* [PATCH 1/5] tools/sched_ext: Don't restart over a pending exit request
2026-07-24 18:21 [PATCHSET sched_ext/for-7.3] sched_ext: Follow-up fixes and missing cap enforcement Tejun Heo
@ 2026-07-24 18:21 ` Tejun Heo
2026-07-24 18:21 ` [PATCH 2/5] sched_ext: Gate local DSQ reenq on baseline cid access Tejun Heo
` (3 subsequent siblings)
4 siblings, 0 replies; 11+ messages in thread
From: Tejun Heo @ 2026-07-24 18:21 UTC (permalink / raw)
To: David Vernet, Andrea Righi, Changwoo Min
Cc: sched-ext, Emil Tsalapatis, linux-kernel, Tejun Heo
The tools restart when the kernel exits the scheduler with
SCX_ECODE_ACT_RESTART. The restart decision doesn't consult exit_req, so an
exit request arriving while the restart condition persists is ignored and
the tool reloads in a tight loop. Test exit_req before restarting.
scx_userland needs more: its main loop never watches the kernel-side exit
and exit_req doubles as the stats printer's stop signal, set by the teardown
and reset on each restart. Add the missing UEI_EXITED() test and give the
printer its own stop flag so that exit_req only means an exit request and
stays latched like in the other tools.
Signed-off-by: Tejun Heo <tj@kernel.org>
---
tools/sched_ext/scx_central.c | 2 +-
tools/sched_ext/scx_cpu0.c | 2 +-
tools/sched_ext/scx_flatcg.c | 2 +-
tools/sched_ext/scx_pair.c | 2 +-
tools/sched_ext/scx_qmap.c | 2 +-
tools/sched_ext/scx_sdt.c | 2 +-
tools/sched_ext/scx_simple.c | 2 +-
tools/sched_ext/scx_userland.c | 11 ++++++-----
8 files changed, 13 insertions(+), 12 deletions(-)
diff --git a/tools/sched_ext/scx_central.c b/tools/sched_ext/scx_central.c
index 4a72df39500d..e1acaed4ec31 100644
--- a/tools/sched_ext/scx_central.c
+++ b/tools/sched_ext/scx_central.c
@@ -120,7 +120,7 @@ int main(int argc, char **argv)
ecode = UEI_REPORT(skel, uei);
scx_central__destroy(skel);
- if (UEI_ECODE_RESTART(ecode))
+ if (!exit_req && UEI_ECODE_RESTART(ecode))
goto restart;
return 0;
}
diff --git a/tools/sched_ext/scx_cpu0.c b/tools/sched_ext/scx_cpu0.c
index 84a47aee2f95..1c2b507b8b34 100644
--- a/tools/sched_ext/scx_cpu0.c
+++ b/tools/sched_ext/scx_cpu0.c
@@ -99,7 +99,7 @@ int main(int argc, char **argv)
ecode = UEI_REPORT(skel, uei);
scx_cpu0__destroy(skel);
- if (UEI_ECODE_RESTART(ecode))
+ if (!exit_req && UEI_ECODE_RESTART(ecode))
goto restart;
return 0;
}
diff --git a/tools/sched_ext/scx_flatcg.c b/tools/sched_ext/scx_flatcg.c
index 7799782b76d1..a223bff3746a 100644
--- a/tools/sched_ext/scx_flatcg.c
+++ b/tools/sched_ext/scx_flatcg.c
@@ -233,7 +233,7 @@ int main(int argc, char **argv)
ecode = UEI_REPORT(skel, uei);
scx_flatcg__destroy(skel);
- if (UEI_ECODE_RESTART(ecode))
+ if (!exit_req && UEI_ECODE_RESTART(ecode))
goto restart;
return 0;
}
diff --git a/tools/sched_ext/scx_pair.c b/tools/sched_ext/scx_pair.c
index 41b136d43a55..00f595b58f97 100644
--- a/tools/sched_ext/scx_pair.c
+++ b/tools/sched_ext/scx_pair.c
@@ -190,7 +190,7 @@ int main(int argc, char **argv)
ecode = UEI_REPORT(skel, uei);
scx_pair__destroy(skel);
- if (UEI_ECODE_RESTART(ecode))
+ if (!exit_req && UEI_ECODE_RESTART(ecode))
goto restart;
return 0;
}
diff --git a/tools/sched_ext/scx_qmap.c b/tools/sched_ext/scx_qmap.c
index 27ffda1c519e..3f54796e48be 100644
--- a/tools/sched_ext/scx_qmap.c
+++ b/tools/sched_ext/scx_qmap.c
@@ -465,7 +465,7 @@ int main(int argc, char **argv)
ecode = UEI_REPORT(skel, uei);
scx_qmap__destroy(skel);
- if (UEI_ECODE_RESTART(ecode))
+ if (!exit_req && UEI_ECODE_RESTART(ecode))
goto restart;
return 0;
}
diff --git a/tools/sched_ext/scx_sdt.c b/tools/sched_ext/scx_sdt.c
index ef197b266a36..2f93a00de548 100644
--- a/tools/sched_ext/scx_sdt.c
+++ b/tools/sched_ext/scx_sdt.c
@@ -96,7 +96,7 @@ int main(int argc, char **argv)
ecode = UEI_REPORT(skel, uei);
scx_sdt__destroy(skel);
- if (UEI_ECODE_RESTART(ecode))
+ if (!exit_req && UEI_ECODE_RESTART(ecode))
goto restart;
return 0;
}
diff --git a/tools/sched_ext/scx_simple.c b/tools/sched_ext/scx_simple.c
index 34f9785335b7..b7589a83f28a 100644
--- a/tools/sched_ext/scx_simple.c
+++ b/tools/sched_ext/scx_simple.c
@@ -104,7 +104,7 @@ int main(int argc, char **argv)
ecode = UEI_REPORT(skel, uei);
scx_simple__destroy(skel);
- if (UEI_ECODE_RESTART(ecode))
+ if (!exit_req && UEI_ECODE_RESTART(ecode))
goto restart;
return 0;
}
diff --git a/tools/sched_ext/scx_userland.c b/tools/sched_ext/scx_userland.c
index 192b79c7e4aa..b16b6db5f365 100644
