perf sample: Make user_regs and intr_regs optional
Ian Rogers
reviewerPATCH 1/1 perf sample: Make user_regs and intr_regs optional
The struct dump_regs contains 512 bytes of cache_regs, meaning the two values in perf_sample contribute 1088 bytes of its total 1384 bytes size. Initializing this much memory has a cost reported by Tavian Barnes <[email protected]> as about 2.5% when running perf script --itrace=i0:
https://lore.kernel.org/lkml/d841b97b3ad2ca8bcab07e4293375fb7c32dfce7.1736618095.git.tavianator@tavianator.com/
Adrian Hunter <[email protected]> replied that the zero initialization was necessary and couldn't simply be removed.
This patch aims to strike a middle ground of still zeroing the perf_sample, but removing 79% of its size by make user_regs and intr_regs optional pointers to zalloc-ed memory. To support the allocation accessors are created for user_regs and intr_regs. To support correct cleanup perf_sample__init and perf_sample__exit functions are created and added throughout the code base.
---
It is likely the size of this patch will be complained about. The majority of the size comes from the addition of calls to init/exit/accessor (and avoiding early returns) which don't need to exist prior to this patch. Whilst adding init/exit/accessor could be done separately the difference in file size would be minimal and the separation of one problem across multiple patches was considered worse than having 1 bigger patch.
--- tools/perf/arch/x86/tests/dwarf-unwind.c | 2 +- tools/perf/arch/x86/util/unwind-libdw.c | 2 +- tools/perf/builtin-record.c | 4 +- tools/perf/builtin-script.c | 10 +- tools/perf/builtin-top.c | 8 +- tools/perf/builtin-trace.c | 5 +- tools/perf/tests/code-reading.c | 12 +- tools/perf/tests/dwarf-unwind.c | 6 +- tools/perf/tests/mmap-basic.c | 3 + tools/perf/tests/openat-syscall-tp-fields.c | 4 +- tools/perf/tests/parse-no-sample-id-all.c | 6 +- tools/perf/tests/perf-record.c | 2 + tools/perf/tests/perf-time-to-tsc.c | 2 + tools/perf/tests/sample-parsing.c | 62 ++++---- tools/perf/tests/sw-clock.c | 3 + tools/perf/tests/switch-tracking.c | 14 +- tools/perf/util/Build | 1 + tools/perf/util/arm-spe.c | 24 +++- .../util/arm64-frame-pointer-unwind-support.c | 29 ++-- tools/perf/util/auxtrace.c | 15 +- tools/perf/util/cs-etm.c | 31 ++-- tools/perf/util/evsel.c | 21 +-- tools/perf/util/intel-bts.c | 4 +- tools/perf/util/intel-pt.c | 136 ++++++++++++------ tools/perf/util/jitdump.c | 10 +- tools/perf/util/machine.c | 4 +- tools/perf/util/python.c | 9 ++ tools/perf/util/s390-cpumsf.c | 6 +- tools/perf/util/sample.c | 43 ++++++ tools/perf/util/sample.h | 9 +- .../scripting-engines/trace-event-python.c | 29 ++-- tools/perf/util/session.c | 94 ++++++++---- tools/perf/util/synthetic-events.c | 24 ++-- tools/perf/util/unwind-libdw.c | 9 +- 34 files changed, 450 insertions(+), 193 deletions(-) create mode 100644 tools/perf/util/sample.c
tools/perf/arch/x86/tests/dwarf-unwind.c +1 -1
@@ -53,7 +53,7 @@ static int sample_ustack(struct perf_sample *sample,int test__arch_unwind_sample (struct perf_sample *sample,struct thread *thread) { -struct regs_dump *regs = &sample->user_regs; +struct regs_dump *regs =perf_sample__user_regs (sample);u64 *buf; buf =malloc (sizeof (u64) *PERF_REGS_MAX );
tools/perf/arch/x86/util/unwind-libdw.c +1 -1
@@ -8,7 +8,7 @@bool libdw__arch_set_initial_registers (Dwfl_Thread *thread,void *arg) {struct unwind_info *ui = arg; -struct regs_dump *user_regs = &ui->sample->user_regs; +struct regs_dump *user_regs =perf_sample__user_regs (ui->sample);Dwarf_Word dwarf_regs[17 ];unsigned nregs;
tools/perf/builtin-record.c +3 -1
@@ -1917,9 +1917,10 @@ static void __record__save_lost_samples(struct record *rec, struct evsel *evsel, u16 misc_flag) {struct perf_sample_id *sid; -struct perf_sample sample = {}; +struct perf_sample sample;int id_hdr_size; +perf_sample__init (&sample,/*all=*/ true); lost->lost = lost_count;if (evsel->core.ids) { sid =xyarray__entry (evsel->core.sample_id, cpu_idx, thread_idx);
@@ -1931,6 +1932,7 @@ static void __record__save_lost_samples(struct record *rec, struct evsel *evsel, lost->header.size =sizeof (*lost) + id_hdr_size; lost->header.misc = misc_flag;record__write (rec,NULL , lost, lost->header.size); +perf_sample__exit (&sample); }static void record__read_lost_samples (struct record *rec)
tools/perf/builtin-script.c +8 -2
@@ -783,14 +783,20 @@ tod_scnprintf(struct perf_script *script, char *buf, int buflen,static int perf_sample__fprintf_iregs (struct perf_sample *sample,struct perf_event_attr *attr,const char *arch,FILE *fp) { -return perf_sample__fprintf_regs (&sample->intr_regs, +if (!sample->intr_regs) +return 0 ; + +return perf_sample__fprintf_regs (perf_sample__intr_regs (sample), attr->sample_regs_intr, arch, fp); }static int perf_sample__fprintf_uregs (struct perf_sample *sample,struct perf_event_attr *attr,const char *arch,FILE *fp) { -return perf_sample__fprintf_regs (&sample->user_regs, +if (!sample->user_regs) +return 0 ; + +return perf_sample__fprintf_regs (perf_sample__user_regs (sample), attr->sample_regs_user, arch, fp); }
