
(previously multiple branch-return events could be coalesced);
* branch-return event now points to the corresponding conditional jump
(previously past the end of the last branch);
* step event is now only generated before an actual instruction
evaluation (previously also for branching and finishing events);
* program counter is no longer allowed to point past the program end;
These changes should simplify branch tracking of the validator by tests
and interactive applications. E.g. branch-enter and branch-return events
can now be used to push and pop branching points to/from some stack.
Signed-off-by: Marat Khalili <marat.khalili@huawei.com>
---
lib/bpf/bpf_validate.c | 28 +++++++++----
lib/bpf/bpf_validate_debug.c | 72 +++++++++++++++++++-------------
lib/bpf/bpf_validate_debug.h | 3 +-
lib/bpf/rte_bpf_validate_debug.h | 24 +++++++----
4 files changed, 78 insertions(+), 49 deletions(-)
diff --git a/lib/bpf/bpf_validate.c b/lib/bpf/bpf_validate.c
index b317abb81172..f4258c2fb54f 100644
--- a/lib/bpf/bpf_validate.c
+++ b/lib/bpf/bpf_validate.c
@@ -2892,6 +2892,23 @@ prune_eval_state(struct bpf_verifier *bvf, const struct inst_node *node,
return rc;
}
+static bool
+is_branch_start(const struct inst_node *node)
+{
+ return node->prev_node != NULL && node->prev_node->nb_edge > 1;
+}
+
+static uint64_t
+step_events(const struct inst_node *node)
+{
+ uint64_t events = RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_STEP);
+
+ if (is_branch_start(node))
+ events |= RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER);
+
+ return events;
+}
+
/* Do second pass through CFG and try to evaluate instructions
* via each possible path. The verifier will try all paths, tracking types of
* registers used as input to instructions, and updating resulting type via
@@ -2918,7 +2935,6 @@ evaluate(struct bpf_verifier *bvf)
const char *err;
const struct ebpf_insn *ins;
struct inst_node *next, *node;
- int prev_nb_edge; /* branching number of the previous instruction */
int rc, debug_rc;
struct rte_bpf_validate_debug *const debug = bvf->prm->debug;
@@ -2954,7 +2970,6 @@ evaluate(struct bpf_verifier *bvf)
ins = bvf->prm->raw.ins;
node = bvf->in;
next = node;
- prev_nb_edge = 1;
memset(&stats, 0, sizeof(stats));
@@ -2990,8 +3005,7 @@ evaluate(struct bpf_verifier *bvf)
}
rc = __rte_bpf_validate_debug_evaluate_update(debug, idx,
- prev_nb_edge > 1 ?
- RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER) : 0);
+ step_events(node));
if (rc < 0)
break;
@@ -3047,7 +3061,6 @@ evaluate(struct bpf_verifier *bvf)
stats.nb_prune++;
} else {
next->prev_node = node;
- prev_nb_edge = node->nb_edge;
node = next;
}
} else {
@@ -3057,9 +3070,9 @@ evaluate(struct bpf_verifier *bvf)
* and proceed with parent.
*/
- if (prev_nb_edge != 0) {
+ if (is_branch_start(node)) {
rc = __rte_bpf_validate_debug_evaluate_update(
- debug, get_node_idx(bvf, node) + 1,
+ debug, get_node_idx(bvf, node->prev_node),
RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN));
if (rc < 0)
break;
@@ -3067,7 +3080,6 @@ evaluate(struct bpf_verifier *bvf)
node->cur_edge = 0;
save_safe_eval_state(bvf, node);
- prev_nb_edge = 0;
node = node->prev_node;
/* first node will not have prev, signalling finish */
diff --git a/lib/bpf/bpf_validate_debug.c b/lib/bpf/bpf_validate_debug.c
index f76303a7fd2a..db6d13f9cb22 100644
--- a/lib/bpf/bpf_validate_debug.c
+++ b/lib/bpf/bpf_validate_debug.c
@@ -225,6 +225,13 @@ debug_points_call_back(struct rte_bpf_validate_debug *debug,
return rc;
}
+/* Call back all breakpoints for the specified program counter. */
+static int
+debug_trigger_breakpoints(struct rte_bpf_validate_debug *debug, uint32_t pc)
+{
+ return debug_points_call_back(debug, &debug->breakpoint_lists[pc]);
+}
+
/* Call back all catchpoints for the specified event. */
static int
debug_send_event(struct rte_bpf_validate_debug *debug, debug_event_t event)
@@ -585,6 +592,21 @@ int
__rte_bpf_validate_debug_evaluate_update(struct rte_bpf_validate_debug *debug,
uint32_t pc, uint64_t events)
{
+ /* Required order of sent events according to the documentation. */
+ static const enum rte_bpf_validate_debug_event ordered_events[] = {
+ RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_START,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_INVALID_STATE,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_STEP,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_FAILURE,
+ };
+ RTE_BUILD_BUG_ON(
+ RTE_DIM(ordered_events) != RTE_BPF_VALIDATE_DEBUG_EVENT_END);
+
int rc;
if (debug == NULL)
@@ -595,29 +617,31 @@ __rte_bpf_validate_debug_evaluate_update(struct rte_bpf_validate_debug *debug,
return -ECHILD;
}
- if (pc > debug->bpf_prm->raw.nb_ins)
+ if (pc >= debug->bpf_prm->raw.nb_ins)
return -EINVAL;
debug->pc = pc;
rc = __rte_bpf_validate_state_is_valid(debug->verifier);
if (rc == 0)
- rc = debug_send_event(debug,
- RTE_BPF_VALIDATE_DEBUG_EVENT_INVALID_STATE);
+ events |= RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_INVALID_STATE);
+
+ for (uint32_t index = 0; index < RTE_DIM(ordered_events); index++) {
+ const enum rte_bpf_validate_debug_event event =
+ ordered_events[index];
+ if ((events & RTE_BIT64(event)) == 0)
+ continue;
- for (enum rte_bpf_validate_debug_event event = 0;
- event != RTE_BPF_VALIDATE_DEBUG_EVENT_END; ++event)
- if (events & RTE_BIT64(event))
- rc = rc < 0 ? rc : debug_send_event(debug, event);
+ if (event == RTE_BPF_VALIDATE_DEBUG_EVENT_STEP)
+ rc = rc < 0 ? rc : debug_trigger_breakpoints(debug, pc);
- if (events == 0 || events == RTE_BIT64(
- RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER))
- /* Stepping into a real instruction to execute. */
- rc = rc < 0 ? rc : debug_points_call_back(debug,
- &debug->breakpoint_lists[pc]);
+ rc = rc < 0 ? rc : debug_send_event(debug, event);
+ events -= RTE_BIT64(event);
+ }
- rc = rc < 0 ? rc : debug_send_event(debug,
- RTE_BPF_VALIDATE_DEBUG_EVENT_STEP);
+ if (events != 0)
+ /* Received unsupported events. */
+ rc = rc < 0 ? rc : -EINVAL;
return rc;
}
@@ -627,8 +651,6 @@ __rte_bpf_validate_debug_evaluate_finish(struct rte_bpf_validate_debug *debug,
int result)
{
int rc = 0;
- uint32_t pc;
- debug_event_t event;
if (debug == NULL)
return 0;
@@ -641,20 +663,10 @@ __rte_bpf_validate_debug_evaluate_finish(struct rte_bpf_validate_debug *debug,
debug->evaluate_finished = true;
debug->evaluate_result = result;
- if (result != -ECANCELED) {
- if (result < 0) {
- /* Last known pc is the place we failed. */
- pc = debug->pc;
- event = RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_FAILURE;
- } else {
- /* Show program end, not particular instruction. */
- pc = debug->bpf_prm->raw.nb_ins;
- event = RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS;
- }
-
- rc = __rte_bpf_validate_debug_evaluate_update(debug, pc,
- RTE_BIT64(event));
- }
+ if (result != -ECANCELED)
+ rc = debug_send_event(debug, result < 0 ?
+ RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_FAILURE :
+ RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS);
debug_evaluate_close(debug);
diff --git a/lib/bpf/bpf_validate_debug.h b/lib/bpf/bpf_validate_debug.h
index 2b0c3bb646a7..4603841db2ed 100644
--- a/lib/bpf/bpf_validate_debug.h
+++ b/lib/bpf/bpf_validate_debug.h
@@ -53,10 +53,9 @@ __rte_bpf_validate_debug_evaluate_start(struct rte_bpf_validate_debug *debug,
* @param debug
* Validate debug instance configured by user, can be NULL.
* @param pc
- * Current value of the program counter, or next after last instruction.
+ * Current value of the program counter.
* @param events
* Bitmask of events.
- * Step bit is always implied and should not be set.
* @return
* Non-negative value: evaluation should continue;
* -ECANCELED: evaluation should fail without calling this API again;
diff --git a/lib/bpf/rte_bpf_validate_debug.h b/lib/bpf/rte_bpf_validate_debug.h
index f30fa926f10a..b3fedd5428bc 100644
--- a/lib/bpf/rte_bpf_validate_debug.h
+++ b/lib/bpf/rte_bpf_validate_debug.h
@@ -31,25 +31,31 @@ extern "C" {
* Supported validate events.
*
* Valid events begin from 0 and end before `RTE_BPF_VALIDATE_DEBUG_EVENT_END`.
+ *
+ * At any given evaluation step, callbacks are fired in the following order:
+ * - Validation start event;
+ * - Branching and invalid-state events (can be interleaved);
+ * - Instruction breakpoints (before evaluating instruction);
+ * - Step (before evaluating instruction) or validation result (if done) event;
*/
enum rte_bpf_validate_debug_event {
- /* Just before every instruction, at branch or validation end. */
+ /* Just before every evaluated instruction. */
RTE_BPF_VALIDATE_DEBUG_EVENT_STEP,
/* Validator has failed its internal self-checks. */
RTE_BPF_VALIDATE_DEBUG_EVENT_INVALID_STATE,
- /* Start of validation. */
+ /* Start of validation (pc points to first instruction). */
RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_START,
- /* Successful finish of validation. */
+ /* Successful finish (pc undefined). */
RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS,
- /* Finish of validation with error. */
+ /* Finish of validation with error (pc points at error). */
RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_FAILURE,
- /* Beginning of a branch just after the jump. */
+ /* Beginning of a branch evaluation (pc points to branch start). */
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER,
- /* Pruning branch as verified earlier. */
+ /* Pruning branch as verified earlier (pc points to branch start). */
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE,
- /* End of branch verification, after the last verified instruction. */
+ /* End of branch verification (pc points to jump instruction). */
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN,
- /* Pruning branch as dynamically unreachable. */
+ /* Pruning branch as dynamically unreachable (pc points to branch start). */
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE,
/* Number of valid event values. */
RTE_BPF_VALIDATE_DEBUG_EVENT_END,
@@ -208,7 +214,7 @@ rte_bpf_validate_debug_get_last_point(const struct rte_bpf_validate_debug *debug
* @param debug
* Debug instance.
* @return
- * Current program counter being validated, or one after last.
+ * Current program counter being validated.
* UINT32_MAX if no program is being validated.
*/
__rte_experimental
--
2.43.0
^ permalink raw reply related [flat|nested] 38+ messages in thread
* [PATCH v2 4/7] bpf/validate: add jump notification events
2026-09-14 14:54 ` [PATCH v2 0/7] validate-bpf: add standalone validation debug tool Marat Khalili
` (2 preceding siblings ...)
2026-09-14 14:54 ` [PATCH v2 3/7] bpf/validate: formalize call back requirements Marat Khalili
@ 2026-09-14 14:54 ` Marat Khalili
2026-09-14 14:54 ` [PATCH v2 5/7] bpf/validate: add get current event API Marat Khalili
` (4 subsequent siblings)
8 siblings, 0 replies; 38+ messages in thread
From: Marat Khalili @ 2026-09-14 14:54 UTC (permalink / raw)
To: Konstantin Ananyev; +Cc: dev
Add events notifying validate debug client of instructions that may
change program counter. This is particularly useful to an interactive
debugger to track conditional and unconditional jumps without trying to
parse the program and interpret instructions independently.
Signed-off-by: Marat Khalili <marat.khalili@huawei.com>
---
lib/bpf/bpf_validate.c | 10 ++++++++--
lib/bpf/bpf_validate_debug.c | 2 ++
lib/bpf/rte_bpf_validate_debug.h | 5 +++++
3 files changed, 15 insertions(+), 2 deletions(-)
diff --git a/lib/bpf/bpf_validate.c b/lib/bpf/bpf_validate.c
index f4258c2fb54f..d503db836105 100644
--- a/lib/bpf/bpf_validate.c
+++ b/lib/bpf/bpf_validate.c
@@ -2899,13 +2899,19 @@ is_branch_start(const struct inst_node *node)
}
static uint64_t
-step_events(const struct inst_node *node)
+step_events(const struct inst_node *node, uint8_t op)
{
uint64_t events = RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_STEP);
if (is_branch_start(node))
events |= RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER);
+ if (op == (BPF_JMP | BPF_JA))
+ events |= RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS);
+
+ if (node->nb_edge > 1)
+ events |= RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL);
+
return events;
}
@@ -3005,7 +3011,7 @@ evaluate(struct bpf_verifier *bvf)
}
rc = __rte_bpf_validate_debug_evaluate_update(debug, idx,
- step_events(node));
+ step_events(node, op));
if (rc < 0)
break;
diff --git a/lib/bpf/bpf_validate_debug.c b/lib/bpf/bpf_validate_debug.c
index db6d13f9cb22..098a88026d6d 100644
--- a/lib/bpf/bpf_validate_debug.c
+++ b/lib/bpf/bpf_validate_debug.c
@@ -600,6 +600,8 @@ __rte_bpf_validate_debug_evaluate_update(struct rte_bpf_validate_debug *debug,
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE,
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN,
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL,
RTE_BPF_VALIDATE_DEBUG_EVENT_STEP,
RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS,
RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_FAILURE,
diff --git a/lib/bpf/rte_bpf_validate_debug.h b/lib/bpf/rte_bpf_validate_debug.h
index b3fedd5428bc..f1ee8769d06a 100644
--- a/lib/bpf/rte_bpf_validate_debug.h
+++ b/lib/bpf/rte_bpf_validate_debug.h
@@ -35,6 +35,7 @@ extern "C" {
* At any given evaluation step, callbacks are fired in the following order:
* - Validation start event;
* - Branching and invalid-state events (can be interleaved);
+ * - Jump-instruction events (before evaluating jump instruction);
* - Instruction breakpoints (before evaluating instruction);
* - Step (before evaluating instruction) or validation result (if done) event;
*/
@@ -57,6 +58,10 @@ enum rte_bpf_validate_debug_event {
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN,
/* Pruning branch as dynamically unreachable (pc points to branch start). */
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE,
+ /* Program counter points to a jump-always instruction. */
+ RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS,
+ /* Program counter points to a conditional jump instruction. */
+ RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL,
/* Number of valid event values. */
RTE_BPF_VALIDATE_DEBUG_EVENT_END,
};
--
2.43.0
^ permalink raw reply related [flat|nested] 38+ messages in thread
* [PATCH v2 5/7] bpf/validate: add get current event API
2026-09-14 14:54 ` [PATCH v2 0/7] validate-bpf: add standalone validation debug tool Marat Khalili
` (3 preceding siblings ...)
2026-09-14 14:54 ` [PATCH v2 4/7] bpf/validate: add jump notification events Marat Khalili
@ 2026-09-14 14:54 ` Marat Khalili
2026-09-14 14:54 ` [PATCH v2 6/7] app/test: add test for bpf validate debug events Marat Khalili
` (3 subsequent siblings)
8 siblings, 0 replies; 38+ messages in thread
From: Marat Khalili @ 2026-09-14 14:54 UTC (permalink / raw)
To: Konstantin Ananyev; +Cc: dev
Add API for retrieving the event currently processed by a callback.
Useful when single callback is registered for multiple events to put
less pressure on the context argument.
Signed-off-by: Marat Khalili <marat.khalili@huawei.com>
---
lib/bpf/bpf_validate_debug.c | 13 +++++++++++++
lib/bpf/rte_bpf_validate_debug.h | 13 +++++++++++++
2 files changed, 26 insertions(+)
diff --git a/lib/bpf/bpf_validate_debug.c b/lib/bpf/bpf_validate_debug.c
index 098a88026d6d..c8e9e5dd9227 100644
--- a/lib/bpf/bpf_validate_debug.c
+++ b/lib/bpf/bpf_validate_debug.c
@@ -43,6 +43,7 @@ struct rte_bpf_validate_debug {
const struct bpf_verifier *verifier;
const struct rte_bpf_prm_ex *bpf_prm;
struct point_list *breakpoint_lists;
+ enum rte_bpf_validate_debug_event current_event;
struct rte_bpf_validate_debug_point *last_point;
uint32_t pc;
/* Evaluate stage (only tracking `evaluate` part at the moment). */
@@ -236,6 +237,7 @@ debug_trigger_breakpoints(struct rte_bpf_validate_debug *debug, uint32_t pc)
static int
debug_send_event(struct rte_bpf_validate_debug *debug, debug_event_t event)
{
+ debug->current_event = event;
return debug_points_call_back(debug, &debug->catchpoint_lists[event]);
}
@@ -342,6 +344,17 @@ rte_bpf_validate_debug_get_ins(const struct rte_bpf_validate_debug *debug,
return 0;
}
+RTE_EXPORT_EXPERIMENTAL_SYMBOL(rte_bpf_validate_debug_get_event, 26.11)
+enum rte_bpf_validate_debug_event
+rte_bpf_validate_debug_get_event(const struct rte_bpf_validate_debug *debug)
+{
+ if (debug == NULL)
+ /* Just to be fool-proof, not really required by API. */
+ return -EINVAL;
+
+ return debug->current_event;
+}
+
RTE_EXPORT_EXPERIMENTAL_SYMBOL(rte_bpf_validate_debug_get_last_point, 26.07)
struct rte_bpf_validate_debug_point *
rte_bpf_validate_debug_get_last_point(const struct rte_bpf_validate_debug *debug)
diff --git a/lib/bpf/rte_bpf_validate_debug.h b/lib/bpf/rte_bpf_validate_debug.h
index f1ee8769d06a..6d4bf53589e0 100644
--- a/lib/bpf/rte_bpf_validate_debug.h
+++ b/lib/bpf/rte_bpf_validate_debug.h
@@ -194,6 +194,19 @@ int
rte_bpf_validate_debug_get_ins(const struct rte_bpf_validate_debug *debug,
const struct ebpf_insn **ins, uint32_t *nb_ins);
+/**
+ * Get event currently processed by a catchpoint callback.
+ *
+ * @param debug
+ * Debug instance.
+ * @return
+ * Event currently processed by a catchpoint callback.
+ * Undefined if no event is currently being processed.
+ */
+__rte_experimental
+enum rte_bpf_validate_debug_event
+rte_bpf_validate_debug_get_event(const struct rte_bpf_validate_debug *debug);
+
/**
* Get last triggered breakpoint or catchpoint.
*
--
2.43.0
^ permalink raw reply related [flat|nested] 38+ messages in thread
* [PATCH v2 6/7] app/test: add test for bpf validate debug events
2026-09-14 14:54 ` [PATCH v2 0/7] validate-bpf: add standalone validation debug tool Marat Khalili
` (4 preceding siblings ...)
2026-09-14 14:54 ` [PATCH v2 5/7] bpf/validate: add get current event API Marat Khalili
@ 2026-09-14 14:54 ` Marat Khalili
2026-09-14 14:54 ` [PATCH v2 7/7] app/validate-bpf: add BPF validation application Marat Khalili
` (2 subsequent siblings)
8 siblings, 0 replies; 38+ messages in thread
From: Marat Khalili @ 2026-09-14 14:54 UTC (permalink / raw)
To: Konstantin Ananyev; +Cc: dev
Add a test that verifies that eBPF validator generates callbacks with
the number, order and program counter specified in the documentation.
Signed-off-by: Marat Khalili <marat.khalili@huawei.com>
---
app/test/test_bpf_validate.c | 341 ++++++++++++++++++++++++++++++++++-
1 file changed, 337 insertions(+), 4 deletions(-)
diff --git a/app/test/test_bpf_validate.c b/app/test/test_bpf_validate.c
index 066f1fa156ae..ea4621b257f4 100644
--- a/app/test/test_bpf_validate.c
+++ b/app/test/test_bpf_validate.c
@@ -9,10 +9,6 @@
#include <rte_bpf_validate_debug.h>
#include <rte_errno.h>
-/*
- * Tests of BPF validation.
- */
-
extern int test_bpf_validate_logtype;
#define RTE_LOGTYPE_TEST_BPF_VALIDATE test_bpf_validate_logtype
#define TEST_LOG_LINE(level, ...) \
@@ -26,6 +22,343 @@ RTE_LOG_REGISTER(test_bpf_validate_logtype, test.bpf_validate, NOTICE);
/* Special value indicating that register variable is not being used. */
#define NO_REGISTER UINT8_MAX
+static const char *const event_names[RTE_BPF_VALIDATE_DEBUG_EVENT_END] = {
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_STEP] = "step",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_INVALID_STATE] = "invalid-state",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_START] = "validation-start",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS] = "validation-success",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_FAILURE] = "validation-failure",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER] = "branch-enter",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE] = "branch-prune",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN] = "branch-return",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE] = "branch-unreachable",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS] = "jump-always",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL] = "jump-conditional",
+};
+
+/*
+ * Self-test of BPF validation debug events.
+ */
+
+#define EVENT_BIT(name) RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_ ## name)
+#define EVENT_BITS_ALL RTE_LEN2MASK(RTE_BPF_VALIDATE_DEBUG_EVENT_END, uint64_t)
+
+struct test_events_context {
+ /* Tested program. */
+ const struct ebpf_insn *ins;
+ uint32_t nb_ins;
+ /* Number of times each of the events was encountered. */
+ int event_counts[RTE_BPF_VALIDATE_DEBUG_EVENT_END];
+ /* Events active within the current step. */
+ uint64_t active_events;
+ /* Breakpoint was called within the current step. */
+ bool breakpoint_called;
+ /* Currently active conditional jump pc. */
+ uint32_t branch_pc;
+ /* Program counters of all conditional jumps in the current code path. */
+ uint32_t branch_pc_stack[4];
+ size_t branch_pc_stack_length;
+};
+
+/* Push current branch_pc to stack and clear it. */
+static int
+test_events_branch_pc_push(struct test_events_context *ctx)
+{
+ TEST_ASSERT_NOT_EQUAL(ctx->branch_pc, NO_PROGRAM_COUNTER,
+ "no branch pc to push");
+ TEST_ASSERT(ctx->branch_pc_stack_length < RTE_DIM(ctx->branch_pc_stack),
+ "branch stack overflow");
+ ctx->branch_pc_stack[ctx->branch_pc_stack_length++] = ctx->branch_pc;
+ ctx->branch_pc = NO_PROGRAM_COUNTER;
+ return TEST_SUCCESS;
+}
+
+/* Pop branch_pc from stack. */
+static int
+test_events_branch_pc_pop(struct test_events_context *ctx)
+{
+ TEST_ASSERT(ctx->branch_pc_stack_length > 0, "branch stack underflow");
+ ctx->branch_pc = ctx->branch_pc_stack[--ctx->branch_pc_stack_length];
+ return TEST_SUCCESS;
+}
+
+/* Verify breakpoint, checking that it is only called once and remembering it. */
+static int
+test_events_break_cb(struct rte_bpf_validate_debug *debug, void *void_ctx)
+{
+ struct test_events_context *const ctx = void_ctx;
+ const uint32_t pc = rte_bpf_validate_debug_get_pc(debug);
+
+ TEST_LOG_LINE(DEBUG, "%2u: break", pc);
+
+ TEST_ASSERT(!ctx->breakpoint_called,
+ "breakpoint called again at pc %u", pc);
+ ctx->breakpoint_called = true;
+
+ return TEST_SUCCESS;
+}
+
+/*
+ * Verify event, checking that:
+ * - validation-start is called at pc 0;
+ * - branch-return is called at a corresponding conditional jump pc;
+ * - branch-enter and branch-return are correctly paired;
+ * - jump-conditional and jump-always are called for corresponding opcodes;
+ * - jump-conditional and only it is followed by branches events;
+ * - events and breakpoints are correctly ordered;
+ *
+ * Increment event counter to compare against expected counts later.
+ */
+static int
+test_events_catch_cb(struct rte_bpf_validate_debug *debug, void *void_ctx)
+{
+ /* Bitmask of events allowed to precede each given event within the step. */
+ static const uint64_t allowed_preceding_events[] = {
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_START] = 0,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE] =
+ EVENT_BIT(BRANCH_PRUNE) |
+ EVENT_BIT(BRANCH_UNREACHABLE) |
+ EVENT_BIT(BRANCH_RETURN),
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE] =
+ EVENT_BIT(BRANCH_PRUNE) |
+ EVENT_BIT(BRANCH_UNREACHABLE) |
+ EVENT_BIT(BRANCH_RETURN),
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER] =
+ EVENT_BIT(BRANCH_PRUNE) |
+ EVENT_BIT(BRANCH_UNREACHABLE) |
+ EVENT_BIT(BRANCH_RETURN),
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN] =
+ EVENT_BIT(BRANCH_PRUNE) |
+ EVENT_BIT(BRANCH_UNREACHABLE) |
+ EVENT_BIT(BRANCH_RETURN),
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL] =
+ EVENT_BIT(BRANCH_ENTER) |
+ EVENT_BIT(BRANCH_PRUNE) |
+ EVENT_BIT(BRANCH_UNREACHABLE) |
+ EVENT_BIT(BRANCH_RETURN),
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS] =
+ EVENT_BIT(BRANCH_ENTER) |
+ EVENT_BIT(BRANCH_PRUNE) |
+ EVENT_BIT(BRANCH_UNREACHABLE) |
+ EVENT_BIT(BRANCH_RETURN),
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_STEP] =
+ EVENT_BITS_ALL - EVENT_BIT(VALIDATION_SUCCESS),
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS] =
+ EVENT_BIT(BRANCH_PRUNE) |
+ EVENT_BIT(BRANCH_UNREACHABLE) |
+ EVENT_BIT(BRANCH_RETURN),
+ };
+
+ struct test_events_context *const ctx = void_ctx;
+ const enum rte_bpf_validate_debug_event event =
+ rte_bpf_validate_debug_get_event(debug);
+ const uint32_t pc = rte_bpf_validate_debug_get_pc(debug);
+
+ TEST_LOG_LINE(DEBUG, "%2u: %s", pc, event_names[event]);
+
+ /* Verify event program counter and underlying instruction. */
+ switch (event) {
+ case RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_START:
+ TEST_ASSERT_EQUAL(pc, 0,
+ "event %s called at pc %u",
+ event_names[event], pc);
+ break;
+ case RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER:
+ TEST_ASSERT_SUCCESS(test_events_branch_pc_push(ctx),
+ "error pushing branch for event %s at pc %u",
+ event_names[event], pc);
+ break;
+ case RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN:
+ TEST_ASSERT_SUCCESS(test_events_branch_pc_pop(ctx),
+ "error popping branch for event %s at pc %u",
+ event_names[event], pc);
+ TEST_ASSERT_EQUAL(pc, ctx->branch_pc,
+ "event %s called at pc %u after jump pc %u",
+ event_names[event], pc, ctx->branch_pc);
+ break;
+ case RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS:
+ TEST_ASSERT_EQUAL(BPF_CLASS(ctx->ins[pc].code), BPF_JMP,
+ "event %s called at non-jump pc %u",
+ event_names[event], pc);
+ TEST_ASSERT_EQUAL(BPF_OP(ctx->ins[pc].code), BPF_JA,
+ "event %s called at non-jump-always pc %u",
+ event_names[event], pc);
+ break;
+ case RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL:
+ TEST_ASSERT_EQUAL(BPF_CLASS(ctx->ins[pc].code), BPF_JMP,
+ "event %s called at non-jump pc %u",
+ event_names[event], pc);
+ TEST_ASSERT_NOT_EQUAL(BPF_OP(ctx->ins[pc].code), BPF_JA,
+ "event %s called at jump-always pc %u",
+ event_names[event], pc);
+ break;
+ default:
+ break;
+ }
+
+ /* Verify events order. */
+ TEST_ASSERT_EQUAL(
+ (ctx->active_events & ~allowed_preceding_events[event]), 0,
+ "event %s at pc %u preceded by %#jx",
+ event_names[event], pc, (uintmax_t)ctx->active_events);
+
+ /* Update events for the current step as well as branch program counter. */
+ if (event != RTE_BPF_VALIDATE_DEBUG_EVENT_STEP) {
+ TEST_ASSERT(!ctx->breakpoint_called,
+ "event %s called after a breakpoint at pc %u",
+ event_names[event], pc);
+ ctx->active_events |= RTE_BIT64(event);
+ } else {
+ TEST_ASSERT_EQUAL(ctx->branch_pc, NO_PROGRAM_COUNTER,
+ "no branch-enter between branch- and next instruction");
+
+ if (ctx->active_events & EVENT_BIT(JUMP_CONDITIONAL))
+ /* Remember branch instruction pointer, branches to follow. */
+ ctx->branch_pc = pc;
+
+ TEST_ASSERT(ctx->breakpoint_called,
+ "step called before or without a breakpoint at pc %u", pc);
+
+ /* Prepare for the next step. */
+ ctx->active_events = 0;
+ ctx->breakpoint_called = false;
+ }
+
+ /* Update total event counts. */
+ ++ctx->event_counts[event];
+
+ return TEST_SUCCESS;
+}
+
+/*
+ * This test verifies that eBPF validator generates expected numbers of events
+ * of each kind in the expected order (e.g. step after jump after branch).
+ */
+static int
+test_events(void)
+{
+ /*
+ * Program demonstrating the following properties:
+ * - prunnable code paths;
+ * - dynamically unreachable code paths;
+ * - conditional jumps landing on conditional jumps;
+ * - conditional jumps landing on unconditional jumps;
+ */
+ static const struct ebpf_insn ins[] = {
+ /* 0: r0 = 0 */
+ { .code = (EBPF_ALU64 | EBPF_MOV | BPF_K), .dst_reg = EBPF_REG_0, .imm = 0 },
+ /* 1: if r1 > 0 goto 4 */
+ { .code = (BPF_JMP | BPF_JGT | BPF_K), .dst_reg = EBPF_REG_1, .imm = 0, .off = 2 },
+ /* 2: r1 = 0 */
+ { .code = (EBPF_ALU64 | EBPF_MOV | BPF_K), .dst_reg = EBPF_REG_1, .imm = 0 },
+ /* 3: if r0 < 1 goto 8 */
+ { .code = (BPF_JMP | EBPF_JLT | BPF_K), .dst_reg = EBPF_REG_0, .imm = 1, .off = 4 },
+ /* 4: if r1 > 0 goto 6 */
+ { .code = (BPF_JMP | BPF_JGT | BPF_K), .dst_reg = EBPF_REG_1, .imm = 0, .off = 1 },
+ /* 5: goto 9 */
+ { .code = (BPF_JMP | BPF_JA), .off = 3 },
+ /* 6: r1 = 0 */
+ { .code = (EBPF_ALU64 | EBPF_MOV | BPF_K), .dst_reg = EBPF_REG_1, .imm = 0 },
+ /* 7: if r0 > 0 goto 9 */
+ { .code = (BPF_JMP | BPF_JGT | BPF_K), .dst_reg = EBPF_REG_0, .imm = 0, .off = 1 },
+ /* 8: goto 10 */
+ { .code = (BPF_JMP | BPF_JA), .off = 1 },
+ /* 9: r0 = 1 */
+ { .code = (EBPF_ALU64 | EBPF_MOV | BPF_K), .dst_reg = EBPF_REG_0, .imm = 1 },
+ /* 10: exit */
+ { .code = (BPF_JMP | EBPF_EXIT) }
+ };
+
+ /*
+ * Verified code paths (order may vary by library implementation):
+ * 0 -- 1 -- 4 -- 6 -- 7 -- unreachable 9
+ * | | `-- 8 -- 10 -- return 7 -- return 4
+ * | `-- unreachable 5 -- return 1
+ * `-- 2 -- 3 -- prune 8
+ * `-- unreachable 4 -- return 1 -- success 11
+ *
+ * Note that 4 encountered conditional jump instructions generate:
+ * - 4 pairs of branch-enter/branch-return events;
+ * - 3 branch-unreachable events;
+ * - 1 branch-prune event;
+ * for a total count of 8 (4 instructions * 2 branches/instruction).
+ */
+ static const int expected_counts[RTE_BPF_VALIDATE_DEBUG_EVENT_END] = {
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_STEP] = 9,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_START] = 1,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS] = 1,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER] = 4,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE] = 1,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN] = 4,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE] = 3,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS] = 1,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL] = 4,
+ };
+
+ struct test_events_context ctx = {
+ .ins = ins,
+ .nb_ins = RTE_DIM(ins),
+ .branch_pc = NO_PROGRAM_COUNTER,
+ };
+
+ struct rte_bpf_validate_debug *debug = rte_bpf_validate_debug_create();
+ TEST_ASSERT_NOT_NULL(debug, "validate debug create error");
+
+ /* Set breakpoints for all instructions. */
+ for (uint32_t pc = 0; pc != RTE_DIM(ins); ++pc)
+ rte_bpf_validate_debug_break(debug, pc,
+ &(struct rte_bpf_validate_debug_callback){
+ .fn = test_events_break_cb,
+ .ctx = &ctx,
+ });
+
+ /* Set catchpoints for all events except invalid-state. */
+ for (enum rte_bpf_validate_debug_event event = 0;
+ event != RTE_BPF_VALIDATE_DEBUG_EVENT_END; ++event)
+ if (event != RTE_BPF_VALIDATE_DEBUG_EVENT_INVALID_STATE)
+ rte_bpf_validate_debug_catch(debug, event,
+ &(struct rte_bpf_validate_debug_callback){
+ .fn = test_events_catch_cb,
+ .ctx = &ctx,
+ });
+
+ struct rte_bpf *const bpf = rte_bpf_load_ex(&(struct rte_bpf_prm_ex){
+ .sz = sizeof(struct rte_bpf_prm_ex),
+ .origin = RTE_BPF_ORIGIN_RAW,
+ .raw.ins = ins,
+ .raw.nb_ins = RTE_DIM(ins),
+ .prog_arg[0] = { .type = RTE_BPF_ARG_RAW },
+ .nb_prog_arg = 1,
+ .debug = debug,
+ });
+
+ rte_bpf_destroy(bpf);
+ rte_bpf_validate_debug_destroy(debug);
+
+ TEST_ASSERT_NOT_NULL(bpf, "validation failed");
+
+ /* Verify event counts. */
+ for (enum rte_bpf_validate_debug_event event = 0;
+ event != RTE_BPF_VALIDATE_DEBUG_EVENT_END; ++event)
+ TEST_ASSERT_EQUAL(
+ ctx.event_counts[event], expected_counts[event],
+ "expected %d %s events, but got %d",
+ expected_counts[event], event_names[event],
+ ctx.event_counts[event]);
+
+ /* Additionally double-check final stack state. */
+ TEST_ASSERT_EQUAL(ctx.branch_pc_stack_length, 0,
+ "unbalanced branch enters and returns");
+
+ return TEST_SUCCESS;
+}
+
+REGISTER_FAST_TEST(bpf_validate_events_autotest, NOHUGE_OK, ASAN_OK, test_events);
+
+/*
+ * Tests of BPF validation.
+ */
+
/* Sizes of text buffers used for formatting various debug outputs. */
#define VALUE_FORMAT_BUFFER_SIZE 32
#define INTERVAL_FORMAT_BUFFER_SIZE 64
--
2.43.0
^ permalink raw reply related [flat|nested] 38+ messages in thread
* [PATCH v2 7/7] app/validate-bpf: add BPF validation application
2026-09-14 14:54 ` [PATCH v2 0/7] validate-bpf: add standalone validation debug tool Marat Khalili
` (5 preceding siblings ...)
2026-09-14 14:54 ` [PATCH v2 6/7] app/test: add test for bpf validate debug events Marat Khalili
@ 2026-09-14 14:54 ` Marat Khalili
2026-09-14 18:41 ` Stephen Hemminger
2026-09-15 13:13 ` Stephen Hemminger
2026-09-14 18:42 ` [PATCH v2 0/7] validate-bpf: add standalone validation debug tool Stephen Hemminger
2026-09-15 19:26 ` [PATCH v3 " Marat Khalili
8 siblings, 2 replies; 38+ messages in thread
From: Marat Khalili @ 2026-09-14 14:54 UTC (permalink / raw)
To: Thomas Monjalon, Konstantin Ananyev; +Cc: dev
Add a new DPDK application, dpdk-validate-bpf, for pre-validating eBPF
programs for compatibility with the lib/bpf execution context.
The application allows evaluating eBPF programs against the DPDK verifier
before loading them into a real application. It includes an interactive
debugging mode to trace state changes per instruction and understand the
validator's decisions.
Signed-off-by: Marat Khalili <marat.khalili@huawei.com>
---
MAINTAINERS | 2 +
app/meson.build | 1 +
app/validate-bpf/alloc_list.c | 53 ++
app/validate-bpf/args.c | 263 ++++++
app/validate-bpf/debug.c | 1099 ++++++++++++++++++++++++
app/validate-bpf/debug_command.c | 383 +++++++++
app/validate-bpf/debug_command.h | 65 ++
app/validate-bpf/eal_init_args.c | 57 ++
app/validate-bpf/internal.h | 151 ++++
app/validate-bpf/main.c | 77 ++
app/validate-bpf/meson.build | 13 +
app/validate-bpf/parse_decl.c | 611 +++++++++++++
doc/guides/rel_notes/release_26_11.rst | 5 +
doc/guides/tools/index.rst | 1 +
doc/guides/tools/validate_bpf.rst | 97 +++
15 files changed, 2878 insertions(+)
create mode 100644 app/validate-bpf/alloc_list.c
create mode 100644 app/validate-bpf/args.c
create mode 100644 app/validate-bpf/debug.c
create mode 100644 app/validate-bpf/debug_command.c
create mode 100644 app/validate-bpf/debug_command.h
create mode 100644 app/validate-bpf/eal_init_args.c
create mode 100644 app/validate-bpf/internal.h
create mode 100644 app/validate-bpf/main.c
create mode 100644 app/validate-bpf/meson.build
create mode 100644 app/validate-bpf/parse_decl.c
create mode 100644 doc/guides/tools/validate_bpf.rst
diff --git a/MAINTAINERS b/MAINTAINERS
index e99a65d1974d..39cd0a3de6fa 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -1923,8 +1923,10 @@ F: lib/bpf/
F: examples/bpf/
F: app/test/test_bpf*
F: app/test/bpf/
+F: app/validate-bpf/
F: app/test-pmd/bpf_cmd.*
F: doc/guides/prog_guide/bpf_lib.rst
+F: doc/guides/tools/validate_bpf.rst
Graph
M: Jerin Jacob <jerinj@marvell.com>
diff --git a/app/meson.build b/app/meson.build
index 1798db3ae43f..9b59ae483056 100644
--- a/app/meson.build
+++ b/app/meson.build
@@ -33,6 +33,7 @@ apps = [
'test-regex',
'test-sad',
'test-security-perf',
+ 'validate-bpf',
]
if get_option('tests')
diff --git a/app/validate-bpf/alloc_list.c b/app/validate-bpf/alloc_list.c
new file mode 100644
index 000000000000..2f4d18ad3df6
--- /dev/null
+++ b/app/validate-bpf/alloc_list.c
@@ -0,0 +1,53 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include "internal.h"
+#include <stdlib.h>
+
+
+/* Needs to be a power of two. */
+#define START_CAPACITY (1u << 3)
+
+size_t
+alloc_list_append(struct alloc_list *alloc_list, void *ptr)
+{
+ if (alloc_list->count == 0) {
+ RTE_ASSERT(alloc_list->ptrs == NULL);
+ alloc_list->ptrs = malloc(
+ sizeof(alloc_list->ptrs[0]) * START_CAPACITY);
+ RTE_VERIFY(alloc_list->ptrs != NULL);
+ } else if (alloc_list->count >= START_CAPACITY &&
+ /* Power of two detection */
+ (alloc_list->count & (alloc_list->count - 1)) == 0) {
+ /*
+ * We allocate in powers of two, and current count is one of
+ * these powers, so need to reallocate to larger capacity.
+ */
+ RTE_ASSERT(alloc_list->ptrs != NULL);
+ const size_t new_capacity = alloc_list->count * 2;
+ alloc_list->ptrs = realloc(alloc_list->ptrs,
+ sizeof(alloc_list->ptrs[0]) * new_capacity);
+ RTE_VERIFY(alloc_list->ptrs != NULL);
+ }
+ alloc_list->ptrs[alloc_list->count] = ptr;
+ return alloc_list->count++;
+}
+
+void alloc_list_replace(struct alloc_list *alloc_list, size_t index, void *ptr)
+{
+ RTE_ASSERT(index < alloc_list->count);
+ alloc_list->ptrs[index] = ptr;
+}
+
+void alloc_list_free_all(struct alloc_list *alloc_list)
+{
+ /* Copy and clear fields first in case alloc_list itself gets freed. */
+ size_t count = alloc_list->count;
+ void ** const ptrs = alloc_list->ptrs;
+ *alloc_list = (struct alloc_list){};
+
+ while (count != 0)
+ free(ptrs[--count]);
+ free(ptrs);
+}
diff --git a/app/validate-bpf/args.c b/app/validate-bpf/args.c
new file mode 100644
index 000000000000..a23b9bfe152a
--- /dev/null
+++ b/app/validate-bpf/args.c
@@ -0,0 +1,263 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include "internal.h"
+
+#include <ctype.h>
+#include <getopt.h>
+#include <stdlib.h>
+
+#include <rte_errno.h>
+
+
+/* Values to be used in getopt_long option.val. */
+enum app_args {
+ ARG_UNRECOGNIZED = '?',
+ ARG_AUTO = 0, /* Value set by getopt_long when flag is non-NULL. */
+ ARG_HELP, /* Keep this one in the beginning for UX reasons. */
+ ARG_DEBUG,
+ ARG_MBUF_BUF_SIZE,
+ ARG_NO_PROG_ARGS,
+ ARG_PROG_ARG,
+ ARG_SECTION,
+ ARG_XSYM,
+};
+
+/* Program options. */
+static struct option OPTIONS[] = {
+ {
+ .name = "help",
+ .val = ARG_HELP,
+ },
+ {
+ .name = "debug",
+ .val = ARG_DEBUG,
+ },
+ {
+ .name = "mbuf-buf-size",
+ .has_arg = required_argument,
+ .val = ARG_MBUF_BUF_SIZE,
+ },
+ {
+ .name = "no-prog-arg",
+ .val = ARG_NO_PROG_ARGS,
+ },
+ {
+ .name = "no-prog-args",
+ .val = ARG_NO_PROG_ARGS,
+ },
+ {
+ .name = "prog-arg",
+ .has_arg = required_argument,
+ .val = ARG_PROG_ARG,
+ },
+ {
+ .name = "section",
+ .has_arg = required_argument,
+ .val = ARG_SECTION,
+ },
+ {
+ .name = "xsym",
+ .has_arg = required_argument,
+ .val = ARG_XSYM,
+ },
+ { /* terminating zero record */ }
+};
+
+/* Default args value. */
+static const struct args ARGS_DEFAULT = {
+ .bpf_prm = {
+ .sz = sizeof(struct rte_bpf_prm_ex),
+ .origin = RTE_BPF_ORIGIN_ELF_FILE,
+ .elf_file.section = ".text",
+ },
+ .mbuf_buf_size = RTE_MBUF_DEFAULT_BUF_SIZE,
+};
+
+/* Default --prog-arg argument value. */
+static const char * const PROG_ARG_DEFAULT = "struct rte_mbuf *";
+
+
+void
+print_usage(const char *program_name)
+{
+ static const char *const options[][2] = {
+ { "--help", "Display help and exit." },
+ { "--debug", "Enable interactive debug mode." },
+ { "--mbuf-buf-size=MBUF_BUF_SIZE", "Size of the mbuf data buffer (in bytes)." },
+ { "--prog-arg=TYPE", "Expected type of the next BPF program argument (up to 5)." },
+ { "--no-prog-args", "BPF program does not take any arguments." },
+ { "--section=SECTION", "ELF section name in the BPF object file to load." },
+ { "--xsym='TYPE NAME[(TYPE, ...)]'", "External symbol BPF program can access." },
+ };
+
+ printf("USAGE: %s [OPTIONS]... BPF_PATH\n", program_name);
+ printf("OPTIONS:\n");
+ for (int oi = 0; oi != RTE_DIM(options); ++oi)
+ printf("\t%-31s %s\n", options[oi][0], options[oi][1]);
+}
+
+void
+print_defaults(void)
+{
+ printf("DEFAULTS:"
+ " --mbuf-buf-size=%#zx"
+ " --prog-arg='%s'"
+ " --section='%s'\n",
+ (size_t)RTE_MBUF_DEFAULT_BUF_SIZE,
+ PROG_ARG_DEFAULT,
+ ARGS_DEFAULT.bpf_prm.elf_file.section);
+}
+
+static int
+parse_size(size_t *result, const char *text)
+{
+ char *parse_end = NULL;
+
+ errno = 0;
+ const uintmax_t strtoumax_result = strtoumax(text, &parse_end, 0);
+ if (errno != 0 || *parse_end != '\0' || strtoumax_result == 0 ||
+ strtoumax_result > SIZE_MAX)
+ return -1;
+
+ *result = strtoumax_result;
+ return 0;
+}
+
+struct args *
+args_parse(int argc, char *argv[])
+{
+ int val;
+ size_t xsym_alloc_index;
+ struct rte_bpf_xsym *xsym = NULL;
+ const char * const program_name = argv[0];
+
+ /* Allocate args and set aliases for some of its members. */
+ struct args * const args = malloc(sizeof(*args));
+ RTE_VERIFY(args != NULL);
+ struct alloc_list * const alloc_list = &args->_alloc_list;
+ struct rte_bpf_prm_ex * const bpf_prm = &args->bpf_prm;
+
+ /* Set default values. */
+ *args = ARGS_DEFAULT;
+ RTE_VERIFY(parse_arg(&bpf_prm->prog_arg[bpf_prm->nb_prog_arg++],
+ PROG_ARG_DEFAULT) >= 0);
+ bool default_prog_args = true;
+
+ /* Reserve space for xsym in alloc_list. */
+ xsym_alloc_index = alloc_list_append(alloc_list, xsym);
+
+ while ((val = getopt_long(argc, argv, "", OPTIONS, NULL)) != EOF) {
+ int rc = 0;
+ switch (val) {
+ case ARG_AUTO:
+ /* getopt_long made the assignment, nothing to do */
+ break;
+ case ARG_HELP:
+ args->show_help = true;
+ break;
+ case ARG_DEBUG:
+ if (bpf_prm->debug == NULL)
+ bpf_prm->debug = debug_create();
+ if (bpf_prm->debug == NULL) {
+ rc = -rte_errno;
+ fprintf(stderr,
+ "%s: error %d creating debug session\n",
+ program_name, -rc);
+ }
+ break;
+ case ARG_MBUF_BUF_SIZE:
+ rc = parse_size(&args->mbuf_buf_size, optarg);
+ if (rc < 0)
+ fprintf(stderr,
+ "%s: invalid mbuf buf size '%s'\n",
+ program_name, optarg);
+ break;
+ case ARG_NO_PROG_ARGS:
+ bpf_prm->nb_prog_arg = 0;
+ default_prog_args = false;
+ break;
+ case ARG_PROG_ARG:
+ if (default_prog_args) {
+ bpf_prm->nb_prog_arg = 0;
+ default_prog_args = false;
+ }
+
+ if (bpf_prm->nb_prog_arg == RTE_DIM(bpf_prm->prog_arg)) {
+ fprintf(stderr,
+ "%s: at most %d program arguments allowed\n",
+ program_name,
+ (int)RTE_DIM(bpf_prm->prog_arg));
+ rc = -EINVAL;
+ break;
+ }
+
+ rc = parse_arg(&bpf_prm->prog_arg[bpf_prm->nb_prog_arg++],
+ optarg);
+ if (rc < 0)
+ fprintf(stderr,
+ "%s: unrecognized prog arg '%s'\n",
+ program_name, optarg);
+
+ break;
+ case ARG_SECTION:
+ bpf_prm->elf_file.section = optarg;
+ break;
+ case ARG_XSYM:
+ bpf_prm->xsym = xsym = realloc(xsym,
+ sizeof(xsym[0]) * (bpf_prm->nb_xsym + 1));
+ RTE_VERIFY(xsym != NULL);
+ alloc_list_replace(alloc_list, xsym_alloc_index, xsym);
+ rc = parse_xsym(&xsym[bpf_prm->nb_xsym++], optarg,
+ alloc_list);
+ if (rc < 0)
+ fprintf(stderr, "%s: unrecognized xsym '%s'\n",
+ program_name, optarg);
+ break;
+ case ARG_UNRECOGNIZED:
+ args_destroy(args);
+ return NULL;
+ default:
+ rte_panic("Unexpected getopt_long return value %d\n",
+ val);
+ }
+ if (rc < 0) {
+ args_destroy(args);
+ return NULL;
+ }
+ }
+
+ /* Set buf_size to the value specified in command line arguments. */
+ adjust_arg_buf_size(bpf_prm->prog_arg, args->mbuf_buf_size);
+ for (size_t xsymi = 0; xsymi != bpf_prm->nb_xsym; ++xsymi)
+ adjust_xsym_buf_size(&xsym[xsymi], args->mbuf_buf_size);
+
+ /* getopt_long moves all non-options to the end, starting at optind. */
+ const int nb_bpf_path = argc - optind;
+ switch (nb_bpf_path) {
+ case 0:
+ break;
+ case 1:
+ bpf_prm->elf_file.path = argv[optind];
+ break;
+ default:
+ fprintf(stderr, "%s: too many positional arguments\n",
+ program_name);
+ args_destroy(args);
+ return NULL;
+ }
+
+ return args;
+}
+
+void
+args_destroy(struct args *args)
+{
+ if (args == NULL)
+ return;
+ if (args->bpf_prm.debug != NULL)
+ debug_destroy(args->bpf_prm.debug);
+ alloc_list_free_all(&args->_alloc_list);
+ free(args);
+}
diff --git a/app/validate-bpf/debug.c b/app/validate-bpf/debug.c
new file mode 100644
index 000000000000..a2d6a44868c8
--- /dev/null
+++ b/app/validate-bpf/debug.c
@@ -0,0 +1,1099 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include "debug_command.h"
+#include "internal.h"
+
+#include <rte_bpf_validate_debug.h>
+#include <rte_errno.h>
+
+#include <stdlib.h>
+
+
+/* Write single line to the user, currently just to stdout. */
+#define PRINTLN(fmt, ...) do { \
+ RTE_LOG_CHECK_NO_NEWLINE(fmt); \
+ printf(fmt "\n", ##__VA_ARGS__); \
+} while (0)
+
+#define PROMPT "(validate) "
+
+#define INITIAL_CAPACITY 8u
+
+static const char *const event_names[] = {
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_INVALID_STATE] = "invalid-state",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER] = "branch-enter",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE] = "branch-prune",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN] = "branch-return",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE] = "branch-unreachable",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS] = "jump-always",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL] = "jump-conditional",
+};
+
+static const char *const register_names[] = {
+ [EBPF_REG_0] = "r0",
+ [EBPF_REG_1] = "r1",
+ [EBPF_REG_2] = "r2",
+ [EBPF_REG_3] = "r3",
+ [EBPF_REG_4] = "r4",
+ [EBPF_REG_5] = "r5",
+ [EBPF_REG_6] = "r6",
+ [EBPF_REG_7] = "r7",
+ [EBPF_REG_8] = "r8",
+ [EBPF_REG_9] = "r9",
+ [EBPF_REG_10] = "r10",
+};
+
+static const char *const comparison_operator_names[] = {
+ [BPF_JEQ] = "==",
+ [BPF_JGT] = ">",
+ [BPF_JGE] = ">=",
+ [EBPF_JNE] = "!=",
+ [EBPF_JSGT] = "s>",
+ [EBPF_JSGE] = "s>=",
+ [EBPF_JLT] = "<",
+ [EBPF_JLE] = "<=",
+ [EBPF_JSLT] = "s<",
+ [EBPF_JSLE] = "s<=",
+};
+
+enum point_type {
+ POINT_TYPE_BREAK,
+ POINT_TYPE_CATCH,
+};
+
+/* Local representation of points with additional info for UI. */
+struct point_info {
+ struct rte_bpf_validate_debug_point *point;
+ enum point_type type;
+ union {
+ uint32_t pc;
+ enum rte_bpf_validate_debug_event event;
+ };
+};
+
+/* Dynamically growing list of point infos. */
+struct point_infos {
+ struct point_info *elements;
+ uint32_t length;
+ uint32_t capacity;
+};
+
+/* Information about a single conditional or unconditional jump in code path. */
+struct branch_info {
+ uint32_t jump_pc;
+ uint32_t target_pc;
+ bool is_conditional;
+};
+
+/* Dynamically growing stack to track code path. */
+struct branch_stack {
+ struct branch_info *branches;
+ uint32_t length;
+ uint32_t capacity;
+};
+
+/* Flags telling if we should validate again, stopping or not at start. */
+static bool validate_again;
+
+/* List of point infos; element index is its ID for UI. */
+static struct point_infos point_infos;
+
+/* Catchpoint invisible to the user used for step-by-step validation. */
+static struct rte_bpf_validate_debug_point *step_point;
+
+/* Tracking entered branches. */
+static struct branch_stack branch_stack;
+static uint32_t pending_jump_pc = UINT32_MAX;
+static struct rte_bpf_validate_debug_point *jump_always_step_point;
+
+static int
+step_cb(struct rte_bpf_validate_debug *debug, void *ctx);
+static int
+point_cb(struct rte_bpf_validate_debug *debug, void *ctx);
+
+/* Find index of the string in the list. */
+static int
+find_name(const char *name, const char *const *names, int nb_names)
+{
+ if (nb_names < 0)
+ return -EINVAL;
+
+ for (int index = 0; index != nb_names; ++index)
+ if (names[index] != NULL && strcmp(names[index], name) == 0)
+ return index;
+
+ return -ENOENT;
+}
+
+/* Convert event name to its enum value. */
+static enum rte_bpf_validate_debug_event
+parse_event_name(const char *name)
+{
+ const int rc = find_name(name, event_names, RTE_DIM(event_names));
+ if (rc < 0)
+ PRINTLN("Error: invalid event name.");
+
+ return rc;
+}
+
+/* Convert comparison operator name to opcode. */
+static int
+parse_comparison_operator(const char *name)
+{
+ const int rc = find_name(name, comparison_operator_names,
+ RTE_DIM(comparison_operator_names));
+ if (rc < 0)
+ PRINTLN("Error: invalid comparison operator name.");
+
+ return rc;
+}
+
+/* Convert register name to its number. */
+static int
+parse_register(const char *name)
+{
+ const int rc = find_name(name, register_names, RTE_DIM(register_names));
+ if (rc < 0)
+ PRINTLN("Error: invalid register name.");
+
+ return rc;
+}
+
+/* Free local list of point infos (do not destroy points). */
+static void
+point_infos_free(void)
+{
+ free(point_infos.elements);
+ point_infos = (struct point_infos){};
+}
+
+/* Free local branch stack. */
+static void
+branch_stack_free(void)
+{
+ free(branch_stack.branches);
+ branch_stack = (struct branch_stack){};
+}
+
+/* Return existing element from the list of point infos. */
+static struct point_info *
+point_infos_at(uint32_t point_number)
+{
+ RTE_ASSERT(point_number < point_infos.length);
+ RTE_ASSERT(point_infos.elements[point_number].point != NULL);
+ return &point_infos.elements[point_number];
+}
+
+/* Print existing element from the list of point infos. */
+static void
+point_infos_print_at(uint32_t point_number)
+{
+ const struct point_info *const point_info = point_infos_at(point_number);
+
+ switch (point_info->type) {
+ case POINT_TYPE_BREAK:
+ PRINTLN("Breakpoint %d at %d.", point_number,
+ point_info->pc);
+ break;
+ case POINT_TYPE_CATCH:
+ PRINTLN("Catchpoint %d on %s.", point_number,
+ event_names[point_info->event]);
+ break;
+ default:
+ PRINTLN("Point %d of unknown type", point_number);
+ break;
+ }
+}
+
+/* Print all point infos. */
+static int
+point_infos_print_all(void)
+{
+ uint32_t nb_printed = 0;
+ for (uint32_t pn = 0; pn != point_infos.length; ++pn) {
+ const struct point_info *const point_info =
+ &point_infos.elements[pn];
+ if (point_info->point != NULL) {
+ point_infos_print_at(pn);
+ ++nb_printed;
+ }
+ }
+
+ if (nb_printed == 0) {
+ PRINTLN("No breakpoints or catchpoints set.");
+ return -ENOENT;
+ }
+
+ return nb_printed;
+}
+
+/* Allocate space for a new element in the list of point infos. */
+static uint32_t
+point_infos_append(void)
+{
+ if (point_infos.length == point_infos.capacity) {
+ /* Set to initial capacity or double previous one. */
+ point_infos.capacity = RTE_MAX(INITIAL_CAPACITY,
+ point_infos.capacity * 2);
+ point_infos.elements = realloc(point_infos.elements,
+ point_infos.capacity * sizeof(point_infos.elements[0]));
+ RTE_VERIFY(point_infos.elements != NULL);
+ }
+ return point_infos.length++;
+}
+
+/* Allocate space for a new element in the branch stack. */
+static void
+branch_stack_append(const struct branch_info *branch)
+{
+ if (branch_stack.length == branch_stack.capacity) {
+ /* Set to initial capacity or double previous one. */
+ branch_stack.capacity = RTE_MAX(INITIAL_CAPACITY,
+ branch_stack.capacity * 2);
+ branch_stack.branches = realloc(branch_stack.branches,
+ branch_stack.capacity * sizeof(branch_stack.branches[0]));
+ RTE_VERIFY(branch_stack.branches != NULL);
+ }
+ branch_stack.branches[branch_stack.length++] = *branch;
+}
+
+/* Destroy existing element from the list of point infos, printing it first. */
+static void
+point_infos_destroy_existing(uint32_t point_number)
+{
+ point_infos_print_at(point_number);
+
+ struct point_info *const point_info = point_infos_at(point_number);
+ rte_bpf_validate_debug_point_destroy(point_info->point);
+ *point_info = (struct point_info){};
+}
+
+/* Destroy point with specified number if it exists. */
+static int
+point_infos_destroy_at(uint32_t point_number)
+{
+ if (point_number >= point_infos.length ||
+ point_infos.elements[point_number].point == NULL) {
+ PRINTLN("No breakpoint number %d.", point_number);
+ return -ENOENT;
+ }
+
+ point_infos_destroy_existing(point_number);
+ return 1;
+}
+
+/* Destroy all point infos. */
+static int64_t
+point_infos_destroy_all(void)
+{
+ uint32_t nb_destroyed = 0;
+ for (uint32_t pn = 0; pn < point_infos.length; ++pn) {
+ const struct point_info *const point_info =
+ &point_infos.elements[pn];
+ if (point_info->point != NULL) {
+ point_infos_destroy_existing(pn);
+ ++nb_destroyed;
+ }
+ }
+
+ if (nb_destroyed == 0) {
+ PRINTLN("No breakpoints or catchpoints set.");
+ return -ENOENT;
+ }
+
+ return nb_destroyed;
+}
+
+/* Destroy all breakpoints at specified location. */
+static int64_t
+point_infos_destroy_breakpoints(uint32_t pc)
+{
+ uint32_t nb_destroyed = 0;
+ for (uint32_t pn = 0; pn < point_infos.length; ++pn) {
+ const struct point_info *const point_info =
+ &point_infos.elements[pn];
+ if (point_info->point != NULL &&
+ point_info->type == POINT_TYPE_BREAK &&
+ point_info->pc == pc) {
+ point_infos_destroy_existing(pn);
+ ++nb_destroyed;
+ }
+ }
+
+ if (nb_destroyed == 0) {
+ PRINTLN("No breakpoint at %u.", pc);
+ return -ENOENT;
+ }
+
+ return nb_destroyed;
+}
+
+/* Destroy all catchpoints for specified event. */
+static int64_t
+point_infos_destroy_catchpoints(int event)
+{
+ if (event < 0)
+ /* Error was already printed by parse_event_name. */
+ return event;
+
+ uint32_t nb_destroyed = 0;
+ for (uint32_t pn = 0; pn < point_infos.length; ++pn) {
+ const struct point_info *const point_info =
+ &point_infos.elements[pn];
+ if (point_info->point != NULL &&
+ point_info->type == POINT_TYPE_CATCH &&
+ (int)point_info->event == event) {
+ point_infos_destroy_existing(pn);
+ ++nb_destroyed;
+ }
+ }
+
+ if (nb_destroyed == 0) {
+ PRINTLN("No catchpoint on %s.", event_names[event]);
+ return -ENOENT;
+ }
+
+ return nb_destroyed;
+}
+
+/* Create new breakpoint at specified location. */
+static int
+add_breakpoint(struct rte_bpf_validate_debug *debug, uint32_t nb_ins, uint32_t pc)
+{
+ const uint32_t point_number = point_infos_append();
+
+ if (pc >= nb_ins) {
+ PRINTLN("Error: program only has %u instructions.", nb_ins);
+ return -ENOENT;
+ }
+
+ struct rte_bpf_validate_debug_point *const point =
+ rte_bpf_validate_debug_break(debug, pc,
+ &(struct rte_bpf_validate_debug_callback){
+ .fn = point_cb,
+ .ctx = (void *)(uintptr_t)point_number,
+ });
+ if (point == NULL) {
+ PRINTLN("Library error %d.", rte_errno);
+ return -rte_errno;
+ }
+
+ point_infos.elements[point_number] = (struct point_info){
+ .point = point,
+ .type = POINT_TYPE_BREAK,
+ .pc = pc,
+ };
+ point_infos_print_at(point_number);
+ return 0;
+}
+
+/* Create new catchpoint at specified location. */
+static int
+add_catchpoint(struct rte_bpf_validate_debug *debug, int event)
+{
+ if (event < 0)
+ /* Error was already printed by parse_event_name. */
+ return event;
+
+ const uint32_t point_number = point_infos_append();
+ struct rte_bpf_validate_debug_point *const point =
+ rte_bpf_validate_debug_catch(debug, event,
+ &(struct rte_bpf_validate_debug_callback){
+ .fn = point_cb,
+ .ctx = (void *)(uintptr_t)point_number,
+ });
+ if (point == NULL) {
+ PRINTLN("Library error %d.", rte_errno);
+ return -rte_errno;
+ }
+
+ point_infos.elements[point_number] = (struct point_info){
+ .point = point,
+ .type = POINT_TYPE_CATCH,
+ .event = event,
+ };
+ point_infos_print_at(point_number);
+ return 0;
+}
+
+static bool
+is_step_enabled(void)
+{
+ return step_point != NULL;
+}
+
+/* Enable step-by-step validation: make sure catchpoint is set on step event. */
+static int
+enable_step(struct rte_bpf_validate_debug *debug)
+{
+ if (is_step_enabled())
+ return 0;
+
+ step_point = rte_bpf_validate_debug_catch(debug,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_STEP,
+ &(struct rte_bpf_validate_debug_callback){ step_cb });
+ if (step_point == NULL)
+ return -rte_errno;
+
+ return 0;
+}
+
+/* Disable step-by-step validation: destroy catchpoint on step event if any. */
+static void
+disable_step(void)
+{
+ rte_bpf_validate_debug_point_destroy(step_point);
+ step_point = NULL;
+}
+
+/* Format and print information about specified frame offset. */
+static int
+print_frame_offset(struct rte_bpf_validate_debug *debug, int32_t offset)
+{
+ char *info;
+ int info_size, rc;
+
+ if (offset >= 0 || offset % sizeof(uint64_t) != 0) {
+ PRINTLN("Invalid frame offset, must be a negative multiple of %zu.",
+ sizeof(uint64_t));
+ return -EINVAL;
+ }
+
+ rc = rte_bpf_validate_debug_format_frame_info(debug, NULL, 0, offset);
+ if (rc == -ERANGE) {
+ PRINTLN("Offset is out of frame range.");
+ return rc;
+ }
+ if (rc < 0) {
+ PRINTLN("Error %d printing information.", -rc);
+ return rc;
+ }
+
+ info_size = rc + 1;
+ info = malloc(info_size);
+ if (info == NULL)
+ return -ENOMEM;
+
+ rc = rte_bpf_validate_debug_format_frame_info(debug, info, info_size,
+ offset);
+ if (rc + 1 != info_size) {
+ if (rc >= 0) {
+ PRINTLN("Expect format return value %d, got %d.",
+ info_size, rc);
+ rc = -EINVAL;
+ } else
+ PRINTLN("Error %d printing information.", -rc);
+ free(info);
+ return rc;
+ }
+
+ printf("%5jd: \t%s\n", (intmax_t)offset, info);
+ free(info);
+ return 0;
+}
+
+/* Format and print informatiion about the frame. */
+static int
+print_frame(struct rte_bpf_validate_debug *debug)
+{
+ int32_t frame_size;
+ int rc;
+
+ frame_size = rte_bpf_validate_debug_get_frame_size(debug);
+ if (frame_size < 0) {
+ PRINTLN("Error %d getting frame size.", -frame_size);
+ return frame_size;
+ }
+
+ for (int32_t frame_offset = 0;;) {
+ frame_offset -= sizeof(uint64_t);
+ if (frame_offset < -frame_size)
+ break;
+
+ rc = print_frame_offset(debug, frame_offset);
+ if (rc < 0)
+ return rc;
+ }
+
+ return 0;
+}
+
+/* Format and print informatiion about specified register. */
+static int
+print_register(struct rte_bpf_validate_debug *debug, int reg)
+{
+ char *info;
+ int info_size, rc;
+
+ if (reg < 0)
+ /* Error was already printed by parse_register. */
+ return reg;
+
+ rc = rte_bpf_validate_debug_format_register_info(debug, NULL, 0, reg);
+ if (rc < 0) {
+ PRINTLN("Error %d printing information.", -rc);
+ return rc;
+ }
+
+ info_size = rc + 1;
+ info = malloc(info_size);
+ if (info == NULL)
+ return -ENOMEM;
+
+ rc = rte_bpf_validate_debug_format_register_info(debug, info, info_size,
+ reg);
+ if (rc + 1 != info_size) {
+ if (rc >= 0) {
+ PRINTLN("Expect format return value %d, got %d.",
+ info_size, rc);
+ rc = -EINVAL;
+ } else
+ PRINTLN("Error %d printing information.", -rc);
+ free(info);
+ return rc;
+ }
+
+ printf("%5s: \t%s\n", register_names[reg], info);
+ free(info);
+ return 0;
+}
+
+/* Format and print informatiion about all registers. */
+static int
+print_registers(struct rte_bpf_validate_debug *debug)
+{
+ int rc = 0;
+
+ for (int reg = 0; reg != EBPF_REG_NUM; ++reg)
+ rc = rc < 0 ? rc : print_register(debug, reg);
+
+ return rc;
+}
+
+/* List one eBPF program instruction. */
+static int
+list_one(const struct ebpf_insn *ins, uint32_t nb_ins, uint32_t offset,
+ uint32_t pc, const char *comment)
+{
+ char hexadecimal[256], disassembly[256];
+
+ if (offset >= nb_ins) {
+ PRINTLN("Error: program only has %u instructions.", nb_ins);
+ return -EINVAL;
+ }
+
+ ins += offset;
+
+ if (offset == nb_ins - 1 && rte_bpf_insn_is_wide(ins)) {
+ PRINTLN("Error: truncated last instruction.");
+ return -EINVAL;
+ }
+
+ rte_bpf_format(hexadecimal, sizeof(hexadecimal), ins, 0,
+ RTE_BPF_FORMAT_FLAG_HEXADECIMAL |
+ RTE_BPF_FORMAT_FLAG_NEVER_WIDE);
+ rte_bpf_format(disassembly, sizeof(disassembly), ins, offset,
+ RTE_BPF_FORMAT_FLAG_DISASSEMBLY |
+ RTE_BPF_FORMAT_FLAG_ABSOLUTE_JUMPS);
+
+ if (comment == NULL)
+ comment = "";
+ PRINTLN("%2s %10u: \t%s \t%s%s%s",
+ offset == pc ? "=>" : "", offset, hexadecimal, disassembly,
+ comment[0] != '\0' ? " \t; " : "", comment);
+
+ if (rte_bpf_insn_is_wide(ins)) {
+ rte_bpf_format(hexadecimal, sizeof(hexadecimal), ins + 1, 0,
+ RTE_BPF_FORMAT_FLAG_HEXADECIMAL |
+ RTE_BPF_FORMAT_FLAG_NEVER_WIDE);
+ PRINTLN("%15s\t%s", "", hexadecimal);
+ }
+
+ return 0;
+}
+
+/* List specified range of eBPF program instructions, updating start offset. */
+static int
+list(const struct ebpf_insn *ins, uint32_t nb_ins, uint32_t *offset,
+ uint32_t count, uint32_t pc)
+{
+ uint32_t local_offset = 0;
+ if (offset == NULL)
+ offset = &local_offset;
+
+ if (*offset > nb_ins) {
+ PRINTLN("Error: program only has %u instructions.", nb_ins);
+ return -EINVAL;
+ }
+
+ const uint32_t end =
+ /* Calculate end in a way preventing overflow: */
+ *offset + RTE_MIN(nb_ins - *offset, count);
+ while (*offset < end) {
+ const int rc = list_one(ins, nb_ins, *offset, pc, NULL);
+ if (rc < 0)
+ return rc;
+
+ *offset += 1 + rte_bpf_insn_is_wide(&ins[*offset]);
+ }
+
+ return 0;
+}
+
+/* Print if specified conditional jump _may_ be executed. */
+static int
+print_if_may(struct rte_bpf_validate_debug *debug, const struct ebpf_insn *jump,
+ uint64_t imm64)
+{
+ const int result = rte_bpf_validate_debug_may_jump(debug, jump, imm64);
+
+ switch (result) {
+ case 0:
+ case RTE_BPF_VALIDATE_DEBUG_MAY_BE_FALSE:
+ PRINTLN("NO");
+ break;
+ case RTE_BPF_VALIDATE_DEBUG_MAY_BE_TRUE:
+ case RTE_BPF_VALIDATE_DEBUG_MAY_BE_FALSE | RTE_BPF_VALIDATE_DEBUG_MAY_BE_TRUE:
+ PRINTLN("YES");
+ break;
+ default:
+ PRINTLN("Error %d getting result.", -result);
+ break;
+ }
+
+ return result;
+}
+
+/* Print if specified condition with literal right hand side _may_ be true. */
+static int
+print_if_may_literal_rhs(struct rte_bpf_validate_debug *debug,
+ const struct debug_command_comparison *comparison)
+{
+ const int lhs = parse_register(comparison->lhs);
+ const int op = parse_comparison_operator(comparison->op);
+ const int64_t rhs = comparison->literal_rhs;
+
+ if (lhs < 0 || op < 0)
+ /* Error was already printed by parse function. */
+ return lhs < 0 ? lhs : op;
+
+ return print_if_may(debug, &(struct ebpf_insn){
+ .code = BPF_JMP | op | BPF_K,
+ .dst_reg = lhs,
+ }, /* imm64 = */ rhs);
+}
+
+/* Print if specified condition with register right hand side _may_ be true. */
+static int
+print_if_may_register_rhs(struct rte_bpf_validate_debug *debug,
+ const struct debug_command_comparison *comparison)
+{
+ const int lhs = parse_register(comparison->lhs);
+ const int op = parse_comparison_operator(comparison->op);
+ const int rhs = parse_register(comparison->register_rhs);
+
+ if (lhs < 0 || op < 0 || rhs < 0)
+ /* Error was already printed by parse function. */
+ return lhs < 0 ? lhs : op < 0 ? op : rhs;
+
+ return print_if_may(debug, &(struct ebpf_insn){
+ .code = BPF_JMP | op | BPF_X,
