Commit | Line | Data |
---|---|---|
0a02ad93 | 1 | #include "builtin.h" |
b1ffe8f3 | 2 | #include "perf.h" |
0a02ad93 IM |
3 | |
4 | #include "util/util.h" | |
5 | #include "util/cache.h" | |
6 | #include "util/symbol.h" | |
7 | #include "util/thread.h" | |
8 | #include "util/header.h" | |
9 | ||
10 | #include "util/parse-options.h" | |
b1ffe8f3 | 11 | #include "util/trace-event.h" |
0a02ad93 | 12 | |
0a02ad93 | 13 | #include "util/debug.h" |
016e92fb | 14 | #include "util/data_map.h" |
0a02ad93 | 15 | |
ec156764 | 16 | #include <sys/types.h> |
b1ffe8f3 | 17 | #include <sys/prctl.h> |
0a02ad93 | 18 | |
b1ffe8f3 IM |
19 | #include <semaphore.h> |
20 | #include <pthread.h> | |
21 | #include <math.h> | |
419ab0d6 | 22 | |
ec156764 | 23 | static char const *input_name = "perf.data"; |
0a02ad93 | 24 | |
ec156764 IM |
25 | static struct perf_header *header; |
26 | static u64 sample_type; | |
0a02ad93 | 27 | |
daa1d7a5 FW |
28 | static char default_sort_order[] = "avg, max, switch, runtime"; |
29 | static char *sort_order = default_sort_order; | |
30 | ||
55ffb7a6 MG |
31 | static int profile_cpu = -1; |
32 | ||
b1ffe8f3 IM |
33 | #define PR_SET_NAME 15 /* Set process name */ |
34 | #define MAX_CPUS 4096 | |
0a02ad93 | 35 | |
b1ffe8f3 IM |
36 | static u64 run_measurement_overhead; |
37 | static u64 sleep_measurement_overhead; | |
ec156764 | 38 | |
b1ffe8f3 IM |
39 | #define COMM_LEN 20 |
40 | #define SYM_LEN 129 | |
ec156764 | 41 | |
b1ffe8f3 | 42 | #define MAX_PID 65536 |
ec156764 | 43 | |
b1ffe8f3 | 44 | static unsigned long nr_tasks; |
ec156764 | 45 | |
39aeb52f | 46 | struct sched_atom; |
ec156764 | 47 | |
b1ffe8f3 IM |
48 | struct task_desc { |
49 | unsigned long nr; | |
50 | unsigned long pid; | |
51 | char comm[COMM_LEN]; | |
ec156764 | 52 | |
b1ffe8f3 IM |
53 | unsigned long nr_events; |
54 | unsigned long curr_event; | |
39aeb52f | 55 | struct sched_atom **atoms; |
b1ffe8f3 IM |
56 | |
57 | pthread_t thread; | |
58 | sem_t sleep_sem; | |
ec156764 | 59 | |
b1ffe8f3 IM |
60 | sem_t ready_for_work; |
61 | sem_t work_done_sem; | |
62 | ||
63 | u64 cpu_usage; | |
64 | }; | |
65 | ||
66 | enum sched_event_type { | |
67 | SCHED_EVENT_RUN, | |
68 | SCHED_EVENT_SLEEP, | |
69 | SCHED_EVENT_WAKEUP, | |
55ffb7a6 | 70 | SCHED_EVENT_MIGRATION, |
b1ffe8f3 IM |
71 | }; |
72 | ||
39aeb52f | 73 | struct sched_atom { |
b1ffe8f3 IM |
74 | enum sched_event_type type; |
75 | u64 timestamp; | |
76 | u64 duration; | |
77 | unsigned long nr; | |
78 | int specific_wait; | |
79 | sem_t *wait_sem; | |
80 | struct task_desc *wakee; | |
81 | }; | |
82 | ||
83 | static struct task_desc *pid_to_task[MAX_PID]; | |
84 | ||
85 | static struct task_desc **tasks; | |
86 | ||
87 | static pthread_mutex_t start_work_mutex = PTHREAD_MUTEX_INITIALIZER; | |
88 | static u64 start_time; | |
89 | ||
90 | static pthread_mutex_t work_done_wait_mutex = PTHREAD_MUTEX_INITIALIZER; | |
ec156764 | 91 | |
b1ffe8f3 IM |
92 | static unsigned long nr_run_events; |
93 | static unsigned long nr_sleep_events; | |
94 | static unsigned long nr_wakeup_events; | |
95 | ||
96 | static unsigned long nr_sleep_corrections; | |
97 | static unsigned long nr_run_events_optimized; | |
98 | ||
99 | static unsigned long targetless_wakeups; | |
100 | static unsigned long multitarget_wakeups; | |
101 | ||
102 | static u64 cpu_usage; | |
103 | static u64 runavg_cpu_usage; | |
104 | static u64 parent_cpu_usage; | |
105 | static u64 runavg_parent_cpu_usage; | |
106 | ||
107 | static unsigned long nr_runs; | |
108 | static u64 sum_runtime; | |
109 | static u64 sum_fluct; | |
110 | static u64 run_avg; | |
111 | ||
112 | static unsigned long replay_repeat = 10; | |
ea57c4f5 | 113 | static unsigned long nr_timestamps; |
dc02bf71 IM |
114 | static unsigned long nr_unordered_timestamps; |
115 | static unsigned long nr_state_machine_bugs; | |
c8a37751 | 116 | static unsigned long nr_context_switch_bugs; |
dc02bf71 IM |
117 | static unsigned long nr_events; |
118 | static unsigned long nr_lost_chunks; | |
119 | static unsigned long nr_lost_events; | |
b1ffe8f3 IM |
120 | |
121 | #define TASK_STATE_TO_CHAR_STR "RSDTtZX" | |
122 | ||
123 | enum thread_state { | |
124 | THREAD_SLEEPING = 0, | |
125 | THREAD_WAIT_CPU, | |
126 | THREAD_SCHED_IN, | |
127 | THREAD_IGNORE | |
128 | }; | |
129 | ||
130 | struct work_atom { | |
131 | struct list_head list; | |
132 | enum thread_state state; | |
aa1ab9d2 | 133 | u64 sched_out_time; |
b1ffe8f3 IM |
134 | u64 wake_up_time; |
135 | u64 sched_in_time; | |
136 | u64 runtime; | |
137 | }; | |
138 | ||
39aeb52f | 139 | struct work_atoms { |
140 | struct list_head work_list; | |
b1ffe8f3 IM |
141 | struct thread *thread; |
142 | struct rb_node node; | |
143 | u64 max_lat; | |
144 | u64 total_lat; | |
145 | u64 nb_atoms; | |
146 | u64 total_runtime; | |
147 | }; | |
148 | ||
39aeb52f | 149 | typedef int (*sort_fn_t)(struct work_atoms *, struct work_atoms *); |
b1ffe8f3 IM |
150 | |
151 | static struct rb_root atom_root, sorted_atom_root; | |
152 | ||
153 | static u64 all_runtime; | |
154 | static u64 all_count; | |
155 | ||
b1ffe8f3 IM |
156 | |
157 | static u64 get_nsecs(void) | |
ec156764 IM |
158 | { |
159 | struct timespec ts; | |
160 | ||
161 | clock_gettime(CLOCK_MONOTONIC, &ts); | |
162 | ||
163 | return ts.tv_sec * 1000000000ULL + ts.tv_nsec; | |
164 | } | |
165 | ||
b1ffe8f3 | 166 | static void burn_nsecs(u64 nsecs) |
ec156764 | 167 | { |
b1ffe8f3 | 168 | u64 T0 = get_nsecs(), T1; |
ec156764 IM |
169 | |
170 | do { | |
171 | T1 = get_nsecs(); | |
172 | } while (T1 + run_measurement_overhead < T0 + nsecs); | |
173 | } | |
174 | ||
b1ffe8f3 | 175 | static void sleep_nsecs(u64 nsecs) |
ec156764 IM |
176 | { |
177 | struct timespec ts; | |
178 | ||
179 | ts.tv_nsec = nsecs % 999999999; | |
180 | ts.tv_sec = nsecs / 999999999; | |
181 | ||
182 | nanosleep(&ts, NULL); | |
183 | } | |
184 | ||
185 | static void calibrate_run_measurement_overhead(void) | |
186 | { | |
b1ffe8f3 | 187 | u64 T0, T1, delta, min_delta = 1000000000ULL; |
ec156764 IM |
188 | int i; |
189 | ||
190 | for (i = 0; i < 10; i++) { | |
191 | T0 = get_nsecs(); | |
192 | burn_nsecs(0); | |
193 | T1 = get_nsecs(); | |
194 | delta = T1-T0; | |
195 | min_delta = min(min_delta, delta); | |
196 | } | |
197 | run_measurement_overhead = min_delta; | |
198 | ||
ad236fd2 | 199 | printf("run measurement overhead: %Ld nsecs\n", min_delta); |
ec156764 IM |
200 | } |
201 | ||
202 | static void calibrate_sleep_measurement_overhead(void) | |
203 | { | |
b1ffe8f3 | 204 | u64 T0, T1, delta, min_delta = 1000000000ULL; |
ec156764 IM |
205 | int i; |
206 | ||
207 | for (i = 0; i < 10; i++) { | |
208 | T0 = get_nsecs(); | |
209 | sleep_nsecs(10000); | |
210 | T1 = get_nsecs(); | |
211 | delta = T1-T0; | |
212 | min_delta = min(min_delta, delta); | |
213 | } | |
214 | min_delta -= 10000; | |
215 | sleep_measurement_overhead = min_delta; | |
216 | ||
ad236fd2 | 217 | printf("sleep measurement overhead: %Ld nsecs\n", min_delta); |
ec156764 IM |
218 | } |
219 | ||
39aeb52f | 220 | static struct sched_atom * |
b1ffe8f3 | 221 | get_new_event(struct task_desc *task, u64 timestamp) |
ec156764 | 222 | { |
36479484 | 223 | struct sched_atom *event = zalloc(sizeof(*event)); |
ec156764 IM |
224 | unsigned long idx = task->nr_events; |
225 | size_t size; | |
226 | ||
227 | event->timestamp = timestamp; | |
228 | event->nr = idx; | |
229 | ||
230 | task->nr_events++; | |
39aeb52f | 231 | size = sizeof(struct sched_atom *) * task->nr_events; |
232 | task->atoms = realloc(task->atoms, size); | |
233 | BUG_ON(!task->atoms); | |
ec156764 | 234 | |
39aeb52f | 235 | task->atoms[idx] = event; |
ec156764 IM |
236 | |
237 | return event; | |
238 | } | |
239 | ||
39aeb52f | 240 | static struct sched_atom *last_event(struct task_desc *task) |
ec156764 IM |
241 | { |
242 | if (!task->nr_events) | |
243 | return NULL; | |
244 | ||
39aeb52f | 245 | return task->atoms[task->nr_events - 1]; |
ec156764 IM |
246 | } |
247 | ||
248 | static void | |
b1ffe8f3 | 249 | add_sched_event_run(struct task_desc *task, u64 timestamp, u64 duration) |
ec156764 | 250 | { |
39aeb52f | 251 | struct sched_atom *event, *curr_event = last_event(task); |
ec156764 IM |
252 | |
253 | /* | |
fbf94829 IM |
254 | * optimize an existing RUN event by merging this one |
255 | * to it: | |
256 | */ | |
ec156764 IM |
257 | if (curr_event && curr_event->type == SCHED_EVENT_RUN) { |
