From: Nicholas Piggin <npiggin@gmail.com>
To: qemu-devel@nongnu.org
Cc: "Nicholas Piggin" <npiggin@gmail.com>,
"Pavel Dovgalyuk" <Pavel.Dovgalyuk@ispras.ru>,
"Philippe Mathieu-Daudé" <philmd@linaro.org>,
"Richard Henderson" <richard.henderson@linaro.org>,
"Alex Bennée" <alex.bennee@linaro.org>,
"Paolo Bonzini" <pbonzini@redhat.com>,
"John Snow" <jsnow@redhat.com>, "Cleber Rosa" <crosa@redhat.com>,
"Wainer dos Santos Moschetta" <wainersm@redhat.com>,
"Beraldo Leal" <bleal@redhat.com>,
"Michael Tokarev" <mjt@tls.msk.ru>,
qemu-ppc@nongnu.org
Subject: [PATCH v4 23/24] spapr: Fix vpa dispatch count for record-replay
Date: Tue, 12 Mar 2024 03:40:25 +1000 [thread overview]
Message-ID: <20240311174026.2177152-24-npiggin@gmail.com> (raw)
In-Reply-To: <20240311174026.2177152-1-npiggin@gmail.com>
The dispatch count is a field in guest memory that the hypervisor
increments when preempting and dispatching the guest. This was not
being done deterministically with respect to icount, because tcg
exec exit is not deterministic (e.g., an async event could cause it).
Change vpa dispatch count increment to keep track of whether the
vCPU is considered dispatched or not, and only consider it preempted
when calling cede / confer / join / stop-self / etc.
Cc: qemu-ppc@nongnu.org
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
---
include/hw/ppc/spapr_cpu_core.h | 3 +++
hw/ppc/spapr.c | 36 ++-------------------------------
hw/ppc/spapr_hcall.c | 33 ++++++++++++++++++++++++++++++
hw/ppc/spapr_rtas.c | 1 +
4 files changed, 39 insertions(+), 34 deletions(-)
diff --git a/include/hw/ppc/spapr_cpu_core.h b/include/hw/ppc/spapr_cpu_core.h
index 69a52e39b8..5feb0ecc2a 100644
--- a/include/hw/ppc/spapr_cpu_core.h
+++ b/include/hw/ppc/spapr_cpu_core.h
@@ -47,6 +47,7 @@ typedef struct SpaprCpuState {
uint64_t vpa_addr;
uint64_t slb_shadow_addr, slb_shadow_size;
uint64_t dtl_addr, dtl_size;
+ bool dispatched; /* for vpa dispatch counter tracking */
bool prod; /* not migrated, only used to improve dispatch latencies */
struct ICPState *icp;
struct XiveTCTX *tctx;
@@ -61,4 +62,6 @@ static inline SpaprCpuState *spapr_cpu_state(PowerPCCPU *cpu)
return (SpaprCpuState *)cpu->machine_data;
}
+void vpa_dispatch(CPUState *cs, SpaprCpuState *spapr_cpu, bool dispatch);
+
#endif
diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c
index 55263f0815..28ce5f263a 100644
--- a/hw/ppc/spapr.c
+++ b/hw/ppc/spapr.c
@@ -4572,47 +4572,15 @@ static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
{
SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
- /* These are only called by TCG, KVM maintains dispatch state */
-
spapr_cpu->prod = false;
- if (spapr_cpu->vpa_addr) {
+ if (!spapr_cpu->dispatched) {
CPUState *cs = CPU(cpu);
- uint32_t dispatch;
-
- dispatch = ldl_be_phys(cs->as,
- spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
- dispatch++;
- if ((dispatch & 1) != 0) {
- qemu_log_mask(LOG_GUEST_ERROR,
- "VPA: incorrect dispatch counter value for "
- "dispatched partition %u, correcting.\n", dispatch);
- dispatch++;
- }
- stl_be_phys(cs->as,
- spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
+ vpa_dispatch(cs, spapr_cpu, true);
}
}
static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
{
- SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
-
- if (spapr_cpu->vpa_addr) {
- CPUState *cs = CPU(cpu);
- uint32_t dispatch;
-
- dispatch = ldl_be_phys(cs->as,
- spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
- dispatch++;
- if ((dispatch & 1) != 1) {
- qemu_log_mask(LOG_GUEST_ERROR,
- "VPA: incorrect dispatch counter value for "
- "preempted partition %u, correcting.\n", dispatch);
- dispatch++;
- }
- stl_be_phys(cs->as,
- spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
- }
}
static void spapr_machine_class_init(ObjectClass *oc, void *data)
diff --git a/hw/ppc/spapr_hcall.c b/hw/ppc/spapr_hcall.c
index 75c2d12978..5f68c02739 100644
--- a/hw/ppc/spapr_hcall.c
+++ b/hw/ppc/spapr_hcall.c
@@ -487,6 +487,36 @@ static target_ulong h_register_vpa(PowerPCCPU *cpu, SpaprMachineState *spapr,
return ret;
}
+void vpa_dispatch(CPUState *cs, SpaprCpuState *spapr_cpu, bool dispatch)
+{
+ uint32_t counter;
+
+ if (!dispatch) {
+ assert(spapr_cpu->dispatched);
+ } else {
+ assert(!spapr_cpu->dispatched);
+ }
+ spapr_cpu->dispatched = dispatch;
+
+ return;
+
+ if (!spapr_cpu->vpa_addr) {
+ return;
+ }
+
+ /* These are only called by TCG, KVM maintains dispatch state */
+ counter = ldl_be_phys(cs->as, spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
+ counter++;
+ if ((counter & 1) != dispatch) {
+ qemu_log_mask(LOG_GUEST_ERROR,
+ "VPA: incorrect dispatch counter value for "
+ "%s partition %u, correcting.\n",
+ dispatch ? "preempted" : "running", counter);
+ counter++;
+ }
+ stl_be_phys(cs->as, spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, counter);
+}
+
static target_ulong h_cede(PowerPCCPU *cpu, SpaprMachineState *spapr,
target_ulong opcode, target_ulong *args)
{
@@ -505,6 +535,7 @@ static target_ulong h_cede(PowerPCCPU *cpu, SpaprMachineState *spapr,
if (!cpu_has_work(cs)) {
cs->halted = 1;
+ vpa_dispatch(cs, spapr_cpu, false);
cs->exception_index = EXCP_HLT;
cs->exit_request = 1;
ppc_maybe_interrupt(env);
@@ -531,6 +562,8 @@ static target_ulong h_confer_self(PowerPCCPU *cpu)
cs->exit_request = 1;
ppc_maybe_interrupt(&cpu->env);
+ vpa_dispatch(cs, spapr_cpu, false);
+
return H_SUCCESS;
}
diff --git a/hw/ppc/spapr_rtas.c b/hw/ppc/spapr_rtas.c
index f329693c55..8ce4230223 100644
--- a/hw/ppc/spapr_rtas.c
+++ b/hw/ppc/spapr_rtas.c
@@ -216,6 +216,7 @@ static void rtas_stop_self(PowerPCCPU *cpu, SpaprMachineState *spapr,
*/
env->spr[SPR_PSSCR] |= PSSCR_EC;
cs->halted = 1;
+ vpa_dispatch(cs, spapr_cpu_state(cpu), false);
ppc_store_lpcr(cpu, env->spr[SPR_LPCR] & ~pcc->lpcr_pm);
kvmppc_set_reg_ppc_online(cpu, 0);
qemu_cpu_kick(cs);
--
2.42.0
next prev parent reply other threads:[~2024-03-11 17:45 UTC|newest]
Thread overview: 43+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-03-11 17:40 [PATCH v4 00/24] replay: fixes and new test cases Nicholas Piggin
2024-03-11 17:40 ` [PATCH v4 01/24] scripts/replay-dump.py: Update to current rr record format Nicholas Piggin
2024-03-11 17:40 ` [PATCH v4 02/24] scripts/replay-dump.py: rejig decoders in event number order Nicholas Piggin
2024-03-11 17:40 ` [PATCH v4 03/24] tests/avocado: excercise scripts/replay-dump.py in replay tests Nicholas Piggin
2024-03-12 13:25 ` Alex Bennée
2024-03-11 17:40 ` [PATCH v4 04/24] replay: allow runstate shutdown->running when replaying trace Nicholas Piggin