+ .dst_reg = lhs,
+ .src_reg = rhs,
+ }, /* imm64 = */ 0);
+}
+
+/* Return 1 on validation success, 0 on failure, -EAGAIN if still running. */
+static int
+get_validation_success(struct rte_bpf_validate_debug *debug)
+{
+ int validation_result, rc;
+
+ rc = rte_bpf_validate_debug_get_validation_result(debug,
+ &validation_result);
+ return rc < 0 ? rc : (validation_result >= 0);
+}
+
+static void
+print_status(const struct ebpf_insn *ins, uint32_t nb_ins,
+ int validation_success, uint32_t pc)
+{
+ if (validation_success == 1) {
+ PRINTLN("Validation succeeded.");
+ return;
+ }
+
+ list_one(ins, nb_ins, pc, pc, NULL);
+
+ if (validation_success == 0)
+ PRINTLN("Validation failed.");
+}
+
+static void
+debug_command_where(const struct ebpf_insn *ins, uint32_t nb_ins,
+ int validation_success, uint32_t pc)
+{
+ for (uint32_t bi = 0; bi != branch_stack.length; ++bi) {
+ const uint32_t jump_pc = branch_stack.branches[bi].jump_pc;
+ const uint32_t target_pc = branch_stack.branches[bi].target_pc;
+ const char *const comment =
+ !branch_stack.branches[bi].is_conditional ? NULL :
+ target_pc == jump_pc + 1 ? "fallen-through" : "taken";
+ list_one(ins, nb_ins, jump_pc, UINT32_MAX, comment);
+ }
+ print_status(ins, nb_ins, validation_success, pc);
+}
+
+/* Return true if pc is defined, otherwise print an error message. */
+static bool
+ensure_pc_defined(uint32_t pc)
+{
+ if (pc != UINT32_MAX)
+ return true;
+
+ PRINTLN("No current instruction.");
+ return false;
+}
+
+/* Return true if still validating, otherwise print an error message. */
+static bool
+ensure_still_validating(int validation_success)
+{
+ if (validation_success == -EAGAIN)
+ return true;
+
+ PRINTLN("Finished, use `start` or `run` to restart, `quit` to quit.");
+ return false;
+}
+
+/* Step-by-step validation callback: read and process user commands. */
+static int
+step_cb(struct rte_bpf_validate_debug *debug, __rte_unused void *ctx)
+{
+ int rc;
+ int validation_success;
+ const struct ebpf_insn *ins;
+ uint32_t nb_ins, pc, list_offset;
+
+ validate_again = false;
+
+ rc = rte_bpf_validate_debug_get_ins(debug, &ins, &nb_ins);
+ if (rc < 0) {
+ PRINTLN("Error %d getting program instructions.", -rc);
+ return rc;
+ }
+
+ validation_success = get_validation_success(debug);
+ if (validation_success < 0 && validation_success != -EAGAIN) {
+ PRINTLN("Error %d getting validation result.",
+ -validation_success);
+ return validation_success;
+ }
+
+ pc = validation_success == 1 ? UINT32_MAX :
+ rte_bpf_validate_debug_get_pc(debug);
+
+ print_status(ins, nb_ins, validation_success, pc);
+
+ list_offset = pc;
+
+ while (true) {
+ switch (debug_command_get(PROMPT)) {
+ case DEBUG_COMMAND_BREAK:
+ if (ensure_pc_defined(pc))
+ add_breakpoint(debug, nb_ins, pc);
+ continue;
+ case DEBUG_COMMAND_BREAK_PC:
+ add_breakpoint(debug, nb_ins, debug_command_parsed.pc);
+ continue;
+ case DEBUG_COMMAND_CATCH:
+ add_catchpoint(debug,
+ parse_event_name(debug_command_parsed.event));
+ continue;
+ case DEBUG_COMMAND_CLEAR:
+ if (ensure_pc_defined(pc))
+ point_infos_destroy_breakpoints(pc);
+ continue;
+ case DEBUG_COMMAND_CLEAR_EVENT:
+ point_infos_destroy_catchpoints(
+ parse_event_name(debug_command_parsed.event));
+ continue;
+ case DEBUG_COMMAND_CLEAR_PC:
+ point_infos_destroy_breakpoints(
+ debug_command_parsed.pc);
+ continue;
+ case DEBUG_COMMAND_CONTINUE:
+ if (ensure_still_validating(validation_success)) {
+ disable_step();
+ return 0;
+ }
+ continue;
+ case DEBUG_COMMAND_DELETE:
+ point_infos_destroy_all();
+ continue;
+ case DEBUG_COMMAND_DELETE_NUMBER:
+ point_infos_destroy_at(
+ debug_command_parsed.point_number);
+ continue;
+ case DEBUG_COMMAND_INFO_FRAME:
+ print_frame(debug);
+ continue;
+ case DEBUG_COMMAND_INFO_FRAME_OFFSET:
+ print_frame_offset(debug,
+ debug_command_parsed.frame_offset);
+ continue;
+ case DEBUG_COMMAND_INFO_POINTS:
+ point_infos_print_all();
+ continue;
+ case DEBUG_COMMAND_INFO_REGISTER:
+ print_register(debug,
+ parse_register(debug_command_parsed.register_));
+ continue;
+ case DEBUG_COMMAND_INFO_REGISTERS:
+ print_registers(debug);
+ continue;
+ case DEBUG_COMMAND_LIST:
+ if (ensure_pc_defined(pc))
+ list(ins, nb_ins, &list_offset, 10, pc);
+ continue;
+ case DEBUG_COMMAND_LIST_COUNT:
+ if (ensure_pc_defined(pc))
+ list(ins, nb_ins, &list_offset,
+ debug_command_parsed.instruction_count, pc);
+ continue;
+ case DEBUG_COMMAND_LIST_PROGRAM:
+ if (ensure_pc_defined(pc))
+ list(ins, nb_ins, NULL, UINT32_MAX, pc);
+ continue;
+ case DEBUG_COMMAND_MAY_LITERAL_RHS:
+ print_if_may_literal_rhs(debug,
+ &debug_command_parsed.comparison);
+ continue;
+ case DEBUG_COMMAND_MAY_REGISTER_RHS:
+ print_if_may_register_rhs(debug,
+ &debug_command_parsed.comparison);
+ continue;
+ case DEBUG_COMMAND_EOF:
+ case DEBUG_COMMAND_QUIT:
+ PRINTLN("Quitting...");
+ return validation_success == -EAGAIN ? -ECANCELED : 0;
+ case DEBUG_COMMAND_RUN:
+ PRINTLN("Re-running...");
+ disable_step();
+ validate_again = true;
+ return -ECANCELED;
+ case DEBUG_COMMAND_START:
+ PRINTLN("Re-starting...");
+ enable_step(debug);
+ validate_again = true;
+ return -ECANCELED;
+ case DEBUG_COMMAND_STEP:
+ if (ensure_still_validating(validation_success))
+ return 0;
+ continue;
+ case DEBUG_COMMAND_WHERE:
+ debug_command_where(ins, nb_ins, validation_success, pc);
+ continue;
+ default:
+ PRINTLN("INTERNAL ERROR");
+ return -ENOTSUP;
+ }
+ }
+}
+
+/* Any point callback: print it and enable step-by-step validation. */
+static int
+point_cb(struct rte_bpf_validate_debug *debug, void *ctx)
+{
+ const uint32_t point_number = (uintptr_t)ctx;
+ point_infos_print_at(point_number);
+ return enable_step(debug);
+}
+
+/* Branch stack machinery. */
+
+static void
+clear_jump_always_step_point(void)
+{
+ rte_bpf_validate_debug_point_destroy(jump_always_step_point);
+ jump_always_step_point = NULL;
+}
+
+static int
+reset_branch_tracking(struct rte_bpf_validate_debug *debug __rte_unused,
+ void *ctx __rte_unused)
+{
+ branch_stack.length = 0;
+ pending_jump_pc = UINT32_MAX;
+ clear_jump_always_step_point();
+ return 0;
+}
+
+/*
+ * Handling the jump-always instructions:
+ * - upon a jump-always event save the pc and set a custom step callback;
+ * - ignore the first custom step callback call (still on the same jump);
+ * - on the second call push the jump pc into stack and delete the callback;
+ */
+
+static int
+jump_always_step_cb(struct rte_bpf_validate_debug *debug, void *ctx __rte_unused)
+{
+ /* Step event is also emitted at the end of the jump instruction itself. */
+ if (rte_bpf_validate_debug_get_pc(debug) == pending_jump_pc)
+ return 0;
+
+ branch_stack_append(&(struct branch_info){
+ .jump_pc = pending_jump_pc,
+ .target_pc = rte_bpf_validate_debug_get_pc(debug),
+ .is_conditional = false,
+ });
+ clear_jump_always_step_point();
+ return 0;
+}
+
+static int
+jump_always_cb(struct rte_bpf_validate_debug *debug, void *ctx __rte_unused)
+{
+ pending_jump_pc = rte_bpf_validate_debug_get_pc(debug);
+ clear_jump_always_step_point();
+ jump_always_step_point = rte_bpf_validate_debug_catch(debug,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_STEP,
+ &(struct rte_bpf_validate_debug_callback){ jump_always_step_cb });
+ return 0;
+}
+
+/*
+ * Handling the jump-conditional instructions:
+ * - upon a jump-conditional event save the pc;
+ * - upon a branch-enter event push the conditional jump pc into stack;
+ * - upon a branch-return event pop conditional jump and all unconditional jumps
+ * preceding it from the stack;
+ * - during step execution notify the user about the step events;
+ */
+
+static int
+jump_conditional_cb(struct rte_bpf_validate_debug *debug, void *ctx __rte_unused)
+{
+ pending_jump_pc = rte_bpf_validate_debug_get_pc(debug);
+ return 0;
+}
+
+static int
+branch_enter_cb(struct rte_bpf_validate_debug *debug, void *ctx __rte_unused)
+{
+ branch_stack_append(&(struct branch_info){
+ .jump_pc = pending_jump_pc,
+ .target_pc = rte_bpf_validate_debug_get_pc(debug),
+ .is_conditional = true,
+ });
+ if (is_step_enabled())
+ PRINTLN("Entered new branch at pc %u.", pending_jump_pc);
+ return 0;
+}
+
+static int
+branch_return_cb(struct rte_bpf_validate_debug *debug __rte_unused,
+ void *ctx __rte_unused)
+{
+ clear_jump_always_step_point();
+
+ while (branch_stack.length > 0) {
+ const struct branch_info branch_info =
+ branch_stack.branches[--branch_stack.length];
+ pending_jump_pc = branch_info.jump_pc;
+ if (branch_info.is_conditional)
+ break;
+ }
+ if (is_step_enabled())
+ PRINTLN("Returned from branch at pc %u.", pending_jump_pc);
+ return 0;
+}
+
+/* Notify user when skipping branches in step mode. */
+
+static int
+branch_prune_cb(struct rte_bpf_validate_debug *debug __rte_unused,
+ void *ctx __rte_unused)
+{
+ if (is_step_enabled())
+ PRINTLN("Prunned branch at pc %u.", pending_jump_pc);
+ return 0;
+}
+
+static int
+branch_unreachable_cb(struct rte_bpf_validate_debug *debug __rte_unused,
+ void *ctx __rte_unused)
+{
+ if (is_step_enabled())
+ PRINTLN("Unreachable branch at pc %u.", pending_jump_pc);
+ return 0;
+}
+
+/* Global initialization and cleanup functions. */
+
+static int
+set_callbacks(struct rte_bpf_validate_debug *debug)
+{
+ static const struct rte_bpf_validate_debug_callback events_callback[
+ RTE_BPF_VALIDATE_DEBUG_EVENT_END] = {
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_START] = { reset_branch_tracking },
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS] = { step_cb },
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_FAILURE] = { step_cb },
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL] = { jump_conditional_cb },
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS] = { jump_always_cb },
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER] = { branch_enter_cb },
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN] = { branch_return_cb },
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE] = { branch_prune_cb },
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE] = { branch_unreachable_cb },
+ };
+
+ for (enum rte_bpf_validate_debug_event event = 0;
+ event < RTE_BPF_VALIDATE_DEBUG_EVENT_END; ++event)
+ if (events_callback[event].fn != NULL &&
+ rte_bpf_validate_debug_catch(debug, event,
+ &events_callback[event]) == NULL)
+ return -rte_errno;
+
+ return 0;
+}
+
+struct rte_bpf_validate_debug *
+debug_create(void)
+{
+
+ int rc = 0;
+
+ struct rte_bpf_validate_debug *const debug = rte_bpf_validate_debug_create();
+ if (debug == NULL)
+ rc = -rte_errno;
+
+ rc = rc < 0 ? rc : set_callbacks(debug);
+
+ rc = rc < 0 ? rc : enable_step(debug);
+
+ if (rc < 0) {
+ debug_destroy(debug);
+ rte_errno = -rc;
+ return NULL;
+ }
+
+ return debug;
+}
+
+void
+debug_destroy(struct rte_bpf_validate_debug *debug)
+{
+ /* No need to destroy created points, destroying debug will do it. */
+ point_infos_free();
+ branch_stack_free();
+ rte_bpf_validate_debug_destroy(debug);
+}
+
+bool
+debug_validate_again(void)
+{
+ return validate_again;
+}
diff --git a/app/validate-bpf/debug_command.c b/app/validate-bpf/debug_command.c
new file mode 100644
index 000000000000..505ae389c94f
--- /dev/null
+++ b/app/validate-bpf/debug_command.c
@@ -0,0 +1,383 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include "debug_command.h"
+
+#include <rte_debug.h>
+#include <cmdline.h>
+#include <cmdline_socket.h>
+#include <cmdline_parse_num.h>
+
+#include <string.h>
+
+
+#define EVENT_PATTERN \
+ "invalid-state#" \
+ "branch-enter#branch-prune#branch-return#branch-unreachable#" \
+ "jump-always#jump-conditional"
+
+#define REGISTER_PATTERN "r0#r1#r2#r3#r4#r5#r6#r7#r8#r9#r10"
+
+#define COMPARISON_OP_PATTERN "==#!=#<#<=#>#>=#s<#s<=#s>#s>="
+
+static void
+handle_command(void *parsed_result, struct cmdline *cl, void *data);
+
+/* Keywords */
+static cmdline_parse_token_string_t break_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "b#break");
+static cmdline_parse_token_string_t may_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "may");
+static cmdline_parse_token_string_t catch_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "catch");
+static cmdline_parse_token_string_t clear_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "clear");
+static cmdline_parse_token_string_t continue_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "c#continue");
+static cmdline_parse_token_string_t delete_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "delete");
+static cmdline_parse_token_string_t info_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "i#info");
+static cmdline_parse_token_string_t info_points_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword,
+ "b#break#breakpoints#points");
+static cmdline_parse_token_string_t info_frame_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "f#frame");
+static cmdline_parse_token_string_t info_registers_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "r#registers");
+static cmdline_parse_token_string_t list_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "l#list");
+static cmdline_parse_token_string_t program_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "program");
+static cmdline_parse_token_string_t quit_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "q#quit");
+static cmdline_parse_token_string_t run_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "run");
+static cmdline_parse_token_string_t start_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "start");
+static cmdline_parse_token_string_t step_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "s#step");
+static cmdline_parse_token_string_t where_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "where");
+
+/* Variable tokens */
+static cmdline_parse_token_string_t comparison_lhs_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, comparison.lhs, REGISTER_PATTERN);
+static cmdline_parse_token_string_t comparison_op_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, comparison.op, COMPARISON_OP_PATTERN);
+static cmdline_parse_token_num_t comparison_rhs_literal_tok =
+ TOKEN_NUM_INITIALIZER(struct debug_command_parsed, comparison.literal_rhs, RTE_INT64);
+static cmdline_parse_token_string_t comparison_rhs_register_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, comparison.register_rhs,
+ REGISTER_PATTERN);
+static cmdline_parse_token_string_t event_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, event, EVENT_PATTERN);
+static cmdline_parse_token_num_t frame_offset_tok =
+ TOKEN_NUM_INITIALIZER(struct debug_command_parsed, frame_offset, RTE_INT64);
+static cmdline_parse_token_num_t instruction_count_tok =
+ TOKEN_NUM_INITIALIZER(struct debug_command_parsed, instruction_count, RTE_UINT32);
+static cmdline_parse_token_num_t pc_tok =
+ TOKEN_NUM_INITIALIZER(struct debug_command_parsed, instruction_offset, RTE_UINT32);
+static cmdline_parse_token_num_t point_number_tok =
+ TOKEN_NUM_INITIALIZER(struct debug_command_parsed, point_number, RTE_UINT32);
+static cmdline_parse_token_string_t register_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, register_, REGISTER_PATTERN);
+
+
+/* Commands */
+static cmdline_parse_inst_t cmd_break = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_BREAK,
+ .help_str = "b|break: break at current instruction",
+ .tokens = {
+ (void *)&break_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_break_pc = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_BREAK_PC,
+ .help_str = "b|break <pc>: break at specified instruction",
+ .tokens = {
+ (void *)&break_tok,
+ (void *)&pc_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_catch = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_CATCH,
+ .help_str = "catch <event>: catch specified event",
+ .tokens = {
+ (void *)&catch_tok,
+ (void *)&event_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_clear = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_CLEAR,
+ .help_str = "clear: delete all breakpoints at current instruction",
+ .tokens = {
+ (void *)&clear_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_clear_event = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_CLEAR_EVENT,
+ .help_str = "clear <event>: delete all catchpoints for specified event",
+ .tokens = {
+ (void *)&clear_tok,
+ (void *)&event_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_clear_pc = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_CLEAR_PC,
+ .help_str = "clear <pc>: delete all breakpoints at specified instruction",
+ .tokens = {
+ (void *)&clear_tok,
+ (void *)&pc_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_continue = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_CONTINUE,
+ .help_str = "c|continue: continue validation",
+ .tokens = {
+ (void *)&continue_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_delete = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_DELETE,
+ .help_str = "delete: delete all breakpoints and catchpoints",
+ .tokens = {
+ (void *)&delete_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_delete_number = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_DELETE_NUMBER,
+ .help_str = "delete <point>: delete specified breakpoint or catchpoint",
+ .tokens = {
+ (void *)&delete_tok,
+ (void *)&point_number_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_info_frame = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_INFO_FRAME,
+ .help_str = "i|info f|frame: show information about all frame locations",
+ .tokens = {
+ (void *)&info_tok,
+ (void *)&info_frame_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_info_frame_offset = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_INFO_FRAME_OFFSET,
+ .help_str = "i|info f|frame -<offset>: show information about specified frame location",
+ .tokens = {
+ (void *)&info_tok,
+ (void *)&info_frame_tok,
+ (void *)&frame_offset_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_info_points = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_INFO_POINTS,
+ .help_str = "i|info b|break|breakpoints|points: "
+ "show information about all breakpoints and catchpoints",
+ .tokens = {
+ (void *)&info_tok,
+ (void *)&info_points_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_info_register = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_INFO_REGISTER,
+ .help_str = "i|info <register>: show information about specified register",
+ .tokens = {
+ (void *)&info_tok,
+ (void *)®ister_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_info_registers = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_INFO_REGISTERS,
+ .help_str = "i|info r|registers: show information about all registers",
+ .tokens = {
+ (void *)&info_tok,
+ (void *)&info_registers_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_list = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_LIST,
+ .help_str = "l|list: list ten instructions",
+ .tokens = {
+ (void *)&list_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_list_count = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_LIST_COUNT,
+ .help_str = "l|list <number>: list specified number of instructions",
+ .tokens = {
+ (void *)&list_tok,
+ (void *)&instruction_count_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_list_program = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_LIST_PROGRAM,
+ .help_str = "l|list program: list whole program",
+ .tokens = {
+ (void *)&list_tok,
+ (void *)&program_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_may_literal_rhs = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_MAY_LITERAL_RHS,
+ .help_str = "may <register> <comparison> <number>: "
+ "check if specified condition _may_ be true",
+ .tokens = {
+ (void *)&may_tok,
+ (void *)&comparison_lhs_tok,
+ (void *)&comparison_op_tok,
+ (void *)&comparison_rhs_literal_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_may_register_rhs = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_MAY_REGISTER_RHS,
+ .help_str = "may <register> <comparison> <register>: "
+ "check if specified condition _may_ be true",
+ .tokens = {
+ (void *)&may_tok,
+ (void *)&comparison_lhs_tok,
+ (void *)&comparison_op_tok,
+ (void *)&comparison_rhs_register_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_quit = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_QUIT,
+ .help_str = "quit: q|quit debugger",
+ .tokens = {
+ (void *)&quit_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_run = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_RUN,
+ .help_str = "run: re-run validation from the start",
+ .tokens = {
+ (void *)&run_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_start = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_START,
+ .help_str = "start: re-start validation and stop at start",
+ .tokens = {
+ (void *)&start_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_step = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_STEP,
+ .help_str = "s|step: validate one instruction",
+ .tokens = {
+ (void *)&step_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_where = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_WHERE,
+ .help_str = "where: show current branch stack",
+ .tokens = {
+ (void *)&where_tok,
+ NULL,
+ }
+};
+
+static cmdline_parse_ctx_t debug_ctx[] = {
+ &cmd_break,
+ &cmd_break_pc,
+ &cmd_catch,
+ &cmd_clear,
+ &cmd_clear_event,
+ &cmd_clear_pc,
+ &cmd_continue,
+ &cmd_delete,
+ &cmd_delete_number,
+ &cmd_info_frame,
+ &cmd_info_frame_offset,
+ &cmd_info_points,
+ &cmd_info_register,
+ &cmd_info_registers,
+ &cmd_list,
+ &cmd_list_count,
+ &cmd_list_program,
+ &cmd_may_literal_rhs,
+ &cmd_may_register_rhs,
+ &cmd_quit,
+ &cmd_run,
+ &cmd_start,
+ &cmd_step,
+ &cmd_where,
+ NULL
+};
+
+/* Receive, fill and return one command. */
+
+static enum debug_command debug_command;
+
+struct debug_command_parsed debug_command_parsed;
+
+static void
+handle_command(void *parsed_result, struct cmdline *cl, void *data)
+{
+ RTE_BUILD_BUG_ON(sizeof(debug_command_parsed) > CMDLINE_PARSE_RESULT_BUFSIZE);
+ memcpy(&debug_command_parsed, parsed_result, sizeof(debug_command_parsed));
+ debug_command = (uintptr_t)data;
+ cmdline_quit(cl);
+}
+
+enum debug_command
+debug_command_get(const char *prompt)
+{
+ debug_command = DEBUG_COMMAND_EOF;
+ struct cmdline *const cmdline = cmdline_stdin_new(debug_ctx, prompt);
+ RTE_VERIFY(cmdline != NULL);
+ cmdline_interact(cmdline);
+ cmdline_stdin_exit(cmdline);
+ /* Clear prompt, or it would prepend first message that follows. */
+ printf("\r%*s\r", (int)strlen(prompt), "");
+ fflush(stdout);
+ return debug_command;
+}
diff --git a/app/validate-bpf/debug_command.h b/app/validate-bpf/debug_command.h
new file mode 100644
index 000000000000..1a59ab251ebb
--- /dev/null
+++ b/app/validate-bpf/debug_command.h
@@ -0,0 +1,65 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include <cmdline_parse_string.h>
+
+#include <stdint.h>
+
+
+struct debug_command_comparison {
+ cmdline_fixed_string_t lhs;
+ cmdline_fixed_string_t op;
+ union {
+ cmdline_fixed_string_t register_rhs;
+ int64_t literal_rhs;
+ };
+};
+
+struct debug_command_parsed {
+ cmdline_fixed_string_t keyword; /* Any keyword we don't need */
+ union {
+ struct debug_command_comparison comparison;
+ cmdline_fixed_string_t event;
+ int32_t frame_offset;
+ uint32_t instruction_count;
+ uint32_t instruction_offset;
+ uint32_t pc;
+ uint32_t point_number;
+ cmdline_fixed_string_t register_;
+ };
+};
+
+enum debug_command {
+ DEBUG_COMMAND_EOF,
+ DEBUG_COMMAND_BREAK,
+ DEBUG_COMMAND_BREAK_PC,
+ DEBUG_COMMAND_CATCH,
+ DEBUG_COMMAND_CLEAR,
+ DEBUG_COMMAND_CLEAR_EVENT,
+ DEBUG_COMMAND_CLEAR_PC,
+ DEBUG_COMMAND_CONTINUE,
+ DEBUG_COMMAND_DELETE,
+ DEBUG_COMMAND_DELETE_NUMBER,
+ DEBUG_COMMAND_INFO_FRAME,
+ DEBUG_COMMAND_INFO_FRAME_OFFSET,
+ DEBUG_COMMAND_INFO_POINTS,
+ DEBUG_COMMAND_INFO_REGISTER,
+ DEBUG_COMMAND_INFO_REGISTERS,
+ DEBUG_COMMAND_LIST,
+ DEBUG_COMMAND_LIST_COUNT,
+ DEBUG_COMMAND_LIST_PROGRAM,
+ DEBUG_COMMAND_MAY_LITERAL_RHS,
+ DEBUG_COMMAND_MAY_REGISTER_RHS,
+ DEBUG_COMMAND_QUIT,
+ DEBUG_COMMAND_RUN,
+ DEBUG_COMMAND_START,
+ DEBUG_COMMAND_STEP,
+ DEBUG_COMMAND_WHERE,
+};
+
+extern struct debug_command_parsed debug_command_parsed;
+
+/** Get and return one command line command, storing its data in the struct above. */
+enum debug_command
+debug_command_get(const char *prompt);
diff --git a/app/validate-bpf/eal_init_args.c b/app/validate-bpf/eal_init_args.c
new file mode 100644
index 000000000000..ba85b7c58807
--- /dev/null
+++ b/app/validate-bpf/eal_init_args.c
@@ -0,0 +1,57 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include "internal.h"
+
+
+#define RTE_EAL_INIT_ARG_SIZE_MAX sizeof("--log-level=lib.eal:warning")
+
+static const char RTE_EAL_INIT_ARGS[][RTE_EAL_INIT_ARG_SIZE_MAX] = {
+ "--log-level=lib.eal:warning",
+ "--no-huge",
+ "--no-pci",
+ "--no-hpet",
+ "--no-shconf",
+};
+
+#define RTE_EAL_INIT_ARGC (/* program name */ 1 + RTE_DIM(RTE_EAL_INIT_ARGS))
+
+int
+get_eal_init_argc(void)
+{
+ return RTE_EAL_INIT_ARGC;
+}
+
+/*
+ * We cannot just return literal strings here, because rte_eal_init accepts
+ * an array of mutable pointers to mutable strings, so literals won't work.
+ * Instead we build a copy of RTE_EAL_INIT_ARGS in a static mutable area.
+ */
+char**
+get_eal_init_argv(char *prog_name)
+{
+ /* Static arrays for mutable copies of actual args */
+ static char mutable_args[RTE_DIM(RTE_EAL_INIT_ARGS)]
+ [RTE_EAL_INIT_ARG_SIZE_MAX];
+ /*
+ * Static array for pointers to args. First element will hold pointer
+ * to the prog_name, last will be set to NULL, the rest will point to
+ * elements of mutable_args with index one smaller.
+ */
+ static char *mutable_ptrs[RTE_EAL_INIT_ARGC + /* terminating NULL */ 1];
+
+ mutable_ptrs[0] = prog_name;
+ for (int argi = 0; argi != RTE_DIM(RTE_EAL_INIT_ARGS); ++argi) {
+ const char * const const_arg = RTE_EAL_INIT_ARGS[argi];
+ char * const mutable_arg = mutable_args[argi];
+ const int snprintf_rc = snprintf(mutable_arg,
+ RTE_EAL_INIT_ARG_SIZE_MAX, "%s", const_arg);
+ RTE_VERIFY(snprintf_rc >= 0 &&
+ (size_t)snprintf_rc < RTE_EAL_INIT_ARG_SIZE_MAX);
+ mutable_ptrs[1 + argi] = mutable_arg;
+ }
+ mutable_ptrs[RTE_DIM(mutable_ptrs) - 1] = NULL;
+
+ return mutable_ptrs;
+}
diff --git a/app/validate-bpf/internal.h b/app/validate-bpf/internal.h
new file mode 100644
index 000000000000..e134e683142a
--- /dev/null
+++ b/app/validate-bpf/internal.h
@@ -0,0 +1,151 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include <rte_bpf.h>
+#include <rte_log.h>
+
+
+extern int validate_bpf_logtype;
+#define RTE_LOGTYPE_VALIDATE_BPF validate_bpf_logtype
+#define VALIDATE_BPF_LOG(level, ...) \
+ RTE_LOG_LINE(level, VALIDATE_BPF, "" __VA_ARGS__)
+
+struct rte_bpf_validate_debug;
+
+/**
+ * List of pointers to allocations, to keep track of things to free.
+ *
+ * May contain NULLs. May contain itself directly or via an outer struct.
+ */
+struct alloc_list {
+ size_t count;
+ void **ptrs;
+};
+
+/** Add new pointer to the alloc list, return its index. */
+size_t alloc_list_append(struct alloc_list *alloc_list, void *ptr);
+
+/** Replace pointer at the specified index in the alloc list. */
+void alloc_list_replace(struct alloc_list *alloc_list, size_t index, void *ptr);
+
+/** Free all allocations in the alloc list */
+void alloc_list_free_all(struct alloc_list *alloc_list);
+
+
+/** Parsed program arguments */
+struct args {
+ /* Allocations list, args.c internal use only. */
+ struct alloc_list _alloc_list;
+
+ /* Set if command line contains --help */
+ bool show_help;
+
+ /* BPF load parameters. */
+ struct rte_bpf_prm_ex bpf_prm;
+
+ /* Size of the mbuf data buffer. */
+ size_t mbuf_buf_size;
+};
+
+/** Print program usage information */
+void
+print_usage(const char *program_name);
+
+/** Print program defaults. */
+void
+print_defaults(void);
+
+/**
+ * Parse command-line arguments
+ *
+ * @param argc
+ * Command-line arguments count, as received by main.
+ * @param argv
+ * Command-line arguments array, as received by main.
+ * Modified during the call due to the use of getopt_long.
+ * Modifying it after the call invalidates the return value.
+ * @return
+ * - parse results in case of success;
+ * - NULL in case of an error (diagnostic will be printed to stderr)
+ */
+struct args *
+args_parse(int argc, char *argv[]);
+
+/** Destroy args struct returned by args_parse */
+void
+args_destroy(struct args *args);
+
+
+/** Value for the rte_eal_init argc argument */
+int
+get_eal_init_argc(void);
+
+/**
+ * Value for the rte_eal_init argv argument
+ *
+ * @param prog_name
+ * Program name, can be obtained from argv[0]
+ */
+char **
+get_eal_init_argv(char *prog_name);
+
+
+/** Print types supported in command-line. */
+void
+print_supported_types(void);
+
+/**
+ * Parse text into arg.
+ * @param arg
+ * Pointer to a variable to put parse result to.
+ * Member `buf_size` of types related to `struct rte_mbuf` is set to
+ * `RTE_MBUF_DEFAULT_BUF_SIZE`, adjust using `adjust_arg_buf_size` if needed.
+ * @param text
+ * Text representation of the type, e.g. "struct rte_mbuf *".
+ * @param alloc_list
+ * Alloc list to use for new allocations.
+ * @return
+ * 0 on success
+ * -1 on failure
+ */
+int
+parse_arg(struct rte_bpf_arg *arg, const char *text);
+
+/**
+ * Parse text into xsym.
+ * @param arg
+ * Pointer to a variable to put parse result to.
+ * Member `buf_size` of types related to `struct rte_mbuf` is set to
+ * `RTE_MBUF_DEFAULT_BUF_SIZE`, adjust using `adjust_xsym_buf_size` if needed.
+ * @param text
+ * Text representation of the external symbol, e.g. "void exit(uint32_t)".
+ * @param alloc_list
+ * Alloc list to use for new allocations.
+ * @return
+ * 0 on success
+ * -1 on failure
+ */
+int
+parse_xsym(struct rte_bpf_xsym *xsym, const char *text,
+ struct alloc_list *alloc_list);
+
+/** If arg->buf_size is non-zero, set it to mbuf_buf_size */
+void
+adjust_arg_buf_size(struct rte_bpf_arg *arg, size_t mbuf_buf_size);
+
+/** If buf_size members are non-zero, set them to mbuf_buf_size */
+void
+adjust_xsym_buf_size(struct rte_bpf_xsym *xsym, size_t mbuf_buf_size);
+
+/** Create and set up global debugging session. */
+struct rte_bpf_validate_debug *
+debug_create(void);
+
+/** Clear and destroy global debugging session. */
+void
+debug_destroy(struct rte_bpf_validate_debug *debug);
+
+/** Tells caller if validation should be re-tried. */
+bool
+debug_validate_again(void);
diff --git a/app/validate-bpf/main.c b/app/validate-bpf/main.c
new file mode 100644
index 000000000000..d8840cb8088e
--- /dev/null
+++ b/app/validate-bpf/main.c
@@ -0,0 +1,77 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include "internal.h"
+
+#include <stdio.h>
+#include <stdlib.h>
+
+#include <rte_bpf.h>
+#include <rte_debug.h>
+#include <rte_eal.h>
+#include <rte_errno.h>
+
+RTE_LOG_REGISTER(validate_bpf_logtype, validate-bpf, NOTICE);
+
+static int
+test_bpf_load(struct rte_bpf_prm_ex *prm)
+{
+ struct rte_bpf * const bpf = rte_bpf_load_ex(prm);
+
+ const int rc = -rte_errno;
+
+ rte_bpf_destroy(bpf);
+
+ return bpf == NULL ? rc : 0;
+}
+
+/* Re-starts validation of asked from interactive debugger. */
+static int
+test_bpf_load_with_restarts(struct rte_bpf_prm_ex *prm)
+{
+ for (;;) {
+ const int rc = test_bpf_load(prm);
+
+ if (rc == -ECANCELED && debug_validate_again())
+ continue;
+
+ if (rc != 0)
+ fprintf(stderr, "Error %d loading BPF: %s\n",
+ -rc, strerror(-rc));
+ else
+ fprintf(stderr, "Validation succeeded.\n");
+
+ return rc;
+ }
+}
+
+int
+main(int argc, char *argv[])
+{
+ struct args * const args = args_parse(argc, argv);
+ if (args == NULL || args->show_help
+ || args->bpf_prm.elf_file.path == NULL) {
+ args_destroy(args);
+ print_usage(argv[0]);
+ if (args == NULL)
+ /* Could not parse arguments. */
+ return 2;
+ print_supported_types();
+ print_defaults();
+ return 0;
+ }
+
+ const int eal_init_argc = get_eal_init_argc();
+ char ** const eal_init_argv = get_eal_init_argv(argv[0]);
+ RTE_VERIFY(rte_eal_init(eal_init_argc, eal_init_argv) ==
+ eal_init_argc - 1);
+
+ const int ret = test_bpf_load_with_restarts(&args->bpf_prm);
+
+ args_destroy(args);
+
+ RTE_VERIFY(rte_eal_cleanup() == 0);
+
+ return ret < 0 ? EXIT_FAILURE : EXIT_SUCCESS;
+}
diff --git a/app/validate-bpf/meson.build b/app/validate-bpf/meson.build
new file mode 100644
index 000000000000..03cfc329efb5
--- /dev/null
+++ b/app/validate-bpf/meson.build
@@ -0,0 +1,13 @@
+# SPDX-License-Identifier: BSD-3-Clause
+# Copyright(c) 2025 Huawei Technologies Co., Ltd
+
+sources = files(
+ 'alloc_list.c',
+ 'args.c',
+ 'debug.c',
+ 'debug_command.c',
+ 'eal_init_args.c',
+ 'main.c',
+ 'parse_decl.c',
+)
+deps = ['bpf', 'cmdline']
diff --git a/app/validate-bpf/parse_decl.c b/app/validate-bpf/parse_decl.c
new file mode 100644
index 000000000000..6bbc20c06798
--- /dev/null
+++ b/app/validate-bpf/parse_decl.c
@@ -0,0 +1,611 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include "internal.h"
+
+#include <ctype.h>
+#include <stdlib.h>
+
+#include <rte_ether.h>
+#include <rte_ip.h>
+#include <rte_mbuf_core.h>
+#include <rte_tcp.h>
+#include <rte_udp.h>
+
+#define RETURN_TEXT_ERROR(text, text_start, message, ...) do { \
+ const int _offset = (text) - (text_start); \
+ VALIDATE_BPF_LOG(ERR, "at offset %d: " message, \
+ _offset, ## __VA_ARGS__); \
+ VALIDATE_BPF_LOG(NOTICE, "%s", text_start); \
+ VALIDATE_BPF_LOG(NOTICE, "%*c", _offset + 1, '^'); \
+ return -1; \
+} while (0)
+
+/* Used in place of any signature, this is not really being checked. */
+static uint64_t
+dummy_function(uint64_t arg1, uint64_t arg2, uint64_t arg3, uint64_t arg4,
+ uint64_t arg5)
+{
+ RTE_SET_USED(arg1);
+ RTE_SET_USED(arg2);
+ RTE_SET_USED(arg3);
+ RTE_SET_USED(arg4);
+ RTE_SET_USED(arg5);
+ return 0;
+}
+
+
+/* TOKENS AND TYPES */
+
+enum token {
+ TOKEN_UNRECOGNIZED = -1,
+ TOKEN_END = 0,
+
+ TOKEN_ASTERISK,
+ TOKEN_BRACKET_CLOSE,
+ TOKEN_BRACKET_OPEN,
+ TOKEN_COMMA,
+ TOKEN_PARENTHESIS_CLOSE,
+ TOKEN_PARENTHESIS_OPEN,
+ TOKEN_STRUCT,
+
+ TOKEN_TYPES_BEGIN,
+
+ TOKEN_TYPE_CHAR,
+ TOKEN_TYPE_ETHER_HEADER,
+ TOKEN_TYPE_INT32_T,
+ TOKEN_TYPE_IP_HEADER,
+ TOKEN_TYPE_IP_HEADERS,
+ TOKEN_TYPE_RTE_MBUF,
+ TOKEN_TYPE_TCP_HEADER,
+ TOKEN_TYPE_TCP_HEADERS,
+ TOKEN_TYPE_UDP_HEADER,
+ TOKEN_TYPE_UDP_HEADERS,
+ TOKEN_TYPE_UINT32_T,
+ TOKEN_TYPE_UINT64_T,
+ TOKEN_TYPE_UINTPTR_T,
+ TOKEN_TYPE_VOID,
+
+ TOKEN_TYPES_END,
+};
+
+/* Return true if the token is a type token */
+static bool
+is_type_token(enum token token)
+{
+ return token > TOKEN_TYPES_BEGIN && token < TOKEN_TYPES_END;
+}
+
+struct text_token {
+ const char *text;
+ unsigned int length; /* Not size_t to fit this struct in 2 regs. */
+ enum token token;
+};
+
+#define TEXT_TOKEN_DEF(text_, token_) { \
+ .text = text_, \
+ .length = sizeof(text_) - 1, \
+ .token = token_, \
+}
+
+/* Token search table, MUST BE SORTED! */
+const struct text_token TEXT_TOKENS[] = {
+ TEXT_TOKEN_DEF("(", TOKEN_PARENTHESIS_OPEN),
+ TEXT_TOKEN_DEF(")", TOKEN_PARENTHESIS_CLOSE),
+ TEXT_TOKEN_DEF("*", TOKEN_ASTERISK),
+ TEXT_TOKEN_DEF(",", TOKEN_COMMA),
+ TEXT_TOKEN_DEF("[", TOKEN_BRACKET_OPEN),
+ TEXT_TOKEN_DEF("]", TOKEN_BRACKET_CLOSE),
+ TEXT_TOKEN_DEF("char", TOKEN_TYPE_CHAR),
+ TEXT_TOKEN_DEF("ether_header", TOKEN_TYPE_ETHER_HEADER),
+ TEXT_TOKEN_DEF("int32_t", TOKEN_TYPE_INT32_T),
+ TEXT_TOKEN_DEF("ip_header", TOKEN_TYPE_IP_HEADER),
+ TEXT_TOKEN_DEF("ip_headers", TOKEN_TYPE_IP_HEADERS),
+ TEXT_TOKEN_DEF("rte_ether_hdr", TOKEN_TYPE_ETHER_HEADER),
+ TEXT_TOKEN_DEF("rte_ipv4_hdr", TOKEN_TYPE_IP_HEADER),
+ TEXT_TOKEN_DEF("rte_mbuf", TOKEN_TYPE_RTE_MBUF),
+ TEXT_TOKEN_DEF("rte_tcp_hdr", TOKEN_TYPE_TCP_HEADER),
+ TEXT_TOKEN_DEF("rte_udp_hdr", TOKEN_TYPE_UDP_HEADER),
+ TEXT_TOKEN_DEF("struct", TOKEN_STRUCT),
+ TEXT_TOKEN_DEF("tcp_header", TOKEN_TYPE_TCP_HEADER),
+ TEXT_TOKEN_DEF("tcp_headers", TOKEN_TYPE_TCP_HEADERS),
+ TEXT_TOKEN_DEF("udp_header", TOKEN_TYPE_UDP_HEADER),
+ TEXT_TOKEN_DEF("udp_headers", TOKEN_TYPE_UDP_HEADERS),
+ TEXT_TOKEN_DEF("uint32_t", TOKEN_TYPE_UINT32_T),
+ TEXT_TOKEN_DEF("uint64_t", TOKEN_TYPE_UINT64_T),
+ TEXT_TOKEN_DEF("uintptr_t", TOKEN_TYPE_UINTPTR_T),
+ TEXT_TOKEN_DEF("void", TOKEN_TYPE_VOID),
+};
+
+const struct text_token TEXT_TOKEN_UNRECOGNIZED = { "", 0, TOKEN_UNRECOGNIZED };
+const struct text_token TEXT_TOKEN_END = { "", 0, TOKEN_END };
+
+/* IP header maximum possible size with options. */
+#define IP_HEADER_MAX_SIZE 60
+#define IP_HEADERS_MAX_SIZE (sizeof(struct rte_ether_hdr) + IP_HEADER_MAX_SIZE)
+#define TCP_HEADERS_MAX_SIZE (IP_HEADERS_MAX_SIZE + sizeof(struct rte_tcp_hdr))
+#define UDP_HEADERS_MAX_SIZE (IP_HEADERS_MAX_SIZE + sizeof(struct rte_udp_hdr))
+
+#define POINTER_SIZE (sizeof(char *))
+
+const size_t TYPE_TOKEN_SIZE[] = {
+ [TOKEN_TYPE_CHAR] = sizeof(char),
+ [TOKEN_TYPE_ETHER_HEADER] = sizeof(struct rte_ether_hdr),
+ [TOKEN_TYPE_INT32_T] = sizeof(int32_t),
+ [TOKEN_TYPE_IP_HEADERS] = IP_HEADERS_MAX_SIZE,
+ [TOKEN_TYPE_IP_HEADER] = sizeof(struct rte_ipv4_hdr),
+ [TOKEN_TYPE_TCP_HEADERS] = TCP_HEADERS_MAX_SIZE,
+ [TOKEN_TYPE_TCP_HEADER] = sizeof(struct rte_tcp_hdr),
+ [TOKEN_TYPE_UDP_HEADERS] = UDP_HEADERS_MAX_SIZE,
+ [TOKEN_TYPE_UDP_HEADER] = sizeof(struct rte_udp_hdr),
+ [TOKEN_TYPE_UINT32_T] = sizeof(uint32_t),
+ [TOKEN_TYPE_UINT64_T] = sizeof(uint64_t),
+ [TOKEN_TYPE_UINTPTR_T] = sizeof(uintptr_t),
+};
+
+/*
+ * Struct rte_bpf_arg augmented with a type of pointer to it, since rte_bpf_arg
+ * by itself may not contain enough information to determine its pointer type.
+ */
+struct arg_info {
+ struct rte_bpf_arg value;
+ enum rte_bpf_arg_type ptr_type;
+ bool is_array;
+};
+
+/* Return arg_info struct described by the specified type token. */
+static struct arg_info
+get_type_token_arg(enum token type)
+{
+ RTE_ASSERT(is_type_token(type));
+ switch (type) {
+ case TOKEN_TYPE_VOID:
+ return (struct arg_info){
+ .value = { .type = RTE_BPF_ARG_UNDEF },
+ .ptr_type = RTE_BPF_ARG_PTR,
+ };
+ case TOKEN_TYPE_RTE_MBUF:
+ return (struct arg_info){
+ .value = {
+ .type = RTE_BPF_ARG_RAW,
+ .size = sizeof(struct rte_mbuf),
+ .buf_size = RTE_MBUF_DEFAULT_BUF_SIZE,
+ },
+ .ptr_type = RTE_BPF_ARG_PTR_MBUF,
+ };
+ default:
+ /* Should only reach here for normal sized types. */
+ RTE_ASSERT((size_t)type < RTE_DIM(TYPE_TOKEN_SIZE) &&
+ TYPE_TOKEN_SIZE[type] != 0);
+ return (struct arg_info){
+ .value = {
+ .type = RTE_BPF_ARG_RAW,
+ .size = TYPE_TOKEN_SIZE[type],
+ },
+ .ptr_type = RTE_BPF_ARG_PTR,
+ };
+ }
+}
+
+
+/* PARSING TEXT INTO TOKENS */
+
+/* Return true if character matches [a-zA-Z0-9_] in regex */
+static bool
+iswordchar(char character)
+{
+ return isalnum(character) || character == '_';
+}
+
+/*
+ * Compare pointer to text with pointer to struct text_token to determine
+ * if text starts with the specified token.
+ */
+static int
+text_and_text_token_cmp(const void *text_void_ptr, const void *text_token_void_ptr)
+{
+ int result;
+ const char * const text = text_void_ptr;
+ const struct text_token * const text_token = text_token_void_ptr;
+
+ if (memchr(text, 0, text_token->length) != NULL)
+ /* Text is shorter than the token. */
+ return strcmp(text, text_token->text);
+
+ /* Cannot use strcmp because we are looking for a prefix. */
+ result = memcmp(text, text_token->text, text_token->length);
+
+ /* Checking the case of a partial word match. */
+ if (result == 0 && iswordchar(text[text_token->length - 1]) &&
+ iswordchar(text[text_token->length]))
+ /* Text word is longer than the token. */
+ result = 1;
+
+ return result;
+}
+
+/* Advance pointed character pointer to the first non-space character. */
+static void
+skip_space(const char **text_ptr)
+{
+ while (isspace(**text_ptr))
+ ++*text_ptr;
+}
+
+/*
+ * Recognize and return a token starting text.
+ * Return TEXT_TOKEN_END if text is empty.
+ * Return TEXT_TOKEN_UNRECOGNIZED if text starts with unknown token.
+ */
+static struct text_token
+peek_text_token(const char *text)
+{
+ if (*text == '\0')
+ return TEXT_TOKEN_END;
+ const struct text_token * const text_token = bsearch(text, TEXT_TOKENS,
+ RTE_DIM(TEXT_TOKENS), sizeof(TEXT_TOKENS[0]),
+ text_and_text_token_cmp);
+ if (text_token == NULL)
+ return TEXT_TOKEN_UNRECOGNIZED;
+ return *text_token;
+}
+
+/*
+ * Advance pointed text starting with the specified token to the first
+ * non-space character after it.
+ * Do nothing if called for TEXT_TOKEN_END or TEXT_TOKEN_UNRECOGNIZED.
+ */
+static void
+consume_text_token(const char **text_ptr, struct text_token text_token)
+{
+ RTE_ASSERT(memcmp(*text_ptr, text_token.text, text_token.length) == 0);
+ *text_ptr += text_token.length;
+ skip_space(text_ptr);
+}
+
+/* Return length of the word starting at `text`. */
+static size_t
+find_word_length(const char *text)
+{
+ size_t word_length = 0;
+ while (iswordchar(text[word_length]))
+ ++word_length;
+ return word_length;
+}
+
+/* Create a copy of the word starting text, advance to next token. */
+static const char *
+take_name(const char **text_ptr, struct alloc_list *alloc_list)
+{
+ const size_t word_length = find_word_length(*text_ptr);
+ if (word_length == 0)
+ /* Text does not start with a word. */
+ return NULL;
+
+ /* Allocate memory for the word and add it to the alloc_list */
+ char *word = malloc(word_length + 1);
+ RTE_VERIFY(word != NULL);
+ alloc_list_append(alloc_list, word);
+
+ /* Copy and terminate word contents. */
+ memcpy(word, *text_ptr, word_length);
+ word[word_length] = '\0';
+
+ /* Advance text pointer. */
+ *text_ptr += word_length;
+ skip_space(text_ptr);
+
+ return word;
+}
+
+/* Read a number starting text, advance to next token. */
+static int
+take_number(size_t *number, const char **text_ptr)
+{
+ /* Read a word and let strtoull to decide if it's a number. */
+ const size_t word_length = find_word_length(*text_ptr);
+ if (word_length == 0)
+ /* Text does not start with a word. */
+ return -ENOENT;
+
+ errno = 0;
+ char *number_end;
+ unsigned long long long_number = strtoull(*text_ptr, &number_end, 0);
+ if (errno > 0)
+ return -errno;
+ if (number_end != *text_ptr + word_length)
+ /* Could not parse whole word. */
+ return -EINVAL;
+ if (long_number > SIZE_MAX)
+ return -ERANGE;
+
+ *number = long_number;
+
+ /* Advance text pointer. */
+ *text_ptr += word_length;
+ skip_space(text_ptr);
+
+ return 0;
+}
+
+
+/* PARSING DECLARATION PARTS */
+
+/* Change arg into a reference. */
+static void
+change_into_reference(struct arg_info *arg)
+{
+ if (RTE_BPF_ARG_PTR_TYPE(arg->value.type) != 0) {
+ VALIDATE_BPF_LOG(WARNING,
+ "After taking reference to a pointer the latter "
+ "will be described as an opaque pointer-size blob.");
+ RTE_ASSERT(arg->ptr_type == RTE_BPF_ARG_PTR);
+ arg->value.size = POINTER_SIZE;
+ }
+ arg->value.type = arg->ptr_type;
+ arg->ptr_type = RTE_BPF_ARG_PTR;
+ arg->is_array = false;
+}
+
+/*
+ * Recognize and consume arg starting text, advance to next token.
+ */
+static int
+take_arg(struct arg_info *arg, const char **text_ptr, const char *text_start)
+{
+ struct text_token next = peek_text_token(*text_ptr);
+
+ if (next.token == TOKEN_STRUCT) {
+ consume_text_token(text_ptr, next);
+ next = peek_text_token(*text_ptr);
+ }
+
+ if (!is_type_token(next.token))
+ RETURN_TEXT_ERROR(*text_ptr, text_start, "expect type");
+ const enum token type = next.token;
+ consume_text_token(text_ptr, next);
+ next = peek_text_token(*text_ptr);
+
+ *arg = get_type_token_arg(type);
+
+ while (next.token == TOKEN_ASTERISK) {
+ consume_text_token(text_ptr, next);
+ next = peek_text_token(*text_ptr);
+ change_into_reference(arg);
+ }
+
+ while (next.token == TOKEN_BRACKET_OPEN) {
+ consume_text_token(text_ptr, next);
+
+ size_t array_length;
+ if (take_number(&array_length, text_ptr) < 0)
+ RETURN_TEXT_ERROR(*text_ptr, text_start, "expect length");
+ next = peek_text_token(*text_ptr);
+
+ if (next.token != TOKEN_BRACKET_CLOSE)
+ RETURN_TEXT_ERROR(*text_ptr, text_start, "expect ']'");
+ consume_text_token(text_ptr, next);
+ next = peek_text_token(*text_ptr);
+
+ change_into_reference(arg);
+ arg->value.size *= array_length;
+ arg->is_array = true;
+ }
+
+ return 0;
+}
+
+/* Fill struct rte_bpf_arg within xsym trying not to unzero the padding. */
+static void
+fill_xsym_arg(struct rte_bpf_arg *target, struct rte_bpf_arg source)
+{
+ /* Copy fields individually to try and prevent copying the padding. */
+ target->type = source.type;
+ target->size = source.size;
+ target->buf_size = source.buf_size;
+}
+
+/* Build and return struct rte_bpf_arg of type RTE_BPF_XTYPE_VAR */
+static void
+fill_var_xsym(struct rte_bpf_xsym *xsym, struct arg_info arg, const char *name,
+ struct alloc_list *alloc_list)
+{
+ /* Variables are passed by reference, except for arrays. */
+ if (!arg.is_array) {
+ if (RTE_BPF_ARG_PTR_TYPE(arg.value.type) != 0)
+ VALIDATE_BPF_LOG(WARNING,
+ "External pointers may not work as expected "
+ "because all external variables are passed by "
+ "reference but there is currently no way to "
+ "describe double pointer to the validator.");
+ change_into_reference(&arg);
+ }
+
+ /* Allocate something to assign to a val pointer. */
+ void * const val = calloc(1, RTE_MAX(1u, arg.value.size));
+ RTE_VERIFY(val != 0);
+ alloc_list_append(alloc_list, val);
+
+ /* Need all padding and unused fields to be zero-filled. */
+ memset(xsym, 0, sizeof(*xsym));
+ xsym->name = name;
+ xsym->type = RTE_BPF_XTYPE_VAR;
+ xsym->var.val = val;
+ fill_xsym_arg(&xsym->var.desc, arg.value);
+}
+
+/* Build and return struct rte_bpf_arg of type RTE_BPF_XTYPE_FUNC */
+static void
+fill_func_xsym(struct rte_bpf_xsym *xsym, struct arg_info arg, const char *name)
+{
+ /* Need all padding and unused fields to be zero-filled. */
+ memset(xsym, 0, sizeof(*xsym));
+ xsym->name = name;
+ xsym->type = RTE_BPF_XTYPE_FUNC;
+ xsym->func.val = &dummy_function;
+ fill_xsym_arg(&xsym->func.ret, arg.value);
+}
+
+/*
+ * Parse and store function arguments, advance to next token after ')'.
+ * Value of *text_ptr should point to the next token after '('.
+ */
+static int
+take_func_xsym_args(struct rte_bpf_xsym *xsym, const char **text_ptr,
+ const char *text_start)
+{
+ struct arg_info arg;
+ struct text_token delimiter = peek_text_token(*text_ptr);
+
+ while (delimiter.token != TOKEN_PARENTHESIS_CLOSE) {
+ if (xsym->func.nb_args == EBPF_FUNC_MAX_ARGS)
+ RETURN_TEXT_ERROR(*text_ptr, text_start,
+ "too many arguments, maximum %d allowed",
+ EBPF_FUNC_MAX_ARGS);
+
+ if (take_arg(&arg, text_ptr, text_start) < 0)
+ return -1;
+ if (arg.value.type == RTE_BPF_ARG_UNDEF &&
+ xsym->func.nb_args != 0)
+ RETURN_TEXT_ERROR(*text_ptr, text_start,
+ "arguments of type void are not allowed");
+ fill_xsym_arg(&xsym->func.args[xsym->func.nb_args++],
+ arg.value);
+
+ delimiter = peek_text_token(*text_ptr);
+ switch (delimiter.token) {
+ case TOKEN_COMMA:
+ consume_text_token(text_ptr, delimiter);
+ continue;
+ case TOKEN_PARENTHESIS_CLOSE:
+ break;
+ default:
+ RETURN_TEXT_ERROR(*text_ptr, text_start,
+ "expect ')' or ','");
+ }
+ }
+ consume_text_token(text_ptr, delimiter);
+
+ /* Special case of single void argument. */
+ if (xsym->func.nb_args == 1 &&
+ xsym->func.args[0].type == RTE_BPF_ARG_UNDEF) {
+ xsym->func.nb_args = 0;
+ /* Need all padding and unused fields to be zero-filled. */
+ fill_xsym_arg(&xsym->func.args[0], (struct rte_bpf_arg){});
+ }
+
+ return 0;
+}
+
+/* Make sure text has ended. */
+static int
+ensure_end(const char *text, const char *text_start)
+{
+ if (peek_text_token(text).token != TOKEN_END)
+ RETURN_TEXT_ERROR(text, text_start, "trailing garbage");
+ return 0;
+}
+
+/* Parse and store xsym, advance to next token. */
+static int
+take_xsym(struct rte_bpf_xsym *xsym, const char **text_ptr,
+ const char *text_start, struct alloc_list *alloc_list)
+{
+ struct arg_info arg;
+
+ if (take_arg(&arg, text_ptr, text_start) < 0)
+ return -1;
+
+ const char * const name = take_name(text_ptr, alloc_list);
+ if (name == NULL)
+ RETURN_TEXT_ERROR(*text_ptr, text_start, "expect name");
+
+ const struct text_token next = peek_text_token(*text_ptr);
+ switch (next.token) {
+ case TOKEN_END:
+ fill_var_xsym(xsym, arg, name, alloc_list);
+ break;
+ case TOKEN_PARENTHESIS_OPEN:
+ consume_text_token(text_ptr, next);
+ fill_func_xsym(xsym, arg, name);
+ if (take_func_xsym_args(xsym, text_ptr, text_start) < 0)
+ return -1;
+ break;
+ default:
+ RETURN_TEXT_ERROR(*text_ptr, text_start,
+ "expect '(' or text end");
+ }
+ return 0;
+}
+
+
+/* PUBLIC FUNCTIONS */
+
+void
+print_supported_types(void)
+{
+ printf("TYPE: [struct] BASIC_TYPE [*]... [[N]]...\n");
+ printf("BASIC_TYPE: one of\n");
+ for (int tti = 0; tti != RTE_DIM(TEXT_TOKENS); ++tti) {
+ if (is_type_token(TEXT_TOKENS[tti].token))
+ printf("\t%s\n", TEXT_TOKENS[tti].text);
+ }
+}
+
+int
+parse_arg(struct rte_bpf_arg *arg, const char *text)
+{
+ struct arg_info arg_info;
+ const char * const text_start = text;
+ skip_space(&text);
+ if (take_arg(&arg_info, &text, text_start) < 0)
+ return -1;
+ if (ensure_end(text, text_start) < 0)
+ return -1;
+ if (arg_info.ptr_type != RTE_BPF_ARG_PTR)
+ VALIDATE_BPF_LOG(WARNING,
+ "`%s` has a special pointer type which was left unused; "
+ "argument was set to an opaque blob.",
+ text_start);
+ *arg = arg_info.value;
+ return 0;
+}
+
+int
+parse_xsym(struct rte_bpf_xsym *xsym, const char *text,
+ struct alloc_list *alloc_list)
+{
+ const char * const text_start = text;
+ skip_space(&text);
+ if (take_xsym(xsym, &text, text_start, alloc_list) < 0)
+ return -1;
+ if (ensure_end(text, text_start) < 0)
+ return -1;
+ return 0;
+}
+
+void
+adjust_arg_buf_size(struct rte_bpf_arg *arg, size_t mbuf_buf_size)
+{
+ if (arg->buf_size != 0)
+ arg->buf_size = mbuf_buf_size;
+}
+
+void
+adjust_xsym_buf_size(struct rte_bpf_xsym *xsym, size_t mbuf_buf_size)
+{
+ switch (xsym->type) {
+ case RTE_BPF_XTYPE_FUNC:
+ for (uint32_t argi = 0; argi != xsym->func.nb_args; ++argi)
+ adjust_arg_buf_size(&xsym->func.args[argi],
+ mbuf_buf_size);
+ adjust_arg_buf_size(&xsym->func.ret, mbuf_buf_size);
+ break;
+ case RTE_BPF_XTYPE_VAR:
+ adjust_arg_buf_size(&xsym->var.desc, mbuf_buf_size);
+ break;
+ default:
+ rte_panic("Unexpected xsym type %d\n", xsym->type);
+ }
+}
diff --git a/doc/guides/rel_notes/release_26_11.rst b/doc/guides/rel_notes/release_26_11.rst
index 87c7e81bdebd..5c0caba50d07 100644
--- a/doc/guides/rel_notes/release_26_11.rst
+++ b/doc/guides/rel_notes/release_26_11.rst
@@ -55,6 +55,11 @@ New Features
Also, make sure to start the actual text at the margin.
=======================================================
+* **Added BPF validation application.**
+
+ Added ``dpdk-validate-bpf`` tool to pre-validate eBPF programs for
+ compatibility with the ``lib/bpf`` execution context.
+
Removed Items
-------------
diff --git a/doc/guides/tools/index.rst b/doc/guides/tools/index.rst
index 13f75a5bc652..1c43060e15cd 100644
--- a/doc/guides/tools/index.rst
+++ b/doc/guides/tools/index.rst
@@ -26,3 +26,4 @@ DPDK Tools User Guides
testmldev
graph
dts
+ validate_bpf
diff --git a/doc/guides/tools/validate_bpf.rst b/doc/guides/tools/validate_bpf.rst
new file mode 100644
index 000000000000..9e2d58cfbe5f
--- /dev/null
+++ b/doc/guides/tools/validate_bpf.rst
@@ -0,0 +1,97 @@
+.. SPDX-License-Identifier: BSD-3-Clause
+ Copyright(c) 2026 The DPDK contributors
+
+dpdk-validate-bpf Application
+=============================
+
+The ``dpdk-validate-bpf`` tool is an application that allows evaluating BPF
+programs against the ``lib/bpf`` library. It can be used to pre-validate BPF
+programs before loading them in a real application to ensure that they pass the
+internal verifier. This makes it a useful tool for integrating into build
+systems and continuous integration (CI) pipelines to verify the correctness
+of BPF programs during compilation, as well as for debugging validation issues.
+
+Running the Application
+-----------------------
+
+The tool requires a path to a compiled BPF object file. It also provides options
+to configure the execution environment and external symbol definitions.
+
+For a comprehensive list of options and their meanings, refer to the
+built-in help by running ``dpdk-validate-bpf --help``.
+
+The most important options for a minimal example are:
+
+* ``--prog-arg <type>``
+
+ Define program arguments, if different from ``struct rte_mbuf *``.
+ This argument may be repeated up to 5 times.
+
+* ``--xsym '<type> <name> | <type> <name>(<type>, ...)'``
+
+ Define an external symbol (variable or function) that the BPF program uses.
+ This allows the validator to recognize external calls and memory accesses.
+ Multiple external symbols can be defined by repeating this option.
+
+* ``--section <name>``
+
+ Specify the ELF section name in the BPF object file to load and validate.
+ The default section name is ``.text``, which is typically fine in cases
+ where the object file contains only one program.
+
+Interactive Debugging Mode
+--------------------------
+
+The interactive mode (enabled with the ``--debug`` flag) allows the user to
+debug the BPF validation process itself. This is particularly useful when
+BPF program fails verification, allowing you to trace the state changes per
+instruction and understand the validator's decisions.
+
+Design
+~~~~~~
+
+The interactive debugger provides a command-line interface similar to GDB.
+It allows setting breakpoints at specific instructions, or catchpoints on
+validator events (such as ``branch-enter`` or ``branch-prune``).
+Users can step through the validation process,
+inspect the state of registers, and query if certain conditional jumps may
+be taken based on the validator's knowledge of the program state.
+
+Usage Example
+~~~~~~~~~~~~~
+
+Launch the tool in debug mode with a compiled BPF object file:
+
+.. code-block:: console
+
+ $ dpdk-validate-bpf bpf_prog.o --prog-arg={'void *',uint64_t} --debug
+ => 0: b7 00 00 00 01 00 00 00 mov r0, #0x1
+ (validate) list 3
+ => 0: b7 00 00 00 01 00 00 00 mov r0, #0x1
+ 1: 25 02 01 00 0e 00 00 00 jgt r2, #0xe, L3
+ 2: b7 00 00 00 00 00 00 00 mov r0, #0x0
+ (validate) break 2
+ Breakpoint 0 at 2.
+ (validate) catch branch-enter
+ Catchpoint 1 on branch-enter.
+ (validate) continue
+ Catchpoint 1 on branch-enter.
+ Entered new branch at pc 1.
+ => 3: 95 00 00 00 00 00 00 00 exit
+ (validate) where
+ 1: 25 02 01 00 0e 00 00 00 jgt r2, #0xe, L3 ; taken
+ => 3: 95 00 00 00 00 00 00 00 exit
+ (validate) info r1
+ r1: %buffer<0> + 0
+ (validate) info r2
+ r2: 0xf..UINT64_MAX
+ (validate) may r2 <= 14
+ NO
+ (validate) continue
+ Catchpoint 1 on branch-enter.
+ Entered new branch at pc 1.
+ Breakpoint 0 at 2.
+ => 2: b7 00 00 00 00 00 00 00 mov r0, #0x0
+ (validate) continue
+ Validation succeeded.
+ (validate) quit
--
2.43.0
^ permalink raw reply related [flat|nested] 38+ messages in thread
* Re: [PATCH v2 7/7] app/validate-bpf: add BPF validation application
2026-09-14 14:54 ` [PATCH v2 7/7] app/validate-bpf: add BPF validation application Marat Khalili
@ 2026-09-14 18:41 ` Stephen Hemminger
2026-09-15 13:13 ` Stephen Hemminger
1 sibling, 0 replies; 38+ messages in thread
From: Stephen Hemminger @ 2026-09-14 18:41 UTC (permalink / raw)
To: Marat Khalili; +Cc: Thomas Monjalon, Konstantin Ananyev, dev
On Mon, 14 Sep 2026 15:54:17 +0100
Marat Khalili <marat.khalili@huawei.com> wrote:
> Add a new DPDK application, dpdk-validate-bpf, for pre-validating eBPF
> programs for compatibility with the lib/bpf execution context.
>
> The application allows evaluating eBPF programs against the DPDK verifier
> before loading them into a real application. It includes an interactive
> debugging mode to trace state changes per instruction and understand the
> validator's decisions.
>
> Signed-off-by: Marat Khalili <marat.khalili@huawei.com>
> ---
I ran review of all outstanding BPF patches together.
This one had this feedback.
Patch 7/7 (app/validate-bpf: add BPF validation application)
app/validate-bpf/parse_decl.c:
+ arg->value.size *= array_length;
array_length comes from take_number() on the --xsym text with no
upper bound, and the multiply is not checked for overflow, so a
large length silently wraps to a small size that is then reported
to the validator as the object size. Reproduced:
dpdk-validate-bpf --xsym='uint64_t[2305843009213693953] v' \
--section=.text prog.o
Validation succeeded.
8 * 2305843009213693953 wraps to 8, so the tool accepts the
declaration and describes an 8-byte object. Clamp array_length,
or check the product, and reject with the usual text error.
^ permalink raw reply [flat|nested] 38+ messages in thread
* Re: [PATCH v2 0/7] validate-bpf: add standalone validation debug tool
2026-09-14 14:54 ` [PATCH v2 0/7] validate-bpf: add standalone validation debug tool Marat Khalili
` (6 preceding siblings ...)
2026-09-14 14:54 ` [PATCH v2 7/7] app/validate-bpf: add BPF validation application Marat Khalili
@ 2026-09-14 18:42 ` Stephen Hemminger
2026-09-15 19:27 ` Marat Khalili
2026-09-15 19:26 ` [PATCH v3 " Marat Khalili
8 siblings, 1 reply; 38+ messages in thread
From: Stephen Hemminger @ 2026-09-14 18:42 UTC (permalink / raw)
To: Marat Khalili; +Cc: dev
On Mon, 14 Sep 2026 15:54:10 +0100
Marat Khalili <marat.khalili@huawei.com> wrote:
> This patchset introduces a new standalone tool for pre-validating eBPF
> programs and debugging validation problems. Its should allow developers
> to trace validator state changes per instruction and understand verifier
> decisions, simplifying the debugging of rejected programs before loading
> them into a real application context.
>
> v2:
> * for FreeBSD compatibility switched from Linux to DPDK network structs;
> * addressed small AI comments:
> * clarified patch 1 commit message;
> * renamed `rte_validate_bpf_logtype` to `validate_bpf_logtype`;
> * corrected comment to an internal-use-only function to reflect that
> next-after-last value of the program counter is no longer allowed;
>
> Marat Khalili (7):
> bpf/validate: fix finished status on restart
> bpf/validate: refactor internal step function
> bpf/validate: formalize call back requirements
> bpf/validate: add jump notification events
> bpf/validate: add get current event API
> app/test: add test for bpf validate debug events
> app/validate-bpf: add BPF validation application
>
> MAINTAINERS | 2 +
> app/meson.build | 1 +
> app/test/test_bpf_validate.c | 341 +++++++-
> app/validate-bpf/alloc_list.c | 53 ++
> app/validate-bpf/args.c | 263 ++++++
> app/validate-bpf/debug.c | 1099 ++++++++++++++++++++++++
> app/validate-bpf/debug_command.c | 383 +++++++++
> app/validate-bpf/debug_command.h | 65 ++
> app/validate-bpf/eal_init_args.c | 57 ++
> app/validate-bpf/internal.h | 151 ++++
> app/validate-bpf/main.c | 77 ++
> app/validate-bpf/meson.build | 13 +
> app/validate-bpf/parse_decl.c | 611 +++++++++++++
> doc/guides/rel_notes/release_26_11.rst | 5 +
> doc/guides/tools/index.rst | 1 +
> doc/guides/tools/validate_bpf.rst | 97 +++
> lib/bpf/bpf_validate.c | 49 +-
> lib/bpf/bpf_validate_debug.c | 92 +-
> lib/bpf/bpf_validate_debug.h | 12 +-
> lib/bpf/rte_bpf_validate_debug.h | 42 +-
> 20 files changed, 3348 insertions(+), 66 deletions(-)
> create mode 100644 app/validate-bpf/alloc_list.c
> create mode 100644 app/validate-bpf/args.c
> create mode 100644 app/validate-bpf/debug.c
> create mode 100644 app/validate-bpf/debug_command.c
> create mode 100644 app/validate-bpf/debug_command.h
> create mode 100644 app/validate-bpf/eal_init_args.c
> create mode 100644 app/validate-bpf/internal.h
> create mode 100644 app/validate-bpf/main.c
> create mode 100644 app/validate-bpf/meson.build
> create mode 100644 app/validate-bpf/parse_decl.c
> create mode 100644 doc/guides/tools/validate_bpf.rst
>
Looks good to me, but AI had some nits to pick.
Re: [PATCH v2 0/7] bpf/validate: debug events and validation app
Errors
------
Patch 7/7 (app/validate-bpf: add BPF validation application)
app/validate-bpf/parse_decl.c:
+ arg->value.size *= array_length;
array_length comes from take_number() on the --xsym text with no
upper bound, and the multiply is not checked for overflow, so a
large length silently wraps to a small size that is then reported
to the validator as the object size. Reproduced:
dpdk-validate-bpf --xsym='uint64_t[2305843009213693953] v' \
--section=.text prog.o
Validation succeeded.
8 * 2305843009213693953 wraps to 8, so the tool accepts the
declaration and describes an 8-byte object. Clamp array_length,
or check the product, and reject with the usual text error.
Warnings
--------
Patch 5/7 (bpf/validate: add get current event API)
lib/bpf/bpf_validate_debug.c:
+ return -EINVAL;
in rte_bpf_validate_debug_get_event(), whose return type is
enum rte_bpf_validate_debug_event. Every enumerator in that enum
is non-negative, so the compatible type is unsigned and -EINVAL
arrives at the caller as 4294967274. It equals no enumerator,
including RTE_BPF_VALIDATE_DEBUG_EVENT_END, so a caller cannot
test for it. Either return
RTE_BPF_VALIDATE_DEBUG_EVENT_END for the NULL case and say so in
the doc comment, or change the signature to return int with the
event in an out-parameter.