258 | nr_run_events_optimized++; | |
259 | curr_event->duration += duration; | |
260 | return; | |
261 | } | |
262 | ||
263 | event = get_new_event(task, timestamp); | |
264 | ||
265 | event->type = SCHED_EVENT_RUN; | |
266 | event->duration = duration; | |
267 | ||
268 | nr_run_events++; | |
269 | } | |
270 | ||
ec156764 | 271 | static void |
b1ffe8f3 | 272 | add_sched_event_wakeup(struct task_desc *task, u64 timestamp, |
ec156764 IM |
273 | struct task_desc *wakee) |
274 | { | |
39aeb52f | 275 | struct sched_atom *event, *wakee_event; |
ec156764 IM |
276 | |
277 | event = get_new_event(task, timestamp); | |
278 | event->type = SCHED_EVENT_WAKEUP; | |
279 | event->wakee = wakee; | |
280 | ||
281 | wakee_event = last_event(wakee); | |
282 | if (!wakee_event || wakee_event->type != SCHED_EVENT_SLEEP) { | |
283 | targetless_wakeups++; | |
284 | return; | |
285 | } | |
286 | if (wakee_event->wait_sem) { | |
287 | multitarget_wakeups++; | |
288 | return; | |
289 | } | |
290 | ||
36479484 | 291 | wakee_event->wait_sem = zalloc(sizeof(*wakee_event->wait_sem)); |
ec156764 IM |
292 | sem_init(wakee_event->wait_sem, 0, 0); |
293 | wakee_event->specific_wait = 1; | |
294 | event->wait_sem = wakee_event->wait_sem; | |
295 | ||
296 | nr_wakeup_events++; | |
297 | } | |
298 | ||
299 | static void | |
b1ffe8f3 | 300 | add_sched_event_sleep(struct task_desc *task, u64 timestamp, |
ad236fd2 | 301 | u64 task_state __used) |
ec156764 | 302 | { |
39aeb52f | 303 | struct sched_atom *event = get_new_event(task, timestamp); |
ec156764 IM |
304 | |
305 | event->type = SCHED_EVENT_SLEEP; | |
306 | ||
307 | nr_sleep_events++; | |
308 | } | |
309 | ||
310 | static struct task_desc *register_pid(unsigned long pid, const char *comm) | |
311 | { | |
312 | struct task_desc *task; | |
313 | ||
314 | BUG_ON(pid >= MAX_PID); | |
315 | ||
316 | task = pid_to_task[pid]; | |
317 | ||
318 | if (task) | |
319 | return task; | |
320 | ||
36479484 | 321 | task = zalloc(sizeof(*task)); |
ec156764 IM |
322 | task->pid = pid; |
323 | task->nr = nr_tasks; | |
324 | strcpy(task->comm, comm); | |
325 | /* | |
326 | * every task starts in sleeping state - this gets ignored | |
327 | * if there's no wakeup pointing to this sleep state: | |
328 | */ | |
329 | add_sched_event_sleep(task, 0, 0); | |
330 | ||
331 | pid_to_task[pid] = task; | |
332 | nr_tasks++; | |
333 | tasks = realloc(tasks, nr_tasks*sizeof(struct task_task *)); | |
334 | BUG_ON(!tasks); | |
335 | tasks[task->nr] = task; | |
336 | ||
ad236fd2 IM |
337 | if (verbose) |
338 | printf("registered task #%ld, PID %ld (%s)\n", nr_tasks, pid, comm); | |
ec156764 IM |
339 | |
340 | return task; | |
341 | } | |
342 | ||
343 | ||
ec156764 IM |
344 | static void print_task_traces(void) |
345 | { | |
346 | struct task_desc *task; | |
347 | unsigned long i; | |
348 | ||
349 | for (i = 0; i < nr_tasks; i++) { | |
350 | task = tasks[i]; | |
ad236fd2 | 351 | printf("task %6ld (%20s:%10ld), nr_events: %ld\n", |
ec156764 IM |
352 | task->nr, task->comm, task->pid, task->nr_events); |
353 | } | |
354 | } | |
355 | ||
356 | static void add_cross_task_wakeups(void) | |
357 | { | |
358 | struct task_desc *task1, *task2; | |
359 | unsigned long i, j; | |
360 | ||
361 | for (i = 0; i < nr_tasks; i++) { | |
362 | task1 = tasks[i]; | |
363 | j = i + 1; | |
364 | if (j == nr_tasks) | |
365 | j = 0; | |
366 | task2 = tasks[j]; | |
367 | add_sched_event_wakeup(task1, 0, task2); | |
368 | } | |
369 | } | |
370 | ||
371 | static void | |
39aeb52f | 372 | process_sched_event(struct task_desc *this_task __used, struct sched_atom *atom) |
ec156764 IM |
373 | { |
374 | int ret = 0; | |
b1ffe8f3 | 375 | u64 now; |
ec156764 IM |
376 | long long delta; |
377 | ||
378 | now = get_nsecs(); | |
39aeb52f | 379 | delta = start_time + atom->timestamp - now; |
ec156764 | 380 | |
39aeb52f | 381 | switch (atom->type) { |
ec156764 | 382 | case SCHED_EVENT_RUN: |
39aeb52f | 383 | burn_nsecs(atom->duration); |
ec156764 IM |
384 | break; |
385 | case SCHED_EVENT_SLEEP: | |
39aeb52f | 386 | if (atom->wait_sem) |
387 | ret = sem_wait(atom->wait_sem); | |
ec156764 IM |
388 | BUG_ON(ret); |
389 | break; | |
390 | case SCHED_EVENT_WAKEUP: | |
39aeb52f | 391 | if (atom->wait_sem) |
392 | ret = sem_post(atom->wait_sem); | |
ec156764 IM |
393 | BUG_ON(ret); |
394 | break; | |
55ffb7a6 MG |
395 | case SCHED_EVENT_MIGRATION: |
396 | break; | |
ec156764 IM |
397 | default: |
398 | BUG_ON(1); | |
399 | } | |
400 | } | |
401 | ||
b1ffe8f3 | 402 | static u64 get_cpu_usage_nsec_parent(void) |
ec156764 IM |
403 | { |
404 | struct rusage ru; | |
b1ffe8f3 | 405 | u64 sum; |
ec156764 IM |
406 | int err; |
407 | ||
408 | err = getrusage(RUSAGE_SELF, &ru); | |
409 | BUG_ON(err); | |
410 | ||
411 | sum = ru.ru_utime.tv_sec*1e9 + ru.ru_utime.tv_usec*1e3; | |
412 | sum += ru.ru_stime.tv_sec*1e9 + ru.ru_stime.tv_usec*1e3; | |
413 | ||
414 | return sum; | |
415 | } | |
416 | ||
b1ffe8f3 | 417 | static u64 get_cpu_usage_nsec_self(void) |
ec156764 IM |
418 | { |
419 | char filename [] = "/proc/1234567890/sched"; | |
420 | unsigned long msecs, nsecs; | |
421 | char *line = NULL; | |
b1ffe8f3 | 422 | u64 total = 0; |
ec156764 IM |
423 | size_t len = 0; |
424 | ssize_t chars; | |
425 | FILE *file; | |
426 | int ret; | |
427 | ||
428 | sprintf(filename, "/proc/%d/sched", getpid()); | |
429 | file = fopen(filename, "r"); | |
430 | BUG_ON(!file); | |
431 | ||
432 | while ((chars = getline(&line, &len, file)) != -1) { | |
ec156764 IM |
433 | ret = sscanf(line, "se.sum_exec_runtime : %ld.%06ld\n", |
434 | &msecs, &nsecs); | |
435 | if (ret == 2) { | |
436 | total = msecs*1e6 + nsecs; | |
ec156764 IM |
437 | break; |
438 | } | |
439 | } | |
440 | if (line) | |
441 | free(line); | |
442 | fclose(file); | |
443 | ||
444 | return total; | |
445 | } | |
446 | ||
447 | static void *thread_func(void *ctx) | |
448 | { | |
449 | struct task_desc *this_task = ctx; | |
b1ffe8f3 | 450 | u64 cpu_usage_0, cpu_usage_1; |
ec156764 IM |
451 | unsigned long i, ret; |
452 | char comm2[22]; | |
453 | ||
ec156764 IM |
454 | sprintf(comm2, ":%s", this_task->comm); |
455 | prctl(PR_SET_NAME, comm2); | |
456 | ||
457 | again: | |
458 | ret = sem_post(&this_task->ready_for_work); | |
459 | BUG_ON(ret); | |
ec156764 IM |
460 | ret = pthread_mutex_lock(&start_work_mutex); |
461 | BUG_ON(ret); | |
462 | ret = pthread_mutex_unlock(&start_work_mutex); | |
463 | BUG_ON(ret); | |
ec156764 IM |
464 | |
465 | cpu_usage_0 = get_cpu_usage_nsec_self(); | |
466 | ||
467 | for (i = 0; i < this_task->nr_events; i++) { | |
468 | this_task->curr_event = i; | |
39aeb52f | 469 | process_sched_event(this_task, this_task->atoms[i]); |
ec156764 IM |
470 | } |
471 | ||
472 | cpu_usage_1 = get_cpu_usage_nsec_self(); | |
473 | this_task->cpu_usage = cpu_usage_1 - cpu_usage_0; | |
474 | ||
ec156764 IM |
475 | ret = sem_post(&this_task->work_done_sem); |
476 | BUG_ON(ret); | |
ec156764 IM |
477 | |
478 | ret = pthread_mutex_lock(&work_done_wait_mutex); | |
479 | BUG_ON(ret); | |
480 | ret = pthread_mutex_unlock(&work_done_wait_mutex); | |
481 | BUG_ON(ret); | |
ec156764 IM |
482 | |
483 | goto again; | |
484 | } | |
485 | ||
486 | static void create_tasks(void) | |
487 | { | |
488 | struct task_desc *task; | |
489 | pthread_attr_t attr; | |
490 | unsigned long i; | |
491 | int err; | |
492 | ||
493 | err = pthread_attr_init(&attr); | |
494 | BUG_ON(err); | |
495 | err = pthread_attr_setstacksize(&attr, (size_t)(16*1024)); | |
496 | BUG_ON(err); | |
497 | err = pthread_mutex_lock(&start_work_mutex); | |
498 | BUG_ON(err); | |
499 | err = pthread_mutex_lock(&work_done_wait_mutex); | |
500 | BUG_ON(err); | |
501 | for (i = 0; i < nr_tasks; i++) { | |
502 | task = tasks[i]; | |
503 | sem_init(&task->sleep_sem, 0, 0); | |
504 | sem_init(&task->ready_for_work, 0, 0); | |
505 | sem_init(&task->work_done_sem, 0, 0); | |
506 | task->curr_event = 0; | |
507 | err = pthread_create(&task->thread, &attr, thread_func, task); | |
508 | BUG_ON(err); | |
509 | } | |
510 | } | |
511 | ||
ec156764 IM |
512 | static void wait_for_tasks(void) |
513 | { | |
b1ffe8f3 | 514 | u64 cpu_usage_0, cpu_usage_1; |
ec156764 IM |
515 | struct task_desc *task; |
516 | unsigned long i, ret; | |
517 | ||
ec156764 | 518 | start_time = get_nsecs(); |
ec156764 IM |
519 | cpu_usage = 0; |
520 | pthread_mutex_unlock(&work_done_wait_mutex); | |
