blob: ea9c15c0cdfe0d96fa7c489956e2beb39df77bbe [file] [log] [blame]
Ingo Molnar0a02ad92009-09-11 12:12:54 +02001#include "builtin.h"
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +02002#include "perf.h"
Ingo Molnar0a02ad92009-09-11 12:12:54 +02003
4#include "util/util.h"
5#include "util/cache.h"
6#include "util/symbol.h"
7#include "util/thread.h"
8#include "util/header.h"
9
10#include "util/parse-options.h"
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +020011#include "util/trace-event.h"
Ingo Molnar0a02ad92009-09-11 12:12:54 +020012
Ingo Molnar0a02ad92009-09-11 12:12:54 +020013#include "util/debug.h"
14
Ingo Molnarec156762009-09-11 12:12:54 +020015#include <sys/types.h>
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +020016#include <sys/prctl.h>
Ingo Molnar0a02ad92009-09-11 12:12:54 +020017
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +020018#include <semaphore.h>
19#include <pthread.h>
20#include <math.h>
Frederic Weisbecker419ab0d2009-09-12 03:59:01 +020021
Ingo Molnarec156762009-09-11 12:12:54 +020022static char const *input_name = "perf.data";
23static int input;
24static unsigned long page_size;
25static unsigned long mmap_window = 32;
Ingo Molnar0a02ad92009-09-11 12:12:54 +020026
Ingo Molnarec156762009-09-11 12:12:54 +020027static unsigned long total_comm = 0;
Ingo Molnar0a02ad92009-09-11 12:12:54 +020028
Ingo Molnarec156762009-09-11 12:12:54 +020029static struct rb_root threads;
30static struct thread *last_match;
Ingo Molnar0a02ad92009-09-11 12:12:54 +020031
Ingo Molnarec156762009-09-11 12:12:54 +020032static struct perf_header *header;
33static u64 sample_type;
Ingo Molnar0a02ad92009-09-11 12:12:54 +020034
Frederic Weisbeckerdaa1d7a2009-09-13 03:36:29 +020035static char default_sort_order[] = "avg, max, switch, runtime";
36static char *sort_order = default_sort_order;
37
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +020038#define PR_SET_NAME 15 /* Set process name */
39#define MAX_CPUS 4096
Ingo Molnar0a02ad92009-09-11 12:12:54 +020040
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +020041#define BUG_ON(x) assert(!(x))
Ingo Molnarec156762009-09-11 12:12:54 +020042
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +020043static u64 run_measurement_overhead;
44static u64 sleep_measurement_overhead;
Ingo Molnarec156762009-09-11 12:12:54 +020045
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +020046#define COMM_LEN 20
47#define SYM_LEN 129
Ingo Molnarec156762009-09-11 12:12:54 +020048
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +020049#define MAX_PID 65536
Ingo Molnarec156762009-09-11 12:12:54 +020050
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +020051static unsigned long nr_tasks;
Ingo Molnarec156762009-09-11 12:12:54 +020052
mingo39aeb522009-09-14 20:04:48 +020053struct sched_atom;
Ingo Molnarec156762009-09-11 12:12:54 +020054
55struct task_desc {
56 unsigned long nr;
57 unsigned long pid;
58 char comm[COMM_LEN];
59
60 unsigned long nr_events;
61 unsigned long curr_event;
mingo39aeb522009-09-14 20:04:48 +020062 struct sched_atom **atoms;
Ingo Molnarec156762009-09-11 12:12:54 +020063
64 pthread_t thread;
65 sem_t sleep_sem;
66
67 sem_t ready_for_work;
68 sem_t work_done_sem;
69
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +020070 u64 cpu_usage;
Ingo Molnarec156762009-09-11 12:12:54 +020071};
72
73enum sched_event_type {
74 SCHED_EVENT_RUN,
75 SCHED_EVENT_SLEEP,
76 SCHED_EVENT_WAKEUP,
77};
78
mingo39aeb522009-09-14 20:04:48 +020079struct sched_atom {
Ingo Molnarec156762009-09-11 12:12:54 +020080 enum sched_event_type type;
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +020081 u64 timestamp;
82 u64 duration;
Ingo Molnarec156762009-09-11 12:12:54 +020083 unsigned long nr;
84 int specific_wait;
85 sem_t *wait_sem;
86 struct task_desc *wakee;
87};
88
89static struct task_desc *pid_to_task[MAX_PID];
90
91static struct task_desc **tasks;
92
93static pthread_mutex_t start_work_mutex = PTHREAD_MUTEX_INITIALIZER;
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +020094static u64 start_time;
Ingo Molnarec156762009-09-11 12:12:54 +020095
96static pthread_mutex_t work_done_wait_mutex = PTHREAD_MUTEX_INITIALIZER;
97
98static unsigned long nr_run_events;
99static unsigned long nr_sleep_events;
100static unsigned long nr_wakeup_events;
101
102static unsigned long nr_sleep_corrections;
103static unsigned long nr_run_events_optimized;
104
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +0200105static unsigned long targetless_wakeups;
106static unsigned long multitarget_wakeups;
107
108static u64 cpu_usage;
109static u64 runavg_cpu_usage;
110static u64 parent_cpu_usage;
111static u64 runavg_parent_cpu_usage;
112
113static unsigned long nr_runs;
114static u64 sum_runtime;
115static u64 sum_fluct;
116static u64 run_avg;
117
118static unsigned long replay_repeat = 10;
Ingo Molnarea57c4f2009-09-13 18:15:54 +0200119static unsigned long nr_timestamps;
Ingo Molnardc02bf72009-09-16 13:45:00 +0200120static unsigned long nr_unordered_timestamps;
121static unsigned long nr_state_machine_bugs;
Ingo Molnarc8a37752009-09-16 14:07:00 +0200122static unsigned long nr_context_switch_bugs;
Ingo Molnardc02bf72009-09-16 13:45:00 +0200123static unsigned long nr_events;
124static unsigned long nr_lost_chunks;
125static unsigned long nr_lost_events;
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +0200126
127#define TASK_STATE_TO_CHAR_STR "RSDTtZX"
128
129enum thread_state {
130 THREAD_SLEEPING = 0,
131 THREAD_WAIT_CPU,
132 THREAD_SCHED_IN,
133 THREAD_IGNORE
134};
135
136struct work_atom {
137 struct list_head list;
138 enum thread_state state;
Frederic Weisbeckeraa1ab9d2009-09-14 03:01:12 +0200139 u64 sched_out_time;
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +0200140 u64 wake_up_time;
141 u64 sched_in_time;
142 u64 runtime;
143};
144
mingo39aeb522009-09-14 20:04:48 +0200145struct work_atoms {
146 struct list_head work_list;
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +0200147 struct thread *thread;
148 struct rb_node node;
149 u64 max_lat;
150 u64 total_lat;
151 u64 nb_atoms;
152 u64 total_runtime;
153};
154
mingo39aeb522009-09-14 20:04:48 +0200155typedef int (*sort_fn_t)(struct work_atoms *, struct work_atoms *);
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +0200156
157static struct rb_root atom_root, sorted_atom_root;
158
159static u64 all_runtime;
160static u64 all_count;
161
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +0200162
163static u64 get_nsecs(void)
164{
165 struct timespec ts;
166
167 clock_gettime(CLOCK_MONOTONIC, &ts);
168
169 return ts.tv_sec * 1000000000ULL + ts.tv_nsec;
170}
171
172static void burn_nsecs(u64 nsecs)
173{
174 u64 T0 = get_nsecs(), T1;
175
176 do {
177 T1 = get_nsecs();
178 } while (T1 + run_measurement_overhead < T0 + nsecs);
179}
180
181static void sleep_nsecs(u64 nsecs)
182{
183 struct timespec ts;
184
185 ts.tv_nsec = nsecs % 999999999;
186 ts.tv_sec = nsecs / 999999999;
187
188 nanosleep(&ts, NULL);
189}
190
191static void calibrate_run_measurement_overhead(void)
192{
193 u64 T0, T1, delta, min_delta = 1000000000ULL;
194 int i;
195
196 for (i = 0; i < 10; i++) {
197 T0 = get_nsecs();
198 burn_nsecs(0);
199 T1 = get_nsecs();
200 delta = T1-T0;
201 min_delta = min(min_delta, delta);
202 }
203 run_measurement_overhead = min_delta;
204
205 printf("run measurement overhead: %Ld nsecs\n", min_delta);
206}
207
208static void calibrate_sleep_measurement_overhead(void)
209{
210 u64 T0, T1, delta, min_delta = 1000000000ULL;
211 int i;
212
213 for (i = 0; i < 10; i++) {
214 T0 = get_nsecs();
215 sleep_nsecs(10000);
216 T1 = get_nsecs();
217 delta = T1-T0;
218 min_delta = min(min_delta, delta);
219 }
220 min_delta -= 10000;
221 sleep_measurement_overhead = min_delta;
222
223 printf("sleep measurement overhead: %Ld nsecs\n", min_delta);
224}
225
mingo39aeb522009-09-14 20:04:48 +0200226static struct sched_atom *
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +0200227get_new_event(struct task_desc *task, u64 timestamp)
Ingo Molnarec156762009-09-11 12:12:54 +0200228{
mingo39aeb522009-09-14 20:04:48 +0200229 struct sched_atom *event = calloc(1, sizeof(*event));
Ingo Molnarec156762009-09-11 12:12:54 +0200230 unsigned long idx = task->nr_events;
231 size_t size;
232
233 event->timestamp = timestamp;
234 event->nr = idx;
235
236 task->nr_events++;
mingo39aeb522009-09-14 20:04:48 +0200237 size = sizeof(struct sched_atom *) * task->nr_events;
238 task->atoms = realloc(task->atoms, size);
239 BUG_ON(!task->atoms);
Ingo Molnarec156762009-09-11 12:12:54 +0200240
mingo39aeb522009-09-14 20:04:48 +0200241 task->atoms[idx] = event;
Ingo Molnarec156762009-09-11 12:12:54 +0200242
243 return event;
244}
245
mingo39aeb522009-09-14 20:04:48 +0200246static struct sched_atom *last_event(struct task_desc *task)