Patch 7/7
app/validate-bpf/parse_decl.c:
+ void * const val = calloc(1, RTE_MAX(1u, arg.value.size));
+ RTE_VERIFY(val != 0);
arg.value.size is user-supplied, so an allocation failure driven by
the command line aborts with a core dump instead of an error
message. Reproduced:
dpdk-validate-bpf --xsym='uint64_t[18446744073709551615] v' \
--section=.text prog.o
EAL: PANIC in fill_var_xsym():
line 427 assert "val != 0" failed
Return an error up through the parse path like the other --xsym
failures do.
app/validate-bpf/main.c:
+ RTE_VERIFY(rte_eal_init(eal_init_argc, eal_init_argv) ==
+ eal_init_argc - 1);
rte_eal_init() returns -1 for ordinary environmental failures such
as a missing runtime directory or insufficient permissions. For a
command-line tool that should be a message and a non-zero exit, not
rte_panic(). Same for
+ RTE_VERIFY(rte_eal_cleanup() == 0);
app/validate-bpf/eal_init_args.c:
+#define RTE_EAL_INIT_ARG_SIZE_MAX sizeof("--log-level=lib.eal:warning")
and RTE_EAL_INIT_ARGS / RTE_EAL_INIT_ARGC below it. The RTE_
prefix is reserved for the libraries; an application should not
define into it. Drop the prefix.
Info
----
Patch 5/7
debug->current_event is set only in debug_send_event(). A
breakpoint callback reached through debug_trigger_breakpoints()
therefore sees whatever event fired last. The doc comment says the
value is undefined when no event is being processed, so this is
consistent, but a callback shared between a breakpoint and a
catchpoint will read a stale event rather than an obviously invalid
one.
rte_bpf_validate_debug_get_event() is new public experimental API
and gets no release note. 7/7 adds a release note, but only for
the application.
Patch 7/7
In --debug mode the result is printed twice: debug.c prints
"Validation succeeded." to stdout from the validation-success
catchpoint, and main.c prints it again to stderr.
Array declarations are written as "uint64_t[4] v", not the C
"uint64_t v[4]". The C spelling is rejected with
at offset 10: expect '(' or text end
which does not point at the real problem. Worth either accepting
the C form or naming the expected form in the message. The
documented form is in --help, so this is a diagnostics nit.
"uint64_t[0] v" is accepted and yields value.size 0 while calloc
still allocates one byte. Probably worth rejecting a zero length.
Verification
------------
Applied on d55ccd4. Each of the seven commits builds on its own with
-Dwerror=true, so bisect is safe. bpf_validate_autotest passes 29/29
and the new bpf_validate_events_autotest passes. valgrind is clean on
the application's normal paths, with no definite leaks.
Checked and found nothing to report: 2/7 is equivalent at all call
sites, so "No functional changes" holds; 3/7's tightening of the pc
check to ">= nb_ins" is consistent with dropping the past-the-end call
from evaluate_finish(); 4/7's nb_edge > 1 test does select exactly the
conditional jumps.
Review-Result: ERROR
^ permalink raw reply [flat|nested] 38+ messages in thread
* Re: [PATCH v2 7/7] app/validate-bpf: add BPF validation application
2026-09-14 14:54 ` [PATCH v2 7/7] app/validate-bpf: add BPF validation application Marat Khalili
2026-09-14 18:41 ` Stephen Hemminger
@ 2026-09-15 13:13 ` Stephen Hemminger
1 sibling, 0 replies; 38+ messages in thread
From: Stephen Hemminger @ 2026-09-15 13:13 UTC (permalink / raw)
To: Marat Khalili; +Cc: Thomas Monjalon, Konstantin Ananyev, dev
On Mon, 14 Sep 2026 15:54:17 +0100
Marat Khalili <marat.khalili@huawei.com> wrote:
> Add a new DPDK application, dpdk-validate-bpf, for pre-validating eBPF
> programs for compatibility with the lib/bpf execution context.
>
> The application allows evaluating eBPF programs against the DPDK verifier
> before loading them into a real application. It includes an interactive
> debugging mode to trace state changes per instruction and understand the
> validator's decisions.
>
> Signed-off-by: Marat Khalili <marat.khalili@huawei.com>
> ---
Build fails on FreeBSD
*Build Failed #1:
OS: FreeBSD15-64
Target: x86_64-native-bsdapp-gcc
FAILED: app/dpdk-validate-bpf.p/validate-bpf_parse_decl.c.o
gcc -Iapp/dpdk-validate-bpf.p -Iapp -I../app -Ilib/bpf -I../lib/bpf -Ilib/eal/common -I../lib/eal/common -I. -I.. -Iconfig -I../config -Ilib/eal/include -I../lib/eal/include -Ilib/eal/freebsd/include -I../lib/eal/freebsd/include -Ilib/eal/x86/include -I../lib/eal/x86/include -Ilib/eal -I../lib/eal -Ilib/kvargs -I../lib/kvargs -Ilib/log -I../lib/log -Ilib/metrics -I../lib/metrics -Ilib/telemetry -I../lib/telemetry -Ilib/argparse -I../lib/argparse -Ilib/mbuf -I../lib/mbuf -Ilib/mempool -I../lib/mempool -Ilib/ring -I../lib/ring -Ilib/net -I../lib/net -Ilib/ethdev -I../lib/ethdev -Ilib/meter -I../lib/meter -Ilib/cmdline -I../lib/cmdline -fdiagnostics-color=always -D_FILE_OFFSET_BITS=64 -Wall -Winvalid-pch -Wextra -Werror -std=c11 -O3 -include rte_config.h -Wvla -Wcast-qual -Wdeprecated -Wformat -Wformat-nonliteral -Wformat-security -Wmissing-declarations -Wmissing-prototypes -Wnested-externs -Wold-style-definition -Wpointer-arith -Wshadow -Wsign-compare -Wstrict-prototypes -Wundef -Wwrite-strings -Wno-packed-not-aligned -Wno-missing-field-initializers -D_GNU_SOURCE -D__BSD_VISIBLE -march=native -mno-avx512f -mrtm -DALLOW_EXPERIMENTAL_API -MD -MQ app/dpdk-validate-bpf.p/validate-bpf_parse_decl.c.o -MF app/dpdk-validate-bpf.p/validate-bpf_parse_decl.c.o.d -o app/dpdk-validate-bpf.p/validate-bpf_parse_decl.c.o -c ../app/validate-bpf/parse_decl.c
../app/validate-bpf/parse_decl.c: In function 'take_arg':
../app/validate-bpf/parse_decl.c:392:33: error: 'array_length' may be used uninitialized [-Werror=maybe-uninitialized]
392 | arg->value.size *= array_length;
| ^~
../app/validate-bpf/parse_decl.c:381:24: note: 'array_length' was declared here
381 | size_t array_length;
| ^~~~~~~~~~~~
cc1: all warnings being treated as errors
^ permalink raw reply [flat|nested] 38+ messages in thread
* [PATCH v3 0/7] validate-bpf: add standalone validation debug tool
2026-09-14 14:54 ` [PATCH v2 0/7] validate-bpf: add standalone validation debug tool Marat Khalili
` (7 preceding siblings ...)
2026-09-14 18:42 ` [PATCH v2 0/7] validate-bpf: add standalone validation debug tool Stephen Hemminger
@ 2026-09-15 19:26 ` Marat Khalili
2026-09-15 19:26 ` [PATCH v3 1/7] bpf/validate: fix finished status on restart Marat Khalili
` (7 more replies)
8 siblings, 8 replies; 38+ messages in thread
From: Marat Khalili @ 2026-09-15 19:26 UTC (permalink / raw)
Cc: dev
This patchset introduces a new standalone tool for pre-validating eBPF
programs and debugging validation problems. Its should allow developers
to trace validator state changes per instruction and understand verifier
decisions, simplifying the debugging of rejected programs before loading
them into a real application context.
v3:
* addressed another FreeBSD issue (spurious uninitialized warning);
* addressed more AI comments:
* fail if library fails to set step catchpoint on start command;
* fail nicer if external variables are too big;
* fail nicer if EAL initialization or cleanup fails;
* explain why 9 steps are expeced in the test;
* explain that non-zero buf_size marks mbuf;
* avoid symbols starting with `RTE_` in the app;
v2:
* for FreeBSD compatibility switched from Linux to DPDK network structs;
* addressed small AI comments:
* clarified patch 1 commit message;
* renamed `rte_validate_bpf_logtype` to `validate_bpf_logtype`;
* corrected comment to an internal-use-only function to reflect that
next-after-last value of the program counter is no longer allowed;
Marat Khalili (7):
bpf/validate: fix finished status on restart
bpf/validate: refactor internal step function
bpf/validate: formalize call back requirements
bpf/validate: add jump notification events
bpf/validate: add get current event API
app/test: add test for bpf validate debug events
app/validate-bpf: add BPF validation application
MAINTAINERS | 2 +
app/meson.build | 1 +
app/test/test_bpf_validate.c | 344 +++++++-
app/validate-bpf/alloc_list.c | 53 ++
app/validate-bpf/args.c | 263 ++++++
app/validate-bpf/debug.c | 1103 ++++++++++++++++++++++++
app/validate-bpf/debug_command.c | 383 ++++++++
app/validate-bpf/debug_command.h | 65 ++
app/validate-bpf/eal_init_args.c | 57 ++
app/validate-bpf/internal.h | 151 ++++
app/validate-bpf/main.c | 90 ++
app/validate-bpf/meson.build | 13 +
app/validate-bpf/parse_decl.c | 623 +++++++++++++
doc/guides/rel_notes/release_26_11.rst | 5 +
doc/guides/tools/index.rst | 1 +
doc/guides/tools/validate_bpf.rst | 97 +++
lib/bpf/bpf_validate.c | 49 +-
lib/bpf/bpf_validate_debug.c | 92 +-
lib/bpf/bpf_validate_debug.h | 12 +-
lib/bpf/rte_bpf_validate_debug.h | 42 +-
20 files changed, 3380 insertions(+), 66 deletions(-)
create mode 100644 app/validate-bpf/alloc_list.c
create mode 100644 app/validate-bpf/args.c
create mode 100644 app/validate-bpf/debug.c
create mode 100644 app/validate-bpf/debug_command.c
create mode 100644 app/validate-bpf/debug_command.h
create mode 100644 app/validate-bpf/eal_init_args.c
create mode 100644 app/validate-bpf/internal.h
create mode 100644 app/validate-bpf/main.c
create mode 100644 app/validate-bpf/meson.build
create mode 100644 app/validate-bpf/parse_decl.c
create mode 100644 doc/guides/tools/validate_bpf.rst
--
2.43.0
^ permalink raw reply [flat|nested] 38+ messages in thread
* [PATCH v3 1/7] bpf/validate: fix finished status on restart
2026-09-15 19:26 ` [PATCH v3 " Marat Khalili
@ 2026-09-15 19:26 ` Marat Khalili
2026-09-15 19:26 ` [PATCH v3 2/7] bpf/validate: refactor internal step function Marat Khalili
` (6 subsequent siblings)
7 siblings, 0 replies; 38+ messages in thread
From: Marat Khalili @ 2026-09-15 19:26 UTC (permalink / raw)
To: Konstantin Ananyev; +Cc: dev
If debug session is reused, requesting evaluation result while
evaluation is still in progress would successfully return result from
the previous evaluation instead of expected -EAGAIN error.
Make evaluation start clear the finished status.
Fixes: 6b0da0602cc7 ("bpf/validate: introduce debugging interface")
Signed-off-by: Marat Khalili <marat.khalili@huawei.com>
---
lib/bpf/bpf_validate_debug.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/lib/bpf/bpf_validate_debug.c b/lib/bpf/bpf_validate_debug.c
index 5d18804a74bc..9a039766322c 100644
--- a/lib/bpf/bpf_validate_debug.c
+++ b/lib/bpf/bpf_validate_debug.c
@@ -568,6 +568,7 @@ __rte_bpf_validate_debug_evaluate_start(struct rte_bpf_validate_debug *debug,
debug->last_point = NULL;
debug->pc = 0;
debug->evaluate_started = true;
+ debug->evaluate_finished = false;
const int rc = debug_send_event(debug,
RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_START);
--
2.43.0
^ permalink raw reply related [flat|nested] 38+ messages in thread
* [PATCH v3 2/7] bpf/validate: refactor internal step function
2026-09-15 19:26 ` [PATCH v3 " Marat Khalili
2026-09-15 19:26 ` [PATCH v3 1/7] bpf/validate: fix finished status on restart Marat Khalili
@ 2026-09-15 19:26 ` Marat Khalili
2026-09-15 19:26 ` [PATCH v3 3/7] bpf/validate: formalize call back requirements Marat Khalili
` (5 subsequent siblings)
7 siblings, 0 replies; 38+ messages in thread
From: Marat Khalili @ 2026-09-15 19:26 UTC (permalink / raw)
To: Konstantin Ananyev; +Cc: dev
Rename `__rte_bpf_validate_debug_evaluate_step` to
`__rte_bpf_validate_debug_evaluate_update` to better reflect planned
usage, and make it accept a bitmask of events instead of a single event.
Step event is still implicitly triggered for now.
No functional changes.
Signed-off-by: Marat Khalili <marat.khalili@huawei.com>
---
lib/bpf/bpf_validate.c | 17 ++++++++---------
lib/bpf/bpf_validate_debug.c | 20 +++++++++++---------
lib/bpf/bpf_validate_debug.h | 11 ++++++-----
3 files changed, 25 insertions(+), 23 deletions(-)
diff --git a/lib/bpf/bpf_validate.c b/lib/bpf/bpf_validate.c
index f9960088a285..b317abb81172 100644
--- a/lib/bpf/bpf_validate.c
+++ b/lib/bpf/bpf_validate.c
@@ -2989,10 +2989,9 @@ evaluate(struct bpf_verifier *bvf)
break;
}
- rc = __rte_bpf_validate_debug_evaluate_step(debug, idx,
+ rc = __rte_bpf_validate_debug_evaluate_update(debug, idx,
prev_nb_edge > 1 ?
- RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER :
- RTE_BPF_VALIDATE_DEBUG_EVENT_STEP);
+ RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER) : 0);
if (rc < 0)
break;
@@ -3024,9 +3023,9 @@ evaluate(struct bpf_verifier *bvf)
}
if (bvf->evst->unreachable) {
- rc = __rte_bpf_validate_debug_evaluate_step(
+ rc = __rte_bpf_validate_debug_evaluate_update(
debug, get_node_idx(bvf, next),
- RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE);
+ RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE));
if (rc < 0)
break;
@@ -3038,9 +3037,9 @@ evaluate(struct bpf_verifier *bvf)
*/
} else if (node->nb_edge > 1 &&
prune_eval_state(bvf, node, next) == 0) {
- rc = __rte_bpf_validate_debug_evaluate_step(
+ rc = __rte_bpf_validate_debug_evaluate_update(
debug, get_node_idx(bvf, next),
- RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE);
+ RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE));
if (rc < 0)
break;
@@ -3059,9 +3058,9 @@ evaluate(struct bpf_verifier *bvf)
*/
if (prev_nb_edge != 0) {
- rc = __rte_bpf_validate_debug_evaluate_step(
+ rc = __rte_bpf_validate_debug_evaluate_update(
debug, get_node_idx(bvf, node) + 1,
- RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN);
+ RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN));
if (rc < 0)
break;
}
diff --git a/lib/bpf/bpf_validate_debug.c b/lib/bpf/bpf_validate_debug.c
index 9a039766322c..f76303a7fd2a 100644
--- a/lib/bpf/bpf_validate_debug.c
+++ b/lib/bpf/bpf_validate_debug.c
@@ -582,8 +582,8 @@ __rte_bpf_validate_debug_evaluate_start(struct rte_bpf_validate_debug *debug,
}
int
-__rte_bpf_validate_debug_evaluate_step(struct rte_bpf_validate_debug *debug,
- uint32_t pc, debug_event_t event)
+__rte_bpf_validate_debug_evaluate_update(struct rte_bpf_validate_debug *debug,
+ uint32_t pc, uint64_t events)
{
int rc;
@@ -595,8 +595,7 @@ __rte_bpf_validate_debug_evaluate_step(struct rte_bpf_validate_debug *debug,
return -ECHILD;
}
- if (pc > debug->bpf_prm->raw.nb_ins || event < 0 ||
- event >= RTE_BPF_VALIDATE_DEBUG_EVENT_END)
+ if (pc > debug->bpf_prm->raw.nb_ins)
return -EINVAL;
debug->pc = pc;
@@ -606,11 +605,13 @@ __rte_bpf_validate_debug_evaluate_step(struct rte_bpf_validate_debug *debug,
rc = debug_send_event(debug,
RTE_BPF_VALIDATE_DEBUG_EVENT_INVALID_STATE);
- if (event != RTE_BPF_VALIDATE_DEBUG_EVENT_STEP)
- rc = rc < 0 ? rc : debug_send_event(debug, event);
+ for (enum rte_bpf_validate_debug_event event = 0;
+ event != RTE_BPF_VALIDATE_DEBUG_EVENT_END; ++event)
+ if (events & RTE_BIT64(event))
+ rc = rc < 0 ? rc : debug_send_event(debug, event);
- if (event == RTE_BPF_VALIDATE_DEBUG_EVENT_STEP ||
- event == RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER)
+ if (events == 0 || events == RTE_BIT64(
+ RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER))
/* Stepping into a real instruction to execute. */
rc = rc < 0 ? rc : debug_points_call_back(debug,
&debug->breakpoint_lists[pc]);
@@ -651,7 +652,8 @@ __rte_bpf_validate_debug_evaluate_finish(struct rte_bpf_validate_debug *debug,
event = RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS;
}
- rc = __rte_bpf_validate_debug_evaluate_step(debug, pc, event);
+ rc = __rte_bpf_validate_debug_evaluate_update(debug, pc,
+ RTE_BIT64(event));
}
debug_evaluate_close(debug);
diff --git a/lib/bpf/bpf_validate_debug.h b/lib/bpf/bpf_validate_debug.h
index a91f3e9c48b2..2b0c3bb646a7 100644
--- a/lib/bpf/bpf_validate_debug.h
+++ b/lib/bpf/bpf_validate_debug.h
@@ -46,7 +46,7 @@ __rte_bpf_validate_debug_evaluate_start(struct rte_bpf_validate_debug *debug,
const struct bpf_verifier *verifier, const struct rte_bpf_prm_ex *bpf_prm);
/*
- * Signal each instruction, branch end, or evaluation end.
+ * Signal evaluation events.
*
* Immediately return 0 if debug is NULL.
*
@@ -54,16 +54,17 @@ __rte_bpf_validate_debug_evaluate_start(struct rte_bpf_validate_debug *debug,
* Validate debug instance configured by user, can be NULL.
* @param pc
* Current value of the program counter, or next after last instruction.
- * @param event
- * Specific evaluation event if any, or RTE_BPF_VALIDATE_DEBUG_EVENT_STEP.
+ * @param events
+ * Bitmask of events.
+ * Step bit is always implied and should not be set.
* @return
* Non-negative value: evaluation should continue;
* -ECANCELED: evaluation should fail without calling this API again;
* Other negative value: evaluation should fail signalling failure;
*/
int
-__rte_bpf_validate_debug_evaluate_step(struct rte_bpf_validate_debug *debug,
- uint32_t pc, debug_event_t event);
+__rte_bpf_validate_debug_evaluate_update(struct rte_bpf_validate_debug *debug,
+ uint32_t pc, uint64_t events);
/*
* Signal end of evaluation process.
--
2.43.0
^ permalink raw reply related [flat|nested] 38+ messages in thread
* [PATCH v3 3/7] bpf/validate: formalize call back requirements
2026-09-15 19:26 ` [PATCH v3 " Marat Khalili
2026-09-15 19:26 ` [PATCH v3 1/7] bpf/validate: fix finished status on restart Marat Khalili
2026-09-15 19:26 ` [PATCH v3 2/7] bpf/validate: refactor internal step function Marat Khalili
@ 2026-09-15 19:26 ` Marat Khalili
2026-09-15 19:26 ` [PATCH v3 4/7] bpf/validate: add jump notification events Marat Khalili
` (4 subsequent siblings)
7 siblings, 0 replies; 38+ messages in thread
From: Marat Khalili @ 2026-09-15 19:26 UTC (permalink / raw)
To: Konstantin Ananyev; +Cc: dev
Formalize requirements to evaluation events in documentation and adjust
their handling in the actual code to match.
Particularly, specify more precisely:
* conditions for each catchpoint callback;
* relative order of callbacks within the step;
* expected value of the program counter;
Behaviour changes:
* one branch-return event is now generated for each branch-enter event
(previously multiple branch-return events could be coalesced);
* branch-return event now points to the corresponding conditional jump
(previously past the end of the last branch);
* step event is now only generated before an actual instruction
evaluation (previously also for branching and finishing events);
* program counter is no longer allowed to point past the program end;
These changes should simplify branch tracking of the validator by tests
and interactive applications. E.g. branch-enter and branch-return events
can now be used to push and pop branching points to/from some stack.
Signed-off-by: Marat Khalili <marat.khalili@huawei.com>
---
lib/bpf/bpf_validate.c | 28 +++++++++----
lib/bpf/bpf_validate_debug.c | 72 +++++++++++++++++++-------------
lib/bpf/bpf_validate_debug.h | 3 +-
lib/bpf/rte_bpf_validate_debug.h | 24 +++++++----
4 files changed, 78 insertions(+), 49 deletions(-)
diff --git a/lib/bpf/bpf_validate.c b/lib/bpf/bpf_validate.c
index b317abb81172..f4258c2fb54f 100644
--- a/lib/bpf/bpf_validate.c
+++ b/lib/bpf/bpf_validate.c
@@ -2892,6 +2892,23 @@ prune_eval_state(struct bpf_verifier *bvf, const struct inst_node *node,
return rc;
}
+static bool
+is_branch_start(const struct inst_node *node)
+{
+ return node->prev_node != NULL && node->prev_node->nb_edge > 1;
+}
+
+static uint64_t
+step_events(const struct inst_node *node)
+{
+ uint64_t events = RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_STEP);
+
+ if (is_branch_start(node))
+ events |= RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER);
+
+ return events;
+}
+
/* Do second pass through CFG and try to evaluate instructions
* via each possible path. The verifier will try all paths, tracking types of
* registers used as input to instructions, and updating resulting type via
@@ -2918,7 +2935,6 @@ evaluate(struct bpf_verifier *bvf)
const char *err;
const struct ebpf_insn *ins;
struct inst_node *next, *node;
- int prev_nb_edge; /* branching number of the previous instruction */
int rc, debug_rc;
struct rte_bpf_validate_debug *const debug = bvf->prm->debug;
@@ -2954,7 +2970,6 @@ evaluate(struct bpf_verifier *bvf)
ins = bvf->prm->raw.ins;
node = bvf->in;
next = node;
- prev_nb_edge = 1;
memset(&stats, 0, sizeof(stats));
@@ -2990,8 +3005,7 @@ evaluate(struct bpf_verifier *bvf)
}
rc = __rte_bpf_validate_debug_evaluate_update(debug, idx,
- prev_nb_edge > 1 ?
- RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER) : 0);
+ step_events(node));
if (rc < 0)
break;
@@ -3047,7 +3061,6 @@ evaluate(struct bpf_verifier *bvf)
stats.nb_prune++;
} else {
next->prev_node = node;
- prev_nb_edge = node->nb_edge;
node = next;
}
} else {
@@ -3057,9 +3070,9 @@ evaluate(struct bpf_verifier *bvf)
* and proceed with parent.
*/
- if (prev_nb_edge != 0) {
+ if (is_branch_start(node)) {
rc = __rte_bpf_validate_debug_evaluate_update(
- debug, get_node_idx(bvf, node) + 1,
+ debug, get_node_idx(bvf, node->prev_node),
RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN));
if (rc < 0)
break;
@@ -3067,7 +3080,6 @@ evaluate(struct bpf_verifier *bvf)
node->cur_edge = 0;
save_safe_eval_state(bvf, node);
- prev_nb_edge = 0;
node = node->prev_node;
/* first node will not have prev, signalling finish */
diff --git a/lib/bpf/bpf_validate_debug.c b/lib/bpf/bpf_validate_debug.c
index f76303a7fd2a..db6d13f9cb22 100644
--- a/lib/bpf/bpf_validate_debug.c
+++ b/lib/bpf/bpf_validate_debug.c
@@ -225,6 +225,13 @@ debug_points_call_back(struct rte_bpf_validate_debug *debug,
return rc;
}
+/* Call back all breakpoints for the specified program counter. */
+static int
+debug_trigger_breakpoints(struct rte_bpf_validate_debug *debug, uint32_t pc)
+{
+ return debug_points_call_back(debug, &debug->breakpoint_lists[pc]);
+}
+
/* Call back all catchpoints for the specified event. */
static int
debug_send_event(struct rte_bpf_validate_debug *debug, debug_event_t event)
@@ -585,6 +592,21 @@ int
__rte_bpf_validate_debug_evaluate_update(struct rte_bpf_validate_debug *debug,
uint32_t pc, uint64_t events)
{
+ /* Required order of sent events according to the documentation. */
+ static const enum rte_bpf_validate_debug_event ordered_events[] = {
+ RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_START,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_INVALID_STATE,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_STEP,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_FAILURE,
+ };
+ RTE_BUILD_BUG_ON(
+ RTE_DIM(ordered_events) != RTE_BPF_VALIDATE_DEBUG_EVENT_END);
+
int rc;
if (debug == NULL)
@@ -595,29 +617,31 @@ __rte_bpf_validate_debug_evaluate_update(struct rte_bpf_validate_debug *debug,
return -ECHILD;
}
- if (pc > debug->bpf_prm->raw.nb_ins)
+ if (pc >= debug->bpf_prm->raw.nb_ins)
return -EINVAL;
debug->pc = pc;
rc = __rte_bpf_validate_state_is_valid(debug->verifier);
if (rc == 0)
- rc = debug_send_event(debug,
- RTE_BPF_VALIDATE_DEBUG_EVENT_INVALID_STATE);
+ events |= RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_INVALID_STATE);
+
+ for (uint32_t index = 0; index < RTE_DIM(ordered_events); index++) {
+ const enum rte_bpf_validate_debug_event event =
+ ordered_events[index];
+ if ((events & RTE_BIT64(event)) == 0)
+ continue;
- for (enum rte_bpf_validate_debug_event event = 0;
- event != RTE_BPF_VALIDATE_DEBUG_EVENT_END; ++event)
- if (events & RTE_BIT64(event))
- rc = rc < 0 ? rc : debug_send_event(debug, event);
+ if (event == RTE_BPF_VALIDATE_DEBUG_EVENT_STEP)
+ rc = rc < 0 ? rc : debug_trigger_breakpoints(debug, pc);
- if (events == 0 || events == RTE_BIT64(
- RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER))
- /* Stepping into a real instruction to execute. */
- rc = rc < 0 ? rc : debug_points_call_back(debug,
- &debug->breakpoint_lists[pc]);
+ rc = rc < 0 ? rc : debug_send_event(debug, event);
+ events -= RTE_BIT64(event);
+ }
- rc = rc < 0 ? rc : debug_send_event(debug,
- RTE_BPF_VALIDATE_DEBUG_EVENT_STEP);
+ if (events != 0)
+ /* Received unsupported events. */
+ rc = rc < 0 ? rc : -EINVAL;
return rc;
}
@@ -627,8 +651,6 @@ __rte_bpf_validate_debug_evaluate_finish(struct rte_bpf_validate_debug *debug,
int result)
{
int rc = 0;
- uint32_t pc;
- debug_event_t event;
if (debug == NULL)
return 0;
@@ -641,20 +663,10 @@ __rte_bpf_validate_debug_evaluate_finish(struct rte_bpf_validate_debug *debug,
debug->evaluate_finished = true;
debug->evaluate_result = result;
- if (result != -ECANCELED) {
- if (result < 0) {
- /* Last known pc is the place we failed. */
- pc = debug->pc;
- event = RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_FAILURE;
- } else {
- /* Show program end, not particular instruction. */
- pc = debug->bpf_prm->raw.nb_ins;
- event = RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS;
- }
-
- rc = __rte_bpf_validate_debug_evaluate_update(debug, pc,
- RTE_BIT64(event));
- }
+ if (result != -ECANCELED)
+ rc = debug_send_event(debug, result < 0 ?
+ RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_FAILURE :
+ RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS);
debug_evaluate_close(debug);
diff --git a/lib/bpf/bpf_validate_debug.h b/lib/bpf/bpf_validate_debug.h
index 2b0c3bb646a7..4603841db2ed 100644
--- a/lib/bpf/bpf_validate_debug.h
+++ b/lib/bpf/bpf_validate_debug.h
@@ -53,10 +53,9 @@ __rte_bpf_validate_debug_evaluate_start(struct rte_bpf_validate_debug *debug,
* @param debug
* Validate debug instance configured by user, can be NULL.
* @param pc
- * Current value of the program counter, or next after last instruction.
+ * Current value of the program counter.
* @param events
* Bitmask of events.
- * Step bit is always implied and should not be set.
* @return
* Non-negative value: evaluation should continue;
* -ECANCELED: evaluation should fail without calling this API again;
diff --git a/lib/bpf/rte_bpf_validate_debug.h b/lib/bpf/rte_bpf_validate_debug.h
index f30fa926f10a..b3fedd5428bc 100644
--- a/lib/bpf/rte_bpf_validate_debug.h
+++ b/lib/bpf/rte_bpf_validate_debug.h
@@ -31,25 +31,31 @@ extern "C" {
* Supported validate events.
*
* Valid events begin from 0 and end before `RTE_BPF_VALIDATE_DEBUG_EVENT_END`.
+ *
+ * At any given evaluation step, callbacks are fired in the following order:
+ * - Validation start event;
+ * - Branching and invalid-state events (can be interleaved);
+ * - Instruction breakpoints (before evaluating instruction);
+ * - Step (before evaluating instruction) or validation result (if done) event;
*/
enum rte_bpf_validate_debug_event {
- /* Just before every instruction, at branch or validation end. */
+ /* Just before every evaluated instruction. */
RTE_BPF_VALIDATE_DEBUG_EVENT_STEP,
/* Validator has failed its internal self-checks. */
RTE_BPF_VALIDATE_DEBUG_EVENT_INVALID_STATE,
- /* Start of validation. */
+ /* Start of validation (pc points to first instruction). */
RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_START,
- /* Successful finish of validation. */
+ /* Successful finish (pc undefined). */
RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS,
- /* Finish of validation with error. */
+ /* Finish of validation with error (pc points at error). */
RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_FAILURE,
- /* Beginning of a branch just after the jump. */
+ /* Beginning of a branch evaluation (pc points to branch start). */
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER,
- /* Pruning branch as verified earlier. */
+ /* Pruning branch as verified earlier (pc points to branch start). */
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE,
- /* End of branch verification, after the last verified instruction. */
+ /* End of branch verification (pc points to jump instruction). */
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN,
- /* Pruning branch as dynamically unreachable. */
+ /* Pruning branch as dynamically unreachable (pc points to branch start). */
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE,
/* Number of valid event values. */
RTE_BPF_VALIDATE_DEBUG_EVENT_END,
@@ -208,7 +214,7 @@ rte_bpf_validate_debug_get_last_point(const struct rte_bpf_validate_debug *debug
* @param debug
* Debug instance.
* @return
- * Current program counter being validated, or one after last.
+ * Current program counter being validated.
* UINT32_MAX if no program is being validated.
*/
__rte_experimental
--
2.43.0
^ permalink raw reply related [flat|nested] 38+ messages in thread
* [PATCH v3 4/7] bpf/validate: add jump notification events
2026-09-15 19:26 ` [PATCH v3 " Marat Khalili
` (2 preceding siblings ...)
2026-09-15 19:26 ` [PATCH v3 3/7] bpf/validate: formalize call back requirements Marat Khalili
@ 2026-09-15 19:26 ` Marat Khalili
2026-09-15 19:26 ` [PATCH v3 5/7] bpf/validate: add get current event API Marat Khalili
` (3 subsequent siblings)
7 siblings, 0 replies; 38+ messages in thread
From: Marat Khalili @ 2026-09-15 19:26 UTC (permalink / raw)
To: Konstantin Ananyev; +Cc: dev
Add events notifying validate debug client of instructions that may
change program counter. This is particularly useful to an interactive
debugger to track conditional and unconditional jumps without trying to
parse the program and interpret instructions independently.
Signed-off-by: Marat Khalili <marat.khalili@huawei.com>
---
lib/bpf/bpf_validate.c | 10 ++++++++--
lib/bpf/bpf_validate_debug.c | 2 ++
lib/bpf/rte_bpf_validate_debug.h | 5 +++++
3 files changed, 15 insertions(+), 2 deletions(-)
diff --git a/lib/bpf/bpf_validate.c b/lib/bpf/bpf_validate.c
index f4258c2fb54f..d503db836105 100644
--- a/lib/bpf/bpf_validate.c
+++ b/lib/bpf/bpf_validate.c
@@ -2899,13 +2899,19 @@ is_branch_start(const struct inst_node *node)
}
static uint64_t
-step_events(const struct inst_node *node)
+step_events(const struct inst_node *node, uint8_t op)
{
uint64_t events = RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_STEP);
if (is_branch_start(node))
events |= RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER);
+ if (op == (BPF_JMP | BPF_JA))
+ events |= RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS);
+
+ if (node->nb_edge > 1)
+ events |= RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL);
+
return events;
}
@@ -3005,7 +3011,7 @@ evaluate(struct bpf_verifier *bvf)
}
rc = __rte_bpf_validate_debug_evaluate_update(debug, idx,
- step_events(node));
+ step_events(node, op));
if (rc < 0)
break;
diff --git a/lib/bpf/bpf_validate_debug.c b/lib/bpf/bpf_validate_debug.c
index db6d13f9cb22..098a88026d6d 100644
--- a/lib/bpf/bpf_validate_debug.c
+++ b/lib/bpf/bpf_validate_debug.c
@@ -600,6 +600,8 @@ __rte_bpf_validate_debug_evaluate_update(struct rte_bpf_validate_debug *debug,
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE,
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN,
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL,
RTE_BPF_VALIDATE_DEBUG_EVENT_STEP,
RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS,
RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_FAILURE,
diff --git a/lib/bpf/rte_bpf_validate_debug.h b/lib/bpf/rte_bpf_validate_debug.h
index b3fedd5428bc..f1ee8769d06a 100644
--- a/lib/bpf/rte_bpf_validate_debug.h
+++ b/lib/bpf/rte_bpf_validate_debug.h
@@ -35,6 +35,7 @@ extern "C" {
* At any given evaluation step, callbacks are fired in the following order:
* - Validation start event;
* - Branching and invalid-state events (can be interleaved);
+ * - Jump-instruction events (before evaluating jump instruction);
* - Instruction breakpoints (before evaluating instruction);
* - Step (before evaluating instruction) or validation result (if done) event;
*/
@@ -57,6 +58,10 @@ enum rte_bpf_validate_debug_event {
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN,
/* Pruning branch as dynamically unreachable (pc points to branch start). */
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE,
+ /* Program counter points to a jump-always instruction. */
+ RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS,
+ /* Program counter points to a conditional jump instruction. */
+ RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL,
/* Number of valid event values. */
RTE_BPF_VALIDATE_DEBUG_EVENT_END,
};
--
2.43.0
^ permalink raw reply related [flat|nested] 38+ messages in thread
* [PATCH v3 5/7] bpf/validate: add get current event API
2026-09-15 19:26 ` [PATCH v3 " Marat Khalili
` (3 preceding siblings ...)
2026-09-15 19:26 ` [PATCH v3 4/7] bpf/validate: add jump notification events Marat Khalili
@ 2026-09-15 19:26 ` Marat Khalili
2026-09-15 19:26 ` [PATCH v3 6/7] app/test: add test for bpf validate debug events Marat Khalili
` (2 subsequent siblings)
7 siblings, 0 replies; 38+ messages in thread
From: Marat Khalili @ 2026-09-15 19:26 UTC (permalink / raw)
To: Konstantin Ananyev; +Cc: dev
Add API for retrieving the event currently processed by a callback.
Useful when single callback is registered for multiple events to put
less pressure on the context argument.
Signed-off-by: Marat Khalili <marat.khalili@huawei.com>
---
lib/bpf/bpf_validate_debug.c | 13 +++++++++++++
lib/bpf/rte_bpf_validate_debug.h | 13 +++++++++++++
2 files changed, 26 insertions(+)
diff --git a/lib/bpf/bpf_validate_debug.c b/lib/bpf/bpf_validate_debug.c
index 098a88026d6d..c8e9e5dd9227 100644
--- a/lib/bpf/bpf_validate_debug.c
+++ b/lib/bpf/bpf_validate_debug.c
@@ -43,6 +43,7 @@ struct rte_bpf_validate_debug {
const struct bpf_verifier *verifier;
const struct rte_bpf_prm_ex *bpf_prm;
struct point_list *breakpoint_lists;
+ enum rte_bpf_validate_debug_event current_event;
struct rte_bpf_validate_debug_point *last_point;
uint32_t pc;
/* Evaluate stage (only tracking `evaluate` part at the moment). */
@@ -236,6 +237,7 @@ debug_trigger_breakpoints(struct rte_bpf_validate_debug *debug, uint32_t pc)
static int
debug_send_event(struct rte_bpf_validate_debug *debug, debug_event_t event)
{
+ debug->current_event = event;
return debug_points_call_back(debug, &debug->catchpoint_lists[event]);
}
@@ -342,6 +344,17 @@ rte_bpf_validate_debug_get_ins(const struct rte_bpf_validate_debug *debug,
return 0;
}
+RTE_EXPORT_EXPERIMENTAL_SYMBOL(rte_bpf_validate_debug_get_event, 26.11)
+enum rte_bpf_validate_debug_event
+rte_bpf_validate_debug_get_event(const struct rte_bpf_validate_debug *debug)
+{
+ if (debug == NULL)
+ /* Just to be fool-proof, not really required by API. */
+ return -EINVAL;
+
+ return debug->current_event;
+}
+
RTE_EXPORT_EXPERIMENTAL_SYMBOL(rte_bpf_validate_debug_get_last_point, 26.07)
struct rte_bpf_validate_debug_point *
rte_bpf_validate_debug_get_last_point(const struct rte_bpf_validate_debug *debug)
diff --git a/lib/bpf/rte_bpf_validate_debug.h b/lib/bpf/rte_bpf_validate_debug.h
index f1ee8769d06a..6d4bf53589e0 100644
--- a/lib/bpf/rte_bpf_validate_debug.h
+++ b/lib/bpf/rte_bpf_validate_debug.h
@@ -194,6 +194,19 @@ int
rte_bpf_validate_debug_get_ins(const struct rte_bpf_validate_debug *debug,
const struct ebpf_insn **ins, uint32_t *nb_ins);
+/**
+ * Get event currently processed by a catchpoint callback.
+ *
+ * @param debug
+ * Debug instance.
+ * @return
+ * Event currently processed by a catchpoint callback.
+ * Undefined if no event is currently being processed.
+ */
+__rte_experimental
+enum rte_bpf_validate_debug_event
+rte_bpf_validate_debug_get_event(const struct rte_bpf_validate_debug *debug);
+
/**
* Get last triggered breakpoint or catchpoint.
*
--
2.43.0
^ permalink raw reply related [flat|nested] 38+ messages in thread
* [PATCH v3 6/7] app/test: add test for bpf validate debug events
2026-09-15 19:26 ` [PATCH v3 " Marat Khalili
` (4 preceding siblings ...)
2026-09-15 19:26 ` [PATCH v3 5/7] bpf/validate: add get current event API Marat Khalili
@ 2026-09-15 19:26 ` Marat Khalili
2026-09-15 19:26 ` [PATCH v3 7/7] app/validate-bpf: add BPF validation application Marat Khalili
2026-09-17 14:35 ` [PATCH v4 0/7] validate-bpf: add standalone validation debug tool Marat Khalili
7 siblings, 0 replies; 38+ messages in thread
From: Marat Khalili @ 2026-09-15 19:26 UTC (permalink / raw)
To: Konstantin Ananyev; +Cc: dev
Add a test that verifies that eBPF validator generates callbacks with
the number, order and program counter specified in the documentation.
Signed-off-by: Marat Khalili <marat.khalili@huawei.com>
---
app/test/test_bpf_validate.c | 344 ++++++++++++++++++++++++++++++++++-
1 file changed, 340 insertions(+), 4 deletions(-)
diff --git a/app/test/test_bpf_validate.c b/app/test/test_bpf_validate.c
index 066f1fa156ae..8b48d8a504dd 100644
--- a/app/test/test_bpf_validate.c
+++ b/app/test/test_bpf_validate.c
@@ -9,10 +9,6 @@
#include <rte_bpf_validate_debug.h>
#include <rte_errno.h>
-/*
- * Tests of BPF validation.
- */
-
extern int test_bpf_validate_logtype;
#define RTE_LOGTYPE_TEST_BPF_VALIDATE test_bpf_validate_logtype
#define TEST_LOG_LINE(level, ...) \
@@ -26,6 +22,346 @@ RTE_LOG_REGISTER(test_bpf_validate_logtype, test.bpf_validate, NOTICE);
/* Special value indicating that register variable is not being used. */
#define NO_REGISTER UINT8_MAX
+static const char *const event_names[RTE_BPF_VALIDATE_DEBUG_EVENT_END] = {
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_STEP] = "step",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_INVALID_STATE] = "invalid-state",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_START] = "validation-start",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS] = "validation-success",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_FAILURE] = "validation-failure",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER] = "branch-enter",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE] = "branch-prune",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN] = "branch-return",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE] = "branch-unreachable",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS] = "jump-always",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL] = "jump-conditional",
+};
+
+/*
+ * Self-test of BPF validation debug events.
+ */
+
+#define EVENT_BIT(name) RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_ ## name)
+#define EVENT_BITS_ALL RTE_LEN2MASK(RTE_BPF_VALIDATE_DEBUG_EVENT_END, uint64_t)
+
+struct test_events_context {
+ /* Tested program. */
+ const struct ebpf_insn *ins;
+ uint32_t nb_ins;
+ /* Number of times each of the events was encountered. */
+ int event_counts[RTE_BPF_VALIDATE_DEBUG_EVENT_END];
+ /* Events active within the current step. */
+ uint64_t active_events;
+ /* Breakpoint was called within the current step. */
+ bool breakpoint_called;
+ /* Currently active conditional jump pc. */
+ uint32_t branch_pc;
+ /* Program counters of all conditional jumps in the current code path. */
+ uint32_t branch_pc_stack[4];
+ size_t branch_pc_stack_length;
+};
+
+/* Push current branch_pc to stack and clear it. */
+static int
+test_events_branch_pc_push(struct test_events_context *ctx)
+{
+ TEST_ASSERT_NOT_EQUAL(ctx->branch_pc, NO_PROGRAM_COUNTER,
+ "no branch pc to push");
+ TEST_ASSERT(ctx->branch_pc_stack_length < RTE_DIM(ctx->branch_pc_stack),
+ "branch stack overflow");
+ ctx->branch_pc_stack[ctx->branch_pc_stack_length++] = ctx->branch_pc;
+ ctx->branch_pc = NO_PROGRAM_COUNTER;
+ return TEST_SUCCESS;
+}
+
+/* Pop branch_pc from stack. */
+static int
+test_events_branch_pc_pop(struct test_events_context *ctx)
+{
+ TEST_ASSERT(ctx->branch_pc_stack_length > 0, "branch stack underflow");
+ ctx->branch_pc = ctx->branch_pc_stack[--ctx->branch_pc_stack_length];
+ return TEST_SUCCESS;
+}
+
+/* Verify breakpoint, checking that it is only called once and remembering it. */
+static int
+test_events_break_cb(struct rte_bpf_validate_debug *debug, void *void_ctx)
+{
+ struct test_events_context *const ctx = void_ctx;
+ const uint32_t pc = rte_bpf_validate_debug_get_pc(debug);
+
+ TEST_LOG_LINE(DEBUG, "%2u: break", pc);
+
+ TEST_ASSERT(!ctx->breakpoint_called,
+ "breakpoint called again at pc %u", pc);
+ ctx->breakpoint_called = true;
+
+ return TEST_SUCCESS;
+}
+
+/*
+ * Verify event, checking that:
+ * - validation-start is called at pc 0;
+ * - branch-return is called at a corresponding conditional jump pc;
+ * - branch-enter and branch-return are correctly paired;
+ * - jump-conditional and jump-always are called for corresponding opcodes;
+ * - jump-conditional and only it is followed by branches events;
+ * - events and breakpoints are correctly ordered;
+ *
+ * Increment event counter to compare against expected counts later.
+ */
+static int
+test_events_catch_cb(struct rte_bpf_validate_debug *debug, void *void_ctx)
+{
+ /* Bitmask of events allowed to precede each given event within the step. */
+ static const uint64_t allowed_preceding_events[] = {
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_START] = 0,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE] =
+ EVENT_BIT(BRANCH_PRUNE) |
+ EVENT_BIT(BRANCH_UNREACHABLE) |
+ EVENT_BIT(BRANCH_RETURN),
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE] =
+ EVENT_BIT(BRANCH_PRUNE) |
+ EVENT_BIT(BRANCH_UNREACHABLE) |
+ EVENT_BIT(BRANCH_RETURN),
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER] =
+ EVENT_BIT(BRANCH_PRUNE) |
+ EVENT_BIT(BRANCH_UNREACHABLE) |
+ EVENT_BIT(BRANCH_RETURN),
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN] =
+ EVENT_BIT(BRANCH_PRUNE) |
+ EVENT_BIT(BRANCH_UNREACHABLE) |
+ EVENT_BIT(BRANCH_RETURN),
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL] =
+ EVENT_BIT(BRANCH_ENTER) |
+ EVENT_BIT(BRANCH_PRUNE) |
+ EVENT_BIT(BRANCH_UNREACHABLE) |
+ EVENT_BIT(BRANCH_RETURN),
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS] =
+ EVENT_BIT(BRANCH_ENTER) |
+ EVENT_BIT(BRANCH_PRUNE) |
+ EVENT_BIT(BRANCH_UNREACHABLE) |
+ EVENT_BIT(BRANCH_RETURN),
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_STEP] =
+ EVENT_BITS_ALL - EVENT_BIT(VALIDATION_SUCCESS),
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS] =
+ EVENT_BIT(BRANCH_PRUNE) |
+ EVENT_BIT(BRANCH_UNREACHABLE) |
+ EVENT_BIT(BRANCH_RETURN),
+ };
+
+ struct test_events_context *const ctx = void_ctx;
+ const enum rte_bpf_validate_debug_event event =
+ rte_bpf_validate_debug_get_event(debug);
+ const uint32_t pc = rte_bpf_validate_debug_get_pc(debug);
+
+ TEST_LOG_LINE(DEBUG, "%2u: %s", pc, event_names[event]);
+
+ /* Verify event program counter and underlying instruction. */
+ switch (event) {
+ case RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_START:
+ TEST_ASSERT_EQUAL(pc, 0,
+ "event %s called at pc %u",
+ event_names[event], pc);
+ break;
+ case RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER:
+ TEST_ASSERT_SUCCESS(test_events_branch_pc_push(ctx),
+ "error pushing branch for event %s at pc %u",
+ event_names[event], pc);
+ break;
+ case RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN:
+ TEST_ASSERT_SUCCESS(test_events_branch_pc_pop(ctx),
+ "error popping branch for event %s at pc %u",
+ event_names[event], pc);
+ TEST_ASSERT_EQUAL(pc, ctx->branch_pc,
+ "event %s called at pc %u after jump pc %u",
+ event_names[event], pc, ctx->branch_pc);
+ break;
+ case RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS:
+ TEST_ASSERT_EQUAL(BPF_CLASS(ctx->ins[pc].code), BPF_JMP,
+ "event %s called at non-jump pc %u",
+ event_names[event], pc);
+ TEST_ASSERT_EQUAL(BPF_OP(ctx->ins[pc].code), BPF_JA,
+ "event %s called at non-jump-always pc %u",
+ event_names[event], pc);
+ break;
+ case RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL:
+ TEST_ASSERT_EQUAL(BPF_CLASS(ctx->ins[pc].code), BPF_JMP,
+ "event %s called at non-jump pc %u",
+ event_names[event], pc);
+ TEST_ASSERT_NOT_EQUAL(BPF_OP(ctx->ins[pc].code), BPF_JA,
+ "event %s called at jump-always pc %u",
+ event_names[event], pc);
+ break;
+ default:
+ break;
+ }
+
+ /* Verify events order. */
+ TEST_ASSERT_EQUAL(
+ (ctx->active_events & ~allowed_preceding_events[event]), 0,
+ "event %s at pc %u preceded by %#jx",
+ event_names[event], pc, (uintmax_t)ctx->active_events);
+
+ /* Update events for the current step as well as branch program counter. */
+ if (event != RTE_BPF_VALIDATE_DEBUG_EVENT_STEP) {
+ TEST_ASSERT(!ctx->breakpoint_called,
+ "event %s called after a breakpoint at pc %u",
+ event_names[event], pc);
+ ctx->active_events |= RTE_BIT64(event);
+ } else {
+ TEST_ASSERT_EQUAL(ctx->branch_pc, NO_PROGRAM_COUNTER,
+ "no branch-enter between branch- and next instruction");
+
+ if (ctx->active_events & EVENT_BIT(JUMP_CONDITIONAL))
+ /* Remember branch instruction pointer, branches to follow. */
+ ctx->branch_pc = pc;
+
+ TEST_ASSERT(ctx->breakpoint_called,
+ "step called before or without a breakpoint at pc %u", pc);
+
+ /* Prepare for the next step. */
+ ctx->active_events = 0;
+ ctx->breakpoint_called = false;
+ }
+
+ /* Update total event counts. */
+ ++ctx->event_counts[event];
+
+ return TEST_SUCCESS;
+}
+
+/*
+ * This test verifies that eBPF validator generates expected numbers of events
+ * of each kind in the expected order (e.g. step after jump after branch).
+ */
+static int
+test_events(void)
+{
+ /*
+ * Program demonstrating the following properties:
+ * - prunnable code paths;
+ * - dynamically unreachable code paths;
+ * - conditional jumps landing on conditional jumps;
+ * - conditional jumps landing on unconditional jumps;
+ */
+ static const struct ebpf_insn ins[] = {
+ /* 0: r0 = 0 */
+ { .code = (EBPF_ALU64 | EBPF_MOV | BPF_K), .dst_reg = EBPF_REG_0, .imm = 0 },
+ /* 1: if r1 > 0 goto 4 */
+ { .code = (BPF_JMP | BPF_JGT | BPF_K), .dst_reg = EBPF_REG_1, .imm = 0, .off = 2 },
+ /* 2: r1 = 0 */
+ { .code = (EBPF_ALU64 | EBPF_MOV | BPF_K), .dst_reg = EBPF_REG_1, .imm = 0 },
+ /* 3: if r0 < 1 goto 8 */
+ { .code = (BPF_JMP | EBPF_JLT | BPF_K), .dst_reg = EBPF_REG_0, .imm = 1, .off = 4 },
+ /* 4: if r1 > 0 goto 6 */
+ { .code = (BPF_JMP | BPF_JGT | BPF_K), .dst_reg = EBPF_REG_1, .imm = 0, .off = 1 },
+ /* 5: goto 9 */
+ { .code = (BPF_JMP | BPF_JA), .off = 3 },
+ /* 6: r1 = 0 */
+ { .code = (EBPF_ALU64 | EBPF_MOV | BPF_K), .dst_reg = EBPF_REG_1, .imm = 0 },
+ /* 7: if r0 > 0 goto 9 */
+ { .code = (BPF_JMP | BPF_JGT | BPF_K), .dst_reg = EBPF_REG_0, .imm = 0, .off = 1 },
+ /* 8: goto 10 */
+ { .code = (BPF_JMP | BPF_JA), .off = 1 },
+ /* 9: r0 = 1 */
+ { .code = (EBPF_ALU64 | EBPF_MOV | BPF_K), .dst_reg = EBPF_REG_0, .imm = 1 },
+ /* 10: exit */
+ { .code = (BPF_JMP | EBPF_EXIT) }
+ };
+
+ /*
+ * Verified code paths (order may vary by library implementation):
+ * 0 -- 1 -- 4 -- 6 -- 7 -- unreachable 9
+ * | | `-- 8 -- 10 -- return 7 -- return 4
+ * | `-- unreachable 5 -- return 1
+ * `-- 2 -- 3 -- prune 8
+ * `-- unreachable 4 -- return 1 -- success 11
+ *
+ * Numbers without preceding word correspond to evaluated instructions,
+ * there are 9 of these in total, so 9 step events are expected.
+ *
+ * Note that 4 encountered conditional jump instructions generate:
+ * - 4 pairs of branch-enter/branch-return events;
+ * - 3 branch-unreachable events;
+ * - 1 branch-prune event;
+ * for a total count of 8 (4 instructions * 2 branches/instruction).
+ */
+ static const int expected_counts[RTE_BPF_VALIDATE_DEBUG_EVENT_END] = {
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_STEP] = 9,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_START] = 1,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS] = 1,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER] = 4,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE] = 1,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN] = 4,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE] = 3,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS] = 1,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL] = 4,
+ };
+
+ struct test_events_context ctx = {
+ .ins = ins,
+ .nb_ins = RTE_DIM(ins),
+ .branch_pc = NO_PROGRAM_COUNTER,
+ };
+
+ struct rte_bpf_validate_debug *debug = rte_bpf_validate_debug_create();
+ TEST_ASSERT_NOT_NULL(debug, "validate debug create error");
+
+ /* Set breakpoints for all instructions. */
+ for (uint32_t pc = 0; pc != RTE_DIM(ins); ++pc)
+ rte_bpf_validate_debug_break(debug, pc,
+ &(struct rte_bpf_validate_debug_callback){
+ .fn = test_events_break_cb,
+ .ctx = &ctx,
+ });
+
+ /* Set catchpoints for all events except invalid-state. */
+ for (enum rte_bpf_validate_debug_event event = 0;
+ event != RTE_BPF_VALIDATE_DEBUG_EVENT_END; ++event)
+ if (event != RTE_BPF_VALIDATE_DEBUG_EVENT_INVALID_STATE)
+ rte_bpf_validate_debug_catch(debug, event,
+ &(struct rte_bpf_validate_debug_callback){
+ .fn = test_events_catch_cb,
+ .ctx = &ctx,
+ });
+
+ struct rte_bpf *const bpf = rte_bpf_load_ex(&(struct rte_bpf_prm_ex){
+ .sz = sizeof(struct rte_bpf_prm_ex),
+ .origin = RTE_BPF_ORIGIN_RAW,
+ .raw.ins = ins,
+ .raw.nb_ins = RTE_DIM(ins),
+ .prog_arg[0] = { .type = RTE_BPF_ARG_RAW },
+ .nb_prog_arg = 1,
+ .debug = debug,
+ });
+
+ rte_bpf_destroy(bpf);
+ rte_bpf_validate_debug_destroy(debug);
+
+ TEST_ASSERT_NOT_NULL(bpf, "validation failed");
+
+ /* Verify event counts. */
+ for (enum rte_bpf_validate_debug_event event = 0;
+ event != RTE_BPF_VALIDATE_DEBUG_EVENT_END; ++event)
+ TEST_ASSERT_EQUAL(
+ ctx.event_counts[event], expected_counts[event],
+ "expected %d %s events, but got %d",
+ expected_counts[event], event_names[event],
+ ctx.event_counts[event]);
+
+ /* Additionally double-check final stack state. */
+ TEST_ASSERT_EQUAL(ctx.branch_pc_stack_length, 0,
+ "unbalanced branch enters and returns");
+
+ return TEST_SUCCESS;
+}
+
+REGISTER_FAST_TEST(bpf_validate_events_autotest, NOHUGE_OK, ASAN_OK, test_events);
+
+/*
+ * Tests of BPF validation.
+ */
+
/* Sizes of text buffers used for formatting various debug outputs. */
#define VALUE_FORMAT_BUFFER_SIZE 32
#define INTERVAL_FORMAT_BUFFER_SIZE 64
--
2.43.0
^ permalink raw reply related [flat|nested] 38+ messages in thread
* [PATCH v3 7/7] app/validate-bpf: add BPF validation application
2026-09-15 19:26 ` [PATCH v3 " Marat Khalili
` (5 preceding siblings ...)
2026-09-15 19:26 ` [PATCH v3 6/7] app/test: add test for bpf validate debug events Marat Khalili
@ 2026-09-15 19:26 ` Marat Khalili
2026-09-16 16:36 ` Konstantin Ananyev
2026-09-17 14:35 ` [PATCH v4 0/7] validate-bpf: add standalone validation debug tool Marat Khalili
7 siblings, 1 reply; 38+ messages in thread
From: Marat Khalili @ 2026-09-15 19:26 UTC (permalink / raw)
To: Thomas Monjalon, Konstantin Ananyev; +Cc: dev
Add a new DPDK application, dpdk-validate-bpf, for pre-validating eBPF
programs for compatibility with the lib/bpf execution context.
The application allows evaluating eBPF programs against the DPDK verifier
before loading them into a real application. It includes an interactive
debugging mode to trace state changes per instruction and understand the
validator's decisions.
Signed-off-by: Marat Khalili <marat.khalili@huawei.com>
---
MAINTAINERS | 2 +
app/meson.build | 1 +
app/validate-bpf/alloc_list.c | 53 ++
app/validate-bpf/args.c | 263 ++++++
app/validate-bpf/debug.c | 1103 ++++++++++++++++++++++++
app/validate-bpf/debug_command.c | 383 ++++++++
app/validate-bpf/debug_command.h | 65 ++
app/validate-bpf/eal_init_args.c | 57 ++
app/validate-bpf/internal.h | 151 ++++
app/validate-bpf/main.c | 90 ++
app/validate-bpf/meson.build | 13 +
app/validate-bpf/parse_decl.c | 623 +++++++++++++
doc/guides/rel_notes/release_26_11.rst | 5 +
doc/guides/tools/index.rst | 1 +
doc/guides/tools/validate_bpf.rst | 97 +++
15 files changed, 2907 insertions(+)
create mode 100644 app/validate-bpf/alloc_list.c
create mode 100644 app/validate-bpf/args.c
create mode 100644 app/validate-bpf/debug.c
create mode 100644 app/validate-bpf/debug_command.c
create mode 100644 app/validate-bpf/debug_command.h
create mode 100644 app/validate-bpf/eal_init_args.c
create mode 100644 app/validate-bpf/internal.h
create mode 100644 app/validate-bpf/main.c
create mode 100644 app/validate-bpf/meson.build
create mode 100644 app/validate-bpf/parse_decl.c
create mode 100644 doc/guides/tools/validate_bpf.rst
diff --git a/MAINTAINERS b/MAINTAINERS
index e99a65d1974d..39cd0a3de6fa 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -1923,8 +1923,10 @@ F: lib/bpf/
F: examples/bpf/
F: app/test/test_bpf*
F: app/test/bpf/
+F: app/validate-bpf/
F: app/test-pmd/bpf_cmd.*
F: doc/guides/prog_guide/bpf_lib.rst
+F: doc/guides/tools/validate_bpf.rst
Graph
M: Jerin Jacob <jerinj@marvell.com>
diff --git a/app/meson.build b/app/meson.build
index 1798db3ae43f..9b59ae483056 100644
--- a/app/meson.build
+++ b/app/meson.build
@@ -33,6 +33,7 @@ apps = [
'test-regex',
'test-sad',
'test-security-perf',
+ 'validate-bpf',
]
if get_option('tests')
diff --git a/app/validate-bpf/alloc_list.c b/app/validate-bpf/alloc_list.c
new file mode 100644
index 000000000000..2f4d18ad3df6
--- /dev/null
+++ b/app/validate-bpf/alloc_list.c
@@ -0,0 +1,53 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include "internal.h"
+#include <stdlib.h>
+
+
+/* Needs to be a power of two. */
+#define START_CAPACITY (1u << 3)
+
+size_t
+alloc_list_append(struct alloc_list *alloc_list, void *ptr)
+{
+ if (alloc_list->count == 0) {
+ RTE_ASSERT(alloc_list->ptrs == NULL);
+ alloc_list->ptrs = malloc(
+ sizeof(alloc_list->ptrs[0]) * START_CAPACITY);
+ RTE_VERIFY(alloc_list->ptrs != NULL);
+ } else if (alloc_list->count >= START_CAPACITY &&
+ /* Power of two detection */
+ (alloc_list->count & (alloc_list->count - 1)) == 0) {
+ /*
+ * We allocate in powers of two, and current count is one of
+ * these powers, so need to reallocate to larger capacity.
