]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
0a02ad93 | 2 | #include "builtin.h" |
b1ffe8f3 | 3 | #include "perf.h" |
91854f9a | 4 | #include "perf-sys.h" |
0a02ad93 | 5 | |
87ffb6c6 | 6 | #include "util/cpumap.h" |
ee29be62 | 7 | #include "util/evlist.h" |
e3f42609 | 8 | #include "util/evsel.h" |
ca125277 | 9 | #include "util/evsel_fprintf.h" |
0a02ad93 IM |
10 | #include "util/symbol.h" |
11 | #include "util/thread.h" | |
12 | #include "util/header.h" | |
94c744b6 | 13 | #include "util/session.h" |
45694aa7 | 14 | #include "util/tool.h" |
57480d2c | 15 | #include "util/cloexec.h" |
a151a37a | 16 | #include "util/thread_map.h" |
8cd91195 | 17 | #include "util/color.h" |
49394a2a | 18 | #include "util/stat.h" |
6a9fa4e3 | 19 | #include "util/string2.h" |
6c973c90 | 20 | #include "util/callchain.h" |
853b7407 | 21 | #include "util/time-utils.h" |
0a02ad93 | 22 | |
fa0d9846 | 23 | #include <subcmd/pager.h> |
4b6ab94e | 24 | #include <subcmd/parse-options.h> |
b1ffe8f3 | 25 | #include "util/trace-event.h" |
0a02ad93 | 26 | |
0a02ad93 | 27 | #include "util/debug.h" |
f12be047 | 28 | #include "util/event.h" |
0a02ad93 | 29 | |
877a7a11 | 30 | #include <linux/kernel.h> |
49394a2a | 31 | #include <linux/log2.h> |
7f7c536f | 32 | #include <linux/zalloc.h> |
b1ffe8f3 | 33 | #include <sys/prctl.h> |
7b78f136 | 34 | #include <sys/resource.h> |
fd20e811 | 35 | #include <inttypes.h> |
0a02ad93 | 36 | |
a43783ae | 37 | #include <errno.h> |
b1ffe8f3 IM |
38 | #include <semaphore.h> |
39 | #include <pthread.h> | |
40 | #include <math.h> | |
cb06ac25 | 41 | #include <api/fs/fs.h> |
87ffb6c6 | 42 | #include <perf/cpumap.h> |
4fc76e49 | 43 | #include <linux/time64.h> |
6ef81c55 | 44 | #include <linux/err.h> |
419ab0d6 | 45 | |
3052ba56 | 46 | #include <linux/ctype.h> |
3d689ed6 | 47 | |
b1ffe8f3 IM |
48 | #define PR_SET_NAME 15 /* Set process name */ |
49 | #define MAX_CPUS 4096 | |
b1ffe8f3 IM |
50 | #define COMM_LEN 20 |
51 | #define SYM_LEN 129 | |
a35e27d0 | 52 | #define MAX_PID 1024000 |
ec156764 | 53 | |
39aeb52f | 54 | struct sched_atom; |
ec156764 | 55 | |
b1ffe8f3 IM |
56 | struct task_desc { |
57 | unsigned long nr; | |
58 | unsigned long pid; | |
59 | char comm[COMM_LEN]; | |
ec156764 | 60 | |
b1ffe8f3 IM |
61 | unsigned long nr_events; |
62 | unsigned long curr_event; | |
39aeb52f | 63 | struct sched_atom **atoms; |
b1ffe8f3 IM |
64 | |
65 | pthread_t thread; | |
66 | sem_t sleep_sem; | |
ec156764 | 67 | |
b1ffe8f3 IM |
68 | sem_t ready_for_work; |
69 | sem_t work_done_sem; | |
70 | ||
71 | u64 cpu_usage; | |
72 | }; | |
73 | ||
74 | enum sched_event_type { | |
75 | SCHED_EVENT_RUN, | |
76 | SCHED_EVENT_SLEEP, | |
77 | SCHED_EVENT_WAKEUP, | |
55ffb7a6 | 78 | SCHED_EVENT_MIGRATION, |
b1ffe8f3 IM |
79 | }; |
80 | ||
39aeb52f | 81 | struct sched_atom { |
b1ffe8f3 | 82 | enum sched_event_type type; |
eed05fe7 | 83 | int specific_wait; |
b1ffe8f3 IM |
84 | u64 timestamp; |
85 | u64 duration; | |
86 | unsigned long nr; | |
b1ffe8f3 IM |
87 | sem_t *wait_sem; |
88 | struct task_desc *wakee; | |
89 | }; | |
90 | ||
e936e8e4 | 91 | #define TASK_STATE_TO_CHAR_STR "RSDTtZXxKWP" |
b1ffe8f3 | 92 | |
941bdea7 NK |
93 | /* task state bitmask, copied from include/linux/sched.h */ |
94 | #define TASK_RUNNING 0 | |
95 | #define TASK_INTERRUPTIBLE 1 | |
96 | #define TASK_UNINTERRUPTIBLE 2 | |
97 | #define __TASK_STOPPED 4 | |
98 | #define __TASK_TRACED 8 | |
99 | /* in tsk->exit_state */ | |
100 | #define EXIT_DEAD 16 | |
101 | #define EXIT_ZOMBIE 32 | |
102 | #define EXIT_TRACE (EXIT_ZOMBIE | EXIT_DEAD) | |
103 | /* in tsk->state again */ | |
104 | #define TASK_DEAD 64 | |
105 | #define TASK_WAKEKILL 128 | |
106 | #define TASK_WAKING 256 | |
107 | #define TASK_PARKED 512 | |
108 | ||
b1ffe8f3 IM |
109 | enum thread_state { |
110 | THREAD_SLEEPING = 0, | |
111 | THREAD_WAIT_CPU, | |
112 | THREAD_SCHED_IN, | |
113 | THREAD_IGNORE | |
114 | }; | |
115 | ||
116 | struct work_atom { | |
117 | struct list_head list; | |
118 | enum thread_state state; | |
aa1ab9d2 | 119 | u64 sched_out_time; |
b1ffe8f3 IM |
120 | u64 wake_up_time; |
121 | u64 sched_in_time; | |
122 | u64 runtime; | |
123 | }; | |
124 | ||
39aeb52f | 125 | struct work_atoms { |
126 | struct list_head work_list; | |
b1ffe8f3 IM |
127 | struct thread *thread; |
128 | struct rb_node node; | |
129 | u64 max_lat; | |
3786310a | 130 | u64 max_lat_at; |
b1ffe8f3 IM |
131 | u64 total_lat; |
132 | u64 nb_atoms; | |
133 | u64 total_runtime; | |
2f80dd44 | 134 | int num_merged; |
b1ffe8f3 IM |
135 | }; |
136 | ||
39aeb52f | 137 | typedef int (*sort_fn_t)(struct work_atoms *, struct work_atoms *); |
b1ffe8f3 | 138 | |
9ec3f4e4 | 139 | struct perf_sched; |
0e9b07e5 | 140 | |
9ec3f4e4 | 141 | struct trace_sched_handler { |
32dcd021 | 142 | int (*switch_event)(struct perf_sched *sched, struct evsel *evsel, |
9ec3f4e4 | 143 | struct perf_sample *sample, struct machine *machine); |
0e9b07e5 | 144 | |
32dcd021 | 145 | int (*runtime_event)(struct perf_sched *sched, struct evsel *evsel, |
9ec3f4e4 | 146 | struct perf_sample *sample, struct machine *machine); |
0e9b07e5 | 147 | |
32dcd021 | 148 | int (*wakeup_event)(struct perf_sched *sched, struct evsel *evsel, |
9ec3f4e4 | 149 | struct perf_sample *sample, struct machine *machine); |
0e9b07e5 | 150 | |
cb627505 DA |
151 | /* PERF_RECORD_FORK event, not sched_process_fork tracepoint */ |
152 | int (*fork_event)(struct perf_sched *sched, union perf_event *event, | |
153 | struct machine *machine); | |
0e9b07e5 ACM |
154 | |
155 | int (*migrate_task_event)(struct perf_sched *sched, | |
32dcd021 | 156 | struct evsel *evsel, |
9ec3f4e4 ACM |
157 | struct perf_sample *sample, |
158 | struct machine *machine); | |
0e9b07e5 ACM |
159 | }; |
160 | ||
a151a37a | 161 | #define COLOR_PIDS PERF_COLOR_BLUE |
cf294f24 | 162 | #define COLOR_CPUS PERF_COLOR_BG_RED |
a151a37a | 163 | |
99623c62 JO |
164 | struct perf_sched_map { |
165 | DECLARE_BITMAP(comp_cpus_mask, MAX_CPUS); | |
166 | int *comp_cpus; | |
167 | bool comp; | |
9749b90e | 168 | struct perf_thread_map *color_pids; |
a151a37a | 169 | const char *color_pids_str; |
f854839b | 170 | struct perf_cpu_map *color_cpus; |
cf294f24 | 171 | const char *color_cpus_str; |
f854839b | 172 | struct perf_cpu_map *cpus; |
73643bb6 | 173 | const char *cpus_str; |
99623c62 JO |
174 | }; |
175 | ||
0e9b07e5 ACM |
176 | struct perf_sched { |
177 | struct perf_tool tool; | |
0e9b07e5 ACM |
178 | const char *sort_order; |
179 | unsigned long nr_tasks; | |
cb06ac25 | 180 | struct task_desc **pid_to_task; |
0e9b07e5 ACM |
181 | struct task_desc **tasks; |
182 | const struct trace_sched_handler *tp_handler; | |
183 | pthread_mutex_t start_work_mutex; | |
184 | pthread_mutex_t work_done_wait_mutex; | |
185 | int profile_cpu; | |
186 | /* | |
187 | * Track the current task - that way we can know whether there's any | |
188 | * weird events, such as a task being switched away that is not current. | |
189 | */ | |
190 | int max_cpu; | |
191 | u32 curr_pid[MAX_CPUS]; | |
192 | struct thread *curr_thread[MAX_CPUS]; | |
193 | char next_shortname1; | |
194 | char next_shortname2; | |
195 | unsigned int replay_repeat; | |
196 | unsigned long nr_run_events; | |
197 | unsigned long nr_sleep_events; | |
198 | unsigned long nr_wakeup_events; | |
199 | unsigned long nr_sleep_corrections; | |
200 | unsigned long nr_run_events_optimized; | |
201 | unsigned long targetless_wakeups; | |
202 | unsigned long multitarget_wakeups; | |
203 | unsigned long nr_runs; | |
204 | unsigned long nr_timestamps; | |
205 | unsigned long nr_unordered_timestamps; | |
0e9b07e5 ACM |
206 | unsigned long nr_context_switch_bugs; |
207 | unsigned long nr_events; | |
208 | unsigned long nr_lost_chunks; | |
209 | unsigned long nr_lost_events; | |
210 | u64 run_measurement_overhead; | |
211 | u64 sleep_measurement_overhead; | |
212 | u64 start_time; | |
213 | u64 cpu_usage; | |
214 | u64 runavg_cpu_usage; | |
215 | u64 parent_cpu_usage; | |
216 | u64 runavg_parent_cpu_usage; | |
217 | u64 sum_runtime; | |
218 | u64 sum_fluct; | |
219 | u64 run_avg; | |
220 | u64 all_runtime; | |
221 | u64 all_count; | |
222 | u64 cpu_last_switched[MAX_CPUS]; | |
cb4c13a5 | 223 | struct rb_root_cached atom_root, sorted_atom_root, merged_atom_root; |
0e9b07e5 | 224 | struct list_head sort_list, cmp_pid; |
939cda52 | 225 | bool force; |
2f80dd44 | 226 | bool skip_merge; |
99623c62 | 227 | struct perf_sched_map map; |
52df138c DA |
228 | |
229 | /* options for timehist command */ | |
230 | bool summary; | |
231 | bool summary_only; | |
699b5b92 | 232 | bool idle_hist; |
6c973c90 DA |
233 | bool show_callchain; |
234 | unsigned int max_stack; | |
a407b067 | 235 | bool show_cpu_visual; |
fc1469f1 | 236 | bool show_wakeups; |
292c4a8f | 237 | bool show_next; |
350f54fa | 238 | bool show_migrations; |
414e050c | 239 | bool show_state; |
52df138c | 240 | u64 skipped_samples; |
853b7407 DA |
241 | const char *time_str; |
242 | struct perf_time_interval ptime; | |
9396c9cb | 243 | struct perf_time_interval hist_time; |
0e9b07e5 | 244 | }; |
b1ffe8f3 | 245 | |
49394a2a DA |
246 | /* per thread run time data */ |
247 | struct thread_runtime { | |
248 | u64 last_time; /* time of previous sched in/out event */ | |
249 | u64 dt_run; /* run time */ | |
941bdea7 NK |
250 | u64 dt_sleep; /* time between CPU access by sleep (off cpu) */ |
251 | u64 dt_iowait; /* time between CPU access by iowait (off cpu) */ | |
252 | u64 dt_preempt; /* time between CPU access by preempt (off cpu) */ | |
49394a2a DA |
253 | u64 dt_delay; /* time between wakeup and sched-in */ |
254 | u64 ready_to_run; /* time of wakeup */ | |
255 | ||
256 | struct stats run_stats; | |
257 | u64 total_run_time; | |
587782c5 NK |
258 | u64 total_sleep_time; |
259 | u64 total_iowait_time; | |
260 | u64 total_preempt_time; | |
261 | u64 total_delay_time; | |
350f54fa | 262 | |
941bdea7 | 263 | int last_state; |
8640da9f CD |
264 | |
265 | char shortname[3]; | |
99a3c3a9 CD |
266 | bool comm_changed; |
267 | ||
350f54fa | 268 | u64 migrations; |
49394a2a DA |
269 | }; |
270 | ||
271 | /* per event run time data */ | |
272 | struct evsel_runtime { | |
273 | u64 *last_time; /* time this event was last seen per cpu */ | |
274 | u32 ncpu; /* highest cpu slot allocated */ | |
275 | }; | |
276 | ||
3bc2fa9c NK |
277 | /* per cpu idle time data */ |
278 | struct idle_thread_runtime { | |
279 | struct thread_runtime tr; | |
280 | struct thread *last_thread; | |
cb4c13a5 | 281 | struct rb_root_cached sorted_root; |
3bc2fa9c NK |
282 | struct callchain_root callchain; |
283 | struct callchain_cursor cursor; | |
284 | }; | |
285 | ||
49394a2a DA |
286 | /* track idle times per cpu */ |
287 | static struct thread **idle_threads; | |
288 | static int idle_max_cpu; | |
289 | static char idle_comm[] = "<idle>"; | |
290 | ||
b1ffe8f3 | 291 | static u64 get_nsecs(void) |
ec156764 IM |
292 | { |
293 | struct timespec ts; | |
294 | ||
295 | clock_gettime(CLOCK_MONOTONIC, &ts); | |
296 | ||
4fc76e49 | 297 | return ts.tv_sec * NSEC_PER_SEC + ts.tv_nsec; |
ec156764 IM |
298 | } |
299 | ||
0e9b07e5 | 300 | static void burn_nsecs(struct perf_sched *sched, u64 nsecs) |
ec156764 | 301 | { |
b1ffe8f3 | 302 | u64 T0 = get_nsecs(), T1; |
ec156764 IM |
303 | |
304 | do { | |
305 | T1 = get_nsecs(); | |
0e9b07e5 | 306 | } while (T1 + sched->run_measurement_overhead < T0 + nsecs); |
ec156764 IM |
307 | } |
308 | ||
b1ffe8f3 | 309 | static void sleep_nsecs(u64 nsecs) |
ec156764 IM |
310 | { |
311 | struct timespec ts; | |
312 | ||
313 | ts.tv_nsec = nsecs % 999999999; | |
314 | ts.tv_sec = nsecs / 999999999; | |
315 | ||
316 | nanosleep(&ts, NULL); | |
317 | } | |
318 | ||
0e9b07e5 | 319 | static void calibrate_run_measurement_overhead(struct perf_sched *sched) |
ec156764 | 320 | { |
4fc76e49 | 321 | u64 T0, T1, delta, min_delta = NSEC_PER_SEC; |
ec156764 IM |
322 | int i; |
323 | ||
324 | for (i = 0; i < 10; i++) { | |
325 | T0 = get_nsecs(); | |
0e9b07e5 | 326 | burn_nsecs(sched, 0); |
ec156764 IM |
327 | T1 = get_nsecs(); |
328 | delta = T1-T0; | |
329 | min_delta = min(min_delta, delta); | |
330 | } | |
0e9b07e5 | 331 | sched->run_measurement_overhead = min_delta; |
ec156764 | 332 | |
9486aa38 | 333 | printf("run measurement overhead: %" PRIu64 " nsecs\n", min_delta); |
ec156764 IM |
334 | } |
335 | ||
0e9b07e5 | 336 | static void calibrate_sleep_measurement_overhead(struct perf_sched *sched) |
ec156764 | 337 | { |
4fc76e49 | 338 | u64 T0, T1, delta, min_delta = NSEC_PER_SEC; |
ec156764 IM |
339 | int i; |
340 | ||
341 | for (i = 0; i < 10; i++) { | |
342 | T0 = get_nsecs(); | |
343 | sleep_nsecs(10000); | |
344 | T1 = get_nsecs(); | |
345 | delta = T1-T0; | |
346 | min_delta = min(min_delta, delta); | |
347 | } | |
348 | min_delta -= 10000; | |
0e9b07e5 | 349 | sched->sleep_measurement_overhead = min_delta; |
ec156764 | 350 | |
9486aa38 | 351 | printf("sleep measurement overhead: %" PRIu64 " nsecs\n", min_delta); |
ec156764 IM |
352 | } |
353 | ||
39aeb52f | 354 | static struct sched_atom * |
b1ffe8f3 | 355 | get_new_event(struct task_desc *task, u64 timestamp) |
ec156764 | 356 | { |
36479484 | 357 | struct sched_atom *event = zalloc(sizeof(*event)); |
ec156764 IM |
358 | unsigned long idx = task->nr_events; |
359 | size_t size; | |
360 | ||
361 | event->timestamp = timestamp; | |
362 | event->nr = idx; | |
363 | ||
364 | task->nr_events++; | |
39aeb52f | 365 | size = sizeof(struct sched_atom *) * task->nr_events; |
366 | task->atoms = realloc(task->atoms, size); | |
367 | BUG_ON(!task->atoms); | |
ec156764 | 368 | |
39aeb52f | 369 | task->atoms[idx] = event; |
ec156764 IM |
370 | |
371 | return event; | |
372 | } | |
373 | ||
39aeb52f | 374 | static struct sched_atom *last_event(struct task_desc *task) |
ec156764 IM |
375 | { |
376 | if (!task->nr_events) | |
377 | return NULL; | |
378 | ||
39aeb52f | 379 | return task->atoms[task->nr_events - 1]; |
ec156764 IM |
380 | } |
381 | ||
0e9b07e5 ACM |
382 | static void add_sched_event_run(struct perf_sched *sched, struct task_desc *task, |
383 | u64 timestamp, u64 duration) | |
ec156764 | 384 | { |
39aeb52f | 385 | struct sched_atom *event, *curr_event = last_event(task); |
ec156764 IM |
386 | |
387 | /* | |
fbf94829 IM |
388 | * optimize an existing RUN event by merging this one |
389 | * to it: | |
390 | */ | |
ec156764 | 391 | if (curr_event && curr_event->type == SCHED_EVENT_RUN) { |
0e9b07e5 | 392 | sched->nr_run_events_optimized++; |
ec156764 IM |
393 | curr_event->duration += duration; |
394 | return; | |
395 | } | |
396 | ||
397 | event = get_new_event(task, timestamp); | |
398 | ||
399 | event->type = SCHED_EVENT_RUN; | |
400 | event->duration = duration; | |
401 | ||
0e9b07e5 | 402 | sched->nr_run_events++; |
ec156764 IM |
403 | } |
404 | ||
0e9b07e5 ACM |
405 | static void add_sched_event_wakeup(struct perf_sched *sched, struct task_desc *task, |
406 | u64 timestamp, struct task_desc *wakee) | |
ec156764 | 407 | { |
39aeb52f | 408 | struct sched_atom *event, *wakee_event; |
ec156764 IM |
409 | |
410 | event = get_new_event(task, timestamp); | |
411 | event->type = SCHED_EVENT_WAKEUP; | |
412 | event->wakee = wakee; | |
413 | ||
414 | wakee_event = last_event(wakee); | |
415 | if (!wakee_event || wakee_event->type != SCHED_EVENT_SLEEP) { | |
0e9b07e5 | 416 | sched->targetless_wakeups++; |
ec156764 IM |
417 | return; |
418 | } | |
419 | if (wakee_event->wait_sem) { | |
0e9b07e5 | 420 | sched->multitarget_wakeups++; |
ec156764 IM |
421 | return; |
422 | } | |
423 | ||
36479484 | 424 | wakee_event->wait_sem = zalloc(sizeof(*wakee_event->wait_sem)); |
ec156764 IM |
425 | sem_init(wakee_event->wait_sem, 0, 0); |
426 | wakee_event->specific_wait = 1; | |
427 | event->wait_sem = wakee_event->wait_sem; | |
428 | ||
0e9b07e5 | 429 | sched->nr_wakeup_events++; |
ec156764 IM |
430 | } |
431 | ||
0e9b07e5 ACM |
432 | static void add_sched_event_sleep(struct perf_sched *sched, struct task_desc *task, |
433 | u64 timestamp, u64 task_state __maybe_unused) | |
ec156764 | 434 | { |
39aeb52f | 435 | struct sched_atom *event = get_new_event(task, timestamp); |
ec156764 IM |
436 | |
437 | event->type = SCHED_EVENT_SLEEP; | |
438 | ||
0e9b07e5 | 439 | sched->nr_sleep_events++; |
ec156764 IM |
440 | } |
441 | ||
0e9b07e5 ACM |
442 | static struct task_desc *register_pid(struct perf_sched *sched, |
443 | unsigned long pid, const char *comm) | |
ec156764 IM |
444 | { |
445 | struct task_desc *task; | |
cb06ac25 | 446 | static int pid_max; |
ec156764 | 447 | |
cb06ac25 YS |
448 | if (sched->pid_to_task == NULL) { |
449 | if (sysctl__read_int("kernel/pid_max", &pid_max) < 0) | |
450 | pid_max = MAX_PID; | |
451 | BUG_ON((sched->pid_to_task = calloc(pid_max, sizeof(struct task_desc *))) == NULL); | |
452 | } | |
3a423a5c YS |
453 | if (pid >= (unsigned long)pid_max) { |
454 | BUG_ON((sched->pid_to_task = realloc(sched->pid_to_task, (pid + 1) * | |
455 | sizeof(struct task_desc *))) == NULL); | |
456 | while (pid >= (unsigned long)pid_max) | |
