perf: Disable extended registers for non-support PMUs
Subsystems (2)
PATCH v3 1/5 perf: Disable extended registers for non-support PMUs
From: Kan Liang <[email protected]>
The perf fuzzer caused skylake machine to crash.
[ 9680.085831] Call Trace: [ 9680.088301] <IRQ> [ 9680.090363] perf_output_sample_regs+0x43/0xa0 [ 9680.094928] perf_output_sample+0x3aa/0x7a0 [ 9680.099181] perf_event_output_forward+0x53/0x80 [ 9680.103917] __perf_event_overflow+0x52/0xf0 [ 9680.108266] ? perf_trace_run_bpf_submit+0xc0/0xc0 [ 9680.113108] perf_swevent_hrtimer+0xe2/0x150 [ 9680.117475] ? check_preempt_wakeup+0x181/0x230 [ 9680.122091] ? check_preempt_curr+0x62/0x90 [ 9680.126361] ? ttwu_do_wakeup+0x19/0x140 [ 9680.130355] ? try_to_wake_up+0x54/0x460 [ 9680.134366] ? reweight_entity+0x15b/0x1a0 [ 9680.138559] ? __queue_work+0x103/0x3f0 [ 9680.142472] ? update_dl_rq_load_avg+0x1cd/0x270 [ 9680.147194] ? timerqueue_del+0x1e/0x40 [ 9680.151092] ? __remove_hrtimer+0x35/0x70 [ 9680.155191] __hrtimer_run_queues+0x100/0x280 [ 9680.159658] hrtimer_interrupt+0x100/0x220 [ 9680.163835] smp_apic_timer_interrupt+0x6a/0x140 [ 9680.168555] apic_timer_interrupt+0xf/0x20 [ 9680.172756] </IRQ>
The XMM registers can only be collected by PEBS hardware events on the platforms with PEBS baseline support, e.g. Icelake, not software/probe events.
Add capabilities flag PERF_PMU_CAP_EXTENDED_REGS to indicate the PMU which support extended registers. For X86, the extended registers are XMM registers.
Add has_extended_regs() to check if extended registers are applied.
The generic code define the mask of extended registers as 0 if arch headers haven't overridden it.
---
Changes since V2:
- Rename PERF_REG_NON_GENERIC_MASK to PERF_REG_EXTENDED_MASK
- Rename has_non_generic_regs() to has_extended_regs()
- Add capabilities flag PERF_PMU_CAP_EXTENDED_REGS
- Don't check separately. Check extended regs and flag in global perf_try_init_event()
arch/x86/events/intel/ds.c | 1 + arch/x86/include/uapi/asm/perf_regs.h | 3 +++ include/linux/perf_event.h | 1 + include/linux/perf_regs.h | 8 ++++++++ kernel/events/core.c | 18 ++++++++++++++---- 5 files changed, 27 insertions(+), 4 deletions(-)
arch/x86/events/intel/ds.c +1 -0
@@ -2020,6 +2020,7 @@ void __init intel_ds_init(void)PERF_SAMPLE_TIME ; x86_pmu.flags |=PMU_FL_PEBS_ALL ; pebs_qual ="-baseline" ; +x86_get_pmu ()->capabilities |=PERF_PMU_CAP_EXTENDED_REGS ; }else {/* Only basic record supported */ x86_pmu.pebs_no_xmm_regs =1 ;
arch/x86/include/uapi/asm/perf_regs.h +3 -0
@@ -52,4 +52,7 @@ enum perf_event_x86_regs {/* These include both GPRs and XMMX registers */ PERF_REG_X86_XMM_MAX = PERF_REG_X86_XMM15 + 2, }; + +#define PERF_REG_EXTENDED_MASK (~((1ULL << PERF_REG_X86_XMM0) - 1)) +#endif /* _ASM_X86_PERF_REGS_H */
include/linux/perf_event.h +1 -0
@@ -241,6 +241,7 @@ struct perf_event;#define PERF_PMU_CAP_NO_INTERRUPT 0x01#define PERF_PMU_CAP_NO_NMI 0x02#define PERF_PMU_CAP_AUX_NO_SG 0x04 +#define PERF_PMU_CAP_EXTENDED_REGS 0x08#define PERF_PMU_CAP_EXCLUSIVE 0x10#define PERF_PMU_CAP_ITRACE 0x20#define PERF_PMU_CAP_HETEROGENEOUS_CPUS 0x40
include/linux/perf_regs.h +8 -0
@@ -11,6 +11,11 @@ struct perf_regs {#ifdef CONFIG_HAVE_PERF_REGS #include <asm/perf_regs.h> + +#ifndef PERF_REG_EXTENDED_MASK +#define PERF_REG_EXTENDED_MASK 0 +#endif +u64 perf_reg_value (struct pt_regs *regs,int idx);int perf_reg_validate (u64 mask);u64 perf_reg_abi (struct task_struct *task);