+ */
+ RTE_ASSERT(alloc_list->ptrs != NULL);
+ const size_t new_capacity = alloc_list->count * 2;
+ alloc_list->ptrs = realloc(alloc_list->ptrs,
+ sizeof(alloc_list->ptrs[0]) * new_capacity);
+ RTE_VERIFY(alloc_list->ptrs != NULL);
+ }
+ alloc_list->ptrs[alloc_list->count] = ptr;
+ return alloc_list->count++;
+}
+
+void alloc_list_replace(struct alloc_list *alloc_list, size_t index, void *ptr)
+{
+ RTE_ASSERT(index < alloc_list->count);
+ alloc_list->ptrs[index] = ptr;
+}
+
+void alloc_list_free_all(struct alloc_list *alloc_list)
+{
+ /* Copy and clear fields first in case alloc_list itself gets freed. */
+ size_t count = alloc_list->count;
+ void ** const ptrs = alloc_list->ptrs;
+ *alloc_list = (struct alloc_list){};
+
+ while (count != 0)
+ free(ptrs[--count]);
+ free(ptrs);
+}
diff --git a/app/validate-bpf/args.c b/app/validate-bpf/args.c
new file mode 100644
index 000000000000..cd7cbd694a3c
--- /dev/null
+++ b/app/validate-bpf/args.c
@@ -0,0 +1,263 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include "internal.h"
+
+#include <ctype.h>
+#include <getopt.h>
+#include <stdlib.h>
+
+#include <rte_errno.h>
+
+
+/* Values to be used in getopt_long option.val. */
+enum app_args {
+ ARG_UNRECOGNIZED = '?',
+ ARG_AUTO = 0, /* Value set by getopt_long when flag is non-NULL. */
+ ARG_HELP, /* Keep this one in the beginning for UX reasons. */
+ ARG_DEBUG,
+ ARG_MBUF_BUF_SIZE,
+ ARG_NO_PROG_ARGS,
+ ARG_PROG_ARG,
+ ARG_SECTION,
+ ARG_XSYM,
+};
+
+/* Program options. */
+static struct option OPTIONS[] = {
+ {
+ .name = "help",
+ .val = ARG_HELP,
+ },
+ {
+ .name = "debug",
+ .val = ARG_DEBUG,
+ },
+ {
+ .name = "mbuf-buf-size",
+ .has_arg = required_argument,
+ .val = ARG_MBUF_BUF_SIZE,
+ },
+ {
+ .name = "no-prog-arg",
+ .val = ARG_NO_PROG_ARGS,
+ },
+ {
+ .name = "no-prog-args",
+ .val = ARG_NO_PROG_ARGS,
+ },
+ {
+ .name = "prog-arg",
+ .has_arg = required_argument,
+ .val = ARG_PROG_ARG,
+ },
+ {
+ .name = "section",
+ .has_arg = required_argument,
+ .val = ARG_SECTION,
+ },
+ {
+ .name = "xsym",
+ .has_arg = required_argument,
+ .val = ARG_XSYM,
+ },
+ { /* terminating zero record */ }
+};
+
+/* Default args value. */
+static const struct args ARGS_DEFAULT = {
+ .bpf_prm = {
+ .sz = sizeof(struct rte_bpf_prm_ex),
+ .origin = RTE_BPF_ORIGIN_ELF_FILE,
+ .elf_file.section = ".text",
+ },
+ .mbuf_buf_size = RTE_MBUF_DEFAULT_BUF_SIZE,
+};
+
+/* Default --prog-arg argument value. */
+static const char * const PROG_ARG_DEFAULT = "struct rte_mbuf *";
+
+
+void
+print_usage(const char *program_name)
+{
+ static const char *const options[][2] = {
+ { "--help", "Display help and exit." },
+ { "--debug", "Enable interactive debug mode." },
+ { "--mbuf-buf-size=MBUF_BUF_SIZE", "Size of the mbuf data buffer (in bytes)." },
+ { "--prog-arg=TYPE", "Expected type of the next BPF program argument (up to 5)." },
+ { "--no-prog-args", "BPF program does not take any arguments." },
+ { "--section=SECTION", "ELF section name in the BPF object file to load." },
+ { "--xsym='TYPE NAME[(TYPE, ...)]'", "External symbol BPF program can access." },
+ };
+
+ printf("USAGE: %s [OPTIONS]... BPF_PATH\n", program_name);
+ printf("OPTIONS:\n");
+ for (int oi = 0; oi != RTE_DIM(options); ++oi)
+ printf("\t%-31s %s\n", options[oi][0], options[oi][1]);
+}
+
+void
+print_defaults(void)
+{
+ printf("DEFAULTS:"
+ " --mbuf-buf-size=%#zx"
+ " --prog-arg='%s'"
+ " --section='%s'\n",
+ (size_t)RTE_MBUF_DEFAULT_BUF_SIZE,
+ PROG_ARG_DEFAULT,
+ ARGS_DEFAULT.bpf_prm.elf_file.section);
+}
+
+static int
+parse_size(size_t *result, const char *text)
+{
+ char *parse_end = NULL;
+
+ errno = 0;
+ const uintmax_t strtoumax_result = strtoumax(text, &parse_end, 0);
+ if (errno != 0 || *parse_end != '\0' || strtoumax_result == 0 ||
+ strtoumax_result > SIZE_MAX)
+ return -1;
+
+ *result = strtoumax_result;
+ return 0;
+}
+
+struct args *
+args_parse(int argc, char *argv[])
+{
+ int val;
+ size_t xsym_alloc_index;
+ struct rte_bpf_xsym *xsym = NULL;
+ const char * const program_name = argv[0];
+
+ /* Allocate args and set aliases for some of its members. */
+ struct args * const args = malloc(sizeof(*args));
+ RTE_VERIFY(args != NULL);
+ struct alloc_list * const alloc_list = &args->_alloc_list;
+ struct rte_bpf_prm_ex * const bpf_prm = &args->bpf_prm;
+
+ /* Set default values. */
+ *args = ARGS_DEFAULT;
+ RTE_VERIFY(parse_arg(&bpf_prm->prog_arg[bpf_prm->nb_prog_arg++],
+ PROG_ARG_DEFAULT) >= 0);
+ bool default_prog_args = true;
+
+ /* Reserve space for xsym in alloc_list. */
+ xsym_alloc_index = alloc_list_append(alloc_list, xsym);
+
+ while ((val = getopt_long(argc, argv, "", OPTIONS, NULL)) != EOF) {
+ int rc = 0;
+ switch (val) {
+ case ARG_AUTO:
+ /* getopt_long made the assignment, nothing to do */
+ break;
+ case ARG_HELP:
+ args->show_help = true;
+ break;
+ case ARG_DEBUG:
+ if (bpf_prm->debug == NULL)
+ bpf_prm->debug = debug_create();
+ if (bpf_prm->debug == NULL) {
+ rc = -rte_errno;
+ fprintf(stderr,
+ "%s: error %d creating debug session\n",
+ program_name, -rc);
+ }
+ break;
+ case ARG_MBUF_BUF_SIZE:
+ rc = parse_size(&args->mbuf_buf_size, optarg);
+ if (rc < 0)
+ fprintf(stderr,
+ "%s: invalid mbuf buf size '%s'\n",
+ program_name, optarg);
+ break;
+ case ARG_NO_PROG_ARGS:
+ bpf_prm->nb_prog_arg = 0;
+ default_prog_args = false;
+ break;
+ case ARG_PROG_ARG:
+ if (default_prog_args) {
+ bpf_prm->nb_prog_arg = 0;
+ default_prog_args = false;
+ }
+
+ if (bpf_prm->nb_prog_arg == RTE_DIM(bpf_prm->prog_arg)) {
+ fprintf(stderr,
+ "%s: at most %d program arguments allowed\n",
+ program_name,
+ (int)RTE_DIM(bpf_prm->prog_arg));
+ rc = -EINVAL;
+ break;
+ }
+
+ rc = parse_arg(&bpf_prm->prog_arg[bpf_prm->nb_prog_arg++],
+ optarg);
+ if (rc < 0)
+ fprintf(stderr,
+ "%s: unrecognized prog arg '%s'\n",
+ program_name, optarg);
+
+ break;
+ case ARG_SECTION:
+ bpf_prm->elf_file.section = optarg;
+ break;
+ case ARG_XSYM:
+ bpf_prm->xsym = xsym = realloc(xsym,
+ sizeof(xsym[0]) * (bpf_prm->nb_xsym + 1));
+ RTE_VERIFY(xsym != NULL);
+ alloc_list_replace(alloc_list, xsym_alloc_index, xsym);
+ rc = parse_xsym(&xsym[bpf_prm->nb_xsym++], optarg,
+ alloc_list);
+ if (rc < 0)
+ fprintf(stderr, "%s: invalid xsym '%s'\n",
+ program_name, optarg);
+ break;
+ case ARG_UNRECOGNIZED:
+ args_destroy(args);
+ return NULL;
+ default:
+ rte_panic("Unexpected getopt_long return value %d\n",
+ val);
+ }
+ if (rc < 0) {
+ args_destroy(args);
+ return NULL;
+ }
+ }
+
+ /* Set buf_size to the value specified in command line arguments. */
+ adjust_arg_buf_size(bpf_prm->prog_arg, args->mbuf_buf_size);
+ for (size_t xsymi = 0; xsymi != bpf_prm->nb_xsym; ++xsymi)
+ adjust_xsym_buf_size(&xsym[xsymi], args->mbuf_buf_size);
+
+ /* getopt_long moves all non-options to the end, starting at optind. */
+ const int nb_bpf_path = argc - optind;
+ switch (nb_bpf_path) {
+ case 0:
+ break;
+ case 1:
+ bpf_prm->elf_file.path = argv[optind];
+ break;
+ default:
+ fprintf(stderr, "%s: too many positional arguments\n",
+ program_name);
+ args_destroy(args);
+ return NULL;
+ }
+
+ return args;
+}
+
+void
+args_destroy(struct args *args)
+{
+ if (args == NULL)
+ return;
+ if (args->bpf_prm.debug != NULL)
+ debug_destroy(args->bpf_prm.debug);
+ alloc_list_free_all(&args->_alloc_list);
+ free(args);
+}
diff --git a/app/validate-bpf/debug.c b/app/validate-bpf/debug.c
new file mode 100644
index 000000000000..ac085b53e3bd
--- /dev/null
+++ b/app/validate-bpf/debug.c
@@ -0,0 +1,1103 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include "debug_command.h"
+#include "internal.h"
+
+#include <rte_bpf_validate_debug.h>
+#include <rte_errno.h>
+
+#include <stdlib.h>
+
+
+/* Write single line to the user, currently just to stdout. */
+#define PRINTLN(fmt, ...) do { \
+ RTE_LOG_CHECK_NO_NEWLINE(fmt); \
+ printf(fmt "\n", ##__VA_ARGS__); \
+} while (0)
+
+#define PROMPT "(validate) "
+
+#define INITIAL_CAPACITY 8u
+
+static const char *const event_names[] = {
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_INVALID_STATE] = "invalid-state",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER] = "branch-enter",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE] = "branch-prune",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN] = "branch-return",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE] = "branch-unreachable",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS] = "jump-always",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL] = "jump-conditional",
+};
+
+static const char *const register_names[] = {
+ [EBPF_REG_0] = "r0",
+ [EBPF_REG_1] = "r1",
+ [EBPF_REG_2] = "r2",
+ [EBPF_REG_3] = "r3",
+ [EBPF_REG_4] = "r4",
+ [EBPF_REG_5] = "r5",
+ [EBPF_REG_6] = "r6",
+ [EBPF_REG_7] = "r7",
+ [EBPF_REG_8] = "r8",
+ [EBPF_REG_9] = "r9",
+ [EBPF_REG_10] = "r10",
+};
+
+static const char *const comparison_operator_names[] = {
+ [BPF_JEQ] = "==",
+ [BPF_JGT] = ">",
+ [BPF_JGE] = ">=",
+ [EBPF_JNE] = "!=",
+ [EBPF_JSGT] = "s>",
+ [EBPF_JSGE] = "s>=",
+ [EBPF_JLT] = "<",
+ [EBPF_JLE] = "<=",
+ [EBPF_JSLT] = "s<",
+ [EBPF_JSLE] = "s<=",
+};
+
+enum point_type {
+ POINT_TYPE_BREAK,
+ POINT_TYPE_CATCH,
+};
+
+/* Local representation of points with additional info for UI. */
+struct point_info {
+ struct rte_bpf_validate_debug_point *point;
+ enum point_type type;
+ union {
+ uint32_t pc;
+ enum rte_bpf_validate_debug_event event;
+ };
+};
+
+/* Dynamically growing list of point infos. */
+struct point_infos {
+ struct point_info *elements;
+ uint32_t length;
+ uint32_t capacity;
+};
+
+/* Information about a single conditional or unconditional jump in code path. */
+struct branch_info {
+ uint32_t jump_pc;
+ uint32_t target_pc;
+ bool is_conditional;
+};
+
+/* Dynamically growing stack to track code path. */
+struct branch_stack {
+ struct branch_info *branches;
+ uint32_t length;
+ uint32_t capacity;
+};
+
+/* Flags telling if we should validate again, stopping or not at start. */
+static bool validate_again;
+
+/* List of point infos; element index is its ID for UI. */
+static struct point_infos point_infos;
+
+/* Catchpoint invisible to the user used for step-by-step validation. */
+static struct rte_bpf_validate_debug_point *step_point;
+
+/* Tracking entered branches. */
+static struct branch_stack branch_stack;
+static uint32_t pending_jump_pc = UINT32_MAX;
+static struct rte_bpf_validate_debug_point *jump_always_step_point;
+
+static int
+step_cb(struct rte_bpf_validate_debug *debug, void *ctx);
+static int
+point_cb(struct rte_bpf_validate_debug *debug, void *ctx);
+
+/* Find index of the string in the list. */
+static int
+find_name(const char *name, const char *const *names, int nb_names)
+{
+ if (nb_names < 0)
+ return -EINVAL;
+
+ for (int index = 0; index != nb_names; ++index)
+ if (names[index] != NULL && strcmp(names[index], name) == 0)
+ return index;
+
+ return -ENOENT;
+}
+
+/* Convert event name to its enum value. */
+static enum rte_bpf_validate_debug_event
+parse_event_name(const char *name)
+{
+ const int rc = find_name(name, event_names, RTE_DIM(event_names));
+ if (rc < 0)
+ PRINTLN("Error: invalid event name.");
+
+ return rc;
+}
+
+/* Convert comparison operator name to opcode. */
+static int
+parse_comparison_operator(const char *name)
+{
+ const int rc = find_name(name, comparison_operator_names,
+ RTE_DIM(comparison_operator_names));
+ if (rc < 0)
+ PRINTLN("Error: invalid comparison operator name.");
+
+ return rc;
+}
+
+/* Convert register name to its number. */
+static int
+parse_register(const char *name)
+{
+ const int rc = find_name(name, register_names, RTE_DIM(register_names));
+ if (rc < 0)
+ PRINTLN("Error: invalid register name.");
+
+ return rc;
+}
+
+/* Free local list of point infos (do not destroy points). */
+static void
+point_infos_free(void)
+{
+ free(point_infos.elements);
+ point_infos = (struct point_infos){};
+}
+
+/* Free local branch stack. */
+static void
+branch_stack_free(void)
+{
+ free(branch_stack.branches);
+ branch_stack = (struct branch_stack){};
+}
+
+/* Return existing element from the list of point infos. */
+static struct point_info *
+point_infos_at(uint32_t point_number)
+{
+ RTE_ASSERT(point_number < point_infos.length);
+ RTE_ASSERT(point_infos.elements[point_number].point != NULL);
+ return &point_infos.elements[point_number];
+}
+
+/* Print existing element from the list of point infos. */
+static void
+point_infos_print_at(uint32_t point_number)
+{
+ const struct point_info *const point_info = point_infos_at(point_number);
+
+ switch (point_info->type) {
+ case POINT_TYPE_BREAK:
+ PRINTLN("Breakpoint %d at %d.", point_number,
+ point_info->pc);
+ break;
+ case POINT_TYPE_CATCH:
+ PRINTLN("Catchpoint %d on %s.", point_number,
+ event_names[point_info->event]);
+ break;
+ default:
+ PRINTLN("Point %d of unknown type", point_number);
+ break;
+ }
+}
+
+/* Print all point infos. */
+static int
+point_infos_print_all(void)
+{
+ uint32_t nb_printed = 0;
+ for (uint32_t pn = 0; pn != point_infos.length; ++pn) {
+ const struct point_info *const point_info =
+ &point_infos.elements[pn];
+ if (point_info->point != NULL) {
+ point_infos_print_at(pn);
+ ++nb_printed;
+ }
+ }
+
+ if (nb_printed == 0) {
+ PRINTLN("No breakpoints or catchpoints set.");
+ return -ENOENT;
+ }
+
+ return nb_printed;
+}
+
+/* Allocate space for a new element in the list of point infos. */
+static uint32_t
+point_infos_append(void)
+{
+ if (point_infos.length == point_infos.capacity) {
+ /* Set to initial capacity or double previous one. */
+ point_infos.capacity = RTE_MAX(INITIAL_CAPACITY,
+ point_infos.capacity * 2);
+ point_infos.elements = realloc(point_infos.elements,
+ point_infos.capacity * sizeof(point_infos.elements[0]));
+ RTE_VERIFY(point_infos.elements != NULL);
+ }
+ return point_infos.length++;
+}
+
+/* Allocate space for a new element in the branch stack. */
+static void
+branch_stack_append(const struct branch_info *branch)
+{
+ if (branch_stack.length == branch_stack.capacity) {
+ /* Set to initial capacity or double previous one. */
+ branch_stack.capacity = RTE_MAX(INITIAL_CAPACITY,
+ branch_stack.capacity * 2);
+ branch_stack.branches = realloc(branch_stack.branches,
+ branch_stack.capacity * sizeof(branch_stack.branches[0]));
+ RTE_VERIFY(branch_stack.branches != NULL);
+ }
+ branch_stack.branches[branch_stack.length++] = *branch;
+}
+
+/* Destroy existing element from the list of point infos, printing it first. */
+static void
+point_infos_destroy_existing(uint32_t point_number)
+{
+ point_infos_print_at(point_number);
+
+ struct point_info *const point_info = point_infos_at(point_number);
+ rte_bpf_validate_debug_point_destroy(point_info->point);
+ *point_info = (struct point_info){};
+}
+
+/* Destroy point with specified number if it exists. */
+static int
+point_infos_destroy_at(uint32_t point_number)
+{
+ if (point_number >= point_infos.length ||
+ point_infos.elements[point_number].point == NULL) {
+ PRINTLN("No breakpoint number %d.", point_number);
+ return -ENOENT;
+ }
+
+ point_infos_destroy_existing(point_number);
+ return 1;
+}
+
+/* Destroy all point infos. */
+static int64_t
+point_infos_destroy_all(void)
+{
+ uint32_t nb_destroyed = 0;
+ for (uint32_t pn = 0; pn < point_infos.length; ++pn) {
+ const struct point_info *const point_info =
+ &point_infos.elements[pn];
+ if (point_info->point != NULL) {
+ point_infos_destroy_existing(pn);
+ ++nb_destroyed;
+ }
+ }
+
+ if (nb_destroyed == 0) {
+ PRINTLN("No breakpoints or catchpoints set.");
+ return -ENOENT;
+ }
+
+ return nb_destroyed;
+}
+
+/* Destroy all breakpoints at specified location. */
+static int64_t
+point_infos_destroy_breakpoints(uint32_t pc)
+{
+ uint32_t nb_destroyed = 0;
+ for (uint32_t pn = 0; pn < point_infos.length; ++pn) {
+ const struct point_info *const point_info =
+ &point_infos.elements[pn];
+ if (point_info->point != NULL &&
+ point_info->type == POINT_TYPE_BREAK &&
+ point_info->pc == pc) {
+ point_infos_destroy_existing(pn);
+ ++nb_destroyed;
+ }
+ }
+
+ if (nb_destroyed == 0) {
+ PRINTLN("No breakpoint at %u.", pc);
+ return -ENOENT;
+ }
+
+ return nb_destroyed;
+}
+
+/* Destroy all catchpoints for specified event. */
+static int64_t
+point_infos_destroy_catchpoints(int event)
+{
+ if (event < 0)
+ /* Error was already printed by parse_event_name. */
+ return event;
+
+ uint32_t nb_destroyed = 0;
+ for (uint32_t pn = 0; pn < point_infos.length; ++pn) {
+ const struct point_info *const point_info =
+ &point_infos.elements[pn];
+ if (point_info->point != NULL &&
+ point_info->type == POINT_TYPE_CATCH &&
+ (int)point_info->event == event) {
+ point_infos_destroy_existing(pn);
+ ++nb_destroyed;
+ }
+ }
+
+ if (nb_destroyed == 0) {
+ PRINTLN("No catchpoint on %s.", event_names[event]);
+ return -ENOENT;
+ }
+
+ return nb_destroyed;
+}
+
+/* Create new breakpoint at specified location. */
+static int
+add_breakpoint(struct rte_bpf_validate_debug *debug, uint32_t nb_ins, uint32_t pc)
+{
+ const uint32_t point_number = point_infos_append();
+
+ if (pc >= nb_ins) {
+ PRINTLN("Error: program only has %u instructions.", nb_ins);
+ return -ENOENT;
+ }
+
+ struct rte_bpf_validate_debug_point *const point =
+ rte_bpf_validate_debug_break(debug, pc,
+ &(struct rte_bpf_validate_debug_callback){
+ .fn = point_cb,
+ .ctx = (void *)(uintptr_t)point_number,
+ });
+ if (point == NULL) {
+ PRINTLN("Library error %d.", rte_errno);
+ return -rte_errno;
+ }
+
+ point_infos.elements[point_number] = (struct point_info){
+ .point = point,
+ .type = POINT_TYPE_BREAK,
+ .pc = pc,
+ };
+ point_infos_print_at(point_number);
+ return 0;
+}
+
+/* Create new catchpoint at specified location. */
+static int
+add_catchpoint(struct rte_bpf_validate_debug *debug, int event)
+{
+ if (event < 0)
+ /* Error was already printed by parse_event_name. */
+ return event;
+
+ const uint32_t point_number = point_infos_append();
+ struct rte_bpf_validate_debug_point *const point =
+ rte_bpf_validate_debug_catch(debug, event,
+ &(struct rte_bpf_validate_debug_callback){
+ .fn = point_cb,
+ .ctx = (void *)(uintptr_t)point_number,
+ });
+ if (point == NULL) {
+ PRINTLN("Library error %d.", rte_errno);
+ return -rte_errno;
+ }
+
+ point_infos.elements[point_number] = (struct point_info){
+ .point = point,
+ .type = POINT_TYPE_CATCH,
+ .event = event,
+ };
+ point_infos_print_at(point_number);
+ return 0;
+}
+
+static bool
+is_step_enabled(void)
+{
+ return step_point != NULL;
+}
+
+/* Enable step-by-step validation: make sure catchpoint is set on step event. */
+static int
+enable_step(struct rte_bpf_validate_debug *debug)
+{
+ if (is_step_enabled())
+ return 0;
+
+ step_point = rte_bpf_validate_debug_catch(debug,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_STEP,
+ &(struct rte_bpf_validate_debug_callback){ step_cb });
+ if (step_point == NULL) {
+ PRINTLN("Library error %d.", rte_errno);
+ return -rte_errno;
+ }
+
+ return 0;
+}
+
+/* Disable step-by-step validation: destroy catchpoint on step event if any. */
+static void
+disable_step(void)
+{
+ rte_bpf_validate_debug_point_destroy(step_point);
+ step_point = NULL;
+}
+
+/* Format and print information about specified frame offset. */
+static int
+print_frame_offset(struct rte_bpf_validate_debug *debug, int32_t offset)
+{
+ char *info;
+ int info_size, rc;
+
+ if (offset >= 0 || offset % sizeof(uint64_t) != 0) {
+ PRINTLN("Invalid frame offset, must be a negative multiple of %zu.",
+ sizeof(uint64_t));
+ return -EINVAL;
+ }
+
+ rc = rte_bpf_validate_debug_format_frame_info(debug, NULL, 0, offset);
+ if (rc == -ERANGE) {
+ PRINTLN("Offset is out of frame range.");
+ return rc;
+ }
+ if (rc < 0) {
+ PRINTLN("Error %d printing information.", -rc);
+ return rc;
+ }
+
+ info_size = rc + 1;
+ info = malloc(info_size);
+ if (info == NULL)
+ return -ENOMEM;
+
+ rc = rte_bpf_validate_debug_format_frame_info(debug, info, info_size,
+ offset);
+ if (rc + 1 != info_size) {
+ if (rc >= 0) {
+ PRINTLN("Expect format return value %d, got %d.",
+ info_size, rc);
+ rc = -EINVAL;
+ } else
+ PRINTLN("Error %d printing information.", -rc);
+ free(info);
+ return rc;
+ }
+
+ printf("%5jd: \t%s\n", (intmax_t)offset, info);
+ free(info);
+ return 0;
+}
+
+/* Format and print informatiion about the frame. */
+static int
+print_frame(struct rte_bpf_validate_debug *debug)
+{
+ int32_t frame_size;
+ int rc;
+
+ frame_size = rte_bpf_validate_debug_get_frame_size(debug);
+ if (frame_size < 0) {
+ PRINTLN("Error %d getting frame size.", -frame_size);
+ return frame_size;
+ }
+
+ for (int32_t frame_offset = 0;;) {
+ frame_offset -= sizeof(uint64_t);
+ if (frame_offset < -frame_size)
+ break;
+
+ rc = print_frame_offset(debug, frame_offset);
+ if (rc < 0)
+ return rc;
+ }
+
+ return 0;
+}
+
+/* Format and print informatiion about specified register. */
+static int
+print_register(struct rte_bpf_validate_debug *debug, int reg)
+{
+ char *info;
+ int info_size, rc;
+
+ if (reg < 0)
+ /* Error was already printed by parse_register. */
+ return reg;
+
+ rc = rte_bpf_validate_debug_format_register_info(debug, NULL, 0, reg);
+ if (rc < 0) {
+ PRINTLN("Error %d printing information.", -rc);
+ return rc;
+ }
+
+ info_size = rc + 1;
+ info = malloc(info_size);
+ if (info == NULL)
+ return -ENOMEM;
+
+ rc = rte_bpf_validate_debug_format_register_info(debug, info, info_size,
+ reg);
+ if (rc + 1 != info_size) {
+ if (rc >= 0) {
+ PRINTLN("Expect format return value %d, got %d.",
+ info_size, rc);
+ rc = -EINVAL;
+ } else
+ PRINTLN("Error %d printing information.", -rc);
+ free(info);
+ return rc;
+ }
+
+ printf("%5s: \t%s\n", register_names[reg], info);
+ free(info);
+ return 0;
+}
+
+/* Format and print informatiion about all registers. */
+static int
+print_registers(struct rte_bpf_validate_debug *debug)
+{
+ int rc = 0;
+
+ for (int reg = 0; reg != EBPF_REG_NUM; ++reg)
+ rc = rc < 0 ? rc : print_register(debug, reg);
+
+ return rc;
+}
+
+/* List one eBPF program instruction. */
+static int
+list_one(const struct ebpf_insn *ins, uint32_t nb_ins, uint32_t offset,
+ uint32_t pc, const char *comment)
+{
+ char hexadecimal[256], disassembly[256];
+
+ if (offset >= nb_ins) {
+ PRINTLN("Error: program only has %u instructions.", nb_ins);
+ return -EINVAL;
+ }
+
+ ins += offset;
+
+ if (offset == nb_ins - 1 && rte_bpf_insn_is_wide(ins)) {
+ PRINTLN("Error: truncated last instruction.");
+ return -EINVAL;
+ }
+
+ rte_bpf_format(hexadecimal, sizeof(hexadecimal), ins, 0,
+ RTE_BPF_FORMAT_FLAG_HEXADECIMAL |
+ RTE_BPF_FORMAT_FLAG_NEVER_WIDE);
+ rte_bpf_format(disassembly, sizeof(disassembly), ins, offset,
+ RTE_BPF_FORMAT_FLAG_DISASSEMBLY |
+ RTE_BPF_FORMAT_FLAG_ABSOLUTE_JUMPS);
+
+ if (comment == NULL)
+ comment = "";
+ PRINTLN("%2s %10u: \t%s \t%s%s%s",
+ offset == pc ? "=>" : "", offset, hexadecimal, disassembly,
+ comment[0] != '\0' ? " \t; " : "", comment);
+
+ if (rte_bpf_insn_is_wide(ins)) {
+ rte_bpf_format(hexadecimal, sizeof(hexadecimal), ins + 1, 0,
+ RTE_BPF_FORMAT_FLAG_HEXADECIMAL |
+ RTE_BPF_FORMAT_FLAG_NEVER_WIDE);
+ PRINTLN("%15s\t%s", "", hexadecimal);
+ }
+
+ return 0;
+}
+
+/* List specified range of eBPF program instructions, updating start offset. */
+static int
+list(const struct ebpf_insn *ins, uint32_t nb_ins, uint32_t *offset,
+ uint32_t count, uint32_t pc)
+{
+ uint32_t local_offset = 0;
+ if (offset == NULL)
+ offset = &local_offset;
+
+ if (*offset > nb_ins) {
+ PRINTLN("Error: program only has %u instructions.", nb_ins);
+ return -EINVAL;
+ }
+
+ const uint32_t end =
+ /* Calculate end in a way preventing overflow: */
+ *offset + RTE_MIN(nb_ins - *offset, count);
+ while (*offset < end) {
+ const int rc = list_one(ins, nb_ins, *offset, pc, NULL);
+ if (rc < 0)
+ return rc;
+
+ *offset += 1 + rte_bpf_insn_is_wide(&ins[*offset]);
+ }
+
+ return 0;
+}
+
+/* Print if specified conditional jump _may_ be executed. */
+static int
+print_if_may(struct rte_bpf_validate_debug *debug, const struct ebpf_insn *jump,
+ uint64_t imm64)
+{
+ const int result = rte_bpf_validate_debug_may_jump(debug, jump, imm64);
+
+ switch (result) {
+ case 0:
+ case RTE_BPF_VALIDATE_DEBUG_MAY_BE_FALSE:
+ PRINTLN("NO");
+ break;
+ case RTE_BPF_VALIDATE_DEBUG_MAY_BE_TRUE:
+ case RTE_BPF_VALIDATE_DEBUG_MAY_BE_FALSE | RTE_BPF_VALIDATE_DEBUG_MAY_BE_TRUE:
+ PRINTLN("YES");
+ break;
+ default:
+ PRINTLN("Error %d getting result.", -result);
+ break;
+ }
+
+ return result;
+}
+
+/* Print if specified condition with literal right hand side _may_ be true. */
+static int
+print_if_may_literal_rhs(struct rte_bpf_validate_debug *debug,
+ const struct debug_command_comparison *comparison)
+{
+ const int lhs = parse_register(comparison->lhs);
+ const int op = parse_comparison_operator(comparison->op);
+ const int64_t rhs = comparison->literal_rhs;
+
+ if (lhs < 0 || op < 0)
+ /* Error was already printed by parse function. */
+ return lhs < 0 ? lhs : op;
+
+ return print_if_may(debug, &(struct ebpf_insn){
+ .code = BPF_JMP | op | BPF_K,
+ .dst_reg = lhs,
+ }, /* imm64 = */ rhs);
+}
+
+/* Print if specified condition with register right hand side _may_ be true. */
+static int
+print_if_may_register_rhs(struct rte_bpf_validate_debug *debug,
+ const struct debug_command_comparison *comparison)
+{
+ const int lhs = parse_register(comparison->lhs);
+ const int op = parse_comparison_operator(comparison->op);
+ const int rhs = parse_register(comparison->register_rhs);
+
+ if (lhs < 0 || op < 0 || rhs < 0)
+ /* Error was already printed by parse function. */
+ return lhs < 0 ? lhs : op < 0 ? op : rhs;
+
+ return print_if_may(debug, &(struct ebpf_insn){
+ .code = BPF_JMP | op | BPF_X,
+ .dst_reg = lhs,
+ .src_reg = rhs,
+ }, /* imm64 = */ 0);
+}
+
+/* Return 1 on validation success, 0 on failure, -EAGAIN if still running. */
+static int
+get_validation_success(struct rte_bpf_validate_debug *debug)
+{
+ int validation_result, rc;
+
+ rc = rte_bpf_validate_debug_get_validation_result(debug,
+ &validation_result);
+ return rc < 0 ? rc : (validation_result >= 0);
+}
+
+static void
+print_status(const struct ebpf_insn *ins, uint32_t nb_ins,
+ int validation_success, uint32_t pc)
+{
+ if (validation_success == 1) {
+ PRINTLN("Validation succeeded.");
+ return;
+ }
+
+ list_one(ins, nb_ins, pc, pc, NULL);
+
+ if (validation_success == 0)
+ PRINTLN("Validation failed.");
+}
+
+static void
+debug_command_where(const struct ebpf_insn *ins, uint32_t nb_ins,
+ int validation_success, uint32_t pc)
+{
+ for (uint32_t bi = 0; bi != branch_stack.length; ++bi) {
+ const uint32_t jump_pc = branch_stack.branches[bi].jump_pc;
+ const uint32_t target_pc = branch_stack.branches[bi].target_pc;
+ const char *const comment =
+ !branch_stack.branches[bi].is_conditional ? NULL :
+ target_pc == jump_pc + 1 ? "fallen-through" : "taken";
+ list_one(ins, nb_ins, jump_pc, UINT32_MAX, comment);
+ }
+ print_status(ins, nb_ins, validation_success, pc);
+}
+
+/* Return true if pc is defined, otherwise print an error message. */
+static bool
+ensure_pc_defined(uint32_t pc)
+{
+ if (pc != UINT32_MAX)
+ return true;
+
+ PRINTLN("No current instruction.");
+ return false;
+}
+
+/* Return true if still validating, otherwise print an error message. */
+static bool
+ensure_still_validating(int validation_success)
+{
+ if (validation_success == -EAGAIN)
+ return true;
+
+ PRINTLN("Finished, use `start` or `run` to restart, `quit` to quit.");
+ return false;
+}
+
+/* Step-by-step validation callback: read and process user commands. */
+static int
+step_cb(struct rte_bpf_validate_debug *debug, __rte_unused void *ctx)
+{
+ int rc;
+ int validation_success;
+ const struct ebpf_insn *ins;
+ uint32_t nb_ins, pc, list_offset;
+
+ validate_again = false;
+
+ rc = rte_bpf_validate_debug_get_ins(debug, &ins, &nb_ins);
+ if (rc < 0) {
+ PRINTLN("Error %d getting program instructions.", -rc);
+ return rc;
+ }
+
+ validation_success = get_validation_success(debug);
+ if (validation_success < 0 && validation_success != -EAGAIN) {
+ PRINTLN("Error %d getting validation result.",
+ -validation_success);
+ return validation_success;
+ }
+
+ pc = validation_success == 1 ? UINT32_MAX :
+ rte_bpf_validate_debug_get_pc(debug);
+
+ print_status(ins, nb_ins, validation_success, pc);
+
+ list_offset = pc;
+
+ while (true) {
+ switch (debug_command_get(PROMPT)) {
+ case DEBUG_COMMAND_BREAK:
+ if (ensure_pc_defined(pc))
+ add_breakpoint(debug, nb_ins, pc);
+ continue;
+ case DEBUG_COMMAND_BREAK_PC:
+ add_breakpoint(debug, nb_ins, debug_command_parsed.pc);
+ continue;
+ case DEBUG_COMMAND_CATCH:
+ add_catchpoint(debug,
+ parse_event_name(debug_command_parsed.event));
+ continue;
+ case DEBUG_COMMAND_CLEAR:
+ if (ensure_pc_defined(pc))
+ point_infos_destroy_breakpoints(pc);
+ continue;
+ case DEBUG_COMMAND_CLEAR_EVENT:
+ point_infos_destroy_catchpoints(
+ parse_event_name(debug_command_parsed.event));
+ continue;
+ case DEBUG_COMMAND_CLEAR_PC:
+ point_infos_destroy_breakpoints(
+ debug_command_parsed.pc);
+ continue;
+ case DEBUG_COMMAND_CONTINUE:
+ if (ensure_still_validating(validation_success)) {
+ disable_step();
+ return 0;
+ }
+ continue;
+ case DEBUG_COMMAND_DELETE:
+ point_infos_destroy_all();
+ continue;
+ case DEBUG_COMMAND_DELETE_NUMBER:
+ point_infos_destroy_at(
+ debug_command_parsed.point_number);
+ continue;
+ case DEBUG_COMMAND_INFO_FRAME:
+ print_frame(debug);
+ continue;
+ case DEBUG_COMMAND_INFO_FRAME_OFFSET:
+ print_frame_offset(debug,
+ debug_command_parsed.frame_offset);
+ continue;
+ case DEBUG_COMMAND_INFO_POINTS:
+ point_infos_print_all();
+ continue;
+ case DEBUG_COMMAND_INFO_REGISTER:
+ print_register(debug,
+ parse_register(debug_command_parsed.register_));
+ continue;
+ case DEBUG_COMMAND_INFO_REGISTERS:
+ print_registers(debug);
+ continue;
+ case DEBUG_COMMAND_LIST:
+ if (ensure_pc_defined(pc))
+ list(ins, nb_ins, &list_offset, 10, pc);
+ continue;
+ case DEBUG_COMMAND_LIST_COUNT:
+ if (ensure_pc_defined(pc))
+ list(ins, nb_ins, &list_offset,
+ debug_command_parsed.instruction_count, pc);
+ continue;
+ case DEBUG_COMMAND_LIST_PROGRAM:
+ if (ensure_pc_defined(pc))
+ list(ins, nb_ins, NULL, UINT32_MAX, pc);
+ continue;
+ case DEBUG_COMMAND_MAY_LITERAL_RHS:
+ print_if_may_literal_rhs(debug,
+ &debug_command_parsed.comparison);
+ continue;
+ case DEBUG_COMMAND_MAY_REGISTER_RHS:
+ print_if_may_register_rhs(debug,
+ &debug_command_parsed.comparison);
+ continue;
+ case DEBUG_COMMAND_EOF:
+ case DEBUG_COMMAND_QUIT:
+ PRINTLN("Quitting...");
+ return validation_success == -EAGAIN ? -ECANCELED : 0;
+ case DEBUG_COMMAND_RUN:
+ PRINTLN("Re-running...");
+ disable_step();
+ validate_again = true;
+ return -ECANCELED;
+ case DEBUG_COMMAND_START:
+ PRINTLN("Re-starting...");
+ rc = enable_step(debug);
+ if (rc < 0)
+ return rc;
+ validate_again = true;
+ return -ECANCELED;
+ case DEBUG_COMMAND_STEP:
+ if (ensure_still_validating(validation_success))
+ return 0;
+ continue;
+ case DEBUG_COMMAND_WHERE:
+ debug_command_where(ins, nb_ins, validation_success, pc);
+ continue;
+ default:
+ PRINTLN("INTERNAL ERROR");
+ return -ENOTSUP;
+ }
+ }
+}
+
+/* Any point callback: print it and enable step-by-step validation. */
+static int
+point_cb(struct rte_bpf_validate_debug *debug, void *ctx)
+{
+ const uint32_t point_number = (uintptr_t)ctx;
+ point_infos_print_at(point_number);
+ return enable_step(debug);
+}
+
+/* Branch stack machinery. */
+
+static void
+clear_jump_always_step_point(void)
+{
+ rte_bpf_validate_debug_point_destroy(jump_always_step_point);
+ jump_always_step_point = NULL;
+}
+
+static int
+reset_branch_tracking(struct rte_bpf_validate_debug *debug __rte_unused,
+ void *ctx __rte_unused)
+{
+ branch_stack.length = 0;
+ pending_jump_pc = UINT32_MAX;
+ clear_jump_always_step_point();
+ return 0;
+}
+
+/*
+ * Handling the jump-always instructions:
+ * - upon a jump-always event save the pc and set a custom step callback;
+ * - ignore the first custom step callback call (still on the same jump);
+ * - on the second call push the jump pc into stack and delete the callback;
+ */
+
+static int
+jump_always_step_cb(struct rte_bpf_validate_debug *debug, void *ctx __rte_unused)
+{
+ /* Step event is also emitted at the end of the jump instruction itself. */
+ if (rte_bpf_validate_debug_get_pc(debug) == pending_jump_pc)
+ return 0;
+
+ branch_stack_append(&(struct branch_info){
+ .jump_pc = pending_jump_pc,
+ .target_pc = rte_bpf_validate_debug_get_pc(debug),
+ .is_conditional = false,
+ });
+ clear_jump_always_step_point();
+ return 0;
+}
+
+static int
+jump_always_cb(struct rte_bpf_validate_debug *debug, void *ctx __rte_unused)
+{
+ pending_jump_pc = rte_bpf_validate_debug_get_pc(debug);
+ clear_jump_always_step_point();
+ jump_always_step_point = rte_bpf_validate_debug_catch(debug,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_STEP,
+ &(struct rte_bpf_validate_debug_callback){ jump_always_step_cb });
+ return 0;
+}
+
+/*
+ * Handling the jump-conditional instructions:
+ * - upon a jump-conditional event save the pc;
+ * - upon a branch-enter event push the conditional jump pc into stack;
+ * - upon a branch-return event pop conditional jump and all unconditional jumps
+ * preceding it from the stack;
+ * - during step execution notify the user about the step events;
+ */
+
+static int
+jump_conditional_cb(struct rte_bpf_validate_debug *debug, void *ctx __rte_unused)
+{
+ pending_jump_pc = rte_bpf_validate_debug_get_pc(debug);
+ return 0;
+}
+
+static int
+branch_enter_cb(struct rte_bpf_validate_debug *debug, void *ctx __rte_unused)
+{
+ branch_stack_append(&(struct branch_info){
+ .jump_pc = pending_jump_pc,
+ .target_pc = rte_bpf_validate_debug_get_pc(debug),
+ .is_conditional = true,
+ });
+ if (is_step_enabled())
+ PRINTLN("Entered new branch at pc %u.", pending_jump_pc);
+ return 0;
+}
+
+static int
+branch_return_cb(struct rte_bpf_validate_debug *debug __rte_unused,
+ void *ctx __rte_unused)
+{
+ clear_jump_always_step_point();
+
+ while (branch_stack.length > 0) {
+ const struct branch_info branch_info =
+ branch_stack.branches[--branch_stack.length];
+ pending_jump_pc = branch_info.jump_pc;
+ if (branch_info.is_conditional)
+ break;
+ }
+ if (is_step_enabled())
+ PRINTLN("Returned from branch at pc %u.", pending_jump_pc);
+ return 0;
+}
+
+/* Notify user when skipping branches in step mode. */
+
+static int
+branch_prune_cb(struct rte_bpf_validate_debug *debug __rte_unused,
+ void *ctx __rte_unused)
+{
+ if (is_step_enabled())
+ PRINTLN("Prunned branch at pc %u.", pending_jump_pc);
+ return 0;
+}
+
+static int
+branch_unreachable_cb(struct rte_bpf_validate_debug *debug __rte_unused,
+ void *ctx __rte_unused)
+{
+ if (is_step_enabled())
+ PRINTLN("Unreachable branch at pc %u.", pending_jump_pc);
+ return 0;
+}
+
+/* Global initialization and cleanup functions. */
+
+static int
+set_callbacks(struct rte_bpf_validate_debug *debug)
+{
+ static const struct rte_bpf_validate_debug_callback events_callback[
+ RTE_BPF_VALIDATE_DEBUG_EVENT_END] = {
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_START] = { reset_branch_tracking },
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS] = { step_cb },
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_FAILURE] = { step_cb },
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL] = { jump_conditional_cb },
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS] = { jump_always_cb },
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER] = { branch_enter_cb },
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN] = { branch_return_cb },
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE] = { branch_prune_cb },
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE] = { branch_unreachable_cb },
+ };
+
+ for (enum rte_bpf_validate_debug_event event = 0;
+ event < RTE_BPF_VALIDATE_DEBUG_EVENT_END; ++event)
+ if (events_callback[event].fn != NULL &&
+ rte_bpf_validate_debug_catch(debug, event,
+ &events_callback[event]) == NULL)
+ return -rte_errno;
+
+ return 0;
+}
+
+struct rte_bpf_validate_debug *
+debug_create(void)
+{
+
+ int rc = 0;
+
+ struct rte_bpf_validate_debug *const debug = rte_bpf_validate_debug_create();
+ if (debug == NULL)
+ rc = -rte_errno;
+
+ rc = rc < 0 ? rc : set_callbacks(debug);
+
+ rc = rc < 0 ? rc : enable_step(debug);
+
+ if (rc < 0) {
+ debug_destroy(debug);
+ rte_errno = -rc;
+ return NULL;
+ }
+
+ return debug;
+}
+
+void
+debug_destroy(struct rte_bpf_validate_debug *debug)
+{
+ /* No need to destroy created points, destroying debug will do it. */
+ point_infos_free();
+ branch_stack_free();
+ rte_bpf_validate_debug_destroy(debug);
+}
+
+bool
+debug_validate_again(void)
+{
+ return validate_again;
+}
diff --git a/app/validate-bpf/debug_command.c b/app/validate-bpf/debug_command.c
new file mode 100644
index 000000000000..505ae389c94f
--- /dev/null
+++ b/app/validate-bpf/debug_command.c
@@ -0,0 +1,383 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include "debug_command.h"
+
+#include <rte_debug.h>
+#include <cmdline.h>
+#include <cmdline_socket.h>
+#include <cmdline_parse_num.h>
+
+#include <string.h>
+
+
+#define EVENT_PATTERN \
+ "invalid-state#" \
+ "branch-enter#branch-prune#branch-return#branch-unreachable#" \
+ "jump-always#jump-conditional"
+
+#define REGISTER_PATTERN "r0#r1#r2#r3#r4#r5#r6#r7#r8#r9#r10"
+
+#define COMPARISON_OP_PATTERN "==#!=#<#<=#>#>=#s<#s<=#s>#s>="
+
+static void
+handle_command(void *parsed_result, struct cmdline *cl, void *data);
+
+/* Keywords */
+static cmdline_parse_token_string_t break_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "b#break");
+static cmdline_parse_token_string_t may_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "may");
+static cmdline_parse_token_string_t catch_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "catch");
+static cmdline_parse_token_string_t clear_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "clear");
+static cmdline_parse_token_string_t continue_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "c#continue");
+static cmdline_parse_token_string_t delete_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "delete");
+static cmdline_parse_token_string_t info_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "i#info");
+static cmdline_parse_token_string_t info_points_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword,
+ "b#break#breakpoints#points");
+static cmdline_parse_token_string_t info_frame_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "f#frame");
+static cmdline_parse_token_string_t info_registers_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "r#registers");
+static cmdline_parse_token_string_t list_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "l#list");
+static cmdline_parse_token_string_t program_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "program");
+static cmdline_parse_token_string_t quit_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "q#quit");
+static cmdline_parse_token_string_t run_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "run");
+static cmdline_parse_token_string_t start_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "start");
+static cmdline_parse_token_string_t step_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "s#step");
+static cmdline_parse_token_string_t where_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "where");
+
+/* Variable tokens */
+static cmdline_parse_token_string_t comparison_lhs_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, comparison.lhs, REGISTER_PATTERN);
+static cmdline_parse_token_string_t comparison_op_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, comparison.op, COMPARISON_OP_PATTERN);
+static cmdline_parse_token_num_t comparison_rhs_literal_tok =
+ TOKEN_NUM_INITIALIZER(struct debug_command_parsed, comparison.literal_rhs, RTE_INT64);
+static cmdline_parse_token_string_t comparison_rhs_register_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, comparison.register_rhs,
+ REGISTER_PATTERN);
+static cmdline_parse_token_string_t event_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, event, EVENT_PATTERN);
+static cmdline_parse_token_num_t frame_offset_tok =
+ TOKEN_NUM_INITIALIZER(struct debug_command_parsed, frame_offset, RTE_INT64);
+static cmdline_parse_token_num_t instruction_count_tok =
+ TOKEN_NUM_INITIALIZER(struct debug_command_parsed, instruction_count, RTE_UINT32);
+static cmdline_parse_token_num_t pc_tok =
+ TOKEN_NUM_INITIALIZER(struct debug_command_parsed, instruction_offset, RTE_UINT32);
+static cmdline_parse_token_num_t point_number_tok =
+ TOKEN_NUM_INITIALIZER(struct debug_command_parsed, point_number, RTE_UINT32);
+static cmdline_parse_token_string_t register_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, register_, REGISTER_PATTERN);
+
+
+/* Commands */
+static cmdline_parse_inst_t cmd_break = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_BREAK,
+ .help_str = "b|break: break at current instruction",
+ .tokens = {
+ (void *)&break_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_break_pc = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_BREAK_PC,
+ .help_str = "b|break <pc>: break at specified instruction",
+ .tokens = {
+ (void *)&break_tok,
+ (void *)&pc_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_catch = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_CATCH,
+ .help_str = "catch <event>: catch specified event",
+ .tokens = {
+ (void *)&catch_tok,
+ (void *)&event_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_clear = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_CLEAR,
+ .help_str = "clear: delete all breakpoints at current instruction",
+ .tokens = {
+ (void *)&clear_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_clear_event = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_CLEAR_EVENT,
+ .help_str = "clear <event>: delete all catchpoints for specified event",
+ .tokens = {
+ (void *)&clear_tok,
+ (void *)&event_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_clear_pc = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_CLEAR_PC,
+ .help_str = "clear <pc>: delete all breakpoints at specified instruction",
+ .tokens = {
+ (void *)&clear_tok,
+ (void *)&pc_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_continue = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_CONTINUE,
+ .help_str = "c|continue: continue validation",
+ .tokens = {
+ (void *)&continue_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_delete = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_DELETE,
+ .help_str = "delete: delete all breakpoints and catchpoints",
+ .tokens = {
+ (void *)&delete_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_delete_number = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_DELETE_NUMBER,
+ .help_str = "delete <point>: delete specified breakpoint or catchpoint",
+ .tokens = {
+ (void *)&delete_tok,
+ (void *)&point_number_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_info_frame = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_INFO_FRAME,
+ .help_str = "i|info f|frame: show information about all frame locations",
+ .tokens = {
+ (void *)&info_tok,
+ (void *)&info_frame_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_info_frame_offset = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_INFO_FRAME_OFFSET,
+ .help_str = "i|info f|frame -<offset>: show information about specified frame location",
+ .tokens = {
+ (void *)&info_tok,
+ (void *)&info_frame_tok,
+ (void *)&frame_offset_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_info_points = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_INFO_POINTS,
+ .help_str = "i|info b|break|breakpoints|points: "
+ "show information about all breakpoints and catchpoints",
+ .tokens = {
+ (void *)&info_tok,
+ (void *)&info_points_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_info_register = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_INFO_REGISTER,
+ .help_str = "i|info <register>: show information about specified register",
+ .tokens = {
+ (void *)&info_tok,
+ (void *)®ister_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_info_registers = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_INFO_REGISTERS,
+ .help_str = "i|info r|registers: show information about all registers",
+ .tokens = {
+ (void *)&info_tok,
+ (void *)&info_registers_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_list = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_LIST,
+ .help_str = "l|list: list ten instructions",
+ .tokens = {
+ (void *)&list_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_list_count = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_LIST_COUNT,
+ .help_str = "l|list <number>: list specified number of instructions",
+ .tokens = {
+ (void *)&list_tok,
+ (void *)&instruction_count_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_list_program = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_LIST_PROGRAM,
+ .help_str = "l|list program: list whole program",
+ .tokens = {
+ (void *)&list_tok,
+ (void *)&program_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_may_literal_rhs = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_MAY_LITERAL_RHS,
+ .help_str = "may <register> <comparison> <number>: "
+ "check if specified condition _may_ be true",
+ .tokens = {
+ (void *)&may_tok,
+ (void *)&comparison_lhs_tok,
+ (void *)&comparison_op_tok,
+ (void *)&comparison_rhs_literal_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_may_register_rhs = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_MAY_REGISTER_RHS,
+ .help_str = "may <register> <comparison> <register>: "
+ "check if specified condition _may_ be true",
+ .tokens = {
+ (void *)&may_tok,
+ (void *)&comparison_lhs_tok,
+ (void *)&comparison_op_tok,
+ (void *)&comparison_rhs_register_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_quit = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_QUIT,
+ .help_str = "quit: q|quit debugger",
+ .tokens = {
+ (void *)&quit_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_run = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_RUN,
+ .help_str = "run: re-run validation from the start",
+ .tokens = {
+ (void *)&run_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_start = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_START,
+ .help_str = "start: re-start validation and stop at start",
+ .tokens = {
+ (void *)&start_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_step = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_STEP,
+ .help_str = "s|step: validate one instruction",
+ .tokens = {
+ (void *)&step_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_where = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_WHERE,
+ .help_str = "where: show current branch stack",
+ .tokens = {
+ (void *)&where_tok,
+ NULL,
+ }
+};
+
+static cmdline_parse_ctx_t debug_ctx[] = {
+ &cmd_break,
+ &cmd_break_pc,
+ &cmd_catch,
+ &cmd_clear,
+ &cmd_clear_event,
+ &cmd_clear_pc,
+ &cmd_continue,
+ &cmd_delete,
+ &cmd_delete_number,
+ &cmd_info_frame,
+ &cmd_info_frame_offset,
+ &cmd_info_points,
+ &cmd_info_register,
+ &cmd_info_registers,
+ &cmd_list,
+ &cmd_list_count,
+ &cmd_list_program,
+ &cmd_may_literal_rhs,
+ &cmd_may_register_rhs,
+ &cmd_quit,
+ &cmd_run,
+ &cmd_start,
+ &cmd_step,
+ &cmd_where,
+ NULL
+};
+
+/* Receive, fill and return one command. */
+
+static enum debug_command debug_command;
+
+struct debug_command_parsed debug_command_parsed;
+
+static void
+handle_command(void *parsed_result, struct cmdline *cl, void *data)
+{
+ RTE_BUILD_BUG_ON(sizeof(debug_command_parsed) > CMDLINE_PARSE_RESULT_BUFSIZE);
+ memcpy(&debug_command_parsed, parsed_result, sizeof(debug_command_parsed));
+ debug_command = (uintptr_t)data;
+ cmdline_quit(cl);
+}
+
+enum debug_command
+debug_command_get(const char *prompt)
+{
+ debug_command = DEBUG_COMMAND_EOF;
+ struct cmdline *const cmdline = cmdline_stdin_new(debug_ctx, prompt);
+ RTE_VERIFY(cmdline != NULL);
+ cmdline_interact(cmdline);
+ cmdline_stdin_exit(cmdline);
+ /* Clear prompt, or it would prepend first message that follows. */
+ printf("\r%*s\r", (int)strlen(prompt), "");
+ fflush(stdout);
+ return debug_command;
+}
diff --git a/app/validate-bpf/debug_command.h b/app/validate-bpf/debug_command.h
new file mode 100644
index 000000000000..1a59ab251ebb
--- /dev/null
+++ b/app/validate-bpf/debug_command.h
@@ -0,0 +1,65 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include <cmdline_parse_string.h>
+
+#include <stdint.h>
+
+
+struct debug_command_comparison {
+ cmdline_fixed_string_t lhs;
+ cmdline_fixed_string_t op;
+ union {
+ cmdline_fixed_string_t register_rhs;
+ int64_t literal_rhs;
+ };
+};
+
+struct debug_command_parsed {
+ cmdline_fixed_string_t keyword; /* Any keyword we don't need */
+ union {
+ struct debug_command_comparison comparison;
+ cmdline_fixed_string_t event;
+ int32_t frame_offset;
+ uint32_t instruction_count;
+ uint32_t instruction_offset;
+ uint32_t pc;
+ uint32_t point_number;
+ cmdline_fixed_string_t register_;
+ };
+};
+
+enum debug_command {
+ DEBUG_COMMAND_EOF,
+ DEBUG_COMMAND_BREAK,
+ DEBUG_COMMAND_BREAK_PC,
+ DEBUG_COMMAND_CATCH,
+ DEBUG_COMMAND_CLEAR,
+ DEBUG_COMMAND_CLEAR_EVENT,
+ DEBUG_COMMAND_CLEAR_PC,
+ DEBUG_COMMAND_CONTINUE,
+ DEBUG_COMMAND_DELETE,
+ DEBUG_COMMAND_DELETE_NUMBER,
+ DEBUG_COMMAND_INFO_FRAME,
+ DEBUG_COMMAND_INFO_FRAME_OFFSET,
+ DEBUG_COMMAND_INFO_POINTS,
+ DEBUG_COMMAND_INFO_REGISTER,
+ DEBUG_COMMAND_INFO_REGISTERS,
+ DEBUG_COMMAND_LIST,
+ DEBUG_COMMAND_LIST_COUNT,
+ DEBUG_COMMAND_LIST_PROGRAM,
+ DEBUG_COMMAND_MAY_LITERAL_RHS,
+ DEBUG_COMMAND_MAY_REGISTER_RHS,
+ DEBUG_COMMAND_QUIT,
+ DEBUG_COMMAND_RUN,
+ DEBUG_COMMAND_START,
+ DEBUG_COMMAND_STEP,
+ DEBUG_COMMAND_WHERE,
+};
+
+extern struct debug_command_parsed debug_command_parsed;
+
+/** Get and return one command line command, storing its data in the struct above. */
+enum debug_command
+debug_command_get(const char *prompt);
diff --git a/app/validate-bpf/eal_init_args.c b/app/validate-bpf/eal_init_args.c
new file mode 100644
index 000000000000..6ce9d561e0ce
--- /dev/null
+++ b/app/validate-bpf/eal_init_args.c
@@ -0,0 +1,57 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include "internal.h"
+
+
+#define APP_EAL_INIT_ARG_SIZE_MAX sizeof("--log-level=lib.eal:warning")
+
+static const char APP_EAL_INIT_ARGS[][APP_EAL_INIT_ARG_SIZE_MAX] = {
+ "--log-level=lib.eal:warning",
+ "--no-huge",
+ "--no-pci",
+ "--no-hpet",
+ "--no-shconf",
+};
+
+#define APP_EAL_INIT_ARGC (/* program name */ 1 + RTE_DIM(APP_EAL_INIT_ARGS))
+
+int
+get_eal_init_argc(void)
+{
+ return APP_EAL_INIT_ARGC;
+}
+
+/*
+ * We cannot just return literal strings here, because rte_eal_init accepts
+ * an array of mutable pointers to mutable strings, so literals won't work.
+ * Instead we build a copy of APP_EAL_INIT_ARGS in a static mutable area.
+ */
+char**
+get_eal_init_argv(char *prog_name)
+{
+ /* Static arrays for mutable copies of actual args */
+ static char mutable_args[RTE_DIM(APP_EAL_INIT_ARGS)]
+ [APP_EAL_INIT_ARG_SIZE_MAX];
+ /*
+ * Static array for pointers to args. First element will hold pointer
+ * to the prog_name, last will be set to NULL, the rest will point to
+ * elements of mutable_args with index one smaller.
+ */
+ static char *mutable_ptrs[APP_EAL_INIT_ARGC + /* terminating NULL */ 1];
+
+ mutable_ptrs[0] = prog_name;
+ for (int argi = 0; argi != RTE_DIM(APP_EAL_INIT_ARGS); ++argi) {
+ const char * const const_arg = APP_EAL_INIT_ARGS[argi];
+ char * const mutable_arg = mutable_args[argi];
+ const int snprintf_rc = snprintf(mutable_arg,
+ APP_EAL_INIT_ARG_SIZE_MAX, "%s", const_arg);
+ RTE_VERIFY(snprintf_rc >= 0 &&
+ (size_t)snprintf_rc < APP_EAL_INIT_ARG_SIZE_MAX);
+ mutable_ptrs[1 + argi] = mutable_arg;
+ }
+ mutable_ptrs[RTE_DIM(mutable_ptrs) - 1] = NULL;
+
+ return mutable_ptrs;
+}
diff --git a/app/validate-bpf/internal.h b/app/validate-bpf/internal.h
new file mode 100644
index 000000000000..e134e683142a
--- /dev/null
+++ b/app/validate-bpf/internal.h
@@ -0,0 +1,151 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include <rte_bpf.h>
+#include <rte_log.h>
+
+
+extern int validate_bpf_logtype;
+#define RTE_LOGTYPE_VALIDATE_BPF validate_bpf_logtype
+#define VALIDATE_BPF_LOG(level, ...) \
+ RTE_LOG_LINE(level, VALIDATE_BPF, "" __VA_ARGS__)
+
+struct rte_bpf_validate_debug;
+
+/**
+ * List of pointers to allocations, to keep track of things to free.
+ *
+ * May contain NULLs. May contain itself directly or via an outer struct.
+ */
+struct alloc_list {
+ size_t count;
+ void **ptrs;
+};
+
+/** Add new pointer to the alloc list, return its index. */
+size_t alloc_list_append(struct alloc_list *alloc_list, void *ptr);
+
+/** Replace pointer at the specified index in the alloc list. */
+void alloc_list_replace(struct alloc_list *alloc_list, size_t index, void *ptr);
+
+/** Free all allocations in the alloc list */
+void alloc_list_free_all(struct alloc_list *alloc_list);
+
+
+/** Parsed program arguments */
+struct args {
+ /* Allocations list, args.c internal use only. */
+ struct alloc_list _alloc_list;
+
+ /* Set if command line contains --help */
+ bool show_help;
+
+ /* BPF load parameters. */
+ struct rte_bpf_prm_ex bpf_prm;
+
+ /* Size of the mbuf data buffer. */
+ size_t mbuf_buf_size;
+};
+
+/** Print program usage information */
+void
+print_usage(const char *program_name);
+
+/** Print program defaults. */
+void
+print_defaults(void);
+
+/**
+ * Parse command-line arguments
+ *
+ * @param argc
+ * Command-line arguments count, as received by main.
+ * @param argv
+ * Command-line arguments array, as received by main.
+ * Modified during the call due to the use of getopt_long.
+ * Modifying it after the call invalidates the return value.
+ * @return
+ * - parse results in case of success;
+ * - NULL in case of an error (diagnostic will be printed to stderr)
+ */
+struct args *
+args_parse(int argc, char *argv[]);
+
+/** Destroy args struct returned by args_parse */
+void
+args_destroy(struct args *args);
+
+
+/** Value for the rte_eal_init argc argument */
+int
+get_eal_init_argc(void);
+
+/**
+ * Value for the rte_eal_init argv argument
+ *
+ * @param prog_name
+ * Program name, can be obtained from argv[0]
+ */
+char **
+get_eal_init_argv(char *prog_name);
+
+
+/** Print types supported in command-line. */
+void
+print_supported_types(void);
+
+/**
+ * Parse text into arg.
+ * @param arg
+ * Pointer to a variable to put parse result to.
+ * Member `buf_size` of types related to `struct rte_mbuf` is set to
+ * `RTE_MBUF_DEFAULT_BUF_SIZE`, adjust using `adjust_arg_buf_size` if needed.
+ * @param text
+ * Text representation of the type, e.g. "struct rte_mbuf *".
+ * @param alloc_list
+ * Alloc list to use for new allocations.
+ * @return
+ * 0 on success
+ * -1 on failure
+ */
+int
+parse_arg(struct rte_bpf_arg *arg, const char *text);
+
+/**
+ * Parse text into xsym.
+ * @param arg
+ * Pointer to a variable to put parse result to.
+ * Member `buf_size` of types related to `struct rte_mbuf` is set to
+ * `RTE_MBUF_DEFAULT_BUF_SIZE`, adjust using `adjust_xsym_buf_size` if needed.
+ * @param text
+ * Text representation of the external symbol, e.g. "void exit(uint32_t)".
+ * @param alloc_list
+ * Alloc list to use for new allocations.
+ * @return
+ * 0 on success
+ * -1 on failure
+ */
+int
+parse_xsym(struct rte_bpf_xsym *xsym, const char *text,
+ struct alloc_list *alloc_list);
+
+/** If arg->buf_size is non-zero, set it to mbuf_buf_size */
+void
+adjust_arg_buf_size(struct rte_bpf_arg *arg, size_t mbuf_buf_size);
+
+/** If buf_size members are non-zero, set them to mbuf_buf_size */
+void
+adjust_xsym_buf_size(struct rte_bpf_xsym *xsym, size_t mbuf_buf_size);
+
+/** Create and set up global debugging session. */
+struct rte_bpf_validate_debug *
+debug_create(void);
+
+/** Clear and destroy global debugging session. */
+void
+debug_destroy(struct rte_bpf_validate_debug *debug);
+
+/** Tells caller if validation should be re-tried. */
+bool
+debug_validate_again(void);
diff --git a/app/validate-bpf/main.c b/app/validate-bpf/main.c
new file mode 100644
index 000000000000..c2f73993767b
--- /dev/null
+++ b/app/validate-bpf/main.c
@@ -0,0 +1,90 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include "internal.h"
+
+#include <stdio.h>
+#include <stdlib.h>
+
+#include <rte_bpf.h>
+#include <rte_debug.h>
+#include <rte_eal.h>
+#include <rte_errno.h>
+
+RTE_LOG_REGISTER(validate_bpf_logtype, validate-bpf, NOTICE);
+
+static int
+test_bpf_load(struct rte_bpf_prm_ex *prm)
+{
+ struct rte_bpf * const bpf = rte_bpf_load_ex(prm);
+
+ const int rc = -rte_errno;
+
+ rte_bpf_destroy(bpf);
+
+ return bpf == NULL ? rc : 0;
+}
+
+/* Re-starts validation of asked from interactive debugger. */
+static int
+test_bpf_load_with_restarts(struct rte_bpf_prm_ex *prm)
+{
+ for (;;) {
+ const int rc = test_bpf_load(prm);
+
+ if (rc == -ECANCELED && debug_validate_again())
+ continue;
+
+ if (rc != 0)
+ fprintf(stderr, "Error %d loading BPF: %s\n",
+ -rc, strerror(-rc));
+ else
+ fprintf(stderr, "Validation succeeded.\n");
+
+ return rc;
+ }
+}
+
+int
+main(int argc, char *argv[])
+{
+ int rc;
+
+ struct args * const args = args_parse(argc, argv);
+ if (args == NULL || args->show_help
+ || args->bpf_prm.elf_file.path == NULL) {
+ args_destroy(args);
+ print_usage(argv[0]);
+ if (args == NULL)
+ /* Could not parse arguments. */
+ return 2;
+ print_supported_types();
+ print_defaults();
+ return 0;
+ }
+
+ const int eal_init_argc = get_eal_init_argc();
+ char ** const eal_init_argv = get_eal_init_argv(argv[0]);
+ rc = rte_eal_init(eal_init_argc, eal_init_argv);
+ if (rc < 0) {
+ fprintf(stderr, "Error %d initializing EAL: %s\n",
+ rte_errno, strerror(rte_errno));
+ args_destroy(args);
+ return EXIT_FAILURE;
+ }
+ RTE_VERIFY(rc == eal_init_argc - 1);
+
+ const int ret = test_bpf_load_with_restarts(&args->bpf_prm);
+
+ args_destroy(args);
+
+ rc = rte_eal_cleanup();
+ if (rc < 0) {
+ fprintf(stderr, "Error %d cleaning up EAL: %s\n",
+ rte_errno, strerror(rte_errno));
+ return EXIT_FAILURE;
+ }
+
+ return ret < 0 ? EXIT_FAILURE : EXIT_SUCCESS;
+}
diff --git a/app/validate-bpf/meson.build b/app/validate-bpf/meson.build
new file mode 100644
index 000000000000..03cfc329efb5
--- /dev/null
+++ b/app/validate-bpf/meson.build
@@ -0,0 +1,13 @@
+# SPDX-License-Identifier: BSD-3-Clause
+# Copyright(c) 2025 Huawei Technologies Co., Ltd
+
+sources = files(
+ 'alloc_list.c',
+ 'args.c',
+ 'debug.c',
+ 'debug_command.c',
+ 'eal_init_args.c',
+ 'main.c',
+ 'parse_decl.c',
+)
+deps = ['bpf', 'cmdline']
diff --git a/app/validate-bpf/parse_decl.c b/app/validate-bpf/parse_decl.c
new file mode 100644
index 000000000000..be733d58cd41
--- /dev/null
+++ b/app/validate-bpf/parse_decl.c
@@ -0,0 +1,623 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include "internal.h"
+
+#include <ctype.h>
+#include <stdlib.h>
+
+#include <rte_ether.h>
+#include <rte_ip.h>
+#include <rte_mbuf_core.h>
+#include <rte_tcp.h>
+#include <rte_udp.h>
+
+#define RETURN_TEXT_ERROR(text, text_start, message, ...) do { \
+ const int _offset = (text) - (text_start); \
+ VALIDATE_BPF_LOG(ERR, "at offset %d: " message, \
+ _offset, ## __VA_ARGS__); \
+ VALIDATE_BPF_LOG(NOTICE, "%s", text_start); \
+ VALIDATE_BPF_LOG(NOTICE, "%*c", _offset + 1, '^'); \
+ return -1; \
+} while (0)
+
+/* Used in place of any signature, this is not really being checked. */
+static uint64_t
+dummy_function(uint64_t arg1, uint64_t arg2, uint64_t arg3, uint64_t arg4,
+ uint64_t arg5)
+{
+ RTE_SET_USED(arg1);
+ RTE_SET_USED(arg2);
+ RTE_SET_USED(arg3);
+ RTE_SET_USED(arg4);
+ RTE_SET_USED(arg5);
+ return 0;
+}
+
+
+/* TOKENS AND TYPES */
+
+enum token {
+ TOKEN_UNRECOGNIZED = -1,
+ TOKEN_END = 0,
+
+ TOKEN_ASTERISK,
+ TOKEN_BRACKET_CLOSE,
+ TOKEN_BRACKET_OPEN,
+ TOKEN_COMMA,
+ TOKEN_PARENTHESIS_CLOSE,
+ TOKEN_PARENTHESIS_OPEN,
+ TOKEN_STRUCT,
+
+ TOKEN_TYPES_BEGIN,
+
+ TOKEN_TYPE_CHAR,
+ TOKEN_TYPE_ETHER_HEADER,
+ TOKEN_TYPE_INT32_T,
+ TOKEN_TYPE_IP_HEADER,
+ TOKEN_TYPE_IP_HEADERS,
+ TOKEN_TYPE_RTE_MBUF,
+ TOKEN_TYPE_TCP_HEADER,
+ TOKEN_TYPE_TCP_HEADERS,
+ TOKEN_TYPE_UDP_HEADER,
+ TOKEN_TYPE_UDP_HEADERS,
+ TOKEN_TYPE_UINT32_T,
+ TOKEN_TYPE_UINT64_T,
+ TOKEN_TYPE_UINTPTR_T,
+ TOKEN_TYPE_VOID,
+
+ TOKEN_TYPES_END,
+};
+
+/* Return true if the token is a type token */
+static bool
+is_type_token(enum token token)
+{
+ return token > TOKEN_TYPES_BEGIN && token < TOKEN_TYPES_END;
+}
+
+struct text_token {
+ const char *text;
+ unsigned int length; /* Not size_t to fit this struct in 2 regs. */
+ enum token token;
+};
+
+#define TEXT_TOKEN_DEF(text_, token_) { \
+ .text = text_, \
+ .length = sizeof(text_) - 1, \
+ .token = token_, \
+}
+
+/* Token search table, MUST BE SORTED! */
+const struct text_token TEXT_TOKENS[] = {
+ TEXT_TOKEN_DEF("(", TOKEN_PARENTHESIS_OPEN),
+ TEXT_TOKEN_DEF(")", TOKEN_PARENTHESIS_CLOSE),
+ TEXT_TOKEN_DEF("*", TOKEN_ASTERISK),
+ TEXT_TOKEN_DEF(",", TOKEN_COMMA),
+ TEXT_TOKEN_DEF("[", TOKEN_BRACKET_OPEN),
+ TEXT_TOKEN_DEF("]", TOKEN_BRACKET_CLOSE),
+ TEXT_TOKEN_DEF("char", TOKEN_TYPE_CHAR),
+ TEXT_TOKEN_DEF("ether_header", TOKEN_TYPE_ETHER_HEADER),
+ TEXT_TOKEN_DEF("int32_t", TOKEN_TYPE_INT32_T),
+ TEXT_TOKEN_DEF("ip_header", TOKEN_TYPE_IP_HEADER),
+ TEXT_TOKEN_DEF("ip_headers", TOKEN_TYPE_IP_HEADERS),
+ TEXT_TOKEN_DEF("rte_ether_hdr", TOKEN_TYPE_ETHER_HEADER),
+ TEXT_TOKEN_DEF("rte_ipv4_hdr", TOKEN_TYPE_IP_HEADER),
+ TEXT_TOKEN_DEF("rte_mbuf", TOKEN_TYPE_RTE_MBUF),
+ TEXT_TOKEN_DEF("rte_tcp_hdr", TOKEN_TYPE_TCP_HEADER),
+ TEXT_TOKEN_DEF("rte_udp_hdr", TOKEN_TYPE_UDP_HEADER),
+ TEXT_TOKEN_DEF("struct", TOKEN_STRUCT),
+ TEXT_TOKEN_DEF("tcp_header", TOKEN_TYPE_TCP_HEADER),
+ TEXT_TOKEN_DEF("tcp_headers", TOKEN_TYPE_TCP_HEADERS),
+ TEXT_TOKEN_DEF("udp_header", TOKEN_TYPE_UDP_HEADER),
+ TEXT_TOKEN_DEF("udp_headers", TOKEN_TYPE_UDP_HEADERS),
+ TEXT_TOKEN_DEF("uint32_t", TOKEN_TYPE_UINT32_T),
+ TEXT_TOKEN_DEF("uint64_t", TOKEN_TYPE_UINT64_T),
+ TEXT_TOKEN_DEF("uintptr_t", TOKEN_TYPE_UINTPTR_T),
+ TEXT_TOKEN_DEF("void", TOKEN_TYPE_VOID),
+};
+
+const struct text_token TEXT_TOKEN_UNRECOGNIZED = { "", 0, TOKEN_UNRECOGNIZED };
+const struct text_token TEXT_TOKEN_END = { "", 0, TOKEN_END };
+
+/* IP header maximum possible size with options. */
+#define IP_HEADER_MAX_SIZE 60
+#define IP_HEADERS_MAX_SIZE (sizeof(struct rte_ether_hdr) + IP_HEADER_MAX_SIZE)
+#define TCP_HEADERS_MAX_SIZE (IP_HEADERS_MAX_SIZE + sizeof(struct rte_tcp_hdr))
+#define UDP_HEADERS_MAX_SIZE (IP_HEADERS_MAX_SIZE + sizeof(struct rte_udp_hdr))
+
+#define POINTER_SIZE (sizeof(char *))
+
+const size_t TYPE_TOKEN_SIZE[] = {
+ [TOKEN_TYPE_CHAR] = sizeof(char),
+ [TOKEN_TYPE_ETHER_HEADER] = sizeof(struct rte_ether_hdr),
+ [TOKEN_TYPE_INT32_T] = sizeof(int32_t),
+ [TOKEN_TYPE_IP_HEADERS] = IP_HEADERS_MAX_SIZE,
+ [TOKEN_TYPE_IP_HEADER] = sizeof(struct rte_ipv4_hdr),
+ [TOKEN_TYPE_TCP_HEADERS] = TCP_HEADERS_MAX_SIZE,
+ [TOKEN_TYPE_TCP_HEADER] = sizeof(struct rte_tcp_hdr),
+ [TOKEN_TYPE_UDP_HEADERS] = UDP_HEADERS_MAX_SIZE,
+ [TOKEN_TYPE_UDP_HEADER] = sizeof(struct rte_udp_hdr),
+ [TOKEN_TYPE_UINT32_T] = sizeof(uint32_t),
+ [TOKEN_TYPE_UINT64_T] = sizeof(uint64_t),
+ [TOKEN_TYPE_UINTPTR_T] = sizeof(uintptr_t),
+};
+
+/*
+ * Struct rte_bpf_arg augmented with a type of pointer to it, since rte_bpf_arg
+ * by itself may not contain enough information to determine its pointer type.
+ */
+struct arg_info {
+ struct rte_bpf_arg value;
+ enum rte_bpf_arg_type ptr_type;
+ bool is_array;
+};
+
+/* Return arg_info struct described by the specified type token. */
+static struct arg_info
+get_type_token_arg(enum token type)
+{
+ RTE_ASSERT(is_type_token(type));
+ switch (type) {
+ case TOKEN_TYPE_VOID:
+ return (struct arg_info){
+ .value = { .type = RTE_BPF_ARG_UNDEF },
+ .ptr_type = RTE_BPF_ARG_PTR,
+ };
+ case TOKEN_TYPE_RTE_MBUF:
+ return (struct arg_info){
+ .value = {
+ .type = RTE_BPF_ARG_RAW,
+ .size = sizeof(struct rte_mbuf),
+ .buf_size = RTE_MBUF_DEFAULT_BUF_SIZE,
+ },
+ .ptr_type = RTE_BPF_ARG_PTR_MBUF,
+ };
+ default:
+ /* Should only reach here for normal sized types. */
+ RTE_ASSERT((size_t)type < RTE_DIM(TYPE_TOKEN_SIZE) &&
+ TYPE_TOKEN_SIZE[type] != 0);
+ return (struct arg_info){
+ .value = {
+ .type = RTE_BPF_ARG_RAW,
+ .size = TYPE_TOKEN_SIZE[type],
+ },
+ .ptr_type = RTE_BPF_ARG_PTR,
+ };
+ }
+}
+
+
+/* PARSING TEXT INTO TOKENS */
+
+/* Return true if character matches [a-zA-Z0-9_] in regex */
+static bool
+iswordchar(char character)
+{
+ return isalnum(character) || character == '_';
+}
+
+/*
+ * Compare pointer to text with pointer to struct text_token to determine
+ * if text starts with the specified token.
+ */
+static int
+text_and_text_token_cmp(const void *text_void_ptr, const void *text_token_void_ptr)
+{
+ int result;
+ const char * const text = text_void_ptr;
+ const struct text_token * const text_token = text_token_void_ptr;
+
+ if (memchr(text, 0, text_token->length) != NULL)
+ /* Text is shorter than the token. */
+ return strcmp(text, text_token->text);
+
+ /* Cannot use strcmp because we are looking for a prefix. */
+ result = memcmp(text, text_token->text, text_token->length);
+
+ /* Checking the case of a partial word match. */
+ if (result == 0 && iswordchar(text[text_token->length - 1]) &&
+ iswordchar(text[text_token->length]))
+ /* Text word is longer than the token. */
+ result = 1;
+
+ return result;
+}
+
+/* Advance pointed character pointer to the first non-space character. */
+static void
+skip_space(const char **text_ptr)
+{
+ while (isspace(**text_ptr))
+ ++*text_ptr;
+}
+
+/*
+ * Recognize and return a token starting text.
+ * Return TEXT_TOKEN_END if text is empty.
+ * Return TEXT_TOKEN_UNRECOGNIZED if text starts with unknown token.
+ */
+static struct text_token
+peek_text_token(const char *text)
+{
+ if (*text == '\0')
+ return TEXT_TOKEN_END;
+ const struct text_token * const text_token = bsearch(text, TEXT_TOKENS,
+ RTE_DIM(TEXT_TOKENS), sizeof(TEXT_TOKENS[0]),
+ text_and_text_token_cmp);
+ if (text_token == NULL)
+ return TEXT_TOKEN_UNRECOGNIZED;
+ return *text_token;
+}
+
+/*
+ * Advance pointed text starting with the specified token to the first
+ * non-space character after it.
+ * Do nothing if called for TEXT_TOKEN_END or TEXT_TOKEN_UNRECOGNIZED.
+ */
+static void
+consume_text_token(const char **text_ptr, struct text_token text_token)
+{
+ RTE_ASSERT(memcmp(*text_ptr, text_token.text, text_token.length) == 0);
+ *text_ptr += text_token.length;
+ skip_space(text_ptr);
+}
+
+/* Return length of the word starting at `text`. */
+static size_t
+find_word_length(const char *text)
+{
+ size_t word_length = 0;
+ while (iswordchar(text[word_length]))
+ ++word_length;
+ return word_length;
+}
+
+/* Create a copy of the word starting text, advance to next token. */
+static const char *
+take_name(const char **text_ptr, struct alloc_list *alloc_list)
+{
+ const size_t word_length = find_word_length(*text_ptr);
+ if (word_length == 0)
+ /* Text does not start with a word. */
+ return NULL;
+
+ /* Allocate memory for the word and add it to the alloc_list */
+ char *word = malloc(word_length + 1);
+ RTE_VERIFY(word != NULL);
+ alloc_list_append(alloc_list, word);
+
+ /* Copy and terminate word contents. */
+ memcpy(word, *text_ptr, word_length);
+ word[word_length] = '\0';
+
+ /* Advance text pointer. */
+ *text_ptr += word_length;
+ skip_space(text_ptr);
+
+ return word;
+}
+
+/* Read a number starting text, advance to next token. */
+static int
+take_number(size_t *number, const char **text_ptr)
+{
+ /* Read a word and let strtoull to decide if it's a number. */
+ const size_t word_length = find_word_length(*text_ptr);
+ if (word_length == 0)
+ /* Text does not start with a word. */
+ return -ENOENT;
+
+ errno = 0;
+ char *number_end;
+ unsigned long long long_number = strtoull(*text_ptr, &number_end, 0);
+ if (errno > 0)
+ return -errno;
+ if (number_end != *text_ptr + word_length)
+ /* Could not parse whole word. */
+ return -EINVAL;
+ if (long_number > SIZE_MAX)
+ return -ERANGE;
+
+ *number = long_number;
+
+ /* Advance text pointer. */
+ *text_ptr += word_length;
+ skip_space(text_ptr);
+
+ return 0;
+}
+
+
+/* PARSING DECLARATION PARTS */
+
+/* Change arg into a reference. */
+static void
+change_into_reference(struct arg_info *arg)
+{
+ if (RTE_BPF_ARG_PTR_TYPE(arg->value.type) != 0) {
+ VALIDATE_BPF_LOG(WARNING,
+ "After taking reference to a pointer the latter "
+ "will be described as an opaque pointer-size blob.");
+ RTE_ASSERT(arg->ptr_type == RTE_BPF_ARG_PTR);
+ arg->value.size = POINTER_SIZE;
+ }
+ arg->value.type = arg->ptr_type;
+ arg->ptr_type = RTE_BPF_ARG_PTR;
+ arg->is_array = false;
+}
+
+/*
+ * Recognize and consume arg starting text, advance to next token.
