* perf: Add duration_time to perf stat
@ 2019-03-26 22:18 Andi Kleen
2019-03-26 22:18 ` [PATCH 1/4] perf, tools: Revert checks for duration_time Andi Kleen
` (4 more replies)
0 siblings, 5 replies; 12+ messages in thread
From: Andi Kleen @ 2019-03-26 22:18 UTC (permalink / raw)
To: acme; +Cc: jolsa, linux-kernel, linux-perf-users
Currently it's not possible to get the wall clock time of a measurement
in CSV mode output, it's only in the non CSV perf stat output.
Often this is needed to normalize event counts to time.
Add a new duration_time event that can be specified and reports
the wall clock time in ns.
^ permalink raw reply [flat|nested] 12+ messages in thread* [PATCH 1/4] perf, tools: Revert checks for duration_time 2019-03-26 22:18 perf: Add duration_time to perf stat Andi Kleen @ 2019-03-26 22:18 ` Andi Kleen 2019-04-05 11:38 ` [tip:perf/core] perf stat: " tip-bot for Andi Kleen 2019-03-26 22:18 ` [PATCH 2/4] perf, tools: Implement duration_time as a proper event Andi Kleen ` (3 subsequent siblings) 4 siblings, 1 reply; 12+ messages in thread From: Andi Kleen @ 2019-03-26 22:18 UTC (permalink / raw) To: acme; +Cc: jolsa, linux-kernel, linux-perf-users, Andi Kleen From: Andi Kleen <ak@linux.intel.com> This reverts e864c5ca145e perf stat: Hide internal duration_time counter but doing it manually since the code has now moved to a different file. The next patch will properly implement duration_time as a full event, so no need to hide it anymore. Signed-off-by: Andi Kleen <ak@linux.intel.com> --- tools/perf/util/stat-display.c | 18 ------------------ 1 file changed, 18 deletions(-) diff --git a/tools/perf/util/stat-display.c b/tools/perf/util/stat-display.c index 6d043c78f3c2..3324f23c7efc 100644 --- a/tools/perf/util/stat-display.c +++ b/tools/perf/util/stat-display.c @@ -18,11 +18,6 @@ #define CNTR_NOT_SUPPORTED "<not supported>" #define CNTR_NOT_COUNTED "<not counted>" -static bool is_duration_time(struct perf_evsel *evsel) -{ - return !strcmp(evsel->name, "duration_time"); -} - static void print_running(struct perf_stat_config *config, u64 run, u64 ena) { @@ -628,9 +623,6 @@ static void print_aggr(struct perf_stat_config *config, ad.id = id = config->aggr_map->map[s]; first = true; evlist__for_each_entry(evlist, counter) { - if (is_duration_time(counter)) - continue; - ad.val = ad.ena = ad.run = 0; ad.nr = 0; if (!collect_data(config, counter, aggr_cb, &ad)) @@ -848,8 +840,6 @@ static void print_no_aggr_metric(struct perf_stat_config *config, if (prefix) fputs(prefix, config->output); evlist__for_each_entry(evlist, counter) { - if (is_duration_time(counter)) - continue; if (first) { aggr_printout(config, counter, cpu, 0); first = false; @@ -906,8 +896,6 @@ static void print_metric_headers(struct perf_stat_config *config, /* Print metrics headers only */ evlist__for_each_entry(evlist, counter) { - if (is_duration_time(counter)) - continue; os.evsel = counter; out.ctx = &os; out.print_metric = print_metric_header; @@ -1136,15 +1124,11 @@ perf_evlist__print_counters(struct perf_evlist *evlist, break; case AGGR_THREAD: evlist__for_each_entry(evlist, counter) { - if (is_duration_time(counter)) - continue; print_aggr_thread(config, _target, counter, prefix); } break; case AGGR_GLOBAL: evlist__for_each_entry(evlist, counter) { - if (is_duration_time(counter)) - continue; print_counter_aggr(config, counter, prefix); } if (metric_only) @@ -1155,8 +1139,6 @@ perf_evlist__print_counters(struct perf_evlist *evlist, print_no_aggr_metric(config, evlist, prefix); else { evlist__for_each_entry(evlist, counter) { - if (is_duration_time(counter)) - continue; print_counter(config, counter, prefix); } } -- 2.20.1 ^ permalink raw reply related [flat|nested] 12+ messages in thread
* [tip:perf/core] perf stat: Revert checks for duration_time 2019-03-26 22:18 ` [PATCH 1/4] perf, tools: Revert checks for duration_time Andi Kleen @ 2019-04-05 11:38 ` tip-bot for Andi Kleen 0 siblings, 0 replies; 12+ messages in thread From: tip-bot for Andi Kleen @ 2019-04-05 11:38 UTC (permalink / raw) To: linux-tip-commits; +Cc: tglx, hpa, jolsa, linux-kernel, acme, mingo, ak Commit-ID: c2b3c170db610896e4e633cba2135045333811c2 Gitweb: https://git.kernel.org/tip/c2b3c170db610896e4e633cba2135045333811c2 Author: Andi Kleen <ak@linux.intel.com> AuthorDate: Tue, 26 Mar 2019 15:18:20 -0700 Committer: Arnaldo Carvalho de Melo <acme@redhat.com> CommitDate: Mon, 1 Apr 2019 14:49:24 -0300 perf stat: Revert checks for duration_time This reverts e864c5ca145e ("perf stat: Hide internal duration_time counter") but doing it manually since the code has now moved to a different file. The next patch will properly implement duration_time as a full event, so no need to hide it anymore. Signed-off-by: Andi Kleen <ak@linux.intel.com> Acked-by: Jiri Olsa <jolsa@kernel.org> Link: http://lkml.kernel.org/r/20190326221823.11518-2-andi@firstfloor.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> --- tools/perf/util/stat-display.c | 18 ------------------ 1 file changed, 18 deletions(-) diff --git a/tools/perf/util/stat-display.c b/tools/perf/util/stat-display.c index 6d043c78f3c2..3324f23c7efc 100644 --- a/tools/perf/util/stat-display.c +++ b/tools/perf/util/stat-display.c @@ -18,11 +18,6 @@ #define CNTR_NOT_SUPPORTED "<not supported>" #define CNTR_NOT_COUNTED "<not counted>" -static bool is_duration_time(struct perf_evsel *evsel) -{ - return !strcmp(evsel->name, "duration_time"); -} - static void print_running(struct perf_stat_config *config, u64 run, u64 ena) { @@ -628,9 +623,6 @@ static void print_aggr(struct perf_stat_config *config, ad.id = id = config->aggr_map->map[s]; first = true; evlist__for_each_entry(evlist, counter) { - if (is_duration_time(counter)) - continue; - ad.val = ad.ena = ad.run = 0; ad.nr = 0; if (!collect_data(config, counter, aggr_cb, &ad)) @@ -848,8 +840,6 @@ static void print_no_aggr_metric(struct perf_stat_config *config, if (prefix) fputs(prefix, config->output); evlist__for_each_entry(evlist, counter) { - if (is_duration_time(counter)) - continue; if (first) { aggr_printout(config, counter, cpu, 0); first = false; @@ -906,8 +896,6 @@ static