@@ -18,6 +23,9 @@ void perf_get_regs_user(struct perf_regs *regs_user,struct pt_regs *regs, struct pt_regs *regs_user_copy);#else + +#define PERF_REG_EXTENDED_MASK 0 +static inline u64 perf_reg_value (struct pt_regs *regs,int idx) {return 0 ;
kernel/events/core.c +14 -4
@@ -10033,6 +10033,12 @@ void perf_pmu_unregister(struct pmu *pmu) }EXPORT_SYMBOL_GPL (perf_pmu_unregister); +static inline bool has_extended_regs (struct perf_event *event) +{ +return (event->attr.sample_regs_user &PERF_REG_EXTENDED_MASK ) || + (event->attr.sample_regs_intr &PERF_REG_EXTENDED_MASK ); +} +static int perf_try_init_event (struct pmu *pmu,struct perf_event *event) {struct perf_event_context *ctx =NULL ;
@@ -10064,12 +10070,16 @@ static int perf_try_init_event(struct pmu *pmu, struct perf_event *event)perf_event_ctx_unlock (event->group_leader, ctx);if (!ret) { +if (!(pmu->capabilities &PERF_PMU_CAP_EXTENDED_REGS ) && +has_extended_regs (event)) + ret = -EOPNOTSUPP ; +if (pmu->capabilities &PERF_PMU_CAP_NO_EXCLUDE && -event_has_any_exclude_flag (event)) { -if (event->destroy) - event->destroy (event); +event_has_any_exclude_flag (event)) ret = -EINVAL ; - } + +if (ret && event->destroy) + event->destroy (event); }if (ret)
Signature · 1 line
2.7.4
PATCH v3 2/5 perf/x86/regs: Check reserved bits
From: Kan Liang <[email protected]>
The perf fuzzer triggers a warning which map to:
if (WARN_ON_ONCE(idx >= ARRAY_SIZE(pt_regs_offset)))
return 0;
The bits between XMM registers and generic registers are reserved. But perf_reg_validate() doesn't check these bits.
Add PERF_REG_X86_RESERVED for reserved bits on X86. Check the reserved bits in perf_reg_validate().
---
No changes since V2
arch/x86/kernel/perf_regs.c | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-)
arch/x86/kernel/perf_regs.c +5 -2
@@ -74,6 +74,9 @@ u64 perf_reg_value(struct pt_regs *regs, int idx)return regs_get_register (regs, pt_regs_offset[idx]); } +#define PERF_REG_X86_RESERVED (((1ULL << PERF_REG_X86_XMM0) - 1) & \ + ~((1ULL << PERF_REG_X86_MAX) - 1)) +#ifdef CONFIG_X86_32 #define REG_NOSUPPORT ((1ULL << PERF_REG_X86_R8) | \ (1ULL << PERF_REG_X86_R9) | \
@@ -86,7 +89,7 @@ u64 perf_reg_value(struct pt_regs *regs, int idx)int perf_reg_validate (u64 mask) { -if (!mask || (mask ®_NOSUPPORT )) +if (!mask || (mask & (REG_NOSUPPORT |PERF_REG_X86_RESERVED )))return -EINVAL ;return 0 ;
@@ -112,7 +115,7 @@ void perf_get_regs_user(struct perf_regs *regs_user,int perf_reg_validate (u64 mask) { -if (!mask || (mask ®_NOSUPPORT )) +if (!mask || (mask & (REG_NOSUPPORT |PERF_REG_X86_RESERVED )))return -EINVAL ;return 0 ;
Signature · 1 line
2.7.4
tip-bot for Kan Liang
merge bot[tip:perf/urgent] perf/x86/regs: Check reserved bits
perf/x86/regs: Check reserved bits
Commit-ID: 90d424915ab6550826d297fd62df8ee255345b95 Gitweb: https://git.kernel.org/tip/90d424915ab6550826d297fd62df8ee255345b95 Author: Kan Liang <[email protected]> AuthorDate: Tue, 28 May 2019 15:08:31 -0700 Committer: Ingo Molnar <[email protected]> CommitDate: Mon, 24 Jun 2019 19:19:24 +0200
perf/x86/regs: Check reserved bits
The perf fuzzer triggers a warning which map to:
if (WARN_ON_ONCE(idx >= ARRAY_SIZE(pt_regs_offset)))
return 0;
The bits between XMM registers and generic registers are reserved. But perf_reg_validate() doesn't check these bits.