+ */
+static int
+take_arg(struct arg_info *arg, const char **text_ptr, const char *text_start)
+{
+ struct text_token next = peek_text_token(*text_ptr);
+
+ if (next.token == TOKEN_STRUCT) {
+ consume_text_token(text_ptr, next);
+ next = peek_text_token(*text_ptr);
+ }
+
+ if (!is_type_token(next.token))
+ RETURN_TEXT_ERROR(*text_ptr, text_start, "expect type");
+ const enum token type = next.token;
+ consume_text_token(text_ptr, next);
+ next = peek_text_token(*text_ptr);
+
+ *arg = get_type_token_arg(type);
+
+ while (next.token == TOKEN_ASTERISK) {
+ consume_text_token(text_ptr, next);
+ next = peek_text_token(*text_ptr);
+ change_into_reference(arg);
+ }
+
+ while (next.token == TOKEN_BRACKET_OPEN) {
+ consume_text_token(text_ptr, next);
+
+ /* Initialize to zero to avoid spurious compiler warnings. */
+ size_t array_length = 0;
+ if (take_number(&array_length, text_ptr) < 0)
+ RETURN_TEXT_ERROR(*text_ptr, text_start, "expect length");
+ next = peek_text_token(*text_ptr);
+
+ if (arg->value.size != 0 &&
+ array_length > SIZE_MAX / arg->value.size)
+ RETURN_TEXT_ERROR(*text_ptr, text_start, "type too big");
+
+ if (next.token != TOKEN_BRACKET_CLOSE)
+ RETURN_TEXT_ERROR(*text_ptr, text_start, "expect ']'");
+ consume_text_token(text_ptr, next);
+ next = peek_text_token(*text_ptr);
+
+ change_into_reference(arg);
+ arg->value.size *= array_length;
+ arg->is_array = true;
+ }
+
+ return 0;
+}
+
+/* Fill struct rte_bpf_arg within xsym trying not to unzero the padding. */
+static void
+fill_xsym_arg(struct rte_bpf_arg *target, struct rte_bpf_arg source)
+{
+ /* Copy fields individually to try and prevent copying the padding. */
+ target->type = source.type;
+ target->size = source.size;
+ target->buf_size = source.buf_size;
+}
+
+/* Build and return struct rte_bpf_arg of type RTE_BPF_XTYPE_VAR */
+static int
+fill_var_xsym(struct rte_bpf_xsym *xsym, struct arg_info arg, const char *name,
+ struct alloc_list *alloc_list)
+{
+ /* Variables are passed by reference, except for arrays. */
+ if (!arg.is_array) {
+ if (RTE_BPF_ARG_PTR_TYPE(arg.value.type) != 0)
+ VALIDATE_BPF_LOG(WARNING,
+ "External pointers may not work as expected "
+ "because all external variables are passed by "
+ "reference but there is currently no way to "
+ "describe double pointer to the validator.");
+ change_into_reference(&arg);
+ }
+
+ /* Allocate something to assign to a val pointer. */
+ void * const val = calloc(1, RTE_MAX(1u, arg.value.size));
+ if (val == NULL) {
+ VALIDATE_BPF_LOG(ERR, "could not allocate enough memory");
+ return -1;
+ }
+ alloc_list_append(alloc_list, val);
+
+ /* Need all padding and unused fields to be zero-filled. */
+ memset(xsym, 0, sizeof(*xsym));
+ xsym->name = name;
+ xsym->type = RTE_BPF_XTYPE_VAR;
+ xsym->var.val = val;
+ fill_xsym_arg(&xsym->var.desc, arg.value);
+
+ return 0;
+}
+
+/* Build and return struct rte_bpf_arg of type RTE_BPF_XTYPE_FUNC */
+static void
+fill_func_xsym(struct rte_bpf_xsym *xsym, struct arg_info arg, const char *name)
+{
+ /* Need all padding and unused fields to be zero-filled. */
+ memset(xsym, 0, sizeof(*xsym));
+ xsym->name = name;
+ xsym->type = RTE_BPF_XTYPE_FUNC;
+ xsym->func.val = &dummy_function;
+ fill_xsym_arg(&xsym->func.ret, arg.value);
+}
+
+/*
+ * Parse and store function arguments, advance to next token after ')'.
+ * Value of *text_ptr should point to the next token after '('.
+ */
+static int
+take_func_xsym_args(struct rte_bpf_xsym *xsym, const char **text_ptr,
+ const char *text_start)
+{
+ struct arg_info arg;
+ struct text_token delimiter = peek_text_token(*text_ptr);
+
+ while (delimiter.token != TOKEN_PARENTHESIS_CLOSE) {
+ if (xsym->func.nb_args == EBPF_FUNC_MAX_ARGS)
+ RETURN_TEXT_ERROR(*text_ptr, text_start,
+ "too many arguments, maximum %d allowed",
+ EBPF_FUNC_MAX_ARGS);
+
+ if (take_arg(&arg, text_ptr, text_start) < 0)
+ return -1;
+ if (arg.value.type == RTE_BPF_ARG_UNDEF &&
+ xsym->func.nb_args != 0)
+ RETURN_TEXT_ERROR(*text_ptr, text_start,
+ "arguments of type void are not allowed");
+ fill_xsym_arg(&xsym->func.args[xsym->func.nb_args++],
+ arg.value);
+
+ delimiter = peek_text_token(*text_ptr);
+ switch (delimiter.token) {
+ case TOKEN_COMMA:
+ consume_text_token(text_ptr, delimiter);
+ continue;
+ case TOKEN_PARENTHESIS_CLOSE:
+ break;
+ default:
+ RETURN_TEXT_ERROR(*text_ptr, text_start,
+ "expect ')' or ','");
+ }
+ }
+ consume_text_token(text_ptr, delimiter);
+
+ /* Special case of single void argument. */
+ if (xsym->func.nb_args == 1 &&
+ xsym->func.args[0].type == RTE_BPF_ARG_UNDEF) {
+ xsym->func.nb_args = 0;
+ /* Need all padding and unused fields to be zero-filled. */
+ fill_xsym_arg(&xsym->func.args[0], (struct rte_bpf_arg){});
+ }
+
+ return 0;
+}
+
+/* Make sure text has ended. */
+static int
+ensure_end(const char *text, const char *text_start)
+{
+ if (peek_text_token(text).token != TOKEN_END)
+ RETURN_TEXT_ERROR(text, text_start, "trailing garbage");
+ return 0;
+}
+
+/* Parse and store xsym, advance to next token. */
+static int
+take_xsym(struct rte_bpf_xsym *xsym, const char **text_ptr,
+ const char *text_start, struct alloc_list *alloc_list)
+{
+ struct arg_info arg;
+
+ if (take_arg(&arg, text_ptr, text_start) < 0)
+ return -1;
+
+ const char * const name = take_name(text_ptr, alloc_list);
+ if (name == NULL)
+ RETURN_TEXT_ERROR(*text_ptr, text_start, "expect name");
+
+ const struct text_token next = peek_text_token(*text_ptr);
+ switch (next.token) {
+ case TOKEN_END:
+ if (fill_var_xsym(xsym, arg, name, alloc_list) < 0)
+ return -1;
+ break;
+ case TOKEN_PARENTHESIS_OPEN:
+ consume_text_token(text_ptr, next);
+ fill_func_xsym(xsym, arg, name);
+ if (take_func_xsym_args(xsym, text_ptr, text_start) < 0)
+ return -1;
+ break;
+ default:
+ RETURN_TEXT_ERROR(*text_ptr, text_start,
+ "expect '(' or text end");
+ }
+ return 0;
+}
+
+
+/* PUBLIC FUNCTIONS */
+
+void
+print_supported_types(void)
+{
+ printf("TYPE: [struct] BASIC_TYPE [*]... [[N]]...\n");
+ printf("BASIC_TYPE: one of\n");
+ for (int tti = 0; tti != RTE_DIM(TEXT_TOKENS); ++tti) {
+ if (is_type_token(TEXT_TOKENS[tti].token))
+ printf("\t%s\n", TEXT_TOKENS[tti].text);
+ }
+}
+
+int
+parse_arg(struct rte_bpf_arg *arg, const char *text)
+{
+ struct arg_info arg_info;
+ const char * const text_start = text;
+ skip_space(&text);
+ if (take_arg(&arg_info, &text, text_start) < 0)
+ return -1;
+ if (ensure_end(text, text_start) < 0)
+ return -1;
+ if (arg_info.ptr_type != RTE_BPF_ARG_PTR)
+ VALIDATE_BPF_LOG(WARNING,
+ "`%s` has a special pointer type which was left unused; "
+ "argument was set to an opaque blob.",
+ text_start);
+ *arg = arg_info.value;
+ return 0;
+}
+
+int
+parse_xsym(struct rte_bpf_xsym *xsym, const char *text,
+ struct alloc_list *alloc_list)
+{
+ const char * const text_start = text;
+ skip_space(&text);
+ if (take_xsym(xsym, &text, text_start, alloc_list) < 0)
+ return -1;
+ if (ensure_end(text, text_start) < 0)
+ return -1;
+ return 0;
+}
+
+void
+adjust_arg_buf_size(struct rte_bpf_arg *arg, size_t mbuf_buf_size)
+{
+ if (arg->buf_size != 0)
+ /* Non-zero buf_size indicates mbuf or a pointer to it. */
+ arg->buf_size = mbuf_buf_size;
+}
+
+void
+adjust_xsym_buf_size(struct rte_bpf_xsym *xsym, size_t mbuf_buf_size)
+{
+ switch (xsym->type) {
+ case RTE_BPF_XTYPE_FUNC:
+ for (uint32_t argi = 0; argi != xsym->func.nb_args; ++argi)
+ adjust_arg_buf_size(&xsym->func.args[argi],
+ mbuf_buf_size);
+ adjust_arg_buf_size(&xsym->func.ret, mbuf_buf_size);
+ break;
+ case RTE_BPF_XTYPE_VAR:
+ adjust_arg_buf_size(&xsym->var.desc, mbuf_buf_size);
+ break;
+ default:
+ rte_panic("Unexpected xsym type %d\n", xsym->type);
+ }
+}
diff --git a/doc/guides/rel_notes/release_26_11.rst b/doc/guides/rel_notes/release_26_11.rst
index 87c7e81bdebd..5c0caba50d07 100644
--- a/doc/guides/rel_notes/release_26_11.rst
+++ b/doc/guides/rel_notes/release_26_11.rst
@@ -55,6 +55,11 @@ New Features
Also, make sure to start the actual text at the margin.
=======================================================
+* **Added BPF validation application.**
+
+ Added ``dpdk-validate-bpf`` tool to pre-validate eBPF programs for
+ compatibility with the ``lib/bpf`` execution context.
+
Removed Items
-------------
diff --git a/doc/guides/tools/index.rst b/doc/guides/tools/index.rst
index 13f75a5bc652..1c43060e15cd 100644
--- a/doc/guides/tools/index.rst
+++ b/doc/guides/tools/index.rst
@@ -26,3 +26,4 @@ DPDK Tools User Guides
testmldev
graph
dts
+ validate_bpf
diff --git a/doc/guides/tools/validate_bpf.rst b/doc/guides/tools/validate_bpf.rst
new file mode 100644
index 000000000000..9e2d58cfbe5f
--- /dev/null
+++ b/doc/guides/tools/validate_bpf.rst
@@ -0,0 +1,97 @@
+.. SPDX-License-Identifier: BSD-3-Clause
+ Copyright(c) 2026 The DPDK contributors
+
+dpdk-validate-bpf Application
+=============================
+
+The ``dpdk-validate-bpf`` tool is an application that allows evaluating BPF
+programs against the ``lib/bpf`` library. It can be used to pre-validate BPF
+programs before loading them in a real application to ensure that they pass the
+internal verifier. This makes it a useful tool for integrating into build
+systems and continuous integration (CI) pipelines to verify the correctness
+of BPF programs during compilation, as well as for debugging validation issues.
+
+Running the Application
+-----------------------
+
+The tool requires a path to a compiled BPF object file. It also provides options
+to configure the execution environment and external symbol definitions.
+
+For a comprehensive list of options and their meanings, refer to the
+built-in help by running ``dpdk-validate-bpf --help``.
+
+The most important options for a minimal example are:
+
+* ``--prog-arg <type>``
+
+ Define program arguments, if different from ``struct rte_mbuf *``.
+ This argument may be repeated up to 5 times.
+
+* ``--xsym '<type> <name> | <type> <name>(<type>, ...)'``
+
+ Define an external symbol (variable or function) that the BPF program uses.
+ This allows the validator to recognize external calls and memory accesses.
+ Multiple external symbols can be defined by repeating this option.
+
+* ``--section <name>``
+
+ Specify the ELF section name in the BPF object file to load and validate.
+ The default section name is ``.text``, which is typically fine in cases
+ where the object file contains only one program.
+
+Interactive Debugging Mode
+--------------------------
+
+The interactive mode (enabled with the ``--debug`` flag) allows the user to
+debug the BPF validation process itself. This is particularly useful when
+BPF program fails verification, allowing you to trace the state changes per
+instruction and understand the validator's decisions.
+
+Design
+~~~~~~
+
+The interactive debugger provides a command-line interface similar to GDB.
+It allows setting breakpoints at specific instructions, or catchpoints on
+validator events (such as ``branch-enter`` or ``branch-prune``).
+Users can step through the validation process,
+inspect the state of registers, and query if certain conditional jumps may
+be taken based on the validator's knowledge of the program state.
+
+Usage Example
+~~~~~~~~~~~~~
+
+Launch the tool in debug mode with a compiled BPF object file:
+
+.. code-block:: console
+
+ $ dpdk-validate-bpf bpf_prog.o --prog-arg={'void *',uint64_t} --debug
+ => 0: b7 00 00 00 01 00 00 00 mov r0, #0x1
+ (validate) list 3
+ => 0: b7 00 00 00 01 00 00 00 mov r0, #0x1
+ 1: 25 02 01 00 0e 00 00 00 jgt r2, #0xe, L3
+ 2: b7 00 00 00 00 00 00 00 mov r0, #0x0
+ (validate) break 2
+ Breakpoint 0 at 2.
+ (validate) catch branch-enter
+ Catchpoint 1 on branch-enter.
+ (validate) continue
+ Catchpoint 1 on branch-enter.
+ Entered new branch at pc 1.
+ => 3: 95 00 00 00 00 00 00 00 exit
+ (validate) where
+ 1: 25 02 01 00 0e 00 00 00 jgt r2, #0xe, L3 ; taken
+ => 3: 95 00 00 00 00 00 00 00 exit
+ (validate) info r1
+ r1: %buffer<0> + 0
+ (validate) info r2
+ r2: 0xf..UINT64_MAX
+ (validate) may r2 <= 14
+ NO
+ (validate) continue
+ Catchpoint 1 on branch-enter.
+ Entered new branch at pc 1.
+ Breakpoint 0 at 2.
+ => 2: b7 00 00 00 00 00 00 00 mov r0, #0x0
+ (validate) continue
+ Validation succeeded.
+ (validate) quit
--
2.43.0
^ permalink raw reply related [flat|nested] 38+ messages in thread
* RE: [PATCH v2 0/7] validate-bpf: add standalone validation debug tool
2026-09-14 18:42 ` [PATCH v2 0/7] validate-bpf: add standalone validation debug tool Stephen Hemminger
@ 2026-09-15 19:27 ` Marat Khalili
0 siblings, 0 replies; 38+ messages in thread
From: Marat Khalili @ 2026-09-15 19:27 UTC (permalink / raw)
To: Stephen Hemminger; +Cc: dev@dpdk.org
Thanks a lot for looking, please see the answers inline below.
> Looks good to me, but AI had some nits to pick.
>
> Re: [PATCH v2 0/7] bpf/validate: debug events and validation app
>
> Errors
> ------
>
> Patch 7/7 (app/validate-bpf: add BPF validation application)
>
> app/validate-bpf/parse_decl.c:
>
> + arg->value.size *= array_length;
>
> array_length comes from take_number() on the --xsym text with no
> upper bound, and the multiply is not checked for overflow, so a
> large length silently wraps to a small size that is then reported
> to the validator as the object size. Reproduced:
>
> dpdk-validate-bpf --xsym='uint64_t[2305843009213693953] v' \
> --section=.text prog.o
> Validation succeeded.
>
> 8 * 2305843009213693953 wraps to 8, so the tool accepts the
> declaration and describes an 8-byte object. Clamp array_length,
> or check the product, and reject with the usual text error.
Addressed in v3.
> Warnings
> --------
>
> Patch 5/7 (bpf/validate: add get current event API)
>
> lib/bpf/bpf_validate_debug.c:
>
> + return -EINVAL;
>
> in rte_bpf_validate_debug_get_event(), whose return type is
> enum rte_bpf_validate_debug_event. Every enumerator in that enum
> is non-negative, so the compatible type is unsigned and -EINVAL
> arrives at the caller as 4294967274. It equals no enumerator,
> including RTE_BPF_VALIDATE_DEBUG_EVENT_END, so a caller cannot
> test for it. Either return
> RTE_BPF_VALIDATE_DEBUG_EVENT_END for the NULL case and say so in
> the doc comment, or change the signature to return int with the
> event in an out-parameter.
Behavior is intentionally not defined and caller is intentionally not required
to check return value from this API for errors to keep its life easier.
The return value is always valid with the catchpoint callback.
> Patch 7/7
>
> app/validate-bpf/parse_decl.c:
>
> + void * const val = calloc(1, RTE_MAX(1u, arg.value.size));
> + RTE_VERIFY(val != 0);
>
> arg.value.size is user-supplied, so an allocation failure driven by
> the command line aborts with a core dump instead of an error
> message. Reproduced:
>
> dpdk-validate-bpf --xsym='uint64_t[18446744073709551615] v' \
> --section=.text prog.o
> EAL: PANIC in fill_var_xsym():
> line 427 assert "val != 0" failed
>
> Return an error up through the parse path like the other --xsym
> failures do.
Addressed in v3.
>
> app/validate-bpf/main.c:
>
> + RTE_VERIFY(rte_eal_init(eal_init_argc, eal_init_argv) ==
> + eal_init_argc - 1);
>
> rte_eal_init() returns -1 for ordinary environmental failures such
> as a missing runtime directory or insufficient permissions. For a
> command-line tool that should be a message and a non-zero exit, not
> rte_panic(). Same for
>
> + RTE_VERIFY(rte_eal_cleanup() == 0);
Addressed in v3.
> app/validate-bpf/eal_init_args.c:
>
> +#define RTE_EAL_INIT_ARG_SIZE_MAX sizeof("--log-level=lib.eal:warning")
>
> and RTE_EAL_INIT_ARGS / RTE_EAL_INIT_ARGC below it. The RTE_
> prefix is reserved for the libraries; an application should not
> define into it. Drop the prefix.
Addressed in v3.
> Info
> ----
>
> Patch 5/7
>
> debug->current_event is set only in debug_send_event(). A
> breakpoint callback reached through debug_trigger_breakpoints()
> therefore sees whatever event fired last. The doc comment says the
> value is undefined when no event is being processed, so this is
> consistent, but a callback shared between a breakpoint and a
> catchpoint will read a stale event rather than an obviously invalid
> one.
There is currently no use case for this, and undefined behaviour allows
implementing this in the future.
> rte_bpf_validate_debug_get_event() is new public experimental API
> and gets no release note. 7/7 adds a release note, but only for
> the application.
Technically true, but practically I suspect there are no users of this API
yet, so noise in release notes is kept to the minimum.
> Patch 7/7
>
> In --debug mode the result is printed twice: debug.c prints
> "Validation succeeded." to stdout from the validation-success
> catchpoint, and main.c prints it again to stderr.
There is a prompt between them, so the result looks ok.
> Array declarations are written as "uint64_t[4] v", not the C
> "uint64_t v[4]". The C spelling is rejected with
>
> at offset 10: expect '(' or text end
>
> which does not point at the real problem. Worth either accepting
> the C form or naming the expected form in the message. The
> documented form is in --help, so this is a diagnostics nit.
Yes, this is a documented current limitation, patches are welcome.
> "uint64_t[0] v" is accepted and yields value.size 0 while calloc
> still allocates one byte. Probably worth rejecting a zero length.
Decisions on validity of external symbols are left to the BPF loader.
> Patch 7/7 (app/validate-bpf: add BPF validation application)
>
> app/validate-bpf/parse_decl.c:
>
> + arg->value.size *= array_length;
>
> array_length comes from take_number() on the --xsym text with no
> upper bound, and the multiply is not checked for overflow, so a
> large length silently wraps to a small size that is then reported
> to the validator as the object size. Reproduced:
>
> dpdk-validate-bpf --xsym='uint64_t[2305843009213693953] v' \
> --section=.text prog.o
> Validation succeeded.
>
> 8 * 2305843009213693953 wraps to 8, so the tool accepts the
> declaration and describes an 8-byte object. Clamp array_length,
> or check the product, and reject with the usual text error.
Addressed in v3.
> Build fails on FreeBSD
// snip
> ../app/validate-bpf/parse_decl.c:392:33: error: 'array_length' may be used uninitialized [-
> Werror=maybe-uninitialized]
Hopefully addressed in v3.
^ permalink raw reply [flat|nested] 38+ messages in thread
* RE: [PATCH v3 7/7] app/validate-bpf: add BPF validation application
2026-09-15 19:26 ` [PATCH v3 7/7] app/validate-bpf: add BPF validation application Marat Khalili
@ 2026-09-16 16:36 ` Konstantin Ananyev
0 siblings, 0 replies; 38+ messages in thread
From: Konstantin Ananyev @ 2026-09-16 16:36 UTC (permalink / raw)
To: Marat Khalili, Thomas Monjalon; +Cc: dev@dpdk.org
> Add a new DPDK application, dpdk-validate-bpf, for pre-validating eBPF
> programs for compatibility with the lib/bpf execution context.
>
> The application allows evaluating eBPF programs against the DPDK verifier
> before loading them into a real application. It includes an interactive
> debugging mode to trace state changes per instruction and understand the
> validator's decisions.
>
> Signed-off-by: Marat Khalili <marat.khalili@huawei.com>
> ---
> MAINTAINERS | 2 +
> app/meson.build | 1 +
> app/validate-bpf/alloc_list.c | 53 ++
> app/validate-bpf/args.c | 263 ++++++
> app/validate-bpf/debug.c | 1103 ++++++++++++++++++++++++
> app/validate-bpf/debug_command.c | 383 ++++++++
> app/validate-bpf/debug_command.h | 65 ++
> app/validate-bpf/eal_init_args.c | 57 ++
> app/validate-bpf/internal.h | 151 ++++
> app/validate-bpf/main.c | 90 ++
> app/validate-bpf/meson.build | 13 +
> app/validate-bpf/parse_decl.c | 623 +++++++++++++
> doc/guides/rel_notes/release_26_11.rst | 5 +
> doc/guides/tools/index.rst | 1 +
> doc/guides/tools/validate_bpf.rst | 97 +++
> 15 files changed, 2907 insertions(+)
> create mode 100644 app/validate-bpf/alloc_list.c
> create mode 100644 app/validate-bpf/args.c
> create mode 100644 app/validate-bpf/debug.c
> create mode 100644 app/validate-bpf/debug_command.c
> create mode 100644 app/validate-bpf/debug_command.h
> create mode 100644 app/validate-bpf/eal_init_args.c
> create mode 100644 app/validate-bpf/internal.h
> create mode 100644 app/validate-bpf/main.c
> create mode 100644 app/validate-bpf/meson.build
> create mode 100644 app/validate-bpf/parse_decl.c
> create mode 100644 doc/guides/tools/validate_bpf.rst
>
> diff --git a/MAINTAINERS b/MAINTAINERS
> index e99a65d1974d..39cd0a3de6fa 100644
> --- a/MAINTAINERS
> +++ b/MAINTAINERS
> @@ -1923,8 +1923,10 @@ F: lib/bpf/
> F: examples/bpf/
> F: app/test/test_bpf*
> F: app/test/bpf/
> +F: app/validate-bpf/
> F: app/test-pmd/bpf_cmd.*
> F: doc/guides/prog_guide/bpf_lib.rst
> +F: doc/guides/tools/validate_bpf.rst
>
> Graph
> M: Jerin Jacob <jerinj@marvell.com>
> diff --git a/app/meson.build b/app/meson.build
> index 1798db3ae43f..9b59ae483056 100644
> --- a/app/meson.build
> +++ b/app/meson.build
> @@ -33,6 +33,7 @@ apps = [
> 'test-regex',
> 'test-sad',
> 'test-security-perf',
> + 'validate-bpf',
> ]
>
> if get_option('tests')
> diff --git a/app/validate-bpf/alloc_list.c b/app/validate-bpf/alloc_list.c
> new file mode 100644
> index 000000000000..2f4d18ad3df6
> --- /dev/null
> +++ b/app/validate-bpf/alloc_list.c
> @@ -0,0 +1,53 @@
> +/* SPDX-License-Identifier: BSD-3-Clause
> + * Copyright(c) 2025 Huawei Technologies Co., Ltd
> + */
> +
> +#include "internal.h"
> +#include <stdlib.h>
> +
> +
> +/* Needs to be a power of two. */
> +#define START_CAPACITY (1u << 3)
> +
> +size_t
> +alloc_list_append(struct alloc_list *alloc_list, void *ptr)
> +{
> + if (alloc_list->count == 0) {
> + RTE_ASSERT(alloc_list->ptrs == NULL);
> + alloc_list->ptrs = malloc(
> + sizeof(alloc_list->ptrs[0]) * START_CAPACITY);
> + RTE_VERIFY(alloc_list->ptrs != NULL);
> + } else if (alloc_list->count >= START_CAPACITY &&
> + /* Power of two detection */
> + (alloc_list->count & (alloc_list->count - 1)) == 0) {
> + /*
> + * We allocate in powers of two, and current count is one of
> + * these powers, so need to reallocate to larger capacity.
> + */
> + RTE_ASSERT(alloc_list->ptrs != NULL);
> + const size_t new_capacity = alloc_list->count * 2;
> + alloc_list->ptrs = realloc(alloc_list->ptrs,
> + sizeof(alloc_list->ptrs[0]) * new_capacity);
> + RTE_VERIFY(alloc_list->ptrs != NULL);
> + }
> + alloc_list->ptrs[alloc_list->count] = ptr;
> + return alloc_list->count++;
> +}
> +
> +void alloc_list_replace(struct alloc_list *alloc_list, size_t index, void *ptr)
> +{
> + RTE_ASSERT(index < alloc_list->count);
> + alloc_list->ptrs[index] = ptr;
> +}
> +
> +void alloc_list_free_all(struct alloc_list *alloc_list)
> +{
> + /* Copy and clear fields first in case alloc_list itself gets freed. */
> + size_t count = alloc_list->count;
> + void ** const ptrs = alloc_list->ptrs;
> + *alloc_list = (struct alloc_list){};
> +
> + while (count != 0)
> + free(ptrs[--count]);
> + free(ptrs);
> +}
> diff --git a/app/validate-bpf/args.c b/app/validate-bpf/args.c
> new file mode 100644
> index 000000000000..cd7cbd694a3c
> --- /dev/null
> +++ b/app/validate-bpf/args.c
> @@ -0,0 +1,263 @@
> +/* SPDX-License-Identifier: BSD-3-Clause
> + * Copyright(c) 2025 Huawei Technologies Co., Ltd
> + */
> +
> +#include "internal.h"
> +
> +#include <ctype.h>
> +#include <getopt.h>
> +#include <stdlib.h>
> +
> +#include <rte_errno.h>
> +
> +
> +/* Values to be used in getopt_long option.val. */
> +enum app_args {
> + ARG_UNRECOGNIZED = '?',
> + ARG_AUTO = 0, /* Value set by getopt_long when flag is non-NULL. */
> + ARG_HELP, /* Keep this one in the beginning for UX reasons. */
> + ARG_DEBUG,
> + ARG_MBUF_BUF_SIZE,
> + ARG_NO_PROG_ARGS,
> + ARG_PROG_ARG,
> + ARG_SECTION,
> + ARG_XSYM,
> +};
> +
> +/* Program options. */
> +static struct option OPTIONS[] = {
Here and everywhere:
By DPDK coding convention variable names are in lowercase:
https://doc.dpdk.org/guides-25.07/contributing/coding_style.html
> + {
> + .name = "help",
> + .val = ARG_HELP,
> + },
> + {
> + .name = "debug",
> + .val = ARG_DEBUG,
> + },
> + {
> + .name = "mbuf-buf-size",
> + .has_arg = required_argument,
> + .val = ARG_MBUF_BUF_SIZE,
> + },
> + {
> + .name = "no-prog-arg",
> + .val = ARG_NO_PROG_ARGS,
> + },
> + {
> + .name = "no-prog-args",
> + .val = ARG_NO_PROG_ARGS,
> + },
> + {
> + .name = "prog-arg",
> + .has_arg = required_argument,
> + .val = ARG_PROG_ARG,
> + },
> + {
> + .name = "section",
> + .has_arg = required_argument,
> + .val = ARG_SECTION,
> + },
> + {
> + .name = "xsym",
> + .has_arg = required_argument,
> + .val = ARG_XSYM,
> + },
> + { /* terminating zero record */ }
> +};
> +
> +/* Default args value. */
> +static const struct args ARGS_DEFAULT = {
> + .bpf_prm = {
> + .sz = sizeof(struct rte_bpf_prm_ex),
> + .origin = RTE_BPF_ORIGIN_ELF_FILE,
> + .elf_file.section = ".text",
> + },
> + .mbuf_buf_size = RTE_MBUF_DEFAULT_BUF_SIZE,
> +};
> +
> +/* Default --prog-arg argument value. */
> +static const char * const PROG_ARG_DEFAULT = "struct rte_mbuf *";
> +
> +
> +void
> +print_usage(const char *program_name)
> +{
> + static const char *const options[][2] = {
> + { "--help", "Display help and exit." },
> + { "--debug", "Enable interactive debug mode." },
> + { "--mbuf-buf-size=MBUF_BUF_SIZE", "Size of the mbuf data buffer
> (in bytes)." },
> + { "--prog-arg=TYPE", "Expected type of the next BPF program
> argument (up to 5)." },
> + { "--no-prog-args", "BPF program does not take any arguments." },
> + { "--section=SECTION", "ELF section name in the BPF object file to
> load." },
> + { "--xsym='TYPE NAME[(TYPE, ...)]'", "External symbol BPF program
> can access." },
> + };
> +
> + printf("USAGE: %s [OPTIONS]... BPF_PATH\n", program_name);
> + printf("OPTIONS:\n");
> + for (int oi = 0; oi != RTE_DIM(options); ++oi)
> + printf("\t%-31s %s\n", options[oi][0], options[oi][1]);
> +}
> +
> +void
> +print_defaults(void)
> +{
> + printf("DEFAULTS:"
> + " --mbuf-buf-size=%#zx"
> + " --prog-arg='%s'"
> + " --section='%s'\n",
> + (size_t)RTE_MBUF_DEFAULT_BUF_SIZE,
> + PROG_ARG_DEFAULT,
> + ARGS_DEFAULT.bpf_prm.elf_file.section);
> +}
> +
> +static int
> +parse_size(size_t *result, const char *text)
> +{
> + char *parse_end = NULL;
> +
> + errno = 0;
> + const uintmax_t strtoumax_result = strtoumax(text, &parse_end, 0);
Curious, why not just stroul()?
> + if (errno != 0 || *parse_end != '\0' || strtoumax_result == 0 ||
> + strtoumax_result > SIZE_MAX)
> + return -1;
> +
> + *result = strtoumax_result;
> + return 0;
> +}
> +
> +struct args *
> +args_parse(int argc, char *argv[])
> +{
> + int val;
> + size_t xsym_alloc_index;
> + struct rte_bpf_xsym *xsym = NULL;
> + const char * const program_name = argv[0];
> +
> + /* Allocate args and set aliases for some of its members. */
> + struct args * const args = malloc(sizeof(*args));
> + RTE_VERIFY(args != NULL);
> + struct alloc_list * const alloc_list = &args->_alloc_list;
> + struct rte_bpf_prm_ex * const bpf_prm = &args->bpf_prm;
> +
> + /* Set default values. */
> + *args = ARGS_DEFAULT;
> + RTE_VERIFY(parse_arg(&bpf_prm->prog_arg[bpf_prm->nb_prog_arg++],
> + PROG_ARG_DEFAULT) >= 0);
> + bool default_prog_args = true;
> +
> + /* Reserve space for xsym in alloc_list. */
> + xsym_alloc_index = alloc_list_append(alloc_list, xsym);
> +
> + while ((val = getopt_long(argc, argv, "", OPTIONS, NULL)) != EOF) {
> + int rc = 0;
> + switch (val) {
> + case ARG_AUTO:
> + /* getopt_long made the assignment, nothing to do */
> + break;
> + case ARG_HELP:
> + args->show_help = true;
> + break;
> + case ARG_DEBUG:
> + if (bpf_prm->debug == NULL)
> + bpf_prm->debug = debug_create();
> + if (bpf_prm->debug == NULL) {
> + rc = -rte_errno;
> + fprintf(stderr,
> + "%s: error %d creating debug session\n",
> + program_name, -rc);
> + }
> + break;
> + case ARG_MBUF_BUF_SIZE:
> + rc = parse_size(&args->mbuf_buf_size, optarg);
> + if (rc < 0)
> + fprintf(stderr,
> + "%s: invalid mbuf buf size '%s'\n",
> + program_name, optarg);
> + break;
> + case ARG_NO_PROG_ARGS:
> + bpf_prm->nb_prog_arg = 0;
> + default_prog_args = false;
> + break;
> + case ARG_PROG_ARG:
> + if (default_prog_args) {
> + bpf_prm->nb_prog_arg = 0;
> + default_prog_args = false;
> + }
> +
> + if (bpf_prm->nb_prog_arg == RTE_DIM(bpf_prm-
> >prog_arg)) {
> + fprintf(stderr,
> + "%s: at most %d program arguments
> allowed\n",
> + program_name,
> + (int)RTE_DIM(bpf_prm->prog_arg));
> + rc = -EINVAL;
> + break;
> + }
> +
> + rc = parse_arg(&bpf_prm->prog_arg[bpf_prm-
> >nb_prog_arg++],
> + optarg);
> + if (rc < 0)
> + fprintf(stderr,
> + "%s: unrecognized prog arg '%s'\n",
> + program_name, optarg);
> +
> + break;
> + case ARG_SECTION:
> + bpf_prm->elf_file.section = optarg;
> + break;
> + case ARG_XSYM:
> + bpf_prm->xsym = xsym = realloc(xsym,
> + sizeof(xsym[0]) * (bpf_prm->nb_xsym + 1));
> + RTE_VERIFY(xsym != NULL);
> + alloc_list_replace(alloc_list, xsym_alloc_index, xsym);
> + rc = parse_xsym(&xsym[bpf_prm->nb_xsym++], optarg,
> + alloc_list);
> + if (rc < 0)
> + fprintf(stderr, "%s: invalid xsym '%s'\n",
> + program_name, optarg);
> + break;
> + case ARG_UNRECOGNIZED:
> + args_destroy(args);
> + return NULL;
> + default:
> + rte_panic("Unexpected getopt_long return value %d\n",
> + val);
> + }
> + if (rc < 0) {
> + args_destroy(args);
> + return NULL;
> + }
> + }
> +
> + /* Set buf_size to the value specified in command line arguments. */
> + adjust_arg_buf_size(bpf_prm->prog_arg, args->mbuf_buf_size);
> + for (size_t xsymi = 0; xsymi != bpf_prm->nb_xsym; ++xsymi)
> + adjust_xsym_buf_size(&xsym[xsymi], args->mbuf_buf_size);
> +
> + /* getopt_long moves all non-options to the end, starting at optind. */
> + const int nb_bpf_path = argc - optind;
> + switch (nb_bpf_path) {
> + case 0:
> + break;
> + case 1:
> + bpf_prm->elf_file.path = argv[optind];
> + break;
> + default:
> + fprintf(stderr, "%s: too many positional arguments\n",
> + program_name);
> + args_destroy(args);
> + return NULL;
> + }
> +
> + return args;
> +}
> +
> +void
> +args_destroy(struct args *args)
> +{
> + if (args == NULL)
> + return;
> + if (args->bpf_prm.debug != NULL)
> + debug_destroy(args->bpf_prm.debug);
> + alloc_list_free_all(&args->_alloc_list);
> + free(args);
> +}
> diff --git a/app/validate-bpf/debug.c b/app/validate-bpf/debug.c
> new file mode 100644
> index 000000000000..ac085b53e3bd
> --- /dev/null
> +++ b/app/validate-bpf/debug.c
> @@ -0,0 +1,1103 @@
> +/* SPDX-License-Identifier: BSD-3-Clause
> + * Copyright(c) 2025 Huawei Technologies Co., Ltd
> + */
> +
> +#include "debug_command.h"
> +#include "internal.h"
> +
> +#include <rte_bpf_validate_debug.h>
> +#include <rte_errno.h>
> +
> +#include <stdlib.h>
> +
> +
> +/* Write single line to the user, currently just to stdout. */
> +#define PRINTLN(fmt, ...) do { \
> + RTE_LOG_CHECK_NO_NEWLINE(fmt); \
> + printf(fmt "\n", ##__VA_ARGS__); \
> +} while (0)
> +
> +#define PROMPT "(validate) "
> +
> +#define INITIAL_CAPACITY 8u
> +
> +static const char *const event_names[] = {
> + [RTE_BPF_VALIDATE_DEBUG_EVENT_INVALID_STATE] = "invalid-state",
> + [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER] = "branch-enter",
> + [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE] = "branch-prune",
> + [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN] = "branch-
> return",
> + [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE] = "branch-
> unreachable",
> + [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS] = "jump-always",
> + [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL] = "jump-
> conditional",
> +};
> +
> +static const char *const register_names[] = {
> + [EBPF_REG_0] = "r0",
> + [EBPF_REG_1] = "r1",
> + [EBPF_REG_2] = "r2",
> + [EBPF_REG_3] = "r3",
> + [EBPF_REG_4] = "r4",
> + [EBPF_REG_5] = "r5",
> + [EBPF_REG_6] = "r6",
> + [EBPF_REG_7] = "r7",
> + [EBPF_REG_8] = "r8",
> + [EBPF_REG_9] = "r9",
> + [EBPF_REG_10] = "r10",
> +};
> +
> +static const char *const comparison_operator_names[] = {
> + [BPF_JEQ] = "==",
> + [BPF_JGT] = ">",
> + [BPF_JGE] = ">=",
> + [EBPF_JNE] = "!=",
> + [EBPF_JSGT] = "s>",
> + [EBPF_JSGE] = "s>=",
> + [EBPF_JLT] = "<",
> + [EBPF_JLE] = "<=",
> + [EBPF_JSLT] = "s<",
> + [EBPF_JSLE] = "s<=",
> +};
> +
> +enum point_type {
> + POINT_TYPE_BREAK,
> + POINT_TYPE_CATCH,
> +};
> +
> +/* Local representation of points with additional info for UI. */
> +struct point_info {
> + struct rte_bpf_validate_debug_point *point;
> + enum point_type type;
> + union {
> + uint32_t pc;
> + enum rte_bpf_validate_debug_event event;
> + };
> +};
> +
> +/* Dynamically growing list of point infos. */
> +struct point_infos {
> + struct point_info *elements;
> + uint32_t length;
> + uint32_t capacity;
> +};
> +
> +/* Information about a single conditional or unconditional jump in code path. */
> +struct branch_info {
> + uint32_t jump_pc;
> + uint32_t target_pc;
> + bool is_conditional;
> +};
> +
> +/* Dynamically growing stack to track code path. */
> +struct branch_stack {
> + struct branch_info *branches;
> + uint32_t length;
> + uint32_t capacity;
> +};
> +
> +/* Flags telling if we should validate again, stopping or not at start. */
> +static bool validate_again;
> +
> +/* List of point infos; element index is its ID for UI. */
> +static struct point_infos point_infos;
> +
> +/* Catchpoint invisible to the user used for step-by-step validation. */
> +static struct rte_bpf_validate_debug_point *step_point;
> +
> +/* Tracking entered branches. */
> +static struct branch_stack branch_stack;
> +static uint32_t pending_jump_pc = UINT32_MAX;
> +static struct rte_bpf_validate_debug_point *jump_always_step_point;
> +
> +static int
> +step_cb(struct rte_bpf_validate_debug *debug, void *ctx);
> +static int
> +point_cb(struct rte_bpf_validate_debug *debug, void *ctx);
> +
> +/* Find index of the string in the list. */
> +static int
> +find_name(const char *name, const char *const *names, int nb_names)
> +{
> + if (nb_names < 0)
> + return -EINVAL;
> +
> + for (int index = 0; index != nb_names; ++index)
> + if (names[index] != NULL && strcmp(names[index], name) == 0)
> + return index;
> +
> + return -ENOENT;
> +}
> +
> +/* Convert event name to its enum value. */
> +static enum rte_bpf_validate_debug_event
> +parse_event_name(const char *name)
> +{
> + const int rc = find_name(name, event_names, RTE_DIM(event_names));
> + if (rc < 0)
> + PRINTLN("Error: invalid event name.");
> +
> + return rc;
> +}
> +
> +/* Convert comparison operator name to opcode. */
> +static int
> +parse_comparison_operator(const char *name)
> +{
> + const int rc = find_name(name, comparison_operator_names,
> + RTE_DIM(comparison_operator_names));
> + if (rc < 0)
> + PRINTLN("Error: invalid comparison operator name.");
> +
> + return rc;
> +}
> +
> +/* Convert register name to its number. */
> +static int
> +parse_register(const char *name)
> +{
> + const int rc = find_name(name, register_names, RTE_DIM(register_names));
> + if (rc < 0)
> + PRINTLN("Error: invalid register name.");
> +
> + return rc;
> +}
> +
> +/* Free local list of point infos (do not destroy points). */
> +static void
> +point_infos_free(void)
> +{
> + free(point_infos.elements);
> + point_infos = (struct point_infos){};
> +}
> +
> +/* Free local branch stack. */
> +static void
> +branch_stack_free(void)
> +{
> + free(branch_stack.branches);
> + branch_stack = (struct branch_stack){};
> +}
> +
> +/* Return existing element from the list of point infos. */
> +static struct point_info *
> +point_infos_at(uint32_t point_number)
> +{
> + RTE_ASSERT(point_number < point_infos.length);
> + RTE_ASSERT(point_infos.elements[point_number].point != NULL);
> + return &point_infos.elements[point_number];
> +}
> +
> +/* Print existing element from the list of point infos. */
> +static void
> +point_infos_print_at(uint32_t point_number)
> +{
> + const struct point_info *const point_info = point_infos_at(point_number);
> +
> + switch (point_info->type) {
> + case POINT_TYPE_BREAK:
> + PRINTLN("Breakpoint %d at %d.", point_number,
> + point_info->pc);
> + break;
> + case POINT_TYPE_CATCH:
> + PRINTLN("Catchpoint %d on %s.", point_number,
> + event_names[point_info->event]);
> + break;
> + default:
> + PRINTLN("Point %d of unknown type", point_number);
> + break;
> + }
> +}
> +
> +/* Print all point infos. */
> +static int
> +point_infos_print_all(void)
> +{
> + uint32_t nb_printed = 0;
> + for (uint32_t pn = 0; pn != point_infos.length; ++pn) {
> + const struct point_info *const point_info =
> + &point_infos.elements[pn];
> + if (point_info->point != NULL) {
> + point_infos_print_at(pn);
> + ++nb_printed;
> + }
> + }
> +
> + if (nb_printed == 0) {
> + PRINTLN("No breakpoints or catchpoints set.");
> + return -ENOENT;
> + }
> +
> + return nb_printed;
> +}
> +
> +/* Allocate space for a new element in the list of point infos. */
> +static uint32_t
> +point_infos_append(void)
> +{
> + if (point_infos.length == point_infos.capacity) {
> + /* Set to initial capacity or double previous one. */
> + point_infos.capacity = RTE_MAX(INITIAL_CAPACITY,
> + point_infos.capacity * 2);
> + point_infos.elements = realloc(point_infos.elements,
> + point_infos.capacity * sizeof(point_infos.elements[0]));
> + RTE_VERIFY(point_infos.elements != NULL);
> + }
> + return point_infos.length++;
> +}
> +
> +/* Allocate space for a new element in the branch stack. */
> +static void
> +branch_stack_append(const struct branch_info *branch)
> +{
> + if (branch_stack.length == branch_stack.capacity) {
> + /* Set to initial capacity or double previous one. */
> + branch_stack.capacity = RTE_MAX(INITIAL_CAPACITY,
> + branch_stack.capacity * 2);
> + branch_stack.branches = realloc(branch_stack.branches,
> + branch_stack.capacity * sizeof(branch_stack.branches[0]));
> + RTE_VERIFY(branch_stack.branches != NULL);
> + }
> + branch_stack.branches[branch_stack.length++] = *branch;
> +}
> +
> +/* Destroy existing element from the list of point infos, printing it first. */
> +static void
> +point_infos_destroy_existing(uint32_t point_number)
> +{
> + point_infos_print_at(point_number);
> +
> + struct point_info *const point_info = point_infos_at(point_number);
> + rte_bpf_validate_debug_point_destroy(point_info->point);
> + *point_info = (struct point_info){};
> +}
> +
> +/* Destroy point with specified number if it exists. */
> +static int
> +point_infos_destroy_at(uint32_t point_number)
> +{
> + if (point_number >= point_infos.length ||
> + point_infos.elements[point_number].point == NULL) {
> + PRINTLN("No breakpoint number %d.", point_number);
> + return -ENOENT;
> + }
> +
> + point_infos_destroy_existing(point_number);
> + return 1;
> +}
> +
> +/* Destroy all point infos. */
> +static int64_t
> +point_infos_destroy_all(void)
> +{
> + uint32_t nb_destroyed = 0;
> + for (uint32_t pn = 0; pn < point_infos.length; ++pn) {
> + const struct point_info *const point_info =
> + &point_infos.elements[pn];
> + if (point_info->point != NULL) {
> + point_infos_destroy_existing(pn);
> + ++nb_destroyed;
> + }
> + }
> +
> + if (nb_destroyed == 0) {
> + PRINTLN("No breakpoints or catchpoints set.");
> + return -ENOENT;
> + }
> +
> + return nb_destroyed;
Curious if function always returns uint32_t, why make it to return 'int64_t'?
For errors? But the only error is -ENOENT when nb_destroyed == 0.
Same question for similar functions below.
> +}
> +
> +/* Destroy all breakpoints at specified location. */
> +static int64_t
> +point_infos_destroy_breakpoints(uint32_t pc)
> +{
> + uint32_t nb_destroyed = 0;
> + for (uint32_t pn = 0; pn < point_infos.length; ++pn) {
> + const struct point_info *const point_info =
> + &point_infos.elements[pn];
> + if (point_info->point != NULL &&
> + point_info->type == POINT_TYPE_BREAK &&
> + point_info->pc == pc) {
> + point_infos_destroy_existing(pn);
> + ++nb_destroyed;
> + }
> + }
> +
> + if (nb_destroyed == 0) {
> + PRINTLN("No breakpoint at %u.", pc);
> + return -ENOENT;
> + }
> +
> + return nb_destroyed;
> +}
> +
> +/* Destroy all catchpoints for specified event. */
> +static int64_t
> +point_infos_destroy_catchpoints(int event)
> +{
> + if (event < 0)
> + /* Error was already printed by parse_event_name. */
> + return event;
> +
> + uint32_t nb_destroyed = 0;
> + for (uint32_t pn = 0; pn < point_infos.length; ++pn) {
> + const struct point_info *const point_info =
> + &point_infos.elements[pn];
> + if (point_info->point != NULL &&
> + point_info->type == POINT_TYPE_CATCH &&
> + (int)point_info->event == event) {
> + point_infos_destroy_existing(pn);
> + ++nb_destroyed;
> + }
> + }
> +
> + if (nb_destroyed == 0) {
> + PRINTLN("No catchpoint on %s.", event_names[event]);
> + return -ENOENT;
> + }
> +
> + return nb_destroyed;
> +}
> +
> +/* Create new breakpoint at specified location. */
> +static int
> +add_breakpoint(struct rte_bpf_validate_debug *debug, uint32_t nb_ins, uint32_t
> pc)
> +{
> + const uint32_t point_number = point_infos_append();
> +
> + if (pc >= nb_ins) {
> + PRINTLN("Error: program only has %u instructions.", nb_ins);
> + return -ENOENT;
> + }
> +
> + struct rte_bpf_validate_debug_point *const point =
> + rte_bpf_validate_debug_break(debug, pc,
> + &(struct rte_bpf_validate_debug_callback){
> + .fn = point_cb,
> + .ctx = (void *)(uintptr_t)point_number,
> + });
> + if (point == NULL) {
> + PRINTLN("Library error %d.", rte_errno);
> + return -rte_errno;
> + }
> +
> + point_infos.elements[point_number] = (struct point_info){
> + .point = point,
> + .type = POINT_TYPE_BREAK,
> + .pc = pc,
> + };
> + point_infos_print_at(point_number);
> + return 0;
> +}
> +
> +/* Create new catchpoint at specified location. */
> +static int
> +add_catchpoint(struct rte_bpf_validate_debug *debug, int event)
> +{
> + if (event < 0)
> + /* Error was already printed by parse_event_name. */
> + return event;
> +
> + const uint32_t point_number = point_infos_append();
> + struct rte_bpf_validate_debug_point *const point =
> + rte_bpf_validate_debug_catch(debug, event,
> + &(struct rte_bpf_validate_debug_callback){
> + .fn = point_cb,
> + .ctx = (void *)(uintptr_t)point_number,
> + });
> + if (point == NULL) {
> + PRINTLN("Library error %d.", rte_errno);
> + return -rte_errno;
> + }
> +
> + point_infos.elements[point_number] = (struct point_info){
> + .point = point,
> + .type = POINT_TYPE_CATCH,
> + .event = event,
> + };
> + point_infos_print_at(point_number);
> + return 0;
> +}
> +
> +static bool
> +is_step_enabled(void)
> +{
> + return step_point != NULL;
> +}
> +
> +/* Enable step-by-step validation: make sure catchpoint is set on step event. */
> +static int
> +enable_step(struct rte_bpf_validate_debug *debug)
> +{
> + if (is_step_enabled())
> + return 0;
> +
> + step_point = rte_bpf_validate_debug_catch(debug,
> + RTE_BPF_VALIDATE_DEBUG_EVENT_STEP,
> + &(struct rte_bpf_validate_debug_callback){ step_cb });
> + if (step_point == NULL) {
> + PRINTLN("Library error %d.", rte_errno);
> + return -rte_errno;
> + }
> +
> + return 0;
> +}
> +
> +/* Disable step-by-step validation: destroy catchpoint on step event if any. */
> +static void
> +disable_step(void)
> +{
> + rte_bpf_validate_debug_point_destroy(step_point);
> + step_point = NULL;
> +}
> +
> +/* Format and print information about specified frame offset. */
> +static int
> +print_frame_offset(struct rte_bpf_validate_debug *debug, int32_t offset)
> +{
> + char *info;
> + int info_size, rc;
> +
> + if (offset >= 0 || offset % sizeof(uint64_t) != 0) {
> + PRINTLN("Invalid frame offset, must be a negative multiple of
> %zu.",
> + sizeof(uint64_t));
> + return -EINVAL;
> + }
> +
> + rc = rte_bpf_validate_debug_format_frame_info(debug, NULL, 0, offset);
> + if (rc == -ERANGE) {
> + PRINTLN("Offset is out of frame range.");
> + return rc;
> + }
> + if (rc < 0) {
> + PRINTLN("Error %d printing information.", -rc);
> + return rc;
> + }
> +
> + info_size = rc + 1;
> + info = malloc(info_size);
> + if (info == NULL)
> + return -ENOMEM;
> +
> + rc = rte_bpf_validate_debug_format_frame_info(debug, info, info_size,
> + offset);
> + if (rc + 1 != info_size) {
> + if (rc >= 0) {
> + PRINTLN("Expect format return value %d, got %d.",
> + info_size, rc);
> + rc = -EINVAL;
> + } else
> + PRINTLN("Error %d printing information.", -rc);
> + free(info);
> + return rc;
> + }
> +
> + printf("%5jd: \t%s\n", (intmax_t)offset, info);
Again, what for intmax_t here, if offset is just 32bit int?
> + free(info);
> + return 0;
> +}
> +
> +/* Format and print informatiion about the frame. */
> +static int
> +print_frame(struct rte_bpf_validate_debug *debug)
> +{
> + int32_t frame_size;
> + int rc;
> +
> + frame_size = rte_bpf_validate_debug_get_frame_size(debug);
> + if (frame_size < 0) {
> + PRINTLN("Error %d getting frame size.", -frame_size);
> + return frame_size;
> + }
> +
> + for (int32_t frame_offset = 0;;) {
That's an unusal loop construction...
Why not something more common, obvious and easy to read?
i.e:
for(ofs = sizeof(uint64_t); ofs < frame_size; ofs += sizeof(uint64_t)) {
rc = print_frame_offset(debug, -ofs);
if (rc < 0) return rc;
}
> + frame_offset -= sizeof(uint64_t);
> + if (frame_offset < -frame_size)
> + break;
> +
> + rc = print_frame_offset(debug, frame_offset);
> + if (rc < 0)
> + return rc;
> + }
> +
> + return 0;
> +}
> +
> +/* Format and print informatiion about specified register. */
> +static int
> +print_register(struct rte_bpf_validate_debug *debug, int reg)
> +{
> + char *info;
> + int info_size, rc;
> +
> + if (reg < 0)
> + /* Error was already printed by parse_register. */
> + return reg;
> +
> + rc = rte_bpf_validate_debug_format_register_info(debug, NULL, 0, reg);
> + if (rc < 0) {
> + PRINTLN("Error %d printing information.", -rc);
> + return rc;
> + }
> +
> + info_size = rc + 1;
> + info = malloc(info_size);
> + if (info == NULL)
> + return -ENOMEM;
In some places you panic (RTE_VERIFY) when malloc() fails, in others return an error?
Why? Any strategy behind?
> +
> + rc = rte_bpf_validate_debug_format_register_info(debug, info, info_size,
> + reg);
> + if (rc + 1 != info_size) {
> + if (rc >= 0) {
> + PRINTLN("Expect format return value %d, got %d.",
> + info_size, rc);
> + rc = -EINVAL;
> + } else
> + PRINTLN("Error %d printing information.", -rc);
> + free(info);
> + return rc;
> + }
> +
> + printf("%5s: \t%s\n", register_names[reg], info);
> + free(info);
> + return 0;
> +}
> +
> +/* Format and print informatiion about all registers. */
> +static int
> +print_registers(struct rte_bpf_validate_debug *debug)
> +{
> + int rc = 0;
> +
> + for (int reg = 0; reg != EBPF_REG_NUM; ++reg)
> + rc = rc < 0 ? rc : print_register(debug, reg);
> +
> + return rc;
> +}
> +
> +/* List one eBPF program instruction. */
> +static int
> +list_one(const struct ebpf_insn *ins, uint32_t nb_ins, uint32_t offset,
> + uint32_t pc, const char *comment)
> +{
> + char hexadecimal[256], disassembly[256];
> +
> + if (offset >= nb_ins) {
> + PRINTLN("Error: program only has %u instructions.", nb_ins);
> + return -EINVAL;
> + }
> +
> + ins += offset;
> +
> + if (offset == nb_ins - 1 && rte_bpf_insn_is_wide(ins)) {
> + PRINTLN("Error: truncated last instruction.");
> + return -EINVAL;
> + }
> +
> + rte_bpf_format(hexadecimal, sizeof(hexadecimal), ins, 0,
> + RTE_BPF_FORMAT_FLAG_HEXADECIMAL |
> + RTE_BPF_FORMAT_FLAG_NEVER_WIDE);
> + rte_bpf_format(disassembly, sizeof(disassembly), ins, offset,
> + RTE_BPF_FORMAT_FLAG_DISASSEMBLY |
> + RTE_BPF_FORMAT_FLAG_ABSOLUTE_JUMPS);
> +
> + if (comment == NULL)
> + comment = "";
> + PRINTLN("%2s %10u: \t%s \t%s%s%s",
> + offset == pc ? "=>" : "", offset, hexadecimal, disassembly,
> + comment[0] != '\0' ? " \t; " : "", comment);
> +
> + if (rte_bpf_insn_is_wide(ins)) {
> + rte_bpf_format(hexadecimal, sizeof(hexadecimal), ins + 1, 0,
> + RTE_BPF_FORMAT_FLAG_HEXADECIMAL |
> + RTE_BPF_FORMAT_FLAG_NEVER_WIDE);
> + PRINTLN("%15s\t%s", "", hexadecimal);
> + }
> +
> + return 0;
> +}
> +
> +/* List specified range of eBPF program instructions, updating start offset. */
> +static int
> +list(const struct ebpf_insn *ins, uint32_t nb_ins, uint32_t *offset,
> + uint32_t count, uint32_t pc)
> +{
> + uint32_t local_offset = 0;
> + if (offset == NULL)
> + offset = &local_offset;
> +
> + if (*offset > nb_ins) {
> + PRINTLN("Error: program only has %u instructions.", nb_ins);
> + return -EINVAL;
> + }
> +
> + const uint32_t end =
> + /* Calculate end in a way preventing overflow: */
> + *offset + RTE_MIN(nb_ins - *offset, count);
> + while (*offset < end) {
> + const int rc = list_one(ins, nb_ins, *offset, pc, NULL);
> + if (rc < 0)
> + return rc;
> +
> + *offset += 1 + rte_bpf_insn_is_wide(&ins[*offset]);
> + }
> +
> + return 0;
> +}
> +
> +/* Print if specified conditional jump _may_ be executed. */
> +static int
> +print_if_may(struct rte_bpf_validate_debug *debug, const struct ebpf_insn
> *jump,
> + uint64_t imm64)
> +{
> + const int result = rte_bpf_validate_debug_may_jump(debug, jump,
> imm64);
> +
> + switch (result) {
> + case 0:
> + case RTE_BPF_VALIDATE_DEBUG_MAY_BE_FALSE:
> + PRINTLN("NO");
> + break;
> + case RTE_BPF_VALIDATE_DEBUG_MAY_BE_TRUE:
> + case RTE_BPF_VALIDATE_DEBUG_MAY_BE_FALSE |
> RTE_BPF_VALIDATE_DEBUG_MAY_BE_TRUE:
> + PRINTLN("YES");
> + break;
> + default:
> + PRINTLN("Error %d getting result.", -result);
> + break;
> + }
> +
> + return result;
> +}
> +
> +/* Print if specified condition with literal right hand side _may_ be true. */
> +static int
> +print_if_may_literal_rhs(struct rte_bpf_validate_debug *debug,
> + const struct debug_command_comparison *comparison)
> +{
> + const int lhs = parse_register(comparison->lhs);
> + const int op = parse_comparison_operator(comparison->op);
> + const int64_t rhs = comparison->literal_rhs;
> +
> + if (lhs < 0 || op < 0)
> + /* Error was already printed by parse function. */
> + return lhs < 0 ? lhs : op;
> +
> + return print_if_may(debug, &(struct ebpf_insn){
> + .code = BPF_JMP | op | BPF_K,
> + .dst_reg = lhs,
> + }, /* imm64 = */ rhs);
> +}
> +
> +/* Print if specified condition with register right hand side _may_ be true. */
> +static int
> +print_if_may_register_rhs(struct rte_bpf_validate_debug *debug,
> + const struct debug_command_comparison *comparison)
> +{
> + const int lhs = parse_register(comparison->lhs);
> + const int op = parse_comparison_operator(comparison->op);
> + const int rhs = parse_register(comparison->register_rhs);
> +
> + if (lhs < 0 || op < 0 || rhs < 0)
> + /* Error was already printed by parse function. */
> + return lhs < 0 ? lhs : op < 0 ? op : rhs;
> +
> + return print_if_may(debug, &(struct ebpf_insn){
> + .code = BPF_JMP | op | BPF_X,
> + .dst_reg = lhs,
> + .src_reg = rhs,
> + }, /* imm64 = */ 0);
> +}
> +
> +/* Return 1 on validation success, 0 on failure, -EAGAIN if still running. */
> +static int
> +get_validation_success(struct rte_bpf_validate_debug *debug)
> +{
> + int validation_result, rc;
> +
> + rc = rte_bpf_validate_debug_get_validation_result(debug,
> + &validation_result);
> + return rc < 0 ? rc : (validation_result >= 0);
> +}
> +
> +static void
> +print_status(const struct ebpf_insn *ins, uint32_t nb_ins,
> + int validation_success, uint32_t pc)
> +{
> + if (validation_success == 1) {
> + PRINTLN("Validation succeeded.");
> + return;
> + }
> +
> + list_one(ins, nb_ins, pc, pc, NULL);
> +
> + if (validation_success == 0)
> + PRINTLN("Validation failed.");
> +}
> +
> +static void
> +debug_command_where(const struct ebpf_insn *ins, uint32_t nb_ins,
> + int validation_success, uint32_t pc)
> +{
> + for (uint32_t bi = 0; bi != branch_stack.length; ++bi) {
> + const uint32_t jump_pc = branch_stack.branches[bi].jump_pc;
> + const uint32_t target_pc = branch_stack.branches[bi].target_pc;
> + const char *const comment =
> + !branch_stack.branches[bi].is_conditional ? NULL :
> + target_pc == jump_pc + 1 ? "fallen-through" : "taken";
> + list_one(ins, nb_ins, jump_pc, UINT32_MAX, comment);
> + }
> + print_status(ins, nb_ins, validation_success, pc);
> +}
> +
> +/* Return true if pc is defined, otherwise print an error message. */
> +static bool
> +ensure_pc_defined(uint32_t pc)
> +{
> + if (pc != UINT32_MAX)
> + return true;
> +
> + PRINTLN("No current instruction.");
> + return false;
> +}
> +
> +/* Return true if still validating, otherwise print an error message. */
> +static bool
> +ensure_still_validating(int validation_success)
> +{
> + if (validation_success == -EAGAIN)
> + return true;
> +
> + PRINTLN("Finished, use `start` or `run` to restart, `quit` to quit.");
> + return false;
> +}
> +
> +/* Step-by-step validation callback: read and process user commands. */
> +static int
> +step_cb(struct rte_bpf_validate_debug *debug, __rte_unused void *ctx)
> +{
> + int rc;
> + int validation_success;
> + const struct ebpf_insn *ins;
> + uint32_t nb_ins, pc, list_offset;
> +
> + validate_again = false;
> +
> + rc = rte_bpf_validate_debug_get_ins(debug, &ins, &nb_ins);
> + if (rc < 0) {
> + PRINTLN("Error %d getting program instructions.", -rc);
> + return rc;
> + }
> +
> + validation_success = get_validation_success(debug);
> + if (validation_success < 0 && validation_success != -EAGAIN) {
> + PRINTLN("Error %d getting validation result.",
> + -validation_success);
> + return validation_success;
> + }
> +
> + pc = validation_success == 1 ? UINT32_MAX :
> + rte_bpf_validate_debug_get_pc(debug);
> +
> + print_status(ins, nb_ins, validation_success, pc);
> +
> + list_offset = pc;
> +
> + while (true) {
> + switch (debug_command_get(PROMPT)) {
> + case DEBUG_COMMAND_BREAK:
> + if (ensure_pc_defined(pc))
> + add_breakpoint(debug, nb_ins, pc);
> + continue;
> + case DEBUG_COMMAND_BREAK_PC:
> + add_breakpoint(debug, nb_ins,
> debug_command_parsed.pc);
> + continue;
> + case DEBUG_COMMAND_CATCH:
> + add_catchpoint(debug,
> +
> parse_event_name(debug_command_parsed.event));
> + continue;
> + case DEBUG_COMMAND_CLEAR:
> + if (ensure_pc_defined(pc))
> + point_infos_destroy_breakpoints(pc);
> + continue;
> + case DEBUG_COMMAND_CLEAR_EVENT:
> + point_infos_destroy_catchpoints(
> +
> parse_event_name(debug_command_parsed.event));
> + continue;
> + case DEBUG_COMMAND_CLEAR_PC:
> + point_infos_destroy_breakpoints(
> + debug_command_parsed.pc);
> + continue;
> + case DEBUG_COMMAND_CONTINUE:
> + if (ensure_still_validating(validation_success)) {
> + disable_step();
> + return 0;
> + }
> + continue;
> + case DEBUG_COMMAND_DELETE:
> + point_infos_destroy_all();
> + continue;
> + case DEBUG_COMMAND_DELETE_NUMBER:
> + point_infos_destroy_at(
> + debug_command_parsed.point_number);
> + continue;
> + case DEBUG_COMMAND_INFO_FRAME:
> + print_frame(debug);
> + continue;
> + case DEBUG_COMMAND_INFO_FRAME_OFFSET:
> + print_frame_offset(debug,
> + debug_command_parsed.frame_offset);
> + continue;
> + case DEBUG_COMMAND_INFO_POINTS:
> + point_infos_print_all();
> + continue;
> + case DEBUG_COMMAND_INFO_REGISTER:
> + print_register(debug,
> +
> parse_register(debug_command_parsed.register_));
> + continue;
> + case DEBUG_COMMAND_INFO_REGISTERS:
> + print_registers(debug);
> + continue;
> + case DEBUG_COMMAND_LIST:
> + if (ensure_pc_defined(pc))
> + list(ins, nb_ins, &list_offset, 10, pc);
> + continue;
> + case DEBUG_COMMAND_LIST_COUNT:
> + if (ensure_pc_defined(pc))
> + list(ins, nb_ins, &list_offset,
> +
> debug_command_parsed.instruction_count, pc);
> + continue;
> + case DEBUG_COMMAND_LIST_PROGRAM:
> + if (ensure_pc_defined(pc))
> + list(ins, nb_ins, NULL, UINT32_MAX, pc);
> + continue;
> + case DEBUG_COMMAND_MAY_LITERAL_RHS:
> + print_if_may_literal_rhs(debug,
> + &debug_command_parsed.comparison);
> + continue;
> + case DEBUG_COMMAND_MAY_REGISTER_RHS:
> + print_if_may_register_rhs(debug,
> + &debug_command_parsed.comparison);
> + continue;
> + case DEBUG_COMMAND_EOF:
> + case DEBUG_COMMAND_QUIT:
> + PRINTLN("Quitting...");
> + return validation_success == -EAGAIN ? -ECANCELED : 0;
> + case DEBUG_COMMAND_RUN:
> + PRINTLN("Re-running...");
> + disable_step();
> + validate_again = true;
> + return -ECANCELED;
> + case DEBUG_COMMAND_START:
> + PRINTLN("Re-starting...");
> + rc = enable_step(debug);
> + if (rc < 0)
> + return rc;
> + validate_again = true;
> + return -ECANCELED;
> + case DEBUG_COMMAND_STEP:
> + if (ensure_still_validating(validation_success))
> + return 0;
> + continue;
> + case DEBUG_COMMAND_WHERE:
> + debug_command_where(ins, nb_ins, validation_success,
> pc);
> + continue;
> + default:
> + PRINTLN("INTERNAL ERROR");
> + return -ENOTSUP;
> + }
> + }
> +}
> +
> +/* Any point callback: print it and enable step-by-step validation. */
> +static int
> +point_cb(struct rte_bpf_validate_debug *debug, void *ctx)
> +{
> + const uint32_t point_number = (uintptr_t)ctx;
> + point_infos_print_at(point_number);
> + return enable_step(debug);
> +}
> +
> +/* Branch stack machinery. */
> +
> +static void
> +clear_jump_always_step_point(void)
> +{
> + rte_bpf_validate_debug_point_destroy(jump_always_step_point);
> + jump_always_step_point = NULL;
> +}
> +
> +static int
> +reset_branch_tracking(struct rte_bpf_validate_debug *debug __rte_unused,
> + void *ctx __rte_unused)
> +{
> + branch_stack.length = 0;
> + pending_jump_pc = UINT32_MAX;
> + clear_jump_always_step_point();
> + return 0;
> +}
> +
> +/*
> + * Handling the jump-always instructions:
> + * - upon a jump-always event save the pc and set a custom step callback;
> + * - ignore the first custom step callback call (still on the same jump);
> + * - on the second call push the jump pc into stack and delete the callback;
> + */
> +
> +static int
> +jump_always_step_cb(struct rte_bpf_validate_debug *debug, void *ctx
> __rte_unused)
> +{
> + /* Step event is also emitted at the end of the jump instruction itself. */
> + if (rte_bpf_validate_debug_get_pc(debug) == pending_jump_pc)
> + return 0;
> +
> + branch_stack_append(&(struct branch_info){
> + .jump_pc = pending_jump_pc,
> + .target_pc = rte_bpf_validate_debug_get_pc(debug),
> + .is_conditional = false,
> + });
> + clear_jump_always_step_point();
> + return 0;
> +}
> +
> +static int
> +jump_always_cb(struct rte_bpf_validate_debug *debug, void *ctx __rte_unused)
> +{
> + pending_jump_pc = rte_bpf_validate_debug_get_pc(debug);
> + clear_jump_always_step_point();
> + jump_always_step_point = rte_bpf_validate_debug_catch(debug,
> + RTE_BPF_VALIDATE_DEBUG_EVENT_STEP,
> + &(struct rte_bpf_validate_debug_callback){ jump_always_step_cb
> });
> + return 0;
> +}
> +
> +/*
> + * Handling the jump-conditional instructions:
> + * - upon a jump-conditional event save the pc;
> + * - upon a branch-enter event push the conditional jump pc into stack;
> + * - upon a branch-return event pop conditional jump and all unconditional jumps
> + * preceding it from the stack;
> + * - during step execution notify the user about the step events;
> + */
> +
> +static int
> +jump_conditional_cb(struct rte_bpf_validate_debug *debug, void *ctx
> __rte_unused)
> +{
> + pending_jump_pc = rte_bpf_validate_debug_get_pc(debug);
> + return 0;
> +}
> +
> +static int
> +branch_enter_cb(struct rte_bpf_validate_debug *debug, void *ctx __rte_unused)
> +{
> + branch_stack_append(&(struct branch_info){
> + .jump_pc = pending_jump_pc,
> + .target_pc = rte_bpf_validate_debug_get_pc(debug),
> + .is_conditional = true,
> + });
> + if (is_step_enabled())
> + PRINTLN("Entered new branch at pc %u.", pending_jump_pc);
> + return 0;
> +}
> +
> +static int
> +branch_return_cb(struct rte_bpf_validate_debug *debug __rte_unused,
> + void *ctx __rte_unused)
> +{
> + clear_jump_always_step_point();
> +
> + while (branch_stack.length > 0) {
> + const struct branch_info branch_info =
> + branch_stack.branches[--branch_stack.length];
> + pending_jump_pc = branch_info.jump_pc;
> + if (branch_info.is_conditional)
> + break;
> + }
> + if (is_step_enabled())
> + PRINTLN("Returned from branch at pc %u.", pending_jump_pc);
> + return 0;
> +}
> +
> +/* Notify user when skipping branches in step mode. */
> +
> +static int
> +branch_prune_cb(struct rte_bpf_validate_debug *debug __rte_unused,
> + void *ctx __rte_unused)
> +{
> + if (is_step_enabled())
> + PRINTLN("Prunned branch at pc %u.", pending_jump_pc);
> + return 0;
> +}
> +
> +static int
> +branch_unreachable_cb(struct rte_bpf_validate_debug *debug __rte_unused,
> + void *ctx __rte_unused)
> +{
> + if (is_step_enabled())
> + PRINTLN("Unreachable branch at pc %u.", pending_jump_pc);
> + return 0;
> +}
> +
> +/* Global initialization and cleanup functions. */
> +
> +static int
> +set_callbacks(struct rte_bpf_validate_debug *debug)
> +{
> + static const struct rte_bpf_validate_debug_callback events_callback[
> + RTE_BPF_VALIDATE_DEBUG_EVENT_END] = {
> + [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_START] = {
> reset_branch_tracking },
> + [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS] = {
> step_cb },
> + [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_FAILURE] = {
> step_cb },
> + [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL] = {
> jump_conditional_cb },
> + [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS] = {
> jump_always_cb },
> + [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER] = {
> branch_enter_cb },
> + [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN] = {
> branch_return_cb },
> + [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE] = {
> branch_prune_cb },
> + [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE] = {
> branch_unreachable_cb },
> + };
> +
> + for (enum rte_bpf_validate_debug_event event = 0;
> + event < RTE_BPF_VALIDATE_DEBUG_EVENT_END;
> ++event)
> + if (events_callback[event].fn != NULL &&
> + rte_bpf_validate_debug_catch(debug, event,
> + &events_callback[event]) == NULL)
> + return -rte_errno;
> +
> + return 0;
> +}
> +
> +struct rte_bpf_validate_debug *
> +debug_create(void)
> +{
> +
> + int rc = 0;
> +
> + struct rte_bpf_validate_debug *const debug =
> rte_bpf_validate_debug_create();
> + if (debug == NULL)
> + rc = -rte_errno;
> +
> + rc = rc < 0 ? rc : set_callbacks(debug);
> +
> + rc = rc < 0 ? rc : enable_step(debug);
> +
> + if (rc < 0) {
> + debug_destroy(debug);
> + rte_errno = -rc;
> + return NULL;
> + }
> +
> + return debug;
> +}
> +
> +void
> +debug_destroy(struct rte_bpf_validate_debug *debug)
> +{
> + /* No need to destroy created points, destroying debug will do it. */
> + point_infos_free();
> + branch_stack_free();
> + rte_bpf_validate_debug_destroy(debug);
> +}
> +
> +bool
> +debug_validate_again(void)
> +{
> + return validate_again;
> +}
> diff --git a/app/validate-bpf/debug_command.c b/app/validate-
> bpf/debug_command.c
> new file mode 100644
> index 000000000000..505ae389c94f
> --- /dev/null
> +++ b/app/validate-bpf/debug_command.c
> @@ -0,0 +1,383 @@
> +/* SPDX-License-Identifier: BSD-3-Clause
> + * Copyright(c) 2025 Huawei Technologies Co., Ltd
> + */
> +
> +#include "debug_command.h"
> +
> +#include <rte_debug.h>
> +#include <cmdline.h>
> +#include <cmdline_socket.h>
> +#include <cmdline_parse_num.h>
> +
> +#include <string.h>
> +
> +
> +#define EVENT_PATTERN \
> + "invalid-state#" \
> + "branch-enter#branch-prune#branch-return#branch-unreachable#" \
> + "jump-always#jump-conditional"
> +
> +#define REGISTER_PATTERN "r0#r1#r2#r3#r4#r5#r6#r7#r8#r9#r10"
> +
> +#define COMPARISON_OP_PATTERN "==#!=#<#<=#>#>=#s<#s<=#s>#s>="
> +
> +static void
> +handle_command(void *parsed_result, struct cmdline *cl, void *data);
> +
> +/* Keywords */
> +static cmdline_parse_token_string_t break_tok =
> + TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword,
> "b#break");
> +static cmdline_parse_token_string_t may_tok =
> + TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword,
> "may");
> +static cmdline_parse_token_string_t catch_tok =
> + TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword,
> "catch");
> +static cmdline_parse_token_string_t clear_tok =
> + TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword,
> "clear");
> +static cmdline_parse_token_string_t continue_tok =
> + TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword,
> "c#continue");
> +static cmdline_parse_token_string_t delete_tok =
> + TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword,
> "delete");
> +static cmdline_parse_token_string_t info_tok =
> + TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword,
> "i#info");
> +static cmdline_parse_token_string_t info_points_tok =
> + TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword,
> + "b#break#breakpoints#points");
> +static cmdline_parse_token_string_t info_frame_tok =
> + TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword,
> "f#frame");
> +static cmdline_parse_token_string_t info_registers_tok =
> + TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword,
> "r#registers");
> +static cmdline_parse_token_string_t list_tok =
> + TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword,
> "l#list");
> +static cmdline_parse_token_string_t program_tok =
> + TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword,
> "program");
> +static cmdline_parse_token_string_t quit_tok =
> + TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword,
> "q#quit");
> +static cmdline_parse_token_string_t run_tok =
> + TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword,
> "run");
> +static cmdline_parse_token_string_t start_tok =
> + TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword,
> "start");
> +static cmdline_parse_token_string_t step_tok =
> + TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword,
> "s#step");
> +static cmdline_parse_token_string_t where_tok =
> + TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword,
> "where");
> +
> +/* Variable tokens */
> +static cmdline_parse_token_string_t comparison_lhs_tok =
> + TOKEN_STRING_INITIALIZER(struct debug_command_parsed,
> comparison.lhs, REGISTER_PATTERN);
> +static cmdline_parse_token_string_t comparison_op_tok =
> + TOKEN_STRING_INITIALIZER(struct debug_command_parsed,
> comparison.op, COMPARISON_OP_PATTERN);
> +static cmdline_parse_token_num_t comparison_rhs_literal_tok =
> + TOKEN_NUM_INITIALIZER(struct debug_command_parsed,
> comparison.literal_rhs, RTE_INT64);
> +static cmdline_parse_token_string_t comparison_rhs_register_tok =
> + TOKEN_STRING_INITIALIZER(struct debug_command_parsed,
> comparison.register_rhs,
> + REGISTER_PATTERN);
> +static cmdline_parse_token_string_t event_tok =
> + TOKEN_STRING_INITIALIZER(struct debug_command_parsed, event,
> EVENT_PATTERN);
> +static cmdline_parse_token_num_t frame_offset_tok =
> + TOKEN_NUM_INITIALIZER(struct debug_command_parsed, frame_offset,
> RTE_INT64);
> +static cmdline_parse_token_num_t instruction_count_tok =
> + TOKEN_NUM_INITIALIZER(struct debug_command_parsed,
> instruction_count, RTE_UINT32);
> +static cmdline_parse_token_num_t pc_tok =
> + TOKEN_NUM_INITIALIZER(struct debug_command_parsed,
> instruction_offset, RTE_UINT32);
> +static cmdline_parse_token_num_t point_number_tok =
> + TOKEN_NUM_INITIALIZER(struct debug_command_parsed, point_number,
> RTE_UINT32);
> +static cmdline_parse_token_string_t register_tok =
> + TOKEN_STRING_INITIALIZER(struct debug_command_parsed, register_,
> REGISTER_PATTERN);
> +
> +
> +/* Commands */
> +static cmdline_parse_inst_t cmd_break = {
> + .f = handle_command,
> + .data = (void *)DEBUG_COMMAND_BREAK,
> + .help_str = "b|break: break at current instruction",
> + .tokens = {
> + (void *)&break_tok,
> + NULL,
> + }
> +};
> +static cmdline_parse_inst_t cmd_break_pc = {
> + .f = handle_command,
> + .data = (void *)DEBUG_COMMAND_BREAK_PC,
> + .help_str = "b|break <pc>: break at specified instruction",
> + .tokens = {
> + (void *)&break_tok,
> + (void *)&pc_tok,
> + NULL,
> + }
> +};
> +static cmdline_parse_inst_t cmd_catch = {
> + .f = handle_command,
> + .data = (void *)DEBUG_COMMAND_CATCH,
> + .help_str = "catch <event>: catch specified event",
> + .tokens = {
> + (void *)&catch_tok,
> + (void *)&event_tok,
> + NULL,
> + }
> +};
> +static cmdline_parse_inst_t cmd_clear = {
> + .f = handle_command,
> + .data = (void *)DEBUG_COMMAND_CLEAR,
> + .help_str = "clear: delete all breakpoints at current instruction",
> + .tokens = {
> + (void *)&clear_tok,
> + NULL,
> + }
> +};
> +static cmdline_parse_inst_t cmd_clear_event = {
> + .f = handle_command,
> + .data = (void *)DEBUG_COMMAND_CLEAR_EVENT,
> + .help_str = "clear <event>: delete all catchpoints for specified event",
> + .tokens = {
> + (void *)&clear_tok,
> + (void *)&event_tok,
> + NULL,
> + }
> +};
> +static cmdline_parse_inst_t cmd_clear_pc = {
> + .f = handle_command,
> + .data = (void *)DEBUG_COMMAND_CLEAR_PC,
> + .help_str = "clear <pc>: delete all breakpoints at specified instruction",
> + .tokens = {
> + (void *)&clear_tok,
> + (void *)&pc_tok,
> + NULL,
> + }
> +};
> +static cmdline_parse_inst_t cmd_continue = {
> + .f = handle_command,
> + .data = (void *)DEBUG_COMMAND_CONTINUE,
> + .help_str = "c|continue: continue validation",
> + .tokens = {
> + (void *)&continue_tok,
> + NULL,
> + }
> +};
> +static cmdline_parse_inst_t cmd_delete = {
> + .f = handle_command,
> + .data = (void *)DEBUG_COMMAND_DELETE,
> + .help_str = "delete: delete all breakpoints and catchpoints",
> + .tokens = {
> + (void *)&delete_tok,
> + NULL,
> + }
> +};
> +static cmdline_parse_inst_t cmd_delete_number = {
> + .f = handle_command,
> + .data = (void *)DEBUG_COMMAND_DELETE_NUMBER,
> + .help_str = "delete <point>: delete specified breakpoint or catchpoint",
> + .tokens = {
> + (void *)&delete_tok,
> + (void *)&point_number_tok,
> + NULL,
> + }
> +};
> +static cmdline_parse_inst_t cmd_info_frame = {
> + .f = handle_command,
> + .data = (void *)DEBUG_COMMAND_INFO_FRAME,
> + .help_str = "i|info f|frame: show information about all frame locations",
> + .tokens = {
> + (void *)&info_tok,
> + (void *)&info_frame_tok,
> + NULL,
> + }
> +};
> +static cmdline_parse_inst_t cmd_info_frame_offset = {
> + .f = handle_command,
> + .data = (void *)DEBUG_COMMAND_INFO_FRAME_OFFSET,
> + .help_str = "i|info f|frame -<offset>: show information about specified
> frame location",
> + .tokens = {
> + (void *)&info_tok,
> + (void *)&info_frame_tok,
> + (void *)&frame_offset_tok,
> + NULL,
> + }
> +};
> +static cmdline_parse_inst_t cmd_info_points = {
> + .f = handle_command,
> + .data = (void *)DEBUG_COMMAND_INFO_POINTS,
> + .help_str = "i|info b|break|breakpoints|points: "
> + "show information about all breakpoints and catchpoints",
> + .tokens = {
> + (void *)&info_tok,
> + (void *)&info_points_tok,
> + NULL,
> + }
> +};
> +static cmdline_parse_inst_t cmd_info_register = {
> + .f = handle_command,
> + .data = (void *)DEBUG_COMMAND_INFO_REGISTER,
> + .help_str = "i|info <register>: show information about specified register",
> + .tokens = {
> + (void *)&info_tok,
> + (void *)®ister_tok,
> + NULL,
> + }
> +};
> +static cmdline_parse_inst_t cmd_info_registers = {
> + .f = handle_command,
> + .data = (void *)DEBUG_COMMAND_INFO_REGISTERS,
> + .help_str = "i|info r|registers: show information about all registers",
> + .tokens = {
> + (void *)&info_tok,
> + (void *)&info_registers_tok,
> + NULL,
> + }
> +};
> +static cmdline_parse_inst_t cmd_list = {
> + .f = handle_command,
> + .data = (void *)DEBUG_COMMAND_LIST,
> + .help_str = "l|list: list ten instructions",
> + .tokens = {
> + (void *)&list_tok,
> + NULL,
> + }
> +};
> +static cmdline_parse_inst_t cmd_list_count = {
> + .f = handle_command,
> + .data = (void *)DEBUG_COMMAND_LIST_COUNT,
> + .help_str = "l|list <number>: list specified number of instructions",
> + .tokens = {
> + (void *)&list_tok,
> + (void *)&instruction_count_tok,
> + NULL,
> + }
> +};
> +static cmdline_parse_inst_t cmd_list_program = {
> + .f = handle_command,
> + .data = (void *)DEBUG_COMMAND_LIST_PROGRAM,
> + .help_str = "l|list program: list whole program",
> + .tokens = {
> + (void *)&list_tok,
> + (void *)&program_tok,
> + NULL,
> + }
> +};
> +static cmdline_parse_inst_t cmd_may_literal_rhs = {
> + .f = handle_command,
> + .data = (void *)DEBUG_COMMAND_MAY_LITERAL_RHS,
> + .help_str = "may <register> <comparison> <number>: "
> + "check if specified condition _may_ be true",
> + .tokens = {
> + (void *)&may_tok,
> + (void *)&comparison_lhs_tok,
> + (void *)&comparison_op_tok,
> + (void *)&comparison_rhs_literal_tok,
> + NULL,
> + }
> +};
> +static cmdline_parse_inst_t cmd_may_register_rhs = {
> + .f = handle_command,
> + .data = (void *)DEBUG_COMMAND_MAY_REGISTER_RHS,
> + .help_str = "may <register> <comparison> <register>: "
> + "check if specified condition _may_ be true",
> + .tokens = {
> + (void *)&may_tok,
> + (void *)&comparison_lhs_tok,
> + (void *)&comparison_op_tok,
> + (void *)&comparison_rhs_register_tok,
> + NULL,
> + }
> +};
> +static cmdline_parse_inst_t cmd_quit = {
> + .f = handle_command,
> + .data = (void *)DEBUG_COMMAND_QUIT,
> + .help_str = "quit: q|quit debugger",
> + .tokens = {
> + (void *)&quit_tok,
> + NULL,
> + }
> +};
> +static cmdline_parse_inst_t cmd_run = {
> + .f = handle_command,
> + .data = (void *)DEBUG_COMMAND_RUN,
> + .help_str = "run: re-run validation from the start",
> + .tokens = {
> + (void *)&run_tok,
> + NULL,
> + }
> +};
> +static cmdline_parse_inst_t cmd_start = {
> + .f = handle_command,
> + .data = (void *)DEBUG_COMMAND_START,
> + .help_str = "start: re-start validation and stop at start",
> + .tokens = {
> + (void *)&start_tok,
> + NULL,
> + }
> +};
> +static cmdline_parse_inst_t cmd_step = {
> + .f = handle_command,
> + .data = (void *)DEBUG_COMMAND_STEP,
> + .help_str = "s|step: validate one instruction",
> + .tokens = {
> + (void *)&step_tok,
> + NULL,
> + }
> +};
> +static cmdline_parse_inst_t cmd_where = {
> + .f = handle_command,
> + .data = (void *)DEBUG_COMMAND_WHERE,
> + .help_str = "where: show current branch stack",
> + .tokens = {
> + (void *)&where_tok,
> + NULL,
> + }
> +};
> +
> +static cmdline_parse_ctx_t debug_ctx[] = {
> + &cmd_break,
> + &cmd_break_pc,
> + &cmd_catch,
> + &cmd_clear,
> + &cmd_clear_event,
> + &cmd_clear_pc,
> + &cmd_continue,
> + &cmd_delete,
> + &cmd_delete_number,
> + &cmd_info_frame,
> + &cmd_info_frame_offset,
> + &cmd_info_points,
> + &cmd_info_register,
> + &cmd_info_registers,
> + &cmd_list,
> + &cmd_list_count,
> + &cmd_list_program,
> + &cmd_may_literal_rhs,
> + &cmd_may_register_rhs,
> + &cmd_quit,
> + &cmd_run,
> + &cmd_start,
> + &cmd_step,
> + &cmd_where,
> + NULL
> +};
> +
> +/* Receive, fill and return one command. */
> +
> +static enum debug_command debug_command;
> +
> +struct debug_command_parsed debug_command_parsed;
> +
> +static void
> +handle_command(void *parsed_result, struct cmdline *cl, void *data)
> +{
> + RTE_BUILD_BUG_ON(sizeof(debug_command_parsed) >
> CMDLINE_PARSE_RESULT_BUFSIZE);
> + memcpy(&debug_command_parsed, parsed_result,
> sizeof(debug_command_parsed));
> + debug_command = (uintptr_t)data;
> + cmdline_quit(cl);
> +}
> +
> +enum debug_command
> +debug_command_get(const char *prompt)
> +{
> + debug_command = DEBUG_COMMAND_EOF;
> + struct cmdline *const cmdline = cmdline_stdin_new(debug_ctx, prompt);
> + RTE_VERIFY(cmdline != NULL);
> + cmdline_interact(cmdline);
> + cmdline_stdin_exit(cmdline);
> + /* Clear prompt, or it would prepend first message that follows. */
> + printf("\r%*s\r", (int)strlen(prompt), "");
> + fflush(stdout);
> + return debug_command;
> +}
> diff --git a/app/validate-bpf/debug_command.h b/app/validate-
> bpf/debug_command.h
> new file mode 100644
> index 000000000000..1a59ab251ebb
> --- /dev/null
> +++ b/app/validate-bpf/debug_command.h
> @@ -0,0 +1,65 @@
> +/* SPDX-License-Identifier: BSD-3-Clause
> + * Copyright(c) 2025 Huawei Technologies Co., Ltd
> + */
> +
> +#include <cmdline_parse_string.h>
> +
> +#include <stdint.h>
> +
> +
> +struct debug_command_comparison {
> + cmdline_fixed_string_t lhs;
> + cmdline_fixed_string_t op;
> + union {
> + cmdline_fixed_string_t register_rhs;
> + int64_t literal_rhs;
> + };
> +};
> +
> +struct debug_command_parsed {
> + cmdline_fixed_string_t keyword; /* Any keyword we don't need */
> + union {
> + struct debug_command_comparison comparison;
> + cmdline_fixed_string_t event;
> + int32_t frame_offset;
> + uint32_t instruction_count;
> + uint32_t instruction_offset;
> + uint32_t pc;
> + uint32_t point_number;
> + cmdline_fixed_string_t register_;
> + };
> +};
> +
> +enum debug_command {
> + DEBUG_COMMAND_EOF,
> + DEBUG_COMMAND_BREAK,
> + DEBUG_COMMAND_BREAK_PC,
> + DEBUG_COMMAND_CATCH,
> + DEBUG_COMMAND_CLEAR,
> + DEBUG_COMMAND_CLEAR_EVENT,
> + DEBUG_COMMAND_CLEAR_PC,
> + DEBUG_COMMAND_CONTINUE,
> + DEBUG_COMMAND_DELETE,
> + DEBUG_COMMAND_DELETE_NUMBER,
> + DEBUG_COMMAND_INFO_FRAME,
> + DEBUG_COMMAND_INFO_FRAME_OFFSET,
> + DEBUG_COMMAND_INFO_POINTS,
> + DEBUG_COMMAND_INFO_REGISTER,
> + DEBUG_COMMAND_INFO_REGISTERS,
> + DEBUG_COMMAND_LIST,
> + DEBUG_COMMAND_LIST_COUNT,
> + DEBUG_COMMAND_LIST_PROGRAM,
> + DEBUG_COMMAND_MAY_LITERAL_RHS,
> + DEBUG_COMMAND_MAY_REGISTER_RHS,
> + DEBUG_COMMAND_QUIT,
> + DEBUG_COMMAND_RUN,
> + DEBUG_COMMAND_START,
> + DEBUG_COMMAND_STEP,
> + DEBUG_COMMAND_WHERE,
> +};
> +
> +extern struct debug_command_parsed debug_command_parsed;
> +
> +/** Get and return one command line command, storing its data in the struct
> above. */
> +enum debug_command
> +debug_command_get(const char *prompt);
> diff --git a/app/validate-bpf/eal_init_args.c b/app/validate-bpf/eal_init_args.c
> new file mode 100644
> index 000000000000..6ce9d561e0ce
> --- /dev/null
> +++ b/app/validate-bpf/eal_init_args.c
> @@ -0,0 +1,57 @@
> +/* SPDX-License-Identifier: BSD-3-Clause
> + * Copyright(c) 2025 Huawei Technologies Co., Ltd
> + */
> +
> +#include "internal.h"
> +
> +
> +#define APP_EAL_INIT_ARG_SIZE_MAX sizeof("--log-level=lib.eal:warning")
> +
> +static const char APP_EAL_INIT_ARGS[][APP_EAL_INIT_ARG_SIZE_MAX] = {
> + "--log-level=lib.eal:warning",
> + "--no-huge",
> + "--no-pci",
> + "--no-hpet",
> + "--no-shconf",
> +};
> +
> +#define APP_EAL_INIT_ARGC (/* program name */ 1 +
> RTE_DIM(APP_EAL_INIT_ARGS))
> +
> +int
> +get_eal_init_argc(void)
> +{
> + return APP_EAL_INIT_ARGC;
> +}
> +
> +/*
> + * We cannot just return literal strings here, because rte_eal_init accepts
> + * an array of mutable pointers to mutable strings, so literals won't work.