void print_metric_headers(struct perf_stat_config *config, /* Print metrics headers only */ evlist__for_each_entry(evlist, counter) { - if (is_duration_time(counter)) - continue; os.evsel = counter; out.ctx = &os; out.print_metric = print_metric_header; @@ -1136,15 +1124,11 @@ perf_evlist__print_counters(struct perf_evlist *evlist, break; case AGGR_THREAD: evlist__for_each_entry(evlist, counter) { - if (is_duration_time(counter)) - continue; print_aggr_thread(config, _target, counter, prefix); } break; case AGGR_GLOBAL: evlist__for_each_entry(evlist, counter) { - if (is_duration_time(counter)) - continue; print_counter_aggr(config, counter, prefix); } if (metric_only) @@ -1155,8 +1139,6 @@ perf_evlist__print_counters(struct perf_evlist *evlist, print_no_aggr_metric(config, evlist, prefix); else { evlist__for_each_entry(evlist, counter) { - if (is_duration_time(counter)) - continue; print_counter(config, counter, prefix); } } ^ permalink raw reply related [flat|nested] 12+ messages in thread
* [PATCH 2/4] perf, tools: Implement duration_time as a proper event 2019-03-26 22:18 perf: Add duration_time to perf stat Andi Kleen 2019-03-26 22:18 ` [PATCH 1/4] perf, tools: Revert checks for duration_time Andi Kleen @ 2019-03-26 22:18 ` Andi Kleen 2019-03-27 8:58 ` Jiri Olsa 2019-04-05 11:39 ` [tip:perf/core] perf stat: " tip-bot for Andi Kleen 2019-03-26 22:18 ` [PATCH 3/4] perf, tools: Support name for duration_time Andi Kleen ` (2 subsequent siblings) 4 siblings, 2 replies; 12+ messages in thread From: Andi Kleen @ 2019-03-26 22:18 UTC (permalink / raw) To: acme; +Cc: jolsa, linux-kernel, linux-perf-users, Andi Kleen From: Andi Kleen <ak@linux.intel.com> The perf metric expression use duration_time internally to normalize events. Normal perf stat without -x also prints the duration time. But when using -x, the interval is not output anywhere, which is inconvenient for any post processing which often wants to normalize values to time. So implement duration_time as a proper perf event that can be specified explicitely with -e. The previous implementation of duration_time only worked for metric processing. This adds the concept of a tool event that is handled by the tool. On the kernel level it is still mapped to the dummy software event, but the values are not read anymore, but instead computed by the tool. Add proper plumbing to handle this in the event parser, and display it in perf stat. We don't want duration_time to be added up, so it's only printed for the first CPU. % perf stat -e duration_time,cycles true Performance counter stats for 'true': 555,476 ns duration_time 771,958 cycles 0.000555476 seconds time elapsed 0.000644000 seconds user 0.000000000 seconds sys Signed-off-by: Andi Kleen <ak@linux.intel.com> --- tools/perf/builtin-stat.c | 28 ++++++++++++++++++------- tools/perf/util/evsel.h | 6 ++++++ tools/perf/util/parse-events.c | 38 +++++++++++++++++++++++++++++----- tools/perf/util/parse-events.h | 4 ++++ tools/perf/util/parse-events.l | 11 +++++++++- tools/perf/util/parse-events.y | 12 +++++++++++ 6 files changed, 86 insertions(+), 13 deletions(-) diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c index 49ee3c2033ec..7f9c4b7f5d69 100644 --- a/tools/perf/builtin-stat.c +++ b/tools/perf/builtin-stat.c @@ -244,11 +244,25 @@ perf_evsel__write_stat_event(struct perf_evsel *counter, u32 cpu, u32 thread, process_synthesized_event, NULL); } +static int read_single_counter(struct perf_evsel *counter, int cpu, + int thread, struct timespec *rs) +{ + if (counter->tool_event == PERF_TOOL_DURATION_TIME) { + u64 val = rs->tv_nsec + rs->tv_sec*1000000000ULL; + struct perf_counts_values *count = + perf_counts(counter->counts, cpu, thread); + count->ena = count->run = val; + count->val = val; + return 0; + } + return perf_evsel__read_counter(counter, cpu, thread); +} + /* * Read out the results of a single counter: * do not aggregate counts across CPUs in system-wide mode */ -static int read_counter(struct perf_evsel *counter) +static int read_counter(struct perf_evsel *counter, struct timespec *rs) { int nthreads = thread_map__nr(evsel_list->threads); int ncpus, cpu, thread; @@ -275,7 +289,7 @@ static int read_counter(struct perf_evsel *counter) * (via perf_evsel__read_counter) and sets threir count->loaded. */ if (!count->loaded && - perf_evsel__read_counter(counter, cpu, thread)) { + read_single_counter(counter, cpu, thread, rs)) { counter->counts->scaled = -1; perf_counts(counter->counts, cpu, thread)->ena = 0; perf_counts(counter->counts, cpu, thread)->run = 0; @@ -304,13 +318,13 @@ static int read_counter(struct perf_evsel *counter) return 0; } -static void read_counters(void) +static void read_counters(struct timespec *rs) { struct perf_evsel *counter; int ret; evlist__for_each_entry(evsel_list, counter) { - ret = read_counter(counter); + ret = read_counter(counter, rs); if (ret) pr_debug("failed to read counter %s\n", counter->name); @@ -323,11 +337,11 @@ static void process_interval(void) { struct timespec ts, rs; - read_counters(); - clock_gettime(CLOCK_MONOTONIC, &ts); diff_timespec(&rs, &ts, &ref_time); + read_counters(&rs); + if (STAT_RECORD) { if (WRITE_STAT_ROUND_EVENT(rs.tv_sec * NSEC_PER_SEC + rs.tv_nsec, INTERVAL)) pr_err("failed to write stat round event\n"); @@ -593,7 +607,7 @@ static int __run_perf_stat(int argc, const char **argv, int run_idx) * avoid arbitrary skew, we must read all counters before closing any * group leaders. */ - read_counters(); + read_counters(&(struct timespec) { .tv_nsec = t1-t0 }); perf_evlist__close(evsel_list); return WEXITSTATUS(status); diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h index 0f2c6c93d721..6d190cbf1070 100644 --- a/tools/perf/util/evsel.h +++ b/tools/perf/util/evsel.h @@ -75,6 +75,11 @@ struct perf_stat_evsel; typedef int (perf_evsel__sb_cb_t)(union perf_event *event, void *data); +enum perf_tool_event { + PERF_TOOL_NONE = 0, + PERF_TOOL_DURATION_TIME = 1, +}; + /** struct perf_evsel - event selector * * @evlist - evlist this evsel is in, if it is in one. @@ -121,6 +126,7 @@ struct perf_evsel { unsigned int sample_size; int id_pos; int