perf/x86/intel: Fix segfault with PEBS-via-PT with sample_freq
Subsystems (2)
Adrian Hunter
reviewerPATCH 1/1 perf/x86/intel: Fix segfault with PEBS-via-PT with sample_freq
Currently, using PEBS-via-PT with a sample frequency instead of a sample period, causes a segfault. For example:
[ 103.607823] BUG: kernel NULL pointer dereference, address: 0000000000000195 [ 103.607876] <NMI> [ 103.607879] ? __die_body.cold+0x19/0x27 [ 103.607885] ? page_fault_oops+0xca/0x290 [ 103.607891] ? exc_page_fault+0x7e/0x1b0 [ 103.607897] ? asm_exc_page_fault+0x26/0x30 [ 103.607901] ? intel_pmu_pebs_event_update_no_drain+0x40/0x60 [ 103.607903] ? intel_pmu_pebs_event_update_no_drain+0x32/0x60 [ 103.607905] intel_pmu_drain_pebs_icl+0x333/0x350 [ 103.607910] handle_pmi_common+0x272/0x3c0 [ 103.607919] intel_pmu_handle_irq+0x10a/0x2e0 [ 103.607922] perf_event_nmi_handler+0x2a/0x50
That happens because intel_pmu_pebs_event_update_no_drain() assumes all the pebs_enabled bits represent counter indexes, which is not always the case. In this particular case, bits 60 and 61 are set for PEBS-via-PT purposes.
The behaviour of PEBS-via-PT with sample frequency is questionable because although a PMI is generated (PEBS_PMI_AFTER_EACH_RECORD), the period is not adjusted anyway.
Putting that aside, fix intel_pmu_pebs_event_update_no_drain() by passing the mask of counter bits instead of 'size'. Note, prior to the Fixes commit, 'size' would be limited to the maximum counter index, so the issue was not hit.
--- arch/x86/events/intel/ds.c | 9 +++++---- 1 file changed, 5 insertions(+), 4 deletions(-)
arch/x86/events/intel/ds.c +5 -4
@@ -2469,8 +2469,9 @@ static void intel_pmu_drain_pebs_core(struct pt_regs *iregs, struct perf_sample_ setup_pebs_fixed_sample_data); } -static void intel_pmu_pebs_event_update_no_drain (struct cpu_hw_events *cpuc,int size) +static void intel_pmu_pebs_event_update_no_drain (struct cpu_hw_events *cpuc,u64 mask) { +u64 pebs_enabled = cpuc->pebs_enabled & mask;struct perf_event *event;int bit;
@@ -2481,7 +2482,7 @@ static void intel_pmu_pebs_event_update_no_drain(struct cpu_hw_events *cpuc, int* It needs to call intel_pmu_save_and_restart_reload() to * update the event->count for this case. */ -for_each_set_bit (bit, (unsigned long *)&cpuc->pebs_enabled, size) { +for_each_set_bit (bit, (unsigned long *)&pebs_enabled,X86_PMC_IDX_MAX ) { event = cpuc->events[bit];if (event->hw.flags &PERF_X86_EVENT_AUTO_RELOAD )intel_pmu_save_and_restart_reload (event,0 );
@@ -2516,7 +2517,7 @@ static void intel_pmu_drain_pebs_nhm(struct pt_regs *iregs, struct perf_sample_d }if (unlikely (base >= top)) { -intel_pmu_pebs_event_update_no_drain (cpuc, size); +intel_pmu_pebs_event_update_no_drain (cpuc, mask);return ; }
@@ -2630,7 +2631,7 @@ static void intel_pmu_drain_pebs_icl(struct pt_regs *iregs, struct perf_sample_d (hybrid (cpuc->pmu, fixed_cntr_mask64) <<INTEL_PMC_IDX_FIXED );if (unlikely (base >= top)) { -intel_pmu_pebs_event_update_no_drain (cpuc,X86_PMC_IDX_MAX ); +intel_pmu_pebs_event_update_no_drain (cpuc, mask);return ; }
Signature · 1 line
2.45.2
