tjh.dev / kernel
Linux kernel mirror (for testing) git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel os linux
1
fork

Configure Feed

Select the types of activity you want to include in your feed.

perf evlist: Special map propagation for tool events that read on 1 CPU

Tool events like duration_time don't need a perf_cpu_map that contains
all online CPUs.

Having such a perf_cpu_map causes overheads when iterating between
events for CPU affinity.

During parsing mark events that just read on a single CPU map index as
such, then during map propagation set up the evsel's CPUs and thereby
the evlists accordingly.

The setting cannot be done early in parsing as user CPUs are only fully
known when evlist__create_maps is called.

Signed-off-by: Ian Rogers <irogers@google.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Andres Freund <andres@anarazel.de>
Cc: Dapeng Mi <dapeng1.mi@linux.intel.com>
Cc: Dr. David Alan Gilbert <linux@treblig.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: James Clark <james.clark@linaro.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Falcon <thomas.falcon@intel.com>
Cc: Thomas Richter <tmricht@linux.ibm.com>
Cc: Yang Li <yang.lee@linux.alibaba.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

authored by

Ian Rogers and committed by
Arnaldo Carvalho de Melo ff854817 63b320aa

+48 -4
+32 -4
tools/lib/perf/evlist.c
··· 101 101 evsel->cpus = perf_cpu_map__get(evlist->user_requested_cpus); 102 102 } 103 103 104 + /* 105 + * Tool events may only read on the first CPU index to avoid double 106 + * counting things like duration_time. Make the evsel->cpus contain just 107 + * that single entry otherwise we may spend time changing affinity to 108 + * CPUs that just have tool events, etc. 109 + */ 110 + if (evsel->reads_only_on_cpu_idx0 && perf_cpu_map__nr(evsel->cpus) > 0) { 111 + struct perf_cpu_map *srcs[3] = { 112 + evlist->all_cpus, 113 + evlist->user_requested_cpus, 114 + evsel->pmu_cpus, 115 + }; 116 + for (size_t i = 0; i < ARRAY_SIZE(srcs); i++) { 117 + if (!srcs[i]) 118 + continue; 119 + 120 + perf_cpu_map__put(evsel->cpus); 121 + evsel->cpus = perf_cpu_map__new_int(perf_cpu_map__cpu(srcs[i], 0).cpu); 122 + break; 123 + } 124 + } 125 + 104 126 /* Sanity check assert before the evsel is potentially removed. */ 105 127 assert(!evsel->requires_cpu || !perf_cpu_map__has_any_cpu(evsel->cpus)); 106 128 ··· 155 133 156 134 static void perf_evlist__propagate_maps(struct perf_evlist *evlist) 157 135 { 158 - struct perf_evsel *evsel, *n; 159 - 160 136 evlist->needs_map_propagation = true; 161 137 162 138 /* Clear the all_cpus set which will be merged into during propagation. */ 163 139 perf_cpu_map__put(evlist->all_cpus); 164 140 evlist->all_cpus = NULL; 165 141 166 - list_for_each_entry_safe(evsel, n, &evlist->entries, node) 167 - __perf_evlist__propagate_maps(evlist, evsel); 142 + /* 2 rounds so that reads_only_on_cpu_idx0 benefit from knowing the other CPU maps. */ 143 + for (int round = 0; round < 2; round++) { 144 + struct perf_evsel *evsel, *n; 145 + 146 + list_for_each_entry_safe(evsel, n, &evlist->entries, node) { 147 + if ((!evsel->reads_only_on_cpu_idx0 && round == 0) || 148 + (evsel->reads_only_on_cpu_idx0 && round == 1)) 149 + __perf_evlist__propagate_maps(evlist, evsel); 150 + } 151 + } 168 152 } 169 153 170 154 void perf_evlist__add(struct perf_evlist *evlist,
+2
tools/lib/perf/include/internal/evsel.h
··· 128 128 bool requires_cpu; 129 129 /** Is the PMU for the event a core one? Effects the handling of own_cpus. */ 130 130 bool is_pmu_core; 131 + /** Does the evsel on read on the first CPU index such as tool time events? */ 132 + bool reads_only_on_cpu_idx0; 131 133 int idx; 132 134 }; 133 135
+1
tools/perf/util/parse-events.c
··· 269 269 evsel->core.pmu_cpus = pmu_cpus; 270 270 evsel->core.requires_cpu = pmu ? pmu->is_uncore : false; 271 271 evsel->core.is_pmu_core = is_pmu_core; 272 + evsel->core.reads_only_on_cpu_idx0 = perf_pmu__reads_only_on_cpu_idx0(attr); 272 273 evsel->pmu = pmu; 273 274 evsel->alternate_hw_config = alternate_hw_config; 274 275 evsel->first_wildcard_match = first_wildcard_match;
+11
tools/perf/util/pmu.c
··· 2718 2718 } 2719 2719 return NULL; 2720 2720 } 2721 + 2722 + bool perf_pmu__reads_only_on_cpu_idx0(const struct perf_event_attr *attr) 2723 + { 2724 + enum tool_pmu_event event; 2725 + 2726 + if (attr->type != PERF_PMU_TYPE_TOOL) 2727 + return false; 2728 + 2729 + event = (enum tool_pmu_event)attr->config; 2730 + return event != TOOL_PMU__EVENT_USER_TIME && event != TOOL_PMU__EVENT_SYSTEM_TIME; 2731 + }
+2
tools/perf/util/pmu.h
··· 350 350 const char *perf_pmu__name_from_config(struct perf_pmu *pmu, u64 config); 351 351 bool perf_pmu__is_fake(const struct perf_pmu *pmu); 352 352 353 + bool perf_pmu__reads_only_on_cpu_idx0(const struct perf_event_attr *attr); 354 + 353 355 static inline enum pmu_kind perf_pmu__kind(const struct perf_pmu *pmu) 354 356 { 355 357 __u32 type;