perf:core: Save raw sample data
Subsystems (7)
v3 cover perf:core: Save raw sample data
Changes since v1:
- Check event->attr.sample_type & PERF_SAMPLE_RAW before calling perf_sample_save_raw_data().
- Subject has been changed to reflect the change of solution.
Changes since v2:
- Move sample_type check into perf_sample_save_raw_data().
- (New patch) Move sample_type check into perf_sample_save_callchain().
- (New patch) Move sample_type check into perf_sample_save_brstack().
Original commit message from v1:
perf/core: Trim dyn_size if raw data is absent
Yabin Cui (3):
perf/core: Save raw sample data conditionally based on sample type
perf: core: Check sample_type in perf_sample_save_callchain
perf: core: Check sample_type in perf_sample_save_brstack
arch/s390/kernel/perf_cpum_cf.c | 2 +- arch/s390/kernel/perf_pai_crypto.c | 2 +- arch/s390/kernel/perf_pai_ext.c | 2 +- arch/x86/events/amd/core.c | 3 +-- arch/x86/events/amd/ibs.c | 5 ++--- arch/x86/events/core.c | 3 +-- arch/x86/events/intel/ds.c | 9 +++----- include/linux/perf_event.h | 10 +++++++++ kernel/events/core.c | 35 +++++++++++++++--------------- kernel/trace/bpf_trace.c | 11 +++++----- 10 files changed, 44 insertions(+), 38 deletions(-)
Signature · 1 line
2.45.0.118.g7fe29c98d7-goog
PATCH v3 1/3 perf/core: Save raw sample data conditionally based on sample type
Currently, space for raw sample data is always allocated within sample records for both BPF output and tracepoint events. This leads to unused space in sample records when raw sample data is not requested.
This patch checks sample type of an event before saving raw sample data in both BPF output and tracepoint event handling logic. Raw sample data will only be saved if explicitly requested, reducing overhead when it is not needed.
--- arch/s390/kernel/perf_cpum_cf.c | 2 +- arch/s390/kernel/perf_pai_crypto.c | 2 +- arch/s390/kernel/perf_pai_ext.c | 2 +- arch/x86/events/amd/ibs.c | 2 +- include/linux/perf_event.h | 4 ++++ kernel/events/core.c | 35 +++++++++++++++--------------- kernel/trace/bpf_trace.c | 11 +++++----- 7 files changed, 32 insertions(+), 26 deletions(-)
arch/s390/kernel/perf_cpum_cf.c +1 -1
@@ -971,7 +971,7 @@ static int cfdiag_push_sample(struct perf_event *event,if (event->attr.sample_type &PERF_SAMPLE_RAW ) { raw.frag.size = cpuhw->usedss; raw.frag.data = cpuhw->stop; -perf_sample_save_raw_data (&data, &raw); +perf_sample_save_raw_data (&data, event, &raw); } overflow =perf_event_overflow (event, &data, ®s);
arch/s390/kernel/perf_pai_crypto.c +1 -1
@@ -444,7 +444,7 @@ static int paicrypt_push_sample(size_t rawsize, struct paicrypt_map *cpump,if (event->attr.sample_type &PERF_SAMPLE_RAW ) { raw.frag.size = rawsize; raw.frag.data = cpump->save; -perf_sample_save_raw_data (&data, &raw); +perf_sample_save_raw_data (&data, event, &raw); } overflow =perf_event_overflow (event, &data, ®s);
arch/s390/kernel/perf_pai_ext.c +1 -1
@@ -458,7 +458,7 @@ static int paiext_push_sample(size_t rawsize, struct paiext_map *cpump,if (event->attr.sample_type &PERF_SAMPLE_RAW ) { raw.frag.size = rawsize; raw.frag.data = cpump->save; -perf_sample_save_raw_data (&data, &raw); +perf_sample_save_raw_data (&data, event, &raw); } overflow =perf_event_overflow (event, &data, ®s);
