perf/x86/intel: Fix OMR snoop information parsing issues
Subsystems (2)
PATCH 1/2 perf/x86/intel: Fix OMR snoop information parsing issues
When omr_source is 0x2, the omr_snoop (bit[6]) and omr_promoted (bit[7]) fields are combined to represent the snoop information. However, the omr_promoted field was not left-shifted by 1 bit, resulting in incorrect snoop information.
Besides, the snoop information parsing is not accurate for some OMR sources, like the snoop information should be SNOOP_NONE for these memory access (omr_source >= 7) instead of SNOOP_HIT.
Fix these issues.
--- arch/x86/events/intel/ds.c | 11 +++++++---- 1 file changed, 7 insertions(+), 4 deletions(-)
arch/x86/events/intel/ds.c +7 -4
@@ -345,12 +345,12 @@ static u64 parse_omr_data_source(u8 dse)if (omr.omr_remote) val |=REM ; - val |= omr.omr_hitm ?P (SNOOP ,HITM ) :P (SNOOP ,HIT ); -if (omr.omr_source ==0x2 ) { -u8 snoop = omr.omr_snoop | omr.omr_promoted; +u8 snoop = omr.omr_snoop | (omr.omr_promoted <<1 ); -if (snoop ==0x0 ) +if (omr.omr_hitm) + val |=P (SNOOP ,HITM ); +else if (snoop ==0x0 ) val |=P (SNOOP ,NA );else if (snoop ==0x1 ) val |=P (SNOOP ,MISS );
@@ -359,7 +359,10 @@ static u64 parse_omr_data_source(u8 dse)else if (snoop ==0x3 ) val |=P (SNOOP ,NONE ); }else if (omr.omr_source >0x2 && omr.omr_source <0x7 ) { + val |= omr.omr_hitm ?P (SNOOP ,HITM ) :P (SNOOP ,HIT ); val |= omr.omr_snoop ?P (SNOOPX ,FWD ) :0 ; + }else { + val |=P (SNOOP ,NONE ); }return val;
base-commit: 73cee0aad1ee2479fde2c9b753a1b66acb7d1b9a
Signature · 1 line
2.34.1
PATCH 2/2 perf/x86: Update cap_user_rdpmc base on rdpmc user disable state
After introducing the RDPMC user disable feature, user-space RDPMC may return 0 instead of the actual event count. This creates an inconsistency with cap_user_rdpmc, where cap_user_rdpmc is set, but user-space RDPMC only returns 0.
To accurately represent the user-space RDPMC capability, update cap_user_rdpmc based on the RDPMC user disable state. If RDPMC user disable is enabled, cap_user_rdpmc is set to false, allowing user-space programs to fall back to the read() syscall to obtain the real event count.
--- arch/x86/events/core.c | 3 +++ 1 file changed, 3 insertions(+)
arch/x86/events/core.c +3 -0
@@ -2807,6 +2807,9 @@ void arch_perf_update_userpage(struct perf_event *event, userpg->cap_user_time_zero =0 ; userpg->cap_user_rdpmc = !!(event->hw.flags &PERF_EVENT_FLAG_USER_READ_CNT ); +if (x86_pmu_has_rdpmc_user_disable (event->pmu) && + event->hw.config &ARCH_PERFMON_EVENTSEL_RDPMC_USER_DISABLE ) + userpg->cap_user_rdpmc =0 ; userpg->pmc_width = x86_pmu.cntval_bits;if (!using_native_sched_clock () || !sched_clock_stable ())
Signature · 1 line
2.34.1
Ian Rogers
reviewerRe: [PATCH 2/2] perf/x86: Update cap_user_rdpmc base on rdpmc user disable state
Quoted 23 lines · Dapeng Mi
After introducing the RDPMC user disable feature, user-space RDPMC may return 0 instead of the actual event count. This creates an inconsistency with
cap_user_rdpmc, wherecap_user_rdpmcis set, but user-space RDPMC only returns 0.To accurately represent the user-space RDPMC capability, update
cap_user_rdpmcbased on the RDPMC user disable state. If RDPMC user disable is enabled,cap_user_rdpmcis set to false, allowing user-space programs to fall back to theread()syscall to obtain the real event count.Fixes 59af95e028d4 ("perf/x86/intel: Add support for rdpmc user disable feature")Signed-off-by Dapeng Mi <[email protected]>--- arch/x86/events/core.c | 3 +++ 1 file changed, 3 insertions(+)arch/x86/events/core.c +0
-0
arch/x86/events/core.c void arch_perf_update_userpage(struct perf_event *event, · line 2810
Quoted 4 lines
arch/x86/events/core.c +1
-0@@ -2807,6 +2807,9 @@ void arch_perf_update_userpage(struct perf_event *event, userpg->cap_user_time_zero =0 ; userpg->cap_user_rdpmc = !!(event->hw.flags &PERF_EVENT_FLAG_USER_READ_CNT ); + if (x86_pmu_has_rdpmc_user_disable(event->pmu) &&
With the AI's help the following bug was spotted:
Places like cpu_clock_event_add call perf_event_update_userpage with a software event:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/kernel/events/core.c#n12314
This then calls arch_perf_update_userpage:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/kernel/events/core.c#n6870
In x86_pmu_has_rdpmc_user_disable:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/x86/events/perf_event.h#n1336
```
static inline bool x86_pmu_has_rdpmc_user_disable(struct pmu *pmu)
{
return !!(hybrid(pmu, config_mask) &
ARCH_PERFMON_EVENTSEL_RDPMC_USER_DISABLE);
}
```The hybrid call does a call to hybrid_pmu:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/x86/events/perf_event.h#n793
and that does a container_of:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/x86/events/perf_event.h#n782
```
static __always_inline struct x86_hybrid_pmu *hybrid_pmu(struct pmu *pmu)
{
return container_of(pmu, struct x86_hybrid_pmu, pmu);
}
```In the case that the event's pmu is a software PMU the container_of should be invalid and this will lead to an out-of-bounds read of the config_mask on hybrid systems.
Unfortunately checking the event is x86 doesn't cover the hybrid case:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/x86/events/core.c#n780
and it seems this bug may exist elsewhere. It'd be nice if in the hybrid_pmu function there were a BUG_ON(!is_x86_pmu(pmu)), but unfortunately that'd only get exposed on hybrid systems.
Thanks,
Ian
Quoted 8 lines
arch/x86/events/core.c +2
-1+ event->hw.config & ARCH_PERFMON_EVENTSEL_RDPMC_USER_DISABLE) + userpg->cap_user_rdpmc = 0; userpg->pmc_width = x86_pmu.cntval_bits;if (!using_native_sched_clock () || !sched_clock_stable ()) --2.34.1
Ian Rogers
reviewerRe: [PATCH 2/2] perf/x86: Update cap_user_rdpmc base on rdpmc user disable state
Quoted 60 lines · Ian Rogers
Quoted 23 lines · Dapeng Mi
After introducing the RDPMC user disable feature, user-space RDPMC may return 0 instead of the actual event count. This creates an inconsistency with
cap_user_rdpmc, wherecap_user_rdpmcis set, but user-space RDPMC only returns 0.To accurately represent the user-space RDPMC capability, update
cap_user_rdpmcbased on the RDPMC user disable state. If RDPMC user disable is enabled,cap_user_rdpmcis set to false, allowing user-space programs to fall back to theread()syscall to obtain the real event count.Fixes 59af95e028d4 ("perf/x86/intel: Add support for rdpmc user disable feature")Signed-off-by Dapeng Mi <[email protected]>--- arch/x86/events/core.c | 3 +++ 1 file changed, 3 insertions(+)arch/x86/events/core.c +0
-0arch/x86/events/core.c void arch_perf_update_userpage(struct perf_event *event, · line 2810
Quoted 4 lines
arch/x86/events/core.c +1
-0@@ -2807,6 +2807,9 @@ void arch_perf_update_userpage(struct perf_event *event, userpg->cap_user_time_zero =0 ; userpg->cap_user_rdpmc = !!(event->hw.flags &PERF_EVENT_FLAG_USER_READ_CNT ); + if (x86_pmu_has_rdpmc_user_disable(event->pmu) &&With the AI's help the following bug was spotted:
Places like
cpu_clock_event_addcallperf_event_update_userpagewith a software event:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/kernel/events/core.c#n12314
This then callsarch_perf_update_userpage:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/kernel/events/core.c#n6870
Inx86_pmu_has_rdpmc_user_disable:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/x86/events/perf_event.h#n1336``` static inline bool x86_pmu_has_rdpmc_user_disable(struct pmu *pmu) { return !!(hybrid(pmu, config_mask) & ARCH_PERFMON_EVENTSEL_RDPMC_USER_DISABLE); } ```The hybrid call does a call to
hybrid_pmu:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/x86/events/perf_event.h#n793
and that does acontainer_of:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/x86/events/perf_event.h#n782``` static __always_inline struct x86_hybrid_pmu *hybrid_pmu(struct pmu *pmu) { return container_of(pmu, struct x86_hybrid_pmu, pmu); } ```In the case that the event's pmu is a software PMU the
container_ofshould be invalid and this will lead to an out-of-bounds read of theconfig_maskon hybrid systems.