Ingo Molnarec156762009-09-11 12:12:54 +0200247{
248 if (!task->nr_events)
249 return NULL;
250
mingo39aeb522009-09-14 20:04:48 +0200251 return task->atoms[task->nr_events - 1];
Ingo Molnarec156762009-09-11 12:12:54 +0200252}
253
254static void
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +0200255add_sched_event_run(struct task_desc *task, u64 timestamp, u64 duration)
Ingo Molnarec156762009-09-11 12:12:54 +0200256{
mingo39aeb522009-09-14 20:04:48 +0200257 struct sched_atom *event, *curr_event = last_event(task);
Ingo Molnarec156762009-09-11 12:12:54 +0200258
259 /*
Ingo Molnarfbf94822009-09-11 12:12:54 +0200260 * optimize an existing RUN event by merging this one
261 * to it:
262 */
Ingo Molnarec156762009-09-11 12:12:54 +0200263 if (curr_event && curr_event->type == SCHED_EVENT_RUN) {
264 nr_run_events_optimized++;
265 curr_event->duration += duration;
266 return;
267 }
268
269 event = get_new_event(task, timestamp);
270
271 event->type = SCHED_EVENT_RUN;
272 event->duration = duration;
273
274 nr_run_events++;
275}
276
Ingo Molnarec156762009-09-11 12:12:54 +0200277static void
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +0200278add_sched_event_wakeup(struct task_desc *task, u64 timestamp,
Ingo Molnarec156762009-09-11 12:12:54 +0200279 struct task_desc *wakee)
280{
mingo39aeb522009-09-14 20:04:48 +0200281 struct sched_atom *event, *wakee_event;
Ingo Molnarec156762009-09-11 12:12:54 +0200282
283 event = get_new_event(task, timestamp);
284 event->type = SCHED_EVENT_WAKEUP;
285 event->wakee = wakee;
286
287 wakee_event = last_event(wakee);
288 if (!wakee_event || wakee_event->type != SCHED_EVENT_SLEEP) {
289 targetless_wakeups++;
290 return;
291 }
292 if (wakee_event->wait_sem) {
293 multitarget_wakeups++;
294 return;
295 }
296
297 wakee_event->wait_sem = calloc(1, sizeof(*wakee_event->wait_sem));
298 sem_init(wakee_event->wait_sem, 0, 0);
299 wakee_event->specific_wait = 1;
300 event->wait_sem = wakee_event->wait_sem;
301
302 nr_wakeup_events++;
303}
304
305static void
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +0200306add_sched_event_sleep(struct task_desc *task, u64 timestamp,
Ingo Molnarad236fd2009-09-11 12:12:54 +0200307 u64 task_state __used)
Ingo Molnarec156762009-09-11 12:12:54 +0200308{
mingo39aeb522009-09-14 20:04:48 +0200309 struct sched_atom *event = get_new_event(task, timestamp);
Ingo Molnarec156762009-09-11 12:12:54 +0200310
311 event->type = SCHED_EVENT_SLEEP;
312
313 nr_sleep_events++;
314}
315
316static struct task_desc *register_pid(unsigned long pid, const char *comm)
317{
318 struct task_desc *task;
319
320 BUG_ON(pid >= MAX_PID);
321
322 task = pid_to_task[pid];
323
324 if (task)
325 return task;
326
327 task = calloc(1, sizeof(*task));
328 task->pid = pid;
329 task->nr = nr_tasks;
330 strcpy(task->comm, comm);
331 /*
332 * every task starts in sleeping state - this gets ignored
333 * if there's no wakeup pointing to this sleep state:
334 */
335 add_sched_event_sleep(task, 0, 0);
336
337 pid_to_task[pid] = task;
338 nr_tasks++;
339 tasks = realloc(tasks, nr_tasks*sizeof(struct task_task *));
340 BUG_ON(!tasks);
341 tasks[task->nr] = task;
342
Ingo Molnarad236fd2009-09-11 12:12:54 +0200343 if (verbose)
344 printf("registered task #%ld, PID %ld (%s)\n", nr_tasks, pid, comm);
Ingo Molnarec156762009-09-11 12:12:54 +0200345
346 return task;
347}
348
349
Ingo Molnarec156762009-09-11 12:12:54 +0200350static void print_task_traces(void)
351{
352 struct task_desc *task;
353 unsigned long i;
354
355 for (i = 0; i < nr_tasks; i++) {
356 task = tasks[i];
Ingo Molnarad236fd2009-09-11 12:12:54 +0200357 printf("task %6ld (%20s:%10ld), nr_events: %ld\n",
Ingo Molnarec156762009-09-11 12:12:54 +0200358 task->nr, task->comm, task->pid, task->nr_events);
359 }
360}
361
362static void add_cross_task_wakeups(void)
363{
364 struct task_desc *task1, *task2;
365 unsigned long i, j;
366
367 for (i = 0; i < nr_tasks; i++) {
368 task1 = tasks[i];
369 j = i + 1;
370 if (j == nr_tasks)
371 j = 0;
372 task2 = tasks[j];
373 add_sched_event_wakeup(task1, 0, task2);
374 }
375}
376
377static void
mingo39aeb522009-09-14 20:04:48 +0200378process_sched_event(struct task_desc *this_task __used, struct sched_atom *atom)
Ingo Molnarec156762009-09-11 12:12:54 +0200379{
380 int ret = 0;
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +0200381 u64 now;
Ingo Molnarec156762009-09-11 12:12:54 +0200382 long long delta;
383
384 now = get_nsecs();
mingo39aeb522009-09-14 20:04:48 +0200385 delta = start_time + atom->timestamp - now;
Ingo Molnarec156762009-09-11 12:12:54 +0200386
mingo39aeb522009-09-14 20:04:48 +0200387 switch (atom->type) {
Ingo Molnarec156762009-09-11 12:12:54 +0200388 case SCHED_EVENT_RUN:
mingo39aeb522009-09-14 20:04:48 +0200389 burn_nsecs(atom->duration);
Ingo Molnarec156762009-09-11 12:12:54 +0200390 break;
391 case SCHED_EVENT_SLEEP:
mingo39aeb522009-09-14 20:04:48 +0200392 if (atom->wait_sem)
393 ret = sem_wait(atom->wait_sem);
Ingo Molnarec156762009-09-11 12:12:54 +0200394 BUG_ON(ret);
395 break;
396 case SCHED_EVENT_WAKEUP:
mingo39aeb522009-09-14 20:04:48 +0200397 if (atom->wait_sem)
398 ret = sem_post(atom->wait_sem);
Ingo Molnarec156762009-09-11 12:12:54 +0200399 BUG_ON(ret);
400 break;
401 default:
402 BUG_ON(1);
403 }
404}
405
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +0200406static u64 get_cpu_usage_nsec_parent(void)
Ingo Molnarec156762009-09-11 12:12:54 +0200407{
408 struct rusage ru;
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +0200409 u64 sum;
Ingo Molnarec156762009-09-11 12:12:54 +0200410 int err;
411
412 err = getrusage(RUSAGE_SELF, &ru);
413 BUG_ON(err);
414
415 sum = ru.ru_utime.tv_sec*1e9 + ru.ru_utime.tv_usec*1e3;
416 sum += ru.ru_stime.tv_sec*1e9 + ru.ru_stime.tv_usec*1e3;
417
418 return sum;
419}
420
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +0200421static u64 get_cpu_usage_nsec_self(void)
Ingo Molnarec156762009-09-11 12:12:54 +0200422{
423 char filename [] = "/proc/1234567890/sched";
424 unsigned long msecs, nsecs;
425 char *line = NULL;
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +0200426 u64 total = 0;
Ingo Molnarec156762009-09-11 12:12:54 +0200427 size_t len = 0;
428 ssize_t chars;
429 FILE *file;
430 int ret;
431
432 sprintf(filename, "/proc/%d/sched", getpid());
433 file = fopen(filename, "r");
434 BUG_ON(!file);
435
436 while ((chars = getline(&line, &len, file)) != -1) {
Ingo Molnarec156762009-09-11 12:12:54 +0200437 ret = sscanf(line, "se.sum_exec_runtime : %ld.%06ld\n",
438 &msecs, &nsecs);
439 if (ret == 2) {
440 total = msecs*1e6 + nsecs;
Ingo Molnarec156762009-09-11 12:12:54 +0200441 break;
442 }
443 }
444 if (line)
445 free(line);
446 fclose(file);
447
448 return total;
449}
450
451static void *thread_func(void *ctx)
452{
453 struct task_desc *this_task = ctx;
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +0200454 u64 cpu_usage_0, cpu_usage_1;
Ingo Molnarec156762009-09-11 12:12:54 +0200455 unsigned long i, ret;
456 char comm2[22];
457
Ingo Molnarec156762009-09-11 12:12:54 +0200458 sprintf(comm2, ":%s", this_task->comm);
459 prctl(PR_SET_NAME, comm2);
460
461again:
462 ret = sem_post(&this_task->ready_for_work);
463 BUG_ON(ret);
Ingo Molnarec156762009-09-11 12:12:54 +0200464 ret = pthread_mutex_lock(&start_work_mutex);
465 BUG_ON(ret);
466 ret = pthread_mutex_unlock(&start_work_mutex);
467 BUG_ON(ret);
Ingo Molnarec156762009-09-11 12:12:54 +0200468
469 cpu_usage_0 = get_cpu_usage_nsec_self();
470
471 for (i = 0; i < this_task->nr_events; i++) {
472 this_task->curr_event = i;
mingo39aeb522009-09-14 20:04:48 +0200473 process_sched_event(this_task, this_task->atoms[i]);
Ingo Molnarec156762009-09-11 12:12:54 +0200474 }
475
476 cpu_usage_1 = get_cpu_usage_nsec_self();
477 this_task->cpu_usage = cpu_usage_1 - cpu_usage_0;
478
Ingo Molnarec156762009-09-11 12:12:54 +0200479 ret = sem_post(&this_task->work_done_sem);
480 BUG_ON(ret);
Ingo Molnarec156762009-09-11 12:12:54 +0200481
482 ret = pthread_mutex_lock(&work_done_wait_mutex);
483 BUG_ON(ret);
484 ret = pthread_mutex_unlock(&work_done_wait_mutex);
485 BUG_ON(ret);
Ingo Molnarec156762009-09-11 12:12:54 +0200486
487 goto again;
488}
489
490static void create_tasks(void)
491{
492 struct task_desc *task;
493 pthread_attr_t attr;
494 unsigned long i;
495 int err;
496
497 err = pthread_attr_init(&attr);
498 BUG_ON(err);
499 err = pthread_attr_setstacksize(&attr, (size_t)(16*1024));
500 BUG_ON(err);
501 err = pthread_mutex_lock(&start_work_mutex);
502 BUG_ON(err);
503 err = pthread_mutex_lock(&work_done_wait_mutex);
504 BUG_ON(err);
505 for (i = 0; i < nr_tasks; i++) {
506 task = tasks[i];
507 sem_init(&task->sleep_sem, 0, 0);
508 sem_init(&task->ready_for_work, 0, 0);
509 sem_init(&task->work_done_sem, 0, 0);
510 task->curr_event = 0;
511 err = pthread_create(&task->thread, &attr, thread_func, task);