457 | sched->pid_to_task[pid_max++] = NULL; | |
458 | } | |
ec156764 | 459 | |
0e9b07e5 | 460 | task = sched->pid_to_task[pid]; |
ec156764 IM |
461 | |
462 | if (task) | |
463 | return task; | |
464 | ||
36479484 | 465 | task = zalloc(sizeof(*task)); |
ec156764 | 466 | task->pid = pid; |
0e9b07e5 | 467 | task->nr = sched->nr_tasks; |
ec156764 IM |
468 | strcpy(task->comm, comm); |
469 | /* | |
470 | * every task starts in sleeping state - this gets ignored | |
471 | * if there's no wakeup pointing to this sleep state: | |
472 | */ | |
0e9b07e5 | 473 | add_sched_event_sleep(sched, task, 0, 0); |
ec156764 | 474 | |
0e9b07e5 ACM |
475 | sched->pid_to_task[pid] = task; |
476 | sched->nr_tasks++; | |
0755bc4d | 477 | sched->tasks = realloc(sched->tasks, sched->nr_tasks * sizeof(struct task_desc *)); |
0e9b07e5 ACM |
478 | BUG_ON(!sched->tasks); |
479 | sched->tasks[task->nr] = task; | |
ec156764 | 480 | |
bb963e16 | 481 | if (verbose > 0) |
0e9b07e5 | 482 | printf("registered task #%ld, PID %ld (%s)\n", sched->nr_tasks, pid, comm); |
ec156764 IM |
483 | |
484 | return task; | |
485 | } | |
486 | ||
487 | ||
0e9b07e5 | 488 | static void print_task_traces(struct perf_sched *sched) |
ec156764 IM |
489 | { |
490 | struct task_desc *task; | |
491 | unsigned long i; | |
492 | ||
0e9b07e5 ACM |
493 | for (i = 0; i < sched->nr_tasks; i++) { |
494 | task = sched->tasks[i]; | |
ad236fd2 | 495 | printf("task %6ld (%20s:%10ld), nr_events: %ld\n", |
ec156764 IM |
496 | task->nr, task->comm, task->pid, task->nr_events); |
497 | } | |
498 | } | |
499 | ||
0e9b07e5 | 500 | static void add_cross_task_wakeups(struct perf_sched *sched) |
ec156764 IM |
501 | { |
502 | struct task_desc *task1, *task2; | |
503 | unsigned long i, j; | |
504 | ||
0e9b07e5 ACM |
505 | for (i = 0; i < sched->nr_tasks; i++) { |
506 | task1 = sched->tasks[i]; | |
ec156764 | 507 | j = i + 1; |
0e9b07e5 | 508 | if (j == sched->nr_tasks) |
ec156764 | 509 | j = 0; |
0e9b07e5 ACM |
510 | task2 = sched->tasks[j]; |
511 | add_sched_event_wakeup(sched, task1, 0, task2); | |
ec156764 IM |
512 | } |
513 | } | |
514 | ||
0e9b07e5 ACM |
515 | static void perf_sched__process_event(struct perf_sched *sched, |
516 | struct sched_atom *atom) | |
ec156764 IM |
517 | { |
518 | int ret = 0; | |
ec156764 | 519 | |
39aeb52f | 520 | switch (atom->type) { |
ec156764 | 521 | case SCHED_EVENT_RUN: |
0e9b07e5 | 522 | burn_nsecs(sched, atom->duration); |
ec156764 IM |
523 | break; |
524 | case SCHED_EVENT_SLEEP: | |
39aeb52f | 525 | if (atom->wait_sem) |
526 | ret = sem_wait(atom->wait_sem); | |
ec156764 IM |
527 | BUG_ON(ret); |
528 | break; | |
529 | case SCHED_EVENT_WAKEUP: | |
39aeb52f | 530 | if (atom->wait_sem) |
531 | ret = sem_post(atom->wait_sem); | |
ec156764 IM |
532 | BUG_ON(ret); |
533 | break; | |
55ffb7a6 MG |
534 | case SCHED_EVENT_MIGRATION: |
535 | break; | |
ec156764 IM |
536 | default: |
537 | BUG_ON(1); | |
538 | } | |
539 | } | |
540 | ||
b1ffe8f3 | 541 | static u64 get_cpu_usage_nsec_parent(void) |
ec156764 IM |
542 | { |
543 | struct rusage ru; | |
b1ffe8f3 | 544 | u64 sum; |
ec156764 IM |
545 | int err; |
546 | ||
547 | err = getrusage(RUSAGE_SELF, &ru); | |
548 | BUG_ON(err); | |
549 | ||
4fc76e49 ACM |
550 | sum = ru.ru_utime.tv_sec * NSEC_PER_SEC + ru.ru_utime.tv_usec * NSEC_PER_USEC; |
551 | sum += ru.ru_stime.tv_sec * NSEC_PER_SEC + ru.ru_stime.tv_usec * NSEC_PER_USEC; | |
ec156764 IM |
552 | |
553 | return sum; | |
554 | } | |
555 | ||
939cda52 | 556 | static int self_open_counters(struct perf_sched *sched, unsigned long cur_task) |
ec156764 | 557 | { |
c0c9e721 | 558 | struct perf_event_attr attr; |
939cda52 | 559 | char sbuf[STRERR_BUFSIZE], info[STRERR_BUFSIZE]; |
c0c9e721 | 560 | int fd; |
939cda52 YS |
561 | struct rlimit limit; |
562 | bool need_privilege = false; | |
ec156764 | 563 | |
c0c9e721 | 564 | memset(&attr, 0, sizeof(attr)); |
ec156764 | 565 | |
c0c9e721 XG |
566 | attr.type = PERF_TYPE_SOFTWARE; |
567 | attr.config = PERF_COUNT_SW_TASK_CLOCK; | |
ec156764 | 568 | |
939cda52 | 569 | force_again: |
57480d2c YD |
570 | fd = sys_perf_event_open(&attr, 0, -1, -1, |
571 | perf_event_open_cloexec_flag()); | |
c0c9e721 | 572 | |
1aff59be | 573 | if (fd < 0) { |
939cda52 YS |
574 | if (errno == EMFILE) { |
575 | if (sched->force) { | |
576 | BUG_ON(getrlimit(RLIMIT_NOFILE, &limit) == -1); | |
577 | limit.rlim_cur += sched->nr_tasks - cur_task; | |
578 | if (limit.rlim_cur > limit.rlim_max) { | |
579 | limit.rlim_max = limit.rlim_cur; | |
580 | need_privilege = true; | |
581 | } | |
582 | if (setrlimit(RLIMIT_NOFILE, &limit) == -1) { | |
583 | if (need_privilege && errno == EPERM) | |
584 | strcpy(info, "Need privilege\n"); | |
585 | } else | |
586 | goto force_again; | |
587 | } else | |
588 | strcpy(info, "Have a try with -f option\n"); | |
589 | } | |
60b7d14a | 590 | pr_err("Error: sys_perf_event_open() syscall returned " |
939cda52 | 591 | "with %d (%s)\n%s", fd, |
c8b5f2c9 | 592 | str_error_r(errno, sbuf, sizeof(sbuf)), info); |
1aff59be YS |
593 | exit(EXIT_FAILURE); |
594 | } | |
c0c9e721 XG |
595 | return fd; |
596 | } | |
597 | ||
598 | static u64 get_cpu_usage_nsec_self(int fd) | |
599 | { | |
600 | u64 runtime; | |
601 | int ret; | |
602 | ||
603 | ret = read(fd, &runtime, sizeof(runtime)); | |
604 | BUG_ON(ret != sizeof(runtime)); | |
605 | ||
606 | return runtime; | |
ec156764 IM |
607 | } |
608 | ||
0e9b07e5 ACM |
609 | struct sched_thread_parms { |
610 | struct task_desc *task; | |
611 | struct perf_sched *sched; | |
08097abc | 612 | int fd; |
0e9b07e5 ACM |
613 | }; |
614 | ||
ec156764 IM |
615 | static void *thread_func(void *ctx) |
616 | { | |
0e9b07e5 ACM |
617 | struct sched_thread_parms *parms = ctx; |
618 | struct task_desc *this_task = parms->task; | |
619 | struct perf_sched *sched = parms->sched; | |
b1ffe8f3 | 620 | u64 cpu_usage_0, cpu_usage_1; |
ec156764 IM |
621 | unsigned long i, ret; |
622 | char comm2[22]; | |
08097abc | 623 | int fd = parms->fd; |
ec156764 | 624 | |
74cf249d | 625 | zfree(&parms); |
0e9b07e5 | 626 | |
ec156764 IM |
627 | sprintf(comm2, ":%s", this_task->comm); |
628 | prctl(PR_SET_NAME, comm2); | |
a116e05d ACM |
629 | if (fd < 0) |
630 | return NULL; | |
ec156764 IM |
631 | again: |
632 | ret = sem_post(&this_task->ready_for_work); | |
633 | BUG_ON(ret); | |
0e9b07e5 | 634 | ret = pthread_mutex_lock(&sched->start_work_mutex); |
ec156764 | 635 | BUG_ON(ret); |
0e9b07e5 | 636 | ret = pthread_mutex_unlock(&sched->start_work_mutex); |
ec156764 | 637 | BUG_ON(ret); |
ec156764 | 638 | |
c0c9e721 | 639 | cpu_usage_0 = get_cpu_usage_nsec_self(fd); |
ec156764 IM |
640 | |
641 | for (i = 0; i < this_task->nr_events; i++) { | |
642 | this_task->curr_event = i; | |
0e9b07e5 | 643 | perf_sched__process_event(sched, this_task->atoms[i]); |
ec156764 IM |
644 | } |
645 | ||
c0c9e721 | 646 | cpu_usage_1 = get_cpu_usage_nsec_self(fd); |
ec156764 | 647 | this_task->cpu_usage = cpu_usage_1 - cpu_usage_0; |
ec156764 IM |
648 | ret = sem_post(&this_task->work_done_sem); |
649 | BUG_ON(ret); | |
ec156764 | 650 | |
0e9b07e5 | 651 | ret = pthread_mutex_lock(&sched->work_done_wait_mutex); |
ec156764 | 652 | BUG_ON(ret); |
0e9b07e5 | 653 | ret = pthread_mutex_unlock(&sched->work_done_wait_mutex); |
ec156764 | 654 | BUG_ON(ret); |
ec156764 IM |
655 | |
656 | goto again; | |
657 | } | |
658 | ||
0e9b07e5 | 659 | static void create_tasks(struct perf_sched *sched) |
ec156764 IM |
660 | { |
661 | struct task_desc *task; | |
662 | pthread_attr_t attr; | |
663 | unsigned long i; | |
664 | int err; | |
665 | ||
666 | err = pthread_attr_init(&attr); | |
667 | BUG_ON(err); | |
12f7e036 JP |
668 | err = pthread_attr_setstacksize(&attr, |
669 | (size_t) max(16 * 1024, PTHREAD_STACK_MIN)); | |
ec156764 | 670 | BUG_ON(err); |
0e9b07e5 | 671 | err = pthread_mutex_lock(&sched->start_work_mutex); |
ec156764 | 672 | BUG_ON(err); |
0e9b07e5 | 673 | err = pthread_mutex_lock(&sched->work_done_wait_mutex); |
ec156764 | 674 | BUG_ON(err); |
0e9b07e5 ACM |
675 | for (i = 0; i < sched->nr_tasks; i++) { |
676 | struct sched_thread_parms *parms = malloc(sizeof(*parms)); | |
677 | BUG_ON(parms == NULL); | |
678 | parms->task = task = sched->tasks[i]; | |
679 | parms->sched = sched; | |
939cda52 | 680 | parms->fd = self_open_counters(sched, i); |
ec156764 IM |
681 | sem_init(&task->sleep_sem, 0, 0); |
682 | sem_init(&task->ready_for_work, 0, 0); | |
683 | sem_init(&task->work_done_sem, 0, 0); | |
684 | task->curr_event = 0; | |
0e9b07e5 | 685 | err = pthread_create(&task->thread, &attr, thread_func, parms); |
ec156764 IM |
686 | BUG_ON(err); |
687 | } | |
688 | } | |
689 | ||
0e9b07e5 | 690 | static void wait_for_tasks(struct perf_sched *sched) |
ec156764 | 691 | { |
b1ffe8f3 | 692 | u64 cpu_usage_0, cpu_usage_1; |
ec156764 IM |
693 | struct task_desc *task; |
694 | unsigned long i, ret; | |
695 | ||
0e9b07e5 ACM |
696 | sched->start_time = get_nsecs(); |
697 | sched->cpu_usage = 0; | |
698 | pthread_mutex_unlock(&sched->work_done_wait_mutex); | |
ec156764 | 699 | |
0e9b07e5 ACM |
700 | for (i = 0; i < sched->nr_tasks; i++) { |
701 | task = sched->tasks[i]; | |
ec156764 IM |
702 | ret = sem_wait(&task->ready_for_work); |
703 | BUG_ON(ret); | |
704 | sem_init(&task->ready_for_work, 0, 0); | |
705 | } | |
0e9b07e5 | 706 | ret = pthread_mutex_lock(&sched->work_done_wait_mutex); |
ec156764 IM |
707 | BUG_ON(ret); |
708 | ||
709 | cpu_usage_0 = get_cpu_usage_nsec_parent(); | |
710 | ||
0e9b07e5 | 711 | pthread_mutex_unlock(&sched->start_work_mutex); |
ec156764 | 712 | |
0e9b07e5 ACM |
713 | for (i = 0; i < sched->nr_tasks; i++) { |
714 | task = sched->tasks[i]; | |
ec156764 IM |
715 | ret = sem_wait(&task->work_done_sem); |
716 | BUG_ON(ret); | |
717 | sem_init(&task->work_done_sem, 0, 0); | |
0e9b07e5 | 718 | sched->cpu_usage += task->cpu_usage; |
ec156764 IM |
719 | task->cpu_usage = 0; |
720 | } | |
721 | ||
722 | cpu_usage_1 = get_cpu_usage_nsec_parent(); | |
0e9b07e5 ACM |
723 | if (!sched->runavg_cpu_usage) |
724 | sched->runavg_cpu_usage = sched->cpu_usage; | |
ff5f3bbd | 725 | sched->runavg_cpu_usage = (sched->runavg_cpu_usage * (sched->replay_repeat - 1) + sched->cpu_usage) / sched->replay_repeat; |
ec156764 | 726 | |
0e9b07e5 ACM |
727 | sched->parent_cpu_usage = cpu_usage_1 - cpu_usage_0; |
728 | if (!sched->runavg_parent_cpu_usage) | |
729 | sched->runavg_parent_cpu_usage = sched->parent_cpu_usage; | |
ff5f3bbd YS |
730 | sched->runavg_parent_cpu_usage = (sched->runavg_parent_cpu_usage * (sched->replay_repeat - 1) + |
731 | sched->parent_cpu_usage)/sched->replay_repeat; | |
ec156764 | 732 | |
0e9b07e5 | 733 | ret = pthread_mutex_lock(&sched->start_work_mutex); |
ec156764 IM |
734 | BUG_ON(ret); |
735 | ||
0e9b07e5 ACM |
736 | for (i = 0; i < sched->nr_tasks; i++) { |
737 | task = sched->tasks[i]; | |
ec156764 IM |
738 | sem_init(&task->sleep_sem, 0, 0); |
739 | task->curr_event = 0; | |
740 | } | |
741 | } | |
742 | ||
0e9b07e5 | 743 | static void run_one_test(struct perf_sched *sched) |
ec156764 | 744 | { |
fb7d0b3c | 745 | u64 T0, T1, delta, avg_delta, fluct; |
ec156764 IM |
746 | |
747 | T0 = get_nsecs(); | |
0e9b07e5 | 748 | wait_for_tasks(sched); |
ec156764 IM |
749 | T1 = get_nsecs(); |
750 | ||
751 | delta = T1 - T0; | |
0e9b07e5 ACM |
752 | sched->sum_runtime += delta; |
753 | sched->nr_runs++; | |
ec156764 | 754 | |
0e9b07e5 | 755 | avg_delta = sched->sum_runtime / sched->nr_runs; |
ec156764 IM |
756 | if (delta < avg_delta) |
757 | fluct = avg_delta - delta; | |
758 | else | |
759 | fluct = delta - avg_delta; | |
0e9b07e5 ACM |
760 | sched->sum_fluct += fluct; |
761 | if (!sched->run_avg) | |
762 | sched->run_avg = delta; | |
ff5f3bbd | 763 | sched->run_avg = (sched->run_avg * (sched->replay_repeat - 1) + delta) / sched->replay_repeat; |
ec156764 | 764 | |
4fc76e49 | 765 | printf("#%-3ld: %0.3f, ", sched->nr_runs, (double)delta / NSEC_PER_MSEC); |
ec156764 | 766 | |
4fc76e49 | 767 | printf("ravg: %0.2f, ", (double)sched->run_avg / NSEC_PER_MSEC); |
ec156764 | 768 | |
ad236fd2 | 769 | printf("cpu: %0.2f / %0.2f", |
4fc76e49 | 770 | (double)sched->cpu_usage / NSEC_PER_MSEC, (double)sched->runavg_cpu_usage / NSEC_PER_MSEC); |
ec156764 IM |
771 | |
772 | #if 0 | |
773 | /* | |
fbf94829 | 774 | * rusage statistics done by the parent, these are less |
0e9b07e5 | 775 | * accurate than the sched->sum_exec_runtime based statistics: |
fbf94829 | 776 | */ |
ad236fd2 | 777 | printf(" [%0.2f / %0.2f]", |
4fc76e49 ACM |
778 | (double)sched->parent_cpu_usage / NSEC_PER_MSEC, |
779 | (double)sched->runavg_parent_cpu_usage / NSEC_PER_MSEC); | |
ec156764 IM |
780 | #endif |
781 | ||
ad236fd2 | 782 | printf("\n"); |
ec156764 | 783 | |
0e9b07e5 ACM |
784 | if (sched->nr_sleep_corrections) |
785 | printf(" (%ld sleep corrections)\n", sched->nr_sleep_corrections); | |
786 | sched->nr_sleep_corrections = 0; | |
ec156764 IM |
787 | } |
788 | ||
0e9b07e5 | 789 | static void test_calibrations(struct perf_sched *sched) |
ec156764 | 790 | { |
b1ffe8f3 | 791 | u64 T0, T1; |
ec156764 IM |
792 | |
793 | T0 = get_nsecs(); | |
4fc76e49 | 794 | burn_nsecs(sched, NSEC_PER_MSEC); |
ec156764 IM |
795 | T1 = get_nsecs(); |
796 | ||
9486aa38 | 797 | printf("the run test took %" PRIu64 " nsecs\n", T1 - T0); |
ec156764 IM |
798 | |
799 | T0 = get_nsecs(); | |
4fc76e49 | 800 | sleep_nsecs(NSEC_PER_MSEC); |
ec156764 IM |
801 | T1 = get_nsecs(); |
802 | ||
9486aa38 | 803 | printf("the sleep test took %" PRIu64 " nsecs\n", T1 - T0); |
ec156764 IM |
804 | } |
805 | ||
a116e05d | 806 | static int |
0e9b07e5 | 807 | replay_wakeup_event(struct perf_sched *sched, |
32dcd021 | 808 | struct evsel *evsel, struct perf_sample *sample, |
9ec3f4e4 | 809 | struct machine *machine __maybe_unused) |
419ab0d6 | 810 | { |
9ec3f4e4 ACM |
811 | const char *comm = perf_evsel__strval(evsel, sample, "comm"); |
812 | const u32 pid = perf_evsel__intval(evsel, sample, "pid"); | |
419ab0d6 | 813 | struct task_desc *waker, *wakee; |
fbf94829 | 814 | |
bb963e16 | 815 | if (verbose > 0) { |
2b7fcbc5 | 816 | printf("sched_wakeup event %p\n", evsel); |
fbf94829 | 817 | |
9ec3f4e4 | 818 | printf(" ... pid %d woke up %s/%d\n", sample->tid, comm, pid); |
ad236fd2 | 819 | } |
fbf94829 | 820 | |
2b7fcbc5 | 821 | waker = register_pid(sched, sample->tid, "<unknown>"); |
9ec3f4e4 | 822 | wakee = register_pid(sched, pid, comm); |
fbf94829 | 823 | |
0e9b07e5 | 824 | add_sched_event_wakeup(sched, waker, sample->time, wakee); |
a116e05d | 825 | return 0; |
ec156764 IM |
826 | } |
827 | ||
9ec3f4e4 | 828 | static int replay_switch_event(struct perf_sched *sched, |
32dcd021 | 829 | struct evsel *evsel, |
9ec3f4e4 ACM |
830 | struct perf_sample *sample, |
831 | struct machine *machine __maybe_unused) | |
ec156764 | 832 | { |
9ec3f4e4 ACM |
833 | const char *prev_comm = perf_evsel__strval(evsel, sample, "prev_comm"), |
834 | *next_comm = perf_evsel__strval(evsel, sample, "next_comm"); | |
835 | const u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"), | |
836 | next_pid = perf_evsel__intval(evsel, sample, "next_pid"); | |
837 | const u64 prev_state = perf_evsel__intval(evsel, sample, "prev_state"); | |
1d037ca1 | 838 | struct task_desc *prev, __maybe_unused *next; |
7f7f8d0b ACM |
839 | u64 timestamp0, timestamp = sample->time; |
840 | int cpu = sample->cpu; | |
fbf94829 IM |
841 | s64 delta; |
842 | ||
bb963e16 | 843 | if (verbose > 0) |
2b7fcbc5 | 844 | printf("sched_switch event %p\n", evsel); |
ad236fd2 | 845 | |
fbf94829 | 846 | if (cpu >= MAX_CPUS || cpu < 0) |
a116e05d | 847 | return 0; |
fbf94829 | 848 | |
0e9b07e5 | 849 | timestamp0 = sched->cpu_last_switched[cpu]; |
fbf94829 IM |
850 | if (timestamp0) |
851 | delta = timestamp - timestamp0; | |
852 | else | |
853 | delta = 0; | |
854 | ||
a116e05d | 855 | if (delta < 0) { |
60b7d14a | 856 | pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta); |
a116e05d ACM |
857 | return -1; |
858 | } | |
fbf94829 | 859 | |
9ec3f4e4 ACM |
860 | pr_debug(" ... switch from %s/%d to %s/%d [ran %" PRIu64 " nsecs]\n", |
861 | prev_comm, prev_pid, next_comm, next_pid, delta); | |
fbf94829 | 862 | |
9ec3f4e4 ACM |
863 | prev = register_pid(sched, prev_pid, prev_comm); |
864 | next = register_pid(sched, next_pid, next_comm); | |
fbf94829 | 865 | |
0e9b07e5 | 866 | sched->cpu_last_switched[cpu] = timestamp; |
fbf94829 | 867 | |
0e9b07e5 | 868 | add_sched_event_run(sched, prev, timestamp, delta); |
9ec3f4e4 | 869 | add_sched_event_sleep(sched, prev, timestamp, prev_state); |
a116e05d ACM |
870 | |
871 | return 0; | |
fbf94829 IM |
872 | } |
873 | ||
cb627505 DA |
874 | static int replay_fork_event(struct perf_sched *sched, |
875 | union perf_event *event, | |
876 | struct machine *machine) | |
419ab0d6 | 877 | { |
cb627505 DA |
878 | struct thread *child, *parent; |
879 | ||
314add6b AH |
880 | child = machine__findnew_thread(machine, event->fork.pid, |
881 | event->fork.tid); | |
882 | parent = machine__findnew_thread(machine, event->fork.ppid, | |
883 | event->fork.ptid); | |
cb627505 DA |
884 | |