Quoted 5 lines
Unfortunately checking the event is x86 doesn't cover the hybrid case:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/x86/events/core.c#n780
and it seems this bug may exist elsewhere. It'd be nice if in thehybrid_pmufunction there were aBUG_ON(!is_x86_pmu(pmu)), but unfortunately that'd only get exposed on hybrid systems.
Actually is_x86_event does work for hybrid, so making this:
``` if (is_x86_event(event) && x86_pmu_has_rdpmc_user_disable(event->pmu) && ... ```
Should fix the issue.
Thanks,
Ian
Quoted 11 lines
Thanks,
IanQuoted 8 lines
arch/x86/events/core.c +2
-1+ event->hw.config & ARCH_PERFMON_EVENTSEL_RDPMC_USER_DISABLE) + userpg->cap_user_rdpmc = 0; userpg->pmc_width = x86_pmu.cntval_bits;if (!using_native_sched_clock () || !sched_clock_stable ()) --2.34.1
Ian Rogers
reviewerPATCH 1/2 perf/x86: Avoid inadvertent casts to x86_hybrid_pmu
The patch:
https://lore.kernel.org/lkml/[email protected]/
showed it was pretty easy to accidentally cast non-x86 PMUs to x86_hybrid_pmus. Add a BUG_ON for that case. Restructure is_x86_event and add an is_x86_pmu to facilitate this.
---
Only build tested.
--- arch/x86/events/core.c | 16 ---------------- arch/x86/events/perf_event.h | 19 ++++++++++++++++++- 2 files changed, 18 insertions(+), 17 deletions(-)
arch/x86/events/core.c +0 -16
@@ -774,22 +774,6 @@ void x86_pmu_enable_all(int added) } } -int is_x86_event (struct perf_event *event) -{ -/* -* For a non-hybrid platforms, the type of X86 pmu is -* always PERF_TYPE_RAW. -* For a hybrid platform, the PERF_PMU_CAP_EXTENDED_HW_TYPE -* is a unique capability for the X86 PMU. -* Use them to detect a X86 event. -*/ -if (event->pmu->type ==PERF_TYPE_RAW || - event->pmu->capabilities &PERF_PMU_CAP_EXTENDED_HW_TYPE ) -return true; - -return false; -} -struct pmu *x86_get_pmu (unsigned int cpu) {struct cpu_hw_events *cpuc = &per_cpu (cpu_hw_events, cpu);
arch/x86/events/perf_event.h +18 -1
@@ -115,7 +115,23 @@ static inline bool is_topdown_event(struct perf_event *event)return is_metric_event (event) ||is_slots_event (event); } -int is_x86_event (struct perf_event *event); +static inline bool is_x86_pmu (struct pmu *pmu) +{ +/* +* For a non-hybrid platforms, the type of X86 pmu is +* always PERF_TYPE_RAW. +* For a hybrid platform, the PERF_PMU_CAP_EXTENDED_HW_TYPE +* is a unique capability for the X86 PMU. +* Use them to detect a X86 event. +*/ +return pmu->type ==PERF_TYPE_RAW || + (pmu->capabilities &PERF_PMU_CAP_EXTENDED_HW_TYPE ); +} + +static inline bool is_x86_event (struct perf_event *event) +{ +return is_x86_pmu (event->pmu); +}static inline bool check_leader_group (struct perf_event *leader,int flags) {
@@ -779,6 +795,7 @@ struct x86_hybrid_pmu { static __always_inline struct x86_hybrid_pmu *hybrid_pmu(struct pmu *pmu) { +BUG_ON (!is_x86_pmu (pmu)); return container_of(pmu, struct x86_hybrid_pmu, pmu); }
Signature · 1 line
2.53.0.851.ga537e3e6e9-goog
Ian Rogers
reviewerPATCH 2/2 perf/x86: Reduce is_hybrid calls and aid ellision of BUG_ON in hybrid_pmu
Use the capabilities of the PMU rather than the global variable perf_is_hybrid to determine if a hybrid pmu has been passed to the main accessors. As the pmu capabilities check mirrors that in is_x86_pmu, the BUG_ON(!is_x86_pmu...) in hybrid_pmu can be elided as it is provably always false (with sufficient function inlining, common sub-expression elimination, etc.) in its most common uses.
---
Only build tested.
--- arch/x86/events/perf_event.h | 52 +++++++++++++++++++----------------- 1 file changed, 28 insertions(+), 24 deletions(-)
arch/x86/events/perf_event.h +28 -24
@@ -802,34 +802,38 @@ static __always_inline struct x86_hybrid_pmu *hybrid_pmu(struct pmu *pmu)extern struct static_key_false perf_is_hybrid;#define is_hybrid () static_branch_unlikely(&perf_is_hybrid) -#define hybrid (_pmu, _field) \ -(*({ \ - typeof(&x86_pmu._field) __Fp = &x86_pmu._field; \ - \ - if (is_hybrid() && (_pmu)) \ - __Fp = &hybrid_pmu(_pmu)->_field; \ - \ - __Fp; \ + +#define hybrid (_pmu, _field) \ +(*({ \ + typeof(&x86_pmu._field) __Fp = &x86_pmu._field; \ + struct pmu *__pmu = _pmu; \ + \ + if (__pmu->capabilities & PERF_PMU_CAP_EXTENDED_HW_TYPE) \ + __Fp = &hybrid_pmu(__pmu)->_field; \ + \ + __Fp; \ })) -#define hybrid_var (_pmu, _var) \ -(*({ \ - typeof(&_var) __Fp = &_var; \ - \ - if (is_hybrid() && (_pmu)) \ - __Fp = &hybrid_pmu(_pmu)->_var; \ - \ - __Fp; \ +#define hybrid_var (_pmu, _var) \ +(*({ \ + typeof(&_var) __Fp = &_var; \ + struct pmu *__pmu = _pmu; \ + \ + if (__pmu->capabilities & PERF_PMU_CAP_EXTENDED_HW_TYPE) \ + __Fp = &hybrid_pmu(__pmu)->_var; \ + \ + __Fp; \ })) -#define hybrid_bit (_pmu, _field) \ -({ \ - bool __Fp = x86_pmu._field; \ - \ - if (is_hybrid() && (_pmu)) \ - __Fp = hybrid_pmu(_pmu)->_field; \ - \ - __Fp; \ +#define hybrid_bit (_pmu, _field) \ +({ \ + bool __Fp = x86_pmu._field; \ + struct pmu *__pmu = _pmu; \ + \ + if (__pmu->capabilities & PERF_PMU_CAP_EXTENDED_HW_TYPE) \ + __Fp = hybrid_pmu(__pmu)->_field; \ + \ + __Fp; \ })/*
Signature · 1 line
2.53.0.851.ga537e3e6e9-goog
Re: [PATCH v1 2/2] perf/x86: Reduce is_hybrid calls and aid ellision of BUG_ON in hybrid_pmu
LGTM.
