Commit | Line | Data |
---|---|---|
0a02ad93 | 1 | #include "builtin.h" |
b1ffe8f3 | 2 | #include "perf.h" |
0a02ad93 IM |
3 | |
4 | #include "util/util.h" | |
ee29be62 | 5 | #include "util/evlist.h" |
0a02ad93 | 6 | #include "util/cache.h" |
e3f42609 | 7 | #include "util/evsel.h" |
0a02ad93 IM |
8 | #include "util/symbol.h" |
9 | #include "util/thread.h" | |
10 | #include "util/header.h" | |
94c744b6 | 11 | #include "util/session.h" |
45694aa7 | 12 | #include "util/tool.h" |
0a02ad93 IM |
13 | |
14 | #include "util/parse-options.h" | |
b1ffe8f3 | 15 | #include "util/trace-event.h" |
0a02ad93 | 16 | |
0a02ad93 IM |
17 | #include "util/debug.h" |
18 | ||
b1ffe8f3 | 19 | #include <sys/prctl.h> |
0a02ad93 | 20 | |
b1ffe8f3 IM |
21 | #include <semaphore.h> |
22 | #include <pthread.h> | |
23 | #include <math.h> | |
419ab0d6 | 24 | |
efad1415 | 25 | static const char *input_name; |
0a02ad93 | 26 | |
daa1d7a5 | 27 | static char default_sort_order[] = "avg, max, switch, runtime"; |
edb7c60e | 28 | static const char *sort_order = default_sort_order; |
daa1d7a5 | 29 | |
55ffb7a6 MG |
30 | static int profile_cpu = -1; |
31 | ||
b1ffe8f3 IM |
32 | #define PR_SET_NAME 15 /* Set process name */ |
33 | #define MAX_CPUS 4096 | |
0a02ad93 | 34 | |
b1ffe8f3 IM |
35 | static u64 run_measurement_overhead; |
36 | static u64 sleep_measurement_overhead; | |
ec156764 | 37 | |
b1ffe8f3 IM |
38 | #define COMM_LEN 20 |
39 | #define SYM_LEN 129 | |
ec156764 | 40 | |
b1ffe8f3 | 41 | #define MAX_PID 65536 |
ec156764 | 42 | |
b1ffe8f3 | 43 | static unsigned long nr_tasks; |
ec156764 | 44 | |
39aeb52f | 45 | struct sched_atom; |
ec156764 | 46 | |
b1ffe8f3 IM |
47 | struct task_desc { |
48 | unsigned long nr; | |
49 | unsigned long pid; | |
50 | char comm[COMM_LEN]; | |
ec156764 | 51 | |
b1ffe8f3 IM |
52 | unsigned long nr_events; |
53 | unsigned long curr_event; | |
39aeb52f | 54 | struct sched_atom **atoms; |
b1ffe8f3 IM |
55 | |
56 | pthread_t thread; | |
57 | sem_t sleep_sem; | |
ec156764 | 58 | |
b1ffe8f3 IM |
59 | sem_t ready_for_work; |
60 | sem_t work_done_sem; | |
61 | ||
62 | u64 cpu_usage; | |
63 | }; | |
64 | ||
65 | enum sched_event_type { | |
66 | SCHED_EVENT_RUN, | |
67 | SCHED_EVENT_SLEEP, | |
68 | SCHED_EVENT_WAKEUP, | |
55ffb7a6 | 69 | SCHED_EVENT_MIGRATION, |
b1ffe8f3 IM |
70 | }; |
71 | ||
39aeb52f | 72 | struct sched_atom { |
b1ffe8f3 | 73 | enum sched_event_type type; |
eed05fe7 | 74 | int specific_wait; |
b1ffe8f3 IM |
75 | u64 timestamp; |
76 | u64 duration; | |
77 | unsigned long nr; | |
b1ffe8f3 IM |
78 | sem_t *wait_sem; |
79 | struct task_desc *wakee; | |
80 | }; | |
81 | ||
82 | static struct task_desc *pid_to_task[MAX_PID]; | |
83 | ||
84 | static struct task_desc **tasks; | |
85 | ||
86 | static pthread_mutex_t start_work_mutex = PTHREAD_MUTEX_INITIALIZER; | |
87 | static u64 start_time; | |
88 | ||
89 | static pthread_mutex_t work_done_wait_mutex = PTHREAD_MUTEX_INITIALIZER; | |
ec156764 | 90 | |
b1ffe8f3 IM |
91 | static unsigned long nr_run_events; |
92 | static unsigned long nr_sleep_events; | |
93 | static unsigned long nr_wakeup_events; | |
94 | ||
95 | static unsigned long nr_sleep_corrections; | |
96 | static unsigned long nr_run_events_optimized; | |
97 | ||
98 | static unsigned long targetless_wakeups; | |
99 | static unsigned long multitarget_wakeups; | |
100 | ||
101 | static u64 cpu_usage; | |
102 | static u64 runavg_cpu_usage; | |
103 | static u64 parent_cpu_usage; | |
104 | static u64 runavg_parent_cpu_usage; | |
105 | ||
106 | static unsigned long nr_runs; | |
107 | static u64 sum_runtime; | |
108 | static u64 sum_fluct; | |
109 | static u64 run_avg; | |
110 | ||
1967936d | 111 | static unsigned int replay_repeat = 10; |
ea57c4f5 | 112 | static unsigned long nr_timestamps; |
dc02bf71 IM |
113 | static unsigned long nr_unordered_timestamps; |
114 | static unsigned long nr_state_machine_bugs; | |
c8a37751 | 115 | static unsigned long nr_context_switch_bugs; |
dc02bf71 IM |
116 | static unsigned long nr_events; |
117 | static unsigned long nr_lost_chunks; | |
118 | static unsigned long nr_lost_events; | |
b1ffe8f3 IM |
119 | |
120 | #define TASK_STATE_TO_CHAR_STR "RSDTtZX" | |
121 | ||
122 | enum thread_state { | |
123 | THREAD_SLEEPING = 0, | |
124 | THREAD_WAIT_CPU, | |
125 | THREAD_SCHED_IN, | |
126 | THREAD_IGNORE | |
127 | }; | |
128 | ||
129 | struct work_atom { | |
130 | struct list_head list; | |
131 | enum thread_state state; | |
aa1ab9d2 | 132 | u64 sched_out_time; |
b1ffe8f3 IM |
133 | u64 wake_up_time; |
134 | u64 sched_in_time; | |
135 | u64 runtime; | |
136 | }; | |
137 | ||
39aeb52f | 138 | struct work_atoms { |
139 | struct list_head work_list; | |
b1ffe8f3 IM |
140 | struct thread *thread; |
141 | struct rb_node node; | |
142 | u64 max_lat; | |
3786310a | 143 | u64 max_lat_at; |
b1ffe8f3 IM |
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 | ||
9486aa38 | 199 | printf("run measurement overhead: %" PRIu64 " 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 | ||
9486aa38 | 217 | printf("sleep measurement overhead: %" PRIu64 " 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; | |
ec156764 | 375 | |
39aeb52f | 376 | switch (atom->type) { |
ec156764 | 377 | case SCHED_EVENT_RUN: |
39aeb52f | 378 | burn_nsecs(atom->duration); |
ec156764 IM |
379 | break; |
380 | case SCHED_EVENT_SLEEP: | |
39aeb52f | 381 | if (atom->wait_sem) |
382 | ret = sem_wait(atom->wait_sem); | |
ec156764 IM |
383 | BUG_ON(ret); |
384 | break; | |
385 | case SCHED_EVENT_WAKEUP: | |
39aeb52f | 386 | if (atom->wait_sem) |
387 | ret = sem_post(atom->wait_sem); | |
ec156764 IM |
388 | BUG_ON(ret); |
389 | break; | |
55ffb7a6 MG |
390 | case SCHED_EVENT_MIGRATION: |
391 | break; | |
ec156764 IM |
392 | default: |
393 | BUG_ON(1); | |
394 | } | |
395 | } | |
396 | ||
b1ffe8f3 | 397 | static u64 get_cpu_usage_nsec_parent(void) |
ec156764 IM |
398 | { |
399 | struct rusage ru; | |
b1ffe8f3 | 400 | u64 sum; |
ec156764 IM |
401 | int err; |
402 | ||
403 | err = getrusage(RUSAGE_SELF, &ru); | |
404 | BUG_ON(err); | |
405 | ||
406 | sum = ru.ru_utime.tv_sec*1e9 + ru.ru_utime.tv_usec*1e3; | |
407 | sum += ru.ru_stime.tv_sec*1e9 + ru.ru_stime.tv_usec*1e3; | |
408 | ||
409 | return sum; | |
410 | } | |
411 | ||
c0c9e721 | 412 | static int self_open_counters(void) |
ec156764 | 413 | { |
c0c9e721 XG |
414 | struct perf_event_attr attr; |
415 | int fd; | |
ec156764 | 416 | |
c0c9e721 | 417 | memset(&attr, 0, sizeof(attr)); |
ec156764 | 418 | |
c0c9e721 XG |
419 | attr.type = PERF_TYPE_SOFTWARE; |
420 | attr.config = PERF_COUNT_SW_TASK_CLOCK; | |
ec156764 | 421 | |
c0c9e721 XG |
422 | fd = sys_perf_event_open(&attr, 0, -1, -1, 0); |
423 | ||
424 | if (fd < 0) | |
425 | die("Error: sys_perf_event_open() syscall returned" | |
426 | "with %d (%s)\n", fd, strerror(errno)); | |
427 | return fd; | |
428 | } | |
429 | ||
430 | static u64 get_cpu_usage_nsec_self(int fd) | |
431 | { | |
432 | u64 runtime; | |
433 | int ret; | |
434 | ||
435 | ret = read(fd, &runtime, sizeof(runtime)); | |
436 | BUG_ON(ret != sizeof(runtime)); | |
437 | ||
438 | return runtime; | |
ec156764 IM |
439 | } |
440 | ||
441 | static void *thread_func(void *ctx) | |
442 | { | |
443 | struct task_desc *this_task = ctx; | |
b1ffe8f3 | 444 | u64 cpu_usage_0, cpu_usage_1; |
ec156764 IM |
445 | unsigned long i, ret; |
446 | char comm2[22]; | |
c0c9e721 | 447 | int fd; |
ec156764 | 448 | |
ec156764 IM |
449 | sprintf(comm2, ":%s", this_task->comm); |
450 | prctl(PR_SET_NAME, comm2); | |
c0c9e721 | 451 | fd = self_open_counters(); |
ec156764 IM |
452 | |
453 | again: | |