885 | if (child == NULL || parent == NULL) { | |
886 | pr_debug("thread does not exist on fork event: child %p, parent %p\n", | |
887 | child, parent); | |
b91fc39f | 888 | goto out_put; |
cb627505 | 889 | } |
9ec3f4e4 | 890 | |
bb963e16 | 891 | if (verbose > 0) { |
cb627505 | 892 | printf("fork event\n"); |
b9c5143a FW |
893 | printf("... parent: %s/%d\n", thread__comm_str(parent), parent->tid); |
894 | printf("... child: %s/%d\n", thread__comm_str(child), child->tid); | |
419ab0d6 | 895 | } |
9ec3f4e4 | 896 | |
b9c5143a FW |
897 | register_pid(sched, parent->tid, thread__comm_str(parent)); |
898 | register_pid(sched, child->tid, thread__comm_str(child)); | |
b91fc39f ACM |
899 | out_put: |
900 | thread__put(child); | |
901 | thread__put(parent); | |
a116e05d | 902 | return 0; |
419ab0d6 | 903 | } |
fbf94829 | 904 | |
b1ffe8f3 IM |
905 | struct sort_dimension { |
906 | const char *name; | |
b5fae128 | 907 | sort_fn_t cmp; |
b1ffe8f3 IM |
908 | struct list_head list; |
909 | }; | |
910 | ||
8640da9f CD |
911 | /* |
912 | * handle runtime stats saved per thread | |
913 | */ | |
914 | static struct thread_runtime *thread__init_runtime(struct thread *thread) | |
915 | { | |
916 | struct thread_runtime *r; | |
917 | ||
918 | r = zalloc(sizeof(struct thread_runtime)); | |
919 | if (!r) | |
920 | return NULL; | |
921 | ||
922 | init_stats(&r->run_stats); | |
923 | thread__set_priv(thread, r); | |
924 | ||
925 | return r; | |
926 | } | |
927 | ||
928 | static struct thread_runtime *thread__get_runtime(struct thread *thread) | |
929 | { | |
930 | struct thread_runtime *tr; | |
931 | ||
932 | tr = thread__priv(thread); | |
933 | if (tr == NULL) { | |
934 | tr = thread__init_runtime(thread); | |
935 | if (tr == NULL) | |
936 | pr_debug("Failed to malloc memory for runtime data.\n"); | |
937 | } | |
938 | ||
939 | return tr; | |
940 | } | |
941 | ||
daa1d7a5 | 942 | static int |
39aeb52f | 943 | thread_lat_cmp(struct list_head *list, struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 FW |
944 | { |
945 | struct sort_dimension *sort; | |
946 | int ret = 0; | |
947 | ||
b5fae128 IM |
948 | BUG_ON(list_empty(list)); |
949 | ||
daa1d7a5 FW |
950 | list_for_each_entry(sort, list, list) { |
951 | ret = sort->cmp(l, r); | |
952 | if (ret) | |
953 | return ret; | |
954 | } | |
955 | ||
956 | return ret; | |
957 | } | |
958 | ||
39aeb52f | 959 | static struct work_atoms * |
cb4c13a5 | 960 | thread_atoms_search(struct rb_root_cached *root, struct thread *thread, |
b5fae128 IM |
961 | struct list_head *sort_list) |
962 | { | |
cb4c13a5 | 963 | struct rb_node *node = root->rb_root.rb_node; |
39aeb52f | 964 | struct work_atoms key = { .thread = thread }; |
b5fae128 IM |
965 | |
966 | while (node) { | |
39aeb52f | 967 | struct work_atoms *atoms; |
b5fae128 IM |
968 | int cmp; |
969 | ||
39aeb52f | 970 | atoms = container_of(node, struct work_atoms, node); |
b5fae128 IM |
971 | |
972 | cmp = thread_lat_cmp(sort_list, &key, atoms); | |
973 | if (cmp > 0) | |
974 | node = node->rb_left; | |
975 | else if (cmp < 0) | |
976 | node = node->rb_right; | |
977 | else { | |
978 | BUG_ON(thread != atoms->thread); | |
979 | return atoms; | |
980 | } | |
981 | } | |
982 | return NULL; | |
983 | } | |
984 | ||
cdce9d73 | 985 | static void |
cb4c13a5 | 986 | __thread_latency_insert(struct rb_root_cached *root, struct work_atoms *data, |
daa1d7a5 | 987 | struct list_head *sort_list) |
cdce9d73 | 988 | { |
cb4c13a5 DB |
989 | struct rb_node **new = &(root->rb_root.rb_node), *parent = NULL; |
990 | bool leftmost = true; | |
cdce9d73 FW |
991 | |
992 | while (*new) { | |
39aeb52f | 993 | struct work_atoms *this; |
daa1d7a5 | 994 | int cmp; |
cdce9d73 | 995 | |
39aeb52f | 996 | this = container_of(*new, struct work_atoms, node); |
cdce9d73 | 997 | parent = *new; |
daa1d7a5 FW |
998 | |
999 | cmp = thread_lat_cmp(sort_list, data, this); | |
1000 | ||
1001 | if (cmp > 0) | |
cdce9d73 | 1002 | new = &((*new)->rb_left); |
cb4c13a5 | 1003 | else { |
daa1d7a5 | 1004 | new = &((*new)->rb_right); |
cb4c13a5 DB |
1005 | leftmost = false; |
1006 | } | |
cdce9d73 FW |
1007 | } |
1008 | ||
1009 | rb_link_node(&data->node, parent, new); | |
cb4c13a5 | 1010 | rb_insert_color_cached(&data->node, root, leftmost); |
cdce9d73 FW |
1011 | } |
1012 | ||
0e9b07e5 | 1013 | static int thread_atoms_insert(struct perf_sched *sched, struct thread *thread) |
cdce9d73 | 1014 | { |
36479484 | 1015 | struct work_atoms *atoms = zalloc(sizeof(*atoms)); |
a116e05d ACM |
1016 | if (!atoms) { |
1017 | pr_err("No memory at %s\n", __func__); | |
1018 | return -1; | |
1019 | } | |
cdce9d73 | 1020 | |
f3b623b8 | 1021 | atoms->thread = thread__get(thread); |
39aeb52f | 1022 | INIT_LIST_HEAD(&atoms->work_list); |
0e9b07e5 | 1023 | __thread_latency_insert(&sched->atom_root, atoms, &sched->cmp_pid); |
a116e05d | 1024 | return 0; |
cdce9d73 FW |
1025 | } |
1026 | ||
9ec3f4e4 | 1027 | static char sched_out_state(u64 prev_state) |
cdce9d73 FW |
1028 | { |
1029 | const char *str = TASK_STATE_TO_CHAR_STR; | |
1030 | ||
9ec3f4e4 | 1031 | return str[prev_state]; |
cdce9d73 FW |
1032 | } |
1033 | ||
a116e05d | 1034 | static int |
39aeb52f | 1035 | add_sched_out_event(struct work_atoms *atoms, |
1036 | char run_state, | |
1037 | u64 timestamp) | |
cdce9d73 | 1038 | { |
36479484 | 1039 | struct work_atom *atom = zalloc(sizeof(*atom)); |
a116e05d ACM |
1040 | if (!atom) { |
1041 | pr_err("Non memory at %s", __func__); | |
1042 | return -1; | |
1043 | } | |
cdce9d73 | 1044 | |
aa1ab9d2 FW |
1045 | atom->sched_out_time = timestamp; |
1046 | ||
39aeb52f | 1047 | if (run_state == 'R') { |
b1ffe8f3 | 1048 | atom->state = THREAD_WAIT_CPU; |
aa1ab9d2 | 1049 | atom->wake_up_time = atom->sched_out_time; |
c6ced611 FW |
1050 | } |
1051 | ||
39aeb52f | 1052 | list_add_tail(&atom->list, &atoms->work_list); |
a116e05d | 1053 | return 0; |
cdce9d73 FW |
1054 | } |
1055 | ||
1056 | static void | |
1d037ca1 IT |
1057 | add_runtime_event(struct work_atoms *atoms, u64 delta, |
1058 | u64 timestamp __maybe_unused) | |
39aeb52f | 1059 | { |
1060 | struct work_atom *atom; | |
1061 | ||
1062 | BUG_ON(list_empty(&atoms->work_list)); | |
1063 | ||
1064 | atom = list_entry(atoms->work_list.prev, struct work_atom, list); | |
1065 | ||
1066 | atom->runtime += delta; | |
1067 | atoms->total_runtime += delta; | |
1068 | } | |
1069 | ||
1070 | static void | |
1071 | add_sched_in_event(struct work_atoms *atoms, u64 timestamp) | |
cdce9d73 | 1072 | { |
b1ffe8f3 | 1073 | struct work_atom *atom; |
66685678 | 1074 | u64 delta; |
cdce9d73 | 1075 | |
39aeb52f | 1076 | if (list_empty(&atoms->work_list)) |
cdce9d73 FW |
1077 | return; |
1078 | ||
39aeb52f | 1079 | atom = list_entry(atoms->work_list.prev, struct work_atom, list); |
cdce9d73 | 1080 | |
b1ffe8f3 | 1081 | if (atom->state != THREAD_WAIT_CPU) |
cdce9d73 FW |
1082 | return; |
1083 | ||
b1ffe8f3 IM |
1084 | if (timestamp < atom->wake_up_time) { |
1085 | atom->state = THREAD_IGNORE; | |
cdce9d73 FW |
1086 | return; |
1087 | } | |
1088 | ||
b1ffe8f3 IM |
1089 | atom->state = THREAD_SCHED_IN; |
1090 | atom->sched_in_time = timestamp; | |
66685678 | 1091 | |
b1ffe8f3 | 1092 | delta = atom->sched_in_time - atom->wake_up_time; |
66685678 | 1093 | atoms->total_lat += delta; |
3786310a | 1094 | if (delta > atoms->max_lat) { |
66685678 | 1095 | atoms->max_lat = delta; |
3786310a FW |
1096 | atoms->max_lat_at = timestamp; |
1097 | } | |
66685678 | 1098 | atoms->nb_atoms++; |
cdce9d73 FW |
1099 | } |
1100 | ||
9ec3f4e4 | 1101 | static int latency_switch_event(struct perf_sched *sched, |
32dcd021 | 1102 | struct evsel *evsel, |
9ec3f4e4 ACM |
1103 | struct perf_sample *sample, |
1104 | struct machine *machine) | |
cdce9d73 | 1105 | { |
9ec3f4e4 ACM |
1106 | const u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"), |
1107 | next_pid = perf_evsel__intval(evsel, sample, "next_pid"); | |
1108 | const u64 prev_state = perf_evsel__intval(evsel, sample, "prev_state"); | |
39aeb52f | 1109 | struct work_atoms *out_events, *in_events; |
cdce9d73 | 1110 | struct thread *sched_out, *sched_in; |
7f7f8d0b | 1111 | u64 timestamp0, timestamp = sample->time; |
b91fc39f | 1112 | int cpu = sample->cpu, err = -1; |
ea92ed5a IM |
1113 | s64 delta; |
1114 | ||
39aeb52f | 1115 | BUG_ON(cpu >= MAX_CPUS || cpu < 0); |
ea92ed5a | 1116 | |
0e9b07e5 ACM |
1117 | timestamp0 = sched->cpu_last_switched[cpu]; |
1118 | sched->cpu_last_switched[cpu] = timestamp; | |
ea92ed5a IM |
1119 | if (timestamp0) |
1120 | delta = timestamp - timestamp0; | |
1121 | else | |
1122 | delta = 0; | |
1123 | ||
a116e05d ACM |
1124 | if (delta < 0) { |
1125 | pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta); | |
1126 | return -1; | |
1127 | } | |
cdce9d73 | 1128 | |
1fcb8768 AH |
1129 | sched_out = machine__findnew_thread(machine, -1, prev_pid); |
1130 | sched_in = machine__findnew_thread(machine, -1, next_pid); | |
b91fc39f ACM |
1131 | if (sched_out == NULL || sched_in == NULL) |
1132 | goto out_put; | |
cdce9d73 | 1133 | |
0e9b07e5 | 1134 | out_events = thread_atoms_search(&sched->atom_root, sched_out, &sched->cmp_pid); |
39aeb52f | 1135 | if (!out_events) { |
0e9b07e5 | 1136 | if (thread_atoms_insert(sched, sched_out)) |
b91fc39f | 1137 | goto out_put; |
0e9b07e5 | 1138 | out_events = thread_atoms_search(&sched->atom_root, sched_out, &sched->cmp_pid); |
a116e05d ACM |
1139 | if (!out_events) { |
1140 | pr_err("out-event: Internal tree error"); | |
b91fc39f | 1141 | goto out_put; |
a116e05d | 1142 | } |
39aeb52f | 1143 | } |
9ec3f4e4 | 1144 | if (add_sched_out_event(out_events, sched_out_state(prev_state), timestamp)) |
a116e05d | 1145 | return -1; |
39aeb52f | 1146 | |
0e9b07e5 | 1147 | in_events = thread_atoms_search(&sched->atom_root, sched_in, &sched->cmp_pid); |
39aeb52f | 1148 | if (!in_events) { |
0e9b07e5 | 1149 | if (thread_atoms_insert(sched, sched_in)) |
b91fc39f | 1150 | goto out_put; |
0e9b07e5 | 1151 | in_events = thread_atoms_search(&sched->atom_root, sched_in, &sched->cmp_pid); |
a116e05d ACM |
1152 | if (!in_events) { |
1153 | pr_err("in-event: Internal tree error"); | |
b91fc39f | 1154 | goto out_put; |
a116e05d | 1155 | } |
39aeb52f | 1156 | /* |
1157 | * Take came in we have not heard about yet, | |
1158 | * add in an initial atom in runnable state: | |
1159 | */ | |
a116e05d | 1160 | if (add_sched_out_event(in_events, 'R', timestamp)) |
b91fc39f | 1161 | goto out_put; |
cdce9d73 | 1162 | } |
39aeb52f | 1163 | add_sched_in_event(in_events, timestamp); |
b91fc39f ACM |
1164 | err = 0; |
1165 | out_put: | |
1166 | thread__put(sched_out); | |
1167 | thread__put(sched_in); | |
1168 | return err; | |
39aeb52f | 1169 | } |
cdce9d73 | 1170 | |
9ec3f4e4 | 1171 | static int latency_runtime_event(struct perf_sched *sched, |
32dcd021 | 1172 | struct evsel *evsel, |
9ec3f4e4 ACM |
1173 | struct perf_sample *sample, |
1174 | struct machine *machine) | |
39aeb52f | 1175 | { |
9ec3f4e4 ACM |
1176 | const u32 pid = perf_evsel__intval(evsel, sample, "pid"); |
1177 | const u64 runtime = perf_evsel__intval(evsel, sample, "runtime"); | |
1fcb8768 | 1178 | struct thread *thread = machine__findnew_thread(machine, -1, pid); |
0e9b07e5 | 1179 | struct work_atoms *atoms = thread_atoms_search(&sched->atom_root, thread, &sched->cmp_pid); |
7f7f8d0b | 1180 | u64 timestamp = sample->time; |
b91fc39f ACM |
1181 | int cpu = sample->cpu, err = -1; |
1182 | ||
1183 | if (thread == NULL) | |
1184 | return -1; | |
39aeb52f | 1185 | |
1186 | BUG_ON(cpu >= MAX_CPUS || cpu < 0); | |
39aeb52f | 1187 | if (!atoms) { |
0e9b07e5 | 1188 | if (thread_atoms_insert(sched, thread)) |
b91fc39f | 1189 | goto out_put; |
0e9b07e5 | 1190 | atoms = thread_atoms_search(&sched->atom_root, thread, &sched->cmp_pid); |
a116e05d | 1191 | if (!atoms) { |
60b7d14a | 1192 | pr_err("in-event: Internal tree error"); |
b91fc39f | 1193 | goto out_put; |
a116e05d ACM |
1194 | } |
1195 | if (add_sched_out_event(atoms, 'R', timestamp)) | |
b91fc39f | 1196 | goto out_put; |
cdce9d73 FW |
1197 | } |
1198 | ||
9ec3f4e4 | 1199 | add_runtime_event(atoms, runtime, timestamp); |
b91fc39f ACM |
1200 | err = 0; |
1201 | out_put: | |
1202 | thread__put(thread); | |
1203 | return err; | |
cdce9d73 FW |
1204 | } |
1205 | ||
9ec3f4e4 | 1206 | static int latency_wakeup_event(struct perf_sched *sched, |
32dcd021 | 1207 | struct evsel *evsel, |
9ec3f4e4 ACM |
1208 | struct perf_sample *sample, |
1209 | struct machine *machine) | |
cdce9d73 | 1210 | { |
0680ee7d | 1211 | const u32 pid = perf_evsel__intval(evsel, sample, "pid"); |
39aeb52f | 1212 | struct work_atoms *atoms; |
b1ffe8f3 | 1213 | struct work_atom *atom; |
cdce9d73 | 1214 | struct thread *wakee; |
7f7f8d0b | 1215 | u64 timestamp = sample->time; |
b91fc39f | 1216 | int err = -1; |
cdce9d73 | 1217 | |
1fcb8768 | 1218 | wakee = machine__findnew_thread(machine, -1, pid); |
b91fc39f ACM |
1219 | if (wakee == NULL) |
1220 | return -1; | |
0e9b07e5 | 1221 | atoms = thread_atoms_search(&sched->atom_root, wakee, &sched->cmp_pid); |
17562205 | 1222 | if (!atoms) { |
0e9b07e5 | 1223 | if (thread_atoms_insert(sched, wakee)) |
b91fc39f | 1224 | goto out_put; |
0e9b07e5 | 1225 | atoms = thread_atoms_search(&sched->atom_root, wakee, &sched->cmp_pid); |
a116e05d | 1226 | if (!atoms) { |
60b7d14a | 1227 | pr_err("wakeup-event: Internal tree error"); |
b91fc39f | 1228 | goto out_put; |
a116e05d ACM |
1229 | } |
1230 | if (add_sched_out_event(atoms, 'S', timestamp)) | |
b91fc39f | 1231 | goto out_put; |
cdce9d73 FW |
1232 | } |
1233 | ||
39aeb52f | 1234 | BUG_ON(list_empty(&atoms->work_list)); |
cdce9d73 | 1235 | |
39aeb52f | 1236 | atom = list_entry(atoms->work_list.prev, struct work_atom, list); |
cdce9d73 | 1237 | |
55ffb7a6 | 1238 | /* |
67d6259d DY |
1239 | * As we do not guarantee the wakeup event happens when |
1240 | * task is out of run queue, also may happen when task is | |
1241 | * on run queue and wakeup only change ->state to TASK_RUNNING, | |
1242 | * then we should not set the ->wake_up_time when wake up a | |
1243 | * task which is on run queue. | |
1244 | * | |
55ffb7a6 MG |
1245 | * You WILL be missing events if you've recorded only |
1246 | * one CPU, or are only looking at only one, so don't | |
67d6259d | 1247 | * skip in this case. |
55ffb7a6 | 1248 | */ |
0e9b07e5 | 1249 | if (sched->profile_cpu == -1 && atom->state != THREAD_SLEEPING) |
b91fc39f | 1250 | goto out_ok; |
cdce9d73 | 1251 | |
0e9b07e5 | 1252 | sched->nr_timestamps++; |
ea57c4f5 | 1253 | if (atom->sched_out_time > timestamp) { |
0e9b07e5 | 1254 | sched->nr_unordered_timestamps++; |
b91fc39f | 1255 | goto out_ok; |
ea57c4f5 | 1256 | } |
aa1ab9d2 | 1257 | |
b1ffe8f3 IM |
1258 | atom->state = THREAD_WAIT_CPU; |
1259 | atom->wake_up_time = timestamp; | |
b91fc39f ACM |
1260 | out_ok: |
1261 | err = 0; | |
1262 | out_put: | |
1263 | thread__put(wakee); | |
1264 | return err; | |
cdce9d73 FW |
1265 | } |
1266 | ||
9ec3f4e4 | 1267 | static int latency_migrate_task_event(struct perf_sched *sched, |
32dcd021 | 1268 | struct evsel *evsel, |
9ec3f4e4 ACM |
1269 | struct perf_sample *sample, |
1270 | struct machine *machine) | |
55ffb7a6 | 1271 | { |
9ec3f4e4 | 1272 | const u32 pid = perf_evsel__intval(evsel, sample, "pid"); |
7f7f8d0b | 1273 | u64 timestamp = sample->time; |
55ffb7a6 MG |
1274 | struct work_atoms *atoms; |
1275 | struct work_atom *atom; | |
1276 | struct thread *migrant; | |
b91fc39f | 1277 | int err = -1; |
55ffb7a6 MG |
1278 | |
1279 | /* | |
1280 | * Only need to worry about migration when profiling one CPU. | |
1281 | */ | |
0e9b07e5 | 1282 | if (sched->profile_cpu == -1) |
a116e05d | 1283 | return 0; |
55ffb7a6 | 1284 | |
1fcb8768 | 1285 | migrant = machine__findnew_thread(machine, -1, pid); |
b91fc39f ACM |
1286 | if (migrant == NULL) |
1287 | return -1; | |
0e9b07e5 | 1288 | atoms = thread_atoms_search(&sched->atom_root, migrant, &sched->cmp_pid); |
55ffb7a6 | 1289 | if (!atoms) { |
0e9b07e5 | 1290 | if (thread_atoms_insert(sched, migrant)) |
b91fc39f | 1291 | goto out_put; |
b9c5143a | 1292 | register_pid(sched, migrant->tid, thread__comm_str(migrant)); |
0e9b07e5 | 1293 | atoms = thread_atoms_search(&sched->atom_root, migrant, &sched->cmp_pid); |
a116e05d | 1294 | if (!atoms) { |
60b7d14a | 1295 | pr_err("migration-event: Internal tree error"); |
b91fc39f | 1296 | goto out_put; |
a116e05d ACM |
1297 | } |
1298 | if (add_sched_out_event(atoms, 'R', timestamp)) | |
b91fc39f | 1299 | goto out_put; |
55ffb7a6 MG |
1300 | } |
1301 | ||
1302 | BUG_ON(list_empty(&atoms->work_list)); | |
1303 | ||
1304 | atom = list_entry(atoms->work_list.prev, struct work_atom, list); | |
1305 | atom->sched_in_time = atom->sched_out_time = atom->wake_up_time = timestamp; | |
1306 | ||
0e9b07e5 | 1307 | sched->nr_timestamps++; |
55ffb7a6 MG |
1308 | |
1309 | if (atom->sched_out_time > timestamp) | |
0e9b07e5 | 1310 | sched->nr_unordered_timestamps++; |
b91fc39f ACM |
1311 | err = 0; |
1312 | out_put: | |
1313 | thread__put(migrant); | |
1314 | return err; | |
55ffb7a6 MG |
1315 | } |
1316 | ||
