]>
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 IM |
13 | #include "util/debug.h" |
14 | ||
ec156764 | 15 | #include <sys/types.h> |
b1ffe8f3 | 16 | #include <sys/prctl.h> |
0a02ad93 | 17 | |
b1ffe8f3 IM |
18 | #include <semaphore.h> |
19 | #include <pthread.h> | |
20 | #include <math.h> | |
419ab0d6 | 21 | |
ec156764 IM |
22 | static char const *input_name = "perf.data"; |
23 | static int input; | |
24 | static unsigned long page_size; | |
25 | static unsigned long mmap_window = 32; | |
0a02ad93 | 26 | |
ec156764 | 27 | static unsigned long total_comm = 0; |
0a02ad93 | 28 | |
ec156764 IM |
29 | static struct rb_root threads; |
30 | static struct thread *last_match; | |
0a02ad93 | 31 | |
ec156764 IM |
32 | static struct perf_header *header; |
33 | static u64 sample_type; | |
0a02ad93 | 34 | |
daa1d7a5 FW |
35 | static char default_sort_order[] = "avg, max, switch, runtime"; |
36 | static char *sort_order = default_sort_order; | |
37 | ||
b1ffe8f3 IM |
38 | #define PR_SET_NAME 15 /* Set process name */ |
39 | #define MAX_CPUS 4096 | |
0a02ad93 | 40 | |
b1ffe8f3 | 41 | #define BUG_ON(x) assert(!(x)) |
ec156764 | 42 | |
b1ffe8f3 IM |
43 | static u64 run_measurement_overhead; |
44 | static u64 sleep_measurement_overhead; | |
ec156764 | 45 | |
b1ffe8f3 IM |
46 | #define COMM_LEN 20 |
47 | #define SYM_LEN 129 | |
ec156764 | 48 | |
b1ffe8f3 | 49 | #define MAX_PID 65536 |
ec156764 | 50 | |
b1ffe8f3 | 51 | static unsigned long nr_tasks; |
ec156764 | 52 | |
39aeb52f | 53 | struct sched_atom; |
ec156764 | 54 | |
b1ffe8f3 IM |
55 | struct task_desc { |
56 | unsigned long nr; | |
57 | unsigned long pid; | |
58 | char comm[COMM_LEN]; | |
ec156764 | 59 | |
b1ffe8f3 IM |
60 | unsigned long nr_events; |
61 | unsigned long curr_event; | |
39aeb52f | 62 | struct sched_atom **atoms; |
b1ffe8f3 IM |
63 | |
64 | pthread_t thread; | |
65 | sem_t sleep_sem; | |
ec156764 | 66 | |
b1ffe8f3 IM |
67 | sem_t ready_for_work; |
68 | sem_t work_done_sem; | |
69 | ||
70 | u64 cpu_usage; | |
71 | }; | |
72 | ||
73 | enum sched_event_type { | |
74 | SCHED_EVENT_RUN, | |
75 | SCHED_EVENT_SLEEP, | |
76 | SCHED_EVENT_WAKEUP, | |
77 | }; | |
78 | ||
39aeb52f | 79 | struct sched_atom { |
b1ffe8f3 IM |
80 | enum sched_event_type type; |
81 | u64 timestamp; | |
82 | u64 duration; | |
83 | unsigned long nr; | |
84 | int specific_wait; | |
85 | sem_t *wait_sem; | |
86 | struct task_desc *wakee; | |
87 | }; | |
88 | ||
89 | static struct task_desc *pid_to_task[MAX_PID]; | |
90 | ||
91 | static struct task_desc **tasks; | |
92 | ||
93 | static pthread_mutex_t start_work_mutex = PTHREAD_MUTEX_INITIALIZER; | |
94 | static u64 start_time; | |
95 | ||
96 | static pthread_mutex_t work_done_wait_mutex = PTHREAD_MUTEX_INITIALIZER; | |
ec156764 | 97 | |
b1ffe8f3 IM |
98 | static unsigned long nr_run_events; |
99 | static unsigned long nr_sleep_events; | |
100 | static unsigned long nr_wakeup_events; | |
101 | ||
102 | static unsigned long nr_sleep_corrections; | |
103 | static unsigned long nr_run_events_optimized; | |
104 | ||
105 | static unsigned long targetless_wakeups; | |
106 | static unsigned long multitarget_wakeups; | |
107 | ||
108 | static u64 cpu_usage; | |
109 | static u64 runavg_cpu_usage; | |
110 | static u64 parent_cpu_usage; | |
111 | static u64 runavg_parent_cpu_usage; | |
112 | ||
113 | static unsigned long nr_runs; | |
114 | static u64 sum_runtime; | |
115 | static u64 sum_fluct; | |
116 | static u64 run_avg; | |
117 | ||
118 | static unsigned long replay_repeat = 10; | |
ea57c4f5 | 119 | static unsigned long nr_timestamps; |
dc02bf71 IM |
120 | static unsigned long nr_unordered_timestamps; |
121 | static unsigned long nr_state_machine_bugs; | |
c8a37751 | 122 | static unsigned long nr_context_switch_bugs; |
dc02bf71 IM |
123 | static unsigned long nr_events; |
124 | static unsigned long nr_lost_chunks; | |
125 | static unsigned long nr_lost_events; | |
b1ffe8f3 IM |
126 | |
127 | #define TASK_STATE_TO_CHAR_STR "RSDTtZX" | |
128 | ||
129 | enum thread_state { | |
130 | THREAD_SLEEPING = 0, | |
131 | THREAD_WAIT_CPU, | |
132 | THREAD_SCHED_IN, | |
133 | THREAD_IGNORE | |
134 | }; | |
135 | ||
136 | struct work_atom { | |
137 | struct list_head list; | |
138 | enum thread_state state; | |
aa1ab9d2 | 139 | u64 sched_out_time; |
b1ffe8f3 IM |
140 | u64 wake_up_time; |
141 | u64 sched_in_time; | |
142 | u64 runtime; | |
143 | }; | |
144 | ||
39aeb52f | 145 | struct work_atoms { |
146 | struct list_head work_list; | |
b1ffe8f3 IM |
147 | struct thread *thread; |
148 | struct rb_node node; | |
149 | u64 max_lat; | |
150 | u64 total_lat; | |
151 | u64 nb_atoms; | |
152 | u64 total_runtime; | |
153 | }; | |
154 | ||
39aeb52f | 155 | typedef int (*sort_fn_t)(struct work_atoms *, struct work_atoms *); |
b1ffe8f3 IM |
156 | |
157 | static struct rb_root atom_root, sorted_atom_root; | |
158 | ||
159 | static u64 all_runtime; | |
160 | static u64 all_count; | |
161 | ||
b1ffe8f3 IM |
162 | |
163 | static u64 get_nsecs(void) | |
ec156764 IM |
164 | { |
165 | struct timespec ts; | |
166 | ||
167 | clock_gettime(CLOCK_MONOTONIC, &ts); | |
168 | ||
169 | return ts.tv_sec * 1000000000ULL + ts.tv_nsec; | |
170 | } | |
171 | ||
b1ffe8f3 | 172 | static void burn_nsecs(u64 nsecs) |
ec156764 | 173 | { |
b1ffe8f3 | 174 | u64 T0 = get_nsecs(), T1; |
ec156764 IM |
175 | |
176 | do { | |
177 | T1 = get_nsecs(); | |
178 | } while (T1 + run_measurement_overhead < T0 + nsecs); | |
179 | } | |
180 | ||
b1ffe8f3 | 181 | static void sleep_nsecs(u64 nsecs) |
ec156764 IM |
182 | { |
183 | struct timespec ts; | |
184 | ||
185 | ts.tv_nsec = nsecs % 999999999; | |
186 | ts.tv_sec = nsecs / 999999999; | |
187 | ||
188 | nanosleep(&ts, NULL); | |
189 | } | |
190 | ||
191 | static void calibrate_run_measurement_overhead(void) | |
192 | { | |
b1ffe8f3 | 193 | u64 T0, T1, delta, min_delta = 1000000000ULL; |
ec156764 IM |
194 | int i; |
195 | ||
196 | for (i = 0; i < 10; i++) { | |
197 | T0 = get_nsecs(); | |
198 | burn_nsecs(0); | |
199 | T1 = get_nsecs(); | |
200 | delta = T1-T0; | |
201 | min_delta = min(min_delta, delta); | |
202 | } | |
203 | run_measurement_overhead = min_delta; | |
204 | ||
ad236fd2 | 205 | printf("run measurement overhead: %Ld nsecs\n", min_delta); |
ec156764 IM |
206 | } |
207 | ||
208 | static void calibrate_sleep_measurement_overhead(void) | |
209 | { | |
b1ffe8f3 | 210 | u64 T0, T1, delta, min_delta = 1000000000ULL; |
ec156764 IM |
211 | int i; |
212 | ||
213 | for (i = 0; i < 10; i++) { | |
214 | T0 = get_nsecs(); | |
215 | sleep_nsecs(10000); | |
216 | T1 = get_nsecs(); | |
217 | delta = T1-T0; | |
218 | min_delta = min(min_delta, delta); | |
219 | } | |
220 | min_delta -= 10000; | |
221 | sleep_measurement_overhead = min_delta; | |
222 | ||
ad236fd2 | 223 | printf("sleep measurement overhead: %Ld nsecs\n", min_delta); |
ec156764 IM |
224 | } |
225 | ||
39aeb52f | 226 | static struct sched_atom * |
b1ffe8f3 | 227 | get_new_event(struct task_desc *task, u64 timestamp) |
ec156764 | 228 | { |
39aeb52f | 229 | struct sched_atom *event = calloc(1, sizeof(*event)); |
ec156764 IM |
230 | unsigned long idx = task->nr_events; |
231 | size_t size; | |
232 | ||
233 | event->timestamp = timestamp; | |
234 | event->nr = idx; | |
235 | ||
236 | task->nr_events++; | |
39aeb52f | 237 | size = sizeof(struct sched_atom *) * task->nr_events; |
238 | task->atoms = realloc(task->atoms, size); | |
239 | BUG_ON(!task->atoms); | |
ec156764 | 240 | |
39aeb52f | 241 | task->atoms[idx] = event; |
ec156764 IM |
242 | |
243 | return event; | |
244 | } | |
245 | ||
39aeb52f | 246 | static struct sched_atom *last_event(struct task_desc *task) |
ec156764 IM |
247 | { |
248 | if (!task->nr_events) | |
249 | return NULL; | |
250 | ||
39aeb52f | 251 | return task->atoms[task->nr_events - 1]; |
ec156764 IM |
252 | } |
253 | ||
254 | static void | |
b1ffe8f3 | 255 | add_sched_event_run(struct task_desc *task, u64 timestamp, u64 duration) |
ec156764 | 256 | { |
39aeb52f | 257 | struct sched_atom *event, *curr_event = last_event(task); |
ec156764 IM |
258 | |
259 | /* | |
fbf94829 IM |
260 | * optimize an existing RUN event by merging this one |
261 | * to it: | |
262 | */ | |
ec156764 IM |
263 | if (curr_event && curr_event->type == SCHED_EVENT_RUN) { |
264 | nr_run_events_optimized++; | |
265 | curr_event->duration += duration; | |
266 | return; | |