521 | ||
522 | for (i = 0; i < nr_tasks; i++) { | |
523 | task = tasks[i]; | |
524 | ret = sem_wait(&task->ready_for_work); | |
525 | BUG_ON(ret); | |
526 | sem_init(&task->ready_for_work, 0, 0); | |
527 | } | |
528 | ret = pthread_mutex_lock(&work_done_wait_mutex); | |
529 | BUG_ON(ret); | |
530 | ||
531 | cpu_usage_0 = get_cpu_usage_nsec_parent(); | |
532 | ||
533 | pthread_mutex_unlock(&start_work_mutex); | |
534 | ||
ec156764 IM |
535 | for (i = 0; i < nr_tasks; i++) { |
536 | task = tasks[i]; | |
537 | ret = sem_wait(&task->work_done_sem); | |
538 | BUG_ON(ret); | |
539 | sem_init(&task->work_done_sem, 0, 0); | |
540 | cpu_usage += task->cpu_usage; | |
541 | task->cpu_usage = 0; | |
542 | } | |
543 | ||
544 | cpu_usage_1 = get_cpu_usage_nsec_parent(); | |
545 | if (!runavg_cpu_usage) | |
546 | runavg_cpu_usage = cpu_usage; | |
547 | runavg_cpu_usage = (runavg_cpu_usage*9 + cpu_usage)/10; | |
548 | ||
549 | parent_cpu_usage = cpu_usage_1 - cpu_usage_0; | |
550 | if (!runavg_parent_cpu_usage) | |
551 | runavg_parent_cpu_usage = parent_cpu_usage; | |
552 | runavg_parent_cpu_usage = (runavg_parent_cpu_usage*9 + | |
553 | parent_cpu_usage)/10; | |
554 | ||
555 | ret = pthread_mutex_lock(&start_work_mutex); | |
556 | BUG_ON(ret); | |
557 | ||
558 | for (i = 0; i < nr_tasks; i++) { | |
559 | task = tasks[i]; | |
560 | sem_init(&task->sleep_sem, 0, 0); | |
561 | task->curr_event = 0; | |
562 | } | |
563 | } | |
564 | ||
ec156764 IM |
565 | static void run_one_test(void) |
566 | { | |
b1ffe8f3 | 567 | u64 T0, T1, delta, avg_delta, fluct, std_dev; |
ec156764 IM |
568 | |
569 | T0 = get_nsecs(); | |
570 | wait_for_tasks(); | |
571 | T1 = get_nsecs(); | |
572 | ||
573 | delta = T1 - T0; | |
574 | sum_runtime += delta; | |
575 | nr_runs++; | |
576 | ||
577 | avg_delta = sum_runtime / nr_runs; | |
578 | if (delta < avg_delta) | |
579 | fluct = avg_delta - delta; | |
580 | else | |
581 | fluct = delta - avg_delta; | |
582 | sum_fluct += fluct; | |
583 | std_dev = sum_fluct / nr_runs / sqrt(nr_runs); | |
584 | if (!run_avg) | |
585 | run_avg = delta; | |
586 | run_avg = (run_avg*9 + delta)/10; | |
587 | ||
ad236fd2 | 588 | printf("#%-3ld: %0.3f, ", |
ec156764 IM |
589 | nr_runs, (double)delta/1000000.0); |
590 | ||
ad236fd2 | 591 | printf("ravg: %0.2f, ", |
ec156764 IM |
592 | (double)run_avg/1e6); |
593 | ||
ad236fd2 | 594 | printf("cpu: %0.2f / %0.2f", |
ec156764 IM |
595 | (double)cpu_usage/1e6, (double)runavg_cpu_usage/1e6); |
596 | ||
597 | #if 0 | |
598 | /* | |
fbf94829 IM |
599 | * rusage statistics done by the parent, these are less |
600 | * accurate than the sum_exec_runtime based statistics: | |
601 | */ | |
ad236fd2 | 602 | printf(" [%0.2f / %0.2f]", |
ec156764 IM |
603 | (double)parent_cpu_usage/1e6, |
604 | (double)runavg_parent_cpu_usage/1e6); | |
605 | #endif | |
606 | ||
ad236fd2 | 607 | printf("\n"); |
ec156764 IM |
608 | |
609 | if (nr_sleep_corrections) | |
ad236fd2 | 610 | printf(" (%ld sleep corrections)\n", nr_sleep_corrections); |
ec156764 IM |
611 | nr_sleep_corrections = 0; |
612 | } | |
613 | ||
614 | static void test_calibrations(void) | |
615 | { | |
b1ffe8f3 | 616 | u64 T0, T1; |
ec156764 IM |
617 | |
618 | T0 = get_nsecs(); | |
619 | burn_nsecs(1e6); | |
620 | T1 = get_nsecs(); | |
621 | ||
ad236fd2 | 622 | printf("the run test took %Ld nsecs\n", T1-T0); |
ec156764 IM |
623 | |
624 | T0 = get_nsecs(); | |
625 | sleep_nsecs(1e6); | |
626 | T1 = get_nsecs(); | |
627 | ||
ad236fd2 | 628 | printf("the sleep test took %Ld nsecs\n", T1-T0); |
ec156764 IM |
629 | } |
630 | ||
46538818 FW |
631 | #define FILL_FIELD(ptr, field, event, data) \ |
632 | ptr.field = (typeof(ptr.field)) raw_field_value(event, #field, data) | |
633 | ||
634 | #define FILL_ARRAY(ptr, array, event, data) \ | |
635 | do { \ | |
636 | void *__array = raw_field_ptr(event, #array, data); \ | |
637 | memcpy(ptr.array, __array, sizeof(ptr.array)); \ | |
638 | } while(0) | |
639 | ||
640 | #define FILL_COMMON_FIELDS(ptr, event, data) \ | |
641 | do { \ | |
642 | FILL_FIELD(ptr, common_type, event, data); \ | |
643 | FILL_FIELD(ptr, common_flags, event, data); \ | |
644 | FILL_FIELD(ptr, common_preempt_count, event, data); \ | |
645 | FILL_FIELD(ptr, common_pid, event, data); \ | |
646 | FILL_FIELD(ptr, common_tgid, event, data); \ | |
647 | } while (0) | |
648 | ||
419ab0d6 FW |
649 | |
650 | ||
651 | struct trace_switch_event { | |
652 | u32 size; | |
653 | ||
654 | u16 common_type; | |
655 | u8 common_flags; | |
656 | u8 common_preempt_count; | |
657 | u32 common_pid; | |
658 | u32 common_tgid; | |
659 | ||
660 | char prev_comm[16]; | |
661 | u32 prev_pid; | |
662 | u32 prev_prio; | |
663 | u64 prev_state; | |
664 | char next_comm[16]; | |
665 | u32 next_pid; | |
666 | u32 next_prio; | |
667 | }; | |
668 | ||
39aeb52f | 669 | struct trace_runtime_event { |
670 | u32 size; | |
671 | ||
672 | u16 common_type; | |
673 | u8 common_flags; | |
674 | u8 common_preempt_count; | |
675 | u32 common_pid; | |
676 | u32 common_tgid; | |
677 | ||
678 | char comm[16]; | |
679 | u32 pid; | |
680 | u64 runtime; | |
681 | u64 vruntime; | |
682 | }; | |
419ab0d6 | 683 | |
fbf94829 IM |
684 | struct trace_wakeup_event { |
685 | u32 size; | |
686 | ||
687 | u16 common_type; | |
688 | u8 common_flags; | |
689 | u8 common_preempt_count; | |
690 | u32 common_pid; | |
691 | u32 common_tgid; | |
692 | ||
693 | char comm[16]; | |
694 | u32 pid; | |
695 | ||
696 | u32 prio; | |
697 | u32 success; | |
698 | u32 cpu; | |
699 | }; | |
700 | ||
419ab0d6 FW |
701 | struct trace_fork_event { |
702 | u32 size; | |
46538818 | 703 | |
419ab0d6 FW |
704 | u16 common_type; |
705 | u8 common_flags; | |
706 | u8 common_preempt_count; | |
707 | u32 common_pid; | |
708 | u32 common_tgid; | |
709 | ||
710 | char parent_comm[16]; | |
711 | u32 parent_pid; | |
712 | char child_comm[16]; | |
713 | u32 child_pid; | |
714 | }; | |
715 | ||
55ffb7a6 MG |
716 | struct trace_migrate_task_event { |
717 | u32 size; | |
718 | ||
719 | u16 common_type; | |
720 | u8 common_flags; | |
721 | u8 common_preempt_count; | |
722 | u32 common_pid; | |
723 | u32 common_tgid; | |
724 | ||
725 | char comm[16]; | |
726 | u32 pid; | |
727 | ||
728 | u32 prio; | |
729 | u32 cpu; | |
730 | }; | |
731 | ||
419ab0d6 FW |
732 | struct trace_sched_handler { |
733 | void (*switch_event)(struct trace_switch_event *, | |
734 | struct event *, | |
735 | int cpu, | |
736 | u64 timestamp, | |
737 | struct thread *thread); | |
738 | ||
39aeb52f | 739 | void (*runtime_event)(struct trace_runtime_event *, |
740 | struct event *, | |
741 | int cpu, | |
742 | u64 timestamp, | |
743 | struct thread *thread); | |
744 | ||
419ab0d6 FW |
745 | void (*wakeup_event)(struct trace_wakeup_event *, |
746 | struct event *, | |
747 | int cpu, | |
748 | u64 timestamp, | |
749 | struct thread *thread); | |
750 | ||
751 | void (*fork_event)(struct trace_fork_event *, | |
752 | struct event *, | |
753 | int cpu, | |
754 | u64 timestamp, | |
755 | struct thread *thread); | |
55ffb7a6 MG |
756 | |
757 | void (*migrate_task_event)(struct trace_migrate_task_event *, | |
758 | struct event *, | |
759 | int cpu, | |
760 | u64 timestamp, | |
761 | struct thread *thread); | |
419ab0d6 | 762 | }; |
46538818 | 763 | |
46538818 | 764 | |
419ab0d6 FW |
765 | static void |
766 | replay_wakeup_event(struct trace_wakeup_event *wakeup_event, | |
767 | struct event *event, | |
768 | int cpu __used, | |
769 | u64 timestamp __used, | |
770 | struct thread *thread __used) | |
771 | { | |
772 | struct task_desc *waker, *wakee; | |
fbf94829 | 773 | |
ad236fd2 IM |
774 | if (verbose) { |
775 | printf("sched_wakeup event %p\n", event); | |
fbf94829 | 776 | |
ad236fd2 | 777 | printf(" ... pid %d woke up %s/%d\n", |
419ab0d6 FW |
778 | wakeup_event->common_pid, |
779 | wakeup_event->comm, | |
780 | wakeup_event->pid); | |
ad236fd2 | 781 | } |
fbf94829 | 782 | |
419ab0d6 FW |
783 | waker = register_pid(wakeup_event->common_pid, "<unknown>"); |
784 | wakee = register_pid(wakeup_event->pid, wakeup_event->comm); | |
fbf94829 IM |
785 | |
786 | add_sched_event_wakeup(waker, timestamp, wakee); | |
ec156764 IM |
787 | } |
788 | ||
d1153389 | 789 | static u64 cpu_last_switched[MAX_CPUS]; |
fbf94829 IM |
790 | |
791 | static void | |
419ab0d6 FW |
792 | replay_switch_event(struct trace_switch_event *switch_event, |