0e9b07e5 | 1317 | static void output_lat_thread(struct perf_sched *sched, struct work_atoms *work_list) |
cdce9d73 | 1318 | { |
cdce9d73 FW |
1319 | int i; |
1320 | int ret; | |
66685678 | 1321 | u64 avg; |
99620a5d | 1322 | char max_lat_at[32]; |
cdce9d73 | 1323 | |
39aeb52f | 1324 | if (!work_list->nb_atoms) |
cdce9d73 | 1325 | return; |
ea57c4f5 IM |
1326 | /* |
1327 | * Ignore idle threads: | |
1328 | */ | |
b9c5143a | 1329 | if (!strcmp(thread__comm_str(work_list->thread), "swapper")) |
ea57c4f5 | 1330 | return; |
cdce9d73 | 1331 | |
0e9b07e5 ACM |
1332 | sched->all_runtime += work_list->total_runtime; |
1333 | sched->all_count += work_list->nb_atoms; | |
66685678 | 1334 | |
2f80dd44 JB |
1335 | if (work_list->num_merged > 1) |
1336 | ret = printf(" %s:(%d) ", thread__comm_str(work_list->thread), work_list->num_merged); | |
1337 | else | |
1338 | ret = printf(" %s:%d ", thread__comm_str(work_list->thread), work_list->thread->tid); | |
cdce9d73 | 1339 | |
08f69e6c | 1340 | for (i = 0; i < 24 - ret; i++) |
cdce9d73 FW |
1341 | printf(" "); |
1342 | ||
39aeb52f | 1343 | avg = work_list->total_lat / work_list->nb_atoms; |
99620a5d | 1344 | timestamp__scnprintf_usec(work_list->max_lat_at, max_lat_at, sizeof(max_lat_at)); |
cdce9d73 | 1345 | |
99620a5d | 1346 | printf("|%11.3f ms |%9" PRIu64 " | avg:%9.3f ms | max:%9.3f ms | max at: %13s s\n", |
4fc76e49 ACM |
1347 | (double)work_list->total_runtime / NSEC_PER_MSEC, |
1348 | work_list->nb_atoms, (double)avg / NSEC_PER_MSEC, | |
1349 | (double)work_list->max_lat / NSEC_PER_MSEC, | |
99620a5d | 1350 | max_lat_at); |
cdce9d73 FW |
1351 | } |
1352 | ||
39aeb52f | 1353 | static int pid_cmp(struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 | 1354 | { |
0014de17 JO |
1355 | if (l->thread == r->thread) |
1356 | return 0; | |
38051234 | 1357 | if (l->thread->tid < r->thread->tid) |
daa1d7a5 | 1358 | return -1; |
38051234 | 1359 | if (l->thread->tid > r->thread->tid) |
daa1d7a5 | 1360 | return 1; |
0014de17 | 1361 | return (int)(l->thread - r->thread); |
daa1d7a5 FW |
1362 | } |
1363 | ||
39aeb52f | 1364 | static int avg_cmp(struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 FW |
1365 | { |
1366 | u64 avgl, avgr; | |
1367 | ||
1368 | if (!l->nb_atoms) | |
1369 | return -1; | |
1370 | ||
1371 | if (!r->nb_atoms) | |
1372 | return 1; | |
1373 | ||
1374 | avgl = l->total_lat / l->nb_atoms; | |
1375 | avgr = r->total_lat / r->nb_atoms; | |
1376 | ||
1377 | if (avgl < avgr) | |
1378 | return -1; | |
1379 | if (avgl > avgr) | |
1380 | return 1; | |
1381 | ||
1382 | return 0; | |
1383 | } | |
1384 | ||
39aeb52f | 1385 | static int max_cmp(struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 FW |
1386 | { |
1387 | if (l->max_lat < r->max_lat) | |
1388 | return -1; | |
1389 | if (l->max_lat > r->max_lat) | |
1390 | return 1; | |
1391 | ||
1392 | return 0; | |
1393 | } | |
1394 | ||
39aeb52f | 1395 | static int switch_cmp(struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 FW |
1396 | { |
1397 | if (l->nb_atoms < r->nb_atoms) | |
1398 | return -1; | |
1399 | if (l->nb_atoms > r->nb_atoms) | |
1400 | return 1; | |
1401 | ||
1402 | return 0; | |
1403 | } | |
1404 | ||
39aeb52f | 1405 | static int runtime_cmp(struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 FW |
1406 | { |
1407 | if (l->total_runtime < r->total_runtime) | |
1408 | return -1; | |
1409 | if (l->total_runtime > r->total_runtime) | |
1410 | return 1; | |
1411 | ||
1412 | return 0; | |
1413 | } | |
1414 | ||
cbef79a8 | 1415 | static int sort_dimension__add(const char *tok, struct list_head *list) |
daa1d7a5 | 1416 | { |
0e9b07e5 ACM |
1417 | size_t i; |
1418 | static struct sort_dimension avg_sort_dimension = { | |
1419 | .name = "avg", | |
1420 | .cmp = avg_cmp, | |
1421 | }; | |
1422 | static struct sort_dimension max_sort_dimension = { | |
1423 | .name = "max", | |
1424 | .cmp = max_cmp, | |
1425 | }; | |
1426 | static struct sort_dimension pid_sort_dimension = { | |
1427 | .name = "pid", | |
1428 | .cmp = pid_cmp, | |
1429 | }; | |
1430 | static struct sort_dimension runtime_sort_dimension = { | |
1431 | .name = "runtime", | |
1432 | .cmp = runtime_cmp, | |
1433 | }; | |
1434 | static struct sort_dimension switch_sort_dimension = { | |
1435 | .name = "switch", | |
1436 | .cmp = switch_cmp, | |
1437 | }; | |
1438 | struct sort_dimension *available_sorts[] = { | |
1439 | &pid_sort_dimension, | |
1440 | &avg_sort_dimension, | |
1441 | &max_sort_dimension, | |
1442 | &switch_sort_dimension, | |
1443 | &runtime_sort_dimension, | |
1444 | }; | |
daa1d7a5 | 1445 | |
0e9b07e5 | 1446 | for (i = 0; i < ARRAY_SIZE(available_sorts); i++) { |
daa1d7a5 FW |
1447 | if (!strcmp(available_sorts[i]->name, tok)) { |
1448 | list_add_tail(&available_sorts[i]->list, list); | |
1449 | ||
1450 | return 0; | |
1451 | } | |
1452 | } | |
1453 | ||
1454 | return -1; | |
1455 | } | |
1456 | ||
0e9b07e5 | 1457 | static void perf_sched__sort_lat(struct perf_sched *sched) |
daa1d7a5 FW |
1458 | { |
1459 | struct rb_node *node; | |
cb4c13a5 | 1460 | struct rb_root_cached *root = &sched->atom_root; |
2f80dd44 | 1461 | again: |
daa1d7a5 | 1462 | for (;;) { |
39aeb52f | 1463 | struct work_atoms *data; |
cb4c13a5 | 1464 | node = rb_first_cached(root); |
daa1d7a5 FW |
1465 | if (!node) |
1466 | break; | |
1467 | ||
cb4c13a5 | 1468 | rb_erase_cached(node, root); |
39aeb52f | 1469 | data = rb_entry(node, struct work_atoms, node); |
0e9b07e5 | 1470 | __thread_latency_insert(&sched->sorted_atom_root, data, &sched->sort_list); |
daa1d7a5 | 1471 | } |
2f80dd44 JB |
1472 | if (root == &sched->atom_root) { |
1473 | root = &sched->merged_atom_root; | |
1474 | goto again; | |
1475 | } | |
daa1d7a5 FW |
1476 | } |
1477 | ||
0e9b07e5 | 1478 | static int process_sched_wakeup_event(struct perf_tool *tool, |
32dcd021 | 1479 | struct evsel *evsel, |
1d037ca1 | 1480 | struct perf_sample *sample, |
4218e673 | 1481 | struct machine *machine) |
419ab0d6 | 1482 | { |
0e9b07e5 | 1483 | struct perf_sched *sched = container_of(tool, struct perf_sched, tool); |
419ab0d6 | 1484 | |
9ec3f4e4 ACM |
1485 | if (sched->tp_handler->wakeup_event) |
1486 | return sched->tp_handler->wakeup_event(sched, evsel, sample, machine); | |
a116e05d | 1487 | |
2b7fcbc5 | 1488 | return 0; |
419ab0d6 FW |
1489 | } |
1490 | ||
a151a37a JO |
1491 | union map_priv { |
1492 | void *ptr; | |
1493 | bool color; | |
1494 | }; | |
1495 | ||
1496 | static bool thread__has_color(struct thread *thread) | |
1497 | { | |
1498 | union map_priv priv = { | |
1499 | .ptr = thread__priv(thread), | |
1500 | }; | |
1501 | ||
1502 | return priv.color; | |
1503 | } | |
1504 | ||
1505 | static struct thread* | |
1506 | map__findnew_thread(struct perf_sched *sched, struct machine *machine, pid_t pid, pid_t tid) | |
1507 | { | |
1508 | struct thread *thread = machine__findnew_thread(machine, pid, tid); | |
1509 | union map_priv priv = { | |
1510 | .color = false, | |
1511 | }; | |
1512 | ||
1513 | if (!sched->map.color_pids || !thread || thread__priv(thread)) | |
1514 | return thread; | |
1515 | ||
1516 | if (thread_map__has(sched->map.color_pids, tid)) | |
1517 | priv.color = true; | |
1518 | ||
1519 | thread__set_priv(thread, priv.ptr); | |
1520 | return thread; | |
1521 | } | |
1522 | ||
32dcd021 | 1523 | static int map_switch_event(struct perf_sched *sched, struct evsel *evsel, |
9ec3f4e4 | 1524 | struct perf_sample *sample, struct machine *machine) |
0ec04e16 | 1525 | { |
9d372ca5 DY |
1526 | const u32 next_pid = perf_evsel__intval(evsel, sample, "next_pid"); |
1527 | struct thread *sched_in; | |
8640da9f | 1528 | struct thread_runtime *tr; |
0ec04e16 | 1529 | int new_shortname; |
7f7f8d0b | 1530 | u64 timestamp0, timestamp = sample->time; |
0ec04e16 | 1531 | s64 delta; |
99623c62 JO |
1532 | int i, this_cpu = sample->cpu; |
1533 | int cpus_nr; | |
1534 | bool new_cpu = false; | |
8cd91195 | 1535 | const char *color = PERF_COLOR_NORMAL; |
99620a5d | 1536 | char stimestamp[32]; |
0ec04e16 IM |
1537 | |
1538 | BUG_ON(this_cpu >= MAX_CPUS || this_cpu < 0); | |
1539 | ||
0e9b07e5 ACM |
1540 | if (this_cpu > sched->max_cpu) |
1541 | sched->max_cpu = this_cpu; | |
0ec04e16 | 1542 | |
99623c62 JO |
1543 | if (sched->map.comp) { |
1544 | cpus_nr = bitmap_weight(sched->map.comp_cpus_mask, MAX_CPUS); | |
1545 | if (!test_and_set_bit(this_cpu, sched->map.comp_cpus_mask)) { | |
1546 | sched->map.comp_cpus[cpus_nr++] = this_cpu; | |
1547 | new_cpu = true; | |
1548 | } | |
1549 | } else | |
1550 | cpus_nr = sched->max_cpu; | |
1551 | ||
0e9b07e5 ACM |
1552 | timestamp0 = sched->cpu_last_switched[this_cpu]; |
1553 | sched->cpu_last_switched[this_cpu] = timestamp; | |
0ec04e16 IM |
1554 | if (timestamp0) |
1555 | delta = timestamp - timestamp0; | |
1556 | else | |
1557 | delta = 0; | |
1558 | ||
a116e05d | 1559 | if (delta < 0) { |
60b7d14a | 1560 | pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta); |
a116e05d ACM |
1561 | return -1; |
1562 | } | |
0ec04e16 | 1563 | |
a151a37a | 1564 | sched_in = map__findnew_thread(sched, machine, -1, next_pid); |
b91fc39f ACM |
1565 | if (sched_in == NULL) |
1566 | return -1; | |
0ec04e16 | 1567 | |
8640da9f CD |
1568 | tr = thread__get_runtime(sched_in); |
1569 | if (tr == NULL) { | |
1570 | thread__put(sched_in); | |
1571 | return -1; | |
1572 | } | |
1573 | ||
b91fc39f | 1574 | sched->curr_thread[this_cpu] = thread__get(sched_in); |
0ec04e16 IM |
1575 | |
1576 | printf(" "); | |
1577 | ||
1578 | new_shortname = 0; | |
8640da9f | 1579 | if (!tr->shortname[0]) { |
6bcab4e1 D |
1580 | if (!strcmp(thread__comm_str(sched_in), "swapper")) { |
1581 | /* | |
1582 | * Don't allocate a letter-number for swapper:0 | |
1583 | * as a shortname. Instead, we use '.' for it. | |
1584 | */ | |
8640da9f CD |
1585 | tr->shortname[0] = '.'; |
1586 | tr->shortname[1] = ' '; | |
0ec04e16 | 1587 | } else { |
8640da9f CD |
1588 | tr->shortname[0] = sched->next_shortname1; |
1589 | tr->shortname[1] = sched->next_shortname2; | |
6bcab4e1 D |
1590 | |
1591 | if (sched->next_shortname1 < 'Z') { | |
1592 | sched->next_shortname1++; | |
0ec04e16 | 1593 | } else { |
6bcab4e1 D |
1594 | sched->next_shortname1 = 'A'; |
1595 | if (sched->next_shortname2 < '9') | |
1596 | sched->next_shortname2++; | |
1597 | else | |
1598 | sched->next_shortname2 = '0'; | |
0ec04e16 IM |
1599 | } |
1600 | } | |
1601 | new_shortname = 1; | |
1602 | } | |
1603 | ||
99623c62 JO |
1604 | for (i = 0; i < cpus_nr; i++) { |
1605 | int cpu = sched->map.comp ? sched->map.comp_cpus[i] : i; | |
a151a37a | 1606 | struct thread *curr_thread = sched->curr_thread[cpu]; |
8640da9f | 1607 | struct thread_runtime *curr_tr; |
a151a37a | 1608 | const char *pid_color = color; |
cf294f24 | 1609 | const char *cpu_color = color; |
a151a37a JO |
1610 | |
1611 | if (curr_thread && thread__has_color(curr_thread)) | |
1612 | pid_color = COLOR_PIDS; | |
99623c62 | 1613 | |
73643bb6 JO |
1614 | if (sched->map.cpus && !cpu_map__has(sched->map.cpus, cpu)) |
1615 | continue; | |
1616 | ||
cf294f24 JO |
1617 | if (sched->map.color_cpus && cpu_map__has(sched->map.color_cpus, cpu)) |
1618 | cpu_color = COLOR_CPUS; | |
1619 | ||
0ec04e16 | 1620 | if (cpu != this_cpu) |
1208bb27 | 1621 | color_fprintf(stdout, color, " "); |
0ec04e16 | 1622 | else |
cf294f24 | 1623 | color_fprintf(stdout, cpu_color, "*"); |
0ec04e16 | 1624 | |
8640da9f CD |
1625 | if (sched->curr_thread[cpu]) { |
1626 | curr_tr = thread__get_runtime(sched->curr_thread[cpu]); | |
1627 | if (curr_tr == NULL) { | |
1628 | thread__put(sched_in); | |
1629 | return -1; | |
1630 | } | |
1631 | color_fprintf(stdout, pid_color, "%2s ", curr_tr->shortname); | |
1632 | } else | |
8cd91195 | 1633 | color_fprintf(stdout, color, " "); |
0ec04e16 IM |
1634 | } |
1635 | ||
73643bb6 JO |
1636 | if (sched->map.cpus && !cpu_map__has(sched->map.cpus, this_cpu)) |
1637 | goto out; | |
1638 | ||
99620a5d NK |
1639 | timestamp__scnprintf_usec(timestamp, stimestamp, sizeof(stimestamp)); |
1640 | color_fprintf(stdout, color, " %12s secs ", stimestamp); | |
99a3c3a9 | 1641 | if (new_shortname || tr->comm_changed || (verbose > 0 && sched_in->tid)) { |
a151a37a JO |
1642 | const char *pid_color = color; |
1643 | ||
1644 | if (thread__has_color(sched_in)) | |
1645 | pid_color = COLOR_PIDS; | |
1646 | ||
1647 | color_fprintf(stdout, pid_color, "%s => %s:%d", | |
8640da9f | 1648 | tr->shortname, thread__comm_str(sched_in), sched_in->tid); |
99a3c3a9 | 1649 | tr->comm_changed = false; |
0ec04e16 | 1650 | } |
a116e05d | 1651 | |
99623c62 | 1652 | if (sched->map.comp && new_cpu) |
8cd91195 | 1653 | color_fprintf(stdout, color, " (CPU %d)", this_cpu); |
99623c62 | 1654 | |
73643bb6 | 1655 | out: |
8cd91195 | 1656 | color_fprintf(stdout, color, "\n"); |
99623c62 | 1657 | |
b91fc39f ACM |
1658 | thread__put(sched_in); |
1659 | ||
a116e05d | 1660 | return 0; |
0ec04e16 IM |
1661 | } |
1662 | ||
0e9b07e5 | 1663 | static int process_sched_switch_event(struct perf_tool *tool, |
32dcd021 | 1664 | struct evsel *evsel, |
1d037ca1 | 1665 | struct perf_sample *sample, |
4218e673 | 1666 | struct machine *machine) |
419ab0d6 | 1667 | { |
0e9b07e5 | 1668 | struct perf_sched *sched = container_of(tool, struct perf_sched, tool); |
a116e05d | 1669 | int this_cpu = sample->cpu, err = 0; |
2b7fcbc5 ACM |
1670 | u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"), |
1671 | next_pid = perf_evsel__intval(evsel, sample, "next_pid"); | |
419ab0d6 | 1672 | |
0e9b07e5 | 1673 | if (sched->curr_pid[this_cpu] != (u32)-1) { |
c8a37751 IM |
1674 | /* |
1675 | * Are we trying to switch away a PID that is | |
1676 | * not current? | |
1677 | */ | |
2b7fcbc5 | 1678 | if (sched->curr_pid[this_cpu] != prev_pid) |
0e9b07e5 | 1679 | sched->nr_context_switch_bugs++; |
c8a37751 | 1680 | } |
c8a37751 | 1681 | |
9ec3f4e4 ACM |
1682 | if (sched->tp_handler->switch_event) |
1683 | err = sched->tp_handler->switch_event(sched, evsel, sample, machine); | |
2b7fcbc5 ACM |
1684 | |
1685 | sched->curr_pid[this_cpu] = next_pid; | |
a116e05d | 1686 | return err; |
419ab0d6 FW |
1687 | } |
1688 | ||
0e9b07e5 | 1689 | static int process_sched_runtime_event(struct perf_tool *tool, |
32dcd021 | 1690 | struct evsel *evsel, |
1d037ca1 | 1691 | struct perf_sample *sample, |
4218e673 | 1692 | struct machine *machine) |
39aeb52f | 1693 | { |
0e9b07e5 | 1694 | struct perf_sched *sched = container_of(tool, struct perf_sched, tool); |
39aeb52f | 1695 | |
9ec3f4e4 ACM |
1696 | if (sched->tp_handler->runtime_event) |
1697 | return sched->tp_handler->runtime_event(sched, evsel, sample, machine); | |
a116e05d | 1698 | |
2b7fcbc5 | 1699 | return 0; |
39aeb52f | 1700 | } |
1701 | ||
cb627505 DA |
1702 | static int perf_sched__process_fork_event(struct perf_tool *tool, |
1703 | union perf_event *event, | |
1704 | struct perf_sample *sample, | |
1705 | struct machine *machine) | |
fbf94829 | 1706 | { |
0e9b07e5 | 1707 | struct perf_sched *sched = container_of(tool, struct perf_sched, tool); |
46538818 | 1708 | |
cb627505 DA |
1709 | /* run the fork event through the perf machineruy */ |
1710 | perf_event__process_fork(tool, event, sample, machine); | |
1711 | ||
1712 | /* and then run additional processing needed for this command */ | |
9ec3f4e4 | 1713 | if (sched->tp_handler->fork_event) |
cb627505 | 1714 | return sched->tp_handler->fork_event(sched, event, machine); |
a116e05d | 1715 | |
2b7fcbc5 | 1716 | return 0; |
fbf94829 IM |
1717 | } |
1718 | ||
0e9b07e5 | 1719 | static int process_sched_migrate_task_event(struct perf_tool *tool, |
32dcd021 | 1720 | struct evsel *evsel, |
1d037ca1 | 1721 | struct perf_sample *sample, |
4218e673 | 1722 | struct machine *machine) |
55ffb7a6 | 1723 | { |
0e9b07e5 | 1724 | struct perf_sched *sched = container_of(tool, struct perf_sched, tool); |
55ffb7a6 | 1725 | |
9ec3f4e4 ACM |
1726 | if (sched->tp_handler->migrate_task_event) |
1727 | return sched->tp_handler->migrate_task_event(sched, evsel, sample, machine); | |
a116e05d | 1728 | |
2b7fcbc5 | 1729 | return 0; |
55ffb7a6 MG |
1730 | } |
1731 | ||
a116e05d | 1732 | typedef int (*tracepoint_handler)(struct perf_tool *tool, |
32dcd021 | 1733 | struct evsel *evsel, |
a116e05d | 1734 | struct perf_sample *sample, |
4218e673 | 1735 | struct machine *machine); |
ec156764 | 1736 | |
1d037ca1 IT |
1737 | static int perf_sched__process_tracepoint_sample(struct perf_tool *tool __maybe_unused, |
1738 | union perf_event *event __maybe_unused, | |
ee29be62 | 1739 | struct perf_sample *sample, |
32dcd021 | 1740 | struct evsel *evsel, |
ee29be62 | 1741 | struct machine *machine) |
0a02ad93 | 1742 | { |
a116e05d | 1743 | int err = 0; |
0a02ad93 | 1744 | |
744a9719 ACM |
1745 | if (evsel->handler != NULL) { |
1746 | tracepoint_handler f = evsel->handler; | |
2b7fcbc5 | 1747 | err = f(tool, evsel, sample, machine); |
ee29be62 | 1748 | } |
0a02ad93 | 1749 | |
a116e05d | 1750 | return err; |
0a02ad93 IM |
1751 | } |
1752 | ||
99a3c3a9 CD |
1753 | static int perf_sched__process_comm(struct perf_tool *tool __maybe_unused, |
1754 | union perf_event *event, | |
