Commit | Line | Data |
---|---|---|
0a02ad93 | 1 | #include "builtin.h" |
b1ffe8f3 | 2 | #include "perf.h" |
0a02ad93 IM |
3 | |
4 | #include "util/util.h" | |
ee29be62 | 5 | #include "util/evlist.h" |
0a02ad93 | 6 | #include "util/cache.h" |
e3f42609 | 7 | #include "util/evsel.h" |
0a02ad93 IM |
8 | #include "util/symbol.h" |
9 | #include "util/thread.h" | |
10 | #include "util/header.h" | |
94c744b6 | 11 | #include "util/session.h" |
45694aa7 | 12 | #include "util/tool.h" |
57480d2c | 13 | #include "util/cloexec.h" |
0a02ad93 IM |
14 | |
15 | #include "util/parse-options.h" | |
b1ffe8f3 | 16 | #include "util/trace-event.h" |
0a02ad93 | 17 | |
0a02ad93 IM |
18 | #include "util/debug.h" |
19 | ||
b1ffe8f3 | 20 | #include <sys/prctl.h> |
7b78f136 | 21 | #include <sys/resource.h> |
0a02ad93 | 22 | |
b1ffe8f3 IM |
23 | #include <semaphore.h> |
24 | #include <pthread.h> | |
25 | #include <math.h> | |
cb06ac25 | 26 | #include <api/fs/fs.h> |
419ab0d6 | 27 | |
b1ffe8f3 IM |
28 | #define PR_SET_NAME 15 /* Set process name */ |
29 | #define MAX_CPUS 4096 | |
b1ffe8f3 IM |
30 | #define COMM_LEN 20 |
31 | #define SYM_LEN 129 | |
a35e27d0 | 32 | #define MAX_PID 1024000 |
ec156764 | 33 | |
39aeb52f | 34 | struct sched_atom; |
ec156764 | 35 | |
b1ffe8f3 IM |
36 | struct task_desc { |
37 | unsigned long nr; | |
38 | unsigned long pid; | |
39 | char comm[COMM_LEN]; | |
ec156764 | 40 | |
b1ffe8f3 IM |
41 | unsigned long nr_events; |
42 | unsigned long curr_event; | |
39aeb52f | 43 | struct sched_atom **atoms; |
b1ffe8f3 IM |
44 | |
45 | pthread_t thread; | |
46 | sem_t sleep_sem; | |
ec156764 | 47 | |
b1ffe8f3 IM |
48 | sem_t ready_for_work; |
49 | sem_t work_done_sem; | |
50 | ||
51 | u64 cpu_usage; | |
52 | }; | |
53 | ||
54 | enum sched_event_type { | |
55 | SCHED_EVENT_RUN, | |
56 | SCHED_EVENT_SLEEP, | |
57 | SCHED_EVENT_WAKEUP, | |
55ffb7a6 | 58 | SCHED_EVENT_MIGRATION, |
b1ffe8f3 IM |
59 | }; |
60 | ||
39aeb52f | 61 | struct sched_atom { |
b1ffe8f3 | 62 | enum sched_event_type type; |
eed05fe7 | 63 | int specific_wait; |
b1ffe8f3 IM |
64 | u64 timestamp; |
65 | u64 duration; | |
66 | unsigned long nr; | |
b1ffe8f3 IM |
67 | sem_t *wait_sem; |
68 | struct task_desc *wakee; | |
69 | }; | |
70 | ||
e936e8e4 | 71 | #define TASK_STATE_TO_CHAR_STR "RSDTtZXxKWP" |
b1ffe8f3 IM |
72 | |
73 | enum thread_state { | |
74 | THREAD_SLEEPING = 0, | |
75 | THREAD_WAIT_CPU, | |
76 | THREAD_SCHED_IN, | |
77 | THREAD_IGNORE | |
78 | }; | |
79 | ||
80 | struct work_atom { | |
81 | struct list_head list; | |
82 | enum thread_state state; | |
aa1ab9d2 | 83 | u64 sched_out_time; |
b1ffe8f3 IM |
84 | u64 wake_up_time; |
85 | u64 sched_in_time; | |
86 | u64 runtime; | |
87 | }; | |
88 | ||
39aeb52f | 89 | struct work_atoms { |
90 | struct list_head work_list; | |
b1ffe8f3 IM |
91 | struct thread *thread; |
92 | struct rb_node node; | |
93 | u64 max_lat; | |
3786310a | 94 | u64 max_lat_at; |
b1ffe8f3 IM |
95 | u64 total_lat; |
96 | u64 nb_atoms; | |
97 | u64 total_runtime; | |
2f80dd44 | 98 | int num_merged; |
b1ffe8f3 IM |
99 | }; |
100 | ||
39aeb52f | 101 | typedef int (*sort_fn_t)(struct work_atoms *, struct work_atoms *); |
b1ffe8f3 | 102 | |
9ec3f4e4 | 103 | struct perf_sched; |
0e9b07e5 | 104 | |
9ec3f4e4 ACM |
105 | struct trace_sched_handler { |
106 | int (*switch_event)(struct perf_sched *sched, struct perf_evsel *evsel, | |
107 | struct perf_sample *sample, struct machine *machine); | |
0e9b07e5 | 108 | |
9ec3f4e4 ACM |
109 | int (*runtime_event)(struct perf_sched *sched, struct perf_evsel *evsel, |
110 | struct perf_sample *sample, struct machine *machine); | |
0e9b07e5 | 111 | |
9ec3f4e4 ACM |
112 | int (*wakeup_event)(struct perf_sched *sched, struct perf_evsel *evsel, |
113 | struct perf_sample *sample, struct machine *machine); | |
0e9b07e5 | 114 | |
cb627505 DA |
115 | /* PERF_RECORD_FORK event, not sched_process_fork tracepoint */ |
116 | int (*fork_event)(struct perf_sched *sched, union perf_event *event, | |
117 | struct machine *machine); | |
0e9b07e5 ACM |
118 | |
119 | int (*migrate_task_event)(struct perf_sched *sched, | |
9ec3f4e4 ACM |
120 | struct perf_evsel *evsel, |
121 | struct perf_sample *sample, | |
122 | struct machine *machine); | |
0e9b07e5 ACM |
123 | }; |
124 | ||
125 | struct perf_sched { | |
126 | struct perf_tool tool; | |
0e9b07e5 ACM |
127 | const char *sort_order; |
128 | unsigned long nr_tasks; | |
cb06ac25 | 129 | struct task_desc **pid_to_task; |
0e9b07e5 ACM |
130 | struct task_desc **tasks; |
131 | const struct trace_sched_handler *tp_handler; | |
132 | pthread_mutex_t start_work_mutex; | |
133 | pthread_mutex_t work_done_wait_mutex; | |
134 | int profile_cpu; | |
135 | /* | |
136 | * Track the current task - that way we can know whether there's any | |
137 | * weird events, such as a task being switched away that is not current. | |
138 | */ | |
139 | int max_cpu; | |
140 | u32 curr_pid[MAX_CPUS]; | |
141 | struct thread *curr_thread[MAX_CPUS]; | |
142 | char next_shortname1; | |
143 | char next_shortname2; | |
144 | unsigned int replay_repeat; | |
145 | unsigned long nr_run_events; | |
146 | unsigned long nr_sleep_events; | |
147 | unsigned long nr_wakeup_events; | |
148 | unsigned long nr_sleep_corrections; | |
149 | unsigned long nr_run_events_optimized; | |
150 | unsigned long targetless_wakeups; | |
151 | unsigned long multitarget_wakeups; | |
152 | unsigned long nr_runs; | |
153 | unsigned long nr_timestamps; | |
154 | unsigned long nr_unordered_timestamps; | |
0e9b07e5 ACM |
155 | unsigned long nr_context_switch_bugs; |
156 | unsigned long nr_events; | |
157 | unsigned long nr_lost_chunks; | |
158 | unsigned long nr_lost_events; | |
159 | u64 run_measurement_overhead; | |
160 | u64 sleep_measurement_overhead; | |
161 | u64 start_time; | |
162 | u64 cpu_usage; | |
163 | u64 runavg_cpu_usage; | |
164 | u64 parent_cpu_usage; | |
165 | u64 runavg_parent_cpu_usage; | |
166 | u64 sum_runtime; | |
167 | u64 sum_fluct; | |
168 | u64 run_avg; | |
169 | u64 all_runtime; | |
170 | u64 all_count; | |
171 | u64 cpu_last_switched[MAX_CPUS]; | |
2f80dd44 | 172 | struct rb_root atom_root, sorted_atom_root, merged_atom_root; |
0e9b07e5 | 173 | struct list_head sort_list, cmp_pid; |
939cda52 | 174 | bool force; |
2f80dd44 | 175 | bool skip_merge; |
0e9b07e5 | 176 | }; |
b1ffe8f3 IM |
177 | |
178 | static u64 get_nsecs(void) | |
ec156764 IM |
179 | { |
180 | struct timespec ts; | |
181 | ||
182 | clock_gettime(CLOCK_MONOTONIC, &ts); | |
183 | ||
184 | return ts.tv_sec * 1000000000ULL + ts.tv_nsec; | |
185 | } | |
186 | ||
0e9b07e5 | 187 | static void burn_nsecs(struct perf_sched *sched, u64 nsecs) |
ec156764 | 188 | { |
b1ffe8f3 | 189 | u64 T0 = get_nsecs(), T1; |
ec156764 IM |
190 | |
191 | do { | |
192 | T1 = get_nsecs(); | |
0e9b07e5 | 193 | } while (T1 + sched->run_measurement_overhead < T0 + nsecs); |
ec156764 IM |
194 | } |
195 | ||
b1ffe8f3 | 196 | static void sleep_nsecs(u64 nsecs) |
ec156764 IM |
197 | { |
198 | struct timespec ts; | |
199 | ||
200 | ts.tv_nsec = nsecs % 999999999; | |
201 | ts.tv_sec = nsecs / 999999999; | |
202 | ||
203 | nanosleep(&ts, NULL); | |
204 | } | |
205 | ||
0e9b07e5 | 206 | static void calibrate_run_measurement_overhead(struct perf_sched *sched) |
ec156764 | 207 | { |
b1ffe8f3 | 208 | u64 T0, T1, delta, min_delta = 1000000000ULL; |
ec156764 IM |
209 | int i; |
210 | ||
211 | for (i = 0; i < 10; i++) { | |
212 | T0 = get_nsecs(); | |
0e9b07e5 | 213 | burn_nsecs(sched, 0); |
ec156764 IM |
214 | T1 = get_nsecs(); |
215 | delta = T1-T0; | |
216 | min_delta = min(min_delta, delta); | |
217 | } | |
0e9b07e5 | 218 | sched->run_measurement_overhead = min_delta; |
ec156764 | 219 | |
9486aa38 | 220 | printf("run measurement overhead: %" PRIu64 " nsecs\n", min_delta); |
ec156764 IM |
221 | } |
222 | ||
0e9b07e5 | 223 | static void calibrate_sleep_measurement_overhead(struct perf_sched *sched) |
ec156764 | 224 | { |
b1ffe8f3 | 225 | u64 T0, T1, delta, min_delta = 1000000000ULL; |
ec156764 IM |
226 | int i; |
227 | ||
228 | for (i = 0; i < 10; i++) { | |
229 | T0 = get_nsecs(); | |
230 | sleep_nsecs(10000); | |
231 | T1 = get_nsecs(); | |
232 | delta = T1-T0; | |
233 | min_delta = min(min_delta, delta); | |
234 | } | |
235 | min_delta -= 10000; | |
0e9b07e5 | 236 | sched->sleep_measurement_overhead = min_delta; |
ec156764 | 237 | |
9486aa38 | 238 | printf("sleep measurement overhead: %" PRIu64 " nsecs\n", min_delta); |
ec156764 IM |
239 | } |
240 | ||
39aeb52f | 241 | static struct sched_atom * |
b1ffe8f3 | 242 | get_new_event(struct task_desc *task, u64 timestamp) |
ec156764 | 243 | { |
36479484 | 244 | struct sched_atom *event = zalloc(sizeof(*event)); |
ec156764 IM |
245 | unsigned long idx = task->nr_events; |
246 | size_t size; | |
247 | ||
248 | event->timestamp = timestamp; | |
249 | event->nr = idx; | |
250 | ||
251 | task->nr_events++; | |
39aeb52f | 252 | size = sizeof(struct sched_atom *) * task->nr_events; |
253 | task->atoms = realloc(task->atoms, size); | |
254 | BUG_ON(!task->atoms); | |
ec156764 | 255 | |
39aeb52f | 256 | task->atoms[idx] = event; |
ec156764 IM |
257 | |
258 | return event; | |
259 | } | |
260 | ||
39aeb52f | 261 | static struct sched_atom *last_event(struct task_desc *task) |
ec156764 IM |
262 | { |
263 | if (!task->nr_events) | |
264 | return NULL; | |
265 | ||
39aeb52f | 266 | return task->atoms[task->nr_events - 1]; |