Quoted 24 lines · Adrian Hunter
Currently, using PEBS-via-PT with a sample frequency instead of a sample period, causes a segfault. For example:
[ 103.607823] BUG: kernel NULL pointer dereference, address: 0000000000000195 [ 103.607876] <NMI> [ 103.607879] ? __die_body.cold+0x19/0x27 [ 103.607885] ? page_fault_oops+0xca/0x290 [ 103.607891] ? exc_page_fault+0x7e/0x1b0 [ 103.607897] ? asm_exc_page_fault+0x26/0x30 [ 103.607901] ? intel_pmu_pebs_event_update_no_drain+0x40/0x60 [ 103.607903] ? intel_pmu_pebs_event_update_no_drain+0x32/0x60 [ 103.607905] intel_pmu_drain_pebs_icl+0x333/0x350 [ 103.607910] handle_pmi_common+0x272/0x3c0 [ 103.607919] intel_pmu_handle_irq+0x10a/0x2e0 [ 103.607922] perf_event_nmi_handler+0x2a/0x50That happens because
intel_pmu_pebs_event_update_no_drain()assumes all thepebs_enabledbits represent counter indexes, which is not always the case. In this particular case, bits 60 and 61 are set for PEBS-via-PT purposes.The behaviour of PEBS-via-PT with sample frequency is questionable because although a PMI is generated (
PEBS_PMI_AFTER_EACH_RECORD), the period is not adjusted anyway.
Quoted 7 lines
Putting that aside, fix
intel_pmu_pebs_event_update_no_drain()by passing the mask of counter bits instead of 'size'. Note, prior to the Fixes commit, 'size' would be limited to the maximum counter index, so the issue was not hit.Fixes 722e42e45c2f1 ("perf/x86: Support counter mask")Signed-off-by Adrian Hunter <[email protected]>
Thanks for the fix.
Thanks,
Kan
Quoted 46 lines
--- arch/x86/events/intel/ds.c | 9 +++++---- 1 file changed, 5 insertions(+), 4 deletions(-)arch/x86/events/intel/ds.c +5
-4@@ -2469,8 +2469,9 @@ static void intel_pmu_drain_pebs_core(struct pt_regs *iregs, struct perf_sample_ setup_pebs_fixed_sample_data); } -static void intel_pmu_pebs_event_update_no_drain (struct cpu_hw_events *cpuc,int size) +static void intel_pmu_pebs_event_update_no_drain (struct cpu_hw_events *cpuc,u64 mask) { +u64 pebs_enabled = cpuc->pebs_enabled & mask;struct perf_event *event;int bit;@@ -2481,7 +2482,7 @@ static void intel_pmu_pebs_event_update_no_drain(struct cpu_hw_events *cpuc, int* It needs to call intel_pmu_save_and_restart_reload() to * update the event->count for this case. */ -for_each_set_bit (bit, (unsigned long *)&cpuc->pebs_enabled, size) { +for_each_set_bit (bit, (unsigned long *)&pebs_enabled,X86_PMC_IDX_MAX ) { event = cpuc->events[bit];if (event->hw.flags &PERF_X86_EVENT_AUTO_RELOAD )intel_pmu_save_and_restart_reload (event,0 );@@ -2516,7 +2517,7 @@ static void intel_pmu_drain_pebs_nhm(struct pt_regs *iregs, struct perf_sample_d }if (unlikely (base >= top)) { -intel_pmu_pebs_event_update_no_drain (cpuc, size); +intel_pmu_pebs_event_update_no_drain (cpuc, mask);return ; }@@ -2630,7 +2631,7 @@ static void intel_pmu_drain_pebs_icl(struct pt_regs *iregs, struct perf_sample_d (hybrid (cpuc->pmu, fixed_cntr_mask64) <<INTEL_PMC_IDX_FIXED );if (unlikely (base >= top)) { -intel_pmu_pebs_event_update_no_drain (cpuc,X86_PMC_IDX_MAX ); +intel_pmu_pebs_event_update_no_drain (cpuc, mask);return ; }