--- a/tools/sched_ext/scx_userland.c
+++ b/tools/sched_ext/scx_userland.c
@@ -52,6 +52,7 @@ static __u32 batch_size = 8;
static bool verbose;
static volatile int exit_req;
+static volatile int stats_stop;
static int enqueued_fd, dispatched_fd;
static pthread_t stats_printer;
@@ -286,7 +287,7 @@ static void dispatch_batch(void)
static void *run_stats_printer(void *arg)
{
- while (!exit_req) {
+ while (!stats_stop) {
__u64 nr_failed_enqueues, nr_kernel_enqueues, nr_user_enqueues, total;
nr_failed_enqueues = skel->bss->nr_failed_enqueues;
@@ -374,7 +375,7 @@ static void pre_bootstrap(int argc, char **argv)
static void bootstrap(char *comm)
{
- exit_req = 0;
+ stats_stop = 0;
min_vruntime = 0.0;
__atomic_store_n(&nr_vruntime_enqueues, 0, __ATOMIC_RELAXED);
__atomic_store_n(&nr_vruntime_dispatches, 0, __ATOMIC_RELAXED);
@@ -404,7 +405,7 @@ static void bootstrap(char *comm)
static void sched_main_loop(void)
{
- while (!exit_req) {
+ while (!exit_req && !UEI_EXITED(skel, uei)) {
/*
* Perform the following work in the main user space scheduler
* loop:
@@ -434,13 +435,13 @@ int main(int argc, char **argv)
bootstrap(argv[0]);
sched_main_loop();
- exit_req = 1;
+ stats_stop = 1;
bpf_link__destroy(ops_link);
pthread_join(stats_printer, NULL);
ecode = UEI_REPORT(skel, uei);
scx_userland__destroy(skel);
- if (UEI_ECODE_RESTART(ecode))
+ if (!exit_req && UEI_ECODE_RESTART(ecode))
goto restart;
return 0;
}
--
2.55.0
^ permalink raw reply related [flat|nested] 11+ messages in thread* [PATCH 2/5] sched_ext: Gate local DSQ reenq on baseline cid access
2026-07-24 18:21 [PATCHSET sched_ext/for-7.3] sched_ext: Follow-up fixes and missing cap enforcement Tejun Heo
2026-07-24 18:21 ` [PATCH 1/5] tools/sched_ext: Don't restart over a pending exit request Tejun Heo
@ 2026-07-24 18:21 ` Tejun Heo
2026-07-24 18:21 ` [PATCH 3/5] sched_ext: Count kicks denied for lacking " Tejun Heo
` (2 subsequent siblings)
4 siblings, 0 replies; 11+ messages in thread
From: Tejun Heo @ 2026-07-24 18:21 UTC (permalink / raw)
To: David Vernet, Andrea Righi, Changwoo Min
Cc: sched-ext, Emil Tsalapatis, linux-kernel, Tejun Heo
scx_bpf_dsq_reenq() with an SCX_DSQ_LOCAL_ON target schedules deferred reenq
work on the cid's cpu, raising an IPI when the target rq isn't the locked
one. Nothing checks caps along the way, so a sub-sched holding no cap at all
on a cid can force its cpu to take IPIs and rq lock cycles at will. The
analogous scx_bpf_kick_cid() path gates delivery on SCX_CAP_BASE in
kick_one_cpu() to prevent exactly this.
Apply the same rule at the reenq scheduling point: if the calling sched
lacks SCX_CAP_BASE on the target cid, drop the reenq and count it in the new
SCX_EV_SUB_REENQ_DENIED event. The check is lockless, which is fine: a reenq
slipping through right after a revoke is harmless, and a wrong denial can't
happen - if the caller has seen its ownership of the cpu, the check sees it
too.
Signed-off-by: Tejun Heo <tj@kernel.org>
---
kernel/sched/ext/ext.c | 12 ++++++++++++
kernel/sched/ext/internal.h | 9 ++++++++-
2 files changed, 20 insertions(+), 1 deletion(-)
diff --git a/kernel/sched/ext/ext.c b/kernel/sched/ext/ext.c
index 81c2d8eeae41..fd88b4d4f12a 100644
--- a/kernel/sched/ext/ext.c
+++ b/kernel/sched/ext/ext.c
@@ -1071,6 +1071,18 @@ void schedule_dsq_reenq(struct scx_sched *sch, struct scx_dispatch_q *dsq,
if (dsq->id == SCX_DSQ_LOCAL) {
rq = container_of(dsq, struct rq, scx.local_dsq);
+ /*
+ * A sub-sched lacking baseline access on the target cid has no
+ * business triggering IPIs. The lockless test is fine: slipping
+ * through right after a revoke is harmless and a wrong denial
+ * can't happen - if the caller has seen its ownership, so does
+ * this test.
+ */
+ if (unlikely(scx_missing_caps(sch, cpu_of(rq), SCX_CAP_BASE))) {
+ __scx_add_event(sch, SCX_EV_SUB_REENQ_DENIED, 1);
+ return;
+ }
+
struct scx_sched_pcpu *sch_pcpu = per_cpu_ptr(sch->pcpu, cpu_of(rq));
struct scx_deferred_reenq_local *drl = &sch_pcpu->deferred_reenq_local;
diff --git a/kernel/sched/ext/internal.h b/kernel/sched/ext/internal.h
index a9a853a71061..1bdf34f3fc3c 100644
--- a/kernel/sched/ext/internal.h
+++ b/kernel/sched/ext/internal.h
@@ -1193,6 +1193,12 @@ struct scx_event_stats {
* kick degrades to a plain reschedule.
*/
s64 SCX_EV_SUB_PREEMPT_DENIED;
+
+ /*
+ * The number of times a local DSQ reenq was dropped because the
+ * sub-sched lacked baseline access on the target cid.