is_pos; + enum perf_tool_event tool_event; bool uniquified_name; bool snapshot; bool supported; diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c index 5ef4939408f2..98c0fadaedb9 100644 --- a/tools/perf/util/parse-events.c +++ b/tools/perf/util/parse-events.c @@ -317,10 +317,12 @@ static struct perf_evsel * __add_event(struct list_head *list, int *idx, struct perf_event_attr *attr, char *name, struct perf_pmu *pmu, - struct list_head *config_terms, bool auto_merge_stats) + struct list_head *config_terms, bool auto_merge_stats, + const char *cpu_list) { struct perf_evsel *evsel; - struct cpu_map *cpus = pmu ? pmu->cpus : NULL; + struct cpu_map *cpus = pmu ? pmu->cpus : + cpu_list ? cpu_map__new(cpu_list) : NULL; event_attr_init(attr); @@ -348,7 +350,25 @@ static int add_event(struct list_head *list, int *idx, struct perf_event_attr *attr, char *name, struct list_head *config_terms) { - return __add_event(list, idx, attr, name, NULL, config_terms, false) ? 0 : -ENOMEM; + return __add_event(list, idx, attr, name, NULL, config_terms, false, NULL) ? 0 : -ENOMEM; +} + +static int add_event_tool(struct list_head *list, int *idx, + enum perf_tool_event tool_event) +{ + struct perf_evsel *evsel; + struct perf_event_attr attr = { + .type = PERF_TYPE_SOFTWARE, + .config = PERF_COUNT_SW_DUMMY, + }; + + evsel = __add_event(list, idx, &attr, NULL, NULL, NULL, false, "0"); + if (!evsel) + return -ENOMEM; + evsel->tool_event = tool_event; + if (tool_event == PERF_TOOL_DURATION_TIME) + evsel->unit = strdup("ns"); + return 0; } static int parse_aliases(char *str, const char *names[][PERF_EVSEL__MAX_ALIASES], int size) @@ -1233,6 +1253,13 @@ int parse_events_add_numeric(struct parse_events_state *parse_state, get_config_name(head_config), &config_terms); } +int parse_events_add_tool(struct parse_events_state *parse_state, + struct list_head *list, + enum perf_tool_event tool_event) +{ + return add_event_tool(list, &parse_state->idx, tool_event); +} + int parse_events_add_pmu(struct parse_events_state *parse_state, struct list_head *list, char *name, struct list_head *head_config, @@ -1267,7 +1294,8 @@ int parse_events_add_pmu(struct parse_events_state *parse_state, if (!head_config) { attr.type = pmu->type; - evsel = __add_event(list, &parse_state->idx, &attr, NULL, pmu, NULL, auto_merge_stats); + evsel = __add_event(list, &parse_state->idx, &attr, NULL, pmu, NULL, + auto_merge_stats, NULL); if (evsel) { evsel->pmu_name = name; evsel->use_uncore_alias = use_uncore_alias; @@ -1295,7 +1323,7 @@ int parse_events_add_pmu(struct parse_events_state *parse_state, evsel = __add_event(list, &parse_state->idx, &attr, get_config_name(head_config), pmu, - &config_terms, auto_merge_stats); + &config_terms, auto_merge_stats, NULL); if (evsel) { evsel->unit = info.unit; evsel->scale = info.scale; diff --git a/tools/perf/util/parse-events.h b/tools/perf/util/parse-events.h index 5ed035cbcbb7..0c1f5b98f636 100644 --- a/tools/perf/util/parse-events.h +++ b/tools/perf/util/parse-events.h @@ -160,6 +160,10 @@ int parse_events_add_numeric(struct parse_events_state *parse_state, struct list_head *list, u32 type, u64 config, struct list_head *head_config); +enum perf_tool_event; +int parse_events_add_tool(struct parse_events_state *parse_state, + struct list_head *list, + enum perf_tool_event tool_event); int parse_events_add_cache(struct list_head *list, int *idx, char *type, char *op_result1, char *op_result2, struct parse_events_error *error, diff --git a/tools/perf/util/parse-events.l b/tools/perf/util/parse-events.l index 7805c71aaae2..c54bfe88626c 100644 --- a/tools/perf/util/parse-events.l +++ b/tools/perf/util/parse-events.l @@ -15,6 +15,7 @@ #include "../perf.h" #include "parse-events.h" #include "parse-events-bison.h" +#include "evsel.h" char *parse_events_get_text(yyscan_t yyscanner); YYSTYPE *parse_events_get_lval(yyscan_t yyscanner); @@ -154,6 +155,14 @@ static int sym(yyscan_t scanner, int type, int config) return type == PERF_TYPE_HARDWARE ? PE_VALUE_SYM_HW : PE_VALUE_SYM_SW; } +static int tool(yyscan_t scanner, enum perf_tool_event event) +{ + YYSTYPE *yylval = parse_events_get_lval(scanner); + + yylval->num = event; + return PE_VALUE_SYM_TOOL; +} + static int term(yyscan_t scanner, int type) { YYSTYPE *yylval = parse_events_get_lval(scanner); @@ -322,7 +331,7 @@ cpu-migrations|migrations { return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COU alignment-faults { return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COUNT_SW_ALIGNMENT_FAULTS); } emulation-faults { return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COUNT_SW_EMULATION_FAULTS); } dummy { return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COUNT_SW_DUMMY); } -duration_time { return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COUNT_SW_DUMMY); } +duration_time { return tool(yyscanner, PERF_TOOL_DURATION_TIME); } bpf-output { return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COUNT_SW_BPF_OUTPUT); } /* diff --git a/tools/perf/util/parse-events.y b/tools/perf/util/parse-events.y index 44819bdb037d..6ad8d4914969 100644 --- a/tools/perf/util/parse-events.y +++ b/tools/perf/util/parse-events.y @@ -14,6 +14,7 @@ #include <linux/types.h> #include "util.h" #include "pmu.h" +#include "evsel.h" #include "debug.h" #include "parse-events.h" #include "parse-events-bison.h" @@ -45,6 +46,7 @@ static void inc_group_count(struct list_head *list, %token PE_START_EVENTS PE_START_TERMS %token PE_VALUE PE_VALUE_SYM_HW PE_VALUE_SYM_SW PE_RAW PE_TERM +%token PE_VALUE_SYM_TOOL %token PE_EVENT_NAME %token PE_NAME %token PE_BPF_OBJECT PE_BPF_SOURCE @@ -58,6 +60,7 @@ static void inc_group_count(struct list_head *list, %type <num> PE_VALUE %type <num> PE_VALUE_SYM_HW %type <num> PE_VALUE_SYM_SW +%type <num> PE_VALUE_SYM_TOOL %type <num> PE_RAW %type <num> PE_TERM %type <str> PE_NAME @@ -321,6 +324,15 @@ value_sym sep_slash_slash_dc ABORT_ON(parse_events_add_numeric(_parse_state, list, type, config, NULL)); $$ = list; } +| +PE_VALUE_SYM_TOOL sep_slash_slash_dc +{ + struct list_head *list; + + ALLOC_LIST(list); + ABORT_ON(parse_events_add_tool(_parse_state, list, $1)); + $$ = list; +} event_legacy_cache: PE_NAME_CACHE_TYPE '-' PE_NAME_CACHE_OP_RESULT '-' PE_NAME_CACHE_OP_RESULT opt_event_config -- 2.20.1 ^ permalink raw reply related [flat|nested] 12+ messages in thread