454 | ret = sem_post(&this_task->ready_for_work); | |
455 | BUG_ON(ret); | |
ec156764 IM |
456 | ret = pthread_mutex_lock(&start_work_mutex); |
457 | BUG_ON(ret); | |
458 | ret = pthread_mutex_unlock(&start_work_mutex); | |
459 | BUG_ON(ret); | |
ec156764 | 460 | |
c0c9e721 | 461 | cpu_usage_0 = get_cpu_usage_nsec_self(fd); |
ec156764 IM |
462 | |
463 | for (i = 0; i < this_task->nr_events; i++) { | |
464 | this_task->curr_event = i; | |
39aeb52f | 465 | process_sched_event(this_task, this_task->atoms[i]); |
ec156764 IM |
466 | } |
467 | ||
c0c9e721 | 468 | cpu_usage_1 = get_cpu_usage_nsec_self(fd); |
ec156764 | 469 | this_task->cpu_usage = cpu_usage_1 - cpu_usage_0; |
ec156764 IM |
470 | ret = sem_post(&this_task->work_done_sem); |
471 | BUG_ON(ret); | |
ec156764 IM |
472 | |
473 | ret = pthread_mutex_lock(&work_done_wait_mutex); | |
474 | BUG_ON(ret); | |
475 | ret = pthread_mutex_unlock(&work_done_wait_mutex); | |
476 | BUG_ON(ret); | |
ec156764 IM |
477 | |
478 | goto again; | |
479 | } | |
480 | ||
481 | static void create_tasks(void) | |
482 | { | |
483 | struct task_desc *task; | |
484 | pthread_attr_t attr; | |
485 | unsigned long i; | |
486 | int err; | |
487 | ||
488 | err = pthread_attr_init(&attr); | |
489 | BUG_ON(err); | |
12f7e036 JP |
490 | err = pthread_attr_setstacksize(&attr, |
491 | (size_t) max(16 * 1024, PTHREAD_STACK_MIN)); | |
ec156764 IM |
492 | BUG_ON(err); |
493 | err = pthread_mutex_lock(&start_work_mutex); | |
494 | BUG_ON(err); | |
495 | err = pthread_mutex_lock(&work_done_wait_mutex); | |
496 | BUG_ON(err); | |
497 | for (i = 0; i < nr_tasks; i++) { | |
498 | task = tasks[i]; | |
499 | sem_init(&task->sleep_sem, 0, 0); | |
500 | sem_init(&task->ready_for_work, 0, 0); | |
501 | sem_init(&task->work_done_sem, 0, 0); | |
502 | task->curr_event = 0; | |
503 | err = pthread_create(&task->thread, &attr, thread_func, task); | |
504 | BUG_ON(err); | |
505 | } | |
506 | } | |
507 | ||
ec156764 IM |
508 | static void wait_for_tasks(void) |
509 | { | |
b1ffe8f3 | 510 | u64 cpu_usage_0, cpu_usage_1; |
ec156764 IM |
511 | struct task_desc *task; |
512 | unsigned long i, ret; | |
513 | ||
ec156764 | 514 | start_time = get_nsecs(); |
ec156764 IM |
515 | cpu_usage = 0; |
516 | pthread_mutex_unlock(&work_done_wait_mutex); | |
517 | ||
518 | for (i = 0; i < nr_tasks; i++) { | |
519 | task = tasks[i]; | |
520 | ret = sem_wait(&task->ready_for_work); | |
521 | BUG_ON(ret); | |
522 | sem_init(&task->ready_for_work, 0, 0); | |
523 | } | |
524 | ret = pthread_mutex_lock(&work_done_wait_mutex); | |
525 | BUG_ON(ret); | |
526 | ||
527 | cpu_usage_0 = get_cpu_usage_nsec_parent(); | |
528 | ||
529 | pthread_mutex_unlock(&start_work_mutex); | |
530 | ||
ec156764 IM |
531 | for (i = 0; i < nr_tasks; i++) { |
532 | task = tasks[i]; | |
533 | ret = sem_wait(&task->work_done_sem); | |
534 | BUG_ON(ret); | |
535 | sem_init(&task->work_done_sem, 0, 0); | |
536 | cpu_usage += task->cpu_usage; | |
537 | task->cpu_usage = 0; | |
538 | } | |
539 | ||
540 | cpu_usage_1 = get_cpu_usage_nsec_parent(); | |
541 | if (!runavg_cpu_usage) | |
542 | runavg_cpu_usage = cpu_usage; | |
543 | runavg_cpu_usage = (runavg_cpu_usage*9 + cpu_usage)/10; | |
544 | ||
545 | parent_cpu_usage = cpu_usage_1 - cpu_usage_0; | |
546 | if (!runavg_parent_cpu_usage) | |
547 | runavg_parent_cpu_usage = parent_cpu_usage; | |
548 | runavg_parent_cpu_usage = (runavg_parent_cpu_usage*9 + | |
549 | parent_cpu_usage)/10; | |
550 | ||
551 | ret = pthread_mutex_lock(&start_work_mutex); | |
552 | BUG_ON(ret); | |
553 | ||
554 | for (i = 0; i < nr_tasks; i++) { | |
555 | task = tasks[i]; | |
556 | sem_init(&task->sleep_sem, 0, 0); | |
557 | task->curr_event = 0; | |
558 | } | |
559 | } | |
560 | ||
ec156764 IM |
561 | static void run_one_test(void) |
562 | { | |
fb7d0b3c | 563 | u64 T0, T1, delta, avg_delta, fluct; |
ec156764 IM |
564 | |
565 | T0 = get_nsecs(); | |
566 | wait_for_tasks(); | |
567 | T1 = get_nsecs(); | |
568 | ||
569 | delta = T1 - T0; | |
570 | sum_runtime += delta; | |
571 | nr_runs++; | |
572 | ||
573 | avg_delta = sum_runtime / nr_runs; | |
574 | if (delta < avg_delta) | |
575 | fluct = avg_delta - delta; | |
576 | else | |
577 | fluct = delta - avg_delta; | |
578 | sum_fluct += fluct; | |
ec156764 IM |
579 | if (!run_avg) |
580 | run_avg = delta; | |
581 | run_avg = (run_avg*9 + delta)/10; | |
582 | ||
ad236fd2 | 583 | printf("#%-3ld: %0.3f, ", |
ec156764 IM |
584 | nr_runs, (double)delta/1000000.0); |
585 | ||
ad236fd2 | 586 | printf("ravg: %0.2f, ", |
ec156764 IM |
587 | (double)run_avg/1e6); |
588 | ||
ad236fd2 | 589 | printf("cpu: %0.2f / %0.2f", |
ec156764 IM |
590 | (double)cpu_usage/1e6, (double)runavg_cpu_usage/1e6); |
591 | ||
592 | #if 0 | |
593 | /* | |
fbf94829 IM |
594 | * rusage statistics done by the parent, these are less |
595 | * accurate than the sum_exec_runtime based statistics: | |
596 | */ | |
ad236fd2 | 597 | printf(" [%0.2f / %0.2f]", |
ec156764 IM |
598 | (double)parent_cpu_usage/1e6, |
599 | (double)runavg_parent_cpu_usage/1e6); | |
600 | #endif | |
601 | ||
ad236fd2 | 602 | printf("\n"); |
ec156764 IM |
603 | |
604 | if (nr_sleep_corrections) | |
ad236fd2 | 605 | printf(" (%ld sleep corrections)\n", nr_sleep_corrections); |
ec156764 IM |
606 | nr_sleep_corrections = 0; |
607 | } | |
608 | ||
609 | static void test_calibrations(void) | |
610 | { | |
b1ffe8f3 | 611 | u64 T0, T1; |
ec156764 IM |
612 | |
613 | T0 = get_nsecs(); | |
614 | burn_nsecs(1e6); | |
615 | T1 = get_nsecs(); | |
616 | ||
9486aa38 | 617 | printf("the run test took %" PRIu64 " nsecs\n", T1 - T0); |
ec156764 IM |
618 | |
619 | T0 = get_nsecs(); | |
620 | sleep_nsecs(1e6); | |
621 | T1 = get_nsecs(); | |
622 | ||
9486aa38 | 623 | printf("the sleep test took %" PRIu64 " nsecs\n", T1 - T0); |
ec156764 IM |
624 | } |
625 | ||
46538818 FW |
626 | #define FILL_FIELD(ptr, field, event, data) \ |
627 | ptr.field = (typeof(ptr.field)) raw_field_value(event, #field, data) | |
628 | ||
629 | #define FILL_ARRAY(ptr, array, event, data) \ | |
630 | do { \ | |
631 | void *__array = raw_field_ptr(event, #array, data); \ | |
632 | memcpy(ptr.array, __array, sizeof(ptr.array)); \ | |
633 | } while(0) | |
634 | ||
635 | #define FILL_COMMON_FIELDS(ptr, event, data) \ | |
636 | do { \ | |
637 | FILL_FIELD(ptr, common_type, event, data); \ | |
638 | FILL_FIELD(ptr, common_flags, event, data); \ | |
639 | FILL_FIELD(ptr, common_preempt_count, event, data); \ | |
640 | FILL_FIELD(ptr, common_pid, event, data); \ | |
641 | FILL_FIELD(ptr, common_tgid, event, data); \ | |
642 | } while (0) | |
643 | ||
419ab0d6 FW |
644 | |
645 | ||
646 | struct trace_switch_event { | |
647 | u32 size; | |
648 | ||
649 | u16 common_type; | |
650 | u8 common_flags; | |
651 | u8 common_preempt_count; | |
652 | u32 common_pid; | |
653 | u32 common_tgid; | |
654 | ||
655 | char prev_comm[16]; | |
656 | u32 prev_pid; | |
657 | u32 prev_prio; | |
658 | u64 prev_state; | |
659 | char next_comm[16]; | |
660 | u32 next_pid; | |
661 | u32 next_prio; | |
662 | }; | |
663 | ||
39aeb52f | 664 | struct trace_runtime_event { |
665 | u32 size; | |
666 | ||
667 | u16 common_type; | |
668 | u8 common_flags; | |
669 | u8 common_preempt_count; | |
670 | u32 common_pid; | |
671 | u32 common_tgid; | |
672 | ||
673 | char comm[16]; | |
674 | u32 pid; | |
675 | u64 runtime; | |
676 | u64 vruntime; | |
677 | }; | |
419ab0d6 | 678 | |
fbf94829 IM |
679 | struct trace_wakeup_event { |
680 | u32 size; | |
681 | ||
682 | u16 common_type; | |
683 | u8 common_flags; | |
684 | u8 common_preempt_count; | |
685 | u32 common_pid; | |
686 | u32 common_tgid; | |