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: 99bcd91fabada0dbb1d5f0de44532d8008db93c6 Gitweb: https://git.kernel.org/tip/99bcd91fabada0dbb1d5f0de44532d8008db93c6 Author: Adrian Hunter <[email protected]> AuthorDate: Thu, 08 May 2025 16:44:52 +03:00 Committer: Ingo Molnar <[email protected]> CommitterDate: Thu, 15 May 2025 18:15:54 +02:00
perf/x86/intel: Fix segfault with PEBS-via-PT with sample_freq
Currently, using PEBS-via-PT with a sample frequency instead of a sample period, causes a segfault. For example:
BUG: kernel NULL pointer dereference, address: 0000000000000195
<NMI>
? __die_body.cold+0x19/0x27
? page_fault_oops+0xca/0x290
? exc_page_fault+0x7e/0x1b0
? asm_exc_page_fault+0x26/0x30
? intel_pmu_pebs_event_update_no_drain+0x40/0x60
? intel_pmu_pebs_event_update_no_drain+0x32/0x60
intel_pmu_drain_pebs_icl+0x333/0x350
handle_pmi_common+0x272/0x3c0
intel_pmu_handle_irq+0x10a/0x2e0
perf_event_nmi_handler+0x2a/0x50That happens because intel_pmu_pebs_event_update_no_drain() assumes all the pebs_enabled bits represent counter indexes, which is not always the case. In this particular case, bits 60 and 61 are set for PEBS-via-PT purposes.
The behaviour of PEBS-via-PT with sample frequency is questionable because although a PMI is generated (PEBS_PMI_AFTER_EACH_RECORD), the period is not adjusted anyway.
Putting that aside, fix intel_pmu_pebs_event_update_no_drain() by passing the mask of counter bits instead of 'size'. Note, prior to the Fixes commit, 'size' would be limited to the maximum counter index, so the issue was not hit.
--- arch/x86/events/intel/ds.c | 9 +++++---- 1 file changed, 5 insertions(+), 4 deletions(-)
arch/x86/events/intel/ds.c +5 -4
@@ -2465,8 +2465,9 @@ static void intel_pmu_drain_pebs_core(struct pt_regs *iregs, struct perf_sample_ setup_pebs_fixed_sample_data); } -static void intel_pmu_pebs_event_update_no_drain (struct cpu_hw_events *cpuc,int size) +static void intel_pmu_pebs_event_update_no_drain (struct cpu_hw_events *cpuc,u64 mask) { +u64 pebs_enabled = cpuc->pebs_enabled & mask;struct perf_event *event;int bit;
@@ -2477,7 +2478,7 @@ static void intel_pmu_pebs_event_update_no_drain(struct cpu_hw_events *cpuc, int* It needs to call intel_pmu_save_and_restart_reload() to * update the event->count for this case. */ -for_each_set_bit (bit, (unsigned long *)&cpuc->pebs_enabled, size) { +for_each_set_bit (bit, (unsigned long *)&pebs_enabled,X86_PMC_IDX_MAX ) { event = cpuc->events[bit];if (event->hw.flags &PERF_X86_EVENT_AUTO_RELOAD )intel_pmu_save_and_restart_reload (event,0 );
@@ -2512,7 +2513,7 @@ static void intel_pmu_drain_pebs_nhm(struct pt_regs *iregs, struct perf_sample_d }if (unlikely (base >= top)) { -intel_pmu_pebs_event_update_no_drain (cpuc, size); +intel_pmu_pebs_event_update_no_drain (cpuc, mask);return ; }
@@ -2626,7 +2627,7 @@ static void intel_pmu_drain_pebs_icl(struct pt_regs *iregs, struct perf_sample_d (hybrid (cpuc->pmu, fixed_cntr_mask64) <<INTEL_PMC_IDX_FIXED );if (unlikely (base >= top)) { -intel_pmu_pebs_event_update_no_drain (cpuc,X86_PMC_IDX_MAX ); +intel_pmu_pebs_event_update_no_drain (cpuc, mask);return ; }