perf/x86/intel/ds: Add PEBS format 6
Subsystems (3)
PATCH v7 1/3 perf/x86/intel/ds: Add PEBS format 6
From: Kan Liang <[email protected]>
The only difference between 5 and 6 is the new counters snapshotting group, without the following counters snapshotting enabling patches, it's impossible to utilize the feature in a PEBS record. It's safe to share the same code path with format 5.
Add format 6, so the end user can at least utilize the legacy PEBS features.
---
No changes since V6
arch/x86/events/intel/ds.c | 1 + 1 file changed, 1 insertion(+)
arch/x86/events/intel/ds.c +1 -0
@@ -2551,6 +2551,7 @@ void __init intel_ds_init(void) x86_pmu.large_pebs_flags |=PERF_SAMPLE_TIME ;break ; +case 6 :case 5 : x86_pmu.pebs_ept =1 ; fallthrough;
Signature · 1 line
2.38.1
PATCH v7 2/3 perf: Avoid the read if the count is already updated
From: "Peter Zijlstra (Intel)" <[email protected]>
The event may have been updated in the PMU-specific implementation, e.g., Intel PEBS counters snapshotting. The common code should not read and overwrite the value.
The PERF_SAMPLE_READ in the data->sample_type can be used to detect whether the PMU-specific value is available. If yes, avoid the pmu->read() in the common code. Add a new flag, skip_read, to track the case.
Factor out a perf_pmu_read() to clean up the code.
---
New patch from Peter Zijlstra.
- Use a flag to avoid the read.
include/linux/perf_event.h | 8 +++++++- kernel/events/core.c | 33 ++++++++++++++++----------------- kernel/events/ring_buffer.c | 1 + 3 files changed, 24 insertions(+), 18 deletions(-)
include/linux/perf_event.h +7 -1
@@ -1062,7 +1062,13 @@ struct perf_output_handle {struct perf_buffer *rb;unsigned long wakeup;unsigned long size; -u64 aux_flags; +union { +u64 flags;/* perf_output*() */ +u64 aux_flags;/* perf_aux_output*() */ +struct { +u64 skip_read :1 ; + }; + };union {void *addr;unsigned long head;
kernel/events/core.c +16 -17
@@ -1195,6 +1195,12 @@ static void perf_assert_pmu_disabled(struct pmu *pmu)WARN_ON_ONCE (*this_cpu_ptr (pmu->pmu_disable_count) ==0 ); } +static inline void perf_pmu_read (struct perf_event *event) +{ +if (event->state ==PERF_EVENT_STATE_ACTIVE ) + event->pmu->read (event); +} +static void get_ctx (struct perf_event_context *ctx) {refcount_inc (&ctx->refcount);
@@ -3477,8 +3483,7 @@ static void __perf_event_sync_stat(struct perf_event *event,* we know the event must be on the current CPU, therefore we * don't need to use it. */ -if (event->state ==PERF_EVENT_STATE_ACTIVE ) - event->pmu->read (event); +perf_pmu_read (event);perf_event_update_time (event);
@@ -4629,15 +4634,8 @@ static void __perf_event_read(void *info) pmu->read (event); -for_each_sibling_event (sub, event) { -if (sub->state ==PERF_EVENT_STATE_ACTIVE ) { -/* -* Use sibling's PMU rather than @event's since -* sibling could be on different (eg: software) PMU. -*/ - sub->pmu->read (sub); - } - } +for_each_sibling_event (sub, event) +perf_pmu_read (sub); data->ret = pmu->commit_txn (pmu);
@@ -7445,9 +7443,8 @@ static void perf_output_read_group(struct perf_output_handle *handle,if (read_format &PERF_FORMAT_TOTAL_TIME_RUNNING ) values[n++] = running; -if ((leader != event) && - (leader->state ==PERF_EVENT_STATE_ACTIVE )) - leader->pmu->read (leader); +if ((leader != event) && !handle->skip_read) +perf_pmu_read (leader); values[n++] =perf_event_count (leader, self);if (read_format &PERF_FORMAT_ID )
@@ -7460,9 +7457,8 @@ static void perf_output_read_group(struct perf_output_handle *handle,for_each_sibling_event (sub, leader) { n =0 ; -if ((sub != event) && - (sub->state ==PERF_EVENT_STATE_ACTIVE )) - sub->pmu->read (sub); +if ((sub != event) && !handle->skip_read) +perf_pmu_read (sub); values[n++] =perf_event_count (sub, self);if (read_format &PERF_FORMAT_ID )
@@ -7521,6 +7517,9 @@ void perf_output_sample(struct perf_output_handle *handle, {u64 sample_type = data->type; +if (data->sample_flags &PERF_SAMPLE_READ ) + handle->skip_read =1 ; +perf_output_put (handle, *header);if (sample_type &PERF_SAMPLE_IDENTIFIER )
kernel/events/ring_buffer.c +1 -0
@@ -185,6 +185,7 @@ __perf_output_begin(struct perf_output_handle *handle, handle->rb = rb; handle->event = event; + handle->flags =0 ; have_lost =local_read (&rb->lost);if (unlikely (have_lost)) {
Signature · 1 line
2.38.1
PATCH v7 3/3 perf/x86/intel: Support PEBS counters snapshotting
In reply to [email protected]From: Kan Liang <[email protected]>
The counters snapshotting is a new adaptive PEBS extension, which can capture programmable counters, fixed-function counters, and performance metrics in a PEBS record. The feature is available in the PEBS format V6.