ec156764 IM |
267 | } |
268 | ||
0e9b07e5 ACM |
269 | static void add_sched_event_run(struct perf_sched *sched, struct task_desc *task, |
270 | u64 timestamp, u64 duration) | |
ec156764 | 271 | { |
39aeb52f | 272 | struct sched_atom *event, *curr_event = last_event(task); |
ec156764 IM |
273 | |
274 | /* | |
fbf94829 IM |
275 | * optimize an existing RUN event by merging this one |
276 | * to it: | |
277 | */ | |
ec156764 | 278 | if (curr_event && curr_event->type == SCHED_EVENT_RUN) { |
0e9b07e5 | 279 | sched->nr_run_events_optimized++; |
ec156764 IM |
280 | curr_event->duration += duration; |
281 | return; | |
282 | } | |
283 | ||
284 | event = get_new_event(task, timestamp); | |
285 | ||
286 | event->type = SCHED_EVENT_RUN; | |
287 | event->duration = duration; | |
288 | ||
0e9b07e5 | 289 | sched->nr_run_events++; |
ec156764 IM |
290 | } |
291 | ||
0e9b07e5 ACM |
292 | static void add_sched_event_wakeup(struct perf_sched *sched, struct task_desc *task, |
293 | u64 timestamp, struct task_desc *wakee) | |
ec156764 | 294 | { |
39aeb52f | 295 | struct sched_atom *event, *wakee_event; |
ec156764 IM |
296 | |
297 | event = get_new_event(task, timestamp); | |
298 | event->type = SCHED_EVENT_WAKEUP; | |
299 | event->wakee = wakee; | |
300 | ||
301 | wakee_event = last_event(wakee); | |
302 | if (!wakee_event || wakee_event->type != SCHED_EVENT_SLEEP) { | |
0e9b07e5 | 303 | sched->targetless_wakeups++; |
ec156764 IM |
304 | return; |
305 | } | |
306 | if (wakee_event->wait_sem) { | |
0e9b07e5 | 307 | sched->multitarget_wakeups++; |
ec156764 IM |
308 | return; |
309 | } | |
310 | ||
36479484 | 311 | wakee_event->wait_sem = zalloc(sizeof(*wakee_event->wait_sem)); |
ec156764 IM |
312 | sem_init(wakee_event->wait_sem, 0, 0); |
313 | wakee_event->specific_wait = 1; | |
314 | event->wait_sem = wakee_event->wait_sem; | |
315 | ||
0e9b07e5 | 316 | sched->nr_wakeup_events++; |
ec156764 IM |
317 | } |
318 | ||
0e9b07e5 ACM |
319 | static void add_sched_event_sleep(struct perf_sched *sched, struct task_desc *task, |
320 | u64 timestamp, u64 task_state __maybe_unused) | |
ec156764 | 321 | { |
39aeb52f | 322 | struct sched_atom *event = get_new_event(task, timestamp); |
ec156764 IM |
323 | |
324 | event->type = SCHED_EVENT_SLEEP; | |
325 | ||
0e9b07e5 | 326 | sched->nr_sleep_events++; |
ec156764 IM |
327 | } |
328 | ||
0e9b07e5 ACM |
329 | static struct task_desc *register_pid(struct perf_sched *sched, |
330 | unsigned long pid, const char *comm) | |
ec156764 IM |
331 | { |
332 | struct task_desc *task; | |
cb06ac25 | 333 | static int pid_max; |
ec156764 | 334 | |
cb06ac25 YS |
335 | if (sched->pid_to_task == NULL) { |
336 | if (sysctl__read_int("kernel/pid_max", &pid_max) < 0) | |
337 | pid_max = MAX_PID; | |
338 | BUG_ON((sched->pid_to_task = calloc(pid_max, sizeof(struct task_desc *))) == NULL); | |
339 | } | |
3a423a5c YS |
340 | if (pid >= (unsigned long)pid_max) { |
341 | BUG_ON((sched->pid_to_task = realloc(sched->pid_to_task, (pid + 1) * | |
342 | sizeof(struct task_desc *))) == NULL); | |
343 | while (pid >= (unsigned long)pid_max) | |
344 | sched->pid_to_task[pid_max++] = NULL; | |
345 | } | |
ec156764 | 346 | |
0e9b07e5 | 347 | task = sched->pid_to_task[pid]; |
ec156764 IM |
348 | |
349 | if (task) | |
350 | return task; | |
351 | ||
36479484 | 352 | task = zalloc(sizeof(*task)); |
ec156764 | 353 | task->pid = pid; |
0e9b07e5 | 354 | task->nr = sched->nr_tasks; |
ec156764 IM |
355 | strcpy(task->comm, comm); |
356 | /* | |
357 | * every task starts in sleeping state - this gets ignored | |
358 | * if there's no wakeup pointing to this sleep state: | |
359 | */ | |
0e9b07e5 | 360 | add_sched_event_sleep(sched, task, 0, 0); |
ec156764 | 361 | |
0e9b07e5 ACM |
362 | sched->pid_to_task[pid] = task; |
363 | sched->nr_tasks++; | |
0755bc4d | 364 | sched->tasks = realloc(sched->tasks, sched->nr_tasks * sizeof(struct task_desc *)); |
0e9b07e5 ACM |
365 | BUG_ON(!sched->tasks); |
366 | sched->tasks[task->nr] = task; | |
ec156764 | 367 | |
ad236fd2 | 368 | if (verbose) |
0e9b07e5 | 369 | printf("registered task #%ld, PID %ld (%s)\n", sched->nr_tasks, pid, comm); |
ec156764 IM |
370 | |
371 | return task; | |
372 | } | |
373 | ||
374 | ||
0e9b07e5 | 375 | static void print_task_traces(struct perf_sched *sched) |
ec156764 IM |
376 | { |
377 | struct task_desc *task; | |
378 | unsigned long i; | |
379 | ||
0e9b07e5 ACM |
380 | for (i = 0; i < sched->nr_tasks; i++) { |
381 | task = sched->tasks[i]; | |
ad236fd2 | 382 | printf("task %6ld (%20s:%10ld), nr_events: %ld\n", |
ec156764 IM |
383 | task->nr, task->comm, task->pid, task->nr_events); |
384 | } | |
385 | } | |
386 | ||
0e9b07e5 | 387 | static void add_cross_task_wakeups(struct perf_sched *sched) |
ec156764 IM |
388 | { |
389 | struct task_desc *task1, *task2; | |
390 | unsigned long i, j; | |
391 | ||
0e9b07e5 ACM |
392 | for (i = 0; i < sched->nr_tasks; i++) { |
393 | task1 = sched->tasks[i]; | |
ec156764 | 394 | j = i + 1; |
0e9b07e5 | 395 | if (j == sched->nr_tasks) |
ec156764 | 396 | j = 0; |
0e9b07e5 ACM |
397 | task2 = sched->tasks[j]; |
398 | add_sched_event_wakeup(sched, task1, 0, task2); | |
ec156764 IM |
399 | } |
400 | } | |
401 | ||
0e9b07e5 ACM |
402 | static void perf_sched__process_event(struct perf_sched *sched, |
403 | struct sched_atom *atom) | |
ec156764 IM |
404 | { |
405 | int ret = 0; | |
ec156764 | 406 | |
39aeb52f | 407 | switch (atom->type) { |
ec156764 | 408 | case SCHED_EVENT_RUN: |
0e9b07e5 | 409 | burn_nsecs(sched, atom->duration); |
ec156764 IM |
410 | break; |
411 | case SCHED_EVENT_SLEEP: | |
39aeb52f | 412 | if (atom->wait_sem) |
413 | ret = sem_wait(atom->wait_sem); | |
ec156764 IM |
414 | BUG_ON(ret); |
415 | break; | |
416 | case SCHED_EVENT_WAKEUP: | |
39aeb52f | 417 | if (atom->wait_sem) |
418 | ret = sem_post(atom->wait_sem); | |
ec156764 IM |
419 | BUG_ON(ret); |
420 | break; | |
55ffb7a6 MG |
421 | case SCHED_EVENT_MIGRATION: |
422 | break; | |
ec156764 IM |
423 | default: |
424 | BUG_ON(1); | |
425 | } | |
426 | } | |
427 | ||
b1ffe8f3 | 428 | static u64 get_cpu_usage_nsec_parent(void) |
ec156764 IM |
429 | { |
430 | struct rusage ru; | |
b1ffe8f3 | 431 | u64 sum; |
ec156764 IM |
432 | int err; |
433 | ||
434 | err = getrusage(RUSAGE_SELF, &ru); | |
435 | BUG_ON(err); | |
436 | ||
437 | sum = ru.ru_utime.tv_sec*1e9 + ru.ru_utime.tv_usec*1e3; | |
438 | sum += ru.ru_stime.tv_sec*1e9 + ru.ru_stime.tv_usec*1e3; | |
439 | ||
440 | return sum; | |
441 | } | |
442 | ||
939cda52 | 443 | static int self_open_counters(struct perf_sched *sched, unsigned long cur_task) |
ec156764 | 444 | { |
c0c9e721 | 445 | struct perf_event_attr attr; |
939cda52 | 446 | char sbuf[STRERR_BUFSIZE], info[STRERR_BUFSIZE]; |
c0c9e721 | 447 | int fd; |
939cda52 YS |
448 | struct rlimit limit; |
449 | bool need_privilege = false; | |
ec156764 | 450 | |
c0c9e721 | 451 | memset(&attr, 0, sizeof(attr)); |
ec156764 | 452 | |
c0c9e721 XG |
453 | attr.type = PERF_TYPE_SOFTWARE; |
454 | attr.config = PERF_COUNT_SW_TASK_CLOCK; | |
ec156764 | 455 | |
939cda52 | 456 | force_again: |
57480d2c YD |
457 | fd = sys_perf_event_open(&attr, 0, -1, -1, |
458 | perf_event_open_cloexec_flag()); | |
c0c9e721 | 459 | |
1aff59be | 460 | if (fd < 0) { |
939cda52 YS |
461 | if (errno == EMFILE) { |
462 | if (sched->force) { | |
463 | BUG_ON(getrlimit(RLIMIT_NOFILE, &limit) == -1); | |
464 | limit.rlim_cur += sched->nr_tasks - cur_task; | |
465 | if (limit.rlim_cur > limit.rlim_max) { | |
466 | limit.rlim_max = limit.rlim_cur; | |
467 | need_privilege = true; | |
468 | } | |
469 | if (setrlimit(RLIMIT_NOFILE, &limit) == -1) { | |
470 | if (need_privilege && errno == EPERM) | |
471 | strcpy(info, "Need privilege\n"); | |
472 | } else | |
473 | goto force_again; | |
474 | } else | |
475 | strcpy(info, "Have a try with -f option\n"); | |
476 | } | |
60b7d14a | 477 | pr_err("Error: sys_perf_event_open() syscall returned " |
939cda52 YS |
478 | "with %d (%s)\n%s", fd, |
479 | strerror_r(errno, sbuf, sizeof(sbuf)), info); | |
1aff59be YS |
480 | exit(EXIT_FAILURE); |
481 | } | |
c0c9e721 XG |
482 | return fd; |
483 | } | |
484 | ||
485 | static u64 get_cpu_usage_nsec_self(int fd) | |
486 | { | |
487 | u64 runtime; | |
488 | int ret; | |
489 | ||
490 | ret = read(fd, &runtime, sizeof(runtime)); | |
491 | BUG_ON(ret != sizeof(runtime)); | |
492 | ||
493 | return runtime; | |
ec156764 IM |
494 | } |
495 | ||
0e9b07e5 ACM |
496 | struct sched_thread_parms { |
497 | struct task_desc *task; | |
498 | struct perf_sched *sched; | |
08097abc | 499 | int fd; |
0e9b07e5 ACM |
500 | }; |
501 | ||
ec156764 IM |
502 | static void *thread_func(void *ctx) |
503 | { | |
0e9b07e5 ACM |
504 | struct sched_thread_parms *parms = ctx; |
505 | struct task_desc *this_task = parms->task; | |
506 | struct perf_sched *sched = parms->sched; | |
b1ffe8f3 | 507 | u64 cpu_usage_0, cpu_usage_1; |
ec156764 IM |
508 | unsigned long i, ret; |
509 | char comm2[22]; | |
08097abc | 510 | int fd = parms->fd; |
ec156764 | 511 | |
74cf249d | 512 | zfree(&parms); |
0e9b07e5 | 513 | |
ec156764 IM |
514 | sprintf(comm2, ":%s", this_task->comm); |
515 | prctl(PR_SET_NAME, comm2); | |
a116e05d ACM |
516 | if (fd < 0) |
517 | return NULL; | |
ec156764 IM |
518 | again: |
519 | ret = sem_post(&this_task->ready_for_work); | |
520 | BUG_ON(ret); | |
0e9b07e5 | 521 | ret = pthread_mutex_lock(&sched->start_work_mutex); |