512 BUG_ON(err);
513 }
514}
515
Ingo Molnarec156762009-09-11 12:12:54 +0200516static void wait_for_tasks(void)
517{
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +0200518 u64 cpu_usage_0, cpu_usage_1;
Ingo Molnarec156762009-09-11 12:12:54 +0200519 struct task_desc *task;
520 unsigned long i, ret;
521
Ingo Molnarec156762009-09-11 12:12:54 +0200522 start_time = get_nsecs();
Ingo Molnarec156762009-09-11 12:12:54 +0200523 cpu_usage = 0;
524 pthread_mutex_unlock(&work_done_wait_mutex);
525
526 for (i = 0; i < nr_tasks; i++) {
527 task = tasks[i];
528 ret = sem_wait(&task->ready_for_work);
529 BUG_ON(ret);
530 sem_init(&task->ready_for_work, 0, 0);
531 }
532 ret = pthread_mutex_lock(&work_done_wait_mutex);
533 BUG_ON(ret);
534
535 cpu_usage_0 = get_cpu_usage_nsec_parent();
536
537 pthread_mutex_unlock(&start_work_mutex);
538
Ingo Molnarec156762009-09-11 12:12:54 +0200539 for (i = 0; i < nr_tasks; i++) {
540 task = tasks[i];
541 ret = sem_wait(&task->work_done_sem);
542 BUG_ON(ret);
543 sem_init(&task->work_done_sem, 0, 0);
544 cpu_usage += task->cpu_usage;
545 task->cpu_usage = 0;
546 }
547
548 cpu_usage_1 = get_cpu_usage_nsec_parent();
549 if (!runavg_cpu_usage)
550 runavg_cpu_usage = cpu_usage;
551 runavg_cpu_usage = (runavg_cpu_usage*9 + cpu_usage)/10;
552
553 parent_cpu_usage = cpu_usage_1 - cpu_usage_0;
554 if (!runavg_parent_cpu_usage)
555 runavg_parent_cpu_usage = parent_cpu_usage;
556 runavg_parent_cpu_usage = (runavg_parent_cpu_usage*9 +
557 parent_cpu_usage)/10;
558
559 ret = pthread_mutex_lock(&start_work_mutex);
560 BUG_ON(ret);
561
562 for (i = 0; i < nr_tasks; i++) {
563 task = tasks[i];
564 sem_init(&task->sleep_sem, 0, 0);
565 task->curr_event = 0;
566 }
567}
568
Ingo Molnarec156762009-09-11 12:12:54 +0200569static void run_one_test(void)
570{
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +0200571 u64 T0, T1, delta, avg_delta, fluct, std_dev;
Ingo Molnarec156762009-09-11 12:12:54 +0200572
573 T0 = get_nsecs();
574 wait_for_tasks();
575 T1 = get_nsecs();
576
577 delta = T1 - T0;
578 sum_runtime += delta;
579 nr_runs++;
580
581 avg_delta = sum_runtime / nr_runs;
582 if (delta < avg_delta)
583 fluct = avg_delta - delta;
584 else
585 fluct = delta - avg_delta;
586 sum_fluct += fluct;
587 std_dev = sum_fluct / nr_runs / sqrt(nr_runs);
588 if (!run_avg)
589 run_avg = delta;
590 run_avg = (run_avg*9 + delta)/10;
591
Ingo Molnarad236fd2009-09-11 12:12:54 +0200592 printf("#%-3ld: %0.3f, ",
Ingo Molnarec156762009-09-11 12:12:54 +0200593 nr_runs, (double)delta/1000000.0);
594
Ingo Molnarad236fd2009-09-11 12:12:54 +0200595 printf("ravg: %0.2f, ",
Ingo Molnarec156762009-09-11 12:12:54 +0200596 (double)run_avg/1e6);
597
Ingo Molnarad236fd2009-09-11 12:12:54 +0200598 printf("cpu: %0.2f / %0.2f",
Ingo Molnarec156762009-09-11 12:12:54 +0200599 (double)cpu_usage/1e6, (double)runavg_cpu_usage/1e6);
600
601#if 0
602 /*
Ingo Molnarfbf94822009-09-11 12:12:54 +0200603 * rusage statistics done by the parent, these are less
604 * accurate than the sum_exec_runtime based statistics:
605 */
Ingo Molnarad236fd2009-09-11 12:12:54 +0200606 printf(" [%0.2f / %0.2f]",
Ingo Molnarec156762009-09-11 12:12:54 +0200607 (double)parent_cpu_usage/1e6,
608 (double)runavg_parent_cpu_usage/1e6);
609#endif
610
Ingo Molnarad236fd2009-09-11 12:12:54 +0200611 printf("\n");
Ingo Molnarec156762009-09-11 12:12:54 +0200612
613 if (nr_sleep_corrections)
Ingo Molnarad236fd2009-09-11 12:12:54 +0200614 printf(" (%ld sleep corrections)\n", nr_sleep_corrections);
Ingo Molnarec156762009-09-11 12:12:54 +0200615 nr_sleep_corrections = 0;
616}
617
618static void test_calibrations(void)
619{
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +0200620 u64 T0, T1;
Ingo Molnarec156762009-09-11 12:12:54 +0200621
622 T0 = get_nsecs();
623 burn_nsecs(1e6);
624 T1 = get_nsecs();
625
Ingo Molnarad236fd2009-09-11 12:12:54 +0200626 printf("the run test took %Ld nsecs\n", T1-T0);
Ingo Molnarec156762009-09-11 12:12:54 +0200627
628 T0 = get_nsecs();
629 sleep_nsecs(1e6);
630 T1 = get_nsecs();
631
Ingo Molnarad236fd2009-09-11 12:12:54 +0200632 printf("the sleep test took %Ld nsecs\n", T1-T0);
Ingo Molnarec156762009-09-11 12:12:54 +0200633}
634
Ingo Molnar0a02ad92009-09-11 12:12:54 +0200635static int
636process_comm_event(event_t *event, unsigned long offset, unsigned long head)
637{
638 struct thread *thread;
639
640 thread = threads__findnew(event->comm.pid, &threads, &last_match);
641
Ingo Molnardc02bf72009-09-16 13:45:00 +0200642 dump_printf("%p [%p]: perf_event_comm: %s:%d\n",
Ingo Molnar0a02ad92009-09-11 12:12:54 +0200643 (void *)(offset + head),
644 (void *)(long)(event->header.size),
645 event->comm.comm, event->comm.pid);
646
647 if (thread == NULL ||
648 thread__set_comm(thread, event->comm.comm)) {
Ingo Molnardc02bf72009-09-16 13:45:00 +0200649 dump_printf("problem processing perf_event_comm, skipping event.\n");
Ingo Molnar0a02ad92009-09-11 12:12:54 +0200650 return -1;
651 }
652 total_comm++;
653
654 return 0;
655}
656
Frederic Weisbecker46538812009-09-12 02:43:45 +0200657
658struct raw_event_sample {
659 u32 size;
660 char data[0];
661};
662
663#define FILL_FIELD(ptr, field, event, data) \
664 ptr.field = (typeof(ptr.field)) raw_field_value(event, #field, data)
665
666#define FILL_ARRAY(ptr, array, event, data) \
667do { \
668 void *__array = raw_field_ptr(event, #array, data); \
669 memcpy(ptr.array, __array, sizeof(ptr.array)); \
670} while(0)
671
672#define FILL_COMMON_FIELDS(ptr, event, data) \
673do { \
674 FILL_FIELD(ptr, common_type, event, data); \
675 FILL_FIELD(ptr, common_flags, event, data); \
676 FILL_FIELD(ptr, common_preempt_count, event, data); \
677 FILL_FIELD(ptr, common_pid, event, data); \
678 FILL_FIELD(ptr, common_tgid, event, data); \
679} while (0)
680
Ingo Molnarfbf94822009-09-11 12:12:54 +0200681
Ingo Molnarec156762009-09-11 12:12:54 +0200682
Ingo Molnarfbf94822009-09-11 12:12:54 +0200683struct trace_switch_event {
684 u32 size;
685
686 u16 common_type;
687 u8 common_flags;
688 u8 common_preempt_count;
689 u32 common_pid;
690 u32 common_tgid;
691
692 char prev_comm[16];
693 u32 prev_pid;
694 u32 prev_prio;
695 u64 prev_state;
696 char next_comm[16];
697 u32 next_pid;
698 u32 next_prio;
699};
700
mingo39aeb522009-09-14 20:04:48 +0200701struct trace_runtime_event {
702 u32 size;
703
704 u16 common_type;
705 u8 common_flags;
706 u8 common_preempt_count;
707 u32 common_pid;
708 u32 common_tgid;
709
710 char comm[16];
711 u32 pid;
712 u64 runtime;
713 u64 vruntime;
714};
Ingo Molnarfbf94822009-09-11 12:12:54 +0200715
Frederic Weisbecker419ab0d2009-09-12 03:59:01 +0200716struct trace_wakeup_event {
717 u32 size;
718
719 u16 common_type;
720 u8 common_flags;
721 u8 common_preempt_count;
722 u32 common_pid;
723 u32 common_tgid;
724
725 char comm[16];
726 u32 pid;
727
728 u32 prio;
729 u32 success;
730 u32 cpu;
731};
732
733struct trace_fork_event {
734 u32 size;
735
736 u16 common_type;
737 u8 common_flags;
738 u8 common_preempt_count;
739 u32 common_pid;
740 u32 common_tgid;
741
742 char parent_comm[16];
743 u32 parent_pid;
744 char child_comm[16];
745 u32 child_pid;
746};
747
748struct trace_sched_handler {
749 void (*switch_event)(struct trace_switch_event *,
750 struct event *,
751 int cpu,
752 u64 timestamp,
753 struct thread *thread);
754
mingo39aeb522009-09-14 20:04:48 +0200755 void (*runtime_event)(struct trace_runtime_event *,
756 struct event *,
757 int cpu,
758 u64 timestamp,
759 struct thread *thread);
760
Frederic Weisbecker419ab0d2009-09-12 03:59:01 +0200761 void (*wakeup_event)(struct trace_wakeup_event *,
762 struct event *,
763 int cpu,
764 u64 timestamp,
765 struct thread *thread);
766
767 void (*fork_event)(struct trace_fork_event *,
768 struct event *,
769 int cpu,
770 u64 timestamp,
771 struct thread *thread);
772};
773
Ingo Molnarfbf94822009-09-11 12:12:54 +0200774
775static void
Frederic Weisbecker419ab0d2009-09-12 03:59:01 +0200776replay_wakeup_event(struct trace_wakeup_event *wakeup_event,
777 struct event *event,
778 int cpu __used,
779 u64 timestamp __used,
780 struct thread *thread __used)
Ingo Molnarec156762009-09-11 12:12:54 +0200781{
Frederic Weisbecker419ab0d2009-09-12 03:59:01 +0200782 struct task_desc *waker, *wakee;
783
784 if (verbose) {
785 printf("sched_wakeup event %p\n", event);
786
787 printf(" ... pid %d woke up %s/%d\n",
788 wakeup_event->common_pid,
789 wakeup_event->comm,
790 wakeup_event->pid);
791 }
792
793 waker = register_pid(wakeup_event->common_pid, "<unknown>");
794 wakee = register_pid(wakeup_event->pid, wakeup_event->comm);
795
796 add_sched_event_wakeup(waker, timestamp, wakee);
797}
798
Ingo Molnard1153382009-09-14 18:22:53 +0200799static u64 cpu_last_switched[MAX_CPUS];
Frederic Weisbecker419ab0d2009-09-12 03:59:01 +0200800
801static void
