Commit | Line | Data |
---|---|---|
ebac4655 JA |
1 | /* |
2 | * fio - the flexible io tester | |
3 | * | |
4 | * Copyright (C) 2005 Jens Axboe <axboe@suse.de> | |
aae22ca7 | 5 | * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk> |
ebac4655 JA |
6 | * |
7 | * This program is free software; you can redistribute it and/or modify | |
8 | * it under the terms of the GNU General Public License as published by | |
9 | * the Free Software Foundation; either version 2 of the License, or | |
10 | * (at your option) any later version. | |
11 | * | |
12 | * This program is distributed in the hope that it will be useful, | |
13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
15 | * GNU General Public License for more details. | |
16 | * | |
17 | * You should have received a copy of the GNU General Public License | |
18 | * along with this program; if not, write to the Free Software | |
19 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
20 | * | |
21 | */ | |
ebac4655 JA |
22 | #include <unistd.h> |
23 | #include <fcntl.h> | |
24 | #include <string.h> | |
ebac4655 JA |
25 | #include <signal.h> |
26 | #include <time.h> | |
ebac4655 | 27 | #include <assert.h> |
ebac4655 JA |
28 | #include <sys/stat.h> |
29 | #include <sys/wait.h> | |
30 | #include <sys/ipc.h> | |
31 | #include <sys/shm.h> | |
32 | #include <sys/ioctl.h> | |
33 | #include <sys/mman.h> | |
34 | ||
35 | #include "fio.h" | |
36 | #include "os.h" | |
37 | ||
38 | #define MASK (4095) | |
39 | ||
40 | #define ALIGN(buf) (char *) (((unsigned long) (buf) + MASK) & ~(MASK)) | |
41 | ||
42 | int groupid = 0; | |
43 | int thread_number = 0; | |
ebac4655 | 44 | int shm_id = 0; |
53cdc686 | 45 | int temp_stall_ts; |
c1d5725e | 46 | char *fio_inst_prefix = _INST_PREFIX; |
ebac4655 | 47 | |
bbfd6b00 | 48 | static volatile int startup_sem; |
ebac4655 JA |
49 | |
50 | #define TERMINATE_ALL (-1) | |
75154845 | 51 | #define JOB_START_TIMEOUT (5 * 1000) |
ebac4655 JA |
52 | |
53 | static void terminate_threads(int group_id) | |
54 | { | |
34572e28 | 55 | struct thread_data *td; |
ebac4655 JA |
56 | int i; |
57 | ||
34572e28 | 58 | for_each_td(td, i) { |
ebac4655 JA |
59 | if (group_id == TERMINATE_ALL || groupid == td->groupid) { |
60 | td->terminate = 1; | |
61 | td->start_delay = 0; | |
62 | } | |
63 | } | |
64 | } | |
65 | ||
66 | static void sig_handler(int sig) | |
67 | { | |
68 | switch (sig) { | |
69 | case SIGALRM: | |
70 | update_io_ticks(); | |
71 | disk_util_timer_arm(); | |
72 | print_thread_status(); | |
73 | break; | |
74 | default: | |
75 | printf("\nfio: terminating on signal\n"); | |
76 | fflush(stdout); | |
77 | terminate_threads(TERMINATE_ALL); | |
78 | break; | |
79 | } | |
80 | } | |
81 | ||
906c8d75 JA |
82 | /* |
83 | * Check if we are above the minimum rate given. | |
84 | */ | |
ebac4655 JA |
85 | static int check_min_rate(struct thread_data *td, struct timeval *now) |
86 | { | |
87 | unsigned long spent; | |
88 | unsigned long rate; | |
89 | int ddir = td->ddir; | |
90 | ||
91 | /* | |
92 | * allow a 2 second settle period in the beginning | |
93 | */ | |
94 | if (mtime_since(&td->start, now) < 2000) | |
95 | return 0; | |
96 | ||
97 | /* | |
98 | * if rate blocks is set, sample is running | |
99 | */ | |
100 | if (td->rate_bytes) { | |
101 | spent = mtime_since(&td->lastrate, now); | |
102 | if (spent < td->ratecycle) | |
103 | return 0; | |
104 | ||
105 | rate = (td->this_io_bytes[ddir] - td->rate_bytes) / spent; | |
106 | if (rate < td->ratemin) { | |
eb8bbf48 | 107 | fprintf(f_out, "%s: min rate %d not met, got %ldKiB/sec\n", td->name, td->ratemin, rate); |
ebac4655 JA |
108 | return 1; |
109 | } | |
110 | } | |
111 | ||
112 | td->rate_bytes = td->this_io_bytes[ddir]; | |
113 | memcpy(&td->lastrate, now, sizeof(*now)); | |
114 | return 0; | |
115 | } | |
116 | ||
117 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) | |
118 | { | |
119 | if (!td->timeout) | |
120 | return 0; | |
121 | if (mtime_since(&td->epoch, t) >= td->timeout * 1000) | |
122 | return 1; | |
123 | ||
124 | return 0; | |
125 | } | |
126 | ||
ebac4655 JA |