ec156764 | 522 | BUG_ON(ret); |
0e9b07e5 | 523 | ret = pthread_mutex_unlock(&sched->start_work_mutex); |
ec156764 | 524 | BUG_ON(ret); |
ec156764 | 525 | |
c0c9e721 | 526 | cpu_usage_0 = get_cpu_usage_nsec_self(fd); |
ec156764 IM |
527 | |
528 | for (i = 0; i < this_task->nr_events; i++) { | |
529 | this_task->curr_event = i; | |
0e9b07e5 | 530 | perf_sched__process_event(sched, this_task->atoms[i]); |
ec156764 IM |
531 | } |
532 | ||
c0c9e721 | 533 | cpu_usage_1 = get_cpu_usage_nsec_self(fd); |
ec156764 | 534 | this_task->cpu_usage = cpu_usage_1 - cpu_usage_0; |
ec156764 IM |
535 | ret = sem_post(&this_task->work_done_sem); |
536 | BUG_ON(ret); | |
ec156764 | 537 | |
0e9b07e5 | 538 | ret = pthread_mutex_lock(&sched->work_done_wait_mutex); |
ec156764 | 539 | BUG_ON(ret); |
0e9b07e5 | 540 | ret = pthread_mutex_unlock(&sched->work_done_wait_mutex); |
ec156764 | 541 | BUG_ON(ret); |
ec156764 IM |
542 | |
543 | goto again; | |
544 | } | |
545 | ||
0e9b07e5 | 546 | static void create_tasks(struct perf_sched *sched) |
ec156764 IM |
547 | { |
548 | struct task_desc *task; | |
549 | pthread_attr_t attr; | |
550 | unsigned long i; | |
551 | int err; | |
552 | ||
553 | err = pthread_attr_init(&attr); | |
554 | BUG_ON(err); | |
12f7e036 JP |
555 | err = pthread_attr_setstacksize(&attr, |
556 | (size_t) max(16 * 1024, PTHREAD_STACK_MIN)); | |
ec156764 | 557 | BUG_ON(err); |
0e9b07e5 | 558 | err = pthread_mutex_lock(&sched->start_work_mutex); |
ec156764 | 559 | BUG_ON(err); |
0e9b07e5 | 560 | err = pthread_mutex_lock(&sched->work_done_wait_mutex); |
ec156764 | 561 | BUG_ON(err); |
0e9b07e5 ACM |
562 | for (i = 0; i < sched->nr_tasks; i++) { |
563 | struct sched_thread_parms *parms = malloc(sizeof(*parms)); | |
564 | BUG_ON(parms == NULL); | |
565 | parms->task = task = sched->tasks[i]; | |
566 | parms->sched = sched; | |
939cda52 | 567 | parms->fd = self_open_counters(sched, i); |
ec156764 IM |
568 | sem_init(&task->sleep_sem, 0, 0); |
569 | sem_init(&task->ready_for_work, 0, 0); | |
570 | sem_init(&task->work_done_sem, 0, 0); | |
571 | task->curr_event = 0; | |
0e9b07e5 | 572 | err = pthread_create(&task->thread, &attr, thread_func, parms); |
ec156764 IM |
573 | BUG_ON(err); |
574 | } | |
575 | } | |
576 | ||
0e9b07e5 | 577 | static void wait_for_tasks(struct perf_sched *sched) |
ec156764 | 578 | { |
b1ffe8f3 | 579 | u64 cpu_usage_0, cpu_usage_1; |
ec156764 IM |
580 | struct task_desc *task; |
581 | unsigned long i, ret; | |
582 | ||
0e9b07e5 ACM |
583 | sched->start_time = get_nsecs(); |
584 | sched->cpu_usage = 0; | |
585 | pthread_mutex_unlock(&sched->work_done_wait_mutex); | |
ec156764 | 586 | |
0e9b07e5 ACM |
587 | for (i = 0; i < sched->nr_tasks; i++) { |
588 | task = sched->tasks[i]; | |
ec156764 IM |
589 | ret = sem_wait(&task->ready_for_work); |
590 | BUG_ON(ret); | |
591 | sem_init(&task->ready_for_work, 0, 0); | |
592 | } | |
0e9b07e5 | 593 | ret = pthread_mutex_lock(&sched->work_done_wait_mutex); |
ec156764 IM |
594 | BUG_ON(ret); |
595 | ||
596 | cpu_usage_0 = get_cpu_usage_nsec_parent(); | |
597 | ||
0e9b07e5 | 598 | pthread_mutex_unlock(&sched->start_work_mutex); |
ec156764 | 599 | |
0e9b07e5 ACM |
600 | for (i = 0; i < sched->nr_tasks; i++) { |
601 | task = sched->tasks[i]; | |
ec156764 IM |
602 | ret = sem_wait(&task->work_done_sem); |
603 | BUG_ON(ret); | |
604 | sem_init(&task->work_done_sem, 0, 0); | |
0e9b07e5 | 605 | sched->cpu_usage += task->cpu_usage; |
ec156764 IM |
606 | task->cpu_usage = 0; |
607 | } | |
608 | ||
609 | cpu_usage_1 = get_cpu_usage_nsec_parent(); | |
0e9b07e5 ACM |
610 | if (!sched->runavg_cpu_usage) |
611 | sched->runavg_cpu_usage = sched->cpu_usage; | |
ff5f3bbd | 612 | sched->runavg_cpu_usage = (sched->runavg_cpu_usage * (sched->replay_repeat - 1) + sched->cpu_usage) / sched->replay_repeat; |
ec156764 | 613 | |
0e9b07e5 ACM |
614 | sched->parent_cpu_usage = cpu_usage_1 - cpu_usage_0; |
615 | if (!sched->runavg_parent_cpu_usage) | |
616 | sched->runavg_parent_cpu_usage = sched->parent_cpu_usage; | |
ff5f3bbd YS |
617 | sched->runavg_parent_cpu_usage = (sched->runavg_parent_cpu_usage * (sched->replay_repeat - 1) + |
618 | sched->parent_cpu_usage)/sched->replay_repeat; | |
ec156764 | 619 | |
0e9b07e5 | 620 | ret = pthread_mutex_lock(&sched->start_work_mutex); |
ec156764 IM |
621 | BUG_ON(ret); |
622 | ||
0e9b07e5 ACM |
623 | for (i = 0; i < sched->nr_tasks; i++) { |
624 | task = sched->tasks[i]; | |
ec156764 IM |
625 | sem_init(&task->sleep_sem, 0, 0); |
626 | task->curr_event = 0; | |
627 | } | |
628 | } | |
629 | ||
0e9b07e5 | 630 | static void run_one_test(struct perf_sched *sched) |
ec156764 | 631 | { |
fb7d0b3c | 632 | u64 T0, T1, delta, avg_delta, fluct; |
ec156764 IM |
633 | |
634 | T0 = get_nsecs(); | |
0e9b07e5 | 635 | wait_for_tasks(sched); |
ec156764 IM |
636 | T1 = get_nsecs(); |
637 | ||
638 | delta = T1 - T0; | |
0e9b07e5 ACM |
639 | sched->sum_runtime += delta; |
640 | sched->nr_runs++; | |
ec156764 | 641 | |
0e9b07e5 | 642 | avg_delta = sched->sum_runtime / sched->nr_runs; |
ec156764 IM |
643 | if (delta < avg_delta) |
644 | fluct = avg_delta - delta; | |
645 | else | |
646 | fluct = delta - avg_delta; | |
0e9b07e5 ACM |
647 | sched->sum_fluct += fluct; |
648 | if (!sched->run_avg) | |
649 | sched->run_avg = delta; | |
ff5f3bbd | 650 | sched->run_avg = (sched->run_avg * (sched->replay_repeat - 1) + delta) / sched->replay_repeat; |
ec156764 | 651 | |
0e9b07e5 | 652 | printf("#%-3ld: %0.3f, ", sched->nr_runs, (double)delta / 1000000.0); |
ec156764 | 653 | |
0e9b07e5 | 654 | printf("ravg: %0.2f, ", (double)sched->run_avg / 1e6); |
ec156764 | 655 | |
ad236fd2 | 656 | printf("cpu: %0.2f / %0.2f", |
0e9b07e5 | 657 | (double)sched->cpu_usage / 1e6, (double)sched->runavg_cpu_usage / 1e6); |
ec156764 IM |
658 | |
659 | #if 0 | |
660 | /* | |
fbf94829 | 661 | * rusage statistics done by the parent, these are less |
0e9b07e5 | 662 | * accurate than the sched->sum_exec_runtime based statistics: |
fbf94829 | 663 | */ |
ad236fd2 | 664 | printf(" [%0.2f / %0.2f]", |
0e9b07e5 ACM |
665 | (double)sched->parent_cpu_usage/1e6, |
666 | (double)sched->runavg_parent_cpu_usage/1e6); | |
ec156764 IM |
667 | #endif |
668 | ||
ad236fd2 | 669 | printf("\n"); |
ec156764 | 670 | |
0e9b07e5 ACM |
671 | if (sched->nr_sleep_corrections) |
672 | printf(" (%ld sleep corrections)\n", sched->nr_sleep_corrections); | |
673 | sched->nr_sleep_corrections = 0; | |
ec156764 IM |
674 | } |
675 | ||
0e9b07e5 | 676 | static void test_calibrations(struct perf_sched *sched) |
ec156764 | 677 | { |
b1ffe8f3 | 678 | u64 T0, T1; |
ec156764 IM |
679 | |
680 | T0 = get_nsecs(); | |
0e9b07e5 | 681 | burn_nsecs(sched, 1e6); |
ec156764 IM |
682 | T1 = get_nsecs(); |
683 | ||
9486aa38 | 684 | printf("the run test took %" PRIu64 " nsecs\n", T1 - T0); |
ec156764 IM |
685 | |
686 | T0 = get_nsecs(); | |
687 | sleep_nsecs(1e6); | |
688 | T1 = get_nsecs(); | |
689 | ||
9486aa38 | 690 | printf("the sleep test took %" PRIu64 " nsecs\n", T1 - T0); |
ec156764 IM |
691 | } |
692 | ||
a116e05d | 693 | static int |
0e9b07e5 | 694 | replay_wakeup_event(struct perf_sched *sched, |
9ec3f4e4 ACM |
695 | struct perf_evsel *evsel, struct perf_sample *sample, |
696 | struct machine *machine __maybe_unused) | |
419ab0d6 | 697 | { |
9ec3f4e4 ACM |
698 | const char *comm = perf_evsel__strval(evsel, sample, "comm"); |
699 | const u32 pid = perf_evsel__intval(evsel, sample, "pid"); | |
419ab0d6 | 700 | struct task_desc *waker, *wakee; |
fbf94829 | 701 | |
ad236fd2 | 702 | if (verbose) { |
2b7fcbc5 | 703 | printf("sched_wakeup event %p\n", evsel); |
fbf94829 | 704 | |
9ec3f4e4 | 705 | printf(" ... pid %d woke up %s/%d\n", sample->tid, comm, pid); |
ad236fd2 | 706 | } |
fbf94829 | 707 | |
2b7fcbc5 | 708 | waker = register_pid(sched, sample->tid, "<unknown>"); |
9ec3f4e4 | 709 | wakee = register_pid(sched, pid, comm); |
fbf94829 | 710 | |
0e9b07e5 | 711 | add_sched_event_wakeup(sched, waker, sample->time, wakee); |
a116e05d | 712 | return 0; |
ec156764 IM |
713 | } |
714 | ||
9ec3f4e4 ACM |
715 | static int replay_switch_event(struct perf_sched *sched, |
716 | struct perf_evsel *evsel, | |
717 | struct perf_sample *sample, | |
718 | struct machine *machine __maybe_unused) | |
ec156764 | 719 | { |
9ec3f4e4 ACM |
720 | const char *prev_comm = perf_evsel__strval(evsel, sample, "prev_comm"), |
721 | *next_comm = perf_evsel__strval(evsel, sample, "next_comm"); | |
722 | const u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"), | |
723 | next_pid = perf_evsel__intval(evsel, sample, "next_pid"); | |
724 | const u64 prev_state = perf_evsel__intval(evsel, sample, "prev_state"); | |
1d037ca1 | 725 | struct task_desc *prev, __maybe_unused *next; |
7f7f8d0b ACM |
726 | u64 timestamp0, timestamp = sample->time; |
727 | int cpu = sample->cpu; | |
fbf94829 IM |
728 | s64 delta; |
729 | ||
ad236fd2 | 730 | if (verbose) |
2b7fcbc5 | 731 | printf("sched_switch event %p\n", evsel); |
ad236fd2 | 732 | |
fbf94829 | 733 | if (cpu >= MAX_CPUS || cpu < 0) |
a116e05d | 734 | return 0; |
fbf94829 | 735 | |
0e9b07e5 | 736 | timestamp0 = sched->cpu_last_switched[cpu]; |
fbf94829 IM |
737 | if (timestamp0) |
738 | delta = timestamp - timestamp0; | |
739 | else | |
740 | delta = 0; | |
741 | ||
a116e05d | 742 | if (delta < 0) { |
60b7d14a | 743 | pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta); |