tools/perf/builtin-top.c +6 -2
@@ -1157,6 +1157,7 @@ static int deliver_event(struct ordered_events *qe,return 0 ; } +perf_sample__init (&sample,/*all=*/ false); ret =evlist__parse_sample (evlist, event, &sample);if (ret) {pr_err ("Can't parse sample, err = %d\n" , ret);
@@ -1167,8 +1168,10 @@ static int deliver_event(struct ordered_events *qe,assert (evsel !=NULL );if (event->header.type ==PERF_RECORD_SAMPLE ) { -if (evswitch__discard (&top->evswitch, evsel)) -return 0 ; +if (evswitch__discard (&top->evswitch, evsel)) { + ret =0 ; + goto next_event; + } ++top->samples; }
@@ -1219,6 +1222,7 @@ static int deliver_event(struct ordered_events *qe, ret =0 ; next_event: +perf_sample__exit (&sample);return ret; }
tools/perf/builtin-trace.c +4 -1
@@ -4015,13 +4015,16 @@ static int __trace__deliver_event(struct trace *trace, union perf_event *event) {struct evlist *evlist = trace->evlist;struct perf_sample sample; -int err =evlist__parse_sample (evlist, event, &sample); +int err; +perf_sample__init (&sample,/*all=*/ false); + err =evlist__parse_sample (evlist, event, &sample);if (err)fprintf (trace->output,"Can't parse sample, err = %d, skipping...\n" , err);else trace__handle_event (trace, event, &sample); +perf_sample__exit (&sample);return 0 ; }
tools/perf/tests/code-reading.c +9 -3
@@ -479,19 +479,25 @@ static int process_sample_event(struct machine *machine,struct thread *thread;int ret; -if (evlist__parse_sample (evlist, event, &sample)) { +perf_sample__init (&sample,/*all=*/ false); + ret =evlist__parse_sample (evlist, event, &sample); +if (ret) {pr_debug ("evlist__parse_sample failed\n" ); -return -1 ; + ret =-1 ; + goto out; } thread =machine__findnew_thread (machine, sample.pid, sample.tid);if (!thread) {pr_debug ("machine__findnew_thread failed\n" ); -return -1 ; + ret =-1 ; + goto out; } ret =read_object_code (sample.ip,READLEN , sample.cpumode, thread, state);thread__put (thread); +out: +perf_sample__exit (&sample);return ret; }
tools/perf/tests/dwarf-unwind.c +3 -3
@@ -115,8 +115,7 @@ NO_TAIL_CALL_ATTRIBUTE noinline int test_dwarf_unwind__thread(struct thread *thrunsigned long cnt =0 ;int err =-1 ; -memset (&sample,0 ,sizeof (sample)); - +perf_sample__init (&sample,/*all=*/ true);if (test__arch_unwind_sample (&sample, thread)) {pr_debug ("failed to get unwind sample\n" ); goto out;
@@ -134,7 +133,8 @@ NO_TAIL_CALL_ATTRIBUTE noinline int test_dwarf_unwind__thread(struct thread *thr out:zfree (&sample.user_stack.data); -zfree (&sample.user_regs.regs); +zfree (&sample.user_regs->regs); +perf_sample__exit (&sample);return err; }
tools/perf/tests/mmap-basic.c +3 -0
@@ -130,14 +130,17 @@ static int test__basic_mmap(struct test_suite *test __maybe_unused, int subtest goto out_delete_evlist; } +perf_sample__init (&sample,/*all=*/ false); err =evlist__parse_sample (evlist, event, &sample);if (err) {pr_err ("Can't parse sample, err = %d\n" , err); +perf_sample__exit (&sample); goto out_delete_evlist; } err =-1 ; evsel =evlist__id2evsel (evlist, sample.id); +perf_sample__exit (&sample);if (evsel ==NULL ) {pr_debug ("event with id %" PRIu64" doesn't map to an evsel\n" , sample.id);
tools/perf/tests/openat-syscall-tp-fields.c +3 -1
@@ -111,14 +111,16 @@ static int test__syscall_openat_tp_fields(struct test_suite *test __maybe_unusedcontinue ; } +perf_sample__init (&sample,/*all=*/ false); err =evsel__parse_sample (evsel, event, &sample);if (err) {pr_debug ("Can't parse sample, err = %d\n" , err); +perf_sample__exit (&sample); goto out_delete_evlist; } tp_flags =evsel__intval (evsel, &sample,"flags" ); - +perf_sample__exit (&sample);if (flags != tp_flags) {pr_debug ("%s: Expected flags=%#x, got %#x\n" , __func__, flags, tp_flags);
tools/perf/tests/parse-no-sample-id-all.c +5 -1
@@ -13,6 +13,7 @@static int process_event (struct evlist **pevlist,union perf_event *event) {struct perf_sample sample; +int ret;if (event->header.type ==PERF_RECORD_HEADER_ATTR ) {if (perf_event__process_attr (NULL , event, pevlist)) {
@@ -28,7 +29,10 @@ static int process_event(struct evlist **pevlist, union perf_event *event)if (!*pevlist)return -1 ; -if (evlist__parse_sample (*pevlist, event, &sample)) { +perf_sample__init (&sample,/*all=*/ false); + ret =evlist__parse_sample (*pevlist, event, &sample); +perf_sample__exit (&sample); +if (ret) {pr_debug ("evlist__parse_sample failed\n" );return -1 ; }
tools/perf/tests/perf-record.c +2 -0
@@ -70,6 +70,7 @@ static int test__PERF_RECORD(struct test_suite *test __maybe_unused, int subtestint total_events =0 , nr_events[PERF_RECORD_MAX ] = {0 , };char sbuf[STRERR_BUFSIZE ]; +perf_sample__init (&sample,/*all=*/ false);if (evlist ==NULL )/* Fallback for kernels lacking PERF_COUNT_SW_DUMMY */ evlist =evlist__new_default ();