793 | struct event *event, | |
794 | int cpu, | |
795 | u64 timestamp, | |
796 | struct thread *thread __used) | |
ec156764 | 797 | { |
fbf94829 IM |
798 | struct task_desc *prev, *next; |
799 | u64 timestamp0; | |
800 | s64 delta; | |
801 | ||
ad236fd2 IM |
802 | if (verbose) |
803 | printf("sched_switch event %p\n", event); | |
804 | ||
fbf94829 IM |
805 | if (cpu >= MAX_CPUS || cpu < 0) |
806 | return; | |
807 | ||
808 | timestamp0 = cpu_last_switched[cpu]; | |
809 | if (timestamp0) | |
810 | delta = timestamp - timestamp0; | |
811 | else | |
812 | delta = 0; | |
813 | ||
814 | if (delta < 0) | |
815 | die("hm, delta: %Ld < 0 ?\n", delta); | |
816 | ||
ad236fd2 IM |
817 | if (verbose) { |
818 | printf(" ... switch from %s/%d to %s/%d [ran %Ld nsecs]\n", | |
419ab0d6 FW |
819 | switch_event->prev_comm, switch_event->prev_pid, |
820 | switch_event->next_comm, switch_event->next_pid, | |
ad236fd2 IM |
821 | delta); |
822 | } | |
fbf94829 | 823 | |
419ab0d6 FW |
824 | prev = register_pid(switch_event->prev_pid, switch_event->prev_comm); |
825 | next = register_pid(switch_event->next_pid, switch_event->next_comm); | |
fbf94829 IM |
826 | |
827 | cpu_last_switched[cpu] = timestamp; | |
828 | ||
829 | add_sched_event_run(prev, timestamp, delta); | |
419ab0d6 | 830 | add_sched_event_sleep(prev, timestamp, switch_event->prev_state); |
fbf94829 IM |
831 | } |
832 | ||
fbf94829 | 833 | |
419ab0d6 FW |
834 | static void |
835 | replay_fork_event(struct trace_fork_event *fork_event, | |
836 | struct event *event, | |
837 | int cpu __used, | |
838 | u64 timestamp __used, | |
839 | struct thread *thread __used) | |
840 | { | |
841 | if (verbose) { | |
842 | printf("sched_fork event %p\n", event); | |
843 | printf("... parent: %s/%d\n", fork_event->parent_comm, fork_event->parent_pid); | |
844 | printf("... child: %s/%d\n", fork_event->child_comm, fork_event->child_pid); | |
845 | } | |
846 | register_pid(fork_event->parent_pid, fork_event->parent_comm); | |
847 | register_pid(fork_event->child_pid, fork_event->child_comm); | |
848 | } | |
fbf94829 | 849 | |
419ab0d6 | 850 | static struct trace_sched_handler replay_ops = { |
ea92ed5a IM |
851 | .wakeup_event = replay_wakeup_event, |
852 | .switch_event = replay_switch_event, | |
853 | .fork_event = replay_fork_event, | |
fbf94829 IM |
854 | }; |
855 | ||
b1ffe8f3 IM |
856 | struct sort_dimension { |
857 | const char *name; | |
b5fae128 | 858 | sort_fn_t cmp; |
b1ffe8f3 IM |
859 | struct list_head list; |
860 | }; | |
861 | ||
862 | static LIST_HEAD(cmp_pid); | |
863 | ||
daa1d7a5 | 864 | static int |
39aeb52f | 865 | thread_lat_cmp(struct list_head *list, struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 FW |
866 | { |
867 | struct sort_dimension *sort; | |
868 | int ret = 0; | |
869 | ||
b5fae128 IM |
870 | BUG_ON(list_empty(list)); |
871 | ||
daa1d7a5 FW |
872 | list_for_each_entry(sort, list, list) { |
873 | ret = sort->cmp(l, r); | |
874 | if (ret) | |
875 | return ret; | |
876 | } | |
877 | ||
878 | return ret; | |
879 | } | |
880 | ||
39aeb52f | 881 | static struct work_atoms * |
b5fae128 IM |
882 | thread_atoms_search(struct rb_root *root, struct thread *thread, |
883 | struct list_head *sort_list) | |
884 | { | |
885 | struct rb_node *node = root->rb_node; | |
39aeb52f | 886 | struct work_atoms key = { .thread = thread }; |
b5fae128 IM |
887 | |
888 | while (node) { | |
39aeb52f | 889 | struct work_atoms *atoms; |
b5fae128 IM |
890 | int cmp; |
891 | ||
39aeb52f | 892 | atoms = container_of(node, struct work_atoms, node); |
b5fae128 IM |
893 | |
894 | cmp = thread_lat_cmp(sort_list, &key, atoms); | |
895 | if (cmp > 0) | |
896 | node = node->rb_left; | |
897 | else if (cmp < 0) | |
898 | node = node->rb_right; | |
899 | else { | |
900 | BUG_ON(thread != atoms->thread); | |
901 | return atoms; | |
902 | } | |
903 | } | |
904 | return NULL; | |
905 | } | |
906 | ||
cdce9d73 | 907 | static void |
39aeb52f | 908 | __thread_latency_insert(struct rb_root *root, struct work_atoms *data, |
daa1d7a5 | 909 | struct list_head *sort_list) |
cdce9d73 FW |
910 | { |
911 | struct rb_node **new = &(root->rb_node), *parent = NULL; | |
912 | ||
913 | while (*new) { | |
39aeb52f | 914 | struct work_atoms *this; |
daa1d7a5 | 915 | int cmp; |
cdce9d73 | 916 | |
39aeb52f | 917 | this = container_of(*new, struct work_atoms, node); |
cdce9d73 | 918 | parent = *new; |
daa1d7a5 FW |
919 | |
920 | cmp = thread_lat_cmp(sort_list, data, this); | |
921 | ||
922 | if (cmp > 0) | |
cdce9d73 | 923 | new = &((*new)->rb_left); |
cdce9d73 | 924 | else |
daa1d7a5 | 925 | new = &((*new)->rb_right); |
cdce9d73 FW |
926 | } |
927 | ||
928 | rb_link_node(&data->node, parent, new); | |
929 | rb_insert_color(&data->node, root); | |
930 | } | |
931 | ||
b1ffe8f3 | 932 | static void thread_atoms_insert(struct thread *thread) |
cdce9d73 | 933 | { |
36479484 | 934 | struct work_atoms *atoms = zalloc(sizeof(*atoms)); |
17562205 | 935 | if (!atoms) |
cdce9d73 FW |
936 | die("No memory"); |
937 | ||
17562205 | 938 | atoms->thread = thread; |
39aeb52f | 939 | INIT_LIST_HEAD(&atoms->work_list); |
b1ffe8f3 | 940 | __thread_latency_insert(&atom_root, atoms, &cmp_pid); |
cdce9d73 FW |
941 | } |
942 | ||
943 | static void | |
944 | latency_fork_event(struct trace_fork_event *fork_event __used, | |
945 | struct event *event __used, | |
946 | int cpu __used, | |
947 | u64 timestamp __used, | |
948 | struct thread *thread __used) | |
949 | { | |
950 | /* should insert the newcomer */ | |
951 | } | |
952 | ||
ea92ed5a | 953 | __used |
cdce9d73 FW |
954 | static char sched_out_state(struct trace_switch_event *switch_event) |
955 | { | |
956 | const char *str = TASK_STATE_TO_CHAR_STR; | |
957 | ||
958 | return str[switch_event->prev_state]; | |
959 | } | |
960 | ||
961 | static void | |
39aeb52f | 962 | add_sched_out_event(struct work_atoms *atoms, |
963 | char run_state, | |
964 | u64 timestamp) | |
cdce9d73 | 965 | { |
36479484 | 966 | struct work_atom *atom = zalloc(sizeof(*atom)); |
b1ffe8f3 | 967 | if (!atom) |
cdce9d73 FW |
968 | die("Non memory"); |
969 | ||
aa1ab9d2 FW |
970 | atom->sched_out_time = timestamp; |
971 | ||
39aeb52f | 972 | if (run_state == 'R') { |
b1ffe8f3 | 973 | atom->state = THREAD_WAIT_CPU; |
aa1ab9d2 | 974 | atom->wake_up_time = atom->sched_out_time; |
c6ced611 FW |
975 | } |
976 | ||
39aeb52f | 977 | list_add_tail(&atom->list, &atoms->work_list); |
cdce9d73 FW |
978 | } |
979 | ||
980 | static void | |
39aeb52f | 981 | add_runtime_event(struct work_atoms *atoms, u64 delta, u64 timestamp __used) |
982 | { | |
983 | struct work_atom *atom; | |
984 | ||
985 | BUG_ON(list_empty(&atoms->work_list)); | |
986 | ||
987 | atom = list_entry(atoms->work_list.prev, struct work_atom, list); | |
988 | ||
989 | atom->runtime += delta; | |
990 | atoms->total_runtime += delta; | |
991 | } | |
992 | ||
993 | static void | |
994 | add_sched_in_event(struct work_atoms *atoms, u64 timestamp) | |
cdce9d73 | 995 | { |
b1ffe8f3 | 996 | struct work_atom *atom; |
66685678 | 997 | u64 delta; |
cdce9d73 | 998 | |
39aeb52f | 999 | if (list_empty(&atoms->work_list)) |
cdce9d73 FW |
1000 | return; |
1001 | ||
39aeb52f | 1002 | atom = list_entry(atoms->work_list.prev, struct work_atom, list); |
cdce9d73 | 1003 | |
b1ffe8f3 | 1004 | if (atom->state != THREAD_WAIT_CPU) |
cdce9d73 FW |
1005 | return; |
1006 | ||
b1ffe8f3 IM |
1007 | if (timestamp < atom->wake_up_time) { |
1008 | atom->state = THREAD_IGNORE; | |
cdce9d73 FW |
1009 | return; |
1010 | } | |
1011 | ||
b1ffe8f3 IM |
1012 | atom->state = THREAD_SCHED_IN; |
1013 | atom->sched_in_time = timestamp; | |
66685678 | 1014 | |
b1ffe8f3 | 1015 | delta = atom->sched_in_time - atom->wake_up_time; |
66685678 FW |
1016 | atoms->total_lat += delta; |
1017 | if (delta > atoms->max_lat) | |
1018 | atoms->max_lat = delta; | |
1019 | atoms->nb_atoms++; | |
cdce9d73 FW |
1020 | } |
1021 | ||
cdce9d73 FW |
1022 | static void |
1023 | latency_switch_event(struct trace_switch_event *switch_event, | |
1024 | struct event *event __used, | |
ea92ed5a | 1025 | int cpu, |
cdce9d73 FW |
1026 | u64 timestamp, |
1027 | struct thread *thread __used) | |
1028 | { | |
39aeb52f | 1029 | struct work_atoms *out_events, *in_events; |
cdce9d73 | 1030 | struct thread *sched_out, *sched_in; |
ea92ed5a IM |
1031 | u64 timestamp0; |
1032 | s64 delta; | |
1033 | ||
39aeb52f | 1034 | BUG_ON(cpu >= MAX_CPUS || cpu < 0); |