1755 | struct perf_sample *sample, | |
1756 | struct machine *machine) | |
1757 | { | |
1758 | struct thread *thread; | |
1759 | struct thread_runtime *tr; | |
1760 | int err; | |
1761 | ||
1762 | err = perf_event__process_comm(tool, event, sample, machine); | |
1763 | if (err) | |
1764 | return err; | |
1765 | ||
1766 | thread = machine__find_thread(machine, sample->pid, sample->tid); | |
1767 | if (!thread) { | |
1768 | pr_err("Internal error: can't find thread\n"); | |
1769 | return -1; | |
1770 | } | |
1771 | ||
1772 | tr = thread__get_runtime(thread); | |
1773 | if (tr == NULL) { | |
1774 | thread__put(thread); | |
1775 | return -1; | |
1776 | } | |
1777 | ||
1778 | tr->comm_changed = true; | |
1779 | thread__put(thread); | |
1780 | ||
1781 | return 0; | |
1782 | } | |
1783 | ||
ae536acf | 1784 | static int perf_sched__read_events(struct perf_sched *sched) |
0a02ad93 | 1785 | { |
32dcd021 | 1786 | const struct evsel_str_handler handlers[] = { |
ee29be62 ACM |
1787 | { "sched:sched_switch", process_sched_switch_event, }, |
1788 | { "sched:sched_stat_runtime", process_sched_runtime_event, }, | |
1789 | { "sched:sched_wakeup", process_sched_wakeup_event, }, | |
1790 | { "sched:sched_wakeup_new", process_sched_wakeup_event, }, | |
ee29be62 ACM |
1791 | { "sched:sched_migrate_task", process_sched_migrate_task_event, }, |
1792 | }; | |
da378962 | 1793 | struct perf_session *session; |
8ceb41d7 | 1794 | struct perf_data data = { |
2d4f2799 JO |
1795 | .path = input_name, |
1796 | .mode = PERF_DATA_MODE_READ, | |
1797 | .force = sched->force, | |
f5fc1412 | 1798 | }; |
ae536acf | 1799 | int rc = -1; |
da378962 | 1800 | |
8ceb41d7 | 1801 | session = perf_session__new(&data, false, &sched->tool); |
6ef81c55 MI |
1802 | if (IS_ERR(session)) { |
1803 | pr_debug("Error creating perf session"); | |
1804 | return PTR_ERR(session); | |
a116e05d | 1805 | } |
94c744b6 | 1806 | |
0a7e6d1b | 1807 | symbol__init(&session->header.env); |
04934106 | 1808 | |
a116e05d ACM |
1809 | if (perf_session__set_tracepoints_handlers(session, handlers)) |
1810 | goto out_delete; | |
ee29be62 | 1811 | |
cee75ac7 | 1812 | if (perf_session__has_traces(session, "record -R")) { |
b7b61cbe | 1813 | int err = perf_session__process_events(session); |
a116e05d ACM |
1814 | if (err) { |
1815 | pr_err("Failed to process events, error %d", err); | |
1816 | goto out_delete; | |
1817 | } | |
4c09bafa | 1818 | |
75be989a ACM |
1819 | sched->nr_events = session->evlist->stats.nr_events[0]; |
1820 | sched->nr_lost_events = session->evlist->stats.total_lost; | |
1821 | sched->nr_lost_chunks = session->evlist->stats.nr_events[PERF_RECORD_LOST]; | |
cee75ac7 | 1822 | } |
d549c769 | 1823 | |
ae536acf | 1824 | rc = 0; |
a116e05d ACM |
1825 | out_delete: |
1826 | perf_session__delete(session); | |
ae536acf | 1827 | return rc; |
0a02ad93 IM |
1828 | } |
1829 | ||
49394a2a DA |
1830 | /* |
1831 | * scheduling times are printed as msec.usec | |
1832 | */ | |
1833 | static inline void print_sched_time(unsigned long long nsecs, int width) | |
1834 | { | |
1835 | unsigned long msecs; | |
1836 | unsigned long usecs; | |
1837 | ||
1838 | msecs = nsecs / NSEC_PER_MSEC; | |
1839 | nsecs -= msecs * NSEC_PER_MSEC; | |
1840 | usecs = nsecs / NSEC_PER_USEC; | |
1841 | printf("%*lu.%03lu ", width, msecs, usecs); | |
1842 | } | |
1843 | ||
1844 | /* | |
1845 | * returns runtime data for event, allocating memory for it the | |
1846 | * first time it is used. | |
1847 | */ | |
32dcd021 | 1848 | static struct evsel_runtime *perf_evsel__get_runtime(struct evsel *evsel) |
49394a2a DA |
1849 | { |
1850 | struct evsel_runtime *r = evsel->priv; | |
1851 | ||
1852 | if (r == NULL) { | |
1853 | r = zalloc(sizeof(struct evsel_runtime)); | |
1854 | evsel->priv = r; | |
1855 | } | |
1856 | ||
1857 | return r; | |
1858 | } | |
1859 | ||
1860 | /* | |
1861 | * save last time event was seen per cpu | |
1862 | */ | |
32dcd021 | 1863 | static void perf_evsel__save_time(struct evsel *evsel, |
49394a2a DA |
1864 | u64 timestamp, u32 cpu) |
1865 | { | |
1866 | struct evsel_runtime *r = perf_evsel__get_runtime(evsel); | |
1867 | ||
1868 | if (r == NULL) | |
1869 | return; | |
1870 | ||
1871 | if ((cpu >= r->ncpu) || (r->last_time == NULL)) { | |
1872 | int i, n = __roundup_pow_of_two(cpu+1); | |
1873 | void *p = r->last_time; | |
1874 | ||
1875 | p = realloc(r->last_time, n * sizeof(u64)); | |
1876 | if (!p) | |
1877 | return; | |
1878 | ||
1879 | r->last_time = p; | |
1880 | for (i = r->ncpu; i < n; ++i) | |
1881 | r->last_time[i] = (u64) 0; | |
1882 | ||
1883 | r->ncpu = n; | |
1884 | } | |
1885 | ||
1886 | r->last_time[cpu] = timestamp; | |
1887 | } | |
1888 | ||
1889 | /* returns last time this event was seen on the given cpu */ | |
32dcd021 | 1890 | static u64 perf_evsel__get_time(struct evsel *evsel, u32 cpu) |
49394a2a DA |
1891 | { |
1892 | struct evsel_runtime *r = perf_evsel__get_runtime(evsel); | |
1893 | ||
1894 | if ((r == NULL) || (r->last_time == NULL) || (cpu >= r->ncpu)) | |
1895 | return 0; | |
1896 | ||
1897 | return r->last_time[cpu]; | |
1898 | } | |
1899 | ||
9b8087d7 | 1900 | static int comm_width = 30; |
49394a2a DA |
1901 | |
1902 | static char *timehist_get_commstr(struct thread *thread) | |
1903 | { | |
1904 | static char str[32]; | |
1905 | const char *comm = thread__comm_str(thread); | |
1906 | pid_t tid = thread->tid; | |
1907 | pid_t pid = thread->pid_; | |
1908 | int n; | |
1909 | ||
1910 | if (pid == 0) | |
1911 | n = scnprintf(str, sizeof(str), "%s", comm); | |
1912 | ||
1913 | else if (tid != pid) | |
1914 | n = scnprintf(str, sizeof(str), "%s[%d/%d]", comm, tid, pid); | |
1915 | ||
1916 | else | |
1917 | n = scnprintf(str, sizeof(str), "%s[%d]", comm, tid); | |
1918 | ||
1919 | if (n > comm_width) | |
1920 | comm_width = n; | |
1921 | ||
1922 | return str; | |
1923 | } | |
1924 | ||
a407b067 | 1925 | static void timehist_header(struct perf_sched *sched) |
49394a2a | 1926 | { |
a407b067 DA |
1927 | u32 ncpus = sched->max_cpu + 1; |
1928 | u32 i, j; | |
1929 | ||
49394a2a DA |
1930 | printf("%15s %6s ", "time", "cpu"); |
1931 | ||
a407b067 DA |
1932 | if (sched->show_cpu_visual) { |
1933 | printf(" "); | |
1934 | for (i = 0, j = 0; i < ncpus; ++i) { | |
1935 | printf("%x", j++); | |
1936 | if (j > 15) | |
1937 | j = 0; | |
1938 | } | |
1939 | printf(" "); | |
1940 | } | |
1941 | ||
0e6758e8 | 1942 | printf(" %-*s %9s %9s %9s", comm_width, |
49394a2a DA |
1943 | "task name", "wait time", "sch delay", "run time"); |
1944 | ||
414e050c NK |
1945 | if (sched->show_state) |
1946 | printf(" %s", "state"); | |
1947 | ||
49394a2a DA |
1948 | printf("\n"); |
1949 | ||
1950 | /* | |
1951 | * units row | |
1952 | */ | |
1953 | printf("%15s %-6s ", "", ""); | |
1954 | ||
a407b067 DA |
1955 | if (sched->show_cpu_visual) |
1956 | printf(" %*s ", ncpus, ""); | |
1957 | ||
414e050c | 1958 | printf(" %-*s %9s %9s %9s", comm_width, |
0e6758e8 | 1959 | "[tid/pid]", "(msec)", "(msec)", "(msec)"); |
49394a2a | 1960 | |
414e050c NK |
1961 | if (sched->show_state) |
1962 | printf(" %5s", ""); | |
1963 | ||
1964 | printf("\n"); | |
1965 | ||
49394a2a DA |
1966 | /* |
1967 | * separator | |
1968 | */ | |
1969 | printf("%.15s %.6s ", graph_dotted_line, graph_dotted_line); | |
1970 | ||
a407b067 DA |
1971 | if (sched->show_cpu_visual) |
1972 | printf(" %.*s ", ncpus, graph_dotted_line); | |
1973 | ||
0e6758e8 | 1974 | printf(" %.*s %.9s %.9s %.9s", comm_width, |
49394a2a DA |
1975 | graph_dotted_line, graph_dotted_line, graph_dotted_line, |
1976 | graph_dotted_line); | |
1977 | ||
414e050c NK |
1978 | if (sched->show_state) |
1979 | printf(" %.5s", graph_dotted_line); | |
1980 | ||
49394a2a DA |
1981 | printf("\n"); |
1982 | } | |
1983 | ||
414e050c NK |
1984 | static char task_state_char(struct thread *thread, int state) |
1985 | { | |
1986 | static const char state_to_char[] = TASK_STATE_TO_CHAR_STR; | |
1987 | unsigned bit = state ? ffs(state) : 0; | |
1988 | ||
1989 | /* 'I' for idle */ | |
1990 | if (thread->tid == 0) | |
1991 | return 'I'; | |
1992 | ||
1993 | return bit < sizeof(state_to_char) - 1 ? state_to_char[bit] : '?'; | |
1994 | } | |
1995 | ||
fc1469f1 | 1996 | static void timehist_print_sample(struct perf_sched *sched, |
32dcd021 | 1997 | struct evsel *evsel, |
fc1469f1 | 1998 | struct perf_sample *sample, |
6c973c90 | 1999 | struct addr_location *al, |
853b7407 | 2000 | struct thread *thread, |
414e050c | 2001 | u64 t, int state) |
49394a2a DA |
2002 | { |
2003 | struct thread_runtime *tr = thread__priv(thread); | |
292c4a8f BG |
2004 | const char *next_comm = perf_evsel__strval(evsel, sample, "next_comm"); |
2005 | const u32 next_pid = perf_evsel__intval(evsel, sample, "next_pid"); | |
a407b067 | 2006 | u32 max_cpus = sched->max_cpu + 1; |
49394a2a | 2007 | char tstr[64]; |
292c4a8f | 2008 | char nstr[30]; |
941bdea7 | 2009 | u64 wait_time; |
49394a2a | 2010 | |
853b7407 | 2011 | timestamp__scnprintf_usec(t, tstr, sizeof(tstr)); |
49394a2a DA |
2012 | printf("%15s [%04d] ", tstr, sample->cpu); |
2013 | ||
a407b067 DA |
2014 | if (sched->show_cpu_visual) { |
2015 | u32 i; | |
2016 | char c; | |
2017 | ||
2018 | printf(" "); | |
2019 | for (i = 0; i < max_cpus; ++i) { | |
2020 | /* flag idle times with 'i'; others are sched events */ | |
2021 | if (i == sample->cpu) | |
2022 | c = (thread->tid == 0) ? 'i' : 's'; | |
2023 | else | |
2024 | c = ' '; | |
2025 | printf("%c", c); | |
2026 | } | |
2027 | printf(" "); | |
2028 | } | |
2029 | ||
49394a2a DA |
2030 | printf(" %-*s ", comm_width, timehist_get_commstr(thread)); |
2031 | ||
941bdea7 NK |
2032 | wait_time = tr->dt_sleep + tr->dt_iowait + tr->dt_preempt; |
2033 | print_sched_time(wait_time, 6); | |
2034 | ||
49394a2a DA |
2035 | print_sched_time(tr->dt_delay, 6); |
2036 | print_sched_time(tr->dt_run, 6); | |
fc1469f1 | 2037 | |
414e050c NK |
2038 | if (sched->show_state) |
2039 | printf(" %5c ", task_state_char(thread, state)); | |
2040 | ||
292c4a8f BG |
2041 | if (sched->show_next) { |
2042 | snprintf(nstr, sizeof(nstr), "next: %s[%d]", next_comm, next_pid); | |
2043 | printf(" %-*s", comm_width, nstr); | |
2044 | } | |
2045 | ||
2046 | if (sched->show_wakeups && !sched->show_next) | |
fc1469f1 DA |
2047 | printf(" %-*s", comm_width, ""); |
2048 | ||
6c973c90 DA |
2049 | if (thread->tid == 0) |
2050 | goto out; | |
2051 | ||
2052 | if (sched->show_callchain) | |
2053 | printf(" "); | |
2054 | ||
2055 | sample__fprintf_sym(sample, al, 0, | |
2056 | EVSEL__PRINT_SYM | EVSEL__PRINT_ONELINE | | |
2d9bbf6e NK |
2057 | EVSEL__PRINT_CALLCHAIN_ARROW | |
2058 | EVSEL__PRINT_SKIP_IGNORED, | |
9620bc36 | 2059 | &callchain_cursor, symbol_conf.bt_stop_list, stdout); |
6c973c90 DA |
2060 | |
2061 | out: | |
49394a2a DA |
2062 | printf("\n"); |
2063 | } | |
2064 | ||
2065 | /* | |
2066 | * Explanation of delta-time stats: | |
2067 | * | |
2068 | * t = time of current schedule out event | |
2069 | * tprev = time of previous sched out event | |
2070 | * also time of schedule-in event for current task | |
2071 | * last_time = time of last sched change event for current task | |
2072 | * (i.e, time process was last scheduled out) | |
2073 | * ready_to_run = time of wakeup for current task | |
2074 | * | |
2075 | * -----|------------|------------|------------|------ | |
2076 | * last ready tprev t | |
2077 | * time to run | |
2078 | * | |
2079 | * |-------- dt_wait --------| | |
2080 | * |- dt_delay -|-- dt_run --| | |
2081 | * | |
2082 | * dt_run = run time of current task | |
2083 | * dt_wait = time between last schedule out event for task and tprev | |
2084 | * represents time spent off the cpu | |
2085 | * dt_delay = time between wakeup and schedule-in of task | |
2086 | */ | |
2087 | ||
2088 | static void timehist_update_runtime_stats(struct thread_runtime *r, | |
2089 | u64 t, u64 tprev) | |
2090 | { | |
2091 | r->dt_delay = 0; | |
941bdea7 NK |
2092 | r->dt_sleep = 0; |
2093 | r->dt_iowait = 0; | |
2094 | r->dt_preempt = 0; | |
49394a2a | 2095 | r->dt_run = 0; |
941bdea7 | 2096 | |
49394a2a DA |
2097 | if (tprev) { |
2098 | r->dt_run = t - tprev; | |
2099 | if (r->ready_to_run) { | |
2100 | if (r->ready_to_run > tprev) | |
2101 | pr_debug("time travel: wakeup time for task > previous sched_switch event\n"); | |
2102 | else | |
2103 | r->dt_delay = tprev - r->ready_to_run; | |
2104 | } | |
2105 | ||
2106 | if (r->last_time > tprev) | |
2107 | pr_debug("time travel: last sched out time for task > previous sched_switch event\n"); | |
941bdea7 NK |
2108 | else if (r->last_time) { |
2109 | u64 dt_wait = tprev - r->last_time; | |
2110 | ||
2111 | if (r->last_state == TASK_RUNNING) | |
2112 | r->dt_preempt = dt_wait; | |
2113 | else if (r->last_state == TASK_UNINTERRUPTIBLE) | |
2114 | r->dt_iowait = dt_wait; | |
2115 | else | |
2116 | r->dt_sleep = dt_wait; | |
2117 | } | |
49394a2a DA |
2118 | } |
2119 | ||
2120 | update_stats(&r->run_stats, r->dt_run); | |
587782c5 NK |
2121 | |
2122 | r->total_run_time += r->dt_run; | |
2123 | r->total_delay_time += r->dt_delay; | |
2124 | r->total_sleep_time += r->dt_sleep; | |
2125 | r->total_iowait_time += r->dt_iowait; | |
2126 | r->total_preempt_time += r->dt_preempt; | |
49394a2a DA |
2127 | } |
2128 | ||
96039c7c | 2129 | static bool is_idle_sample(struct perf_sample *sample, |
32dcd021 | 2130 | struct evsel *evsel) |
49394a2a DA |
2131 | { |
2132 | /* pid 0 == swapper == idle task */ | |
96039c7c NK |
2133 | if (strcmp(perf_evsel__name(evsel), "sched:sched_switch") == 0) |
2134 | return perf_evsel__intval(evsel, sample, "prev_pid") == 0; | |
49394a2a | 2135 | |
96039c7c NK |
2136 | return sample->pid == 0; |
2137 | } | |
2138 | ||
2139 | static void save_task_callchain(struct perf_sched *sched, | |
2140 | struct perf_sample *sample, | |
32dcd021 | 2141 | struct evsel *evsel, |
96039c7c NK |
2142 | struct machine *machine) |
2143 | { | |
2144 | struct callchain_cursor *cursor = &callchain_cursor; | |
2145 | struct thread *thread; | |
6c973c90 DA |
2146 | |
2147 | /* want main thread for process - has maps */ | |
2148 | thread = machine__findnew_thread(machine, sample->pid, sample->pid); | |
2149 | if (thread == NULL) { | |
2150 | pr_debug("Failed to get thread for pid %d.\n", sample->pid); | |
96039c7c | 2151 | return; |
6c973c90 DA |
2152 | } |
2153 | ||
4c50563d | 2154 | if (!sched->show_callchain || sample->callchain == NULL) |
96039c7c | 2155 | return; |
6c973c90 DA |
2156 | |
2157 | if (thread__resolve_callchain(thread, cursor, evsel, sample, | |
8388deb3 | 2158 | NULL, NULL, sched->max_stack + 2) != 0) { |
bb963e16 | 2159 | if (verbose > 0) |
62d94b00 | 2160 | pr_err("Failed to resolve callchain. Skipping\n"); |
6c973c90 | 2161 | |
96039c7c | 2162 | return; |
6c973c90 | 2163 | } |
cdeb01bf | 2164 | |
6c973c90 | 2165 | callchain_cursor_commit(cursor); |
cdeb01bf NK |
2166 | |
2167 | while (true) { | |
2168 | struct callchain_cursor_node *node; | |
2169 | struct symbol *sym; | |
2170 | ||
2171 | node = callchain_cursor_current(cursor); | |
2172 | if (node == NULL) | |
2173 | break; | |
2174 | ||
2175 | sym = node->sym; | |
a7c3899c | 2176 | if (sym) { |
cdeb01bf NK |
2177 | if (!strcmp(sym->name, "schedule") || |
2178 | !strcmp(sym->name, "__schedule") || | |
2179 | !strcmp(sym->name, "preempt_schedule")) | |
2180 | sym->ignore = 1; | |
2181 | } | |
2182 | ||
2183 | callchain_cursor_advance(cursor); | |
2184 | } | |
49394a2a DA |
2185 | } |
2186 | ||
3bc2fa9c NK |
2187 | static int init_idle_thread(struct thread *thread) |
2188 | { | |
2189 | struct idle_thread_runtime *itr; | |
2190 | ||
2191 | thread__set_comm(thread, idle_comm, 0); | |
2192 | ||
2193 | itr = zalloc(sizeof(*itr)); | |
2194 | if (itr == NULL) | |
2195 | return -ENOMEM; | |
2196 | ||
2197 | init_stats(&itr->tr.run_stats); | |
2198 | callchain_init(&itr->callchain); | |
2199 | callchain_cursor_reset(&itr->cursor); | |
2200 | thread__set_priv(thread, itr); | |
2201 | ||
2202 | return 0; | |
2203 | } | |
2204 | ||
49394a2a DA |
2205 | /* |
2206 | * Track idle stats per cpu by maintaining a local thread | |
2207 | * struct for the idle task on each cpu. | |
2208 | */ | |
2209 | static int init_idle_threads(int ncpu) | |
2210 | { | |
3bc2fa9c | 2211 | int i, ret; |
49394a2a DA |
2212 | |
2213 | idle_threads = zalloc(ncpu * sizeof(struct thread *)); | |
2214 | if (!idle_threads) | |
2215 | return -ENOMEM; | |
2216 | ||
b336352b | 2217 | idle_max_cpu = ncpu; |
49394a2a DA |
2218 | |
2219 | /* allocate the actual thread struct if needed */ | |
2220 | for (i = 0; i < ncpu; ++i) { | |
2221 | idle_threads[i] = thread__new(0, 0); | |
2222 | if (idle_threads[i] == NULL) | |
2223 | return -ENOMEM; | |
2224 | ||
3bc2fa9c NK |
2225 | ret = init_idle_thread(idle_threads[i]); |
2226 | if (ret < 0) | |
2227 | return ret; | |
49394a2a DA |
2228 | } |
2229 | ||
2230 | return 0; | |
2231 | } | |
2232 | ||
2233 | static void free_idle_threads(void) | |
2234 | { | |
2235 | int i; | |
2236 | ||
2237 | if (idle_threads == NULL) | |
2238 | return; | |
2239 | ||