@@ -330,6 +331,7 @@ static int test__PERF_RECORD(struct test_suite *test __maybe_unused, int subtest out_delete_evlist:evlist__delete (evlist); out: +perf_sample__exit (&sample);if (err == -EACCES )return TEST_SKIP ;if (err <0 || errs !=0 )
tools/perf/tests/perf-time-to-tsc.c +2 -0
@@ -153,6 +153,7 @@ static int test__perf_time_to_tsc(struct test_suite *test __maybe_unused, int suwhile ((event =perf_mmap__read_event (&md->core)) !=NULL ) {struct perf_sample sample; +perf_sample__init (&sample,/*all=*/ false);if (event->header.type !=PERF_RECORD_COMM || (pid_t )event->comm.pid !=getpid () || (pid_t )event->comm.tid !=getpid ())
@@ -170,6 +171,7 @@ static int test__perf_time_to_tsc(struct test_suite *test __maybe_unused, int su } next_event:perf_mmap__consume (&md->core); +perf_sample__exit (&sample); }perf_mmap__read_done (&md->core); }
tools/perf/tests/sample-parsing.c +36 -26
110 lines not on this page · raw message
tools/perf/tests/sw-clock.c +3 -0
@@ -104,12 +104,14 @@ static int __test__sw_clock_freq(enum perf_sw_ids clock_id)while ((event =perf_mmap__read_event (&md->core)) !=NULL ) {struct perf_sample sample; +perf_sample__init (&sample,/*all=*/ false);if (event->header.type !=PERF_RECORD_SAMPLE ) goto next_event; err =evlist__parse_sample (evlist, event, &sample);if (err <0 ) {pr_debug ("Error during parse sample\n" ); +perf_sample__exit (&sample); goto out_delete_evlist; }
@@ -117,6 +119,7 @@ static int __test__sw_clock_freq(enum perf_sw_ids clock_id) nr_samples++; next_event:perf_mmap__consume (&md->core); +perf_sample__exit (&sample); }perf_mmap__read_done (&md->core);
tools/perf/tests/switch-tracking.c +10 -4
@@ -131,9 +131,11 @@ static int process_sample_event(struct evlist *evlist,pid_t next_tid, prev_tid;int cpu, err; +perf_sample__init (&sample,/*all=*/ false);if (evlist__parse_sample (evlist, event, &sample)) {pr_debug ("evlist__parse_sample failed\n" ); -return -1 ; + err =-1 ; + goto out; } evsel =evlist__id2evsel (evlist, sample.id);
@@ -145,7 +147,7 @@ static int process_sample_event(struct evlist *evlist, cpu, prev_tid, next_tid); err =check_cpu (switch_tracking, cpu);if (err) -return err; + goto out;/* * Check for no missing sched_switch events i.e. that the * evsel->core.system_wide flag has worked.
@@ -153,7 +155,8 @@ static int process_sample_event(struct evlist *evlist,if (switch_tracking->tids[cpu] !=-1 && switch_tracking->tids[cpu] != prev_tid) {pr_debug ("Missing sched_switch events\n" ); -return -1 ; + err =-1 ; + goto out; } switch_tracking->tids[cpu] = next_tid; }
@@ -169,7 +172,10 @@ static int process_sample_event(struct evlist *evlist, switch_tracking->cycles_after_comm_4 =1 ; } -return 0 ; + err =0 ; +out: +perf_sample__exit (&sample); +return err; } static int process_event(struct evlist *evlist, union perf_event *event,
tools/perf/util/Build +1 -0
@@ -67,6 +67,7 @@ perf-util-y += maps.o perf-util-y += pstack.o perf-util-y += session.o perf-util-y += tool.o +perf-util-y += sample.o perf-util-y += sample-raw.o perf-util-y += s390-sample-raw.o perf-util-y += amd-sample-raw.o
tools/perf/util/arm-spe.c +18 -6
@@ -379,8 +379,10 @@ static int arm_spe__synth_mem_sample(struct arm_spe_queue *speq,struct arm_spe *spe = speq->spe;struct arm_spe_record *record = &speq->decoder->record;union perf_event *event = speq->event_buf; -struct perf_sample sample = { .ip =0 , }; +struct perf_sample sample; +int ret; +perf_sample__init (&sample,/*all=*/ true);arm_spe_prep_sample (spe, speq, event, &sample); sample.id = spe_events_id;
@@ -390,7 +392,9 @@ static int arm_spe__synth_mem_sample(struct arm_spe_queue *speq, sample.data_src = data_src; sample.weight = record->latency; -return arm_spe_deliver_synth_event (spe, speq, event, &sample); + ret =arm_spe_deliver_synth_event (spe, speq, event, &sample); +perf_sample__exit (&sample); +return ret; } static int arm_spe__synth_branch_sample(struct arm_spe_queue *speq,
@@ -399,8 +403,10 @@ static int arm_spe__synth_branch_sample(struct arm_spe_queue *speq,struct arm_spe *spe = speq->spe;struct arm_spe_record *record = &speq->decoder->record;union perf_event *event = speq->event_buf; -struct perf_sample sample = { .ip =0 , }; +struct perf_sample sample; +int ret; +perf_sample__init (&sample,/*all=*/ true);arm_spe_prep_sample (spe, speq, event, &sample); sample.id = spe_events_id;
@@ -409,7 +415,9 @@ static int arm_spe__synth_branch_sample(struct arm_spe_queue *speq, sample.weight = record->latency; sample.flags = speq->flags; -return arm_spe_deliver_synth_event (spe, speq, event, &sample); + ret =arm_spe_deliver_synth_event (spe, speq, event, &sample); +perf_sample__exit (&sample); +return ret; } static int arm_spe__synth_instruction_sample(struct arm_spe_queue *speq,
@@ -418,7 +426,8 @@ static int arm_spe__synth_instruction_sample(struct arm_spe_queue *speq,struct arm_spe *spe = speq->spe;struct arm_spe_record *record = &speq->decoder->record;union perf_event *event = speq->event_buf; -struct perf_sample sample = { .ip =0 , }; +struct perf_sample sample; +int ret;/* * Handles perf instruction sampling period.