The target counters can be configured in the new fields of MSR_PEBS_CFG. Then the PEBS HW will generate the bit mask of counters (Counters Group Header) followed by the content of all the requested counters into a PEBS record.
The current Linux perf sample read feature intends to read the counters of other member events when the leader event is overflowing. But the current read is in the NMI handler, which may has a small gap from overflow. Using the counters snapshotting feature for the sample read.
Extend intel_update_topdown_event() to accept the value from PEBS records.
Add a new PEBS_CNTR flag to indicate a sample read group that utilizes the counters snapshotting feature. When the group is scheduled, the PEBS configure can be updated accordingly.
To prevent the case that a PEBS record value might be in the past relative to what is already in the event, perf always stops the PMU and drains the PEBS buffer before updating the corresponding event->count.
---
No changes since V6
arch/x86/events/intel/core.c | 95 ++++++++++++++++---- arch/x86/events/intel/ds.c | 137 ++++++++++++++++++++++++++--- arch/x86/events/perf_event.h | 15 +++- arch/x86/events/perf_event_flags.h | 2 +- arch/x86/include/asm/perf_event.h | 15 ++++ 5 files changed, 235 insertions(+), 29 deletions(-)
arch/x86/events/intel/core.c +79 -16
@@ -2710,7 +2710,7 @@ static void update_saved_topdown_regs(struct perf_event *event, u64 slots,* modify by a NMI. PMU has to be disabled before calling this function. */ -static u64 intel_update_topdown_event (struct perf_event *event,int metric_end) +static u64 intel_update_topdown_event (struct perf_event *event,int metric_end,u64 *val) {struct cpu_hw_events *cpuc =this_cpu_ptr (&cpu_hw_events);struct perf_event *other;
@@ -2718,13 +2718,18 @@ static u64 intel_update_topdown_event(struct perf_event *event, int metric_end)bool reset = true;int idx; -/* read Fixed counter 3 */ -rdpmcl ((3 |INTEL_PMC_FIXED_RDPMC_BASE ), slots); -if (!slots) -return 0 ; +if (!val) { +/* read Fixed counter 3 */ +rdpmcl ((3 |INTEL_PMC_FIXED_RDPMC_BASE ), slots); +if (!slots) +return 0 ; -/* read PERF_METRICS */ -rdpmcl (INTEL_PMC_FIXED_RDPMC_METRICS , metrics); +/* read PERF_METRICS */ +rdpmcl (INTEL_PMC_FIXED_RDPMC_METRICS , metrics); + }else { + slots = val[0 ]; + metrics = val[1 ]; + }for_each_set_bit (idx, cpuc->active_mask, metric_end +1 ) {if (!is_topdown_idx (idx))
@@ -2767,13 +2772,14 @@ static u64 intel_update_topdown_event(struct perf_event *event, int metric_end)return slots; } -static u64 icl_update_topdown_event (struct perf_event *event) +static u64 icl_update_topdown_event (struct perf_event *event,u64 *val) {return intel_update_topdown_event (event,INTEL_PMC_IDX_METRIC_BASE + - x86_pmu.num_topdown_events -1 ); + x86_pmu.num_topdown_events -1 , + val); } -DEFINE_STATIC_CALL (intel_pmu_update_topdown_event, x86_perf_event_update); +DEFINE_STATIC_CALL (intel_pmu_update_topdown_event, intel_pmu_topdown_event_update);static void intel_pmu_read_topdown_event (struct perf_event *event) {