Quoted 81 lines · Ian Rogers
Use the capabilities of the PMU rather than the global variable
perf_is_hybridto determine if a hybrid pmu has been passed to the main accessors. As the pmu capabilities check mirrors that inis_x86_pmu, theBUG_ON(!is_x86_pmu...)inhybrid_pmucan be elided as it is provably always false (with sufficient function inlining, common sub-expression elimination, etc.) in its most common uses.Signed-off-by Ian Rogers <[email protected]>---
Only build tested.--- arch/x86/events/perf_event.h | 52 +++++++++++++++++++----------------- 1 file changed, 28 insertions(+), 24 deletions(-)arch/x86/events/perf_event.h +28
-24@@ -802,34 +802,38 @@ static __always_inline struct x86_hybrid_pmu *hybrid_pmu(struct pmu *pmu)extern struct static_key_false perf_is_hybrid;#define is_hybrid () static_branch_unlikely(&perf_is_hybrid) -#define hybrid (_pmu, _field) \ -(*({ \ - typeof(&x86_pmu._field) __Fp = &x86_pmu._field; \ - \ - if (is_hybrid() && (_pmu)) \ - __Fp = &hybrid_pmu(_pmu)->_field; \ - \ - __Fp; \ + +#define hybrid (_pmu, _field) \ +(*({ \ + typeof(&x86_pmu._field) __Fp = &x86_pmu._field; \ + struct pmu *__pmu = _pmu; \ + \ + if (__pmu->capabilities & PERF_PMU_CAP_EXTENDED_HW_TYPE) \ + __Fp = &hybrid_pmu(__pmu)->_field; \ + \ + __Fp; \ })) -#define hybrid_var (_pmu, _var) \ -(*({ \ - typeof(&_var) __Fp = &_var; \ - \ - if (is_hybrid() && (_pmu)) \ - __Fp = &hybrid_pmu(_pmu)->_var; \ - \ - __Fp; \ +#define hybrid_var (_pmu, _var) \ +(*({ \ + typeof(&_var) __Fp = &_var; \ + struct pmu *__pmu = _pmu; \ + \ + if (__pmu->capabilities & PERF_PMU_CAP_EXTENDED_HW_TYPE) \ + __Fp = &hybrid_pmu(__pmu)->_var; \ + \ + __Fp; \ })) -#define hybrid_bit (_pmu, _field) \ -({ \ - bool __Fp = x86_pmu._field; \ - \ - if (is_hybrid() && (_pmu)) \ - __Fp = hybrid_pmu(_pmu)->_field; \ - \ - __Fp; \ +#define hybrid_bit (_pmu, _field) \ +({ \ + bool __Fp = x86_pmu._field; \ + struct pmu *__pmu = _pmu; \ + \ + if (__pmu->capabilities & PERF_PMU_CAP_EXTENDED_HW_TYPE) \ + __Fp = hybrid_pmu(__pmu)->_field; \ + \ + __Fp; \ })/*
Peter Zijlstra
maintainerRe: [PATCH v1 2/2] perf/x86: Reduce is_hybrid calls and aid ellision of BUG_ON in hybrid_pmu
In reply to Ian RogersQuoted 6 lines · Ian Rogers
Use the capabilities of the PMU rather than the global variable
perf_is_hybridto determine if a hybrid pmu has been passed to the main accessors. As the pmu capabilities check mirrors that inis_x86_pmu, theBUG_ON(!is_x86_pmu...)inhybrid_pmucan be elided as it is provably always false (with sufficient function inlining, common sub-expression elimination, etc.) in its most common uses.
perf_is_hybrid is not a variable, its a static_branch. You're adding a runtime branch here, for no appreciable benefit.
And while (Intel) client seems flooded with this hybrid nonsense, servers are still sane. Also AMD have uniform PMU across their regular and compact cores and don't need this either.
Ian Rogers
reviewerRe: [PATCH v1 2/2] perf/x86: Reduce is_hybrid calls and aid ellision of BUG_ON in hybrid_pmu
Quoted 9 lines · Peter Zijlstra
Quoted 6 lines · Ian Rogers
Use the capabilities of the PMU rather than the global variable
perf_is_hybridto determine if a hybrid pmu has been passed to the main accessors. As the pmu capabilities check mirrors that inis_x86_pmu, theBUG_ON(!is_x86_pmu...)inhybrid_pmucan be elided as it is provably always false (with sufficient function inlining, common sub-expression elimination, etc.) in its most common uses.
Quoted 6 lines
perf_is_hybridis not a variable, its astatic_branch. You're adding a runtime branch here, for no appreciable benefit.And while (Intel) client seems flooded with this hybrid nonsense, servers are still sane. Also AMD have uniform PMU across their regular and compact cores and don't need this either.
Agreed, and the AI review (set up similarly to how Dapeng suggested) also agrees with you:
Additionally, does replacing the is_hybrid() static branch with a dynamic pointer dereference and bitwise check introduce unnecessary overhead on hot paths (like NMIs and context switches) for non-hybrid x86 CPUs?
This is missing what the next patch does: it introduces a dominating branch, meaning the BUG_ON condition is known to be false at compile time if reasonable compiler optimizations are enabled - possibly better than a nop, although perhaps the static_branch makes the initial capabilities check a nop too. Anyway, this series was primarily meant to catch other inadvertent container_ofs and isn't something that needs to be in production code. I just couldn't see an easy way to do the equivalent of "ifndef NDEBUG". Given the comments on the next patch I suggest we don't adopt the changes, but I'll add some more feedback in that patch.
Thanks,
Ian
Re: [PATCH v1 1/2] perf/x86: Avoid inadvertent casts to x86_hybrid_pmu
In reply to Ian RogersQuoted 71 lines · Ian Rogers
The patch:
https://lore.kernel.org/lkml/[email protected]/
showed it was pretty easy to accidentally cast non-x86 PMUs tox86_hybrid_pmus. Add aBUG_ONfor that case. Restructureis_x86_eventand add anis_x86_pmuto facilitate this.Signed-off-by Ian Rogers <[email protected]>---
Only build tested.--- arch/x86/events/core.c | 16 ---------------- arch/x86/events/perf_event.h | 19 ++++++++++++++++++- 2 files changed, 18 insertions(+), 17 deletions(-)arch/x86/events/core.c +0
-16@@ -774,22 +774,6 @@ void x86_pmu_enable_all(int added) } } -int is_x86_event (struct perf_event *event) -{ -/* -* For a non-hybrid platforms, the type of X86 pmu is -* always PERF_TYPE_RAW. -* For a hybrid platform, the PERF_PMU_CAP_EXTENDED_HW_TYPE -* is a unique capability for the X86 PMU. -* Use them to detect a X86 event. -*/ -if (event->pmu->type ==PERF_TYPE_RAW || - event->pmu->capabilities &PERF_PMU_CAP_EXTENDED_HW_TYPE ) -return true; - -return false; -} -struct pmu *x86_get_pmu (unsigned int cpu) {struct cpu_hw_events *cpuc = &per_cpu (cpu_hw_events, cpu);arch/x86/events/perf_event.h +17
-1@@ -115,7 +115,23 @@ static inline bool is_topdown_event(struct perf_event *event)return is_metric_event (event) ||is_slots_event (event); } -int is_x86_event (struct perf_event *event); +static inline bool is_x86_pmu (struct pmu *pmu) +{ +/* +* For a non-hybrid platforms, the type of X86 pmu is +* always PERF_TYPE_RAW. +* For a hybrid platform, the PERF_PMU_CAP_EXTENDED_HW_TYPE +* is a unique capability for the X86 PMU. +* Use them to detect a X86 event. +*/ +return pmu->type ==PERF_TYPE_RAW || + (pmu->capabilities &PERF_PMU_CAP_EXTENDED_HW_TYPE ); +} + +static inline bool is_x86_event (struct perf_event *event) +{ +return is_x86_pmu (event->pmu); +}static inline bool check_leader_group (struct perf_event *leader,int flags) {
arch/x86/events/perf_event.h struct x86_hybrid_pmu { · line 801
Quoted 7 lines
arch/x86/events/perf_event.h +1
-0@@ -779,6 +795,7 @@ struct x86_hybrid_pmu { static __always_inline struct x86_hybrid_pmu *hybrid_pmu(struct pmu *pmu) { +BUG_ON (!is_x86_pmu (pmu)); return container_of(pmu, struct x86_hybrid_pmu, pmu); }
LGTM.