@@ -428,6 +437,7 @@ static int arm_spe__synth_instruction_sample(struct arm_spe_queue *speq,return 0 ; speq->period_instructions =0 ; +perf_sample__init (&sample,/*all=*/ true);arm_spe_prep_sample (spe, speq, event, &sample); sample.id = spe_events_id;
@@ -439,7 +449,9 @@ static int arm_spe__synth_instruction_sample(struct arm_spe_queue *speq, sample.weight = record->latency; sample.flags = speq->flags; -return arm_spe_deliver_synth_event (spe, speq, event, &sample); + ret =arm_spe_deliver_synth_event (spe, speq, event, &sample); +perf_sample__exit (&sample); +return ret; }static const struct midr_range common_ds_encoding_cpus[] = {
tools/perf/util/arm64-frame-pointer-unwind-support.c +18 -11
@@ -4,6 +4,7 @@#include "event.h" #include "perf_regs.h" // SMPL_REG_MASK #include "unwind.h" +#include <string.h> #define perf_event_arm_regs perf_event_arm64_regs#include "../../arch/arm64/include/uapi/asm/perf_regs.h"
@@ -16,8 +17,13 @@ struct entries {static bool get_leaf_frame_caller_enabled (struct perf_sample *sample) { -return callchain_param.record_mode ==CALLCHAIN_FP && sample->user_regs.regs - && sample->user_regs.mask &SMPL_REG_MASK (PERF_REG_ARM64_LR ); +struct regs_dump *regs; + +if (callchain_param.record_mode !=CALLCHAIN_FP ) +return false; + + regs =perf_sample__user_regs (sample); +return regs->regs && regs->mask &SMPL_REG_MASK (PERF_REG_ARM64_LR ); }static int add_entry (struct unwind_entry *entry,void *arg)
@@ -32,7 +38,7 @@ u64 get_leaf_frame_caller_aarch64(struct perf_sample *sample, struct thread *thr {int ret;struct entries entries = {}; -struct regs_dump old_regs = sample->user_regs; +struct regs_dump old_regs, *regs;if (!get_leaf_frame_caller_enabled (sample))return 0 ;
@@ -42,19 +48,20 @@ u64 get_leaf_frame_caller_aarch64(struct perf_sample *sample, struct thread *thr* and set its mask. SP is not used when doing the unwinding but it * still needs to be set to prevent failures. */ - -if (!(sample->user_regs.mask &SMPL_REG_MASK (PERF_REG_ARM64_PC ))) { - sample->user_regs.cache_mask |=SMPL_REG_MASK (PERF_REG_ARM64_PC ); - sample->user_regs.cache_regs[PERF_REG_ARM64_PC ] = sample->callchain->ips[usr_idx+1 ]; + regs =perf_sample__user_regs (sample); +memcpy (&old_regs, regs,sizeof (*regs)); +if (!(regs->mask &SMPL_REG_MASK (PERF_REG_ARM64_PC ))) { + regs->cache_mask |=SMPL_REG_MASK (PERF_REG_ARM64_PC ); + regs->cache_regs[PERF_REG_ARM64_PC ] = sample->callchain->ips[usr_idx+1 ]; } -if (!(sample->user_regs.mask &SMPL_REG_MASK (PERF_REG_ARM64_SP ))) { - sample->user_regs.cache_mask |=SMPL_REG_MASK (PERF_REG_ARM64_SP ); - sample->user_regs.cache_regs[PERF_REG_ARM64_SP ] =0 ; +if (!(regs->mask &SMPL_REG_MASK (PERF_REG_ARM64_SP ))) { + regs->cache_mask |=SMPL_REG_MASK (PERF_REG_ARM64_SP ); + regs->cache_regs[PERF_REG_ARM64_SP ] =0 ; } ret =unwind__get_entries (add_entry, &entries, thread, sample,2 , true); - sample->user_regs = old_regs; +memcpy (regs, &old_regs,sizeof (*regs));if (ret || entries.length !=2 )return ret;
tools/perf/util/auxtrace.c +9 -6
@@ -1173,16 +1173,19 @@ static int auxtrace_queue_data_cb(struct perf_session *session,if (!qd->samples || event->header.type !=PERF_RECORD_SAMPLE )return 0 ; +perf_sample__init (&sample,/*all=*/ false); err =evlist__parse_sample (session->evlist, event, &sample);if (err) -return err; - -if (!sample.aux_sample.size) -return 0 ; + goto out; - offset += sample.aux_sample.data - (void *)event; +if (sample.aux_sample.size) { + offset += sample.aux_sample.data - (void *)event; -return session->auxtrace->queue_data (session, &sample,NULL , offset); + err = session->auxtrace->queue_data (session, &sample,NULL , offset); + } +out: +perf_sample__exit (&sample); +return err; }int auxtrace_queue_data (struct perf_session *session,bool samples,bool events)
tools/perf/util/cs-etm.c +22 -9
@@ -506,20 +506,27 @@ static int cs_etm__process_aux_output_hw_id(struct perf_session *session, evsel =evlist__event2evsel (session->evlist, event);if (!evsel)return -EINVAL ; +perf_sample__init (&sample,/*all=*/ false); err =evsel__parse_sample (evsel, event, &sample);if (err) -return err; + goto out; cpu = sample.cpu;if (cpu ==-1 ) {/* no CPU in the sample - possibly recorded with an old version of perf */ pr_err ("CS_ETM: no CPU AUX_OUTPUT_HW_ID sample. Use compatible perf to record." ); -return -EINVAL ; + err = -EINVAL ; + goto out; } -if (FIELD_GET (CS_AUX_HW_ID_MINOR_VERSION_MASK , hw_id) ==0 ) -return cs_etm__process_trace_id_v0 (etm, cpu, hw_id); +if (FIELD_GET (CS_AUX_HW_ID_MINOR_VERSION_MASK , hw_id) ==0 ) { + err =cs_etm__process_trace_id_v0 (etm, cpu, hw_id); + goto out; + } -return cs_etm__process_trace_id_v0_1 (etm, cpu, hw_id); + err =cs_etm__process_trace_id_v0_1 (etm, cpu, hw_id); +out: +perf_sample__exit (&sample); +return err; } void cs_etm__etmq_set_traceid_queue_timestamp(struct cs_etm_queue *etmq,
@@ -1560,8 +1567,9 @@ static int cs_etm__synth_instruction_sample(struct cs_etm_queue *etmq,int ret =0 ;struct cs_etm_auxtrace *etm = etmq->etm;union perf_event *event = tidq->event_buf; -struct perf_sample sample = {.ip =0 ,}; +struct perf_sample sample; +perf_sample__init (&sample,/*all=*/ true); event->sample.header.type =PERF_RECORD_SAMPLE ; event->sample.header.misc =cs_etm__cpu_mode (etmq, addr, tidq->el); event->sample.header.size =sizeof (struct perf_event_header );
@@ -1598,6 +1606,7 @@ static int cs_etm__synth_instruction_sample(struct cs_etm_queue *etmq,"CS ETM Trace: failed to deliver instruction event, error %d\n" , ret); +perf_sample__exit (&sample);return ret; }
@@ -3151,9 +3160,10 @@ static int cs_etm__queue_aux_records_cb(struct perf_session *session, union perf evsel =evlist__event2evsel (session->evlist, event);if (!evsel)return -EINVAL ; +perf_sample__init (&sample,/*all=*/ false); ret =evsel__parse_sample (evsel, event, &sample);if (ret) -return ret; + goto out;/* * Loop through the auxtrace index to find the buffer that matches up with this aux event.