687 | ||
688 | char comm[16]; | |
689 | u32 pid; | |
690 | ||
691 | u32 prio; | |
692 | u32 success; | |
693 | u32 cpu; | |
694 | }; | |
695 | ||
419ab0d6 FW |
696 | struct trace_fork_event { |
697 | u32 size; | |
46538818 | 698 | |
419ab0d6 FW |
699 | u16 common_type; |
700 | u8 common_flags; | |
701 | u8 common_preempt_count; | |
702 | u32 common_pid; | |
703 | u32 common_tgid; | |
704 | ||
705 | char parent_comm[16]; | |
706 | u32 parent_pid; | |
707 | char child_comm[16]; | |
708 | u32 child_pid; | |
709 | }; | |
710 | ||
55ffb7a6 MG |
711 | struct trace_migrate_task_event { |
712 | u32 size; | |
713 | ||
714 | u16 common_type; | |
715 | u8 common_flags; | |
716 | u8 common_preempt_count; | |
717 | u32 common_pid; | |
718 | u32 common_tgid; | |
719 | ||
720 | char comm[16]; | |
721 | u32 pid; | |
722 | ||
723 | u32 prio; | |
724 | u32 cpu; | |
725 | }; | |
726 | ||
419ab0d6 FW |
727 | struct trace_sched_handler { |
728 | void (*switch_event)(struct trace_switch_event *, | |
743eb868 | 729 | struct machine *, |
419ab0d6 FW |
730 | struct event *, |
731 | int cpu, | |
732 | u64 timestamp, | |
733 | struct thread *thread); | |
734 | ||
39aeb52f | 735 | void (*runtime_event)(struct trace_runtime_event *, |
743eb868 | 736 | struct machine *, |
39aeb52f | 737 | struct event *, |
738 | int cpu, | |
739 | u64 timestamp, | |
740 | struct thread *thread); | |
741 | ||
419ab0d6 | 742 | void (*wakeup_event)(struct trace_wakeup_event *, |
743eb868 | 743 | struct machine *, |
419ab0d6 FW |
744 | struct event *, |
745 | int cpu, | |
746 | u64 timestamp, | |
747 | struct thread *thread); | |
748 | ||
749 | void (*fork_event)(struct trace_fork_event *, | |
750 | struct event *, | |
751 | int cpu, | |
752 | u64 timestamp, | |
753 | struct thread *thread); | |
55ffb7a6 MG |
754 | |
755 | void (*migrate_task_event)(struct trace_migrate_task_event *, | |
743eb868 | 756 | struct machine *machine, |
55ffb7a6 MG |
757 | struct event *, |
758 | int cpu, | |
759 | u64 timestamp, | |
760 | struct thread *thread); | |
419ab0d6 | 761 | }; |
46538818 | 762 | |
46538818 | 763 | |
419ab0d6 FW |
764 | static void |
765 | replay_wakeup_event(struct trace_wakeup_event *wakeup_event, | |
743eb868 | 766 | struct machine *machine __used, |
419ab0d6 FW |
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 | 792 | replay_switch_event(struct trace_switch_event *switch_event, |
743eb868 | 793 | struct machine *machine __used, |
419ab0d6 FW |
794 | struct event *event, |
795 | int cpu, | |
796 | u64 timestamp, | |
797 | struct thread *thread __used) | |
ec156764 | 798 | { |
fb7d0b3c | 799 | struct task_desc *prev, __used *next; |
fbf94829 IM |
800 | u64 timestamp0; |
801 | s64 delta; | |
802 | ||
ad236fd2 IM |
803 | if (verbose) |
804 | printf("sched_switch event %p\n", event); | |
805 | ||
fbf94829 IM |
806 | if (cpu >= MAX_CPUS || cpu < 0) |
807 | return; | |
808 | ||
809 | timestamp0 = cpu_last_switched[cpu]; | |
810 | if (timestamp0) | |
811 | delta = timestamp - timestamp0; | |
812 | else | |
813 | delta = 0; | |
814 | ||
815 | if (delta < 0) | |
9486aa38 | 816 | die("hm, delta: %" PRIu64 " < 0 ?\n", delta); |
fbf94829 | 817 | |
ad236fd2 | 818 | if (verbose) { |
9486aa38 | 819 | printf(" ... switch from %s/%d to %s/%d [ran %" PRIu64 " nsecs]\n", |
419ab0d6 FW |
820 | switch_event->prev_comm, switch_event->prev_pid, |
821 | switch_event->next_comm, switch_event->next_pid, | |
ad236fd2 IM |
822 | delta); |
823 | } | |
fbf94829 | 824 | |
419ab0d6 FW |
825 | prev = register_pid(switch_event->prev_pid, switch_event->prev_comm); |
826 | next = register_pid(switch_event->next_pid, switch_event->next_comm); | |
fbf94829 IM |
827 | |
828 | cpu_last_switched[cpu] = timestamp; | |
829 | ||
830 | add_sched_event_run(prev, timestamp, delta); | |
419ab0d6 | 831 | add_sched_event_sleep(prev, timestamp, switch_event->prev_state); |
fbf94829 IM |
832 | } |
833 | ||
fbf94829 | 834 | |
419ab0d6 FW |
835 | static void |
836 | replay_fork_event(struct trace_fork_event *fork_event, | |
837 | struct event *event, | |
838 | int cpu __used, | |
839 | u64 timestamp __used, | |
840 | struct thread *thread __used) | |
841 | { | |
842 | if (verbose) { | |
843 | printf("sched_fork event %p\n", event); | |
844 | printf("... parent: %s/%d\n", fork_event->parent_comm, fork_event->parent_pid); | |
845 | printf("... child: %s/%d\n", fork_event->child_comm, fork_event->child_pid); | |
846 | } | |
847 | register_pid(fork_event->parent_pid, fork_event->parent_comm); | |
848 | register_pid(fork_event->child_pid, fork_event->child_comm); | |
849 | } | |
fbf94829 | 850 | |
419ab0d6 | 851 | static struct trace_sched_handler replay_ops = { |
ea92ed5a IM |
852 | .wakeup_event = replay_wakeup_event, |
853 | .switch_event = replay_switch_event, | |
854 | .fork_event = replay_fork_event, | |
fbf94829 IM |
855 | }; |
856 | ||
b1ffe8f3 IM |
857 | struct sort_dimension { |
858 | const char *name; | |
b5fae128 | 859 | sort_fn_t cmp; |
b1ffe8f3 IM |
860 | struct list_head list; |
861 | }; | |
862 | ||
863 | static LIST_HEAD(cmp_pid); | |
864 | ||
daa1d7a5 | 865 | static int |
39aeb52f | 866 | thread_lat_cmp(struct list_head *list, struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 FW |
867 | { |
868 | struct sort_dimension *sort; | |
869 | int ret = 0; | |
870 | ||
b5fae128 IM |
871 | BUG_ON(list_empty(list)); |
872 | ||
daa1d7a5 FW |
873 | list_for_each_entry(sort, list, list) { |
874 | ret = sort->cmp(l, r); | |
875 | if (ret) | |
876 | return ret; | |
877 | } | |
878 | ||
879 | return ret; | |
880 | } | |
881 | ||
39aeb52f | 882 | static struct work_atoms * |
b5fae128 IM |
883 | thread_atoms_search(struct rb_root *root, struct thread *thread, |
884 | struct list_head *sort_list) | |
885 | { | |
886 | struct rb_node *node = root->rb_node; | |
39aeb52f | 887 | struct work_atoms key = { .thread = thread }; |
b5fae128 IM |
888 | |
889 | while (node) { | |
39aeb52f | 890 | struct work_atoms *atoms; |
b5fae128 IM |
891 | int cmp; |
892 | ||
39aeb52f | 893 | atoms = container_of(node, struct work_atoms, node); |
b5fae128 IM |
894 | |
895 | cmp = thread_lat_cmp(sort_list, &key, atoms); | |
896 | if (cmp > 0) | |
897 | node = node->rb_left; | |
898 | else if (cmp < 0) | |
899 | node = node->rb_right; | |
900 | else { | |
901 | BUG_ON(thread != atoms->thread); | |
902 | return atoms; | |
903 | } | |
904 | } | |
905 | return NULL; | |
906 | } | |
907 | ||
cdce9d73 | 908 | static void |
39aeb52f | 909 | __thread_latency_insert(struct rb_root *root, struct work_atoms *data, |
daa1d7a5 | 910 | struct list_head *sort_list) |
cdce9d73 FW |
911 | { |
912 | struct rb_node **new = &(root->rb_node), *parent = NULL; | |
913 | ||
914 | while (*new) { | |
39aeb52f | 915 | struct work_atoms *this; |
daa1d7a5 | 916 | int cmp; |
cdce9d73 | 917 | |
39aeb52f | 918 | this = container_of(*new, struct work_atoms, node); |
cdce9d73 | 919 | parent = *new; |
daa1d7a5 FW |
920 | |
921 | cmp = thread_lat_cmp(sort_list, data, this); | |
922 | ||
923 | if (cmp > 0) | |
cdce9d73 | 924 | new = &((*new)->rb_left); |
cdce9d73 | 925 | else |
daa1d7a5 | 926 | new = &((*new)->rb_right); |
cdce9d73 FW |
927 | } |
928 | ||
929 | rb_link_node(&data->node, parent, new); | |
930 | rb_insert_color(&data->node, root); | |
931 | } | |
932 | ||
b1ffe8f3 | 933 | static void thread_atoms_insert(struct thread *thread) |
cdce9d73 | 934 | { |
36479484 | 935 | struct work_atoms *atoms = zalloc(sizeof(*atoms)); |
17562205 | 936 | if (!atoms) |
cdce9d73 FW |
937 | die("No memory"); |
938 | ||
17562205 | 939 | atoms->thread = thread; |
39aeb52f | 940 | INIT_LIST_HEAD(&atoms->work_list); |