267 | } | |
268 | ||
269 | event = get_new_event(task, timestamp); | |
270 | ||
271 | event->type = SCHED_EVENT_RUN; | |
272 | event->duration = duration; | |
273 | ||
274 | nr_run_events++; | |
275 | } | |
276 | ||
ec156764 | 277 | static void |
b1ffe8f3 | 278 | add_sched_event_wakeup(struct task_desc *task, u64 timestamp, |
ec156764 IM |
279 | struct task_desc *wakee) |
280 | { | |
39aeb52f | 281 | struct sched_atom *event, *wakee_event; |
ec156764 IM |
282 | |
283 | event = get_new_event(task, timestamp); | |
284 | event->type = SCHED_EVENT_WAKEUP; | |
285 | event->wakee = wakee; | |
286 | ||
287 | wakee_event = last_event(wakee); | |
288 | if (!wakee_event || wakee_event->type != SCHED_EVENT_SLEEP) { | |
289 | targetless_wakeups++; | |
290 | return; | |
291 | } | |
292 | if (wakee_event->wait_sem) { | |
293 | multitarget_wakeups++; | |
294 | return; | |
295 | } | |
296 | ||
297 | wakee_event->wait_sem = calloc(1, sizeof(*wakee_event->wait_sem)); | |
298 | sem_init(wakee_event->wait_sem, 0, 0); | |
299 | wakee_event->specific_wait = 1; | |
300 | event->wait_sem = wakee_event->wait_sem; | |
301 | ||
302 | nr_wakeup_events++; | |
303 | } | |
304 | ||
305 | static void | |
b1ffe8f3 | 306 | add_sched_event_sleep(struct task_desc *task, u64 timestamp, |
ad236fd2 | 307 | u64 task_state __used) |
ec156764 | 308 | { |
39aeb52f | 309 | struct sched_atom *event = get_new_event(task, timestamp); |
ec156764 IM |
310 | |
311 | event->type = SCHED_EVENT_SLEEP; | |
312 | ||
313 | nr_sleep_events++; | |
314 | } | |
315 | ||
316 | static struct task_desc *register_pid(unsigned long pid, const char *comm) | |
317 | { | |
318 | struct task_desc *task; | |
319 | ||
320 | BUG_ON(pid >= MAX_PID); | |
321 | ||
322 | task = pid_to_task[pid]; | |
323 | ||
324 | if (task) | |
325 | return task; | |
326 | ||
327 | task = calloc(1, sizeof(*task)); | |
328 | task->pid = pid; | |
329 | task->nr = nr_tasks; | |
330 | strcpy(task->comm, comm); | |
331 | /* | |
332 | * every task starts in sleeping state - this gets ignored | |
333 | * if there's no wakeup pointing to this sleep state: | |
334 | */ | |
335 | add_sched_event_sleep(task, 0, 0); | |
336 | ||
337 | pid_to_task[pid] = task; | |
338 | nr_tasks++; | |
339 | tasks = realloc(tasks, nr_tasks*sizeof(struct task_task *)); | |
340 | BUG_ON(!tasks); | |
341 | tasks[task->nr] = task; | |
342 | ||
ad236fd2 IM |
343 | if (verbose) |
344 | printf("registered task #%ld, PID %ld (%s)\n", nr_tasks, pid, comm); | |
ec156764 IM |
345 | |
346 | return task; | |
347 | } | |
348 | ||
349 | ||
ec156764 IM |
350 | static void print_task_traces(void) |
351 | { | |
352 | struct task_desc *task; | |
353 | unsigned long i; | |
354 | ||
355 | for (i = 0; i < nr_tasks; i++) { | |
356 | task = tasks[i]; | |
ad236fd2 | 357 | printf("task %6ld (%20s:%10ld), nr_events: %ld\n", |
ec156764 IM |
358 | task->nr, task->comm, task->pid, task->nr_events); |
359 | } | |
360 | } | |
361 | ||
362 | static void add_cross_task_wakeups(void) | |
363 | { | |
364 | struct task_desc *task1, *task2; | |
365 | unsigned long i, j; | |
366 | ||
367 | for (i = 0; i < nr_tasks; i++) { | |
368 | task1 = tasks[i]; | |
369 | j = i + 1; | |
370 | if (j == nr_tasks) | |
371 | j = 0; | |
372 | task2 = tasks[j]; | |
373 | add_sched_event_wakeup(task1, 0, task2); | |
374 | } | |
375 | } | |
376 | ||
377 | static void | |
39aeb52f | 378 | process_sched_event(struct task_desc *this_task __used, struct sched_atom *atom) |
ec156764 IM |
379 | { |
380 | int ret = 0; | |
b1ffe8f3 | 381 | u64 now; |
ec156764 IM |
382 | long long delta; |
383 | ||
384 | now = get_nsecs(); | |
39aeb52f | 385 | delta = start_time + atom->timestamp - now; |
ec156764 | 386 | |
39aeb52f | 387 | switch (atom->type) { |
ec156764 | 388 | case SCHED_EVENT_RUN: |
39aeb52f | 389 | burn_nsecs(atom->duration); |
ec156764 IM |
390 | break; |
391 | case SCHED_EVENT_SLEEP: | |
39aeb52f | 392 | if (atom->wait_sem) |
393 | ret = sem_wait(atom->wait_sem); | |
ec156764 IM |
394 | BUG_ON(ret); |
395 | break; | |
396 | case SCHED_EVENT_WAKEUP: | |
39aeb52f | 397 | if (atom->wait_sem) |
398 | ret = sem_post(atom->wait_sem); | |
ec156764 IM |
399 | BUG_ON(ret); |
400 | break; | |
401 | default: | |
402 | BUG_ON(1); | |
403 | } | |
404 | } | |
405 | ||
b1ffe8f3 | 406 | static u64 get_cpu_usage_nsec_parent(void) |
ec156764 IM |
407 | { |
408 | struct rusage ru; | |
b1ffe8f3 | 409 | u64 sum; |
ec156764 IM |
410 | int err; |
411 | ||
412 | err = getrusage(RUSAGE_SELF, &ru); | |
413 | BUG_ON(err); | |
414 | ||
415 | sum = ru.ru_utime.tv_sec*1e9 + ru.ru_utime.tv_usec*1e3; | |
416 | sum += ru.ru_stime.tv_sec*1e9 + ru.ru_stime.tv_usec*1e3; | |
417 | ||
418 | return sum; | |
419 | } | |
420 | ||
b1ffe8f3 | 421 | static u64 get_cpu_usage_nsec_self(void) |
ec156764 IM |
422 | { |
423 | char filename [] = "/proc/1234567890/sched"; | |
424 | unsigned long msecs, nsecs; | |
425 | char *line = NULL; | |
b1ffe8f3 | 426 | u64 total = 0; |
ec156764 IM |
427 | size_t len = 0; |
428 | ssize_t chars; | |
429 | FILE *file; | |
430 | int ret; | |
431 | ||
432 | sprintf(filename, "/proc/%d/sched", getpid()); | |
433 | file = fopen(filename, "r"); | |
434 | BUG_ON(!file); | |
435 | ||
436 | while ((chars = getline(&line, &len, file)) != -1) { | |
ec156764 IM |
437 | ret = sscanf(line, "se.sum_exec_runtime : %ld.%06ld\n", |
438 | &msecs, &nsecs); | |
439 | if (ret == 2) { | |
440 | total = msecs*1e6 + nsecs; | |
ec156764 IM |
441 | break; |
442 | } | |
443 | } | |
444 | if (line) | |
445 | free(line); | |
446 | fclose(file); | |
447 | ||
448 | return total; | |
449 | } | |
450 | ||
451 | static void *thread_func(void *ctx) | |
452 | { | |
453 | struct task_desc *this_task = ctx; | |
b1ffe8f3 | 454 | u64 cpu_usage_0, cpu_usage_1; |
ec156764 IM |
455 | unsigned long i, ret; |
456 | char comm2[22]; | |
457 | ||
ec156764 IM |
458 | sprintf(comm2, ":%s", this_task->comm); |
459 | prctl(PR_SET_NAME, comm2); | |
460 | ||
461 | again: | |
462 | ret = sem_post(&this_task->ready_for_work); | |
463 | BUG_ON(ret); | |
ec156764 IM |
464 | ret = pthread_mutex_lock(&start_work_mutex); |
465 | BUG_ON(ret); | |
466 | ret = pthread_mutex_unlock(&start_work_mutex); | |
467 | BUG_ON(ret); | |
ec156764 IM |
468 | |
469 | cpu_usage_0 = get_cpu_usage_nsec_self(); | |
470 | ||
471 | for (i = 0; i < this_task->nr_events; i++) { | |
472 | this_task->curr_event = i; | |
39aeb52f | 473 | process_sched_event(this_task, this_task->atoms[i]); |
ec156764 IM |
474 | } |
475 | ||
476 | cpu_usage_1 = get_cpu_usage_nsec_self(); | |
477 | this_task->cpu_usage = cpu_usage_1 - cpu_usage_0; | |
478 | ||
ec156764 IM |
479 | ret = sem_post(&this_task->work_done_sem); |
480 | BUG_ON(ret); | |
ec156764 IM |
481 | |
482 | ret = pthread_mutex_lock(&work_done_wait_mutex); | |
483 | BUG_ON(ret); | |
484 | ret = pthread_mutex_unlock(&work_done_wait_mutex); | |
485 | BUG_ON(ret); | |
ec156764 IM |
486 | |
487 | goto again; | |
488 | } | |
489 | ||
490 | static void create_tasks(void) | |
491 | { | |
492 | struct task_desc *task; | |
493 | pthread_attr_t attr; | |
494 | unsigned long i; | |
495 | int err; | |
496 | ||
497 | err = pthread_attr_init(&attr); | |
498 | BUG_ON(err); | |
499 | err = pthread_attr_setstacksize(&attr, (size_t)(16*1024)); | |
500 | BUG_ON(err); | |
501 | err = pthread_mutex_lock(&start_work_mutex); | |
502 | BUG_ON(err); | |
503 | err = pthread_mutex_lock(&work_done_wait_mutex); | |
504 | BUG_ON(err); | |
505 | for (i = 0; i < nr_tasks; i++) { | |
506 | task = tasks[i]; | |
507 | sem_init(&task->sleep_sem, 0, 0); | |
508 | sem_init(&task->ready_for_work, 0, 0); | |
509 | sem_init(&task->work_done_sem, 0, 0); | |
510 | task->curr_event = 0; | |
511 | err = pthread_create(&task->thread, &attr, thread_func, task); | |
512 | BUG_ON(err); | |
513 | } | |
514 | } | |
515 | ||
ec156764 IM |
516 | static void wait_for_tasks(void) |
517 | { | |
b1ffe8f3 | 518 | u64 cpu_usage_0, cpu_usage_1; |
ec156764 IM |
519 | struct task_desc *task; |
520 | unsigned long i, ret; | |
521 | ||
ec156764 | 522 | start_time = get_nsecs(); |
ec156764 IM |
523 | cpu_usage = 0; |
524 | pthread_mutex_unlock(&work_done_wait_mutex); | |
525 | ||
526 | for (i = 0; i < nr_tasks; i++) { | |
527 | task = tasks[i]; | |
528 | ret = sem_wait(&task->ready_for_work); | |
529 | BUG_ON(ret); | |
530 | sem_init(&task->ready_for_work, 0, 0); | |
531 | } | |