> + * Instead we build a copy of APP_EAL_INIT_ARGS in a static mutable area.
> + */
> +char**
> +get_eal_init_argv(char *prog_name)
> +{
> + /* Static arrays for mutable copies of actual args */
> + static char mutable_args[RTE_DIM(APP_EAL_INIT_ARGS)]
> + [APP_EAL_INIT_ARG_SIZE_MAX];
> + /*
> + * Static array for pointers to args. First element will hold pointer
> + * to the prog_name, last will be set to NULL, the rest will point to
> + * elements of mutable_args with index one smaller.
> + */
> + static char *mutable_ptrs[APP_EAL_INIT_ARGC + /* terminating NULL */ 1];
> +
> + mutable_ptrs[0] = prog_name;
> + for (int argi = 0; argi != RTE_DIM(APP_EAL_INIT_ARGS); ++argi) {
> + const char * const const_arg = APP_EAL_INIT_ARGS[argi];
> + char * const mutable_arg = mutable_args[argi];
> + const int snprintf_rc = snprintf(mutable_arg,
> + APP_EAL_INIT_ARG_SIZE_MAX, "%s", const_arg);
> + RTE_VERIFY(snprintf_rc >= 0 &&
> + (size_t)snprintf_rc < APP_EAL_INIT_ARG_SIZE_MAX);
> + mutable_ptrs[1 + argi] = mutable_arg;
> + }
> + mutable_ptrs[RTE_DIM(mutable_ptrs) - 1] = NULL;
> +
> + return mutable_ptrs;
> +}
> diff --git a/app/validate-bpf/internal.h b/app/validate-bpf/internal.h
> new file mode 100644
> index 000000000000..e134e683142a
> --- /dev/null
> +++ b/app/validate-bpf/internal.h
> @@ -0,0 +1,151 @@
> +/* SPDX-License-Identifier: BSD-3-Clause
> + * Copyright(c) 2025 Huawei Technologies Co., Ltd
> + */
> +
> +#include <rte_bpf.h>
> +#include <rte_log.h>
> +
> +
> +extern int validate_bpf_logtype;
> +#define RTE_LOGTYPE_VALIDATE_BPF validate_bpf_logtype
> +#define VALIDATE_BPF_LOG(level, ...) \
> + RTE_LOG_LINE(level, VALIDATE_BPF, "" __VA_ARGS__)
> +
> +struct rte_bpf_validate_debug;
> +
> +/**
> + * List of pointers to allocations, to keep track of things to free.
> + *
> + * May contain NULLs. May contain itself directly or via an outer struct.
> + */
> +struct alloc_list {
> + size_t count;
> + void **ptrs;
> +};
> +
> +/** Add new pointer to the alloc list, return its index. */
> +size_t alloc_list_append(struct alloc_list *alloc_list, void *ptr);
> +
> +/** Replace pointer at the specified index in the alloc list. */
> +void alloc_list_replace(struct alloc_list *alloc_list, size_t index, void *ptr);
> +
> +/** Free all allocations in the alloc list */
> +void alloc_list_free_all(struct alloc_list *alloc_list);
> +
> +
> +/** Parsed program arguments */
> +struct args {
> + /* Allocations list, args.c internal use only. */
> + struct alloc_list _alloc_list;
> +
> + /* Set if command line contains --help */
> + bool show_help;
> +
> + /* BPF load parameters. */
> + struct rte_bpf_prm_ex bpf_prm;
> +
> + /* Size of the mbuf data buffer. */
> + size_t mbuf_buf_size;
> +};
> +
> +/** Print program usage information */
> +void
> +print_usage(const char *program_name);
> +
> +/** Print program defaults. */
> +void
> +print_defaults(void);
> +
> +/**
> + * Parse command-line arguments
> + *
> + * @param argc
> + * Command-line arguments count, as received by main.
> + * @param argv
> + * Command-line arguments array, as received by main.
> + * Modified during the call due to the use of getopt_long.
> + * Modifying it after the call invalidates the return value.
> + * @return
> + * - parse results in case of success;
> + * - NULL in case of an error (diagnostic will be printed to stderr)
> + */
> +struct args *
> +args_parse(int argc, char *argv[]);
> +
> +/** Destroy args struct returned by args_parse */
> +void
> +args_destroy(struct args *args);
> +
> +
> +/** Value for the rte_eal_init argc argument */
> +int
> +get_eal_init_argc(void);
> +
> +/**
> + * Value for the rte_eal_init argv argument
> + *
> + * @param prog_name
> + * Program name, can be obtained from argv[0]
> + */
> +char **
> +get_eal_init_argv(char *prog_name);
> +
> +
> +/** Print types supported in command-line. */
> +void
> +print_supported_types(void);
> +
> +/**
> + * Parse text into arg.
> + * @param arg
> + * Pointer to a variable to put parse result to.
> + * Member `buf_size` of types related to `struct rte_mbuf` is set to
> + * `RTE_MBUF_DEFAULT_BUF_SIZE`, adjust using `adjust_arg_buf_size` if
> needed.
> + * @param text
> + * Text representation of the type, e.g. "struct rte_mbuf *".
> + * @param alloc_list
> + * Alloc list to use for new allocations.
> + * @return
> + * 0 on success
> + * -1 on failure
> + */
> +int
> +parse_arg(struct rte_bpf_arg *arg, const char *text);
> +
> +/**
> + * Parse text into xsym.
> + * @param arg
> + * Pointer to a variable to put parse result to.
> + * Member `buf_size` of types related to `struct rte_mbuf` is set to
> + * `RTE_MBUF_DEFAULT_BUF_SIZE`, adjust using `adjust_xsym_buf_size` if
> needed.
> + * @param text
> + * Text representation of the external symbol, e.g. "void exit(uint32_t)".
> + * @param alloc_list
> + * Alloc list to use for new allocations.
> + * @return
> + * 0 on success
> + * -1 on failure
> + */
> +int
> +parse_xsym(struct rte_bpf_xsym *xsym, const char *text,
> + struct alloc_list *alloc_list);
> +
> +/** If arg->buf_size is non-zero, set it to mbuf_buf_size */
> +void
> +adjust_arg_buf_size(struct rte_bpf_arg *arg, size_t mbuf_buf_size);
> +
> +/** If buf_size members are non-zero, set them to mbuf_buf_size */
> +void
> +adjust_xsym_buf_size(struct rte_bpf_xsym *xsym, size_t mbuf_buf_size);
> +
> +/** Create and set up global debugging session. */
> +struct rte_bpf_validate_debug *
> +debug_create(void);
> +
> +/** Clear and destroy global debugging session. */
> +void
> +debug_destroy(struct rte_bpf_validate_debug *debug);
> +
> +/** Tells caller if validation should be re-tried. */
> +bool
> +debug_validate_again(void);
> diff --git a/app/validate-bpf/main.c b/app/validate-bpf/main.c
> new file mode 100644
> index 000000000000..c2f73993767b
> --- /dev/null
> +++ b/app/validate-bpf/main.c
> @@ -0,0 +1,90 @@
> +/* SPDX-License-Identifier: BSD-3-Clause
> + * Copyright(c) 2025 Huawei Technologies Co., Ltd
> + */
> +
> +#include "internal.h"
> +
> +#include <stdio.h>
> +#include <stdlib.h>
> +
> +#include <rte_bpf.h>
> +#include <rte_debug.h>
> +#include <rte_eal.h>
> +#include <rte_errno.h>
> +
> +RTE_LOG_REGISTER(validate_bpf_logtype, validate-bpf, NOTICE);
> +
> +static int
> +test_bpf_load(struct rte_bpf_prm_ex *prm)
> +{
> + struct rte_bpf * const bpf = rte_bpf_load_ex(prm);
> +
> + const int rc = -rte_errno;
> +
> + rte_bpf_destroy(bpf);
> +
> + return bpf == NULL ? rc : 0;
> +}
> +
> +/* Re-starts validation of asked from interactive debugger. */
> +static int
> +test_bpf_load_with_restarts(struct rte_bpf_prm_ex *prm)
> +{
> + for (;;) {
> + const int rc = test_bpf_load(prm);
> +
> + if (rc == -ECANCELED && debug_validate_again())
> + continue;
> +
> + if (rc != 0)
> + fprintf(stderr, "Error %d loading BPF: %s\n",
> + -rc, strerror(-rc));
> + else
> + fprintf(stderr, "Validation succeeded.\n");
> +
> + return rc;
> + }
> +}
> +
> +int
> +main(int argc, char *argv[])
> +{
> + int rc;
> +
> + struct args * const args = args_parse(argc, argv);
> + if (args == NULL || args->show_help
> + || args->bpf_prm.elf_file.path == NULL) {
> + args_destroy(args);
> + print_usage(argv[0]);
> + if (args == NULL)
> + /* Could not parse arguments. */
> + return 2;
> + print_supported_types();
> + print_defaults();
> + return 0;
> + }
> +
> + const int eal_init_argc = get_eal_init_argc();
> + char ** const eal_init_argv = get_eal_init_argv(argv[0]);
> + rc = rte_eal_init(eal_init_argc, eal_init_argv);
> + if (rc < 0) {
> + fprintf(stderr, "Error %d initializing EAL: %s\n",
> + rte_errno, strerror(rte_errno));
> + args_destroy(args);
> + return EXIT_FAILURE;
> + }
> + RTE_VERIFY(rc == eal_init_argc - 1);
> +
> + const int ret = test_bpf_load_with_restarts(&args->bpf_prm);
> +
> + args_destroy(args);
> +
> + rc = rte_eal_cleanup();
> + if (rc < 0) {
> + fprintf(stderr, "Error %d cleaning up EAL: %s\n",
> + rte_errno, strerror(rte_errno));
> + return EXIT_FAILURE;
> + }
> +
> + return ret < 0 ? EXIT_FAILURE : EXIT_SUCCESS;
> +}
> diff --git a/app/validate-bpf/meson.build b/app/validate-bpf/meson.build
> new file mode 100644
> index 000000000000..03cfc329efb5
> --- /dev/null
> +++ b/app/validate-bpf/meson.build
> @@ -0,0 +1,13 @@
> +# SPDX-License-Identifier: BSD-3-Clause
> +# Copyright(c) 2025 Huawei Technologies Co., Ltd
> +
> +sources = files(
> + 'alloc_list.c',
> + 'args.c',
> + 'debug.c',
> + 'debug_command.c',
> + 'eal_init_args.c',
> + 'main.c',
> + 'parse_decl.c',
> +)
> +deps = ['bpf', 'cmdline']
> diff --git a/app/validate-bpf/parse_decl.c b/app/validate-bpf/parse_decl.c
> new file mode 100644
> index 000000000000..be733d58cd41
> --- /dev/null
> +++ b/app/validate-bpf/parse_decl.c
> @@ -0,0 +1,623 @@
> +/* SPDX-License-Identifier: BSD-3-Clause
> + * Copyright(c) 2025 Huawei Technologies Co., Ltd
> + */
> +
> +#include "internal.h"
> +
> +#include <ctype.h>
> +#include <stdlib.h>
> +
> +#include <rte_ether.h>
> +#include <rte_ip.h>
> +#include <rte_mbuf_core.h>
> +#include <rte_tcp.h>
> +#include <rte_udp.h>
> +
> +#define RETURN_TEXT_ERROR(text, text_start, message, ...) do { \
> + const int _offset = (text) - (text_start); \
> + VALIDATE_BPF_LOG(ERR, "at offset %d: " message, \
> + _offset, ## __VA_ARGS__); \
> + VALIDATE_BPF_LOG(NOTICE, "%s", text_start); \
> + VALIDATE_BPF_LOG(NOTICE, "%*c", _offset + 1, '^'); \
> + return -1; \
> +} while (0)
> +
> +/* Used in place of any signature, this is not really being checked. */
> +static uint64_t
> +dummy_function(uint64_t arg1, uint64_t arg2, uint64_t arg3, uint64_t arg4,
> + uint64_t arg5)
> +{
> + RTE_SET_USED(arg1);
> + RTE_SET_USED(arg2);
> + RTE_SET_USED(arg3);
> + RTE_SET_USED(arg4);
> + RTE_SET_USED(arg5);
> + return 0;
> +}
> +
> +
> +/* TOKENS AND TYPES */
> +
> +enum token {
> + TOKEN_UNRECOGNIZED = -1,
> + TOKEN_END = 0,
> +
> + TOKEN_ASTERISK,
> + TOKEN_BRACKET_CLOSE,
> + TOKEN_BRACKET_OPEN,
> + TOKEN_COMMA,
> + TOKEN_PARENTHESIS_CLOSE,
> + TOKEN_PARENTHESIS_OPEN,
> + TOKEN_STRUCT,
> +
> + TOKEN_TYPES_BEGIN,
> +
> + TOKEN_TYPE_CHAR,
> + TOKEN_TYPE_ETHER_HEADER,
> + TOKEN_TYPE_INT32_T,
> + TOKEN_TYPE_IP_HEADER,
> + TOKEN_TYPE_IP_HEADERS,
> + TOKEN_TYPE_RTE_MBUF,
> + TOKEN_TYPE_TCP_HEADER,
> + TOKEN_TYPE_TCP_HEADERS,
> + TOKEN_TYPE_UDP_HEADER,
> + TOKEN_TYPE_UDP_HEADERS,
> + TOKEN_TYPE_UINT32_T,
> + TOKEN_TYPE_UINT64_T,
> + TOKEN_TYPE_UINTPTR_T,
> + TOKEN_TYPE_VOID,
> +
> + TOKEN_TYPES_END,
> +};
> +
> +/* Return true if the token is a type token */
> +static bool
> +is_type_token(enum token token)
> +{
> + return token > TOKEN_TYPES_BEGIN && token < TOKEN_TYPES_END;
> +}
> +
> +struct text_token {
> + const char *text;
> + unsigned int length; /* Not size_t to fit this struct in 2 regs. */
> + enum token token;
> +};
> +
> +#define TEXT_TOKEN_DEF(text_, token_) { \
> + .text = text_, \
> + .length = sizeof(text_) - 1, \
> + .token = token_, \
> +}
> +
> +/* Token search table, MUST BE SORTED! */
> +const struct text_token TEXT_TOKENS[] = {
> + TEXT_TOKEN_DEF("(", TOKEN_PARENTHESIS_OPEN),
> + TEXT_TOKEN_DEF(")", TOKEN_PARENTHESIS_CLOSE),
> + TEXT_TOKEN_DEF("*", TOKEN_ASTERISK),
> + TEXT_TOKEN_DEF(",", TOKEN_COMMA),
> + TEXT_TOKEN_DEF("[", TOKEN_BRACKET_OPEN),
> + TEXT_TOKEN_DEF("]", TOKEN_BRACKET_CLOSE),
> + TEXT_TOKEN_DEF("char", TOKEN_TYPE_CHAR),
> + TEXT_TOKEN_DEF("ether_header",
> TOKEN_TYPE_ETHER_HEADER),
> + TEXT_TOKEN_DEF("int32_t", TOKEN_TYPE_INT32_T),
> + TEXT_TOKEN_DEF("ip_header", TOKEN_TYPE_IP_HEADER),
> + TEXT_TOKEN_DEF("ip_headers",
> TOKEN_TYPE_IP_HEADERS),
> + TEXT_TOKEN_DEF("rte_ether_hdr",
> TOKEN_TYPE_ETHER_HEADER),
> + TEXT_TOKEN_DEF("rte_ipv4_hdr",
> TOKEN_TYPE_IP_HEADER),
> + TEXT_TOKEN_DEF("rte_mbuf", TOKEN_TYPE_RTE_MBUF),
> + TEXT_TOKEN_DEF("rte_tcp_hdr",
> TOKEN_TYPE_TCP_HEADER),
> + TEXT_TOKEN_DEF("rte_udp_hdr",
> TOKEN_TYPE_UDP_HEADER),
> + TEXT_TOKEN_DEF("struct", TOKEN_STRUCT),
> + TEXT_TOKEN_DEF("tcp_header",
> TOKEN_TYPE_TCP_HEADER),
> + TEXT_TOKEN_DEF("tcp_headers",
> TOKEN_TYPE_TCP_HEADERS),
> + TEXT_TOKEN_DEF("udp_header",
> TOKEN_TYPE_UDP_HEADER),
> + TEXT_TOKEN_DEF("udp_headers",
> TOKEN_TYPE_UDP_HEADERS),
> + TEXT_TOKEN_DEF("uint32_t", TOKEN_TYPE_UINT32_T),
> + TEXT_TOKEN_DEF("uint64_t", TOKEN_TYPE_UINT64_T),
> + TEXT_TOKEN_DEF("uintptr_t", TOKEN_TYPE_UINTPTR_T),
> + TEXT_TOKEN_DEF("void", TOKEN_TYPE_VOID),
> +};
> +
> +const struct text_token TEXT_TOKEN_UNRECOGNIZED = { "", 0,
> TOKEN_UNRECOGNIZED };
> +const struct text_token TEXT_TOKEN_END = { "", 0, TOKEN_END };
> +
> +/* IP header maximum possible size with options. */
> +#define IP_HEADER_MAX_SIZE 60
> +#define IP_HEADERS_MAX_SIZE (sizeof(struct rte_ether_hdr) +
> IP_HEADER_MAX_SIZE)
> +#define TCP_HEADERS_MAX_SIZE (IP_HEADERS_MAX_SIZE + sizeof(struct
> rte_tcp_hdr))
> +#define UDP_HEADERS_MAX_SIZE (IP_HEADERS_MAX_SIZE + sizeof(struct
> rte_udp_hdr))
> +
> +#define POINTER_SIZE (sizeof(char *))
> +
> +const size_t TYPE_TOKEN_SIZE[] = {
> + [TOKEN_TYPE_CHAR] = sizeof(char),
> + [TOKEN_TYPE_ETHER_HEADER] = sizeof(struct rte_ether_hdr),
> + [TOKEN_TYPE_INT32_T] = sizeof(int32_t),
> + [TOKEN_TYPE_IP_HEADERS] = IP_HEADERS_MAX_SIZE,
> + [TOKEN_TYPE_IP_HEADER] = sizeof(struct rte_ipv4_hdr),
> + [TOKEN_TYPE_TCP_HEADERS] = TCP_HEADERS_MAX_SIZE,
> + [TOKEN_TYPE_TCP_HEADER] = sizeof(struct rte_tcp_hdr),
> + [TOKEN_TYPE_UDP_HEADERS] = UDP_HEADERS_MAX_SIZE,
> + [TOKEN_TYPE_UDP_HEADER] = sizeof(struct rte_udp_hdr),
> + [TOKEN_TYPE_UINT32_T] = sizeof(uint32_t),
> + [TOKEN_TYPE_UINT64_T] = sizeof(uint64_t),
> + [TOKEN_TYPE_UINTPTR_T] = sizeof(uintptr_t),
> +};
> +
> +/*
> + * Struct rte_bpf_arg augmented with a type of pointer to it, since rte_bpf_arg
> + * by itself may not contain enough information to determine its pointer type.
> + */
> +struct arg_info {
> + struct rte_bpf_arg value;
> + enum rte_bpf_arg_type ptr_type;
> + bool is_array;
> +};
> +
> +/* Return arg_info struct described by the specified type token. */
> +static struct arg_info
> +get_type_token_arg(enum token type)
> +{
> + RTE_ASSERT(is_type_token(type));
> + switch (type) {
> + case TOKEN_TYPE_VOID:
> + return (struct arg_info){
> + .value = { .type = RTE_BPF_ARG_UNDEF },
> + .ptr_type = RTE_BPF_ARG_PTR,
> + };
> + case TOKEN_TYPE_RTE_MBUF:
> + return (struct arg_info){
> + .value = {
> + .type = RTE_BPF_ARG_RAW,
> + .size = sizeof(struct rte_mbuf),
> + .buf_size = RTE_MBUF_DEFAULT_BUF_SIZE,
> + },
> + .ptr_type = RTE_BPF_ARG_PTR_MBUF,
> + };
> + default:
> + /* Should only reach here for normal sized types. */
> + RTE_ASSERT((size_t)type < RTE_DIM(TYPE_TOKEN_SIZE) &&
> + TYPE_TOKEN_SIZE[type] != 0);
> + return (struct arg_info){
> + .value = {
> + .type = RTE_BPF_ARG_RAW,
> + .size = TYPE_TOKEN_SIZE[type],
> + },
> + .ptr_type = RTE_BPF_ARG_PTR,
> + };
> + }
> +}
> +
> +
> +/* PARSING TEXT INTO TOKENS */
> +
> +/* Return true if character matches [a-zA-Z0-9_] in regex */
> +static bool
> +iswordchar(char character)
> +{
> + return isalnum(character) || character == '_';
> +}
> +
> +/*
> + * Compare pointer to text with pointer to struct text_token to determine
> + * if text starts with the specified token.
> + */
> +static int
> +text_and_text_token_cmp(const void *text_void_ptr, const void
> *text_token_void_ptr)
> +{
> + int result;
> + const char * const text = text_void_ptr;
> + const struct text_token * const text_token = text_token_void_ptr;
> +
> + if (memchr(text, 0, text_token->length) != NULL)
> + /* Text is shorter than the token. */
> + return strcmp(text, text_token->text);
> +
> + /* Cannot use strcmp because we are looking for a prefix. */
> + result = memcmp(text, text_token->text, text_token->length);
I am puzzled here..
Why not simply strncmp(text, text_token->text, text_token->length);
?
> +
> + /* Checking the case of a partial word match. */
> + if (result == 0 && iswordchar(text[text_token->length - 1]) &&
> + iswordchar(text[text_token->length]))
> + /* Text word is longer than the token. */
> + result = 1;
> +
> + return result;
> +}
> +
> +/* Advance pointed character pointer to the first non-space character. */
> +static void
> +skip_space(const char **text_ptr)
> +{
> + while (isspace(**text_ptr))
> + ++*text_ptr;
> +}
> +
> +/*
> + * Recognize and return a token starting text.
> + * Return TEXT_TOKEN_END if text is empty.
> + * Return TEXT_TOKEN_UNRECOGNIZED if text starts with unknown token.
> + */
> +static struct text_token
> +peek_text_token(const char *text)
> +{
> + if (*text == '\0')
> + return TEXT_TOKEN_END;
> + const struct text_token * const text_token = bsearch(text, TEXT_TOKENS,
> + RTE_DIM(TEXT_TOKENS), sizeof(TEXT_TOKENS[0]),
> + text_and_text_token_cmp);
> + if (text_token == NULL)
> + return TEXT_TOKEN_UNRECOGNIZED;
> + return *text_token;
> +}
> +
> +/*
> + * Advance pointed text starting with the specified token to the first
> + * non-space character after it.
> + * Do nothing if called for TEXT_TOKEN_END or TEXT_TOKEN_UNRECOGNIZED.
> + */
> +static void
> +consume_text_token(const char **text_ptr, struct text_token text_token)
> +{
> + RTE_ASSERT(memcmp(*text_ptr, text_token.text, text_token.length) == 0);
> + *text_ptr += text_token.length;
> + skip_space(text_ptr);
> +}
> +
> +/* Return length of the word starting at `text`. */
> +static size_t
> +find_word_length(const char *text)
> +{
> + size_t word_length = 0;
> + while (iswordchar(text[word_length]))
> + ++word_length;
> + return word_length;
> +}
> +
> +/* Create a copy of the word starting text, advance to next token. */
> +static const char *
> +take_name(const char **text_ptr, struct alloc_list *alloc_list)
> +{
> + const size_t word_length = find_word_length(*text_ptr);
> + if (word_length == 0)
> + /* Text does not start with a word. */
> + return NULL;
> +
> + /* Allocate memory for the word and add it to the alloc_list */
> + char *word = malloc(word_length + 1);
> + RTE_VERIFY(word != NULL);
> + alloc_list_append(alloc_list, word);
> +
> + /* Copy and terminate word contents. */
> + memcpy(word, *text_ptr, word_length);
> + word[word_length] = '\0';
> +
> + /* Advance text pointer. */
> + *text_ptr += word_length;
> + skip_space(text_ptr);
> +
> + return word;
> +}
> +
> +/* Read a number starting text, advance to next token. */
> +static int
> +take_number(size_t *number, const char **text_ptr)
> +{
> + /* Read a word and let strtoull to decide if it's a number. */
> + const size_t word_length = find_word_length(*text_ptr);
> + if (word_length == 0)
> + /* Text does not start with a word. */
> + return -ENOENT;
> +
> + errno = 0;
> + char *number_end;
> + unsigned long long long_number = strtoull(*text_ptr, &number_end, 0);
> + if (errno > 0)
> + return -errno;
> + if (number_end != *text_ptr + word_length)
> + /* Could not parse whole word. */
> + return -EINVAL;
> + if (long_number > SIZE_MAX)
I think stroul() will return ERANGE itself if the resulting value iss out of range.
So not sure that precaution (strtoull()l then check range is necessary).
> + return -ERANGE;
> +
> + *number = long_number;
> +
> + /* Advance text pointer. */
> + *text_ptr += word_length;
> + skip_space(text_ptr);
> +
> + return 0;
> +}
> +
> +
> +/* PARSING DECLARATION PARTS */
> +
> +/* Change arg into a reference. */
> +static void
> +change_into_reference(struct arg_info *arg)
> +{
> + if (RTE_BPF_ARG_PTR_TYPE(arg->value.type) != 0) {
> + VALIDATE_BPF_LOG(WARNING,
> + "After taking reference to a pointer the latter "
> + "will be described as an opaque pointer-size blob.");
> + RTE_ASSERT(arg->ptr_type == RTE_BPF_ARG_PTR);
> + arg->value.size = POINTER_SIZE;
> + }
> + arg->value.type = arg->ptr_type;
> + arg->ptr_type = RTE_BPF_ARG_PTR;
> + arg->is_array = false;
> +}
> +
> +/*
> + * Recognize and consume arg starting text, advance to next token.
> + */
> +static int
> +take_arg(struct arg_info *arg, const char **text_ptr, const char *text_start)
> +{
> + struct text_token next = peek_text_token(*text_ptr);
> +
> + if (next.token == TOKEN_STRUCT) {
> + consume_text_token(text_ptr, next);
> + next = peek_text_token(*text_ptr);
> + }
> +
> + if (!is_type_token(next.token))
> + RETURN_TEXT_ERROR(*text_ptr, text_start, "expect type");
> + const enum token type = next.token;
> + consume_text_token(text_ptr, next);
> + next = peek_text_token(*text_ptr);
> +
> + *arg = get_type_token_arg(type);
> +
> + while (next.token == TOKEN_ASTERISK) {
> + consume_text_token(text_ptr, next);
> + next = peek_text_token(*text_ptr);
> + change_into_reference(arg);
> + }
> +
> + while (next.token == TOKEN_BRACKET_OPEN) {
> + consume_text_token(text_ptr, next);
> +
> + /* Initialize to zero to avoid spurious compiler warnings. */
> + size_t array_length = 0;
> + if (take_number(&array_length, text_ptr) < 0)
> + RETURN_TEXT_ERROR(*text_ptr, text_start, "expect
> length");
> + next = peek_text_token(*text_ptr);
> +
> + if (arg->value.size != 0 &&
> + array_length > SIZE_MAX / arg->value.size)
> + RETURN_TEXT_ERROR(*text_ptr, text_start, "type too
> big");
> +
> + if (next.token != TOKEN_BRACKET_CLOSE)
> + RETURN_TEXT_ERROR(*text_ptr, text_start, "expect ']'");
> + consume_text_token(text_ptr, next);
> + next = peek_text_token(*text_ptr);
> +
> + change_into_reference(arg);
> + arg->value.size *= array_length;
> + arg->is_array = true;
> + }
> +
> + return 0;
> +}
> +
> +/* Fill struct rte_bpf_arg within xsym trying not to unzero the padding. */
> +static void
> +fill_xsym_arg(struct rte_bpf_arg *target, struct rte_bpf_arg source)
> +{
> + /* Copy fields individually to try and prevent copying the padding. */
> + target->type = source.type;
> + target->size = source.size;
> + target->buf_size = source.buf_size;
> +}
> +
> +/* Build and return struct rte_bpf_arg of type RTE_BPF_XTYPE_VAR */
> +static int
> +fill_var_xsym(struct rte_bpf_xsym *xsym, struct arg_info arg, const char *name,
> + struct alloc_list *alloc_list)
> +{
> + /* Variables are passed by reference, except for arrays. */
> + if (!arg.is_array) {
> + if (RTE_BPF_ARG_PTR_TYPE(arg.value.type) != 0)
> + VALIDATE_BPF_LOG(WARNING,
> + "External pointers may not work as expected "
> + "because all external variables are passed by "
> + "reference but there is currently no way to "
> + "describe double pointer to the validator.");
> + change_into_reference(&arg);
> + }
> +
> + /* Allocate something to assign to a val pointer. */
> + void * const val = calloc(1, RTE_MAX(1u, arg.value.size));
> + if (val == NULL) {
> + VALIDATE_BPF_LOG(ERR, "could not allocate enough memory");
> + return -1;
> + }
> + alloc_list_append(alloc_list, val);
> +
> + /* Need all padding and unused fields to be zero-filled. */
> + memset(xsym, 0, sizeof(*xsym));
> + xsym->name = name;
> + xsym->type = RTE_BPF_XTYPE_VAR;
> + xsym->var.val = val;
> + fill_xsym_arg(&xsym->var.desc, arg.value);
> +
> + return 0;
> +}
> +
> +/* Build and return struct rte_bpf_arg of type RTE_BPF_XTYPE_FUNC */
> +static void
> +fill_func_xsym(struct rte_bpf_xsym *xsym, struct arg_info arg, const char *name)
> +{
> + /* Need all padding and unused fields to be zero-filled. */
> + memset(xsym, 0, sizeof(*xsym));
> + xsym->name = name;
> + xsym->type = RTE_BPF_XTYPE_FUNC;
> + xsym->func.val = &dummy_function;
> + fill_xsym_arg(&xsym->func.ret, arg.value);
> +}
> +
> +/*
> + * Parse and store function arguments, advance to next token after ')'.
> + * Value of *text_ptr should point to the next token after '('.
> + */
> +static int
> +take_func_xsym_args(struct rte_bpf_xsym *xsym, const char **text_ptr,
> + const char *text_start)
> +{
> + struct arg_info arg;
> + struct text_token delimiter = peek_text_token(*text_ptr);
> +
> + while (delimiter.token != TOKEN_PARENTHESIS_CLOSE) {
> + if (xsym->func.nb_args == EBPF_FUNC_MAX_ARGS)
> + RETURN_TEXT_ERROR(*text_ptr, text_start,
> + "too many arguments, maximum %d allowed",
> + EBPF_FUNC_MAX_ARGS);
> +
> + if (take_arg(&arg, text_ptr, text_start) < 0)
> + return -1;
> + if (arg.value.type == RTE_BPF_ARG_UNDEF &&
> + xsym->func.nb_args != 0)
> + RETURN_TEXT_ERROR(*text_ptr, text_start,
> + "arguments of type void are not allowed");
> + fill_xsym_arg(&xsym->func.args[xsym->func.nb_args++],
> + arg.value);
> +
> + delimiter = peek_text_token(*text_ptr);
> + switch (delimiter.token) {
> + case TOKEN_COMMA:
> + consume_text_token(text_ptr, delimiter);
> + continue;
> + case TOKEN_PARENTHESIS_CLOSE:
> + break;
> + default:
> + RETURN_TEXT_ERROR(*text_ptr, text_start,
> + "expect ')' or ','");
> + }
> + }
> + consume_text_token(text_ptr, delimiter);
> +
> + /* Special case of single void argument. */
> + if (xsym->func.nb_args == 1 &&
> + xsym->func.args[0].type == RTE_BPF_ARG_UNDEF) {
> + xsym->func.nb_args = 0;
> + /* Need all padding and unused fields to be zero-filled. */
> + fill_xsym_arg(&xsym->func.args[0], (struct rte_bpf_arg){});
> + }
> +
> + return 0;
> +}
> +
> +/* Make sure text has ended. */
> +static int
> +ensure_end(const char *text, const char *text_start)
> +{
> + if (peek_text_token(text).token != TOKEN_END)
> + RETURN_TEXT_ERROR(text, text_start, "trailing garbage");
> + return 0;
> +}
> +
> +/* Parse and store xsym, advance to next token. */
> +static int
> +take_xsym(struct rte_bpf_xsym *xsym, const char **text_ptr,
> + const char *text_start, struct alloc_list *alloc_list)
> +{
> + struct arg_info arg;
> +
> + if (take_arg(&arg, text_ptr, text_start) < 0)
> + return -1;
> +
> + const char * const name = take_name(text_ptr, alloc_list);
> + if (name == NULL)
> + RETURN_TEXT_ERROR(*text_ptr, text_start, "expect name");
> +
> + const struct text_token next = peek_text_token(*text_ptr);
> + switch (next.token) {
> + case TOKEN_END:
> + if (fill_var_xsym(xsym, arg, name, alloc_list) < 0)
> + return -1;
> + break;
> + case TOKEN_PARENTHESIS_OPEN:
> + consume_text_token(text_ptr, next);
> + fill_func_xsym(xsym, arg, name);
> + if (take_func_xsym_args(xsym, text_ptr, text_start) < 0)
> + return -1;
> + break;
> + default:
> + RETURN_TEXT_ERROR(*text_ptr, text_start,
> + "expect '(' or text end");
> + }
> + return 0;
> +}
> +
> +
> +/* PUBLIC FUNCTIONS */
> +
> +void
> +print_supported_types(void)
> +{
> + printf("TYPE: [struct] BASIC_TYPE [*]... [[N]]...\n");
> + printf("BASIC_TYPE: one of\n");
> + for (int tti = 0; tti != RTE_DIM(TEXT_TOKENS); ++tti) {
> + if (is_type_token(TEXT_TOKENS[tti].token))
> + printf("\t%s\n", TEXT_TOKENS[tti].text);
> + }
> +}
> +
> +int
> +parse_arg(struct rte_bpf_arg *arg, const char *text)
> +{
> + struct arg_info arg_info;
> + const char * const text_start = text;
> + skip_space(&text);
> + if (take_arg(&arg_info, &text, text_start) < 0)
> + return -1;
> + if (ensure_end(text, text_start) < 0)
> + return -1;
> + if (arg_info.ptr_type != RTE_BPF_ARG_PTR)
> + VALIDATE_BPF_LOG(WARNING,
> + "`%s` has a special pointer type which was left unused; "
> + "argument was set to an opaque blob.",
> + text_start);
> + *arg = arg_info.value;
> + return 0;
> +}
> +
> +int
> +parse_xsym(struct rte_bpf_xsym *xsym, const char *text,
> + struct alloc_list *alloc_list)
> +{
> + const char * const text_start = text;
> + skip_space(&text);
> + if (take_xsym(xsym, &text, text_start, alloc_list) < 0)
> + return -1;
> + if (ensure_end(text, text_start) < 0)
> + return -1;
> + return 0;
> +}
> +
> +void
> +adjust_arg_buf_size(struct rte_bpf_arg *arg, size_t mbuf_buf_size)
> +{
> + if (arg->buf_size != 0)
> + /* Non-zero buf_size indicates mbuf or a pointer to it. */
> + arg->buf_size = mbuf_buf_size;
> +}
> +
> +void
> +adjust_xsym_buf_size(struct rte_bpf_xsym *xsym, size_t mbuf_buf_size)
> +{
> + switch (xsym->type) {
> + case RTE_BPF_XTYPE_FUNC:
> + for (uint32_t argi = 0; argi != xsym->func.nb_args; ++argi)
> + adjust_arg_buf_size(&xsym->func.args[argi],
> + mbuf_buf_size);
> + adjust_arg_buf_size(&xsym->func.ret, mbuf_buf_size);
> + break;
> + case RTE_BPF_XTYPE_VAR:
> + adjust_arg_buf_size(&xsym->var.desc, mbuf_buf_size);
> + break;
> + default:
> + rte_panic("Unexpected xsym type %d\n", xsym->type);
> + }
> +}
^ permalink raw reply [flat|nested] 38+ messages in thread
* [PATCH v4 0/7] validate-bpf: add standalone validation debug tool
2026-09-15 19:26 ` [PATCH v3 " Marat Khalili
` (6 preceding siblings ...)
2026-09-15 19:26 ` [PATCH v3 7/7] app/validate-bpf: add BPF validation application Marat Khalili
@ 2026-09-17 14:35 ` Marat Khalili
2026-09-17 14:35 ` [PATCH v4 1/7] bpf/validate: fix finished status on restart Marat Khalili
` (7 more replies)
7 siblings, 8 replies; 38+ messages in thread
From: Marat Khalili @ 2026-09-17 14:35 UTC (permalink / raw)
Cc: dev, Marat Khalili, Stephen Hemminger, Konstantin Ananyev,
Marat Khalili
This patchset introduces a new standalone tool for pre-validating eBPF
programs and debugging validation problems. Its should allow developers
to trace validator state changes per instruction and understand verifier
decisions, simplifying the debugging of rejected programs before loading
them into a real application context.
v4:
* addressed reviewer feedback from Konstantin Ananyev:
* changed variable and array names to lowercase to follow conventions;
* replaced strtoull()/strtoumax() with strtoul() and removed redundant
SIZE_MAX checks;
* changed return types of point_infos_destroy_* functions to void;
* replaced %jd and (intmax_t) cast with PRId32 when printing frame
offsets;
* refactored the inverted loop in print_frame() for clarity;
* replaced return -ENOMEM with RTE_VERIFY() for allocations in debug.c
print functions;
* simplified token matching logic in parse_decl.c by using strncmp().
v3:
* addressed another FreeBSD issue (spurious uninitialized warning);
* addressed more AI comments:
* fail if library fails to set step catchpoint on start command;
* fail nicer if external variables are too big;
* fail nicer if EAL initialization or cleanup fails;
* explain why 9 steps are expeced in the test;
* explain that non-zero buf_size marks mbuf;
* avoid symbols starting with `RTE_` in the app;
v2:
* for FreeBSD compatibility switched from Linux to DPDK network structs;
* addressed small AI comments:
* clarified patch 1 commit message;
* renamed `rte_validate_bpf_logtype` to `validate_bpf_logtype`;
* corrected comment to an internal-use-only function to reflect that
next-after-last value of the program counter is no longer allowed;
Marat Khalili (7):
bpf/validate: fix finished status on restart
bpf/validate: refactor internal step function
bpf/validate: formalize call back requirements
bpf/validate: add jump notification events
bpf/validate: add get current event API
app/test: add test for bpf validate debug events
app/validate-bpf: add BPF validation application
MAINTAINERS | 2 +
app/meson.build | 1 +
app/test/test_bpf_validate.c | 344 +++++++-
app/validate-bpf/alloc_list.c | 53 ++
app/validate-bpf/args.c | 262 ++++++
app/validate-bpf/debug.c | 1087 ++++++++++++++++++++++++
app/validate-bpf/debug_command.c | 383 +++++++++
app/validate-bpf/debug_command.h | 65 ++
app/validate-bpf/eal_init_args.c | 57 ++
app/validate-bpf/internal.h | 151 ++++
app/validate-bpf/main.c | 90 ++
app/validate-bpf/meson.build | 13 +
app/validate-bpf/parse_decl.c | 616 ++++++++++++++
doc/guides/rel_notes/release_26_11.rst | 5 +
doc/guides/tools/index.rst | 1 +
doc/guides/tools/validate_bpf.rst | 97 +++
lib/bpf/bpf_validate.c | 49 +-
lib/bpf/bpf_validate_debug.c | 92 +-
lib/bpf/bpf_validate_debug.h | 12 +-
lib/bpf/rte_bpf_validate_debug.h | 42 +-
20 files changed, 3356 insertions(+), 66 deletions(-)
create mode 100644 app/validate-bpf/alloc_list.c
create mode 100644 app/validate-bpf/args.c
create mode 100644 app/validate-bpf/debug.c
create mode 100644 app/validate-bpf/debug_command.c
create mode 100644 app/validate-bpf/debug_command.h
create mode 100644 app/validate-bpf/eal_init_args.c
create mode 100644 app/validate-bpf/internal.h
create mode 100644 app/validate-bpf/main.c
create mode 100644 app/validate-bpf/meson.build
create mode 100644 app/validate-bpf/parse_decl.c
create mode 100644 doc/guides/tools/validate_bpf.rst
--
2.43.0
^ permalink raw reply [flat|nested] 38+ messages in thread
* [PATCH v4 1/7] bpf/validate: fix finished status on restart
2026-09-17 14:35 ` [PATCH v4 0/7] validate-bpf: add standalone validation debug tool Marat Khalili
@ 2026-09-17 14:35 ` Marat Khalili
2026-09-17 14:35 ` [PATCH v4 2/7] bpf/validate: refactor internal step function Marat Khalili
` (6 subsequent siblings)
7 siblings, 0 replies; 38+ messages in thread
From: Marat Khalili @ 2026-09-17 14:35 UTC (permalink / raw)
To: Konstantin Ananyev, Marat Khalili; +Cc: dev, Marat Khalili, Stephen Hemminger
From: Marat Khalili <marat.khalili@huawei.com>
If debug session is reused, requesting evaluation result while
evaluation is still in progress would successfully return result from
the previous evaluation instead of expected -EAGAIN error.
Make evaluation start clear the finished status.
Fixes: 6b0da0602cc7 ("bpf/validate: introduce debugging interface")
Signed-off-by: Marat Khalili <marat.khalili@huawei.com>
---
lib/bpf/bpf_validate_debug.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/lib/bpf/bpf_validate_debug.c b/lib/bpf/bpf_validate_debug.c
index 5d18804a74..9a03976632 100644
--- a/lib/bpf/bpf_validate_debug.c
+++ b/lib/bpf/bpf_validate_debug.c
@@ -568,6 +568,7 @@ __rte_bpf_validate_debug_evaluate_start(struct rte_bpf_validate_debug *debug,
debug->last_point = NULL;
debug->pc = 0;
debug->evaluate_started = true;
+ debug->evaluate_finished = false;
const int rc = debug_send_event(debug,
RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_START);
--
2.43.0
^ permalink raw reply related [flat|nested] 38+ messages in thread
* [PATCH v4 2/7] bpf/validate: refactor internal step function
2026-09-17 14:35 ` [PATCH v4 0/7] validate-bpf: add standalone validation debug tool Marat Khalili
2026-09-17 14:35 ` [PATCH v4 1/7] bpf/validate: fix finished status on restart Marat Khalili
@ 2026-09-17 14:35 ` Marat Khalili
2026-09-17 14:35 ` [PATCH v4 3/7] bpf/validate: formalize call back requirements Marat Khalili
` (5 subsequent siblings)
7 siblings, 0 replies; 38+ messages in thread
From: Marat Khalili @ 2026-09-17 14:35 UTC (permalink / raw)
To: Konstantin Ananyev, Marat Khalili; +Cc: dev, Marat Khalili, Stephen Hemminger
From: Marat Khalili <marat.khalili@huawei.com>
Rename `__rte_bpf_validate_debug_evaluate_step` to
`__rte_bpf_validate_debug_evaluate_update` to better reflect planned
usage, and make it accept a bitmask of events instead of a single event.
Step event is still implicitly triggered for now.
No functional changes.
Signed-off-by: Marat Khalili <marat.khalili@huawei.com>
---
lib/bpf/bpf_validate.c | 17 ++++++++---------
lib/bpf/bpf_validate_debug.c | 20 +++++++++++---------
lib/bpf/bpf_validate_debug.h | 11 ++++++-----
3 files changed, 25 insertions(+), 23 deletions(-)
diff --git a/lib/bpf/bpf_validate.c b/lib/bpf/bpf_validate.c
index f9960088a2..b317abb811 100644
--- a/lib/bpf/bpf_validate.c
+++ b/lib/bpf/bpf_validate.c
@@ -2989,10 +2989,9 @@ evaluate(struct bpf_verifier *bvf)
break;
}
- rc = __rte_bpf_validate_debug_evaluate_step(debug, idx,
+ rc = __rte_bpf_validate_debug_evaluate_update(debug, idx,
prev_nb_edge > 1 ?
- RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER :
- RTE_BPF_VALIDATE_DEBUG_EVENT_STEP);
+ RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER) : 0);
if (rc < 0)
break;
@@ -3024,9 +3023,9 @@ evaluate(struct bpf_verifier *bvf)
}
if (bvf->evst->unreachable) {
- rc = __rte_bpf_validate_debug_evaluate_step(
+ rc = __rte_bpf_validate_debug_evaluate_update(
debug, get_node_idx(bvf, next),
- RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE);
+ RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE));
if (rc < 0)
break;
@@ -3038,9 +3037,9 @@ evaluate(struct bpf_verifier *bvf)
*/
} else if (node->nb_edge > 1 &&
prune_eval_state(bvf, node, next) == 0) {
- rc = __rte_bpf_validate_debug_evaluate_step(
+ rc = __rte_bpf_validate_debug_evaluate_update(
debug, get_node_idx(bvf, next),
- RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE);
+ RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE));
if (rc < 0)
break;
@@ -3059,9 +3058,9 @@ evaluate(struct bpf_verifier *bvf)
*/
if (prev_nb_edge != 0) {
- rc = __rte_bpf_validate_debug_evaluate_step(
+ rc = __rte_bpf_validate_debug_evaluate_update(
debug, get_node_idx(bvf, node) + 1,
- RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN);
+ RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN));
if (rc < 0)
break;
}
diff --git a/lib/bpf/bpf_validate_debug.c b/lib/bpf/bpf_validate_debug.c
index 9a03976632..f76303a7fd 100644
--- a/lib/bpf/bpf_validate_debug.c
+++ b/lib/bpf/bpf_validate_debug.c
@@ -582,8 +582,8 @@ __rte_bpf_validate_debug_evaluate_start(struct rte_bpf_validate_debug *debug,
}
int
-__rte_bpf_validate_debug_evaluate_step(struct rte_bpf_validate_debug *debug,
- uint32_t pc, debug_event_t event)
+__rte_bpf_validate_debug_evaluate_update(struct rte_bpf_validate_debug *debug,
+ uint32_t pc, uint64_t events)
{
int rc;
@@ -595,8 +595,7 @@ __rte_bpf_validate_debug_evaluate_step(struct rte_bpf_validate_debug *debug,
return -ECHILD;
}
- if (pc > debug->bpf_prm->raw.nb_ins || event < 0 ||
- event >= RTE_BPF_VALIDATE_DEBUG_EVENT_END)
+ if (pc > debug->bpf_prm->raw.nb_ins)
return -EINVAL;
debug->pc = pc;
@@ -606,11 +605,13 @@ __rte_bpf_validate_debug_evaluate_step(struct rte_bpf_validate_debug *debug,
rc = debug_send_event(debug,
RTE_BPF_VALIDATE_DEBUG_EVENT_INVALID_STATE);
- if (event != RTE_BPF_VALIDATE_DEBUG_EVENT_STEP)
- rc = rc < 0 ? rc : debug_send_event(debug, event);
+ for (enum rte_bpf_validate_debug_event event = 0;
+ event != RTE_BPF_VALIDATE_DEBUG_EVENT_END; ++event)
+ if (events & RTE_BIT64(event))
+ rc = rc < 0 ? rc : debug_send_event(debug, event);
- if (event == RTE_BPF_VALIDATE_DEBUG_EVENT_STEP ||
- event == RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER)
+ if (events == 0 || events == RTE_BIT64(
+ RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER))
/* Stepping into a real instruction to execute. */
rc = rc < 0 ? rc : debug_points_call_back(debug,
&debug->breakpoint_lists[pc]);
@@ -651,7 +652,8 @@ __rte_bpf_validate_debug_evaluate_finish(struct rte_bpf_validate_debug *debug,
event = RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS;
}
- rc = __rte_bpf_validate_debug_evaluate_step(debug, pc, event);
+ rc = __rte_bpf_validate_debug_evaluate_update(debug, pc,
+ RTE_BIT64(event));
}
debug_evaluate_close(debug);
diff --git a/lib/bpf/bpf_validate_debug.h b/lib/bpf/bpf_validate_debug.h
index a91f3e9c48..2b0c3bb646 100644
--- a/lib/bpf/bpf_validate_debug.h
+++ b/lib/bpf/bpf_validate_debug.h
@@ -46,7 +46,7 @@ __rte_bpf_validate_debug_evaluate_start(struct rte_bpf_validate_debug *debug,
const struct bpf_verifier *verifier, const struct rte_bpf_prm_ex *bpf_prm);
/*
- * Signal each instruction, branch end, or evaluation end.
+ * Signal evaluation events.
*
* Immediately return 0 if debug is NULL.
*
@@ -54,16 +54,17 @@ __rte_bpf_validate_debug_evaluate_start(struct rte_bpf_validate_debug *debug,
* Validate debug instance configured by user, can be NULL.
* @param pc
* Current value of the program counter, or next after last instruction.
- * @param event
- * Specific evaluation event if any, or RTE_BPF_VALIDATE_DEBUG_EVENT_STEP.
+ * @param events
+ * Bitmask of events.
+ * Step bit is always implied and should not be set.
* @return
* Non-negative value: evaluation should continue;
* -ECANCELED: evaluation should fail without calling this API again;
* Other negative value: evaluation should fail signalling failure;
*/
int
-__rte_bpf_validate_debug_evaluate_step(struct rte_bpf_validate_debug *debug,
- uint32_t pc, debug_event_t event);
+__rte_bpf_validate_debug_evaluate_update(struct rte_bpf_validate_debug *debug,
+ uint32_t pc, uint64_t events);
/*
* Signal end of evaluation process.
--
2.43.0
^ permalink raw reply related [flat|nested] 38+ messages in thread
* [PATCH v4 3/7] bpf/validate: formalize call back requirements
2026-09-17 14:35 ` [PATCH v4 0/7] validate-bpf: add standalone validation debug tool Marat Khalili
2026-09-17 14:35 ` [PATCH v4 1/7] bpf/validate: fix finished status on restart Marat Khalili
2026-09-17 14:35 ` [PATCH v4 2/7] bpf/validate: refactor internal step function Marat Khalili
@ 2026-09-17 14:35 ` Marat Khalili
2026-09-17 14:35 ` [PATCH v4 4/7] bpf/validate: add jump notification events Marat Khalili
` (4 subsequent siblings)
7 siblings, 0 replies; 38+ messages in thread
From: Marat Khalili @ 2026-09-17 14:35 UTC (permalink / raw)
To: Konstantin Ananyev, Marat Khalili; +Cc: dev, Marat Khalili, Stephen Hemminger
From: Marat Khalili <marat.khalili@huawei.com>
Formalize requirements to evaluation events in documentation and adjust
their handling in the actual code to match.
Particularly, specify more precisely:
* conditions for each catchpoint callback;
* relative order of callbacks within the step;
* expected value of the program counter;
Behaviour changes:
* one branch-return event is now generated for each branch-enter event
(previously multiple branch-return events could be coalesced);
* branch-return event now points to the corresponding conditional jump
(previously past the end of the last branch);
* step event is now only generated before an actual instruction
evaluation (previously also for branching and finishing events);
* program counter is no longer allowed to point past the program end;
These changes should simplify branch tracking of the validator by tests
and interactive applications. E.g. branch-enter and branch-return events
can now be used to push and pop branching points to/from some stack.
Signed-off-by: Marat Khalili <marat.khalili@huawei.com>
---
lib/bpf/bpf_validate.c | 28 +++++++++----
lib/bpf/bpf_validate_debug.c | 72 +++++++++++++++++++-------------
lib/bpf/bpf_validate_debug.h | 3 +-
lib/bpf/rte_bpf_validate_debug.h | 24 +++++++----
4 files changed, 78 insertions(+), 49 deletions(-)
diff --git a/lib/bpf/bpf_validate.c b/lib/bpf/bpf_validate.c
index b317abb811..f4258c2fb5 100644
--- a/lib/bpf/bpf_validate.c
+++ b/lib/bpf/bpf_validate.c
@@ -2892,6 +2892,23 @@ prune_eval_state(struct bpf_verifier *bvf, const struct inst_node *node,
return rc;
}
+static bool
+is_branch_start(const struct inst_node *node)
+{
+ return node->prev_node != NULL && node->prev_node->nb_edge > 1;
+}
+
+static uint64_t
+step_events(const struct inst_node *node)
+{
+ uint64_t events = RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_STEP);
+
+ if (is_branch_start(node))
+ events |= RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER);
+
+ return events;
+}
+
/* Do second pass through CFG and try to evaluate instructions
* via each possible path. The verifier will try all paths, tracking types of
* registers used as input to instructions, and updating resulting type via
@@ -2918,7 +2935,6 @@ evaluate(struct bpf_verifier *bvf)
const char *err;
const struct ebpf_insn *ins;
struct inst_node *next, *node;
- int prev_nb_edge; /* branching number of the previous instruction */
int rc, debug_rc;
struct rte_bpf_validate_debug *const debug = bvf->prm->debug;
@@ -2954,7 +2970,6 @@ evaluate(struct bpf_verifier *bvf)
ins = bvf->prm->raw.ins;
node = bvf->in;
next = node;
- prev_nb_edge = 1;
memset(&stats, 0, sizeof(stats));
@@ -2990,8 +3005,7 @@ evaluate(struct bpf_verifier *bvf)
}
rc = __rte_bpf_validate_debug_evaluate_update(debug, idx,
- prev_nb_edge > 1 ?
- RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER) : 0);
+ step_events(node));
if (rc < 0)
break;
@@ -3047,7 +3061,6 @@ evaluate(struct bpf_verifier *bvf)
stats.nb_prune++;
} else {
next->prev_node = node;
- prev_nb_edge = node->nb_edge;
node = next;
}
} else {
@@ -3057,9 +3070,9 @@ evaluate(struct bpf_verifier *bvf)
* and proceed with parent.
*/
- if (prev_nb_edge != 0) {
+ if (is_branch_start(node)) {
rc = __rte_bpf_validate_debug_evaluate_update(
- debug, get_node_idx(bvf, node) + 1,
+ debug, get_node_idx(bvf, node->prev_node),
RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN));
if (rc < 0)
break;
@@ -3067,7 +3080,6 @@ evaluate(struct bpf_verifier *bvf)
node->cur_edge = 0;
save_safe_eval_state(bvf, node);
- prev_nb_edge = 0;
node = node->prev_node;
/* first node will not have prev, signalling finish */
diff --git a/lib/bpf/bpf_validate_debug.c b/lib/bpf/bpf_validate_debug.c
index f76303a7fd..db6d13f9cb 100644
--- a/lib/bpf/bpf_validate_debug.c
+++ b/lib/bpf/bpf_validate_debug.c
@@ -225,6 +225,13 @@ debug_points_call_back(struct rte_bpf_validate_debug *debug,
return rc;
}
+/* Call back all breakpoints for the specified program counter. */
+static int
+debug_trigger_breakpoints(struct rte_bpf_validate_debug *debug, uint32_t pc)
+{
+ return debug_points_call_back(debug, &debug->breakpoint_lists[pc]);
+}
+
/* Call back all catchpoints for the specified event. */
static int
debug_send_event(struct rte_bpf_validate_debug *debug, debug_event_t event)
@@ -585,6 +592,21 @@ int
__rte_bpf_validate_debug_evaluate_update(struct rte_bpf_validate_debug *debug,
uint32_t pc, uint64_t events)
{
+ /* Required order of sent events according to the documentation. */
+ static const enum rte_bpf_validate_debug_event ordered_events[] = {
+ RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_START,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_INVALID_STATE,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_STEP,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_FAILURE,
+ };
+ RTE_BUILD_BUG_ON(
+ RTE_DIM(ordered_events) != RTE_BPF_VALIDATE_DEBUG_EVENT_END);
+
int rc;
if (debug == NULL)
@@ -595,29 +617,31 @@ __rte_bpf_validate_debug_evaluate_update(struct rte_bpf_validate_debug *debug,
return -ECHILD;
}
- if (pc > debug->bpf_prm->raw.nb_ins)
+ if (pc >= debug->bpf_prm->raw.nb_ins)
return -EINVAL;
debug->pc = pc;
rc = __rte_bpf_validate_state_is_valid(debug->verifier);
if (rc == 0)
- rc = debug_send_event(debug,
- RTE_BPF_VALIDATE_DEBUG_EVENT_INVALID_STATE);
+ events |= RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_INVALID_STATE);
+
+ for (uint32_t index = 0; index < RTE_DIM(ordered_events); index++) {
+ const enum rte_bpf_validate_debug_event event =
+ ordered_events[index];
+ if ((events & RTE_BIT64(event)) == 0)
+ continue;
- for (enum rte_bpf_validate_debug_event event = 0;
- event != RTE_BPF_VALIDATE_DEBUG_EVENT_END; ++event)
- if (events & RTE_BIT64(event))
- rc = rc < 0 ? rc : debug_send_event(debug, event);
+ if (event == RTE_BPF_VALIDATE_DEBUG_EVENT_STEP)
+ rc = rc < 0 ? rc : debug_trigger_breakpoints(debug, pc);
- if (events == 0 || events == RTE_BIT64(
- RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER))
- /* Stepping into a real instruction to execute. */
- rc = rc < 0 ? rc : debug_points_call_back(debug,
- &debug->breakpoint_lists[pc]);
+ rc = rc < 0 ? rc : debug_send_event(debug, event);
+ events -= RTE_BIT64(event);
+ }
- rc = rc < 0 ? rc : debug_send_event(debug,
- RTE_BPF_VALIDATE_DEBUG_EVENT_STEP);
+ if (events != 0)
+ /* Received unsupported events. */
+ rc = rc < 0 ? rc : -EINVAL;
return rc;
}
@@ -627,8 +651,6 @@ __rte_bpf_validate_debug_evaluate_finish(struct rte_bpf_validate_debug *debug,
int result)
{
int rc = 0;
- uint32_t pc;
- debug_event_t event;
if (debug == NULL)
return 0;
@@ -641,20 +663,10 @@ __rte_bpf_validate_debug_evaluate_finish(struct rte_bpf_validate_debug *debug,
debug->evaluate_finished = true;
debug->evaluate_result = result;
- if (result != -ECANCELED) {
- if (result < 0) {
- /* Last known pc is the place we failed. */
- pc = debug->pc;
- event = RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_FAILURE;
- } else {
- /* Show program end, not particular instruction. */
- pc = debug->bpf_prm->raw.nb_ins;
- event = RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS;
- }
-
- rc = __rte_bpf_validate_debug_evaluate_update(debug, pc,
- RTE_BIT64(event));
- }
+ if (result != -ECANCELED)
+ rc = debug_send_event(debug, result < 0 ?
+ RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_FAILURE :
+ RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS);
debug_evaluate_close(debug);
diff --git a/lib/bpf/bpf_validate_debug.h b/lib/bpf/bpf_validate_debug.h
index 2b0c3bb646..4603841db2 100644
--- a/lib/bpf/bpf_validate_debug.h
+++ b/lib/bpf/bpf_validate_debug.h
@@ -53,10 +53,9 @@ __rte_bpf_validate_debug_evaluate_start(struct rte_bpf_validate_debug *debug,
* @param debug
* Validate debug instance configured by user, can be NULL.
* @param pc
- * Current value of the program counter, or next after last instruction.
+ * Current value of the program counter.
* @param events
* Bitmask of events.
- * Step bit is always implied and should not be set.
* @return
* Non-negative value: evaluation should continue;
* -ECANCELED: evaluation should fail without calling this API again;
diff --git a/lib/bpf/rte_bpf_validate_debug.h b/lib/bpf/rte_bpf_validate_debug.h
index f30fa926f1..b3fedd5428 100644
--- a/lib/bpf/rte_bpf_validate_debug.h
+++ b/lib/bpf/rte_bpf_validate_debug.h
@@ -31,25 +31,31 @@ extern "C" {
* Supported validate events.
*
* Valid events begin from 0 and end before `RTE_BPF_VALIDATE_DEBUG_EVENT_END`.
+ *
+ * At any given evaluation step, callbacks are fired in the following order:
+ * - Validation start event;
+ * - Branching and invalid-state events (can be interleaved);
+ * - Instruction breakpoints (before evaluating instruction);
+ * - Step (before evaluating instruction) or validation result (if done) event;
*/
enum rte_bpf_validate_debug_event {
- /* Just before every instruction, at branch or validation end. */
+ /* Just before every evaluated instruction. */
RTE_BPF_VALIDATE_DEBUG_EVENT_STEP,
/* Validator has failed its internal self-checks. */
RTE_BPF_VALIDATE_DEBUG_EVENT_INVALID_STATE,
- /* Start of validation. */
+ /* Start of validation (pc points to first instruction). */
RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_START,
- /* Successful finish of validation. */
+ /* Successful finish (pc undefined). */
RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS,
- /* Finish of validation with error. */
+ /* Finish of validation with error (pc points at error). */
RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_FAILURE,
- /* Beginning of a branch just after the jump. */
+ /* Beginning of a branch evaluation (pc points to branch start). */
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER,
- /* Pruning branch as verified earlier. */
+ /* Pruning branch as verified earlier (pc points to branch start). */
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE,
- /* End of branch verification, after the last verified instruction. */
+ /* End of branch verification (pc points to jump instruction). */
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN,
- /* Pruning branch as dynamically unreachable. */
+ /* Pruning branch as dynamically unreachable (pc points to branch start). */
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE,
/* Number of valid event values. */
RTE_BPF_VALIDATE_DEBUG_EVENT_END,
@@ -208,7 +214,7 @@ rte_bpf_validate_debug_get_last_point(const struct rte_bpf_validate_debug *debug
* @param debug
* Debug instance.
* @return
- * Current program counter being validated, or one after last.
+ * Current program counter being validated.
* UINT32_MAX if no program is being validated.
*/
__rte_experimental
--
2.43.0
^ permalink raw reply related [flat|nested] 38+ messages in thread
* [PATCH v4 4/7] bpf/validate: add jump notification events
2026-09-17 14:35 ` [PATCH v4 0/7] validate-bpf: add standalone validation debug tool Marat Khalili
` (2 preceding siblings ...)
2026-09-17 14:35 ` [PATCH v4 3/7] bpf/validate: formalize call back requirements Marat Khalili
@ 2026-09-17 14:35 ` Marat Khalili
2026-09-17 14:35 ` [PATCH v4 5/7] bpf/validate: add get current event API Marat Khalili
` (3 subsequent siblings)
7 siblings, 0 replies; 38+ messages in thread
From: Marat Khalili @ 2026-09-17 14:35 UTC (permalink / raw)
To: Konstantin Ananyev, Marat Khalili; +Cc: dev, Marat Khalili, Stephen Hemminger
From: Marat Khalili <marat.khalili@huawei.com>
Add events notifying validate debug client of instructions that may
change program counter. This is particularly useful to an interactive
debugger to track conditional and unconditional jumps without trying to
parse the program and interpret instructions independently.
Signed-off-by: Marat Khalili <marat.khalili@huawei.com>
---
lib/bpf/bpf_validate.c | 10 ++++++++--
lib/bpf/bpf_validate_debug.c | 2 ++
lib/bpf/rte_bpf_validate_debug.h | 5 +++++
3 files changed, 15 insertions(+), 2 deletions(-)
diff --git a/lib/bpf/bpf_validate.c b/lib/bpf/bpf_validate.c
index f4258c2fb5..d503db8361 100644
--- a/lib/bpf/bpf_validate.c
+++ b/lib/bpf/bpf_validate.c
@@ -2899,13 +2899,19 @@ is_branch_start(const struct inst_node *node)
}
static uint64_t
-step_events(const struct inst_node *node)
+step_events(const struct inst_node *node, uint8_t op)
{
uint64_t events = RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_STEP);
if (is_branch_start(node))
events |= RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER);
+ if (op == (BPF_JMP | BPF_JA))
+ events |= RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS);
+
+ if (node->nb_edge > 1)
+ events |= RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL);
+
return events;
}
@@ -3005,7 +3011,7 @@ evaluate(struct bpf_verifier *bvf)
}
rc = __rte_bpf_validate_debug_evaluate_update(debug, idx,
- step_events(node));
+ step_events(node, op));
if (rc < 0)
break;
diff --git a/lib/bpf/bpf_validate_debug.c b/lib/bpf/bpf_validate_debug.c
index db6d13f9cb..098a88026d 100644
--- a/lib/bpf/bpf_validate_debug.c
+++ b/lib/bpf/bpf_validate_debug.c
@@ -600,6 +600,8 @@ __rte_bpf_validate_debug_evaluate_update(struct rte_bpf_validate_debug *debug,
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE,
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN,
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL,
RTE_BPF_VALIDATE_DEBUG_EVENT_STEP,
RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS,
RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_FAILURE,
diff --git a/lib/bpf/rte_bpf_validate_debug.h b/lib/bpf/rte_bpf_validate_debug.h
index b3fedd5428..f1ee8769d0 100644
--- a/lib/bpf/rte_bpf_validate_debug.h
+++ b/lib/bpf/rte_bpf_validate_debug.h
@@ -35,6 +35,7 @@ extern "C" {
* At any given evaluation step, callbacks are fired in the following order:
* - Validation start event;
* - Branching and invalid-state events (can be interleaved);
+ * - Jump-instruction events (before evaluating jump instruction);
* - Instruction breakpoints (before evaluating instruction);
* - Step (before evaluating instruction) or validation result (if done) event;
*/
@@ -57,6 +58,10 @@ enum rte_bpf_validate_debug_event {
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN,
/* Pruning branch as dynamically unreachable (pc points to branch start). */
RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE,
+ /* Program counter points to a jump-always instruction. */
+ RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS,
+ /* Program counter points to a conditional jump instruction. */
+ RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL,
/* Number of valid event values. */
RTE_BPF_VALIDATE_DEBUG_EVENT_END,
};
--
2.43.0
^ permalink raw reply related [flat|nested] 38+ messages in thread
* [PATCH v4 5/7] bpf/validate: add get current event API
2026-09-17 14:35 ` [PATCH v4 0/7] validate-bpf: add standalone validation debug tool Marat Khalili
` (3 preceding siblings ...)
2026-09-17 14:35 ` [PATCH v4 4/7] bpf/validate: add jump notification events Marat Khalili
@ 2026-09-17 14:35 ` Marat Khalili
2026-09-17 14:35 ` [PATCH v4 6/7] app/test: add test for bpf validate debug events Marat Khalili
` (2 subsequent siblings)
7 siblings, 0 replies; 38+ messages in thread
From: Marat Khalili @ 2026-09-17 14:35 UTC (permalink / raw)
To: Konstantin Ananyev, Marat Khalili; +Cc: dev, Marat Khalili, Stephen Hemminger
From: Marat Khalili <marat.khalili@huawei.com>
Add API for retrieving the event currently processed by a callback.
Useful when single callback is registered for multiple events to put
less pressure on the context argument.
Signed-off-by: Marat Khalili <marat.khalili@huawei.com>
---
lib/bpf/bpf_validate_debug.c | 13 +++++++++++++
lib/bpf/rte_bpf_validate_debug.h | 13 +++++++++++++
2 files changed, 26 insertions(+)
diff --git a/lib/bpf/bpf_validate_debug.c b/lib/bpf/bpf_validate_debug.c
index 098a88026d..c8e9e5dd92 100644
--- a/lib/bpf/bpf_validate_debug.c
+++ b/lib/bpf/bpf_validate_debug.c
@@ -43,6 +43,7 @@ struct rte_bpf_validate_debug {
const struct bpf_verifier *verifier;
const struct rte_bpf_prm_ex *bpf_prm;
struct point_list *breakpoint_lists;
+ enum rte_bpf_validate_debug_event current_event;
struct rte_bpf_validate_debug_point *last_point;
uint32_t pc;
/* Evaluate stage (only tracking `evaluate` part at the moment). */
@@ -236,6 +237,7 @@ debug_trigger_breakpoints(struct rte_bpf_validate_debug *debug, uint32_t pc)
static int
debug_send_event(struct rte_bpf_validate_debug *debug, debug_event_t event)
{
+ debug->current_event = event;
return debug_points_call_back(debug, &debug->catchpoint_lists[event]);
}
@@ -342,6 +344,17 @@ rte_bpf_validate_debug_get_ins(const struct rte_bpf_validate_debug *debug,
return 0;
}
+RTE_EXPORT_EXPERIMENTAL_SYMBOL(rte_bpf_validate_debug_get_event, 26.11)
+enum rte_bpf_validate_debug_event
+rte_bpf_validate_debug_get_event(const struct rte_bpf_validate_debug *debug)
+{
+ if (debug == NULL)
+ /* Just to be fool-proof, not really required by API. */
+ return -EINVAL;
+
+ return debug->current_event;
+}
+
RTE_EXPORT_EXPERIMENTAL_SYMBOL(rte_bpf_validate_debug_get_last_point, 26.07)
struct rte_bpf_validate_debug_point *
rte_bpf_validate_debug_get_last_point(const struct rte_bpf_validate_debug *debug)
diff --git a/lib/bpf/rte_bpf_validate_debug.h b/lib/bpf/rte_bpf_validate_debug.h
index f1ee8769d0..6d4bf53589 100644
--- a/lib/bpf/rte_bpf_validate_debug.h
+++ b/lib/bpf/rte_bpf_validate_debug.h
@@ -194,6 +194,19 @@ int
rte_bpf_validate_debug_get_ins(const struct rte_bpf_validate_debug *debug,
const struct ebpf_insn **ins, uint32_t *nb_ins);
+/**
+ * Get event currently processed by a catchpoint callback.
+ *
+ * @param debug
+ * Debug instance.
+ * @return
+ * Event currently processed by a catchpoint callback.
+ * Undefined if no event is currently being processed.
+ */
+__rte_experimental
+enum rte_bpf_validate_debug_event
+rte_bpf_validate_debug_get_event(const struct rte_bpf_validate_debug *debug);
+
/**
* Get last triggered breakpoint or catchpoint.
*
--
2.43.0
^ permalink raw reply related [flat|nested] 38+ messages in thread
* [PATCH v4 6/7] app/test: add test for bpf validate debug events
2026-09-17 14:35 ` [PATCH v4 0/7] validate-bpf: add standalone validation debug tool Marat Khalili
` (4 preceding siblings ...)
2026-09-17 14:35 ` [PATCH v4 5/7] bpf/validate: add get current event API Marat Khalili
@ 2026-09-17 14:35 ` Marat Khalili
2026-09-17 14:35 ` [PATCH v4 7/7] app/validate-bpf: add BPF validation application Marat Khalili
2026-09-17 16:16 ` [PATCH v4 0/7] validate-bpf: add standalone validation debug tool Konstantin Ananyev
7 siblings, 0 replies; 38+ messages in thread
From: Marat Khalili @ 2026-09-17 14:35 UTC (permalink / raw)
To: Konstantin Ananyev, Marat Khalili; +Cc: dev, Marat Khalili, Stephen Hemminger
From: Marat Khalili <marat.khalili@huawei.com>
Add a test that verifies that eBPF validator generates callbacks with
the number, order and program counter specified in the documentation.
Signed-off-by: Marat Khalili <marat.khalili@huawei.com>
---
app/test/test_bpf_validate.c | 344 ++++++++++++++++++++++++++++++++++-
1 file changed, 340 insertions(+), 4 deletions(-)
diff --git a/app/test/test_bpf_validate.c b/app/test/test_bpf_validate.c
index 066f1fa156..8b48d8a504 100644
--- a/app/test/test_bpf_validate.c
+++ b/app/test/test_bpf_validate.c
@@ -9,10 +9,6 @@
#include <rte_bpf_validate_debug.h>
#include <rte_errno.h>
-/*
- * Tests of BPF validation.
- */
-
extern int test_bpf_validate_logtype;
#define RTE_LOGTYPE_TEST_BPF_VALIDATE test_bpf_validate_logtype
#define TEST_LOG_LINE(level, ...) \
@@ -26,6 +22,346 @@ RTE_LOG_REGISTER(test_bpf_validate_logtype, test.bpf_validate, NOTICE);
/* Special value indicating that register variable is not being used. */
#define NO_REGISTER UINT8_MAX
+static const char *const event_names[RTE_BPF_VALIDATE_DEBUG_EVENT_END] = {
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_STEP] = "step",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_INVALID_STATE] = "invalid-state",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_START] = "validation-start",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS] = "validation-success",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_FAILURE] = "validation-failure",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER] = "branch-enter",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE] = "branch-prune",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN] = "branch-return",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE] = "branch-unreachable",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS] = "jump-always",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL] = "jump-conditional",
+};
+
+/*
+ * Self-test of BPF validation debug events.
+ */
+
+#define EVENT_BIT(name) RTE_BIT64(RTE_BPF_VALIDATE_DEBUG_EVENT_ ## name)
+#define EVENT_BITS_ALL RTE_LEN2MASK(RTE_BPF_VALIDATE_DEBUG_EVENT_END, uint64_t)
+
+struct test_events_context {
+ /* Tested program. */
+ const struct ebpf_insn *ins;
+ uint32_t nb_ins;
+ /* Number of times each of the events was encountered. */
+ int event_counts[RTE_BPF_VALIDATE_DEBUG_EVENT_END];
+ /* Events active within the current step. */
+ uint64_t active_events;
+ /* Breakpoint was called within the current step. */
+ bool breakpoint_called;
+ /* Currently active conditional jump pc. */
+ uint32_t branch_pc;
+ /* Program counters of all conditional jumps in the current code path. */
+ uint32_t branch_pc_stack[4];
+ size_t branch_pc_stack_length;
+};
+
+/* Push current branch_pc to stack and clear it. */
+static int
+test_events_branch_pc_push(struct test_events_context *ctx)
+{
+ TEST_ASSERT_NOT_EQUAL(ctx->branch_pc, NO_PROGRAM_COUNTER,
+ "no branch pc to push");
+ TEST_ASSERT(ctx->branch_pc_stack_length < RTE_DIM(ctx->branch_pc_stack),
+ "branch stack overflow");
+ ctx->branch_pc_stack[ctx->branch_pc_stack_length++] = ctx->branch_pc;
+ ctx->branch_pc = NO_PROGRAM_COUNTER;
+ return TEST_SUCCESS;
+}
+
+/* Pop branch_pc from stack. */
+static int
+test_events_branch_pc_pop(struct test_events_context *ctx)
+{
+ TEST_ASSERT(ctx->branch_pc_stack_length > 0, "branch stack underflow");
+ ctx->branch_pc = ctx->branch_pc_stack[--ctx->branch_pc_stack_length];
+ return TEST_SUCCESS;
+}
+
+/* Verify breakpoint, checking that it is only called once and remembering it. */
+static int
+test_events_break_cb(struct rte_bpf_validate_debug *debug, void *void_ctx)
+{
+ struct test_events_context *const ctx = void_ctx;
+ const uint32_t pc = rte_bpf_validate_debug_get_pc(debug);
+
+ TEST_LOG_LINE(DEBUG, "%2u: break", pc);
+
+ TEST_ASSERT(!ctx->breakpoint_called,
+ "breakpoint called again at pc %u", pc);
+ ctx->breakpoint_called = true;
+
+ return TEST_SUCCESS;
+}
+
+/*
+ * Verify event, checking that:
+ * - validation-start is called at pc 0;
+ * - branch-return is called at a corresponding conditional jump pc;
+ * - branch-enter and branch-return are correctly paired;
+ * - jump-conditional and jump-always are called for corresponding opcodes;
+ * - jump-conditional and only it is followed by branches events;
+ * - events and breakpoints are correctly ordered;
+ *
+ * Increment event counter to compare against expected counts later.
+ */
+static int
+test_events_catch_cb(struct rte_bpf_validate_debug *debug, void *void_ctx)
+{
+ /* Bitmask of events allowed to precede each given event within the step. */
+ static const uint64_t allowed_preceding_events[] = {
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_START] = 0,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE] =
+ EVENT_BIT(BRANCH_PRUNE) |
+ EVENT_BIT(BRANCH_UNREACHABLE) |
+ EVENT_BIT(BRANCH_RETURN),
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE] =
+ EVENT_BIT(BRANCH_PRUNE) |
+ EVENT_BIT(BRANCH_UNREACHABLE) |
+ EVENT_BIT(BRANCH_RETURN),
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER] =
+ EVENT_BIT(BRANCH_PRUNE) |
+ EVENT_BIT(BRANCH_UNREACHABLE) |
+ EVENT_BIT(BRANCH_RETURN),
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN] =
+ EVENT_BIT(BRANCH_PRUNE) |
+ EVENT_BIT(BRANCH_UNREACHABLE) |
+ EVENT_BIT(BRANCH_RETURN),
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL] =
+ EVENT_BIT(BRANCH_ENTER) |
+ EVENT_BIT(BRANCH_PRUNE) |
+ EVENT_BIT(BRANCH_UNREACHABLE) |
+ EVENT_BIT(BRANCH_RETURN),
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS] =
+ EVENT_BIT(BRANCH_ENTER) |
+ EVENT_BIT(BRANCH_PRUNE) |
+ EVENT_BIT(BRANCH_UNREACHABLE) |
+ EVENT_BIT(BRANCH_RETURN),
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_STEP] =
+ EVENT_BITS_ALL - EVENT_BIT(VALIDATION_SUCCESS),
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS] =
+ EVENT_BIT(BRANCH_PRUNE) |
+ EVENT_BIT(BRANCH_UNREACHABLE) |
+ EVENT_BIT(BRANCH_RETURN),
+ };
+
+ struct test_events_context *const ctx = void_ctx;
+ const enum rte_bpf_validate_debug_event event =
+ rte_bpf_validate_debug_get_event(debug);
+ const uint32_t pc = rte_bpf_validate_debug_get_pc(debug);
+
+ TEST_LOG_LINE(DEBUG, "%2u: %s", pc, event_names[event]);
+
+ /* Verify event program counter and underlying instruction. */
+ switch (event) {
+ case RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_START:
+ TEST_ASSERT_EQUAL(pc, 0,
+ "event %s called at pc %u",
+ event_names[event], pc);
+ break;
+ case RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER:
+ TEST_ASSERT_SUCCESS(test_events_branch_pc_push(ctx),
+ "error pushing branch for event %s at pc %u",
+ event_names[event], pc);
+ break;
+ case RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN:
+ TEST_ASSERT_SUCCESS(test_events_branch_pc_pop(ctx),
+ "error popping branch for event %s at pc %u",
+ event_names[event], pc);
+ TEST_ASSERT_EQUAL(pc, ctx->branch_pc,
+ "event %s called at pc %u after jump pc %u",
+ event_names[event], pc, ctx->branch_pc);
+ break;
+ case RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS:
+ TEST_ASSERT_EQUAL(BPF_CLASS(ctx->ins[pc].code), BPF_JMP,
+ "event %s called at non-jump pc %u",
+ event_names[event], pc);
+ TEST_ASSERT_EQUAL(BPF_OP(ctx->ins[pc].code), BPF_JA,
+ "event %s called at non-jump-always pc %u",
+ event_names[event], pc);
+ break;
+ case RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL:
+ TEST_ASSERT_EQUAL(BPF_CLASS(ctx->ins[pc].code), BPF_JMP,
+ "event %s called at non-jump pc %u",
+ event_names[event], pc);
+ TEST_ASSERT_NOT_EQUAL(BPF_OP(ctx->ins[pc].code), BPF_JA,
+ "event %s called at jump-always pc %u",
+ event_names[event], pc);
+ break;
+ default:
+ break;
+ }
+
+ /* Verify events order. */
+ TEST_ASSERT_EQUAL(
+ (ctx->active_events & ~allowed_preceding_events[event]), 0,
+ "event %s at pc %u preceded by %#jx",
+ event_names[event], pc, (uintmax_t)ctx->active_events);
+
+ /* Update events for the current step as well as branch program counter. */
+ if (event != RTE_BPF_VALIDATE_DEBUG_EVENT_STEP) {
+ TEST_ASSERT(!ctx->breakpoint_called,
+ "event %s called after a breakpoint at pc %u",
+ event_names[event], pc);
+ ctx->active_events |= RTE_BIT64(event);
+ } else {
+ TEST_ASSERT_EQUAL(ctx->branch_pc, NO_PROGRAM_COUNTER,
+ "no branch-enter between branch- and next instruction");
+
+ if (ctx->active_events & EVENT_BIT(JUMP_CONDITIONAL))
+ /* Remember branch instruction pointer, branches to follow. */
+ ctx->branch_pc = pc;
+
+ TEST_ASSERT(ctx->breakpoint_called,
+ "step called before or without a breakpoint at pc %u", pc);
+
+ /* Prepare for the next step. */
+ ctx->active_events = 0;
+ ctx->breakpoint_called = false;
+ }
+
+ /* Update total event counts. */
+ ++ctx->event_counts[event];
+
+ return TEST_SUCCESS;
+}
+
+/*
+ * This test verifies that eBPF validator generates expected numbers of events
+ * of each kind in the expected order (e.g. step after jump after branch).
+ */
+static int
+test_events(void)
+{
+ /*
+ * Program demonstrating the following properties:
+ * - prunnable code paths;
+ * - dynamically unreachable code paths;
+ * - conditional jumps landing on conditional jumps;
+ * - conditional jumps landing on unconditional jumps;
+ */
+ static const struct ebpf_insn ins[] = {
+ /* 0: r0 = 0 */
+ { .code = (EBPF_ALU64 | EBPF_MOV | BPF_K), .dst_reg = EBPF_REG_0, .imm = 0 },
+ /* 1: if r1 > 0 goto 4 */
+ { .code = (BPF_JMP | BPF_JGT | BPF_K), .dst_reg = EBPF_REG_1, .imm = 0, .off = 2 },
+ /* 2: r1 = 0 */
+ { .code = (EBPF_ALU64 | EBPF_MOV | BPF_K), .dst_reg = EBPF_REG_1, .imm = 0 },
+ /* 3: if r0 < 1 goto 8 */
+ { .code = (BPF_JMP | EBPF_JLT | BPF_K), .dst_reg = EBPF_REG_0, .imm = 1, .off = 4 },
+ /* 4: if r1 > 0 goto 6 */
+ { .code = (BPF_JMP | BPF_JGT | BPF_K), .dst_reg = EBPF_REG_1, .imm = 0, .off = 1 },
+ /* 5: goto 9 */
+ { .code = (BPF_JMP | BPF_JA), .off = 3 },
+ /* 6: r1 = 0 */
+ { .code = (EBPF_ALU64 | EBPF_MOV | BPF_K), .dst_reg = EBPF_REG_1, .imm = 0 },
+ /* 7: if r0 > 0 goto 9 */
+ { .code = (BPF_JMP | BPF_JGT | BPF_K), .dst_reg = EBPF_REG_0, .imm = 0, .off = 1 },
+ /* 8: goto 10 */
+ { .code = (BPF_JMP | BPF_JA), .off = 1 },
+ /* 9: r0 = 1 */
+ { .code = (EBPF_ALU64 | EBPF_MOV | BPF_K), .dst_reg = EBPF_REG_0, .imm = 1 },
+ /* 10: exit */
+ { .code = (BPF_JMP | EBPF_EXIT) }
+ };
+
+ /*
+ * Verified code paths (order may vary by library implementation):
+ * 0 -- 1 -- 4 -- 6 -- 7 -- unreachable 9
+ * | | `-- 8 -- 10 -- return 7 -- return 4
+ * | `-- unreachable 5 -- return 1
+ * `-- 2 -- 3 -- prune 8
+ * `-- unreachable 4 -- return 1 -- success 11
+ *
+ * Numbers without preceding word correspond to evaluated instructions,
+ * there are 9 of these in total, so 9 step events are expected.
+ *
+ * Note that 4 encountered conditional jump instructions generate:
+ * - 4 pairs of branch-enter/branch-return events;
+ * - 3 branch-unreachable events;
+ * - 1 branch-prune event;
+ * for a total count of 8 (4 instructions * 2 branches/instruction).
+ */
+ static const int expected_counts[RTE_BPF_VALIDATE_DEBUG_EVENT_END] = {
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_STEP] = 9,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_START] = 1,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS] = 1,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER] = 4,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE] = 1,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN] = 4,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE] = 3,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS] = 1,
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL] = 4,
+ };
+
+ struct test_events_context ctx = {
+ .ins = ins,
+ .nb_ins = RTE_DIM(ins),
+ .branch_pc = NO_PROGRAM_COUNTER,
+ };
+
+ struct rte_bpf_validate_debug *debug = rte_bpf_validate_debug_create();
+ TEST_ASSERT_NOT_NULL(debug, "validate debug create error");
+
+ /* Set breakpoints for all instructions. */
+ for (uint32_t pc = 0; pc != RTE_DIM(ins); ++pc)
+ rte_bpf_validate_debug_break(debug, pc,
+ &(struct rte_bpf_validate_debug_callback){
+ .fn = test_events_break_cb,
+ .ctx = &ctx,
+ });
+
+ /* Set catchpoints for all events except invalid-state. */
+ for (enum rte_bpf_validate_debug_event event = 0;
+ event != RTE_BPF_VALIDATE_DEBUG_EVENT_END; ++event)
+ if (event != RTE_BPF_VALIDATE_DEBUG_EVENT_INVALID_STATE)
+ rte_bpf_validate_debug_catch(debug, event,
+ &(struct rte_bpf_validate_debug_callback){
+ .fn = test_events_catch_cb,
+ .ctx = &ctx,
+ });
+
+ struct rte_bpf *const bpf = rte_bpf_load_ex(&(struct rte_bpf_prm_ex){
+ .sz = sizeof(struct rte_bpf_prm_ex),
+ .origin = RTE_BPF_ORIGIN_RAW,
+ .raw.ins = ins,
+ .raw.nb_ins = RTE_DIM(ins),
+ .prog_arg[0] = { .type = RTE_BPF_ARG_RAW },
+ .nb_prog_arg = 1,
+ .debug = debug,
+ });
+
+ rte_bpf_destroy(bpf);
+ rte_bpf_validate_debug_destroy(debug);
+
+ TEST_ASSERT_NOT_NULL(bpf, "validation failed");
+
+ /* Verify event counts. */
+ for (enum rte_bpf_validate_debug_event event = 0;
+ event != RTE_BPF_VALIDATE_DEBUG_EVENT_END; ++event)
+ TEST_ASSERT_EQUAL(
+ ctx.event_counts[event], expected_counts[event],
+ "expected %d %s events, but got %d",
+ expected_counts[event], event_names[event],
+ ctx.event_counts[event]);
+
+ /* Additionally double-check final stack state. */
+ TEST_ASSERT_EQUAL(ctx.branch_pc_stack_length, 0,
+ "unbalanced branch enters and returns");
+
+ return TEST_SUCCESS;
+}
+
+REGISTER_FAST_TEST(bpf_validate_events_autotest, NOHUGE_OK, ASAN_OK, test_events);
+
+/*
+ * Tests of BPF validation.
+ */
+
/* Sizes of text buffers used for formatting various debug outputs. */
#define VALUE_FORMAT_BUFFER_SIZE 32
#define INTERVAL_FORMAT_BUFFER_SIZE 64
--
2.43.0
^ permalink raw reply related [flat|nested] 38+ messages in thread
* [PATCH v4 7/7] app/validate-bpf: add BPF validation application
2026-09-17 14:35 ` [PATCH v4 0/7] validate-bpf: add standalone validation debug tool Marat Khalili
` (5 preceding siblings ...)
2026-09-17 14:35 ` [PATCH v4 6/7] app/test: add test for bpf validate debug events Marat Khalili
@ 2026-09-17 14:35 ` Marat Khalili
2026-09-17 16:16 ` [PATCH v4 0/7] validate-bpf: add standalone validation debug tool Konstantin Ananyev
7 siblings, 0 replies; 38+ messages in thread
From: Marat Khalili @ 2026-09-17 14:35 UTC (permalink / raw)
To: Thomas Monjalon, Konstantin Ananyev, Marat Khalili
Cc: dev, Marat Khalili, Stephen Hemminger
From: Marat Khalili <marat.khalili@huawei.com>
Add a new DPDK application, dpdk-validate-bpf, for pre-validating eBPF
programs for compatibility with the lib/bpf execution context.
The application allows evaluating eBPF programs against the DPDK verifier
before loading them into a real application. It includes an interactive
debugging mode to trace state changes per instruction and understand the
validator's decisions.
Signed-off-by: Marat Khalili <marat.khalili@huawei.com>
---
MAINTAINERS | 2 +
app/meson.build | 1 +
app/validate-bpf/alloc_list.c | 53 ++
app/validate-bpf/args.c | 262 ++++++
app/validate-bpf/debug.c | 1087 ++++++++++++++++++++++++
app/validate-bpf/debug_command.c | 383 +++++++++
app/validate-bpf/debug_command.h | 65 ++
app/validate-bpf/eal_init_args.c | 57 ++
app/validate-bpf/internal.h | 151 ++++
app/validate-bpf/main.c | 90 ++
app/validate-bpf/meson.build | 13 +
app/validate-bpf/parse_decl.c | 616 ++++++++++++++
doc/guides/rel_notes/release_26_11.rst | 5 +
doc/guides/tools/index.rst | 1 +
doc/guides/tools/validate_bpf.rst | 97 +++
15 files changed, 2883 insertions(+)
create mode 100644 app/validate-bpf/alloc_list.c
create mode 100644 app/validate-bpf/args.c
create mode 100644 app/validate-bpf/debug.c
create mode 100644 app/validate-bpf/debug_command.c
create mode 100644 app/validate-bpf/debug_command.h
create mode 100644 app/validate-bpf/eal_init_args.c
create mode 100644 app/validate-bpf/internal.h
create mode 100644 app/validate-bpf/main.c
create mode 100644 app/validate-bpf/meson.build
create mode 100644 app/validate-bpf/parse_decl.c
create mode 100644 doc/guides/tools/validate_bpf.rst
diff --git a/MAINTAINERS b/MAINTAINERS
index 8c50c52933..affaf57492 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -1917,8 +1917,10 @@ F: lib/bpf/
F: examples/bpf/
F: app/test/test_bpf*
F: app/test/bpf/
+F: app/validate-bpf/
F: app/test-pmd/bpf_cmd.*
F: doc/guides/prog_guide/bpf_lib.rst
+F: doc/guides/tools/validate_bpf.rst
Graph
M: Jerin Jacob <jerinj@marvell.com>
diff --git a/app/meson.build b/app/meson.build
index 1798db3ae4..9b59ae4830 100644
--- a/app/meson.build
+++ b/app/meson.build
@@ -33,6 +33,7 @@ apps = [
'test-regex',
'test-sad',
'test-security-perf',
+ 'validate-bpf',
]
if get_option('tests')
diff --git a/app/validate-bpf/alloc_list.c b/app/validate-bpf/alloc_list.c
new file mode 100644
index 0000000000..2f4d18ad3d
--- /dev/null
+++ b/app/validate-bpf/alloc_list.c
@@ -0,0 +1,53 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include "internal.h"
+#include <stdlib.h>
+
+
+/* Needs to be a power of two. */
+#define START_CAPACITY (1u << 3)
+
+size_t
+alloc_list_append(struct alloc_list *alloc_list, void *ptr)
+{
+ if (alloc_list->count == 0) {
+ RTE_ASSERT(alloc_list->ptrs == NULL);
+ alloc_list->ptrs = malloc(
+ sizeof(alloc_list->ptrs[0]) * START_CAPACITY);
+ RTE_VERIFY(alloc_list->ptrs != NULL);
+ } else if (alloc_list->count >= START_CAPACITY &&
+ /* Power of two detection */
+ (alloc_list->count & (alloc_list->count - 1)) == 0) {
+ /*
+ * We allocate in powers of two, and current count is one of
+ * these powers, so need to reallocate to larger capacity.
+ */
+ RTE_ASSERT(alloc_list->ptrs != NULL);
+ const size_t new_capacity = alloc_list->count * 2;
+ alloc_list->ptrs = realloc(alloc_list->ptrs,
+ sizeof(alloc_list->ptrs[0]) * new_capacity);
+ RTE_VERIFY(alloc_list->ptrs != NULL);
+ }
+ alloc_list->ptrs[alloc_list->count] = ptr;
+ return alloc_list->count++;
+}
+
+void alloc_list_replace(struct alloc_list *alloc_list, size_t index, void *ptr)
+{
+ RTE_ASSERT(index < alloc_list->count);
+ alloc_list->ptrs[index] = ptr;
+}
+
+void alloc_list_free_all(struct alloc_list *alloc_list)
+{
+ /* Copy and clear fields first in case alloc_list itself gets freed. */
+ size_t count = alloc_list->count;
+ void ** const ptrs = alloc_list->ptrs;
+ *alloc_list = (struct alloc_list){};
+
+ while (count != 0)
+ free(ptrs[--count]);
+ free(ptrs);
+}
diff --git a/app/validate-bpf/args.c b/app/validate-bpf/args.c
new file mode 100644
index 0000000000..b11591ee70
--- /dev/null
+++ b/app/validate-bpf/args.c
@@ -0,0 +1,262 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include "internal.h"
+
+#include <ctype.h>
+#include <getopt.h>
+#include <stdlib.h>
+
+#include <rte_errno.h>
+
+
+/* Values to be used in getopt_long option.val. */
+enum app_args {
+ ARG_UNRECOGNIZED = '?',
+ ARG_AUTO = 0, /* Value set by getopt_long when flag is non-NULL. */
+ ARG_HELP, /* Keep this one in the beginning for UX reasons. */
+ ARG_DEBUG,
+ ARG_MBUF_BUF_SIZE,
+ ARG_NO_PROG_ARGS,
+ ARG_PROG_ARG,
+ ARG_SECTION,
+ ARG_XSYM,
+};
+
+/* Program options. */
+static struct option app_options[] = {
+ {
+ .name = "help",
+ .val = ARG_HELP,
+ },
+ {
+ .name = "debug",
+ .val = ARG_DEBUG,
+ },
+ {
+ .name = "mbuf-buf-size",
+ .has_arg = required_argument,
+ .val = ARG_MBUF_BUF_SIZE,
+ },
+ {
+ .name = "no-prog-arg",
+ .val = ARG_NO_PROG_ARGS,
+ },
+ {
+ .name = "no-prog-args",
+ .val = ARG_NO_PROG_ARGS,
+ },
+ {
+ .name = "prog-arg",
+ .has_arg = required_argument,
+ .val = ARG_PROG_ARG,
+ },
+ {
+ .name = "section",
+ .has_arg = required_argument,
+ .val = ARG_SECTION,
+ },
+ {
+ .name = "xsym",
+ .has_arg = required_argument,
+ .val = ARG_XSYM,
+ },
+ { /* terminating zero record */ }
+};
+
+/* Default args value. */
+static const struct args args_default = {
+ .bpf_prm = {
+ .sz = sizeof(struct rte_bpf_prm_ex),
+ .origin = RTE_BPF_ORIGIN_ELF_FILE,
+ .elf_file.section = ".text",
+ },
+ .mbuf_buf_size = RTE_MBUF_DEFAULT_BUF_SIZE,
+};
+
+/* Default --prog-arg argument value. */
+static const char * const prog_arg_default = "struct rte_mbuf *";
+
+
+void
+print_usage(const char *program_name)
+{
+ static const char *const options[][2] = {
+ { "--help", "Display help and exit." },
+ { "--debug", "Enable interactive debug mode." },
+ { "--mbuf-buf-size=MBUF_BUF_SIZE", "Size of the mbuf data buffer (in bytes)." },
+ { "--prog-arg=TYPE", "Expected type of the next BPF program argument (up to 5)." },
+ { "--no-prog-args", "BPF program does not take any arguments." },
+ { "--section=SECTION", "ELF section name in the BPF object file to load." },
+ { "--xsym='TYPE NAME[(TYPE, ...)]'", "External symbol BPF program can access." },
+ };
+
+ printf("USAGE: %s [OPTIONS]... BPF_PATH\n", program_name);
+ printf("OPTIONS:\n");
+ for (int oi = 0; oi != RTE_DIM(options); ++oi)
+ printf("\t%-31s %s\n", options[oi][0], options[oi][1]);
+}
+
+void
+print_defaults(void)
+{
+ printf("DEFAULTS:"
+ " --mbuf-buf-size=%#zx"
+ " --prog-arg='%s'"
+ " --section='%s'\n",
+ (size_t)RTE_MBUF_DEFAULT_BUF_SIZE,
+ prog_arg_default,
+ args_default.bpf_prm.elf_file.section);
+}
+
+static int
+parse_size(size_t *result, const char *text)
+{
+ char *parse_end = NULL;
+
+ errno = 0;
+ const unsigned long strtoul_result = strtoul(text, &parse_end, 0);
+ if (errno != 0 || *parse_end != '\0' || strtoul_result == 0)
+ return -1;
+
+ *result = strtoul_result;
+ return 0;
+}
+
+struct args *
+args_parse(int argc, char *argv[])
+{
+ int val;
+ size_t xsym_alloc_index;
+ struct rte_bpf_xsym *xsym = NULL;
+ const char * const program_name = argv[0];
+
+ /* Allocate args and set aliases for some of its members. */
+ struct args * const args = malloc(sizeof(*args));
+ RTE_VERIFY(args != NULL);
+ struct alloc_list * const alloc_list = &args->_alloc_list;
+ struct rte_bpf_prm_ex * const bpf_prm = &args->bpf_prm;
+
+ /* Set default values. */
+ *args = args_default;
+ RTE_VERIFY(parse_arg(&bpf_prm->prog_arg[bpf_prm->nb_prog_arg++],
+ prog_arg_default) >= 0);
+ bool default_prog_args = true;
+
+ /* Reserve space for xsym in alloc_list. */
+ xsym_alloc_index = alloc_list_append(alloc_list, xsym);
+
+ while ((val = getopt_long(argc, argv, "", app_options, NULL)) != EOF) {
+ int rc = 0;
+ switch (val) {
+ case ARG_AUTO:
+ /* getopt_long made the assignment, nothing to do */
+ break;
+ case ARG_HELP:
+ args->show_help = true;
+ break;
+ case ARG_DEBUG:
+ if (bpf_prm->debug == NULL)
+ bpf_prm->debug = debug_create();
+ if (bpf_prm->debug == NULL) {
+ rc = -rte_errno;
+ fprintf(stderr,
+ "%s: error %d creating debug session\n",
+ program_name, -rc);
+ }
+ break;
+ case ARG_MBUF_BUF_SIZE:
+ rc = parse_size(&args->mbuf_buf_size, optarg);
+ if (rc < 0)
+ fprintf(stderr,
+ "%s: invalid mbuf buf size '%s'\n",
+ program_name, optarg);
+ break;
+ case ARG_NO_PROG_ARGS:
+ bpf_prm->nb_prog_arg = 0;
+ default_prog_args = false;
+ break;
+ case ARG_PROG_ARG:
+ if (default_prog_args) {
+ bpf_prm->nb_prog_arg = 0;
+ default_prog_args = false;
+ }
+
+ if (bpf_prm->nb_prog_arg == RTE_DIM(bpf_prm->prog_arg)) {
+ fprintf(stderr,
+ "%s: at most %d program arguments allowed\n",
+ program_name,
+ (int)RTE_DIM(bpf_prm->prog_arg));
+ rc = -EINVAL;
+ break;
+ }
+
+ rc = parse_arg(&bpf_prm->prog_arg[bpf_prm->nb_prog_arg++],
+ optarg);
+ if (rc < 0)
+ fprintf(stderr,
+ "%s: unrecognized prog arg '%s'\n",
+ program_name, optarg);
+
+ break;
+ case ARG_SECTION:
+ bpf_prm->elf_file.section = optarg;
+ break;
+ case ARG_XSYM:
+ bpf_prm->xsym = xsym = realloc(xsym,
+ sizeof(xsym[0]) * (bpf_prm->nb_xsym + 1));
+ RTE_VERIFY(xsym != NULL);
+ alloc_list_replace(alloc_list, xsym_alloc_index, xsym);
+ rc = parse_xsym(&xsym[bpf_prm->nb_xsym++], optarg,
+ alloc_list);
+ if (rc < 0)
+ fprintf(stderr, "%s: invalid xsym '%s'\n",
+ program_name, optarg);
+ break;
+ case ARG_UNRECOGNIZED:
+ args_destroy(args);
+ return NULL;
+ default:
+ rte_panic("Unexpected getopt_long return value %d\n",
+ val);
+ }
+ if (rc < 0) {
+ args_destroy(args);
+ return NULL;
+ }
+ }
+
+ /* Set buf_size to the value specified in command line arguments. */
+ adjust_arg_buf_size(bpf_prm->prog_arg, args->mbuf_buf_size);
+ for (size_t xsymi = 0; xsymi != bpf_prm->nb_xsym; ++xsymi)
+ adjust_xsym_buf_size(&xsym[xsymi], args->mbuf_buf_size);
+
+ /* getopt_long moves all non-options to the end, starting at optind. */
+ const int nb_bpf_path = argc - optind;
+ switch (nb_bpf_path) {
+ case 0:
+ break;
+ case 1:
+ bpf_prm->elf_file.path = argv[optind];
+ break;
+ default:
+ fprintf(stderr, "%s: too many positional arguments\n",
+ program_name);
+ args_destroy(args);
+ return NULL;
+ }
+
+ return args;
+}
+
+void
+args_destroy(struct args *args)
+{
+ if (args == NULL)
+ return;
+ if (args->bpf_prm.debug != NULL)
+ debug_destroy(args->bpf_prm.debug);
+ alloc_list_free_all(&args->_alloc_list);
+ free(args);
+}
diff --git a/app/validate-bpf/debug.c b/app/validate-bpf/debug.c
new file mode 100644
index 0000000000..33fabad511
--- /dev/null
+++ b/app/validate-bpf/debug.c
@@ -0,0 +1,1087 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include "debug_command.h"
+#include "internal.h"
+
+#include <rte_bpf_validate_debug.h>
+#include <rte_errno.h>
+
+#include <inttypes.h>
+#include <stdlib.h>
+
+
+/* Write single line to the user, currently just to stdout. */
+#define PRINTLN(fmt, ...) do { \
+ RTE_LOG_CHECK_NO_NEWLINE(fmt); \
+ printf(fmt "\n", ##__VA_ARGS__); \
+} while (0)
+
+#define PROMPT "(validate) "
+
+#define INITIAL_CAPACITY 8u
+
+static const char *const event_names[] = {
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_INVALID_STATE] = "invalid-state",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER] = "branch-enter",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE] = "branch-prune",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN] = "branch-return",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE] = "branch-unreachable",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS] = "jump-always",
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL] = "jump-conditional",
+};
+
+static const char *const register_names[] = {
+ [EBPF_REG_0] = "r0",
+ [EBPF_REG_1] = "r1",
+ [EBPF_REG_2] = "r2",
+ [EBPF_REG_3] = "r3",
+ [EBPF_REG_4] = "r4",
+ [EBPF_REG_5] = "r5",
+ [EBPF_REG_6] = "r6",
+ [EBPF_REG_7] = "r7",
+ [EBPF_REG_8] = "r8",
+ [EBPF_REG_9] = "r9",
+ [EBPF_REG_10] = "r10",
+};
+
+static const char *const comparison_operator_names[] = {
+ [BPF_JEQ] = "==",
+ [BPF_JGT] = ">",
+ [BPF_JGE] = ">=",
+ [EBPF_JNE] = "!=",
+ [EBPF_JSGT] = "s>",
+ [EBPF_JSGE] = "s>=",
+ [EBPF_JLT] = "<",
+ [EBPF_JLE] = "<=",
+ [EBPF_JSLT] = "s<",
+ [EBPF_JSLE] = "s<=",
+};
+
+enum point_type {
+ POINT_TYPE_BREAK,
+ POINT_TYPE_CATCH,
+};
+
+/* Local representation of points with additional info for UI. */
+struct point_info {
+ struct rte_bpf_validate_debug_point *point;
+ enum point_type type;
+ union {
+ uint32_t pc;
+ enum rte_bpf_validate_debug_event event;
+ };
+};
+
+/* Dynamically growing list of point infos. */
+struct point_infos {
+ struct point_info *elements;
+ uint32_t length;
+ uint32_t capacity;
+};
+
+/* Information about a single conditional or unconditional jump in code path. */
+struct branch_info {
+ uint32_t jump_pc;
+ uint32_t target_pc;
+ bool is_conditional;
+};
+
+/* Dynamically growing stack to track code path. */
+struct branch_stack {
+ struct branch_info *branches;
+ uint32_t length;
+ uint32_t capacity;
+};
+
+/* Flags telling if we should validate again, stopping or not at start. */
+static bool validate_again;
+
+/* List of point infos; element index is its ID for UI. */
+static struct point_infos point_infos;
+
+/* Catchpoint invisible to the user used for step-by-step validation. */
+static struct rte_bpf_validate_debug_point *step_point;
+
+/* Tracking entered branches. */
+static struct branch_stack branch_stack;
+static uint32_t pending_jump_pc = UINT32_MAX;
+static struct rte_bpf_validate_debug_point *jump_always_step_point;
+
+static int
+step_cb(struct rte_bpf_validate_debug *debug, void *ctx);
+static int
+point_cb(struct rte_bpf_validate_debug *debug, void *ctx);
+
+/* Find index of the string in the list. */
+static int
+find_name(const char *name, const char *const *names, int nb_names)
+{
+ if (nb_names < 0)
+ return -EINVAL;
+
+ for (int index = 0; index != nb_names; ++index)
+ if (names[index] != NULL && strcmp(names[index], name) == 0)
+ return index;
+
+ return -ENOENT;
+}
+
+/* Convert event name to its enum value. */
+static enum rte_bpf_validate_debug_event
+parse_event_name(const char *name)
+{
+ const int rc = find_name(name, event_names, RTE_DIM(event_names));
+ if (rc < 0)
+ PRINTLN("Error: invalid event name.");
+
+ return rc;
+}
+
+/* Convert comparison operator name to opcode. */
+static int
+parse_comparison_operator(const char *name)
+{
+ const int rc = find_name(name, comparison_operator_names,
+ RTE_DIM(comparison_operator_names));
+ if (rc < 0)
+ PRINTLN("Error: invalid comparison operator name.");
+
+ return rc;
+}
+
+/* Convert register name to its number. */
+static int
+parse_register(const char *name)
+{
+ const int rc = find_name(name, register_names, RTE_DIM(register_names));
+ if (rc < 0)
+ PRINTLN("Error: invalid register name.");
+
+ return rc;
+}
+
+/* Free local list of point infos (do not destroy points). */
+static void
+point_infos_free(void)
+{
+ free(point_infos.elements);
+ point_infos = (struct point_infos){};
+}
+
+/* Free local branch stack. */
+static void
+branch_stack_free(void)
+{
+ free(branch_stack.branches);
+ branch_stack = (struct branch_stack){};
+}
+
+/* Return existing element from the list of point infos. */
+static struct point_info *
+point_infos_at(uint32_t point_number)
+{
+ RTE_ASSERT(point_number < point_infos.length);
+ RTE_ASSERT(point_infos.elements[point_number].point != NULL);
+ return &point_infos.elements[point_number];
+}
+
+/* Print existing element from the list of point infos. */
+static void
+point_infos_print_at(uint32_t point_number)
+{
+ const struct point_info *const point_info = point_infos_at(point_number);
+
+ switch (point_info->type) {
+ case POINT_TYPE_BREAK:
+ PRINTLN("Breakpoint %d at %d.", point_number,
+ point_info->pc);
+ break;
+ case POINT_TYPE_CATCH:
+ PRINTLN("Catchpoint %d on %s.", point_number,
+ event_names[point_info->event]);
+ break;
+ default:
+ PRINTLN("Point %d of unknown type", point_number);
+ break;
+ }
+}
+
+/* Print all point infos. */
+static int
+point_infos_print_all(void)
+{
+ uint32_t nb_printed = 0;
+ for (uint32_t pn = 0; pn != point_infos.length; ++pn) {
+ const struct point_info *const point_info =
+ &point_infos.elements[pn];
+ if (point_info->point != NULL) {
+ point_infos_print_at(pn);
+ ++nb_printed;
+ }
+ }
+
+ if (nb_printed == 0) {
+ PRINTLN("No breakpoints or catchpoints set.");
+ return -ENOENT;
+ }
+
+ return nb_printed;
+}
+
+/* Allocate space for a new element in the list of point infos. */
+static uint32_t
+point_infos_append(void)
+{
+ if (point_infos.length == point_infos.capacity) {
+ /* Set to initial capacity or double previous one. */
+ point_infos.capacity = RTE_MAX(INITIAL_CAPACITY,
+ point_infos.capacity * 2);
+ point_infos.elements = realloc(point_infos.elements,
+ point_infos.capacity * sizeof(point_infos.elements[0]));
+ RTE_VERIFY(point_infos.elements != NULL);
+ }
+ return point_infos.length++;
+}
+
+/* Allocate space for a new element in the branch stack. */
+static void
+branch_stack_append(const struct branch_info *branch)
+{
+ if (branch_stack.length == branch_stack.capacity) {
+ /* Set to initial capacity or double previous one. */
+ branch_stack.capacity = RTE_MAX(INITIAL_CAPACITY,
+ branch_stack.capacity * 2);
+ branch_stack.branches = realloc(branch_stack.branches,
+ branch_stack.capacity * sizeof(branch_stack.branches[0]));
+ RTE_VERIFY(branch_stack.branches != NULL);
+ }
+ branch_stack.branches[branch_stack.length++] = *branch;
+}
+
+/* Destroy existing element from the list of point infos, printing it first. */
+static void
+point_infos_destroy_existing(uint32_t point_number)
+{
+ point_infos_print_at(point_number);
+
+ struct point_info *const point_info = point_infos_at(point_number);
+ rte_bpf_validate_debug_point_destroy(point_info->point);
+ *point_info = (struct point_info){};
+}
+
+/* Destroy point with specified number if it exists. */
+static int
+point_infos_destroy_at(uint32_t point_number)
+{
+ if (point_number >= point_infos.length ||
+ point_infos.elements[point_number].point == NULL) {
+ PRINTLN("No breakpoint number %d.", point_number);
+ return -ENOENT;
+ }
+
+ point_infos_destroy_existing(point_number);
+ return 1;
+}
+
+/* Destroy all point infos. */
+static void
+point_infos_destroy_all(void)
+{
+ uint32_t nb_destroyed = 0;
+ for (uint32_t pn = 0; pn < point_infos.length; ++pn) {
+ const struct point_info *const point_info =
+ &point_infos.elements[pn];
+ if (point_info->point != NULL) {
+ point_infos_destroy_existing(pn);
+ ++nb_destroyed;
+ }
+ }
+
+ if (nb_destroyed == 0)
+ PRINTLN("No breakpoints or catchpoints set.");
+}
+
+/* Destroy all breakpoints at specified location. */
+static void
+point_infos_destroy_breakpoints(uint32_t pc)
+{
+ uint32_t nb_destroyed = 0;
+ for (uint32_t pn = 0; pn < point_infos.length; ++pn) {
+ const struct point_info *const point_info =
+ &point_infos.elements[pn];
+ if (point_info->point != NULL &&
+ point_info->type == POINT_TYPE_BREAK &&
+ point_info->pc == pc) {
+ point_infos_destroy_existing(pn);
+ ++nb_destroyed;
+ }
+ }
+
+ if (nb_destroyed == 0)
+ PRINTLN("No breakpoint at %u.", pc);
+}
+
+/* Destroy all catchpoints for specified event. */
+static void
+point_infos_destroy_catchpoints(int event)
+{
+ if (event < 0)
+ /* Error was already printed by parse_event_name. */
+ return;
+
+ uint32_t nb_destroyed = 0;
+ for (uint32_t pn = 0; pn < point_infos.length; ++pn) {
+ const struct point_info *const point_info =
+ &point_infos.elements[pn];
+ if (point_info->point != NULL &&
+ point_info->type == POINT_TYPE_CATCH &&
+ (int)point_info->event == event) {
+ point_infos_destroy_existing(pn);
+ ++nb_destroyed;
+ }
+ }
+
+ if (nb_destroyed == 0)
+ PRINTLN("No catchpoint on %s.", event_names[event]);
+}
+
+/* Create new breakpoint at specified location. */
+static int
+add_breakpoint(struct rte_bpf_validate_debug *debug, uint32_t nb_ins, uint32_t pc)
+{
+ const uint32_t point_number = point_infos_append();
+
+ if (pc >= nb_ins) {
+ PRINTLN("Error: program only has %u instructions.", nb_ins);
+ return -ENOENT;
+ }
+
+ struct rte_bpf_validate_debug_point *const point =
+ rte_bpf_validate_debug_break(debug, pc,
+ &(struct rte_bpf_validate_debug_callback){
+ .fn = point_cb,
+ .ctx = (void *)(uintptr_t)point_number,
+ });
+ if (point == NULL) {
+ PRINTLN("Library error %d.", rte_errno);
+ return -rte_errno;
+ }
+
+ point_infos.elements[point_number] = (struct point_info){
+ .point = point,
+ .type = POINT_TYPE_BREAK,
+ .pc = pc,
+ };
+ point_infos_print_at(point_number);
+ return 0;
+}
+
+/* Create new catchpoint at specified location. */
+static int
+add_catchpoint(struct rte_bpf_validate_debug *debug, int event)
+{
+ if (event < 0)
+ /* Error was already printed by parse_event_name. */
+ return event;
+
+ const uint32_t point_number = point_infos_append();
+ struct rte_bpf_validate_debug_point *const point =
+ rte_bpf_validate_debug_catch(debug, event,
+ &(struct rte_bpf_validate_debug_callback){
+ .fn = point_cb,
+ .ctx = (void *)(uintptr_t)point_number,
+ });
+ if (point == NULL) {
+ PRINTLN("Library error %d.", rte_errno);
+ return -rte_errno;
+ }
+
+ point_infos.elements[point_number] = (struct point_info){
+ .point = point,
+ .type = POINT_TYPE_CATCH,
+ .event = event,
+ };
+ point_infos_print_at(point_number);
+ return 0;
+}
+
+static bool
+is_step_enabled(void)
+{
+ return step_point != NULL;
+}
+
+/* Enable step-by-step validation: make sure catchpoint is set on step event. */
+static int
+enable_step(struct rte_bpf_validate_debug *debug)
+{
+ if (is_step_enabled())
+ return 0;
+
+ step_point = rte_bpf_validate_debug_catch(debug,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_STEP,
+ &(struct rte_bpf_validate_debug_callback){ step_cb });
+ if (step_point == NULL) {
+ PRINTLN("Library error %d.", rte_errno);
+ return -rte_errno;
+ }
+
+ return 0;
+}
+
+/* Disable step-by-step validation: destroy catchpoint on step event if any. */
+static void
+disable_step(void)
+{
+ rte_bpf_validate_debug_point_destroy(step_point);
+ step_point = NULL;
+}
+
+/* Format and print information about specified frame offset. */
+static int
+print_frame_offset(struct rte_bpf_validate_debug *debug, int32_t offset)
+{
+ char *info;
+ int info_size, rc;
+
+ if (offset >= 0 || offset % sizeof(uint64_t) != 0) {
+ PRINTLN("Invalid frame offset, must be a negative multiple of %zu.",
+ sizeof(uint64_t));
+ return -EINVAL;
+ }
+
+ rc = rte_bpf_validate_debug_format_frame_info(debug, NULL, 0, offset);
+ if (rc == -ERANGE) {
+ PRINTLN("Offset is out of frame range.");
+ return rc;
+ }
+ if (rc < 0) {
+ PRINTLN("Error %d printing information.", -rc);
+ return rc;
+ }
+
+ info_size = rc + 1;
+ info = malloc(info_size);
+ RTE_VERIFY(info != NULL);
+
+ rc = rte_bpf_validate_debug_format_frame_info(debug, info, info_size,
+ offset);
+ if (rc + 1 != info_size) {
+ if (rc >= 0) {
+ PRINTLN("Expect format return value %d, got %d.",
+ info_size, rc);
+ rc = -EINVAL;
+ } else
+ PRINTLN("Error %d printing information.", -rc);
+ free(info);
+ return rc;
+ }
+
+ printf("%5" PRId32 ": \t%s\n", offset, info);
+ free(info);
+ return 0;
+}
+
+/* Format and print informatiion about the frame. */
+static int
+print_frame(struct rte_bpf_validate_debug *debug)
+{
+ int32_t frame_size;
+ int rc;
+
+ frame_size = rte_bpf_validate_debug_get_frame_size(debug);
+ if (frame_size < 0) {
+ PRINTLN("Error %d getting frame size.", -frame_size);
+ return frame_size;
+ }
+
+ for (int32_t neg_ofs = sizeof(uint64_t); neg_ofs <= frame_size;
+ neg_ofs += sizeof(uint64_t)) {
+ rc = print_frame_offset(debug, -neg_ofs);
+ if (rc < 0)
+ return rc;
+ }
+
+ return 0;
+}
+
+/* Format and print informatiion about specified register. */
+static int
+print_register(struct rte_bpf_validate_debug *debug, int reg)
+{
+ char *info;
+ int info_size, rc;
+
+ if (reg < 0)
+ /* Error was already printed by parse_register. */
+ return reg;
+
+ rc = rte_bpf_validate_debug_format_register_info(debug, NULL, 0, reg);
+ if (rc < 0) {
+ PRINTLN("Error %d printing information.", -rc);
+ return rc;
+ }
+
+ info_size = rc + 1;
+ info = malloc(info_size);
+ RTE_VERIFY(info != NULL);
+
+ rc = rte_bpf_validate_debug_format_register_info(debug, info, info_size,
+ reg);
+ if (rc + 1 != info_size) {
+ if (rc >= 0) {
+ PRINTLN("Expect format return value %d, got %d.",
+ info_size, rc);
+ rc = -EINVAL;
+ } else
+ PRINTLN("Error %d printing information.", -rc);
+ free(info);
+ return rc;
+ }
+
+ printf("%5s: \t%s\n", register_names[reg], info);
+ free(info);
+ return 0;
+}
+
+/* Format and print informatiion about all registers. */
+static int
+print_registers(struct rte_bpf_validate_debug *debug)
+{
+ int rc = 0;
+
+ for (int reg = 0; reg != EBPF_REG_NUM; ++reg)
+ rc = rc < 0 ? rc : print_register(debug, reg);
+
+ return rc;
+}
+
+/* List one eBPF program instruction. */
+static int
+list_one(const struct ebpf_insn *ins, uint32_t nb_ins, uint32_t offset,
+ uint32_t pc, const char *comment)
+{
+ char hexadecimal[256], disassembly[256];
+
+ if (offset >= nb_ins) {
+ PRINTLN("Error: program only has %u instructions.", nb_ins);
+ return -EINVAL;
+ }
+
+ ins += offset;
+
+ if (offset == nb_ins - 1 && rte_bpf_insn_is_wide(ins)) {
+ PRINTLN("Error: truncated last instruction.");
+ return -EINVAL;
+ }
+
+ rte_bpf_format(hexadecimal, sizeof(hexadecimal), ins, 0,
+ RTE_BPF_FORMAT_FLAG_HEXADECIMAL |
+ RTE_BPF_FORMAT_FLAG_NEVER_WIDE);
+ rte_bpf_format(disassembly, sizeof(disassembly), ins, offset,
+ RTE_BPF_FORMAT_FLAG_DISASSEMBLY |
+ RTE_BPF_FORMAT_FLAG_ABSOLUTE_JUMPS);
+
+ if (comment == NULL)
+ comment = "";
+ PRINTLN("%2s %10u: \t%s \t%s%s%s",
+ offset == pc ? "=>" : "", offset, hexadecimal, disassembly,
+ comment[0] != '\0' ? " \t; " : "", comment);
+
+ if (rte_bpf_insn_is_wide(ins)) {
+ rte_bpf_format(hexadecimal, sizeof(hexadecimal), ins + 1, 0,
+ RTE_BPF_FORMAT_FLAG_HEXADECIMAL |
+ RTE_BPF_FORMAT_FLAG_NEVER_WIDE);
+ PRINTLN("%15s\t%s", "", hexadecimal);
+ }
+
+ return 0;
+}
+
+/* List specified range of eBPF program instructions, updating start offset. */
+static int
+list(const struct ebpf_insn *ins, uint32_t nb_ins, uint32_t *offset,
+ uint32_t count, uint32_t pc)
+{
+ uint32_t local_offset = 0;
+ if (offset == NULL)
+ offset = &local_offset;
+
+ if (*offset > nb_ins) {
+ PRINTLN("Error: program only has %u instructions.", nb_ins);
+ return -EINVAL;
+ }
+
+ const uint32_t end =
+ /* Calculate end in a way preventing overflow: */
+ *offset + RTE_MIN(nb_ins - *offset, count);
+ while (*offset < end) {
+ const int rc = list_one(ins, nb_ins, *offset, pc, NULL);
+ if (rc < 0)
+ return rc;
+
+ *offset += 1 + rte_bpf_insn_is_wide(&ins[*offset]);
+ }
+
+ return 0;
+}
+
+/* Print if specified conditional jump _may_ be executed. */
+static int
+print_if_may(struct rte_bpf_validate_debug *debug, const struct ebpf_insn *jump,
+ uint64_t imm64)
+{
+ const int result = rte_bpf_validate_debug_may_jump(debug, jump, imm64);
+
+ switch (result) {
+ case 0:
+ case RTE_BPF_VALIDATE_DEBUG_MAY_BE_FALSE:
+ PRINTLN("NO");
+ break;
+ case RTE_BPF_VALIDATE_DEBUG_MAY_BE_TRUE:
+ case RTE_BPF_VALIDATE_DEBUG_MAY_BE_FALSE | RTE_BPF_VALIDATE_DEBUG_MAY_BE_TRUE:
+ PRINTLN("YES");
+ break;
+ default:
+ PRINTLN("Error %d getting result.", -result);
+ break;
+ }
+
+ return result;
+}
+
+/* Print if specified condition with literal right hand side _may_ be true. */
+static int
+print_if_may_literal_rhs(struct rte_bpf_validate_debug *debug,
+ const struct debug_command_comparison *comparison)
+{
+ const int lhs = parse_register(comparison->lhs);
+ const int op = parse_comparison_operator(comparison->op);
+ const int64_t rhs = comparison->literal_rhs;
+
+ if (lhs < 0 || op < 0)
+ /* Error was already printed by parse function. */
+ return lhs < 0 ? lhs : op;
+
+ return print_if_may(debug, &(struct ebpf_insn){
+ .code = BPF_JMP | op | BPF_K,
+ .dst_reg = lhs,
+ }, /* imm64 = */ rhs);
+}
+
+/* Print if specified condition with register right hand side _may_ be true. */
+static int
+print_if_may_register_rhs(struct rte_bpf_validate_debug *debug,
+ const struct debug_command_comparison *comparison)
+{
+ const int lhs = parse_register(comparison->lhs);
+ const int op = parse_comparison_operator(comparison->op);
+ const int rhs = parse_register(comparison->register_rhs);
+
+ if (lhs < 0 || op < 0 || rhs < 0)
+ /* Error was already printed by parse function. */
+ return lhs < 0 ? lhs : op < 0 ? op : rhs;
+
+ return print_if_may(debug, &(struct ebpf_insn){
+ .code = BPF_JMP | op | BPF_X,
+ .dst_reg = lhs,
+ .src_reg = rhs,
+ }, /* imm64 = */ 0);
+}
+
+/* Return 1 on validation success, 0 on failure, -EAGAIN if still running. */
+static int
+get_validation_success(struct rte_bpf_validate_debug *debug)
+{
+ int validation_result, rc;
+
+ rc = rte_bpf_validate_debug_get_validation_result(debug,
+ &validation_result);
+ return rc < 0 ? rc : (validation_result >= 0);
+}
+
+static void
+print_status(const struct ebpf_insn *ins, uint32_t nb_ins,
+ int validation_success, uint32_t pc)
+{
+ if (validation_success == 1) {
+ PRINTLN("Validation succeeded.");
+ return;
+ }
+
+ list_one(ins, nb_ins, pc, pc, NULL);
+
+ if (validation_success == 0)
+ PRINTLN("Validation failed.");
+}
+
+static void
+debug_command_where(const struct ebpf_insn *ins, uint32_t nb_ins,
+ int validation_success, uint32_t pc)
+{
+ for (uint32_t bi = 0; bi != branch_stack.length; ++bi) {
+ const uint32_t jump_pc = branch_stack.branches[bi].jump_pc;
+ const uint32_t target_pc = branch_stack.branches[bi].target_pc;
+ const char *const comment =
+ !branch_stack.branches[bi].is_conditional ? NULL :
+ target_pc == jump_pc + 1 ? "fallen-through" : "taken";
+ list_one(ins, nb_ins, jump_pc, UINT32_MAX, comment);
+ }
+ print_status(ins, nb_ins, validation_success, pc);
+}
+
+/* Return true if pc is defined, otherwise print an error message. */
+static bool
+ensure_pc_defined(uint32_t pc)
+{
+ if (pc != UINT32_MAX)
+ return true;
+
+ PRINTLN("No current instruction.");
+ return false;
+}
+
+/* Return true if still validating, otherwise print an error message. */
+static bool
+ensure_still_validating(int validation_success)
+{
+ if (validation_success == -EAGAIN)
+ return true;
+
+ PRINTLN("Finished, use `start` or `run` to restart, `quit` to quit.");
+ return false;
+}
+
+/* Step-by-step validation callback: read and process user commands. */
+static int
+step_cb(struct rte_bpf_validate_debug *debug, __rte_unused void *ctx)
+{
+ int rc;
+ int validation_success;
+ const struct ebpf_insn *ins;