802replay_switch_event(struct trace_switch_event *switch_event,
803 struct event *event,
804 int cpu,
805 u64 timestamp,
806 struct thread *thread __used)
807{
Ingo Molnarfbf94822009-09-11 12:12:54 +0200808 struct task_desc *prev, *next;
809 u64 timestamp0;
810 s64 delta;
811
Ingo Molnarad236fd2009-09-11 12:12:54 +0200812 if (verbose)
813 printf("sched_switch event %p\n", event);
814
Ingo Molnarfbf94822009-09-11 12:12:54 +0200815 if (cpu >= MAX_CPUS || cpu < 0)
816 return;
817
818 timestamp0 = cpu_last_switched[cpu];
819 if (timestamp0)
820 delta = timestamp - timestamp0;
821 else
822 delta = 0;
823
824 if (delta < 0)
825 die("hm, delta: %Ld < 0 ?\n", delta);
826
Ingo Molnarad236fd2009-09-11 12:12:54 +0200827 if (verbose) {
828 printf(" ... switch from %s/%d to %s/%d [ran %Ld nsecs]\n",
Frederic Weisbecker419ab0d2009-09-12 03:59:01 +0200829 switch_event->prev_comm, switch_event->prev_pid,
830 switch_event->next_comm, switch_event->next_pid,
Ingo Molnarad236fd2009-09-11 12:12:54 +0200831 delta);
832 }
Ingo Molnarfbf94822009-09-11 12:12:54 +0200833
Frederic Weisbecker419ab0d2009-09-12 03:59:01 +0200834 prev = register_pid(switch_event->prev_pid, switch_event->prev_comm);
835 next = register_pid(switch_event->next_pid, switch_event->next_comm);
Ingo Molnarfbf94822009-09-11 12:12:54 +0200836
837 cpu_last_switched[cpu] = timestamp;
838
839 add_sched_event_run(prev, timestamp, delta);
Frederic Weisbecker419ab0d2009-09-12 03:59:01 +0200840 add_sched_event_sleep(prev, timestamp, switch_event->prev_state);
Ingo Molnarfbf94822009-09-11 12:12:54 +0200841}
842
Ingo Molnarfbf94822009-09-11 12:12:54 +0200843
844static void
Frederic Weisbecker419ab0d2009-09-12 03:59:01 +0200845replay_fork_event(struct trace_fork_event *fork_event,
846 struct event *event,
847 int cpu __used,
848 u64 timestamp __used,
849 struct thread *thread __used)
850{
851 if (verbose) {
852 printf("sched_fork event %p\n", event);
853 printf("... parent: %s/%d\n", fork_event->parent_comm, fork_event->parent_pid);
854 printf("... child: %s/%d\n", fork_event->child_comm, fork_event->child_pid);
855 }
856 register_pid(fork_event->parent_pid, fork_event->parent_comm);
857 register_pid(fork_event->child_pid, fork_event->child_comm);
858}
859
860static struct trace_sched_handler replay_ops = {
Ingo Molnarea92ed52009-09-12 10:08:34 +0200861 .wakeup_event = replay_wakeup_event,
862 .switch_event = replay_switch_event,
863 .fork_event = replay_fork_event,
Frederic Weisbecker419ab0d2009-09-12 03:59:01 +0200864};
865
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +0200866struct sort_dimension {
867 const char *name;
Ingo Molnarb5fae122009-09-11 12:12:54 +0200868 sort_fn_t cmp;
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +0200869 struct list_head list;
870};
871
872static LIST_HEAD(cmp_pid);
873
Frederic Weisbeckerdaa1d7a2009-09-13 03:36:29 +0200874static int
mingo39aeb522009-09-14 20:04:48 +0200875thread_lat_cmp(struct list_head *list, struct work_atoms *l, struct work_atoms *r)
Frederic Weisbeckerdaa1d7a2009-09-13 03:36:29 +0200876{
877 struct sort_dimension *sort;
878 int ret = 0;
879
Ingo Molnarb5fae122009-09-11 12:12:54 +0200880 BUG_ON(list_empty(list));
881
Frederic Weisbeckerdaa1d7a2009-09-13 03:36:29 +0200882 list_for_each_entry(sort, list, list) {
883 ret = sort->cmp(l, r);
884 if (ret)
885 return ret;
886 }
887
888 return ret;
889}
890
mingo39aeb522009-09-14 20:04:48 +0200891static struct work_atoms *
Ingo Molnarb5fae122009-09-11 12:12:54 +0200892thread_atoms_search(struct rb_root *root, struct thread *thread,
893 struct list_head *sort_list)
894{
895 struct rb_node *node = root->rb_node;
mingo39aeb522009-09-14 20:04:48 +0200896 struct work_atoms key = { .thread = thread };
Ingo Molnarb5fae122009-09-11 12:12:54 +0200897
898 while (node) {
mingo39aeb522009-09-14 20:04:48 +0200899 struct work_atoms *atoms;
Ingo Molnarb5fae122009-09-11 12:12:54 +0200900 int cmp;
901
mingo39aeb522009-09-14 20:04:48 +0200902 atoms = container_of(node, struct work_atoms, node);
Ingo Molnarb5fae122009-09-11 12:12:54 +0200903
904 cmp = thread_lat_cmp(sort_list, &key, atoms);
905 if (cmp > 0)
906 node = node->rb_left;
907 else if (cmp < 0)
908 node = node->rb_right;
909 else {
910 BUG_ON(thread != atoms->thread);
911 return atoms;
912 }
913 }
914 return NULL;
915}
916
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +0200917static void
mingo39aeb522009-09-14 20:04:48 +0200918__thread_latency_insert(struct rb_root *root, struct work_atoms *data,
Frederic Weisbeckerdaa1d7a2009-09-13 03:36:29 +0200919 struct list_head *sort_list)
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +0200920{
921 struct rb_node **new = &(root->rb_node), *parent = NULL;
922
923 while (*new) {
mingo39aeb522009-09-14 20:04:48 +0200924 struct work_atoms *this;
Frederic Weisbeckerdaa1d7a2009-09-13 03:36:29 +0200925 int cmp;
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +0200926
mingo39aeb522009-09-14 20:04:48 +0200927 this = container_of(*new, struct work_atoms, node);
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +0200928 parent = *new;
Frederic Weisbeckerdaa1d7a2009-09-13 03:36:29 +0200929
930 cmp = thread_lat_cmp(sort_list, data, this);
931
932 if (cmp > 0)
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +0200933 new = &((*new)->rb_left);
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +0200934 else
Frederic Weisbeckerdaa1d7a2009-09-13 03:36:29 +0200935 new = &((*new)->rb_right);
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +0200936 }
937
938 rb_link_node(&data->node, parent, new);
939 rb_insert_color(&data->node, root);
940}
941
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +0200942static void thread_atoms_insert(struct thread *thread)
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +0200943{
mingo39aeb522009-09-14 20:04:48 +0200944 struct work_atoms *atoms;
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +0200945
Frederic Weisbecker17562202009-09-12 23:11:32 +0200946 atoms = calloc(sizeof(*atoms), 1);
947 if (!atoms)
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +0200948 die("No memory");
949
Frederic Weisbecker17562202009-09-12 23:11:32 +0200950 atoms->thread = thread;
mingo39aeb522009-09-14 20:04:48 +0200951 INIT_LIST_HEAD(&atoms->work_list);
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +0200952 __thread_latency_insert(&atom_root, atoms, &cmp_pid);
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +0200953}
954
955static void
956latency_fork_event(struct trace_fork_event *fork_event __used,
957 struct event *event __used,
958 int cpu __used,
959 u64 timestamp __used,
960 struct thread *thread __used)
961{
962 /* should insert the newcomer */
963}
964
Ingo Molnarea92ed52009-09-12 10:08:34 +0200965__used
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +0200966static char sched_out_state(struct trace_switch_event *switch_event)
967{
968 const char *str = TASK_STATE_TO_CHAR_STR;
969
970 return str[switch_event->prev_state];
971}
972
973static void
mingo39aeb522009-09-14 20:04:48 +0200974add_sched_out_event(struct work_atoms *atoms,
975 char run_state,
976 u64 timestamp)
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +0200977{
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +0200978 struct work_atom *atom;
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +0200979
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +0200980 atom = calloc(sizeof(*atom), 1);
981 if (!atom)
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +0200982 die("Non memory");
983
Frederic Weisbeckeraa1ab9d2009-09-14 03:01:12 +0200984 atom->sched_out_time = timestamp;
985
mingo39aeb522009-09-14 20:04:48 +0200986 if (run_state == 'R') {
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +0200987 atom->state = THREAD_WAIT_CPU;
Frederic Weisbeckeraa1ab9d2009-09-14 03:01:12 +0200988 atom->wake_up_time = atom->sched_out_time;
Frederic Weisbeckerc6ced612009-09-13 00:46:19 +0200989 }
990
mingo39aeb522009-09-14 20:04:48 +0200991 list_add_tail(&atom->list, &atoms->work_list);
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +0200992}
993
994static void
mingo39aeb522009-09-14 20:04:48 +0200995add_runtime_event(struct work_atoms *atoms, u64 delta, u64 timestamp __used)
996{
997 struct work_atom *atom;
998
999 BUG_ON(list_empty(&atoms->work_list));
1000
1001 atom = list_entry(atoms->work_list.prev, struct work_atom, list);
1002
1003 atom->runtime += delta;
1004 atoms->total_runtime += delta;
1005}
1006
1007static void
1008add_sched_in_event(struct work_atoms *atoms, u64 timestamp)
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +02001009{
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +02001010 struct work_atom *atom;
Frederic Weisbecker66685672009-09-13 01:56:25 +02001011 u64 delta;
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +02001012