532 | ret = pthread_mutex_lock(&work_done_wait_mutex); | |
533 | BUG_ON(ret); | |
534 | ||
535 | cpu_usage_0 = get_cpu_usage_nsec_parent(); | |
536 | ||
537 | pthread_mutex_unlock(&start_work_mutex); | |
538 | ||
ec156764 IM |
539 | for (i = 0; i < nr_tasks; i++) { |
540 | task = tasks[i]; | |
541 | ret = sem_wait(&task->work_done_sem); | |
542 | BUG_ON(ret); | |
543 | sem_init(&task->work_done_sem, 0, 0); | |
544 | cpu_usage += task->cpu_usage; | |
545 | task->cpu_usage = 0; | |
546 | } | |
547 | ||
548 | cpu_usage_1 = get_cpu_usage_nsec_parent(); | |
549 | if (!runavg_cpu_usage) | |
550 | runavg_cpu_usage = cpu_usage; | |
551 | runavg_cpu_usage = (runavg_cpu_usage*9 + cpu_usage)/10; | |
552 | ||
553 | parent_cpu_usage = cpu_usage_1 - cpu_usage_0; | |
554 | if (!runavg_parent_cpu_usage) | |
555 | runavg_parent_cpu_usage = parent_cpu_usage; | |
556 | runavg_parent_cpu_usage = (runavg_parent_cpu_usage*9 + | |
557 | parent_cpu_usage)/10; | |
558 | ||
559 | ret = pthread_mutex_lock(&start_work_mutex); | |
560 | BUG_ON(ret); | |
561 | ||
562 | for (i = 0; i < nr_tasks; i++) { | |
563 | task = tasks[i]; | |
564 | sem_init(&task->sleep_sem, 0, 0); | |
565 | task->curr_event = 0; | |
566 | } | |
567 | } | |
568 | ||
ec156764 IM |
569 | static void run_one_test(void) |
570 | { | |
b1ffe8f3 | 571 | u64 T0, T1, delta, avg_delta, fluct, std_dev; |
ec156764 IM |
572 | |
573 | T0 = get_nsecs(); | |
574 | wait_for_tasks(); | |
575 | T1 = get_nsecs(); | |
576 | ||
577 | delta = T1 - T0; | |
578 | sum_runtime += delta; | |
579 | nr_runs++; | |
580 | ||
581 | avg_delta = sum_runtime / nr_runs; | |
582 | if (delta < avg_delta) | |
583 | fluct = avg_delta - delta; | |
584 | else | |
585 | fluct = delta - avg_delta; | |
586 | sum_fluct += fluct; | |
587 | std_dev = sum_fluct / nr_runs / sqrt(nr_runs); | |
588 | if (!run_avg) | |
589 | run_avg = delta; | |
590 | run_avg = (run_avg*9 + delta)/10; | |
591 | ||
ad236fd2 | 592 | printf("#%-3ld: %0.3f, ", |
ec156764 IM |
593 | nr_runs, (double)delta/1000000.0); |
594 | ||
ad236fd2 | 595 | printf("ravg: %0.2f, ", |
ec156764 IM |
596 | (double)run_avg/1e6); |
597 | ||
ad236fd2 | 598 | printf("cpu: %0.2f / %0.2f", |
ec156764 IM |
599 | (double)cpu_usage/1e6, (double)runavg_cpu_usage/1e6); |
600 | ||
601 | #if 0 | |
602 | /* | |
fbf94829 IM |
603 | * rusage statistics done by the parent, these are less |
604 | * accurate than the sum_exec_runtime based statistics: | |
605 | */ | |
ad236fd2 | 606 | printf(" [%0.2f / %0.2f]", |
ec156764 IM |
607 | (double)parent_cpu_usage/1e6, |
608 | (double)runavg_parent_cpu_usage/1e6); | |
609 | #endif | |
610 | ||
ad236fd2 | 611 | printf("\n"); |
ec156764 IM |
612 | |
613 | if (nr_sleep_corrections) | |
ad236fd2 | 614 | printf(" (%ld sleep corrections)\n", nr_sleep_corrections); |
ec156764 IM |
615 | nr_sleep_corrections = 0; |
616 | } | |
617 | ||
618 | static void test_calibrations(void) | |
619 | { | |
b1ffe8f3 | 620 | u64 T0, T1; |
ec156764 IM |
621 | |
622 | T0 = get_nsecs(); | |
623 | burn_nsecs(1e6); | |
624 | T1 = get_nsecs(); | |
625 | ||
ad236fd2 | 626 | printf("the run test took %Ld nsecs\n", T1-T0); |
ec156764 IM |
627 | |
628 | T0 = get_nsecs(); | |
629 | sleep_nsecs(1e6); | |
630 | T1 = get_nsecs(); | |
631 | ||
ad236fd2 | 632 | printf("the sleep test took %Ld nsecs\n", T1-T0); |
ec156764 IM |
633 | } |
634 | ||
0a02ad93 IM |
635 | static int |
636 | process_comm_event(event_t *event, unsigned long offset, unsigned long head) | |
637 | { | |
638 | struct thread *thread; | |
639 | ||
640 | thread = threads__findnew(event->comm.pid, &threads, &last_match); | |
641 | ||
dc02bf71 | 642 | dump_printf("%p [%p]: perf_event_comm: %s:%d\n", |
0a02ad93 IM |
643 | (void *)(offset + head), |
644 | (void *)(long)(event->header.size), | |
645 | event->comm.comm, event->comm.pid); | |
646 | ||
647 | if (thread == NULL || | |
648 | thread__set_comm(thread, event->comm.comm)) { | |
dc02bf71 | 649 | dump_printf("problem processing perf_event_comm, skipping event.\n"); |
0a02ad93 IM |
650 | return -1; |
651 | } | |
652 | total_comm++; | |
653 | ||
654 | return 0; | |
655 | } | |
656 | ||
46538818 FW |
657 | |
658 | struct raw_event_sample { | |
659 | u32 size; | |
660 | char data[0]; | |
661 | }; | |
662 | ||
663 | #define FILL_FIELD(ptr, field, event, data) \ | |
664 | ptr.field = (typeof(ptr.field)) raw_field_value(event, #field, data) | |
665 | ||
666 | #define FILL_ARRAY(ptr, array, event, data) \ | |
667 | do { \ | |
668 | void *__array = raw_field_ptr(event, #array, data); \ | |
669 | memcpy(ptr.array, __array, sizeof(ptr.array)); \ | |
670 | } while(0) | |
671 | ||
672 | #define FILL_COMMON_FIELDS(ptr, event, data) \ | |
673 | do { \ | |
674 | FILL_FIELD(ptr, common_type, event, data); \ | |
675 | FILL_FIELD(ptr, common_flags, event, data); \ | |
676 | FILL_FIELD(ptr, common_preempt_count, event, data); \ | |
677 | FILL_FIELD(ptr, common_pid, event, data); \ | |
678 | FILL_FIELD(ptr, common_tgid, event, data); \ | |
679 | } while (0) | |
680 | ||
419ab0d6 FW |
681 | |
682 | ||
683 | struct trace_switch_event { | |
684 | u32 size; | |
685 | ||
686 | u16 common_type; | |
687 | u8 common_flags; | |
688 | u8 common_preempt_count; | |
689 | u32 common_pid; | |
690 | u32 common_tgid; | |
691 | ||
692 | char prev_comm[16]; | |
693 | u32 prev_pid; | |
694 | u32 prev_prio; | |
695 | u64 prev_state; | |
696 | char next_comm[16]; | |
697 | u32 next_pid; | |
698 | u32 next_prio; | |
699 | }; | |
700 | ||
39aeb52f | 701 | struct trace_runtime_event { |
702 | u32 size; | |
703 | ||
704 | u16 common_type; | |
705 | u8 common_flags; | |
706 | u8 common_preempt_count; | |
707 | u32 common_pid; | |
708 | u32 common_tgid; | |
709 | ||
710 | char comm[16]; | |
711 | u32 pid; | |
712 | u64 runtime; | |
713 | u64 vruntime; | |
714 | }; | |
419ab0d6 | 715 | |
fbf94829 IM |
716 | struct trace_wakeup_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 success; | |
730 | u32 cpu; | |
731 | }; | |
732 | ||
419ab0d6 FW |
733 | struct trace_fork_event { |
734 | u32 size; | |
46538818 | 735 | |
419ab0d6 FW |
736 | u16 common_type; |
737 | u8 common_flags; | |
738 | u8 common_preempt_count; | |
739 | u32 common_pid; | |
740 | u32 common_tgid; | |
741 | ||
742 | char parent_comm[16]; | |
743 | u32 parent_pid; | |
744 | char child_comm[16]; | |
745 | u32 child_pid; | |
746 | }; | |
747 | ||
748 | struct trace_sched_handler { | |
749 | void (*switch_event)(struct trace_switch_event *, | |
750 | struct event *, | |
751 | int cpu, | |
752 | u64 timestamp, | |
753 | struct thread *thread); | |
754 | ||
39aeb52f | 755 | void (*runtime_event)(struct trace_runtime_event *, |
756 | struct event *, | |
757 | int cpu, | |
758 | u64 timestamp, | |
759 | struct thread *thread); | |
760 | ||
419ab0d6 FW |
761 | void (*wakeup_event)(struct trace_wakeup_event *, |
762 | struct event *, | |
763 | int cpu, | |
764 | u64 timestamp, | |
765 | struct thread *thread); | |
766 | ||
767 | void (*fork_event)(struct trace_fork_event *, | |
768 | struct event *, | |
769 | int cpu, | |
770 | u64 timestamp, | |
771 | struct thread *thread); | |
772 | }; | |
46538818 | 773 | |
46538818 | 774 | |
419ab0d6 FW |
775 | static void |
776 | replay_wakeup_event(struct trace_wakeup_event *wakeup_event, | |
777 | struct event *event, | |
778 | int cpu __used, | |
779 | u64 timestamp __used, | |
780 | struct thread *thread __used) | |
781 | { | |
782 | struct task_desc *waker, *wakee; | |
fbf94829 | 783 | |
ad236fd2 IM |
784 | if (verbose) { |
785 | printf("sched_wakeup event %p\n", event); | |
fbf94829 | 786 | |
ad236fd2 | 787 | printf(" ... pid %d woke up %s/%d\n", |
419ab0d6 FW |
788 | wakeup_event->common_pid, |
789 | wakeup_event->comm, | |
790 | wakeup_event->pid); | |
ad236fd2 | 791 | } |
fbf94829 | 792 | |
419ab0d6 FW |
793 | waker = register_pid(wakeup_event->common_pid, "<unknown>"); |
794 | wakee = register_pid(wakeup_event->pid, wakeup_event->comm); | |
fbf94829 IM |
795 | |
796 | add_sched_event_wakeup(waker, timestamp, wakee); | |
ec156764 IM |
797 | } |
798 | ||
d1153389 | 799 | static u64 cpu_last_switched[MAX_CPUS]; |
fbf94829 IM |
800 | |
801 | static void | |
419ab0d6 FW |
802 | replay_switch_event(struct trace_switch_event *switch_event, |
803 | struct event *event, | |
804 | int cpu, | |
805 | u64 timestamp, | |
806 | struct thread *thread __used) | |
ec156764 | 807 | { |