Add PERF_REG_X86_RESERVED for reserved bits on X86. Check the reserved bits in perf_reg_validate().
--- arch/x86/kernel/perf_regs.c | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-)
arch/x86/kernel/perf_regs.c +5 -2
@@ -74,6 +74,9 @@ u64 perf_reg_value(struct pt_regs *regs, int idx)return regs_get_register (regs, pt_regs_offset[idx]); } +#define PERF_REG_X86_RESERVED (((1ULL << PERF_REG_X86_XMM0) - 1) & \ + ~((1ULL << PERF_REG_X86_MAX) - 1)) +#ifdef CONFIG_X86_32 #define REG_NOSUPPORT ((1ULL << PERF_REG_X86_R8) | \ (1ULL << PERF_REG_X86_R9) | \
@@ -86,7 +89,7 @@ u64 perf_reg_value(struct pt_regs *regs, int idx)int perf_reg_validate (u64 mask) { -if (!mask || (mask ®_NOSUPPORT )) +if (!mask || (mask & (REG_NOSUPPORT |PERF_REG_X86_RESERVED )))return -EINVAL ;return 0 ;
@@ -112,7 +115,7 @@ void perf_get_regs_user(struct perf_regs *regs_user,int perf_reg_validate (u64 mask) { -if (!mask || (mask ®_NOSUPPORT )) +if (!mask || (mask & (REG_NOSUPPORT |PERF_REG_X86_RESERVED )))return -EINVAL ;return 0 ;
PATCH v3 3/5 perf/x86: Clean up PEBS_XMM_REGS
In reply to [email protected]From: Kan Liang <[email protected]>
Use generic macro PERF_REG_EXTENDED_MASK to replace PEBS_XMM_REGS to avoid duplication.
---
New for V3
arch/x86/events/core.c | 4 ++-- arch/x86/events/intel/ds.c | 2 +- arch/x86/events/perf_event.h | 18 ------------------ 3 files changed, 3 insertions(+), 21 deletions(-)
arch/x86/events/core.c +2 -2
@@ -561,13 +561,13 @@ int x86_pmu_hw_config(struct perf_event *event) }/* sample_regs_user never support XMM registers */ -if (unlikely (event->attr.sample_regs_user &PEBS_XMM_REGS )) +if (unlikely (event->attr.sample_regs_user &PERF_REG_EXTENDED_MASK ))return -EINVAL ;/* * Besides the general purpose registers, XMM registers may * be collected in PEBS on some platforms, e.g. Icelake */ -if (unlikely (event->attr.sample_regs_intr &PEBS_XMM_REGS )) { +if (unlikely (event->attr.sample_regs_intr &PERF_REG_EXTENDED_MASK )) {if (x86_pmu.pebs_no_xmm_regs)return -EINVAL ;
arch/x86/events/intel/ds.c +1 -1
@@ -987,7 +987,7 @@ static u64 pebs_update_adaptive_cfg(struct perf_event *event) pebs_data_cfg |=PEBS_DATACFG_GP ;if ((sample_type &PERF_SAMPLE_REGS_INTR ) && - (attr->sample_regs_intr &PEBS_XMM_REGS )) + (attr->sample_regs_intr &PERF_REG_EXTENDED_MASK )) pebs_data_cfg |=PEBS_DATACFG_XMMS ;if (sample_type &PERF_SAMPLE_BRANCH_STACK ) {
arch/x86/events/perf_event.h +0 -18
@@ -121,24 +121,6 @@ struct amd_nb { (1ULL <<PERF_REG_X86_R14 ) | \ (1ULL << PERF_REG_X86_R15)) -#define PEBS_XMM_REGS \ - ((1ULL << PERF_REG_X86_XMM0) | \ - (1ULL << PERF_REG_X86_XMM1) | \ - (1ULL << PERF_REG_X86_XMM2) | \ - (1ULL << PERF_REG_X86_XMM3) | \ - (1ULL << PERF_REG_X86_XMM4) | \ - (1ULL << PERF_REG_X86_XMM5) | \ - (1ULL << PERF_REG_X86_XMM6) | \ - (1ULL << PERF_REG_X86_XMM7) | \ - (1ULL << PERF_REG_X86_XMM8) | \ - (1ULL << PERF_REG_X86_XMM9) | \ - (1ULL << PERF_REG_X86_XMM10) | \ - (1ULL << PERF_REG_X86_XMM11) | \ - (1ULL << PERF_REG_X86_XMM12) | \ - (1ULL << PERF_REG_X86_XMM13) | \ - (1ULL << PERF_REG_X86_XMM14) | \ - (1ULL << PERF_REG_X86_XMM15)) -/* * Per register state. */
Signature · 1 line
2.7.4
tip-bot for Kan Liang
merge bot[tip:perf/urgent] perf/x86: Clean up PEBS_XMM_REGS
perf/x86: Clean up PEBS_XMM_REGS
Commit-ID: dce86ac75d772047e9bc606154704aa73bfd4c83 Gitweb: https://git.kernel.org/tip/dce86ac75d772047e9bc606154704aa73bfd4c83 Author: Kan Liang <[email protected]> AuthorDate: Tue, 28 May 2019 15:08:32 -0700 Committer: Ingo Molnar <[email protected]> CommitDate: Mon, 24 Jun 2019 19:19:24 +0200
perf/x86: Clean up PEBS_XMM_REGS
Use generic macro PERF_REG_EXTENDED_MASK to replace PEBS_XMM_REGS to avoid duplication.