@@ -3168,7 +3178,7 @@ static int cs_etm__queue_aux_records_cb(struct perf_session *session, union perf* 1 ('not found') */ if (ret !=1 ) -return ret; + goto out; } }
@@ -3178,7 +3188,10 @@ static int cs_etm__queue_aux_records_cb(struct perf_session *session, union perf*/ pr_err ("CS ETM: Couldn't find auxtrace buffer for aux_offset: %#" PRI_lx64" tid: %d cpu: %d\n" , event->aux.aux_offset, sample.tid, sample.cpu); -return 0 ; + ret =0 ; +out: +perf_sample__exit (&sample); +return ret; }static int cs_etm__queue_aux_records (struct perf_session *session)
tools/perf/util/evsel.c +12 -9
@@ -3164,17 +3164,19 @@ int evsel__parse_sample(struct evsel *evsel, union perf_event *event, }if (type &PERF_SAMPLE_REGS_USER ) { +struct regs_dump *regs =perf_sample__user_regs (data); +OVERFLOW_CHECK_u64 (array); - data->user_regs.abi = *array; + regs->abi = *array; array++; -if (data->user_regs.abi) { +if (regs->abi) {u64 mask = evsel->core.attr.sample_regs_user; sz =hweight64 (mask) *sizeof (u64);OVERFLOW_CHECK (array, sz, max_size); - data->user_regs.mask = mask; - data->user_regs.regs = (u64 *)array; + regs->mask = mask; + regs->regs = (u64 *)array; array = (void *)array + sz; } }
@@ -3218,19 +3220,20 @@ int evsel__parse_sample(struct evsel *evsel, union perf_event *event, array++; } - data->intr_regs.abi =PERF_SAMPLE_REGS_ABI_NONE ;if (type &PERF_SAMPLE_REGS_INTR ) { +struct regs_dump *regs =perf_sample__intr_regs (data); +OVERFLOW_CHECK_u64 (array); - data->intr_regs.abi = *array; + regs->abi = *array; array++; -if (data->intr_regs.abi !=PERF_SAMPLE_REGS_ABI_NONE ) { +if (regs->abi !=PERF_SAMPLE_REGS_ABI_NONE ) {u64 mask = evsel->core.attr.sample_regs_intr; sz =hweight64 (mask) *sizeof (u64);OVERFLOW_CHECK (array, sz, max_size); - data->intr_regs.mask = mask; - data->intr_regs.regs = (u64 *)array; + regs->mask = mask; + regs->regs = (u64 *)array; array = (void *)array + sz; } }
tools/perf/util/intel-bts.c +3 -1
@@ -275,12 +275,13 @@ static int intel_bts_synth_branch_sample(struct intel_bts_queue *btsq,int ret;struct intel_bts *bts = btsq->bts;union perf_event event; -struct perf_sample sample = { .ip =0 , }; +struct perf_sample sample;if (bts->synth_opts.initial_skip && bts->num_events++ <= bts->synth_opts.initial_skip)return 0 ; +perf_sample__init (&sample,/*all=*/ true); sample.ip =le64_to_cpu (branch->from); sample.cpumode =intel_bts_cpumode (bts, sample.ip); sample.pid = btsq->pid;
@@ -312,6 +313,7 @@ static int intel_bts_synth_branch_sample(struct intel_bts_queue *btsq,pr_err ("Intel BTS: failed to deliver branch event, error %d\n" , ret); +perf_sample__exit (&sample);return ret; }
tools/perf/util/intel-pt.c +94 -42
383 lines not on this page · raw message
tools/perf/util/jitdump.c +6 -4
@@ -516,7 +516,7 @@ static int jit_repipe_code_load(struct jit_buf_desc *jd, union jr_entry *jr)* create pseudo sample to induce dso hit increment * use first address as sample address */ -memset (&sample,0 ,sizeof (sample)); +perf_sample__init (&sample,/*all=*/ true); sample.cpumode =PERF_RECORD_MISC_USER ; sample.pid = pid; sample.tid = tid;
@@ -535,6 +535,7 @@ static int jit_repipe_code_load(struct jit_buf_desc *jd, union jr_entry *jr)build_id__mark_dso_hit (tool, event, &sample,NULL , jd->machine); out: +perf_sample__exit (&sample);free (event);return ret; }
@@ -611,7 +612,7 @@ static int jit_repipe_code_move(struct jit_buf_desc *jd, union jr_entry *jr)* create pseudo sample to induce dso hit increment * use first address as sample address */ -memset (&sample,0 ,sizeof (sample)); +perf_sample__init (&sample,/*all=*/ true); sample.cpumode =PERF_RECORD_MISC_USER ; sample.pid = pid; sample.tid = tid;
@@ -620,12 +621,13 @@ static int jit_repipe_code_move(struct jit_buf_desc *jd, union jr_entry *jr) ret =perf_event__process_mmap2 (tool, event, &sample, jd->machine);if (ret) -return ret; + goto out; ret =jit_inject_event (jd, event);if (!ret)build_id__mark_dso_hit (tool, event, &sample,NULL , jd->machine); - +out: +perf_sample__exit (&sample);return ret; }
tools/perf/util/machine.c +2 -2
@@ -2929,8 +2929,8 @@ static int thread__resolve_callchain_unwind(struct thread *thread,return 0 ;/* Bail out if nothing was captured. */ -if ((!sample->user_regs.regs) || - (!sample->user_stack.size)) +if (!sample->user_regs || !sample->user_regs->regs || + !sample->user_stack.size)return 0 ;if (!symbols)
tools/perf/util/python.c +9 -0
@@ -270,6 +270,12 @@ static PyMemberDef pyrf_sample_event__members[] = { { .name = NULL, }, }; +static void pyrf_sample_event__delete (struct pyrf_event *pevent) +{ +perf_sample__exit (&pevent->sample); +Py_TYPE (pevent)->tp_free ((PyObject *)pevent); +} +static PyObject *pyrf_sample_event__repr (const struct pyrf_event *pevent) {PyObject *ret;