ea92ed5a IM |
1035 | |
1036 | timestamp0 = cpu_last_switched[cpu]; | |
1037 | cpu_last_switched[cpu] = timestamp; | |
1038 | if (timestamp0) | |
1039 | delta = timestamp - timestamp0; | |
1040 | else | |
1041 | delta = 0; | |
1042 | ||
1043 | if (delta < 0) | |
1044 | die("hm, delta: %Ld < 0 ?\n", delta); | |
1045 | ||
cdce9d73 | 1046 | |
d5b889f2 ACM |
1047 | sched_out = threads__findnew(switch_event->prev_pid); |
1048 | sched_in = threads__findnew(switch_event->next_pid); | |
cdce9d73 | 1049 | |
39aeb52f | 1050 | out_events = thread_atoms_search(&atom_root, sched_out, &cmp_pid); |
1051 | if (!out_events) { | |
1052 | thread_atoms_insert(sched_out); | |
1053 | out_events = thread_atoms_search(&atom_root, sched_out, &cmp_pid); | |
1054 | if (!out_events) | |
1055 | die("out-event: Internal tree error"); | |
1056 | } | |
1057 | add_sched_out_event(out_events, sched_out_state(switch_event), timestamp); | |
1058 | ||
1059 | in_events = thread_atoms_search(&atom_root, sched_in, &cmp_pid); | |
1060 | if (!in_events) { | |
b1ffe8f3 | 1061 | thread_atoms_insert(sched_in); |
39aeb52f | 1062 | in_events = thread_atoms_search(&atom_root, sched_in, &cmp_pid); |
1063 | if (!in_events) | |
1064 | die("in-event: Internal tree error"); | |
1065 | /* | |
1066 | * Take came in we have not heard about yet, | |
1067 | * add in an initial atom in runnable state: | |
1068 | */ | |
1069 | add_sched_out_event(in_events, 'R', timestamp); | |
cdce9d73 | 1070 | } |
39aeb52f | 1071 | add_sched_in_event(in_events, timestamp); |
1072 | } | |
cdce9d73 | 1073 | |
39aeb52f | 1074 | static void |
1075 | latency_runtime_event(struct trace_runtime_event *runtime_event, | |
1076 | struct event *event __used, | |
1077 | int cpu, | |
1078 | u64 timestamp, | |
1079 | struct thread *this_thread __used) | |
1080 | { | |
d5b889f2 ACM |
1081 | struct thread *thread = threads__findnew(runtime_event->pid); |
1082 | struct work_atoms *atoms = thread_atoms_search(&atom_root, thread, &cmp_pid); | |
39aeb52f | 1083 | |
1084 | BUG_ON(cpu >= MAX_CPUS || cpu < 0); | |
39aeb52f | 1085 | if (!atoms) { |
1086 | thread_atoms_insert(thread); | |
1087 | atoms = thread_atoms_search(&atom_root, thread, &cmp_pid); | |
1088 | if (!atoms) | |
1089 | die("in-event: Internal tree error"); | |
1090 | add_sched_out_event(atoms, 'R', timestamp); | |
cdce9d73 FW |
1091 | } |
1092 | ||
39aeb52f | 1093 | add_runtime_event(atoms, runtime_event->runtime, timestamp); |
cdce9d73 FW |
1094 | } |
1095 | ||
1096 | static void | |
1097 | latency_wakeup_event(struct trace_wakeup_event *wakeup_event, | |
39aeb52f | 1098 | struct event *__event __used, |
cdce9d73 FW |
1099 | int cpu __used, |
1100 | u64 timestamp, | |
1101 | struct thread *thread __used) | |
1102 | { | |
39aeb52f | 1103 | struct work_atoms *atoms; |
b1ffe8f3 | 1104 | struct work_atom *atom; |
cdce9d73 FW |
1105 | struct thread *wakee; |
1106 | ||
1107 | /* Note for later, it may be interesting to observe the failing cases */ | |
1108 | if (!wakeup_event->success) | |
1109 | return; | |
1110 | ||
d5b889f2 | 1111 | wakee = threads__findnew(wakeup_event->pid); |
b5fae128 | 1112 | atoms = thread_atoms_search(&atom_root, wakee, &cmp_pid); |
17562205 | 1113 | if (!atoms) { |
b1ffe8f3 | 1114 | thread_atoms_insert(wakee); |
39aeb52f | 1115 | atoms = thread_atoms_search(&atom_root, wakee, &cmp_pid); |
1116 | if (!atoms) | |
1117 | die("wakeup-event: Internal tree error"); | |
1118 | add_sched_out_event(atoms, 'S', timestamp); | |
cdce9d73 FW |
1119 | } |
1120 | ||
39aeb52f | 1121 | BUG_ON(list_empty(&atoms->work_list)); |
cdce9d73 | 1122 | |
39aeb52f | 1123 | atom = list_entry(atoms->work_list.prev, struct work_atom, list); |
cdce9d73 | 1124 | |
55ffb7a6 MG |
1125 | /* |
1126 | * You WILL be missing events if you've recorded only | |
1127 | * one CPU, or are only looking at only one, so don't | |
1128 | * make useless noise. | |
1129 | */ | |
1130 | if (profile_cpu == -1 && atom->state != THREAD_SLEEPING) | |
dc02bf71 | 1131 | nr_state_machine_bugs++; |
cdce9d73 | 1132 | |
ea57c4f5 IM |
1133 | nr_timestamps++; |
1134 | if (atom->sched_out_time > timestamp) { | |
dc02bf71 | 1135 | nr_unordered_timestamps++; |
aa1ab9d2 | 1136 | return; |
ea57c4f5 | 1137 | } |
aa1ab9d2 | 1138 | |
b1ffe8f3 IM |
1139 | atom->state = THREAD_WAIT_CPU; |
1140 | atom->wake_up_time = timestamp; | |
cdce9d73 FW |
1141 | } |
1142 | ||
55ffb7a6 MG |
1143 | static void |
1144 | latency_migrate_task_event(struct trace_migrate_task_event *migrate_task_event, | |
1145 | struct event *__event __used, | |
1146 | int cpu __used, | |
1147 | u64 timestamp, | |
1148 | struct thread *thread __used) | |
1149 | { | |
1150 | struct work_atoms *atoms; | |
1151 | struct work_atom *atom; | |
1152 | struct thread *migrant; | |
1153 | ||
1154 | /* | |
1155 | * Only need to worry about migration when profiling one CPU. | |
1156 | */ | |
1157 | if (profile_cpu == -1) | |
1158 | return; | |
1159 | ||
d5b889f2 | 1160 | migrant = threads__findnew(migrate_task_event->pid); |
55ffb7a6 MG |
1161 | atoms = thread_atoms_search(&atom_root, migrant, &cmp_pid); |
1162 | if (!atoms) { | |
1163 | thread_atoms_insert(migrant); | |
1164 | register_pid(migrant->pid, migrant->comm); | |
1165 | atoms = thread_atoms_search(&atom_root, migrant, &cmp_pid); | |
1166 | if (!atoms) | |
1167 | die("migration-event: Internal tree error"); | |
1168 | add_sched_out_event(atoms, 'R', timestamp); | |
1169 | } | |
1170 | ||
1171 | BUG_ON(list_empty(&atoms->work_list)); | |
1172 | ||
1173 | atom = list_entry(atoms->work_list.prev, struct work_atom, list); | |
1174 | atom->sched_in_time = atom->sched_out_time = atom->wake_up_time = timestamp; | |
1175 | ||
1176 | nr_timestamps++; | |
1177 | ||
1178 | if (atom->sched_out_time > timestamp) | |
1179 | nr_unordered_timestamps++; | |
1180 | } | |
1181 | ||
cdce9d73 | 1182 | static struct trace_sched_handler lat_ops = { |
ea92ed5a IM |
1183 | .wakeup_event = latency_wakeup_event, |
1184 | .switch_event = latency_switch_event, | |
39aeb52f | 1185 | .runtime_event = latency_runtime_event, |
ea92ed5a | 1186 | .fork_event = latency_fork_event, |
55ffb7a6 | 1187 | .migrate_task_event = latency_migrate_task_event, |
cdce9d73 FW |
1188 | }; |
1189 | ||
39aeb52f | 1190 | static void output_lat_thread(struct work_atoms *work_list) |
cdce9d73 | 1191 | { |
cdce9d73 FW |
1192 | int i; |
1193 | int ret; | |
66685678 | 1194 | u64 avg; |
cdce9d73 | 1195 | |
39aeb52f | 1196 | if (!work_list->nb_atoms) |
cdce9d73 | 1197 | return; |
ea57c4f5 IM |
1198 | /* |
1199 | * Ignore idle threads: | |
1200 | */ | |
80ed0987 | 1201 | if (!strcmp(work_list->thread->comm, "swapper")) |
ea57c4f5 | 1202 | return; |
cdce9d73 | 1203 | |
39aeb52f | 1204 | all_runtime += work_list->total_runtime; |
1205 | all_count += work_list->nb_atoms; | |
66685678 | 1206 | |
80ed0987 | 1207 | ret = printf(" %s:%d ", work_list->thread->comm, work_list->thread->pid); |
cdce9d73 | 1208 | |
08f69e6c | 1209 | for (i = 0; i < 24 - ret; i++) |
cdce9d73 FW |
1210 | printf(" "); |
1211 | ||
39aeb52f | 1212 | avg = work_list->total_lat / work_list->nb_atoms; |
cdce9d73 | 1213 | |
dc02bf71 | 1214 | printf("|%11.3f ms |%9llu | avg:%9.3f ms | max:%9.3f ms |\n", |
39aeb52f | 1215 | (double)work_list->total_runtime / 1e6, |
1216 | work_list->nb_atoms, (double)avg / 1e6, | |
1217 | (double)work_list->max_lat / 1e6); | |
cdce9d73 FW |
1218 | } |
1219 | ||
39aeb52f | 1220 | static int pid_cmp(struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 | 1221 | { |
daa1d7a5 FW |
1222 | if (l->thread->pid < r->thread->pid) |
1223 | return -1; | |
1224 | if (l->thread->pid > r->thread->pid) | |
1225 | return 1; | |
1226 | ||
1227 | return 0; | |
1228 | } | |
1229 | ||
1230 | static struct sort_dimension pid_sort_dimension = { | |
b5fae128 IM |
1231 | .name = "pid", |
1232 | .cmp = pid_cmp, | |
daa1d7a5 FW |
1233 | }; |
1234 | ||
39aeb52f | 1235 | static int avg_cmp(struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 FW |
1236 | { |
1237 | u64 avgl, avgr; | |
1238 | ||
1239 | if (!l->nb_atoms) | |
1240 | return -1; | |
1241 | ||
1242 | if (!r->nb_atoms) | |
1243 | return 1; | |
1244 | ||
1245 | avgl = l->total_lat / l->nb_atoms; | |
1246 | avgr = r->total_lat / r->nb_atoms; | |
1247 | ||
1248 | if (avgl < avgr) | |
1249 | return -1; | |
1250 | if (avgl > avgr) | |
1251 | return 1; | |
1252 | ||