+ */
+ s64 SCX_EV_SUB_REENQ_DENIED;
};
#define SCX_EVENTS_LIST(SCX_EVENT) \
@@ -1212,7 +1218,8 @@ struct scx_event_stats {
SCX_EVENT(SCX_EV_INSERT_NOT_OWNED); \
SCX_EVENT(SCX_EV_SUB_BYPASS_DISPATCH); \
SCX_EVENT(SCX_EV_SUB_FORCED_ADMIT); \
- SCX_EVENT(SCX_EV_SUB_PREEMPT_DENIED)
+ SCX_EVENT(SCX_EV_SUB_PREEMPT_DENIED); \
+ SCX_EVENT(SCX_EV_SUB_REENQ_DENIED)
struct scx_sched;
--
2.55.0
^ permalink raw reply related [flat|nested] 11+ messages in thread* [PATCH 3/5] sched_ext: Count kicks denied for lacking baseline cid access
2026-07-24 18:21 [PATCHSET sched_ext/for-7.3] sched_ext: Follow-up fixes and missing cap enforcement Tejun Heo
2026-07-24 18:21 ` [PATCH 1/5] tools/sched_ext: Don't restart over a pending exit request Tejun Heo
2026-07-24 18:21 ` [PATCH 2/5] sched_ext: Gate local DSQ reenq on baseline cid access Tejun Heo
@ 2026-07-24 18:21 ` Tejun Heo
2026-07-24 18:21 ` [PATCH 4/5] sched_ext: Factor out scx_cpuperf_set() Tejun Heo
2026-07-24 18:21 ` [PATCH 5/5] sched_ext: Gate scx_bpf_cidperf_set() behind a new SCX_CAP_PERF Tejun Heo
4 siblings, 0 replies; 11+ messages in thread
From: Tejun Heo @ 2026-07-24 18:21 UTC (permalink / raw)
To: David Vernet, Andrea Righi, Changwoo Min
Cc: sched-ext, Emil Tsalapatis, linux-kernel, Tejun Heo
kick_one_cpu() silently skips a kick when the kicking sub-sched lacks
SCX_CAP_BASE on the target cid, as does kick_one_cpu_if_idle() for idle
kicks. The skips are sound with the same logic as the reenq gate but are
invisible today, unlike the preempt degradation counted in
SCX_EV_SUB_PREEMPT_DENIED. Count them in a new SCX_EV_SUB_KICK_DENIED event
so every cap denial is observable.
Signed-off-by: Tejun Heo <tj@kernel.org>
---
kernel/sched/ext/ext.c | 20 ++++++++++++++------
kernel/sched/ext/internal.h | 8 ++++++++
2 files changed, 22 insertions(+), 6 deletions(-)
diff --git a/kernel/sched/ext/ext.c b/kernel/sched/ext/ext.c
index fd88b4d4f12a..5d02bb99c09d 100644
--- a/kernel/sched/ext/ext.c
+++ b/kernel/sched/ext/ext.c
@@ -8118,6 +8118,7 @@ static bool kick_one_cpu(s32 cpu, struct scx_sched_pcpu *pcpu, struct rq *this_r
struct scx_rq *this_scx = &this_rq->scx;
const struct sched_class *cur_class;
bool should_wait = false;
+ bool kickable;
unsigned long flags;
raw_spin_rq_lock_irqsave(rq, flags);
@@ -8131,9 +8132,10 @@ static bool kick_one_cpu(s32 cpu, struct scx_sched_pcpu *pcpu, struct rq *this_r
* business forcing a reschedule there - skip. This is the authoritative
* cap check: ecaps is read here under @rq's lock.
*/
- if ((cpu_online(cpu) || cpu == cpu_of(this_rq)) &&
- !sched_class_above(cur_class, &ext_sched_class) &&
- !scx_missing_caps(pcpu->sch, cpu, SCX_CAP_BASE)) {
+ kickable = (cpu_online(cpu) || cpu == cpu_of(this_rq)) &&
+ !sched_class_above(cur_class, &ext_sched_class);
+
+ if (kickable && !scx_missing_caps(pcpu->sch, cpu, SCX_CAP_BASE)) {
if (cpumask_test_cpu(cpu, pcpu->cpus_to_preempt)) {
if (cur_class == &ext_sched_class) {
if (likely(!scx_missing_caps(pcpu->sch, cpu,
@@ -8157,6 +8159,9 @@ static bool kick_one_cpu(s32 cpu, struct scx_sched_pcpu *pcpu, struct rq *this_r
resched_curr(rq);
} else {
+ /* a kickable cpu was skipped solely for the missing caps */
+ if (kickable)
+ __scx_add_event(pcpu->sch, SCX_EV_SUB_KICK_DENIED, 1);
cpumask_clear_cpu(cpu, pcpu->cpus_to_preempt);
cpumask_clear_cpu(cpu, pcpu->cpus_to_wait);
}
@@ -8176,9 +8181,12 @@ static void kick_one_cpu_if_idle(s32 cpu, struct scx_sched_pcpu *pcpu,
/* idle kicks need baseline access too, see kick_one_cpu() */
if (!can_skip_idle_kick(rq) &&
- (cpu_online(cpu) || cpu == cpu_of(this_rq)) &&
- !scx_missing_caps(pcpu->sch, cpu, SCX_CAP_BASE))
- resched_curr(rq);
+ (cpu_online(cpu) || cpu == cpu_of(this_rq))) {
+ if (likely(!scx_missing_caps(pcpu->sch, cpu, SCX_CAP_BASE)))
+ resched_curr(rq);
+ else
+ __scx_add_event(pcpu->sch, SCX_EV_SUB_KICK_DENIED, 1);
+ }
raw_spin_rq_unlock_irqrestore(rq, flags);
}
diff --git a/kernel/sched/ext/internal.h b/kernel/sched/ext/internal.h
index 1bdf34f3fc3c..308d16320818 100644
--- a/kernel/sched/ext/internal.h
+++ b/kernel/sched/ext/internal.h
@@ -1194,6 +1194,13 @@ struct scx_event_stats {
*/
s64 SCX_EV_SUB_PREEMPT_DENIED;
+ /*
+ * The number of times a kick was skipped because the sub-sched lacked
+ * baseline access on the target cid. The preempt-part degradation of a
+ * delivered kick is counted in SCX_EV_SUB_PREEMPT_DENIED instead.
+ */
+ s64 SCX_EV_SUB_KICK_DENIED;
+
/*
* The number of times a local DSQ reenq was dropped because the
* sub-sched lacked baseline access on the target cid.
@@ -1219,6 +1226,7 @@ struct scx_event_stats {
SCX_EVENT(SCX_EV_SUB_BYPASS_DISPATCH); \
SCX_EVENT(SCX_EV_SUB_FORCED_ADMIT); \
SCX_EVENT(SCX_EV_SUB_PREEMPT_DENIED); \
+ SCX_EVENT(SCX_EV_SUB_KICK_DENIED); \
SCX_EVENT(SCX_EV_SUB_REENQ_DENIED)
struct scx_sched;
--
2.55.0
^ permalink raw reply related [flat|nested] 11+ messages in thread* [PATCH 4/5] sched_ext: Factor out scx_cpuperf_set()
2026-07-24 18:21 [PATCHSET sched_ext/for-7.3] sched_ext: Follow-up fixes and missing cap enforcement Tejun Heo
` (2 preceding siblings ...)