arch/x86/events/amd/ibs.c +1 -1
@@ -1118,7 +1118,7 @@ static int perf_ibs_handle_irq(struct perf_ibs *perf_ibs, struct pt_regs *iregs) .data = ibs_data.data, }, }; -perf_sample_save_raw_data (&data, &raw); +perf_sample_save_raw_data (&data, event, &raw); }if (perf_ibs == &perf_ibs_op)
include/linux/perf_event.h +4 -0
@@ -1240,12 +1240,16 @@ static inline void perf_sample_save_callchain(struct perf_sample_data *data, }static inline void perf_sample_save_raw_data (struct perf_sample_data *data, +struct perf_event *event,struct perf_raw_record *raw) {struct perf_raw_frag *frag = &raw->frag;u32 sum =0 ;int size; +if (!(event->attr.sample_type &PERF_SAMPLE_RAW )) +return ; + do { sum += frag->size; if (perf_raw_frag_last(frag))
kernel/events/core.c +18 -17
@@ -10120,9 +10120,9 @@ static struct pmu perf_tracepoint = { };static int perf_tp_filter_match (struct perf_event *event, -struct perf_sample_data *data) +struct perf_raw_record *raw) { -void *record = data->raw->frag.data; +void *record = raw->frag.data;/* only top level events have filters set */ if (event->parent)
@@ -10134,7 +10134,7 @@ static int perf_tp_filter_match(struct perf_event *event, }static int perf_tp_event_match (struct perf_event *event, -struct perf_sample_data *data, +struct perf_raw_record *raw,struct pt_regs *regs) {if (event->hw.state &PERF_HES_STOPPED )
@@ -10145,7 +10145,7 @@ static int perf_tp_event_match(struct perf_event *event,if (event->attr.exclude_kernel && !user_mode (regs))return 0 ; -if (!perf_tp_filter_match (event, data)) +if (!perf_tp_filter_match (event, raw))return 0 ;return 1 ;
@@ -10171,6 +10171,7 @@ EXPORT_SYMBOL_GPL(perf_trace_run_bpf_submit);static void __perf_tp_event_target_task (u64 count,void *record,struct pt_regs *regs,struct perf_sample_data *data, +struct perf_raw_record *raw,struct perf_event *event) {struct trace_entry *entry = record;
@@ -10180,13 +10181,17 @@ static void __perf_tp_event_target_task(u64 count, void *record,/* Cannot deliver synchronous signal to other task. */ if (event->attr.sigtrap)return ; -if (perf_tp_event_match (event, data, regs)) +if (perf_tp_event_match (event, raw, regs)) { +perf_sample_data_init (data,0 ,0 ); +perf_sample_save_raw_data (data, event, raw);perf_swevent_event (event, count, data, regs); + } }static void perf_tp_event_target_task (u64 count,void *record,struct pt_regs *regs,struct perf_sample_data *data, +struct perf_raw_record *raw,struct perf_event_context *ctx) {unsigned int cpu =smp_processor_id ();
@@ -10194,15 +10199,15 @@ static void perf_tp_event_target_task(u64 count, void *record,struct perf_event *event, *sibling; perf_event_groups_for_cpu_pmu(event, &ctx->pinned_groups, cpu, pmu) { -__perf_tp_event_target_task (count, record, regs, data, event); +__perf_tp_event_target_task (count, record, regs, data, raw, event);for_each_sibling_event (sibling, event) -__perf_tp_event_target_task (count, record, regs, data, sibling); +__perf_tp_event_target_task (count, record, regs, data, raw, sibling); } perf_event_groups_for_cpu_pmu(event, &ctx->flexible_groups, cpu, pmu) { -__perf_tp_event_target_task (count, record, regs, data, event); +__perf_tp_event_target_task (count, record, regs, data, raw, event);for_each_sibling_event (sibling, event) -__perf_tp_event_target_task (count, record, regs, data, sibling); +__perf_tp_event_target_task (count, record, regs, data, raw, sibling); } }