@@ -428,6 +434,9 @@ static int pyrf_event__setup_types(void) pyrf_sample_event__type.tp_new = pyrf_context_switch_event__type.tp_new = pyrf_throttle_event__type.tp_new = PyType_GenericNew; + + pyrf_sample_event__type.tp_dealloc = (destructor )pyrf_sample_event__delete, + err =PyType_Ready (&pyrf_mmap_event__type);if (err <0 ) goto out;
tools/perf/util/s390-cpumsf.c +4 -2
@@ -513,6 +513,7 @@ static bool s390_cpumsf_make_event(size_t pos, .period = 1 };union perf_event event; +int ret;memset (&event,0 ,sizeof (event));if (basic->CL ==1 )/* Native LPAR mode */
@@ -536,8 +537,9 @@ static bool s390_cpumsf_make_event(size_t pos,pr_debug4 ("%s pos:%#zx ip:%#" PRIx64" P:%d CL:%d pid:%d.%d cpumode:%d cpu:%d\n" , __func__, pos, sample.ip, basic->P, basic->CL, sample.pid, sample.tid, sample.cpumode, sample.cpu); -if (perf_session__deliver_synth_event (sfq->sf->session, &event, - &sample)) { + ret =perf_session__deliver_synth_event (sfq->sf->session, &event, &sample); +perf_sample__exit (&sample); +if (ret) {pr_err ("s390 Auxiliary Trace: failed to deliver event\n" );return false; }
tools/perf/util/sample.c new file +43 -0
43 lines not on this page · raw message
tools/perf/util/sample.h +7 -2
@@ -114,14 +114,19 @@ struct perf_sample {struct ip_callchain *callchain;struct branch_stack *branch_stack;u64 *branch_stack_cntr; -struct regs_dump user_regs; -struct regs_dump intr_regs; +struct regs_dump *user_regs; +struct regs_dump *intr_regs;struct stack_dump user_stack;struct sample_read read;struct aux_sample aux_sample;struct simd_flags simd_flags; }; +void perf_sample__init (struct perf_sample *sample,bool all); +void perf_sample__exit (struct perf_sample *sample); +struct regs_dump *perf_sample__user_regs (struct perf_sample *sample); +struct regs_dump *perf_sample__intr_regs (struct perf_sample *sample); +/* * raw_data is always 4 bytes from an 8-byte boundary, so subtract 4 to get * 8-byte alignment.
tools/perf/util/scripting-engines/trace-event-python.c +20 -9
@@ -745,19 +745,30 @@ static int set_regs_in_dict(PyObject *dict,const char *arch =perf_env__arch (evsel__env (evsel));int size = (__sw_hweight64 (attr->sample_regs_intr) *MAX_REG_SIZE ) +1 ; -char *bf =malloc (size); -if (!bf) -return -1 ; +char *bf =NULL ; -regs_map (&sample->intr_regs, attr->sample_regs_intr, arch, bf, size); +if (sample->intr_regs) { + bf =malloc (size); +if (!bf) +return -1 ; -pydict_set_item_string_decref (dict,"iregs" , -_PyUnicode_FromString (bf)); +regs_map (sample->intr_regs, attr->sample_regs_intr, arch, bf, size); -regs_map (&sample->user_regs, attr->sample_regs_user, arch, bf, size); +pydict_set_item_string_decref (dict,"iregs" , +_PyUnicode_FromString (bf)); + } -pydict_set_item_string_decref (dict,"uregs" , -_PyUnicode_FromString (bf)); +if (sample->user_regs) { +if (!bf) { + bf =malloc (size); +if (!bf) +return -1 ; + } +regs_map (sample->user_regs, attr->sample_regs_user, arch, bf, size); + +pydict_set_item_string_decref (dict,"uregs" , +_PyUnicode_FromString (bf)); + }free (bf);return 0 ;
tools/perf/util/session.c +65 -29
179 lines not on this page · raw message
tools/perf/util/synthetic-events.c +12 -12
@@ -1508,9 +1508,9 @@ size_t perf_event__sample_event_size(const struct perf_sample *sample, u64 type, }if (type &PERF_SAMPLE_REGS_USER ) { -if (sample->user_regs.abi) { +if (sample->user_regs && sample->user_regs->abi) { result +=sizeof (u64); - sz =hweight64 (sample->user_regs.mask) *sizeof (u64); + sz =hweight64 (sample->user_regs->mask) *sizeof (u64); result += sz; }else { result +=sizeof (u64);
@@ -1536,9 +1536,9 @@ size_t perf_event__sample_event_size(const struct perf_sample *sample, u64 type, result +=sizeof (u64);if (type &PERF_SAMPLE_REGS_INTR ) { -if (sample->intr_regs.abi) { +if (sample->intr_regs && sample->intr_regs->abi) { result +=sizeof (u64); - sz =hweight64 (sample->intr_regs.mask) *sizeof (u64); + sz =hweight64 (sample->intr_regs->mask) *sizeof (u64); result += sz; }else { result +=sizeof (u64);
@@ -1707,10 +1707,10 @@ int perf_event__synthesize_sample(union perf_event *event, u64 type, u64 read_fo }if (type &PERF_SAMPLE_REGS_USER ) { -if (sample->user_regs.abi) { - *array++ = sample->user_regs.abi; - sz =hweight64 (sample->user_regs.mask) *sizeof (u64); -memcpy (array, sample->user_regs.regs, sz); +if (sample->user_regs && sample->user_regs->abi) { + *array++ = sample->user_regs->abi; + sz =hweight64 (sample->user_regs->mask) *sizeof (u64); +memcpy (array, sample->user_regs->regs, sz); array = (void *)array + sz; }else { *array++ =0 ;