a116e05d ACM |
744 | return -1; |
745 | } | |
fbf94829 | 746 | |
9ec3f4e4 ACM |
747 | pr_debug(" ... switch from %s/%d to %s/%d [ran %" PRIu64 " nsecs]\n", |
748 | prev_comm, prev_pid, next_comm, next_pid, delta); | |
fbf94829 | 749 | |
9ec3f4e4 ACM |
750 | prev = register_pid(sched, prev_pid, prev_comm); |
751 | next = register_pid(sched, next_pid, next_comm); | |
fbf94829 | 752 | |
0e9b07e5 | 753 | sched->cpu_last_switched[cpu] = timestamp; |
fbf94829 | 754 | |
0e9b07e5 | 755 | add_sched_event_run(sched, prev, timestamp, delta); |
9ec3f4e4 | 756 | add_sched_event_sleep(sched, prev, timestamp, prev_state); |
a116e05d ACM |
757 | |
758 | return 0; | |
fbf94829 IM |
759 | } |
760 | ||
cb627505 DA |
761 | static int replay_fork_event(struct perf_sched *sched, |
762 | union perf_event *event, | |
763 | struct machine *machine) | |
419ab0d6 | 764 | { |
cb627505 DA |
765 | struct thread *child, *parent; |
766 | ||
314add6b AH |
767 | child = machine__findnew_thread(machine, event->fork.pid, |
768 | event->fork.tid); | |
769 | parent = machine__findnew_thread(machine, event->fork.ppid, | |
770 | event->fork.ptid); | |
cb627505 DA |
771 | |
772 | if (child == NULL || parent == NULL) { | |
773 | pr_debug("thread does not exist on fork event: child %p, parent %p\n", | |
774 | child, parent); | |
b91fc39f | 775 | goto out_put; |
cb627505 | 776 | } |
9ec3f4e4 | 777 | |
419ab0d6 | 778 | if (verbose) { |
cb627505 | 779 | printf("fork event\n"); |
b9c5143a FW |
780 | printf("... parent: %s/%d\n", thread__comm_str(parent), parent->tid); |
781 | printf("... child: %s/%d\n", thread__comm_str(child), child->tid); | |
419ab0d6 | 782 | } |
9ec3f4e4 | 783 | |
b9c5143a FW |
784 | register_pid(sched, parent->tid, thread__comm_str(parent)); |
785 | register_pid(sched, child->tid, thread__comm_str(child)); | |
b91fc39f ACM |
786 | out_put: |
787 | thread__put(child); | |
788 | thread__put(parent); | |
a116e05d | 789 | return 0; |
419ab0d6 | 790 | } |
fbf94829 | 791 | |
b1ffe8f3 IM |
792 | struct sort_dimension { |
793 | const char *name; | |
b5fae128 | 794 | sort_fn_t cmp; |
b1ffe8f3 IM |
795 | struct list_head list; |
796 | }; | |
797 | ||
daa1d7a5 | 798 | static int |
39aeb52f | 799 | thread_lat_cmp(struct list_head *list, struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 FW |
800 | { |
801 | struct sort_dimension *sort; | |
802 | int ret = 0; | |
803 | ||
b5fae128 IM |
804 | BUG_ON(list_empty(list)); |
805 | ||
daa1d7a5 FW |
806 | list_for_each_entry(sort, list, list) { |
807 | ret = sort->cmp(l, r); | |
808 | if (ret) | |
809 | return ret; | |
810 | } | |
811 | ||
812 | return ret; | |
813 | } | |
814 | ||
39aeb52f | 815 | static struct work_atoms * |
b5fae128 IM |
816 | thread_atoms_search(struct rb_root *root, struct thread *thread, |
817 | struct list_head *sort_list) | |
818 | { | |
819 | struct rb_node *node = root->rb_node; | |
39aeb52f | 820 | struct work_atoms key = { .thread = thread }; |
b5fae128 IM |
821 | |
822 | while (node) { | |
39aeb52f | 823 | struct work_atoms *atoms; |
b5fae128 IM |
824 | int cmp; |
825 | ||
39aeb52f | 826 | atoms = container_of(node, struct work_atoms, node); |
b5fae128 IM |
827 | |
828 | cmp = thread_lat_cmp(sort_list, &key, atoms); | |
829 | if (cmp > 0) | |
830 | node = node->rb_left; | |
831 | else if (cmp < 0) | |
832 | node = node->rb_right; | |
833 | else { | |
834 | BUG_ON(thread != atoms->thread); | |
835 | return atoms; | |
836 | } | |
837 | } | |
838 | return NULL; | |
839 | } | |
840 | ||
cdce9d73 | 841 | static void |
39aeb52f | 842 | __thread_latency_insert(struct rb_root *root, struct work_atoms *data, |
daa1d7a5 | 843 | struct list_head *sort_list) |
cdce9d73 FW |
844 | { |
845 | struct rb_node **new = &(root->rb_node), *parent = NULL; | |
846 | ||
847 | while (*new) { | |
39aeb52f | 848 | struct work_atoms *this; |
daa1d7a5 | 849 | int cmp; |
cdce9d73 | 850 | |
39aeb52f | 851 | this = container_of(*new, struct work_atoms, node); |
cdce9d73 | 852 | parent = *new; |
daa1d7a5 FW |
853 | |
854 | cmp = thread_lat_cmp(sort_list, data, this); | |
855 | ||
856 | if (cmp > 0) | |
cdce9d73 | 857 | new = &((*new)->rb_left); |
cdce9d73 | 858 | else |
daa1d7a5 | 859 | new = &((*new)->rb_right); |
cdce9d73 FW |
860 | } |
861 | ||
862 | rb_link_node(&data->node, parent, new); | |
863 | rb_insert_color(&data->node, root); | |
864 | } | |
865 | ||
0e9b07e5 | 866 | static int thread_atoms_insert(struct perf_sched *sched, struct thread *thread) |
cdce9d73 | 867 | { |
36479484 | 868 | struct work_atoms *atoms = zalloc(sizeof(*atoms)); |
a116e05d ACM |
869 | if (!atoms) { |
870 | pr_err("No memory at %s\n", __func__); | |
871 | return -1; | |
872 | } | |
cdce9d73 | 873 | |
f3b623b8 | 874 | atoms->thread = thread__get(thread); |
39aeb52f | 875 | INIT_LIST_HEAD(&atoms->work_list); |
0e9b07e5 | 876 | __thread_latency_insert(&sched->atom_root, atoms, &sched->cmp_pid); |
a116e05d | 877 | return 0; |
cdce9d73 FW |
878 | } |
879 | ||
9ec3f4e4 | 880 | static char sched_out_state(u64 prev_state) |
cdce9d73 FW |
881 | { |
882 | const char *str = TASK_STATE_TO_CHAR_STR; | |
883 | ||
9ec3f4e4 | 884 | return str[prev_state]; |
cdce9d73 FW |
885 | } |
886 | ||
a116e05d | 887 | static int |
39aeb52f | 888 | add_sched_out_event(struct work_atoms *atoms, |
889 | char run_state, | |
890 | u64 timestamp) | |
cdce9d73 | 891 | { |
36479484 | 892 | struct work_atom *atom = zalloc(sizeof(*atom)); |
a116e05d ACM |
893 | if (!atom) { |
894 | pr_err("Non memory at %s", __func__); | |
895 | return -1; | |
896 | } | |
cdce9d73 | 897 | |
aa1ab9d2 FW |
898 | atom->sched_out_time = timestamp; |
899 | ||
39aeb52f | 900 | if (run_state == 'R') { |
b1ffe8f3 | 901 | atom->state = THREAD_WAIT_CPU; |
aa1ab9d2 | 902 | atom->wake_up_time = atom->sched_out_time; |
c6ced611 FW |
903 | } |
904 | ||
39aeb52f | 905 | list_add_tail(&atom->list, &atoms->work_list); |
a116e05d | 906 | return 0; |
cdce9d73 FW |
907 | } |
908 | ||
909 | static void | |
1d037ca1 IT |
910 | add_runtime_event(struct work_atoms *atoms, u64 delta, |
911 | u64 timestamp __maybe_unused) | |
39aeb52f | 912 | { |
913 | struct work_atom *atom; | |
914 | ||
915 | BUG_ON(list_empty(&atoms->work_list)); | |
916 | ||
917 | atom = list_entry(atoms->work_list.prev, struct work_atom, list); | |
918 | ||
919 | atom->runtime += delta; | |
920 | atoms->total_runtime += delta; | |
921 | } | |
922 | ||
923 | static void | |
924 | add_sched_in_event(struct work_atoms *atoms, u64 timestamp) | |
cdce9d73 | 925 | { |
b1ffe8f3 | 926 | struct work_atom *atom; |
66685678 | 927 | u64 delta; |
cdce9d73 | 928 | |
39aeb52f | 929 | if (list_empty(&atoms->work_list)) |
cdce9d73 FW |
930 | return; |
931 | ||
39aeb52f | 932 | atom = list_entry(atoms->work_list.prev, struct work_atom, list); |
cdce9d73 | 933 | |
b1ffe8f3 | 934 | if (atom->state != THREAD_WAIT_CPU) |
cdce9d73 FW |
935 | return; |
936 | ||
b1ffe8f3 IM |
937 | if (timestamp < atom->wake_up_time) { |
938 | atom->state = THREAD_IGNORE; | |
cdce9d73 FW |
939 | return; |
940 | } | |
941 | ||
b1ffe8f3 IM |
942 | atom->state = THREAD_SCHED_IN; |
943 | atom->sched_in_time = timestamp; | |
66685678 | 944 | |
b1ffe8f3 | 945 | delta = atom->sched_in_time - atom->wake_up_time; |
66685678 | 946 | atoms->total_lat += delta; |
3786310a | 947 | if (delta > atoms->max_lat) { |
66685678 | 948 | atoms->max_lat = delta; |
3786310a FW |
949 | atoms->max_lat_at = timestamp; |
950 | } | |
66685678 | 951 | atoms->nb_atoms++; |
cdce9d73 FW |
952 | } |
953 | ||
9ec3f4e4 ACM |
954 | static int latency_switch_event(struct perf_sched *sched, |
955 | struct perf_evsel *evsel, | |
956 | struct perf_sample *sample, | |
957 | struct machine *machine) | |
cdce9d73 | 958 | { |
9ec3f4e4 ACM |
959 | const u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"), |
960 | next_pid = perf_evsel__intval(evsel, sample, "next_pid"); | |
961 | const u64 prev_state = perf_evsel__intval(evsel, sample, "prev_state"); | |
39aeb52f | 962 | struct work_atoms *out_events, *in_events; |
cdce9d73 | 963 | struct thread *sched_out, *sched_in; |
7f7f8d0b | 964 | u64 timestamp0, timestamp = sample->time; |
b91fc39f | 965 | int cpu = sample->cpu, err = -1; |
ea92ed5a IM |
966 | s64 delta; |
967 | ||
39aeb52f | 968 | BUG_ON(cpu >= MAX_CPUS || cpu < 0); |
ea92ed5a | 969 | |
0e9b07e5 ACM |
970 | timestamp0 = sched->cpu_last_switched[cpu]; |
971 | sched->cpu_last_switched[cpu] = timestamp; | |
ea92ed5a IM |
972 | if (timestamp0) |
973 | delta = timestamp - timestamp0; | |
974 | else | |
975 | delta = 0; | |
976 | ||
a116e05d ACM |
977 | if (delta < 0) { |
978 | pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta); | |
979 | return -1; | |
980 | } | |
cdce9d73 | 981 | |
1fcb8768 AH |
982 | sched_out = machine__findnew_thread(machine, -1, prev_pid); |
983 | sched_in = machine__findnew_thread(machine, -1, next_pid); | |
b91fc39f ACM |
984 | if (sched_out == NULL || sched_in == NULL) |
985 | goto out_put; | |
cdce9d73 | 986 | |
0e9b07e5 | 987 | out_events = thread_atoms_search(&sched->atom_root, sched_out, &sched->cmp_pid); |
39aeb52f | 988 | if (!out_events) { |
0e9b07e5 | 989 | if (thread_atoms_insert(sched, sched_out)) |
b91fc39f | 990 | goto out_put; |
0e9b07e5 | 991 | out_events = thread_atoms_search(&sched->atom_root, sched_out, &sched->cmp_pid); |
a116e05d ACM |
992 | if (!out_events) { |
993 | pr_err("out-event: Internal tree error"); | |