* Re: [PATCH 2/4] perf, tools: Implement duration_time as a proper event 2019-03-26 22:18 ` [PATCH 2/4] perf, tools: Implement duration_time as a proper event Andi Kleen @ 2019-03-27 8:58 ` Jiri Olsa 2019-03-27 13:54 ` Andi Kleen 2019-04-05 11:39 ` [tip:perf/core] perf stat: " tip-bot for Andi Kleen 1 sibling, 1 reply; 12+ messages in thread From: Jiri Olsa @ 2019-03-27 8:58 UTC (permalink / raw) To: Andi Kleen; +Cc: acme, jolsa, linux-kernel, linux-perf-users, Andi Kleen On Tue, Mar 26, 2019 at 03:18:21PM -0700, Andi Kleen wrote: > From: Andi Kleen <ak@linux.intel.com> > > The perf metric expression use duration_time internally to normalize > events. Normal perf stat without -x also prints the duration time. > But when using -x, the interval is not output anywhere, which > is inconvenient for any post processing which often wants to > normalize values to time. > > So implement duration_time as a proper perf event that > can be specified explicitely with -e. > > The previous implementation of duration_time only worked > for metric processing. This adds the concept of a tool > event that is handled by the tool. On the kernel level > it is still mapped to the dummy software event, but the > values are not read anymore, but instead computed by the tool. > > Add proper plumbing to handle this in the event parser, > and display it in perf stat. We don't want duration_time to be added up, > so it's only printed for the first CPU. > > % perf stat -e duration_time,cycles true > > Performance counter stats for 'true': > > 555,476 ns duration_time > 771,958 cycles > > 0.000555476 seconds time elapsed > > 0.000644000 seconds user > 0.000000000 seconds sys > > Signed-off-by: Andi Kleen <ak@linux.intel.com> > --- > tools/perf/builtin-stat.c | 28 ++++++++++++++++++------- > tools/perf/util/evsel.h | 6 ++++++ > tools/perf/util/parse-events.c | 38 +++++++++++++++++++++++++++++----- > tools/perf/util/parse-events.h | 4 ++++ > tools/perf/util/parse-events.l | 11 +++++++++- > tools/perf/util/parse-events.y | 12 +++++++++++ > 6 files changed, 86 insertions(+), 13 deletions(-) > > diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c > index 49ee3c2033ec..7f9c4b7f5d69 100644 > --- a/tools/perf/builtin-stat.c > +++ b/tools/perf/builtin-stat.c > @@ -244,11 +244,25 @@ perf_evsel__write_stat_event(struct perf_evsel *counter, u32 cpu, u32 thread, > process_synthesized_event, NULL); > } > > +static int read_single_counter(struct perf_evsel *counter, int cpu, > + int thread, struct timespec *rs) > +{ > + if (counter->tool_event == PERF_TOOL_DURATION_TIME) { > + u64 val = rs->tv_nsec + rs->tv_sec*1000000000ULL; > + struct perf_counts_values *count = > + perf_counts(counter->counts, cpu, thread); > + count->ena = count->run = val; > + count->val = val; > + return 0; > + } > + return perf_evsel__read_counter(counter, cpu, thread); > +} so now that we have time in a separate event, we could get rid of the isolated update_stats(&walltime_nsecs_stats) calls and move them to perf_stat__update_shadow_stats? jirka ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH 2/4] perf, tools: Implement duration_time as a proper event 2019-03-27 8:58 ` Jiri Olsa @ 2019-03-27 13:54 ` Andi Kleen 0 siblings, 0 replies; 12+ messages in thread From: Andi Kleen @ 2019-03-27 13:54 UTC (permalink / raw) To: Jiri Olsa; +Cc: Andi Kleen, acme, jolsa, linux-kernel, linux-perf-users > > so now that we have time in a separate event, we could > get rid of the isolated update_stats(&walltime_nsecs_stats) calls > and move them to perf_stat__update_shadow_stats? In theory yes, but it would require perf stat always setting up the duration_time event implicitely and then special handling. For stat it would need to be hidden to keep the same output. It's probably simpler to keep the old way for that. -Andi ^ permalink raw reply [flat|nested] 12+ messages in thread
* [tip:perf/core] perf stat: Implement duration_time as a proper event 2019-03-26 22:18 ` [PATCH 2/4] perf, tools: Implement duration_time as a proper event Andi Kleen 2019-03-27 8:58 ` Jiri Olsa @ 2019-04-05 11:39 ` tip-bot for Andi Kleen 1 sibling, 0 replies; 12+ messages in thread From: tip-bot for Andi Kleen @ 2019-04-05 11:39 UTC (permalink / raw) To: linux-tip-commits; +Cc: linux-kernel, jolsa, ak, acme, mingo, tglx, hpa Commit-ID: f0fbb114e3025f3f737a1e1c5c39c5b2b2e671bd Gitweb: https://git.kernel.org/tip/f0fbb114e3025f3f737a1e1c5c39c5b2b2e671bd Author: Andi Kleen <ak@linux.intel.com> AuthorDate: Tue, 26 Mar 2019 15:18:21 -0700 Committer: Arnaldo Carvalho de Melo <acme@redhat.com> CommitDate: Mon, 1 Apr 2019 14:49:24 -0300 perf stat: Implement duration_time as a proper event The perf metric expression use 'duration_time' internally to normalize events. Normal 'perf stat' without -x also prints the duration time. But when using -x, the interval is not output anywhere, which is inconvenient for any post processing which often wants to normalize values to time. So implement 'duration_time' as a proper perf event that can be specified explicitely with -e. The previous implementation of 'duration_time' only worked for metric processing. This adds the concept of a tool event that is handled by the tool. On the kernel level it is still mapped to the dummy software event, but the values are not read anymore, but instead computed by the tool. Add proper plumbing to handle this in the event parser, and display it in 'perf stat'. We don't want 'duration_time' to be added up, so it's only printed for the first CPU. % perf stat -e duration_time,cycles true Performance counter stats for 'true': 555,476 ns duration_time 771,958 cycles 0.000555476 seconds time elapsed 0.000644000 seconds user 0.000000000 seconds sys Signed-off-by: Andi Kleen <ak@linux.intel.com> Acked-by: Jiri Olsa <jolsa@kernel.org> Link: http://lkml.kernel.org/r/20190326221823.11518-3-andi@firstfloor.