--- arch/x86/events/core.c | 4 ++-- arch/x86/events/intel/ds.c | 2 +- arch/x86/events/perf_event.h | 18 ------------------ 3 files changed, 3 insertions(+), 21 deletions(-)
arch/x86/events/core.c +2 -2
@@ -561,13 +561,13 @@ int x86_pmu_hw_config(struct perf_event *event) }/* sample_regs_user never support XMM registers */ -if (unlikely (event->attr.sample_regs_user &PEBS_XMM_REGS )) +if (unlikely (event->attr.sample_regs_user &PERF_REG_EXTENDED_MASK ))return -EINVAL ;/* * Besides the general purpose registers, XMM registers may * be collected in PEBS on some platforms, e.g. Icelake */ -if (unlikely (event->attr.sample_regs_intr &PEBS_XMM_REGS )) { +if (unlikely (event->attr.sample_regs_intr &PERF_REG_EXTENDED_MASK )) {if (x86_pmu.pebs_no_xmm_regs)return -EINVAL ;
arch/x86/events/intel/ds.c +1 -1
@@ -987,7 +987,7 @@ static u64 pebs_update_adaptive_cfg(struct perf_event *event) pebs_data_cfg |=PEBS_DATACFG_GP ;if ((sample_type &PERF_SAMPLE_REGS_INTR ) && - (attr->sample_regs_intr &PEBS_XMM_REGS )) + (attr->sample_regs_intr &PERF_REG_EXTENDED_MASK )) pebs_data_cfg |=PEBS_DATACFG_XMMS ;if (sample_type &PERF_SAMPLE_BRANCH_STACK ) {
arch/x86/events/perf_event.h +0 -18
@@ -121,24 +121,6 @@ struct amd_nb { (1ULL <<PERF_REG_X86_R14 ) | \ (1ULL << PERF_REG_X86_R15)) -#define PEBS_XMM_REGS \ - ((1ULL << PERF_REG_X86_XMM0) | \ - (1ULL << PERF_REG_X86_XMM1) | \ - (1ULL << PERF_REG_X86_XMM2) | \ - (1ULL << PERF_REG_X86_XMM3) | \ - (1ULL << PERF_REG_X86_XMM4) | \ - (1ULL << PERF_REG_X86_XMM5) | \ - (1ULL << PERF_REG_X86_XMM6) | \ - (1ULL << PERF_REG_X86_XMM7) | \ - (1ULL << PERF_REG_X86_XMM8) | \ - (1ULL << PERF_REG_X86_XMM9) | \ - (1ULL << PERF_REG_X86_XMM10) | \ - (1ULL << PERF_REG_X86_XMM11) | \ - (1ULL << PERF_REG_X86_XMM12) | \ - (1ULL << PERF_REG_X86_XMM13) | \ - (1ULL << PERF_REG_X86_XMM14) | \ - (1ULL << PERF_REG_X86_XMM15)) -/* * Per register state. */
PATCH v3 4/5 perf/x86: Clean up pebs_no_xmm_regs
In reply to [email protected]From: Kan Liang <[email protected]>
Don't need pebs_no_xmm_regs anymore. The capabilities PERF_PMU_CAP_EXTENDED_REGS can be used to check if XMM registers collection is supported.