b91fc39f | 994 | goto out_put; |
a116e05d | 995 | } |
39aeb52f | 996 | } |
9ec3f4e4 | 997 | if (add_sched_out_event(out_events, sched_out_state(prev_state), timestamp)) |
a116e05d | 998 | return -1; |
39aeb52f | 999 | |
0e9b07e5 | 1000 | in_events = thread_atoms_search(&sched->atom_root, sched_in, &sched->cmp_pid); |
39aeb52f | 1001 | if (!in_events) { |
0e9b07e5 | 1002 | if (thread_atoms_insert(sched, sched_in)) |
b91fc39f | 1003 | goto out_put; |
0e9b07e5 | 1004 | in_events = thread_atoms_search(&sched->atom_root, sched_in, &sched->cmp_pid); |
a116e05d ACM |
1005 | if (!in_events) { |
1006 | pr_err("in-event: Internal tree error"); | |
b91fc39f | 1007 | goto out_put; |
a116e05d | 1008 | } |
39aeb52f | 1009 | /* |
1010 | * Take came in we have not heard about yet, | |
1011 | * add in an initial atom in runnable state: | |
1012 | */ | |
a116e05d | 1013 | if (add_sched_out_event(in_events, 'R', timestamp)) |
b91fc39f | 1014 | goto out_put; |
cdce9d73 | 1015 | } |
39aeb52f | 1016 | add_sched_in_event(in_events, timestamp); |
b91fc39f ACM |
1017 | err = 0; |
1018 | out_put: | |
1019 | thread__put(sched_out); | |
1020 | thread__put(sched_in); | |
1021 | return err; | |
39aeb52f | 1022 | } |
cdce9d73 | 1023 | |
9ec3f4e4 ACM |
1024 | static int latency_runtime_event(struct perf_sched *sched, |
1025 | struct perf_evsel *evsel, | |
1026 | struct perf_sample *sample, | |
1027 | struct machine *machine) | |
39aeb52f | 1028 | { |
9ec3f4e4 ACM |
1029 | const u32 pid = perf_evsel__intval(evsel, sample, "pid"); |
1030 | const u64 runtime = perf_evsel__intval(evsel, sample, "runtime"); | |
1fcb8768 | 1031 | struct thread *thread = machine__findnew_thread(machine, -1, pid); |
0e9b07e5 | 1032 | struct work_atoms *atoms = thread_atoms_search(&sched->atom_root, thread, &sched->cmp_pid); |
7f7f8d0b | 1033 | u64 timestamp = sample->time; |
b91fc39f ACM |
1034 | int cpu = sample->cpu, err = -1; |
1035 | ||
1036 | if (thread == NULL) | |
1037 | return -1; | |
39aeb52f | 1038 | |
1039 | BUG_ON(cpu >= MAX_CPUS || cpu < 0); | |
39aeb52f | 1040 | if (!atoms) { |
0e9b07e5 | 1041 | if (thread_atoms_insert(sched, thread)) |
b91fc39f | 1042 | goto out_put; |
0e9b07e5 | 1043 | atoms = thread_atoms_search(&sched->atom_root, thread, &sched->cmp_pid); |
a116e05d | 1044 | if (!atoms) { |
60b7d14a | 1045 | pr_err("in-event: Internal tree error"); |
b91fc39f | 1046 | goto out_put; |
a116e05d ACM |
1047 | } |
1048 | if (add_sched_out_event(atoms, 'R', timestamp)) | |
b91fc39f | 1049 | goto out_put; |
cdce9d73 FW |
1050 | } |
1051 | ||
9ec3f4e4 | 1052 | add_runtime_event(atoms, runtime, timestamp); |
b91fc39f ACM |
1053 | err = 0; |
1054 | out_put: | |
1055 | thread__put(thread); | |
1056 | return err; | |
cdce9d73 FW |
1057 | } |
1058 | ||
9ec3f4e4 ACM |
1059 | static int latency_wakeup_event(struct perf_sched *sched, |
1060 | struct perf_evsel *evsel, | |
1061 | struct perf_sample *sample, | |
1062 | struct machine *machine) | |
cdce9d73 | 1063 | { |
0680ee7d | 1064 | const u32 pid = perf_evsel__intval(evsel, sample, "pid"); |
39aeb52f | 1065 | struct work_atoms *atoms; |
b1ffe8f3 | 1066 | struct work_atom *atom; |
cdce9d73 | 1067 | struct thread *wakee; |
7f7f8d0b | 1068 | u64 timestamp = sample->time; |
b91fc39f | 1069 | int err = -1; |
cdce9d73 | 1070 | |
1fcb8768 | 1071 | wakee = machine__findnew_thread(machine, -1, pid); |
b91fc39f ACM |
1072 | if (wakee == NULL) |
1073 | return -1; | |
0e9b07e5 | 1074 | atoms = thread_atoms_search(&sched->atom_root, wakee, &sched->cmp_pid); |
17562205 | 1075 | if (!atoms) { |
0e9b07e5 | 1076 | if (thread_atoms_insert(sched, wakee)) |
b91fc39f | 1077 | goto out_put; |
0e9b07e5 | 1078 | atoms = thread_atoms_search(&sched->atom_root, wakee, &sched->cmp_pid); |
a116e05d | 1079 | if (!atoms) { |
60b7d14a | 1080 | pr_err("wakeup-event: Internal tree error"); |
b91fc39f | 1081 | goto out_put; |
a116e05d ACM |
1082 | } |
1083 | if (add_sched_out_event(atoms, 'S', timestamp)) | |
b91fc39f | 1084 | goto out_put; |
cdce9d73 FW |
1085 | } |
1086 | ||
39aeb52f | 1087 | BUG_ON(list_empty(&atoms->work_list)); |
cdce9d73 | 1088 | |
39aeb52f | 1089 | atom = list_entry(atoms->work_list.prev, struct work_atom, list); |
cdce9d73 | 1090 | |
55ffb7a6 | 1091 | /* |
67d6259d DY |
1092 | * As we do not guarantee the wakeup event happens when |
1093 | * task is out of run queue, also may happen when task is | |
1094 | * on run queue and wakeup only change ->state to TASK_RUNNING, | |
1095 | * then we should not set the ->wake_up_time when wake up a | |
1096 | * task which is on run queue. | |
1097 | * | |
55ffb7a6 MG |
1098 | * You WILL be missing events if you've recorded only |
1099 | * one CPU, or are only looking at only one, so don't | |
67d6259d | 1100 | * skip in this case. |
55ffb7a6 | 1101 | */ |
0e9b07e5 | 1102 | if (sched->profile_cpu == -1 && atom->state != THREAD_SLEEPING) |
b91fc39f | 1103 | goto out_ok; |
cdce9d73 | 1104 | |
0e9b07e5 | 1105 | sched->nr_timestamps++; |
ea57c4f5 | 1106 | if (atom->sched_out_time > timestamp) { |
0e9b07e5 | 1107 | sched->nr_unordered_timestamps++; |
b91fc39f | 1108 | goto out_ok; |
ea57c4f5 | 1109 | } |
aa1ab9d2 | 1110 | |
b1ffe8f3 IM |
1111 | atom->state = THREAD_WAIT_CPU; |
1112 | atom->wake_up_time = timestamp; | |
b91fc39f ACM |
1113 | out_ok: |
1114 | err = 0; | |
1115 | out_put: | |
1116 | thread__put(wakee); | |
1117 | return err; | |
cdce9d73 FW |
1118 | } |
1119 | ||
9ec3f4e4 ACM |
1120 | static int latency_migrate_task_event(struct perf_sched *sched, |
1121 | struct perf_evsel *evsel, | |
1122 | struct perf_sample *sample, | |
1123 | struct machine *machine) | |
55ffb7a6 | 1124 | { |
9ec3f4e4 | 1125 | const u32 pid = perf_evsel__intval(evsel, sample, "pid"); |
7f7f8d0b | 1126 | u64 timestamp = sample->time; |
55ffb7a6 MG |
1127 | struct work_atoms *atoms; |
1128 | struct work_atom *atom; | |
1129 | struct thread *migrant; | |
b91fc39f | 1130 | int err = -1; |
55ffb7a6 MG |
1131 | |
1132 | /* | |
1133 | * Only need to worry about migration when profiling one CPU. | |
1134 | */ | |
0e9b07e5 | 1135 | if (sched->profile_cpu == -1) |
a116e05d | 1136 | return 0; |
55ffb7a6 | 1137 | |
1fcb8768 | 1138 | migrant = machine__findnew_thread(machine, -1, pid); |
b91fc39f ACM |
1139 | if (migrant == NULL) |
1140 | return -1; | |
0e9b07e5 | 1141 | atoms = thread_atoms_search(&sched->atom_root, migrant, &sched->cmp_pid); |
55ffb7a6 | 1142 | if (!atoms) { |
0e9b07e5 | 1143 | if (thread_atoms_insert(sched, migrant)) |
b91fc39f | 1144 | goto out_put; |
b9c5143a | 1145 | register_pid(sched, migrant->tid, thread__comm_str(migrant)); |
0e9b07e5 | 1146 | atoms = thread_atoms_search(&sched->atom_root, migrant, &sched->cmp_pid); |
a116e05d | 1147 | if (!atoms) { |
60b7d14a | 1148 | pr_err("migration-event: Internal tree error"); |
b91fc39f | 1149 | goto out_put; |
a116e05d ACM |
1150 | } |
1151 | if (add_sched_out_event(atoms, 'R', timestamp)) | |
b91fc39f | 1152 | goto out_put; |
55ffb7a6 MG |
1153 | } |
1154 | ||
1155 | BUG_ON(list_empty(&atoms->work_list)); | |
1156 | ||
1157 | atom = list_entry(atoms->work_list.prev, struct work_atom, list); | |
1158 | atom->sched_in_time = atom->sched_out_time = atom->wake_up_time = timestamp; | |
1159 | ||
0e9b07e5 | 1160 | sched->nr_timestamps++; |
55ffb7a6 MG |
1161 | |
1162 | if (atom->sched_out_time > timestamp) | |
0e9b07e5 | 1163 | sched->nr_unordered_timestamps++; |
b91fc39f ACM |
1164 | err = 0; |
1165 | out_put: | |
1166 | thread__put(migrant); | |
1167 | return err; | |
55ffb7a6 MG |
1168 | } |
1169 | ||
0e9b07e5 | 1170 | static void output_lat_thread(struct perf_sched *sched, struct work_atoms *work_list) |
cdce9d73 | 1171 | { |
cdce9d73 FW |
1172 | int i; |
1173 | int ret; | |
66685678 | 1174 | u64 avg; |
cdce9d73 | 1175 | |
39aeb52f | 1176 | if (!work_list->nb_atoms) |
cdce9d73 | 1177 | return; |
ea57c4f5 IM |
1178 | /* |
1179 | * Ignore idle threads: | |
1180 | */ | |
b9c5143a | 1181 | if (!strcmp(thread__comm_str(work_list->thread), "swapper")) |
ea57c4f5 | 1182 | return; |
cdce9d73 | 1183 | |
0e9b07e5 ACM |
1184 | sched->all_runtime += work_list->total_runtime; |
1185 | sched->all_count += work_list->nb_atoms; | |
66685678 | 1186 | |
2f80dd44 JB |
1187 | if (work_list->num_merged > 1) |
1188 | ret = printf(" %s:(%d) ", thread__comm_str(work_list->thread), work_list->num_merged); | |
1189 | else | |
1190 | ret = printf(" %s:%d ", thread__comm_str(work_list->thread), work_list->thread->tid); | |
cdce9d73 | 1191 | |
08f69e6c | 1192 | for (i = 0; i < 24 - ret; i++) |
cdce9d73 FW |
1193 | printf(" "); |
1194 | ||
39aeb52f | 1195 | avg = work_list->total_lat / work_list->nb_atoms; |
cdce9d73 | 1196 | |
80790e0b | 1197 | printf("|%11.3f ms |%9" PRIu64 " | avg:%9.3f ms | max:%9.3f ms | max at: %13.6f s\n", |
39aeb52f | 1198 | (double)work_list->total_runtime / 1e6, |
1199 | work_list->nb_atoms, (double)avg / 1e6, | |
3786310a FW |
1200 | (double)work_list->max_lat / 1e6, |
1201 | (double)work_list->max_lat_at / 1e9); | |
cdce9d73 FW |
1202 | } |
1203 | ||
39aeb52f | 1204 | static int pid_cmp(struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 | 1205 | { |
38051234 | 1206 | if (l->thread->tid < r->thread->tid) |
daa1d7a5 | 1207 | return -1; |
38051234 | 1208 | if (l->thread->tid > r->thread->tid) |
daa1d7a5 FW |
1209 | return 1; |
1210 | ||
1211 | return 0; | |
1212 | } | |
1213 | ||