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> --- tools/perf/builtin-stat.c | 28 +++++++++++++++++++++------- tools/perf/util/evsel.h | 6 ++++++ tools/perf/util/parse-events.c | 38 +++++++++++++++++++++++++++++++++----- tools/perf/util/parse-events.h | 4 ++++ tools/perf/util/parse-events.l | 11 ++++++++++- tools/perf/util/parse-events.y | 12 ++++++++++++ 6 files changed, 86 insertions(+), 13 deletions(-) diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c index 49ee3c2033ec..7f9c4b7f5d69 100644 --- a/tools/perf/builtin-stat.c +++ b/tools/perf/builtin-stat.c @@ -244,11 +244,25 @@ perf_evsel__write_stat_event(struct perf_evsel *counter, u32 cpu, u32 thread, process_synthesized_event, NULL); } +static int read_single_counter(struct perf_evsel *counter, int cpu, + int thread, struct timespec *rs) +{ + if (counter->tool_event == PERF_TOOL_DURATION_TIME) { + u64 val = rs->tv_nsec + rs->tv_sec*1000000000ULL; + struct perf_counts_values *count = + perf_counts(counter->counts, cpu, thread); + count->ena = count->run = val; + count->val = val; + return 0; + } + return perf_evsel__read_counter(counter, cpu, thread); +} + /* * Read out the results of a single counter: * do not aggregate counts across CPUs in system-wide mode */ -static int read_counter(struct perf_evsel *counter) +static int read_counter(struct perf_evsel *counter, struct timespec *rs) { int nthreads = thread_map__nr(evsel_list->threads); int ncpus, cpu, thread; @@ -275,7 +289,7 @@ static int read_counter(struct perf_evsel *counter) * (via perf_evsel__read_counter) and sets threir count->loaded. */ if (!count->loaded && - perf_evsel__read_counter(counter, cpu, thread)) { + read_single_counter(counter, cpu, thread, rs)) { counter->counts->scaled = -1; perf_counts(counter->counts, cpu, thread)->ena = 0; perf_counts(counter->counts, cpu, thread)->run = 0; @@ -304,13 +318,13 @@ static int read_counter(struct perf_evsel *counter) return 0; } -static void read_counters(void) +static void read_counters(struct timespec *rs) { struct perf_evsel *counter; int ret; evlist__for_each_entry(evsel_list, counter) { - ret = read_counter(counter); + ret = read_counter(counter, rs); if (ret) pr_debug("failed to read counter %s\n", counter->name); @@ -323,11 +337,11 @@ static void process_interval(void) { struct timespec ts, rs; - read_counters(); - clock_gettime(CLOCK_MONOTONIC, &ts); diff_timespec(&rs, &ts, &ref_time); + read_counters(&rs); + if (STAT_RECORD) { if (WRITE_STAT_ROUND_EVENT(rs.tv_sec * NSEC_PER_SEC + rs.tv_nsec, INTERVAL)) pr_err("failed to write stat round event\n"); @@ -593,7 +607,7 @@ try_again: * avoid arbitrary skew, we must read all counters before closing any * group leaders. */ - read_counters(); + read_counters(&(struct timespec) { .tv_nsec = t1-t0 }); perf_evlist__close(evsel_list); return WEXITSTATUS(status); diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h index 0f2c6c93d721..6d190cbf1070 100644 --- a/tools/perf/util/evsel.h +++ b/tools/perf/util/evsel.h @@ -75,6 +75,11 @@ struct perf_stat_evsel; typedef int (perf_evsel__sb_cb_t)(union perf_event *event, void *data); +enum perf_tool_event { + PERF_TOOL_NONE = 0, + PERF_TOOL_DURATION_TIME = 1, +}; + /** struct perf_evsel - event selector * * @evlist - evlist this evsel is in, if it is in one. @@ -121,6 +126,7 @@ struct perf_evsel { unsigned int sample_size; int id_pos; int is_pos; + enum perf_tool_event tool_event; bool uniquified_name; bool snapshot; bool supported; diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c index 5ef4939408f2..98c0fadaedb9 100644 --- a/tools/perf/util/parse-events.c +++ b/tools/perf/util/parse-events.c @@ -317,10 +317,12 @@ static struct perf_evsel * __add_event(struct list_head *list, int *idx, struct perf_event_attr *attr, char *name, struct perf_pmu *pmu, - struct list_head *config_terms, bool auto_merge_stats) + struct list_head *config_terms, bool auto_merge_stats, + const char *cpu_list) { struct perf_evsel *evsel; - struct cpu_map *cpus = pmu ? pmu->cpus : NULL; + struct cpu_map *cpus = pmu ? pmu->cpus : + cpu_list ? cpu_map__new(cpu_list) : NULL; event_attr_init(attr); @@ -348,7 +350,25 @@ static int add_event(struct list_head *list, int *idx, struct perf_event_attr *attr, char *name, struct list_head *config_terms) { - return __add_event(list, idx, attr, name, NULL, config_terms, false) ? 0 : -ENOMEM; + return __add_event(list, idx, attr, name, NULL, config_terms, false, NULL) ? 0 : -ENOMEM; +} + +static int add_event_tool(struct list_head *list, int *idx, + enum perf_tool_event tool_event) +{ + struct perf_evsel *evsel; + struct perf_event_attr attr = { + .type = PERF_TYPE_SOFTWARE, + .config = PERF_COUNT_SW_DUMMY, + }; + + evsel = __add_event(list, idx, &attr, NULL, NULL, NULL, false, "0"); + if (!evsel) + return -ENOMEM; + evsel->tool_event = tool_event; + if (tool_event == PERF_TOOL_DURATION_TIME) + evsel->unit = strdup("ns"); + return 0; } static int parse_aliases(char *str, const char *names[][PERF_EVSEL__MAX_ALIASES], int size) @@ -1233,6 +1253,13 @@ int parse_events_add_numeric(struct parse_events_state *parse_state, get_config_name(head_config), &config_terms); } +int parse_events_add_tool(struct parse_events_state *parse_state, + struct list_head *list, + enum perf_tool_event tool_event) +{ + return add_event_tool(list, &parse_state->idx, tool_event); +} + int parse_events_add_pmu(struct parse_events_state *parse_state, struct list_head *list, char *name, struct list_head *head_config, @@ -1267,7 +1294,8 @@ int parse_events_add_pmu(struct parse_events_state *parse_state, if (!head_config) { attr.type = pmu->type; - evsel = __add_event(list, &parse_state->idx, &attr, NULL, pmu, NULL, auto_merge_stats); + evsel = __add_event(list, &parse_state->idx, &attr, NULL, pmu, NULL, + auto_merge_stats, NULL); if (evsel) { evsel->pmu_name = name; evsel->use_uncore_alias = use_uncore_alias; @@ -1295,7 +1323,7 @@ int parse_events_add_pmu(struct parse_events_state *parse_state, evsel = __add_event(list, &parse_state->idx, &attr, get_config_name(head_config), pmu, - &config_terms, auto_merge_stats); + &config_terms, auto_merge_stats, NULL); if (evsel) { evsel->unit = info.unit; evsel->scale = info.scale; diff --git a/tools/perf/util/parse-events.h b/tools/perf/util/parse-events.h index 5ed035cbcbb7..0c1f5b98f636 100644 --- a/tools/perf/util/parse-events.h +++ b/tools/perf/util/parse-events.h @@ -160,6 +160,10 @@ int parse_events_add_numeric(struct parse_events_state *parse_state, struct list_head *list, u32 type, u64 config, struct list_head *head_config); +enum perf_tool_event; +int parse_events_add_tool(struct parse_events_state *parse_state, + struct list_head *list, + enum perf_tool_event tool_event); int parse_events_add_cache(struct list_head *list, int *idx, char *type, char *op_result1, char *op_result2, struct parse_events_error *error, diff --git a/tools/perf/util/parse-events.l b/tools/perf/util/parse-events.l index 7805c71aaae2..c54bfe88626c 100644 --- a/tools/perf/util/parse-events.l +++ b/tools/perf/util/parse-events.l @@ -15,6 +15,7 @@ #include "../perf.h" #include "parse-events.h" #include "parse-events-bison.h" +#include "evsel.h" char *parse_events_get_text(yyscan_t yyscanner); YYSTYPE *parse_events_get_lval(yyscan_t yyscanner); @@ -154,6 +155,14 @@ static int sym(yyscan_t scanner, int type, int config) return type == PERF_TYPE_HARDWARE ? PE_VALUE_SYM_HW : PE_VALUE_SYM_SW; } +static int tool(yyscan_t scanner, enum perf_tool_event event) +{ + YYSTYPE *yylval = parse_events_get_lval(scanner); + + yylval->num = event; + return PE_VALUE_SYM_TOOL; +} + static int term(yyscan_t scanner, int type) { YYSTYPE *yylval = parse_events_get_lval(scanner); @@ -322,7 +331,7 @@ cpu-migrations|migrations { return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COU alignment-faults { return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COUNT_SW_ALIGNMENT_FAULTS); } emulation-faults { return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COUNT_SW_EMULATION_FAULTS); } dummy { return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COUNT_SW_DUMMY); } -duration_time { return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COUNT_SW_DUMMY); } +duration_time { return tool(yyscanner, PERF_TOOL_DURATION_TIME); } bpf-output { return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COUNT_SW_BPF_OUTPUT); } /* diff --git a/tools/perf/util/parse-events.y b/tools/perf/util/parse-events.y index 44819bdb037d..6ad8d4914969 100644 --- a/tools/perf/util/parse-events.y +++ b/tools/perf/util/parse-events.y @@ -14,6 +14,7 @@ #include <linux/types.h> #include "util.h" #include "pmu.h" +#include "evsel.h" #include "debug.h" #include "parse-events.h" #include "parse-events-bison.h" @@ -45,6 +46,7 @@ static void inc_group_count(struct list_head *list, %token PE_START_EVENTS PE_START_TERMS %token PE_VALUE PE_VALUE_SYM_HW PE_VALUE_SYM_SW PE_RAW PE_TERM +%token PE_VALUE_SYM_TOOL %token PE_EVENT_NAME %token PE_NAME %token PE_BPF_OBJECT PE_BPF_SOURCE @@ -58,6 +60,7 @@ static void inc_group_count(struct list_head *list, %type <num> PE_VALUE %type <num> PE_VALUE_SYM_HW %type <num> PE_VALUE_SYM_SW +%type <num> PE_VALUE_SYM_TOOL %type <num> PE_RAW %type <num> PE_TERM %type <str> PE_NAME @@ -321,6 +324,15 @@ value_sym sep_slash_slash_dc ABORT_ON(parse_events_add_numeric(_parse_state, list, type, config, NULL)); $$ = list; } +| +PE_VALUE_SYM_TOOL sep_slash_slash_dc +{ + struct list_head *list; + + ALLOC_LIST(list); + ABORT_ON(parse_events_add_tool(_parse_state, list, $1)); + $$ = list; +} event_legacy_cache: PE_NAME_CACHE_TYPE '-' PE_NAME_CACHE_OP_RESULT '-' PE_NAME_CACHE_OP_RESULT opt_event_config ^ permalink raw reply related [flat|nested] 12+ messages in thread
* [PATCH 3/4] perf, tools: Support name for duration_time 2019-03-26 22:18 perf: Add duration_time to perf stat Andi Kleen 2019-03-26 22:18 ` [PATCH 1/4] perf, tools: Revert checks for duration_time Andi Kleen 2019-03-26 22:18 ` [PATCH 2/4] perf, tools: Implement duration_time as a proper event Andi Kleen @ 2019-03-26 22:18 ` Andi Kleen 2019-04-05 11:39 ` [tip:perf/core] perf evsel: Support printing evsel name for 'duration_time' tip-bot for Andi Kleen 2019-03-26 22:18 ` [PATCH 4/4] perf, tools, list: Output tool events Andi Kleen 2019-03-27 19:51 ` perf: Add duration_time to perf stat Jiri Olsa 4 siblings, 1 reply; 12+ messages in thread From: Andi Kleen @ 2019-03-26 22:18 UTC (permalink / raw) To: acme; +Cc: jolsa, linux-kernel, linux-perf-users, Andi Kleen From: Andi Kleen <ak@linux.intel.com> Implement printing the correct name for duration_time Signed-off-by: Andi Kleen <ak@linux.intel.com> --- tools/perf/util/evsel.c | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-) diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c index 7835e05f0c0a..6872394b6f55 100644 --- a/tools/perf/util/evsel.c +++ b/tools/perf/util/evsel.c @@ -579,6 +579,12 @@ static int perf_evsel__raw_name(struct perf_evsel *evsel, char *bf, size_t size) return ret + perf_evsel__add_modifiers(evsel, bf + ret, size - ret); } +static int perf_evsel__tool_name(char *bf, size_t size) +{ + int ret = scnprintf(bf, size, "duration_time"); + return ret; +} + const char *perf_evsel__name(struct perf_evsel *evsel) { char bf[128]; @@ -600,7 +606,10 @@ const char *perf_evsel__name(struct perf_evsel *evsel) break; case PERF_TYPE_SOFTWARE: - perf_evsel__sw_name(evsel, bf, sizeof(bf)); + if (evsel->tool_event) + perf_evsel__tool_name(bf, sizeof(bf)); + else + perf_evsel__sw_name(evsel, bf, sizeof(bf)); break; case PERF_TYPE_TRACEPOINT: -- 2.20.1 ^ permalink raw reply related [flat|nested] 12+ messages in thread