---
New for V3
arch/x86/events/core.c | 2 +- arch/x86/events/intel/ds.c | 6 ++---- arch/x86/events/perf_event.h | 3 +-- 3 files changed, 4 insertions(+), 7 deletions(-)
arch/x86/events/core.c +1 -1
@@ -568,7 +568,7 @@ int x86_pmu_hw_config(struct perf_event *event)* be collected in PEBS on some platforms, e.g. Icelake */ if (unlikely (event->attr.sample_regs_intr &PERF_REG_EXTENDED_MASK )) { -if (x86_pmu.pebs_no_xmm_regs) +if (!(event->pmu->capabilities &PERF_PMU_CAP_EXTENDED_REGS ))return -EINVAL ;if (!event->attr.precise_ip)
arch/x86/events/intel/ds.c +2 -4
@@ -1964,10 +1964,9 @@ void __init intel_ds_init(void) x86_pmu.bts =boot_cpu_has (X86_FEATURE_BTS ); x86_pmu.pebs =boot_cpu_has (X86_FEATURE_PEBS ); x86_pmu.pebs_buffer_size =PEBS_BUFFER_SIZE ; -if (x86_pmu.version <=4 ) { +if (x86_pmu.version <=4 ) x86_pmu.pebs_no_isolation =1 ; - x86_pmu.pebs_no_xmm_regs =1 ; - } +if (x86_pmu.pebs) {char pebs_type = x86_pmu.intel_cap.pebs_trap ?'+' :'-' ;char *pebs_qual ="" ;
@@ -2023,7 +2022,6 @@ void __init intel_ds_init(void)x86_get_pmu ()->capabilities |=PERF_PMU_CAP_EXTENDED_REGS ; }else {/* Only basic record supported */ - x86_pmu.pebs_no_xmm_regs =1 ; x86_pmu.large_pebs_flags &= ~(PERF_SAMPLE_ADDR | PERF_SAMPLE_TIME |
arch/x86/events/perf_event.h +1 -2
@@ -650,8 +650,7 @@ struct x86_pmu { pebs_broken :1, pebs_prec_dist :1, pebs_no_tlb :1, - pebs_no_isolation :1, - pebs_no_xmm_regs :1; + pebs_no_isolation :1;int pebs_record_size;int pebs_buffer_size;int max_pebs_events;
Signature · 1 line
2.7.4
tip-bot for Kan Liang
merge bot[tip:perf/urgent] perf/x86: Remove pmu->pebs_no_xmm_regs
perf/x86: Remove pmu->pebs_no_xmm_regs
Commit-ID: cd6b984f6d8cd615755b5404a51b7efe45215f28 Gitweb: https://git.kernel.org/tip/cd6b984f6d8cd615755b5404a51b7efe45215f28 Author: Kan Liang <[email protected]> AuthorDate: Tue, 28 May 2019 15:08:33 -0700 Committer: Ingo Molnar <[email protected]> CommitDate: Mon, 24 Jun 2019 19:19:25 +0200
perf/x86: Remove pmu->pebs_no_xmm_regs
We don't need pmu->pebs_no_xmm_regs anymore, the capabilities PERF_PMU_CAP_EXTENDED_REGS can be used to check if XMM registers collection is supported.
--- arch/x86/events/core.c | 2 +- arch/x86/events/intel/ds.c | 6 ++---- arch/x86/events/perf_event.h | 3 +-- 3 files changed, 4 insertions(+), 7 deletions(-)
arch/x86/events/core.c +1 -1
@@ -568,7 +568,7 @@ int x86_pmu_hw_config(struct perf_event *event)* be collected in PEBS on some platforms, e.g. Icelake */ if (unlikely (event->attr.sample_regs_intr &PERF_REG_EXTENDED_MASK )) { -if (x86_pmu.pebs_no_xmm_regs) +if (!(event->pmu->capabilities &PERF_PMU_CAP_EXTENDED_REGS ))return -EINVAL ;if (!event->attr.precise_ip)
arch/x86/events/intel/ds.c +2 -4
@@ -1964,10 +1964,9 @@ void __init intel_ds_init(void) x86_pmu.bts =boot_cpu_has (X86_FEATURE_BTS ); x86_pmu.pebs =boot_cpu_has (X86_FEATURE_PEBS ); x86_pmu.pebs_buffer_size =PEBS_BUFFER_SIZE ; -if (x86_pmu.version <=4 ) { +if (x86_pmu.version <=4 ) x86_pmu.pebs_no_isolation =1 ; - x86_pmu.pebs_no_xmm_regs =1 ; - } +if (x86_pmu.pebs) {char pebs_type = x86_pmu.intel_cap.pebs_trap ?'+' :'-' ;char *pebs_qual ="" ;
@@ -2023,7 +2022,6 @@ void __init intel_ds_init(void)x86_get_pmu ()->capabilities |=PERF_PMU_CAP_EXTENDED_REGS ; }else {/* Only basic record supported */ - x86_pmu.pebs_no_xmm_regs =1 ; x86_pmu.large_pebs_flags &= ~(PERF_SAMPLE_ADDR | PERF_SAMPLE_TIME |
arch/x86/events/perf_event.h +1 -2
@@ -650,8 +650,7 @@ struct x86_pmu { pebs_broken :1, pebs_prec_dist :1, pebs_no_tlb :1, - pebs_no_isolation :1, - pebs_no_xmm_regs :1; + pebs_no_isolation :1;int pebs_record_size;int pebs_buffer_size;int max_pebs_events;
PATCH v3 5/5 perf regs x86: Use PERF_REG_EXTENDED_MASK
In reply to [email protected]From: Kan Liang <[email protected]>
Use the macro defined in kernel ABI header to replace the local name.