mingo39aeb522009-09-14 20:04:48 +02001013 if (list_empty(&atoms->work_list))
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +02001014 return;
1015
mingo39aeb522009-09-14 20:04:48 +02001016 atom = list_entry(atoms->work_list.prev, struct work_atom, list);
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +02001017
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +02001018 if (atom->state != THREAD_WAIT_CPU)
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +02001019 return;
1020
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +02001021 if (timestamp < atom->wake_up_time) {
1022 atom->state = THREAD_IGNORE;
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +02001023 return;
1024 }
1025
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +02001026 atom->state = THREAD_SCHED_IN;
1027 atom->sched_in_time = timestamp;
Frederic Weisbecker66685672009-09-13 01:56:25 +02001028
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +02001029 delta = atom->sched_in_time - atom->wake_up_time;
Frederic Weisbecker66685672009-09-13 01:56:25 +02001030 atoms->total_lat += delta;
1031 if (delta > atoms->max_lat)
1032 atoms->max_lat = delta;
1033 atoms->nb_atoms++;
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +02001034}
1035
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +02001036static void
1037latency_switch_event(struct trace_switch_event *switch_event,
1038 struct event *event __used,
Ingo Molnarea92ed52009-09-12 10:08:34 +02001039 int cpu,
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +02001040 u64 timestamp,
1041 struct thread *thread __used)
1042{
mingo39aeb522009-09-14 20:04:48 +02001043 struct work_atoms *out_events, *in_events;
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +02001044 struct thread *sched_out, *sched_in;
Ingo Molnarea92ed52009-09-12 10:08:34 +02001045 u64 timestamp0;
1046 s64 delta;
1047
mingo39aeb522009-09-14 20:04:48 +02001048 BUG_ON(cpu >= MAX_CPUS || cpu < 0);
Ingo Molnarea92ed52009-09-12 10:08:34 +02001049
1050 timestamp0 = cpu_last_switched[cpu];
1051 cpu_last_switched[cpu] = timestamp;
1052 if (timestamp0)
1053 delta = timestamp - timestamp0;
1054 else
1055 delta = 0;
1056
1057 if (delta < 0)
1058 die("hm, delta: %Ld < 0 ?\n", delta);
1059
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +02001060
1061 sched_out = threads__findnew(switch_event->prev_pid, &threads, &last_match);
1062 sched_in = threads__findnew(switch_event->next_pid, &threads, &last_match);
1063
mingo39aeb522009-09-14 20:04:48 +02001064 out_events = thread_atoms_search(&atom_root, sched_out, &cmp_pid);
1065 if (!out_events) {
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +02001066 thread_atoms_insert(sched_out);
mingo39aeb522009-09-14 20:04:48 +02001067 out_events = thread_atoms_search(&atom_root, sched_out, &cmp_pid);
1068 if (!out_events)
1069 die("out-event: Internal tree error");
1070 }
1071 add_sched_out_event(out_events, sched_out_state(switch_event), timestamp);
1072
1073 in_events = thread_atoms_search(&atom_root, sched_in, &cmp_pid);
1074 if (!in_events) {
1075 thread_atoms_insert(sched_in);
1076 in_events = thread_atoms_search(&atom_root, sched_in, &cmp_pid);
1077 if (!in_events)
1078 die("in-event: Internal tree error");
1079 /*
1080 * Take came in we have not heard about yet,
1081 * add in an initial atom in runnable state:
1082 */
1083 add_sched_out_event(in_events, 'R', timestamp);
1084 }
1085 add_sched_in_event(in_events, timestamp);
1086}
1087
1088static void
1089latency_runtime_event(struct trace_runtime_event *runtime_event,
1090 struct event *event __used,
1091 int cpu,
1092 u64 timestamp,
1093 struct thread *this_thread __used)
1094{
1095 struct work_atoms *atoms;
1096 struct thread *thread;
1097
1098 BUG_ON(cpu >= MAX_CPUS || cpu < 0);
1099
1100 thread = threads__findnew(runtime_event->pid, &threads, &last_match);
1101 atoms = thread_atoms_search(&atom_root, thread, &cmp_pid);
1102 if (!atoms) {
1103 thread_atoms_insert(thread);
1104 atoms = thread_atoms_search(&atom_root, thread, &cmp_pid);
1105 if (!atoms)
1106 die("in-event: Internal tree error");
1107 add_sched_out_event(atoms, 'R', timestamp);
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +02001108 }
1109
mingo39aeb522009-09-14 20:04:48 +02001110 add_runtime_event(atoms, runtime_event->runtime, timestamp);
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +02001111}
1112
1113static void
1114latency_wakeup_event(struct trace_wakeup_event *wakeup_event,
mingo39aeb522009-09-14 20:04:48 +02001115 struct event *__event __used,
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +02001116 int cpu __used,
1117 u64 timestamp,
1118 struct thread *thread __used)
1119{
mingo39aeb522009-09-14 20:04:48 +02001120 struct work_atoms *atoms;
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +02001121 struct work_atom *atom;
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +02001122 struct thread *wakee;
1123
1124 /* Note for later, it may be interesting to observe the failing cases */
1125 if (!wakeup_event->success)
1126 return;
1127
1128 wakee = threads__findnew(wakeup_event->pid, &threads, &last_match);
Ingo Molnarb5fae122009-09-11 12:12:54 +02001129 atoms = thread_atoms_search(&atom_root, wakee, &cmp_pid);
Frederic Weisbecker17562202009-09-12 23:11:32 +02001130 if (!atoms) {
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +02001131 thread_atoms_insert(wakee);
mingo39aeb522009-09-14 20:04:48 +02001132 atoms = thread_atoms_search(&atom_root, wakee, &cmp_pid);
1133 if (!atoms)
1134 die("wakeup-event: Internal tree error");
1135 add_sched_out_event(atoms, 'S', timestamp);
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +02001136 }
1137
mingo39aeb522009-09-14 20:04:48 +02001138 BUG_ON(list_empty(&atoms->work_list));
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +02001139
mingo39aeb522009-09-14 20:04:48 +02001140 atom = list_entry(atoms->work_list.prev, struct work_atom, list);
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +02001141
Ingo Molnardc02bf72009-09-16 13:45:00 +02001142 if (atom->state != THREAD_SLEEPING)
1143 nr_state_machine_bugs++;
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +02001144
Ingo Molnarea57c4f2009-09-13 18:15:54 +02001145 nr_timestamps++;
1146 if (atom->sched_out_time > timestamp) {
Ingo Molnardc02bf72009-09-16 13:45:00 +02001147 nr_unordered_timestamps++;
Frederic Weisbeckeraa1ab9d2009-09-14 03:01:12 +02001148 return;
Ingo Molnarea57c4f2009-09-13 18:15:54 +02001149 }
Frederic Weisbeckeraa1ab9d2009-09-14 03:01:12 +02001150
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +02001151 atom->state = THREAD_WAIT_CPU;
1152 atom->wake_up_time = timestamp;
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +02001153}
1154
1155static struct trace_sched_handler lat_ops = {
Ingo Molnarea92ed52009-09-12 10:08:34 +02001156 .wakeup_event = latency_wakeup_event,
1157 .switch_event = latency_switch_event,
mingo39aeb522009-09-14 20:04:48 +02001158 .runtime_event = latency_runtime_event,
Ingo Molnarea92ed52009-09-12 10:08:34 +02001159 .fork_event = latency_fork_event,
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +02001160};
1161
mingo39aeb522009-09-14 20:04:48 +02001162static void output_lat_thread(struct work_atoms *work_list)
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +02001163{
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +02001164 int i;
1165 int ret;
Frederic Weisbecker66685672009-09-13 01:56:25 +02001166 u64 avg;
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +02001167
mingo39aeb522009-09-14 20:04:48 +02001168 if (!work_list->nb_atoms)
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +02001169 return;
Ingo Molnarea57c4f2009-09-13 18:15:54 +02001170 /*
1171 * Ignore idle threads:
1172 */
Ingo Molnar80ed0982009-09-16 14:12:36 +02001173 if (!strcmp(work_list->thread->comm, "swapper"))
Ingo Molnarea57c4f2009-09-13 18:15:54 +02001174 return;
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +02001175
mingo39aeb522009-09-14 20:04:48 +02001176 all_runtime += work_list->total_runtime;
1177 all_count += work_list->nb_atoms;
Frederic Weisbecker66685672009-09-13 01:56:25 +02001178
Ingo Molnar80ed0982009-09-16 14:12:36 +02001179 ret = printf(" %s:%d ", work_list->thread->comm, work_list->thread->pid);
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +02001180
mingo08f69e62009-09-14 18:30:44 +02001181 for (i = 0; i < 24 - ret; i++)
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +02001182 printf(" ");
1183
mingo39aeb522009-09-14 20:04:48 +02001184 avg = work_list->total_lat / work_list->nb_atoms;
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +02001185
Ingo Molnardc02bf72009-09-16 13:45:00 +02001186 printf("|%11.3f ms |%9llu | avg:%9.3f ms | max:%9.3f ms |\n",
mingo39aeb522009-09-14 20:04:48 +02001187 (double)work_list->total_runtime / 1e6,