b336352b | 2240 | for (i = 0; i < idle_max_cpu; ++i) { |
49394a2a DA |
2241 | if ((idle_threads[i])) |
2242 | thread__delete(idle_threads[i]); | |
2243 | } | |
2244 | ||
2245 | free(idle_threads); | |
2246 | } | |
2247 | ||
2248 | static struct thread *get_idle_thread(int cpu) | |
2249 | { | |
2250 | /* | |
2251 | * expand/allocate array of pointers to local thread | |
2252 | * structs if needed | |
2253 | */ | |
2254 | if ((cpu >= idle_max_cpu) || (idle_threads == NULL)) { | |
2255 | int i, j = __roundup_pow_of_two(cpu+1); | |
2256 | void *p; | |
2257 | ||
2258 | p = realloc(idle_threads, j * sizeof(struct thread *)); | |
2259 | if (!p) | |
2260 | return NULL; | |
2261 | ||
2262 | idle_threads = (struct thread **) p; | |
b336352b | 2263 | for (i = idle_max_cpu; i < j; ++i) |
49394a2a DA |
2264 | idle_threads[i] = NULL; |
2265 | ||
2266 | idle_max_cpu = j; | |
2267 | } | |
2268 | ||
2269 | /* allocate a new thread struct if needed */ | |
2270 | if (idle_threads[cpu] == NULL) { | |
2271 | idle_threads[cpu] = thread__new(0, 0); | |
2272 | ||
2273 | if (idle_threads[cpu]) { | |
3bc2fa9c NK |
2274 | if (init_idle_thread(idle_threads[cpu]) < 0) |
2275 | return NULL; | |
49394a2a DA |
2276 | } |
2277 | } | |
2278 | ||
2279 | return idle_threads[cpu]; | |
2280 | } | |
2281 | ||
4c50563d ACM |
2282 | static void save_idle_callchain(struct perf_sched *sched, |
2283 | struct idle_thread_runtime *itr, | |
699b5b92 NK |
2284 | struct perf_sample *sample) |
2285 | { | |
4c50563d | 2286 | if (!sched->show_callchain || sample->callchain == NULL) |
699b5b92 NK |
2287 | return; |
2288 | ||
2289 | callchain_cursor__copy(&itr->cursor, &callchain_cursor); | |
2290 | } | |
2291 | ||
6c973c90 DA |
2292 | static struct thread *timehist_get_thread(struct perf_sched *sched, |
2293 | struct perf_sample *sample, | |
49394a2a | 2294 | struct machine *machine, |
32dcd021 | 2295 | struct evsel *evsel) |
49394a2a DA |
2296 | { |
2297 | struct thread *thread; | |
2298 | ||
96039c7c | 2299 | if (is_idle_sample(sample, evsel)) { |
49394a2a DA |
2300 | thread = get_idle_thread(sample->cpu); |
2301 | if (thread == NULL) | |
2302 | pr_err("Failed to get idle thread for cpu %d.\n", sample->cpu); | |
2303 | ||
2304 | } else { | |
5d92d96a NK |
2305 | /* there were samples with tid 0 but non-zero pid */ |
2306 | thread = machine__findnew_thread(machine, sample->pid, | |
2307 | sample->tid ?: sample->pid); | |
49394a2a DA |
2308 | if (thread == NULL) { |
2309 | pr_debug("Failed to get thread for tid %d. skipping sample.\n", | |
2310 | sample->tid); | |
2311 | } | |
96039c7c NK |
2312 | |
2313 | save_task_callchain(sched, sample, evsel, machine); | |
699b5b92 NK |
2314 | if (sched->idle_hist) { |
2315 | struct thread *idle; | |
2316 | struct idle_thread_runtime *itr; | |
2317 | ||
2318 | idle = get_idle_thread(sample->cpu); | |
2319 | if (idle == NULL) { | |
2320 | pr_err("Failed to get idle thread for cpu %d.\n", sample->cpu); | |
2321 | return NULL; | |
2322 | } | |
2323 | ||
2324 | itr = thread__priv(idle); | |
2325 | if (itr == NULL) | |
2326 | return NULL; | |
2327 | ||
2328 | itr->last_thread = thread; | |
2329 | ||
2330 | /* copy task callchain when entering to idle */ | |
2331 | if (perf_evsel__intval(evsel, sample, "next_pid") == 0) | |
4c50563d | 2332 | save_idle_callchain(sched, itr, sample); |
699b5b92 | 2333 | } |
49394a2a DA |
2334 | } |
2335 | ||
2336 | return thread; | |
2337 | } | |
2338 | ||
52df138c | 2339 | static bool timehist_skip_sample(struct perf_sched *sched, |
a4b2b6f5 | 2340 | struct thread *thread, |
32dcd021 | 2341 | struct evsel *evsel, |
a4b2b6f5 | 2342 | struct perf_sample *sample) |
49394a2a DA |
2343 | { |
2344 | bool rc = false; | |
2345 | ||
52df138c | 2346 | if (thread__is_filtered(thread)) { |
49394a2a | 2347 | rc = true; |
52df138c DA |
2348 | sched->skipped_samples++; |
2349 | } | |
49394a2a | 2350 | |
a4b2b6f5 NK |
2351 | if (sched->idle_hist) { |
2352 | if (strcmp(perf_evsel__name(evsel), "sched:sched_switch")) | |
2353 | rc = true; | |
2354 | else if (perf_evsel__intval(evsel, sample, "prev_pid") != 0 && | |
2355 | perf_evsel__intval(evsel, sample, "next_pid") != 0) | |
2356 | rc = true; | |
2357 | } | |
2358 | ||
49394a2a DA |
2359 | return rc; |
2360 | } | |
2361 | ||
fc1469f1 | 2362 | static void timehist_print_wakeup_event(struct perf_sched *sched, |
32dcd021 | 2363 | struct evsel *evsel, |
fc1469f1 DA |
2364 | struct perf_sample *sample, |
2365 | struct machine *machine, | |
2366 | struct thread *awakened) | |
2367 | { | |
2368 | struct thread *thread; | |
2369 | char tstr[64]; | |
2370 | ||
2371 | thread = machine__findnew_thread(machine, sample->pid, sample->tid); | |
2372 | if (thread == NULL) | |
2373 | return; | |
2374 | ||
2375 | /* show wakeup unless both awakee and awaker are filtered */ | |
a4b2b6f5 NK |
2376 | if (timehist_skip_sample(sched, thread, evsel, sample) && |
2377 | timehist_skip_sample(sched, awakened, evsel, sample)) { | |
fc1469f1 DA |
2378 | return; |
2379 | } | |
2380 | ||
2381 | timestamp__scnprintf_usec(sample->time, tstr, sizeof(tstr)); | |
2382 | printf("%15s [%04d] ", tstr, sample->cpu); | |
a407b067 DA |
2383 | if (sched->show_cpu_visual) |
2384 | printf(" %*s ", sched->max_cpu + 1, ""); | |
fc1469f1 DA |
2385 | |
2386 | printf(" %-*s ", comm_width, timehist_get_commstr(thread)); | |
2387 | ||
2388 | /* dt spacer */ | |
2389 | printf(" %9s %9s %9s ", "", "", ""); | |
2390 | ||
2391 | printf("awakened: %s", timehist_get_commstr(awakened)); | |
2392 | ||
2393 | printf("\n"); | |
2394 | } | |
2395 | ||
2396 | static int timehist_sched_wakeup_event(struct perf_tool *tool, | |
49394a2a | 2397 | union perf_event *event __maybe_unused, |
32dcd021 | 2398 | struct evsel *evsel, |
49394a2a DA |
2399 | struct perf_sample *sample, |
2400 | struct machine *machine) | |
2401 | { | |
fc1469f1 | 2402 | struct perf_sched *sched = container_of(tool, struct perf_sched, tool); |
49394a2a DA |
2403 | struct thread *thread; |
2404 | struct thread_runtime *tr = NULL; | |
2405 | /* want pid of awakened task not pid in sample */ | |
2406 | const u32 pid = perf_evsel__intval(evsel, sample, "pid"); | |
2407 | ||
2408 | thread = machine__findnew_thread(machine, 0, pid); | |
2409 | if (thread == NULL) | |
2410 | return -1; | |
2411 | ||
2412 | tr = thread__get_runtime(thread); | |
2413 | if (tr == NULL) | |
2414 | return -1; | |
2415 | ||
2416 | if (tr->ready_to_run == 0) | |
2417 | tr->ready_to_run = sample->time; | |
2418 | ||
fc1469f1 | 2419 | /* show wakeups if requested */ |
853b7407 DA |
2420 | if (sched->show_wakeups && |
2421 | !perf_time__skip_sample(&sched->ptime, sample->time)) | |
a4b2b6f5 | 2422 | timehist_print_wakeup_event(sched, evsel, sample, machine, thread); |
fc1469f1 | 2423 | |
49394a2a DA |
2424 | return 0; |
2425 | } | |
2426 | ||
350f54fa | 2427 | static void timehist_print_migration_event(struct perf_sched *sched, |
32dcd021 | 2428 | struct evsel *evsel, |
350f54fa DA |
2429 | struct perf_sample *sample, |
2430 | struct machine *machine, | |
2431 | struct thread *migrated) | |
2432 | { | |
2433 | struct thread *thread; | |
2434 | char tstr[64]; | |
2435 | u32 max_cpus = sched->max_cpu + 1; | |
2436 | u32 ocpu, dcpu; | |
2437 | ||
2438 | if (sched->summary_only) | |
2439 | return; | |
2440 | ||
2441 | max_cpus = sched->max_cpu + 1; | |
2442 | ocpu = perf_evsel__intval(evsel, sample, "orig_cpu"); | |
2443 | dcpu = perf_evsel__intval(evsel, sample, "dest_cpu"); | |
2444 | ||
2445 | thread = machine__findnew_thread(machine, sample->pid, sample->tid); | |
2446 | if (thread == NULL) | |
2447 | return; | |
2448 | ||
a4b2b6f5 NK |
2449 | if (timehist_skip_sample(sched, thread, evsel, sample) && |
2450 | timehist_skip_sample(sched, migrated, evsel, sample)) { | |
350f54fa DA |
2451 | return; |
2452 | } | |
2453 | ||
2454 | timestamp__scnprintf_usec(sample->time, tstr, sizeof(tstr)); | |
2455 | printf("%15s [%04d] ", tstr, sample->cpu); | |
2456 | ||
2457 | if (sched->show_cpu_visual) { | |
2458 | u32 i; | |
2459 | char c; | |
2460 | ||
2461 | printf(" "); | |
2462 | for (i = 0; i < max_cpus; ++i) { | |
2463 | c = (i == sample->cpu) ? 'm' : ' '; | |
2464 | printf("%c", c); | |
2465 | } | |
2466 | printf(" "); | |
2467 | } | |
2468 | ||
2469 | printf(" %-*s ", comm_width, timehist_get_commstr(thread)); | |
2470 | ||
2471 | /* dt spacer */ | |
2472 | printf(" %9s %9s %9s ", "", "", ""); | |
2473 | ||
2474 | printf("migrated: %s", timehist_get_commstr(migrated)); | |
2475 | printf(" cpu %d => %d", ocpu, dcpu); | |
2476 | ||
2477 | printf("\n"); | |
2478 | } | |
2479 | ||
2480 | static int timehist_migrate_task_event(struct perf_tool *tool, | |
2481 | union perf_event *event __maybe_unused, | |
32dcd021 | 2482 | struct evsel *evsel, |
350f54fa DA |
2483 | struct perf_sample *sample, |
2484 | struct machine *machine) | |
2485 | { | |
2486 | struct perf_sched *sched = container_of(tool, struct perf_sched, tool); | |
2487 | struct thread *thread; | |
2488 | struct thread_runtime *tr = NULL; | |
2489 | /* want pid of migrated task not pid in sample */ | |
2490 | const u32 pid = perf_evsel__intval(evsel, sample, "pid"); | |
2491 | ||
2492 | thread = machine__findnew_thread(machine, 0, pid); | |
2493 | if (thread == NULL) | |
2494 | return -1; | |
2495 | ||
2496 | tr = thread__get_runtime(thread); | |
2497 | if (tr == NULL) | |
2498 | return -1; | |
2499 | ||
2500 | tr->migrations++; | |
2501 | ||
2502 | /* show migrations if requested */ | |
2503 | timehist_print_migration_event(sched, evsel, sample, machine, thread); | |
2504 | ||
2505 | return 0; | |
2506 | } | |
2507 | ||
52df138c | 2508 | static int timehist_sched_change_event(struct perf_tool *tool, |
49394a2a | 2509 | union perf_event *event, |
32dcd021 | 2510 | struct evsel *evsel, |
49394a2a DA |
2511 | struct perf_sample *sample, |
2512 | struct machine *machine) | |
2513 | { | |
fc1469f1 | 2514 | struct perf_sched *sched = container_of(tool, struct perf_sched, tool); |
853b7407 | 2515 | struct perf_time_interval *ptime = &sched->ptime; |
49394a2a DA |
2516 | struct addr_location al; |
2517 | struct thread *thread; | |
2518 | struct thread_runtime *tr = NULL; | |
853b7407 | 2519 | u64 tprev, t = sample->time; |
49394a2a | 2520 | int rc = 0; |
414e050c NK |
2521 | int state = perf_evsel__intval(evsel, sample, "prev_state"); |
2522 | ||
49394a2a DA |
2523 | |
2524 | if (machine__resolve(machine, &al, sample) < 0) { | |
2525 | pr_err("problem processing %d event. skipping it\n", | |
2526 | event->header.type); | |
2527 | rc = -1; | |
2528 | goto out; | |
2529 | } | |
2530 | ||
6c973c90 | 2531 | thread = timehist_get_thread(sched, sample, machine, evsel); |
49394a2a DA |
2532 | if (thread == NULL) { |
2533 | rc = -1; | |
2534 | goto out; | |
2535 | } | |
2536 | ||
a4b2b6f5 | 2537 | if (timehist_skip_sample(sched, thread, evsel, sample)) |
49394a2a DA |
2538 | goto out; |
2539 | ||
2540 | tr = thread__get_runtime(thread); | |
2541 | if (tr == NULL) { | |
2542 | rc = -1; | |
2543 | goto out; | |
2544 | } | |
2545 | ||
2546 | tprev = perf_evsel__get_time(evsel, sample->cpu); | |
2547 | ||
853b7407 DA |
2548 | /* |
2549 | * If start time given: | |
2550 | * - sample time is under window user cares about - skip sample | |
2551 | * - tprev is under window user cares about - reset to start of window | |
2552 | */ | |
2553 | if (ptime->start && ptime->start > t) | |
2554 | goto out; | |
2555 | ||
bdd75729 | 2556 | if (tprev && ptime->start > tprev) |
853b7407 DA |
2557 | tprev = ptime->start; |
2558 | ||
2559 | /* | |
2560 | * If end time given: | |
2561 | * - previous sched event is out of window - we are done | |
2562 | * - sample time is beyond window user cares about - reset it | |
2563 | * to close out stats for time window interest | |
2564 | */ | |
2565 | if (ptime->end) { | |
2566 | if (tprev > ptime->end) | |
2567 | goto out; | |
2568 | ||
2569 | if (t > ptime->end) | |
2570 | t = ptime->end; | |
2571 | } | |
2572 | ||
07235f84 NK |
2573 | if (!sched->idle_hist || thread->tid == 0) { |
2574 | timehist_update_runtime_stats(tr, t, tprev); | |
2575 | ||
2576 | if (sched->idle_hist) { | |
2577 | struct idle_thread_runtime *itr = (void *)tr; | |
2578 | struct thread_runtime *last_tr; | |
2579 | ||
2580 | BUG_ON(thread->tid != 0); | |
2581 | ||
2582 | if (itr->last_thread == NULL) | |
2583 | goto out; | |
2584 | ||
2585 | /* add current idle time as last thread's runtime */ | |
2586 | last_tr = thread__get_runtime(itr->last_thread); | |
2587 | if (last_tr == NULL) | |
2588 | goto out; | |
2589 | ||
2590 | timehist_update_runtime_stats(last_tr, t, tprev); | |
2591 | /* | |
2592 | * remove delta time of last thread as it's not updated | |
2593 | * and otherwise it will show an invalid value next | |
2594 | * time. we only care total run time and run stat. | |
2595 | */ | |
2596 | last_tr->dt_run = 0; | |
07235f84 | 2597 | last_tr->dt_delay = 0; |
941bdea7 NK |
2598 | last_tr->dt_sleep = 0; |
2599 | last_tr->dt_iowait = 0; | |
2600 | last_tr->dt_preempt = 0; | |
07235f84 | 2601 | |
ba957ebb NK |
2602 | if (itr->cursor.nr) |
2603 | callchain_append(&itr->callchain, &itr->cursor, t - tprev); | |
2604 | ||
07235f84 NK |
2605 | itr->last_thread = NULL; |
2606 | } | |
2607 | } | |
853b7407 | 2608 | |
52df138c | 2609 | if (!sched->summary_only) |
292c4a8f | 2610 | timehist_print_sample(sched, evsel, sample, &al, thread, t, state); |
49394a2a DA |
2611 | |
2612 | out: | |
9396c9cb NK |
2613 | if (sched->hist_time.start == 0 && t >= ptime->start) |
2614 | sched->hist_time.start = t; | |
2615 | if (ptime->end == 0 || t <= ptime->end) | |
2616 | sched->hist_time.end = t; | |
2617 | ||
49394a2a DA |
2618 | if (tr) { |
2619 | /* time of this sched_switch event becomes last time task seen */ | |
2620 | tr->last_time = sample->time; | |
2621 | ||
941bdea7 | 2622 | /* last state is used to determine where to account wait time */ |
414e050c | 2623 | tr->last_state = state; |
941bdea7 | 2624 | |
49394a2a DA |
2625 | /* sched out event for task so reset ready to run time */ |
2626 | tr->ready_to_run = 0; | |
2627 | } | |
2628 | ||
2629 | perf_evsel__save_time(evsel, sample->time, sample->cpu); | |
2630 | ||
2631 | return rc; | |
2632 | } | |
2633 | ||
2634 | static int timehist_sched_switch_event(struct perf_tool *tool, | |
2635 | union perf_event *event, | |
32dcd021 | 2636 | struct evsel *evsel, |
49394a2a DA |
2637 | struct perf_sample *sample, |
2638 | struct machine *machine __maybe_unused) | |
2639 | { | |
2640 | return timehist_sched_change_event(tool, event, evsel, sample, machine); | |
2641 | } | |
2642 | ||
2643 | static int process_lost(struct perf_tool *tool __maybe_unused, | |
2644 | union perf_event *event, | |
2645 | struct perf_sample *sample, | |
2646 | struct machine *machine __maybe_unused) | |
2647 | { | |
2648 | char tstr[64]; | |
2649 | ||
2650 | timestamp__scnprintf_usec(sample->time, tstr, sizeof(tstr)); | |
2651 | printf("%15s ", tstr); | |
5290ed69 | 2652 | printf("lost %" PRI_lu64 " events on cpu %d\n", event->lost.lost, sample->cpu); |
49394a2a DA |
2653 | |
2654 | return 0; | |
2655 | } | |
2656 | ||
2657 | ||
52df138c DA |
2658 | static void print_thread_runtime(struct thread *t, |
2659 | struct thread_runtime *r) | |
2660 | { | |
2661 | double mean = avg_stats(&r->run_stats); | |
2662 | float stddev; | |
2663 | ||
2664 | printf("%*s %5d %9" PRIu64 " ", | |
2665 | comm_width, timehist_get_commstr(t), t->ppid, | |
2666 | (u64) r->run_stats.n); | |
2667 | ||
2668 | print_sched_time(r->total_run_time, 8); | |
2669 | stddev = rel_stddev_stats(stddev_stats(&r->run_stats), mean); | |
2670 | print_sched_time(r->run_stats.min, 6); | |
2671 | printf(" "); | |
2672 | print_sched_time((u64) mean, 6); | |
2673 | printf(" "); | |
2674 | print_sched_time(r->run_stats.max, 6); | |
2675 | printf(" "); | |
2676 | printf("%5.2f", stddev); | |
350f54fa | 2677 | printf(" %5" PRIu64, r->migrations); |
52df138c DA |
2678 | printf("\n"); |
2679 | } | |
2680 | ||
587782c5 NK |
2681 | static void print_thread_waittime(struct thread *t, |
2682 | struct thread_runtime *r) | |
2683 | { | |
2684 | printf("%*s %5d %9" PRIu64 " ", | |
2685 | comm_width, timehist_get_commstr(t), t->ppid, | |
2686 | (u64) r->run_stats.n); | |
2687 | ||
2688 | print_sched_time(r->total_run_time, 8); | |
2689 | print_sched_time(r->total_sleep_time, 6); | |
2690 | printf(" "); | |
2691 | print_sched_time(r->total_iowait_time, 6); | |
2692 | printf(" "); | |
2693 | print_sched_time(r->total_preempt_time, 6); | |
2694 | printf(" "); | |
2695 | print_sched_time(r->total_delay_time, 6); | |
2696 | printf("\n"); | |
2697 | } | |
2698 | ||
52df138c | 2699 | struct total_run_stats { |
587782c5 | 2700 | struct perf_sched *sched; |
52df138c DA |
2701 | u64 sched_count; |
2702 | u64 task_count; | |
2703 | u64 total_run_time; | |
2704 | }; | |
2705 | ||
2706 | static int __show_thread_runtime(struct thread *t, void *priv) | |
2707 | { | |
2708 | struct total_run_stats *stats = priv; | |
2709 | struct thread_runtime *r; | |