2024-03-12 13:26 ` Alex Bennée
2024-03-11 17:40 ` [PATCH v4 05/24] Revert "replay: stop us hanging in rr_wait_io_event" Nicholas Piggin
2024-03-12 13:33 ` Alex Bennée
2024-03-12 14:03 ` Nicholas Piggin
2024-03-12 21:03 ` Alex Bennée
2024-03-13 5:27 ` Nicholas Piggin
2024-03-14 5:19 ` Nicholas Piggin
2024-03-11 17:40 ` [PATCH v4 06/24] chardev: set record/replay on the base device of a muxed device Nicholas Piggin
2024-03-12 12:39 ` Marc-André Lureau
2024-03-12 14:11 ` Nicholas Piggin
2024-03-11 17:40 ` [PATCH v4 07/24] replay: Fix migration use of clock Nicholas Piggin
2024-03-11 17:40 ` [PATCH v4 08/24] replay: Fix migration replay_mutex locking Nicholas Piggin
2024-03-11 17:40 ` [PATCH v4 09/24] virtio-net: Use replay_schedule_bh_event for bhs that affect machine state Nicholas Piggin
2024-03-11 17:40 ` [PATCH v4 10/24] virtio-net: Use virtual time for RSC timers Nicholas Piggin
2024-03-11 17:40 ` [PATCH v4 11/24] net: Use virtual time for net announce Nicholas Piggin
2024-03-12 9:09 ` Pavel Dovgalyuk
2024-03-12 11:05 ` Nicholas Piggin
2024-03-12 11:12 ` Pavel Dovgalyuk
2024-03-13 5:38 ` Nicholas Piggin
2024-03-13 7:09 ` Nicholas Piggin
2024-03-11 17:40 ` [PATCH v4 12/24] savevm: Fix load_snapshot error path crash Nicholas Piggin
2024-03-11 17:40 ` [PATCH v4 13/24] tests/avocado: replay_linux.py remove the timeout expected guards Nicholas Piggin
2024-03-11 17:40 ` [PATCH v4 14/24] tests/avocado/reverse_debugging.py: mark aarch64 and pseries as not flaky Nicholas Piggin
2024-03-11 17:40 ` [PATCH v4 15/24] tests/avocado: reverse_debugging.py add test for x86-64 q35 machine Nicholas Piggin
2024-03-11 17:40 ` [PATCH v4 16/24] tests/avocado: reverse_debugging.py verify addresses between record and replay Nicholas Piggin
2024-03-11 17:40 ` [PATCH v4 17/24] tests/avocado: reverse_debugging.py stop VM before sampling icount Nicholas Piggin
2024-03-11 17:40 ` [PATCH v4 18/24] tests/avocado: reverse_debugging reverse-step at the end of the trace Nicholas Piggin
2024-03-11 17:40 ` [PATCH v4 19/24] tests/avocado: reverse_debugging.py add snapshot testing Nicholas Piggin
2024-03-11 17:40 ` [PATCH v4 20/24] replay: simple auto-snapshot mode for record Nicholas Piggin
2024-03-12 9:00 ` Pavel Dovgalyuk
2024-03-12 10:43 ` Nicholas Piggin
2024-03-11 17:40 ` [PATCH v4 21/24] tests/avocado: reverse_debugging.py test auto-snapshot mode Nicholas Piggin
2024-03-11 17:40 ` [PATCH v4 22/24] target/ppc: fix timebase register reset state Nicholas Piggin
2024-03-12 13:24 ` Alex Bennée
2024-03-12 13:47 ` Nicholas Piggin
2024-03-11 17:40 ` Nicholas Piggin [this message]
2024-03-11 17:40 ` [PATCH v4 24/24] tests/avocado: replay_linux.py add ppc64 pseries test Nicholas Piggin
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20240311174026.2177152-24-npiggin@gmail.com \
--to=npiggin@gmail.com \
--cc=Pavel.Dovgalyuk@ispras.ru \
--cc=alex.bennee@linaro.org \
--cc=bleal@redhat.com \
--cc=crosa@redhat.com \
--cc=jsnow@redhat.com \
--cc=mjt@tls.msk.ru \
--cc=pbonzini@redhat.com \
--cc=philmd@linaro.org \
--cc=qemu-devel@nongnu.org \
--cc=qemu-ppc@nongnu.org \
--cc=richard.henderson@linaro.org \
--cc=wainersm@redhat.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).