2026-07-24 18:21 ` [PATCH 3/5] sched_ext: Count kicks denied for lacking " Tejun Heo
@ 2026-07-24 18:21 ` Tejun Heo
2026-07-24 18:30 ` sashiko-bot
2026-07-24 18:21 ` [PATCH 5/5] sched_ext: Gate scx_bpf_cidperf_set() behind a new SCX_CAP_PERF Tejun Heo
4 siblings, 1 reply; 11+ messages in thread
From: Tejun Heo @ 2026-07-24 18:21 UTC (permalink / raw)
To: David Vernet, Andrea Righi, Changwoo Min
Cc: sched-ext, Emil Tsalapatis, linux-kernel, Tejun Heo
Factor the cpuperf target write out of scx_bpf_cpuperf_set() into
scx_cpuperf_set() which takes the acting sched and returns 0 or -errno, and
flatten the nested validation into early returns. No functional change.
Prep for gating the write behind a cap and reporting the outcome from the
cid-form kfunc.
Signed-off-by: Tejun Heo <tj@kernel.org>
---
kernel/sched/ext/ext.c | 77 ++++++++++++++++++++++++------------------
1 file changed, 44 insertions(+), 33 deletions(-)
diff --git a/kernel/sched/ext/ext.c b/kernel/sched/ext/ext.c
index 5d02bb99c09d..366245daab68 100644
--- a/kernel/sched/ext/ext.c
+++ b/kernel/sched/ext/ext.c
@@ -9796,49 +9796,60 @@ __bpf_kfunc u32 scx_bpf_cidperf_cur(s32 cid, const struct bpf_prog_aux *aux)
* use. Consult hardware and cpufreq documentation for more information. The
* current performance level can be monitored using scx_bpf_cpuperf_cur().
*/
-__bpf_kfunc void scx_bpf_cpuperf_set(s32 cpu, u32 perf, const struct bpf_prog_aux *aux)
+static s32 scx_cpuperf_set(struct scx_sched *sch, s32 cpu, u32 perf)
{
- struct scx_sched *sch;
-
- guard(rcu)();
-
- sch = scx_prog_sched(aux);
- if (unlikely(!sch))
- return;
+ struct rq *rq, *locked_rq;
+ struct rq_flags rf;
if (unlikely(perf > SCX_CPUPERF_ONE)) {
scx_error(sch, "Invalid cpuperf target %u for CPU %d", perf, cpu);
- return;
+ return -EINVAL;
}
- if (scx_cpu_valid(sch, cpu, NULL)) {
- struct rq *rq = cpu_rq(cpu), *locked_rq = scx_locked_rq();
- struct rq_flags rf;
-
- /*
- * When called with an rq lock held, restrict the operation
- * to the corresponding CPU to prevent ABBA deadlocks.
- */
- if (locked_rq && rq != locked_rq) {
- scx_error(sch, "Invalid target CPU %d", cpu);
- return;
- }
+ if (!scx_cpu_valid(sch, cpu, NULL))
+ return -EINVAL;
- /*
- * If no rq lock is held, allow to operate on any CPU by
- * acquiring the corresponding rq lock.
- */
- if (!locked_rq) {
- rq_lock_irqsave(rq, &rf);
- update_rq_clock(rq);
- }
+ rq = cpu_rq(cpu);
+ locked_rq = scx_locked_rq();
- rq->scx.cpuperf_target = perf;
- cpufreq_update_util(rq, 0);
+ /*
+ * When called with an rq lock held, restrict the operation to the
+ * corresponding CPU to prevent ABBA deadlocks.
+ */
+ if (locked_rq && rq != locked_rq) {
+ scx_error(sch, "Invalid target CPU %d", cpu);
+ return -EINVAL;
+ }
- if (!locked_rq)
- rq_unlock_irqrestore(rq, &rf);
+ /*
+ * If no rq lock is held, allow to operate on any CPU by acquiring
+ * the corresponding rq lock.
+ */
+ if (!locked_rq) {
+ rq_lock_irqsave(rq, &rf);
+ update_rq_clock(rq);
}
+
+ rq->scx.cpuperf_target = perf;
+ cpufreq_update_util(rq, 0);
+
+ if (!locked_rq)
+ rq_unlock_irqrestore(rq, &rf);
+
+ return 0;
+}
+
+__bpf_kfunc void scx_bpf_cpuperf_set(s32 cpu, u32 perf, const struct bpf_prog_aux *aux)
+{
+ struct scx_sched *sch;
+
+ guard(rcu)();
+
+ sch = scx_prog_sched(aux);
+ if (unlikely(!sch))
+ return;
+
+ scx_cpuperf_set(sch, cpu, perf);
}
/**
--
2.55.0
^ permalink raw reply related [flat|nested] 11+ messages in thread* Re: [PATCH 4/5] sched_ext: Factor out scx_cpuperf_set()
2026-07-24 18:21 ` [PATCH 4/5] sched_ext: Factor out scx_cpuperf_set() Tejun Heo
@ 2026-07-24 18:30 ` sashiko-bot
2026-07-24 19:07 ` Tejun Heo
0 siblings, 1 reply; 11+ messages in thread
From: sashiko-bot @ 2026-07-24 18:30 UTC (permalink / raw)
To: Tejun Heo; +Cc: sched-ext
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Low] Kernel-doc comment disconnected from its intended function.
--
--- Patch [4]: [PATCH 4/5] sched_ext: Factor out scx_cpuperf_set() ---
commit b27126d56fa54ad2d995a8ef1399b04f6ce7c044
Author: Tejun Heo <tj@kernel.org>
sched_ext: Factor out scx_cpuperf_set()
This commit factors out the CPU performance target write from
scx_bpf_cpuperf_set() into a new internal scx_cpuperf_set() helper. It also
flattens the nested validation into early returns to prepare for future
changes.