fbf94829 IM |
808 | struct task_desc *prev, *next; |
809 | u64 timestamp0; | |
810 | s64 delta; | |
811 | ||
ad236fd2 IM |
812 | if (verbose) |
813 | printf("sched_switch event %p\n", event); | |
814 | ||
fbf94829 IM |
815 | if (cpu >= MAX_CPUS || cpu < 0) |
816 | return; | |
817 | ||
818 | timestamp0 = cpu_last_switched[cpu]; | |
819 | if (timestamp0) | |
820 | delta = timestamp - timestamp0; | |
821 | else | |
822 | delta = 0; | |
823 | ||
824 | if (delta < 0) | |
825 | die("hm, delta: %Ld < 0 ?\n", delta); | |
826 | ||
ad236fd2 IM |
827 | if (verbose) { |
828 | printf(" ... switch from %s/%d to %s/%d [ran %Ld nsecs]\n", | |
419ab0d6 FW |
829 | switch_event->prev_comm, switch_event->prev_pid, |
830 | switch_event->next_comm, switch_event->next_pid, | |
ad236fd2 IM |
831 | delta); |
832 | } | |
fbf94829 | 833 | |
419ab0d6 FW |
834 | prev = register_pid(switch_event->prev_pid, switch_event->prev_comm); |
835 | next = register_pid(switch_event->next_pid, switch_event->next_comm); | |
fbf94829 IM |
836 | |
837 | cpu_last_switched[cpu] = timestamp; | |
838 | ||
839 | add_sched_event_run(prev, timestamp, delta); | |
419ab0d6 | 840 | add_sched_event_sleep(prev, timestamp, switch_event->prev_state); |
fbf94829 IM |
841 | } |
842 | ||
fbf94829 | 843 | |
419ab0d6 FW |
844 | static void |
845 | replay_fork_event(struct trace_fork_event *fork_event, | |
846 | struct event *event, | |
847 | int cpu __used, | |
848 | u64 timestamp __used, | |
849 | struct thread *thread __used) | |
850 | { | |
851 | if (verbose) { | |
852 | printf("sched_fork event %p\n", event); | |
853 | printf("... parent: %s/%d\n", fork_event->parent_comm, fork_event->parent_pid); | |
854 | printf("... child: %s/%d\n", fork_event->child_comm, fork_event->child_pid); | |
855 | } | |
856 | register_pid(fork_event->parent_pid, fork_event->parent_comm); | |
857 | register_pid(fork_event->child_pid, fork_event->child_comm); | |
858 | } | |
fbf94829 | 859 | |
419ab0d6 | 860 | static struct trace_sched_handler replay_ops = { |
ea92ed5a IM |
861 | .wakeup_event = replay_wakeup_event, |
862 | .switch_event = replay_switch_event, | |
863 | .fork_event = replay_fork_event, | |
fbf94829 IM |
864 | }; |
865 | ||
b1ffe8f3 IM |
866 | struct sort_dimension { |
867 | const char *name; | |
b5fae128 | 868 | sort_fn_t cmp; |
b1ffe8f3 IM |
869 | struct list_head list; |
870 | }; | |
871 | ||
872 | static LIST_HEAD(cmp_pid); | |
873 | ||
daa1d7a5 | 874 | static int |
39aeb52f | 875 | thread_lat_cmp(struct list_head *list, struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 FW |
876 | { |
877 | struct sort_dimension *sort; | |
878 | int ret = 0; | |
879 | ||
b5fae128 IM |
880 | BUG_ON(list_empty(list)); |
881 | ||
daa1d7a5 FW |
882 | list_for_each_entry(sort, list, list) { |
883 | ret = sort->cmp(l, r); | |
884 | if (ret) | |
885 | return ret; | |
886 | } | |
887 | ||
888 | return ret; | |
889 | } | |
890 | ||
39aeb52f | 891 | static struct work_atoms * |
b5fae128 IM |
892 | thread_atoms_search(struct rb_root *root, struct thread *thread, |
893 | struct list_head *sort_list) | |
894 | { | |
895 | struct rb_node *node = root->rb_node; | |
39aeb52f | 896 | struct work_atoms key = { .thread = thread }; |
b5fae128 IM |
897 | |
898 | while (node) { | |
39aeb52f | 899 | struct work_atoms *atoms; |
b5fae128 IM |
900 | int cmp; |
901 | ||
39aeb52f | 902 | atoms = container_of(node, struct work_atoms, node); |
b5fae128 IM |
903 | |
904 | cmp = thread_lat_cmp(sort_list, &key, atoms); | |
905 | if (cmp > 0) | |
906 | node = node->rb_left; | |
907 | else if (cmp < 0) | |
908 | node = node->rb_right; | |
909 | else { | |
910 | BUG_ON(thread != atoms->thread); | |
911 | return atoms; | |
912 | } | |
913 | } | |
914 | return NULL; | |
915 | } | |
916 | ||
cdce9d73 | 917 | static void |
39aeb52f | 918 | __thread_latency_insert(struct rb_root *root, struct work_atoms *data, |
daa1d7a5 | 919 | struct list_head *sort_list) |
cdce9d73 FW |
920 | { |
921 | struct rb_node **new = &(root->rb_node), *parent = NULL; | |
922 | ||
923 | while (*new) { | |
39aeb52f | 924 | struct work_atoms *this; |
daa1d7a5 | 925 | int cmp; |
cdce9d73 | 926 | |
39aeb52f | 927 | this = container_of(*new, struct work_atoms, node); |
cdce9d73 | 928 | parent = *new; |
daa1d7a5 FW |
929 | |
930 | cmp = thread_lat_cmp(sort_list, data, this); | |
931 | ||
932 | if (cmp > 0) | |
cdce9d73 | 933 | new = &((*new)->rb_left); |
cdce9d73 | 934 | else |
daa1d7a5 | 935 | new = &((*new)->rb_right); |
cdce9d73 FW |
936 | } |
937 | ||
938 | rb_link_node(&data->node, parent, new); | |
939 | rb_insert_color(&data->node, root); | |
940 | } | |
941 | ||
b1ffe8f3 | 942 | static void thread_atoms_insert(struct thread *thread) |
cdce9d73 | 943 | { |
39aeb52f | 944 | struct work_atoms *atoms; |
b1ffe8f3 | 945 | |
17562205 FW |
946 | atoms = calloc(sizeof(*atoms), 1); |
947 | if (!atoms) | |
cdce9d73 FW |
948 | die("No memory"); |
949 | ||
17562205 | 950 | atoms->thread = thread; |
39aeb52f | 951 | INIT_LIST_HEAD(&atoms->work_list); |
b1ffe8f3 | 952 | __thread_latency_insert(&atom_root, atoms, &cmp_pid); |
cdce9d73 FW |
953 | } |
954 | ||
955 | static void | |
956 | latency_fork_event(struct trace_fork_event *fork_event __used, | |
957 | struct event *event __used, | |
958 | int cpu __used, | |
959 | u64 timestamp __used, | |
960 | struct thread *thread __used) | |
961 | { | |
962 | /* should insert the newcomer */ | |
963 | } | |
964 | ||
ea92ed5a | 965 | __used |
cdce9d73 FW |
966 | static char sched_out_state(struct trace_switch_event *switch_event) |
967 | { | |
968 | const char *str = TASK_STATE_TO_CHAR_STR; | |
969 | ||
970 | return str[switch_event->prev_state]; | |
971 | } | |
972 | ||
973 | static void | |
39aeb52f | 974 | add_sched_out_event(struct work_atoms *atoms, |
975 | char run_state, | |
976 | u64 timestamp) | |
cdce9d73 | 977 | { |
b1ffe8f3 | 978 | struct work_atom *atom; |
cdce9d73 | 979 | |
b1ffe8f3 IM |
980 | atom = calloc(sizeof(*atom), 1); |
981 | if (!atom) | |
cdce9d73 FW |
982 | die("Non memory"); |
983 | ||
aa1ab9d2 FW |
984 | atom->sched_out_time = timestamp; |
985 | ||
39aeb52f | 986 | if (run_state == 'R') { |
b1ffe8f3 | 987 | atom->state = THREAD_WAIT_CPU; |
aa1ab9d2 | 988 | atom->wake_up_time = atom->sched_out_time; |
c6ced611 FW |
989 | } |
990 | ||
39aeb52f | 991 | list_add_tail(&atom->list, &atoms->work_list); |
cdce9d73 FW |
992 | } |
993 | ||
994 | static void | |
39aeb52f | 995 | add_runtime_event(struct work_atoms *atoms, u64 delta, u64 timestamp __used) |
996 | { | |
997 | struct work_atom *atom; | |
998 | ||
999 | BUG_ON(list_empty(&atoms->work_list)); | |
1000 | ||
1001 | atom = list_entry(atoms->work_list.prev, struct work_atom, list); | |
1002 | ||
1003 | atom->runtime += delta; | |
1004 | atoms->total_runtime += delta; | |
1005 | } | |
1006 | ||
1007 | static void | |
1008 | add_sched_in_event(struct work_atoms *atoms, u64 timestamp) | |
cdce9d73 | 1009 | { |
b1ffe8f3 | 1010 | struct work_atom *atom; |
66685678 | 1011 | u64 delta; |
cdce9d73 | 1012 | |
39aeb52f | 1013 | if (list_empty(&atoms->work_list)) |
cdce9d73 FW |
1014 | return; |
1015 | ||
39aeb52f | 1016 | atom = list_entry(atoms->work_list.prev, struct work_atom, list); |
cdce9d73 | 1017 | |
b1ffe8f3 | 1018 | if (atom->state != THREAD_WAIT_CPU) |
cdce9d73 FW |
1019 | return; |
1020 | ||
b1ffe8f3 IM |
1021 | if (timestamp < atom->wake_up_time) { |
1022 | atom->state = THREAD_IGNORE; | |
cdce9d73 FW |
1023 | return; |
1024 | } | |
1025 | ||
b1ffe8f3 IM |
1026 | atom->state = THREAD_SCHED_IN; |
1027 | atom->sched_in_time = timestamp; | |
66685678 | 1028 | |
b1ffe8f3 | 1029 | delta = atom->sched_in_time - atom->wake_up_time; |
66685678 FW |
1030 | atoms->total_lat += delta; |
1031 | if (delta > atoms->max_lat) | |
1032 | atoms->max_lat = delta; | |
1033 | atoms->nb_atoms++; | |
cdce9d73 FW |
1034 | } |
1035 | ||
cdce9d73 FW |
1036 | static void |
1037 | latency_switch_event(struct trace_switch_event *switch_event, | |
1038 | struct event *event __used, | |
ea92ed5a | 1039 | int cpu, |
cdce9d73 FW |
1040 | u64 timestamp, |
1041 | struct thread *thread __used) | |
1042 | { | |
39aeb52f | 1043 | struct work_atoms *out_events, *in_events; |
cdce9d73 | 1044 | struct thread *sched_out, *sched_in; |
ea92ed5a IM |
1045 | u64 timestamp0; |
1046 | s64 delta; | |
1047 | ||
39aeb52f | 1048 | BUG_ON(cpu >= MAX_CPUS || cpu < 0); |