1188 work_list->nb_atoms, (double)avg / 1e6,
1189 (double)work_list->max_lat / 1e6);
Frederic Weisbeckercdce9d72009-09-12 08:06:14 +02001190}
1191
mingo39aeb522009-09-14 20:04:48 +02001192static int pid_cmp(struct work_atoms *l, struct work_atoms *r)
Frederic Weisbeckerdaa1d7a2009-09-13 03:36:29 +02001193{
Frederic Weisbeckerdaa1d7a2009-09-13 03:36:29 +02001194 if (l->thread->pid < r->thread->pid)
1195 return -1;
1196 if (l->thread->pid > r->thread->pid)
1197 return 1;
1198
1199 return 0;
1200}
1201
1202static struct sort_dimension pid_sort_dimension = {
Ingo Molnarb5fae122009-09-11 12:12:54 +02001203 .name = "pid",
1204 .cmp = pid_cmp,
Frederic Weisbeckerdaa1d7a2009-09-13 03:36:29 +02001205};
1206
mingo39aeb522009-09-14 20:04:48 +02001207static int avg_cmp(struct work_atoms *l, struct work_atoms *r)
Frederic Weisbeckerdaa1d7a2009-09-13 03:36:29 +02001208{
1209 u64 avgl, avgr;
1210
1211 if (!l->nb_atoms)
1212 return -1;
1213
1214 if (!r->nb_atoms)
1215 return 1;
1216
1217 avgl = l->total_lat / l->nb_atoms;
1218 avgr = r->total_lat / r->nb_atoms;
1219
1220 if (avgl < avgr)
1221 return -1;
1222 if (avgl > avgr)
1223 return 1;
1224
1225 return 0;
1226}
1227
1228static struct sort_dimension avg_sort_dimension = {
Ingo Molnarb5fae122009-09-11 12:12:54 +02001229 .name = "avg",
1230 .cmp = avg_cmp,
Frederic Weisbeckerdaa1d7a2009-09-13 03:36:29 +02001231};
1232
mingo39aeb522009-09-14 20:04:48 +02001233static int max_cmp(struct work_atoms *l, struct work_atoms *r)
Frederic Weisbeckerdaa1d7a2009-09-13 03:36:29 +02001234{
1235 if (l->max_lat < r->max_lat)
1236 return -1;
1237 if (l->max_lat > r->max_lat)
1238 return 1;
1239
1240 return 0;
1241}
1242
1243static struct sort_dimension max_sort_dimension = {
Ingo Molnarb5fae122009-09-11 12:12:54 +02001244 .name = "max",
1245 .cmp = max_cmp,
Frederic Weisbeckerdaa1d7a2009-09-13 03:36:29 +02001246};
1247
mingo39aeb522009-09-14 20:04:48 +02001248static int switch_cmp(struct work_atoms *l, struct work_atoms *r)
Frederic Weisbeckerdaa1d7a2009-09-13 03:36:29 +02001249{
1250 if (l->nb_atoms < r->nb_atoms)
1251 return -1;
1252 if (l->nb_atoms > r->nb_atoms)
1253 return 1;
1254
1255 return 0;
1256}
1257
1258static struct sort_dimension switch_sort_dimension = {
Ingo Molnarb5fae122009-09-11 12:12:54 +02001259 .name = "switch",
1260 .cmp = switch_cmp,
Frederic Weisbeckerdaa1d7a2009-09-13 03:36:29 +02001261};
1262
mingo39aeb522009-09-14 20:04:48 +02001263static int runtime_cmp(struct work_atoms *l, struct work_atoms *r)
Frederic Weisbeckerdaa1d7a2009-09-13 03:36:29 +02001264{
1265 if (l->total_runtime < r->total_runtime)
1266 return -1;
1267 if (l->total_runtime > r->total_runtime)
1268 return 1;
1269
1270 return 0;
1271}
1272
1273static struct sort_dimension runtime_sort_dimension = {
Ingo Molnarb5fae122009-09-11 12:12:54 +02001274 .name = "runtime",
1275 .cmp = runtime_cmp,
Frederic Weisbeckerdaa1d7a2009-09-13 03:36:29 +02001276};
1277
1278static struct sort_dimension *available_sorts[] = {
1279 &pid_sort_dimension,
1280 &avg_sort_dimension,
1281 &max_sort_dimension,
1282 &switch_sort_dimension,
1283 &runtime_sort_dimension,
1284};
1285
1286#define NB_AVAILABLE_SORTS (int)(sizeof(available_sorts) / sizeof(struct sort_dimension *))
1287
1288static LIST_HEAD(sort_list);
1289
1290static int sort_dimension__add(char *tok, struct list_head *list)
1291{
1292 int i;
1293
1294 for (i = 0; i < NB_AVAILABLE_SORTS; i++) {
1295 if (!strcmp(available_sorts[i]->name, tok)) {
1296 list_add_tail(&available_sorts[i]->list, list);
1297
1298 return 0;
1299 }
1300 }
1301
1302 return -1;
1303}
1304
1305static void setup_sorting(void);
1306
1307static void sort_lat(void)
1308{
1309 struct rb_node *node;
1310
1311 for (;;) {
mingo39aeb522009-09-14 20:04:48 +02001312 struct work_atoms *data;
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +02001313 node = rb_first(&atom_root);
Frederic Weisbeckerdaa1d7a2009-09-13 03:36:29 +02001314 if (!node)
1315 break;
1316
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +02001317 rb_erase(node, &atom_root);
mingo39aeb522009-09-14 20:04:48 +02001318 data = rb_entry(node, struct work_atoms, node);
Ingo Molnarb1ffe8f2009-09-11 12:12:54 +02001319 __thread_latency_insert(&sorted_atom_root, data, &sort_list);
Frederic Weisbeckerdaa1d7a2009-09-13 03:36:29 +02001320 }
1321}
1322
Frederic Weisbecker419ab0d2009-09-12 03:59:01 +02001323static struct trace_sched_handler *trace_handler;
1324
1325static void
1326process_sched_wakeup_event(struct raw_event_sample *raw,
1327 struct event *event,
1328 int cpu __used,
1329 u64 timestamp __used,
1330 struct thread *thread __used)
1331{
1332 struct trace_wakeup_event wakeup_event;
1333
1334 FILL_COMMON_FIELDS(wakeup_event, event, raw->data);
1335
1336 FILL_ARRAY(wakeup_event, comm, event, raw->data);
1337 FILL_FIELD(wakeup_event, pid, event, raw->data);
1338 FILL_FIELD(wakeup_event, prio, event, raw->data);
1339 FILL_FIELD(wakeup_event, success, event, raw->data);
1340 FILL_FIELD(wakeup_event, cpu, event, raw->data);
1341
Ingo Molnar0ec04e12009-09-16 17:40:48 +02001342 if (trace_handler->wakeup_event)
1343 trace_handler->wakeup_event(&wakeup_event, event, cpu, timestamp, thread);
Frederic Weisbecker419ab0d2009-09-12 03:59:01 +02001344}
1345
Ingo Molnarc8a37752009-09-16 14:07:00 +02001346/*
1347 * Track the current task - that way we can know whether there's any
1348 * weird events, such as a task being switched away that is not current.
1349 */
Ingo Molnar40749d02009-09-17 18:24:55 +02001350static int max_cpu;
Ingo Molnar0ec04e12009-09-16 17:40:48 +02001351
Ingo Molnarc8a37752009-09-16 14:07:00 +02001352static u32 curr_pid[MAX_CPUS] = { [0 ... MAX_CPUS-1] = -1 };
1353
Ingo Molnar0ec04e12009-09-16 17:40:48 +02001354static struct thread *curr_thread[MAX_CPUS];
1355
1356static char next_shortname1 = 'A';
1357static char next_shortname2 = '0';
1358
1359static void
1360map_switch_event(struct trace_switch_event *switch_event,
1361 struct event *event __used,
1362 int this_cpu,
1363 u64 timestamp,
1364 struct thread *thread __used)
1365{
1366 struct thread *sched_out, *sched_in;
1367 int new_shortname;
1368 u64 timestamp0;
1369 s64 delta;
1370 int cpu;
1371
1372 BUG_ON(this_cpu >= MAX_CPUS || this_cpu < 0);
1373
1374 if (this_cpu > max_cpu)
1375 max_cpu = this_cpu;
1376
1377 timestamp0 = cpu_last_switched[this_cpu];
1378 cpu_last_switched[this_cpu] = timestamp;
1379 if (timestamp0)
1380 delta = timestamp - timestamp0;
1381 else
1382 delta = 0;
1383
1384 if (delta < 0)
1385 die("hm, delta: %Ld < 0 ?\n", delta);
1386
1387
1388 sched_out = threads__findnew(switch_event->prev_pid, &threads, &last_match);
1389 sched_in = threads__findnew(switch_event->next_pid, &threads, &last_match);
1390
1391 curr_thread[this_cpu] = sched_in;
1392
1393 printf(" ");
1394
1395 new_shortname = 0;
1396 if (!sched_in->shortname[0]) {
1397 sched_in->shortname[0] = next_shortname1;
1398 sched_in->shortname[1] = next_shortname2;
1399
1400 if (next_shortname1 < 'Z') {
1401 next_shortname1++;
1402 } else {
1403 next_shortname1='A';
1404 if (next_shortname2 < '9') {
1405 next_shortname2++;
1406 } else {
1407 next_shortname2='0';
1408 }
1409 }
1410 new_shortname = 1;
1411 }
1412
1413 for (cpu = 0; cpu <= max_cpu; cpu++) {
1414 if (cpu != this_cpu)
1415 printf(" ");
1416 else
1417 printf("*");
1418
1419 if (curr_thread[cpu]) {
1420 if (curr_thread[cpu]->pid)
1421 printf("%2s ", curr_thread[cpu]->shortname);
1422 else
1423 printf(". ");
1424 } else
1425 printf(" ");
1426 }
1427
1428 printf(" %12.6f secs ", (double)timestamp/1e9);
1429 if (new_shortname) {
1430 printf("%s => %s:%d\n",
1431 sched_in->shortname, sched_in->comm, sched_in->pid);
1432 } else {
1433 printf("\n");
1434 }
1435}
1436
1437
Frederic Weisbecker419ab0d2009-09-12 03:59:01 +02001438static void
1439process_sched_switch_event(struct raw_event_sample *raw,
1440 struct event *event,
Ingo Molnar0ec04e12009-09-16 17:40:48 +02001441 int this_cpu,
Frederic Weisbecker419ab0d2009-09-12 03:59:01 +02001442 u64 timestamp __used,
1443 struct thread *thread __used)
1444{
1445 struct trace_switch_event switch_event;
1446
1447 FILL_COMMON_FIELDS(switch_event, event, raw->data);
1448
1449 FILL_ARRAY(switch_event, prev_comm, event, raw->data);
1450 FILL_FIELD(switch_event, prev_pid, event, raw->data);
1451 FILL_FIELD(switch_event, prev_prio, event, raw->data);
1452 FILL_FIELD(switch_event, prev_state, event, raw->data);
1453 FILL_ARRAY(switch_event, next_comm, event, raw->data);
1454 FILL_FIELD(switch_event, next_pid, event, raw->data);
1455 FILL_FIELD(switch_event, next_prio, event, raw->data);
1456
Ingo Molnar0ec04e12009-09-16 17:40:48 +02001457 if (curr_pid[this_cpu] != (u32)-1) {
Ingo Molnarc8a37752009-09-16 14:07:00 +02001458 /*
1459 * Are we trying to switch away a PID that is
1460 * not current?