> diff --git a/kernel/sched/ext/ext.c b/kernel/sched/ext/ext.c
> index d78b28d39d46c..00fe8490050df 100644
> --- a/kernel/sched/ext/ext.c
> +++ b/kernel/sched/ext/ext.c
> @@ -9793,49 +9793,60 @@ __bpf_kfunc u32 scx_bpf_cidperf_cur(s32 cid, const struct bpf_prog_aux *aux)
> * use. Consult hardware and cpufreq documentation for more information. The
> * current performance level can be monitored using scx_bpf_cpuperf_cur().
> */
> -__bpf_kfunc void scx_bpf_cpuperf_set(s32 cpu, u32 perf, const struct bpf_prog_aux *aux)
> +static s32 scx_cpuperf_set(struct scx_sched *sch, s32 cpu, u32 perf)
[Severity: Low]
Does this disconnect the kernel-doc comment from its intended function?
By placing the new static helper scx_cpuperf_set() directly under the
kernel-doc block for scx_bpf_cpuperf_set(), kernel-doc will parse this as
the documentation for the helper instead of the kfunc.
This will trigger kernel-doc warnings during documentation builds about
mismatched function names and arguments (such as missing @sch and @aux),
and leaves the public kfunc scx_bpf_cpuperf_set() undocumented in the
generated kernel documentation.
> {
> - struct scx_sched *sch;
> -
> - guard(rcu)();
> -
> - sch = scx_prog_sched(aux);
> - if (unlikely(!sch))
> - return;
[ ... ]
> + if (!locked_rq)
> + rq_unlock_irqrestore(rq, &rf);
> +
> + return 0;
> +}
> +
> +__bpf_kfunc void scx_bpf_cpuperf_set(s32 cpu, u32 perf, const struct bpf_prog_aux *aux)
> +{
> + struct scx_sched *sch;
--
Sashiko AI review · https://sashiko.dev/#/patchset/20260724182125.985061-1-tj@kernel.org?part=4
^ permalink raw reply [flat|nested] 11+ messages in thread* Re: [PATCH 4/5] sched_ext: Factor out scx_cpuperf_set()
2026-07-24 18:30 ` sashiko-bot
@ 2026-07-24 19:07 ` Tejun Heo
0 siblings, 0 replies; 11+ messages in thread
From: Tejun Heo @ 2026-07-24 19:07 UTC (permalink / raw)
To: sashiko-reviews; +Cc: sched-ext
On Fri, Jul 24, 2026 at 06:30:56PM +0000, sashiko-bot@kernel.org wrote:
> [Severity: Low]
> Does this disconnect the kernel-doc comment from its intended function?
>
> By placing the new static helper scx_cpuperf_set() directly under the
> kernel-doc block for scx_bpf_cpuperf_set(), kernel-doc will parse this as
> the documentation for the helper instead of the kfunc.
Indeed, the factoring left the kernel-doc comment attached to the
helper. Will move it to the kfunc when applying.
Thanks.
--
tejun
^ permalink raw reply [flat|nested] 11+ messages in thread
* [PATCH 5/5] sched_ext: Gate scx_bpf_cidperf_set() behind a new SCX_CAP_PERF
2026-07-24 18:21 [PATCHSET sched_ext/for-7.3] sched_ext: Follow-up fixes and missing cap enforcement Tejun Heo
` (3 preceding siblings ...)
2026-07-24 18:21 ` [PATCH 4/5] sched_ext: Factor out scx_cpuperf_set() Tejun Heo
@ 2026-07-24 18:21 ` Tejun Heo
2026-07-24 18:46 ` sashiko-bot
4 siblings, 1 reply; 11+ messages in thread
From: Tejun Heo @ 2026-07-24 18:21 UTC (permalink / raw)
To: David Vernet, Andrea Righi, Changwoo Min
Cc: sched-ext, Emil Tsalapatis, linux-kernel, Tejun Heo
scx_bpf_cidperf_set() reaches cpufreq with no cap check, so any cid-form
sub-sched can steer the frequency of any cid in its view, including ones it
holds nothing on.
Gate it behind a new SCX_CAP_PERF rather than SCX_CAP_BASE: hardware control
is a separate axis from queue access - a parent may well delegate scheduling
on a cid without handing over its frequency. PERF neither implies nor is
implied by the other caps. The check runs under the target rq's lock, which
ecaps updates are also folded under, so it is authoritative - a write can
never land after a revoke has taken effect. Denials are counted in
SCX_EV_SUB_CIDPERF_DENIED.
The operation is synchronous and the outcome is reported to the caller.
scx_bpf_cidperf_set() shipped in v7.2 returning void, so add
scx_bpf_cidperf_set___v2() returning 0 or -errno, -EACCES on denial, and
keep the old kfunc as a gated void wrapper. The tools compat wrapper prefers
the v2 and returns 0 on pre-gate kernels, which always apply the write.
scx_qmap grants PERF alongside its existing cid grants so the cpuperf demo
keeps working in sub-scheds.
Signed-off-by: Tejun Heo <tj@kernel.org>
---
kernel/sched/ext/ext.c | 53 +++++++++++++++++++-----
kernel/sched/ext/internal.h | 16 ++++++-
tools/sched_ext/include/scx/common.bpf.h | 2 +-
tools/sched_ext/include/scx/compat.bpf.h | 18 ++++++++
tools/sched_ext/scx_qmap.bpf.c | 17 +++++---
5 files changed, 88 insertions(+), 18 deletions(-)
diff --git a/kernel/sched/ext/ext.c b/kernel/sched/ext/ext.c
index 366245daab68..4b7cb0665798 100644
--- a/kernel/sched/ext/ext.c
+++ b/kernel/sched/ext/ext.c
@@ -5144,6 +5144,7 @@ static const char *scx_cap_names[__SCX_NR_CAPS] = {
[__SCX_CAP_ENQ_IMMED] = "enq_immed",
[__SCX_CAP_ENQ] = "enq",
[__SCX_CAP_PREEMPT] = "preempt",
+ [__SCX_CAP_PERF] = "perf",
};
static ssize_t scx_attr_caps_show(struct kobject *kobj,
@@ -9800,6 +9801,7 @@ static s32 scx_cpuperf_set(struct scx_sched *sch, s32 cpu, u32 perf)
{
struct rq *rq, *locked_rq;
struct rq_flags rf;
+ s32 ret;
if (unlikely(perf > SCX_CPUPERF_ONE)) {
scx_error(sch, "Invalid cpuperf target %u for CPU %d", perf, cpu);
@@ -9830,13 +9832,24 @@ static s32 scx_cpuperf_set(struct scx_sched *sch, s32 cpu, u32 perf)
update_rq_clock(rq);
}
- rq->scx.cpuperf_target = perf;
- cpufreq_update_util(rq, 0);
+ /*
+ * ecaps updates are folded under the rq lock, making this test
+ * authoritative: a write can never land after a revoke has taken
+ * effect on @cpu.