b1ffe8f3 | 941 | __thread_latency_insert(&atom_root, atoms, &cmp_pid); |
cdce9d73 FW |
942 | } |
943 | ||
944 | static void | |
945 | latency_fork_event(struct trace_fork_event *fork_event __used, | |
946 | struct event *event __used, | |
947 | int cpu __used, | |
948 | u64 timestamp __used, | |
949 | struct thread *thread __used) | |
950 | { | |
951 | /* should insert the newcomer */ | |
952 | } | |
953 | ||
ea92ed5a | 954 | __used |
cdce9d73 FW |
955 | static char sched_out_state(struct trace_switch_event *switch_event) |
956 | { | |
957 | const char *str = TASK_STATE_TO_CHAR_STR; | |
958 | ||
959 | return str[switch_event->prev_state]; | |
960 | } | |
961 | ||
962 | static void | |
39aeb52f | 963 | add_sched_out_event(struct work_atoms *atoms, |
964 | char run_state, | |
965 | u64 timestamp) | |
cdce9d73 | 966 | { |
36479484 | 967 | struct work_atom *atom = zalloc(sizeof(*atom)); |
b1ffe8f3 | 968 | if (!atom) |
cdce9d73 FW |
969 | die("Non memory"); |
970 | ||
aa1ab9d2 FW |
971 | atom->sched_out_time = timestamp; |
972 | ||
39aeb52f | 973 | if (run_state == 'R') { |
b1ffe8f3 | 974 | atom->state = THREAD_WAIT_CPU; |
aa1ab9d2 | 975 | atom->wake_up_time = atom->sched_out_time; |
c6ced611 FW |
976 | } |
977 | ||
39aeb52f | 978 | list_add_tail(&atom->list, &atoms->work_list); |
cdce9d73 FW |
979 | } |
980 | ||
981 | static void | |
39aeb52f | 982 | add_runtime_event(struct work_atoms *atoms, u64 delta, u64 timestamp __used) |
983 | { | |
984 | struct work_atom *atom; | |
985 | ||
986 | BUG_ON(list_empty(&atoms->work_list)); | |
987 | ||
988 | atom = list_entry(atoms->work_list.prev, struct work_atom, list); | |
989 | ||
990 | atom->runtime += delta; | |
991 | atoms->total_runtime += delta; | |
992 | } | |
993 | ||
994 | static void | |
995 | add_sched_in_event(struct work_atoms *atoms, u64 timestamp) | |
cdce9d73 | 996 | { |
b1ffe8f3 | 997 | struct work_atom *atom; |
66685678 | 998 | u64 delta; |
cdce9d73 | 999 | |
39aeb52f | 1000 | if (list_empty(&atoms->work_list)) |
cdce9d73 FW |
1001 | return; |
1002 | ||
39aeb52f | 1003 | atom = list_entry(atoms->work_list.prev, struct work_atom, list); |
cdce9d73 | 1004 | |
b1ffe8f3 | 1005 | if (atom->state != THREAD_WAIT_CPU) |
cdce9d73 FW |
1006 | return; |
1007 | ||
b1ffe8f3 IM |
1008 | if (timestamp < atom->wake_up_time) { |
1009 | atom->state = THREAD_IGNORE; | |
cdce9d73 FW |
1010 | return; |
1011 | } | |
1012 | ||
b1ffe8f3 IM |
1013 | atom->state = THREAD_SCHED_IN; |
1014 | atom->sched_in_time = timestamp; | |
66685678 | 1015 | |
b1ffe8f3 | 1016 | delta = atom->sched_in_time - atom->wake_up_time; |
66685678 | 1017 | atoms->total_lat += delta; |
3786310a | 1018 | if (delta > atoms->max_lat) { |
66685678 | 1019 | atoms->max_lat = delta; |
3786310a FW |
1020 | atoms->max_lat_at = timestamp; |
1021 | } | |
66685678 | 1022 | atoms->nb_atoms++; |
cdce9d73 FW |
1023 | } |
1024 | ||
cdce9d73 FW |
1025 | static void |
1026 | latency_switch_event(struct trace_switch_event *switch_event, | |
743eb868 | 1027 | struct machine *machine, |
cdce9d73 | 1028 | struct event *event __used, |
ea92ed5a | 1029 | int cpu, |
cdce9d73 FW |
1030 | u64 timestamp, |
1031 | struct thread *thread __used) | |
1032 | { | |
39aeb52f | 1033 | struct work_atoms *out_events, *in_events; |
cdce9d73 | 1034 | struct thread *sched_out, *sched_in; |
ea92ed5a IM |
1035 | u64 timestamp0; |
1036 | s64 delta; | |
1037 | ||
39aeb52f | 1038 | BUG_ON(cpu >= MAX_CPUS || cpu < 0); |
ea92ed5a IM |
1039 | |
1040 | timestamp0 = cpu_last_switched[cpu]; | |
1041 | cpu_last_switched[cpu] = timestamp; | |
1042 | if (timestamp0) | |
1043 | delta = timestamp - timestamp0; | |
1044 | else | |
1045 | delta = 0; | |
1046 | ||
1047 | if (delta < 0) | |
9486aa38 | 1048 | die("hm, delta: %" PRIu64 " < 0 ?\n", delta); |
ea92ed5a | 1049 | |
cdce9d73 | 1050 | |
743eb868 ACM |
1051 | sched_out = machine__findnew_thread(machine, switch_event->prev_pid); |
1052 | sched_in = machine__findnew_thread(machine, switch_event->next_pid); | |
cdce9d73 | 1053 | |
39aeb52f | 1054 | out_events = thread_atoms_search(&atom_root, sched_out, &cmp_pid); |
1055 | if (!out_events) { | |
1056 | thread_atoms_insert(sched_out); | |
1057 | out_events = thread_atoms_search(&atom_root, sched_out, &cmp_pid); | |
1058 | if (!out_events) | |
1059 | die("out-event: Internal tree error"); | |
1060 | } | |
1061 | add_sched_out_event(out_events, sched_out_state(switch_event), timestamp); | |
1062 | ||
1063 | in_events = thread_atoms_search(&atom_root, sched_in, &cmp_pid); | |
1064 | if (!in_events) { | |
b1ffe8f3 | 1065 | thread_atoms_insert(sched_in); |
39aeb52f | 1066 | in_events = thread_atoms_search(&atom_root, sched_in, &cmp_pid); |
1067 | if (!in_events) | |
1068 | die("in-event: Internal tree error"); | |
1069 | /* | |
1070 | * Take came in we have not heard about yet, | |
1071 | * add in an initial atom in runnable state: | |
1072 | */ | |
1073 | add_sched_out_event(in_events, 'R', timestamp); | |
cdce9d73 | 1074 | } |
39aeb52f | 1075 | add_sched_in_event(in_events, timestamp); |
1076 | } | |
cdce9d73 | 1077 | |
39aeb52f | 1078 | static void |
1079 | latency_runtime_event(struct trace_runtime_event *runtime_event, | |
743eb868 | 1080 | struct machine *machine, |
39aeb52f | 1081 | struct event *event __used, |
1082 | int cpu, | |
1083 | u64 timestamp, | |
1084 | struct thread *this_thread __used) | |
1085 | { | |
743eb868 | 1086 | struct thread *thread = machine__findnew_thread(machine, runtime_event->pid); |
d5b889f2 | 1087 | struct work_atoms *atoms = thread_atoms_search(&atom_root, thread, &cmp_pid); |
39aeb52f | 1088 | |
1089 | BUG_ON(cpu >= MAX_CPUS || cpu < 0); | |
39aeb52f | 1090 | if (!atoms) { |
1091 | thread_atoms_insert(thread); | |
1092 | atoms = thread_atoms_search(&atom_root, thread, &cmp_pid); | |
1093 | if (!atoms) | |
1094 | die("in-event: Internal tree error"); | |
1095 | add_sched_out_event(atoms, 'R', timestamp); | |
cdce9d73 FW |
1096 | } |
1097 | ||
39aeb52f | 1098 | add_runtime_event(atoms, runtime_event->runtime, timestamp); |
cdce9d73 FW |
1099 | } |
1100 | ||
1101 | static void | |
1102 | latency_wakeup_event(struct trace_wakeup_event *wakeup_event, | |
743eb868 | 1103 | struct machine *machine, |
39aeb52f | 1104 | struct event *__event __used, |
cdce9d73 FW |
1105 | int cpu __used, |
1106 | u64 timestamp, | |
1107 | struct thread *thread __used) | |
1108 | { | |
39aeb52f | 1109 | struct work_atoms *atoms; |
b1ffe8f3 | 1110 | struct work_atom *atom; |
cdce9d73 FW |
1111 | struct thread *wakee; |
1112 | ||
1113 | /* Note for later, it may be interesting to observe the failing cases */ | |
1114 | if (!wakeup_event->success) | |
1115 | return; | |
1116 | ||
743eb868 | 1117 | wakee = machine__findnew_thread(machine, wakeup_event->pid); |
b5fae128 | 1118 | atoms = thread_atoms_search(&atom_root, wakee, &cmp_pid); |
17562205 | 1119 | if (!atoms) { |
b1ffe8f3 | 1120 | thread_atoms_insert(wakee); |
39aeb52f | 1121 | atoms = thread_atoms_search(&atom_root, wakee, &cmp_pid); |
1122 | if (!atoms) | |
1123 | die("wakeup-event: Internal tree error"); | |
1124 | add_sched_out_event(atoms, 'S', timestamp); | |
cdce9d73 FW |
1125 | } |
1126 | ||
39aeb52f | 1127 | BUG_ON(list_empty(&atoms->work_list)); |
cdce9d73 | 1128 | |
39aeb52f | 1129 | atom = list_entry(atoms->work_list.prev, struct work_atom, list); |
cdce9d73 | 1130 | |
55ffb7a6 MG |
1131 | /* |
1132 | * You WILL be missing events if you've recorded only | |
1133 | * one CPU, or are only looking at only one, so don't | |
1134 | * make useless noise. | |
1135 | */ | |
1136 | if (profile_cpu == -1 && atom->state != THREAD_SLEEPING) | |