1253 | return 0; | |
1254 | } | |
1255 | ||
1256 | static struct sort_dimension avg_sort_dimension = { | |
b5fae128 IM |
1257 | .name = "avg", |
1258 | .cmp = avg_cmp, | |
daa1d7a5 FW |
1259 | }; |
1260 | ||
39aeb52f | 1261 | static int max_cmp(struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 FW |
1262 | { |
1263 | if (l->max_lat < r->max_lat) | |
1264 | return -1; | |
1265 | if (l->max_lat > r->max_lat) | |
1266 | return 1; | |
1267 | ||
1268 | return 0; | |
1269 | } | |
1270 | ||
1271 | static struct sort_dimension max_sort_dimension = { | |
b5fae128 IM |
1272 | .name = "max", |
1273 | .cmp = max_cmp, | |
daa1d7a5 FW |
1274 | }; |
1275 | ||
39aeb52f | 1276 | static int switch_cmp(struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 FW |
1277 | { |
1278 | if (l->nb_atoms < r->nb_atoms) | |
1279 | return -1; | |
1280 | if (l->nb_atoms > r->nb_atoms) | |
1281 | return 1; | |
1282 | ||
1283 | return 0; | |
1284 | } | |
1285 | ||
1286 | static struct sort_dimension switch_sort_dimension = { | |
b5fae128 IM |
1287 | .name = "switch", |
1288 | .cmp = switch_cmp, | |
daa1d7a5 FW |
1289 | }; |
1290 | ||
39aeb52f | 1291 | static int runtime_cmp(struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 FW |
1292 | { |
1293 | if (l->total_runtime < r->total_runtime) | |
1294 | return -1; | |
1295 | if (l->total_runtime > r->total_runtime) | |
1296 | return 1; | |
1297 | ||
1298 | return 0; | |
1299 | } | |
1300 | ||
1301 | static struct sort_dimension runtime_sort_dimension = { | |
b5fae128 IM |
1302 | .name = "runtime", |
1303 | .cmp = runtime_cmp, | |
daa1d7a5 FW |
1304 | }; |
1305 | ||
1306 | static struct sort_dimension *available_sorts[] = { | |
1307 | &pid_sort_dimension, | |
1308 | &avg_sort_dimension, | |
1309 | &max_sort_dimension, | |
1310 | &switch_sort_dimension, | |
1311 | &runtime_sort_dimension, | |
1312 | }; | |
1313 | ||
1314 | #define NB_AVAILABLE_SORTS (int)(sizeof(available_sorts) / sizeof(struct sort_dimension *)) | |
1315 | ||
1316 | static LIST_HEAD(sort_list); | |
1317 | ||
cbef79a8 | 1318 | static int sort_dimension__add(const char *tok, struct list_head *list) |
daa1d7a5 FW |
1319 | { |
1320 | int i; | |
1321 | ||
1322 | for (i = 0; i < NB_AVAILABLE_SORTS; i++) { | |
1323 | if (!strcmp(available_sorts[i]->name, tok)) { | |
1324 | list_add_tail(&available_sorts[i]->list, list); | |
1325 | ||
1326 | return 0; | |
1327 | } | |
1328 | } | |
1329 | ||
1330 | return -1; | |
1331 | } | |
1332 | ||
1333 | static void setup_sorting(void); | |
1334 | ||
1335 | static void sort_lat(void) | |
1336 | { | |
1337 | struct rb_node *node; | |
1338 | ||
1339 | for (;;) { | |
39aeb52f | 1340 | struct work_atoms *data; |
b1ffe8f3 | 1341 | node = rb_first(&atom_root); |
daa1d7a5 FW |
1342 | if (!node) |
1343 | break; | |
1344 | ||
b1ffe8f3 | 1345 | rb_erase(node, &atom_root); |
39aeb52f | 1346 | data = rb_entry(node, struct work_atoms, node); |
b1ffe8f3 | 1347 | __thread_latency_insert(&sorted_atom_root, data, &sort_list); |
daa1d7a5 FW |
1348 | } |
1349 | } | |
1350 | ||
419ab0d6 FW |
1351 | static struct trace_sched_handler *trace_handler; |
1352 | ||
fbf94829 | 1353 | static void |
f48f669d | 1354 | process_sched_wakeup_event(void *data, |
419ab0d6 FW |
1355 | struct event *event, |
1356 | int cpu __used, | |
1357 | u64 timestamp __used, | |
1358 | struct thread *thread __used) | |
1359 | { | |
1360 | struct trace_wakeup_event wakeup_event; | |
1361 | ||
f48f669d | 1362 | FILL_COMMON_FIELDS(wakeup_event, event, data); |
419ab0d6 | 1363 | |
f48f669d XG |
1364 | FILL_ARRAY(wakeup_event, comm, event, data); |
1365 | FILL_FIELD(wakeup_event, pid, event, data); | |
1366 | FILL_FIELD(wakeup_event, prio, event, data); | |
1367 | FILL_FIELD(wakeup_event, success, event, data); | |
1368 | FILL_FIELD(wakeup_event, cpu, event, data); | |
419ab0d6 | 1369 | |
0ec04e16 IM |
1370 | if (trace_handler->wakeup_event) |
1371 | trace_handler->wakeup_event(&wakeup_event, event, cpu, timestamp, thread); | |
419ab0d6 FW |
1372 | } |
1373 | ||
c8a37751 IM |
1374 | /* |
1375 | * Track the current task - that way we can know whether there's any | |
1376 | * weird events, such as a task being switched away that is not current. | |
1377 | */ | |
40749d0f | 1378 | static int max_cpu; |
0ec04e16 | 1379 | |
c8a37751 IM |
1380 | static u32 curr_pid[MAX_CPUS] = { [0 ... MAX_CPUS-1] = -1 }; |
1381 | ||
0ec04e16 IM |
1382 | static struct thread *curr_thread[MAX_CPUS]; |
1383 | ||
1384 | static char next_shortname1 = 'A'; | |
1385 | static char next_shortname2 = '0'; | |
1386 | ||
1387 | static void | |
1388 | map_switch_event(struct trace_switch_event *switch_event, | |
1389 | struct event *event __used, | |
1390 | int this_cpu, | |
1391 | u64 timestamp, | |
1392 | struct thread *thread __used) | |
1393 | { | |
1394 | struct thread *sched_out, *sched_in; | |
1395 | int new_shortname; | |
1396 | u64 timestamp0; | |
1397 | s64 delta; | |
1398 | int cpu; | |
1399 | ||
1400 | BUG_ON(this_cpu >= MAX_CPUS || this_cpu < 0); | |
1401 | ||
1402 | if (this_cpu > max_cpu) | |
1403 | max_cpu = this_cpu; | |
1404 | ||
1405 | timestamp0 = cpu_last_switched[this_cpu]; | |
1406 | cpu_last_switched[this_cpu] = timestamp; | |
1407 | if (timestamp0) | |
1408 | delta = timestamp - timestamp0; | |
1409 | else | |
1410 | delta = 0; | |
1411 | ||
1412 | if (delta < 0) | |
1413 | die("hm, delta: %Ld < 0 ?\n", delta); | |
1414 | ||
1415 | ||
d5b889f2 ACM |
1416 | sched_out = threads__findnew(switch_event->prev_pid); |
1417 | sched_in = threads__findnew(switch_event->next_pid); | |
0ec04e16 IM |
1418 | |
1419 | curr_thread[this_cpu] = sched_in; | |
1420 | ||
1421 | printf(" "); | |
1422 | ||
1423 | new_shortname = 0; | |
1424 | if (!sched_in->shortname[0]) { | |
1425 | sched_in->shortname[0] = next_shortname1; | |
1426 | sched_in->shortname[1] = next_shortname2; | |
1427 | ||
1428 | if (next_shortname1 < 'Z') { | |
1429 | next_shortname1++; | |
1430 | } else { | |
1431 | next_shortname1='A'; | |
1432 | if (next_shortname2 < '9') { | |
1433 | next_shortname2++; | |
1434 | } else { | |
1435 | next_shortname2='0'; | |
1436 | } | |
1437 | } | |
1438 | new_shortname = 1; | |
1439 | } | |
1440 | ||
1441 | for (cpu = 0; cpu <= max_cpu; cpu++) { | |
1442 | if (cpu != this_cpu) | |
1443 | printf(" "); | |
1444 | else | |
1445 | printf("*"); | |
1446 | ||
1447 | if (curr_thread[cpu]) { | |
1448 | if (curr_thread[cpu]->pid) | |
1449 | printf("%2s ", curr_thread[cpu]->shortname); | |
1450 | else | |
1451 | printf(". "); | |
1452 | } else | |
1453 | printf(" "); | |
1454 | } | |
1455 | ||
1456 | printf(" %12.6f secs ", (double)timestamp/1e9); | |
1457 | if (new_shortname) { | |
1458 | printf("%s => %s:%d\n", | |
1459 | sched_in->shortname, sched_in->comm, sched_in->pid); | |
1460 | } else { | |
1461 | printf("\n"); | |
1462 | } | |
1463 | } | |
1464 | ||
1465 | ||
419ab0d6 | 1466 | static void |
f48f669d | 1467 | process_sched_switch_event(void *data, |
419ab0d6 | 1468 | struct event *event, |
0ec04e16 | 1469 | int this_cpu, |
419ab0d6 FW |
1470 | u64 timestamp __used, |
1471 | struct thread *thread __used) | |
1472 | { | |
1473 | struct trace_switch_event switch_event; | |
1474 | ||
f48f669d | 1475 | FILL_COMMON_FIELDS(switch_event, event, data); |
419ab0d6 | 1476 | |
f48f669d XG |
1477 | FILL_ARRAY(switch_event, prev_comm, event, data); |
1478 | FILL_FIELD(switch_event, prev_pid, event, data); | |
1479 | FILL_FIELD(switch_event, prev_prio, event, data); | |
1480 | FILL_FIELD(switch_event, prev_state, event, data); | |
1481 | FILL_ARRAY(switch_event, next_comm, event, data); | |
1482 | FILL_FIELD(switch_event, next_pid, event, data); | |
1483 | FILL_FIELD(switch_event, next_prio, event, data); | |
419ab0d6 | 1484 | |
0ec04e16 | 1485 | if (curr_pid[this_cpu] != (u32)-1) { |
c8a37751 IM |
1486 | /* |
1487 | * Are we trying to switch away a PID that is | |
1488 | * not current? | |
1489 | */ | |
0ec04e16 | 1490 | if (curr_pid[this_cpu] != switch_event.prev_pid) |
c8a37751 IM |
1491 | nr_context_switch_bugs++; |
1492 | } | |
0ec04e16 IM |
1493 | if (trace_handler->switch_event) |
1494 | trace_handler->switch_event(&switch_event, event, this_cpu, timestamp, thread); | |
c8a37751 | 1495 | |
0ec04e16 | 1496 | curr_pid[this_cpu] = switch_event.next_pid; |
419ab0d6 FW |
1497 | } |
1498 | ||
39aeb52f | 1499 | static void |
f48f669d | 1500 | process_sched_runtime_event(void *data, |