@@ -1743,10 +1743,10 @@ int perf_event__synthesize_sample(union perf_event *event, u64 type, u64 read_fo }if (type &PERF_SAMPLE_REGS_INTR ) { -if (sample->intr_regs.abi) { - *array++ = sample->intr_regs.abi; - sz =hweight64 (sample->intr_regs.mask) *sizeof (u64); -memcpy (array, sample->intr_regs.regs, sz); +if (sample->intr_regs && sample->intr_regs->abi) { + *array++ = sample->intr_regs->abi; + sz =hweight64 (sample->intr_regs->mask) *sizeof (u64); +memcpy (array, sample->intr_regs->regs, sz); array = (void *)array + sz; }else { *array++ =0 ;
tools/perf/util/unwind-libdw.c +6 -3
@@ -190,7 +190,10 @@ static bool memory_read(Dwfl *dwfl __maybe_unused, Dwarf_Addr addr, Dwarf_Word *int offset;int ret; - ret =perf_reg_value (&start, &ui->sample->user_regs, +if (!ui->sample->user_regs) +return false; + + ret =perf_reg_value (&start, ui->sample->user_regs,perf_arch_reg_sp (arch));if (ret)return false;
@@ -273,7 +276,7 @@ int unwind__get_entries(unwind_entry_cb_t cb, void *arg,Dwarf_Word ip;int err = -EINVAL , i; -if (!data->user_regs.regs) +if (!data->user_regs || !data->user_regs->regs)return -EINVAL ; ui =zalloc (sizeof (ui_buf) +sizeof (ui_buf.entries[0 ]) * max_stack);
@@ -286,7 +289,7 @@ int unwind__get_entries(unwind_entry_cb_t cb, void *arg,if (!ui->dwfl) goto out; - err =perf_reg_value (&ip, &data->user_regs,perf_arch_reg_ip (arch)); + err =perf_reg_value (&ip, data->user_regs,perf_arch_reg_ip (arch));if (err) goto out;
Signature · 1 line
2.47.1.688.g23fc6f90ad-goog
Quoted 9 lines · Ian Rogers
The struct
dump_regscontains 512 bytes ofcache_regs, meaning the two values inperf_samplecontribute 1088 bytes of its total 1384 bytes size. Initializing this much memory has a cost reported by Tavian Barnes <[email protected]> as about 2.5% when runningperf script --itrace=i0:
https://lore.kernel.org/lkml/d841b97b3ad2ca8bcab07e4293375fb7c32dfce7.1736618095.git.tavianator@tavianator.com/Adrian Hunter <[email protected]> replied that the zero initialization was necessary and couldn't simply be removed.
A much easier fix is to keep a global/heap allocate perf event around that has these parts zeroed and only override the fields needed and clear them afterwards.
(similar strategy as a slab constructor in the kernel)
-Andi
Ian Rogers
reviewerQuoted 11 lines · Ian Rogers
The struct
dump_regscontains 512 bytes ofcache_regs, meaning the two values inperf_samplecontribute 1088 bytes of its total 1384 bytes size. Initializing this much memory has a cost reported by Tavian Barnes <[email protected]> as about 2.5% when runningperf script --itrace=i0:
https://lore.kernel.org/lkml/d841b97b3ad2ca8bcab07e4293375fb7c32dfce7.1736618095.git.tavianator@tavianator.com/Adrian Hunter <[email protected]> replied that the zero initialization was necessary and couldn't simply be removed.
Quoted 6 lines
This patch aims to strike a middle ground of still zeroing the
perf_sample, but removing 79% of its size by makeuser_regsandintr_regsoptional pointers to zalloc-ed memory. To support the allocation accessors are created foruser_regsandintr_regs. To support correct cleanupperf_sample__initandperf_sample__exitfunctions are created and added throughout the code base.
Ping. Given the memory savings and performance wins it would be nice to see this land. Andi Kleen commented on doing a reimplementation, which is fine but out-of-scope of what I'm doing here.
Thanks,
Ian
Namhyung Kim
maintainerQuoted 19 lines · Ian Rogers
Quoted 11 lines · Ian Rogers
The struct
dump_regscontains 512 bytes ofcache_regs, meaning the two values inperf_samplecontribute 1088 bytes of its total 1384 bytes size. Initializing this much memory has a cost reported by Tavian Barnes <[email protected]> as about 2.5% when runningperf script --itrace=i0:
https://lore.kernel.org/lkml/d841b97b3ad2ca8bcab07e4293375fb7c32dfce7.1736618095.git.tavianator@tavianator.com/Adrian Hunter <[email protected]> replied that the zero initialization was necessary and couldn't simply be removed.
Quoted 6 lines
This patch aims to strike a middle ground of still zeroing the
perf_sample, but removing 79% of its size by makeuser_regsandintr_regsoptional pointers to zalloc-ed memory. To support the allocation accessors are created foruser_regsandintr_regs. To support correct cleanupperf_sample__initandperf_sample__exitfunctions are created and added throughout the code base.
Quoted 3 lines
Ping. Given the memory savings and performance wins it would be nice to see this land. Andi Kleen commented on doing a reimplementation, which is fine but out-of-scope of what I'm doing here.
Yeah, I like the core of the change. Andi's concern is that it touches too many places. It'd be nice if we can do that without allocating memory for regs and eliminating the perf_sample__{init,exit}. But I'm not if it's possible.