2710 | ||
2711 | if (thread__is_filtered(t)) | |
2712 | return 0; | |
2713 | ||
2714 | r = thread__priv(t); | |
2715 | if (r && r->run_stats.n) { | |
2716 | stats->task_count++; | |
2717 | stats->sched_count += r->run_stats.n; | |
2718 | stats->total_run_time += r->total_run_time; | |
587782c5 NK |
2719 | |
2720 | if (stats->sched->show_state) | |
2721 | print_thread_waittime(t, r); | |
2722 | else | |
2723 | print_thread_runtime(t, r); | |
52df138c DA |
2724 | } |
2725 | ||
2726 | return 0; | |
2727 | } | |
2728 | ||
2729 | static int show_thread_runtime(struct thread *t, void *priv) | |
2730 | { | |
2731 | if (t->dead) | |
2732 | return 0; | |
2733 | ||
2734 | return __show_thread_runtime(t, priv); | |
2735 | } | |
2736 | ||
2737 | static int show_deadthread_runtime(struct thread *t, void *priv) | |
2738 | { | |
2739 | if (!t->dead) | |
2740 | return 0; | |
2741 | ||
2742 | return __show_thread_runtime(t, priv); | |
2743 | } | |
2744 | ||
ba957ebb NK |
2745 | static size_t callchain__fprintf_folded(FILE *fp, struct callchain_node *node) |
2746 | { | |
2747 | const char *sep = " <- "; | |
2748 | struct callchain_list *chain; | |
2749 | size_t ret = 0; | |
2750 | char bf[1024]; | |
2751 | bool first; | |
2752 | ||
2753 | if (node == NULL) | |
2754 | return 0; | |
2755 | ||
2756 | ret = callchain__fprintf_folded(fp, node->parent); | |
2757 | first = (ret == 0); | |
2758 | ||
2759 | list_for_each_entry(chain, &node->val, list) { | |
2760 | if (chain->ip >= PERF_CONTEXT_MAX) | |
2761 | continue; | |
2762 | if (chain->ms.sym && chain->ms.sym->ignore) | |
2763 | continue; | |
2764 | ret += fprintf(fp, "%s%s", first ? "" : sep, | |
2765 | callchain_list__sym_name(chain, bf, sizeof(bf), | |
2766 | false)); | |
2767 | first = false; | |
2768 | } | |
2769 | ||
2770 | return ret; | |
2771 | } | |
2772 | ||
cb4c13a5 | 2773 | static size_t timehist_print_idlehist_callchain(struct rb_root_cached *root) |
ba957ebb NK |
2774 | { |
2775 | size_t ret = 0; | |
2776 | FILE *fp = stdout; | |
2777 | struct callchain_node *chain; | |
cb4c13a5 | 2778 | struct rb_node *rb_node = rb_first_cached(root); |
ba957ebb NK |
2779 | |
2780 | printf(" %16s %8s %s\n", "Idle time (msec)", "Count", "Callchains"); | |
2781 | printf(" %.16s %.8s %.50s\n", graph_dotted_line, graph_dotted_line, | |
2782 | graph_dotted_line); | |
2783 | ||
2784 | while (rb_node) { | |
2785 | chain = rb_entry(rb_node, struct callchain_node, rb_node); | |
2786 | rb_node = rb_next(rb_node); | |
2787 | ||
2788 | ret += fprintf(fp, " "); | |
2789 | print_sched_time(chain->hit, 12); | |
2790 | ret += 16; /* print_sched_time returns 2nd arg + 4 */ | |
2791 | ret += fprintf(fp, " %8d ", chain->count); | |
2792 | ret += callchain__fprintf_folded(fp, chain); | |
2793 | ret += fprintf(fp, "\n"); | |
2794 | } | |
2795 | ||
2796 | return ret; | |
2797 | } | |
2798 | ||
52df138c DA |
2799 | static void timehist_print_summary(struct perf_sched *sched, |
2800 | struct perf_session *session) | |
2801 | { | |
2802 | struct machine *m = &session->machines.host; | |
2803 | struct total_run_stats totals; | |
2804 | u64 task_count; | |
2805 | struct thread *t; | |
2806 | struct thread_runtime *r; | |
2807 | int i; | |
9396c9cb | 2808 | u64 hist_time = sched->hist_time.end - sched->hist_time.start; |
52df138c DA |
2809 | |
2810 | memset(&totals, 0, sizeof(totals)); | |
587782c5 | 2811 | totals.sched = sched; |
52df138c | 2812 | |
07235f84 NK |
2813 | if (sched->idle_hist) { |
2814 | printf("\nIdle-time summary\n"); | |
2815 | printf("%*s parent sched-out ", comm_width, "comm"); | |
2816 | printf(" idle-time min-idle avg-idle max-idle stddev migrations\n"); | |
587782c5 NK |
2817 | } else if (sched->show_state) { |
2818 | printf("\nWait-time summary\n"); | |
2819 | printf("%*s parent sched-in ", comm_width, "comm"); | |
2820 | printf(" run-time sleep iowait preempt delay\n"); | |
07235f84 NK |
2821 | } else { |
2822 | printf("\nRuntime summary\n"); | |
2823 | printf("%*s parent sched-in ", comm_width, "comm"); | |
2824 | printf(" run-time min-run avg-run max-run stddev migrations\n"); | |
2825 | } | |
52df138c | 2826 | printf("%*s (count) ", comm_width, ""); |
587782c5 NK |
2827 | printf(" (msec) (msec) (msec) (msec) %s\n", |
2828 | sched->show_state ? "(msec)" : "%"); | |
350f54fa | 2829 | printf("%.117s\n", graph_dotted_line); |
52df138c DA |
2830 | |
2831 | machine__for_each_thread(m, show_thread_runtime, &totals); | |
2832 | task_count = totals.task_count; | |
2833 | if (!task_count) | |
2834 | printf("<no still running tasks>\n"); | |
2835 | ||
2836 | printf("\nTerminated tasks:\n"); | |
2837 | machine__for_each_thread(m, show_deadthread_runtime, &totals); | |
2838 | if (task_count == totals.task_count) | |
2839 | printf("<no terminated tasks>\n"); | |
2840 | ||
2841 | /* CPU idle stats not tracked when samples were skipped */ | |
07235f84 | 2842 | if (sched->skipped_samples && !sched->idle_hist) |
52df138c DA |
2843 | return; |
2844 | ||
2845 | printf("\nIdle stats:\n"); | |
b336352b | 2846 | for (i = 0; i < idle_max_cpu; ++i) { |
52df138c DA |
2847 | t = idle_threads[i]; |
2848 | if (!t) | |
2849 | continue; | |
2850 | ||
2851 | r = thread__priv(t); | |
2852 | if (r && r->run_stats.n) { | |
2853 | totals.sched_count += r->run_stats.n; | |
2854 | printf(" CPU %2d idle for ", i); | |
2855 | print_sched_time(r->total_run_time, 6); | |
9396c9cb | 2856 | printf(" msec (%6.2f%%)\n", 100.0 * r->total_run_time / hist_time); |
52df138c DA |
2857 | } else |
2858 | printf(" CPU %2d idle entire time window\n", i); | |
2859 | } | |
2860 | ||
4c50563d | 2861 | if (sched->idle_hist && sched->show_callchain) { |
ba957ebb NK |
2862 | callchain_param.mode = CHAIN_FOLDED; |
2863 | callchain_param.value = CCVAL_PERIOD; | |
2864 | ||
2865 | callchain_register_param(&callchain_param); | |
2866 | ||
2867 | printf("\nIdle stats by callchain:\n"); | |
2868 | for (i = 0; i < idle_max_cpu; ++i) { | |
2869 | struct idle_thread_runtime *itr; | |
2870 | ||
2871 | t = idle_threads[i]; | |
2872 | if (!t) | |
2873 | continue; | |
2874 | ||
2875 | itr = thread__priv(t); | |
2876 | if (itr == NULL) | |
2877 | continue; | |
2878 | ||
cb4c13a5 | 2879 | callchain_param.sort(&itr->sorted_root.rb_root, &itr->callchain, |
ba957ebb NK |
2880 | 0, &callchain_param); |
2881 | ||
2882 | printf(" CPU %2d:", i); | |
2883 | print_sched_time(itr->tr.total_run_time, 6); | |
2884 | printf(" msec\n"); | |
2885 | timehist_print_idlehist_callchain(&itr->sorted_root); | |
2886 | printf("\n"); | |
2887 | } | |
2888 | } | |
2889 | ||
52df138c DA |
2890 | printf("\n" |
2891 | " Total number of unique tasks: %" PRIu64 "\n" | |
9396c9cb | 2892 | "Total number of context switches: %" PRIu64 "\n", |
52df138c DA |
2893 | totals.task_count, totals.sched_count); |
2894 | ||
9396c9cb | 2895 | printf(" Total run time (msec): "); |
52df138c DA |
2896 | print_sched_time(totals.total_run_time, 2); |
2897 | printf("\n"); | |
9396c9cb NK |
2898 | |
2899 | printf(" Total scheduling time (msec): "); | |
2900 | print_sched_time(hist_time, 2); | |
2901 | printf(" (x %d)\n", sched->max_cpu); | |
52df138c DA |
2902 | } |
2903 | ||
49394a2a DA |
2904 | typedef int (*sched_handler)(struct perf_tool *tool, |
2905 | union perf_event *event, | |
32dcd021 | 2906 | struct evsel *evsel, |
49394a2a DA |
2907 | struct perf_sample *sample, |
2908 | struct machine *machine); | |
2909 | ||
2910 | static int perf_timehist__process_sample(struct perf_tool *tool, | |
2911 | union perf_event *event, | |
2912 | struct perf_sample *sample, | |
32dcd021 | 2913 | struct evsel *evsel, |
49394a2a DA |
2914 | struct machine *machine) |
2915 | { | |
2916 | struct perf_sched *sched = container_of(tool, struct perf_sched, tool); | |
2917 | int err = 0; | |
2918 | int this_cpu = sample->cpu; | |
2919 | ||
2920 | if (this_cpu > sched->max_cpu) | |
2921 | sched->max_cpu = this_cpu; | |
2922 | ||
2923 | if (evsel->handler != NULL) { | |
2924 | sched_handler f = evsel->handler; | |
2925 | ||
2926 | err = f(tool, event, evsel, sample, machine); | |
2927 | } | |
2928 | ||
2929 | return err; | |
2930 | } | |
2931 | ||
6c973c90 | 2932 | static int timehist_check_attr(struct perf_sched *sched, |
63503dba | 2933 | struct evlist *evlist) |
6c973c90 | 2934 | { |
32dcd021 | 2935 | struct evsel *evsel; |
6c973c90 DA |
2936 | struct evsel_runtime *er; |
2937 | ||
ce9036a6 | 2938 | list_for_each_entry(evsel, &evlist->core.entries, core.node) { |
6c973c90 DA |
2939 | er = perf_evsel__get_runtime(evsel); |
2940 | if (er == NULL) { | |
2941 | pr_err("Failed to allocate memory for evsel runtime data\n"); | |
2942 | return -1; | |
2943 | } | |
2944 | ||
27de9b2b | 2945 | if (sched->show_callchain && !evsel__has_callchain(evsel)) { |
6c973c90 DA |
2946 | pr_info("Samples do not have callchains.\n"); |
2947 | sched->show_callchain = 0; | |
2948 | symbol_conf.use_callchain = 0; | |
2949 | } | |
2950 | } | |
2951 | ||
2952 | return 0; | |
2953 | } | |
2954 | ||
49394a2a DA |
2955 | static int perf_sched__timehist(struct perf_sched *sched) |
2956 | { | |
32dcd021 | 2957 | const struct evsel_str_handler handlers[] = { |
49394a2a DA |
2958 | { "sched:sched_switch", timehist_sched_switch_event, }, |
2959 | { "sched:sched_wakeup", timehist_sched_wakeup_event, }, | |
2960 | { "sched:sched_wakeup_new", timehist_sched_wakeup_event, }, | |
2961 | }; | |
32dcd021 | 2962 | const struct evsel_str_handler migrate_handlers[] = { |
350f54fa DA |
2963 | { "sched:sched_migrate_task", timehist_migrate_task_event, }, |
2964 | }; | |
8ceb41d7 | 2965 | struct perf_data data = { |
2d4f2799 JO |
2966 | .path = input_name, |
2967 | .mode = PERF_DATA_MODE_READ, | |
2968 | .force = sched->force, | |
49394a2a DA |
2969 | }; |
2970 | ||
2971 | struct perf_session *session; | |
63503dba | 2972 | struct evlist *evlist; |
49394a2a DA |
2973 | int err = -1; |
2974 | ||
2975 | /* | |
2976 | * event handlers for timehist option | |
2977 | */ | |
2978 | sched->tool.sample = perf_timehist__process_sample; | |
2979 | sched->tool.mmap = perf_event__process_mmap; | |
2980 | sched->tool.comm = perf_event__process_comm; | |
2981 | sched->tool.exit = perf_event__process_exit; | |
2982 | sched->tool.fork = perf_event__process_fork; | |
2983 | sched->tool.lost = process_lost; | |
2984 | sched->tool.attr = perf_event__process_attr; | |
2985 | sched->tool.tracing_data = perf_event__process_tracing_data; | |
2986 | sched->tool.build_id = perf_event__process_build_id; | |
2987 | ||
2988 | sched->tool.ordered_events = true; | |
2989 | sched->tool.ordering_requires_timestamps = true; | |
2990 | ||
6c973c90 DA |
2991 | symbol_conf.use_callchain = sched->show_callchain; |
2992 | ||
8ceb41d7 | 2993 | session = perf_session__new(&data, false, &sched->tool); |
6ef81c55 MI |
2994 | if (IS_ERR(session)) |
2995 | return PTR_ERR(session); | |
49394a2a | 2996 | |
52df138c DA |
2997 | evlist = session->evlist; |
2998 | ||
49394a2a DA |
2999 | symbol__init(&session->header.env); |
3000 | ||
853b7407 DA |
3001 | if (perf_time__parse_str(&sched->ptime, sched->time_str) != 0) { |
3002 | pr_err("Invalid time string\n"); | |
3003 | return -EINVAL; | |
3004 | } | |
3005 | ||
6c973c90 DA |
3006 | if (timehist_check_attr(sched, evlist) != 0) |
3007 | goto out; | |
3008 | ||
49394a2a DA |
3009 | setup_pager(); |
3010 | ||
3011 | /* setup per-evsel handlers */ | |
3012 | if (perf_session__set_tracepoints_handlers(session, handlers)) | |
3013 | goto out; | |
3014 | ||
f45bf8d3 DA |
3015 | /* sched_switch event at a minimum needs to exist */ |
3016 | if (!perf_evlist__find_tracepoint_by_name(session->evlist, | |
3017 | "sched:sched_switch")) { | |
3018 | pr_err("No sched_switch events found. Have you run 'perf sched record'?\n"); | |
49394a2a | 3019 | goto out; |
f45bf8d3 | 3020 | } |
49394a2a | 3021 | |
350f54fa DA |
3022 | if (sched->show_migrations && |
3023 | perf_session__set_tracepoints_handlers(session, migrate_handlers)) | |
3024 | goto out; | |
3025 | ||
49394a2a DA |
3026 | /* pre-allocate struct for per-CPU idle stats */ |
3027 | sched->max_cpu = session->header.env.nr_cpus_online; | |
3028 | if (sched->max_cpu == 0) | |
3029 | sched->max_cpu = 4; | |
3030 | if (init_idle_threads(sched->max_cpu)) | |
3031 | goto out; | |
3032 | ||
52df138c DA |
3033 | /* summary_only implies summary option, but don't overwrite summary if set */ |
3034 | if (sched->summary_only) | |
3035 | sched->summary = sched->summary_only; | |
3036 | ||
3037 | if (!sched->summary_only) | |
a407b067 | 3038 | timehist_header(sched); |
49394a2a DA |
3039 | |
3040 | err = perf_session__process_events(session); | |
3041 | if (err) { | |
3042 | pr_err("Failed to process events, error %d", err); | |
3043 | goto out; | |
3044 | } | |
3045 | ||
52df138c DA |
3046 | sched->nr_events = evlist->stats.nr_events[0]; |
3047 | sched->nr_lost_events = evlist->stats.total_lost; | |
3048 | sched->nr_lost_chunks = evlist->stats.nr_events[PERF_RECORD_LOST]; | |
3049 | ||
3050 | if (sched->summary) | |
3051 | timehist_print_summary(sched, session); | |
3052 | ||
49394a2a DA |
3053 | out: |
3054 | free_idle_threads(); | |
3055 | perf_session__delete(session); | |
3056 | ||
3057 | return err; | |
3058 | } | |
3059 | ||
3060 | ||
0e9b07e5 | 3061 | static void print_bad_events(struct perf_sched *sched) |
0ec04e16 | 3062 | { |
0e9b07e5 | 3063 | if (sched->nr_unordered_timestamps && sched->nr_timestamps) { |
0ec04e16 | 3064 | printf(" INFO: %.3f%% unordered timestamps (%ld out of %ld)\n", |
0e9b07e5 ACM |
3065 | (double)sched->nr_unordered_timestamps/(double)sched->nr_timestamps*100.0, |
3066 | sched->nr_unordered_timestamps, sched->nr_timestamps); | |
0ec04e16 | 3067 | } |
0e9b07e5 | 3068 | if (sched->nr_lost_events && sched->nr_events) { |
0ec04e16 | 3069 | printf(" INFO: %.3f%% lost events (%ld out of %ld, in %ld chunks)\n", |
0e9b07e5 ACM |
3070 | (double)sched->nr_lost_events/(double)sched->nr_events * 100.0, |
3071 | sched->nr_lost_events, sched->nr_events, sched->nr_lost_chunks); | |
0ec04e16 | 3072 | } |
0e9b07e5 | 3073 | if (sched->nr_context_switch_bugs && sched->nr_timestamps) { |
0ec04e16 | 3074 | printf(" INFO: %.3f%% context switch bugs (%ld out of %ld)", |
0e9b07e5 ACM |
3075 | (double)sched->nr_context_switch_bugs/(double)sched->nr_timestamps*100.0, |
3076 | sched->nr_context_switch_bugs, sched->nr_timestamps); | |
3077 | if (sched->nr_lost_events) | |
0ec04e16 IM |
3078 | printf(" (due to lost events?)"); |
3079 | printf("\n"); | |
3080 | } | |
3081 | } | |
3082 | ||
cb4c13a5 | 3083 | static void __merge_work_atoms(struct rb_root_cached *root, struct work_atoms *data) |
2f80dd44 | 3084 | { |
cb4c13a5 | 3085 | struct rb_node **new = &(root->rb_root.rb_node), *parent = NULL; |
2f80dd44 JB |
3086 | struct work_atoms *this; |
3087 | const char *comm = thread__comm_str(data->thread), *this_comm; | |
cb4c13a5 | 3088 | bool leftmost = true; |
2f80dd44 JB |
3089 | |
3090 | while (*new) { | |
3091 | int cmp; | |
3092 | ||
3093 | this = container_of(*new, struct work_atoms, node); | |
3094 | parent = *new; | |
3095 | ||
3096 | this_comm = thread__comm_str(this->thread); | |
3097 | cmp = strcmp(comm, this_comm); | |
3098 | if (cmp > 0) { | |
3099 | new = &((*new)->rb_left); | |
3100 | } else if (cmp < 0) { | |
3101 | new = &((*new)->rb_right); | |
cb4c13a5 | 3102 | leftmost = false; |
2f80dd44 JB |
3103 | } else { |
3104 | this->num_merged++; | |
3105 | this->total_runtime += data->total_runtime; | |
3106 | this->nb_atoms += data->nb_atoms; | |
3107 | this->total_lat += data->total_lat; | |
3108 | list_splice(&data->work_list, &this->work_list); | |
3109 | if (this->max_lat < data->max_lat) { | |
3110 | this->max_lat = data->max_lat; | |
3111 | this->max_lat_at = data->max_lat_at; | |
3112 | } | |
3113 | zfree(&data); | |
3114 | return; | |
3115 | } | |
3116 | } | |
3117 | ||
3118 | data->num_merged++; | |
3119 | rb_link_node(&data->node, parent, new); | |
cb4c13a5 | 3120 | rb_insert_color_cached(&data->node, root, leftmost); |
2f80dd44 JB |
3121 | } |
3122 | ||
3123 | static void perf_sched__merge_lat(struct perf_sched *sched) | |
3124 | { | |
3125 | struct work_atoms *data; | |
3126 | struct rb_node *node; | |
3127 | ||
3128 | if (sched->skip_merge) | |
3129 | return; | |
3130 | ||
cb4c13a5 DB |
3131 | while ((node = rb_first_cached(&sched->atom_root))) { |
3132 | rb_erase_cached(node, &sched->atom_root); | |
2f80dd44 JB |
3133 | data = rb_entry(node, struct work_atoms, node); |
3134 | __merge_work_atoms(&sched->merged_atom_root, data); | |
3135 | } | |
3136 | } | |
3137 | ||
0e9b07e5 | 3138 | static int perf_sched__lat(struct perf_sched *sched) |
0ec04e16 IM |
3139 | { |
3140 | struct rb_node *next; | |
3141 | ||
3142 | setup_pager(); | |
ad9def7c | 3143 | |
ae536acf | 3144 | if (perf_sched__read_events(sched)) |
a116e05d | 3145 | return -1; |
ad9def7c | 3146 | |
2f80dd44 | 3147 | perf_sched__merge_lat(sched); |
0e9b07e5 | 3148 | perf_sched__sort_lat(sched); |
0ec04e16 | 3149 | |
80790e0b RR |