ea92ed5a IM |
1049 | |
1050 | timestamp0 = cpu_last_switched[cpu]; | |
1051 | cpu_last_switched[cpu] = timestamp; | |
1052 | if (timestamp0) | |
1053 | delta = timestamp - timestamp0; | |
1054 | else | |
1055 | delta = 0; | |
1056 | ||
1057 | if (delta < 0) | |
1058 | die("hm, delta: %Ld < 0 ?\n", delta); | |
1059 | ||
cdce9d73 FW |
1060 | |
1061 | sched_out = threads__findnew(switch_event->prev_pid, &threads, &last_match); | |
1062 | sched_in = threads__findnew(switch_event->next_pid, &threads, &last_match); | |
1063 | ||
39aeb52f | 1064 | out_events = thread_atoms_search(&atom_root, sched_out, &cmp_pid); |
1065 | if (!out_events) { | |
1066 | thread_atoms_insert(sched_out); | |
1067 | out_events = thread_atoms_search(&atom_root, sched_out, &cmp_pid); | |
1068 | if (!out_events) | |
1069 | die("out-event: Internal tree error"); | |
1070 | } | |
1071 | add_sched_out_event(out_events, sched_out_state(switch_event), timestamp); | |
1072 | ||
1073 | in_events = thread_atoms_search(&atom_root, sched_in, &cmp_pid); | |
1074 | if (!in_events) { | |
b1ffe8f3 | 1075 | thread_atoms_insert(sched_in); |
39aeb52f | 1076 | in_events = thread_atoms_search(&atom_root, sched_in, &cmp_pid); |
1077 | if (!in_events) | |
1078 | die("in-event: Internal tree error"); | |
1079 | /* | |
1080 | * Take came in we have not heard about yet, | |
1081 | * add in an initial atom in runnable state: | |
1082 | */ | |
1083 | add_sched_out_event(in_events, 'R', timestamp); | |
cdce9d73 | 1084 | } |
39aeb52f | 1085 | add_sched_in_event(in_events, timestamp); |
1086 | } | |
cdce9d73 | 1087 | |
39aeb52f | 1088 | static void |
1089 | latency_runtime_event(struct trace_runtime_event *runtime_event, | |
1090 | struct event *event __used, | |
1091 | int cpu, | |
1092 | u64 timestamp, | |
1093 | struct thread *this_thread __used) | |
1094 | { | |
1095 | struct work_atoms *atoms; | |
1096 | struct thread *thread; | |
1097 | ||
1098 | BUG_ON(cpu >= MAX_CPUS || cpu < 0); | |
1099 | ||
1100 | thread = threads__findnew(runtime_event->pid, &threads, &last_match); | |
1101 | atoms = thread_atoms_search(&atom_root, thread, &cmp_pid); | |
1102 | if (!atoms) { | |
1103 | thread_atoms_insert(thread); | |
1104 | atoms = thread_atoms_search(&atom_root, thread, &cmp_pid); | |
1105 | if (!atoms) | |
1106 | die("in-event: Internal tree error"); | |
1107 | add_sched_out_event(atoms, 'R', timestamp); | |
cdce9d73 FW |
1108 | } |
1109 | ||
39aeb52f | 1110 | add_runtime_event(atoms, runtime_event->runtime, timestamp); |
cdce9d73 FW |
1111 | } |
1112 | ||
1113 | static void | |
1114 | latency_wakeup_event(struct trace_wakeup_event *wakeup_event, | |
39aeb52f | 1115 | struct event *__event __used, |
cdce9d73 FW |
1116 | int cpu __used, |
1117 | u64 timestamp, | |
1118 | struct thread *thread __used) | |
1119 | { | |
39aeb52f | 1120 | struct work_atoms *atoms; |
b1ffe8f3 | 1121 | struct work_atom *atom; |
cdce9d73 FW |
1122 | struct thread *wakee; |
1123 | ||
1124 | /* Note for later, it may be interesting to observe the failing cases */ | |
1125 | if (!wakeup_event->success) | |
1126 | return; | |
1127 | ||
1128 | wakee = threads__findnew(wakeup_event->pid, &threads, &last_match); | |
b5fae128 | 1129 | atoms = thread_atoms_search(&atom_root, wakee, &cmp_pid); |
17562205 | 1130 | if (!atoms) { |
b1ffe8f3 | 1131 | thread_atoms_insert(wakee); |
39aeb52f | 1132 | atoms = thread_atoms_search(&atom_root, wakee, &cmp_pid); |
1133 | if (!atoms) | |
1134 | die("wakeup-event: Internal tree error"); | |
1135 | add_sched_out_event(atoms, 'S', timestamp); | |
cdce9d73 FW |
1136 | } |
1137 | ||
39aeb52f | 1138 | BUG_ON(list_empty(&atoms->work_list)); |
cdce9d73 | 1139 | |
39aeb52f | 1140 | atom = list_entry(atoms->work_list.prev, struct work_atom, list); |
cdce9d73 | 1141 | |
dc02bf71 IM |
1142 | if (atom->state != THREAD_SLEEPING) |
1143 | nr_state_machine_bugs++; | |
cdce9d73 | 1144 | |
ea57c4f5 IM |
1145 | nr_timestamps++; |
1146 | if (atom->sched_out_time > timestamp) { | |
dc02bf71 | 1147 | nr_unordered_timestamps++; |
aa1ab9d2 | 1148 | return; |
ea57c4f5 | 1149 | } |
aa1ab9d2 | 1150 | |
b1ffe8f3 IM |
1151 | atom->state = THREAD_WAIT_CPU; |
1152 | atom->wake_up_time = timestamp; | |
cdce9d73 FW |
1153 | } |
1154 | ||
1155 | static struct trace_sched_handler lat_ops = { | |
ea92ed5a IM |
1156 | .wakeup_event = latency_wakeup_event, |
1157 | .switch_event = latency_switch_event, | |
39aeb52f | 1158 | .runtime_event = latency_runtime_event, |
ea92ed5a | 1159 | .fork_event = latency_fork_event, |
cdce9d73 FW |
1160 | }; |
1161 | ||
39aeb52f | 1162 | static void output_lat_thread(struct work_atoms *work_list) |
cdce9d73 | 1163 | { |
cdce9d73 FW |
1164 | int i; |
1165 | int ret; | |
66685678 | 1166 | u64 avg; |
cdce9d73 | 1167 | |
39aeb52f | 1168 | if (!work_list->nb_atoms) |
cdce9d73 | 1169 | return; |
ea57c4f5 IM |
1170 | /* |
1171 | * Ignore idle threads: | |
1172 | */ | |
80ed0987 | 1173 | if (!strcmp(work_list->thread->comm, "swapper")) |
ea57c4f5 | 1174 | return; |
cdce9d73 | 1175 | |
39aeb52f | 1176 | all_runtime += work_list->total_runtime; |
1177 | all_count += work_list->nb_atoms; | |
66685678 | 1178 | |
80ed0987 | 1179 | ret = printf(" %s:%d ", work_list->thread->comm, work_list->thread->pid); |
cdce9d73 | 1180 | |
08f69e6c | 1181 | for (i = 0; i < 24 - ret; i++) |
cdce9d73 FW |
1182 | printf(" "); |
1183 | ||
39aeb52f | 1184 | avg = work_list->total_lat / work_list->nb_atoms; |
cdce9d73 | 1185 | |
dc02bf71 | 1186 | printf("|%11.3f ms |%9llu | avg:%9.3f ms | max:%9.3f ms |\n", |
39aeb52f | 1187 | (double)work_list->total_runtime / 1e6, |
1188 | work_list->nb_atoms, (double)avg / 1e6, | |
1189 | (double)work_list->max_lat / 1e6); | |
cdce9d73 FW |
1190 | } |
1191 | ||
39aeb52f | 1192 | static int pid_cmp(struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 | 1193 | { |
daa1d7a5 FW |
1194 | if (l->thread->pid < r->thread->pid) |
1195 | return -1; | |
1196 | if (l->thread->pid > r->thread->pid) | |
1197 | return 1; | |
1198 | ||
1199 | return 0; | |
1200 | } | |
1201 | ||
1202 | static struct sort_dimension pid_sort_dimension = { | |
b5fae128 IM |
1203 | .name = "pid", |
1204 | .cmp = pid_cmp, | |
daa1d7a5 FW |
1205 | }; |
1206 | ||
39aeb52f | 1207 | static int avg_cmp(struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 FW |
1208 | { |
1209 | u64 avgl, avgr; | |
1210 | ||
1211 | if (!l->nb_atoms) | |
1212 | return -1; | |
1213 | ||
1214 | if (!r->nb_atoms) | |
1215 | return 1; | |
1216 | ||
1217 | avgl = l->total_lat / l->nb_atoms; | |
1218 | avgr = r->total_lat / r->nb_atoms; | |
1219 | ||
1220 | if (avgl < avgr) | |
1221 | return -1; | |
1222 | if (avgl > avgr) | |
1223 | return 1; | |
1224 | ||
1225 | return 0; | |
1226 | } | |
1227 | ||
1228 | static struct sort_dimension avg_sort_dimension = { | |
b5fae128 IM |
1229 | .name = "avg", |
1230 | .cmp = avg_cmp, | |
daa1d7a5 FW |
1231 | }; |
1232 | ||
39aeb52f | 1233 | static int max_cmp(struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 FW |
1234 | { |
1235 | if (l->max_lat < r->max_lat) | |
1236 | return -1; | |
1237 | if (l->max_lat > r->max_lat) | |
1238 | return 1; | |
1239 | ||
1240 | return 0; | |
1241 | } | |
1242 | ||
1243 | static struct sort_dimension max_sort_dimension = { | |
b5fae128 IM |
1244 | .name = "max", |
1245 | .cmp = max_cmp, | |
daa1d7a5 FW |
1246 | }; |
1247 | ||
39aeb52f | 1248 | static int switch_cmp(struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 FW |
1249 | { |
1250 | if (l->nb_atoms < r->nb_atoms) | |
1251 | return -1; | |
1252 | if (l->nb_atoms > r->nb_atoms) | |
1253 | return 1; | |
1254 | ||
1255 | return 0; | |
1256 | } | |
1257 | ||
1258 | static struct sort_dimension switch_sort_dimension = { | |
b5fae128 IM |
1259 | .name = "switch", |
1260 | .cmp = switch_cmp, | |
daa1d7a5 FW |
1261 | }; |
1262 | ||
39aeb52f | 1263 | static int runtime_cmp(struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 FW |
1264 | { |
1265 | if (l->total_runtime < r->total_runtime) | |
1266 | return -1; | |
1267 | if (l->total_runtime > r->total_runtime) | |
1268 | return 1; | |
1269 | ||
1270 | return 0; | |
1271 | } | |
1272 | ||
1273 | static struct sort_dimension runtime_sort_dimension = { | |
b5fae128 IM |
1274 | .name = "runtime", |
1275 | .cmp = runtime_cmp, | |
daa1d7a5 FW |
1276 | }; |
1277 | ||
1278 | static struct sort_dimension *available_sorts[] = { | |
