]>
Commit | Line | Data |
---|---|---|
ddcacfa0 | 1 | /* |
bf9e1876 IM |
2 | * builtin-stat.c |
3 | * | |
4 | * Builtin stat command: Give a precise performance counters summary | |
5 | * overview about any workload, CPU or specific PID. | |
6 | * | |
7 | * Sample output: | |
ddcacfa0 | 8 | |
2cba3ffb | 9 | $ perf stat ./hackbench 10 |
ddcacfa0 | 10 | |
2cba3ffb | 11 | Time: 0.118 |
ddcacfa0 | 12 | |
2cba3ffb | 13 | Performance counter stats for './hackbench 10': |
ddcacfa0 | 14 | |
2cba3ffb IM |
15 | 1708.761321 task-clock # 11.037 CPUs utilized |
16 | 41,190 context-switches # 0.024 M/sec | |
17 | 6,735 CPU-migrations # 0.004 M/sec | |
18 | 17,318 page-faults # 0.010 M/sec | |
19 | 5,205,202,243 cycles # 3.046 GHz | |
20 | 3,856,436,920 stalled-cycles-frontend # 74.09% frontend cycles idle | |
21 | 1,600,790,871 stalled-cycles-backend # 30.75% backend cycles idle | |
22 | 2,603,501,247 instructions # 0.50 insns per cycle | |
23 | # 1.48 stalled cycles per insn | |
24 | 484,357,498 branches # 283.455 M/sec | |
25 | 6,388,934 branch-misses # 1.32% of all branches | |
26 | ||
27 | 0.154822978 seconds time elapsed | |
ddcacfa0 | 28 | |
5242519b | 29 | * |
2cba3ffb | 30 | * Copyright (C) 2008-2011, Red Hat Inc, Ingo Molnar <mingo@redhat.com> |
5242519b IM |
31 | * |
32 | * Improvements and fixes by: | |
33 | * | |
34 | * Arjan van de Ven <arjan@linux.intel.com> | |
35 | * Yanmin Zhang <yanmin.zhang@intel.com> | |
36 | * Wu Fengguang <fengguang.wu@intel.com> | |
37 | * Mike Galbraith <efault@gmx.de> | |
38 | * Paul Mackerras <paulus@samba.org> | |
6e750a8f | 39 | * Jaswinder Singh Rajput <jaswinder@kernel.org> |
5242519b IM |
40 | * |
41 | * Released under the GPL v2. (and only v2, not any later version) | |
ddcacfa0 IM |
42 | */ |
43 | ||
1a482f38 | 44 | #include "perf.h" |
16f762a2 | 45 | #include "builtin.h" |
148be2c1 | 46 | #include "util/util.h" |
5242519b IM |
47 | #include "util/parse-options.h" |
48 | #include "util/parse-events.h" | |
8f28827a | 49 | #include "util/event.h" |
361c99a6 | 50 | #include "util/evlist.h" |
69aad6f1 | 51 | #include "util/evsel.h" |
8f28827a | 52 | #include "util/debug.h" |
a5d243d0 | 53 | #include "util/color.h" |
0007ecea | 54 | #include "util/stat.h" |
60666c63 | 55 | #include "util/header.h" |
a12b51c4 | 56 | #include "util/cpumap.h" |
d6d901c2 | 57 | #include "util/thread.h" |
fd78260b | 58 | #include "util/thread_map.h" |
ddcacfa0 | 59 | |
1f16c575 | 60 | #include <stdlib.h> |
ddcacfa0 | 61 | #include <sys/prctl.h> |
5af52b51 | 62 | #include <locale.h> |
16c8a109 | 63 | |
d7470b6a | 64 | #define DEFAULT_SEPARATOR " " |
2cee77c4 DA |
65 | #define CNTR_NOT_SUPPORTED "<not supported>" |
66 | #define CNTR_NOT_COUNTED "<not counted>" | |
d7470b6a | 67 | |
13370a9b SE |
68 | static void print_stat(int argc, const char **argv); |
69 | static void print_counter_aggr(struct perf_evsel *counter, char *prefix); | |
70 | static void print_counter(struct perf_evsel *counter, char *prefix); | |
86ee6e18 | 71 | static void print_aggr(char *prefix); |
13370a9b | 72 | |
666e6d48 | 73 | static struct perf_evlist *evsel_list; |
361c99a6 | 74 | |
77a6f014 NK |
75 | static struct perf_target target = { |
76 | .uid = UINT_MAX, | |
77 | }; | |
ddcacfa0 | 78 | |
86ee6e18 SE |
79 | enum aggr_mode { |
80 | AGGR_NONE, | |
81 | AGGR_GLOBAL, | |
82 | AGGR_SOCKET, | |
83 | }; | |
84 | ||
3d632595 | 85 | static int run_count = 1; |
2e6cdf99 | 86 | static bool no_inherit = false; |
c0555642 | 87 | static bool scale = true; |
86ee6e18 | 88 | static enum aggr_mode aggr_mode = AGGR_GLOBAL; |
933da83a | 89 | static pid_t child_pid = -1; |
c0555642 | 90 | static bool null_run = false; |
2cba3ffb | 91 | static int detailed_run = 0; |
201e0b06 | 92 | static bool big_num = true; |
d7470b6a | 93 | static int big_num_opt = -1; |
d7470b6a SE |
94 | static const char *csv_sep = NULL; |
95 | static bool csv_output = false; | |
43bece79 | 96 | static bool group = false; |
4aa9015f | 97 | static FILE *output = NULL; |
1f16c575 PZ |
98 | static const char *pre_cmd = NULL; |
99 | static const char *post_cmd = NULL; | |
100 | static bool sync_run = false; | |
13370a9b | 101 | static unsigned int interval = 0; |
a7e191c3 | 102 | static bool forever = false; |
13370a9b | 103 | static struct timespec ref_time; |
86ee6e18 SE |
104 | static struct cpu_map *aggr_map; |
105 | static int (*aggr_get_id)(struct cpu_map *m, int cpu); | |
5af52b51 | 106 | |
60666c63 LW |
107 | static volatile int done = 0; |
108 | ||
69aad6f1 ACM |
109 | struct perf_stat { |
110 | struct stats res_stats[3]; | |
69aad6f1 ACM |
111 | }; |
112 | ||
13370a9b SE |
113 | static inline void diff_timespec(struct timespec *r, struct timespec *a, |
114 | struct timespec *b) | |
115 | { | |
116 | r->tv_sec = a->tv_sec - b->tv_sec; | |
117 | if (a->tv_nsec < b->tv_nsec) { | |
118 | r->tv_nsec = a->tv_nsec + 1000000000L - b->tv_nsec; | |
119 | r->tv_sec--; | |
120 | } else { | |
121 | r->tv_nsec = a->tv_nsec - b->tv_nsec ; | |
122 | } | |
123 | } | |
124 | ||
125 | static inline struct cpu_map *perf_evsel__cpus(struct perf_evsel *evsel) | |
126 | { | |
127 | return (evsel->cpus && !target.cpu_list) ? evsel->cpus : evsel_list->cpus; | |
128 | } | |
129 | ||
130 | static inline int perf_evsel__nr_cpus(struct perf_evsel *evsel) | |
131 | { | |
132 | return perf_evsel__cpus(evsel)->nr; | |
133 | } | |
134 | ||
a7e191c3 FD |
135 | static void perf_evsel__reset_stat_priv(struct perf_evsel *evsel) |
136 | { | |
137 | memset(evsel->priv, 0, sizeof(struct perf_stat)); | |
138 | } | |
139 | ||
c52b12ed | 140 | static int perf_evsel__alloc_stat_priv(struct perf_evsel *evsel) |
69aad6f1 | 141 | { |
c52b12ed | 142 | evsel->priv = zalloc(sizeof(struct perf_stat)); |
69aad6f1 ACM |
143 | return evsel->priv == NULL ? -ENOMEM : 0; |
144 | } | |
145 | ||
146 | static void perf_evsel__free_stat_priv(struct perf_evsel *evsel) | |
147 | { | |
148 | free(evsel->priv); | |
149 | evsel->priv = NULL; | |
150 | } | |
151 | ||
13370a9b | 152 | static int perf_evsel__alloc_prev_raw_counts(struct perf_evsel *evsel) |
7ae92e74 | 153 | { |
13370a9b SE |
154 | void *addr; |
155 | size_t sz; | |
156 | ||
157 | sz = sizeof(*evsel->counts) + | |
158 | (perf_evsel__nr_cpus(evsel) * sizeof(struct perf_counts_values)); | |
159 | ||
160 | addr = zalloc(sz); | |
161 | if (!addr) | |
162 | return -ENOMEM; | |
163 | ||
164 | evsel->prev_raw_counts = addr; | |
165 | ||
166 | return 0; | |
7ae92e74 YZ |
167 | } |
168 | ||
13370a9b | 169 | static void perf_evsel__free_prev_raw_counts(struct perf_evsel *evsel) |
7ae92e74 | 170 | { |
13370a9b SE |
171 | free(evsel->prev_raw_counts); |
172 | evsel->prev_raw_counts = NULL; | |
7ae92e74 YZ |
173 | } |
174 | ||
d134ffb9 ACM |
175 | static void perf_evlist__free_stats(struct perf_evlist *evlist) |
176 | { | |
177 | struct perf_evsel *evsel; | |
178 | ||
179 | list_for_each_entry(evsel, &evlist->entries, node) { | |
180 | perf_evsel__free_stat_priv(evsel); | |
181 | perf_evsel__free_counts(evsel); | |
182 | perf_evsel__free_prev_raw_counts(evsel); | |
183 | } | |
184 | } | |
185 | ||
186 | static int perf_evlist__alloc_stats(struct perf_evlist *evlist, bool alloc_raw) | |
187 | { | |
188 | struct perf_evsel *evsel; | |
189 | ||
190 | list_for_each_entry(evsel, &evlist->entries, node) { | |
191 | if (perf_evsel__alloc_stat_priv(evsel) < 0 || | |
192 | perf_evsel__alloc_counts(evsel, perf_evsel__nr_cpus(evsel)) < 0 || | |