@@ -2785,14 +2791,15 @@ static void intel_pmu_read_topdown_event(struct perf_event *event)return ;perf_pmu_disable (event->pmu); -static_call (intel_pmu_update_topdown_event)(event); +static_call (intel_pmu_update_topdown_event)(event,NULL );perf_pmu_enable (event->pmu); }static void intel_pmu_read_event (struct perf_event *event) { -if (event->hw.flags &PERF_X86_EVENT_AUTO_RELOAD ) -intel_pmu_auto_reload_read (event); +if ((event->hw.flags &PERF_X86_EVENT_AUTO_RELOAD ) || +is_pebs_counter_event (event)) +intel_pmu_pebs_read (event);else if (is_topdown_count (event))intel_pmu_read_topdown_event (event);else
@@ -2925,7 +2932,7 @@ static int intel_pmu_set_period(struct perf_event *event)static u64 intel_pmu_update (struct perf_event *event) {if (unlikely (is_topdown_count (event))) -return static_call (intel_pmu_update_topdown_event)(event); +return static_call (intel_pmu_update_topdown_event)(event,NULL );return x86_perf_event_update (event); }
@@ -3091,7 +3098,7 @@ static int handle_pmi_common(struct pt_regs *regs, u64 status)*/ if (__test_and_clear_bit (GLOBAL_STATUS_PERF_METRICS_OVF_BIT , (unsigned long *)&status)) { handled++; -static_call (intel_pmu_update_topdown_event)(NULL ); +static_call (intel_pmu_update_topdown_event)(NULL ,NULL ); }/*
@@ -3109,6 +3116,27 @@ static int handle_pmi_common(struct pt_regs *regs, u64 status)if (!test_bit (bit, cpuc->active_mask))continue ; +/* +* There may be unprocessed PEBS records in the PEBS buffer, +* which still stores the previous values. +* Process those records first before handling the latest value. +* For example, +* A is a regular counter +* B is a PEBS event which reads A +* C is a PEBS event +* +* The following can happen: +* B-assist A=1 +* C A=2 +* B-assist A=3 +* A-overflow-PMI A=4 +* C-assist-PMI (PEBS buffer) A=5 +* +* The PEBS buffer has to be drained before handling the A-PMI +*/ +if (is_pebs_counter_event (event)) + x86_pmu.drain_pebs (regs, &data); +if (!intel_pmu_save_and_restart (event))continue ;
@@ -4056,6 +4084,23 @@ static int intel_pmu_hw_config(struct perf_event *event) event->hw.flags |=PERF_X86_EVENT_PEBS_VIA_PT ; } +if ((event->attr.sample_type &PERF_SAMPLE_READ ) && + (x86_pmu.intel_cap.pebs_format >=6 )) { +struct perf_event *leader = event->group_leader; +bool slots_leader =is_slots_event (leader); + +if (slots_leader) + leader =list_next_entry (leader, sibling_list); + +if (leader->attr.precise_ip) { + event->hw.flags |=PERF_X86_EVENT_PEBS_CNTR ; +if (slots_leader) { + leader->hw.flags |=PERF_X86_EVENT_PEBS_CNTR ; + event->group_leader->hw.flags |=PERF_X86_EVENT_PEBS_CNTR ; + } + } + } +if ((event->attr.type ==PERF_TYPE_HARDWARE ) || (event->attr.type ==PERF_TYPE_HW_CACHE ))return 0 ;
@@ -4159,6 +4204,24 @@ static int intel_pmu_hw_config(struct perf_event *event)return 0 ; } +static int intel_pmu_schedule_events (struct cpu_hw_events *cpuc,int n,int *assign) +{ +struct perf_event *event; +int ret =x86_schedule_events (cpuc, n, assign); + +if (ret) +return ret; + +if (cpuc->is_fake) +return ret; + + event = cpuc->event_list[n -1 ]; +if (event && (event->group_leader->hw.flags &PERF_X86_EVENT_PEBS_CNTR )) +intel_pmu_pebs_update_cfg (cpuc, n, assign); + +return 0 ; +} +/* * Currently, the only caller of this function is the atomic_switch_perf_msrs(). * The host perf context helps to prepare the values of the real hardware for