Re: [PATCH v1 1/2] perf/x86: Avoid inadvertent casts to x86_hybrid_pmu
In reply to Ian RogersQuoted 71 lines · Ian Rogers
The patch:
https://lore.kernel.org/lkml/[email protected]/
showed it was pretty easy to accidentally cast non-x86 PMUs tox86_hybrid_pmus. Add aBUG_ONfor that case. Restructureis_x86_eventand add anis_x86_pmuto facilitate this.Signed-off-by Ian Rogers <[email protected]>---
Only build tested.--- arch/x86/events/core.c | 16 ---------------- arch/x86/events/perf_event.h | 19 ++++++++++++++++++- 2 files changed, 18 insertions(+), 17 deletions(-)arch/x86/events/core.c +0
-16@@ -774,22 +774,6 @@ void x86_pmu_enable_all(int added) } } -int is_x86_event (struct perf_event *event) -{ -/* -* For a non-hybrid platforms, the type of X86 pmu is -* always PERF_TYPE_RAW. -* For a hybrid platform, the PERF_PMU_CAP_EXTENDED_HW_TYPE -* is a unique capability for the X86 PMU. -* Use them to detect a X86 event. -*/ -if (event->pmu->type ==PERF_TYPE_RAW || - event->pmu->capabilities &PERF_PMU_CAP_EXTENDED_HW_TYPE ) -return true; - -return false; -} -struct pmu *x86_get_pmu (unsigned int cpu) {struct cpu_hw_events *cpuc = &per_cpu (cpu_hw_events, cpu);arch/x86/events/perf_event.h +17
-1@@ -115,7 +115,23 @@ static inline bool is_topdown_event(struct perf_event *event)return is_metric_event (event) ||is_slots_event (event); } -int is_x86_event (struct perf_event *event); +static inline bool is_x86_pmu (struct pmu *pmu) +{ +/* +* For a non-hybrid platforms, the type of X86 pmu is +* always PERF_TYPE_RAW. +* For a hybrid platform, the PERF_PMU_CAP_EXTENDED_HW_TYPE +* is a unique capability for the X86 PMU. +* Use them to detect a X86 event. +*/ +return pmu->type ==PERF_TYPE_RAW || + (pmu->capabilities &PERF_PMU_CAP_EXTENDED_HW_TYPE ); +} + +static inline bool is_x86_event (struct perf_event *event) +{ +return is_x86_pmu (event->pmu); +}static inline bool check_leader_group (struct perf_event *leader,int flags) {
arch/x86/events/perf_event.h struct x86_hybrid_pmu { · line 798
Quoted 4 lines
arch/x86/events/perf_event.h +1
-0@@ -779,6 +795,7 @@ struct x86_hybrid_pmu { static __always_inline struct x86_hybrid_pmu *hybrid_pmu(struct pmu *pmu) { +BUG_ON (!is_x86_pmu (pmu));
With the change, I see the kernel crash is triggered on NVL, I still don't figure out which exact place triggers it. I would debug it later. Thanks.
Quoted 3 lines
arch/x86/events/perf_event.h +0
-0return container_of(pmu, struct x86_hybrid_pmu, pmu); }
Peter Zijlstra
maintainerRe: [PATCH v1 1/2] perf/x86: Avoid inadvertent casts to x86_hybrid_pmu
In reply to Ian RogersQuoted 8 lines · Ian Rogers
The patch:
https://lore.kernel.org/lkml/[email protected]/
showed it was pretty easy to accidentally cast non-x86 PMUs tox86_hybrid_pmus. Add aBUG_ONfor that case. Restructureis_x86_eventand add anis_x86_pmuto facilitate this.
Review struct x86_hybrid_pmu { · line 800
Quoted 6 lines
(quoted diff) +1
-0@@ -779,6 +795,7 @@ struct x86_hybrid_pmu { static __always_inline struct x86_hybrid_pmu *hybrid_pmu(struct pmu *pmu) { + BUG_ON(!is_x86_pmu(pmu)); return container_of(pmu, struct x86_hybrid_pmu, pmu); }
Given that hybrid_pmu will have PERF_PMU_CAP_EXTENDED_HW_TYPE, and we should really only use hyrid_pmu() on one of those, would not the simpler patch be so?
arch/x86/events/perf_event.h +1 -0
@@ -779,6 +779,7 @@ struct x86_hybrid_pmu { static __always_inline struct x86_hybrid_pmu *hybrid_pmu(struct pmu *pmu) { +BUG_ON (!(pmu->capabilities &PERF_PMU_CAP_EXTENDED_HW_TYPE )); return container_of(pmu, struct x86_hybrid_pmu, pmu); }
Re: [PATCH v1 1/2] perf/x86: Avoid inadvertent casts to x86_hybrid_pmu
Quoted 24 lines · Peter Zijlstra
arch/x86/events/perf_event.h struct x86_hybrid_pmu { · line 800
Quoted 13 lines · Ian Rogers
The patch:
https://lore.kernel.org/lkml/[email protected]/
showed it was pretty easy to accidentally cast non-x86 PMUs tox86_hybrid_pmus. Add aBUG_ONfor that case. Restructureis_x86_eventand add anis_x86_pmuto facilitate this.arch/x86/events/perf_event.h +1
-0@@ -779,6 +795,7 @@ struct x86_hybrid_pmu { static __always_inline struct x86_hybrid_pmu *hybrid_pmu(struct pmu *pmu) { +BUG_ON (!is_x86_pmu (pmu)); return container_of(pmu, struct x86_hybrid_pmu, pmu); }Given that
hybrid_pmuwill havePERF_PMU_CAP_EXTENDED_HW_TYPE, and we should really only usehyrid_pmu()on one of those, would not the simpler patch be so?arch/x86/events/perf_event.h +0
-0
arch/x86/events/perf_event.h struct x86_hybrid_pmu { · line 782
Quoted 4 lines
arch/x86/events/perf_event.h +1
-0@@ -779,6 +779,7 @@ struct x86_hybrid_pmu { static __always_inline struct x86_hybrid_pmu *hybrid_pmu(struct pmu *pmu) { +BUG_ON (!(pmu->capabilities &PERF_PMU_CAP_EXTENDED_HW_TYPE ));
It looks we can't add either !is_x86_pmu(pmu) or !(pmu->capabilities & PERF_PMU_CAP_EXTENDED_HW_TYPE) here. hybrid_pmu() is called by the hybrid() marco or other variants, and hybrid() macro is called in many places of the intel_pmu_init(), like the update_pmu_cap() , but the flag PERF_PMU_CAP_EXTENDED_HW_TYPE is still not set for the hybrid pmu->capabilities until intel_pmu_init() ends and the hybrid pmus are registered. Then it would cause the unexpected kernel crash.