dc02bf71 | 1137 | nr_state_machine_bugs++; |
cdce9d73 | 1138 | |
ea57c4f5 IM |
1139 | nr_timestamps++; |
1140 | if (atom->sched_out_time > timestamp) { | |
dc02bf71 | 1141 | nr_unordered_timestamps++; |
aa1ab9d2 | 1142 | return; |
ea57c4f5 | 1143 | } |
aa1ab9d2 | 1144 | |
b1ffe8f3 IM |
1145 | atom->state = THREAD_WAIT_CPU; |
1146 | atom->wake_up_time = timestamp; | |
cdce9d73 FW |
1147 | } |
1148 | ||
55ffb7a6 MG |
1149 | static void |
1150 | latency_migrate_task_event(struct trace_migrate_task_event *migrate_task_event, | |
743eb868 | 1151 | struct machine *machine, |
55ffb7a6 MG |
1152 | struct event *__event __used, |
1153 | int cpu __used, | |
1154 | u64 timestamp, | |
1155 | struct thread *thread __used) | |
1156 | { | |
1157 | struct work_atoms *atoms; | |
1158 | struct work_atom *atom; | |
1159 | struct thread *migrant; | |
1160 | ||
1161 | /* | |
1162 | * Only need to worry about migration when profiling one CPU. | |
1163 | */ | |
1164 | if (profile_cpu == -1) | |
1165 | return; | |
1166 | ||
743eb868 | 1167 | migrant = machine__findnew_thread(machine, migrate_task_event->pid); |
55ffb7a6 MG |
1168 | atoms = thread_atoms_search(&atom_root, migrant, &cmp_pid); |
1169 | if (!atoms) { | |
1170 | thread_atoms_insert(migrant); | |
1171 | register_pid(migrant->pid, migrant->comm); | |
1172 | atoms = thread_atoms_search(&atom_root, migrant, &cmp_pid); | |
1173 | if (!atoms) | |
1174 | die("migration-event: Internal tree error"); | |
1175 | add_sched_out_event(atoms, 'R', timestamp); | |
1176 | } | |
1177 | ||
1178 | BUG_ON(list_empty(&atoms->work_list)); | |
1179 | ||
1180 | atom = list_entry(atoms->work_list.prev, struct work_atom, list); | |
1181 | atom->sched_in_time = atom->sched_out_time = atom->wake_up_time = timestamp; | |
1182 | ||
1183 | nr_timestamps++; | |
1184 | ||
1185 | if (atom->sched_out_time > timestamp) | |
1186 | nr_unordered_timestamps++; | |
1187 | } | |
1188 | ||
cdce9d73 | 1189 | static struct trace_sched_handler lat_ops = { |
ea92ed5a IM |
1190 | .wakeup_event = latency_wakeup_event, |
1191 | .switch_event = latency_switch_event, | |
39aeb52f | 1192 | .runtime_event = latency_runtime_event, |
ea92ed5a | 1193 | .fork_event = latency_fork_event, |
55ffb7a6 | 1194 | .migrate_task_event = latency_migrate_task_event, |
cdce9d73 FW |
1195 | }; |
1196 | ||
39aeb52f | 1197 | static void output_lat_thread(struct work_atoms *work_list) |
cdce9d73 | 1198 | { |
cdce9d73 FW |
1199 | int i; |
1200 | int ret; | |
66685678 | 1201 | u64 avg; |
cdce9d73 | 1202 | |
39aeb52f | 1203 | if (!work_list->nb_atoms) |
cdce9d73 | 1204 | return; |
ea57c4f5 IM |
1205 | /* |
1206 | * Ignore idle threads: | |
1207 | */ | |
80ed0987 | 1208 | if (!strcmp(work_list->thread->comm, "swapper")) |
ea57c4f5 | 1209 | return; |
cdce9d73 | 1210 | |
39aeb52f | 1211 | all_runtime += work_list->total_runtime; |
1212 | all_count += work_list->nb_atoms; | |
66685678 | 1213 | |
80ed0987 | 1214 | ret = printf(" %s:%d ", work_list->thread->comm, work_list->thread->pid); |
cdce9d73 | 1215 | |
08f69e6c | 1216 | for (i = 0; i < 24 - ret; i++) |
cdce9d73 FW |
1217 | printf(" "); |
1218 | ||
39aeb52f | 1219 | avg = work_list->total_lat / work_list->nb_atoms; |
cdce9d73 | 1220 | |
9486aa38 | 1221 | printf("|%11.3f ms |%9" PRIu64 " | avg:%9.3f ms | max:%9.3f ms | max at: %9.6f s\n", |
39aeb52f | 1222 | (double)work_list->total_runtime / 1e6, |
1223 | work_list->nb_atoms, (double)avg / 1e6, | |
3786310a FW |
1224 | (double)work_list->max_lat / 1e6, |
1225 | (double)work_list->max_lat_at / 1e9); | |
cdce9d73 FW |
1226 | } |
1227 | ||
39aeb52f | 1228 | static int pid_cmp(struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 | 1229 | { |
daa1d7a5 FW |
1230 | if (l->thread->pid < r->thread->pid) |
1231 | return -1; | |
1232 | if (l->thread->pid > r->thread->pid) | |
1233 | return 1; | |
1234 | ||
1235 | return 0; | |
1236 | } | |
1237 | ||
1238 | static struct sort_dimension pid_sort_dimension = { | |
b5fae128 IM |
1239 | .name = "pid", |
1240 | .cmp = pid_cmp, | |
daa1d7a5 FW |
1241 | }; |
1242 | ||
39aeb52f | 1243 | static int avg_cmp(struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 FW |
1244 | { |
1245 | u64 avgl, avgr; | |
1246 | ||
1247 | if (!l->nb_atoms) | |
1248 | return -1; | |
1249 | ||
1250 | if (!r->nb_atoms) | |
1251 | return 1; | |
1252 | ||
1253 | avgl = l->total_lat / l->nb_atoms; | |
1254 | avgr = r->total_lat / r->nb_atoms; | |
1255 | ||
1256 | if (avgl < avgr) | |
1257 | return -1; | |
1258 | if (avgl > avgr) | |
1259 | return 1; | |
1260 | ||
1261 | return 0; | |
1262 | } | |
1263 | ||
1264 | static struct sort_dimension avg_sort_dimension = { | |
b5fae128 IM |
1265 | .name = "avg", |
1266 | .cmp = avg_cmp, | |
daa1d7a5 FW |
1267 | }; |
1268 | ||
39aeb52f | 1269 | static int max_cmp(struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 FW |
1270 | { |
1271 | if (l->max_lat < r->max_lat) | |
1272 | return -1; | |
1273 | if (l->max_lat > r->max_lat) | |
1274 | return 1; | |
1275 | ||
1276 | return 0; | |
1277 | } | |
1278 | ||
1279 | static struct sort_dimension max_sort_dimension = { | |
b5fae128 IM |
1280 | .name = "max", |
1281 | .cmp = max_cmp, | |
daa1d7a5 FW |
1282 | }; |
1283 | ||
39aeb52f | 1284 | static int switch_cmp(struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 FW |
1285 | { |
1286 | if (l->nb_atoms < r->nb_atoms) | |
1287 | return -1; | |
1288 | if (l->nb_atoms > r->nb_atoms) | |
1289 | return 1; | |
1290 | ||
1291 | return 0; | |
1292 | } | |
1293 | ||
1294 | static struct sort_dimension switch_sort_dimension = { | |
b5fae128 IM |
1295 | .name = "switch", |
1296 | .cmp = switch_cmp, | |
daa1d7a5 FW |
1297 | }; |
1298 | ||
39aeb52f | 1299 | static int runtime_cmp(struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 FW |
1300 | { |
1301 | if (l->total_runtime < r->total_runtime) | |
1302 | return -1; | |
1303 | if (l->total_runtime > r->total_runtime) | |
1304 | return 1; | |
1305 | ||
1306 | return 0; | |
1307 | } | |
1308 | ||
1309 | static struct sort_dimension runtime_sort_dimension = { | |
b5fae128 IM |
1310 | .name = "runtime", |
1311 | .cmp = runtime_cmp, | |
daa1d7a5 FW |
1312 | }; |
1313 | ||
1314 | static struct sort_dimension *available_sorts[] = { | |
1315 | &pid_sort_dimension, | |
1316 | &avg_sort_dimension, | |
1317 | &max_sort_dimension, | |
1318 | &switch_sort_dimension, | |
1319 | &runtime_sort_dimension, | |
1320 | }; | |
1321 | ||
1322 | #define NB_AVAILABLE_SORTS (int)(sizeof(available_sorts) / sizeof(struct sort_dimension *)) | |
1323 | ||
1324 | static LIST_HEAD(sort_list); | |
1325 | ||
cbef79a8 | 1326 | static int sort_dimension__add(const char *tok, struct list_head *list) |
daa1d7a5 FW |
1327 | { |
1328 | int i; | |
1329 | ||
1330 | for (i = 0; i < NB_AVAILABLE_SORTS; i++) { | |
1331 | if (!strcmp(available_sorts[i]->name, tok)) { | |
1332 | list_add_tail(&available_sorts[i]->list, list); | |
1333 | ||
1334 | return 0; | |
1335 | } | |
1336 | } | |
1337 | ||
1338 | return -1; | |
1339 | } | |
1340 | ||
1341 | static void setup_sorting(void); | |
1342 | ||
1343 | static void sort_lat(void) | |
1344 | { | |
1345 | struct rb_node *node; | |
1346 | ||
1347 | for (;;) { | |
39aeb52f | 1348 | struct work_atoms *data; |
b1ffe8f3 | 1349 | node = rb_first(&atom_root); |
daa1d7a5 FW |
1350 | if (!node) |
1351 | break; | |
1352 | ||
b1ffe8f3 | 1353 | rb_erase(node, &atom_root); |
39aeb52f | 1354 | data = rb_entry(node, struct work_atoms, node); |
b1ffe8f3 | 1355 | __thread_latency_insert(&sorted_atom_root, data, &sort_list); |
daa1d7a5 FW |
1356 | } |
1357 | } | |
1358 | ||