No functional change.
---
Changes since V2:
- Rename PERF_REG_NON_GENERIC_MASK to PERF_REG_EXTENDED_MASK
tools/arch/x86/include/uapi/asm/perf_regs.h | 3 +++ tools/perf/arch/x86/include/perf_regs.h | 1 - tools/perf/arch/x86/util/perf_regs.c | 4 ++-- 3 files changed, 5 insertions(+), 3 deletions(-)
tools/arch/x86/include/uapi/asm/perf_regs.h +3 -0
@@ -52,4 +52,7 @@ enum perf_event_x86_regs {/* These include both GPRs and XMMX registers */ PERF_REG_X86_XMM_MAX = PERF_REG_X86_XMM15 + 2, }; + +#define PERF_REG_EXTENDED_MASK (~((1ULL << PERF_REG_X86_XMM0) - 1)) +#endif /* _ASM_X86_PERF_REGS_H */
tools/perf/arch/x86/include/perf_regs.h +0 -1
@@ -9,7 +9,6 @@void perf_regs_load (u64 *regs);#define PERF_REGS_MAX PERF_REG_X86_XMM_MAX -#define PERF_XMM_REGS_MASK (~((1ULL << PERF_REG_X86_XMM0) - 1))#ifndef HAVE_ARCH_X86_64_SUPPORT #define PERF_REGS_MASK ((1ULL << PERF_REG_X86_32_MAX) - 1)#define PERF_SAMPLE_REGS_ABI PERF_SAMPLE_REGS_ABI_32
tools/perf/arch/x86/util/perf_regs.c +2 -2
@@ -277,7 +277,7 @@ uint64_t arch__intr_reg_mask(void) .type = PERF_TYPE_HARDWARE, .config = PERF_COUNT_HW_CPU_CYCLES, .sample_type = PERF_SAMPLE_REGS_INTR, - .sample_regs_intr = PERF_XMM_REGS_MASK, + .sample_regs_intr = PERF_REG_EXTENDED_MASK, .precise_ip = 1, .disabled = 1, .exclude_kernel = 1,
@@ -293,7 +293,7 @@ uint64_t arch__intr_reg_mask(void) fd =sys_perf_event_open (&attr,0 ,-1 ,-1 ,0 );if (fd !=-1 ) {close (fd); -return (PERF_XMM_REGS_MASK |PERF_REGS_MASK ); +return (PERF_REG_EXTENDED_MASK |PERF_REGS_MASK ); }return PERF_REGS_MASK ;
Signature · 1 line
2.7.4
tip-bot for Kan Liang
merge bot[tip:perf/urgent] perf/x86/regs: Use PERF_REG_EXTENDED_MASK
perf/x86/regs: Use PERF_REG_EXTENDED_MASK
Commit-ID: 8b12b812f5367c2469fb937da7e28dd321ad8d7b Gitweb: https://git.kernel.org/tip/8b12b812f5367c2469fb937da7e28dd321ad8d7b Author: Kan Liang <[email protected]> AuthorDate: Tue, 28 May 2019 15:08:34 -0700 Committer: Ingo Molnar <[email protected]> CommitDate: Mon, 24 Jun 2019 19:19:26 +0200
perf/x86/regs: Use PERF_REG_EXTENDED_MASK
Use the macro defined in kernel ABI header to replace the local name.
No functional change.