1279 | &pid_sort_dimension, | |
1280 | &avg_sort_dimension, | |
1281 | &max_sort_dimension, | |
1282 | &switch_sort_dimension, | |
1283 | &runtime_sort_dimension, | |
1284 | }; | |
1285 | ||
1286 | #define NB_AVAILABLE_SORTS (int)(sizeof(available_sorts) / sizeof(struct sort_dimension *)) | |
1287 | ||
1288 | static LIST_HEAD(sort_list); | |
1289 | ||
1290 | static int sort_dimension__add(char *tok, struct list_head *list) | |
1291 | { | |
1292 | int i; | |
1293 | ||
1294 | for (i = 0; i < NB_AVAILABLE_SORTS; i++) { | |
1295 | if (!strcmp(available_sorts[i]->name, tok)) { | |
1296 | list_add_tail(&available_sorts[i]->list, list); | |
1297 | ||
1298 | return 0; | |
1299 | } | |
1300 | } | |
1301 | ||
1302 | return -1; | |
1303 | } | |
1304 | ||
1305 | static void setup_sorting(void); | |
1306 | ||
1307 | static void sort_lat(void) | |
1308 | { | |
1309 | struct rb_node *node; | |
1310 | ||
1311 | for (;;) { | |
39aeb52f | 1312 | struct work_atoms *data; |
b1ffe8f3 | 1313 | node = rb_first(&atom_root); |
daa1d7a5 FW |
1314 | if (!node) |
1315 | break; | |
1316 | ||
b1ffe8f3 | 1317 | rb_erase(node, &atom_root); |
39aeb52f | 1318 | data = rb_entry(node, struct work_atoms, node); |
b1ffe8f3 | 1319 | __thread_latency_insert(&sorted_atom_root, data, &sort_list); |
daa1d7a5 FW |
1320 | } |
1321 | } | |
1322 | ||
419ab0d6 FW |
1323 | static struct trace_sched_handler *trace_handler; |
1324 | ||
fbf94829 | 1325 | static void |
419ab0d6 FW |
1326 | process_sched_wakeup_event(struct raw_event_sample *raw, |
1327 | struct event *event, | |
1328 | int cpu __used, | |
1329 | u64 timestamp __used, | |
1330 | struct thread *thread __used) | |
1331 | { | |
1332 | struct trace_wakeup_event wakeup_event; | |
1333 | ||
1334 | FILL_COMMON_FIELDS(wakeup_event, event, raw->data); | |
1335 | ||
1336 | FILL_ARRAY(wakeup_event, comm, event, raw->data); | |
1337 | FILL_FIELD(wakeup_event, pid, event, raw->data); | |
1338 | FILL_FIELD(wakeup_event, prio, event, raw->data); | |
1339 | FILL_FIELD(wakeup_event, success, event, raw->data); | |
1340 | FILL_FIELD(wakeup_event, cpu, event, raw->data); | |
1341 | ||
0ec04e16 IM |
1342 | if (trace_handler->wakeup_event) |
1343 | trace_handler->wakeup_event(&wakeup_event, event, cpu, timestamp, thread); | |
419ab0d6 FW |
1344 | } |
1345 | ||
c8a37751 IM |
1346 | /* |
1347 | * Track the current task - that way we can know whether there's any | |
1348 | * weird events, such as a task being switched away that is not current. | |
1349 | */ | |
40749d0f | 1350 | static int max_cpu; |
0ec04e16 | 1351 | |
c8a37751 IM |
1352 | static u32 curr_pid[MAX_CPUS] = { [0 ... MAX_CPUS-1] = -1 }; |
1353 | ||
0ec04e16 IM |
1354 | static struct thread *curr_thread[MAX_CPUS]; |
1355 | ||
1356 | static char next_shortname1 = 'A'; | |
1357 | static char next_shortname2 = '0'; | |
1358 | ||
1359 | static void | |
1360 | map_switch_event(struct trace_switch_event *switch_event, | |
1361 | struct event *event __used, | |
1362 | int this_cpu, | |
1363 | u64 timestamp, | |
1364 | struct thread *thread __used) | |
1365 | { | |
1366 | struct thread *sched_out, *sched_in; | |
1367 | int new_shortname; | |
1368 | u64 timestamp0; | |
1369 | s64 delta; | |
1370 | int cpu; | |
1371 | ||
1372 | BUG_ON(this_cpu >= MAX_CPUS || this_cpu < 0); | |
1373 | ||
1374 | if (this_cpu > max_cpu) | |
1375 | max_cpu = this_cpu; | |
1376 | ||
1377 | timestamp0 = cpu_last_switched[this_cpu]; | |
1378 | cpu_last_switched[this_cpu] = timestamp; | |
1379 | if (timestamp0) | |
1380 | delta = timestamp - timestamp0; | |
1381 | else | |
1382 | delta = 0; | |
1383 | ||
1384 | if (delta < 0) | |
1385 | die("hm, delta: %Ld < 0 ?\n", delta); | |
1386 | ||
1387 | ||
1388 | sched_out = threads__findnew(switch_event->prev_pid, &threads, &last_match); | |
1389 | sched_in = threads__findnew(switch_event->next_pid, &threads, &last_match); | |
1390 | ||
1391 | curr_thread[this_cpu] = sched_in; | |
1392 | ||
1393 | printf(" "); | |
1394 | ||
1395 | new_shortname = 0; | |
1396 | if (!sched_in->shortname[0]) { | |
1397 | sched_in->shortname[0] = next_shortname1; | |
1398 | sched_in->shortname[1] = next_shortname2; | |
1399 | ||
1400 | if (next_shortname1 < 'Z') { | |
1401 | next_shortname1++; | |
1402 | } else { | |
1403 | next_shortname1='A'; | |
1404 | if (next_shortname2 < '9') { | |
1405 | next_shortname2++; | |
1406 | } else { | |
1407 | next_shortname2='0'; | |
1408 | } | |
1409 | } | |
1410 | new_shortname = 1; | |
1411 | } | |
1412 | ||
1413 | for (cpu = 0; cpu <= max_cpu; cpu++) { | |
1414 | if (cpu != this_cpu) | |
1415 | printf(" "); | |
1416 | else | |
1417 | printf("*"); | |
1418 | ||
1419 | if (curr_thread[cpu]) { | |
1420 | if (curr_thread[cpu]->pid) | |
1421 | printf("%2s ", curr_thread[cpu]->shortname); | |
1422 | else | |
1423 | printf(". "); | |
1424 | } else | |
1425 | printf(" "); | |
1426 | } | |
1427 | ||
1428 | printf(" %12.6f secs ", (double)timestamp/1e9); | |
1429 | if (new_shortname) { | |
1430 | printf("%s => %s:%d\n", | |
1431 | sched_in->shortname, sched_in->comm, sched_in->pid); | |
1432 | } else { | |
1433 | printf("\n"); | |
1434 | } | |
1435 | } | |
1436 | ||
1437 | ||
419ab0d6 FW |
1438 | static void |
1439 | process_sched_switch_event(struct raw_event_sample *raw, | |
1440 | struct event *event, | |
0ec04e16 | 1441 | int this_cpu, |
419ab0d6 FW |
1442 | u64 timestamp __used, |
1443 | struct thread *thread __used) | |
1444 | { | |
1445 | struct trace_switch_event switch_event; | |
1446 | ||
1447 | FILL_COMMON_FIELDS(switch_event, event, raw->data); | |
1448 | ||
1449 | FILL_ARRAY(switch_event, prev_comm, event, raw->data); | |
1450 | FILL_FIELD(switch_event, prev_pid, event, raw->data); | |
1451 | FILL_FIELD(switch_event, prev_prio, event, raw->data); | |
1452 | FILL_FIELD(switch_event, prev_state, event, raw->data); | |
1453 | FILL_ARRAY(switch_event, next_comm, event, raw->data); | |
1454 | FILL_FIELD(switch_event, next_pid, event, raw->data); | |
1455 | FILL_FIELD(switch_event, next_prio, event, raw->data); | |
1456 | ||
0ec04e16 | 1457 | if (curr_pid[this_cpu] != (u32)-1) { |
c8a37751 IM |
1458 | /* |
1459 | * Are we trying to switch away a PID that is | |
1460 | * not current? | |
1461 | */ | |
0ec04e16 | 1462 | if (curr_pid[this_cpu] != switch_event.prev_pid) |
c8a37751 IM |
1463 | nr_context_switch_bugs++; |
1464 | } | |
0ec04e16 IM |
1465 | if (trace_handler->switch_event) |
1466 | trace_handler->switch_event(&switch_event, event, this_cpu, timestamp, thread); | |
c8a37751 | 1467 | |
0ec04e16 | 1468 | curr_pid[this_cpu] = switch_event.next_pid; |
419ab0d6 FW |
1469 | } |
1470 | ||
39aeb52f | 1471 | static void |
1472 | process_sched_runtime_event(struct raw_event_sample *raw, | |
1473 | struct event *event, | |
1474 | int cpu __used, | |
1475 | u64 timestamp __used, | |
1476 | struct thread *thread __used) | |
1477 | { | |
1478 | struct trace_runtime_event runtime_event; | |
1479 | ||
1480 | FILL_ARRAY(runtime_event, comm, event, raw->data); | |
1481 | FILL_FIELD(runtime_event, pid, event, raw->data); | |
1482 | FILL_FIELD(runtime_event, runtime, event, raw->data); | |
1483 | FILL_FIELD(runtime_event, vruntime, event, raw->data); | |
1484 | ||
0ec04e16 IM |
1485 | if (trace_handler->runtime_event) |
1486 | trace_handler->runtime_event(&runtime_event, event, cpu, timestamp, thread); | |
39aeb52f | 1487 | } |
1488 | ||
419ab0d6 FW |
1489 | static void |
1490 | process_sched_fork_event(struct raw_event_sample *raw, | |
1491 | struct event *event, | |
1492 | int cpu __used, | |
1493 | u64 timestamp __used, | |
1494 | struct thread *thread __used) | |
fbf94829 | 1495 | { |
46538818 FW |
1496 | struct trace_fork_event fork_event; |
1497 | ||
1498 | FILL_COMMON_FIELDS(fork_event, event, raw->data); | |
1499 | ||
1500 | FILL_ARRAY(fork_event, parent_comm, event, raw->data); | |
1501 | FILL_FIELD(fork_event, parent_pid, event, raw->data); | |
1502 | FILL_ARRAY(fork_event, child_comm, event, raw->data); | |
1503 | FILL_FIELD(fork_event, child_pid, event, raw->data); | |
1504 | ||
0ec04e16 IM |
1505 | if (trace_handler->fork_event) |
1506 | trace_handler->fork_event(&fork_event, event, cpu, timestamp, thread); | |
fbf94829 IM |
1507 | } |
1508 | ||
419ab0d6 FW |
1509 | static void |
1510 | process_sched_exit_event(struct event *event, | |
1511 | int cpu __used, | |
1512 | u64 timestamp __used, | |