39aeb52f | 1501 | struct event *event, |
1502 | int cpu __used, | |
1503 | u64 timestamp __used, | |
1504 | struct thread *thread __used) | |
1505 | { | |
1506 | struct trace_runtime_event runtime_event; | |
1507 | ||
f48f669d XG |
1508 | FILL_ARRAY(runtime_event, comm, event, data); |
1509 | FILL_FIELD(runtime_event, pid, event, data); | |
1510 | FILL_FIELD(runtime_event, runtime, event, data); | |
1511 | FILL_FIELD(runtime_event, vruntime, event, data); | |
39aeb52f | 1512 | |
0ec04e16 IM |
1513 | if (trace_handler->runtime_event) |
1514 | trace_handler->runtime_event(&runtime_event, event, cpu, timestamp, thread); | |
39aeb52f | 1515 | } |
1516 | ||
419ab0d6 | 1517 | static void |
f48f669d | 1518 | process_sched_fork_event(void *data, |
419ab0d6 FW |
1519 | struct event *event, |
1520 | int cpu __used, | |
1521 | u64 timestamp __used, | |
1522 | struct thread *thread __used) | |
fbf94829 | 1523 | { |
46538818 FW |
1524 | struct trace_fork_event fork_event; |
1525 | ||
f48f669d | 1526 | FILL_COMMON_FIELDS(fork_event, event, data); |
46538818 | 1527 | |
f48f669d XG |
1528 | FILL_ARRAY(fork_event, parent_comm, event, data); |
1529 | FILL_FIELD(fork_event, parent_pid, event, data); | |
1530 | FILL_ARRAY(fork_event, child_comm, event, data); | |
1531 | FILL_FIELD(fork_event, child_pid, event, data); | |
46538818 | 1532 | |
0ec04e16 IM |
1533 | if (trace_handler->fork_event) |
1534 | trace_handler->fork_event(&fork_event, event, cpu, timestamp, thread); | |
fbf94829 IM |
1535 | } |
1536 | ||
419ab0d6 FW |
1537 | static void |
1538 | process_sched_exit_event(struct event *event, | |
1539 | int cpu __used, | |
1540 | u64 timestamp __used, | |
1541 | struct thread *thread __used) | |
fbf94829 | 1542 | { |
ad236fd2 IM |
1543 | if (verbose) |
1544 | printf("sched_exit event %p\n", event); | |
ec156764 IM |
1545 | } |
1546 | ||
55ffb7a6 | 1547 | static void |
f48f669d | 1548 | process_sched_migrate_task_event(void *data, |
55ffb7a6 MG |
1549 | struct event *event, |
1550 | int cpu __used, | |
1551 | u64 timestamp __used, | |
1552 | struct thread *thread __used) | |
1553 | { | |
1554 | struct trace_migrate_task_event migrate_task_event; | |
1555 | ||
f48f669d | 1556 | FILL_COMMON_FIELDS(migrate_task_event, event, data); |
55ffb7a6 | 1557 | |
f48f669d XG |
1558 | FILL_ARRAY(migrate_task_event, comm, event, data); |
1559 | FILL_FIELD(migrate_task_event, pid, event, data); | |
1560 | FILL_FIELD(migrate_task_event, prio, event, data); | |
1561 | FILL_FIELD(migrate_task_event, cpu, event, data); | |
55ffb7a6 MG |
1562 | |
1563 | if (trace_handler->migrate_task_event) | |
1564 | trace_handler->migrate_task_event(&migrate_task_event, event, cpu, timestamp, thread); | |
1565 | } | |
1566 | ||
ec156764 | 1567 | static void |
f48f669d | 1568 | process_raw_event(event_t *raw_event __used, void *data, |
ec156764 IM |
1569 | int cpu, u64 timestamp, struct thread *thread) |
1570 | { | |
ec156764 IM |
1571 | struct event *event; |
1572 | int type; | |
1573 | ||
d8bd9e0a | 1574 | |
f48f669d | 1575 | type = trace_parse_common_type(data); |
ec156764 IM |
1576 | event = trace_find_event(type); |
1577 | ||
ec156764 | 1578 | if (!strcmp(event->name, "sched_switch")) |
f48f669d | 1579 | process_sched_switch_event(data, event, cpu, timestamp, thread); |
39aeb52f | 1580 | if (!strcmp(event->name, "sched_stat_runtime")) |
f48f669d | 1581 | process_sched_runtime_event(data, event, cpu, timestamp, thread); |
ec156764 | 1582 | if (!strcmp(event->name, "sched_wakeup")) |
f48f669d | 1583 | process_sched_wakeup_event(data, event, cpu, timestamp, thread); |
fbf94829 | 1584 | if (!strcmp(event->name, "sched_wakeup_new")) |
f48f669d | 1585 | process_sched_wakeup_event(data, event, cpu, timestamp, thread); |
fbf94829 | 1586 | if (!strcmp(event->name, "sched_process_fork")) |
f48f669d | 1587 | process_sched_fork_event(data, event, cpu, timestamp, thread); |
fbf94829 IM |
1588 | if (!strcmp(event->name, "sched_process_exit")) |
1589 | process_sched_exit_event(event, cpu, timestamp, thread); | |
55ffb7a6 | 1590 | if (!strcmp(event->name, "sched_migrate_task")) |
f48f669d | 1591 | process_sched_migrate_task_event(data, event, cpu, timestamp, thread); |
ec156764 IM |
1592 | } |
1593 | ||
62daacb5 | 1594 | static int process_sample_event(event_t *event) |
0a02ad93 | 1595 | { |
180f95e2 | 1596 | struct sample_data data; |
0a02ad93 | 1597 | struct thread *thread; |
a80deb62 ACM |
1598 | |
1599 | if (!(sample_type & PERF_SAMPLE_RAW)) | |
1600 | return 0; | |
0a02ad93 | 1601 | |
180f95e2 OH |
1602 | memset(&data, 0, sizeof(data)); |
1603 | data.time = -1; | |
1604 | data.cpu = -1; | |
1605 | data.period = -1; | |
0a02ad93 | 1606 | |
180f95e2 | 1607 | event__parse_sample(event, sample_type, &data); |
0a02ad93 | 1608 | |
62daacb5 | 1609 | dump_printf("(IP, %d): %d/%d: %p period: %Ld\n", |
0a02ad93 | 1610 | event->header.misc, |
180f95e2 OH |
1611 | data.pid, data.tid, |
1612 | (void *)(long)data.ip, | |
1613 | (long long)data.period); | |
0a02ad93 | 1614 | |
180f95e2 | 1615 | thread = threads__findnew(data.pid); |
0a02ad93 | 1616 | if (thread == NULL) { |
6beba7ad ACM |
1617 | pr_debug("problem processing %d event, skipping it.\n", |
1618 | event->header.type); | |
0a02ad93 IM |
1619 | return -1; |
1620 | } | |
1621 | ||
f39cdf25 JL |
1622 | dump_printf(" ... thread: %s:%d\n", thread->comm, thread->pid); |
1623 | ||
180f95e2 | 1624 | if (profile_cpu != -1 && profile_cpu != (int)data.cpu) |
55ffb7a6 MG |
1625 | return 0; |
1626 | ||
f48f669d | 1627 | process_raw_event(event, data.raw_data, data.cpu, data.time, thread); |
0a02ad93 IM |
1628 | |
1629 | return 0; | |
1630 | } | |
1631 | ||
62daacb5 | 1632 | static int process_lost_event(event_t *event __used) |
0a02ad93 | 1633 | { |
016e92fb FW |
1634 | nr_lost_chunks++; |
1635 | nr_lost_events += event->lost.lost; | |
0a02ad93 | 1636 | |
016e92fb FW |
1637 | return 0; |
1638 | } | |
0a02ad93 | 1639 | |
016e92fb FW |
1640 | static int sample_type_check(u64 type) |
1641 | { | |
1642 | sample_type = type; | |
0a02ad93 | 1643 | |
016e92fb FW |
1644 | if (!(sample_type & PERF_SAMPLE_RAW)) { |
1645 | fprintf(stderr, | |
1646 | "No trace sample to read. Did you call perf record " | |
1647 | "without -R?"); | |
0a02ad93 IM |
1648 | return -1; |
1649 | } | |
1650 | ||
1651 | return 0; | |
1652 | } | |
1653 | ||
016e92fb FW |
1654 | static struct perf_file_handler file_handler = { |
1655 | .process_sample_event = process_sample_event, | |
62daacb5 | 1656 | .process_comm_event = event__process_comm, |
016e92fb FW |
1657 | .process_lost_event = process_lost_event, |
1658 | .sample_type_check = sample_type_check, | |
1659 | }; | |
1660 | ||
46f392c9 | 1661 | static int read_events(void) |
0a02ad93 | 1662 | { |
d5b889f2 | 1663 | register_idle_thread(); |
016e92fb | 1664 | register_perf_file_handler(&file_handler); |
0a02ad93 | 1665 | |
b32d133a | 1666 | return mmap_dispatch_perf_file(&header, input_name, 0, 0, |
62daacb5 | 1667 | &event__cwdlen, &event__cwd); |
0a02ad93 IM |
1668 | } |
1669 | ||
0ec04e16 IM |
1670 | static void print_bad_events(void) |
1671 | { | |
1672 | if (nr_unordered_timestamps && nr_timestamps) { | |
1673 | printf(" INFO: %.3f%% unordered timestamps (%ld out of %ld)\n", | |
1674 | (double)nr_unordered_timestamps/(double)nr_timestamps*100.0, | |
1675 | nr_unordered_timestamps, nr_timestamps); | |
1676 | } | |
1677 | if (nr_lost_events && nr_events) { | |
1678 | printf(" INFO: %.3f%% lost events (%ld out of %ld, in %ld chunks)\n", | |
1679 | (double)nr_lost_events/(double)nr_events*100.0, | |
1680 | nr_lost_events, nr_events, nr_lost_chunks); | |
1681 | } | |
1682 | if (nr_state_machine_bugs && nr_timestamps) { | |
1683 | printf(" INFO: %.3f%% state machine bugs (%ld out of %ld)", | |
1684 | (double)nr_state_machine_bugs/(double)nr_timestamps*100.0, | |
1685 | nr_state_machine_bugs, nr_timestamps); | |
1686 | if (nr_lost_events) | |
1687 | printf(" (due to lost events?)"); | |
1688 | printf("\n"); | |
1689 | } | |
1690 | if (nr_context_switch_bugs && nr_timestamps) { | |
1691 | printf(" INFO: %.3f%% context switch bugs (%ld out of %ld)", | |
1692 | (double)nr_context_switch_bugs/(double)nr_timestamps*100.0, | |
1693 | nr_context_switch_bugs, nr_timestamps); | |
1694 | if (nr_lost_events) | |
1695 | printf(" (due to lost events?)"); | |
1696 | printf("\n"); | |
1697 | } | |