* [tip:perf/core] perf evsel: Support printing evsel name for 'duration_time' 2019-03-26 22:18 ` [PATCH 3/4] perf, tools: Support name for duration_time Andi Kleen @ 2019-04-05 11:39 ` tip-bot for Andi Kleen 0 siblings, 0 replies; 12+ messages in thread From: tip-bot for Andi Kleen @ 2019-04-05 11:39 UTC (permalink / raw) To: linux-tip-commits; +Cc: mingo, acme, linux-kernel, tglx, jolsa, ak, hpa Commit-ID: 3371f389e4be6efc496ca395b21911a8f2c2d23f Gitweb: https://git.kernel.org/tip/3371f389e4be6efc496ca395b21911a8f2c2d23f Author: Andi Kleen <ak@linux.intel.com> AuthorDate: Tue, 26 Mar 2019 15:18:22 -0700 Committer: Arnaldo Carvalho de Melo <acme@redhat.com> CommitDate: Mon, 1 Apr 2019 14:49:24 -0300 perf evsel: Support printing evsel name for 'duration_time' Implement printing the correct name for duration_time Signed-off-by: Andi Kleen <ak@linux.intel.com> Acked-by: Jiri Olsa <jolsa@kernel.org> Link: http://lkml.kernel.org/r/20190326221823.11518-4-andi@firstfloor.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> --- tools/perf/util/evsel.c | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-) diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c index 66d066f18b5b..84cfb9fe2fc6 100644 --- a/tools/perf/util/evsel.c +++ b/tools/perf/util/evsel.c @@ -580,6 +580,12 @@ static int perf_evsel__raw_name(struct perf_evsel *evsel, char *bf, size_t size) return ret + perf_evsel__add_modifiers(evsel, bf + ret, size - ret); } +static int perf_evsel__tool_name(char *bf, size_t size) +{ + int ret = scnprintf(bf, size, "duration_time"); + return ret; +} + const char *perf_evsel__name(struct perf_evsel *evsel) { char bf[128]; @@ -601,7 +607,10 @@ const char *perf_evsel__name(struct perf_evsel *evsel) break; case PERF_TYPE_SOFTWARE: - perf_evsel__sw_name(evsel, bf, sizeof(bf)); + if (evsel->tool_event) + perf_evsel__tool_name(bf, sizeof(bf)); + else + perf_evsel__sw_name(evsel, bf, sizeof(bf)); break; case PERF_TYPE_TRACEPOINT: ^ permalink raw reply related [flat|nested] 12+ messages in thread
* [PATCH 4/4] perf, tools, list: Output tool events 2019-03-26 22:18 perf: Add duration_time to perf stat Andi Kleen ` (2 preceding siblings ...) 2019-03-26 22:18 ` [PATCH 3/4] perf, tools: Support name for duration_time Andi Kleen @ 2019-03-26 22:18 ` Andi Kleen 2019-04-05 11:40 ` [tip:perf/core] perf " tip-bot for Andi Kleen 2019-03-27 19:51 ` perf: Add duration_time to perf stat Jiri Olsa 4 siblings, 1 reply; 12+ messages in thread From: Andi Kleen @ 2019-03-26 22:18 UTC (permalink / raw) To: acme; +Cc: jolsa, linux-kernel, linux-perf-users, Andi Kleen From: Andi Kleen <ak@linux.intel.com> Add support to perf list to output tool internal events, currently only duration_time. Signed-off-by: Andi Kleen <ak@linux.intel.com> --- tools/perf/builtin-list.c | 6 ++++-- tools/perf/util/parse-events.c | 20 ++++++++++++++++++++ tools/perf/util/parse-events.h | 1 + 3 files changed, 25 insertions(+), 2 deletions(-) diff --git a/tools/perf/builtin-list.c b/tools/perf/builtin-list.c index a8394b4f1167..e0312a1c4792 100644 --- a/tools/perf/builtin-list.c +++ b/tools/perf/builtin-list.c @@ -70,10 +70,11 @@ int cmd_list(int argc, const char **argv) print_symbol_events(NULL, PERF_TYPE_HARDWARE, event_symbols_hw, PERF_COUNT_HW_MAX, raw_dump); else if (strcmp(argv[i], "sw") == 0 || - strcmp(argv[i], "software") == 0) + strcmp(argv[i], "software") == 0) { print_symbol_events(NULL, PERF_TYPE_SOFTWARE, event_symbols_sw, PERF_COUNT_SW_MAX, raw_dump); - else if (strcmp(argv[i], "cache") == 0 || + print_tool_events(NULL, raw_dump); + } else if (strcmp(argv[i], "cache") == 0 || strcmp(argv[i], "hwcache") == 0) print_hwcache_events(NULL, raw_dump); else if (strcmp(argv[i], "pmu") == 0) @@ -113,6 +114,7 @@ int cmd_list(int argc, const char **argv) event_symbols_hw, PERF_COUNT_HW_MAX, raw_dump); print_symbol_events(s, PERF_TYPE_SOFTWARE, event_symbols_sw, PERF_COUNT_SW_MAX, raw_dump); + print_tool_events(s, raw_dump); print_hwcache_events(s, raw_dump); print_pmu_events(s, raw_dump, !desc_flag, long_desc_flag, diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c index 98c0fadaedb9..4432bfe039fd 100644 --- a/tools/perf/util/parse-events.c +++ b/tools/perf/util/parse-events.c @@ -2457,6 +2457,25 @@ int print_hwcache_events(const char *event_glob, bool name_only) return evt_num; } +static void print_tool_event(const char *name, const char *event_glob, + bool name_only) +{ + if (event_glob && !strglobmatch(name, event_glob)) + return; + if (name_only) + printf("%s ", name); + else + printf(" %-50s [%s]\n", name, "Tool event"); + +} + +void print_tool_events(const char *event_glob, bool name_only) +{ + print_tool_event("duration_time", event_glob, name_only); + if (pager_in_use()) + printf("\n"); +} + void print_symbol_events(const char *event_glob, unsigned type, struct event_symbol *syms, unsigned max, bool name_only) @@ -2540,6 +2559,7 @@ void print_events(const char *event_glob, bool name_only, bool quiet_flag, print_symbol_events(event_glob, PERF_TYPE_SOFTWARE, event_symbols_sw, PERF_COUNT_SW_MAX, name_only); + print_tool_events(event_glob, name_only); print_hwcache_events(event_glob, name_only); diff --git a/tools/perf/util/parse-events.h b/tools/perf/util/parse-events.h index 0c1f5b98f636..a052cd6ac63e 100644 --- a/tools/perf/util/parse-events.h +++ b/tools/perf/util/parse-events.h @@ -204,6 +204,7 @@ extern struct event_symbol event_symbols_sw[]; void print_symbol_events(const char *event_glob, unsigned type, struct event_symbol *syms, unsigned max, bool name_only); +void print_tool_events(const char *event_glob, bool name_only); void print_tracepoint_events(const char *subsys_glob, const char *event_glob, bool name_only); int print_hwcache_events(const char *event_glob, bool name_only); -- 2.20.1 ^ permalink raw reply related [flat|nested] 12+ messages in thread
* [tip:perf/core] perf list: Output tool events 2019-03-26 22:18 ` [PATCH 4/4] perf, tools, list: Output tool events Andi Kleen @ 2019-04-05 11:40 ` tip-bot for Andi Kleen 0 siblings, 0 replies; 12+ messages in thread From: tip-bot for Andi Kleen @ 2019-04-05 11:40 UTC (permalink / raw) To: linux-tip-commits; +Cc: linux-kernel, mingo, acme, jolsa, hpa, tglx, ak Commit-ID: 5e0861baa3fa73e8bd861a4b7ba7fa992b1dff82 Gitweb: https://git.kernel.org/tip/5e0861baa3fa73e8bd861a4b7ba7fa992b1dff82 Author: Andi Kleen <ak@linux.intel.com> AuthorDate: Tue, 26 Mar 2019 15:18:23 -0700 Committer: Arnaldo Carvalho de Melo <acme@redhat.com> CommitDate: Mon, 1 Apr 2019 14:49:25 -0300 perf list: Output tool events Add support in 'perf list' to output tool internal events, currently only 'duration_time'. Committer testing: $ perf list dur* List of pre-defined events (to be used in -e): duration_time [Tool event] Metric Groups: $ perf list sw List of pre-defined events (to be used in -e): alignment-faults [Software event] bpf-output [Software event] context-switches OR cs [Software event] cpu-clock [Software event] cpu-migrations OR migrations [Software event] dummy [Software event] emulation-faults [Software event] major-faults [Software event] minor-faults [Software event] page-faults OR faults [Software event] task-clock [Software event] duration_time [Tool event] $ perf list | grep duration duration_time [Tool event] [L1D miss outstandings duration in cycles] page walk duration are excluded in Skylake] load. EPT page walk duration are excluded in Skylake] page walk duration are excluded in Skylake] store. EPT page walk duration are excluded in Skylake] (instruction fetch) request. EPT page walk duration are excluded in instruction fetch request. EPT page walk duration are excluded in $ Signed-off-by: Andi Kleen <ak@linux.intel.com> Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com> Acked-by: Jiri Olsa <jolsa@kernel.org> Link: http://lkml.kernel.org/r/20190326221823.11518-5-andi@firstfloor.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> --- tools/perf/builtin-list.c | 6 ++++-- tools/perf/util/parse-events.c | 20 ++++++++++++++++++++ tools/perf/util/parse-events.h | 1 + 3 files changed, 25 insertions(+), 2 deletions(-) diff --git a/tools/perf/builtin-list.c b/tools/perf/builtin-list.c index a8394b4f1167..e0312a1c4792 100644 --- a/tools/perf/builtin-list.c +++ b/tools/perf/builtin-list.c @@ -70,10 +70,11 @@ int cmd_list(int argc, const char **argv) print_symbol_events(NULL, PERF_TYPE_HARDWARE, event_symbols_hw, PERF_COUNT_HW_MAX, raw_dump); else if (strcmp(argv[i], "sw") == 0 || - strcmp(argv[i], "software") == 0) + strcmp(argv[i], "software") == 0) { print_symbol_events(NULL, PERF_TYPE_SOFTWARE, event_symbols_sw, PERF_COUNT_SW_MAX, raw_dump); - else if (strcmp(argv[i], "cache") == 0 || + print_tool_events(NULL, raw_dump); + } else if (strcmp(argv[i], "cache") == 0 || strcmp(argv[i], "hwcache") == 0) print_hwcache_events(NULL, raw_dump); else if (strcmp(argv[i], "pmu") == 0) @@ -113,6 +114,7 @@ int cmd_list(int argc, const char **argv) event_symbols_hw, PERF_COUNT_HW_MAX, raw_dump); print_symbol_events(s, PERF_TYPE_SOFTWARE, event_symbols_sw, PERF_COUNT_SW_MAX, raw_dump); + print_tool_events(s, raw_dump); print_hwcache_events(s, raw_dump); print_pmu_events(s, raw_dump, !desc_flag, long_desc_flag, diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c index 98c0fadaedb9..4432bfe039fd 100644 --- a/tools/perf/util/parse-events.c +++ b/tools/perf/util/parse-events.c @@ -2457,6 +2457,25 @@ out_enomem: return evt_num; } +static void print_tool_event(const char *name, const char *event_glob, + bool name_only) +{ + if (event_glob && !strglobmatch(name, event_glob)) + return; + if (name_only) + printf("%s ", name); + else + printf(" %-50s [%s]\n", name, "Tool event"); + +} + +void print_tool_events(const char *event_glob, bool name_only) +{ + print_tool_event("duration_time", event_glob, name_only); + if (pager_in_use()) + printf("\n"); +} + void print_symbol_events(const char *event_glob, unsigned type, struct event_symbol *syms, unsigned max, bool name_only) @@ -2540,6 +2559,7 @@ void print_events(const char *event_glob, bool name_only, bool quiet_flag, print_symbol_events(event_glob, PERF_TYPE_SOFTWARE, event_symbols_sw, PERF_COUNT_SW_MAX, name_only); + print_tool_events(event_glob, name_only); print_hwcache_events(event_glob, name_only); diff --git a/tools/perf/util/parse-events.h b/tools/perf/util/parse-events.h index 0c1f5b98f636..a052cd6ac63e 100644 --- a/tools/perf/util/parse-events.h +++ b/tools/perf/util/parse-events.h @@ -204,6 +204,7 @@ extern struct event_symbol event_symbols_sw[]; void print_symbol_events(const char *event_glob, unsigned type, struct event_symbol *syms, unsigned max, bool name_only); +void print_tool_events(const char *event_glob, bool name_only); void print_tracepoint_events(const char *subsys_glob, const char *event_glob, bool name_only); int print_hwcache_events(const char *event_glob, bool name_only); ^ permalink raw reply related [flat|nested] 12+ messages in thread
* Re: perf: Add duration_time to perf stat 2019-03-26 22:18 perf: Add duration_time to perf stat Andi Kleen ` (3 preceding siblings ...) 2019-03-26 22:18 ` [PATCH 4/4] perf, tools, list: Output tool events Andi Kleen @ 2019-03-27 19:51 ` Jiri Olsa 4 siblings, 0 replies; 12+ messages in thread From: Jiri Olsa @ 2019-03-27 19:51 UTC (permalink / raw) To: Andi Kleen; +Cc: acme, jolsa, linux-kernel, linux-perf-users On Tue, Mar 26, 2019 at 03:18:19PM -0700, Andi Kleen wrote: > Currently it's not possible to get the wall clock time of a measurement > in CSV mode output, it's only in the non CSV perf stat output. > Often this is needed to normalize event counts to time. > > Add a new duration_time event that can be specified and reports > the wall clock time in ns. Acked-by: Jiri Olsa <jolsa@redhat.com> thanks, jirka ^ permalink raw reply [flat|nested] 12+ messages in thread
end of thread, other threads:[~2019-04-05 11:41 UTC | newest] Thread overview: 12+ messages (download: mbox.gz / follow: Atom feed) -- links below jump to the message on this page -- 2019-03-26 22:18 perf: Add duration_time to perf stat Andi Kleen 2019-03-26 22:18 ` [PATCH 1/4] perf, tools: Revert checks for duration_time Andi Kleen 2019-04-05 11:38 ` [tip:perf/core] perf stat: " tip-bot for Andi Kleen 2019-03-26 22:18 ` [PATCH 2/4] perf, tools: Implement duration_time as a proper event Andi Kleen 2019-03-27 8:58 ` Jiri Olsa 2019-03-27 13:54 ` Andi Kleen 2019-04-05 11:39 ` [tip:perf/core] perf stat: " tip-bot for Andi Kleen 2019-03-26 22:18 ` [PATCH 3/4] perf, tools: Support name for duration_time Andi Kleen 2019-04-05 11:39 ` [tip:perf/core] perf evsel: Support printing evsel name for 'duration_time' tip-bot for Andi Kleen 2019-03-26 22:18 ` [PATCH 4/4] perf, tools, list: Output tool events Andi Kleen 2019-04-05 11:40 ` [tip:perf/core] perf " tip-bot for Andi Kleen 2019-03-27 19:51 ` perf: Add duration_time to perf stat Jiri Olsa
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox; as well as URLs for NNTP newsgroup(s).