[ 1.945128] kernel BUG at arch/x86/events/intel/../perf_event.h:798! [ 1.946131] Oops: invalid opcode: 0000 [#1] SMP NOPTI
[ 1.947127] CPU: 0 UID: 0 PID: 1 Comm: swapper/0 Not tainted 7.0.0-rc3-perf-urgent-gc8b4b538960c #460 PREEMPT(full)
[ 1.947127] Hardware name: Intel Corporation Panther Lake Client Platform/PTL-UH LP5 T3 RVP1, BIOS PTLPFWI1.R00.3171.D00.2504220409 04/22/2025 [ 1.947127] RIP: 0010:intel_pmu_init+0x25c9/0x5fd0
[ 1.947127] Code: db 44 ff 4c 89 35 c7 da 44 ff 48 89 2d 80 da 44 ff e9 49 df ff ff 83 7a 68 04 0f 84 1b f9 ff ff f6 42 6d 01 0f 85 11 f9 ff ff <0f> 0b 31 d2 48 89 df
[ 1.947127] RSP: 0000:ffffd5dc800f7db8 EFLAGS: 00010246
[ 1.947127] RAX: 0000000000000001 RBX: 00000000000abfff RCX:
0000000000000000
[ 1.947127] RDX: ffff8f40856bc000 RSI: 0000000000000001 RDI:00000000000000ff
[ 1.947127] RBP: 0000000000000001 R08: ffffffffffffffff R09:
0000000000000004
[ 1.947127] R10: ffffffffbd4e2500 R11: 0000000000000006 R12:ffffffffbc26438b
[ 1.947127] R13: 0000000000000000 R14: 0000000000000000 R15:
0000000000000000
[ 1.947127] FS: 0000000000000000(0000) GS:ffff8f482214f000(0000) knlGS:0000000000000000
[ 1.947127] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 1.947127] CR2: ffff8f47ff7ff000 CR3: 00000004c1434001 CR4:0000000000f70ef0
[ 1.947127] PKRU: 55555554
[ 1.947127] Call Trace:
[ 1.947127] <TASK>
[ 1.947127] ? __pfx_init_hw_perf_events+0x10/0x10
[ 1.947127] init_hw_perf_events+0x2af/0x4b0
[ 1.947127] ? __pfx_init_hw_perf_events+0x10/0x10
[ 1.947127] do_one_initcall+0x52/0x250
[ 1.947127] ? _raw_spin_unlock+0x18/0x40
[ 1.947127] ? __register_sysctl_table+0x143/0x1a0
[ 1.947127] kernel_init_freeable+0x21d/0x340
[ 1.947127] ? __pfx_kernel_init+0x10/0x10
[ 1.947127] kernel_init+0x1a/0x1c0
[ 1.947127] ret_from_fork+0xcb/0x1c0
[ 1.947127] ? __pfx_kernel_init+0x10/0x10
[ 1.947127] ret_from_fork_asm+0x1a/0x30
[ 1.947127] </TASK>
[ 1.947127] Modules linked in:
[ 1.947127] ---[ end trace 0000000000000000 ]---
[ 1.948128] RIP: 0010:intel_pmu_init+0x25c9/0x5fd0
[ 1.949128] Code: db 44 ff 4c 89 35 c7 da 44 ff 48 89 2d 80 da 44 ff e9 49 df ff ff 83 7a 68 04 0f 84 1b f9 ff ff f6 42 6d 01 0f 85 11 f9 ff ff <0f> 0b 31 d2 48 89 df
[ 1.950129] RSP: 0000:ffffd5dc800f7db8 EFLAGS: 00010246
[ 1.951128] RAX: 0000000000000001 RBX: 00000000000abfff RCX:
0000000000000000
[ 1.952128] RDX: ffff8f40856bc000 RSI: 0000000000000001 RDI:00000000000000ff
[ 1.953128] RBP: 0000000000000001 R08: ffffffffffffffff R09:
0000000000000004
[ 1.954129] R10: ffffffffbd4e2500 R11: 0000000000000006 R12:ffffffffbc26438b
[ 1.955128] R13: 0000000000000000 R14: 0000000000000000 R15:
0000000000000000
[ 1.956128] FS: 0000000000000000(0000) GS:ffff8f482214f000(0000) knlGS:0000000000000000
[ 1.957128] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 1.958128] CR2: ffff8f47ff7ff000 CR3: 00000004c1434001 CR4:0000000000f70ef0
[ 1.959128] PKRU: 55555554
[ 1.960128] Kernel panic - not syncing: Attempted to kill init! exitcode=0x0000000b
I'm not sure if we can move the flag PERF_PMU_CAP_EXTENDED_HW_TYPE setting earlier and eventually find a good place to set the flag. Even it's possible, but could be risky ...
Ian, if you don't object, I would suggest to drop the bug_on(). I would adopt other changes and add the is_x86_pmu() check in the x86_pmu_has_rdpmc_user_disable() to fix the issue.
Thanks.
Quoted 3 lines
arch/x86/events/perf_event.h +0
-0return container_of(pmu, struct x86_hybrid_pmu, pmu); }
Ian Rogers
reviewerRe: [PATCH v1 1/2] perf/x86: Avoid inadvertent casts to x86_hybrid_pmu
Quoted 108 lines · Dapeng
Quoted 24 lines · Peter Zijlstra
arch/x86/events/perf_event.h struct x86_hybrid_pmu { · line 800
Quoted 13 lines · Ian Rogers
The patch:
https://lore.kernel.org/lkml/[email protected]/
showed it was pretty easy to accidentally cast non-x86 PMUs tox86_hybrid_pmus. Add aBUG_ONfor that case. Restructureis_x86_eventand add anis_x86_pmuto facilitate this.arch/x86/events/perf_event.h +1
-0@@ -779,6 +795,7 @@ struct x86_hybrid_pmu { static __always_inline struct x86_hybrid_pmu *hybrid_pmu(struct pmu *pmu) { +BUG_ON (!is_x86_pmu (pmu)); return container_of(pmu, struct x86_hybrid_pmu, pmu); }Given that
hybrid_pmuwill havePERF_PMU_CAP_EXTENDED_HW_TYPE, and we should really only usehyrid_pmu()on one of those, would not the simpler patch be so?arch/x86/events/perf_event.h +0
-0arch/x86/events/perf_event.h struct x86_hybrid_pmu { · line 782
Quoted 4 lines
arch/x86/events/perf_event.h +1
-0@@ -779,6 +779,7 @@ struct x86_hybrid_pmu { static __always_inline struct x86_hybrid_pmu *hybrid_pmu(struct pmu *pmu) { +BUG_ON (!(pmu->capabilities &PERF_PMU_CAP_EXTENDED_HW_TYPE ));It looks we can't add either !is_x86_pmu(pmu) or !(pmu->capabilities &
PERF_PMU_CAP_EXTENDED_HW_TYPE) here.hybrid_pmu()is called by thehybrid()marco or other variants, andhybrid()macro is called in many places of theintel_pmu_init(), like theupdate_pmu_cap(), but the flagPERF_PMU_CAP_EXTENDED_HW_TYPEis still not set for the hybridpmu->capabilitiesuntilintel_pmu_init()ends and the hybrid pmus are registered. Then it would cause the unexpected kernel crash.[ 1.945128] kernel BUG at arch/x86/events/intel/../perf_event.h:798! [ 1.946131] Oops: invalid opcode: 0000 [#1] SMP NOPTI [ 1.947127] CPU: 0 UID: 0 PID: 1 Comm: swapper/0 Not tainted 7.0.0-rc3-perf-urgent-gc8b4b538960c #460 PREEMPT(full) [ 1.947127] Hardware name: Intel Corporation Panther Lake Client Platform/PTL-UH LP5 T3 RVP1, BIOS PTLPFWI1.R00.3171.D00.2504220409 04/22/2025 [ 1.947127] RIP: 0010:intel_pmu_init+0x25c9/0x5fd0 [ 1.947127] Code: db 44 ff 4c 89 35 c7 da 44 ff 48 89 2d 80 da 44 ff e9 49 df ff ff 83 7a 68 04 0f 84 1b f9 ff ff f6 42 6d 01 0f 85 11 f9 ff ff <0f> 0b 31 d2 48 89 df [ 1.947127] RSP: 0000:ffffd5dc800f7db8 EFLAGS: 00010246 [ 1.947127] RAX: 0000000000000001 RBX: 00000000000abfff RCX: 0000000000000000 [ 1.947127] RDX: ffff8f40856bc000 RSI: 0000000000000001 RDI: 00000000000000ff [ 1.947127] RBP: 0000000000000001 R08: ffffffffffffffff R09: 0000000000000004 [ 1.947127] R10: ffffffffbd4e2500 R11: 0000000000000006 R12: ffffffffbc26438b [ 1.947127] R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000000 [ 1.947127] FS: 0000000000000000(0000) GS:ffff8f482214f000(0000) knlGS:0000000000000000 [ 1.947127] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 1.947127] CR2: ffff8f47ff7ff000 CR3: 00000004c1434001 CR4: 0000000000f70ef0 [ 1.947127] PKRU: 55555554 [ 1.947127] Call Trace: [ 1.947127] <TASK> [ 1.947127] ? __pfx_init_hw_perf_events+0x10/0x10 [ 1.947127] init_hw_perf_events+0x2af/0x4b0 [ 1.947127] ? __pfx_init_hw_perf_events+0x10/0x10 [ 1.947127] do_one_initcall+0x52/0x250 [ 1.947127] ? _raw_spin_unlock+0x18/0x40 [ 1.947127] ? __register_sysctl_table+0x143/0x1a0 [ 1.947127] kernel_init_freeable+0x21d/0x340 [ 1.947127] ? __pfx_kernel_init+0x10/0x10 [ 1.947127] kernel_init+0x1a/0x1c0 [ 1.947127] ret_from_fork+0xcb/0x1c0 [ 1.947127] ? __pfx_kernel_init+0x10/0x10 [ 1.947127] ret_from_fork_asm+0x1a/0x30 [ 1.947127] </TASK> [ 