@@ -5292,7 +5355,7 @@ static __initconst const struct x86_pmu intel_pmu = { .set_period = intel_pmu_set_period, .update = intel_pmu_update, .hw_config = intel_pmu_hw_config, - .schedule_events = x86_schedule_events, + .schedule_events = intel_pmu_schedule_events, .eventsel = MSR_ARCH_PERFMON_EVENTSEL0, .perfctr = MSR_ARCH_PERFMON_PERFCTR0, .fixedctr = MSR_ARCH_PERFMON_FIXED_CTR0,
arch/x86/events/intel/ds.c +126 -11
184 lines not on this page · raw message
arch/x86/events/perf_event.h +14 -1
@@ -115,6 +115,11 @@ static inline bool is_branch_counters_group(struct perf_event *event)return event->group_leader->hw.flags &PERF_X86_EVENT_BRANCH_COUNTERS ; } +static inline bool is_pebs_counter_event (struct perf_event *event) +{ +return event->hw.flags &PERF_X86_EVENT_PEBS_CNTR ; +} +struct amd_nb {int nb_id;/* NorthBridge id */ int refcnt;/* reference count */
@@ -1147,6 +1152,12 @@ extern u64 __read_mostly hw_cache_extra_regsu64 x86_perf_event_update (struct perf_event *event); +static inline u64 intel_pmu_topdown_event_update (struct perf_event *event,u64 *val) +{ +return x86_perf_event_update (event); +} +DECLARE_STATIC_CALL (intel_pmu_update_topdown_event, intel_pmu_topdown_event_update); +static inline unsigned int x86_pmu_config_addr (int index) { return x86_pmu.eventsel + (x86_pmu.addr_offset ?
@@ -1642,7 +1653,9 @@ void intel_pmu_pebs_disable_all(void);void intel_pmu_pebs_sched_task (struct perf_event_pmu_context *pmu_ctx,bool sched_in); -void intel_pmu_auto_reload_read (struct perf_event *event); +void intel_pmu_pebs_update_cfg (struct cpu_hw_events *cpuc,int n,int *assign); + +void intel_pmu_pebs_read (struct perf_event *event);void intel_pmu_store_pebs_lbrs (struct lbr_entry *lbr);
arch/x86/events/perf_event_flags.h +1 -1
@@ -9,7 +9,7 @@ PERF_ARCH(PEBS_LD_HSW, 0x00008) /* haswell style datala, load */ PERF_ARCH(PEBS_NA_HSW, 0x00010) /* haswell style datala, unknown */ PERF_ARCH(EXCL, 0x00020) /* HT exclusivity on counter */ PERF_ARCH(DYNAMIC, 0x00040) /* dynamic alloc'd constraint */ - /* 0x00080 */ +PERF_ARCH(PEBS_CNTR, 0x00080) /* PEBS counters snapshot */ PERF_ARCH(EXCL_ACCT, 0x00100) /* accounted EXCL event */ PERF_ARCH(AUTO_RELOAD, 0x00200) /* use PEBS auto-reload */PERF_ARCH (LARGE_PEBS ,0x00400 )/* use large PEBS */
arch/x86/include/asm/perf_event.h +15 -0
@@ -140,6 +140,12 @@#define PEBS_DATACFG_XMMS BIT_ULL(2)#define PEBS_DATACFG_LBRS BIT_ULL(3)#define PEBS_DATACFG_LBR_SHIFT 24 +#define PEBS_DATACFG_CNTR BIT_ULL(4) +#define PEBS_DATACFG_CNTR_SHIFT 32 +#define PEBS_DATACFG_CNTR_MASK GENMASK_ULL(15, 0) +#define PEBS_DATACFG_FIX_SHIFT 48 +#define PEBS_DATACFG_FIX_MASK GENMASK_ULL(7, 0) +#define PEBS_DATACFG_METRICS BIT_ULL(5)/* Steal the highest bit of pebs_data_cfg for SW usage */ #define PEBS_UPDATE_DS_SW BIT_ULL(63)
@@ -467,6 +473,15 @@ struct pebs_xmm {#define IBS_CPUID_FEATURES 0x8000001b +struct pebs_cntr_header { +u32 cntr; +u32 fixed; +u32 metrics; +u32 reserved; +}; + +#define INTEL_CNTR_METRICS 0x3 +/* * Same bit mask as for IBS cpuid feature flags (Fn8000_001B_EAX), but * bit 0 is used to indicate the existence of IBS.
Signature · 1 line
2.38.1