Thanks,
Namhyung
Ian Rogers
reviewerQuoted 21 lines · Namhyung Kim
Quoted 19 lines · Ian Rogers
Quoted 17 lines · Ian Rogers
The struct
dump_regscontains 512 bytes ofcache_regs, meaning the two values inperf_samplecontribute 1088 bytes of its total 1384 bytes size. Initializing this much memory has a cost reported by Tavian Barnes <[email protected]> as about 2.5% when runningperf script --itrace=i0:
https://lore.kernel.org/lkml/d841b97b3ad2ca8bcab07e4293375fb7c32dfce7.1736618095.git.tavianator@tavianator.com/Adrian Hunter <[email protected]> replied that the zero initialization was necessary and couldn't simply be removed.
This patch aims to strike a middle ground of still zeroing the
perf_sample, but removing 79% of its size by makeuser_regsandintr_regsoptional pointers to zalloc-ed memory. To support the allocation accessors are created foruser_regsandintr_regs. To support correct cleanupperf_sample__initandperf_sample__exitfunctions are created and added throughout the code base.
Quoted 8 lines
Quoted 3 lines
Ping. Given the memory savings and performance wins it would be nice to see this land. Andi Kleen commented on doing a reimplementation, which is fine but out-of-scope of what I'm doing here.
Yeah, I like the core of the change. Andi's concern is that it touches too many places. It'd be nice if we can do that without allocating memory for regs and eliminating the perf_sample__{init,exit}. But I'm not if it's possible.
Moving from no allocations to 2 possible allocations means there has to be corresponding frees. Putting the frees into an __exit function is the norm for this kind of cleanup. I don't see how you can move to the approach presented without adding the frees and not introduce a memory leak. I don't see what's actionable for me to do here.
Thanks,
Ian
Namhyung Kim
maintainerQuoted 31 lines · Ian Rogers
Quoted 21 lines · Namhyung Kim
Quoted 19 lines · Ian Rogers
Quoted 17 lines · Ian Rogers
The struct
dump_regscontains 512 bytes ofcache_regs, meaning the two values inperf_samplecontribute 1088 bytes of its total 1384 bytes size. Initializing this much memory has a cost reported by Tavian Barnes <[email protected]> as about 2.5% when runningperf script --itrace=i0:
https://lore.kernel.org/lkml/d841b97b3ad2ca8bcab07e4293375fb7c32dfce7.1736618095.git.tavianator@tavianator.com/Adrian Hunter <[email protected]> replied that the zero initialization was necessary and couldn't simply be removed.
This patch aims to strike a middle ground of still zeroing the
perf_sample, but removing 79% of its size by makeuser_regsandintr_regsoptional pointers to zalloc-ed memory. To support the allocation accessors are created foruser_regsandintr_regs. To support correct cleanupperf_sample__initandperf_sample__exitfunctions are created and added throughout the code base.Quoted 8 lines
Quoted 3 lines
Ping. Given the memory savings and performance wins it would be nice to see this land. Andi Kleen commented on doing a reimplementation, which is fine but out-of-scope of what I'm doing here.
Yeah, I like the core of the change. Andi's concern is that it touches too many places. It'd be nice if we can do that without allocating memory for regs and eliminating the perf_sample__{init,exit}. But I'm not if it's possible.
Quoted 5 lines
Moving from no allocations to 2 possible allocations means there has to be corresponding frees. Putting the frees into an
__exitfunction is the norm for this kind of cleanup. I don't see how you can move to the approach presented without adding the frees and not introduce a memory leak. I don't see what's actionable for me to do here.
Right, I'm inclined to merge this patch. But I need to think a bit more about the Andi's approach before that.
Thanks,
Namhyung
Namhyung Kim
maintainerQuoted 32 lines · Namhyung Kim
Quoted 31 lines · Ian Rogers
Quoted 29 lines · Namhyung Kim
Quoted 19 lines · Ian Rogers
Quoted 11 lines · Ian Rogers
The struct
dump_regscontains 512 bytes ofcache_regs, meaning the two values inperf_samplecontribute 1088 bytes of its total 1384 bytes size. Initializing this much memory has a cost reported by Tavian Barnes <[email protected]> as about 2.5% when runningperf script --itrace=i0:
https://lore.kernel.org/lkml/d841b97b3ad2ca8bcab07e4293375fb7c32dfce7.1736618095.git.tavianator@tavianator.com/Adrian Hunter <[email protected]> replied that the zero initialization was necessary and couldn't simply be removed.
Quoted 6 lines
This patch aims to strike a middle ground of still zeroing the
perf_sample, but removing 79% of its size by makeuser_regsandintr_regsoptional pointers to zalloc-ed memory. To support the allocation accessors are created foruser_regsandintr_regs. To support correct cleanupperf_sample__initandperf_sample__exitfunctions are created and added throughout the code base.Quoted 3 lines
Ping. Given the memory savings and performance wins it would be nice to see this land. Andi Kleen commented on doing a reimplementation, which is fine but out-of-scope of what I'm doing here.
Yeah, I like the core of the change. Andi's concern is that it touches too many places. It'd be nice if we can do that without allocating memory for regs and eliminating the perf_sample__{init,exit}. But I'm not if it's possible.
Quoted 8 lines
Quoted 5 lines
Moving from no allocations to 2 possible allocations means there has to be corresponding frees. Putting the frees into an
__exitfunction is the norm for this kind of cleanup. I don't see how you can move to the approach presented without adding the frees and not introduce a memory leak. I don't see what's actionable for me to do here.Right, I'm inclined to merge this patch. But I need to think a bit more about the Andi's approach before that.
Probably we can use a global (or per-thread) variable, but I think it could grow to another pain point in the future. Using __init/exit will make it easier for potential future changes.
Thanks,
Namhyung
Quoted 8 lines · Ian Rogers
The struct
dump_regscontains 512 bytes ofcache_regs, meaning the two values inperf_samplecontribute 1088 bytes of its total 1384 bytes size. Initializing this much memory has a cost reported by Tavian Barnes <[email protected]> as about 2.5% when runningperf script --itrace=i0:
https://lore.kernel.org/lkml/d841b97b3ad2ca8bcab07e4293375fb7c32dfce7.1736618095.git.tavianator@tavianator.com/[...]
Applied to perf-tools-next, thanks!
Best regards,
Namhyung