+ */
+ if (likely(!scx_missing_caps(sch, cpu, SCX_CAP_PERF))) {
+ rq->scx.cpuperf_target = perf;
+ cpufreq_update_util(rq, 0);
+ ret = 0;
+ } else {
+ __scx_add_event(sch, SCX_EV_SUB_CIDPERF_DENIED, 1);
+ ret = -EACCES;
+ }
if (!locked_rq)
rq_unlock_irqrestore(rq, &rf);
- return 0;
+ return ret;
}
__bpf_kfunc void scx_bpf_cpuperf_set(s32 cpu, u32 perf, const struct bpf_prog_aux *aux)
@@ -9853,15 +9866,18 @@ __bpf_kfunc void scx_bpf_cpuperf_set(s32 cpu, u32 perf, const struct bpf_prog_au
}
/**
- * scx_bpf_cidperf_set - Set the performance target of the CPU at @cid
+ * scx_bpf_cidperf_set___v2 - Set the performance target of the CPU at @cid
* @cid: cid of the CPU to target
* @perf: target performance level [0, %SCX_CPUPERF_ONE]
* @aux: implicit BPF argument to access bpf_prog_aux hidden from BPF progs
*
- * cid-addressed equivalent of scx_bpf_cpuperf_set().
+ * cid-addressed equivalent of scx_bpf_cpuperf_set(). A sub-sched needs
+ * SCX_CAP_PERF on @cid. Returns 0 if the target was applied, -%EACCES if
+ * the write was denied for missing caps, other -errnos if @cid didn't
+ * resolve.
*/
-__bpf_kfunc void scx_bpf_cidperf_set(s32 cid, u32 perf,
- const struct bpf_prog_aux *aux)
+__bpf_kfunc s32 scx_bpf_cidperf_set___v2(s32 cid, u32 perf,
+ const struct bpf_prog_aux *aux)
{
struct scx_sched *sch;
s32 cpu;
@@ -9870,11 +9886,27 @@ __bpf_kfunc void scx_bpf_cidperf_set(s32 cid, u32 perf,
sch = scx_prog_sched(aux);
if (unlikely(!sch))
- return;
+ return -ENODEV;
cpu = scx_cid_to_cpu(sch, cid);
if (cpu < 0)
- return;
- scx_bpf_cpuperf_set(cpu, perf, aux);
+ return cpu;
+
+ return scx_cpuperf_set(sch, cpu, perf);
+}
+
+/**
+ * scx_bpf_cidperf_set - Set the performance target of the CPU at @cid
+ * @cid: cid of the CPU to target
+ * @perf: target performance level [0, %SCX_CPUPERF_ONE]
+ * @aux: implicit BPF argument to access bpf_prog_aux hidden from BPF progs
+ *
+ * Pre-v7.3 variant of scx_bpf_cidperf_set___v2() without the result. Kept
+ * for schedulers built against older kernels.
+ */
+__bpf_kfunc void scx_bpf_cidperf_set(s32 cid, u32 perf,
+ const struct bpf_prog_aux *aux)
+{
+ scx_bpf_cidperf_set___v2(cid, perf, aux);
}
/**
@@ -10306,6 +10338,7 @@ BTF_ID_FLAGS(func, scx_bpf_cpuperf_set, KF_IMPLICIT_ARGS)
BTF_ID_FLAGS(func, scx_bpf_cidperf_cap, KF_IMPLICIT_ARGS)
BTF_ID_FLAGS(func, scx_bpf_cidperf_cur, KF_IMPLICIT_ARGS)
BTF_ID_FLAGS(func, scx_bpf_cidperf_set, KF_IMPLICIT_ARGS)
+BTF_ID_FLAGS(func, scx_bpf_cidperf_set___v2, KF_IMPLICIT_ARGS)
BTF_ID_FLAGS(func, scx_bpf_nr_node_ids)
BTF_ID_FLAGS(func, scx_bpf_nr_cpu_ids)
BTF_ID_FLAGS(func, scx_bpf_nr_cids)
diff --git a/kernel/sched/ext/internal.h b/kernel/sched/ext/internal.h
index 308d16320818..886f1d132e6b 100644
--- a/kernel/sched/ext/internal.h
+++ b/kernel/sched/ext/internal.h
@@ -1206,6 +1206,12 @@ struct scx_event_stats {
* sub-sched lacked baseline access on the target cid.
*/
s64 SCX_EV_SUB_REENQ_DENIED;
+
+ /*
+ * The number of times scx_bpf_cidperf_set() was denied because the
+ * sub-sched lacked SCX_CAP_PERF on the target cid.
+ */
+ s64 SCX_EV_SUB_CIDPERF_DENIED;
};
#define SCX_EVENTS_LIST(SCX_EVENT) \
@@ -1227,7 +1233,8 @@ struct scx_event_stats {
SCX_EVENT(SCX_EV_SUB_FORCED_ADMIT); \
SCX_EVENT(SCX_EV_SUB_PREEMPT_DENIED); \
SCX_EVENT(SCX_EV_SUB_KICK_DENIED); \
- SCX_EVENT(SCX_EV_SUB_REENQ_DENIED)
+ SCX_EVENT(SCX_EV_SUB_REENQ_DENIED); \
+ SCX_EVENT(SCX_EV_SUB_CIDPERF_DENIED)
struct scx_sched;
@@ -1353,6 +1360,11 @@ struct scx_sched_pnode {
* - SCX_ENQ_PREEMPT inserts
* - SCX_KICK_PREEMPT kicks
*
+ * PERF control the cid's cpu power/perf management state, currently the
+ * cpufreq target set through scx_bpf_cidperf_set(). Hardware
+ * control is a separate axis from queue access: PERF neither
+ * implies nor is implied by the caps above.
+ *
* Implied caps apply to the holder's own use of a cid, not to delegation.