193 | (alloc_raw && perf_evsel__alloc_prev_raw_counts(evsel) < 0)) | |
194 | goto out_free; | |
195 | } | |
196 | ||
197 | return 0; | |
198 | ||
199 | out_free: | |
200 | perf_evlist__free_stats(evlist); | |
201 | return -1; | |
202 | } | |
203 | ||
666e6d48 RR |
204 | static struct stats runtime_nsecs_stats[MAX_NR_CPUS]; |
205 | static struct stats runtime_cycles_stats[MAX_NR_CPUS]; | |
206 | static struct stats runtime_stalled_cycles_front_stats[MAX_NR_CPUS]; | |
207 | static struct stats runtime_stalled_cycles_back_stats[MAX_NR_CPUS]; | |
208 | static struct stats runtime_branches_stats[MAX_NR_CPUS]; | |
209 | static struct stats runtime_cacherefs_stats[MAX_NR_CPUS]; | |
210 | static struct stats runtime_l1_dcache_stats[MAX_NR_CPUS]; | |
211 | static struct stats runtime_l1_icache_stats[MAX_NR_CPUS]; | |
212 | static struct stats runtime_ll_cache_stats[MAX_NR_CPUS]; | |
213 | static struct stats runtime_itlb_cache_stats[MAX_NR_CPUS]; | |
214 | static struct stats runtime_dtlb_cache_stats[MAX_NR_CPUS]; | |
215 | static struct stats walltime_nsecs_stats; | |
be1ac0d8 | 216 | |
d134ffb9 | 217 | static void perf_stat__reset_stats(struct perf_evlist *evlist) |
a7e191c3 | 218 | { |
d134ffb9 ACM |
219 | struct perf_evsel *evsel; |
220 | ||
221 | list_for_each_entry(evsel, &evlist->entries, node) { | |
222 | perf_evsel__reset_stat_priv(evsel); | |
223 | perf_evsel__reset_counts(evsel, perf_evsel__nr_cpus(evsel)); | |
224 | } | |
225 | ||
a7e191c3 FD |
226 | memset(runtime_nsecs_stats, 0, sizeof(runtime_nsecs_stats)); |
227 | memset(runtime_cycles_stats, 0, sizeof(runtime_cycles_stats)); | |
228 | memset(runtime_stalled_cycles_front_stats, 0, sizeof(runtime_stalled_cycles_front_stats)); | |
229 | memset(runtime_stalled_cycles_back_stats, 0, sizeof(runtime_stalled_cycles_back_stats)); | |
230 | memset(runtime_branches_stats, 0, sizeof(runtime_branches_stats)); | |
231 | memset(runtime_cacherefs_stats, 0, sizeof(runtime_cacherefs_stats)); | |
232 | memset(runtime_l1_dcache_stats, 0, sizeof(runtime_l1_dcache_stats)); | |
233 | memset(runtime_l1_icache_stats, 0, sizeof(runtime_l1_icache_stats)); | |
234 | memset(runtime_ll_cache_stats, 0, sizeof(runtime_ll_cache_stats)); | |
235 | memset(runtime_itlb_cache_stats, 0, sizeof(runtime_itlb_cache_stats)); | |
236 | memset(runtime_dtlb_cache_stats, 0, sizeof(runtime_dtlb_cache_stats)); | |
237 | memset(&walltime_nsecs_stats, 0, sizeof(walltime_nsecs_stats)); | |
238 | } | |
239 | ||
cac21425 | 240 | static int create_perf_stat_counter(struct perf_evsel *evsel) |
ddcacfa0 | 241 | { |
69aad6f1 | 242 | struct perf_event_attr *attr = &evsel->attr; |
727ab04e | 243 | |
ddcacfa0 | 244 | if (scale) |
a21ca2ca IM |
245 | attr->read_format = PERF_FORMAT_TOTAL_TIME_ENABLED | |
246 | PERF_FORMAT_TOTAL_TIME_RUNNING; | |
ddcacfa0 | 247 | |
5d2cd909 ACM |
248 | attr->inherit = !no_inherit; |
249 | ||
594ac61a ACM |
250 | if (perf_target__has_cpu(&target)) |
251 | return perf_evsel__open_per_cpu(evsel, perf_evsel__cpus(evsel)); | |
5622c07b | 252 | |
07ac002f | 253 | if (!perf_target__has_task(&target) && |
823254ed | 254 | perf_evsel__is_group_leader(evsel)) { |
48290609 ACM |
255 | attr->disabled = 1; |
256 | attr->enable_on_exec = 1; | |
ddcacfa0 | 257 | } |
084ab9f8 | 258 | |
594ac61a | 259 | return perf_evsel__open_per_thread(evsel, evsel_list->threads); |
ddcacfa0 IM |
260 | } |
261 | ||
c04f5e5d IM |
262 | /* |
263 | * Does the counter have nsecs as a unit? | |
264 | */ | |
daec78a0 | 265 | static inline int nsec_counter(struct perf_evsel *evsel) |
c04f5e5d | 266 | { |
daec78a0 ACM |
267 | if (perf_evsel__match(evsel, SOFTWARE, SW_CPU_CLOCK) || |
268 | perf_evsel__match(evsel, SOFTWARE, SW_TASK_CLOCK)) | |
c04f5e5d IM |
269 | return 1; |
270 | ||
271 | return 0; | |
272 | } | |
273 | ||
dcd9936a IM |
274 | /* |
275 | * Update various tracking values we maintain to print | |
276 | * more semantic information such as miss/hit ratios, | |
277 | * instruction rates, etc: | |
278 | */ | |
279 | static void update_shadow_stats(struct perf_evsel *counter, u64 *count) | |
280 | { | |
281 | if (perf_evsel__match(counter, SOFTWARE, SW_TASK_CLOCK)) | |
282 | update_stats(&runtime_nsecs_stats[0], count[0]); | |
283 | else if (perf_evsel__match(counter, HARDWARE, HW_CPU_CYCLES)) | |
284 | update_stats(&runtime_cycles_stats[0], count[0]); | |
d3d1e86d IM |
285 | else if (perf_evsel__match(counter, HARDWARE, HW_STALLED_CYCLES_FRONTEND)) |
286 | update_stats(&runtime_stalled_cycles_front_stats[0], count[0]); | |
129c04cb | 287 | else if (perf_evsel__match(counter, HARDWARE, HW_STALLED_CYCLES_BACKEND)) |
d3d1e86d | 288 | update_stats(&runtime_stalled_cycles_back_stats[0], count[0]); |
dcd9936a IM |
289 | else if (perf_evsel__match(counter, HARDWARE, HW_BRANCH_INSTRUCTIONS)) |
290 | update_stats(&runtime_branches_stats[0], count[0]); | |
291 | else if (perf_evsel__match(counter, HARDWARE, HW_CACHE_REFERENCES)) | |
292 | update_stats(&runtime_cacherefs_stats[0], count[0]); | |
8bb6c79f IM |
293 | else if (perf_evsel__match(counter, HW_CACHE, HW_CACHE_L1D)) |
294 | update_stats(&runtime_l1_dcache_stats[0], count[0]); | |
c3305257 IM |
295 | else if (perf_evsel__match(counter, HW_CACHE, HW_CACHE_L1I)) |
296 | update_stats(&runtime_l1_icache_stats[0], count[0]); | |
297 | else if (perf_evsel__match(counter, HW_CACHE, HW_CACHE_LL)) | |
298 | update_stats(&runtime_ll_cache_stats[0], count[0]); | |
299 | else if (perf_evsel__match(counter, HW_CACHE, HW_CACHE_DTLB)) | |
300 | update_stats(&runtime_dtlb_cache_stats[0], count[0]); | |
301 | else if (perf_evsel__match(counter, HW_CACHE, HW_CACHE_ITLB)) | |
302 | update_stats(&runtime_itlb_cache_stats[0], count[0]); | |
dcd9936a IM |
303 | } |
304 | ||
c04f5e5d | 305 | /* |
2996f5dd | 306 | * Read out the results of a single counter: |
f5b4a9c3 | 307 | * aggregate counts across CPUs in system-wide mode |
c04f5e5d | 308 | */ |
c52b12ed | 309 | static int read_counter_aggr(struct perf_evsel *counter) |
c04f5e5d | 310 | { |
69aad6f1 | 311 | struct perf_stat *ps = counter->priv; |
c52b12ed ACM |
312 | u64 *count = counter->counts->aggr.values; |
313 | int i; | |
2996f5dd | 314 | |
7ae92e74 | 315 | if (__perf_evsel__read(counter, perf_evsel__nr_cpus(counter), |
b3a319d5 | 316 | thread_map__nr(evsel_list->threads), scale) < 0) |
c52b12ed | 317 | return -1; |
9e9772c4 PZ |
318 | |
319 | for (i = 0; i < 3; i++) | |
69aad6f1 | 320 | update_stats(&ps->res_stats[i], count[i]); |
9e9772c4 PZ |
321 | |
322 | if (verbose) { | |
4aa9015f | 323 | fprintf(output, "%s: %" PRIu64 " %" PRIu64 " %" PRIu64 "\n", |
7289f83c | 324 | perf_evsel__name(counter), count[0], count[1], count[2]); |
9e9772c4 PZ |
325 | } |
326 | ||
be1ac0d8 IM |
327 | /* |
328 | * Save the full runtime - to allow normalization during printout: | |
329 | */ | |
dcd9936a | 330 | update_shadow_stats(counter, count); |
c52b12ed ACM |
331 | |
332 | return 0; | |
f5b4a9c3 SE |
333 | } |
334 | ||
335 | /* | |
336 | * Read out the results of a single counter: | |
337 | * do not aggregate counts across CPUs in system-wide mode | |
338 | */ | |
c52b12ed | 339 | static int read_counter(struct perf_evsel *counter) |
f5b4a9c3 | 340 | { |
c52b12ed | 341 | u64 *count; |
f5b4a9c3 | 342 | int cpu; |
f5b4a9c3 | 343 | |
7ae92e74 | 344 | for (cpu = 0; cpu < perf_evsel__nr_cpus(counter); cpu++) { |
c52b12ed ACM |
345 | if (__perf_evsel__read_on_cpu(counter, cpu, 0, scale) < 0) |
346 | return -1; | |
f5b4a9c3 | 347 | |
c52b12ed | 348 | count = counter->counts->cpu[cpu].values; |
f5b4a9c3 | 349 | |
dcd9936a | 350 | update_shadow_stats(counter, count); |
f5b4a9c3 | 351 | } |
c52b12ed ACM |
352 | |
353 | return 0; | |
2996f5dd IM |
354 | } |
355 | ||
13370a9b SE |
356 | static void print_interval(void) |
357 | { | |
358 | static int num_print_interval; | |
359 | struct perf_evsel *counter; | |
360 | struct perf_stat *ps; | |
361 | struct timespec ts, rs; | |
362 | char prefix[64]; | |
363 | ||
86ee6e18 | 364 | if (aggr_mode == AGGR_GLOBAL) { |
13370a9b SE |
365 | list_for_each_entry(counter, &evsel_list->entries, node) { |
366 | ps = counter->priv; | |
367 | memset(ps->res_stats, 0, sizeof(ps->res_stats)); | |
86ee6e18 | 368 | read_counter_aggr(counter); |
13370a9b | 369 | } |
86ee6e18 | 370 | } else { |
13370a9b SE |
371 | list_for_each_entry(counter, &evsel_list->entries, node) { |
372 | ps = counter->priv; | |
373 | memset(ps->res_stats, 0, sizeof(ps->res_stats)); | |
86ee6e18 | 374 | read_counter(counter); |
13370a9b SE |
375 | } |
376 | } | |
86ee6e18 | 377 | |
13370a9b SE |
378 | clock_gettime(CLOCK_MONOTONIC, &ts); |
379 | diff_timespec(&rs, &ts, &ref_time); | |
380 | sprintf(prefix, "%6lu.%09lu%s", rs.tv_sec, rs.tv_nsec, csv_sep); | |
381 | ||
382 | if (num_print_interval == 0 && !csv_output) { | |
86ee6e18 SE |
383 | switch (aggr_mode) { |
384 | case AGGR_SOCKET: | |
d7e7a451 | 385 | fprintf(output, "# time socket cpus counts events\n"); |
86ee6e18 SE |
386 | break; |
387 | case AGGR_NONE: | |
13370a9b | 388 | fprintf(output, "# time CPU counts events\n"); |
86ee6e18 SE |
389 | break; |
390 | case AGGR_GLOBAL: | |
391 | default: | |
13370a9b | 392 | fprintf(output, "# time counts events\n"); |
86ee6e18 | 393 | } |
13370a9b SE |
394 | } |
395 | ||
396 | if (++num_print_interval == 25) | |
397 | num_print_interval = 0; | |
398 | ||
86ee6e18 SE |
399 | switch (aggr_mode) { |
400 | case AGGR_SOCKET: | |
401 | print_aggr(prefix); | |
402 | break; | |
403 | case AGGR_NONE: | |
13370a9b SE |
404 | list_for_each_entry(counter, &evsel_list->entries, node) |
405 | print_counter(counter, prefix); | |
86ee6e18 SE |
406 | break; |
407 | case AGGR_GLOBAL: | |
408 | default: | |
13370a9b SE |
409 | list_for_each_entry(counter, &evsel_list->entries, node) |
410 | print_counter_aggr(counter, prefix); | |
411 | } | |
412 | } | |
413 | ||
acf28922 | 414 | static int __run_perf_stat(int argc, const char **argv) |
42202dd5 | 415 | { |
56e52e85 | 416 | char msg[512]; |
42202dd5 | 417 | unsigned long long t0, t1; |
cac21425 | 418 | struct perf_evsel *counter; |
13370a9b | 419 | struct timespec ts; |
42202dd5 | 420 | int status = 0; |
6be2850e | 421 | const bool forks = (argc > 0); |
42202dd5 | 422 | |
13370a9b SE |
423 | if (interval) { |
424 | ts.tv_sec = interval / 1000; | |
425 | ts.tv_nsec = (interval % 1000) * 1000000; | |
426 | } else { | |
427 | ts.tv_sec = 1; | |
428 | ts.tv_nsec = 0; | |
429 | } | |
430 | ||
60666c63 | 431 | if (forks) { |
acf28922 NK |
432 | if (perf_evlist__prepare_workload(evsel_list, &target, argv, |
433 | false, false) < 0) { | |
434 | perror("failed to prepare workload"); | |
435 | return -1; | |
60666c63 | 436 | } |
051ae7f7 PM |
437 | } |
438 | ||
6a4bb04c | 439 | if (group) |
63dab225 | 440 | perf_evlist__set_leader(evsel_list); |
6a4bb04c | 441 | |
361c99a6 | 442 | list_for_each_entry(counter, &evsel_list->entries, node) { |
cac21425 | 443 | if (create_perf_stat_counter(counter) < 0) { |
979987a5 DA |
444 | /* |
445 | * PPC returns ENXIO for HW counters until 2.6.37 | |
446 | * (behavior changed with commit b0a873e). | |
447 | */ | |
38f6ae1e | 448 | if (errno == EINVAL || errno == ENOSYS || |
979987a5 DA |
449 | errno == ENOENT || errno == EOPNOTSUPP || |
450 | errno == ENXIO) { | |
c63ca0c0 DA |
451 | if (verbose) |
452 | ui__warning("%s event is not supported by the kernel.\n", | |
7289f83c | 453 | perf_evsel__name(counter)); |
2cee77c4 | 454 | counter->supported = false; |
ede70290 | 455 | continue; |
c63ca0c0 | 456 | } |
ede70290 | 457 | |
56e52e85 ACM |
458 | perf_evsel__open_strerror(counter, &target, |
459 | errno, msg, sizeof(msg)); | |
460 | ui__error("%s\n", msg); | |
461 | ||
48290609 ACM |
462 | if (child_pid != -1) |
463 | kill(child_pid, SIGTERM); | |
fceda7fe | 464 | |
48290609 ACM |
465 | return -1; |
466 | } | |
2cee77c4 | 467 | counter->supported = true; |
084ab9f8 | 468 | } |
42202dd5 | 469 | |
1491a632 | 470 | if (perf_evlist__apply_filters(evsel_list)) { |
cfd748ae FW |
471 | error("failed to set filter with %d (%s)\n", errno, |
472 | strerror(errno)); | |
473 | return -1; | |
474 | } | |
475 | ||
42202dd5 IM |
476 | /* |
477 | * Enable counters and exec the command: | |
478 | */ | |
479 | t0 = rdclock(); | |
13370a9b | 480 | clock_gettime(CLOCK_MONOTONIC, &ref_time); |
42202dd5 | 481 | |
60666c63 | 482 | if (forks) { |
acf28922 NK |
483 | perf_evlist__start_workload(evsel_list); |
484 | ||
13370a9b SE |
485 | if (interval) { |
486 | while (!waitpid(child_pid, &status, WNOHANG)) { | |
487 | nanosleep(&ts, NULL); | |
488 | print_interval(); | |
489 | } | |
490 | } | |
60666c63 | 491 | wait(&status); |
33e49ea7 AK |
492 | if (WIFSIGNALED(status)) |
493 | psignal(WTERMSIG(status), argv[0]); | |
60666c63 | 494 | } else { |
13370a9b SE |
495 | while (!done) { |
496 | nanosleep(&ts, NULL); | |
497 | if (interval) | |
498 | print_interval(); | |
499 | } | |
60666c63 | 500 | } |
42202dd5 | 501 | |
42202dd5 IM |
502 | t1 = rdclock(); |
503 | ||
9e9772c4 | 504 | update_stats(&walltime_nsecs_stats, t1 - t0); |
42202dd5 | 505 | |
86ee6e18 | 506 | if (aggr_mode == AGGR_GLOBAL) { |
361c99a6 | 507 | list_for_each_entry(counter, &evsel_list->entries, node) { |
f5b4a9c3 | 508 | read_counter_aggr(counter); |
7ae92e74 | 509 | perf_evsel__close_fd(counter, perf_evsel__nr_cpus(counter), |
b3a319d5 | 510 | thread_map__nr(evsel_list->threads)); |
c52b12ed | 511 | } |
86ee6e18 SE |
512 | } else { |
513 | list_for_each_entry(counter, &evsel_list->entries, node) { | |
514 | read_counter(counter); | |
515 | perf_evsel__close_fd(counter, perf_evsel__nr_cpus(counter), 1); | |
516 | } | |
f5b4a9c3 | 517 | } |
c52b12ed | 518 | |
42202dd5 IM |
519 | return WEXITSTATUS(status); |
520 | } | |
521 | ||
1f16c575 PZ |
522 | static int run_perf_stat(int argc __maybe_unused, const char **argv) |
523 | { | |
524 | int ret; | |
525 | ||
526 | if (pre_cmd) { | |
527 | ret = system(pre_cmd); | |
528 | if (ret) | |
529 | return ret; | |
530 | } | |
531 | ||
532 | if (sync_run) | |
533 | sync(); | |
534 | ||
535 | ret = __run_perf_stat(argc, argv); | |
536 | if (ret) | |
537 | return ret; | |
538 | ||
539 | if (post_cmd) { | |
540 | ret = system(post_cmd); | |
541 | if (ret) | |
542 | return ret; | |
543 | } | |
544 | ||
545 | return ret; | |
546 | } | |
547 | ||
f99844cb IM |
548 | static void print_noise_pct(double total, double avg) |
549 | { | |
0007ecea | 550 | double pct = rel_stddev_stats(total, avg); |
f99844cb | 551 | |
3ae9a34d | 552 | if (csv_output) |
4aa9015f | 553 | fprintf(output, "%s%.2f%%", csv_sep, pct); |
a1bca6cc | 554 | else if (pct) |
4aa9015f | 555 | fprintf(output, " ( +-%6.2f%% )", pct); |
f99844cb IM |
556 | } |
557 | ||
69aad6f1 | 558 | static void print_noise(struct perf_evsel *evsel, double avg) |
42202dd5 | 559 | { |
69aad6f1 ACM |
560 | struct perf_stat *ps; |
561 | ||
849abde9 PZ |
562 | if (run_count == 1) |
563 | return; | |
564 | ||
69aad6f1 | 565 | ps = evsel->priv; |
f99844cb | 566 | print_noise_pct(stddev_stats(&ps->res_stats[0]), avg); |
42202dd5 IM |
567 | } |
568 | ||
86ee6e18 | 569 | static void aggr_printout(struct perf_evsel *evsel, int cpu, int nr) |
44175b6f | 570 | { |
86ee6e18 SE |
571 | switch (aggr_mode) { |
572 | case AGGR_SOCKET: | |
573 | fprintf(output, "S%*d%s%*d%s", | |
d7e7a451 SE |
574 | csv_output ? 0 : -5, |
575 | cpu, | |
576 | csv_sep, | |
577 | csv_output ? 0 : 4, | |
578 | nr, | |
579 | csv_sep); | |
86ee6e18 SE |
580 | break; |
581 | case AGGR_NONE: | |
582 | fprintf(output, "CPU%*d%s", | |
d7470b6a | 583 | csv_output ? 0 : -4, |
7ae92e74 | 584 | perf_evsel__cpus(evsel)->map[cpu], csv_sep); |
86ee6e18 SE |
585 | break; |
586 | case AGGR_GLOBAL: | |
587 | default: | |
588 | break; | |
589 | } | |
590 | } | |
591 | ||
592 | static void nsec_printout(int cpu, int nr, struct perf_evsel *evsel, double avg) | |
593 | { | |
594 | double msecs = avg / 1e6; | |
595 | const char *fmt = csv_output ? "%.6f%s%s" : "%18.6f%s%-25s"; | |
596 | ||
597 | aggr_printout(evsel, cpu, nr); | |
d7470b6a | 598 | |
86ee6e18 | 599 | fprintf(output, fmt, msecs, csv_sep, perf_evsel__name(evsel)); |
d7470b6a | 600 | |
023695d9 | 601 | if (evsel->cgrp) |
4aa9015f | 602 | fprintf(output, "%s%s", csv_sep, evsel->cgrp->name); |
023695d9 | 603 | |
13370a9b | 604 | if (csv_output || interval) |
d7470b6a | 605 | return; |
44175b6f | 606 | |
daec78a0 | 607 | if (perf_evsel__match(evsel, SOFTWARE, SW_TASK_CLOCK)) |
4aa9015f SE |
608 | fprintf(output, " # %8.3f CPUs utilized ", |
609 | avg / avg_stats(&walltime_nsecs_stats)); | |
9dac6a29 NK |
610 | else |
611 | fprintf(output, " "); | |
44175b6f IM |
612 | } |
613 | ||
15e6392f NK |
614 | /* used for get_ratio_color() */ |
615 | enum grc_type { | |
616 | GRC_STALLED_CYCLES_FE, | |
617 | GRC_STALLED_CYCLES_BE, | |
618 | GRC_CACHE_MISSES, | |
619 | GRC_MAX_NR | |
620 | }; | |
621 | ||
622 | static const char *get_ratio_color(enum grc_type type, double ratio) | |
623 | { | |
624 | static const double grc_table[GRC_MAX_NR][3] = { | |
625 | [GRC_STALLED_CYCLES_FE] = { 50.0, 30.0, 10.0 }, | |
626 | [GRC_STALLED_CYCLES_BE] = { 75.0, 50.0, 20.0 }, | |
627 | [GRC_CACHE_MISSES] = { 20.0, 10.0, 5.0 }, | |
628 | }; | |
629 | const char *color = PERF_COLOR_NORMAL; | |
630 | ||
631 | if (ratio > grc_table[type][0]) | |
632 | color = PERF_COLOR_RED; | |
633 | else if (ratio > grc_table[type][1]) | |
634 | color = PERF_COLOR_MAGENTA; | |
635 | else if (ratio > grc_table[type][2]) | |
636 | color = PERF_COLOR_YELLOW; | |
637 | ||
638 | return color; | |
639 | } | |
640 | ||
1d037ca1 IT |
641 | static void print_stalled_cycles_frontend(int cpu, |
642 | struct perf_evsel *evsel | |
643 | __maybe_unused, double avg) | |
d3d1e86d IM |
644 | { |
645 | double total, ratio = 0.0; | |
646 | const char *color; | |
647 | ||
648 | total = avg_stats(&runtime_cycles_stats[cpu]); | |
649 | ||
650 | if (total) | |
651 | ratio = avg / total * 100.0; | |
652 | ||
15e6392f | 653 | color = get_ratio_color(GRC_STALLED_CYCLES_FE, ratio); |
d3d1e86d | 654 | |
4aa9015f SE |
655 | fprintf(output, " # "); |
656 | color_fprintf(output, color, "%6.2f%%", ratio); | |
657 | fprintf(output, " frontend cycles idle "); | |
d3d1e86d IM |
658 | } |
659 | ||
1d037ca1 IT |
660 | static void print_stalled_cycles_backend(int cpu, |
661 | struct perf_evsel *evsel | |
662 | __maybe_unused, double avg) | |
a5d243d0 IM |
663 | { |
664 | double total, ratio = 0.0; | |
665 | const char *color; | |
666 | ||
667 | total = avg_stats(&runtime_cycles_stats[cpu]); | |
668 | ||
669 | if (total) | |
670 | ratio = avg / total * 100.0; | |
671 | ||
15e6392f | 672 | color = get_ratio_color(GRC_STALLED_CYCLES_BE, ratio); |
a5d243d0 | 673 | |
4aa9015f SE |
674 | fprintf(output, " # "); |
675 | color_fprintf(output, color, "%6.2f%%", ratio); | |
676 | fprintf(output, " backend cycles idle "); | |
a5d243d0 IM |
677 | } |
678 | ||
1d037ca1 IT |
679 | static void print_branch_misses(int cpu, |
680 | struct perf_evsel *evsel __maybe_unused, | |
681 | double avg) | |
c78df6c1 IM |
682 | { |
683 | double total, ratio = 0.0; | |
684 | const char *color; | |
685 | ||
686 | total = avg_stats(&runtime_branches_stats[cpu]); | |
687 | ||
688 | if (total) | |
689 | ratio = avg / total * 100.0; | |
690 | ||
15e6392f | 691 | color = get_ratio_color(GRC_CACHE_MISSES, ratio); |
c78df6c1 | 692 | |
4aa9015f SE |
693 | fprintf(output, " # "); |
694 | color_fprintf(output, color, "%6.2f%%", ratio); | |
695 | fprintf(output, " of all branches "); | |
c78df6c1 IM |
696 | } |
697 | ||
1d037ca1 IT |
698 | static void print_l1_dcache_misses(int cpu, |
699 | struct perf_evsel *evsel __maybe_unused, | |
700 | double avg) | |
8bb6c79f IM |
701 | { |
702 | double total, ratio = 0.0; | |
703 | const char *color; | |
704 | ||
705 | total = avg_stats(&runtime_l1_dcache_stats[cpu]); | |
706 | ||
707 | if (total) | |
708 | ratio = avg / total * 100.0; | |
709 | ||
15e6392f | 710 | color = get_ratio_color(GRC_CACHE_MISSES, ratio); |
8bb6c79f | 711 | |
4aa9015f SE |
712 | fprintf(output, " # "); |
713 | color_fprintf(output, color, "%6.2f%%", ratio); | |
714 | fprintf(output, " of all L1-dcache hits "); | |
8bb6c79f IM |
715 | } |
716 | ||
1d037ca1 IT |
717 | static void print_l1_icache_misses(int cpu, |
718 | struct perf_evsel *evsel __maybe_unused, | |
719 | double avg) | |
c3305257 IM |
720 | { |
721 | double total, ratio = 0.0; | |
722 | const char *color; | |
723 | ||
724 | total = avg_stats(&runtime_l1_icache_stats[cpu]); | |
725 | ||
726 | if (total) | |
727 | ratio = avg / total * 100.0; | |
728 | ||
15e6392f | 729 | color = get_ratio_color(GRC_CACHE_MISSES, ratio); |
c3305257 | 730 | |
4aa9015f SE |
731 | fprintf(output, " # "); |
732 | color_fprintf(output, color, "%6.2f%%", ratio); | |
733 | fprintf(output, " of all L1-icache hits "); | |
c3305257 IM |
734 | } |
735 | ||
1d037ca1 IT |
736 | static void print_dtlb_cache_misses(int cpu, |
737 | struct perf_evsel *evsel __maybe_unused, | |
738 | double avg) | |
c3305257 IM |
739 | { |
740 | double total, ratio = 0.0; | |
741 | const char *color; | |
742 | ||
743 | total = avg_stats(&runtime_dtlb_cache_stats[cpu]); | |
744 | ||
745 | if (total) | |
746 | ratio = avg / total * 100.0; | |
747 | ||
15e6392f | 748 | color = get_ratio_color(GRC_CACHE_MISSES, ratio); |
c3305257 | 749 | |
4aa9015f SE |
750 | fprintf(output, " # "); |
751 | color_fprintf(output, color, "%6.2f%%", ratio); | |
752 | fprintf(output, " of all dTLB cache hits "); | |
c3305257 IM |
753 | } |
754 | ||
1d037ca1 IT |
755 | static void print_itlb_cache_misses(int cpu, |
756 | struct perf_evsel *evsel __maybe_unused, | |
757 | double avg) | |
c3305257 IM |
758 | { |
759 | double total, ratio = 0.0; | |
760 | const char *color; | |
761 | ||
762 | total = avg_stats(&runtime_itlb_cache_stats[cpu]); | |
763 | ||
764 | if (total) | |
765 | ratio = avg / total * 100.0; | |
766 | ||
15e6392f | 767 | color = get_ratio_color(GRC_CACHE_MISSES, ratio); |
c3305257 | 768 | |
4aa9015f SE |
769 | fprintf(output, " # "); |
770 | color_fprintf(output, color, "%6.2f%%", ratio); | |
771 | fprintf(output, " of all iTLB cache hits "); | |
c3305257 IM |
772 | } |
773 | ||
1d037ca1 IT |
774 | static void print_ll_cache_misses(int cpu, |
775 | struct perf_evsel *evsel __maybe_unused, | |
776 | double avg) | |
c3305257 IM |
777 | { |
778 | double total, ratio = 0.0; | |
779 | const char *color; | |
780 | ||
781 | total = avg_stats(&runtime_ll_cache_stats[cpu]); | |
782 | ||
783 | if (total) | |
784 | ratio = avg / total * 100.0; | |
785 | ||
15e6392f | 786 | color = get_ratio_color(GRC_CACHE_MISSES, ratio); |
c3305257 | 787 | |
4aa9015f SE |
788 | fprintf(output, " # "); |
789 | color_fprintf(output, color, "%6.2f%%", ratio); | |
790 | fprintf(output, " of all LL-cache hits "); | |
c3305257 IM |
791 | } |
792 | ||
d7e7a451 | 793 | static void abs_printout(int cpu, int nr, struct perf_evsel *evsel, double avg) |
44175b6f | 794 | { |
c7f7fea3 | 795 | double total, ratio = 0.0; |
d7470b6a SE |
796 | const char *fmt; |
797 | ||
798 | if (csv_output) | |
86ee6e18 | 799 | fmt = "%.0f%s%s"; |
d7470b6a | 800 | else if (big_num) |
86ee6e18 | 801 | fmt = "%'18.0f%s%-25s"; |
d7470b6a | 802 | else |
86ee6e18 | 803 | fmt = "%18.0f%s%-25s"; |
f5b4a9c3 | 804 | |
86ee6e18 SE |
805 | aggr_printout(evsel, cpu, nr); |
806 | ||
807 | if (aggr_mode == AGGR_GLOBAL) | |
f5b4a9c3 | 808 | cpu = 0; |
c7f7fea3 | 809 | |
86ee6e18 | 810 | fprintf(output, fmt, avg, csv_sep, perf_evsel__name(evsel)); |
d7470b6a | 811 | |
023695d9 | 812 | if (evsel->cgrp) |
4aa9015f | 813 | fprintf(output, "%s%s", csv_sep, evsel->cgrp->name); |
023695d9 | 814 | |
13370a9b | 815 | if (csv_output || interval) |
d7470b6a | 816 | return; |
44175b6f | 817 | |
daec78a0 | 818 | if (perf_evsel__match(evsel, HARDWARE, HW_INSTRUCTIONS)) { |
f5b4a9c3 | 819 | total = avg_stats(&runtime_cycles_stats[cpu]); |
c7f7fea3 IM |
820 | if (total) |
821 | ratio = avg / total; | |
822 | ||
4aa9015f | 823 | fprintf(output, " # %5.2f insns per cycle ", ratio); |
481f988a | 824 | |
d3d1e86d IM |
825 | total = avg_stats(&runtime_stalled_cycles_front_stats[cpu]); |
826 | total = max(total, avg_stats(&runtime_stalled_cycles_back_stats[cpu])); | |
481f988a IM |
827 | |
828 | if (total && avg) { | |
829 | ratio = total / avg; | |
4aa9015f | 830 | fprintf(output, "\n # %5.2f stalled cycles per insn", ratio); |
481f988a IM |
831 | } |
832 | ||
daec78a0 | 833 | } else if (perf_evsel__match(evsel, HARDWARE, HW_BRANCH_MISSES) && |
f5b4a9c3 | 834 | runtime_branches_stats[cpu].n != 0) { |
c78df6c1 | 835 | print_branch_misses(cpu, evsel, avg); |
8bb6c79f IM |
836 | } else if ( |
837 | evsel->attr.type == PERF_TYPE_HW_CACHE && | |
838 | evsel->attr.config == ( PERF_COUNT_HW_CACHE_L1D | | |
839 | ((PERF_COUNT_HW_CACHE_OP_READ) << 8) | | |
840 | ((PERF_COUNT_HW_CACHE_RESULT_MISS) << 16)) && | |
c6264def | 841 | runtime_l1_dcache_stats[cpu].n != 0) { |
8bb6c79f | 842 | print_l1_dcache_misses(cpu, evsel, avg); |
c3305257 IM |
843 | } else if ( |
844 | evsel->attr.type == PERF_TYPE_HW_CACHE && | |
845 | evsel->attr.config == ( PERF_COUNT_HW_CACHE_L1I | | |
846 | ((PERF_COUNT_HW_CACHE_OP_READ) << 8) | | |
847 | ((PERF_COUNT_HW_CACHE_RESULT_MISS) << 16)) && | |
848 | runtime_l1_icache_stats[cpu].n != 0) { | |
849 | print_l1_icache_misses(cpu, evsel, avg); | |
850 | } else if ( | |
851 | evsel->attr.type == PERF_TYPE_HW_CACHE && | |
852 | evsel->attr.config == ( PERF_COUNT_HW_CACHE_DTLB | | |
853 | ((PERF_COUNT_HW_CACHE_OP_READ) << 8) | | |
854 | ((PERF_COUNT_HW_CACHE_RESULT_MISS) << 16)) && | |
855 | runtime_dtlb_cache_stats[cpu].n != 0) { | |
856 | print_dtlb_cache_misses(cpu, evsel, avg); | |
857 | } else if ( | |
858 | evsel->attr.type == PERF_TYPE_HW_CACHE && | |
859 | evsel->attr.config == ( PERF_COUNT_HW_CACHE_ITLB | | |
860 | ((PERF_COUNT_HW_CACHE_OP_READ) << 8) | | |
861 | ((PERF_COUNT_HW_CACHE_RESULT_MISS) << 16)) && | |
862 | runtime_itlb_cache_stats[cpu].n != 0) { | |
863 | print_itlb_cache_misses(cpu, evsel, avg); | |
864 | } else if ( | |
865 | evsel->attr.type == PERF_TYPE_HW_CACHE && | |
866 | evsel->attr.config == ( PERF_COUNT_HW_CACHE_LL | | |
867 | ((PERF_COUNT_HW_CACHE_OP_READ) << 8) | | |
868 | ((PERF_COUNT_HW_CACHE_RESULT_MISS) << 16)) && | |
869 | runtime_ll_cache_stats[cpu].n != 0) { | |
870 | print_ll_cache_misses(cpu, evsel, avg); | |
d58f4c82 IM |
871 | } else if (perf_evsel__match(evsel, HARDWARE, HW_CACHE_MISSES) && |
872 | runtime_cacherefs_stats[cpu].n != 0) { | |
873 | total = avg_stats(&runtime_cacherefs_stats[cpu]); | |
874 | ||
875 | if (total) | |
876 | ratio = avg * 100 / total; | |
877 | ||
4aa9015f | 878 | fprintf(output, " # %8.3f %% of all cache refs ", ratio); |
d58f4c82 | 879 | |
d3d1e86d IM |
880 | } else if (perf_evsel__match(evsel, HARDWARE, HW_STALLED_CYCLES_FRONTEND)) { |
881 | print_stalled_cycles_frontend(cpu, evsel, avg); | |
129c04cb | 882 | } else if (perf_evsel__match(evsel, HARDWARE, HW_STALLED_CYCLES_BACKEND)) { |
d3d1e86d | 883 | print_stalled_cycles_backend(cpu, evsel, avg); |
481f988a | 884 | } else if (perf_evsel__match(evsel, HARDWARE, HW_CPU_CYCLES)) { |
f5b4a9c3 | 885 | total = avg_stats(&runtime_nsecs_stats[cpu]); |
c7f7fea3 IM |
886 | |
887 | if (total) | |
481f988a | 888 | ratio = 1.0 * avg / total; |
c7f7fea3 | 889 | |
4aa9015f | 890 | fprintf(output, " # %8.3f GHz ", ratio); |
481f988a | 891 | } else if (runtime_nsecs_stats[cpu].n != 0) { |
5fde2523 NK |
892 | char unit = 'M'; |
893 | ||
481f988a | 894 | total = avg_stats(&runtime_nsecs_stats[cpu]); |
11ba2b85 IM |
895 | |
896 | if (total) | |
481f988a | 897 | ratio = 1000.0 * avg / total; |
5fde2523 NK |
898 | if (ratio < 0.001) { |
899 | ratio *= 1000; | |
900 | unit = 'K'; | |
901 | } | |
11ba2b85 | 902 | |
5fde2523 | 903 | fprintf(output, " # %8.3f %c/sec ", ratio, unit); |
a5d243d0 | 904 | } else { |
4aa9015f | 905 | fprintf(output, " "); |
44175b6f | 906 | } |
44175b6f IM |
907 | } |
908 | ||
86ee6e18 | 909 | static void print_aggr(char *prefix) |
d7e7a451 SE |
910 | { |
911 | struct perf_evsel *counter; | |
86ee6e18 | 912 | int cpu, s, s2, id, nr; |
d7e7a451 | 913 | u64 ena, run, val; |
d7e7a451 | 914 | |
86ee6e18 | 915 | if (!(aggr_map || aggr_get_id)) |
d7e7a451 SE |
916 | return; |
917 | ||
86ee6e18 SE |
918 | for (s = 0; s < aggr_map->nr; s++) { |
919 | id = aggr_map->map[s]; | |
d7e7a451 SE |
920 | list_for_each_entry(counter, &evsel_list->entries, node) { |
921 | val = ena = run = 0; | |
922 | nr = 0; | |
923 | for (cpu = 0; cpu < perf_evsel__nr_cpus(counter); cpu++) { | |
86ee6e18 SE |
924 | s2 = aggr_get_id(evsel_list->cpus, cpu); |
925 | if (s2 != id) | |
d7e7a451 SE |
926 | continue; |
927 | val += counter->counts->cpu[cpu].val; | |
928 | ena += counter->counts->cpu[cpu].ena; | |
929 | run += counter->counts->cpu[cpu].run; | |
930 | nr++; | |
931 | } | |
932 | if (prefix) | |
933 | fprintf(output, "%s", prefix); | |
934 | ||
935 | if (run == 0 || ena == 0) { | |
86ee6e18 SE |
936 | aggr_printout(counter, cpu, nr); |
937 | ||
938 | fprintf(output, "%*s%s%*s", | |
d7e7a451 SE |
939 | csv_output ? 0 : 18, |
940 | counter->supported ? CNTR_NOT_COUNTED : CNTR_NOT_SUPPORTED, | |
941 | csv_sep, | |
942 | csv_output ? 0 : -24, | |
943 | perf_evsel__name(counter)); | |
86ee6e18 | 944 | |
d7e7a451 SE |
945 | if (counter->cgrp) |
946 | fprintf(output, "%s%s", | |
947 | csv_sep, counter->cgrp->name); | |
948 | ||
949 | fputc('\n', output); | |
950 | continue; | |
951 | } | |
952 | ||
953 | if (nsec_counter(counter)) | |
86ee6e18 | 954 | nsec_printout(id, nr, counter, val); |
d7e7a451 | 955 | else |
86ee6e18 | 956 | abs_printout(id, nr, counter, val); |
d7e7a451 SE |
957 | |
958 | if (!csv_output) { | |
959 | print_noise(counter, 1.0); | |
960 | ||
961 | if (run != ena) | |
962 | fprintf(output, " (%.2f%%)", | |
963 | 100.0 * run / ena); | |
964 | } | |
965 | fputc('\n', output); | |
966 | } | |
967 | } | |
968 | } | |
969 | ||
2996f5dd IM |
970 | /* |
971 | * Print out the results of a single counter: | |
f5b4a9c3 | 972 | * aggregated counts in system-wide mode |
2996f5dd | 973 | */ |
13370a9b | 974 | static void print_counter_aggr(struct perf_evsel *counter, char *prefix) |
2996f5dd | 975 | { |
69aad6f1 ACM |
976 | struct perf_stat *ps = counter->priv; |
977 | double avg = avg_stats(&ps->res_stats[0]); | |
c52b12ed | 978 | int scaled = counter->counts->scaled; |
2996f5dd | 979 | |
13370a9b SE |
980 | if (prefix) |
981 | fprintf(output, "%s", prefix); | |
982 | ||
2996f5dd | 983 | if (scaled == -1) { |
4aa9015f | 984 | fprintf(output, "%*s%s%*s", |
d7470b6a | 985 | csv_output ? 0 : 18, |
2cee77c4 | 986 | counter->supported ? CNTR_NOT_COUNTED : CNTR_NOT_SUPPORTED, |
023695d9 SE |
987 | csv_sep, |
988 | csv_output ? 0 : -24, | |
7289f83c | 989 | perf_evsel__name(counter)); |
023695d9 SE |
990 | |
991 | if (counter->cgrp) | |
4aa9015f | 992 | fprintf(output, "%s%s", csv_sep, counter->cgrp->name); |
023695d9 | 993 | |
4aa9015f | 994 | fputc('\n', output); |
2996f5dd IM |
995 | return; |
996 | } | |
c04f5e5d | 997 | |
44175b6f | 998 | if (nsec_counter(counter)) |
d7e7a451 | 999 | nsec_printout(-1, 0, counter, avg); |
44175b6f | 1000 | else |
d7e7a451 | 1001 | abs_printout(-1, 0, counter, avg); |
849abde9 | 1002 | |
3ae9a34d ZH |
1003 | print_noise(counter, avg); |
1004 | ||
d7470b6a | 1005 | if (csv_output) { |
4aa9015f | 1006 | fputc('\n', output); |
d7470b6a SE |
1007 | return; |
1008 | } | |
1009 | ||
506d4bc8 PZ |
1010 | if (scaled) { |
1011 | double avg_enabled, avg_running; | |
1012 | ||
69aad6f1 ACM |
1013 | avg_enabled = avg_stats(&ps->res_stats[1]); |
1014 | avg_running = avg_stats(&ps->res_stats[2]); | |
d7c29318 | 1015 | |
4aa9015f | 1016 | fprintf(output, " [%5.2f%%]", 100 * avg_running / avg_enabled); |
506d4bc8 | 1017 | } |
4aa9015f | 1018 | fprintf(output, "\n"); |
c04f5e5d IM |
1019 | } |
1020 | ||
f5b4a9c3 SE |
1021 | /* |
1022 | * Print out the results of a single counter: | |
1023 | * does not use aggregated count in system-wide | |
1024 | */ | |
13370a9b | 1025 | static void print_counter(struct perf_evsel *counter, char *prefix) |
f5b4a9c3 SE |
1026 | { |
1027 | u64 ena, run, val; | |
1028 | int cpu; | |
1029 | ||
7ae92e74 | 1030 | for (cpu = 0; cpu < perf_evsel__nr_cpus(counter); cpu++) { |
c52b12ed ACM |
1031 | val = counter->counts->cpu[cpu].val; |
1032 | ena = counter->counts->cpu[cpu].ena; | |
1033 | run = counter->counts->cpu[cpu].run; | |
13370a9b SE |
1034 | |
1035 | if (prefix) | |
1036 | fprintf(output, "%s", prefix); | |
1037 | ||
f5b4a9c3 | 1038 | if (run == 0 || ena == 0) { |
4aa9015f | 1039 | fprintf(output, "CPU%*d%s%*s%s%*s", |
d7470b6a | 1040 | csv_output ? 0 : -4, |
7ae92e74 | 1041 | perf_evsel__cpus(counter)->map[cpu], csv_sep, |
d7470b6a | 1042 | csv_output ? 0 : 18, |
2cee77c4 DA |
1043 | counter->supported ? CNTR_NOT_COUNTED : CNTR_NOT_SUPPORTED, |
1044 | csv_sep, | |
023695d9 | 1045 | csv_output ? 0 : -24, |
7289f83c | 1046 | perf_evsel__name(counter)); |
f5b4a9c3 | 1047 | |
023695d9 | 1048 | if (counter->cgrp) |
4aa9015f SE |
1049 | fprintf(output, "%s%s", |
1050 | csv_sep, counter->cgrp->name); | |
023695d9 | 1051 | |
4aa9015f | 1052 | fputc('\n', output); |
f5b4a9c3 SE |
1053 | continue; |
1054 | } | |
1055 | ||
1056 | if (nsec_counter(counter)) | |
d7e7a451 | 1057 | nsec_printout(cpu, 0, counter, val); |
f5b4a9c3 | 1058 | else |
d7e7a451 | 1059 | abs_printout(cpu, 0, counter, val); |
f5b4a9c3 | 1060 | |
d7470b6a SE |
1061 | if (!csv_output) { |
1062 | print_noise(counter, 1.0); | |
f5b4a9c3 | 1063 | |
c6264def | 1064 | if (run != ena) |
4aa9015f SE |
1065 | fprintf(output, " (%.2f%%)", |
1066 | 100.0 * run / ena); | |
f5b4a9c3 | 1067 | } |
4aa9015f | 1068 | fputc('\n', output); |
f5b4a9c3 SE |
1069 | } |
1070 | } | |
1071 | ||
42202dd5 IM |
1072 | static void print_stat(int argc, const char **argv) |
1073 | { | |
69aad6f1 ACM |
1074 | struct perf_evsel *counter; |
1075 | int i; | |
42202dd5 | 1076 | |
ddcacfa0 IM |
1077 | fflush(stdout); |
1078 | ||
d7470b6a | 1079 | if (!csv_output) { |
4aa9015f SE |
1080 | fprintf(output, "\n"); |
1081 | fprintf(output, " Performance counter stats for "); | |
aa22dd49 | 1082 | if (!perf_target__has_task(&target)) { |
4aa9015f | 1083 | fprintf(output, "\'%s", argv[0]); |
d7470b6a | 1084 | for (i = 1; i < argc; i++) |
4aa9015f | 1085 | fprintf(output, " %s", argv[i]); |
20f946b4 NK |
1086 | } else if (target.pid) |
1087 | fprintf(output, "process id \'%s", target.pid); | |
d7470b6a | 1088 | else |
20f946b4 | 1089 | fprintf(output, "thread id \'%s", target.tid); |
44db76c8 | 1090 | |
4aa9015f | 1091 | fprintf(output, "\'"); |
d7470b6a | 1092 | if (run_count > 1) |
4aa9015f SE |
1093 | fprintf(output, " (%d runs)", run_count); |
1094 | fprintf(output, ":\n\n"); | |
d7470b6a | 1095 | } |
2996f5dd | 1096 | |
86ee6e18 SE |
1097 | switch (aggr_mode) { |
1098 | case AGGR_SOCKET: | |
1099 | print_aggr(NULL); | |
1100 | break; | |
1101 | case AGGR_GLOBAL: | |
361c99a6 | 1102 | list_for_each_entry(counter, &evsel_list->entries, node) |
13370a9b | 1103 | print_counter_aggr(counter, NULL); |
86ee6e18 SE |
1104 | break; |
1105 | case AGGR_NONE: | |
1106 | list_for_each_entry(counter, &evsel_list->entries, node) | |
1107 | print_counter(counter, NULL); | |
1108 | break; | |
1109 | default: | |
1110 | break; | |
f5b4a9c3 | 1111 | } |
ddcacfa0 | 1112 | |
d7470b6a | 1113 | if (!csv_output) { |
c3305257 | 1114 | if (!null_run) |
4aa9015f SE |
1115 | fprintf(output, "\n"); |
1116 | fprintf(output, " %17.9f seconds time elapsed", | |
d7470b6a SE |
1117 | avg_stats(&walltime_nsecs_stats)/1e9); |
1118 | if (run_count > 1) { | |
4aa9015f | 1119 | fprintf(output, " "); |
f99844cb IM |
1120 | print_noise_pct(stddev_stats(&walltime_nsecs_stats), |
1121 | avg_stats(&walltime_nsecs_stats)); | |
d7470b6a | 1122 | } |
4aa9015f | 1123 | fprintf(output, "\n\n"); |
566747e6 | 1124 | } |
ddcacfa0 IM |
1125 | } |
1126 | ||
f7b7c26e PZ |
1127 | static volatile int signr = -1; |
1128 | ||
5242519b | 1129 | static void skip_signal(int signo) |
ddcacfa0 | 1130 | { |
13370a9b | 1131 | if ((child_pid == -1) || interval) |
60666c63 LW |
1132 | done = 1; |
1133 | ||
f7b7c26e PZ |
1134 | signr = signo; |
1135 | } | |
1136 | ||
1137 | static void sig_atexit(void) | |
1138 | { | |
933da83a CW |
1139 | if (child_pid != -1) |
1140 | kill(child_pid, SIGTERM); | |
1141 | ||
f7b7c26e PZ |
1142 | if (signr == -1) |
1143 | return; | |
1144 | ||
1145 | signal(signr, SIG_DFL); | |
1146 | kill(getpid(), signr); | |
5242519b IM |
1147 | } |
1148 | ||
1d037ca1 IT |
1149 | static int stat__set_big_num(const struct option *opt __maybe_unused, |
1150 | const char *s __maybe_unused, int unset) | |
d7470b6a SE |
1151 | { |
1152 | big_num_opt = unset ? 0 : 1; | |
1153 | return 0; | |
1154 | } | |
1155 | ||
86ee6e18 SE |
1156 | static int perf_stat_init_aggr_mode(void) |
1157 | { | |
1158 | switch (aggr_mode) { | |
1159 | case AGGR_SOCKET: | |
1160 | if (cpu_map__build_socket_map(evsel_list->cpus, &aggr_map)) { | |
1161 | perror("cannot build socket map"); | |
1162 | return -1; | |
1163 | } | |
1164 | aggr_get_id = cpu_map__get_socket; | |
1165 | break; | |
1166 | case AGGR_NONE: | |
1167 | case AGGR_GLOBAL: | |
1168 | default: | |
1169 | break; | |
1170 | } | |
1171 | return 0; | |
1172 | } | |
1173 | ||
1174 | ||
2cba3ffb IM |
1175 | /* |
1176 | * Add default attributes, if there were no attributes specified or | |
1177 | * if -d/--detailed, -d -d or -d -d -d is used: | |
1178 | */ | |
1179 | static int add_default_attributes(void) | |
1180 | { | |
b070a547 ACM |
1181 | struct perf_event_attr default_attrs[] = { |
1182 | ||
1183 | { .type = PERF_TYPE_SOFTWARE, .config = PERF_COUNT_SW_TASK_CLOCK }, | |
1184 | { .type = PERF_TYPE_SOFTWARE, .config = PERF_COUNT_SW_CONTEXT_SWITCHES }, | |
1185 | { .type = PERF_TYPE_SOFTWARE, .config = PERF_COUNT_SW_CPU_MIGRATIONS }, | |
1186 | { .type = PERF_TYPE_SOFTWARE, .config = PERF_COUNT_SW_PAGE_FAULTS }, | |
1187 | ||
1188 | { .type = PERF_TYPE_HARDWARE, .config = PERF_COUNT_HW_CPU_CYCLES }, | |
1189 | { .type = PERF_TYPE_HARDWARE, .config = PERF_COUNT_HW_STALLED_CYCLES_FRONTEND }, | |
1190 | { .type = PERF_TYPE_HARDWARE, .config = PERF_COUNT_HW_STALLED_CYCLES_BACKEND }, | |
1191 | { .type = PERF_TYPE_HARDWARE, .config = PERF_COUNT_HW_INSTRUCTIONS }, | |
1192 | { .type = PERF_TYPE_HARDWARE, .config = PERF_COUNT_HW_BRANCH_INSTRUCTIONS }, | |
1193 | { .type = PERF_TYPE_HARDWARE, .config = PERF_COUNT_HW_BRANCH_MISSES }, | |
1194 | ||
1195 | }; | |
1196 | ||
1197 | /* | |
1198 | * Detailed stats (-d), covering the L1 and last level data caches: | |
1199 | */ | |
1200 | struct perf_event_attr detailed_attrs[] = { | |
1201 | ||
1202 | { .type = PERF_TYPE_HW_CACHE, | |
1203 | .config = | |
1204 | PERF_COUNT_HW_CACHE_L1D << 0 | | |
1205 | (PERF_COUNT_HW_CACHE_OP_READ << 8) | | |
1206 | (PERF_COUNT_HW_CACHE_RESULT_ACCESS << 16) }, | |
1207 | ||
1208 | { .type = PERF_TYPE_HW_CACHE, | |
1209 | .config = | |
1210 | PERF_COUNT_HW_CACHE_L1D << 0 | | |
1211 | (PERF_COUNT_HW_CACHE_OP_READ << 8) | | |
1212 | (PERF_COUNT_HW_CACHE_RESULT_MISS << 16) }, | |
1213 | ||
1214 | { .type = PERF_TYPE_HW_CACHE, | |
1215 | .config = | |
1216 | PERF_COUNT_HW_CACHE_LL << 0 | | |
1217 | (PERF_COUNT_HW_CACHE_OP_READ << 8) | | |
1218 | (PERF_COUNT_HW_CACHE_RESULT_ACCESS << 16) }, | |
1219 | ||
1220 | { .type = PERF_TYPE_HW_CACHE, | |
1221 | .config = | |
1222 | PERF_COUNT_HW_CACHE_LL << 0 | | |
1223 | (PERF_COUNT_HW_CACHE_OP_READ << 8) | | |
1224 | (PERF_COUNT_HW_CACHE_RESULT_MISS << 16) }, | |
1225 | }; | |
1226 | ||
1227 | /* | |
1228 | * Very detailed stats (-d -d), covering the instruction cache and the TLB caches: | |
1229 | */ | |
1230 | struct perf_event_attr very_detailed_attrs[] = { | |
1231 | ||
1232 | { .type = PERF_TYPE_HW_CACHE, | |
1233 | .config = | |
1234 | PERF_COUNT_HW_CACHE_L1I << 0 | | |
1235 | (PERF_COUNT_HW_CACHE_OP_READ << 8) | | |
1236 | (PERF_COUNT_HW_CACHE_RESULT_ACCESS << 16) }, | |
1237 | ||
1238 | { .type = PERF_TYPE_HW_CACHE, | |
1239 | .config = | |
1240 | PERF_COUNT_HW_CACHE_L1I << 0 | | |
1241 | (PERF_COUNT_HW_CACHE_OP_READ << 8) | | |
1242 | (PERF_COUNT_HW_CACHE_RESULT_MISS << 16) }, | |
1243 | ||
1244 | { .type = PERF_TYPE_HW_CACHE, | |
1245 | .config = | |
1246 | PERF_COUNT_HW_CACHE_DTLB << 0 | | |
1247 | (PERF_COUNT_HW_CACHE_OP_READ << 8) | | |
1248 | (PERF_COUNT_HW_CACHE_RESULT_ACCESS << 16) }, | |
1249 | ||
1250 | { .type = PERF_TYPE_HW_CACHE, | |
1251 | .config = | |
1252 | PERF_COUNT_HW_CACHE_DTLB << 0 | | |
1253 | (PERF_COUNT_HW_CACHE_OP_READ << 8) | | |
1254 | (PERF_COUNT_HW_CACHE_RESULT_MISS << 16) }, | |
1255 | ||
1256 | { .type = PERF_TYPE_HW_CACHE, | |
1257 | .config = | |
1258 | PERF_COUNT_HW_CACHE_ITLB << 0 | | |
1259 | (PERF_COUNT_HW_CACHE_OP_READ << 8) | | |
1260 | (PERF_COUNT_HW_CACHE_RESULT_ACCESS << 16) }, | |
1261 | ||
1262 | { .type = PERF_TYPE_HW_CACHE, | |
1263 | .config = | |
1264 | PERF_COUNT_HW_CACHE_ITLB << 0 | | |
1265 | (PERF_COUNT_HW_CACHE_OP_READ << 8) | | |
1266 | (PERF_COUNT_HW_CACHE_RESULT_MISS << 16) }, | |
1267 | ||
1268 | }; | |
1269 | ||
1270 | /* | |
1271 | * Very, very detailed stats (-d -d -d), adding prefetch events: | |
1272 | */ | |
1273 | struct perf_event_attr very_very_detailed_attrs[] = { | |
1274 | ||
1275 | { .type = PERF_TYPE_HW_CACHE, | |
1276 | .config = | |
1277 | PERF_COUNT_HW_CACHE_L1D << 0 | | |
1278 | (PERF_COUNT_HW_CACHE_OP_PREFETCH << 8) | | |
1279 | (PERF_COUNT_HW_CACHE_RESULT_ACCESS << 16) }, | |
1280 | ||
1281 | { .type = PERF_TYPE_HW_CACHE, | |
1282 | .config = | |
1283 | PERF_COUNT_HW_CACHE_L1D << 0 | | |
1284 | (PERF_COUNT_HW_CACHE_OP_PREFETCH << 8) | | |
1285 | (PERF_COUNT_HW_CACHE_RESULT_MISS << 16) }, | |
1286 | }; | |
1287 | ||
2cba3ffb IM |
1288 | /* Set attrs if no event is selected and !null_run: */ |
1289 | if (null_run) | |
1290 | return 0; | |
1291 | ||
1292 | if (!evsel_list->nr_entries) { | |
79695e1b | 1293 | if (perf_evlist__add_default_attrs(evsel_list, default_attrs) < 0) |
50d08e47 | 1294 | return -1; |
2cba3ffb IM |
1295 | } |
1296 | ||
1297 | /* Detailed events get appended to the event list: */ | |
1298 | ||
1299 | if (detailed_run < 1) | |
1300 | return 0; | |
1301 | ||
1302 | /* Append detailed run extra attributes: */ | |
79695e1b | 1303 | if (perf_evlist__add_default_attrs(evsel_list, detailed_attrs) < 0) |
50d08e47 | 1304 | return -1; |
2cba3ffb IM |
1305 | |
1306 | if (detailed_run < 2) | |
1307 | return 0; | |
1308 | ||
1309 | /* Append very detailed run extra attributes: */ | |
79695e1b | 1310 | if (perf_evlist__add_default_attrs(evsel_list, very_detailed_attrs) < 0) |
50d08e47 | 1311 | return -1; |
2cba3ffb IM |
1312 | |
1313 | if (detailed_run < 3) | |
1314 | return 0; | |
1315 | ||
1316 | /* Append very, very detailed run extra attributes: */ | |
79695e1b | 1317 | return perf_evlist__add_default_attrs(evsel_list, very_very_detailed_attrs); |
2cba3ffb IM |
1318 | } |
1319 | ||
1d037ca1 | 1320 | int cmd_stat(int argc, const char **argv, const char *prefix __maybe_unused) |
5242519b | 1321 | { |
1f16c575 | 1322 | bool append_file = false; |
b070a547 ACM |
1323 | int output_fd = 0; |
1324 | const char *output_name = NULL; | |
1325 | const struct option options[] = { | |
1326 | OPT_CALLBACK('e', "event", &evsel_list, "event", | |
1327 | "event selector. use 'perf list' to list available events", | |
1328 | parse_events_option), | |
1329 | OPT_CALLBACK(0, "filter", &evsel_list, "filter", | |
1330 | "event filter", parse_filter), | |
1331 | OPT_BOOLEAN('i', "no-inherit", &no_inherit, | |
1332 | "child tasks do not inherit counters"), | |
1333 | OPT_STRING('p', "pid", &target.pid, "pid", | |
1334 | "stat events on existing process id"), | |
1335 | OPT_STRING('t', "tid", &target.tid, "tid", | |
1336 | "stat events on existing thread id"), | |
1337 | OPT_BOOLEAN('a', "all-cpus", &target.system_wide, | |
1338 | "system-wide collection from all CPUs"), | |
1339 | OPT_BOOLEAN('g', "group", &group, | |
1340 | "put the counters into a counter group"), | |
1341 | OPT_BOOLEAN('c', "scale", &scale, "scale/normalize counters"), | |
1342 | OPT_INCR('v', "verbose", &verbose, | |
1343 | "be more verbose (show counter open errors, etc)"), | |
1344 | OPT_INTEGER('r', "repeat", &run_count, | |
a7e191c3 | 1345 | "repeat command and print average + stddev (max: 100, forever: 0)"), |
b070a547 ACM |
1346 | OPT_BOOLEAN('n', "null", &null_run, |
1347 | "null run - dont start any counters"), | |
1348 | OPT_INCR('d', "detailed", &detailed_run, | |
1349 | "detailed run - start a lot of events"), | |
1350 | OPT_BOOLEAN('S', "sync", &sync_run, | |
1351 | "call sync() before starting a run"), | |
1352 | OPT_CALLBACK_NOOPT('B', "big-num", NULL, NULL, | |
1353 | "print large numbers with thousands\' separators", | |
1354 | stat__set_big_num), | |
1355 | OPT_STRING('C', "cpu", &target.cpu_list, "cpu", | |
1356 | "list of cpus to monitor in system-wide"), | |
86ee6e18 SE |
1357 | OPT_SET_UINT('A', "no-aggr", &aggr_mode, |
1358 | "disable CPU count aggregation", AGGR_NONE), | |
b070a547 ACM |
1359 | OPT_STRING('x', "field-separator", &csv_sep, "separator", |
1360 | "print counts with custom separator"), | |
1361 | OPT_CALLBACK('G', "cgroup", &evsel_list, "name", | |
1362 | "monitor event in cgroup name only", parse_cgroups), | |
1363 | OPT_STRING('o', "output", &output_name, "file", "output file name"), | |
1364 | OPT_BOOLEAN(0, "append", &append_file, "append to the output file"), | |
1365 | OPT_INTEGER(0, "log-fd", &output_fd, | |
1366 | "log output to fd, instead of stderr"), | |
1f16c575 PZ |
1367 | OPT_STRING(0, "pre", &pre_cmd, "command", |
1368 | "command to run prior to the measured command"), | |
1369 | OPT_STRING(0, "post", &post_cmd, "command", | |
1370 | "command to run after to the measured command"), | |
13370a9b SE |
1371 | OPT_UINTEGER('I', "interval-print", &interval, |
1372 | "print counts at regular interval in ms (>= 100)"), | |
86ee6e18 SE |
1373 | OPT_SET_UINT(0, "aggr-socket", &aggr_mode, |
1374 | "aggregate counts per processor socket", AGGR_SOCKET), | |
b070a547 ACM |
1375 | OPT_END() |
1376 | }; | |
1377 | const char * const stat_usage[] = { | |
1378 | "perf stat [<options>] [<command>]", | |
1379 | NULL | |
1380 | }; | |
b070a547 | 1381 | int status = -ENOMEM, run_idx; |
4aa9015f | 1382 | const char *mode; |
42202dd5 | 1383 | |
5af52b51 SE |
1384 | setlocale(LC_ALL, ""); |
1385 | ||
334fe7a3 | 1386 | evsel_list = perf_evlist__new(); |
361c99a6 ACM |
1387 | if (evsel_list == NULL) |
1388 | return -ENOMEM; | |
1389 | ||
a0541234 AB |
1390 | argc = parse_options(argc, argv, options, stat_usage, |
1391 | PARSE_OPT_STOP_AT_NON_OPTION); | |
d7470b6a | 1392 | |
4aa9015f SE |
1393 | output = stderr; |
1394 | if (output_name && strcmp(output_name, "-")) | |
1395 | output = NULL; | |
1396 | ||
56f3bae7 JC |
1397 | if (output_name && output_fd) { |
1398 | fprintf(stderr, "cannot use both --output and --log-fd\n"); | |
1399 | usage_with_options(stat_usage, options); | |
1400 | } | |
fc3e4d07 SE |
1401 | |
1402 | if (output_fd < 0) { | |
1403 | fprintf(stderr, "argument to --log-fd must be a > 0\n"); | |
1404 | usage_with_options(stat_usage, options); | |
1405 | } | |
1406 | ||
4aa9015f SE |
1407 | if (!output) { |
1408 | struct timespec tm; | |
1409 | mode = append_file ? "a" : "w"; | |
1410 | ||
1411 | output = fopen(output_name, mode); | |
1412 | if (!output) { | |
1413 | perror("failed to create output file"); | |
fceda7fe | 1414 | return -1; |
4aa9015f SE |
1415 | } |
1416 | clock_gettime(CLOCK_REALTIME, &tm); | |
1417 | fprintf(output, "# started on %s\n", ctime(&tm.tv_sec)); | |
fc3e4d07 | 1418 | } else if (output_fd > 0) { |
56f3bae7 JC |
1419 | mode = append_file ? "a" : "w"; |
1420 | output = fdopen(output_fd, mode); | |
1421 | if (!output) { | |
1422 | perror("Failed opening logfd"); | |
1423 | return -errno; | |
1424 | } | |
4aa9015f SE |
1425 | } |
1426 | ||
d4ffd04d | 1427 | if (csv_sep) { |
d7470b6a | 1428 | csv_output = true; |
d4ffd04d JC |
1429 | if (!strcmp(csv_sep, "\\t")) |
1430 | csv_sep = "\t"; | |
1431 | } else | |
d7470b6a SE |
1432 | csv_sep = DEFAULT_SEPARATOR; |
1433 | ||
1434 | /* | |
1435 | * let the spreadsheet do the pretty-printing | |
1436 | */ | |
1437 | if (csv_output) { | |
61a9f324 | 1438 | /* User explicitly passed -B? */ |
d7470b6a SE |
1439 | if (big_num_opt == 1) { |
1440 | fprintf(stderr, "-B option not supported with -x\n"); | |
1441 | usage_with_options(stat_usage, options); | |
1442 | } else /* Nope, so disable big number formatting */ | |
1443 | big_num = false; | |
1444 | } else if (big_num_opt == 0) /* User passed --no-big-num */ | |
1445 | big_num = false; | |
1446 | ||
aa22dd49 | 1447 | if (!argc && !perf_target__has_task(&target)) |
5242519b | 1448 | usage_with_options(stat_usage, options); |
a7e191c3 | 1449 | if (run_count < 0) { |
42202dd5 | 1450 | usage_with_options(stat_usage, options); |
a7e191c3 FD |
1451 | } else if (run_count == 0) { |
1452 | forever = true; | |
1453 | run_count = 1; | |
1454 | } | |
ddcacfa0 | 1455 | |
023695d9 | 1456 | /* no_aggr, cgroup are for system-wide only */ |
86ee6e18 SE |
1457 | if ((aggr_mode != AGGR_GLOBAL || nr_cgroups) |
1458 | && !perf_target__has_cpu(&target)) { | |
023695d9 SE |
1459 | fprintf(stderr, "both cgroup and no-aggregation " |
1460 | "modes only available in system-wide mode\n"); | |
1461 | ||
f5b4a9c3 | 1462 | usage_with_options(stat_usage, options); |
86ee6e18 | 1463 | return -1; |
d7e7a451 SE |
1464 | } |
1465 | ||
2cba3ffb IM |
1466 | if (add_default_attributes()) |
1467 | goto out; | |
ddcacfa0 | 1468 | |
4bd0f2d2 | 1469 | perf_target__validate(&target); |
5c98d466 | 1470 | |
77a6f014 | 1471 | if (perf_evlist__create_maps(evsel_list, &target) < 0) { |
aa22dd49 | 1472 | if (perf_target__has_task(&target)) |
77a6f014 | 1473 | pr_err("Problems finding threads of monitor\n"); |
aa22dd49 | 1474 | if (perf_target__has_cpu(&target)) |
77a6f014 | 1475 | perror("failed to parse CPUs map"); |
ddcacfa0 | 1476 | |
c45c6ea2 | 1477 | usage_with_options(stat_usage, options); |
60d567e2 ACM |
1478 | return -1; |
1479 | } | |
13370a9b SE |
1480 | if (interval && interval < 100) { |
1481 | pr_err("print interval must be >= 100ms\n"); | |
1482 | usage_with_options(stat_usage, options); | |
1483 | return -1; | |
1484 | } | |
c45c6ea2 | 1485 | |
d134ffb9 ACM |
1486 | if (perf_evlist__alloc_stats(evsel_list, interval)) |
1487 | goto out_free_maps; | |
d6d901c2 | 1488 | |
86ee6e18 SE |
1489 | if (perf_stat_init_aggr_mode()) |
1490 | goto out; | |
1491 | ||
58d7e993 IM |
1492 | /* |
1493 | * We dont want to block the signals - that would cause | |
1494 | * child tasks to inherit that and Ctrl-C would not work. | |
1495 | * What we want is for Ctrl-C to work in the exec()-ed | |
1496 | * task, but being ignored by perf stat itself: | |
1497 | */ | |
f7b7c26e | 1498 | atexit(sig_atexit); |
a7e191c3 FD |
1499 | if (!forever) |
1500 | signal(SIGINT, skip_signal); | |
13370a9b | 1501 | signal(SIGCHLD, skip_signal); |
58d7e993 IM |
1502 | signal(SIGALRM, skip_signal); |
1503 | signal(SIGABRT, skip_signal); | |
1504 | ||
42202dd5 | 1505 | status = 0; |
a7e191c3 | 1506 | for (run_idx = 0; forever || run_idx < run_count; run_idx++) { |
42202dd5 | 1507 | if (run_count != 1 && verbose) |
4aa9015f SE |
1508 | fprintf(output, "[ perf stat: executing run #%d ... ]\n", |
1509 | run_idx + 1); | |
f9cef0a9 | 1510 | |
42202dd5 | 1511 | status = run_perf_stat(argc, argv); |
a7e191c3 FD |
1512 | if (forever && status != -1) { |
1513 | print_stat(argc, argv); | |
d134ffb9 | 1514 | perf_stat__reset_stats(evsel_list); |
a7e191c3 | 1515 | } |
42202dd5 IM |
1516 | } |
1517 | ||
a7e191c3 | 1518 | if (!forever && status != -1 && !interval) |
084ab9f8 | 1519 | print_stat(argc, argv); |
d134ffb9 ACM |
1520 | |
1521 | perf_evlist__free_stats(evsel_list); | |
1522 | out_free_maps: | |
7e2ed097 | 1523 | perf_evlist__delete_maps(evsel_list); |
0015e2e1 ACM |
1524 | out: |
1525 | perf_evlist__delete(evsel_list); | |
42202dd5 | 1526 | return status; |
ddcacfa0 | 1527 | } |