--- tools/arch/x86/include/uapi/asm/perf_regs.h | 3 +++ tools/perf/arch/x86/include/perf_regs.h | 1 - tools/perf/arch/x86/util/perf_regs.c | 4 ++-- 3 files changed, 5 insertions(+), 3 deletions(-)
tools/arch/x86/include/uapi/asm/perf_regs.h +3 -0
@@ -52,4 +52,7 @@ enum perf_event_x86_regs {/* These include both GPRs and XMMX registers */ PERF_REG_X86_XMM_MAX = PERF_REG_X86_XMM15 + 2, }; + +#define PERF_REG_EXTENDED_MASK (~((1ULL << PERF_REG_X86_XMM0) - 1)) +#endif /* _ASM_X86_PERF_REGS_H */
tools/perf/arch/x86/include/perf_regs.h +0 -1
@@ -9,7 +9,6 @@void perf_regs_load (u64 *regs);#define PERF_REGS_MAX PERF_REG_X86_XMM_MAX -#define PERF_XMM_REGS_MASK (~((1ULL << PERF_REG_X86_XMM0) - 1))#ifndef HAVE_ARCH_X86_64_SUPPORT #define PERF_REGS_MASK ((1ULL << PERF_REG_X86_32_MAX) - 1)#define PERF_SAMPLE_REGS_ABI PERF_SAMPLE_REGS_ABI_32
tools/perf/arch/x86/util/perf_regs.c +2 -2
@@ -277,7 +277,7 @@ uint64_t arch__intr_reg_mask(void) .type = PERF_TYPE_HARDWARE, .config = PERF_COUNT_HW_CPU_CYCLES, .sample_type = PERF_SAMPLE_REGS_INTR, - .sample_regs_intr = PERF_XMM_REGS_MASK, + .sample_regs_intr = PERF_REG_EXTENDED_MASK, .precise_ip = 1, .disabled = 1, .exclude_kernel = 1,
@@ -293,7 +293,7 @@ uint64_t arch__intr_reg_mask(void) fd =sys_perf_event_open (&attr,0 ,-1 ,-1 ,0 );if (fd !=-1 ) {close (fd); -return (PERF_XMM_REGS_MASK |PERF_REGS_MASK ); +return (PERF_REG_EXTENDED_MASK |PERF_REGS_MASK ); }return PERF_REGS_MASK ;
Peter Zijlstra
maintainerQuoted 1 line · On Tue, May 28, 2019 at 03:08:30PM -0700, [email protected] wrote:
From: Kan Liang <[email protected]>
Ok, have them now.
tip-bot for Kan Liang
merge bot[tip:perf/urgent] perf/x86: Disable extended registers for non-supported PMUs
In reply to [email protected]perf/x86: Disable extended registers for non-supported PMUs
Commit-ID: e321d02db87af7840da29ef833a2a71fc0eab198 Gitweb: https://git.kernel.org/tip/e321d02db87af7840da29ef833a2a71fc0eab198 Author: Kan Liang <[email protected]> AuthorDate: Tue, 28 May 2019 15:08:30 -0700 Committer: Ingo Molnar <[email protected]> CommitDate: Mon, 24 Jun 2019 19:19:23 +0200
perf/x86: Disable extended registers for non-supported PMUs
The perf fuzzer caused Skylake machine to crash:
[ 9680.085831] Call Trace: [ 9680.088301] <IRQ> [ 9680.090363] perf_output_sample_regs+0x43/0xa0 [ 9680.094928] perf_output_sample+0x3aa/0x7a0 [ 9680.099181] perf_event_output_forward+0x53/0x80 [ 9680.103917] __perf_event_overflow+0x52/0xf0 [ 9680.108266] ? perf_trace_run_bpf_submit+0xc0/0xc0 [ 9680.113108] perf_swevent_hrtimer+0xe2/0x150 [ 9680.117475] ? check_preempt_wakeup+0x181/0x230 [ 9680.122091] ? check_preempt_curr+0x62/0x90 [ 9680.126361] ? ttwu_do_wakeup+0x19/0x140 [ 9680.130355] ? try_to_wake_up+0x54/0x460 [ 9680.134366] ? reweight_entity+0x15b/0x1a0 [ 9680.138559] ? __queue_work+0x103/0x3f0 [ 9680.142472] ? update_dl_rq_load_avg+0x1cd/0x270 [ 9680.147194] ? timerqueue_del+0x1e/0x40 [ 9680.151092] ? __remove_hrtimer+0x35/0x70 [ 9680.155191] __hrtimer_run_queues+0x100/0x280 [ 9680.159658] hrtimer_interrupt+0x100/0x220 [ 9680.163835] smp_apic_timer_interrupt+0x6a/0x140 [ 9680.168555] apic_timer_interrupt+0xf/0x20 [ 9680.172756] </IRQ>
The XMM registers can only be collected by PEBS hardware events on the platforms with PEBS baseline support, e.g. Icelake, not software/probe events.