* scx_bpf_sub_grant() delegates literally-held caps, so a cap held only through
* implication is usable but cannot be re-delegated to a child. When granting a
@@ -1363,6 +1375,7 @@ enum scx_cap_flags {
__SCX_CAP_ENQ_IMMED = 0,
__SCX_CAP_ENQ = 1,
__SCX_CAP_PREEMPT = 2,
+ __SCX_CAP_PERF = 3,
__SCX_NR_CAPS,
__SCX_CAP_ALL = BIT_U64(__SCX_NR_CAPS) - 1,
@@ -1370,6 +1383,7 @@ enum scx_cap_flags {
SCX_CAP_ENQ_IMMED = BIT_U64(__SCX_CAP_ENQ_IMMED),
SCX_CAP_ENQ = BIT_U64(__SCX_CAP_ENQ),
SCX_CAP_PREEMPT = BIT_U64(__SCX_CAP_PREEMPT),
+ SCX_CAP_PERF = BIT_U64(__SCX_CAP_PERF),
/* alias for minimal cap to make any use of a cpu */
SCX_CAP_BASE = SCX_CAP_ENQ_IMMED,
diff --git a/tools/sched_ext/include/scx/common.bpf.h b/tools/sched_ext/include/scx/common.bpf.h
index acc2b131ea8f..c8809887cbde 100644
--- a/tools/sched_ext/include/scx/common.bpf.h
+++ b/tools/sched_ext/include/scx/common.bpf.h
@@ -112,7 +112,7 @@ u32 scx_bpf_nr_cids(void) __ksym __weak;
u32 scx_bpf_nr_online_cids(void) __ksym __weak;
u32 scx_bpf_cidperf_cap(s32 cid) __ksym __weak;
u32 scx_bpf_cidperf_cur(s32 cid) __ksym __weak;
-void scx_bpf_cidperf_set(s32 cid, u32 perf) __ksym __weak;
+/* scx_bpf_cidperf_set() is wrapped in compat.bpf.h */
/* sub-scheduler cap control, scx_bpf_sub_caps() cgroup_id 0 == self */
s32 scx_bpf_sub_grant(u64 cgroup_id, u64 caps, const struct scx_cmask *cmask,
diff --git a/tools/sched_ext/include/scx/compat.bpf.h b/tools/sched_ext/include/scx/compat.bpf.h
index cf469d5ff9ca..1a798213105b 100644
--- a/tools/sched_ext/include/scx/compat.bpf.h
+++ b/tools/sched_ext/include/scx/compat.bpf.h
@@ -420,6 +420,24 @@ static inline void scx_bpf_dsq_reenq(u64 dsq_id, u64 reenq_flags)
scx_bpf_error("kernel too old to reenqueue foreign local or user DSQs");
}
+/*
+ * v7.3: scx_bpf_cidperf_set() now returns 0 or -errno - -EACCES when a
+ * sub-sched lacks SCX_CAP_PERF on the cid and the write was denied. Pre-gate
+ * kernels always apply the write. Drop the wrapper and move the decl to
+ * common.bpf.h after v7.6.
+ */
+s32 scx_bpf_cidperf_set___v2___compat(s32 cid, u32 perf) __ksym __weak;
+void scx_bpf_cidperf_set___compat(s32 cid, u32 perf) __ksym __weak;
+
+static inline s32 scx_bpf_cidperf_set(s32 cid, u32 perf)
+{
+ if (bpf_ksym_exists(scx_bpf_cidperf_set___v2___compat))
+ return scx_bpf_cidperf_set___v2___compat(cid, perf);
+
+ scx_bpf_cidperf_set___compat(cid, perf);
+ return 0;
+}
+
/*
* Define sched_ext_ops. See compat.h::SCX_OPS_OPEN() for how backward
* compatibility is handled (this macro can be expanded to emit multiple
diff --git a/tools/sched_ext/scx_qmap.bpf.c b/tools/sched_ext/scx_qmap.bpf.c
index aead17658573..8822ed11c0d8 100644
--- a/tools/sched_ext/scx_qmap.bpf.c
+++ b/tools/sched_ext/scx_qmap.bpf.c
@@ -1585,11 +1585,13 @@ __noinline void apply_partition(void)
cmask_copy(&qa.to_grant_cids.mask, &ssc->granted_cids.mask);
cmask_andnot(&qa.to_grant_cids.mask, &ssc->prev_granted.mask);
- scx_bpf_sub_revoke(cgid, SCX_CAP_ENQ_IMMED,
+ scx_bpf_sub_revoke(cgid, SCX_CAP_ENQ_IMMED | SCX_CAP_PERF,
(void *)(long)&qa.prev_rr_cids.mask);
- scx_bpf_sub_revoke(cgid, SCX_CAP_ENQ | SCX_CAP_PREEMPT | SCX_CAP_ENQ_IMMED,
+ scx_bpf_sub_revoke(cgid, SCX_CAP_ENQ | SCX_CAP_PREEMPT |
+ SCX_CAP_ENQ_IMMED | SCX_CAP_PERF,
(void *)(long)&qa.to_revoke_cids.mask);
- scx_bpf_sub_grant(cgid, SCX_CAP_ENQ | SCX_CAP_PREEMPT | SCX_CAP_ENQ_IMMED,
+ scx_bpf_sub_grant(cgid, SCX_CAP_ENQ | SCX_CAP_PREEMPT |
+ SCX_CAP_ENQ_IMMED | SCX_CAP_PERF,
(void *)(long)&qa.to_grant_cids.mask, NULL);
}
@@ -1605,7 +1607,8 @@ __noinline void apply_partition(void)
holder_cgid = qa.part.rr_slots[pos]; /* 0 = self, nothing to grant */
if (holder_cgid)
- scx_bpf_sub_grant(holder_cgid, SCX_CAP_ENQ_IMMED,
+ scx_bpf_sub_grant(holder_cgid,
+ SCX_CAP_ENQ_IMMED | SCX_CAP_PERF,
(void *)(long)&qa.rr_cids.mask, NULL);
}
}
@@ -1696,10 +1699,12 @@ static void rr_advance(void)
* time-share.