419ab0d6 FW |
1359 | static struct trace_sched_handler *trace_handler; |
1360 | ||
fbf94829 | 1361 | static void |
ee29be62 | 1362 | process_sched_wakeup_event(struct perf_tool *tool __used, |
419ab0d6 | 1363 | struct event *event, |
ee29be62 ACM |
1364 | struct perf_sample *sample, |
1365 | struct machine *machine, | |
1366 | struct thread *thread) | |
419ab0d6 | 1367 | { |
ee29be62 | 1368 | void *data = sample->raw_data; |
419ab0d6 FW |
1369 | struct trace_wakeup_event wakeup_event; |
1370 | ||
f48f669d | 1371 | FILL_COMMON_FIELDS(wakeup_event, event, data); |
419ab0d6 | 1372 | |
f48f669d XG |
1373 | FILL_ARRAY(wakeup_event, comm, event, data); |
1374 | FILL_FIELD(wakeup_event, pid, event, data); | |
1375 | FILL_FIELD(wakeup_event, prio, event, data); | |
1376 | FILL_FIELD(wakeup_event, success, event, data); | |
1377 | FILL_FIELD(wakeup_event, cpu, event, data); | |
419ab0d6 | 1378 | |
0ec04e16 | 1379 | if (trace_handler->wakeup_event) |
743eb868 | 1380 | trace_handler->wakeup_event(&wakeup_event, machine, event, |
ee29be62 | 1381 | sample->cpu, sample->time, thread); |
419ab0d6 FW |
1382 | } |
1383 | ||
c8a37751 IM |
1384 | /* |
1385 | * Track the current task - that way we can know whether there's any | |
1386 | * weird events, such as a task being switched away that is not current. | |
1387 | */ | |
40749d0f | 1388 | static int max_cpu; |
0ec04e16 | 1389 | |
c8a37751 IM |
1390 | static u32 curr_pid[MAX_CPUS] = { [0 ... MAX_CPUS-1] = -1 }; |
1391 | ||
0ec04e16 IM |
1392 | static struct thread *curr_thread[MAX_CPUS]; |
1393 | ||
1394 | static char next_shortname1 = 'A'; | |
1395 | static char next_shortname2 = '0'; | |
1396 | ||
1397 | static void | |
1398 | map_switch_event(struct trace_switch_event *switch_event, | |
743eb868 | 1399 | struct machine *machine, |
0ec04e16 IM |
1400 | struct event *event __used, |
1401 | int this_cpu, | |
1402 | u64 timestamp, | |
1403 | struct thread *thread __used) | |
1404 | { | |
fb7d0b3c | 1405 | struct thread *sched_out __used, *sched_in; |
0ec04e16 IM |
1406 | int new_shortname; |
1407 | u64 timestamp0; | |
1408 | s64 delta; | |
1409 | int cpu; | |
1410 | ||
1411 | BUG_ON(this_cpu >= MAX_CPUS || this_cpu < 0); | |
1412 | ||
1413 | if (this_cpu > max_cpu) | |
1414 | max_cpu = this_cpu; | |
1415 | ||
1416 | timestamp0 = cpu_last_switched[this_cpu]; | |
1417 | cpu_last_switched[this_cpu] = timestamp; | |
1418 | if (timestamp0) | |
1419 | delta = timestamp - timestamp0; | |
1420 | else | |
1421 | delta = 0; | |
1422 | ||
1423 | if (delta < 0) | |
9486aa38 | 1424 | die("hm, delta: %" PRIu64 " < 0 ?\n", delta); |
0ec04e16 IM |
1425 | |
1426 | ||
743eb868 ACM |
1427 | sched_out = machine__findnew_thread(machine, switch_event->prev_pid); |
1428 | sched_in = machine__findnew_thread(machine, switch_event->next_pid); | |
0ec04e16 IM |
1429 | |
1430 | curr_thread[this_cpu] = sched_in; | |
1431 | ||
1432 | printf(" "); | |
1433 | ||
1434 | new_shortname = 0; | |
1435 | if (!sched_in->shortname[0]) { | |
1436 | sched_in->shortname[0] = next_shortname1; | |
1437 | sched_in->shortname[1] = next_shortname2; | |
1438 | ||
1439 | if (next_shortname1 < 'Z') { | |
1440 | next_shortname1++; | |
1441 | } else { | |
1442 | next_shortname1='A'; | |
1443 | if (next_shortname2 < '9') { | |
1444 | next_shortname2++; | |
1445 | } else { | |
1446 | next_shortname2='0'; | |
1447 | } | |
1448 | } | |
1449 | new_shortname = 1; | |
1450 | } | |
1451 | ||
1452 | for (cpu = 0; cpu <= max_cpu; cpu++) { | |
1453 | if (cpu != this_cpu) | |
1454 | printf(" "); | |
1455 | else | |
1456 | printf("*"); | |
1457 | ||
1458 | if (curr_thread[cpu]) { | |
1459 | if (curr_thread[cpu]->pid) | |
1460 | printf("%2s ", curr_thread[cpu]->shortname); | |
1461 | else | |
1462 | printf(". "); | |
1463 | } else | |
1464 | printf(" "); | |
1465 | } | |
1466 | ||
1467 | printf(" %12.6f secs ", (double)timestamp/1e9); | |
1468 | if (new_shortname) { | |
1469 | printf("%s => %s:%d\n", | |
1470 | sched_in->shortname, sched_in->comm, sched_in->pid); | |
1471 | } else { | |
1472 | printf("\n"); | |
1473 | } | |
1474 | } | |
1475 | ||
419ab0d6 | 1476 | static void |
ee29be62 | 1477 | process_sched_switch_event(struct perf_tool *tool __used, |
419ab0d6 | 1478 | struct event *event, |
ee29be62 ACM |
1479 | struct perf_sample *sample, |
1480 | struct machine *machine, | |
1481 | struct thread *thread) | |
419ab0d6 | 1482 | { |
ee29be62 ACM |
1483 | int this_cpu = sample->cpu; |
1484 | void *data = sample->raw_data; | |
419ab0d6 FW |
1485 | struct trace_switch_event switch_event; |
1486 | ||
f48f669d | 1487 | FILL_COMMON_FIELDS(switch_event, event, data); |
419ab0d6 | 1488 | |
f48f669d XG |
1489 | FILL_ARRAY(switch_event, prev_comm, event, data); |
1490 | FILL_FIELD(switch_event, prev_pid, event, data); | |
1491 | FILL_FIELD(switch_event, prev_prio, event, data); | |
1492 | FILL_FIELD(switch_event, prev_state, event, data); | |
1493 | FILL_ARRAY(switch_event, next_comm, event, data); | |
1494 | FILL_FIELD(switch_event, next_pid, event, data); | |
1495 | FILL_FIELD(switch_event, next_prio, event, data); | |
419ab0d6 | 1496 | |
0ec04e16 | 1497 | if (curr_pid[this_cpu] != (u32)-1) { |
c8a37751 IM |
1498 | /* |
1499 | * Are we trying to switch away a PID that is | |
1500 | * not current? | |
1501 | */ | |
0ec04e16 | 1502 | if (curr_pid[this_cpu] != switch_event.prev_pid) |
c8a37751 IM |
1503 | nr_context_switch_bugs++; |
1504 | } | |
0ec04e16 | 1505 | if (trace_handler->switch_event) |
743eb868 | 1506 | trace_handler->switch_event(&switch_event, machine, event, |
ee29be62 | 1507 | this_cpu, sample->time, thread); |
c8a37751 | 1508 | |
0ec04e16 | 1509 | curr_pid[this_cpu] = switch_event.next_pid; |
419ab0d6 FW |
1510 | } |
1511 | ||
39aeb52f | 1512 | static void |
ee29be62 ACM |
1513 | process_sched_runtime_event(struct perf_tool *tool __used, |
1514 | struct event *event, | |
1515 | struct perf_sample *sample, | |
1516 | struct machine *machine, | |
1517 | struct thread *thread) | |
39aeb52f | 1518 | { |
ee29be62 | 1519 | void *data = sample->raw_data; |
39aeb52f | 1520 | struct trace_runtime_event runtime_event; |
1521 | ||
f48f669d XG |
1522 | FILL_ARRAY(runtime_event, comm, event, data); |
1523 | FILL_FIELD(runtime_event, pid, event, data); | |
1524 | FILL_FIELD(runtime_event, runtime, event, data); | |
1525 | FILL_FIELD(runtime_event, vruntime, event, data); | |
39aeb52f | 1526 | |
0ec04e16 | 1527 | if (trace_handler->runtime_event) |
ee29be62 ACM |
1528 | trace_handler->runtime_event(&runtime_event, machine, event, |
1529 | sample->cpu, sample->time, thread); | |
39aeb52f | 1530 | } |
1531 | ||
419ab0d6 | 1532 | static void |
ee29be62 | 1533 | process_sched_fork_event(struct perf_tool *tool __used, |
419ab0d6 | 1534 | struct event *event, |
ee29be62 ACM |
1535 | struct perf_sample *sample, |
1536 | struct machine *machine __used, | |
1537 | struct thread *thread) | |
fbf94829 | 1538 | { |
ee29be62 | 1539 | void *data = sample->raw_data; |
46538818 FW |
1540 | struct trace_fork_event fork_event; |
1541 | ||
f48f669d | 1542 | FILL_COMMON_FIELDS(fork_event, event, data); |
46538818 | 1543 | |
f48f669d XG |
1544 | FILL_ARRAY(fork_event, parent_comm, event, data); |
1545 | FILL_FIELD(fork_event, parent_pid, event, data); | |
1546 | FILL_ARRAY(fork_event, child_comm, event, data); | |
1547 | FILL_FIELD(fork_event, child_pid, event, data); | |
46538818 | 1548 | |
0ec04e16 | 1549 | if (trace_handler->fork_event) |
b3165f41 | 1550 | trace_handler->fork_event(&fork_event, event, |
ee29be62 | 1551 | sample->cpu, sample->time, thread); |
fbf94829 IM |
1552 | } |
1553 | ||