1.947127] Modules linked in: [ 1.947127] ---[ end trace 0000000000000000 ]--- [ 1.948128] RIP: 0010:intel_pmu_init+0x25c9/0x5fd0 [ 1.949128] Code: db 44 ff 4c 89 35 c7 da 44 ff 48 89 2d 80 da 44 ff e9 49 df ff ff 83 7a 68 04 0f 84 1b f9 ff ff f6 42 6d 01 0f 85 11 f9 ff ff <0f> 0b 31 d2 48 89 df [ 1.950129] RSP: 0000:ffffd5dc800f7db8 EFLAGS: 00010246 [ 1.951128] RAX: 0000000000000001 RBX: 00000000000abfff RCX: 0000000000000000 [ 1.952128] RDX: ffff8f40856bc000 RSI: 0000000000000001 RDI: 00000000000000ff [ 1.953128] RBP: 0000000000000001 R08: ffffffffffffffff R09: 0000000000000004 [ 1.954129] R10: ffffffffbd4e2500 R11: 0000000000000006 R12: ffffffffbc26438b [ 1.955128] R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000000 [ 1.956128] FS: 0000000000000000(0000) GS:ffff8f482214f000(0000) knlGS:0000000000000000 [ 1.957128] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 1.958128] CR2: ffff8f47ff7ff000 CR3: 00000004c1434001 CR4: 0000000000f70ef0 [ 1.959128] PKRU: 55555554 [ 1.960128] Kernel panic - not syncing: Attempted to kill init!exitcode=0x0000000b
Quoted 7 lines
I'm not sure if we can move the flag
PERF_PMU_CAP_EXTENDED_HW_TYPEsetting earlier and eventually find a good place to set the flag. Even it's possible, but could be risky ...Ian, if you don't object, I would suggest to drop the
bug_on(). I would adopt other changes and add theis_x86_pmu()check in thex86_pmu_has_rdpmc_user_disable()to fix the issue.
No objections from me, these patches aim to improve the code's typing and were intended more as a suggestion easily expressed in code than by email :-)
I feel that the x86_pmu and x86_hybrid_pmu are something of a mess, making features like counter partitioning harder than necessary if the code were structured better - ie 1 partition per PMU, which means multiple PMUs even without hybrid. It'd be nice if we could implement something like the BUG_ON to at least ensure correctness in the current code. I'll try to find other broken casts/container_ofs in the code by visual inspection.
Thanks,
Ian
Quoted 5 lines
Thanks.
Quoted 3 lines
arch/x86/events/perf_event.h +0
-0return container_of(pmu, struct x86_hybrid_pmu, pmu); }
Re: [PATCH v1 1/2] perf/x86: Avoid inadvertent casts to x86_hybrid_pmu
Quoted 111 lines · Ian Rogers
Quoted 110 lines · Dapeng
Quoted 24 lines · Peter Zijlstra
arch/x86/events/perf_event.h struct x86_hybrid_pmu { · line 800
Quoted 13 lines · Ian Rogers
The patch:
https://lore.kernel.org/lkml/[email protected]/
showed it was pretty easy to accidentally cast non-x86 PMUs tox86_hybrid_pmus. Add aBUG_ONfor that case. Restructureis_x86_eventand add anis_x86_pmuto facilitate this.arch/x86/events/perf_event.h +1
-0@@ -779,6 +795,7 @@ struct x86_hybrid_pmu { static __always_inline struct x86_hybrid_pmu *hybrid_pmu(struct pmu *pmu) { +BUG_ON (!is_x86_pmu (pmu)); return container_of(pmu, struct x86_hybrid_pmu, pmu); }Given that
hybrid_pmuwill havePERF_PMU_CAP_EXTENDED_HW_TYPE, and we should really only usehyrid_pmu()on one of those, would not the simpler patch be so?arch/x86/events/perf_event.h +0
-0arch/x86/events/perf_event.h struct x86_hybrid_pmu { · line 782
Quoted 4 lines
arch/x86/events/perf_event.h +1
-0@@ -779,6 +779,7 @@ struct x86_hybrid_pmu { static __always_inline struct x86_hybrid_pmu *hybrid_pmu(struct pmu *pmu) { +BUG_ON (!(pmu->capabilities &PERF_PMU_CAP_EXTENDED_HW_TYPE ));It looks we can't add either !is_x86_pmu(pmu) or !(pmu->capabilities &
PERF_PMU_CAP_EXTENDED_HW_TYPE) here.hybrid_pmu()is called by thehybrid()marco or other variants, andhybrid()macro is called in many places of theintel_pmu_init(), like theupdate_pmu_cap(), but the flagPERF_PMU_CAP_EXTENDED_HW_TYPEis still not set for the hybridpmu->capabilitiesuntilintel_pmu_init()ends and the hybrid pmus are registered. Then it would cause the unexpected kernel crash.[ 1.945128] kernel BUG at arch/x86/events/intel/../perf_event.h:798! [ 1.946131] Oops: invalid opcode: 0000 [#1] SMP NOPTI [ 1.947127] CPU: 0 UID: 0 PID: 1 Comm: swapper/0 Not tainted 7.0.0-rc3-perf-urgent-gc8b4b538960c #460 PREEMPT(full) [ 1.947127] Hardware name: Intel Corporation Panther Lake Client Platform/PTL-UH LP5 T3 RVP1, BIOS PTLPFWI1.R00.3171.D00.2504220409 04/22/2025 [ 1.947127] RIP: 0010:intel_pmu_init+0x25c9/0x5fd0 [ 1.947127] Code: db 44 ff 4c 89 35 c7 da 44 ff 48 89 2d 80 da 44 ff e9 49 df ff ff 83 7a 68 04 0f 84 1b f9 ff ff f6 42 6d 01 0f 85 11 f9 ff ff <0f> 0b 31 d2 48 89 df [ 1.947127] RSP: 0000:ffffd5dc800f7db8 EFLAGS: 00010246 [ 1.947127] RAX: 0000000000000001 RBX: 00000000000abfff RCX: 0000000000000000 [ 1.947127] RDX: ffff8f40856bc000 RSI: 0000000000000001 RDI: 00000000000000ff [ 1.947127] RBP: 0000000000000001 R08: ffffffffffffffff R09: 0000000000000004 [ 1.947127] R10: ffffffffbd4e2500 R11: 0000000000000006 R12: ffffffffbc26438b [ 1.947127] R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000000 [ 1.947127] FS: 0000000000000000(0000) GS:ffff8f482214f000(0000) knlGS:0000000000000000 [ 1.947127] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 1.947127] CR2: ffff8f47ff7ff000 CR3: 00000004c1434001 CR4: 0000000000f70ef0 [ 1.947127] PKRU: 55555554 [ 1.947127] Call Trace: [ 1.947127] <TASK> [ 1.947127] ? __pfx_init_hw_perf_events+0x10/0x10 [ 1.947127] init_hw_perf_events+0x2af/0x4b0 [ 1.947127] ? __pfx_init_hw_perf_events+0x10/0x10 [ 1.947127] do_one_initcall+0x52/0x250 [ 1.947127] ? _raw_spin_unlock+0x18/0x40 [ 1.947127] ? __register_sysctl_table+0x143/0x1a0 [ 1.947127] kernel_init_freeable+0x21d/0x340 [ 1.947127] ? __pfx_kernel_init+0x10/0x10 [ 1.947127] kernel_init+0x1a/0x1c0 [ 1.947127] ret_from_fork+0xcb/0x1c0 [ 1.947127] ? __pfx_kernel_init+0x10/0x10 [ 1.947127] ret_from_fork_asm+0x1a/0x30 [ 1.947127] </TASK> [ 1.947127] Modules linked in: [ 1.947127] ---[ end trace 0000000000000000 ]--- [ 1.948128] RIP: 0010:intel_pmu_init+0x25c9/0x5fd0 [ 1.949128] Code: db 44 ff 4c 89 35 c7 da 44 ff 48 89 2d 80 da 44 ff e9 49 df ff ff 83 7a 68 04 0f 84 1b f9 ff ff f6 42 6d 01 0f 85 11 f9 ff ff <0f> 0b 31 d2 48 89 df [ 1.950129] RSP: 0000:ffffd5dc800f7db8 EFLAGS: 00010246 [ 1.951128] RAX: 0000000000000001 RBX: 00000000000abfff RCX: 0000000000000000 [ 1.952128] RDX: ffff8f40856bc000 RSI: 0000000000000001 RDI: 00000000000000ff [ 1.953128] RBP: 0000000000000001 R08: ffffffffffffffff R09: 0000000000000004 [ 1.954129] R10: ffffffffbd4e2500 R11: 0000000000000006 R12: ffffffffbc26438b [ 1.955128] R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000000 [ 1.956128] FS: 0000000000000000(0000) GS:ffff8f482214f000(0000) knlGS:0000000000000000 [ 1.957128] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 1.958128] CR2: ffff8f47ff7ff000 CR3: 00000004c1434001 CR4: 0000000000f70ef0 [ 1.959128] PKRU: 55555554 [ 1.960128] Kernel panic - not syncing: Attempted to kill init!exitcode=0x0000000b
I'm not sure if we can move the flag
PERF_PMU_CAP_EXTENDED_HW_TYPEsetting earlier and eventually find a good place to set the flag. Even it's possible, but could be risky ...