1513 | struct thread *thread __used) | |
fbf94829 | 1514 | { |
ad236fd2 IM |
1515 | if (verbose) |
1516 | printf("sched_exit event %p\n", event); | |
ec156764 IM |
1517 | } |
1518 | ||
1519 | static void | |
ad236fd2 | 1520 | process_raw_event(event_t *raw_event __used, void *more_data, |
ec156764 IM |
1521 | int cpu, u64 timestamp, struct thread *thread) |
1522 | { | |
46538818 | 1523 | struct raw_event_sample *raw = more_data; |
ec156764 IM |
1524 | struct event *event; |
1525 | int type; | |
1526 | ||
1527 | type = trace_parse_common_type(raw->data); | |
1528 | event = trace_find_event(type); | |
1529 | ||
ec156764 | 1530 | if (!strcmp(event->name, "sched_switch")) |
46538818 | 1531 | process_sched_switch_event(raw, event, cpu, timestamp, thread); |
39aeb52f | 1532 | if (!strcmp(event->name, "sched_stat_runtime")) |
1533 | process_sched_runtime_event(raw, event, cpu, timestamp, thread); | |
ec156764 | 1534 | if (!strcmp(event->name, "sched_wakeup")) |
46538818 | 1535 | process_sched_wakeup_event(raw, event, cpu, timestamp, thread); |
fbf94829 | 1536 | if (!strcmp(event->name, "sched_wakeup_new")) |
46538818 | 1537 | process_sched_wakeup_event(raw, event, cpu, timestamp, thread); |
fbf94829 | 1538 | if (!strcmp(event->name, "sched_process_fork")) |
46538818 | 1539 | process_sched_fork_event(raw, event, cpu, timestamp, thread); |
fbf94829 IM |
1540 | if (!strcmp(event->name, "sched_process_exit")) |
1541 | process_sched_exit_event(event, cpu, timestamp, thread); | |
ec156764 IM |
1542 | } |
1543 | ||
0a02ad93 IM |
1544 | static int |
1545 | process_sample_event(event_t *event, unsigned long offset, unsigned long head) | |
1546 | { | |
0a02ad93 IM |
1547 | struct thread *thread; |
1548 | u64 ip = event->ip.ip; | |
1549 | u64 timestamp = -1; | |
1550 | u32 cpu = -1; | |
1551 | u64 period = 1; | |
1552 | void *more_data = event->ip.__more_data; | |
a80deb62 ACM |
1553 | |
1554 | if (!(sample_type & PERF_SAMPLE_RAW)) | |
1555 | return 0; | |
0a02ad93 IM |
1556 | |
1557 | thread = threads__findnew(event->ip.pid, &threads, &last_match); | |
1558 | ||
1559 | if (sample_type & PERF_SAMPLE_TIME) { | |
1560 | timestamp = *(u64 *)more_data; | |
1561 | more_data += sizeof(u64); | |
1562 | } | |
1563 | ||
1564 | if (sample_type & PERF_SAMPLE_CPU) { | |
1565 | cpu = *(u32 *)more_data; | |
1566 | more_data += sizeof(u32); | |
1567 | more_data += sizeof(u32); /* reserved */ | |
1568 | } | |
1569 | ||
1570 | if (sample_type & PERF_SAMPLE_PERIOD) { | |
1571 | period = *(u64 *)more_data; | |
1572 | more_data += sizeof(u64); | |
1573 | } | |
1574 | ||
cdd6c482 | 1575 | dump_printf("%p [%p]: PERF_RECORD_SAMPLE (IP, %d): %d/%d: %p period: %Ld\n", |
0a02ad93 IM |
1576 | (void *)(offset + head), |
1577 | (void *)(long)(event->header.size), | |
1578 | event->header.misc, | |
1579 | event->ip.pid, event->ip.tid, | |
1580 | (void *)(long)ip, | |
1581 | (long long)period); | |
1582 | ||
1583 | dump_printf(" ... thread: %s:%d\n", thread->comm, thread->pid); | |
1584 | ||
1585 | if (thread == NULL) { | |
1586 | eprintf("problem processing %d event, skipping it.\n", | |
1587 | event->header.type); | |
1588 | return -1; | |
1589 | } | |
1590 | ||
a80deb62 | 1591 | process_raw_event(event, more_data, cpu, timestamp, thread); |
0a02ad93 IM |
1592 | |
1593 | return 0; | |
1594 | } | |
1595 | ||
1596 | static int | |
1597 | process_event(event_t *event, unsigned long offset, unsigned long head) | |
1598 | { | |
1599 | trace_event(event); | |
1600 | ||
dc02bf71 | 1601 | nr_events++; |
0a02ad93 | 1602 | switch (event->header.type) { |
cdd6c482 | 1603 | case PERF_RECORD_MMAP: |
dc02bf71 | 1604 | return 0; |
cdd6c482 | 1605 | case PERF_RECORD_LOST: |
dc02bf71 IM |
1606 | nr_lost_chunks++; |
1607 | nr_lost_events += event->lost.lost; | |
0a02ad93 IM |
1608 | return 0; |
1609 | ||
cdd6c482 | 1610 | case PERF_RECORD_COMM: |
0a02ad93 IM |
1611 | return process_comm_event(event, offset, head); |
1612 | ||
cdd6c482 | 1613 | case PERF_RECORD_EXIT ... PERF_RECORD_READ: |
0a02ad93 IM |
1614 | return 0; |
1615 | ||
cdd6c482 | 1616 | case PERF_RECORD_SAMPLE: |
0a02ad93 IM |
1617 | return process_sample_event(event, offset, head); |
1618 | ||
cdd6c482 | 1619 | case PERF_RECORD_MAX: |
0a02ad93 IM |
1620 | default: |
1621 | return -1; | |
1622 | } | |
1623 | ||
1624 | return 0; | |
1625 | } | |
1626 | ||
46f392c9 | 1627 | static int read_events(void) |
0a02ad93 IM |
1628 | { |
1629 | int ret, rc = EXIT_FAILURE; | |
1630 | unsigned long offset = 0; | |
1631 | unsigned long head = 0; | |
1632 | struct stat perf_stat; | |
1633 | event_t *event; | |
1634 | uint32_t size; | |
1635 | char *buf; | |
1636 | ||
0a02ad93 IM |
1637 | register_idle_thread(&threads, &last_match); |
1638 | ||
1639 | input = open(input_name, O_RDONLY); | |
1640 | if (input < 0) { | |
1641 | perror("failed to open file"); | |
1642 | exit(-1); | |
1643 | } | |
1644 | ||
1645 | ret = fstat(input, &perf_stat); | |
1646 | if (ret < 0) { | |
1647 | perror("failed to stat file"); | |
1648 | exit(-1); | |
1649 | } | |
1650 | ||
1651 | if (!perf_stat.st_size) { | |
1652 | fprintf(stderr, "zero-sized file, nothing to do!\n"); | |
1653 | exit(0); | |
1654 | } | |
1655 | header = perf_header__read(input); | |
1656 | head = header->data_offset; | |
1657 | sample_type = perf_header__sample_type(header); | |
1658 | ||
1659 | if (!(sample_type & PERF_SAMPLE_RAW)) | |
1660 | die("No trace sample to read. Did you call perf record " | |
1661 | "without -R?"); | |
1662 | ||
1663 | if (load_kernel() < 0) { | |
1664 | perror("failed to load kernel symbols"); | |
1665 | return EXIT_FAILURE; | |
1666 | } | |
1667 | ||
1668 | remap: | |
1669 | buf = (char *)mmap(NULL, page_size * mmap_window, PROT_READ, | |
1670 | MAP_SHARED, input, offset); | |
1671 | if (buf == MAP_FAILED) { | |
1672 | perror("failed to mmap file"); | |
1673 | exit(-1); | |
1674 | } | |
1675 | ||
1676 | more: | |
1677 | event = (event_t *)(buf + head); | |
1678 | ||
1679 | size = event->header.size; | |
1680 | if (!size) | |
1681 | size = 8; | |
1682 | ||
1683 | if (head + event->header.size >= page_size * mmap_window) { | |
1684 | unsigned long shift = page_size * (head / page_size); | |
1685 | int res; | |
1686 | ||
1687 | res = munmap(buf, page_size * mmap_window); | |
1688 | assert(res == 0); | |
1689 | ||
1690 | offset += shift; | |
1691 | head -= shift; | |
1692 | goto remap; | |
1693 | } | |
1694 | ||
1695 | size = event->header.size; | |
1696 | ||
1697 | ||
1698 | if (!size || process_event(event, offset, head) < 0) { | |
1699 | ||
1700 | /* | |
1701 | * assume we lost track of the stream, check alignment, and | |
1702 | * increment a single u64 in the hope to catch on again 'soon'. | |
1703 | */ | |
1704 | ||
1705 | if (unlikely(head & 7)) | |
1706 | head &= ~7ULL; | |
1707 | ||
1708 | size = 8; | |
1709 | } | |
1710 | ||
1711 | head += size; | |
1712 | ||
1713 | if (offset + head < (unsigned long)perf_stat.st_size) | |
1714 | goto more; | |
1715 | ||
1716 | rc = EXIT_SUCCESS; | |
1717 | close(input); | |
1718 | ||
1719 | return rc; | |
1720 | } | |
1721 | ||
0ec04e16 IM |
1722 | static void print_bad_events(void) |
1723 | { | |
1724 | if (nr_unordered_timestamps && nr_timestamps) { | |
1725 | printf(" INFO: %.3f%% unordered timestamps (%ld out of %ld)\n", | |
1726 | (double)nr_unordered_timestamps/(double)nr_timestamps*100.0, | |
1727 | nr_unordered_timestamps, nr_timestamps); | |
1728 | } | |
1729 | if (nr_lost_events && nr_events) { | |
1730 | printf(" INFO: %.3f%% lost events (%ld out of %ld, in %ld chunks)\n", | |
1731 | (double)nr_lost_events/(double)nr_events*100.0, | |
1732 | nr_lost_events, nr_events, nr_lost_chunks); | |
1733 | } | |
1734 | if (nr_state_machine_bugs && nr_timestamps) { | |
1735 | printf(" INFO: %.3f%% state machine bugs (%ld out of %ld)", | |
1736 | (double)nr_state_machine_bugs/(double)nr_timestamps*100.0, | |
1737 | nr_state_machine_bugs, nr_timestamps); | |
1738 | if (nr_lost_events) | |
1739 | printf(" (due to lost events?)"); | |
1740 | printf("\n"); | |
1741 | } | |
1742 | if (nr_context_switch_bugs && nr_timestamps) { | |
1743 | printf(" INFO: %.3f%% context switch bugs (%ld out of %ld)", | |
1744 | (double)nr_context_switch_bugs/(double)nr_timestamps*100.0, | |
1745 | nr_context_switch_bugs, nr_timestamps); | |
1746 | if (nr_lost_events) | |
1747 | printf(" (due to lost events?)"); | |