+ uint32_t nb_ins, pc, list_offset;
+
+ validate_again = false;
+
+ rc = rte_bpf_validate_debug_get_ins(debug, &ins, &nb_ins);
+ if (rc < 0) {
+ PRINTLN("Error %d getting program instructions.", -rc);
+ return rc;
+ }
+
+ validation_success = get_validation_success(debug);
+ if (validation_success < 0 && validation_success != -EAGAIN) {
+ PRINTLN("Error %d getting validation result.",
+ -validation_success);
+ return validation_success;
+ }
+
+ pc = validation_success == 1 ? UINT32_MAX :
+ rte_bpf_validate_debug_get_pc(debug);
+
+ print_status(ins, nb_ins, validation_success, pc);
+
+ list_offset = pc;
+
+ while (true) {
+ switch (debug_command_get(PROMPT)) {
+ case DEBUG_COMMAND_BREAK:
+ if (ensure_pc_defined(pc))
+ add_breakpoint(debug, nb_ins, pc);
+ continue;
+ case DEBUG_COMMAND_BREAK_PC:
+ add_breakpoint(debug, nb_ins, debug_command_parsed.pc);
+ continue;
+ case DEBUG_COMMAND_CATCH:
+ add_catchpoint(debug,
+ parse_event_name(debug_command_parsed.event));
+ continue;
+ case DEBUG_COMMAND_CLEAR:
+ if (ensure_pc_defined(pc))
+ point_infos_destroy_breakpoints(pc);
+ continue;
+ case DEBUG_COMMAND_CLEAR_EVENT:
+ point_infos_destroy_catchpoints(
+ parse_event_name(debug_command_parsed.event));
+ continue;
+ case DEBUG_COMMAND_CLEAR_PC:
+ point_infos_destroy_breakpoints(
+ debug_command_parsed.pc);
+ continue;
+ case DEBUG_COMMAND_CONTINUE:
+ if (ensure_still_validating(validation_success)) {
+ disable_step();
+ return 0;
+ }
+ continue;
+ case DEBUG_COMMAND_DELETE:
+ point_infos_destroy_all();
+ continue;
+ case DEBUG_COMMAND_DELETE_NUMBER:
+ point_infos_destroy_at(
+ debug_command_parsed.point_number);
+ continue;
+ case DEBUG_COMMAND_INFO_FRAME:
+ print_frame(debug);
+ continue;
+ case DEBUG_COMMAND_INFO_FRAME_OFFSET:
+ print_frame_offset(debug,
+ debug_command_parsed.frame_offset);
+ continue;
+ case DEBUG_COMMAND_INFO_POINTS:
+ point_infos_print_all();
+ continue;
+ case DEBUG_COMMAND_INFO_REGISTER:
+ print_register(debug,
+ parse_register(debug_command_parsed.register_));
+ continue;
+ case DEBUG_COMMAND_INFO_REGISTERS:
+ print_registers(debug);
+ continue;
+ case DEBUG_COMMAND_LIST:
+ if (ensure_pc_defined(pc))
+ list(ins, nb_ins, &list_offset, 10, pc);
+ continue;
+ case DEBUG_COMMAND_LIST_COUNT:
+ if (ensure_pc_defined(pc))
+ list(ins, nb_ins, &list_offset,
+ debug_command_parsed.instruction_count, pc);
+ continue;
+ case DEBUG_COMMAND_LIST_PROGRAM:
+ if (ensure_pc_defined(pc))
+ list(ins, nb_ins, NULL, UINT32_MAX, pc);
+ continue;
+ case DEBUG_COMMAND_MAY_LITERAL_RHS:
+ print_if_may_literal_rhs(debug,
+ &debug_command_parsed.comparison);
+ continue;
+ case DEBUG_COMMAND_MAY_REGISTER_RHS:
+ print_if_may_register_rhs(debug,
+ &debug_command_parsed.comparison);
+ continue;
+ case DEBUG_COMMAND_EOF:
+ case DEBUG_COMMAND_QUIT:
+ PRINTLN("Quitting...");
+ return validation_success == -EAGAIN ? -ECANCELED : 0;
+ case DEBUG_COMMAND_RUN:
+ PRINTLN("Re-running...");
+ disable_step();
+ validate_again = true;
+ return -ECANCELED;
+ case DEBUG_COMMAND_START:
+ PRINTLN("Re-starting...");
+ rc = enable_step(debug);
+ if (rc < 0)
+ return rc;
+ validate_again = true;
+ return -ECANCELED;
+ case DEBUG_COMMAND_STEP:
+ if (ensure_still_validating(validation_success))
+ return 0;
+ continue;
+ case DEBUG_COMMAND_WHERE:
+ debug_command_where(ins, nb_ins, validation_success, pc);
+ continue;
+ default:
+ PRINTLN("INTERNAL ERROR");
+ return -ENOTSUP;
+ }
+ }
+}
+
+/* Any point callback: print it and enable step-by-step validation. */
+static int
+point_cb(struct rte_bpf_validate_debug *debug, void *ctx)
+{
+ const uint32_t point_number = (uintptr_t)ctx;
+ point_infos_print_at(point_number);
+ return enable_step(debug);
+}
+
+/* Branch stack machinery. */
+
+static void
+clear_jump_always_step_point(void)
+{
+ rte_bpf_validate_debug_point_destroy(jump_always_step_point);
+ jump_always_step_point = NULL;
+}
+
+static int
+reset_branch_tracking(struct rte_bpf_validate_debug *debug __rte_unused,
+ void *ctx __rte_unused)
+{
+ branch_stack.length = 0;
+ pending_jump_pc = UINT32_MAX;
+ clear_jump_always_step_point();
+ return 0;
+}
+
+/*
+ * Handling the jump-always instructions:
+ * - upon a jump-always event save the pc and set a custom step callback;
+ * - ignore the first custom step callback call (still on the same jump);
+ * - on the second call push the jump pc into stack and delete the callback;
+ */
+
+static int
+jump_always_step_cb(struct rte_bpf_validate_debug *debug, void *ctx __rte_unused)
+{
+ /* Step event is also emitted at the end of the jump instruction itself. */
+ if (rte_bpf_validate_debug_get_pc(debug) == pending_jump_pc)
+ return 0;
+
+ branch_stack_append(&(struct branch_info){
+ .jump_pc = pending_jump_pc,
+ .target_pc = rte_bpf_validate_debug_get_pc(debug),
+ .is_conditional = false,
+ });
+ clear_jump_always_step_point();
+ return 0;
+}
+
+static int
+jump_always_cb(struct rte_bpf_validate_debug *debug, void *ctx __rte_unused)
+{
+ pending_jump_pc = rte_bpf_validate_debug_get_pc(debug);
+ clear_jump_always_step_point();
+ jump_always_step_point = rte_bpf_validate_debug_catch(debug,
+ RTE_BPF_VALIDATE_DEBUG_EVENT_STEP,
+ &(struct rte_bpf_validate_debug_callback){ jump_always_step_cb });
+ return 0;
+}
+
+/*
+ * Handling the jump-conditional instructions:
+ * - upon a jump-conditional event save the pc;
+ * - upon a branch-enter event push the conditional jump pc into stack;
+ * - upon a branch-return event pop conditional jump and all unconditional jumps
+ * preceding it from the stack;
+ * - during step execution notify the user about the step events;
+ */
+
+static int
+jump_conditional_cb(struct rte_bpf_validate_debug *debug, void *ctx __rte_unused)
+{
+ pending_jump_pc = rte_bpf_validate_debug_get_pc(debug);
+ return 0;
+}
+
+static int
+branch_enter_cb(struct rte_bpf_validate_debug *debug, void *ctx __rte_unused)
+{
+ branch_stack_append(&(struct branch_info){
+ .jump_pc = pending_jump_pc,
+ .target_pc = rte_bpf_validate_debug_get_pc(debug),
+ .is_conditional = true,
+ });
+ if (is_step_enabled())
+ PRINTLN("Entered new branch at pc %u.", pending_jump_pc);
+ return 0;
+}
+
+static int
+branch_return_cb(struct rte_bpf_validate_debug *debug __rte_unused,
+ void *ctx __rte_unused)
+{
+ clear_jump_always_step_point();
+
+ while (branch_stack.length > 0) {
+ const struct branch_info branch_info =
+ branch_stack.branches[--branch_stack.length];
+ pending_jump_pc = branch_info.jump_pc;
+ if (branch_info.is_conditional)
+ break;
+ }
+ if (is_step_enabled())
+ PRINTLN("Returned from branch at pc %u.", pending_jump_pc);
+ return 0;
+}
+
+/* Notify user when skipping branches in step mode. */
+
+static int
+branch_prune_cb(struct rte_bpf_validate_debug *debug __rte_unused,
+ void *ctx __rte_unused)
+{
+ if (is_step_enabled())
+ PRINTLN("Prunned branch at pc %u.", pending_jump_pc);
+ return 0;
+}
+
+static int
+branch_unreachable_cb(struct rte_bpf_validate_debug *debug __rte_unused,
+ void *ctx __rte_unused)
+{
+ if (is_step_enabled())
+ PRINTLN("Unreachable branch at pc %u.", pending_jump_pc);
+ return 0;
+}
+
+/* Global initialization and cleanup functions. */
+
+static int
+set_callbacks(struct rte_bpf_validate_debug *debug)
+{
+ static const struct rte_bpf_validate_debug_callback events_callback[
+ RTE_BPF_VALIDATE_DEBUG_EVENT_END] = {
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_START] = { reset_branch_tracking },
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_SUCCESS] = { step_cb },
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_VALIDATION_FAILURE] = { step_cb },
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_CONDITIONAL] = { jump_conditional_cb },
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_JUMP_ALWAYS] = { jump_always_cb },
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_ENTER] = { branch_enter_cb },
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_RETURN] = { branch_return_cb },
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_PRUNE] = { branch_prune_cb },
+ [RTE_BPF_VALIDATE_DEBUG_EVENT_BRANCH_UNREACHABLE] = { branch_unreachable_cb },
+ };
+
+ for (enum rte_bpf_validate_debug_event event = 0;
+ event < RTE_BPF_VALIDATE_DEBUG_EVENT_END; ++event)
+ if (events_callback[event].fn != NULL &&
+ rte_bpf_validate_debug_catch(debug, event,
+ &events_callback[event]) == NULL)
+ return -rte_errno;
+
+ return 0;
+}
+
+struct rte_bpf_validate_debug *
+debug_create(void)
+{
+
+ int rc = 0;
+
+ struct rte_bpf_validate_debug *const debug = rte_bpf_validate_debug_create();
+ if (debug == NULL)
+ rc = -rte_errno;
+
+ rc = rc < 0 ? rc : set_callbacks(debug);
+
+ rc = rc < 0 ? rc : enable_step(debug);
+
+ if (rc < 0) {
+ debug_destroy(debug);
+ rte_errno = -rc;
+ return NULL;
+ }
+
+ return debug;
+}
+
+void
+debug_destroy(struct rte_bpf_validate_debug *debug)
+{
+ /* No need to destroy created points, destroying debug will do it. */
+ point_infos_free();
+ branch_stack_free();
+ rte_bpf_validate_debug_destroy(debug);
+}
+
+bool
+debug_validate_again(void)
+{
+ return validate_again;
+}
diff --git a/app/validate-bpf/debug_command.c b/app/validate-bpf/debug_command.c
new file mode 100644
index 0000000000..505ae389c9
--- /dev/null
+++ b/app/validate-bpf/debug_command.c
@@ -0,0 +1,383 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include "debug_command.h"
+
+#include <rte_debug.h>
+#include <cmdline.h>
+#include <cmdline_socket.h>
+#include <cmdline_parse_num.h>
+
+#include <string.h>
+
+
+#define EVENT_PATTERN \
+ "invalid-state#" \
+ "branch-enter#branch-prune#branch-return#branch-unreachable#" \
+ "jump-always#jump-conditional"
+
+#define REGISTER_PATTERN "r0#r1#r2#r3#r4#r5#r6#r7#r8#r9#r10"
+
+#define COMPARISON_OP_PATTERN "==#!=#<#<=#>#>=#s<#s<=#s>#s>="
+
+static void
+handle_command(void *parsed_result, struct cmdline *cl, void *data);
+
+/* Keywords */
+static cmdline_parse_token_string_t break_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "b#break");
+static cmdline_parse_token_string_t may_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "may");
+static cmdline_parse_token_string_t catch_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "catch");
+static cmdline_parse_token_string_t clear_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "clear");
+static cmdline_parse_token_string_t continue_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "c#continue");
+static cmdline_parse_token_string_t delete_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "delete");
+static cmdline_parse_token_string_t info_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "i#info");
+static cmdline_parse_token_string_t info_points_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword,
+ "b#break#breakpoints#points");
+static cmdline_parse_token_string_t info_frame_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "f#frame");
+static cmdline_parse_token_string_t info_registers_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "r#registers");
+static cmdline_parse_token_string_t list_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "l#list");
+static cmdline_parse_token_string_t program_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "program");
+static cmdline_parse_token_string_t quit_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "q#quit");
+static cmdline_parse_token_string_t run_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "run");
+static cmdline_parse_token_string_t start_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "start");
+static cmdline_parse_token_string_t step_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "s#step");
+static cmdline_parse_token_string_t where_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, keyword, "where");
+
+/* Variable tokens */
+static cmdline_parse_token_string_t comparison_lhs_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, comparison.lhs, REGISTER_PATTERN);
+static cmdline_parse_token_string_t comparison_op_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, comparison.op, COMPARISON_OP_PATTERN);
+static cmdline_parse_token_num_t comparison_rhs_literal_tok =
+ TOKEN_NUM_INITIALIZER(struct debug_command_parsed, comparison.literal_rhs, RTE_INT64);
+static cmdline_parse_token_string_t comparison_rhs_register_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, comparison.register_rhs,
+ REGISTER_PATTERN);
+static cmdline_parse_token_string_t event_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, event, EVENT_PATTERN);
+static cmdline_parse_token_num_t frame_offset_tok =
+ TOKEN_NUM_INITIALIZER(struct debug_command_parsed, frame_offset, RTE_INT64);
+static cmdline_parse_token_num_t instruction_count_tok =
+ TOKEN_NUM_INITIALIZER(struct debug_command_parsed, instruction_count, RTE_UINT32);
+static cmdline_parse_token_num_t pc_tok =
+ TOKEN_NUM_INITIALIZER(struct debug_command_parsed, instruction_offset, RTE_UINT32);
+static cmdline_parse_token_num_t point_number_tok =
+ TOKEN_NUM_INITIALIZER(struct debug_command_parsed, point_number, RTE_UINT32);
+static cmdline_parse_token_string_t register_tok =
+ TOKEN_STRING_INITIALIZER(struct debug_command_parsed, register_, REGISTER_PATTERN);
+
+
+/* Commands */
+static cmdline_parse_inst_t cmd_break = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_BREAK,
+ .help_str = "b|break: break at current instruction",
+ .tokens = {
+ (void *)&break_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_break_pc = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_BREAK_PC,
+ .help_str = "b|break <pc>: break at specified instruction",
+ .tokens = {
+ (void *)&break_tok,
+ (void *)&pc_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_catch = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_CATCH,
+ .help_str = "catch <event>: catch specified event",
+ .tokens = {
+ (void *)&catch_tok,
+ (void *)&event_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_clear = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_CLEAR,
+ .help_str = "clear: delete all breakpoints at current instruction",
+ .tokens = {
+ (void *)&clear_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_clear_event = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_CLEAR_EVENT,
+ .help_str = "clear <event>: delete all catchpoints for specified event",
+ .tokens = {
+ (void *)&clear_tok,
+ (void *)&event_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_clear_pc = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_CLEAR_PC,
+ .help_str = "clear <pc>: delete all breakpoints at specified instruction",
+ .tokens = {
+ (void *)&clear_tok,
+ (void *)&pc_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_continue = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_CONTINUE,
+ .help_str = "c|continue: continue validation",
+ .tokens = {
+ (void *)&continue_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_delete = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_DELETE,
+ .help_str = "delete: delete all breakpoints and catchpoints",
+ .tokens = {
+ (void *)&delete_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_delete_number = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_DELETE_NUMBER,
+ .help_str = "delete <point>: delete specified breakpoint or catchpoint",
+ .tokens = {
+ (void *)&delete_tok,
+ (void *)&point_number_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_info_frame = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_INFO_FRAME,
+ .help_str = "i|info f|frame: show information about all frame locations",
+ .tokens = {
+ (void *)&info_tok,
+ (void *)&info_frame_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_info_frame_offset = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_INFO_FRAME_OFFSET,
+ .help_str = "i|info f|frame -<offset>: show information about specified frame location",
+ .tokens = {
+ (void *)&info_tok,
+ (void *)&info_frame_tok,
+ (void *)&frame_offset_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_info_points = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_INFO_POINTS,
+ .help_str = "i|info b|break|breakpoints|points: "
+ "show information about all breakpoints and catchpoints",
+ .tokens = {
+ (void *)&info_tok,
+ (void *)&info_points_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_info_register = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_INFO_REGISTER,
+ .help_str = "i|info <register>: show information about specified register",
+ .tokens = {
+ (void *)&info_tok,
+ (void *)®ister_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_info_registers = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_INFO_REGISTERS,
+ .help_str = "i|info r|registers: show information about all registers",
+ .tokens = {
+ (void *)&info_tok,
+ (void *)&info_registers_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_list = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_LIST,
+ .help_str = "l|list: list ten instructions",
+ .tokens = {
+ (void *)&list_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_list_count = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_LIST_COUNT,
+ .help_str = "l|list <number>: list specified number of instructions",
+ .tokens = {
+ (void *)&list_tok,
+ (void *)&instruction_count_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_list_program = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_LIST_PROGRAM,
+ .help_str = "l|list program: list whole program",
+ .tokens = {
+ (void *)&list_tok,
+ (void *)&program_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_may_literal_rhs = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_MAY_LITERAL_RHS,
+ .help_str = "may <register> <comparison> <number>: "
+ "check if specified condition _may_ be true",
+ .tokens = {
+ (void *)&may_tok,
+ (void *)&comparison_lhs_tok,
+ (void *)&comparison_op_tok,
+ (void *)&comparison_rhs_literal_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_may_register_rhs = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_MAY_REGISTER_RHS,
+ .help_str = "may <register> <comparison> <register>: "
+ "check if specified condition _may_ be true",
+ .tokens = {
+ (void *)&may_tok,
+ (void *)&comparison_lhs_tok,
+ (void *)&comparison_op_tok,
+ (void *)&comparison_rhs_register_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_quit = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_QUIT,
+ .help_str = "quit: q|quit debugger",
+ .tokens = {
+ (void *)&quit_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_run = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_RUN,
+ .help_str = "run: re-run validation from the start",
+ .tokens = {
+ (void *)&run_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_start = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_START,
+ .help_str = "start: re-start validation and stop at start",
+ .tokens = {
+ (void *)&start_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_step = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_STEP,
+ .help_str = "s|step: validate one instruction",
+ .tokens = {
+ (void *)&step_tok,
+ NULL,
+ }
+};
+static cmdline_parse_inst_t cmd_where = {
+ .f = handle_command,
+ .data = (void *)DEBUG_COMMAND_WHERE,
+ .help_str = "where: show current branch stack",
+ .tokens = {
+ (void *)&where_tok,
+ NULL,
+ }
+};
+
+static cmdline_parse_ctx_t debug_ctx[] = {
+ &cmd_break,
+ &cmd_break_pc,
+ &cmd_catch,
+ &cmd_clear,
+ &cmd_clear_event,
+ &cmd_clear_pc,
+ &cmd_continue,
+ &cmd_delete,
+ &cmd_delete_number,
+ &cmd_info_frame,
+ &cmd_info_frame_offset,
+ &cmd_info_points,
+ &cmd_info_register,
+ &cmd_info_registers,
+ &cmd_list,
+ &cmd_list_count,
+ &cmd_list_program,
+ &cmd_may_literal_rhs,
+ &cmd_may_register_rhs,
+ &cmd_quit,
+ &cmd_run,
+ &cmd_start,
+ &cmd_step,
+ &cmd_where,
+ NULL
+};
+
+/* Receive, fill and return one command. */
+
+static enum debug_command debug_command;
+
+struct debug_command_parsed debug_command_parsed;
+
+static void
+handle_command(void *parsed_result, struct cmdline *cl, void *data)
+{
+ RTE_BUILD_BUG_ON(sizeof(debug_command_parsed) > CMDLINE_PARSE_RESULT_BUFSIZE);
+ memcpy(&debug_command_parsed, parsed_result, sizeof(debug_command_parsed));
+ debug_command = (uintptr_t)data;
+ cmdline_quit(cl);
+}
+
+enum debug_command
+debug_command_get(const char *prompt)
+{
+ debug_command = DEBUG_COMMAND_EOF;
+ struct cmdline *const cmdline = cmdline_stdin_new(debug_ctx, prompt);
+ RTE_VERIFY(cmdline != NULL);
+ cmdline_interact(cmdline);
+ cmdline_stdin_exit(cmdline);
+ /* Clear prompt, or it would prepend first message that follows. */
+ printf("\r%*s\r", (int)strlen(prompt), "");
+ fflush(stdout);
+ return debug_command;
+}
diff --git a/app/validate-bpf/debug_command.h b/app/validate-bpf/debug_command.h
new file mode 100644
index 0000000000..1a59ab251e
--- /dev/null
+++ b/app/validate-bpf/debug_command.h
@@ -0,0 +1,65 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include <cmdline_parse_string.h>
+
+#include <stdint.h>
+
+
+struct debug_command_comparison {
+ cmdline_fixed_string_t lhs;
+ cmdline_fixed_string_t op;
+ union {
+ cmdline_fixed_string_t register_rhs;
+ int64_t literal_rhs;
+ };
+};
+
+struct debug_command_parsed {
+ cmdline_fixed_string_t keyword; /* Any keyword we don't need */
+ union {
+ struct debug_command_comparison comparison;
+ cmdline_fixed_string_t event;
+ int32_t frame_offset;
+ uint32_t instruction_count;
+ uint32_t instruction_offset;
+ uint32_t pc;
+ uint32_t point_number;
+ cmdline_fixed_string_t register_;
+ };
+};
+
+enum debug_command {
+ DEBUG_COMMAND_EOF,
+ DEBUG_COMMAND_BREAK,
+ DEBUG_COMMAND_BREAK_PC,
+ DEBUG_COMMAND_CATCH,
+ DEBUG_COMMAND_CLEAR,
+ DEBUG_COMMAND_CLEAR_EVENT,
+ DEBUG_COMMAND_CLEAR_PC,
+ DEBUG_COMMAND_CONTINUE,
+ DEBUG_COMMAND_DELETE,
+ DEBUG_COMMAND_DELETE_NUMBER,
+ DEBUG_COMMAND_INFO_FRAME,
+ DEBUG_COMMAND_INFO_FRAME_OFFSET,
+ DEBUG_COMMAND_INFO_POINTS,
+ DEBUG_COMMAND_INFO_REGISTER,
+ DEBUG_COMMAND_INFO_REGISTERS,
+ DEBUG_COMMAND_LIST,
+ DEBUG_COMMAND_LIST_COUNT,
+ DEBUG_COMMAND_LIST_PROGRAM,
+ DEBUG_COMMAND_MAY_LITERAL_RHS,
+ DEBUG_COMMAND_MAY_REGISTER_RHS,
+ DEBUG_COMMAND_QUIT,
+ DEBUG_COMMAND_RUN,
+ DEBUG_COMMAND_START,
+ DEBUG_COMMAND_STEP,
+ DEBUG_COMMAND_WHERE,
+};
+
+extern struct debug_command_parsed debug_command_parsed;
+
+/** Get and return one command line command, storing its data in the struct above. */
+enum debug_command
+debug_command_get(const char *prompt);
diff --git a/app/validate-bpf/eal_init_args.c b/app/validate-bpf/eal_init_args.c
new file mode 100644
index 0000000000..6ce9d561e0
--- /dev/null
+++ b/app/validate-bpf/eal_init_args.c
@@ -0,0 +1,57 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include "internal.h"
+
+
+#define APP_EAL_INIT_ARG_SIZE_MAX sizeof("--log-level=lib.eal:warning")
+
+static const char APP_EAL_INIT_ARGS[][APP_EAL_INIT_ARG_SIZE_MAX] = {
+ "--log-level=lib.eal:warning",
+ "--no-huge",
+ "--no-pci",
+ "--no-hpet",
+ "--no-shconf",
+};
+
+#define APP_EAL_INIT_ARGC (/* program name */ 1 + RTE_DIM(APP_EAL_INIT_ARGS))
+
+int
+get_eal_init_argc(void)
+{
+ return APP_EAL_INIT_ARGC;
+}
+
+/*
+ * We cannot just return literal strings here, because rte_eal_init accepts
+ * an array of mutable pointers to mutable strings, so literals won't work.
+ * Instead we build a copy of APP_EAL_INIT_ARGS in a static mutable area.
+ */
+char**
+get_eal_init_argv(char *prog_name)
+{
+ /* Static arrays for mutable copies of actual args */
+ static char mutable_args[RTE_DIM(APP_EAL_INIT_ARGS)]
+ [APP_EAL_INIT_ARG_SIZE_MAX];
+ /*
+ * Static array for pointers to args. First element will hold pointer
+ * to the prog_name, last will be set to NULL, the rest will point to
+ * elements of mutable_args with index one smaller.
+ */
+ static char *mutable_ptrs[APP_EAL_INIT_ARGC + /* terminating NULL */ 1];
+
+ mutable_ptrs[0] = prog_name;
+ for (int argi = 0; argi != RTE_DIM(APP_EAL_INIT_ARGS); ++argi) {
+ const char * const const_arg = APP_EAL_INIT_ARGS[argi];
+ char * const mutable_arg = mutable_args[argi];
+ const int snprintf_rc = snprintf(mutable_arg,
+ APP_EAL_INIT_ARG_SIZE_MAX, "%s", const_arg);
+ RTE_VERIFY(snprintf_rc >= 0 &&
+ (size_t)snprintf_rc < APP_EAL_INIT_ARG_SIZE_MAX);
+ mutable_ptrs[1 + argi] = mutable_arg;
+ }
+ mutable_ptrs[RTE_DIM(mutable_ptrs) - 1] = NULL;
+
+ return mutable_ptrs;
+}
diff --git a/app/validate-bpf/internal.h b/app/validate-bpf/internal.h
new file mode 100644
index 0000000000..e134e68314
--- /dev/null
+++ b/app/validate-bpf/internal.h
@@ -0,0 +1,151 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include <rte_bpf.h>
+#include <rte_log.h>
+
+
+extern int validate_bpf_logtype;
+#define RTE_LOGTYPE_VALIDATE_BPF validate_bpf_logtype
+#define VALIDATE_BPF_LOG(level, ...) \
+ RTE_LOG_LINE(level, VALIDATE_BPF, "" __VA_ARGS__)
+
+struct rte_bpf_validate_debug;
+
+/**
+ * List of pointers to allocations, to keep track of things to free.
+ *
+ * May contain NULLs. May contain itself directly or via an outer struct.
+ */
+struct alloc_list {
+ size_t count;
+ void **ptrs;
+};
+
+/** Add new pointer to the alloc list, return its index. */
+size_t alloc_list_append(struct alloc_list *alloc_list, void *ptr);
+
+/** Replace pointer at the specified index in the alloc list. */
+void alloc_list_replace(struct alloc_list *alloc_list, size_t index, void *ptr);
+
+/** Free all allocations in the alloc list */
+void alloc_list_free_all(struct alloc_list *alloc_list);
+
+
+/** Parsed program arguments */
+struct args {
+ /* Allocations list, args.c internal use only. */
+ struct alloc_list _alloc_list;
+
+ /* Set if command line contains --help */
+ bool show_help;
+
+ /* BPF load parameters. */
+ struct rte_bpf_prm_ex bpf_prm;
+
+ /* Size of the mbuf data buffer. */
+ size_t mbuf_buf_size;
+};
+
+/** Print program usage information */
+void
+print_usage(const char *program_name);
+
+/** Print program defaults. */
+void
+print_defaults(void);
+
+/**
+ * Parse command-line arguments
+ *
+ * @param argc
+ * Command-line arguments count, as received by main.
+ * @param argv
+ * Command-line arguments array, as received by main.
+ * Modified during the call due to the use of getopt_long.
+ * Modifying it after the call invalidates the return value.
+ * @return
+ * - parse results in case of success;
+ * - NULL in case of an error (diagnostic will be printed to stderr)
+ */
+struct args *
+args_parse(int argc, char *argv[]);
+
+/** Destroy args struct returned by args_parse */
+void
+args_destroy(struct args *args);
+
+
+/** Value for the rte_eal_init argc argument */
+int
+get_eal_init_argc(void);
+
+/**
+ * Value for the rte_eal_init argv argument
+ *
+ * @param prog_name
+ * Program name, can be obtained from argv[0]
+ */
+char **
+get_eal_init_argv(char *prog_name);
+
+
+/** Print types supported in command-line. */
+void
+print_supported_types(void);
+
+/**
+ * Parse text into arg.
+ * @param arg
+ * Pointer to a variable to put parse result to.
+ * Member `buf_size` of types related to `struct rte_mbuf` is set to
+ * `RTE_MBUF_DEFAULT_BUF_SIZE`, adjust using `adjust_arg_buf_size` if needed.
+ * @param text
+ * Text representation of the type, e.g. "struct rte_mbuf *".
+ * @param alloc_list
+ * Alloc list to use for new allocations.
+ * @return
+ * 0 on success
+ * -1 on failure
+ */
+int
+parse_arg(struct rte_bpf_arg *arg, const char *text);
+
+/**
+ * Parse text into xsym.
+ * @param arg
+ * Pointer to a variable to put parse result to.
+ * Member `buf_size` of types related to `struct rte_mbuf` is set to
+ * `RTE_MBUF_DEFAULT_BUF_SIZE`, adjust using `adjust_xsym_buf_size` if needed.
+ * @param text
+ * Text representation of the external symbol, e.g. "void exit(uint32_t)".
+ * @param alloc_list
+ * Alloc list to use for new allocations.
+ * @return
+ * 0 on success
+ * -1 on failure
+ */
+int
+parse_xsym(struct rte_bpf_xsym *xsym, const char *text,
+ struct alloc_list *alloc_list);
+
+/** If arg->buf_size is non-zero, set it to mbuf_buf_size */
+void
+adjust_arg_buf_size(struct rte_bpf_arg *arg, size_t mbuf_buf_size);
+
+/** If buf_size members are non-zero, set them to mbuf_buf_size */
+void
+adjust_xsym_buf_size(struct rte_bpf_xsym *xsym, size_t mbuf_buf_size);
+
+/** Create and set up global debugging session. */
+struct rte_bpf_validate_debug *
+debug_create(void);
+
+/** Clear and destroy global debugging session. */
+void
+debug_destroy(struct rte_bpf_validate_debug *debug);
+
+/** Tells caller if validation should be re-tried. */
+bool
+debug_validate_again(void);
diff --git a/app/validate-bpf/main.c b/app/validate-bpf/main.c
new file mode 100644
index 0000000000..c2f7399376
--- /dev/null
+++ b/app/validate-bpf/main.c
@@ -0,0 +1,90 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include "internal.h"
+
+#include <stdio.h>
+#include <stdlib.h>
+
+#include <rte_bpf.h>
+#include <rte_debug.h>
+#include <rte_eal.h>
+#include <rte_errno.h>
+
+RTE_LOG_REGISTER(validate_bpf_logtype, validate-bpf, NOTICE);
+
+static int
+test_bpf_load(struct rte_bpf_prm_ex *prm)
+{
+ struct rte_bpf * const bpf = rte_bpf_load_ex(prm);
+
+ const int rc = -rte_errno;
+
+ rte_bpf_destroy(bpf);
+
+ return bpf == NULL ? rc : 0;
+}
+
+/* Re-starts validation of asked from interactive debugger. */
+static int
+test_bpf_load_with_restarts(struct rte_bpf_prm_ex *prm)
+{
+ for (;;) {
+ const int rc = test_bpf_load(prm);
+
+ if (rc == -ECANCELED && debug_validate_again())
+ continue;
+
+ if (rc != 0)
+ fprintf(stderr, "Error %d loading BPF: %s\n",
+ -rc, strerror(-rc));
+ else
+ fprintf(stderr, "Validation succeeded.\n");
+
+ return rc;
+ }
+}
+
+int
+main(int argc, char *argv[])
+{
+ int rc;
+
+ struct args * const args = args_parse(argc, argv);
+ if (args == NULL || args->show_help
+ || args->bpf_prm.elf_file.path == NULL) {
+ args_destroy(args);
+ print_usage(argv[0]);
+ if (args == NULL)
+ /* Could not parse arguments. */
+ return 2;
+ print_supported_types();
+ print_defaults();
+ return 0;
+ }
+
+ const int eal_init_argc = get_eal_init_argc();
+ char ** const eal_init_argv = get_eal_init_argv(argv[0]);
+ rc = rte_eal_init(eal_init_argc, eal_init_argv);
+ if (rc < 0) {
+ fprintf(stderr, "Error %d initializing EAL: %s\n",
+ rte_errno, strerror(rte_errno));
+ args_destroy(args);
+ return EXIT_FAILURE;
+ }
+ RTE_VERIFY(rc == eal_init_argc - 1);
+
+ const int ret = test_bpf_load_with_restarts(&args->bpf_prm);
+
+ args_destroy(args);
+
+ rc = rte_eal_cleanup();
+ if (rc < 0) {
+ fprintf(stderr, "Error %d cleaning up EAL: %s\n",
+ rte_errno, strerror(rte_errno));
+ return EXIT_FAILURE;
+ }
+
+ return ret < 0 ? EXIT_FAILURE : EXIT_SUCCESS;
+}
diff --git a/app/validate-bpf/meson.build b/app/validate-bpf/meson.build
new file mode 100644
index 0000000000..03cfc329ef
--- /dev/null
+++ b/app/validate-bpf/meson.build
@@ -0,0 +1,13 @@
+# SPDX-License-Identifier: BSD-3-Clause
+# Copyright(c) 2025 Huawei Technologies Co., Ltd
+
+sources = files(
+ 'alloc_list.c',
+ 'args.c',
+ 'debug.c',
+ 'debug_command.c',
+ 'eal_init_args.c',
+ 'main.c',
+ 'parse_decl.c',
+)
+deps = ['bpf', 'cmdline']
diff --git a/app/validate-bpf/parse_decl.c b/app/validate-bpf/parse_decl.c
new file mode 100644
index 0000000000..1f95d90d37
--- /dev/null
+++ b/app/validate-bpf/parse_decl.c
@@ -0,0 +1,616 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2025 Huawei Technologies Co., Ltd
+ */
+
+#include "internal.h"
+
+#include <ctype.h>
+#include <stdlib.h>
+
+#include <rte_ether.h>
+#include <rte_ip.h>
+#include <rte_mbuf_core.h>
+#include <rte_tcp.h>
+#include <rte_udp.h>
+
+#define RETURN_TEXT_ERROR(text, text_start, message, ...) do { \
+ const int _offset = (text) - (text_start); \
+ VALIDATE_BPF_LOG(ERR, "at offset %d: " message, \
+ _offset, ## __VA_ARGS__); \
+ VALIDATE_BPF_LOG(NOTICE, "%s", text_start); \
+ VALIDATE_BPF_LOG(NOTICE, "%*c", _offset + 1, '^'); \
+ return -1; \
+} while (0)
+
+/* Used in place of any signature, this is not really being checked. */
+static uint64_t
+dummy_function(uint64_t arg1, uint64_t arg2, uint64_t arg3, uint64_t arg4,
+ uint64_t arg5)
+{
+ RTE_SET_USED(arg1);
+ RTE_SET_USED(arg2);
+ RTE_SET_USED(arg3);
+ RTE_SET_USED(arg4);
+ RTE_SET_USED(arg5);
+ return 0;
+}
+
+
+/* TOKENS AND TYPES */
+
+enum token {
+ TOKEN_UNRECOGNIZED = -1,
+ TOKEN_END = 0,
+
+ TOKEN_ASTERISK,
+ TOKEN_BRACKET_CLOSE,
+ TOKEN_BRACKET_OPEN,
+ TOKEN_COMMA,
+ TOKEN_PARENTHESIS_CLOSE,
+ TOKEN_PARENTHESIS_OPEN,
+ TOKEN_STRUCT,
+
+ TOKEN_TYPES_BEGIN,
+
+ TOKEN_TYPE_CHAR,
+ TOKEN_TYPE_ETHER_HEADER,
+ TOKEN_TYPE_INT32_T,
+ TOKEN_TYPE_IP_HEADER,
+ TOKEN_TYPE_IP_HEADERS,
+ TOKEN_TYPE_RTE_MBUF,
+ TOKEN_TYPE_TCP_HEADER,
+ TOKEN_TYPE_TCP_HEADERS,
+ TOKEN_TYPE_UDP_HEADER,
+ TOKEN_TYPE_UDP_HEADERS,
+ TOKEN_TYPE_UINT32_T,
+ TOKEN_TYPE_UINT64_T,
+ TOKEN_TYPE_UINTPTR_T,
+ TOKEN_TYPE_VOID,
+
+ TOKEN_TYPES_END,
+};
+
+/* Return true if the token is a type token */
+static bool
+is_type_token(enum token token)
+{
+ return token > TOKEN_TYPES_BEGIN && token < TOKEN_TYPES_END;
+}
+
+struct text_token {
+ const char *text;
+ unsigned int length; /* Not size_t to fit this struct in 2 regs. */
+ enum token token;
+};
+
+#define TEXT_TOKEN_DEF(text_, token_) { \
+ .text = text_, \
+ .length = sizeof(text_) - 1, \
+ .token = token_, \
+}
+
+/* Token search table, MUST BE SORTED! */
+const struct text_token text_tokens[] = {
+ TEXT_TOKEN_DEF("(", TOKEN_PARENTHESIS_OPEN),
+ TEXT_TOKEN_DEF(")", TOKEN_PARENTHESIS_CLOSE),
+ TEXT_TOKEN_DEF("*", TOKEN_ASTERISK),
+ TEXT_TOKEN_DEF(",", TOKEN_COMMA),
+ TEXT_TOKEN_DEF("[", TOKEN_BRACKET_OPEN),
+ TEXT_TOKEN_DEF("]", TOKEN_BRACKET_CLOSE),
+ TEXT_TOKEN_DEF("char", TOKEN_TYPE_CHAR),
+ TEXT_TOKEN_DEF("ether_header", TOKEN_TYPE_ETHER_HEADER),
+ TEXT_TOKEN_DEF("int32_t", TOKEN_TYPE_INT32_T),
+ TEXT_TOKEN_DEF("ip_header", TOKEN_TYPE_IP_HEADER),
+ TEXT_TOKEN_DEF("ip_headers", TOKEN_TYPE_IP_HEADERS),
+ TEXT_TOKEN_DEF("rte_ether_hdr", TOKEN_TYPE_ETHER_HEADER),
+ TEXT_TOKEN_DEF("rte_ipv4_hdr", TOKEN_TYPE_IP_HEADER),
+ TEXT_TOKEN_DEF("rte_mbuf", TOKEN_TYPE_RTE_MBUF),
+ TEXT_TOKEN_DEF("rte_tcp_hdr", TOKEN_TYPE_TCP_HEADER),
+ TEXT_TOKEN_DEF("rte_udp_hdr", TOKEN_TYPE_UDP_HEADER),
+ TEXT_TOKEN_DEF("struct", TOKEN_STRUCT),
+ TEXT_TOKEN_DEF("tcp_header", TOKEN_TYPE_TCP_HEADER),
+ TEXT_TOKEN_DEF("tcp_headers", TOKEN_TYPE_TCP_HEADERS),
+ TEXT_TOKEN_DEF("udp_header", TOKEN_TYPE_UDP_HEADER),
+ TEXT_TOKEN_DEF("udp_headers", TOKEN_TYPE_UDP_HEADERS),
+ TEXT_TOKEN_DEF("uint32_t", TOKEN_TYPE_UINT32_T),
+ TEXT_TOKEN_DEF("uint64_t", TOKEN_TYPE_UINT64_T),
+ TEXT_TOKEN_DEF("uintptr_t", TOKEN_TYPE_UINTPTR_T),
+ TEXT_TOKEN_DEF("void", TOKEN_TYPE_VOID),
+};
+
+const struct text_token text_token_unrecognized = { "", 0, TOKEN_UNRECOGNIZED };
+const struct text_token text_token_end = { "", 0, TOKEN_END };
+
+/* IP header maximum possible size with options. */
+#define IP_HEADER_MAX_SIZE 60
+#define IP_HEADERS_MAX_SIZE (sizeof(struct rte_ether_hdr) + IP_HEADER_MAX_SIZE)
+#define TCP_HEADERS_MAX_SIZE (IP_HEADERS_MAX_SIZE + sizeof(struct rte_tcp_hdr))
+#define UDP_HEADERS_MAX_SIZE (IP_HEADERS_MAX_SIZE + sizeof(struct rte_udp_hdr))
+
+#define POINTER_SIZE (sizeof(char *))
+
+const size_t type_token_size[] = {
+ [TOKEN_TYPE_CHAR] = sizeof(char),
+ [TOKEN_TYPE_ETHER_HEADER] = sizeof(struct rte_ether_hdr),
+ [TOKEN_TYPE_INT32_T] = sizeof(int32_t),
+ [TOKEN_TYPE_IP_HEADERS] = IP_HEADERS_MAX_SIZE,
+ [TOKEN_TYPE_IP_HEADER] = sizeof(struct rte_ipv4_hdr),
+ [TOKEN_TYPE_TCP_HEADERS] = TCP_HEADERS_MAX_SIZE,
+ [TOKEN_TYPE_TCP_HEADER] = sizeof(struct rte_tcp_hdr),
+ [TOKEN_TYPE_UDP_HEADERS] = UDP_HEADERS_MAX_SIZE,
+ [TOKEN_TYPE_UDP_HEADER] = sizeof(struct rte_udp_hdr),
+ [TOKEN_TYPE_UINT32_T] = sizeof(uint32_t),
+ [TOKEN_TYPE_UINT64_T] = sizeof(uint64_t),
+ [TOKEN_TYPE_UINTPTR_T] = sizeof(uintptr_t),
+};
+
+/*
+ * Struct rte_bpf_arg augmented with a type of pointer to it, since rte_bpf_arg
+ * by itself may not contain enough information to determine its pointer type.
+ */
+struct arg_info {
+ struct rte_bpf_arg value;
+ enum rte_bpf_arg_type ptr_type;
+ bool is_array;
+};
+
+/* Return arg_info struct described by the specified type token. */
+static struct arg_info
+get_type_token_arg(enum token type)
+{
+ RTE_ASSERT(is_type_token(type));
+ switch (type) {
+ case TOKEN_TYPE_VOID:
+ return (struct arg_info){
+ .value = { .type = RTE_BPF_ARG_UNDEF },
+ .ptr_type = RTE_BPF_ARG_PTR,
+ };
+ case TOKEN_TYPE_RTE_MBUF:
+ return (struct arg_info){
+ .value = {
+ .type = RTE_BPF_ARG_RAW,
+ .size = sizeof(struct rte_mbuf),
+ .buf_size = RTE_MBUF_DEFAULT_BUF_SIZE,
+ },
+ .ptr_type = RTE_BPF_ARG_PTR_MBUF,
+ };
+ default:
+ /* Should only reach here for normal sized types. */
+ RTE_ASSERT((size_t)type < RTE_DIM(type_token_size) &&
+ type_token_size[type] != 0);
+ return (struct arg_info){
+ .value = {
+ .type = RTE_BPF_ARG_RAW,
+ .size = type_token_size[type],
+ },
+ .ptr_type = RTE_BPF_ARG_PTR,
+ };
+ }
+}
+
+
+/* PARSING TEXT INTO TOKENS */
+
+/* Return true if character matches [a-zA-Z0-9_] in regex */
+static bool
+iswordchar(char character)
+{
+ return isalnum(character) || character == '_';
+}
+
+/*
+ * Compare pointer to text with pointer to struct text_token to determine
+ * if text starts with the specified token.
+ */
+static int
+text_and_text_token_cmp(const void *text_void_ptr, const void *text_token_void_ptr)
+{
+ int result;
+ const char * const text = text_void_ptr;
+ const struct text_token * const text_token = text_token_void_ptr;
+
+ result = strncmp(text, text_token->text, text_token->length);
+
+ /* Checking the case of a partial word match. */
+ if (result == 0 && iswordchar(text[text_token->length - 1]) &&
+ iswordchar(text[text_token->length]))
+ /* Text word is longer than the token. */
+ result = 1;
+
+ return result;
+}
+
+/* Advance pointed character pointer to the first non-space character. */
+static void
+skip_space(const char **text_ptr)
+{
+ while (isspace(**text_ptr))
+ ++*text_ptr;
+}
+
+/*
+ * Recognize and return a token starting text.
+ * Return text_token_end if text is empty.
+ * Return text_token_unrecognized if text starts with unknown token.
+ */
+static struct text_token
+peek_text_token(const char *text)
+{
+ if (*text == '\0')
+ return text_token_end;
+ const struct text_token * const text_token = bsearch(text, text_tokens,
+ RTE_DIM(text_tokens), sizeof(text_tokens[0]),
+ text_and_text_token_cmp);
+ if (text_token == NULL)
+ return text_token_unrecognized;
+ return *text_token;
+}
+
+/*
+ * Advance pointed text starting with the specified token to the first
+ * non-space character after it.
+ * Do nothing if called for text_token_end or text_token_unrecognized.
+ */
+static void
+consume_text_token(const char **text_ptr, struct text_token text_token)
+{
+ RTE_ASSERT(memcmp(*text_ptr, text_token.text, text_token.length) == 0);
+ *text_ptr += text_token.length;
+ skip_space(text_ptr);
+}
+
+/* Return length of the word starting at `text`. */
+static size_t
+find_word_length(const char *text)
+{
+ size_t word_length = 0;
+ while (iswordchar(text[word_length]))
+ ++word_length;
+ return word_length;
+}
+
+/* Create a copy of the word starting text, advance to next token. */
+static const char *
+take_name(const char **text_ptr, struct alloc_list *alloc_list)
+{
+ const size_t word_length = find_word_length(*text_ptr);
+ if (word_length == 0)
+ /* Text does not start with a word. */
+ return NULL;
+
+ /* Allocate memory for the word and add it to the alloc_list */
+ char *word = malloc(word_length + 1);
+ RTE_VERIFY(word != NULL);
+ alloc_list_append(alloc_list, word);
+
+ /* Copy and terminate word contents. */
+ memcpy(word, *text_ptr, word_length);
+ word[word_length] = '\0';
+
+ /* Advance text pointer. */
+ *text_ptr += word_length;
+ skip_space(text_ptr);
+
+ return word;
+}
+
+/* Read a number starting text, advance to next token. */
+static int
+take_number(size_t *number, const char **text_ptr)
+{
+ /* Read a word and let strtoull to decide if it's a number. */
+ const size_t word_length = find_word_length(*text_ptr);
+ if (word_length == 0)
+ /* Text does not start with a word. */
+ return -ENOENT;
+
+ errno = 0;
+ char *number_end;
+ unsigned long long_number = strtoul(*text_ptr, &number_end, 0);
+ if (errno > 0)
+ return -errno;
+ if (number_end != *text_ptr + word_length)
+ /* Could not parse whole word. */
+ return -EINVAL;
+
+ *number = long_number;
+
+ /* Advance text pointer. */
+ *text_ptr += word_length;
+ skip_space(text_ptr);
+
+ return 0;
+}
+
+
+/* PARSING DECLARATION PARTS */
+
+/* Change arg into a reference. */
+static void
+change_into_reference(struct arg_info *arg)
+{
+ if (RTE_BPF_ARG_PTR_TYPE(arg->value.type) != 0) {
+ VALIDATE_BPF_LOG(WARNING,
+ "After taking reference to a pointer the latter "
+ "will be described as an opaque pointer-size blob.");
+ RTE_ASSERT(arg->ptr_type == RTE_BPF_ARG_PTR);
+ arg->value.size = POINTER_SIZE;
+ }
+ arg->value.type = arg->ptr_type;
+ arg->ptr_type = RTE_BPF_ARG_PTR;
+ arg->is_array = false;
+}
+
+/*
+ * Recognize and consume arg starting text, advance to next token.
+ */
+static int
+take_arg(struct arg_info *arg, const char **text_ptr, const char *text_start)
+{
+ struct text_token next = peek_text_token(*text_ptr);
+
+ if (next.token == TOKEN_STRUCT) {
+ consume_text_token(text_ptr, next);
+ next = peek_text_token(*text_ptr);
+ }
+
+ if (!is_type_token(next.token))
+ RETURN_TEXT_ERROR(*text_ptr, text_start, "expect type");
+ const enum token type = next.token;
+ consume_text_token(text_ptr, next);
+ next = peek_text_token(*text_ptr);
+
+ *arg = get_type_token_arg(type);
+
+ while (next.token == TOKEN_ASTERISK) {
+ consume_text_token(text_ptr, next);
+ next = peek_text_token(*text_ptr);
+ change_into_reference(arg);
+ }
+
+ while (next.token == TOKEN_BRACKET_OPEN) {
+ consume_text_token(text_ptr, next);
+
+ /* Initialize to zero to avoid spurious compiler warnings. */
+ size_t array_length = 0;
+ if (take_number(&array_length, text_ptr) < 0)
+ RETURN_TEXT_ERROR(*text_ptr, text_start, "expect length");
+ next = peek_text_token(*text_ptr);
+
+ if (arg->value.size != 0 &&
+ array_length > SIZE_MAX / arg->value.size)
+ RETURN_TEXT_ERROR(*text_ptr, text_start, "type too big");
+
+ if (next.token != TOKEN_BRACKET_CLOSE)
+ RETURN_TEXT_ERROR(*text_ptr, text_start, "expect ']'");
+ consume_text_token(text_ptr, next);
+ next = peek_text_token(*text_ptr);
+
+ change_into_reference(arg);
+ arg->value.size *= array_length;
+ arg->is_array = true;
+ }
+
+ return 0;
+}
+
+/* Fill struct rte_bpf_arg within xsym trying not to unzero the padding. */
+static void
+fill_xsym_arg(struct rte_bpf_arg *target, struct rte_bpf_arg source)
+{
+ /* Copy fields individually to try and prevent copying the padding. */
+ target->type = source.type;
+ target->size = source.size;
+ target->buf_size = source.buf_size;
+}
+
+/* Build and return struct rte_bpf_arg of type RTE_BPF_XTYPE_VAR */
+static int
+fill_var_xsym(struct rte_bpf_xsym *xsym, struct arg_info arg, const char *name,
+ struct alloc_list *alloc_list)
+{
+ /* Variables are passed by reference, except for arrays. */
+ if (!arg.is_array) {
+ if (RTE_BPF_ARG_PTR_TYPE(arg.value.type) != 0)
+ VALIDATE_BPF_LOG(WARNING,
+ "External pointers may not work as expected "
+ "because all external variables are passed by "
+ "reference but there is currently no way to "
+ "describe double pointer to the validator.");
+ change_into_reference(&arg);
+ }
+
+ /* Allocate something to assign to a val pointer. */
+ void * const val = calloc(1, RTE_MAX(1u, arg.value.size));
+ if (val == NULL) {
+ VALIDATE_BPF_LOG(ERR, "could not allocate enough memory");
+ return -1;
+ }
+ alloc_list_append(alloc_list, val);
+
+ /* Need all padding and unused fields to be zero-filled. */
+ memset(xsym, 0, sizeof(*xsym));
+ xsym->name = name;
+ xsym->type = RTE_BPF_XTYPE_VAR;
+ xsym->var.val = val;
+ fill_xsym_arg(&xsym->var.desc, arg.value);
+
+ return 0;
+}
+
+/* Build and return struct rte_bpf_arg of type RTE_BPF_XTYPE_FUNC */
+static void
+fill_func_xsym(struct rte_bpf_xsym *xsym, struct arg_info arg, const char *name)
+{
+ /* Need all padding and unused fields to be zero-filled. */
+ memset(xsym, 0, sizeof(*xsym));
+ xsym->name = name;
+ xsym->type = RTE_BPF_XTYPE_FUNC;
+ xsym->func.val = &dummy_function;
+ fill_xsym_arg(&xsym->func.ret, arg.value);
+}
+
+/*
+ * Parse and store function arguments, advance to next token after ')'.
+ * Value of *text_ptr should point to the next token after '('.
+ */
+static int
+take_func_xsym_args(struct rte_bpf_xsym *xsym, const char **text_ptr,
+ const char *text_start)
+{
+ struct arg_info arg;
+ struct text_token delimiter = peek_text_token(*text_ptr);
+
+ while (delimiter.token != TOKEN_PARENTHESIS_CLOSE) {
+ if (xsym->func.nb_args == EBPF_FUNC_MAX_ARGS)
+ RETURN_TEXT_ERROR(*text_ptr, text_start,
+ "too many arguments, maximum %d allowed",
+ EBPF_FUNC_MAX_ARGS);
+
+ if (take_arg(&arg, text_ptr, text_start) < 0)
+ return -1;
+ if (arg.value.type == RTE_BPF_ARG_UNDEF &&
+ xsym->func.nb_args != 0)
+ RETURN_TEXT_ERROR(*text_ptr, text_start,
+ "arguments of type void are not allowed");
+ fill_xsym_arg(&xsym->func.args[xsym->func.nb_args++],
+ arg.value);
+
+ delimiter = peek_text_token(*text_ptr);
+ switch (delimiter.token) {
+ case TOKEN_COMMA:
+ consume_text_token(text_ptr, delimiter);
+ continue;
+ case TOKEN_PARENTHESIS_CLOSE:
+ break;
+ default:
+ RETURN_TEXT_ERROR(*text_ptr, text_start,
+ "expect ')' or ','");
+ }
+ }
+ consume_text_token(text_ptr, delimiter);
+
+ /* Special case of single void argument. */
+ if (xsym->func.nb_args == 1 &&
+ xsym->func.args[0].type == RTE_BPF_ARG_UNDEF) {
+ xsym->func.nb_args = 0;
+ /* Need all padding and unused fields to be zero-filled. */
+ fill_xsym_arg(&xsym->func.args[0], (struct rte_bpf_arg){});
+ }
+
+ return 0;
+}
+
+/* Make sure text has ended. */
+static int
+ensure_end(const char *text, const char *text_start)
+{
+ if (peek_text_token(text).token != TOKEN_END)
+ RETURN_TEXT_ERROR(text, text_start, "trailing garbage");
+ return 0;
+}
+
+/* Parse and store xsym, advance to next token. */
+static int
+take_xsym(struct rte_bpf_xsym *xsym, const char **text_ptr,
+ const char *text_start, struct alloc_list *alloc_list)
+{
+ struct arg_info arg;
+
+ if (take_arg(&arg, text_ptr, text_start) < 0)
+ return -1;
+
+ const char * const name = take_name(text_ptr, alloc_list);
+ if (name == NULL)
+ RETURN_TEXT_ERROR(*text_ptr, text_start, "expect name");
+
+ const struct text_token next = peek_text_token(*text_ptr);
+ switch (next.token) {
+ case TOKEN_END:
+ if (fill_var_xsym(xsym, arg, name, alloc_list) < 0)
+ return -1;
+ break;
+ case TOKEN_PARENTHESIS_OPEN:
+ consume_text_token(text_ptr, next);
+ fill_func_xsym(xsym, arg, name);
+ if (take_func_xsym_args(xsym, text_ptr, text_start) < 0)
+ return -1;
+ break;
+ default:
+ RETURN_TEXT_ERROR(*text_ptr, text_start,
+ "expect '(' or text end");
+ }
+ return 0;
+}
+
+
+/* PUBLIC FUNCTIONS */
+
+void
+print_supported_types(void)
+{
+ printf("TYPE: [struct] BASIC_TYPE [*]... [[N]]...\n");
+ printf("BASIC_TYPE: one of\n");
+ for (int tti = 0; tti != RTE_DIM(text_tokens); ++tti) {
+ if (is_type_token(text_tokens[tti].token))
+ printf("\t%s\n", text_tokens[tti].text);
+ }
+}
+
+int
+parse_arg(struct rte_bpf_arg *arg, const char *text)
+{
+ struct arg_info arg_info;
+ const char * const text_start = text;
+ skip_space(&text);
+ if (take_arg(&arg_info, &text, text_start) < 0)
+ return -1;
+ if (ensure_end(text, text_start) < 0)
+ return -1;
+ if (arg_info.ptr_type != RTE_BPF_ARG_PTR)
+ VALIDATE_BPF_LOG(WARNING,
+ "`%s` has a special pointer type which was left unused; "
+ "argument was set to an opaque blob.",
+ text_start);
+ *arg = arg_info.value;
+ return 0;
+}
+
+int
+parse_xsym(struct rte_bpf_xsym *xsym, const char *text,
+ struct alloc_list *alloc_list)
+{
+ const char * const text_start = text;
+ skip_space(&text);
+ if (take_xsym(xsym, &text, text_start, alloc_list) < 0)
+ return -1;
+ if (ensure_end(text, text_start) < 0)
+ return -1;
+ return 0;
+}
+
+void
+adjust_arg_buf_size(struct rte_bpf_arg *arg, size_t mbuf_buf_size)
+{
+ if (arg->buf_size != 0)
+ /* Non-zero buf_size indicates mbuf or a pointer to it. */
+ arg->buf_size = mbuf_buf_size;
+}
+
+void
+adjust_xsym_buf_size(struct rte_bpf_xsym *xsym, size_t mbuf_buf_size)
+{
+ switch (xsym->type) {
+ case RTE_BPF_XTYPE_FUNC:
+ for (uint32_t argi = 0; argi != xsym->func.nb_args; ++argi)
+ adjust_arg_buf_size(&xsym->func.args[argi],
+ mbuf_buf_size);
+ adjust_arg_buf_size(&xsym->func.ret, mbuf_buf_size);
+ break;
+ case RTE_BPF_XTYPE_VAR:
+ adjust_arg_buf_size(&xsym->var.desc, mbuf_buf_size);
+ break;
+ default:
+ rte_panic("Unexpected xsym type %d\n", xsym->type);
+ }
+}
diff --git a/doc/guides/rel_notes/release_26_11.rst b/doc/guides/rel_notes/release_26_11.rst
index 4b3e5d995c..7bd4ae773f 100644
--- a/doc/guides/rel_notes/release_26_11.rst
+++ b/doc/guides/rel_notes/release_26_11.rst
@@ -55,6 +55,11 @@ New Features
Also, make sure to start the actual text at the margin.
=======================================================
+* **Added BPF validation application.**
+
+ Added ``dpdk-validate-bpf`` tool to pre-validate eBPF programs for
+ compatibility with the ``lib/bpf`` execution context.
+
Removed Items
-------------
diff --git a/doc/guides/tools/index.rst b/doc/guides/tools/index.rst
index 13f75a5bc6..1c43060e15 100644
--- a/doc/guides/tools/index.rst
+++ b/doc/guides/tools/index.rst
@@ -26,3 +26,4 @@ DPDK Tools User Guides
testmldev
graph
dts
+ validate_bpf
diff --git a/doc/guides/tools/validate_bpf.rst b/doc/guides/tools/validate_bpf.rst
new file mode 100644
index 0000000000..9e2d58cfbe
--- /dev/null
+++ b/doc/guides/tools/validate_bpf.rst
@@ -0,0 +1,97 @@
+.. SPDX-License-Identifier: BSD-3-Clause
+ Copyright(c) 2026 The DPDK contributors
+
+dpdk-validate-bpf Application
+=============================
+
+The ``dpdk-validate-bpf`` tool is an application that allows evaluating BPF
+programs against the ``lib/bpf`` library. It can be used to pre-validate BPF
+programs before loading them in a real application to ensure that they pass the
+internal verifier. This makes it a useful tool for integrating into build
+systems and continuous integration (CI) pipelines to verify the correctness
+of BPF programs during compilation, as well as for debugging validation issues.
+
+Running the Application
+-----------------------
+
+The tool requires a path to a compiled BPF object file. It also provides options
+to configure the execution environment and external symbol definitions.
+
+For a comprehensive list of options and their meanings, refer to the
+built-in help by running ``dpdk-validate-bpf --help``.
+
+The most important options for a minimal example are:
+
+* ``--prog-arg <type>``
+
+ Define program arguments, if different from ``struct rte_mbuf *``.
+ This argument may be repeated up to 5 times.
+
+* ``--xsym '<type> <name> | <type> <name>(<type>, ...)'``
+
+ Define an external symbol (variable or function) that the BPF program uses.
+ This allows the validator to recognize external calls and memory accesses.
+ Multiple external symbols can be defined by repeating this option.
+
+* ``--section <name>``
+
+ Specify the ELF section name in the BPF object file to load and validate.
+ The default section name is ``.text``, which is typically fine in cases
+ where the object file contains only one program.
+
+Interactive Debugging Mode
+--------------------------
+
+The interactive mode (enabled with the ``--debug`` flag) allows the user to
+debug the BPF validation process itself. This is particularly useful when
+BPF program fails verification, allowing you to trace the state changes per
+instruction and understand the validator's decisions.
+
+Design
+~~~~~~
+
+The interactive debugger provides a command-line interface similar to GDB.
+It allows setting breakpoints at specific instructions, or catchpoints on
+validator events (such as ``branch-enter`` or ``branch-prune``).
+Users can step through the validation process,
+inspect the state of registers, and query if certain conditional jumps may
+be taken based on the validator's knowledge of the program state.
+
+Usage Example
+~~~~~~~~~~~~~
+
+Launch the tool in debug mode with a compiled BPF object file:
+
+.. code-block:: console
+
+ $ dpdk-validate-bpf bpf_prog.o --prog-arg={'void *',uint64_t} --debug
+ => 0: b7 00 00 00 01 00 00 00 mov r0, #0x1
+ (validate) list 3
+ => 0: b7 00 00 00 01 00 00 00 mov r0, #0x1
+ 1: 25 02 01 00 0e 00 00 00 jgt r2, #0xe, L3
+ 2: b7 00 00 00 00 00 00 00 mov r0, #0x0
+ (validate) break 2
+ Breakpoint 0 at 2.
+ (validate) catch branch-enter
+ Catchpoint 1 on branch-enter.
+ (validate) continue
+ Catchpoint 1 on branch-enter.
+ Entered new branch at pc 1.
+ => 3: 95 00 00 00 00 00 00 00 exit
+ (validate) where
+ 1: 25 02 01 00 0e 00 00 00 jgt r2, #0xe, L3 ; taken
+ => 3: 95 00 00 00 00 00 00 00 exit
+ (validate) info r1
+ r1: %buffer<0> + 0
+ (validate) info r2
+ r2: 0xf..UINT64_MAX
+ (validate) may r2 <= 14
+ NO
+ (validate) continue
+ Catchpoint 1 on branch-enter.
+ Entered new branch at pc 1.
+ Breakpoint 0 at 2.
+ => 2: b7 00 00 00 00 00 00 00 mov r0, #0x0
+ (validate) continue
+ Validation succeeded.
+ (validate) quit
--
2.43.0
^ permalink raw reply related [flat|nested] 38+ messages in thread
* RE: [PATCH v4 0/7] validate-bpf: add standalone validation debug tool
2026-09-17 14:35 ` [PATCH v4 0/7] validate-bpf: add standalone validation debug tool Marat Khalili
` (6 preceding siblings ...)
2026-09-17 14:35 ` [PATCH v4 7/7] app/validate-bpf: add BPF validation application Marat Khalili
@ 2026-09-17 16:16 ` Konstantin Ananyev
7 siblings, 0 replies; 38+ messages in thread
From: Konstantin Ananyev @ 2026-09-17 16:16 UTC (permalink / raw)
To: Marat Khalili; +Cc: dev@dpdk.org, Marat Khalili, Stephen Hemminger
> This patchset introduces a new standalone tool for pre-validating eBPF
> programs and debugging validation problems. Its should allow developers
> to trace validator state changes per instruction and understand verifier
> decisions, simplifying the debugging of rejected programs before loading
> them into a real application context.
>
> v4:
> * addressed reviewer feedback from Konstantin Ananyev:
> * changed variable and array names to lowercase to follow conventions;
> * replaced strtoull()/strtoumax() with strtoul() and removed redundant
> SIZE_MAX checks;
> * changed return types of point_infos_destroy_* functions to void;
> * replaced %jd and (intmax_t) cast with PRId32 when printing frame
> offsets;
> * refactored the inverted loop in print_frame() for clarity;
> * replaced return -ENOMEM with RTE_VERIFY() for allocations in debug.c
> print functions;
> * simplified token matching logic in parse_decl.c by using strncmp().
>
> v3:
> * addressed another FreeBSD issue (spurious uninitialized warning);
> * addressed more AI comments:
> * fail if library fails to set step catchpoint on start command;
> * fail nicer if external variables are too big;
> * fail nicer if EAL initialization or cleanup fails;
> * explain why 9 steps are expeced in the test;
> * explain that non-zero buf_size marks mbuf;
> * avoid symbols starting with `RTE_` in the app;
>
> v2:
> * for FreeBSD compatibility switched from Linux to DPDK network structs;
> * addressed small AI comments:
> * clarified patch 1 commit message;
> * renamed `rte_validate_bpf_logtype` to `validate_bpf_logtype`;
> * corrected comment to an internal-use-only function to reflect that
> next-after-last value of the program counter is no longer allowed;
>
> Marat Khalili (7):
> bpf/validate: fix finished status on restart
> bpf/validate: refactor internal step function
> bpf/validate: formalize call back requirements
> bpf/validate: add jump notification events
> bpf/validate: add get current event API
> app/test: add test for bpf validate debug events
> app/validate-bpf: add BPF validation application
>
> MAINTAINERS | 2 +
> app/meson.build | 1 +
> app/test/test_bpf_validate.c | 344 +++++++-
> app/validate-bpf/alloc_list.c | 53 ++
> app/validate-bpf/args.c | 262 ++++++
> app/validate-bpf/debug.c | 1087 ++++++++++++++++++++++++
> app/validate-bpf/debug_command.c | 383 +++++++++
> app/validate-bpf/debug_command.h | 65 ++
> app/validate-bpf/eal_init_args.c | 57 ++
> app/validate-bpf/internal.h | 151 ++++
> app/validate-bpf/main.c | 90 ++
> app/validate-bpf/meson.build | 13 +
> app/validate-bpf/parse_decl.c | 616 ++++++++++++++
> doc/guides/rel_notes/release_26_11.rst | 5 +
> doc/guides/tools/index.rst | 1 +
> doc/guides/tools/validate_bpf.rst | 97 +++
> lib/bpf/bpf_validate.c | 49 +-
> lib/bpf/bpf_validate_debug.c | 92 +-
> lib/bpf/bpf_validate_debug.h | 12 +-
> lib/bpf/rte_bpf_validate_debug.h | 42 +-
> 20 files changed, 3356 insertions(+), 66 deletions(-)
> create mode 100644 app/validate-bpf/alloc_list.c
> create mode 100644 app/validate-bpf/args.c
> create mode 100644 app/validate-bpf/debug.c
> create mode 100644 app/validate-bpf/debug_command.c
> create mode 100644 app/validate-bpf/debug_command.h
> create mode 100644 app/validate-bpf/eal_init_args.c
> create mode 100644 app/validate-bpf/internal.h
> create mode 100644 app/validate-bpf/main.c
> create mode 100644 app/validate-bpf/meson.build
> create mode 100644 app/validate-bpf/parse_decl.c
> create mode 100644 doc/guides/tools/validate_bpf.rst
>
> --
Series-Acked-by: Konstantin Ananyev <konstantin.ananyev@huawei.com>
> 2.43.0
^ permalink raw reply [flat|nested] 38+ messages in thread
end of thread, other threads:[~2026-09-17 16:16 UTC | newest]
Thread overview: 38+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-09-11 10:39 [PATCH 0/7] validate-bpf: add standalone validation debug tool Marat Khalili
2026-09-11 10:39 ` [PATCH 1/7] bpf/validate: fix finished status on restart Marat Khalili
2026-09-11 10:40 ` [PATCH 2/7] bpf/validate: refactor internal step function Marat Khalili
2026-09-11 10:40 ` [PATCH 3/7] bpf/validate: formalize call back requirements Marat Khalili
2026-09-11 10:40 ` [PATCH 4/7] bpf/validate: add jump notification events Marat Khalili
2026-09-11 10:40 ` [PATCH 5/7] bpf/validate: add get current event API Marat Khalili
2026-09-11 10:40 ` [PATCH 6/7] app/test: add test for bpf validate debug events Marat Khalili
2026-09-11 10:40 ` [PATCH 7/7] app/validate-bpf: add BPF validation application Marat Khalili
2026-09-14 14:54 ` [PATCH v2 0/7] validate-bpf: add standalone validation debug tool Marat Khalili
2026-09-14 14:54 ` [PATCH v2 1/7] bpf/validate: fix finished status on restart Marat Khalili
2026-09-14 14:54 ` [PATCH v2 2/7] bpf/validate: refactor internal step function Marat Khalili
2026-09-14 14:54 ` [PATCH v2 3/7] bpf/validate: formalize call back requirements Marat Khalili
2026-09-14 14:54 ` [PATCH v2 4/7] bpf/validate: add jump notification events Marat Khalili
2026-09-14 14:54 ` [PATCH v2 5/7] bpf/validate: add get current event API Marat Khalili
2026-09-14 14:54 ` [PATCH v2 6/7] app/test: add test for bpf validate debug events Marat Khalili
2026-09-14 14:54 ` [PATCH v2 7/7] app/validate-bpf: add BPF validation application Marat Khalili
2026-09-14 18:41 ` Stephen Hemminger
2026-09-15 13:13 ` Stephen Hemminger
2026-09-14 18:42 ` [PATCH v2 0/7] validate-bpf: add standalone validation debug tool Stephen Hemminger
2026-09-15 19:27 ` Marat Khalili
2026-09-15 19:26 ` [PATCH v3 " Marat Khalili
2026-09-15 19:26 ` [PATCH v3 1/7] bpf/validate: fix finished status on restart Marat Khalili
2026-09-15 19:26 ` [PATCH v3 2/7] bpf/validate: refactor internal step function Marat Khalili
2026-09-15 19:26 ` [PATCH v3 3/7] bpf/validate: formalize call back requirements Marat Khalili
2026-09-15 19:26 ` [PATCH v3 4/7] bpf/validate: add jump notification events Marat Khalili
2026-09-15 19:26 ` [PATCH v3 5/7] bpf/validate: add get current event API Marat Khalili
2026-09-15 19:26 ` [PATCH v3 6/7] app/test: add test for bpf validate debug events Marat Khalili
2026-09-15 19:26 ` [PATCH v3 7/7] app/validate-bpf: add BPF validation application Marat Khalili
2026-09-16 16:36 ` Konstantin Ananyev
2026-09-17 14:35 ` [PATCH v4 0/7] validate-bpf: add standalone validation debug tool Marat Khalili
2026-09-17 14:35 ` [PATCH v4 1/7] bpf/validate: fix finished status on restart Marat Khalili
2026-09-17 14:35 ` [PATCH v4 2/7] bpf/validate: refactor internal step function Marat Khalili
2026-09-17 14:35 ` [PATCH v4 3/7] bpf/validate: formalize call back requirements Marat Khalili
2026-09-17 14:35 ` [PATCH v4 4/7] bpf/validate: add jump notification events Marat Khalili
2026-09-17 14:35 ` [PATCH v4 5/7] bpf/validate: add get current event API Marat Khalili
2026-09-17 14:35 ` [PATCH v4 6/7] app/test: add test for bpf validate debug events Marat Khalili
2026-09-17 14:35 ` [PATCH v4 7/7] app/validate-bpf: add BPF validation application Marat Khalili
2026-09-17 16:16 ` [PATCH v4 0/7] validate-bpf: add standalone validation debug tool Konstantin Ananyev
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox