]>
Commit | Line | Data |
---|---|---|
0a02ad93 | 1 | #include "builtin.h" |
b1ffe8f3 | 2 | #include "perf.h" |
0a02ad93 IM |
3 | |
4 | #include "util/util.h" | |
ee29be62 | 5 | #include "util/evlist.h" |
0a02ad93 | 6 | #include "util/cache.h" |
e3f42609 | 7 | #include "util/evsel.h" |
0a02ad93 IM |
8 | #include "util/symbol.h" |
9 | #include "util/thread.h" | |
10 | #include "util/header.h" | |
94c744b6 | 11 | #include "util/session.h" |
45694aa7 | 12 | #include "util/tool.h" |
57480d2c | 13 | #include "util/cloexec.h" |
a151a37a | 14 | #include "util/thread_map.h" |
8cd91195 | 15 | #include "util/color.h" |
49394a2a | 16 | #include "util/stat.h" |
6c973c90 | 17 | #include "util/callchain.h" |
853b7407 | 18 | #include "util/time-utils.h" |
0a02ad93 | 19 | |
4b6ab94e | 20 | #include <subcmd/parse-options.h> |
b1ffe8f3 | 21 | #include "util/trace-event.h" |
0a02ad93 | 22 | |
0a02ad93 IM |
23 | #include "util/debug.h" |
24 | ||
49394a2a | 25 | #include <linux/log2.h> |
b1ffe8f3 | 26 | #include <sys/prctl.h> |
7b78f136 | 27 | #include <sys/resource.h> |
0a02ad93 | 28 | |
b1ffe8f3 IM |
29 | #include <semaphore.h> |
30 | #include <pthread.h> | |
31 | #include <math.h> | |
cb06ac25 | 32 | #include <api/fs/fs.h> |
4fc76e49 | 33 | #include <linux/time64.h> |
419ab0d6 | 34 | |
b1ffe8f3 IM |
35 | #define PR_SET_NAME 15 /* Set process name */ |
36 | #define MAX_CPUS 4096 | |
b1ffe8f3 IM |
37 | #define COMM_LEN 20 |
38 | #define SYM_LEN 129 | |
a35e27d0 | 39 | #define MAX_PID 1024000 |
ec156764 | 40 | |
39aeb52f | 41 | struct sched_atom; |
ec156764 | 42 | |
b1ffe8f3 IM |
43 | struct task_desc { |
44 | unsigned long nr; | |
45 | unsigned long pid; | |
46 | char comm[COMM_LEN]; | |
ec156764 | 47 | |
b1ffe8f3 IM |
48 | unsigned long nr_events; |
49 | unsigned long curr_event; | |
39aeb52f | 50 | struct sched_atom **atoms; |
b1ffe8f3 IM |
51 | |
52 | pthread_t thread; | |
53 | sem_t sleep_sem; | |
ec156764 | 54 | |
b1ffe8f3 IM |
55 | sem_t ready_for_work; |
56 | sem_t work_done_sem; | |
57 | ||
58 | u64 cpu_usage; | |
59 | }; | |
60 | ||
61 | enum sched_event_type { | |
62 | SCHED_EVENT_RUN, | |
63 | SCHED_EVENT_SLEEP, | |
64 | SCHED_EVENT_WAKEUP, | |
55ffb7a6 | 65 | SCHED_EVENT_MIGRATION, |
b1ffe8f3 IM |
66 | }; |
67 | ||
39aeb52f | 68 | struct sched_atom { |
b1ffe8f3 | 69 | enum sched_event_type type; |
eed05fe7 | 70 | int specific_wait; |
b1ffe8f3 IM |
71 | u64 timestamp; |
72 | u64 duration; | |
73 | unsigned long nr; | |
b1ffe8f3 IM |
74 | sem_t *wait_sem; |
75 | struct task_desc *wakee; | |
76 | }; | |
77 | ||
e936e8e4 | 78 | #define TASK_STATE_TO_CHAR_STR "RSDTtZXxKWP" |
b1ffe8f3 IM |
79 | |
80 | enum thread_state { | |
81 | THREAD_SLEEPING = 0, | |
82 | THREAD_WAIT_CPU, | |
83 | THREAD_SCHED_IN, | |
84 | THREAD_IGNORE | |
85 | }; | |
86 | ||
87 | struct work_atom { | |
88 | struct list_head list; | |
89 | enum thread_state state; | |
aa1ab9d2 | 90 | u64 sched_out_time; |
b1ffe8f3 IM |
91 | u64 wake_up_time; |
92 | u64 sched_in_time; | |
93 | u64 runtime; | |
94 | }; | |
95 | ||
39aeb52f | 96 | struct work_atoms { |
97 | struct list_head work_list; | |
b1ffe8f3 IM |
98 | struct thread *thread; |
99 | struct rb_node node; | |
100 | u64 max_lat; | |
3786310a | 101 | u64 max_lat_at; |
b1ffe8f3 IM |
102 | u64 total_lat; |
103 | u64 nb_atoms; | |
104 | u64 total_runtime; | |
2f80dd44 | 105 | int num_merged; |
b1ffe8f3 IM |
106 | }; |
107 | ||
39aeb52f | 108 | typedef int (*sort_fn_t)(struct work_atoms *, struct work_atoms *); |
b1ffe8f3 | 109 | |
9ec3f4e4 | 110 | struct perf_sched; |
0e9b07e5 | 111 | |
9ec3f4e4 ACM |
112 | struct trace_sched_handler { |
113 | int (*switch_event)(struct perf_sched *sched, struct perf_evsel *evsel, | |
114 | struct perf_sample *sample, struct machine *machine); | |
0e9b07e5 | 115 | |
9ec3f4e4 ACM |
116 | int (*runtime_event)(struct perf_sched *sched, struct perf_evsel *evsel, |
117 | struct perf_sample *sample, struct machine *machine); | |
0e9b07e5 | 118 | |
9ec3f4e4 ACM |
119 | int (*wakeup_event)(struct perf_sched *sched, struct perf_evsel *evsel, |
120 | struct perf_sample *sample, struct machine *machine); | |
0e9b07e5 | 121 | |
cb627505 DA |
122 | /* PERF_RECORD_FORK event, not sched_process_fork tracepoint */ |
123 | int (*fork_event)(struct perf_sched *sched, union perf_event *event, | |
124 | struct machine *machine); | |
0e9b07e5 ACM |
125 | |
126 | int (*migrate_task_event)(struct perf_sched *sched, | |
9ec3f4e4 ACM |
127 | struct perf_evsel *evsel, |
128 | struct perf_sample *sample, | |
129 | struct machine *machine); | |
0e9b07e5 ACM |
130 | }; |
131 | ||
a151a37a | 132 | #define COLOR_PIDS PERF_COLOR_BLUE |
cf294f24 | 133 | #define COLOR_CPUS PERF_COLOR_BG_RED |
a151a37a | 134 | |
99623c62 JO |
135 | struct perf_sched_map { |
136 | DECLARE_BITMAP(comp_cpus_mask, MAX_CPUS); | |
137 | int *comp_cpus; | |
138 | bool comp; | |
a151a37a JO |
139 | struct thread_map *color_pids; |
140 | const char *color_pids_str; | |
cf294f24 JO |
141 | struct cpu_map *color_cpus; |
142 | const char *color_cpus_str; | |
73643bb6 JO |
143 | struct cpu_map *cpus; |
144 | const char *cpus_str; | |
99623c62 JO |
145 | }; |
146 | ||
0e9b07e5 ACM |
147 | struct perf_sched { |
148 | struct perf_tool tool; | |
0e9b07e5 ACM |
149 | const char *sort_order; |
150 | unsigned long nr_tasks; | |
cb06ac25 | 151 | struct task_desc **pid_to_task; |
0e9b07e5 ACM |
152 | struct task_desc **tasks; |
153 | const struct trace_sched_handler *tp_handler; | |
154 | pthread_mutex_t start_work_mutex; | |
155 | pthread_mutex_t work_done_wait_mutex; | |
156 | int profile_cpu; | |
157 | /* | |
158 | * Track the current task - that way we can know whether there's any | |
159 | * weird events, such as a task being switched away that is not current. | |
160 | */ | |
161 | int max_cpu; | |
162 | u32 curr_pid[MAX_CPUS]; | |
163 | struct thread *curr_thread[MAX_CPUS]; | |
164 | char next_shortname1; | |
165 | char next_shortname2; | |
166 | unsigned int replay_repeat; | |
167 | unsigned long nr_run_events; | |
168 | unsigned long nr_sleep_events; | |
169 | unsigned long nr_wakeup_events; | |
170 | unsigned long nr_sleep_corrections; | |
171 | unsigned long nr_run_events_optimized; | |
172 | unsigned long targetless_wakeups; | |
173 | unsigned long multitarget_wakeups; | |
174 | unsigned long nr_runs; | |
175 | unsigned long nr_timestamps; | |
176 | unsigned long nr_unordered_timestamps; | |
0e9b07e5 ACM |
177 | unsigned long nr_context_switch_bugs; |
178 | unsigned long nr_events; | |
179 | unsigned long nr_lost_chunks; | |
180 | unsigned long nr_lost_events; | |
181 | u64 run_measurement_overhead; | |
182 | u64 sleep_measurement_overhead; | |
183 | u64 start_time; | |
184 | u64 cpu_usage; | |
185 | u64 runavg_cpu_usage; | |
186 | u64 parent_cpu_usage; | |
187 | u64 runavg_parent_cpu_usage; | |
188 | u64 sum_runtime; | |
189 | u64 sum_fluct; | |
190 | u64 run_avg; | |
191 | u64 all_runtime; | |
192 | u64 all_count; | |
193 | u64 cpu_last_switched[MAX_CPUS]; | |
2f80dd44 | 194 | struct rb_root atom_root, sorted_atom_root, merged_atom_root; |
0e9b07e5 | 195 | struct list_head sort_list, cmp_pid; |
939cda52 | 196 | bool force; |
2f80dd44 | 197 | bool skip_merge; |
99623c62 | 198 | struct perf_sched_map map; |
52df138c DA |
199 | |
200 | /* options for timehist command */ | |
201 | bool summary; | |
202 | bool summary_only; | |
699b5b92 | 203 | bool idle_hist; |
6c973c90 DA |
204 | bool show_callchain; |
205 | unsigned int max_stack; | |
a407b067 | 206 | bool show_cpu_visual; |
fc1469f1 | 207 | bool show_wakeups; |
350f54fa | 208 | bool show_migrations; |
52df138c | 209 | u64 skipped_samples; |
853b7407 DA |
210 | const char *time_str; |
211 | struct perf_time_interval ptime; | |
9396c9cb | 212 | struct perf_time_interval hist_time; |
0e9b07e5 | 213 | }; |
b1ffe8f3 | 214 | |
49394a2a DA |
215 | /* per thread run time data */ |
216 | struct thread_runtime { | |
217 | u64 last_time; /* time of previous sched in/out event */ | |
218 | u64 dt_run; /* run time */ | |
219 | u64 dt_wait; /* time between CPU access (off cpu) */ | |
220 | u64 dt_delay; /* time between wakeup and sched-in */ | |
221 | u64 ready_to_run; /* time of wakeup */ | |
222 | ||
223 | struct stats run_stats; | |
224 | u64 total_run_time; | |
350f54fa DA |
225 | |
226 | u64 migrations; | |
49394a2a DA |
227 | }; |
228 | ||
229 | /* per event run time data */ | |
230 | struct evsel_runtime { | |
231 | u64 *last_time; /* time this event was last seen per cpu */ | |
232 | u32 ncpu; /* highest cpu slot allocated */ | |
233 | }; | |
234 | ||
3bc2fa9c NK |
235 | /* per cpu idle time data */ |
236 | struct idle_thread_runtime { | |
237 | struct thread_runtime tr; | |
238 | struct thread *last_thread; | |
239 | struct rb_root sorted_root; | |
240 | struct callchain_root callchain; | |
241 | struct callchain_cursor cursor; | |
242 | }; | |
243 | ||
49394a2a DA |
244 | /* track idle times per cpu */ |
245 | static struct thread **idle_threads; | |
246 | static int idle_max_cpu; | |
247 | static char idle_comm[] = "<idle>"; | |
248 | ||
b1ffe8f3 | 249 | static u64 get_nsecs(void) |
ec156764 IM |
250 | { |
251 | struct timespec ts; | |
252 | ||
253 | clock_gettime(CLOCK_MONOTONIC, &ts); | |
254 | ||
4fc76e49 | 255 | return ts.tv_sec * NSEC_PER_SEC + ts.tv_nsec; |
ec156764 IM |
256 | } |
257 | ||
0e9b07e5 | 258 | static void burn_nsecs(struct perf_sched *sched, u64 nsecs) |
ec156764 | 259 | { |
b1ffe8f3 | 260 | u64 T0 = get_nsecs(), T1; |
ec156764 IM |
261 | |
262 | do { | |
263 | T1 = get_nsecs(); | |
0e9b07e5 | 264 | } while (T1 + sched->run_measurement_overhead < T0 + nsecs); |
ec156764 IM |
265 | } |
266 | ||
b1ffe8f3 | 267 | static void sleep_nsecs(u64 nsecs) |
ec156764 IM |
268 | { |
269 | struct timespec ts; | |
270 | ||
271 | ts.tv_nsec = nsecs % 999999999; | |
272 | ts.tv_sec = nsecs / 999999999; | |
273 | ||
274 | nanosleep(&ts, NULL); | |
275 | } | |
276 | ||
0e9b07e5 | 277 | static void calibrate_run_measurement_overhead(struct perf_sched *sched) |
ec156764 | 278 | { |
4fc76e49 | 279 | u64 T0, T1, delta, min_delta = NSEC_PER_SEC; |
ec156764 IM |
280 | int i; |
281 | ||
282 | for (i = 0; i < 10; i++) { | |
283 | T0 = get_nsecs(); | |
0e9b07e5 | 284 | burn_nsecs(sched, 0); |
ec156764 IM |
285 | T1 = get_nsecs(); |
286 | delta = T1-T0; | |
287 | min_delta = min(min_delta, delta); | |
288 | } | |
0e9b07e5 | 289 | sched->run_measurement_overhead = min_delta; |
ec156764 | 290 | |
9486aa38 | 291 | printf("run measurement overhead: %" PRIu64 " nsecs\n", min_delta); |
ec156764 IM |
292 | } |
293 | ||
0e9b07e5 | 294 | static void calibrate_sleep_measurement_overhead(struct perf_sched *sched) |
ec156764 | 295 | { |
4fc76e49 | 296 | u64 T0, T1, delta, min_delta = NSEC_PER_SEC; |
ec156764 IM |
297 | int i; |
298 | ||
299 | for (i = 0; i < 10; i++) { | |
300 | T0 = get_nsecs(); | |
301 | sleep_nsecs(10000); | |
302 | T1 = get_nsecs(); | |
303 | delta = T1-T0; | |
304 | min_delta = min(min_delta, delta); | |
305 | } | |
306 | min_delta -= 10000; | |
0e9b07e5 | 307 | sched->sleep_measurement_overhead = min_delta; |
ec156764 | 308 | |
9486aa38 | 309 | printf("sleep measurement overhead: %" PRIu64 " nsecs\n", min_delta); |
ec156764 IM |
310 | } |
311 | ||
39aeb52f | 312 | static struct sched_atom * |
b1ffe8f3 | 313 | get_new_event(struct task_desc *task, u64 timestamp) |
ec156764 | 314 | { |
36479484 | 315 | struct sched_atom *event = zalloc(sizeof(*event)); |
ec156764 IM |
316 | unsigned long idx = task->nr_events; |
317 | size_t size; | |
318 | ||
319 | event->timestamp = timestamp; | |
320 | event->nr = idx; | |
321 | ||
322 | task->nr_events++; | |
39aeb52f | 323 | size = sizeof(struct sched_atom *) * task->nr_events; |
324 | task->atoms = realloc(task->atoms, size); | |
325 | BUG_ON(!task->atoms); | |
ec156764 | 326 | |
39aeb52f | 327 | task->atoms[idx] = event; |
ec156764 IM |
328 | |
329 | return event; | |
330 | } | |
331 | ||
39aeb52f | 332 | static struct sched_atom *last_event(struct task_desc *task) |
ec156764 IM |
333 | { |
334 | if (!task->nr_events) | |
335 | return NULL; | |
336 | ||
39aeb52f | 337 | return task->atoms[task->nr_events - 1]; |
ec156764 IM |
338 | } |
339 | ||
0e9b07e5 ACM |
340 | static void add_sched_event_run(struct perf_sched *sched, struct task_desc *task, |
341 | u64 timestamp, u64 duration) | |
ec156764 | 342 | { |
39aeb52f | 343 | struct sched_atom *event, *curr_event = last_event(task); |
ec156764 IM |
344 | |
345 | /* | |
fbf94829 IM |
346 | * optimize an existing RUN event by merging this one |
347 | * to it: | |
348 | */ | |
ec156764 | 349 | if (curr_event && curr_event->type == SCHED_EVENT_RUN) { |
0e9b07e5 | 350 | sched->nr_run_events_optimized++; |
ec156764 IM |
351 | curr_event->duration += duration; |
352 | return; | |
353 | } | |
354 | ||
355 | event = get_new_event(task, timestamp); | |
356 | ||
357 | event->type = SCHED_EVENT_RUN; | |
358 | event->duration = duration; | |
359 | ||
0e9b07e5 | 360 | sched->nr_run_events++; |
ec156764 IM |
361 | } |
362 | ||
0e9b07e5 ACM |
363 | static void add_sched_event_wakeup(struct perf_sched *sched, struct task_desc *task, |
364 | u64 timestamp, struct task_desc *wakee) | |
ec156764 | 365 | { |
39aeb52f | 366 | struct sched_atom *event, *wakee_event; |
ec156764 IM |
367 | |
368 | event = get_new_event(task, timestamp); | |
369 | event->type = SCHED_EVENT_WAKEUP; | |
370 | event->wakee = wakee; | |
371 | ||
372 | wakee_event = last_event(wakee); | |
373 | if (!wakee_event || wakee_event->type != SCHED_EVENT_SLEEP) { | |
0e9b07e5 | 374 | sched->targetless_wakeups++; |
ec156764 IM |
375 | return; |
376 | } | |
377 | if (wakee_event->wait_sem) { | |
0e9b07e5 | 378 | sched->multitarget_wakeups++; |
ec156764 IM |
379 | return; |
380 | } | |
381 | ||
36479484 | 382 | wakee_event->wait_sem = zalloc(sizeof(*wakee_event->wait_sem)); |
ec156764 IM |
383 | sem_init(wakee_event->wait_sem, 0, 0); |
384 | wakee_event->specific_wait = 1; | |
385 | event->wait_sem = wakee_event->wait_sem; | |
386 | ||
0e9b07e5 | 387 | sched->nr_wakeup_events++; |
ec156764 IM |
388 | } |
389 | ||
0e9b07e5 ACM |
390 | static void add_sched_event_sleep(struct perf_sched *sched, struct task_desc *task, |
391 | u64 timestamp, u64 task_state __maybe_unused) | |
ec156764 | 392 | { |
39aeb52f | 393 | struct sched_atom *event = get_new_event(task, timestamp); |
ec156764 IM |
394 | |
395 | event->type = SCHED_EVENT_SLEEP; | |
396 | ||
0e9b07e5 | 397 | sched->nr_sleep_events++; |
ec156764 IM |
398 | } |
399 | ||
0e9b07e5 ACM |
400 | static struct task_desc *register_pid(struct perf_sched *sched, |
401 | unsigned long pid, const char *comm) | |
ec156764 IM |
402 | { |
403 | struct task_desc *task; | |
cb06ac25 | 404 | static int pid_max; |
ec156764 | 405 | |
cb06ac25 YS |
406 | if (sched->pid_to_task == NULL) { |
407 | if (sysctl__read_int("kernel/pid_max", &pid_max) < 0) | |
408 | pid_max = MAX_PID; | |
409 | BUG_ON((sched->pid_to_task = calloc(pid_max, sizeof(struct task_desc *))) == NULL); | |
410 | } | |
3a423a5c YS |
411 | if (pid >= (unsigned long)pid_max) { |
412 | BUG_ON((sched->pid_to_task = realloc(sched->pid_to_task, (pid + 1) * | |
413 | sizeof(struct task_desc *))) == NULL); | |
414 | while (pid >= (unsigned long)pid_max) | |
415 | sched->pid_to_task[pid_max++] = NULL; | |
416 | } | |
ec156764 | 417 | |
0e9b07e5 | 418 | task = sched->pid_to_task[pid]; |
ec156764 IM |
419 | |
420 | if (task) | |
421 | return task; | |
422 | ||
36479484 | 423 | task = zalloc(sizeof(*task)); |
ec156764 | 424 | task->pid = pid; |
0e9b07e5 | 425 | task->nr = sched->nr_tasks; |
ec156764 IM |
426 | strcpy(task->comm, comm); |
427 | /* | |
428 | * every task starts in sleeping state - this gets ignored | |
429 | * if there's no wakeup pointing to this sleep state: | |
430 | */ | |
0e9b07e5 | 431 | add_sched_event_sleep(sched, task, 0, 0); |
ec156764 | 432 | |
0e9b07e5 ACM |
433 | sched->pid_to_task[pid] = task; |
434 | sched->nr_tasks++; | |
0755bc4d | 435 | sched->tasks = realloc(sched->tasks, sched->nr_tasks * sizeof(struct task_desc *)); |
0e9b07e5 ACM |
436 | BUG_ON(!sched->tasks); |
437 | sched->tasks[task->nr] = task; | |
ec156764 | 438 | |
ad236fd2 | 439 | if (verbose) |
0e9b07e5 | 440 | printf("registered task #%ld, PID %ld (%s)\n", sched->nr_tasks, pid, comm); |
ec156764 IM |
441 | |
442 | return task; | |
443 | } | |
444 | ||
445 | ||
0e9b07e5 | 446 | static void print_task_traces(struct perf_sched *sched) |
ec156764 IM |
447 | { |
448 | struct task_desc *task; | |
449 | unsigned long i; | |
450 | ||
0e9b07e5 ACM |
451 | for (i = 0; i < sched->nr_tasks; i++) { |
452 | task = sched->tasks[i]; | |
ad236fd2 | 453 | printf("task %6ld (%20s:%10ld), nr_events: %ld\n", |
ec156764 IM |
454 | task->nr, task->comm, task->pid, task->nr_events); |
455 | } | |
456 | } | |
457 | ||
0e9b07e5 | 458 | static void add_cross_task_wakeups(struct perf_sched *sched) |
ec156764 IM |
459 | { |
460 | struct task_desc *task1, *task2; | |
461 | unsigned long i, j; | |
462 | ||
0e9b07e5 ACM |
463 | for (i = 0; i < sched->nr_tasks; i++) { |
464 | task1 = sched->tasks[i]; | |
ec156764 | 465 | j = i + 1; |
0e9b07e5 | 466 | if (j == sched->nr_tasks) |
ec156764 | 467 | j = 0; |
0e9b07e5 ACM |
468 | task2 = sched->tasks[j]; |
469 | add_sched_event_wakeup(sched, task1, 0, task2); | |
ec156764 IM |
470 | } |
471 | } | |
472 | ||
0e9b07e5 ACM |
473 | static void perf_sched__process_event(struct perf_sched *sched, |
474 | struct sched_atom *atom) | |
ec156764 IM |
475 | { |
476 | int ret = 0; | |
ec156764 | 477 | |
39aeb52f | 478 | switch (atom->type) { |
ec156764 | 479 | case SCHED_EVENT_RUN: |
0e9b07e5 | 480 | burn_nsecs(sched, atom->duration); |
ec156764 IM |
481 | break; |
482 | case SCHED_EVENT_SLEEP: | |
39aeb52f | 483 | if (atom->wait_sem) |
484 | ret = sem_wait(atom->wait_sem); | |
ec156764 IM |
485 | BUG_ON(ret); |
486 | break; | |
487 | case SCHED_EVENT_WAKEUP: | |
39aeb52f | 488 | if (atom->wait_sem) |
489 | ret = sem_post(atom->wait_sem); | |
ec156764 IM |
490 | BUG_ON(ret); |
491 | break; | |
55ffb7a6 MG |
492 | case SCHED_EVENT_MIGRATION: |
493 | break; | |
ec156764 IM |
494 | default: |
495 | BUG_ON(1); | |
496 | } | |
497 | } | |
498 | ||
b1ffe8f3 | 499 | static u64 get_cpu_usage_nsec_parent(void) |
ec156764 IM |
500 | { |
501 | struct rusage ru; | |
b1ffe8f3 | 502 | u64 sum; |
ec156764 IM |
503 | int err; |
504 | ||
505 | err = getrusage(RUSAGE_SELF, &ru); | |
506 | BUG_ON(err); | |
507 | ||
4fc76e49 ACM |
508 | sum = ru.ru_utime.tv_sec * NSEC_PER_SEC + ru.ru_utime.tv_usec * NSEC_PER_USEC; |
509 | sum += ru.ru_stime.tv_sec * NSEC_PER_SEC + ru.ru_stime.tv_usec * NSEC_PER_USEC; | |
ec156764 IM |
510 | |
511 | return sum; | |
512 | } | |
513 | ||
939cda52 | 514 | static int self_open_counters(struct perf_sched *sched, unsigned long cur_task) |
ec156764 | 515 | { |
c0c9e721 | 516 | struct perf_event_attr attr; |
939cda52 | 517 | char sbuf[STRERR_BUFSIZE], info[STRERR_BUFSIZE]; |
c0c9e721 | 518 | int fd; |
939cda52 YS |
519 | struct rlimit limit; |
520 | bool need_privilege = false; | |
ec156764 | 521 | |
c0c9e721 | 522 | memset(&attr, 0, sizeof(attr)); |
ec156764 | 523 | |
c0c9e721 XG |
524 | attr.type = PERF_TYPE_SOFTWARE; |
525 | attr.config = PERF_COUNT_SW_TASK_CLOCK; | |
ec156764 | 526 | |
939cda52 | 527 | force_again: |
57480d2c YD |
528 | fd = sys_perf_event_open(&attr, 0, -1, -1, |
529 | perf_event_open_cloexec_flag()); | |
c0c9e721 | 530 | |
1aff59be | 531 | if (fd < 0) { |
939cda52 YS |
532 | if (errno == EMFILE) { |
533 | if (sched->force) { | |
534 | BUG_ON(getrlimit(RLIMIT_NOFILE, &limit) == -1); | |
535 | limit.rlim_cur += sched->nr_tasks - cur_task; | |
536 | if (limit.rlim_cur > limit.rlim_max) { | |
537 | limit.rlim_max = limit.rlim_cur; | |
538 | need_privilege = true; | |
539 | } | |
540 | if (setrlimit(RLIMIT_NOFILE, &limit) == -1) { | |
541 | if (need_privilege && errno == EPERM) | |
542 | strcpy(info, "Need privilege\n"); | |
543 | } else | |
544 | goto force_again; | |
545 | } else | |
546 | strcpy(info, "Have a try with -f option\n"); | |
547 | } | |
60b7d14a | 548 | pr_err("Error: sys_perf_event_open() syscall returned " |
939cda52 | 549 | "with %d (%s)\n%s", fd, |
c8b5f2c9 | 550 | str_error_r(errno, sbuf, sizeof(sbuf)), info); |
1aff59be YS |
551 | exit(EXIT_FAILURE); |
552 | } | |
c0c9e721 XG |
553 | return fd; |
554 | } | |
555 | ||
556 | static u64 get_cpu_usage_nsec_self(int fd) | |
557 | { | |
558 | u64 runtime; | |
559 | int ret; | |
560 | ||
561 | ret = read(fd, &runtime, sizeof(runtime)); | |
562 | BUG_ON(ret != sizeof(runtime)); | |
563 | ||
564 | return runtime; | |
ec156764 IM |
565 | } |
566 | ||
0e9b07e5 ACM |
567 | struct sched_thread_parms { |
568 | struct task_desc *task; | |
569 | struct perf_sched *sched; | |
08097abc | 570 | int fd; |
0e9b07e5 ACM |
571 | }; |
572 | ||
ec156764 IM |
573 | static void *thread_func(void *ctx) |
574 | { | |
0e9b07e5 ACM |
575 | struct sched_thread_parms *parms = ctx; |
576 | struct task_desc *this_task = parms->task; | |
577 | struct perf_sched *sched = parms->sched; | |
b1ffe8f3 | 578 | u64 cpu_usage_0, cpu_usage_1; |
ec156764 IM |
579 | unsigned long i, ret; |
580 | char comm2[22]; | |
08097abc | 581 | int fd = parms->fd; |
ec156764 | 582 | |
74cf249d | 583 | zfree(&parms); |
0e9b07e5 | 584 | |
ec156764 IM |
585 | sprintf(comm2, ":%s", this_task->comm); |
586 | prctl(PR_SET_NAME, comm2); | |
a116e05d ACM |
587 | if (fd < 0) |
588 | return NULL; | |
ec156764 IM |
589 | again: |
590 | ret = sem_post(&this_task->ready_for_work); | |
591 | BUG_ON(ret); | |
0e9b07e5 | 592 | ret = pthread_mutex_lock(&sched->start_work_mutex); |
ec156764 | 593 | BUG_ON(ret); |
0e9b07e5 | 594 | ret = pthread_mutex_unlock(&sched->start_work_mutex); |
ec156764 | 595 | BUG_ON(ret); |
ec156764 | 596 | |
c0c9e721 | 597 | cpu_usage_0 = get_cpu_usage_nsec_self(fd); |
ec156764 IM |
598 | |
599 | for (i = 0; i < this_task->nr_events; i++) { | |
600 | this_task->curr_event = i; | |
0e9b07e5 | 601 | perf_sched__process_event(sched, this_task->atoms[i]); |
ec156764 IM |
602 | } |
603 | ||
c0c9e721 | 604 | cpu_usage_1 = get_cpu_usage_nsec_self(fd); |
ec156764 | 605 | this_task->cpu_usage = cpu_usage_1 - cpu_usage_0; |
ec156764 IM |
606 | ret = sem_post(&this_task->work_done_sem); |
607 | BUG_ON(ret); | |
ec156764 | 608 | |
0e9b07e5 | 609 | ret = pthread_mutex_lock(&sched->work_done_wait_mutex); |
ec156764 | 610 | BUG_ON(ret); |
0e9b07e5 | 611 | ret = pthread_mutex_unlock(&sched->work_done_wait_mutex); |
ec156764 | 612 | BUG_ON(ret); |
ec156764 IM |
613 | |
614 | goto again; | |
615 | } | |
616 | ||
0e9b07e5 | 617 | static void create_tasks(struct perf_sched *sched) |
ec156764 IM |
618 | { |
619 | struct task_desc *task; | |
620 | pthread_attr_t attr; | |
621 | unsigned long i; | |
622 | int err; | |
623 | ||
624 | err = pthread_attr_init(&attr); | |
625 | BUG_ON(err); | |
12f7e036 JP |
626 | err = pthread_attr_setstacksize(&attr, |
627 | (size_t) max(16 * 1024, PTHREAD_STACK_MIN)); | |
ec156764 | 628 | BUG_ON(err); |
0e9b07e5 | 629 | err = pthread_mutex_lock(&sched->start_work_mutex); |
ec156764 | 630 | BUG_ON(err); |
0e9b07e5 | 631 | err = pthread_mutex_lock(&sched->work_done_wait_mutex); |
ec156764 | 632 | BUG_ON(err); |
0e9b07e5 ACM |
633 | for (i = 0; i < sched->nr_tasks; i++) { |
634 | struct sched_thread_parms *parms = malloc(sizeof(*parms)); | |
635 | BUG_ON(parms == NULL); | |
636 | parms->task = task = sched->tasks[i]; | |
637 | parms->sched = sched; | |
939cda52 | 638 | parms->fd = self_open_counters(sched, i); |
ec156764 IM |
639 | sem_init(&task->sleep_sem, 0, 0); |
640 | sem_init(&task->ready_for_work, 0, 0); | |
641 | sem_init(&task->work_done_sem, 0, 0); | |
642 | task->curr_event = 0; | |
0e9b07e5 | 643 | err = pthread_create(&task->thread, &attr, thread_func, parms); |
ec156764 IM |
644 | BUG_ON(err); |
645 | } | |
646 | } | |
647 | ||
0e9b07e5 | 648 | static void wait_for_tasks(struct perf_sched *sched) |
ec156764 | 649 | { |
b1ffe8f3 | 650 | u64 cpu_usage_0, cpu_usage_1; |
ec156764 IM |
651 | struct task_desc *task; |
652 | unsigned long i, ret; | |
653 | ||
0e9b07e5 ACM |
654 | sched->start_time = get_nsecs(); |
655 | sched->cpu_usage = 0; | |
656 | pthread_mutex_unlock(&sched->work_done_wait_mutex); | |
ec156764 | 657 | |
0e9b07e5 ACM |
658 | for (i = 0; i < sched->nr_tasks; i++) { |
659 | task = sched->tasks[i]; | |
ec156764 IM |
660 | ret = sem_wait(&task->ready_for_work); |
661 | BUG_ON(ret); | |
662 | sem_init(&task->ready_for_work, 0, 0); | |
663 | } | |
0e9b07e5 | 664 | ret = pthread_mutex_lock(&sched->work_done_wait_mutex); |
ec156764 IM |
665 | BUG_ON(ret); |
666 | ||
667 | cpu_usage_0 = get_cpu_usage_nsec_parent(); | |
668 | ||
0e9b07e5 | 669 | pthread_mutex_unlock(&sched->start_work_mutex); |
ec156764 | 670 | |
0e9b07e5 ACM |
671 | for (i = 0; i < sched->nr_tasks; i++) { |
672 | task = sched->tasks[i]; | |
ec156764 IM |
673 | ret = sem_wait(&task->work_done_sem); |
674 | BUG_ON(ret); | |
675 | sem_init(&task->work_done_sem, 0, 0); | |
0e9b07e5 | 676 | sched->cpu_usage += task->cpu_usage; |
ec156764 IM |
677 | task->cpu_usage = 0; |
678 | } | |
679 | ||
680 | cpu_usage_1 = get_cpu_usage_nsec_parent(); | |
0e9b07e5 ACM |
681 | if (!sched->runavg_cpu_usage) |
682 | sched->runavg_cpu_usage = sched->cpu_usage; | |
ff5f3bbd | 683 | sched->runavg_cpu_usage = (sched->runavg_cpu_usage * (sched->replay_repeat - 1) + sched->cpu_usage) / sched->replay_repeat; |
ec156764 | 684 | |
0e9b07e5 ACM |
685 | sched->parent_cpu_usage = cpu_usage_1 - cpu_usage_0; |
686 | if (!sched->runavg_parent_cpu_usage) | |
687 | sched->runavg_parent_cpu_usage = sched->parent_cpu_usage; | |
ff5f3bbd YS |
688 | sched->runavg_parent_cpu_usage = (sched->runavg_parent_cpu_usage * (sched->replay_repeat - 1) + |
689 | sched->parent_cpu_usage)/sched->replay_repeat; | |
ec156764 | 690 | |
0e9b07e5 | 691 | ret = pthread_mutex_lock(&sched->start_work_mutex); |
ec156764 IM |
692 | BUG_ON(ret); |
693 | ||
0e9b07e5 ACM |
694 | for (i = 0; i < sched->nr_tasks; i++) { |
695 | task = sched->tasks[i]; | |
ec156764 IM |
696 | sem_init(&task->sleep_sem, 0, 0); |
697 | task->curr_event = 0; | |
698 | } | |
699 | } | |
700 | ||
0e9b07e5 | 701 | static void run_one_test(struct perf_sched *sched) |
ec156764 | 702 | { |
fb7d0b3c | 703 | u64 T0, T1, delta, avg_delta, fluct; |
ec156764 IM |
704 | |
705 | T0 = get_nsecs(); | |
0e9b07e5 | 706 | wait_for_tasks(sched); |
ec156764 IM |
707 | T1 = get_nsecs(); |
708 | ||
709 | delta = T1 - T0; | |
0e9b07e5 ACM |
710 | sched->sum_runtime += delta; |
711 | sched->nr_runs++; | |
ec156764 | 712 | |
0e9b07e5 | 713 | avg_delta = sched->sum_runtime / sched->nr_runs; |
ec156764 IM |
714 | if (delta < avg_delta) |
715 | fluct = avg_delta - delta; | |
716 | else | |
717 | fluct = delta - avg_delta; | |
0e9b07e5 ACM |
718 | sched->sum_fluct += fluct; |
719 | if (!sched->run_avg) | |
720 | sched->run_avg = delta; | |
ff5f3bbd | 721 | sched->run_avg = (sched->run_avg * (sched->replay_repeat - 1) + delta) / sched->replay_repeat; |
ec156764 | 722 | |
4fc76e49 | 723 | printf("#%-3ld: %0.3f, ", sched->nr_runs, (double)delta / NSEC_PER_MSEC); |
ec156764 | 724 | |
4fc76e49 | 725 | printf("ravg: %0.2f, ", (double)sched->run_avg / NSEC_PER_MSEC); |
ec156764 | 726 | |
ad236fd2 | 727 | printf("cpu: %0.2f / %0.2f", |
4fc76e49 | 728 | (double)sched->cpu_usage / NSEC_PER_MSEC, (double)sched->runavg_cpu_usage / NSEC_PER_MSEC); |
ec156764 IM |
729 | |
730 | #if 0 | |
731 | /* | |
fbf94829 | 732 | * rusage statistics done by the parent, these are less |
0e9b07e5 | 733 | * accurate than the sched->sum_exec_runtime based statistics: |
fbf94829 | 734 | */ |
ad236fd2 | 735 | printf(" [%0.2f / %0.2f]", |
4fc76e49 ACM |
736 | (double)sched->parent_cpu_usage / NSEC_PER_MSEC, |
737 | (double)sched->runavg_parent_cpu_usage / NSEC_PER_MSEC); | |
ec156764 IM |
738 | #endif |
739 | ||
ad236fd2 | 740 | printf("\n"); |
ec156764 | 741 | |
0e9b07e5 ACM |
742 | if (sched->nr_sleep_corrections) |
743 | printf(" (%ld sleep corrections)\n", sched->nr_sleep_corrections); | |
744 | sched->nr_sleep_corrections = 0; | |
ec156764 IM |
745 | } |
746 | ||
0e9b07e5 | 747 | static void test_calibrations(struct perf_sched *sched) |
ec156764 | 748 | { |
b1ffe8f3 | 749 | u64 T0, T1; |
ec156764 IM |
750 | |
751 | T0 = get_nsecs(); | |
4fc76e49 | 752 | burn_nsecs(sched, NSEC_PER_MSEC); |
ec156764 IM |
753 | T1 = get_nsecs(); |
754 | ||
9486aa38 | 755 | printf("the run test took %" PRIu64 " nsecs\n", T1 - T0); |
ec156764 IM |
756 | |
757 | T0 = get_nsecs(); | |
4fc76e49 | 758 | sleep_nsecs(NSEC_PER_MSEC); |
ec156764 IM |
759 | T1 = get_nsecs(); |
760 | ||
9486aa38 | 761 | printf("the sleep test took %" PRIu64 " nsecs\n", T1 - T0); |
ec156764 IM |
762 | } |
763 | ||
a116e05d | 764 | static int |
0e9b07e5 | 765 | replay_wakeup_event(struct perf_sched *sched, |
9ec3f4e4 ACM |
766 | struct perf_evsel *evsel, struct perf_sample *sample, |
767 | struct machine *machine __maybe_unused) | |
419ab0d6 | 768 | { |
9ec3f4e4 ACM |
769 | const char *comm = perf_evsel__strval(evsel, sample, "comm"); |
770 | const u32 pid = perf_evsel__intval(evsel, sample, "pid"); | |
419ab0d6 | 771 | struct task_desc *waker, *wakee; |
fbf94829 | 772 | |
ad236fd2 | 773 | if (verbose) { |
2b7fcbc5 | 774 | printf("sched_wakeup event %p\n", evsel); |
fbf94829 | 775 | |
9ec3f4e4 | 776 | printf(" ... pid %d woke up %s/%d\n", sample->tid, comm, pid); |
ad236fd2 | 777 | } |
fbf94829 | 778 | |
2b7fcbc5 | 779 | waker = register_pid(sched, sample->tid, "<unknown>"); |
9ec3f4e4 | 780 | wakee = register_pid(sched, pid, comm); |
fbf94829 | 781 | |
0e9b07e5 | 782 | add_sched_event_wakeup(sched, waker, sample->time, wakee); |
a116e05d | 783 | return 0; |
ec156764 IM |
784 | } |
785 | ||
9ec3f4e4 ACM |
786 | static int replay_switch_event(struct perf_sched *sched, |
787 | struct perf_evsel *evsel, | |
788 | struct perf_sample *sample, | |
789 | struct machine *machine __maybe_unused) | |
ec156764 | 790 | { |
9ec3f4e4 ACM |
791 | const char *prev_comm = perf_evsel__strval(evsel, sample, "prev_comm"), |
792 | *next_comm = perf_evsel__strval(evsel, sample, "next_comm"); | |
793 | const u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"), | |
794 | next_pid = perf_evsel__intval(evsel, sample, "next_pid"); | |
795 | const u64 prev_state = perf_evsel__intval(evsel, sample, "prev_state"); | |
1d037ca1 | 796 | struct task_desc *prev, __maybe_unused *next; |
7f7f8d0b ACM |
797 | u64 timestamp0, timestamp = sample->time; |
798 | int cpu = sample->cpu; | |
fbf94829 IM |
799 | s64 delta; |
800 | ||
ad236fd2 | 801 | if (verbose) |
2b7fcbc5 | 802 | printf("sched_switch event %p\n", evsel); |
ad236fd2 | 803 | |
fbf94829 | 804 | if (cpu >= MAX_CPUS || cpu < 0) |
a116e05d | 805 | return 0; |
fbf94829 | 806 | |
0e9b07e5 | 807 | timestamp0 = sched->cpu_last_switched[cpu]; |
fbf94829 IM |
808 | if (timestamp0) |
809 | delta = timestamp - timestamp0; | |
810 | else | |
811 | delta = 0; | |
812 | ||
a116e05d | 813 | if (delta < 0) { |
60b7d14a | 814 | pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta); |
a116e05d ACM |
815 | return -1; |
816 | } | |
fbf94829 | 817 | |
9ec3f4e4 ACM |
818 | pr_debug(" ... switch from %s/%d to %s/%d [ran %" PRIu64 " nsecs]\n", |
819 | prev_comm, prev_pid, next_comm, next_pid, delta); | |
fbf94829 | 820 | |
9ec3f4e4 ACM |
821 | prev = register_pid(sched, prev_pid, prev_comm); |
822 | next = register_pid(sched, next_pid, next_comm); | |
fbf94829 | 823 | |
0e9b07e5 | 824 | sched->cpu_last_switched[cpu] = timestamp; |
fbf94829 | 825 | |
0e9b07e5 | 826 | add_sched_event_run(sched, prev, timestamp, delta); |
9ec3f4e4 | 827 | add_sched_event_sleep(sched, prev, timestamp, prev_state); |
a116e05d ACM |
828 | |
829 | return 0; | |
fbf94829 IM |
830 | } |
831 | ||
cb627505 DA |
832 | static int replay_fork_event(struct perf_sched *sched, |
833 | union perf_event *event, | |
834 | struct machine *machine) | |
419ab0d6 | 835 | { |
cb627505 DA |
836 | struct thread *child, *parent; |
837 | ||
314add6b AH |
838 | child = machine__findnew_thread(machine, event->fork.pid, |
839 | event->fork.tid); | |
840 | parent = machine__findnew_thread(machine, event->fork.ppid, | |
841 | event->fork.ptid); | |
cb627505 DA |
842 | |
843 | if (child == NULL || parent == NULL) { | |
844 | pr_debug("thread does not exist on fork event: child %p, parent %p\n", | |
845 | child, parent); | |
b91fc39f | 846 | goto out_put; |
cb627505 | 847 | } |
9ec3f4e4 | 848 | |
419ab0d6 | 849 | if (verbose) { |
cb627505 | 850 | printf("fork event\n"); |
b9c5143a FW |
851 | printf("... parent: %s/%d\n", thread__comm_str(parent), parent->tid); |
852 | printf("... child: %s/%d\n", thread__comm_str(child), child->tid); | |
419ab0d6 | 853 | } |
9ec3f4e4 | 854 | |
b9c5143a FW |
855 | register_pid(sched, parent->tid, thread__comm_str(parent)); |
856 | register_pid(sched, child->tid, thread__comm_str(child)); | |
b91fc39f ACM |
857 | out_put: |
858 | thread__put(child); | |
859 | thread__put(parent); | |
a116e05d | 860 | return 0; |
419ab0d6 | 861 | } |
fbf94829 | 862 | |
b1ffe8f3 IM |
863 | struct sort_dimension { |
864 | const char *name; | |
b5fae128 | 865 | sort_fn_t cmp; |
b1ffe8f3 IM |
866 | struct list_head list; |
867 | }; | |
868 | ||
daa1d7a5 | 869 | static int |
39aeb52f | 870 | thread_lat_cmp(struct list_head *list, struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 FW |
871 | { |
872 | struct sort_dimension *sort; | |
873 | int ret = 0; | |
874 | ||
b5fae128 IM |
875 | BUG_ON(list_empty(list)); |
876 | ||
daa1d7a5 FW |
877 | list_for_each_entry(sort, list, list) { |
878 | ret = sort->cmp(l, r); | |
879 | if (ret) | |
880 | return ret; | |
881 | } | |
882 | ||
883 | return ret; | |
884 | } | |
885 | ||
39aeb52f | 886 | static struct work_atoms * |
b5fae128 IM |
887 | thread_atoms_search(struct rb_root *root, struct thread *thread, |
888 | struct list_head *sort_list) | |
889 | { | |
890 | struct rb_node *node = root->rb_node; | |
39aeb52f | 891 | struct work_atoms key = { .thread = thread }; |
b5fae128 IM |
892 | |
893 | while (node) { | |
39aeb52f | 894 | struct work_atoms *atoms; |
b5fae128 IM |
895 | int cmp; |
896 | ||
39aeb52f | 897 | atoms = container_of(node, struct work_atoms, node); |
b5fae128 IM |
898 | |
899 | cmp = thread_lat_cmp(sort_list, &key, atoms); | |
900 | if (cmp > 0) | |
901 | node = node->rb_left; | |
902 | else if (cmp < 0) | |
903 | node = node->rb_right; | |
904 | else { | |
905 | BUG_ON(thread != atoms->thread); | |
906 | return atoms; | |
907 | } | |
908 | } | |
909 | return NULL; | |
910 | } | |
911 | ||
cdce9d73 | 912 | static void |
39aeb52f | 913 | __thread_latency_insert(struct rb_root *root, struct work_atoms *data, |
daa1d7a5 | 914 | struct list_head *sort_list) |
cdce9d73 FW |
915 | { |
916 | struct rb_node **new = &(root->rb_node), *parent = NULL; | |
917 | ||
918 | while (*new) { | |
39aeb52f | 919 | struct work_atoms *this; |
daa1d7a5 | 920 | int cmp; |
cdce9d73 | 921 | |
39aeb52f | 922 | this = container_of(*new, struct work_atoms, node); |
cdce9d73 | 923 | parent = *new; |
daa1d7a5 FW |
924 | |
925 | cmp = thread_lat_cmp(sort_list, data, this); | |
926 | ||
927 | if (cmp > 0) | |
cdce9d73 | 928 | new = &((*new)->rb_left); |
cdce9d73 | 929 | else |
daa1d7a5 | 930 | new = &((*new)->rb_right); |
cdce9d73 FW |
931 | } |
932 | ||
933 | rb_link_node(&data->node, parent, new); | |
934 | rb_insert_color(&data->node, root); | |
935 | } | |
936 | ||
0e9b07e5 | 937 | static int thread_atoms_insert(struct perf_sched *sched, struct thread *thread) |
cdce9d73 | 938 | { |
36479484 | 939 | struct work_atoms *atoms = zalloc(sizeof(*atoms)); |
a116e05d ACM |
940 | if (!atoms) { |
941 | pr_err("No memory at %s\n", __func__); | |
942 | return -1; | |
943 | } | |
cdce9d73 | 944 | |
f3b623b8 | 945 | atoms->thread = thread__get(thread); |
39aeb52f | 946 | INIT_LIST_HEAD(&atoms->work_list); |
0e9b07e5 | 947 | __thread_latency_insert(&sched->atom_root, atoms, &sched->cmp_pid); |
a116e05d | 948 | return 0; |
cdce9d73 FW |
949 | } |
950 | ||
9ec3f4e4 | 951 | static char sched_out_state(u64 prev_state) |
cdce9d73 FW |
952 | { |
953 | const char *str = TASK_STATE_TO_CHAR_STR; | |
954 | ||
9ec3f4e4 | 955 | return str[prev_state]; |
cdce9d73 FW |
956 | } |
957 | ||
a116e05d | 958 | static int |
39aeb52f | 959 | add_sched_out_event(struct work_atoms *atoms, |
960 | char run_state, | |
961 | u64 timestamp) | |
cdce9d73 | 962 | { |
36479484 | 963 | struct work_atom *atom = zalloc(sizeof(*atom)); |
a116e05d ACM |
964 | if (!atom) { |
965 | pr_err("Non memory at %s", __func__); | |
966 | return -1; | |
967 | } | |
cdce9d73 | 968 | |
aa1ab9d2 FW |
969 | atom->sched_out_time = timestamp; |
970 | ||
39aeb52f | 971 | if (run_state == 'R') { |
b1ffe8f3 | 972 | atom->state = THREAD_WAIT_CPU; |
aa1ab9d2 | 973 | atom->wake_up_time = atom->sched_out_time; |
c6ced611 FW |
974 | } |
975 | ||
39aeb52f | 976 | list_add_tail(&atom->list, &atoms->work_list); |
a116e05d | 977 | return 0; |
cdce9d73 FW |
978 | } |
979 | ||
980 | static void | |
1d037ca1 IT |
981 | add_runtime_event(struct work_atoms *atoms, u64 delta, |
982 | u64 timestamp __maybe_unused) | |
39aeb52f | 983 | { |
984 | struct work_atom *atom; | |
985 | ||
986 | BUG_ON(list_empty(&atoms->work_list)); | |
987 | ||
988 | atom = list_entry(atoms->work_list.prev, struct work_atom, list); | |
989 | ||
990 | atom->runtime += delta; | |
991 | atoms->total_runtime += delta; | |
992 | } | |
993 | ||
994 | static void | |
995 | add_sched_in_event(struct work_atoms *atoms, u64 timestamp) | |
cdce9d73 | 996 | { |
b1ffe8f3 | 997 | struct work_atom *atom; |
66685678 | 998 | u64 delta; |
cdce9d73 | 999 | |
39aeb52f | 1000 | if (list_empty(&atoms->work_list)) |
cdce9d73 FW |
1001 | return; |
1002 | ||
39aeb52f | 1003 | atom = list_entry(atoms->work_list.prev, struct work_atom, list); |
cdce9d73 | 1004 | |
b1ffe8f3 | 1005 | if (atom->state != THREAD_WAIT_CPU) |
cdce9d73 FW |
1006 | return; |
1007 | ||
b1ffe8f3 IM |
1008 | if (timestamp < atom->wake_up_time) { |
1009 | atom->state = THREAD_IGNORE; | |
cdce9d73 FW |
1010 | return; |
1011 | } | |
1012 | ||
b1ffe8f3 IM |
1013 | atom->state = THREAD_SCHED_IN; |
1014 | atom->sched_in_time = timestamp; | |
66685678 | 1015 | |
b1ffe8f3 | 1016 | delta = atom->sched_in_time - atom->wake_up_time; |
66685678 | 1017 | atoms->total_lat += delta; |
3786310a | 1018 | if (delta > atoms->max_lat) { |
66685678 | 1019 | atoms->max_lat = delta; |
3786310a FW |
1020 | atoms->max_lat_at = timestamp; |
1021 | } | |
66685678 | 1022 | atoms->nb_atoms++; |
cdce9d73 FW |
1023 | } |
1024 | ||
9ec3f4e4 ACM |
1025 | static int latency_switch_event(struct perf_sched *sched, |
1026 | struct perf_evsel *evsel, | |
1027 | struct perf_sample *sample, | |
1028 | struct machine *machine) | |
cdce9d73 | 1029 | { |
9ec3f4e4 ACM |
1030 | const u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"), |
1031 | next_pid = perf_evsel__intval(evsel, sample, "next_pid"); | |
1032 | const u64 prev_state = perf_evsel__intval(evsel, sample, "prev_state"); | |
39aeb52f | 1033 | struct work_atoms *out_events, *in_events; |
cdce9d73 | 1034 | struct thread *sched_out, *sched_in; |
7f7f8d0b | 1035 | u64 timestamp0, timestamp = sample->time; |
b91fc39f | 1036 | int cpu = sample->cpu, err = -1; |
ea92ed5a IM |
1037 | s64 delta; |
1038 | ||
39aeb52f | 1039 | BUG_ON(cpu >= MAX_CPUS || cpu < 0); |
ea92ed5a | 1040 | |
0e9b07e5 ACM |
1041 | timestamp0 = sched->cpu_last_switched[cpu]; |
1042 | sched->cpu_last_switched[cpu] = timestamp; | |
ea92ed5a IM |
1043 | if (timestamp0) |
1044 | delta = timestamp - timestamp0; | |
1045 | else | |
1046 | delta = 0; | |
1047 | ||
a116e05d ACM |
1048 | if (delta < 0) { |
1049 | pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta); | |
1050 | return -1; | |
1051 | } | |
cdce9d73 | 1052 | |
1fcb8768 AH |
1053 | sched_out = machine__findnew_thread(machine, -1, prev_pid); |
1054 | sched_in = machine__findnew_thread(machine, -1, next_pid); | |
b91fc39f ACM |
1055 | if (sched_out == NULL || sched_in == NULL) |
1056 | goto out_put; | |
cdce9d73 | 1057 | |
0e9b07e5 | 1058 | out_events = thread_atoms_search(&sched->atom_root, sched_out, &sched->cmp_pid); |
39aeb52f | 1059 | if (!out_events) { |
0e9b07e5 | 1060 | if (thread_atoms_insert(sched, sched_out)) |
b91fc39f | 1061 | goto out_put; |
0e9b07e5 | 1062 | out_events = thread_atoms_search(&sched->atom_root, sched_out, &sched->cmp_pid); |
a116e05d ACM |
1063 | if (!out_events) { |
1064 | pr_err("out-event: Internal tree error"); | |
b91fc39f | 1065 | goto out_put; |
a116e05d | 1066 | } |
39aeb52f | 1067 | } |
9ec3f4e4 | 1068 | if (add_sched_out_event(out_events, sched_out_state(prev_state), timestamp)) |
a116e05d | 1069 | return -1; |
39aeb52f | 1070 | |
0e9b07e5 | 1071 | in_events = thread_atoms_search(&sched->atom_root, sched_in, &sched->cmp_pid); |
39aeb52f | 1072 | if (!in_events) { |
0e9b07e5 | 1073 | if (thread_atoms_insert(sched, sched_in)) |
b91fc39f | 1074 | goto out_put; |
0e9b07e5 | 1075 | in_events = thread_atoms_search(&sched->atom_root, sched_in, &sched->cmp_pid); |
a116e05d ACM |
1076 | if (!in_events) { |
1077 | pr_err("in-event: Internal tree error"); | |
b91fc39f | 1078 | goto out_put; |
a116e05d | 1079 | } |
39aeb52f | 1080 | /* |
1081 | * Take came in we have not heard about yet, | |
1082 | * add in an initial atom in runnable state: | |
1083 | */ | |
a116e05d | 1084 | if (add_sched_out_event(in_events, 'R', timestamp)) |
b91fc39f | 1085 | goto out_put; |
cdce9d73 | 1086 | } |
39aeb52f | 1087 | add_sched_in_event(in_events, timestamp); |
b91fc39f ACM |
1088 | err = 0; |
1089 | out_put: | |
1090 | thread__put(sched_out); | |
1091 | thread__put(sched_in); | |
1092 | return err; | |
39aeb52f | 1093 | } |
cdce9d73 | 1094 | |
9ec3f4e4 ACM |
1095 | static int latency_runtime_event(struct perf_sched *sched, |
1096 | struct perf_evsel *evsel, | |
1097 | struct perf_sample *sample, | |
1098 | struct machine *machine) | |
39aeb52f | 1099 | { |
9ec3f4e4 ACM |
1100 | const u32 pid = perf_evsel__intval(evsel, sample, "pid"); |
1101 | const u64 runtime = perf_evsel__intval(evsel, sample, "runtime"); | |
1fcb8768 | 1102 | struct thread *thread = machine__findnew_thread(machine, -1, pid); |
0e9b07e5 | 1103 | struct work_atoms *atoms = thread_atoms_search(&sched->atom_root, thread, &sched->cmp_pid); |
7f7f8d0b | 1104 | u64 timestamp = sample->time; |
b91fc39f ACM |
1105 | int cpu = sample->cpu, err = -1; |
1106 | ||
1107 | if (thread == NULL) | |
1108 | return -1; | |
39aeb52f | 1109 | |
1110 | BUG_ON(cpu >= MAX_CPUS || cpu < 0); | |
39aeb52f | 1111 | if (!atoms) { |
0e9b07e5 | 1112 | if (thread_atoms_insert(sched, thread)) |
b91fc39f | 1113 | goto out_put; |
0e9b07e5 | 1114 | atoms = thread_atoms_search(&sched->atom_root, thread, &sched->cmp_pid); |
a116e05d | 1115 | if (!atoms) { |
60b7d14a | 1116 | pr_err("in-event: Internal tree error"); |
b91fc39f | 1117 | goto out_put; |
a116e05d ACM |
1118 | } |
1119 | if (add_sched_out_event(atoms, 'R', timestamp)) | |
b91fc39f | 1120 | goto out_put; |
cdce9d73 FW |
1121 | } |
1122 | ||
9ec3f4e4 | 1123 | add_runtime_event(atoms, runtime, timestamp); |
b91fc39f ACM |
1124 | err = 0; |
1125 | out_put: | |
1126 | thread__put(thread); | |
1127 | return err; | |
cdce9d73 FW |
1128 | } |
1129 | ||
9ec3f4e4 ACM |
1130 | static int latency_wakeup_event(struct perf_sched *sched, |
1131 | struct perf_evsel *evsel, | |
1132 | struct perf_sample *sample, | |
1133 | struct machine *machine) | |
cdce9d73 | 1134 | { |
0680ee7d | 1135 | const u32 pid = perf_evsel__intval(evsel, sample, "pid"); |
39aeb52f | 1136 | struct work_atoms *atoms; |
b1ffe8f3 | 1137 | struct work_atom *atom; |
cdce9d73 | 1138 | struct thread *wakee; |
7f7f8d0b | 1139 | u64 timestamp = sample->time; |
b91fc39f | 1140 | int err = -1; |
cdce9d73 | 1141 | |
1fcb8768 | 1142 | wakee = machine__findnew_thread(machine, -1, pid); |
b91fc39f ACM |
1143 | if (wakee == NULL) |
1144 | return -1; | |
0e9b07e5 | 1145 | atoms = thread_atoms_search(&sched->atom_root, wakee, &sched->cmp_pid); |
17562205 | 1146 | if (!atoms) { |
0e9b07e5 | 1147 | if (thread_atoms_insert(sched, wakee)) |
b91fc39f | 1148 | goto out_put; |
0e9b07e5 | 1149 | atoms = thread_atoms_search(&sched->atom_root, wakee, &sched->cmp_pid); |
a116e05d | 1150 | if (!atoms) { |
60b7d14a | 1151 | pr_err("wakeup-event: Internal tree error"); |
b91fc39f | 1152 | goto out_put; |
a116e05d ACM |
1153 | } |
1154 | if (add_sched_out_event(atoms, 'S', timestamp)) | |
b91fc39f | 1155 | goto out_put; |
cdce9d73 FW |
1156 | } |
1157 | ||
39aeb52f | 1158 | BUG_ON(list_empty(&atoms->work_list)); |
cdce9d73 | 1159 | |
39aeb52f | 1160 | atom = list_entry(atoms->work_list.prev, struct work_atom, list); |
cdce9d73 | 1161 | |
55ffb7a6 | 1162 | /* |
67d6259d DY |
1163 | * As we do not guarantee the wakeup event happens when |
1164 | * task is out of run queue, also may happen when task is | |
1165 | * on run queue and wakeup only change ->state to TASK_RUNNING, | |
1166 | * then we should not set the ->wake_up_time when wake up a | |
1167 | * task which is on run queue. | |
1168 | * | |
55ffb7a6 MG |
1169 | * You WILL be missing events if you've recorded only |
1170 | * one CPU, or are only looking at only one, so don't | |
67d6259d | 1171 | * skip in this case. |
55ffb7a6 | 1172 | */ |
0e9b07e5 | 1173 | if (sched->profile_cpu == -1 && atom->state != THREAD_SLEEPING) |
b91fc39f | 1174 | goto out_ok; |
cdce9d73 | 1175 | |
0e9b07e5 | 1176 | sched->nr_timestamps++; |
ea57c4f5 | 1177 | if (atom->sched_out_time > timestamp) { |
0e9b07e5 | 1178 | sched->nr_unordered_timestamps++; |
b91fc39f | 1179 | goto out_ok; |
ea57c4f5 | 1180 | } |
aa1ab9d2 | 1181 | |
b1ffe8f3 IM |
1182 | atom->state = THREAD_WAIT_CPU; |
1183 | atom->wake_up_time = timestamp; | |
b91fc39f ACM |
1184 | out_ok: |
1185 | err = 0; | |
1186 | out_put: | |
1187 | thread__put(wakee); | |
1188 | return err; | |
cdce9d73 FW |
1189 | } |
1190 | ||
9ec3f4e4 ACM |
1191 | static int latency_migrate_task_event(struct perf_sched *sched, |
1192 | struct perf_evsel *evsel, | |
1193 | struct perf_sample *sample, | |
1194 | struct machine *machine) | |
55ffb7a6 | 1195 | { |
9ec3f4e4 | 1196 | const u32 pid = perf_evsel__intval(evsel, sample, "pid"); |
7f7f8d0b | 1197 | u64 timestamp = sample->time; |
55ffb7a6 MG |
1198 | struct work_atoms *atoms; |
1199 | struct work_atom *atom; | |
1200 | struct thread *migrant; | |
b91fc39f | 1201 | int err = -1; |
55ffb7a6 MG |
1202 | |
1203 | /* | |
1204 | * Only need to worry about migration when profiling one CPU. | |
1205 | */ | |
0e9b07e5 | 1206 | if (sched->profile_cpu == -1) |
a116e05d | 1207 | return 0; |
55ffb7a6 | 1208 | |
1fcb8768 | 1209 | migrant = machine__findnew_thread(machine, -1, pid); |
b91fc39f ACM |
1210 | if (migrant == NULL) |
1211 | return -1; | |
0e9b07e5 | 1212 | atoms = thread_atoms_search(&sched->atom_root, migrant, &sched->cmp_pid); |
55ffb7a6 | 1213 | if (!atoms) { |
0e9b07e5 | 1214 | if (thread_atoms_insert(sched, migrant)) |
b91fc39f | 1215 | goto out_put; |
b9c5143a | 1216 | register_pid(sched, migrant->tid, thread__comm_str(migrant)); |
0e9b07e5 | 1217 | atoms = thread_atoms_search(&sched->atom_root, migrant, &sched->cmp_pid); |
a116e05d | 1218 | if (!atoms) { |
60b7d14a | 1219 | pr_err("migration-event: Internal tree error"); |
b91fc39f | 1220 | goto out_put; |
a116e05d ACM |
1221 | } |
1222 | if (add_sched_out_event(atoms, 'R', timestamp)) | |
b91fc39f | 1223 | goto out_put; |
55ffb7a6 MG |
1224 | } |
1225 | ||
1226 | BUG_ON(list_empty(&atoms->work_list)); | |
1227 | ||
1228 | atom = list_entry(atoms->work_list.prev, struct work_atom, list); | |
1229 | atom->sched_in_time = atom->sched_out_time = atom->wake_up_time = timestamp; | |
1230 | ||
0e9b07e5 | 1231 | sched->nr_timestamps++; |
55ffb7a6 MG |
1232 | |
1233 | if (atom->sched_out_time > timestamp) | |
0e9b07e5 | 1234 | sched->nr_unordered_timestamps++; |
b91fc39f ACM |
1235 | err = 0; |
1236 | out_put: | |
1237 | thread__put(migrant); | |
1238 | return err; | |
55ffb7a6 MG |
1239 | } |
1240 | ||
0e9b07e5 | 1241 | static void output_lat_thread(struct perf_sched *sched, struct work_atoms *work_list) |
cdce9d73 | 1242 | { |
cdce9d73 FW |
1243 | int i; |
1244 | int ret; | |
66685678 | 1245 | u64 avg; |
99620a5d | 1246 | char max_lat_at[32]; |
cdce9d73 | 1247 | |
39aeb52f | 1248 | if (!work_list->nb_atoms) |
cdce9d73 | 1249 | return; |
ea57c4f5 IM |
1250 | /* |
1251 | * Ignore idle threads: | |
1252 | */ | |
b9c5143a | 1253 | if (!strcmp(thread__comm_str(work_list->thread), "swapper")) |
ea57c4f5 | 1254 | return; |
cdce9d73 | 1255 | |
0e9b07e5 ACM |
1256 | sched->all_runtime += work_list->total_runtime; |
1257 | sched->all_count += work_list->nb_atoms; | |
66685678 | 1258 | |
2f80dd44 JB |
1259 | if (work_list->num_merged > 1) |
1260 | ret = printf(" %s:(%d) ", thread__comm_str(work_list->thread), work_list->num_merged); | |
1261 | else | |
1262 | ret = printf(" %s:%d ", thread__comm_str(work_list->thread), work_list->thread->tid); | |
cdce9d73 | 1263 | |
08f69e6c | 1264 | for (i = 0; i < 24 - ret; i++) |
cdce9d73 FW |
1265 | printf(" "); |
1266 | ||
39aeb52f | 1267 | avg = work_list->total_lat / work_list->nb_atoms; |
99620a5d | 1268 | timestamp__scnprintf_usec(work_list->max_lat_at, max_lat_at, sizeof(max_lat_at)); |
cdce9d73 | 1269 | |
99620a5d | 1270 | printf("|%11.3f ms |%9" PRIu64 " | avg:%9.3f ms | max:%9.3f ms | max at: %13s s\n", |
4fc76e49 ACM |
1271 | (double)work_list->total_runtime / NSEC_PER_MSEC, |
1272 | work_list->nb_atoms, (double)avg / NSEC_PER_MSEC, | |
1273 | (double)work_list->max_lat / NSEC_PER_MSEC, | |
99620a5d | 1274 | max_lat_at); |
cdce9d73 FW |
1275 | } |
1276 | ||
39aeb52f | 1277 | static int pid_cmp(struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 | 1278 | { |
0014de17 JO |
1279 | if (l->thread == r->thread) |
1280 | return 0; | |
38051234 | 1281 | if (l->thread->tid < r->thread->tid) |
daa1d7a5 | 1282 | return -1; |
38051234 | 1283 | if (l->thread->tid > r->thread->tid) |
daa1d7a5 | 1284 | return 1; |
0014de17 | 1285 | return (int)(l->thread - r->thread); |
daa1d7a5 FW |
1286 | } |
1287 | ||
39aeb52f | 1288 | static int avg_cmp(struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 FW |
1289 | { |
1290 | u64 avgl, avgr; | |
1291 | ||
1292 | if (!l->nb_atoms) | |
1293 | return -1; | |
1294 | ||
1295 | if (!r->nb_atoms) | |
1296 | return 1; | |
1297 | ||
1298 | avgl = l->total_lat / l->nb_atoms; | |
1299 | avgr = r->total_lat / r->nb_atoms; | |
1300 | ||
1301 | if (avgl < avgr) | |
1302 | return -1; | |
1303 | if (avgl > avgr) | |
1304 | return 1; | |
1305 | ||
1306 | return 0; | |
1307 | } | |
1308 | ||
39aeb52f | 1309 | static int max_cmp(struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 FW |
1310 | { |
1311 | if (l->max_lat < r->max_lat) | |
1312 | return -1; | |
1313 | if (l->max_lat > r->max_lat) | |
1314 | return 1; | |
1315 | ||
1316 | return 0; | |
1317 | } | |
1318 | ||
39aeb52f | 1319 | static int switch_cmp(struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 FW |
1320 | { |
1321 | if (l->nb_atoms < r->nb_atoms) | |
1322 | return -1; | |
1323 | if (l->nb_atoms > r->nb_atoms) | |
1324 | return 1; | |
1325 | ||
1326 | return 0; | |
1327 | } | |
1328 | ||
39aeb52f | 1329 | static int runtime_cmp(struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 FW |
1330 | { |
1331 | if (l->total_runtime < r->total_runtime) | |
1332 | return -1; | |
1333 | if (l->total_runtime > r->total_runtime) | |
1334 | return 1; | |
1335 | ||
1336 | return 0; | |
1337 | } | |
1338 | ||
cbef79a8 | 1339 | static int sort_dimension__add(const char *tok, struct list_head *list) |
daa1d7a5 | 1340 | { |
0e9b07e5 ACM |
1341 | size_t i; |
1342 | static struct sort_dimension avg_sort_dimension = { | |
1343 | .name = "avg", | |
1344 | .cmp = avg_cmp, | |
1345 | }; | |
1346 | static struct sort_dimension max_sort_dimension = { | |
1347 | .name = "max", | |
1348 | .cmp = max_cmp, | |
1349 | }; | |
1350 | static struct sort_dimension pid_sort_dimension = { | |
1351 | .name = "pid", | |
1352 | .cmp = pid_cmp, | |
1353 | }; | |
1354 | static struct sort_dimension runtime_sort_dimension = { | |
1355 | .name = "runtime", | |
1356 | .cmp = runtime_cmp, | |
1357 | }; | |
1358 | static struct sort_dimension switch_sort_dimension = { | |
1359 | .name = "switch", | |
1360 | .cmp = switch_cmp, | |
1361 | }; | |
1362 | struct sort_dimension *available_sorts[] = { | |
1363 | &pid_sort_dimension, | |
1364 | &avg_sort_dimension, | |
1365 | &max_sort_dimension, | |
1366 | &switch_sort_dimension, | |
1367 | &runtime_sort_dimension, | |
1368 | }; | |
daa1d7a5 | 1369 | |
0e9b07e5 | 1370 | for (i = 0; i < ARRAY_SIZE(available_sorts); i++) { |
daa1d7a5 FW |
1371 | if (!strcmp(available_sorts[i]->name, tok)) { |
1372 | list_add_tail(&available_sorts[i]->list, list); | |
1373 | ||
1374 | return 0; | |
1375 | } | |
1376 | } | |
1377 | ||
1378 | return -1; | |
1379 | } | |
1380 | ||
0e9b07e5 | 1381 | static void perf_sched__sort_lat(struct perf_sched *sched) |
daa1d7a5 FW |
1382 | { |
1383 | struct rb_node *node; | |
2f80dd44 JB |
1384 | struct rb_root *root = &sched->atom_root; |
1385 | again: | |
daa1d7a5 | 1386 | for (;;) { |
39aeb52f | 1387 | struct work_atoms *data; |
2f80dd44 | 1388 | node = rb_first(root); |
daa1d7a5 FW |
1389 | if (!node) |
1390 | break; | |
1391 | ||
2f80dd44 | 1392 | rb_erase(node, root); |
39aeb52f | 1393 | data = rb_entry(node, struct work_atoms, node); |
0e9b07e5 | 1394 | __thread_latency_insert(&sched->sorted_atom_root, data, &sched->sort_list); |
daa1d7a5 | 1395 | } |
2f80dd44 JB |
1396 | if (root == &sched->atom_root) { |
1397 | root = &sched->merged_atom_root; | |
1398 | goto again; | |
1399 | } | |
daa1d7a5 FW |
1400 | } |
1401 | ||
0e9b07e5 | 1402 | static int process_sched_wakeup_event(struct perf_tool *tool, |
2b7fcbc5 | 1403 | struct perf_evsel *evsel, |
1d037ca1 | 1404 | struct perf_sample *sample, |
4218e673 | 1405 | struct machine *machine) |
419ab0d6 | 1406 | { |
0e9b07e5 | 1407 | struct perf_sched *sched = container_of(tool, struct perf_sched, tool); |
419ab0d6 | 1408 | |
9ec3f4e4 ACM |
1409 | if (sched->tp_handler->wakeup_event) |
1410 | return sched->tp_handler->wakeup_event(sched, evsel, sample, machine); | |
a116e05d | 1411 | |
2b7fcbc5 | 1412 | return 0; |
419ab0d6 FW |
1413 | } |
1414 | ||
a151a37a JO |
1415 | union map_priv { |
1416 | void *ptr; | |
1417 | bool color; | |
1418 | }; | |
1419 | ||
1420 | static bool thread__has_color(struct thread *thread) | |
1421 | { | |
1422 | union map_priv priv = { | |
1423 | .ptr = thread__priv(thread), | |
1424 | }; | |
1425 | ||
1426 | return priv.color; | |
1427 | } | |
1428 | ||
1429 | static struct thread* | |
1430 | map__findnew_thread(struct perf_sched *sched, struct machine *machine, pid_t pid, pid_t tid) | |
1431 | { | |
1432 | struct thread *thread = machine__findnew_thread(machine, pid, tid); | |
1433 | union map_priv priv = { | |
1434 | .color = false, | |
1435 | }; | |
1436 | ||
1437 | if (!sched->map.color_pids || !thread || thread__priv(thread)) | |
1438 | return thread; | |
1439 | ||
1440 | if (thread_map__has(sched->map.color_pids, tid)) | |
1441 | priv.color = true; | |
1442 | ||
1443 | thread__set_priv(thread, priv.ptr); | |
1444 | return thread; | |
1445 | } | |
1446 | ||
9ec3f4e4 ACM |
1447 | static int map_switch_event(struct perf_sched *sched, struct perf_evsel *evsel, |
1448 | struct perf_sample *sample, struct machine *machine) | |
0ec04e16 | 1449 | { |
9d372ca5 DY |
1450 | const u32 next_pid = perf_evsel__intval(evsel, sample, "next_pid"); |
1451 | struct thread *sched_in; | |
0ec04e16 | 1452 | int new_shortname; |
7f7f8d0b | 1453 | u64 timestamp0, timestamp = sample->time; |
0ec04e16 | 1454 | s64 delta; |
99623c62 JO |
1455 | int i, this_cpu = sample->cpu; |
1456 | int cpus_nr; | |
1457 | bool new_cpu = false; | |
8cd91195 | 1458 | const char *color = PERF_COLOR_NORMAL; |
99620a5d | 1459 | char stimestamp[32]; |
0ec04e16 IM |
1460 | |
1461 | BUG_ON(this_cpu >= MAX_CPUS || this_cpu < 0); | |
1462 | ||
0e9b07e5 ACM |
1463 | if (this_cpu > sched->max_cpu) |
1464 | sched->max_cpu = this_cpu; | |
0ec04e16 | 1465 | |
99623c62 JO |
1466 | if (sched->map.comp) { |
1467 | cpus_nr = bitmap_weight(sched->map.comp_cpus_mask, MAX_CPUS); | |
1468 | if (!test_and_set_bit(this_cpu, sched->map.comp_cpus_mask)) { | |
1469 | sched->map.comp_cpus[cpus_nr++] = this_cpu; | |
1470 | new_cpu = true; | |
1471 | } | |
1472 | } else | |
1473 | cpus_nr = sched->max_cpu; | |
1474 | ||
0e9b07e5 ACM |
1475 | timestamp0 = sched->cpu_last_switched[this_cpu]; |
1476 | sched->cpu_last_switched[this_cpu] = timestamp; | |
0ec04e16 IM |
1477 | if (timestamp0) |
1478 | delta = timestamp - timestamp0; | |
1479 | else | |
1480 | delta = 0; | |
1481 | ||
a116e05d | 1482 | if (delta < 0) { |
60b7d14a | 1483 | pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta); |
a116e05d ACM |
1484 | return -1; |
1485 | } | |
0ec04e16 | 1486 | |
a151a37a | 1487 | sched_in = map__findnew_thread(sched, machine, -1, next_pid); |
b91fc39f ACM |
1488 | if (sched_in == NULL) |
1489 | return -1; | |
0ec04e16 | 1490 | |
b91fc39f | 1491 | sched->curr_thread[this_cpu] = thread__get(sched_in); |
0ec04e16 IM |
1492 | |
1493 | printf(" "); | |
1494 | ||
1495 | new_shortname = 0; | |
1496 | if (!sched_in->shortname[0]) { | |
6bcab4e1 D |
1497 | if (!strcmp(thread__comm_str(sched_in), "swapper")) { |
1498 | /* | |
1499 | * Don't allocate a letter-number for swapper:0 | |
1500 | * as a shortname. Instead, we use '.' for it. | |
1501 | */ | |
1502 | sched_in->shortname[0] = '.'; | |
1503 | sched_in->shortname[1] = ' '; | |
0ec04e16 | 1504 | } else { |
6bcab4e1 D |
1505 | sched_in->shortname[0] = sched->next_shortname1; |
1506 | sched_in->shortname[1] = sched->next_shortname2; | |
1507 | ||
1508 | if (sched->next_shortname1 < 'Z') { | |
1509 | sched->next_shortname1++; | |
0ec04e16 | 1510 | } else { |
6bcab4e1 D |
1511 | sched->next_shortname1 = 'A'; |
1512 | if (sched->next_shortname2 < '9') | |
1513 | sched->next_shortname2++; | |
1514 | else | |
1515 | sched->next_shortname2 = '0'; | |
0ec04e16 IM |
1516 | } |
1517 | } | |
1518 | new_shortname = 1; | |
1519 | } | |
1520 | ||
99623c62 JO |
1521 | for (i = 0; i < cpus_nr; i++) { |
1522 | int cpu = sched->map.comp ? sched->map.comp_cpus[i] : i; | |
a151a37a JO |
1523 | struct thread *curr_thread = sched->curr_thread[cpu]; |
1524 | const char *pid_color = color; | |
cf294f24 | 1525 | const char *cpu_color = color; |
a151a37a JO |
1526 | |
1527 | if (curr_thread && thread__has_color(curr_thread)) | |
1528 | pid_color = COLOR_PIDS; | |
99623c62 | 1529 | |
73643bb6 JO |
1530 | if (sched->map.cpus && !cpu_map__has(sched->map.cpus, cpu)) |
1531 | continue; | |
1532 | ||
cf294f24 JO |
1533 | if (sched->map.color_cpus && cpu_map__has(sched->map.color_cpus, cpu)) |
1534 | cpu_color = COLOR_CPUS; | |
1535 | ||
0ec04e16 | 1536 | if (cpu != this_cpu) |
1208bb27 | 1537 | color_fprintf(stdout, color, " "); |
0ec04e16 | 1538 | else |
cf294f24 | 1539 | color_fprintf(stdout, cpu_color, "*"); |
0ec04e16 | 1540 | |
6bcab4e1 | 1541 | if (sched->curr_thread[cpu]) |
a151a37a | 1542 | color_fprintf(stdout, pid_color, "%2s ", sched->curr_thread[cpu]->shortname); |
6bcab4e1 | 1543 | else |
8cd91195 | 1544 | color_fprintf(stdout, color, " "); |
0ec04e16 IM |
1545 | } |
1546 | ||
73643bb6 JO |
1547 | if (sched->map.cpus && !cpu_map__has(sched->map.cpus, this_cpu)) |
1548 | goto out; | |
1549 | ||
99620a5d NK |
1550 | timestamp__scnprintf_usec(timestamp, stimestamp, sizeof(stimestamp)); |
1551 | color_fprintf(stdout, color, " %12s secs ", stimestamp); | |
e107f129 | 1552 | if (new_shortname || (verbose && sched_in->tid)) { |
a151a37a JO |
1553 | const char *pid_color = color; |
1554 | ||
1555 | if (thread__has_color(sched_in)) | |
1556 | pid_color = COLOR_PIDS; | |
1557 | ||
1558 | color_fprintf(stdout, pid_color, "%s => %s:%d", | |
b9c5143a | 1559 | sched_in->shortname, thread__comm_str(sched_in), sched_in->tid); |
0ec04e16 | 1560 | } |
a116e05d | 1561 | |
99623c62 | 1562 | if (sched->map.comp && new_cpu) |
8cd91195 | 1563 | color_fprintf(stdout, color, " (CPU %d)", this_cpu); |
99623c62 | 1564 | |
73643bb6 | 1565 | out: |
8cd91195 | 1566 | color_fprintf(stdout, color, "\n"); |
99623c62 | 1567 | |
b91fc39f ACM |
1568 | thread__put(sched_in); |
1569 | ||
a116e05d | 1570 | return 0; |
0ec04e16 IM |
1571 | } |
1572 | ||
0e9b07e5 | 1573 | static int process_sched_switch_event(struct perf_tool *tool, |
2b7fcbc5 | 1574 | struct perf_evsel *evsel, |
1d037ca1 | 1575 | struct perf_sample *sample, |
4218e673 | 1576 | struct machine *machine) |
419ab0d6 | 1577 | { |
0e9b07e5 | 1578 | struct perf_sched *sched = container_of(tool, struct perf_sched, tool); |
a116e05d | 1579 | int this_cpu = sample->cpu, err = 0; |
2b7fcbc5 ACM |
1580 | u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"), |
1581 | next_pid = perf_evsel__intval(evsel, sample, "next_pid"); | |
419ab0d6 | 1582 | |
0e9b07e5 | 1583 | if (sched->curr_pid[this_cpu] != (u32)-1) { |
c8a37751 IM |
1584 | /* |
1585 | * Are we trying to switch away a PID that is | |
1586 | * not current? | |
1587 | */ | |
2b7fcbc5 | 1588 | if (sched->curr_pid[this_cpu] != prev_pid) |
0e9b07e5 | 1589 | sched->nr_context_switch_bugs++; |
c8a37751 | 1590 | } |
c8a37751 | 1591 | |
9ec3f4e4 ACM |
1592 | if (sched->tp_handler->switch_event) |
1593 | err = sched->tp_handler->switch_event(sched, evsel, sample, machine); | |
2b7fcbc5 ACM |
1594 | |
1595 | sched->curr_pid[this_cpu] = next_pid; | |
a116e05d | 1596 | return err; |
419ab0d6 FW |
1597 | } |
1598 | ||
0e9b07e5 | 1599 | static int process_sched_runtime_event(struct perf_tool *tool, |
2b7fcbc5 | 1600 | struct perf_evsel *evsel, |
1d037ca1 | 1601 | struct perf_sample *sample, |
4218e673 | 1602 | struct machine *machine) |
39aeb52f | 1603 | { |
0e9b07e5 | 1604 | struct perf_sched *sched = container_of(tool, struct perf_sched, tool); |
39aeb52f | 1605 | |
9ec3f4e4 ACM |
1606 | if (sched->tp_handler->runtime_event) |
1607 | return sched->tp_handler->runtime_event(sched, evsel, sample, machine); | |
a116e05d | 1608 | |
2b7fcbc5 | 1609 | return 0; |
39aeb52f | 1610 | } |
1611 | ||
cb627505 DA |
1612 | static int perf_sched__process_fork_event(struct perf_tool *tool, |
1613 | union perf_event *event, | |
1614 | struct perf_sample *sample, | |
1615 | struct machine *machine) | |
fbf94829 | 1616 | { |
0e9b07e5 | 1617 | struct perf_sched *sched = container_of(tool, struct perf_sched, tool); |
46538818 | 1618 | |
cb627505 DA |
1619 | /* run the fork event through the perf machineruy */ |
1620 | perf_event__process_fork(tool, event, sample, machine); | |
1621 | ||
1622 | /* and then run additional processing needed for this command */ | |
9ec3f4e4 | 1623 | if (sched->tp_handler->fork_event) |
cb627505 | 1624 | return sched->tp_handler->fork_event(sched, event, machine); |
a116e05d | 1625 | |
2b7fcbc5 | 1626 | return 0; |
fbf94829 IM |
1627 | } |
1628 | ||
0e9b07e5 | 1629 | static int process_sched_migrate_task_event(struct perf_tool *tool, |
2b7fcbc5 | 1630 | struct perf_evsel *evsel, |
1d037ca1 | 1631 | struct perf_sample *sample, |
4218e673 | 1632 | struct machine *machine) |
55ffb7a6 | 1633 | { |
0e9b07e5 | 1634 | struct perf_sched *sched = container_of(tool, struct perf_sched, tool); |
55ffb7a6 | 1635 | |
9ec3f4e4 ACM |
1636 | if (sched->tp_handler->migrate_task_event) |
1637 | return sched->tp_handler->migrate_task_event(sched, evsel, sample, machine); | |
a116e05d | 1638 | |
2b7fcbc5 | 1639 | return 0; |
55ffb7a6 MG |
1640 | } |
1641 | ||
a116e05d | 1642 | typedef int (*tracepoint_handler)(struct perf_tool *tool, |
2b7fcbc5 | 1643 | struct perf_evsel *evsel, |
a116e05d | 1644 | struct perf_sample *sample, |
4218e673 | 1645 | struct machine *machine); |
ec156764 | 1646 | |
1d037ca1 IT |
1647 | static int perf_sched__process_tracepoint_sample(struct perf_tool *tool __maybe_unused, |
1648 | union perf_event *event __maybe_unused, | |
ee29be62 ACM |
1649 | struct perf_sample *sample, |
1650 | struct perf_evsel *evsel, | |
1651 | struct machine *machine) | |
0a02ad93 | 1652 | { |
a116e05d | 1653 | int err = 0; |
0a02ad93 | 1654 | |
744a9719 ACM |
1655 | if (evsel->handler != NULL) { |
1656 | tracepoint_handler f = evsel->handler; | |
2b7fcbc5 | 1657 | err = f(tool, evsel, sample, machine); |
ee29be62 | 1658 | } |
0a02ad93 | 1659 | |
a116e05d | 1660 | return err; |
0a02ad93 IM |
1661 | } |
1662 | ||
ae536acf | 1663 | static int perf_sched__read_events(struct perf_sched *sched) |
0a02ad93 | 1664 | { |
ee29be62 ACM |
1665 | const struct perf_evsel_str_handler handlers[] = { |
1666 | { "sched:sched_switch", process_sched_switch_event, }, | |
1667 | { "sched:sched_stat_runtime", process_sched_runtime_event, }, | |
1668 | { "sched:sched_wakeup", process_sched_wakeup_event, }, | |
1669 | { "sched:sched_wakeup_new", process_sched_wakeup_event, }, | |
ee29be62 ACM |
1670 | { "sched:sched_migrate_task", process_sched_migrate_task_event, }, |
1671 | }; | |
da378962 | 1672 | struct perf_session *session; |
f5fc1412 JO |
1673 | struct perf_data_file file = { |
1674 | .path = input_name, | |
1675 | .mode = PERF_DATA_MODE_READ, | |
f0dd330f | 1676 | .force = sched->force, |
f5fc1412 | 1677 | }; |
ae536acf | 1678 | int rc = -1; |
da378962 | 1679 | |
f5fc1412 | 1680 | session = perf_session__new(&file, false, &sched->tool); |
a116e05d ACM |
1681 | if (session == NULL) { |
1682 | pr_debug("No Memory for session\n"); | |
1683 | return -1; | |
1684 | } | |
94c744b6 | 1685 | |
0a7e6d1b | 1686 | symbol__init(&session->header.env); |
04934106 | 1687 | |
a116e05d ACM |
1688 | if (perf_session__set_tracepoints_handlers(session, handlers)) |
1689 | goto out_delete; | |
ee29be62 | 1690 | |
cee75ac7 | 1691 | if (perf_session__has_traces(session, "record -R")) { |
b7b61cbe | 1692 | int err = perf_session__process_events(session); |
a116e05d ACM |
1693 | if (err) { |
1694 | pr_err("Failed to process events, error %d", err); | |
1695 | goto out_delete; | |
1696 | } | |
4c09bafa | 1697 | |
75be989a ACM |
1698 | sched->nr_events = session->evlist->stats.nr_events[0]; |
1699 | sched->nr_lost_events = session->evlist->stats.total_lost; | |
1700 | sched->nr_lost_chunks = session->evlist->stats.nr_events[PERF_RECORD_LOST]; | |
cee75ac7 | 1701 | } |
d549c769 | 1702 | |
ae536acf | 1703 | rc = 0; |
a116e05d ACM |
1704 | out_delete: |
1705 | perf_session__delete(session); | |
ae536acf | 1706 | return rc; |
0a02ad93 IM |
1707 | } |
1708 | ||
49394a2a DA |
1709 | /* |
1710 | * scheduling times are printed as msec.usec | |
1711 | */ | |
1712 | static inline void print_sched_time(unsigned long long nsecs, int width) | |
1713 | { | |
1714 | unsigned long msecs; | |
1715 | unsigned long usecs; | |
1716 | ||
1717 | msecs = nsecs / NSEC_PER_MSEC; | |
1718 | nsecs -= msecs * NSEC_PER_MSEC; | |
1719 | usecs = nsecs / NSEC_PER_USEC; | |
1720 | printf("%*lu.%03lu ", width, msecs, usecs); | |
1721 | } | |
1722 | ||
1723 | /* | |
1724 | * returns runtime data for event, allocating memory for it the | |
1725 | * first time it is used. | |
1726 | */ | |
1727 | static struct evsel_runtime *perf_evsel__get_runtime(struct perf_evsel *evsel) | |
1728 | { | |
1729 | struct evsel_runtime *r = evsel->priv; | |
1730 | ||
1731 | if (r == NULL) { | |
1732 | r = zalloc(sizeof(struct evsel_runtime)); | |
1733 | evsel->priv = r; | |
1734 | } | |
1735 | ||
1736 | return r; | |
1737 | } | |
1738 | ||
1739 | /* | |
1740 | * save last time event was seen per cpu | |
1741 | */ | |
1742 | static void perf_evsel__save_time(struct perf_evsel *evsel, | |
1743 | u64 timestamp, u32 cpu) | |
1744 | { | |
1745 | struct evsel_runtime *r = perf_evsel__get_runtime(evsel); | |
1746 | ||
1747 | if (r == NULL) | |
1748 | return; | |
1749 | ||
1750 | if ((cpu >= r->ncpu) || (r->last_time == NULL)) { | |
1751 | int i, n = __roundup_pow_of_two(cpu+1); | |
1752 | void *p = r->last_time; | |
1753 | ||
1754 | p = realloc(r->last_time, n * sizeof(u64)); | |
1755 | if (!p) | |
1756 | return; | |
1757 | ||
1758 | r->last_time = p; | |
1759 | for (i = r->ncpu; i < n; ++i) | |
1760 | r->last_time[i] = (u64) 0; | |
1761 | ||
1762 | r->ncpu = n; | |
1763 | } | |
1764 | ||
1765 | r->last_time[cpu] = timestamp; | |
1766 | } | |
1767 | ||
1768 | /* returns last time this event was seen on the given cpu */ | |
1769 | static u64 perf_evsel__get_time(struct perf_evsel *evsel, u32 cpu) | |
1770 | { | |
1771 | struct evsel_runtime *r = perf_evsel__get_runtime(evsel); | |
1772 | ||
1773 | if ((r == NULL) || (r->last_time == NULL) || (cpu >= r->ncpu)) | |
1774 | return 0; | |
1775 | ||
1776 | return r->last_time[cpu]; | |
1777 | } | |
1778 | ||
9b8087d7 | 1779 | static int comm_width = 30; |
49394a2a DA |
1780 | |
1781 | static char *timehist_get_commstr(struct thread *thread) | |
1782 | { | |
1783 | static char str[32]; | |
1784 | const char *comm = thread__comm_str(thread); | |
1785 | pid_t tid = thread->tid; | |
1786 | pid_t pid = thread->pid_; | |
1787 | int n; | |
1788 | ||
1789 | if (pid == 0) | |
1790 | n = scnprintf(str, sizeof(str), "%s", comm); | |
1791 | ||
1792 | else if (tid != pid) | |
1793 | n = scnprintf(str, sizeof(str), "%s[%d/%d]", comm, tid, pid); | |
1794 | ||
1795 | else | |
1796 | n = scnprintf(str, sizeof(str), "%s[%d]", comm, tid); | |
1797 | ||
1798 | if (n > comm_width) | |
1799 | comm_width = n; | |
1800 | ||
1801 | return str; | |
1802 | } | |
1803 | ||
a407b067 | 1804 | static void timehist_header(struct perf_sched *sched) |
49394a2a | 1805 | { |
a407b067 DA |
1806 | u32 ncpus = sched->max_cpu + 1; |
1807 | u32 i, j; | |
1808 | ||
49394a2a DA |
1809 | printf("%15s %6s ", "time", "cpu"); |
1810 | ||
a407b067 DA |
1811 | if (sched->show_cpu_visual) { |
1812 | printf(" "); | |
1813 | for (i = 0, j = 0; i < ncpus; ++i) { | |
1814 | printf("%x", j++); | |
1815 | if (j > 15) | |
1816 | j = 0; | |
1817 | } | |
1818 | printf(" "); | |
1819 | } | |
1820 | ||
0e6758e8 | 1821 | printf(" %-*s %9s %9s %9s", comm_width, |
49394a2a DA |
1822 | "task name", "wait time", "sch delay", "run time"); |
1823 | ||
1824 | printf("\n"); | |
1825 | ||
1826 | /* | |
1827 | * units row | |
1828 | */ | |
1829 | printf("%15s %-6s ", "", ""); | |
1830 | ||
a407b067 DA |
1831 | if (sched->show_cpu_visual) |
1832 | printf(" %*s ", ncpus, ""); | |
1833 | ||
0e6758e8 NK |
1834 | printf(" %-*s %9s %9s %9s\n", comm_width, |
1835 | "[tid/pid]", "(msec)", "(msec)", "(msec)"); | |
49394a2a DA |
1836 | |
1837 | /* | |
1838 | * separator | |
1839 | */ | |
1840 | printf("%.15s %.6s ", graph_dotted_line, graph_dotted_line); | |
1841 | ||
a407b067 DA |
1842 | if (sched->show_cpu_visual) |
1843 | printf(" %.*s ", ncpus, graph_dotted_line); | |
1844 | ||
0e6758e8 | 1845 | printf(" %.*s %.9s %.9s %.9s", comm_width, |
49394a2a DA |
1846 | graph_dotted_line, graph_dotted_line, graph_dotted_line, |
1847 | graph_dotted_line); | |
1848 | ||
1849 | printf("\n"); | |
1850 | } | |
1851 | ||
fc1469f1 DA |
1852 | static void timehist_print_sample(struct perf_sched *sched, |
1853 | struct perf_sample *sample, | |
6c973c90 | 1854 | struct addr_location *al, |
853b7407 DA |
1855 | struct thread *thread, |
1856 | u64 t) | |
49394a2a DA |
1857 | { |
1858 | struct thread_runtime *tr = thread__priv(thread); | |
a407b067 | 1859 | u32 max_cpus = sched->max_cpu + 1; |
49394a2a DA |
1860 | char tstr[64]; |
1861 | ||
853b7407 | 1862 | timestamp__scnprintf_usec(t, tstr, sizeof(tstr)); |
49394a2a DA |
1863 | printf("%15s [%04d] ", tstr, sample->cpu); |
1864 | ||
a407b067 DA |
1865 | if (sched->show_cpu_visual) { |
1866 | u32 i; | |
1867 | char c; | |
1868 | ||
1869 | printf(" "); | |
1870 | for (i = 0; i < max_cpus; ++i) { | |
1871 | /* flag idle times with 'i'; others are sched events */ | |
1872 | if (i == sample->cpu) | |
1873 | c = (thread->tid == 0) ? 'i' : 's'; | |
1874 | else | |
1875 | c = ' '; | |
1876 | printf("%c", c); | |
1877 | } | |
1878 | printf(" "); | |
1879 | } | |
1880 | ||
49394a2a DA |
1881 | printf(" %-*s ", comm_width, timehist_get_commstr(thread)); |
1882 | ||
1883 | print_sched_time(tr->dt_wait, 6); | |
1884 | print_sched_time(tr->dt_delay, 6); | |
1885 | print_sched_time(tr->dt_run, 6); | |
fc1469f1 DA |
1886 | |
1887 | if (sched->show_wakeups) | |
1888 | printf(" %-*s", comm_width, ""); | |
1889 | ||
6c973c90 DA |
1890 | if (thread->tid == 0) |
1891 | goto out; | |
1892 | ||
1893 | if (sched->show_callchain) | |
1894 | printf(" "); | |
1895 | ||
1896 | sample__fprintf_sym(sample, al, 0, | |
1897 | EVSEL__PRINT_SYM | EVSEL__PRINT_ONELINE | | |
2d9bbf6e NK |
1898 | EVSEL__PRINT_CALLCHAIN_ARROW | |
1899 | EVSEL__PRINT_SKIP_IGNORED, | |
6c973c90 DA |
1900 | &callchain_cursor, stdout); |
1901 | ||
1902 | out: | |
49394a2a DA |
1903 | printf("\n"); |
1904 | } | |
1905 | ||
1906 | /* | |
1907 | * Explanation of delta-time stats: | |
1908 | * | |
1909 | * t = time of current schedule out event | |
1910 | * tprev = time of previous sched out event | |
1911 | * also time of schedule-in event for current task | |
1912 | * last_time = time of last sched change event for current task | |
1913 | * (i.e, time process was last scheduled out) | |
1914 | * ready_to_run = time of wakeup for current task | |
1915 | * | |
1916 | * -----|------------|------------|------------|------ | |
1917 | * last ready tprev t | |
1918 | * time to run | |
1919 | * | |
1920 | * |-------- dt_wait --------| | |
1921 | * |- dt_delay -|-- dt_run --| | |
1922 | * | |
1923 | * dt_run = run time of current task | |
1924 | * dt_wait = time between last schedule out event for task and tprev | |
1925 | * represents time spent off the cpu | |
1926 | * dt_delay = time between wakeup and schedule-in of task | |
1927 | */ | |
1928 | ||
1929 | static void timehist_update_runtime_stats(struct thread_runtime *r, | |
1930 | u64 t, u64 tprev) | |
1931 | { | |
1932 | r->dt_delay = 0; | |
1933 | r->dt_wait = 0; | |
1934 | r->dt_run = 0; | |
1935 | if (tprev) { | |
1936 | r->dt_run = t - tprev; | |
1937 | if (r->ready_to_run) { | |
1938 | if (r->ready_to_run > tprev) | |
1939 | pr_debug("time travel: wakeup time for task > previous sched_switch event\n"); | |
1940 | else | |
1941 | r->dt_delay = tprev - r->ready_to_run; | |
1942 | } | |
1943 | ||
1944 | if (r->last_time > tprev) | |
1945 | pr_debug("time travel: last sched out time for task > previous sched_switch event\n"); | |
1946 | else if (r->last_time) | |
1947 | r->dt_wait = tprev - r->last_time; | |
1948 | } | |
1949 | ||
1950 | update_stats(&r->run_stats, r->dt_run); | |
1951 | r->total_run_time += r->dt_run; | |
1952 | } | |
1953 | ||
96039c7c NK |
1954 | static bool is_idle_sample(struct perf_sample *sample, |
1955 | struct perf_evsel *evsel) | |
49394a2a DA |
1956 | { |
1957 | /* pid 0 == swapper == idle task */ | |
96039c7c NK |
1958 | if (strcmp(perf_evsel__name(evsel), "sched:sched_switch") == 0) |
1959 | return perf_evsel__intval(evsel, sample, "prev_pid") == 0; | |
49394a2a | 1960 | |
96039c7c NK |
1961 | return sample->pid == 0; |
1962 | } | |
1963 | ||
1964 | static void save_task_callchain(struct perf_sched *sched, | |
1965 | struct perf_sample *sample, | |
1966 | struct perf_evsel *evsel, | |
1967 | struct machine *machine) | |
1968 | { | |
1969 | struct callchain_cursor *cursor = &callchain_cursor; | |
1970 | struct thread *thread; | |
6c973c90 DA |
1971 | |
1972 | /* want main thread for process - has maps */ | |
1973 | thread = machine__findnew_thread(machine, sample->pid, sample->pid); | |
1974 | if (thread == NULL) { | |
1975 | pr_debug("Failed to get thread for pid %d.\n", sample->pid); | |
96039c7c | 1976 | return; |
6c973c90 DA |
1977 | } |
1978 | ||
1979 | if (!symbol_conf.use_callchain || sample->callchain == NULL) | |
96039c7c | 1980 | return; |
6c973c90 DA |
1981 | |
1982 | if (thread__resolve_callchain(thread, cursor, evsel, sample, | |
8388deb3 | 1983 | NULL, NULL, sched->max_stack + 2) != 0) { |
6c973c90 DA |
1984 | if (verbose) |
1985 | error("Failed to resolve callchain. Skipping\n"); | |
1986 | ||
96039c7c | 1987 | return; |
6c973c90 | 1988 | } |
cdeb01bf | 1989 | |
6c973c90 | 1990 | callchain_cursor_commit(cursor); |
cdeb01bf NK |
1991 | |
1992 | while (true) { | |
1993 | struct callchain_cursor_node *node; | |
1994 | struct symbol *sym; | |
1995 | ||
1996 | node = callchain_cursor_current(cursor); | |
1997 | if (node == NULL) | |
1998 | break; | |
1999 | ||
2000 | sym = node->sym; | |
2001 | if (sym && sym->name) { | |
2002 | if (!strcmp(sym->name, "schedule") || | |
2003 | !strcmp(sym->name, "__schedule") || | |
2004 | !strcmp(sym->name, "preempt_schedule")) | |
2005 | sym->ignore = 1; | |
2006 | } | |
2007 | ||
2008 | callchain_cursor_advance(cursor); | |
2009 | } | |
49394a2a DA |
2010 | } |
2011 | ||
3bc2fa9c NK |
2012 | static int init_idle_thread(struct thread *thread) |
2013 | { | |
2014 | struct idle_thread_runtime *itr; | |
2015 | ||
2016 | thread__set_comm(thread, idle_comm, 0); | |
2017 | ||
2018 | itr = zalloc(sizeof(*itr)); | |
2019 | if (itr == NULL) | |
2020 | return -ENOMEM; | |
2021 | ||
2022 | init_stats(&itr->tr.run_stats); | |
2023 | callchain_init(&itr->callchain); | |
2024 | callchain_cursor_reset(&itr->cursor); | |
2025 | thread__set_priv(thread, itr); | |
2026 | ||
2027 | return 0; | |
2028 | } | |
2029 | ||
49394a2a DA |
2030 | /* |
2031 | * Track idle stats per cpu by maintaining a local thread | |
2032 | * struct for the idle task on each cpu. | |
2033 | */ | |
2034 | static int init_idle_threads(int ncpu) | |
2035 | { | |
3bc2fa9c | 2036 | int i, ret; |
49394a2a DA |
2037 | |
2038 | idle_threads = zalloc(ncpu * sizeof(struct thread *)); | |
2039 | if (!idle_threads) | |
2040 | return -ENOMEM; | |
2041 | ||
b336352b | 2042 | idle_max_cpu = ncpu; |
49394a2a DA |
2043 | |
2044 | /* allocate the actual thread struct if needed */ | |
2045 | for (i = 0; i < ncpu; ++i) { | |
2046 | idle_threads[i] = thread__new(0, 0); | |
2047 | if (idle_threads[i] == NULL) | |
2048 | return -ENOMEM; | |
2049 | ||
3bc2fa9c NK |
2050 | ret = init_idle_thread(idle_threads[i]); |
2051 | if (ret < 0) | |
2052 | return ret; | |
49394a2a DA |
2053 | } |
2054 | ||
2055 | return 0; | |
2056 | } | |
2057 | ||
2058 | static void free_idle_threads(void) | |
2059 | { | |
2060 | int i; | |
2061 | ||
2062 | if (idle_threads == NULL) | |
2063 | return; | |
2064 | ||
b336352b | 2065 | for (i = 0; i < idle_max_cpu; ++i) { |
49394a2a DA |
2066 | if ((idle_threads[i])) |
2067 | thread__delete(idle_threads[i]); | |
2068 | } | |
2069 | ||
2070 | free(idle_threads); | |
2071 | } | |
2072 | ||
2073 | static struct thread *get_idle_thread(int cpu) | |
2074 | { | |
2075 | /* | |
2076 | * expand/allocate array of pointers to local thread | |
2077 | * structs if needed | |
2078 | */ | |
2079 | if ((cpu >= idle_max_cpu) || (idle_threads == NULL)) { | |
2080 | int i, j = __roundup_pow_of_two(cpu+1); | |
2081 | void *p; | |
2082 | ||
2083 | p = realloc(idle_threads, j * sizeof(struct thread *)); | |
2084 | if (!p) | |
2085 | return NULL; | |
2086 | ||
2087 | idle_threads = (struct thread **) p; | |
b336352b | 2088 | for (i = idle_max_cpu; i < j; ++i) |
49394a2a DA |
2089 | idle_threads[i] = NULL; |
2090 | ||
2091 | idle_max_cpu = j; | |
2092 | } | |
2093 | ||
2094 | /* allocate a new thread struct if needed */ | |
2095 | if (idle_threads[cpu] == NULL) { | |
2096 | idle_threads[cpu] = thread__new(0, 0); | |
2097 | ||
2098 | if (idle_threads[cpu]) { | |
3bc2fa9c NK |
2099 | if (init_idle_thread(idle_threads[cpu]) < 0) |
2100 | return NULL; | |
49394a2a DA |
2101 | } |
2102 | } | |
2103 | ||
2104 | return idle_threads[cpu]; | |
2105 | } | |
2106 | ||
699b5b92 NK |
2107 | static void save_idle_callchain(struct idle_thread_runtime *itr, |
2108 | struct perf_sample *sample) | |
2109 | { | |
2110 | if (!symbol_conf.use_callchain || sample->callchain == NULL) | |
2111 | return; | |
2112 | ||
2113 | callchain_cursor__copy(&itr->cursor, &callchain_cursor); | |
2114 | } | |
2115 | ||
49394a2a DA |
2116 | /* |
2117 | * handle runtime stats saved per thread | |
2118 | */ | |
2119 | static struct thread_runtime *thread__init_runtime(struct thread *thread) | |
2120 | { | |
2121 | struct thread_runtime *r; | |
2122 | ||
2123 | r = zalloc(sizeof(struct thread_runtime)); | |
2124 | if (!r) | |
2125 | return NULL; | |
2126 | ||
2127 | init_stats(&r->run_stats); | |
2128 | thread__set_priv(thread, r); | |
2129 | ||
2130 | return r; | |
2131 | } | |
2132 | ||
2133 | static struct thread_runtime *thread__get_runtime(struct thread *thread) | |
2134 | { | |
2135 | struct thread_runtime *tr; | |
2136 | ||
2137 | tr = thread__priv(thread); | |
2138 | if (tr == NULL) { | |
2139 | tr = thread__init_runtime(thread); | |
2140 | if (tr == NULL) | |
2141 | pr_debug("Failed to malloc memory for runtime data.\n"); | |
2142 | } | |
2143 | ||
2144 | return tr; | |
2145 | } | |
2146 | ||
6c973c90 DA |
2147 | static struct thread *timehist_get_thread(struct perf_sched *sched, |
2148 | struct perf_sample *sample, | |
49394a2a DA |
2149 | struct machine *machine, |
2150 | struct perf_evsel *evsel) | |
2151 | { | |
2152 | struct thread *thread; | |
2153 | ||
96039c7c | 2154 | if (is_idle_sample(sample, evsel)) { |
49394a2a DA |
2155 | thread = get_idle_thread(sample->cpu); |
2156 | if (thread == NULL) | |
2157 | pr_err("Failed to get idle thread for cpu %d.\n", sample->cpu); | |
2158 | ||
2159 | } else { | |
5d92d96a NK |
2160 | /* there were samples with tid 0 but non-zero pid */ |
2161 | thread = machine__findnew_thread(machine, sample->pid, | |
2162 | sample->tid ?: sample->pid); | |
49394a2a DA |
2163 | if (thread == NULL) { |
2164 | pr_debug("Failed to get thread for tid %d. skipping sample.\n", | |
2165 | sample->tid); | |
2166 | } | |
96039c7c NK |
2167 | |
2168 | save_task_callchain(sched, sample, evsel, machine); | |
699b5b92 NK |
2169 | if (sched->idle_hist) { |
2170 | struct thread *idle; | |
2171 | struct idle_thread_runtime *itr; | |
2172 | ||
2173 | idle = get_idle_thread(sample->cpu); | |
2174 | if (idle == NULL) { | |
2175 | pr_err("Failed to get idle thread for cpu %d.\n", sample->cpu); | |
2176 | return NULL; | |
2177 | } | |
2178 | ||
2179 | itr = thread__priv(idle); | |
2180 | if (itr == NULL) | |
2181 | return NULL; | |
2182 | ||
2183 | itr->last_thread = thread; | |
2184 | ||
2185 | /* copy task callchain when entering to idle */ | |
2186 | if (perf_evsel__intval(evsel, sample, "next_pid") == 0) | |
2187 | save_idle_callchain(itr, sample); | |
2188 | } | |
49394a2a DA |
2189 | } |
2190 | ||
2191 | return thread; | |
2192 | } | |
2193 | ||
52df138c | 2194 | static bool timehist_skip_sample(struct perf_sched *sched, |
a4b2b6f5 NK |
2195 | struct thread *thread, |
2196 | struct perf_evsel *evsel, | |
2197 | struct perf_sample *sample) | |
49394a2a DA |
2198 | { |
2199 | bool rc = false; | |
2200 | ||
52df138c | 2201 | if (thread__is_filtered(thread)) { |
49394a2a | 2202 | rc = true; |
52df138c DA |
2203 | sched->skipped_samples++; |
2204 | } | |
49394a2a | 2205 | |
a4b2b6f5 NK |
2206 | if (sched->idle_hist) { |
2207 | if (strcmp(perf_evsel__name(evsel), "sched:sched_switch")) | |
2208 | rc = true; | |
2209 | else if (perf_evsel__intval(evsel, sample, "prev_pid") != 0 && | |
2210 | perf_evsel__intval(evsel, sample, "next_pid") != 0) | |
2211 | rc = true; | |
2212 | } | |
2213 | ||
49394a2a DA |
2214 | return rc; |
2215 | } | |
2216 | ||
fc1469f1 | 2217 | static void timehist_print_wakeup_event(struct perf_sched *sched, |
a4b2b6f5 | 2218 | struct perf_evsel *evsel, |
fc1469f1 DA |
2219 | struct perf_sample *sample, |
2220 | struct machine *machine, | |
2221 | struct thread *awakened) | |
2222 | { | |
2223 | struct thread *thread; | |
2224 | char tstr[64]; | |
2225 | ||
2226 | thread = machine__findnew_thread(machine, sample->pid, sample->tid); | |
2227 | if (thread == NULL) | |
2228 | return; | |
2229 | ||
2230 | /* show wakeup unless both awakee and awaker are filtered */ | |
a4b2b6f5 NK |
2231 | if (timehist_skip_sample(sched, thread, evsel, sample) && |
2232 | timehist_skip_sample(sched, awakened, evsel, sample)) { | |
fc1469f1 DA |
2233 | return; |
2234 | } | |
2235 | ||
2236 | timestamp__scnprintf_usec(sample->time, tstr, sizeof(tstr)); | |
2237 | printf("%15s [%04d] ", tstr, sample->cpu); | |
a407b067 DA |
2238 | if (sched->show_cpu_visual) |
2239 | printf(" %*s ", sched->max_cpu + 1, ""); | |
fc1469f1 DA |
2240 | |
2241 | printf(" %-*s ", comm_width, timehist_get_commstr(thread)); | |
2242 | ||
2243 | /* dt spacer */ | |
2244 | printf(" %9s %9s %9s ", "", "", ""); | |
2245 | ||
2246 | printf("awakened: %s", timehist_get_commstr(awakened)); | |
2247 | ||
2248 | printf("\n"); | |
2249 | } | |
2250 | ||
2251 | static int timehist_sched_wakeup_event(struct perf_tool *tool, | |
49394a2a DA |
2252 | union perf_event *event __maybe_unused, |
2253 | struct perf_evsel *evsel, | |
2254 | struct perf_sample *sample, | |
2255 | struct machine *machine) | |
2256 | { | |
fc1469f1 | 2257 | struct perf_sched *sched = container_of(tool, struct perf_sched, tool); |
49394a2a DA |
2258 | struct thread *thread; |
2259 | struct thread_runtime *tr = NULL; | |
2260 | /* want pid of awakened task not pid in sample */ | |
2261 | const u32 pid = perf_evsel__intval(evsel, sample, "pid"); | |
2262 | ||
2263 | thread = machine__findnew_thread(machine, 0, pid); | |
2264 | if (thread == NULL) | |
2265 | return -1; | |
2266 | ||
2267 | tr = thread__get_runtime(thread); | |
2268 | if (tr == NULL) | |
2269 | return -1; | |
2270 | ||
2271 | if (tr->ready_to_run == 0) | |
2272 | tr->ready_to_run = sample->time; | |
2273 | ||
fc1469f1 | 2274 | /* show wakeups if requested */ |
853b7407 DA |
2275 | if (sched->show_wakeups && |
2276 | !perf_time__skip_sample(&sched->ptime, sample->time)) | |
a4b2b6f5 | 2277 | timehist_print_wakeup_event(sched, evsel, sample, machine, thread); |
fc1469f1 | 2278 | |
49394a2a DA |
2279 | return 0; |
2280 | } | |
2281 | ||
350f54fa DA |
2282 | static void timehist_print_migration_event(struct perf_sched *sched, |
2283 | struct perf_evsel *evsel, | |
2284 | struct perf_sample *sample, | |
2285 | struct machine *machine, | |
2286 | struct thread *migrated) | |
2287 | { | |
2288 | struct thread *thread; | |
2289 | char tstr[64]; | |
2290 | u32 max_cpus = sched->max_cpu + 1; | |
2291 | u32 ocpu, dcpu; | |
2292 | ||
2293 | if (sched->summary_only) | |
2294 | return; | |
2295 | ||
2296 | max_cpus = sched->max_cpu + 1; | |
2297 | ocpu = perf_evsel__intval(evsel, sample, "orig_cpu"); | |
2298 | dcpu = perf_evsel__intval(evsel, sample, "dest_cpu"); | |
2299 | ||
2300 | thread = machine__findnew_thread(machine, sample->pid, sample->tid); | |
2301 | if (thread == NULL) | |
2302 | return; | |
2303 | ||
a4b2b6f5 NK |
2304 | if (timehist_skip_sample(sched, thread, evsel, sample) && |
2305 | timehist_skip_sample(sched, migrated, evsel, sample)) { | |
350f54fa DA |
2306 | return; |
2307 | } | |
2308 | ||
2309 | timestamp__scnprintf_usec(sample->time, tstr, sizeof(tstr)); | |
2310 | printf("%15s [%04d] ", tstr, sample->cpu); | |
2311 | ||
2312 | if (sched->show_cpu_visual) { | |
2313 | u32 i; | |
2314 | char c; | |
2315 | ||
2316 | printf(" "); | |
2317 | for (i = 0; i < max_cpus; ++i) { | |
2318 | c = (i == sample->cpu) ? 'm' : ' '; | |
2319 | printf("%c", c); | |
2320 | } | |
2321 | printf(" "); | |
2322 | } | |
2323 | ||
2324 | printf(" %-*s ", comm_width, timehist_get_commstr(thread)); | |
2325 | ||
2326 | /* dt spacer */ | |
2327 | printf(" %9s %9s %9s ", "", "", ""); | |
2328 | ||
2329 | printf("migrated: %s", timehist_get_commstr(migrated)); | |
2330 | printf(" cpu %d => %d", ocpu, dcpu); | |
2331 | ||
2332 | printf("\n"); | |
2333 | } | |
2334 | ||
2335 | static int timehist_migrate_task_event(struct perf_tool *tool, | |
2336 | union perf_event *event __maybe_unused, | |
2337 | struct perf_evsel *evsel, | |
2338 | struct perf_sample *sample, | |
2339 | struct machine *machine) | |
2340 | { | |
2341 | struct perf_sched *sched = container_of(tool, struct perf_sched, tool); | |
2342 | struct thread *thread; | |
2343 | struct thread_runtime *tr = NULL; | |
2344 | /* want pid of migrated task not pid in sample */ | |
2345 | const u32 pid = perf_evsel__intval(evsel, sample, "pid"); | |
2346 | ||
2347 | thread = machine__findnew_thread(machine, 0, pid); | |
2348 | if (thread == NULL) | |
2349 | return -1; | |
2350 | ||
2351 | tr = thread__get_runtime(thread); | |
2352 | if (tr == NULL) | |
2353 | return -1; | |
2354 | ||
2355 | tr->migrations++; | |
2356 | ||
2357 | /* show migrations if requested */ | |
2358 | timehist_print_migration_event(sched, evsel, sample, machine, thread); | |
2359 | ||
2360 | return 0; | |
2361 | } | |
2362 | ||
52df138c | 2363 | static int timehist_sched_change_event(struct perf_tool *tool, |
49394a2a DA |
2364 | union perf_event *event, |
2365 | struct perf_evsel *evsel, | |
2366 | struct perf_sample *sample, | |
2367 | struct machine *machine) | |
2368 | { | |
fc1469f1 | 2369 | struct perf_sched *sched = container_of(tool, struct perf_sched, tool); |
853b7407 | 2370 | struct perf_time_interval *ptime = &sched->ptime; |
49394a2a DA |
2371 | struct addr_location al; |
2372 | struct thread *thread; | |
2373 | struct thread_runtime *tr = NULL; | |
853b7407 | 2374 | u64 tprev, t = sample->time; |
49394a2a DA |
2375 | int rc = 0; |
2376 | ||
2377 | if (machine__resolve(machine, &al, sample) < 0) { | |
2378 | pr_err("problem processing %d event. skipping it\n", | |
2379 | event->header.type); | |
2380 | rc = -1; | |
2381 | goto out; | |
2382 | } | |
2383 | ||
6c973c90 | 2384 | thread = timehist_get_thread(sched, sample, machine, evsel); |
49394a2a DA |
2385 | if (thread == NULL) { |
2386 | rc = -1; | |
2387 | goto out; | |
2388 | } | |
2389 | ||
a4b2b6f5 | 2390 | if (timehist_skip_sample(sched, thread, evsel, sample)) |
49394a2a DA |
2391 | goto out; |
2392 | ||
2393 | tr = thread__get_runtime(thread); | |
2394 | if (tr == NULL) { | |
2395 | rc = -1; | |
2396 | goto out; | |
2397 | } | |
2398 | ||
2399 | tprev = perf_evsel__get_time(evsel, sample->cpu); | |
2400 | ||
853b7407 DA |
2401 | /* |
2402 | * If start time given: | |
2403 | * - sample time is under window user cares about - skip sample | |
2404 | * - tprev is under window user cares about - reset to start of window | |
2405 | */ | |
2406 | if (ptime->start && ptime->start > t) | |
2407 | goto out; | |
2408 | ||
bdd75729 | 2409 | if (tprev && ptime->start > tprev) |
853b7407 DA |
2410 | tprev = ptime->start; |
2411 | ||
2412 | /* | |
2413 | * If end time given: | |
2414 | * - previous sched event is out of window - we are done | |
2415 | * - sample time is beyond window user cares about - reset it | |
2416 | * to close out stats for time window interest | |
2417 | */ | |
2418 | if (ptime->end) { | |
2419 | if (tprev > ptime->end) | |
2420 | goto out; | |
2421 | ||
2422 | if (t > ptime->end) | |
2423 | t = ptime->end; | |
2424 | } | |
2425 | ||
07235f84 NK |
2426 | if (!sched->idle_hist || thread->tid == 0) { |
2427 | timehist_update_runtime_stats(tr, t, tprev); | |
2428 | ||
2429 | if (sched->idle_hist) { | |
2430 | struct idle_thread_runtime *itr = (void *)tr; | |
2431 | struct thread_runtime *last_tr; | |
2432 | ||
2433 | BUG_ON(thread->tid != 0); | |
2434 | ||
2435 | if (itr->last_thread == NULL) | |
2436 | goto out; | |
2437 | ||
2438 | /* add current idle time as last thread's runtime */ | |
2439 | last_tr = thread__get_runtime(itr->last_thread); | |
2440 | if (last_tr == NULL) | |
2441 | goto out; | |
2442 | ||
2443 | timehist_update_runtime_stats(last_tr, t, tprev); | |
2444 | /* | |
2445 | * remove delta time of last thread as it's not updated | |
2446 | * and otherwise it will show an invalid value next | |
2447 | * time. we only care total run time and run stat. | |
2448 | */ | |
2449 | last_tr->dt_run = 0; | |
2450 | last_tr->dt_wait = 0; | |
2451 | last_tr->dt_delay = 0; | |
2452 | ||
ba957ebb NK |
2453 | if (itr->cursor.nr) |
2454 | callchain_append(&itr->callchain, &itr->cursor, t - tprev); | |
2455 | ||
07235f84 NK |
2456 | itr->last_thread = NULL; |
2457 | } | |
2458 | } | |
853b7407 | 2459 | |
52df138c | 2460 | if (!sched->summary_only) |
853b7407 | 2461 | timehist_print_sample(sched, sample, &al, thread, t); |
49394a2a DA |
2462 | |
2463 | out: | |
9396c9cb NK |
2464 | if (sched->hist_time.start == 0 && t >= ptime->start) |
2465 | sched->hist_time.start = t; | |
2466 | if (ptime->end == 0 || t <= ptime->end) | |
2467 | sched->hist_time.end = t; | |
2468 | ||
49394a2a DA |
2469 | if (tr) { |
2470 | /* time of this sched_switch event becomes last time task seen */ | |
2471 | tr->last_time = sample->time; | |
2472 | ||
2473 | /* sched out event for task so reset ready to run time */ | |
2474 | tr->ready_to_run = 0; | |
2475 | } | |
2476 | ||
2477 | perf_evsel__save_time(evsel, sample->time, sample->cpu); | |
2478 | ||
2479 | return rc; | |
2480 | } | |
2481 | ||
2482 | static int timehist_sched_switch_event(struct perf_tool *tool, | |
2483 | union perf_event *event, | |
2484 | struct perf_evsel *evsel, | |
2485 | struct perf_sample *sample, | |
2486 | struct machine *machine __maybe_unused) | |
2487 | { | |
2488 | return timehist_sched_change_event(tool, event, evsel, sample, machine); | |
2489 | } | |
2490 | ||
2491 | static int process_lost(struct perf_tool *tool __maybe_unused, | |
2492 | union perf_event *event, | |
2493 | struct perf_sample *sample, | |
2494 | struct machine *machine __maybe_unused) | |
2495 | { | |
2496 | char tstr[64]; | |
2497 | ||
2498 | timestamp__scnprintf_usec(sample->time, tstr, sizeof(tstr)); | |
2499 | printf("%15s ", tstr); | |
2500 | printf("lost %" PRIu64 " events on cpu %d\n", event->lost.lost, sample->cpu); | |
2501 | ||
2502 | return 0; | |
2503 | } | |
2504 | ||
2505 | ||
52df138c DA |
2506 | static void print_thread_runtime(struct thread *t, |
2507 | struct thread_runtime *r) | |
2508 | { | |
2509 | double mean = avg_stats(&r->run_stats); | |
2510 | float stddev; | |
2511 | ||
2512 | printf("%*s %5d %9" PRIu64 " ", | |
2513 | comm_width, timehist_get_commstr(t), t->ppid, | |
2514 | (u64) r->run_stats.n); | |
2515 | ||
2516 | print_sched_time(r->total_run_time, 8); | |
2517 | stddev = rel_stddev_stats(stddev_stats(&r->run_stats), mean); | |
2518 | print_sched_time(r->run_stats.min, 6); | |
2519 | printf(" "); | |
2520 | print_sched_time((u64) mean, 6); | |
2521 | printf(" "); | |
2522 | print_sched_time(r->run_stats.max, 6); | |
2523 | printf(" "); | |
2524 | printf("%5.2f", stddev); | |
350f54fa | 2525 | printf(" %5" PRIu64, r->migrations); |
52df138c DA |
2526 | printf("\n"); |
2527 | } | |
2528 | ||
2529 | struct total_run_stats { | |
2530 | u64 sched_count; | |
2531 | u64 task_count; | |
2532 | u64 total_run_time; | |
2533 | }; | |
2534 | ||
2535 | static int __show_thread_runtime(struct thread *t, void *priv) | |
2536 | { | |
2537 | struct total_run_stats *stats = priv; | |
2538 | struct thread_runtime *r; | |
2539 | ||
2540 | if (thread__is_filtered(t)) | |
2541 | return 0; | |
2542 | ||
2543 | r = thread__priv(t); | |
2544 | if (r && r->run_stats.n) { | |
2545 | stats->task_count++; | |
2546 | stats->sched_count += r->run_stats.n; | |
2547 | stats->total_run_time += r->total_run_time; | |
2548 | print_thread_runtime(t, r); | |
2549 | } | |
2550 | ||
2551 | return 0; | |
2552 | } | |
2553 | ||
2554 | static int show_thread_runtime(struct thread *t, void *priv) | |
2555 | { | |
2556 | if (t->dead) | |
2557 | return 0; | |
2558 | ||
2559 | return __show_thread_runtime(t, priv); | |
2560 | } | |
2561 | ||
2562 | static int show_deadthread_runtime(struct thread *t, void *priv) | |
2563 | { | |
2564 | if (!t->dead) | |
2565 | return 0; | |
2566 | ||
2567 | return __show_thread_runtime(t, priv); | |
2568 | } | |
2569 | ||
ba957ebb NK |
2570 | static size_t callchain__fprintf_folded(FILE *fp, struct callchain_node *node) |
2571 | { | |
2572 | const char *sep = " <- "; | |
2573 | struct callchain_list *chain; | |
2574 | size_t ret = 0; | |
2575 | char bf[1024]; | |
2576 | bool first; | |
2577 | ||
2578 | if (node == NULL) | |
2579 | return 0; | |
2580 | ||
2581 | ret = callchain__fprintf_folded(fp, node->parent); | |
2582 | first = (ret == 0); | |
2583 | ||
2584 | list_for_each_entry(chain, &node->val, list) { | |
2585 | if (chain->ip >= PERF_CONTEXT_MAX) | |
2586 | continue; | |
2587 | if (chain->ms.sym && chain->ms.sym->ignore) | |
2588 | continue; | |
2589 | ret += fprintf(fp, "%s%s", first ? "" : sep, | |
2590 | callchain_list__sym_name(chain, bf, sizeof(bf), | |
2591 | false)); | |
2592 | first = false; | |
2593 | } | |
2594 | ||
2595 | return ret; | |
2596 | } | |
2597 | ||
2598 | static size_t timehist_print_idlehist_callchain(struct rb_root *root) | |
2599 | { | |
2600 | size_t ret = 0; | |
2601 | FILE *fp = stdout; | |
2602 | struct callchain_node *chain; | |
2603 | struct rb_node *rb_node = rb_first(root); | |
2604 | ||
2605 | printf(" %16s %8s %s\n", "Idle time (msec)", "Count", "Callchains"); | |
2606 | printf(" %.16s %.8s %.50s\n", graph_dotted_line, graph_dotted_line, | |
2607 | graph_dotted_line); | |
2608 | ||
2609 | while (rb_node) { | |
2610 | chain = rb_entry(rb_node, struct callchain_node, rb_node); | |
2611 | rb_node = rb_next(rb_node); | |
2612 | ||
2613 | ret += fprintf(fp, " "); | |
2614 | print_sched_time(chain->hit, 12); | |
2615 | ret += 16; /* print_sched_time returns 2nd arg + 4 */ | |
2616 | ret += fprintf(fp, " %8d ", chain->count); | |
2617 | ret += callchain__fprintf_folded(fp, chain); | |
2618 | ret += fprintf(fp, "\n"); | |
2619 | } | |
2620 | ||
2621 | return ret; | |
2622 | } | |
2623 | ||
52df138c DA |
2624 | static void timehist_print_summary(struct perf_sched *sched, |
2625 | struct perf_session *session) | |
2626 | { | |
2627 | struct machine *m = &session->machines.host; | |
2628 | struct total_run_stats totals; | |
2629 | u64 task_count; | |
2630 | struct thread *t; | |
2631 | struct thread_runtime *r; | |
2632 | int i; | |
9396c9cb | 2633 | u64 hist_time = sched->hist_time.end - sched->hist_time.start; |
52df138c DA |
2634 | |
2635 | memset(&totals, 0, sizeof(totals)); | |
2636 | ||
07235f84 NK |
2637 | if (sched->idle_hist) { |
2638 | printf("\nIdle-time summary\n"); | |
2639 | printf("%*s parent sched-out ", comm_width, "comm"); | |
2640 | printf(" idle-time min-idle avg-idle max-idle stddev migrations\n"); | |
2641 | } else { | |
2642 | printf("\nRuntime summary\n"); | |
2643 | printf("%*s parent sched-in ", comm_width, "comm"); | |
2644 | printf(" run-time min-run avg-run max-run stddev migrations\n"); | |
2645 | } | |
52df138c DA |
2646 | printf("%*s (count) ", comm_width, ""); |
2647 | printf(" (msec) (msec) (msec) (msec) %%\n"); | |
350f54fa | 2648 | printf("%.117s\n", graph_dotted_line); |
52df138c DA |
2649 | |
2650 | machine__for_each_thread(m, show_thread_runtime, &totals); | |
2651 | task_count = totals.task_count; | |
2652 | if (!task_count) | |
2653 | printf("<no still running tasks>\n"); | |
2654 | ||
2655 | printf("\nTerminated tasks:\n"); | |
2656 | machine__for_each_thread(m, show_deadthread_runtime, &totals); | |
2657 | if (task_count == totals.task_count) | |
2658 | printf("<no terminated tasks>\n"); | |
2659 | ||
2660 | /* CPU idle stats not tracked when samples were skipped */ | |
07235f84 | 2661 | if (sched->skipped_samples && !sched->idle_hist) |
52df138c DA |
2662 | return; |
2663 | ||
2664 | printf("\nIdle stats:\n"); | |
b336352b | 2665 | for (i = 0; i < idle_max_cpu; ++i) { |
52df138c DA |
2666 | t = idle_threads[i]; |
2667 | if (!t) | |
2668 | continue; | |
2669 | ||
2670 | r = thread__priv(t); | |
2671 | if (r && r->run_stats.n) { | |
2672 | totals.sched_count += r->run_stats.n; | |
2673 | printf(" CPU %2d idle for ", i); | |
2674 | print_sched_time(r->total_run_time, 6); | |
9396c9cb | 2675 | printf(" msec (%6.2f%%)\n", 100.0 * r->total_run_time / hist_time); |
52df138c DA |
2676 | } else |
2677 | printf(" CPU %2d idle entire time window\n", i); | |
2678 | } | |
2679 | ||
ba957ebb NK |
2680 | if (sched->idle_hist && symbol_conf.use_callchain) { |
2681 | callchain_param.mode = CHAIN_FOLDED; | |
2682 | callchain_param.value = CCVAL_PERIOD; | |
2683 | ||
2684 | callchain_register_param(&callchain_param); | |
2685 | ||
2686 | printf("\nIdle stats by callchain:\n"); | |
2687 | for (i = 0; i < idle_max_cpu; ++i) { | |
2688 | struct idle_thread_runtime *itr; | |
2689 | ||
2690 | t = idle_threads[i]; | |
2691 | if (!t) | |
2692 | continue; | |
2693 | ||
2694 | itr = thread__priv(t); | |
2695 | if (itr == NULL) | |
2696 | continue; | |
2697 | ||
2698 | callchain_param.sort(&itr->sorted_root, &itr->callchain, | |
2699 | 0, &callchain_param); | |
2700 | ||
2701 | printf(" CPU %2d:", i); | |
2702 | print_sched_time(itr->tr.total_run_time, 6); | |
2703 | printf(" msec\n"); | |
2704 | timehist_print_idlehist_callchain(&itr->sorted_root); | |
2705 | printf("\n"); | |
2706 | } | |
2707 | } | |
2708 | ||
52df138c DA |
2709 | printf("\n" |
2710 | " Total number of unique tasks: %" PRIu64 "\n" | |
9396c9cb | 2711 | "Total number of context switches: %" PRIu64 "\n", |
52df138c DA |
2712 | totals.task_count, totals.sched_count); |
2713 | ||
9396c9cb | 2714 | printf(" Total run time (msec): "); |
52df138c DA |
2715 | print_sched_time(totals.total_run_time, 2); |
2716 | printf("\n"); | |
9396c9cb NK |
2717 | |
2718 | printf(" Total scheduling time (msec): "); | |
2719 | print_sched_time(hist_time, 2); | |
2720 | printf(" (x %d)\n", sched->max_cpu); | |
52df138c DA |
2721 | } |
2722 | ||
49394a2a DA |
2723 | typedef int (*sched_handler)(struct perf_tool *tool, |
2724 | union perf_event *event, | |
2725 | struct perf_evsel *evsel, | |
2726 | struct perf_sample *sample, | |
2727 | struct machine *machine); | |
2728 | ||
2729 | static int perf_timehist__process_sample(struct perf_tool *tool, | |
2730 | union perf_event *event, | |
2731 | struct perf_sample *sample, | |
2732 | struct perf_evsel *evsel, | |
2733 | struct machine *machine) | |
2734 | { | |
2735 | struct perf_sched *sched = container_of(tool, struct perf_sched, tool); | |
2736 | int err = 0; | |
2737 | int this_cpu = sample->cpu; | |
2738 | ||
2739 | if (this_cpu > sched->max_cpu) | |
2740 | sched->max_cpu = this_cpu; | |
2741 | ||
2742 | if (evsel->handler != NULL) { | |
2743 | sched_handler f = evsel->handler; | |
2744 | ||
2745 | err = f(tool, event, evsel, sample, machine); | |
2746 | } | |
2747 | ||
2748 | return err; | |
2749 | } | |
2750 | ||
6c973c90 DA |
2751 | static int timehist_check_attr(struct perf_sched *sched, |
2752 | struct perf_evlist *evlist) | |
2753 | { | |
2754 | struct perf_evsel *evsel; | |
2755 | struct evsel_runtime *er; | |
2756 | ||
2757 | list_for_each_entry(evsel, &evlist->entries, node) { | |
2758 | er = perf_evsel__get_runtime(evsel); | |
2759 | if (er == NULL) { | |
2760 | pr_err("Failed to allocate memory for evsel runtime data\n"); | |
2761 | return -1; | |
2762 | } | |
2763 | ||
2764 | if (sched->show_callchain && | |
2765 | !(evsel->attr.sample_type & PERF_SAMPLE_CALLCHAIN)) { | |
2766 | pr_info("Samples do not have callchains.\n"); | |
2767 | sched->show_callchain = 0; | |
2768 | symbol_conf.use_callchain = 0; | |
2769 | } | |
2770 | } | |
2771 | ||
2772 | return 0; | |
2773 | } | |
2774 | ||
49394a2a DA |
2775 | static int perf_sched__timehist(struct perf_sched *sched) |
2776 | { | |
2777 | const struct perf_evsel_str_handler handlers[] = { | |
2778 | { "sched:sched_switch", timehist_sched_switch_event, }, | |
2779 | { "sched:sched_wakeup", timehist_sched_wakeup_event, }, | |
2780 | { "sched:sched_wakeup_new", timehist_sched_wakeup_event, }, | |
2781 | }; | |
350f54fa DA |
2782 | const struct perf_evsel_str_handler migrate_handlers[] = { |
2783 | { "sched:sched_migrate_task", timehist_migrate_task_event, }, | |
2784 | }; | |
49394a2a DA |
2785 | struct perf_data_file file = { |
2786 | .path = input_name, | |
2787 | .mode = PERF_DATA_MODE_READ, | |
6fa94258 | 2788 | .force = sched->force, |
49394a2a DA |
2789 | }; |
2790 | ||
2791 | struct perf_session *session; | |
52df138c | 2792 | struct perf_evlist *evlist; |
49394a2a DA |
2793 | int err = -1; |
2794 | ||
2795 | /* | |
2796 | * event handlers for timehist option | |
2797 | */ | |
2798 | sched->tool.sample = perf_timehist__process_sample; | |
2799 | sched->tool.mmap = perf_event__process_mmap; | |
2800 | sched->tool.comm = perf_event__process_comm; | |
2801 | sched->tool.exit = perf_event__process_exit; | |
2802 | sched->tool.fork = perf_event__process_fork; | |
2803 | sched->tool.lost = process_lost; | |
2804 | sched->tool.attr = perf_event__process_attr; | |
2805 | sched->tool.tracing_data = perf_event__process_tracing_data; | |
2806 | sched->tool.build_id = perf_event__process_build_id; | |
2807 | ||
2808 | sched->tool.ordered_events = true; | |
2809 | sched->tool.ordering_requires_timestamps = true; | |
2810 | ||
6c973c90 DA |
2811 | symbol_conf.use_callchain = sched->show_callchain; |
2812 | ||
49394a2a DA |
2813 | session = perf_session__new(&file, false, &sched->tool); |
2814 | if (session == NULL) | |
2815 | return -ENOMEM; | |
2816 | ||
52df138c DA |
2817 | evlist = session->evlist; |
2818 | ||
49394a2a DA |
2819 | symbol__init(&session->header.env); |
2820 | ||
853b7407 DA |
2821 | if (perf_time__parse_str(&sched->ptime, sched->time_str) != 0) { |
2822 | pr_err("Invalid time string\n"); | |
2823 | return -EINVAL; | |
2824 | } | |
2825 | ||
6c973c90 DA |
2826 | if (timehist_check_attr(sched, evlist) != 0) |
2827 | goto out; | |
2828 | ||
49394a2a DA |
2829 | setup_pager(); |
2830 | ||
2831 | /* setup per-evsel handlers */ | |
2832 | if (perf_session__set_tracepoints_handlers(session, handlers)) | |
2833 | goto out; | |
2834 | ||
f45bf8d3 DA |
2835 | /* sched_switch event at a minimum needs to exist */ |
2836 | if (!perf_evlist__find_tracepoint_by_name(session->evlist, | |
2837 | "sched:sched_switch")) { | |
2838 | pr_err("No sched_switch events found. Have you run 'perf sched record'?\n"); | |
49394a2a | 2839 | goto out; |
f45bf8d3 | 2840 | } |
49394a2a | 2841 | |
350f54fa DA |
2842 | if (sched->show_migrations && |
2843 | perf_session__set_tracepoints_handlers(session, migrate_handlers)) | |
2844 | goto out; | |
2845 | ||
49394a2a DA |
2846 | /* pre-allocate struct for per-CPU idle stats */ |
2847 | sched->max_cpu = session->header.env.nr_cpus_online; | |
2848 | if (sched->max_cpu == 0) | |
2849 | sched->max_cpu = 4; | |
2850 | if (init_idle_threads(sched->max_cpu)) | |
2851 | goto out; | |
2852 | ||
52df138c DA |
2853 | /* summary_only implies summary option, but don't overwrite summary if set */ |
2854 | if (sched->summary_only) | |
2855 | sched->summary = sched->summary_only; | |
2856 | ||
2857 | if (!sched->summary_only) | |
a407b067 | 2858 | timehist_header(sched); |
49394a2a DA |
2859 | |
2860 | err = perf_session__process_events(session); | |
2861 | if (err) { | |
2862 | pr_err("Failed to process events, error %d", err); | |
2863 | goto out; | |
2864 | } | |
2865 | ||
52df138c DA |
2866 | sched->nr_events = evlist->stats.nr_events[0]; |
2867 | sched->nr_lost_events = evlist->stats.total_lost; | |
2868 | sched->nr_lost_chunks = evlist->stats.nr_events[PERF_RECORD_LOST]; | |
2869 | ||
2870 | if (sched->summary) | |
2871 | timehist_print_summary(sched, session); | |
2872 | ||
49394a2a DA |
2873 | out: |
2874 | free_idle_threads(); | |
2875 | perf_session__delete(session); | |
2876 | ||
2877 | return err; | |
2878 | } | |
2879 | ||
2880 | ||
0e9b07e5 | 2881 | static void print_bad_events(struct perf_sched *sched) |
0ec04e16 | 2882 | { |
0e9b07e5 | 2883 | if (sched->nr_unordered_timestamps && sched->nr_timestamps) { |
0ec04e16 | 2884 | printf(" INFO: %.3f%% unordered timestamps (%ld out of %ld)\n", |
0e9b07e5 ACM |
2885 | (double)sched->nr_unordered_timestamps/(double)sched->nr_timestamps*100.0, |
2886 | sched->nr_unordered_timestamps, sched->nr_timestamps); | |
0ec04e16 | 2887 | } |
0e9b07e5 | 2888 | if (sched->nr_lost_events && sched->nr_events) { |
0ec04e16 | 2889 | printf(" INFO: %.3f%% lost events (%ld out of %ld, in %ld chunks)\n", |
0e9b07e5 ACM |
2890 | (double)sched->nr_lost_events/(double)sched->nr_events * 100.0, |
2891 | sched->nr_lost_events, sched->nr_events, sched->nr_lost_chunks); | |
0ec04e16 | 2892 | } |
0e9b07e5 | 2893 | if (sched->nr_context_switch_bugs && sched->nr_timestamps) { |
0ec04e16 | 2894 | printf(" INFO: %.3f%% context switch bugs (%ld out of %ld)", |
0e9b07e5 ACM |
2895 | (double)sched->nr_context_switch_bugs/(double)sched->nr_timestamps*100.0, |
2896 | sched->nr_context_switch_bugs, sched->nr_timestamps); | |
2897 | if (sched->nr_lost_events) | |
0ec04e16 IM |
2898 | printf(" (due to lost events?)"); |
2899 | printf("\n"); | |
2900 | } | |
2901 | } | |
2902 | ||
2f80dd44 JB |
2903 | static void __merge_work_atoms(struct rb_root *root, struct work_atoms *data) |
2904 | { | |
2905 | struct rb_node **new = &(root->rb_node), *parent = NULL; | |
2906 | struct work_atoms *this; | |
2907 | const char *comm = thread__comm_str(data->thread), *this_comm; | |
2908 | ||
2909 | while (*new) { | |
2910 | int cmp; | |
2911 | ||
2912 | this = container_of(*new, struct work_atoms, node); | |
2913 | parent = *new; | |
2914 | ||
2915 | this_comm = thread__comm_str(this->thread); | |
2916 | cmp = strcmp(comm, this_comm); | |
2917 | if (cmp > 0) { | |
2918 | new = &((*new)->rb_left); | |
2919 | } else if (cmp < 0) { | |
2920 | new = &((*new)->rb_right); | |
2921 | } else { | |
2922 | this->num_merged++; | |
2923 | this->total_runtime += data->total_runtime; | |
2924 | this->nb_atoms += data->nb_atoms; | |
2925 | this->total_lat += data->total_lat; | |
2926 | list_splice(&data->work_list, &this->work_list); | |
2927 | if (this->max_lat < data->max_lat) { | |
2928 | this->max_lat = data->max_lat; | |
2929 | this->max_lat_at = data->max_lat_at; | |
2930 | } | |
2931 | zfree(&data); | |
2932 | return; | |
2933 | } | |
2934 | } | |
2935 | ||
2936 | data->num_merged++; | |
2937 | rb_link_node(&data->node, parent, new); | |
2938 | rb_insert_color(&data->node, root); | |
2939 | } | |
2940 | ||
2941 | static void perf_sched__merge_lat(struct perf_sched *sched) | |
2942 | { | |
2943 | struct work_atoms *data; | |
2944 | struct rb_node *node; | |
2945 | ||
2946 | if (sched->skip_merge) | |
2947 | return; | |
2948 | ||
2949 | while ((node = rb_first(&sched->atom_root))) { | |
2950 | rb_erase(node, &sched->atom_root); | |
2951 | data = rb_entry(node, struct work_atoms, node); | |
2952 | __merge_work_atoms(&sched->merged_atom_root, data); | |
2953 | } | |
2954 | } | |
2955 | ||
0e9b07e5 | 2956 | static int perf_sched__lat(struct perf_sched *sched) |
0ec04e16 IM |
2957 | { |
2958 | struct rb_node *next; | |
2959 | ||
2960 | setup_pager(); | |
ad9def7c | 2961 | |
ae536acf | 2962 | if (perf_sched__read_events(sched)) |
a116e05d | 2963 | return -1; |
ad9def7c | 2964 | |
2f80dd44 | 2965 | perf_sched__merge_lat(sched); |
0e9b07e5 | 2966 | perf_sched__sort_lat(sched); |
0ec04e16 | 2967 | |
80790e0b RR |
2968 | printf("\n -----------------------------------------------------------------------------------------------------------------\n"); |
2969 | printf(" Task | Runtime ms | Switches | Average delay ms | Maximum delay ms | Maximum delay at |\n"); | |
2970 | printf(" -----------------------------------------------------------------------------------------------------------------\n"); | |
0ec04e16 | 2971 | |
0e9b07e5 | 2972 | next = rb_first(&sched->sorted_atom_root); |
0ec04e16 IM |
2973 | |
2974 | while (next) { | |
2975 | struct work_atoms *work_list; | |
2976 | ||
2977 | work_list = rb_entry(next, struct work_atoms, node); | |
0e9b07e5 | 2978 | output_lat_thread(sched, work_list); |
0ec04e16 | 2979 | next = rb_next(next); |
ae536acf | 2980 | thread__zput(work_list->thread); |
0ec04e16 IM |
2981 | } |
2982 | ||
80790e0b | 2983 | printf(" -----------------------------------------------------------------------------------------------------------------\n"); |
9486aa38 | 2984 | printf(" TOTAL: |%11.3f ms |%9" PRIu64 " |\n", |
4fc76e49 | 2985 | (double)sched->all_runtime / NSEC_PER_MSEC, sched->all_count); |
0ec04e16 IM |
2986 | |
2987 | printf(" ---------------------------------------------------\n"); | |
2988 | ||
0e9b07e5 | 2989 | print_bad_events(sched); |
0ec04e16 IM |
2990 | printf("\n"); |
2991 | ||
a116e05d | 2992 | return 0; |
0ec04e16 IM |
2993 | } |
2994 | ||
99623c62 JO |
2995 | static int setup_map_cpus(struct perf_sched *sched) |
2996 | { | |
73643bb6 JO |
2997 | struct cpu_map *map; |
2998 | ||
99623c62 JO |
2999 | sched->max_cpu = sysconf(_SC_NPROCESSORS_CONF); |
3000 | ||
3001 | if (sched->map.comp) { | |
3002 | sched->map.comp_cpus = zalloc(sched->max_cpu * sizeof(int)); | |
cf294f24 JO |
3003 | if (!sched->map.comp_cpus) |
3004 | return -1; | |
99623c62 JO |
3005 | } |
3006 | ||
73643bb6 JO |
3007 | if (!sched->map.cpus_str) |
3008 | return 0; | |
3009 | ||
3010 | map = cpu_map__new(sched->map.cpus_str); | |
3011 | if (!map) { | |
3012 | pr_err("failed to get cpus map from %s\n", sched->map.cpus_str); | |
3013 | return -1; | |
3014 | } | |
3015 | ||
3016 | sched->map.cpus = map; | |
99623c62 JO |
3017 | return 0; |
3018 | } | |
3019 | ||
a151a37a JO |
3020 | static int setup_color_pids(struct perf_sched *sched) |
3021 | { | |
3022 | struct thread_map *map; | |
3023 | ||
3024 | if (!sched->map.color_pids_str) | |
3025 | return 0; | |
3026 | ||
3027 | map = thread_map__new_by_tid_str(sched->map.color_pids_str); | |
3028 | if (!map) { | |
3029 | pr_err("failed to get thread map from %s\n", sched->map.color_pids_str); | |
3030 | return -1; | |
3031 | } | |
3032 | ||
3033 | sched->map.color_pids = map; | |
3034 | return 0; | |
3035 | } | |
3036 | ||
cf294f24 JO |
3037 | static int setup_color_cpus(struct perf_sched *sched) |
3038 | { | |
3039 | struct cpu_map *map; | |
3040 | ||
3041 | if (!sched->map.color_cpus_str) | |
3042 | return 0; | |
3043 | ||
3044 | map = cpu_map__new(sched->map.color_cpus_str); | |
3045 | if (!map) { | |
3046 | pr_err("failed to get thread map from %s\n", sched->map.color_cpus_str); | |
3047 | return -1; | |
3048 | } | |
3049 | ||
3050 | sched->map.color_cpus = map; | |
3051 | return 0; | |
3052 | } | |
3053 | ||
0e9b07e5 | 3054 | static int perf_sched__map(struct perf_sched *sched) |
0ec04e16 | 3055 | { |
99623c62 JO |
3056 | if (setup_map_cpus(sched)) |
3057 | return -1; | |
40749d0f | 3058 | |
a151a37a JO |
3059 | if (setup_color_pids(sched)) |
3060 | return -1; | |
3061 | ||
cf294f24 JO |
3062 | if (setup_color_cpus(sched)) |
3063 | return -1; | |
3064 | ||
0ec04e16 | 3065 | setup_pager(); |
ae536acf | 3066 | if (perf_sched__read_events(sched)) |
a116e05d | 3067 | return -1; |
0e9b07e5 | 3068 | print_bad_events(sched); |
a116e05d | 3069 | return 0; |
0ec04e16 IM |
3070 | } |
3071 | ||
0e9b07e5 | 3072 | static int perf_sched__replay(struct perf_sched *sched) |
0ec04e16 IM |
3073 | { |
3074 | unsigned long i; | |
3075 | ||
0e9b07e5 ACM |
3076 | calibrate_run_measurement_overhead(sched); |
3077 | calibrate_sleep_measurement_overhead(sched); | |
0ec04e16 | 3078 | |
0e9b07e5 | 3079 | test_calibrations(sched); |
0ec04e16 | 3080 | |
ae536acf | 3081 | if (perf_sched__read_events(sched)) |
a116e05d | 3082 | return -1; |
0ec04e16 | 3083 | |
0e9b07e5 ACM |
3084 | printf("nr_run_events: %ld\n", sched->nr_run_events); |
3085 | printf("nr_sleep_events: %ld\n", sched->nr_sleep_events); | |
3086 | printf("nr_wakeup_events: %ld\n", sched->nr_wakeup_events); | |
0ec04e16 | 3087 | |
0e9b07e5 ACM |
3088 | if (sched->targetless_wakeups) |
3089 | printf("target-less wakeups: %ld\n", sched->targetless_wakeups); | |
3090 | if (sched->multitarget_wakeups) | |
3091 | printf("multi-target wakeups: %ld\n", sched->multitarget_wakeups); | |
3092 | if (sched->nr_run_events_optimized) | |
0ec04e16 | 3093 | printf("run atoms optimized: %ld\n", |
0e9b07e5 | 3094 | sched->nr_run_events_optimized); |
0ec04e16 | 3095 | |
0e9b07e5 ACM |
3096 | print_task_traces(sched); |
3097 | add_cross_task_wakeups(sched); | |
0ec04e16 | 3098 | |
0e9b07e5 | 3099 | create_tasks(sched); |
0ec04e16 | 3100 | printf("------------------------------------------------------------\n"); |
0e9b07e5 ACM |
3101 | for (i = 0; i < sched->replay_repeat; i++) |
3102 | run_one_test(sched); | |
a116e05d ACM |
3103 | |
3104 | return 0; | |
0ec04e16 IM |
3105 | } |
3106 | ||
0e9b07e5 ACM |
3107 | static void setup_sorting(struct perf_sched *sched, const struct option *options, |
3108 | const char * const usage_msg[]) | |
daa1d7a5 | 3109 | { |
0e9b07e5 | 3110 | char *tmp, *tok, *str = strdup(sched->sort_order); |
daa1d7a5 FW |
3111 | |
3112 | for (tok = strtok_r(str, ", ", &tmp); | |
3113 | tok; tok = strtok_r(NULL, ", ", &tmp)) { | |
0e9b07e5 | 3114 | if (sort_dimension__add(tok, &sched->sort_list) < 0) { |
c7118369 NK |
3115 | usage_with_options_msg(usage_msg, options, |
3116 | "Unknown --sort key: `%s'", tok); | |
daa1d7a5 FW |
3117 | } |
3118 | } | |
3119 | ||
3120 | free(str); | |
3121 | ||
0e9b07e5 | 3122 | sort_dimension__add("pid", &sched->cmp_pid); |
daa1d7a5 FW |
3123 | } |
3124 | ||
1fc35b29 IM |
3125 | static int __cmd_record(int argc, const char **argv) |
3126 | { | |
3127 | unsigned int rec_argc, i, j; | |
3128 | const char **rec_argv; | |
0e9b07e5 ACM |
3129 | const char * const record_args[] = { |
3130 | "record", | |
3131 | "-a", | |
3132 | "-R", | |
0e9b07e5 ACM |
3133 | "-m", "1024", |
3134 | "-c", "1", | |
3135 | "-e", "sched:sched_switch", | |
3136 | "-e", "sched:sched_stat_wait", | |
3137 | "-e", "sched:sched_stat_sleep", | |
3138 | "-e", "sched:sched_stat_iowait", | |
3139 | "-e", "sched:sched_stat_runtime", | |
0e9b07e5 ACM |
3140 | "-e", "sched:sched_process_fork", |
3141 | "-e", "sched:sched_wakeup", | |
7fff9597 | 3142 | "-e", "sched:sched_wakeup_new", |
0e9b07e5 ACM |
3143 | "-e", "sched:sched_migrate_task", |
3144 | }; | |
1fc35b29 IM |
3145 | |
3146 | rec_argc = ARRAY_SIZE(record_args) + argc - 1; | |
3147 | rec_argv = calloc(rec_argc + 1, sizeof(char *)); | |
3148 | ||
e462dc55 | 3149 | if (rec_argv == NULL) |
ce47dc56 CS |
3150 | return -ENOMEM; |
3151 | ||
1fc35b29 IM |
3152 | for (i = 0; i < ARRAY_SIZE(record_args); i++) |
3153 | rec_argv[i] = strdup(record_args[i]); | |
3154 | ||
3155 | for (j = 1; j < (unsigned int)argc; j++, i++) | |
3156 | rec_argv[i] = argv[j]; | |
3157 | ||
3158 | BUG_ON(i != rec_argc); | |
3159 | ||
3160 | return cmd_record(i, rec_argv, NULL); | |
3161 | } | |
3162 | ||
1d037ca1 | 3163 | int cmd_sched(int argc, const char **argv, const char *prefix __maybe_unused) |
0a02ad93 | 3164 | { |
8a39df8f AH |
3165 | const char default_sort_order[] = "avg, max, switch, runtime"; |
3166 | struct perf_sched sched = { | |
3167 | .tool = { | |
3168 | .sample = perf_sched__process_tracepoint_sample, | |
3169 | .comm = perf_event__process_comm, | |
3170 | .lost = perf_event__process_lost, | |
3171 | .fork = perf_sched__process_fork_event, | |
0a8cb85c | 3172 | .ordered_events = true, |
8a39df8f AH |
3173 | }, |
3174 | .cmp_pid = LIST_HEAD_INIT(sched.cmp_pid), | |
3175 | .sort_list = LIST_HEAD_INIT(sched.sort_list), | |
3176 | .start_work_mutex = PTHREAD_MUTEX_INITIALIZER, | |
3177 | .work_done_wait_mutex = PTHREAD_MUTEX_INITIALIZER, | |
8a39df8f AH |
3178 | .sort_order = default_sort_order, |
3179 | .replay_repeat = 10, | |
3180 | .profile_cpu = -1, | |
3181 | .next_shortname1 = 'A', | |
3182 | .next_shortname2 = '0', | |
2f80dd44 | 3183 | .skip_merge = 0, |
6c973c90 DA |
3184 | .show_callchain = 1, |
3185 | .max_stack = 5, | |
8a39df8f | 3186 | }; |
77f02f44 NK |
3187 | const struct option sched_options[] = { |
3188 | OPT_STRING('i', "input", &input_name, "file", | |
3189 | "input file name"), | |
3190 | OPT_INCR('v', "verbose", &verbose, | |
3191 | "be more verbose (show symbol address, etc)"), | |
3192 | OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, | |
3193 | "dump raw trace in ASCII"), | |
6fa94258 | 3194 | OPT_BOOLEAN('f', "force", &sched.force, "don't complain, do it"), |
77f02f44 NK |
3195 | OPT_END() |
3196 | }; | |
0e9b07e5 ACM |
3197 | const struct option latency_options[] = { |
3198 | OPT_STRING('s', "sort", &sched.sort_order, "key[,key2...]", | |
3199 | "sort by key(s): runtime, switch, avg, max"), | |
0e9b07e5 ACM |
3200 | OPT_INTEGER('C', "CPU", &sched.profile_cpu, |
3201 | "CPU to profile on"), | |
2f80dd44 JB |
3202 | OPT_BOOLEAN('p', "pids", &sched.skip_merge, |
3203 | "latency stats per pid instead of per comm"), | |
77f02f44 | 3204 | OPT_PARENT(sched_options) |
0e9b07e5 ACM |
3205 | }; |
3206 | const struct option replay_options[] = { | |
3207 | OPT_UINTEGER('r', "repeat", &sched.replay_repeat, | |
3208 | "repeat the workload replay N times (-1: infinite)"), | |
77f02f44 | 3209 | OPT_PARENT(sched_options) |
0e9b07e5 | 3210 | }; |
99623c62 JO |
3211 | const struct option map_options[] = { |
3212 | OPT_BOOLEAN(0, "compact", &sched.map.comp, | |
3213 | "map output in compact mode"), | |
a151a37a JO |
3214 | OPT_STRING(0, "color-pids", &sched.map.color_pids_str, "pids", |
3215 | "highlight given pids in map"), | |
cf294f24 JO |
3216 | OPT_STRING(0, "color-cpus", &sched.map.color_cpus_str, "cpus", |
3217 | "highlight given CPUs in map"), | |
73643bb6 JO |
3218 | OPT_STRING(0, "cpus", &sched.map.cpus_str, "cpus", |
3219 | "display given CPUs in map"), | |
77f02f44 | 3220 | OPT_PARENT(sched_options) |
99623c62 | 3221 | }; |
49394a2a DA |
3222 | const struct option timehist_options[] = { |
3223 | OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name, | |
3224 | "file", "vmlinux pathname"), | |
3225 | OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name, | |
3226 | "file", "kallsyms pathname"), | |
6c973c90 DA |
3227 | OPT_BOOLEAN('g', "call-graph", &sched.show_callchain, |
3228 | "Display call chains if present (default on)"), | |
3229 | OPT_UINTEGER(0, "max-stack", &sched.max_stack, | |
3230 | "Maximum number of functions to display backtrace."), | |
49394a2a DA |
3231 | OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory", |
3232 | "Look for files with symbols relative to this directory"), | |
52df138c DA |
3233 | OPT_BOOLEAN('s', "summary", &sched.summary_only, |
3234 | "Show only syscall summary with statistics"), | |
3235 | OPT_BOOLEAN('S', "with-summary", &sched.summary, | |
3236 | "Show all syscalls and summary with statistics"), | |
fc1469f1 | 3237 | OPT_BOOLEAN('w', "wakeups", &sched.show_wakeups, "Show wakeup events"), |
350f54fa | 3238 | OPT_BOOLEAN('M', "migrations", &sched.show_migrations, "Show migration events"), |
a407b067 | 3239 | OPT_BOOLEAN('V', "cpu-visual", &sched.show_cpu_visual, "Add CPU visual"), |
07235f84 | 3240 | OPT_BOOLEAN('I', "idle-hist", &sched.idle_hist, "Show idle events only"), |
853b7407 DA |
3241 | OPT_STRING(0, "time", &sched.time_str, "str", |
3242 | "Time span for analysis (start,stop)"), | |
49394a2a DA |
3243 | OPT_PARENT(sched_options) |
3244 | }; | |
3245 | ||
0e9b07e5 ACM |
3246 | const char * const latency_usage[] = { |
3247 | "perf sched latency [<options>]", | |
3248 | NULL | |
3249 | }; | |
3250 | const char * const replay_usage[] = { | |
3251 | "perf sched replay [<options>]", | |
3252 | NULL | |
3253 | }; | |
99623c62 JO |
3254 | const char * const map_usage[] = { |
3255 | "perf sched map [<options>]", | |
3256 | NULL | |
3257 | }; | |
49394a2a DA |
3258 | const char * const timehist_usage[] = { |
3259 | "perf sched timehist [<options>]", | |
3260 | NULL | |
3261 | }; | |
a83edb2d | 3262 | const char *const sched_subcommands[] = { "record", "latency", "map", |
49394a2a DA |
3263 | "replay", "script", |
3264 | "timehist", NULL }; | |
a83edb2d RR |
3265 | const char *sched_usage[] = { |
3266 | NULL, | |
0e9b07e5 ACM |
3267 | NULL |
3268 | }; | |
3269 | struct trace_sched_handler lat_ops = { | |
3270 | .wakeup_event = latency_wakeup_event, | |
3271 | .switch_event = latency_switch_event, | |
3272 | .runtime_event = latency_runtime_event, | |
0e9b07e5 ACM |
3273 | .migrate_task_event = latency_migrate_task_event, |
3274 | }; | |
3275 | struct trace_sched_handler map_ops = { | |
3276 | .switch_event = map_switch_event, | |
3277 | }; | |
3278 | struct trace_sched_handler replay_ops = { | |
3279 | .wakeup_event = replay_wakeup_event, | |
3280 | .switch_event = replay_switch_event, | |
3281 | .fork_event = replay_fork_event, | |
3282 | }; | |
156a2b02 AH |
3283 | unsigned int i; |
3284 | ||
3285 | for (i = 0; i < ARRAY_SIZE(sched.curr_pid); i++) | |
3286 | sched.curr_pid[i] = -1; | |
0e9b07e5 | 3287 | |
a83edb2d RR |
3288 | argc = parse_options_subcommand(argc, argv, sched_options, sched_subcommands, |
3289 | sched_usage, PARSE_OPT_STOP_AT_NON_OPTION); | |
f2858d8a IM |
3290 | if (!argc) |
3291 | usage_with_options(sched_usage, sched_options); | |
0a02ad93 | 3292 | |
c0777c5a | 3293 | /* |
133dc4c3 | 3294 | * Aliased to 'perf script' for now: |
c0777c5a | 3295 | */ |
133dc4c3 IM |
3296 | if (!strcmp(argv[0], "script")) |
3297 | return cmd_script(argc, argv, prefix); | |
c0777c5a | 3298 | |
1fc35b29 IM |
3299 | if (!strncmp(argv[0], "rec", 3)) { |
3300 | return __cmd_record(argc, argv); | |
3301 | } else if (!strncmp(argv[0], "lat", 3)) { | |
0e9b07e5 | 3302 | sched.tp_handler = &lat_ops; |
f2858d8a IM |
3303 | if (argc > 1) { |
3304 | argc = parse_options(argc, argv, latency_options, latency_usage, 0); | |
3305 | if (argc) | |
3306 | usage_with_options(latency_usage, latency_options); | |
f2858d8a | 3307 | } |
0e9b07e5 ACM |
3308 | setup_sorting(&sched, latency_options, latency_usage); |
3309 | return perf_sched__lat(&sched); | |
0ec04e16 | 3310 | } else if (!strcmp(argv[0], "map")) { |
99623c62 | 3311 | if (argc) { |
a151a37a | 3312 | argc = parse_options(argc, argv, map_options, map_usage, 0); |
99623c62 JO |
3313 | if (argc) |
3314 | usage_with_options(map_usage, map_options); | |
3315 | } | |
0e9b07e5 ACM |
3316 | sched.tp_handler = &map_ops; |
3317 | setup_sorting(&sched, latency_options, latency_usage); | |
3318 | return perf_sched__map(&sched); | |
f2858d8a | 3319 | } else if (!strncmp(argv[0], "rep", 3)) { |
0e9b07e5 | 3320 | sched.tp_handler = &replay_ops; |
f2858d8a IM |
3321 | if (argc) { |
3322 | argc = parse_options(argc, argv, replay_options, replay_usage, 0); | |
3323 | if (argc) | |
3324 | usage_with_options(replay_usage, replay_options); | |
3325 | } | |
0e9b07e5 | 3326 | return perf_sched__replay(&sched); |
49394a2a DA |
3327 | } else if (!strcmp(argv[0], "timehist")) { |
3328 | if (argc) { | |
3329 | argc = parse_options(argc, argv, timehist_options, | |
3330 | timehist_usage, 0); | |
3331 | if (argc) | |
3332 | usage_with_options(timehist_usage, timehist_options); | |
3333 | } | |
fc1469f1 DA |
3334 | if (sched.show_wakeups && sched.summary_only) { |
3335 | pr_err(" Error: -s and -w are mutually exclusive.\n"); | |
3336 | parse_options_usage(timehist_usage, timehist_options, "s", true); | |
3337 | parse_options_usage(NULL, timehist_options, "w", true); | |
3338 | return -EINVAL; | |
3339 | } | |
3340 | ||
49394a2a | 3341 | return perf_sched__timehist(&sched); |
f2858d8a IM |
3342 | } else { |
3343 | usage_with_options(sched_usage, sched_options); | |
3344 | } | |
3345 | ||
ec156764 | 3346 | return 0; |
0a02ad93 | 3347 | } |