419ab0d6 | 1554 | static void |
ee29be62 ACM |
1555 | process_sched_exit_event(struct perf_tool *tool __used, |
1556 | struct event *event, | |
1557 | struct perf_sample *sample __used, | |
1558 | struct machine *machine __used, | |
419ab0d6 | 1559 | struct thread *thread __used) |
fbf94829 | 1560 | { |
ad236fd2 IM |
1561 | if (verbose) |
1562 | printf("sched_exit event %p\n", event); | |
ec156764 IM |
1563 | } |
1564 | ||
55ffb7a6 | 1565 | static void |
ee29be62 ACM |
1566 | process_sched_migrate_task_event(struct perf_tool *tool __used, |
1567 | struct event *event, | |
1568 | struct perf_sample *sample, | |
1569 | struct machine *machine, | |
1570 | struct thread *thread) | |
55ffb7a6 | 1571 | { |
ee29be62 | 1572 | void *data = sample->raw_data; |
55ffb7a6 MG |
1573 | struct trace_migrate_task_event migrate_task_event; |
1574 | ||
f48f669d | 1575 | FILL_COMMON_FIELDS(migrate_task_event, event, data); |
55ffb7a6 | 1576 | |
f48f669d XG |
1577 | FILL_ARRAY(migrate_task_event, comm, event, data); |
1578 | FILL_FIELD(migrate_task_event, pid, event, data); | |
1579 | FILL_FIELD(migrate_task_event, prio, event, data); | |
1580 | FILL_FIELD(migrate_task_event, cpu, event, data); | |
55ffb7a6 MG |
1581 | |
1582 | if (trace_handler->migrate_task_event) | |
743eb868 | 1583 | trace_handler->migrate_task_event(&migrate_task_event, machine, |
ee29be62 ACM |
1584 | event, sample->cpu, |
1585 | sample->time, thread); | |
55ffb7a6 MG |
1586 | } |
1587 | ||
ee29be62 ACM |
1588 | typedef void (*tracepoint_handler)(struct perf_tool *tool, struct event *event, |
1589 | struct perf_sample *sample, | |
1590 | struct machine *machine, | |
1591 | struct thread *thread); | |
ec156764 | 1592 | |
ee29be62 ACM |
1593 | static int perf_sched__process_tracepoint_sample(struct perf_tool *tool, |
1594 | union perf_event *event __used, | |
1595 | struct perf_sample *sample, | |
1596 | struct perf_evsel *evsel, | |
1597 | struct machine *machine) | |
0a02ad93 | 1598 | { |
ee29be62 | 1599 | struct thread *thread = machine__findnew_thread(machine, sample->pid); |
0a02ad93 | 1600 | |
0a02ad93 | 1601 | if (thread == NULL) { |
ee29be62 ACM |
1602 | pr_debug("problem processing %s event, skipping it.\n", |
1603 | evsel->name); | |
0a02ad93 IM |
1604 | return -1; |
1605 | } | |
1606 | ||
ee29be62 ACM |
1607 | evsel->hists.stats.total_period += sample->period; |
1608 | hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE); | |
f39cdf25 | 1609 | |
ee29be62 ACM |
1610 | if (evsel->handler.func != NULL) { |
1611 | tracepoint_handler f = evsel->handler.func; | |
55ffb7a6 | 1612 | |
ee29be62 ACM |
1613 | if (evsel->handler.data == NULL) |
1614 | evsel->handler.data = trace_find_event(evsel->attr.config); | |
1615 | ||
1616 | f(tool, evsel->handler.data, sample, machine, thread); | |
1617 | } | |
0a02ad93 IM |
1618 | |
1619 | return 0; | |
1620 | } | |
1621 | ||
45694aa7 | 1622 | static struct perf_tool perf_sched = { |
ee29be62 | 1623 | .sample = perf_sched__process_tracepoint_sample, |
8115d60c ACM |
1624 | .comm = perf_event__process_comm, |
1625 | .lost = perf_event__process_lost, | |
1626 | .fork = perf_event__process_task, | |
a64eae70 | 1627 | .ordered_samples = true, |
016e92fb FW |
1628 | }; |
1629 | ||
4c09bafa | 1630 | static void read_events(bool destroy, struct perf_session **psession) |
0a02ad93 | 1631 | { |
d549c769 | 1632 | int err = -EINVAL; |
ee29be62 ACM |
1633 | const struct perf_evsel_str_handler handlers[] = { |
1634 | { "sched:sched_switch", process_sched_switch_event, }, | |
1635 | { "sched:sched_stat_runtime", process_sched_runtime_event, }, | |
1636 | { "sched:sched_wakeup", process_sched_wakeup_event, }, | |
1637 | { "sched:sched_wakeup_new", process_sched_wakeup_event, }, | |
1638 | { "sched:sched_process_fork", process_sched_fork_event, }, | |
1639 | { "sched:sched_process_exit", process_sched_exit_event, }, | |
1640 | { "sched:sched_migrate_task", process_sched_migrate_task_event, }, | |
1641 | }; | |
21ef97f0 | 1642 | struct perf_session *session = perf_session__new(input_name, O_RDONLY, |
45694aa7 | 1643 | 0, false, &perf_sched); |
94c744b6 | 1644 | if (session == NULL) |
4c09bafa | 1645 | die("No Memory"); |
94c744b6 | 1646 | |
ee29be62 ACM |
1647 | err = perf_evlist__set_tracepoints_handlers_array(session->evlist, handlers); |
1648 | assert(err == 0); | |
1649 | ||
cee75ac7 | 1650 | if (perf_session__has_traces(session, "record -R")) { |
45694aa7 | 1651 | err = perf_session__process_events(session, &perf_sched); |
4c09bafa JO |
1652 | if (err) |
1653 | die("Failed to process events, error %d", err); | |
1654 | ||
cee75ac7 ACM |
1655 | nr_events = session->hists.stats.nr_events[0]; |
1656 | nr_lost_events = session->hists.stats.total_lost; | |
1657 | nr_lost_chunks = session->hists.stats.nr_events[PERF_RECORD_LOST]; | |
1658 | } | |
d549c769 | 1659 | |
4c09bafa JO |
1660 | if (destroy) |
1661 | perf_session__delete(session); | |
1662 | ||
1663 | if (psession) | |
1664 | *psession = session; | |
0a02ad93 IM |
1665 | } |
1666 | ||
0ec04e16 IM |
1667 | static void print_bad_events(void) |
1668 | { | |
1669 | if (nr_unordered_timestamps && nr_timestamps) { | |
1670 | printf(" INFO: %.3f%% unordered timestamps (%ld out of %ld)\n", | |
1671 | (double)nr_unordered_timestamps/(double)nr_timestamps*100.0, | |
1672 | nr_unordered_timestamps, nr_timestamps); | |
1673 | } | |
1674 | if (nr_lost_events && nr_events) { | |
1675 | printf(" INFO: %.3f%% lost events (%ld out of %ld, in %ld chunks)\n", | |
1676 | (double)nr_lost_events/(double)nr_events*100.0, | |
1677 | nr_lost_events, nr_events, nr_lost_chunks); | |
1678 | } | |
1679 | if (nr_state_machine_bugs && nr_timestamps) { | |
1680 | printf(" INFO: %.3f%% state machine bugs (%ld out of %ld)", | |
1681 | (double)nr_state_machine_bugs/(double)nr_timestamps*100.0, | |
1682 | nr_state_machine_bugs, nr_timestamps); | |
1683 | if (nr_lost_events) | |
1684 | printf(" (due to lost events?)"); | |
1685 | printf("\n"); | |
1686 | } | |
1687 | if (nr_context_switch_bugs && nr_timestamps) { | |
1688 | printf(" INFO: %.3f%% context switch bugs (%ld out of %ld)", | |
1689 | (double)nr_context_switch_bugs/(double)nr_timestamps*100.0, | |
1690 | nr_context_switch_bugs, nr_timestamps); | |
1691 | if (nr_lost_events) | |
1692 | printf(" (due to lost events?)"); | |
1693 | printf("\n"); | |
1694 | } | |
1695 | } | |
1696 | ||
1697 | static void __cmd_lat(void) | |
1698 | { | |
1699 | struct rb_node *next; | |
4c09bafa | 1700 | struct perf_session *session; |
0ec04e16 IM |
1701 | |
1702 | setup_pager(); | |
4c09bafa | 1703 | read_events(false, &session); |
0ec04e16 IM |
1704 | sort_lat(); |
1705 | ||
3786310a FW |
1706 | printf("\n ---------------------------------------------------------------------------------------------------------------\n"); |
1707 | printf(" Task | Runtime ms | Switches | Average delay ms | Maximum delay ms | Maximum delay at |\n"); | |
1708 | printf(" ---------------------------------------------------------------------------------------------------------------\n"); | |
0ec04e16 IM |
1709 | |
1710 | next = rb_first(&sorted_atom_root); | |
1711 | ||
1712 | while (next) { | |
1713 | struct work_atoms *work_list; | |
1714 | ||
1715 | work_list = rb_entry(next, struct work_atoms, node); | |
1716 | output_lat_thread(work_list); | |
1717 | next = rb_next(next); | |
1718 | } | |
1719 | ||
1720 | printf(" -----------------------------------------------------------------------------------------\n"); | |
9486aa38 | 1721 | printf(" TOTAL: |%11.3f ms |%9" PRIu64 " |\n", |
0ec04e16 IM |
1722 | (double)all_runtime/1e6, all_count); |