1461 */
Ingo Molnar0ec04e12009-09-16 17:40:48 +02001462 if (curr_pid[this_cpu] != switch_event.prev_pid)
Ingo Molnarc8a37752009-09-16 14:07:00 +02001463 nr_context_switch_bugs++;
1464 }
Ingo Molnar0ec04e12009-09-16 17:40:48 +02001465 if (trace_handler->switch_event)
1466 trace_handler->switch_event(&switch_event, event, this_cpu, timestamp, thread);
Ingo Molnarc8a37752009-09-16 14:07:00 +02001467
Ingo Molnar0ec04e12009-09-16 17:40:48 +02001468 curr_pid[this_cpu] = switch_event.next_pid;
Frederic Weisbecker419ab0d2009-09-12 03:59:01 +02001469}
1470
1471static void
mingo39aeb522009-09-14 20:04:48 +02001472process_sched_runtime_event(struct raw_event_sample *raw,
1473 struct event *event,
1474 int cpu __used,
1475 u64 timestamp __used,
1476 struct thread *thread __used)
1477{
1478 struct trace_runtime_event runtime_event;
1479
1480 FILL_ARRAY(runtime_event, comm, event, raw->data);
1481 FILL_FIELD(runtime_event, pid, event, raw->data);
1482 FILL_FIELD(runtime_event, runtime, event, raw->data);
1483 FILL_FIELD(runtime_event, vruntime, event, raw->data);
1484
Ingo Molnar0ec04e12009-09-16 17:40:48 +02001485 if (trace_handler->runtime_event)
1486 trace_handler->runtime_event(&runtime_event, event, cpu, timestamp, thread);
mingo39aeb522009-09-14 20:04:48 +02001487}
1488
1489static void
Frederic Weisbecker419ab0d2009-09-12 03:59:01 +02001490process_sched_fork_event(struct raw_event_sample *raw,
1491 struct event *event,
1492 int cpu __used,
1493 u64 timestamp __used,
1494 struct thread *thread __used)
Ingo Molnarfbf94822009-09-11 12:12:54 +02001495{
Frederic Weisbecker46538812009-09-12 02:43:45 +02001496 struct trace_fork_event fork_event;
1497
1498 FILL_COMMON_FIELDS(fork_event, event, raw->data);
1499
1500 FILL_ARRAY(fork_event, parent_comm, event, raw->data);
1501 FILL_FIELD(fork_event, parent_pid, event, raw->data);
1502 FILL_ARRAY(fork_event, child_comm, event, raw->data);
1503 FILL_FIELD(fork_event, child_pid, event, raw->data);
1504
Ingo Molnar0ec04e12009-09-16 17:40:48 +02001505 if (trace_handler->fork_event)
1506 trace_handler->fork_event(&fork_event, event, cpu, timestamp, thread);
Ingo Molnarfbf94822009-09-11 12:12:54 +02001507}
1508
Frederic Weisbecker419ab0d2009-09-12 03:59:01 +02001509static void
1510process_sched_exit_event(struct event *event,
1511 int cpu __used,
1512 u64 timestamp __used,
1513 struct thread *thread __used)
Ingo Molnarfbf94822009-09-11 12:12:54 +02001514{
Ingo Molnarad236fd2009-09-11 12:12:54 +02001515 if (verbose)
1516 printf("sched_exit event %p\n", event);
Ingo Molnarec156762009-09-11 12:12:54 +02001517}
1518
1519static void
Ingo Molnarad236fd2009-09-11 12:12:54 +02001520process_raw_event(event_t *raw_event __used, void *more_data,
Ingo Molnarec156762009-09-11 12:12:54 +02001521 int cpu, u64 timestamp, struct thread *thread)
1522{
Frederic Weisbecker46538812009-09-12 02:43:45 +02001523 struct raw_event_sample *raw = more_data;
Ingo Molnarec156762009-09-11 12:12:54 +02001524 struct event *event;
1525 int type;
1526
1527 type = trace_parse_common_type(raw->data);
1528 event = trace_find_event(type);
1529
Ingo Molnarec156762009-09-11 12:12:54 +02001530 if (!strcmp(event->name, "sched_switch"))
Frederic Weisbecker46538812009-09-12 02:43:45 +02001531 process_sched_switch_event(raw, event, cpu, timestamp, thread);
mingo39aeb522009-09-14 20:04:48 +02001532 if (!strcmp(event->name, "sched_stat_runtime"))
1533 process_sched_runtime_event(raw, event, cpu, timestamp, thread);
Ingo Molnarec156762009-09-11 12:12:54 +02001534 if (!strcmp(event->name, "sched_wakeup"))
Frederic Weisbecker46538812009-09-12 02:43:45 +02001535 process_sched_wakeup_event(raw, event, cpu, timestamp, thread);
Ingo Molnarfbf94822009-09-11 12:12:54 +02001536 if (!strcmp(event->name, "sched_wakeup_new"))
Frederic Weisbecker46538812009-09-12 02:43:45 +02001537 process_sched_wakeup_event(raw, event, cpu, timestamp, thread);
Ingo Molnarfbf94822009-09-11 12:12:54 +02001538 if (!strcmp(event->name, "sched_process_fork"))
Frederic Weisbecker46538812009-09-12 02:43:45 +02001539 process_sched_fork_event(raw, event, cpu, timestamp, thread);
Ingo Molnarfbf94822009-09-11 12:12:54 +02001540 if (!strcmp(event->name, "sched_process_exit"))
1541 process_sched_exit_event(event, cpu, timestamp, thread);
Ingo Molnarec156762009-09-11 12:12:54 +02001542}
1543
Ingo Molnar0a02ad92009-09-11 12:12:54 +02001544static int
1545process_sample_event(event_t *event, unsigned long offset, unsigned long head)
1546{
1547 char level;
1548 int show = 0;
1549 struct dso *dso = NULL;
1550 struct thread *thread;
1551 u64 ip = event->ip.ip;
1552 u64 timestamp = -1;
1553 u32 cpu = -1;
1554 u64 period = 1;
1555 void *more_data = event->ip.__more_data;
1556 int cpumode;
1557
1558 thread = threads__findnew(event->ip.pid, &threads, &last_match);
1559
1560 if (sample_type & PERF_SAMPLE_TIME) {
1561 timestamp = *(u64 *)more_data;
1562 more_data += sizeof(u64);
1563 }
1564
1565 if (sample_type & PERF_SAMPLE_CPU) {
1566 cpu = *(u32 *)more_data;
1567 more_data += sizeof(u32);
1568 more_data += sizeof(u32); /* reserved */
1569 }
1570
1571 if (sample_type & PERF_SAMPLE_PERIOD) {
1572 period = *(u64 *)more_data;
1573 more_data += sizeof(u64);
1574 }
1575
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001576 dump_printf("%p [%p]: PERF_RECORD_SAMPLE (IP, %d): %d/%d: %p period: %Ld\n",
Ingo Molnar0a02ad92009-09-11 12:12:54 +02001577 (void *)(offset + head),
1578 (void *)(long)(event->header.size),
1579 event->header.misc,
1580 event->ip.pid, event->ip.tid,
1581 (void *)(long)ip,
1582 (long long)period);
1583
1584 dump_printf(" ... thread: %s:%d\n", thread->comm, thread->pid);
1585
1586 if (thread == NULL) {
1587 eprintf("problem processing %d event, skipping it.\n",
1588 event->header.type);
1589 return -1;
1590 }
1591
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001592 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
Ingo Molnar0a02ad92009-09-11 12:12:54 +02001593
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001594 if (cpumode == PERF_RECORD_MISC_KERNEL) {
Ingo Molnar0a02ad92009-09-11 12:12:54 +02001595 show = SHOW_KERNEL;
1596 level = 'k';
1597
1598 dso = kernel_dso;
1599
1600 dump_printf(" ...... dso: %s\n", dso->name);
1601
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001602 } else if (cpumode == PERF_RECORD_MISC_USER) {
Ingo Molnar0a02ad92009-09-11 12:12:54 +02001603
1604 show = SHOW_USER;
1605 level = '.';
1606
1607 } else {
1608 show = SHOW_HV;
1609 level = 'H';
1610
1611 dso = hypervisor_dso;
1612
1613 dump_printf(" ...... dso: [hypervisor]\n");
1614 }
1615
Ingo Molnarec156762009-09-11 12:12:54 +02001616 if (sample_type & PERF_SAMPLE_RAW)
1617 process_raw_event(event, more_data, cpu, timestamp, thread);
Ingo Molnar0a02ad92009-09-11 12:12:54 +02001618
1619 return 0;
1620}
1621
1622static int
1623process_event(event_t *event, unsigned long offset, unsigned long head)
1624{
1625 trace_event(event);
1626
Ingo Molnardc02bf72009-09-16 13:45:00 +02001627 nr_events++;
Ingo Molnar0a02ad92009-09-11 12:12:54 +02001628 switch (event->header.type) {
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001629 case PERF_RECORD_MMAP:
Ingo Molnardc02bf72009-09-16 13:45:00 +02001630 return 0;
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001631 case PERF_RECORD_LOST:
Ingo Molnardc02bf72009-09-16 13:45:00 +02001632 nr_lost_chunks++;
1633 nr_lost_events += event->lost.lost;
Ingo Molnar0a02ad92009-09-11 12:12:54 +02001634 return 0;
1635
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001636 case PERF_RECORD_COMM:
Ingo Molnar0a02ad92009-09-11 12:12:54 +02001637 return process_comm_event(event, offset, head);
1638
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001639 case PERF_RECORD_EXIT ... PERF_RECORD_READ:
Ingo Molnar0a02ad92009-09-11 12:12:54 +02001640 return 0;
1641
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001642 case PERF_RECORD_SAMPLE:
Ingo Molnar0a02ad92009-09-11 12:12:54 +02001643 return process_sample_event(event, offset, head);
1644
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001645 case PERF_RECORD_MAX:
Ingo Molnar0a02ad92009-09-11 12:12:54 +02001646 default:
1647 return -1;
1648 }
1649
1650 return 0;
1651}
1652
Ingo Molnar46f392c2009-09-12 10:08:34 +02001653static int read_events(void)
Ingo Molnar0a02ad92009-09-11 12:12:54 +02001654{
1655 int ret, rc = EXIT_FAILURE;
1656 unsigned long offset = 0;
1657 unsigned long head = 0;
1658 struct stat perf_stat;
1659 event_t *event;
1660 uint32_t size;
1661 char *buf;
1662
1663 trace_report();
1664 register_idle_thread(&threads, &last_match);
1665
1666 input = open(input_name, O_RDONLY);
1667 if (input < 0) {
1668 perror("failed to open file");
1669 exit(-1);
1670 }
1671
1672 ret = fstat(input, &perf_stat);
1673 if (ret < 0) {
1674 perror("failed to stat file");
1675 exit(-1);
1676 }
1677
1678 if (!perf_stat.st_size) {
1679 fprintf(stderr, "zero-sized file, nothing to do!\n");
1680 exit(0);
1681 }
1682 header = perf_header__read(input);
1683 head = header->data_offset;
1684 sample_type = perf_header__sample_type(header);
1685
1686 if (!(sample_type & PERF_SAMPLE_RAW))
1687 die("No trace sample to read. Did you call perf record "
1688 "without -R?");
1689
1690 if (load_kernel() < 0) {
1691 perror("failed to load kernel symbols");
1692 return EXIT_FAILURE;
1693 }
1694
1695remap:
1696 buf = (char *)mmap(NULL, page_size * mmap_window, PROT_READ,
1697 MAP_SHARED, input, offset);
1698 if (buf == MAP_FAILED) {
1699 perror("failed to mmap file");
1700 exit(-1);
1701 }
1702
1703more:
1704 event = (event_t *)(buf + head);
1705
1706 size = event->header.size;
1707 if (!size)
1708 size = 8;
1709
1710 if (head + event->header.size >= page_size * mmap_window) {
1711 unsigned long shift = page_size * (head / page_size);
1712 int res;
1713
1714 res = munmap(buf, page_size * mmap_window);
1715 assert(res == 0);
1716
1717 offset += shift;
1718 head -= shift;
1719 goto remap;
1720 }
1721
1722 size = event->header.size;
1723
1724
1725 if (!size || process_event(event, offset, head) < 0) {
1726
1727 /*
1728 * assume we lost track of the stream, check alignment, and
1729 * increment a single u64 in the hope to catch on again 'soon'.