3150 | printf("\n -----------------------------------------------------------------------------------------------------------------\n"); |
3151 | printf(" Task | Runtime ms | Switches | Average delay ms | Maximum delay ms | Maximum delay at |\n"); | |
3152 | printf(" -----------------------------------------------------------------------------------------------------------------\n"); | |
0ec04e16 | 3153 | |
cb4c13a5 | 3154 | next = rb_first_cached(&sched->sorted_atom_root); |
0ec04e16 IM |
3155 | |
3156 | while (next) { | |
3157 | struct work_atoms *work_list; | |
3158 | ||
3159 | work_list = rb_entry(next, struct work_atoms, node); | |
0e9b07e5 | 3160 | output_lat_thread(sched, work_list); |
0ec04e16 | 3161 | next = rb_next(next); |
ae536acf | 3162 | thread__zput(work_list->thread); |
0ec04e16 IM |
3163 | } |
3164 | ||
80790e0b | 3165 | printf(" -----------------------------------------------------------------------------------------------------------------\n"); |
9486aa38 | 3166 | printf(" TOTAL: |%11.3f ms |%9" PRIu64 " |\n", |
4fc76e49 | 3167 | (double)sched->all_runtime / NSEC_PER_MSEC, sched->all_count); |
0ec04e16 IM |
3168 | |
3169 | printf(" ---------------------------------------------------\n"); | |
3170 | ||
0e9b07e5 | 3171 | print_bad_events(sched); |
0ec04e16 IM |
3172 | printf("\n"); |
3173 | ||
a116e05d | 3174 | return 0; |
0ec04e16 IM |
3175 | } |
3176 | ||
99623c62 JO |
3177 | static int setup_map_cpus(struct perf_sched *sched) |
3178 | { | |
f854839b | 3179 | struct perf_cpu_map *map; |
73643bb6 | 3180 | |
99623c62 JO |
3181 | sched->max_cpu = sysconf(_SC_NPROCESSORS_CONF); |
3182 | ||
3183 | if (sched->map.comp) { | |
3184 | sched->map.comp_cpus = zalloc(sched->max_cpu * sizeof(int)); | |
cf294f24 JO |
3185 | if (!sched->map.comp_cpus) |
3186 | return -1; | |
99623c62 JO |
3187 | } |
3188 | ||
73643bb6 JO |
3189 | if (!sched->map.cpus_str) |
3190 | return 0; | |
3191 | ||
9c3516d1 | 3192 | map = perf_cpu_map__new(sched->map.cpus_str); |
73643bb6 JO |
3193 | if (!map) { |
3194 | pr_err("failed to get cpus map from %s\n", sched->map.cpus_str); | |
3195 | return -1; | |
3196 | } | |
3197 | ||
3198 | sched->map.cpus = map; | |
99623c62 JO |
3199 | return 0; |
3200 | } | |
3201 | ||
a151a37a JO |
3202 | static int setup_color_pids(struct perf_sched *sched) |
3203 | { | |
9749b90e | 3204 | struct perf_thread_map *map; |
a151a37a JO |
3205 | |
3206 | if (!sched->map.color_pids_str) | |
3207 | return 0; | |
3208 | ||
3209 | map = thread_map__new_by_tid_str(sched->map.color_pids_str); | |
3210 | if (!map) { | |
3211 | pr_err("failed to get thread map from %s\n", sched->map.color_pids_str); | |
3212 | return -1; | |
3213 | } | |
3214 | ||
3215 | sched->map.color_pids = map; | |
3216 | return 0; | |
3217 | } | |
3218 | ||
cf294f24 JO |
3219 | static int setup_color_cpus(struct perf_sched *sched) |
3220 | { | |
f854839b | 3221 | struct perf_cpu_map *map; |
cf294f24 JO |
3222 | |
3223 | if (!sched->map.color_cpus_str) | |
3224 | return 0; | |
3225 | ||
9c3516d1 | 3226 | map = perf_cpu_map__new(sched->map.color_cpus_str); |
cf294f24 JO |
3227 | if (!map) { |
3228 | pr_err("failed to get thread map from %s\n", sched->map.color_cpus_str); | |
3229 | return -1; | |
3230 | } | |
3231 | ||
3232 | sched->map.color_cpus = map; | |
3233 | return 0; | |
3234 | } | |
3235 | ||
0e9b07e5 | 3236 | static int perf_sched__map(struct perf_sched *sched) |
0ec04e16 | 3237 | { |
99623c62 JO |
3238 | if (setup_map_cpus(sched)) |
3239 | return -1; | |
40749d0f | 3240 | |
a151a37a JO |
3241 | if (setup_color_pids(sched)) |
3242 | return -1; | |
3243 | ||
cf294f24 JO |
3244 | if (setup_color_cpus(sched)) |
3245 | return -1; | |
3246 | ||
0ec04e16 | 3247 | setup_pager(); |
ae536acf | 3248 | if (perf_sched__read_events(sched)) |
a116e05d | 3249 | return -1; |
0e9b07e5 | 3250 | print_bad_events(sched); |
a116e05d | 3251 | return 0; |
0ec04e16 IM |
3252 | } |
3253 | ||
0e9b07e5 | 3254 | static int perf_sched__replay(struct perf_sched *sched) |
0ec04e16 IM |
3255 | { |
3256 | unsigned long i; | |
3257 | ||
0e9b07e5 ACM |
3258 | calibrate_run_measurement_overhead(sched); |
3259 | calibrate_sleep_measurement_overhead(sched); | |
0ec04e16 | 3260 | |
0e9b07e5 | 3261 | test_calibrations(sched); |
0ec04e16 | 3262 | |
ae536acf | 3263 | if (perf_sched__read_events(sched)) |
a116e05d | 3264 | return -1; |
0ec04e16 | 3265 | |
0e9b07e5 ACM |
3266 | printf("nr_run_events: %ld\n", sched->nr_run_events); |
3267 | printf("nr_sleep_events: %ld\n", sched->nr_sleep_events); | |
3268 | printf("nr_wakeup_events: %ld\n", sched->nr_wakeup_events); | |
0ec04e16 | 3269 | |
0e9b07e5 ACM |
3270 | if (sched->targetless_wakeups) |
3271 | printf("target-less wakeups: %ld\n", sched->targetless_wakeups); | |
3272 | if (sched->multitarget_wakeups) | |
3273 | printf("multi-target wakeups: %ld\n", sched->multitarget_wakeups); | |
3274 | if (sched->nr_run_events_optimized) | |
0ec04e16 | 3275 | printf("run atoms optimized: %ld\n", |
0e9b07e5 | 3276 | sched->nr_run_events_optimized); |
0ec04e16 | 3277 | |
0e9b07e5 ACM |
3278 | print_task_traces(sched); |
3279 | add_cross_task_wakeups(sched); | |
0ec04e16 | 3280 | |
0e9b07e5 | 3281 | create_tasks(sched); |
0ec04e16 | 3282 | printf("------------------------------------------------------------\n"); |
0e9b07e5 ACM |
3283 | for (i = 0; i < sched->replay_repeat; i++) |
3284 | run_one_test(sched); | |
a116e05d ACM |
3285 | |
3286 | return 0; | |
0ec04e16 IM |
3287 | } |
3288 | ||
0e9b07e5 ACM |
3289 | static void setup_sorting(struct perf_sched *sched, const struct option *options, |
3290 | const char * const usage_msg[]) | |
daa1d7a5 | 3291 | { |
0e9b07e5 | 3292 | char *tmp, *tok, *str = strdup(sched->sort_order); |
daa1d7a5 FW |
3293 | |
3294 | for (tok = strtok_r(str, ", ", &tmp); | |
3295 | tok; tok = strtok_r(NULL, ", ", &tmp)) { | |
0e9b07e5 | 3296 | if (sort_dimension__add(tok, &sched->sort_list) < 0) { |
c7118369 NK |
3297 | usage_with_options_msg(usage_msg, options, |
3298 | "Unknown --sort key: `%s'", tok); | |
daa1d7a5 FW |
3299 | } |
3300 | } | |
3301 | ||
3302 | free(str); | |
3303 | ||
0e9b07e5 | 3304 | sort_dimension__add("pid", &sched->cmp_pid); |
daa1d7a5 FW |
3305 | } |
3306 | ||
1fc35b29 IM |
3307 | static int __cmd_record(int argc, const char **argv) |
3308 | { | |
3309 | unsigned int rec_argc, i, j; | |
3310 | const char **rec_argv; | |
0e9b07e5 ACM |
3311 | const char * const record_args[] = { |
3312 | "record", | |
3313 | "-a", | |
3314 | "-R", | |
0e9b07e5 ACM |
3315 | "-m", "1024", |
3316 | "-c", "1", | |
3317 | "-e", "sched:sched_switch", | |
3318 | "-e", "sched:sched_stat_wait", | |
3319 | "-e", "sched:sched_stat_sleep", | |
3320 | "-e", "sched:sched_stat_iowait", | |
3321 | "-e", "sched:sched_stat_runtime", | |
0e9b07e5 ACM |
3322 | "-e", "sched:sched_process_fork", |
3323 | "-e", "sched:sched_wakeup", | |
7fff9597 | 3324 | "-e", "sched:sched_wakeup_new", |
0e9b07e5 ACM |
3325 | "-e", "sched:sched_migrate_task", |
3326 | }; | |
1fc35b29 IM |
3327 | |
3328 | rec_argc = ARRAY_SIZE(record_args) + argc - 1; | |
3329 | rec_argv = calloc(rec_argc + 1, sizeof(char *)); | |
3330 | ||
e462dc55 | 3331 | if (rec_argv == NULL) |
ce47dc56 CS |
3332 | return -ENOMEM; |
3333 | ||
1fc35b29 IM |
3334 | for (i = 0; i < ARRAY_SIZE(record_args); i++) |
3335 | rec_argv[i] = strdup(record_args[i]); | |
3336 | ||
3337 | for (j = 1; j < (unsigned int)argc; j++, i++) | |
3338 | rec_argv[i] = argv[j]; | |
3339 | ||
3340 | BUG_ON(i != rec_argc); | |
3341 | ||
b0ad8ea6 | 3342 | return cmd_record(i, rec_argv); |
1fc35b29 IM |
3343 | } |
3344 | ||
b0ad8ea6 | 3345 | int cmd_sched(int argc, const char **argv) |
0a02ad93 | 3346 | { |
49b8e2be | 3347 | static const char default_sort_order[] = "avg, max, switch, runtime"; |
8a39df8f AH |
3348 | struct perf_sched sched = { |
3349 | .tool = { | |
3350 | .sample = perf_sched__process_tracepoint_sample, | |
99a3c3a9 | 3351 | .comm = perf_sched__process_comm, |
f3b3614a | 3352 | .namespaces = perf_event__process_namespaces, |
8a39df8f AH |
3353 | .lost = perf_event__process_lost, |
3354 | .fork = perf_sched__process_fork_event, | |
0a8cb85c | 3355 | .ordered_events = true, |
8a39df8f AH |
3356 | }, |
3357 | .cmp_pid = LIST_HEAD_INIT(sched.cmp_pid), | |
3358 | .sort_list = LIST_HEAD_INIT(sched.sort_list), | |
3359 | .start_work_mutex = PTHREAD_MUTEX_INITIALIZER, | |
3360 | .work_done_wait_mutex = PTHREAD_MUTEX_INITIALIZER, | |
8a39df8f AH |
3361 | .sort_order = default_sort_order, |
3362 | .replay_repeat = 10, | |
3363 | .profile_cpu = -1, | |
3364 | .next_shortname1 = 'A', | |
3365 | .next_shortname2 = '0', | |
2f80dd44 | 3366 | .skip_merge = 0, |
6c973c90 DA |
3367 | .show_callchain = 1, |
3368 | .max_stack = 5, | |
8a39df8f | 3369 | }; |
77f02f44 NK |
3370 | const struct option sched_options[] = { |
3371 | OPT_STRING('i', "input", &input_name, "file", | |
3372 | "input file name"), | |
3373 | OPT_INCR('v', "verbose", &verbose, | |
3374 | "be more verbose (show symbol address, etc)"), | |
3375 | OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, | |
3376 | "dump raw trace in ASCII"), | |
6fa94258 | 3377 | OPT_BOOLEAN('f', "force", &sched.force, "don't complain, do it"), |
77f02f44 NK |
3378 | OPT_END() |
3379 | }; | |
0e9b07e5 ACM |
3380 | const struct option latency_options[] = { |
3381 | OPT_STRING('s', "sort", &sched.sort_order, "key[,key2...]", | |
3382 | "sort by key(s): runtime, switch, avg, max"), | |
0e9b07e5 ACM |
3383 | OPT_INTEGER('C', "CPU", &sched.profile_cpu, |
3384 | "CPU to profile on"), | |
2f80dd44 JB |
3385 | OPT_BOOLEAN('p', "pids", &sched.skip_merge, |
3386 | "latency stats per pid instead of per comm"), | |
77f02f44 | 3387 | OPT_PARENT(sched_options) |
0e9b07e5 ACM |
3388 | }; |
3389 | const struct option replay_options[] = { | |
3390 | OPT_UINTEGER('r', "repeat", &sched.replay_repeat, | |
3391 | "repeat the workload replay N times (-1: infinite)"), | |
77f02f44 | 3392 | OPT_PARENT(sched_options) |
0e9b07e5 | 3393 | }; |
99623c62 JO |
3394 | const struct option map_options[] = { |
3395 | OPT_BOOLEAN(0, "compact", &sched.map.comp, | |
3396 | "map output in compact mode"), | |
a151a37a JO |
3397 | OPT_STRING(0, "color-pids", &sched.map.color_pids_str, "pids", |
3398 | "highlight given pids in map"), | |
cf294f24 JO |
3399 | OPT_STRING(0, "color-cpus", &sched.map.color_cpus_str, "cpus", |
3400 | "highlight given CPUs in map"), | |
73643bb6 JO |
3401 | OPT_STRING(0, "cpus", &sched.map.cpus_str, "cpus", |
3402 | "display given CPUs in map"), | |
77f02f44 | 3403 | OPT_PARENT(sched_options) |
99623c62 | 3404 | }; |
49394a2a DA |
3405 | const struct option timehist_options[] = { |
3406 | OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name, | |
3407 | "file", "vmlinux pathname"), | |
3408 | OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name, | |
3409 | "file", "kallsyms pathname"), | |
6c973c90 DA |
3410 | OPT_BOOLEAN('g', "call-graph", &sched.show_callchain, |
3411 | "Display call chains if present (default on)"), | |
3412 | OPT_UINTEGER(0, "max-stack", &sched.max_stack, | |
3413 | "Maximum number of functions to display backtrace."), | |
49394a2a DA |
3414 | OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory", |
3415 | "Look for files with symbols relative to this directory"), | |
52df138c DA |
3416 | OPT_BOOLEAN('s', "summary", &sched.summary_only, |
3417 | "Show only syscall summary with statistics"), | |
3418 | OPT_BOOLEAN('S', "with-summary", &sched.summary, | |
3419 | "Show all syscalls and summary with statistics"), | |
fc1469f1 | 3420 | OPT_BOOLEAN('w', "wakeups", &sched.show_wakeups, "Show wakeup events"), |
292c4a8f | 3421 | OPT_BOOLEAN('n', "next", &sched.show_next, "Show next task"), |
350f54fa | 3422 | OPT_BOOLEAN('M', "migrations", &sched.show_migrations, "Show migration events"), |
a407b067 | 3423 | OPT_BOOLEAN('V', "cpu-visual", &sched.show_cpu_visual, "Add CPU visual"), |
07235f84 | 3424 | OPT_BOOLEAN('I', "idle-hist", &sched.idle_hist, "Show idle events only"), |
853b7407 DA |
3425 | OPT_STRING(0, "time", &sched.time_str, "str", |
3426 | "Time span for analysis (start,stop)"), | |
414e050c | 3427 | OPT_BOOLEAN(0, "state", &sched.show_state, "Show task state when sched-out"), |
0f59d7a3 DA |
3428 | OPT_STRING('p', "pid", &symbol_conf.pid_list_str, "pid[,pid...]", |
3429 | "analyze events only for given process id(s)"), | |
3430 | OPT_STRING('t', "tid", &symbol_conf.tid_list_str, "tid[,tid...]", | |
3431 | "analyze events only for given thread id(s)"), | |
49394a2a DA |
3432 | OPT_PARENT(sched_options) |
3433 | }; | |
3434 | ||
0e9b07e5 ACM |
3435 | const char * const latency_usage[] = { |
3436 | "perf sched latency [<options>]", | |
3437 | NULL | |
3438 | }; | |
3439 | const char * const replay_usage[] = { | |
3440 | "perf sched replay [<options>]", | |
3441 | NULL | |
3442 | }; | |
99623c62 JO |
3443 | const char * const map_usage[] = { |
3444 | "perf sched map [<options>]", | |
3445 | NULL | |
3446 | }; | |
49394a2a DA |
3447 | const char * const timehist_usage[] = { |
3448 | "perf sched timehist [<options>]", | |
3449 | NULL | |
3450 | }; | |
a83edb2d | 3451 | const char *const sched_subcommands[] = { "record", "latency", "map", |
49394a2a DA |
3452 | "replay", "script", |
3453 | "timehist", NULL }; | |
a83edb2d RR |
3454 | const char *sched_usage[] = { |
3455 | NULL, | |
0e9b07e5 ACM |
3456 | NULL |
3457 | }; | |
3458 | struct trace_sched_handler lat_ops = { | |
3459 | .wakeup_event = latency_wakeup_event, | |
3460 | .switch_event = latency_switch_event, | |
3461 | .runtime_event = latency_runtime_event, | |
0e9b07e5 ACM |
3462 | .migrate_task_event = latency_migrate_task_event, |
3463 | }; | |
3464 | struct trace_sched_handler map_ops = { | |
3465 | .switch_event = map_switch_event, | |
3466 | }; | |
3467 | struct trace_sched_handler replay_ops = { | |
3468 | .wakeup_event = replay_wakeup_event, | |
3469 | .switch_event = replay_switch_event, | |
3470 | .fork_event = replay_fork_event, | |
3471 | }; | |
156a2b02 AH |
3472 | unsigned int i; |
3473 | ||
3474 | for (i = 0; i < ARRAY_SIZE(sched.curr_pid); i++) | |
3475 | sched.curr_pid[i] = -1; | |
0e9b07e5 | 3476 | |
a83edb2d RR |
3477 | argc = parse_options_subcommand(argc, argv, sched_options, sched_subcommands, |
3478 | sched_usage, PARSE_OPT_STOP_AT_NON_OPTION); | |
f2858d8a IM |
3479 | if (!argc) |
3480 | usage_with_options(sched_usage, sched_options); | |
0a02ad93 | 3481 | |
c0777c5a | 3482 | /* |
133dc4c3 | 3483 | * Aliased to 'perf script' for now: |
c0777c5a | 3484 | */ |
133dc4c3 | 3485 | if (!strcmp(argv[0], "script")) |
b0ad8ea6 | 3486 | return cmd_script(argc, argv); |
c0777c5a | 3487 | |
1fc35b29 IM |
3488 | if (!strncmp(argv[0], "rec", 3)) { |
3489 | return __cmd_record(argc, argv); | |
3490 | } else if (!strncmp(argv[0], "lat", 3)) { | |
0e9b07e5 | 3491 | sched.tp_handler = &lat_ops; |
f2858d8a IM |
3492 | if (argc > 1) { |
3493 | argc = parse_options(argc, argv, latency_options, latency_usage, 0); | |
3494 | if (argc) | |
3495 | usage_with_options(latency_usage, latency_options); | |
f2858d8a | 3496 | } |
0e9b07e5 ACM |
3497 | setup_sorting(&sched, latency_options, latency_usage); |
3498 | return perf_sched__lat(&sched); | |
0ec04e16 | 3499 | } else if (!strcmp(argv[0], "map")) { |
99623c62 | 3500 | if (argc) { |
a151a37a | 3501 | argc = parse_options(argc, argv, map_options, map_usage, 0); |
99623c62 JO |
3502 | if (argc) |
3503 | usage_with_options(map_usage, map_options); | |
3504 | } | |
0e9b07e5 ACM |
3505 | sched.tp_handler = &map_ops; |
3506 | setup_sorting(&sched, latency_options, latency_usage); | |
3507 | return perf_sched__map(&sched); | |
f2858d8a | 3508 | } else if (!strncmp(argv[0], "rep", 3)) { |
0e9b07e5 | 3509 | sched.tp_handler = &replay_ops; |
f2858d8a IM |
3510 | if (argc) { |
3511 | argc = parse_options(argc, argv, replay_options, replay_usage, 0); | |
3512 | if (argc) | |
3513 | usage_with_options(replay_usage, replay_options); | |
3514 | } | |
0e9b07e5 | 3515 | return perf_sched__replay(&sched); |
49394a2a DA |
3516 | } else if (!strcmp(argv[0], "timehist")) { |
3517 | if (argc) { | |
3518 | argc = parse_options(argc, argv, timehist_options, | |
3519 | timehist_usage, 0); | |
3520 | if (argc) | |
3521 | usage_with_options(timehist_usage, timehist_options); | |
3522 | } | |
292c4a8f BG |
3523 | if ((sched.show_wakeups || sched.show_next) && |
3524 | sched.summary_only) { | |
3525 | pr_err(" Error: -s and -[n|w] are mutually exclusive.\n"); | |
fc1469f1 | 3526 | parse_options_usage(timehist_usage, timehist_options, "s", true); |
292c4a8f BG |
3527 | if (sched.show_wakeups) |
3528 | parse_options_usage(NULL, timehist_options, "w", true); | |
3529 | if (sched.show_next) | |
3530 | parse_options_usage(NULL, timehist_options, "n", true); | |
fc1469f1 DA |
3531 | return -EINVAL; |
3532 | } | |
3533 | ||
49394a2a | 3534 | return perf_sched__timehist(&sched); |
f2858d8a IM |
3535 | } else { |
3536 | usage_with_options(sched_usage, sched_options); | |
3537 | } | |
3538 | ||
ec156764 | 3539 | return 0; |
0a02ad93 | 3540 | } |