@@ -10220,15 +10225,10 @@ void perf_tp_event(u16 event_type, u64 count, void *record, int entry_size, }, }; -perf_sample_data_init (&data,0 ,0 ); -perf_sample_save_raw_data (&data, &raw); -perf_trace_buf_update (record, event_type);hlist_for_each_entry_rcu (event, head, hlist_entry) { -if (perf_tp_event_match (event, &data, regs)) { -perf_swevent_event (event, count, &data, regs); - +if (perf_tp_event_match (event, &raw, regs)) {/* * Here use the same on-stack perf_sample_data, * some members in data are event-specific and
@@ -10238,7 +10238,8 @@ void perf_tp_event(u16 event_type, u64 count, void *record, int entry_size,* because data->sample_flags is set. */ perf_sample_data_init (&data,0 ,0 ); -perf_sample_save_raw_data (&data, &raw); +perf_sample_save_raw_data (&data, event, &raw); +perf_swevent_event (event, count, &data, regs); } }
@@ -10255,7 +10256,7 @@ void perf_tp_event(u16 event_type, u64 count, void *record, int entry_size, goto unlock;raw_spin_lock (&ctx->lock); -perf_tp_event_target_task (count, record, regs, &data, ctx); +perf_tp_event_target_task (count, record, regs, &data, &raw, ctx);raw_spin_unlock (&ctx->lock); unlock:rcu_read_unlock ();
kernel/trace/bpf_trace.c +6 -5
@@ -620,7 +620,8 @@ static const struct bpf_func_proto bpf_perf_event_read_value_proto = { static __always_inline u64__bpf_perf_event_output (struct pt_regs *regs,struct bpf_map *map, -u64 flags,struct perf_sample_data *sd) +u64 flags,struct perf_raw_record *raw, +struct perf_sample_data *sd) { struct bpf_array *array = container_of(map, struct bpf_array, map);unsigned int cpu =smp_processor_id ();
@@ -645,6 +646,8 @@ __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,if (unlikely (event->oncpu != cpu))return -EOPNOTSUPP ; +perf_sample_save_raw_data (sd, event, raw); +return perf_event_output (event, sd, regs); }
@@ -688,9 +691,8 @@ BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map, }perf_sample_data_init (sd,0 ,0 ); -perf_sample_save_raw_data (sd, &raw); - err =__bpf_perf_event_output (regs, map, flags, sd); + err =__bpf_perf_event_output (regs, map, flags, &raw, sd); out:this_cpu_dec (bpf_trace_nest_level);preempt_enable ();
@@ -749,9 +751,8 @@ u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,perf_fetch_caller_regs (regs);perf_sample_data_init (sd,0 ,0 ); -perf_sample_save_raw_data (sd, &raw); - ret =__bpf_perf_event_output (regs, map, flags, sd); + ret =__bpf_perf_event_output (regs, map, flags, &raw, sd); out:this_cpu_dec (bpf_event_output_nest_level);preempt_enable ();
Signature · 1 line
2.45.0.118.g7fe29c98d7-goog
PATCH v3 2/3 perf: core: Check sample_type in perf_sample_save_callchain
In reply to Yabin CuiCheck sample_type in perf_sample_save_callchain() to prevent saving callchain data when it isn't required.