1698 | } | |
1699 | ||
1700 | static void __cmd_lat(void) | |
1701 | { | |
1702 | struct rb_node *next; | |
1703 | ||
1704 | setup_pager(); | |
1705 | read_events(); | |
1706 | sort_lat(); | |
1707 | ||
1708 | printf("\n -----------------------------------------------------------------------------------------\n"); | |
1709 | printf(" Task | Runtime ms | Switches | Average delay ms | Maximum delay ms |\n"); | |
1710 | printf(" -----------------------------------------------------------------------------------------\n"); | |
1711 | ||
1712 | next = rb_first(&sorted_atom_root); | |
1713 | ||
1714 | while (next) { | |
1715 | struct work_atoms *work_list; | |
1716 | ||
1717 | work_list = rb_entry(next, struct work_atoms, node); | |
1718 | output_lat_thread(work_list); | |
1719 | next = rb_next(next); | |
1720 | } | |
1721 | ||
1722 | printf(" -----------------------------------------------------------------------------------------\n"); | |
1723 | printf(" TOTAL: |%11.3f ms |%9Ld |\n", | |
1724 | (double)all_runtime/1e6, all_count); | |
1725 | ||
1726 | printf(" ---------------------------------------------------\n"); | |
1727 | ||
1728 | print_bad_events(); | |
1729 | printf("\n"); | |
1730 | ||
1731 | } | |
1732 | ||
1733 | static struct trace_sched_handler map_ops = { | |
1734 | .wakeup_event = NULL, | |
1735 | .switch_event = map_switch_event, | |
1736 | .runtime_event = NULL, | |
1737 | .fork_event = NULL, | |
1738 | }; | |
1739 | ||
1740 | static void __cmd_map(void) | |
1741 | { | |
40749d0f IM |
1742 | max_cpu = sysconf(_SC_NPROCESSORS_CONF); |
1743 | ||
0ec04e16 IM |
1744 | setup_pager(); |
1745 | read_events(); | |
1746 | print_bad_events(); | |
1747 | } | |
1748 | ||
1749 | static void __cmd_replay(void) | |
1750 | { | |
1751 | unsigned long i; | |
1752 | ||
1753 | calibrate_run_measurement_overhead(); | |
1754 | calibrate_sleep_measurement_overhead(); | |
1755 | ||
1756 | test_calibrations(); | |
1757 | ||
1758 | read_events(); | |
1759 | ||
1760 | printf("nr_run_events: %ld\n", nr_run_events); | |
1761 | printf("nr_sleep_events: %ld\n", nr_sleep_events); | |
1762 | printf("nr_wakeup_events: %ld\n", nr_wakeup_events); | |
1763 | ||
1764 | if (targetless_wakeups) | |
1765 | printf("target-less wakeups: %ld\n", targetless_wakeups); | |
1766 | if (multitarget_wakeups) | |
1767 | printf("multi-target wakeups: %ld\n", multitarget_wakeups); | |
1768 | if (nr_run_events_optimized) | |
1769 | printf("run atoms optimized: %ld\n", | |
1770 | nr_run_events_optimized); | |
1771 | ||
1772 | print_task_traces(); | |
1773 | add_cross_task_wakeups(); | |
1774 | ||
1775 | create_tasks(); | |
1776 | printf("------------------------------------------------------------\n"); | |
1777 | for (i = 0; i < replay_repeat; i++) | |
1778 | run_one_test(); | |
1779 | } | |
1780 | ||
1781 | ||
46f392c9 | 1782 | static const char * const sched_usage[] = { |
4b77a729 | 1783 | "perf sched [<options>] {record|latency|map|replay|trace}", |
0a02ad93 IM |
1784 | NULL |
1785 | }; | |
1786 | ||
f2858d8a | 1787 | static const struct option sched_options[] = { |
4b77a729 MG |
1788 | OPT_STRING('i', "input", &input_name, "file", |
1789 | "input file name"), | |
f2858d8a IM |
1790 | OPT_BOOLEAN('v', "verbose", &verbose, |
1791 | "be more verbose (show symbol address, etc)"), | |
0a02ad93 IM |
1792 | OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, |
1793 | "dump raw trace in ASCII"), | |
f2858d8a IM |
1794 | OPT_END() |
1795 | }; | |
1796 | ||
1797 | static const char * const latency_usage[] = { | |
1798 | "perf sched latency [<options>]", | |
1799 | NULL | |
1800 | }; | |
1801 | ||
1802 | static const struct option latency_options[] = { | |
daa1d7a5 FW |
1803 | OPT_STRING('s', "sort", &sort_order, "key[,key2...]", |
1804 | "sort by key(s): runtime, switch, avg, max"), | |
0a02ad93 IM |
1805 | OPT_BOOLEAN('v', "verbose", &verbose, |
1806 | "be more verbose (show symbol address, etc)"), | |
55ffb7a6 MG |
1807 | OPT_INTEGER('C', "CPU", &profile_cpu, |
1808 | "CPU to profile on"), | |
f2858d8a IM |
1809 | OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, |
1810 | "dump raw trace in ASCII"), | |
1811 | OPT_END() | |
1812 | }; | |
1813 | ||
1814 | static const char * const replay_usage[] = { | |
1815 | "perf sched replay [<options>]", | |
1816 | NULL | |
1817 | }; | |
1818 | ||
1819 | static const struct option replay_options[] = { | |
1820 | OPT_INTEGER('r', "repeat", &replay_repeat, | |
1821 | "repeat the workload replay N times (-1: infinite)"), | |
1822 | OPT_BOOLEAN('v', "verbose", &verbose, | |
1823 | "be more verbose (show symbol address, etc)"), | |
1824 | OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, | |
1825 | "dump raw trace in ASCII"), | |
0a02ad93 IM |
1826 | OPT_END() |
1827 | }; | |
1828 | ||
daa1d7a5 FW |
1829 | static void setup_sorting(void) |
1830 | { | |
1831 | char *tmp, *tok, *str = strdup(sort_order); | |
1832 | ||
1833 | for (tok = strtok_r(str, ", ", &tmp); | |
1834 | tok; tok = strtok_r(NULL, ", ", &tmp)) { | |
1835 | if (sort_dimension__add(tok, &sort_list) < 0) { | |
1836 | error("Unknown --sort key: `%s'", tok); | |
f2858d8a | 1837 | usage_with_options(latency_usage, latency_options); |
daa1d7a5 FW |
1838 | } |
1839 | } | |
1840 | ||
1841 | free(str); | |
1842 | ||
cbef79a8 | 1843 | sort_dimension__add("pid", &cmp_pid); |
daa1d7a5 FW |
1844 | } |
1845 | ||
1fc35b29 IM |
1846 | static const char *record_args[] = { |
1847 | "record", | |
1848 | "-a", | |
1849 | "-R", | |
d1302522 | 1850 | "-M", |
ea57c4f5 | 1851 | "-f", |
dc02bf71 | 1852 | "-m", "1024", |
1fc35b29 IM |
1853 | "-c", "1", |
1854 | "-e", "sched:sched_switch:r", | |
1855 | "-e", "sched:sched_stat_wait:r", | |
1856 | "-e", "sched:sched_stat_sleep:r", | |
1857 | "-e", "sched:sched_stat_iowait:r", | |
ea57c4f5 | 1858 | "-e", "sched:sched_stat_runtime:r", |
1fc35b29 IM |
1859 | "-e", "sched:sched_process_exit:r", |
1860 | "-e", "sched:sched_process_fork:r", | |
1861 | "-e", "sched:sched_wakeup:r", | |
1862 | "-e", "sched:sched_migrate_task:r", | |
1863 | }; | |
1864 | ||
1865 | static int __cmd_record(int argc, const char **argv) | |
1866 | { | |
1867 | unsigned int rec_argc, i, j; | |
1868 | const char **rec_argv; | |
1869 | ||
1870 | rec_argc = ARRAY_SIZE(record_args) + argc - 1; | |
1871 | rec_argv = calloc(rec_argc + 1, sizeof(char *)); | |
1872 | ||
1873 | for (i = 0; i < ARRAY_SIZE(record_args); i++) | |
1874 | rec_argv[i] = strdup(record_args[i]); | |
1875 | ||
1876 | for (j = 1; j < (unsigned int)argc; j++, i++) | |
1877 | rec_argv[i] = argv[j]; | |
1878 | ||
1879 | BUG_ON(i != rec_argc); | |
1880 | ||
1881 | return cmd_record(i, rec_argv, NULL); | |
1882 | } | |
1883 | ||
0a02ad93 IM |
1884 | int cmd_sched(int argc, const char **argv, const char *prefix __used) |
1885 | { | |
f2858d8a IM |
1886 | argc = parse_options(argc, argv, sched_options, sched_usage, |
1887 | PARSE_OPT_STOP_AT_NON_OPTION); | |
1888 | if (!argc) | |
1889 | usage_with_options(sched_usage, sched_options); | |
0a02ad93 | 1890 | |
c0777c5a XG |
1891 | /* |
1892 | * Aliased to 'perf trace' for now: | |
1893 | */ | |
1894 | if (!strcmp(argv[0], "trace")) | |
1895 | return cmd_trace(argc, argv, prefix); | |
1896 | ||
1897 | symbol__init(0); | |
1fc35b29 IM |
1898 | if (!strncmp(argv[0], "rec", 3)) { |
1899 | return __cmd_record(argc, argv); | |
1900 | } else if (!strncmp(argv[0], "lat", 3)) { | |
cdce9d73 | 1901 | trace_handler = &lat_ops; |
f2858d8a IM |
1902 | if (argc > 1) { |
1903 | argc = parse_options(argc, argv, latency_options, latency_usage, 0); | |
1904 | if (argc) | |
1905 | usage_with_options(latency_usage, latency_options); | |
f2858d8a | 1906 | } |
b5fae128 | 1907 | setup_sorting(); |
46f392c9 | 1908 | __cmd_lat(); |
0ec04e16 IM |
1909 | } else if (!strcmp(argv[0], "map")) { |
1910 | trace_handler = &map_ops; | |
1911 | setup_sorting(); | |
1912 | __cmd_map(); | |
f2858d8a IM |
1913 | } else if (!strncmp(argv[0], "rep", 3)) { |
1914 | trace_handler = &replay_ops; | |
1915 | if (argc) { | |
1916 | argc = parse_options(argc, argv, replay_options, replay_usage, 0); | |
1917 | if (argc) | |
1918 | usage_with_options(replay_usage, replay_options); | |
1919 | } | |
1920 | __cmd_replay(); | |
1921 | } else { | |
1922 | usage_with_options(sched_usage, sched_options); | |
1923 | } | |
1924 | ||
ec156764 | 1925 | return 0; |
0a02ad93 | 1926 | } |