Add capabilities flag PERF_PMU_CAP_EXTENDED_REGS to indicate the PMU which support extended registers. For X86, the extended registers are XMM registers.
Add has_extended_regs() to check if extended registers are applied.
The generic code define the mask of extended registers as 0 if arch headers haven't overridden it.
--- arch/x86/events/intel/ds.c | 1 + arch/x86/include/uapi/asm/perf_regs.h | 3 +++ include/linux/perf_event.h | 1 + include/linux/perf_regs.h | 8 ++++++++ kernel/events/core.c | 18 ++++++++++++++---- 5 files changed, 27 insertions(+), 4 deletions(-)
arch/x86/events/intel/ds.c +1 -0
@@ -2020,6 +2020,7 @@ void __init intel_ds_init(void)PERF_SAMPLE_TIME ; x86_pmu.flags |=PMU_FL_PEBS_ALL ; pebs_qual ="-baseline" ; +x86_get_pmu ()->capabilities |=PERF_PMU_CAP_EXTENDED_REGS ; }else {/* Only basic record supported */ x86_pmu.pebs_no_xmm_regs =1 ;
arch/x86/include/uapi/asm/perf_regs.h +3 -0
@@ -52,4 +52,7 @@ enum perf_event_x86_regs {/* These include both GPRs and XMMX registers */ PERF_REG_X86_XMM_MAX = PERF_REG_X86_XMM15 + 2, }; + +#define PERF_REG_EXTENDED_MASK (~((1ULL << PERF_REG_X86_XMM0) - 1)) +#endif /* _ASM_X86_PERF_REGS_H */
include/linux/perf_event.h +1 -0
@@ -241,6 +241,7 @@ struct perf_event;#define PERF_PMU_CAP_NO_INTERRUPT 0x01#define PERF_PMU_CAP_NO_NMI 0x02#define PERF_PMU_CAP_AUX_NO_SG 0x04 +#define PERF_PMU_CAP_EXTENDED_REGS 0x08#define PERF_PMU_CAP_EXCLUSIVE 0x10#define PERF_PMU_CAP_ITRACE 0x20#define PERF_PMU_CAP_HETEROGENEOUS_CPUS 0x40
include/linux/perf_regs.h +8 -0
@@ -11,6 +11,11 @@ struct perf_regs {#ifdef CONFIG_HAVE_PERF_REGS #include <asm/perf_regs.h> + +#ifndef PERF_REG_EXTENDED_MASK +#define PERF_REG_EXTENDED_MASK 0 +#endif +u64 perf_reg_value (struct pt_regs *regs,int idx);int perf_reg_validate (u64 mask);u64 perf_reg_abi (struct task_struct *task);
@@ -18,6 +23,9 @@ void perf_get_regs_user(struct perf_regs *regs_user,struct pt_regs *regs, struct pt_regs *regs_user_copy);#else + +#define PERF_REG_EXTENDED_MASK 0 +static inline u64 perf_reg_value (struct pt_regs *regs,int idx) {return 0 ;
kernel/events/core.c +14 -4
@@ -10036,6 +10036,12 @@ void perf_pmu_unregister(struct pmu *pmu) }EXPORT_SYMBOL_GPL (perf_pmu_unregister); +static inline bool has_extended_regs (struct perf_event *event) +{ +return (event->attr.sample_regs_user &PERF_REG_EXTENDED_MASK ) || + (event->attr.sample_regs_intr &PERF_REG_EXTENDED_MASK ); +} +static int perf_try_init_event (struct pmu *pmu,struct perf_event *event) {struct perf_event_context *ctx =NULL ;
@@ -10067,12 +10073,16 @@ static int perf_try_init_event(struct pmu *pmu, struct perf_event *event)perf_event_ctx_unlock (event->group_leader, ctx);if (!ret) { +if (!(pmu->capabilities &PERF_PMU_CAP_EXTENDED_REGS ) && +has_extended_regs (event)) + ret = -EOPNOTSUPP ; +if (pmu->capabilities &PERF_PMU_CAP_NO_EXCLUDE && -event_has_any_exclude_flag (event)) { -if (event->destroy) - event->destroy (event); +event_has_any_exclude_flag (event)) ret = -EINVAL ; - } + +if (ret && event->destroy) + event->destroy (event); }if (ret)