--- arch/x86/events/amd/ibs.c | 3 +-- arch/x86/events/intel/ds.c | 6 ++---- include/linux/perf_event.h | 3 +++ 3 files changed, 6 insertions(+), 6 deletions(-)
arch/x86/events/amd/ibs.c +1 -2
@@ -1129,8 +1129,7 @@ static int perf_ibs_handle_irq(struct perf_ibs *perf_ibs, struct pt_regs *iregs)* recorded as part of interrupt regs. Thus we need to use rip from * interrupt regs while unwinding call stack. */ -if (event->attr.sample_type &PERF_SAMPLE_CALLCHAIN ) -perf_sample_save_callchain (&data, event, iregs); +perf_sample_save_callchain (&data, event, iregs); throttle =perf_event_overflow (event, &data, ®s); out:
arch/x86/events/intel/ds.c +2 -4
@@ -1655,8 +1655,7 @@ static void setup_pebs_fixed_sample_data(struct perf_event *event,* previous PMI context or an (I)RET happened between the record and * PMI. */ -if (sample_type &PERF_SAMPLE_CALLCHAIN ) -perf_sample_save_callchain (data, event, iregs); +perf_sample_save_callchain (data, event, iregs);/* * We use the interrupt regs as a base because the PEBS record does not
@@ -1823,8 +1822,7 @@ static void setup_pebs_adaptive_sample_data(struct perf_event *event,* previous PMI context or an (I)RET happened between the record and * PMI. */ -if (sample_type &PERF_SAMPLE_CALLCHAIN ) -perf_sample_save_callchain (data, event, iregs); +perf_sample_save_callchain (data, event, iregs); *regs = *iregs;/* The ip in basic is EventingIP */
include/linux/perf_event.h +3 -0
@@ -1232,6 +1232,9 @@ static inline void perf_sample_save_callchain(struct perf_sample_data *data, {int size =1 ; +if (!(event->attr.sample_type &PERF_SAMPLE_CALLCHAIN )) +return ; + data->callchain =perf_callchain (event, regs); size += data->callchain->nr;
Signature · 1 line
2.45.0.118.g7fe29c98d7-goog
PATCH v3 3/3 perf: core: Check sample_type in perf_sample_save_brstack
In reply to Yabin CuiCheck sample_type in perf_sample_save_brstack() to prevent saving branch stack data when it isn't required.
--- arch/x86/events/amd/core.c | 3 +-- arch/x86/events/core.c | 3 +-- arch/x86/events/intel/ds.c | 3 +-- include/linux/perf_event.h | 3 +++ 4 files changed, 6 insertions(+), 6 deletions(-)
arch/x86/events/amd/core.c +1 -2
@@ -967,8 +967,7 @@ static int amd_pmu_v2_handle_irq(struct pt_regs *regs)if (!x86_perf_event_set_period (event))continue ; -if (has_branch_stack (event)) -perf_sample_save_brstack (&data, event, &cpuc->lbr_stack,NULL ); +perf_sample_save_brstack (&data, event, &cpuc->lbr_stack,NULL );if (perf_event_overflow (event, &data, regs))x86_pmu_stop (event,0 );
arch/x86/events/core.c +1 -2
@@ -1702,8 +1702,7 @@ int x86_pmu_handle_irq(struct pt_regs *regs)perf_sample_data_init (&data,0 , event->hw.last_period); -if (has_branch_stack (event)) -perf_sample_save_brstack (&data, event, &cpuc->lbr_stack,NULL ); +perf_sample_save_brstack (&data, event, &cpuc->lbr_stack,NULL );if (perf_event_overflow (event, &data, regs))x86_pmu_stop (event,0 );
arch/x86/events/intel/ds.c +1 -2
@@ -1754,8 +1754,7 @@ static void setup_pebs_fixed_sample_data(struct perf_event *event,if (x86_pmu.intel_cap.pebs_format >=3 )setup_pebs_time (event, data, pebs->tsc); -if (has_branch_stack (event)) -perf_sample_save_brstack (data, event, &cpuc->lbr_stack,NULL ); +perf_sample_save_brstack (data, event, &cpuc->lbr_stack,NULL ); } static void adaptive_pebs_save_regs(struct pt_regs *regs,
include/linux/perf_event.h +3 -0
@@ -1276,6 +1276,9 @@ static inline void perf_sample_save_brstack(struct perf_sample_data *data, {int size =sizeof (u64);/* nr */ +if (!has_branch_stack (event)) +return ; +if (branch_sample_hw_index (event)) size +=sizeof (u64); size += brs->nr *sizeof (struct perf_branch_entry );
Signature · 1 line
2.45.0.118.g7fe29c98d7-goog