*/
if (old_cgid)
- scx_bpf_sub_revoke(old_cgid, SCX_CAP_ENQ_IMMED,
+ scx_bpf_sub_revoke(old_cgid,
+ SCX_CAP_ENQ_IMMED | SCX_CAP_PERF,
(void *)(long)&qa.rr_cids.mask);
if (new_cgid)
- scx_bpf_sub_grant(new_cgid, SCX_CAP_ENQ_IMMED,
+ scx_bpf_sub_grant(new_cgid,
+ SCX_CAP_ENQ_IMMED | SCX_CAP_PERF,
(void *)(long)&qa.rr_cids.mask, NULL);
}
--
2.55.0
^ permalink raw reply related [flat|nested] 11+ messages in thread* Re: [PATCH 5/5] sched_ext: Gate scx_bpf_cidperf_set() behind a new SCX_CAP_PERF
2026-07-24 18:21 ` [PATCH 5/5] sched_ext: Gate scx_bpf_cidperf_set() behind a new SCX_CAP_PERF Tejun Heo
@ 2026-07-24 18:46 ` sashiko-bot
2026-07-24 19:07 ` Tejun Heo
0 siblings, 1 reply; 11+ messages in thread
From: sashiko-bot @ 2026-07-24 18:46 UTC (permalink / raw)
To: Tejun Heo; +Cc: sched-ext
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] The scx_qmap scheduler breaks backward compatibility on older kernels by unconditionally passing SCX_CAP_PERF to scx_bpf_sub_grant() and scx_bpf_sub_revoke().
--
--- Patch [5]: [PATCH 5/5] sched_ext: Gate scx_bpf_cidperf_set() behind a new SCX_CAP_PERF ---
commit 3bc27e508735d52ab01e9e24c0c210b6cb92366e
Author: Tejun Heo <tj@kernel.org>
sched_ext: Gate scx_bpf_cidperf_set() behind a new SCX_CAP_PERF
scx_bpf_cidperf_set() reaches cpufreq with no cap check, so any cid-form
sub-sched can steer the frequency of any cid in its view. This gates the
operation behind a new SCX_CAP_PERF capability check.
> diff --git a/tools/sched_ext/scx_qmap.bpf.c b/tools/sched_ext/scx_qmap.bpf.c
> index aead17658573b..8822ed11c0d8b 100644
> --- a/tools/sched_ext/scx_qmap.bpf.c
> +++ b/tools/sched_ext/scx_qmap.bpf.c
> @@ -1585,11 +1585,13 @@ __noinline void apply_partition(void)
> cmask_copy(&qa.to_grant_cids.mask, &ssc->granted_cids.mask);
> cmask_andnot(&qa.to_grant_cids.mask, &ssc->prev_granted.mask);
>
> - scx_bpf_sub_revoke(cgid, SCX_CAP_ENQ_IMMED,
> + scx_bpf_sub_revoke(cgid, SCX_CAP_ENQ_IMMED | SCX_CAP_PERF,
> (void *)(long)&qa.prev_rr_cids.mask);
> - scx_bpf_sub_revoke(cgid, SCX_CAP_ENQ | SCX_CAP_PREEMPT | SCX_CAP_ENQ_IMMED,
> + scx_bpf_sub_revoke(cgid, SCX_CAP_ENQ | SCX_CAP_PREEMPT |
> + SCX_CAP_ENQ_IMMED | SCX_CAP_PERF,
> (void *)(long)&qa.to_revoke_cids.mask);
> - scx_bpf_sub_grant(cgid, SCX_CAP_ENQ | SCX_CAP_PREEMPT | SCX_CAP_ENQ_IMMED,
> + scx_bpf_sub_grant(cgid, SCX_CAP_ENQ | SCX_CAP_PREEMPT |
> + SCX_CAP_ENQ_IMMED | SCX_CAP_PERF,
> (void *)(long)&qa.to_grant_cids.mask, NULL);
> }
[Severity: High]
Does unconditionally passing SCX_CAP_PERF break backward compatibility on
older kernels?
If scx_qmap is loaded on a pre-gate kernel where SCX_CAP_PERF is not part of
__SCX_CAP_ALL, won't the kernel's scx_bpf_sub_grant() and scx_bpf_sub_revoke()
reject the unknown cap bit?
Since they call scx_error() on invalid caps, would this immediately abort
the BPF scheduler and revert the system to CFS on older kernels?
[ ... ]
> @@ -1696,10 +1699,12 @@ static void rr_advance(void)
> * time-share.
> */
> if (old_cgid)
> - scx_bpf_sub_revoke(old_cgid, SCX_CAP_ENQ_IMMED,
> + scx_bpf_sub_revoke(old_cgid,
> + SCX_CAP_ENQ_IMMED | SCX_CAP_PERF,
> (void *)(long)&qa.rr_cids.mask);
> if (new_cgid)
> - scx_bpf_sub_grant(new_cgid, SCX_CAP_ENQ_IMMED,
> + scx_bpf_sub_grant(new_cgid,
> + SCX_CAP_ENQ_IMMED | SCX_CAP_PERF,
> (void *)(long)&qa.rr_cids.mask, NULL);
> }
[Severity: High]
Does the same compatibility issue apply here during the round-robin advance
in rr_advance()?
--
Sashiko AI review · https://sashiko.dev/#/patchset/20260724182125.985061-1-tj@kernel.org?part=5
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: [PATCH 5/5] sched_ext: Gate scx_bpf_cidperf_set() behind a new SCX_CAP_PERF
2026-07-24 18:46 ` sashiko-bot
@ 2026-07-24 19:07 ` Tejun Heo
0 siblings, 0 replies; 11+ messages in thread
From: Tejun Heo @ 2026-07-24 19:07 UTC (permalink / raw)
To: sashiko-reviews; +Cc: sched-ext
On Fri, Jul 24, 2026 at 06:46:05PM +0000, sashiko-bot@kernel.org wrote:
> [Severity: High]
> Does unconditionally passing SCX_CAP_PERF break backward compatibility on
> older kernels?
>
> If scx_qmap is loaded on a pre-gate kernel where SCX_CAP_PERF is not part of
> __SCX_CAP_ALL, won't the kernel's scx_bpf_sub_grant() and scx_bpf_sub_revoke()
> reject the unknown cap bit?
It would, but that's fine. The sub-scheduler caps interface is still
going through initial development and there is no cross-version
compatibility commitment for it yet. The in-tree schedulers track the
in-tree kernel for these interfaces.
Thanks.
--
tejun
^ permalink raw reply [flat|nested] 11+ messages in thread