39aeb52f | 1214 | static int avg_cmp(struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 FW |
1215 | { |
1216 | u64 avgl, avgr; | |
1217 | ||
1218 | if (!l->nb_atoms) | |
1219 | return -1; | |
1220 | ||
1221 | if (!r->nb_atoms) | |
1222 | return 1; | |
1223 | ||
1224 | avgl = l->total_lat / l->nb_atoms; | |
1225 | avgr = r->total_lat / r->nb_atoms; | |
1226 | ||
1227 | if (avgl < avgr) | |
1228 | return -1; | |
1229 | if (avgl > avgr) | |
1230 | return 1; | |
1231 | ||
1232 | return 0; | |
1233 | } | |
1234 | ||
39aeb52f | 1235 | static int max_cmp(struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 FW |
1236 | { |
1237 | if (l->max_lat < r->max_lat) | |
1238 | return -1; | |
1239 | if (l->max_lat > r->max_lat) | |
1240 | return 1; | |
1241 | ||
1242 | return 0; | |
1243 | } | |
1244 | ||
39aeb52f | 1245 | static int switch_cmp(struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 FW |
1246 | { |
1247 | if (l->nb_atoms < r->nb_atoms) | |
1248 | return -1; | |
1249 | if (l->nb_atoms > r->nb_atoms) | |
1250 | return 1; | |
1251 | ||
1252 | return 0; | |
1253 | } | |
1254 | ||
39aeb52f | 1255 | static int runtime_cmp(struct work_atoms *l, struct work_atoms *r) |
daa1d7a5 FW |
1256 | { |
1257 | if (l->total_runtime < r->total_runtime) | |
1258 | return -1; | |
1259 | if (l->total_runtime > r->total_runtime) | |
1260 | return 1; | |
1261 | ||
1262 | return 0; | |
1263 | } | |
1264 | ||
cbef79a8 | 1265 | static int sort_dimension__add(const char *tok, struct list_head *list) |
daa1d7a5 | 1266 | { |
0e9b07e5 ACM |
1267 | size_t i; |
1268 | static struct sort_dimension avg_sort_dimension = { | |
1269 | .name = "avg", | |
1270 | .cmp = avg_cmp, | |
1271 | }; | |
1272 | static struct sort_dimension max_sort_dimension = { | |
1273 | .name = "max", | |
1274 | .cmp = max_cmp, | |
1275 | }; | |
1276 | static struct sort_dimension pid_sort_dimension = { | |
1277 | .name = "pid", | |
1278 | .cmp = pid_cmp, | |
1279 | }; | |
1280 | static struct sort_dimension runtime_sort_dimension = { | |
1281 | .name = "runtime", | |
1282 | .cmp = runtime_cmp, | |
1283 | }; | |
1284 | static struct sort_dimension switch_sort_dimension = { | |
1285 | .name = "switch", | |
1286 | .cmp = switch_cmp, | |
1287 | }; | |
1288 | struct sort_dimension *available_sorts[] = { | |
1289 | &pid_sort_dimension, | |
1290 | &avg_sort_dimension, | |
1291 | &max_sort_dimension, | |
1292 | &switch_sort_dimension, | |
1293 | &runtime_sort_dimension, | |
1294 | }; | |
daa1d7a5 | 1295 | |
0e9b07e5 | 1296 | for (i = 0; i < ARRAY_SIZE(available_sorts); i++) { |
daa1d7a5 FW |
1297 | if (!strcmp(available_sorts[i]->name, tok)) { |
1298 | list_add_tail(&available_sorts[i]->list, list); | |
1299 | ||
1300 | return 0; | |
1301 | } | |
1302 | } | |
1303 | ||
1304 | return -1; | |
1305 | } | |
1306 | ||
0e9b07e5 | 1307 | static void perf_sched__sort_lat(struct perf_sched *sched) |
daa1d7a5 FW |
1308 | { |
1309 | struct rb_node *node; | |
2f80dd44 JB |
1310 | struct rb_root *root = &sched->atom_root; |
1311 | again: | |
daa1d7a5 | 1312 | for (;;) { |
39aeb52f | 1313 | struct work_atoms *data; |
2f80dd44 | 1314 | node = rb_first(root); |
daa1d7a5 FW |
1315 | if (!node) |
1316 | break; | |
1317 | ||
2f80dd44 | 1318 | rb_erase(node, root); |
39aeb52f | 1319 | data = rb_entry(node, struct work_atoms, node); |
0e9b07e5 | 1320 | __thread_latency_insert(&sched->sorted_atom_root, data, &sched->sort_list); |
daa1d7a5 | 1321 | } |
2f80dd44 JB |
1322 | if (root == &sched->atom_root) { |
1323 | root = &sched->merged_atom_root; | |
1324 | goto again; | |
1325 | } | |
daa1d7a5 FW |
1326 | } |
1327 | ||
0e9b07e5 | 1328 | static int process_sched_wakeup_event(struct perf_tool *tool, |
2b7fcbc5 | 1329 | struct perf_evsel *evsel, |
1d037ca1 | 1330 | struct perf_sample *sample, |
4218e673 | 1331 | struct machine *machine) |
419ab0d6 | 1332 | { |
0e9b07e5 | 1333 | struct perf_sched *sched = container_of(tool, struct perf_sched, tool); |
419ab0d6 | 1334 | |
9ec3f4e4 ACM |
1335 | if (sched->tp_handler->wakeup_event) |
1336 | return sched->tp_handler->wakeup_event(sched, evsel, sample, machine); | |
a116e05d | 1337 | |
2b7fcbc5 | 1338 | return 0; |
419ab0d6 FW |
1339 | } |
1340 | ||
9ec3f4e4 ACM |
1341 | static int map_switch_event(struct perf_sched *sched, struct perf_evsel *evsel, |
1342 | struct perf_sample *sample, struct machine *machine) | |
0ec04e16 | 1343 | { |
9d372ca5 DY |
1344 | const u32 next_pid = perf_evsel__intval(evsel, sample, "next_pid"); |
1345 | struct thread *sched_in; | |
0ec04e16 | 1346 | int new_shortname; |
7f7f8d0b | 1347 | u64 timestamp0, timestamp = sample->time; |
0ec04e16 | 1348 | s64 delta; |
7f7f8d0b | 1349 | int cpu, this_cpu = sample->cpu; |
0ec04e16 IM |
1350 | |
1351 | BUG_ON(this_cpu >= MAX_CPUS || this_cpu < 0); | |
1352 | ||
0e9b07e5 ACM |
1353 | if (this_cpu > sched->max_cpu) |
1354 | sched->max_cpu = this_cpu; | |
0ec04e16 | 1355 | |
0e9b07e5 ACM |
1356 | timestamp0 = sched->cpu_last_switched[this_cpu]; |
1357 | sched->cpu_last_switched[this_cpu] = timestamp; | |
0ec04e16 IM |
1358 | if (timestamp0) |
1359 | delta = timestamp - timestamp0; | |
1360 | else | |
1361 | delta = 0; | |
1362 | ||
a116e05d | 1363 | if (delta < 0) { |
60b7d14a | 1364 | pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta); |
a116e05d ACM |
1365 | return -1; |
1366 | } | |
0ec04e16 | 1367 | |
1fcb8768 | 1368 | sched_in = machine__findnew_thread(machine, -1, next_pid); |
b91fc39f ACM |
1369 | if (sched_in == NULL) |
1370 | return -1; | |
0ec04e16 | 1371 | |
b91fc39f | 1372 | sched->curr_thread[this_cpu] = thread__get(sched_in); |
0ec04e16 IM |
1373 | |
1374 | printf(" "); | |
1375 | ||
1376 | new_shortname = 0; | |
1377 | if (!sched_in->shortname[0]) { | |
6bcab4e1 D |
1378 | if (!strcmp(thread__comm_str(sched_in), "swapper")) { |
1379 | /* | |
1380 | * Don't allocate a letter-number for swapper:0 | |
1381 | * as a shortname. Instead, we use '.' for it. | |
1382 | */ | |
1383 | sched_in->shortname[0] = '.'; | |
1384 | sched_in->shortname[1] = ' '; | |
0ec04e16 | 1385 | } else { |
6bcab4e1 D |
1386 | sched_in->shortname[0] = sched->next_shortname1; |
1387 | sched_in->shortname[1] = sched->next_shortname2; | |
1388 | ||
1389 | if (sched->next_shortname1 < 'Z') { | |
1390 | sched->next_shortname1++; | |
0ec04e16 | 1391 | } else { |
6bcab4e1 D |
1392 | sched->next_shortname1 = 'A'; |
1393 | if (sched->next_shortname2 < '9') | |
1394 | sched->next_shortname2++; | |
1395 | else | |
1396 | sched->next_shortname2 = '0'; | |
0ec04e16 IM |
1397 | } |
1398 | } | |
1399 | new_shortname = 1; | |
1400 | } | |
1401 | ||
0e9b07e5 | 1402 | for (cpu = 0; cpu <= sched->max_cpu; cpu++) { |
0ec04e16 IM |
1403 | if (cpu != this_cpu) |
1404 | printf(" "); | |
1405 | else | |
1406 | printf("*"); | |
1407 | ||
6bcab4e1 D |
1408 | if (sched->curr_thread[cpu]) |
1409 | printf("%2s ", sched->curr_thread[cpu]->shortname); | |
1410 | else | |
0ec04e16 IM |
1411 | printf(" "); |
1412 | } | |
1413 | ||
1414 | printf(" %12.6f secs ", (double)timestamp/1e9); | |
1415 | if (new_shortname) { | |
1416 | printf("%s => %s:%d\n", | |
b9c5143a | 1417 | sched_in->shortname, thread__comm_str(sched_in), sched_in->tid); |
0ec04e16 IM |
1418 | } else { |
1419 | printf("\n"); | |
1420 | } | |
a116e05d | 1421 | |
b91fc39f ACM |
1422 | thread__put(sched_in); |
1423 | ||
a116e05d | 1424 | return 0; |
0ec04e16 IM |
1425 | } |
1426 | ||
0e9b07e5 | 1427 | static int process_sched_switch_event(struct perf_tool *tool, |
2b7fcbc5 | 1428 | struct perf_evsel *evsel, |
1d037ca1 | 1429 | struct perf_sample *sample, |
4218e673 | 1430 | struct machine *machine) |
419ab0d6 | 1431 | { |
0e9b07e5 | 1432 | struct perf_sched *sched = container_of(tool, struct perf_sched, tool); |
a116e05d | 1433 | int this_cpu = sample->cpu, err = 0; |
2b7fcbc5 ACM |
1434 | u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"), |
1435 | next_pid = perf_evsel__intval(evsel, sample, "next_pid"); | |
419ab0d6 | 1436 | |
0e9b07e5 | 1437 | if (sched->curr_pid[this_cpu] != (u32)-1) { |
c8a37751 IM |
1438 | /* |
1439 | * Are we trying to switch away a PID that is | |
1440 | * not current? | |
1441 | */ | |
2b7fcbc5 | 1442 | if (sched->curr_pid[this_cpu] != prev_pid) |
0e9b07e5 | 1443 | sched->nr_context_switch_bugs++; |
c8a37751 | 1444 | } |
c8a37751 | 1445 | |
9ec3f4e4 ACM |
1446 | if (sched->tp_handler->switch_event) |
1447 | err = sched->tp_handler->switch_event(sched, evsel, sample, machine); | |
2b7fcbc5 ACM |
1448 | |
1449 | sched->curr_pid[this_cpu] = next_pid; | |
a116e05d | 1450 | return err; |
419ab0d6 FW |
1451 | } |
1452 | ||
0e9b07e5 | 1453 | static int process_sched_runtime_event(struct perf_tool *tool, |
2b7fcbc5 | 1454 | struct perf_evsel *evsel, |
1d037ca1 | 1455 | struct perf_sample *sample, |
4218e673 | 1456 | struct machine *machine) |
39aeb52f | 1457 | { |
0e9b07e5 | 1458 | struct perf_sched *sched = container_of(tool, struct perf_sched, tool); |
39aeb52f | 1459 | |
9ec3f4e4 ACM |
1460 | if (sched->tp_handler->runtime_event) |
1461 | return sched->tp_handler->runtime_event(sched, evsel, sample, machine); | |
a116e05d | 1462 | |
2b7fcbc5 | 1463 | return 0; |
39aeb52f | 1464 | } |
1465 | ||
cb627505 DA |
1466 | static int perf_sched__process_fork_event(struct perf_tool *tool, |
1467 | union perf_event *event, | |
1468 | struct perf_sample *sample, | |
1469 | struct machine *machine) | |
fbf94829 | 1470 | { |