1723 | ||
1724 | printf(" ---------------------------------------------------\n"); | |
1725 | ||
1726 | print_bad_events(); | |
1727 | printf("\n"); | |
1728 | ||
4c09bafa | 1729 | perf_session__delete(session); |
0ec04e16 IM |
1730 | } |
1731 | ||
1732 | static struct trace_sched_handler map_ops = { | |
1733 | .wakeup_event = NULL, | |
1734 | .switch_event = map_switch_event, | |
1735 | .runtime_event = NULL, | |
1736 | .fork_event = NULL, | |
1737 | }; | |
1738 | ||
1739 | static void __cmd_map(void) | |
1740 | { | |
40749d0f IM |
1741 | max_cpu = sysconf(_SC_NPROCESSORS_CONF); |
1742 | ||
0ec04e16 | 1743 | setup_pager(); |
4c09bafa | 1744 | read_events(true, NULL); |
0ec04e16 IM |
1745 | print_bad_events(); |
1746 | } | |
1747 | ||
1748 | static void __cmd_replay(void) | |
1749 | { | |
1750 | unsigned long i; | |
1751 | ||
1752 | calibrate_run_measurement_overhead(); | |
1753 | calibrate_sleep_measurement_overhead(); | |
1754 | ||
1755 | test_calibrations(); | |
1756 | ||
4c09bafa | 1757 | read_events(true, NULL); |
0ec04e16 IM |
1758 | |
1759 | printf("nr_run_events: %ld\n", nr_run_events); | |
1760 | printf("nr_sleep_events: %ld\n", nr_sleep_events); | |
1761 | printf("nr_wakeup_events: %ld\n", nr_wakeup_events); | |
1762 | ||
1763 | if (targetless_wakeups) | |
1764 | printf("target-less wakeups: %ld\n", targetless_wakeups); | |
1765 | if (multitarget_wakeups) | |
1766 | printf("multi-target wakeups: %ld\n", multitarget_wakeups); | |
1767 | if (nr_run_events_optimized) | |
1768 | printf("run atoms optimized: %ld\n", | |
1769 | nr_run_events_optimized); | |
1770 | ||
1771 | print_task_traces(); | |
1772 | add_cross_task_wakeups(); | |
1773 | ||
1774 | create_tasks(); | |
1775 | printf("------------------------------------------------------------\n"); | |
1776 | for (i = 0; i < replay_repeat; i++) | |
1777 | run_one_test(); | |
1778 | } | |
1779 | ||
1780 | ||
46f392c9 | 1781 | static const char * const sched_usage[] = { |
580cabed | 1782 | "perf sched [<options>] {record|latency|map|replay|script}", |
0a02ad93 IM |
1783 | NULL |
1784 | }; | |
1785 | ||
f2858d8a | 1786 | static const struct option sched_options[] = { |
4b77a729 MG |
1787 | OPT_STRING('i', "input", &input_name, "file", |
1788 | "input file name"), | |
c0555642 | 1789 | OPT_INCR('v', "verbose", &verbose, |
f2858d8a | 1790 | "be more verbose (show symbol address, etc)"), |
0a02ad93 IM |
1791 | OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, |
1792 | "dump raw trace in ASCII"), | |
f2858d8a IM |
1793 | OPT_END() |
1794 | }; | |
1795 | ||
1796 | static const char * const latency_usage[] = { | |
1797 | "perf sched latency [<options>]", | |
1798 | NULL | |
1799 | }; | |
1800 | ||
1801 | static const struct option latency_options[] = { | |
daa1d7a5 FW |
1802 | OPT_STRING('s', "sort", &sort_order, "key[,key2...]", |
1803 | "sort by key(s): runtime, switch, avg, max"), | |
c0555642 | 1804 | OPT_INCR('v', "verbose", &verbose, |
0a02ad93 | 1805 | "be more verbose (show symbol address, etc)"), |
55ffb7a6 MG |
1806 | OPT_INTEGER('C', "CPU", &profile_cpu, |
1807 | "CPU to profile on"), | |
f2858d8a IM |
1808 | OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, |
1809 | "dump raw trace in ASCII"), | |
1810 | OPT_END() | |
1811 | }; | |
1812 | ||
1813 | static const char * const replay_usage[] = { | |
1814 | "perf sched replay [<options>]", | |
1815 | NULL | |
1816 | }; | |
1817 | ||
1818 | static const struct option replay_options[] = { | |
1967936d ACM |
1819 | OPT_UINTEGER('r', "repeat", &replay_repeat, |
1820 | "repeat the workload replay N times (-1: infinite)"), | |
c0555642 | 1821 | OPT_INCR('v', "verbose", &verbose, |
f2858d8a IM |
1822 | "be more verbose (show symbol address, etc)"), |
1823 | OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, | |
1824 | "dump raw trace in ASCII"), | |
0a02ad93 IM |
1825 | OPT_END() |
1826 | }; | |
1827 | ||
daa1d7a5 FW |
1828 | static void setup_sorting(void) |
1829 | { | |
1830 | char *tmp, *tok, *str = strdup(sort_order); | |
1831 | ||
1832 | for (tok = strtok_r(str, ", ", &tmp); | |
1833 | tok; tok = strtok_r(NULL, ", ", &tmp)) { | |
1834 | if (sort_dimension__add(tok, &sort_list) < 0) { | |
1835 | error("Unknown --sort key: `%s'", tok); | |
f2858d8a | 1836 | usage_with_options(latency_usage, latency_options); |
daa1d7a5 FW |
1837 | } |
1838 | } | |
1839 | ||
1840 | free(str); | |
1841 | ||
cbef79a8 | 1842 | sort_dimension__add("pid", &cmp_pid); |
daa1d7a5 FW |
1843 | } |
1844 | ||
1fc35b29 IM |
1845 | static const char *record_args[] = { |
1846 | "record", | |
1847 | "-a", | |
1848 | "-R", | |
ea57c4f5 | 1849 | "-f", |
dc02bf71 | 1850 | "-m", "1024", |
1fc35b29 | 1851 | "-c", "1", |
9710118b SE |
1852 | "-e", "sched:sched_switch", |
1853 | "-e", "sched:sched_stat_wait", | |
1854 | "-e", "sched:sched_stat_sleep", | |
1855 | "-e", "sched:sched_stat_iowait", | |
1856 | "-e", "sched:sched_stat_runtime", | |
1857 | "-e", "sched:sched_process_exit", | |
1858 | "-e", "sched:sched_process_fork", | |
1859 | "-e", "sched:sched_wakeup", | |
1860 | "-e", "sched:sched_migrate_task", | |
1fc35b29 IM |
1861 | }; |
1862 | ||
1863 | static int __cmd_record(int argc, const char **argv) | |
1864 | { | |
1865 | unsigned int rec_argc, i, j; | |
1866 | const char **rec_argv; | |
1867 | ||
1868 | rec_argc = ARRAY_SIZE(record_args) + argc - 1; | |
1869 | rec_argv = calloc(rec_argc + 1, sizeof(char *)); | |
1870 | ||
e462dc55 | 1871 | if (rec_argv == NULL) |
ce47dc56 CS |
1872 | return -ENOMEM; |
1873 | ||
1fc35b29 IM |
1874 | for (i = 0; i < ARRAY_SIZE(record_args); i++) |
1875 | rec_argv[i] = strdup(record_args[i]); | |
1876 | ||
1877 | for (j = 1; j < (unsigned int)argc; j++, i++) | |
1878 | rec_argv[i] = argv[j]; | |
1879 | ||
1880 | BUG_ON(i != rec_argc); | |
1881 | ||
1882 | return cmd_record(i, rec_argv, NULL); | |
1883 | } | |
1884 | ||
0a02ad93 IM |
1885 | int cmd_sched(int argc, const char **argv, const char *prefix __used) |
1886 | { | |
f2858d8a IM |
1887 | argc = parse_options(argc, argv, sched_options, sched_usage, |
1888 | PARSE_OPT_STOP_AT_NON_OPTION); | |
1889 | if (!argc) | |
1890 | usage_with_options(sched_usage, sched_options); | |
0a02ad93 | 1891 | |
c0777c5a | 1892 | /* |
133dc4c3 | 1893 | * Aliased to 'perf script' for now: |
c0777c5a | 1894 | */ |
133dc4c3 IM |
1895 | if (!strcmp(argv[0], "script")) |
1896 | return cmd_script(argc, argv, prefix); | |
c0777c5a | 1897 | |
75be6cf4 | 1898 | symbol__init(); |
1fc35b29 IM |
1899 | if (!strncmp(argv[0], "rec", 3)) { |
1900 | return __cmd_record(argc, argv); | |
1901 | } else if (!strncmp(argv[0], "lat", 3)) { | |
cdce9d73 | 1902 | trace_handler = &lat_ops; |
f2858d8a IM |
1903 | if (argc > 1) { |
1904 | argc = parse_options(argc, argv, latency_options, latency_usage, 0); | |
1905 | if (argc) | |
1906 | usage_with_options(latency_usage, latency_options); | |
f2858d8a | 1907 | } |
b5fae128 | 1908 | setup_sorting(); |
46f392c9 | 1909 | __cmd_lat(); |
0ec04e16 IM |
1910 | } else if (!strcmp(argv[0], "map")) { |
1911 | trace_handler = &map_ops; | |
1912 | setup_sorting(); | |
1913 | __cmd_map(); | |
f2858d8a IM |
1914 | } else if (!strncmp(argv[0], "rep", 3)) { |
1915 | trace_handler = &replay_ops; | |
1916 | if (argc) { | |
1917 | argc = parse_options(argc, argv, replay_options, replay_usage, 0); | |
1918 | if (argc) | |
1919 | usage_with_options(replay_usage, replay_options); | |
1920 | } | |
1921 | __cmd_replay(); | |
1922 | } else { | |
1923 | usage_with_options(sched_usage, sched_options); | |
1924 | } | |
1925 | ||
ec156764 | 1926 | return 0; |
0a02ad93 | 1927 | } |