1748 | printf("\n"); | |
1749 | } | |
1750 | } | |
1751 | ||
1752 | static void __cmd_lat(void) | |
1753 | { | |
1754 | struct rb_node *next; | |
1755 | ||
1756 | setup_pager(); | |
1757 | read_events(); | |
1758 | sort_lat(); | |
1759 | ||
1760 | printf("\n -----------------------------------------------------------------------------------------\n"); | |
1761 | printf(" Task | Runtime ms | Switches | Average delay ms | Maximum delay ms |\n"); | |
1762 | printf(" -----------------------------------------------------------------------------------------\n"); | |
1763 | ||
1764 | next = rb_first(&sorted_atom_root); | |
1765 | ||
1766 | while (next) { | |
1767 | struct work_atoms *work_list; | |
1768 | ||
1769 | work_list = rb_entry(next, struct work_atoms, node); | |
1770 | output_lat_thread(work_list); | |
1771 | next = rb_next(next); | |
1772 | } | |
1773 | ||
1774 | printf(" -----------------------------------------------------------------------------------------\n"); | |
1775 | printf(" TOTAL: |%11.3f ms |%9Ld |\n", | |
1776 | (double)all_runtime/1e6, all_count); | |
1777 | ||
1778 | printf(" ---------------------------------------------------\n"); | |
1779 | ||
1780 | print_bad_events(); | |
1781 | printf("\n"); | |
1782 | ||
1783 | } | |
1784 | ||
1785 | static struct trace_sched_handler map_ops = { | |
1786 | .wakeup_event = NULL, | |
1787 | .switch_event = map_switch_event, | |
1788 | .runtime_event = NULL, | |
1789 | .fork_event = NULL, | |
1790 | }; | |
1791 | ||
1792 | static void __cmd_map(void) | |
1793 | { | |
40749d0f IM |
1794 | max_cpu = sysconf(_SC_NPROCESSORS_CONF); |
1795 | ||
0ec04e16 IM |
1796 | setup_pager(); |
1797 | read_events(); | |
1798 | print_bad_events(); | |
1799 | } | |
1800 | ||
1801 | static void __cmd_replay(void) | |
1802 | { | |
1803 | unsigned long i; | |
1804 | ||
1805 | calibrate_run_measurement_overhead(); | |
1806 | calibrate_sleep_measurement_overhead(); | |
1807 | ||
1808 | test_calibrations(); | |
1809 | ||
1810 | read_events(); | |
1811 | ||
1812 | printf("nr_run_events: %ld\n", nr_run_events); | |
1813 | printf("nr_sleep_events: %ld\n", nr_sleep_events); | |
1814 | printf("nr_wakeup_events: %ld\n", nr_wakeup_events); | |
1815 | ||
1816 | if (targetless_wakeups) | |
1817 | printf("target-less wakeups: %ld\n", targetless_wakeups); | |
1818 | if (multitarget_wakeups) | |
1819 | printf("multi-target wakeups: %ld\n", multitarget_wakeups); | |
1820 | if (nr_run_events_optimized) | |
1821 | printf("run atoms optimized: %ld\n", | |
1822 | nr_run_events_optimized); | |
1823 | ||
1824 | print_task_traces(); | |
1825 | add_cross_task_wakeups(); | |
1826 | ||
1827 | create_tasks(); | |
1828 | printf("------------------------------------------------------------\n"); | |
1829 | for (i = 0; i < replay_repeat; i++) | |
1830 | run_one_test(); | |
1831 | } | |
1832 | ||
1833 | ||
46f392c9 | 1834 | static const char * const sched_usage[] = { |
4b77a729 | 1835 | "perf sched [<options>] {record|latency|map|replay|trace}", |
0a02ad93 IM |
1836 | NULL |
1837 | }; | |
1838 | ||
f2858d8a | 1839 | static const struct option sched_options[] = { |
4b77a729 MG |
1840 | OPT_STRING('i', "input", &input_name, "file", |
1841 | "input file name"), | |
f2858d8a IM |
1842 | OPT_BOOLEAN('v', "verbose", &verbose, |
1843 | "be more verbose (show symbol address, etc)"), | |
0a02ad93 IM |
1844 | OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, |
1845 | "dump raw trace in ASCII"), | |
f2858d8a IM |
1846 | OPT_END() |
1847 | }; | |
1848 | ||
1849 | static const char * const latency_usage[] = { | |
1850 | "perf sched latency [<options>]", | |
1851 | NULL | |
1852 | }; | |
1853 | ||
1854 | static const struct option latency_options[] = { | |
daa1d7a5 FW |
1855 | OPT_STRING('s', "sort", &sort_order, "key[,key2...]", |
1856 | "sort by key(s): runtime, switch, avg, max"), | |
0a02ad93 IM |
1857 | OPT_BOOLEAN('v', "verbose", &verbose, |
1858 | "be more verbose (show symbol address, etc)"), | |
f2858d8a IM |
1859 | OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, |
1860 | "dump raw trace in ASCII"), | |
1861 | OPT_END() | |
1862 | }; | |
1863 | ||
1864 | static const char * const replay_usage[] = { | |
1865 | "perf sched replay [<options>]", | |
1866 | NULL | |
1867 | }; | |
1868 | ||
1869 | static const struct option replay_options[] = { | |
1870 | OPT_INTEGER('r', "repeat", &replay_repeat, | |
1871 | "repeat the workload replay N times (-1: infinite)"), | |
1872 | OPT_BOOLEAN('v', "verbose", &verbose, | |
1873 | "be more verbose (show symbol address, etc)"), | |
1874 | OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, | |
1875 | "dump raw trace in ASCII"), | |
0a02ad93 IM |
1876 | OPT_END() |
1877 | }; | |
1878 | ||
daa1d7a5 FW |
1879 | static void setup_sorting(void) |
1880 | { | |
1881 | char *tmp, *tok, *str = strdup(sort_order); | |
1882 | ||
1883 | for (tok = strtok_r(str, ", ", &tmp); | |
1884 | tok; tok = strtok_r(NULL, ", ", &tmp)) { | |
1885 | if (sort_dimension__add(tok, &sort_list) < 0) { | |
1886 | error("Unknown --sort key: `%s'", tok); | |
f2858d8a | 1887 | usage_with_options(latency_usage, latency_options); |
daa1d7a5 FW |
1888 | } |
1889 | } | |
1890 | ||
1891 | free(str); | |
1892 | ||
1893 | sort_dimension__add((char *)"pid", &cmp_pid); | |
1894 | } | |
1895 | ||
1fc35b29 IM |
1896 | static const char *record_args[] = { |
1897 | "record", | |
1898 | "-a", | |
1899 | "-R", | |
d1302522 | 1900 | "-M", |
ea57c4f5 | 1901 | "-f", |
dc02bf71 | 1902 | "-m", "1024", |
1fc35b29 IM |
1903 | "-c", "1", |
1904 | "-e", "sched:sched_switch:r", | |
1905 | "-e", "sched:sched_stat_wait:r", | |
1906 | "-e", "sched:sched_stat_sleep:r", | |
1907 | "-e", "sched:sched_stat_iowait:r", | |
ea57c4f5 | 1908 | "-e", "sched:sched_stat_runtime:r", |
1fc35b29 IM |
1909 | "-e", "sched:sched_process_exit:r", |
1910 | "-e", "sched:sched_process_fork:r", | |
1911 | "-e", "sched:sched_wakeup:r", | |
1912 | "-e", "sched:sched_migrate_task:r", | |
1913 | }; | |
1914 | ||
1915 | static int __cmd_record(int argc, const char **argv) | |
1916 | { | |
1917 | unsigned int rec_argc, i, j; | |
1918 | const char **rec_argv; | |
1919 | ||
1920 | rec_argc = ARRAY_SIZE(record_args) + argc - 1; | |
1921 | rec_argv = calloc(rec_argc + 1, sizeof(char *)); | |
1922 | ||
1923 | for (i = 0; i < ARRAY_SIZE(record_args); i++) | |
1924 | rec_argv[i] = strdup(record_args[i]); | |
1925 | ||
1926 | for (j = 1; j < (unsigned int)argc; j++, i++) | |
1927 | rec_argv[i] = argv[j]; | |
1928 | ||
1929 | BUG_ON(i != rec_argc); | |
1930 | ||
1931 | return cmd_record(i, rec_argv, NULL); | |
1932 | } | |
1933 | ||
0a02ad93 IM |
1934 | int cmd_sched(int argc, const char **argv, const char *prefix __used) |
1935 | { | |
1936 | symbol__init(); | |
1937 | page_size = getpagesize(); | |
1938 | ||
f2858d8a IM |
1939 | argc = parse_options(argc, argv, sched_options, sched_usage, |
1940 | PARSE_OPT_STOP_AT_NON_OPTION); | |
1941 | if (!argc) | |
1942 | usage_with_options(sched_usage, sched_options); | |
0a02ad93 | 1943 | |
1fc35b29 IM |
1944 | if (!strncmp(argv[0], "rec", 3)) { |
1945 | return __cmd_record(argc, argv); | |
1946 | } else if (!strncmp(argv[0], "lat", 3)) { | |
cdce9d73 | 1947 | trace_handler = &lat_ops; |
f2858d8a IM |
1948 | if (argc > 1) { |
1949 | argc = parse_options(argc, argv, latency_options, latency_usage, 0); | |
1950 | if (argc) | |
1951 | usage_with_options(latency_usage, latency_options); | |
f2858d8a | 1952 | } |
b5fae128 | 1953 | setup_sorting(); |
46f392c9 | 1954 | __cmd_lat(); |
0ec04e16 IM |
1955 | } else if (!strcmp(argv[0], "map")) { |
1956 | trace_handler = &map_ops; | |
1957 | setup_sorting(); | |
1958 | __cmd_map(); | |
f2858d8a IM |
1959 | } else if (!strncmp(argv[0], "rep", 3)) { |
1960 | trace_handler = &replay_ops; | |
1961 | if (argc) { | |
1962 | argc = parse_options(argc, argv, replay_options, replay_usage, 0); | |
1963 | if (argc) | |
1964 | usage_with_options(replay_usage, replay_options); | |
1965 | } | |
1966 | __cmd_replay(); | |
c13f0d3c IM |
1967 | } else if (!strcmp(argv[0], "trace")) { |
1968 | /* | |
1969 | * Aliased to 'perf trace' for now: | |
1970 | */ | |
1971 | return cmd_trace(argc, argv, prefix); | |
f2858d8a IM |
1972 | } else { |
1973 | usage_with_options(sched_usage, sched_options); | |
1974 | } | |
1975 | ||
ec156764 | 1976 | return 0; |
0a02ad93 | 1977 | } |