0e9b07e5 | 1471 | struct perf_sched *sched = container_of(tool, struct perf_sched, tool); |
46538818 | 1472 | |
cb627505 DA |
1473 | /* run the fork event through the perf machineruy */ |
1474 | perf_event__process_fork(tool, event, sample, machine); | |
1475 | ||
1476 | /* and then run additional processing needed for this command */ | |
9ec3f4e4 | 1477 | if (sched->tp_handler->fork_event) |
cb627505 | 1478 | return sched->tp_handler->fork_event(sched, event, machine); |
a116e05d | 1479 | |
2b7fcbc5 | 1480 | return 0; |
fbf94829 IM |
1481 | } |
1482 | ||
0e9b07e5 | 1483 | static int process_sched_migrate_task_event(struct perf_tool *tool, |
2b7fcbc5 | 1484 | struct perf_evsel *evsel, |
1d037ca1 | 1485 | struct perf_sample *sample, |
4218e673 | 1486 | struct machine *machine) |
55ffb7a6 | 1487 | { |
0e9b07e5 | 1488 | struct perf_sched *sched = container_of(tool, struct perf_sched, tool); |
55ffb7a6 | 1489 | |
9ec3f4e4 ACM |
1490 | if (sched->tp_handler->migrate_task_event) |
1491 | return sched->tp_handler->migrate_task_event(sched, evsel, sample, machine); | |
a116e05d | 1492 | |
2b7fcbc5 | 1493 | return 0; |
55ffb7a6 MG |
1494 | } |
1495 | ||
a116e05d | 1496 | typedef int (*tracepoint_handler)(struct perf_tool *tool, |
2b7fcbc5 | 1497 | struct perf_evsel *evsel, |
a116e05d | 1498 | struct perf_sample *sample, |
4218e673 | 1499 | struct machine *machine); |
ec156764 | 1500 | |
1d037ca1 IT |
1501 | static int perf_sched__process_tracepoint_sample(struct perf_tool *tool __maybe_unused, |
1502 | union perf_event *event __maybe_unused, | |
ee29be62 ACM |
1503 | struct perf_sample *sample, |
1504 | struct perf_evsel *evsel, | |
1505 | struct machine *machine) | |
0a02ad93 | 1506 | { |
a116e05d | 1507 | int err = 0; |
0a02ad93 | 1508 | |
744a9719 ACM |
1509 | if (evsel->handler != NULL) { |
1510 | tracepoint_handler f = evsel->handler; | |
2b7fcbc5 | 1511 | err = f(tool, evsel, sample, machine); |
ee29be62 | 1512 | } |
0a02ad93 | 1513 | |
a116e05d | 1514 | return err; |
0a02ad93 IM |
1515 | } |
1516 | ||
ae536acf | 1517 | static int perf_sched__read_events(struct perf_sched *sched) |
0a02ad93 | 1518 | { |
ee29be62 ACM |
1519 | const struct perf_evsel_str_handler handlers[] = { |
1520 | { "sched:sched_switch", process_sched_switch_event, }, | |
1521 | { "sched:sched_stat_runtime", process_sched_runtime_event, }, | |
1522 | { "sched:sched_wakeup", process_sched_wakeup_event, }, | |
1523 | { "sched:sched_wakeup_new", process_sched_wakeup_event, }, | |
ee29be62 ACM |
1524 | { "sched:sched_migrate_task", process_sched_migrate_task_event, }, |
1525 | }; | |
da378962 | 1526 | struct perf_session *session; |
f5fc1412 JO |
1527 | struct perf_data_file file = { |
1528 | .path = input_name, | |
1529 | .mode = PERF_DATA_MODE_READ, | |
f0dd330f | 1530 | .force = sched->force, |
f5fc1412 | 1531 | }; |
ae536acf | 1532 | int rc = -1; |
da378962 | 1533 | |
f5fc1412 | 1534 | session = perf_session__new(&file, false, &sched->tool); |
a116e05d ACM |
1535 | if (session == NULL) { |
1536 | pr_debug("No Memory for session\n"); | |
1537 | return -1; | |
1538 | } | |
94c744b6 | 1539 | |
0a7e6d1b | 1540 | symbol__init(&session->header.env); |
04934106 | 1541 | |
a116e05d ACM |
1542 | if (perf_session__set_tracepoints_handlers(session, handlers)) |
1543 | goto out_delete; | |
ee29be62 | 1544 | |
cee75ac7 | 1545 | if (perf_session__has_traces(session, "record -R")) { |
b7b61cbe | 1546 | int err = perf_session__process_events(session); |
a116e05d ACM |
1547 | if (err) { |
1548 | pr_err("Failed to process events, error %d", err); | |
1549 | goto out_delete; | |
1550 | } | |
4c09bafa | 1551 | |
75be989a ACM |
1552 | sched->nr_events = session->evlist->stats.nr_events[0]; |
1553 | sched->nr_lost_events = session->evlist->stats.total_lost; | |
1554 | sched->nr_lost_chunks = session->evlist->stats.nr_events[PERF_RECORD_LOST]; | |
cee75ac7 | 1555 | } |
d549c769 | 1556 | |
ae536acf | 1557 | rc = 0; |
a116e05d ACM |
1558 | out_delete: |
1559 | perf_session__delete(session); | |
ae536acf | 1560 | return rc; |
0a02ad93 IM |
1561 | } |
1562 | ||
0e9b07e5 | 1563 | static void print_bad_events(struct perf_sched *sched) |
0ec04e16 | 1564 | { |
0e9b07e5 | 1565 | if (sched->nr_unordered_timestamps && sched->nr_timestamps) { |
0ec04e16 | 1566 | printf(" INFO: %.3f%% unordered timestamps (%ld out of %ld)\n", |
0e9b07e5 ACM |
1567 | (double)sched->nr_unordered_timestamps/(double)sched->nr_timestamps*100.0, |
1568 | sched->nr_unordered_timestamps, sched->nr_timestamps); | |
0ec04e16 | 1569 | } |
0e9b07e5 | 1570 | if (sched->nr_lost_events && sched->nr_events) { |
0ec04e16 | 1571 | printf(" INFO: %.3f%% lost events (%ld out of %ld, in %ld chunks)\n", |
0e9b07e5 ACM |
1572 | (double)sched->nr_lost_events/(double)sched->nr_events * 100.0, |
1573 | sched->nr_lost_events, sched->nr_events, sched->nr_lost_chunks); | |
0ec04e16 | 1574 | } |
0e9b07e5 | 1575 | if (sched->nr_context_switch_bugs && sched->nr_timestamps) { |
0ec04e16 | 1576 | printf(" INFO: %.3f%% context switch bugs (%ld out of %ld)", |
0e9b07e5 ACM |
1577 | (double)sched->nr_context_switch_bugs/(double)sched->nr_timestamps*100.0, |
1578 | sched->nr_context_switch_bugs, sched->nr_timestamps); | |
1579 | if (sched->nr_lost_events) | |
0ec04e16 IM |
1580 | printf(" (due to lost events?)"); |
1581 | printf("\n"); | |
1582 | } | |
1583 | } | |
1584 | ||
2f80dd44 JB |
1585 | static void __merge_work_atoms(struct rb_root *root, struct work_atoms *data) |
1586 | { | |
1587 | struct rb_node **new = &(root->rb_node), *parent = NULL; | |
1588 | struct work_atoms *this; | |
1589 | const char *comm = thread__comm_str(data->thread), *this_comm; | |
1590 | ||
1591 | while (*new) { | |
1592 | int cmp; | |
1593 | ||
1594 | this = container_of(*new, struct work_atoms, node); | |
1595 | parent = *new; | |
1596 | ||
1597 | this_comm = thread__comm_str(this->thread); | |
1598 | cmp = strcmp(comm, this_comm); | |
1599 | if (cmp > 0) { | |
1600 | new = &((*new)->rb_left); | |
1601 | } else if (cmp < 0) { | |
1602 | new = &((*new)->rb_right); | |
1603 | } else { | |
1604 | this->num_merged++; | |
1605 | this->total_runtime += data->total_runtime; | |
1606 | this->nb_atoms += data->nb_atoms; | |
1607 | this->total_lat += data->total_lat; | |
1608 | list_splice(&data->work_list, &this->work_list); | |
1609 | if (this->max_lat < data->max_lat) { | |
1610 | this->max_lat = data->max_lat; | |
1611 | this->max_lat_at = data->max_lat_at; | |
1612 | } | |
1613 | zfree(&data); | |
1614 | return; | |
1615 | } | |
1616 | } | |
1617 | ||
1618 | data->num_merged++; | |
1619 | rb_link_node(&data->node, parent, new); | |
1620 | rb_insert_color(&data->node, root); | |
1621 | } | |
1622 | ||
1623 | static void perf_sched__merge_lat(struct perf_sched *sched) | |
1624 | { | |
1625 | struct work_atoms *data; | |
1626 | struct rb_node *node; | |
1627 | ||
1628 | if (sched->skip_merge) | |
1629 | return; | |
1630 | ||
1631 | while ((node = rb_first(&sched->atom_root))) { | |
1632 | rb_erase(node, &sched->atom_root); | |
1633 | data = rb_entry(node, struct work_atoms, node); | |
1634 | __merge_work_atoms(&sched->merged_atom_root, data); | |
1635 | } | |
1636 | } | |
1637 | ||
0e9b07e5 | 1638 | static int perf_sched__lat(struct perf_sched *sched) |
0ec04e16 IM |
1639 | { |
1640 | struct rb_node *next; | |
1641 | ||
1642 | setup_pager(); | |
ad9def7c | 1643 | |
ae536acf | 1644 | if (perf_sched__read_events(sched)) |
a116e05d | 1645 | return -1; |
ad9def7c | 1646 | |
2f80dd44 | 1647 | perf_sched__merge_lat(sched); |
0e9b07e5 | 1648 | perf_sched__sort_lat(sched); |
0ec04e16 | 1649 | |
80790e0b RR |
1650 | printf("\n -----------------------------------------------------------------------------------------------------------------\n"); |
1651 | printf(" Task | Runtime ms | Switches | Average delay ms | Maximum delay ms | Maximum delay at |\n"); | |
1652 | printf(" -----------------------------------------------------------------------------------------------------------------\n"); | |
0ec04e16 | 1653 | |
0e9b07e5 | 1654 | next = rb_first(&sched->sorted_atom_root); |
0ec04e16 IM |
1655 | |
1656 | while (next) { | |
1657 | struct work_atoms *work_list; | |
1658 | ||
1659 | work_list = rb_entry(next, struct work_atoms, node); | |
0e9b07e5 | 1660 | output_lat_thread(sched, work_list); |
0ec04e16 | 1661 | next = rb_next(next); |
ae536acf | 1662 | thread__zput(work_list->thread); |
0ec04e16 IM |
1663 | } |
1664 | ||
80790e0b | 1665 | printf(" -----------------------------------------------------------------------------------------------------------------\n"); |
9486aa38 | 1666 | printf(" TOTAL: |%11.3f ms |%9" PRIu64 " |\n", |
0e9b07e5 | 1667 | (double)sched->all_runtime / 1e6, sched->all_count); |
0ec04e16 IM |
1668 | |
1669 | printf(" ---------------------------------------------------\n"); | |
1670 | ||
0e9b07e5 | 1671 | print_bad_events(sched); |
0ec04e16 IM |
1672 | printf("\n"); |
1673 | ||
a116e05d | 1674 | return 0; |
0ec04e16 IM |
1675 | } |
1676 | ||
0e9b07e5 | 1677 | static int perf_sched__map(struct perf_sched *sched) |
0ec04e16 | 1678 | { |
0e9b07e5 | 1679 | sched->max_cpu = sysconf(_SC_NPROCESSORS_CONF); |
40749d0f | 1680 | |
0ec04e16 | 1681 | setup_pager(); |
ae536acf | 1682 | if (perf_sched__read_events(sched)) |
a116e05d | 1683 | return -1; |
0e9b07e5 | 1684 | print_bad_events(sched); |
a116e05d | 1685 | return 0; |