1730 */
1731
1732 if (unlikely(head & 7))
1733 head &= ~7ULL;
1734
1735 size = 8;
1736 }
1737
1738 head += size;
1739
1740 if (offset + head < (unsigned long)perf_stat.st_size)
1741 goto more;
1742
1743 rc = EXIT_SUCCESS;
1744 close(input);
1745
1746 return rc;
1747}
1748
Ingo Molnar0ec04e12009-09-16 17:40:48 +02001749static void print_bad_events(void)
1750{
1751 if (nr_unordered_timestamps && nr_timestamps) {
1752 printf(" INFO: %.3f%% unordered timestamps (%ld out of %ld)\n",
1753 (double)nr_unordered_timestamps/(double)nr_timestamps*100.0,
1754 nr_unordered_timestamps, nr_timestamps);
1755 }
1756 if (nr_lost_events && nr_events) {
1757 printf(" INFO: %.3f%% lost events (%ld out of %ld, in %ld chunks)\n",
1758 (double)nr_lost_events/(double)nr_events*100.0,
1759 nr_lost_events, nr_events, nr_lost_chunks);
1760 }
1761 if (nr_state_machine_bugs && nr_timestamps) {
1762 printf(" INFO: %.3f%% state machine bugs (%ld out of %ld)",
1763 (double)nr_state_machine_bugs/(double)nr_timestamps*100.0,
1764 nr_state_machine_bugs, nr_timestamps);
1765 if (nr_lost_events)
1766 printf(" (due to lost events?)");
1767 printf("\n");
1768 }
1769 if (nr_context_switch_bugs && nr_timestamps) {
1770 printf(" INFO: %.3f%% context switch bugs (%ld out of %ld)",
1771 (double)nr_context_switch_bugs/(double)nr_timestamps*100.0,
1772 nr_context_switch_bugs, nr_timestamps);
1773 if (nr_lost_events)
1774 printf(" (due to lost events?)");
1775 printf("\n");
1776 }
1777}
1778
1779static void __cmd_lat(void)
1780{
1781 struct rb_node *next;
1782
1783 setup_pager();
1784 read_events();
1785 sort_lat();
1786
1787 printf("\n -----------------------------------------------------------------------------------------\n");
1788 printf(" Task | Runtime ms | Switches | Average delay ms | Maximum delay ms |\n");
1789 printf(" -----------------------------------------------------------------------------------------\n");
1790
1791 next = rb_first(&sorted_atom_root);
1792
1793 while (next) {
1794 struct work_atoms *work_list;
1795
1796 work_list = rb_entry(next, struct work_atoms, node);
1797 output_lat_thread(work_list);
1798 next = rb_next(next);
1799 }
1800
1801 printf(" -----------------------------------------------------------------------------------------\n");
1802 printf(" TOTAL: |%11.3f ms |%9Ld |\n",
1803 (double)all_runtime/1e6, all_count);
1804
1805 printf(" ---------------------------------------------------\n");
1806
1807 print_bad_events();
1808 printf("\n");
1809
1810}
1811
1812static struct trace_sched_handler map_ops = {
1813 .wakeup_event = NULL,
1814 .switch_event = map_switch_event,
1815 .runtime_event = NULL,
1816 .fork_event = NULL,
1817};
1818
1819static void __cmd_map(void)
1820{
Ingo Molnar40749d02009-09-17 18:24:55 +02001821 max_cpu = sysconf(_SC_NPROCESSORS_CONF);
1822
Ingo Molnar0ec04e12009-09-16 17:40:48 +02001823 setup_pager();
1824 read_events();
1825 print_bad_events();
1826}
1827
1828static void __cmd_replay(void)
1829{
1830 unsigned long i;
1831
1832 calibrate_run_measurement_overhead();
1833 calibrate_sleep_measurement_overhead();
1834
1835 test_calibrations();
1836
1837 read_events();
1838
1839 printf("nr_run_events: %ld\n", nr_run_events);
1840 printf("nr_sleep_events: %ld\n", nr_sleep_events);
1841 printf("nr_wakeup_events: %ld\n", nr_wakeup_events);
1842
1843 if (targetless_wakeups)
1844 printf("target-less wakeups: %ld\n", targetless_wakeups);
1845 if (multitarget_wakeups)
1846 printf("multi-target wakeups: %ld\n", multitarget_wakeups);
1847 if (nr_run_events_optimized)
1848 printf("run atoms optimized: %ld\n",
1849 nr_run_events_optimized);
1850
1851 print_task_traces();
1852 add_cross_task_wakeups();
1853
1854 create_tasks();
1855 printf("------------------------------------------------------------\n");
1856 for (i = 0; i < replay_repeat; i++)
1857 run_one_test();
1858}
1859
1860
Ingo Molnar46f392c2009-09-12 10:08:34 +02001861static const char * const sched_usage[] = {
Mike Galbraith4b77a722009-09-18 08:22:24 +02001862 "perf sched [<options>] {record|latency|map|replay|trace}",
Ingo Molnar0a02ad92009-09-11 12:12:54 +02001863 NULL
1864};
1865
Ingo Molnarf2858d82009-09-11 12:12:54 +02001866static const struct option sched_options[] = {
Mike Galbraith4b77a722009-09-18 08:22:24 +02001867 OPT_STRING('i', "input", &input_name, "file",
1868 "input file name"),
Ingo Molnarf2858d82009-09-11 12:12:54 +02001869 OPT_BOOLEAN('v', "verbose", &verbose,
1870 "be more verbose (show symbol address, etc)"),
Ingo Molnar0a02ad92009-09-11 12:12:54 +02001871 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1872 "dump raw trace in ASCII"),
Ingo Molnarf2858d82009-09-11 12:12:54 +02001873 OPT_END()
1874};
1875
1876static const char * const latency_usage[] = {
1877 "perf sched latency [<options>]",
1878 NULL
1879};
1880
1881static const struct option latency_options[] = {
Frederic Weisbeckerdaa1d7a2009-09-13 03:36:29 +02001882 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1883 "sort by key(s): runtime, switch, avg, max"),
Ingo Molnar0a02ad92009-09-11 12:12:54 +02001884 OPT_BOOLEAN('v', "verbose", &verbose,
1885 "be more verbose (show symbol address, etc)"),
Ingo Molnarf2858d82009-09-11 12:12:54 +02001886 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1887 "dump raw trace in ASCII"),
1888 OPT_END()
1889};
1890
1891static const char * const replay_usage[] = {
1892 "perf sched replay [<options>]",
1893 NULL
1894};
1895
1896static const struct option replay_options[] = {
1897 OPT_INTEGER('r', "repeat", &replay_repeat,
1898 "repeat the workload replay N times (-1: infinite)"),
1899 OPT_BOOLEAN('v', "verbose", &verbose,
1900 "be more verbose (show symbol address, etc)"),
1901 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1902 "dump raw trace in ASCII"),
Ingo Molnar0a02ad92009-09-11 12:12:54 +02001903 OPT_END()
1904};
1905
Frederic Weisbeckerdaa1d7a2009-09-13 03:36:29 +02001906static void setup_sorting(void)
1907{
1908 char *tmp, *tok, *str = strdup(sort_order);
1909
1910 for (tok = strtok_r(str, ", ", &tmp);
1911 tok; tok = strtok_r(NULL, ", ", &tmp)) {
1912 if (sort_dimension__add(tok, &sort_list) < 0) {
1913 error("Unknown --sort key: `%s'", tok);
Ingo Molnarf2858d82009-09-11 12:12:54 +02001914 usage_with_options(latency_usage, latency_options);
Frederic Weisbeckerdaa1d7a2009-09-13 03:36:29 +02001915 }
1916 }
1917
1918 free(str);
1919
1920 sort_dimension__add((char *)"pid", &cmp_pid);
1921}
1922
Ingo Molnar1fc35b22009-09-13 09:44:29 +02001923static const char *record_args[] = {
1924 "record",
1925 "-a",
1926 "-R",
Frederic Weisbeckerd1302522009-09-14 08:57:15 +02001927 "-M",
Ingo Molnarea57c4f2009-09-13 18:15:54 +02001928 "-f",
Ingo Molnardc02bf72009-09-16 13:45:00 +02001929 "-m", "1024",
Ingo Molnar1fc35b22009-09-13 09:44:29 +02001930 "-c", "1",
1931 "-e", "sched:sched_switch:r",
1932 "-e", "sched:sched_stat_wait:r",
1933 "-e", "sched:sched_stat_sleep:r",
1934 "-e", "sched:sched_stat_iowait:r",
Ingo Molnarea57c4f2009-09-13 18:15:54 +02001935 "-e", "sched:sched_stat_runtime:r",
Ingo Molnar1fc35b22009-09-13 09:44:29 +02001936 "-e", "sched:sched_process_exit:r",
1937 "-e", "sched:sched_process_fork:r",
1938 "-e", "sched:sched_wakeup:r",
1939 "-e", "sched:sched_migrate_task:r",
1940};
1941
1942static int __cmd_record(int argc, const char **argv)
1943{
1944 unsigned int rec_argc, i, j;
1945 const char **rec_argv;
1946
1947 rec_argc = ARRAY_SIZE(record_args) + argc - 1;
1948 rec_argv = calloc(rec_argc + 1, sizeof(char *));
1949
1950 for (i = 0; i < ARRAY_SIZE(record_args); i++)
1951 rec_argv[i] = strdup(record_args[i]);
1952
1953 for (j = 1; j < (unsigned int)argc; j++, i++)
1954 rec_argv[i] = argv[j];
1955
1956 BUG_ON(i != rec_argc);
1957
1958 return cmd_record(i, rec_argv, NULL);
1959}
1960
Ingo Molnar0a02ad92009-09-11 12:12:54 +02001961int cmd_sched(int argc, const char **argv, const char *prefix __used)
1962{
1963 symbol__init();
1964 page_size = getpagesize();
1965
Ingo Molnarf2858d82009-09-11 12:12:54 +02001966 argc = parse_options(argc, argv, sched_options, sched_usage,
1967 PARSE_OPT_STOP_AT_NON_OPTION);
1968 if (!argc)
1969 usage_with_options(sched_usage, sched_options);
1970
Ingo Molnar1fc35b22009-09-13 09:44:29 +02001971 if (!strncmp(argv[0], "rec", 3)) {
1972 return __cmd_record(argc, argv);
1973 } else if (!strncmp(argv[0], "lat", 3)) {
Ingo Molnarf2858d82009-09-11 12:12:54 +02001974 trace_handler = &lat_ops;
1975 if (argc > 1) {
1976 argc = parse_options(argc, argv, latency_options, latency_usage, 0);
1977 if (argc)
1978 usage_with_options(latency_usage, latency_options);
Ingo Molnarf2858d82009-09-11 12:12:54 +02001979 }
Ingo Molnarb5fae122009-09-11 12:12:54 +02001980 setup_sorting();
Ingo Molnarf2858d82009-09-11 12:12:54 +02001981 __cmd_lat();
Ingo Molnar0ec04e12009-09-16 17:40:48 +02001982 } else if (!strcmp(argv[0], "map")) {
1983 trace_handler = &map_ops;
1984 setup_sorting();
1985 __cmd_map();
Ingo Molnarf2858d82009-09-11 12:12:54 +02001986 } else if (!strncmp(argv[0], "rep", 3)) {
1987 trace_handler = &replay_ops;
1988 if (argc) {
1989 argc = parse_options(argc, argv, replay_options, replay_usage, 0);
1990 if (argc)
1991 usage_with_options(replay_usage, replay_options);
1992 }
1993 __cmd_replay();
Ingo Molnarc13f0d32009-09-13 16:51:04 +02001994 } else if (!strcmp(argv[0], "trace")) {
1995 /*
1996 * Aliased to 'perf trace' for now:
1997 */
1998 return cmd_trace(argc, argv, prefix);
Ingo Molnarf2858d82009-09-11 12:12:54 +02001999 } else {
2000 usage_with_options(sched_usage, sched_options);
Ingo Molnar0a02ad92009-09-11 12:12:54 +02002001 }
2002
Ingo Molnarec156762009-09-11 12:12:54 +02002003 return 0;
Ingo Molnar0a02ad92009-09-11 12:12:54 +02002004}