Quoted 6 lines
Quoted 3 lines
Ian, if you don't object, I would suggest to drop the
bug_on(). I would adopt other changes and add theis_x86_pmu()check in thex86_pmu_has_rdpmc_user_disable()to fix the issue.No objections from me, these patches aim to improve the code's typing and were intended more as a suggestion easily expressed in code than by email :-)
Got it.
Quoted 8 lines
I feel that the
x86_pmuandx86_hybrid_pmuare something of a mess, making features like counter partitioning harder than necessary if the code were structured better - ie 1 partition per PMU, which means multiple PMUs even without hybrid. It'd be nice if we could implement something like theBUG_ONto at least ensure correctness in the current code. I'll try to find other broken casts/container_ofs in the code by visual inspection.
Agree. The intel_pmu_init() indeed becomes over large and complicated along with supporting more and more platforms. I ever wanted to do some optimization for it, but always interrupted by other higher priority tasks. Anyway, I would put it into my todo list and think how could we reconstruct it.
Thanks.
Quoted 9 lines
Thanks,
IanQuoted 5 lines
Thanks.
Quoted 3 lines
arch/x86/events/perf_event.h +0
-0return container_of(pmu, struct x86_hybrid_pmu, pmu); }
Re: [PATCH 2/2] perf/x86: Update cap_user_rdpmc base on rdpmc user disable state
In reply to Ian RogersQuoted 63 lines · Ian Rogers
Quoted 57 lines · Ian Rogers
Quoted 22 lines · Dapeng Mi
After introducing the RDPMC user disable feature, user-space RDPMC may return 0 instead of the actual event count. This creates an inconsistency with
cap_user_rdpmc, wherecap_user_rdpmcis set, but user-space RDPMC only returns 0.To accurately represent the user-space RDPMC capability, update
cap_user_rdpmcbased on the RDPMC user disable state. If RDPMC user disable is enabled,cap_user_rdpmcis set to false, allowing user-space programs to fall back to theread()syscall to obtain the real event count.Fixes 59af95e028d4 ("perf/x86/intel: Add support for rdpmc user disable feature")Signed-off-by Dapeng Mi <[email protected]>--- arch/x86/events/core.c | 3 +++ 1 file changed, 3 insertions(+)arch/x86/events/core.c +0
-0arch/x86/events/core.c void arch_perf_update_userpage(struct perf_event *event, · line 2810
Quoted 4 lines
arch/x86/events/core.c +1
-0@@ -2807,6 +2807,9 @@ void arch_perf_update_userpage(struct perf_event *event, userpg->cap_user_time_zero =0 ; userpg->cap_user_rdpmc = !!(event->hw.flags &PERF_EVENT_FLAG_USER_READ_CNT ); + if (x86_pmu_has_rdpmc_user_disable(event->pmu) &&With the AI's help the following bug was spotted:
Places like
cpu_clock_event_addcallperf_event_update_userpagewith a software event:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/kernel/events/core.c#n12314
This then callsarch_perf_update_userpage:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/kernel/events/core.c#n6870
Inx86_pmu_has_rdpmc_user_disable:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/x86/events/perf_event.h#n1336``` static inline bool x86_pmu_has_rdpmc_user_disable(struct pmu *pmu) { return !!(hybrid(pmu, config_mask) & ARCH_PERFMON_EVENTSEL_RDPMC_USER_DISABLE); } ```The hybrid call does a call to
hybrid_pmu:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/x86/events/perf_event.h#n793
and that does acontainer_of:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/x86/events/perf_event.h#n782``` static __always_inline struct x86_hybrid_pmu *hybrid_pmu(struct pmu *pmu) { return container_of(pmu, struct x86_hybrid_pmu, pmu); } ```In the case that the event's pmu is a software PMU the
container_ofshould be invalid and this will lead to an out-of-bounds read of theconfig_maskon hybrid systems.Quoted 5 lines
Unfortunately checking the event is x86 doesn't cover the hybrid case:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/x86/events/core.c#n780
and it seems this bug may exist elsewhere. It'd be nice if in thehybrid_pmufunction there were aBUG_ON(!is_x86_pmu(pmu)), but unfortunately that'd only get exposed on hybrid systems.
Quoted 5 lines
Actually
is_x86_eventdoes work for hybrid, so making this:``` if (is_x86_event(event) && x86_pmu_has_rdpmc_user_disable(event->pmu) && ... ```Should fix the issue.
Yes, I'd like move the check is_x86_event(event) into x86_pmu_has_rdpmc_user_disable(), then the check "is_x86_event(event)" won't be missed in next call for x86_pmu_has_rdpmc_user_disable(). Thanks.