0ec04e16 IM |
1686 | } |
1687 | ||
0e9b07e5 | 1688 | static int perf_sched__replay(struct perf_sched *sched) |
0ec04e16 IM |
1689 | { |
1690 | unsigned long i; | |
1691 | ||
0e9b07e5 ACM |
1692 | calibrate_run_measurement_overhead(sched); |
1693 | calibrate_sleep_measurement_overhead(sched); | |
0ec04e16 | 1694 | |
0e9b07e5 | 1695 | test_calibrations(sched); |
0ec04e16 | 1696 | |
ae536acf | 1697 | if (perf_sched__read_events(sched)) |
a116e05d | 1698 | return -1; |
0ec04e16 | 1699 | |
0e9b07e5 ACM |
1700 | printf("nr_run_events: %ld\n", sched->nr_run_events); |
1701 | printf("nr_sleep_events: %ld\n", sched->nr_sleep_events); | |
1702 | printf("nr_wakeup_events: %ld\n", sched->nr_wakeup_events); | |
0ec04e16 | 1703 | |
0e9b07e5 ACM |
1704 | if (sched->targetless_wakeups) |
1705 | printf("target-less wakeups: %ld\n", sched->targetless_wakeups); | |
1706 | if (sched->multitarget_wakeups) | |
1707 | printf("multi-target wakeups: %ld\n", sched->multitarget_wakeups); | |
1708 | if (sched->nr_run_events_optimized) | |
0ec04e16 | 1709 | printf("run atoms optimized: %ld\n", |
0e9b07e5 | 1710 | sched->nr_run_events_optimized); |
0ec04e16 | 1711 | |
0e9b07e5 ACM |
1712 | print_task_traces(sched); |
1713 | add_cross_task_wakeups(sched); | |
0ec04e16 | 1714 | |
0e9b07e5 | 1715 | create_tasks(sched); |
0ec04e16 | 1716 | printf("------------------------------------------------------------\n"); |
0e9b07e5 ACM |
1717 | for (i = 0; i < sched->replay_repeat; i++) |
1718 | run_one_test(sched); | |
a116e05d ACM |
1719 | |
1720 | return 0; | |
0ec04e16 IM |
1721 | } |
1722 | ||
0e9b07e5 ACM |
1723 | static void setup_sorting(struct perf_sched *sched, const struct option *options, |
1724 | const char * const usage_msg[]) | |
daa1d7a5 | 1725 | { |
0e9b07e5 | 1726 | char *tmp, *tok, *str = strdup(sched->sort_order); |
daa1d7a5 FW |
1727 | |
1728 | for (tok = strtok_r(str, ", ", &tmp); | |
1729 | tok; tok = strtok_r(NULL, ", ", &tmp)) { | |
0e9b07e5 | 1730 | if (sort_dimension__add(tok, &sched->sort_list) < 0) { |
daa1d7a5 | 1731 | error("Unknown --sort key: `%s'", tok); |
0e9b07e5 | 1732 | usage_with_options(usage_msg, options); |
daa1d7a5 FW |
1733 | } |
1734 | } | |
1735 | ||
1736 | free(str); | |
1737 | ||
0e9b07e5 | 1738 | sort_dimension__add("pid", &sched->cmp_pid); |
daa1d7a5 FW |
1739 | } |
1740 | ||
1fc35b29 IM |
1741 | static int __cmd_record(int argc, const char **argv) |
1742 | { | |
1743 | unsigned int rec_argc, i, j; | |
1744 | const char **rec_argv; | |
0e9b07e5 ACM |
1745 | const char * const record_args[] = { |
1746 | "record", | |
1747 | "-a", | |
1748 | "-R", | |
0e9b07e5 ACM |
1749 | "-m", "1024", |
1750 | "-c", "1", | |
1751 | "-e", "sched:sched_switch", | |
1752 | "-e", "sched:sched_stat_wait", | |
1753 | "-e", "sched:sched_stat_sleep", | |
1754 | "-e", "sched:sched_stat_iowait", | |
1755 | "-e", "sched:sched_stat_runtime", | |
0e9b07e5 ACM |
1756 | "-e", "sched:sched_process_fork", |
1757 | "-e", "sched:sched_wakeup", | |
7fff9597 | 1758 | "-e", "sched:sched_wakeup_new", |
0e9b07e5 ACM |
1759 | "-e", "sched:sched_migrate_task", |
1760 | }; | |
1fc35b29 IM |
1761 | |
1762 | rec_argc = ARRAY_SIZE(record_args) + argc - 1; | |
1763 | rec_argv = calloc(rec_argc + 1, sizeof(char *)); | |
1764 | ||
e462dc55 | 1765 | if (rec_argv == NULL) |
ce47dc56 CS |
1766 | return -ENOMEM; |
1767 | ||
1fc35b29 IM |
1768 | for (i = 0; i < ARRAY_SIZE(record_args); i++) |
1769 | rec_argv[i] = strdup(record_args[i]); | |
1770 | ||
1771 | for (j = 1; j < (unsigned int)argc; j++, i++) | |
1772 | rec_argv[i] = argv[j]; | |
1773 | ||
1774 | BUG_ON(i != rec_argc); | |
1775 | ||
1776 | return cmd_record(i, rec_argv, NULL); | |
1777 | } | |
1778 | ||
1d037ca1 | 1779 | int cmd_sched(int argc, const char **argv, const char *prefix __maybe_unused) |
0a02ad93 | 1780 | { |
8a39df8f AH |
1781 | const char default_sort_order[] = "avg, max, switch, runtime"; |
1782 | struct perf_sched sched = { | |
1783 | .tool = { | |
1784 | .sample = perf_sched__process_tracepoint_sample, | |
1785 | .comm = perf_event__process_comm, | |
1786 | .lost = perf_event__process_lost, | |
1787 | .fork = perf_sched__process_fork_event, | |
0a8cb85c | 1788 | .ordered_events = true, |
8a39df8f AH |
1789 | }, |
1790 | .cmp_pid = LIST_HEAD_INIT(sched.cmp_pid), | |
1791 | .sort_list = LIST_HEAD_INIT(sched.sort_list), | |
1792 | .start_work_mutex = PTHREAD_MUTEX_INITIALIZER, | |
1793 | .work_done_wait_mutex = PTHREAD_MUTEX_INITIALIZER, | |
8a39df8f AH |
1794 | .sort_order = default_sort_order, |
1795 | .replay_repeat = 10, | |
1796 | .profile_cpu = -1, | |
1797 | .next_shortname1 = 'A', | |
1798 | .next_shortname2 = '0', | |
2f80dd44 | 1799 | .skip_merge = 0, |
8a39df8f | 1800 | }; |
0e9b07e5 ACM |
1801 | const struct option latency_options[] = { |
1802 | OPT_STRING('s', "sort", &sched.sort_order, "key[,key2...]", | |
1803 | "sort by key(s): runtime, switch, avg, max"), | |
1804 | OPT_INCR('v', "verbose", &verbose, | |
1805 | "be more verbose (show symbol address, etc)"), | |
1806 | OPT_INTEGER('C', "CPU", &sched.profile_cpu, | |
1807 | "CPU to profile on"), | |
1808 | OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, | |
1809 | "dump raw trace in ASCII"), | |
2f80dd44 JB |
1810 | OPT_BOOLEAN('p', "pids", &sched.skip_merge, |
1811 | "latency stats per pid instead of per comm"), | |
0e9b07e5 ACM |
1812 | OPT_END() |
1813 | }; | |
1814 | const struct option replay_options[] = { | |
1815 | OPT_UINTEGER('r', "repeat", &sched.replay_repeat, | |
1816 | "repeat the workload replay N times (-1: infinite)"), | |
1817 | OPT_INCR('v', "verbose", &verbose, | |
1818 | "be more verbose (show symbol address, etc)"), | |
1819 | OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, | |
1820 | "dump raw trace in ASCII"), | |
939cda52 | 1821 | OPT_BOOLEAN('f', "force", &sched.force, "don't complain, do it"), |
0e9b07e5 ACM |
1822 | OPT_END() |
1823 | }; | |
1824 | const struct option sched_options[] = { | |
70cb4e96 | 1825 | OPT_STRING('i', "input", &input_name, "file", |
0e9b07e5 ACM |
1826 | "input file name"), |
1827 | OPT_INCR('v', "verbose", &verbose, | |
1828 | "be more verbose (show symbol address, etc)"), | |
1829 | OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, | |
1830 | "dump raw trace in ASCII"), | |
1831 | OPT_END() | |
1832 | }; | |
1833 | const char * const latency_usage[] = { | |
1834 | "perf sched latency [<options>]", | |
1835 | NULL | |
1836 | }; | |
1837 | const char * const replay_usage[] = { | |
1838 | "perf sched replay [<options>]", | |
1839 | NULL | |
1840 | }; | |
a83edb2d RR |
1841 | const char *const sched_subcommands[] = { "record", "latency", "map", |
1842 | "replay", "script", NULL }; | |
1843 | const char *sched_usage[] = { | |
1844 | NULL, | |
0e9b07e5 ACM |
1845 | NULL |
1846 | }; | |
1847 | struct trace_sched_handler lat_ops = { | |
1848 | .wakeup_event = latency_wakeup_event, | |
1849 | .switch_event = latency_switch_event, | |
1850 | .runtime_event = latency_runtime_event, | |
0e9b07e5 ACM |
1851 | .migrate_task_event = latency_migrate_task_event, |
1852 | }; | |
1853 | struct trace_sched_handler map_ops = { | |
1854 | .switch_event = map_switch_event, | |
1855 | }; | |
1856 | struct trace_sched_handler replay_ops = { | |
1857 | .wakeup_event = replay_wakeup_event, | |
1858 | .switch_event = replay_switch_event, | |
1859 | .fork_event = replay_fork_event, | |
1860 | }; | |
156a2b02 AH |
1861 | unsigned int i; |
1862 | ||
1863 | for (i = 0; i < ARRAY_SIZE(sched.curr_pid); i++) | |
1864 | sched.curr_pid[i] = -1; | |
0e9b07e5 | 1865 | |
a83edb2d RR |
1866 | argc = parse_options_subcommand(argc, argv, sched_options, sched_subcommands, |
1867 | sched_usage, PARSE_OPT_STOP_AT_NON_OPTION); | |
f2858d8a IM |
1868 | if (!argc) |
1869 | usage_with_options(sched_usage, sched_options); | |
0a02ad93 | 1870 | |
c0777c5a | 1871 | /* |
133dc4c3 | 1872 | * Aliased to 'perf script' for now: |
c0777c5a | 1873 | */ |
133dc4c3 IM |
1874 | if (!strcmp(argv[0], "script")) |
1875 | return cmd_script(argc, argv, prefix); | |
c0777c5a | 1876 | |
1fc35b29 IM |
1877 | if (!strncmp(argv[0], "rec", 3)) { |
1878 | return __cmd_record(argc, argv); | |
1879 | } else if (!strncmp(argv[0], "lat", 3)) { | |
0e9b07e5 | 1880 | sched.tp_handler = &lat_ops; |
f2858d8a IM |
1881 | if (argc > 1) { |
1882 | argc = parse_options(argc, argv, latency_options, latency_usage, 0); | |
1883 | if (argc) | |
1884 | usage_with_options(latency_usage, latency_options); | |
f2858d8a | 1885 | } |
0e9b07e5 ACM |
1886 | setup_sorting(&sched, latency_options, latency_usage); |
1887 | return perf_sched__lat(&sched); | |
0ec04e16 | 1888 | } else if (!strcmp(argv[0], "map")) { |
0e9b07e5 ACM |
1889 | sched.tp_handler = &map_ops; |
1890 | setup_sorting(&sched, latency_options, latency_usage); | |
1891 | return perf_sched__map(&sched); | |
f2858d8a | 1892 | } else if (!strncmp(argv[0], "rep", 3)) { |
0e9b07e5 | 1893 | sched.tp_handler = &replay_ops; |
f2858d8a IM |
1894 | if (argc) { |
1895 | argc = parse_options(argc, argv, replay_options, replay_usage, 0); | |
1896 | if (argc) | |
1897 | usage_with_options(replay_usage, replay_options); | |
1898 | } | |
0e9b07e5 | 1899 | return perf_sched__replay(&sched); |
f2858d8a IM |
1900 | } else { |
1901 | usage_with_options(sched_usage, sched_options); | |
1902 | } | |
1903 | ||
ec156764 | 1904 | return 0; |
0a02ad93 | 1905 | } |