Quoted 15 lines
Thanks,
IanQuoted 11 lines
Thanks,
IanQuoted 8 lines
arch/x86/events/core.c +2
-1+ event->hw.config & ARCH_PERFMON_EVENTSEL_RDPMC_USER_DISABLE) + userpg->cap_user_rdpmc = 0; userpg->pmc_width = x86_pmu.cntval_bits;if (!using_native_sched_clock () || !sched_clock_stable ()) --2.34.1
Re: [PATCH 2/2] perf/x86: Update cap_user_rdpmc base on rdpmc user disable state
In reply to Ian RogersQuoted 57 lines · Ian Rogers
Quoted 22 lines · Dapeng Mi
After introducing the RDPMC user disable feature, user-space RDPMC may return 0 instead of the actual event count. This creates an inconsistency with
cap_user_rdpmc, wherecap_user_rdpmcis set, but user-space RDPMC only returns 0.To accurately represent the user-space RDPMC capability, update
cap_user_rdpmcbased on the RDPMC user disable state. If RDPMC user disable is enabled,cap_user_rdpmcis set to false, allowing user-space programs to fall back to theread()syscall to obtain the real event count.Fixes 59af95e028d4 ("perf/x86/intel: Add support for rdpmc user disable feature")Signed-off-by Dapeng Mi <[email protected]>--- arch/x86/events/core.c | 3 +++ 1 file changed, 3 insertions(+)arch/x86/events/core.c +0
-0arch/x86/events/core.c void arch_perf_update_userpage(struct perf_event *event, · line 2810
Quoted 4 lines
arch/x86/events/core.c +1
-0@@ -2807,6 +2807,9 @@ void arch_perf_update_userpage(struct perf_event *event, userpg->cap_user_time_zero =0 ; userpg->cap_user_rdpmc = !!(event->hw.flags &PERF_EVENT_FLAG_USER_READ_CNT ); + if (x86_pmu_has_rdpmc_user_disable(event->pmu) &&With the AI's help the following bug was spotted:
Places like
cpu_clock_event_addcallperf_event_update_userpagewith a software event:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/kernel/events/core.c#n12314
This then callsarch_perf_update_userpage:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/kernel/events/core.c#n6870
Inx86_pmu_has_rdpmc_user_disable:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/x86/events/perf_event.h#n1336``` static inline bool x86_pmu_has_rdpmc_user_disable(struct pmu *pmu) { return !!(hybrid(pmu, config_mask) & ARCH_PERFMON_EVENTSEL_RDPMC_USER_DISABLE); } ```The hybrid call does a call to
hybrid_pmu:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/x86/events/perf_event.h#n793
and that does acontainer_of:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/x86/events/perf_event.h#n782``` static __always_inline struct x86_hybrid_pmu *hybrid_pmu(struct pmu *pmu) { return container_of(pmu, struct x86_hybrid_pmu, pmu); } ```In the case that the event's pmu is a software PMU the
container_ofshould be invalid and this will lead to an out-of-bounds read of theconfig_maskon hybrid systems.
Quoted 5 lines
Unfortunately checking the event is x86 doesn't cover the hybrid case:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/x86/events/core.c#n780
and it seems this bug may exist elsewhere. It'd be nice if in thehybrid_pmufunction there were aBUG_ON(!is_x86_pmu(pmu)), but unfortunately that'd only get exposed on hybrid systems.
Good catch. I seems I put too much time on x86 PMU and always ignored there were other event types. Thanks a lot.
Quoted 12 lines
Thanks,
IanQuoted 8 lines
arch/x86/events/core.c +2
-1+ event->hw.config & ARCH_PERFMON_EVENTSEL_RDPMC_USER_DISABLE) + userpg->cap_user_rdpmc = 0; userpg->pmc_width = x86_pmu.cntval_bits;if (!using_native_sched_clock () || !sched_clock_stable ()) --2.34.1
Quoted 10 lines · Dapeng Mi
When
omr_sourceis0x2, theomr_snoop(bit[6]) andomr_promoted(bit[7]) fields are combined to represent the snoop information. However, theomr_promotedfield was not left-shifted by 1 bit, resulting in incorrect snoop information.Besides, the snoop information parsing is not accurate for some OMR sources, like the snoop information should be
SNOOP_NONEfor these memory access (omr_source>= 7) instead ofSNOOP_HIT.
Quoted 6 lines
Fix these issues.
Reported-by Ian Rogers <[email protected]>Closes https://lore.kernel.org/all/CAP-5=fW4zLWFw1v38zCzB9-cseNSTTCtup=p2SDxZq7dPayVww@mail.gmail.com/ open hereFixes d2bdcde9626c ("perf/x86/intel: Add support for PEBS memory auxiliary info field in DMR")Signed-off-by Dapeng Mi <[email protected]>
Quoted 26 lines
--- arch/x86/events/intel/ds.c | 11 +++++++---- 1 file changed, 7 insertions(+), 4 deletions(-)arch/x86/events/intel/ds.c +4
-4@@ -345,12 +345,12 @@ static u64 parse_omr_data_source(u8 dse)if (omr.omr_remote) val |=REM ; - val |= omr.omr_hitm ?P (SNOOP ,HITM ) :P (SNOOP ,HIT ); -if (omr.omr_source ==0x2 ) { -u8 snoop = omr.omr_snoop | omr.omr_promoted; +u8 snoop = omr.omr_snoop | (omr.omr_promoted <<1 ); -if (snoop ==0x0 ) +if (omr.omr_hitm) + val |=P (SNOOP ,HITM ); +else if (snoop ==0x0 ) val |=P (SNOOP ,NA );else if (snoop ==0x1 ) val |=P (SNOOP ,MISS );
arch/x86/events/intel/ds.c static u64 parse_omr_data_source(u8 dse) · line 365
Quoted 7 lines
arch/x86/events/intel/ds.c +3
-0@@ -359,7 +359,10 @@ static u64 parse_omr_data_source(u8 dse)else if (snoop ==0x3 ) val |=P (SNOOP ,NONE ); }else if (omr.omr_source >0x2 && omr.omr_source <0x7 ) { + val |= omr.omr_hitm ?P (SNOOP ,HITM ) :P (SNOOP ,HIT ); val |= omr.omr_snoop ?P (SNOOPX ,FWD ) :0 ; + }else { + val |=P (SNOOP ,NONE );
nit: the else here is fired for omr.omr_source of 0 (invalid) and 1 (reserved) but the commit message says it handles omr_source >= 7. In the case of 0 there's an overlap between P(SNOOP, NA) (from omr_data_source[0])and the P(SNOOP,NONE) here.
nit: why not make omr_data_source const?
Quoted 8 lines
arch/x86/events/intel/ds.c +0
-0}return val;base-commit:
73cee0aad1ee2479fde2c9b753a1b66acb7d1b9a--
2.34.1
The following commit has been merged into the perf/urgent branch of tip:
The following commit has been merged into the perf/urgent branch of tip:
Commit-ID: e7fcc54524f04e42641de99028edd9c69dc19f8c Gitweb: https://git.kernel.org/tip/e7fcc54524f04e42641de99028edd9c69dc19f8c Author: Dapeng Mi <[email protected]> AuthorDate: Wed, 11 Mar 2026 15:52:00 +08:00 Committer: Peter Zijlstra <[email protected]> CommitterDate: Thu, 12 Mar 2026 11:29:16 +01:00
perf/x86/intel: Fix OMR snoop information parsing issues
When omr_source is 0x2, the omr_snoop (bit[6]) and omr_promoted (bit[7]) fields are combined to represent the snoop information. However, the omr_promoted field was not left-shifted by 1 bit, resulting in incorrect snoop information.
Besides, the snoop information parsing is not accurate for some OMR sources, like the snoop information should be SNOOP_NONE for these memory access (omr_source >= 7) instead of SNOOP_HIT.
Fix these issues.
--- arch/x86/events/intel/ds.c | 11 +++++++---- 1 file changed, 7 insertions(+), 4 deletions(-)
arch/x86/events/intel/ds.c +7 -4
@@ -345,12 +345,12 @@ static u64 parse_omr_data_source(u8 dse)if (omr.omr_remote) val |=REM ; - val |= omr.omr_hitm ?P (SNOOP ,HITM ) :P (SNOOP ,HIT ); -if (omr.omr_source ==0x2 ) { -u8 snoop = omr.omr_snoop | omr.omr_promoted; +u8 snoop = omr.omr_snoop | (omr.omr_promoted <<1 ); -if (snoop ==0x0 ) +if (omr.omr_hitm) + val |=P (SNOOP ,HITM ); +else if (snoop ==0x0 ) val |=P (SNOOP ,NA );else if (snoop ==0x1 ) val |=P (SNOOP ,MISS );
@@ -359,7 +359,10 @@ static u64 parse_omr_data_source(u8 dse)else if (snoop ==0x3 ) val |=P (SNOOP ,NONE ); }else if (omr.omr_source >0x2 && omr.omr_source <0x7 ) { + val |= omr.omr_hitm ?P (SNOOP ,HITM ) :P (SNOOP ,HIT ); val |= omr.omr_snoop ?P (SNOOPX ,FWD ) :0 ; + }else { + val |=P (SNOOP ,NONE ); }return val;