127 | static inline void td_set_runstate(struct thread_data *td, int runstate) |
128 | { | |
ebac4655 JA |
129 | td->runstate = runstate; |
130 | } | |
131 | ||
53cdc686 JA |
132 | static struct fio_file *get_next_file(struct thread_data *td) |
133 | { | |
0ab8db89 | 134 | unsigned int old_next_file = td->next_file; |
b2a15192 JA |
135 | struct fio_file *f; |
136 | ||
137 | do { | |
138 | f = &td->files[td->next_file]; | |
139 | ||
140 | td->next_file++; | |
141 | if (td->next_file >= td->nr_files) | |
142 | td->next_file = 0; | |
143 | ||
144 | if (f->fd != -1) | |
145 | break; | |
53cdc686 | 146 | |
b2a15192 JA |
147 | f = NULL; |
148 | } while (td->next_file != old_next_file); | |
53cdc686 JA |
149 | |
150 | return f; | |
151 | } | |
152 | ||
906c8d75 JA |
153 | /* |
154 | * When job exits, we can cancel the in-flight IO if we are using async | |
155 | * io. Attempt to do so. | |
156 | */ | |
ebac4655 JA |
157 | static void cleanup_pending_aio(struct thread_data *td) |
158 | { | |
159 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0}; | |
160 | struct list_head *entry, *n; | |
161 | struct io_completion_data icd; | |
162 | struct io_u *io_u; | |
163 | int r; | |
164 | ||
165 | /* | |
166 | * get immediately available events, if any | |
167 | */ | |
45bee283 | 168 | r = td_io_getevents(td, 0, td->cur_depth, &ts); |
ebac4655 JA |
169 | if (r > 0) { |
170 | icd.nr = r; | |
171 | ios_completed(td, &icd); | |
172 | } | |
173 | ||
174 | /* | |
175 | * now cancel remaining active events | |
176 | */ | |
2866c82d | 177 | if (td->io_ops->cancel) { |
ebac4655 JA |
178 | list_for_each_safe(entry, n, &td->io_u_busylist) { |
179 | io_u = list_entry(entry, struct io_u, list); | |
180 | ||
2866c82d | 181 | r = td->io_ops->cancel(td, io_u); |
ebac4655 JA |
182 | if (!r) |
183 | put_io_u(td, io_u); | |
184 | } | |
185 | } | |
186 | ||
187 | if (td->cur_depth) { | |
45bee283 | 188 | r = td_io_getevents(td, td->cur_depth, td->cur_depth, NULL); |
ebac4655 JA |
189 | if (r > 0) { |
190 | icd.nr = r; | |
191 | ios_completed(td, &icd); | |
192 | } | |
193 | } | |
194 | } | |
195 | ||
906c8d75 JA |
196 | /* |
197 | * The main verify engine. Runs over the writes we previusly submitted, | |
198 | * reads the blocks back in, and checks the crc/md5 of the data. | |
199 | */ | |
a9619d44 | 200 | void do_verify(struct thread_data *td) |
ebac4655 JA |
201 | { |
202 | struct timeval t; | |
203 | struct io_u *io_u, *v_io_u = NULL; | |
204 | struct io_completion_data icd; | |
53cdc686 | 205 | struct fio_file *f; |
e5b401d4 JA |
206 | int ret, i; |
207 | ||
208 | /* | |
209 | * sync io first and invalidate cache, to make sure we really | |
210 | * read from disk. | |
211 | */ | |
212 | for_each_file(td, f, i) { | |
213 | td_io_sync(td, f); | |
214 | file_invalidate_cache(td, f); | |
215 | } | |
ebac4655 JA |
216 | |
217 | td_set_runstate(td, TD_VERIFYING); | |
218 | ||
219 | do { | |
220 | if (td->terminate) | |
221 | break; | |
222 | ||
223 | gettimeofday(&t, NULL); | |
224 | if (runtime_exceeded(td, &t)) | |
225 | break; | |
226 | ||
227 | io_u = __get_io_u(td); | |
228 | if (!io_u) | |
229 | break; | |
230 | ||
aea47d44 | 231 | if (get_next_verify(td, io_u)) { |
ebac4655 JA |
232 | put_io_u(td, io_u); |
233 | break; | |
234 | } | |
235 | ||
53cdc686 JA |
236 | f = get_next_file(td); |
237 | if (!f) | |
238 | break; | |
239 | ||
240 | io_u->file = f; | |
241 | ||
aea47d44 | 242 | if (td_io_prep(td, io_u)) { |
ebac4655 JA |
243 | put_io_u(td, io_u); |
244 | break; | |
245 | } | |
246 | ||
45bee283 | 247 | ret = td_io_queue(td, io_u); |
ebac4655 JA |
248 | if (ret) { |
249 | put_io_u(td, io_u); | |
250 | td_verror(td, ret); | |
251 | break; | |
252 | } | |
253 | ||
254 | /* | |
255 | * we have one pending to verify, do that while | |
256 | * we are doing io on the next one | |
257 | */ | |
258 | if (do_io_u_verify(td, &v_io_u)) | |
259 | break; | |
260 | ||
45bee283 | 261 | ret = td_io_getevents(td, 1, 1, NULL); |
ebac4655 JA |
262 | if (ret != 1) { |
263 | if (ret < 0) | |
264 | td_verror(td, ret); | |
265 | break; | |
266 | } | |
267 | ||
2866c82d | 268 | v_io_u = td->io_ops->event(td, 0); |
ebac4655 JA |
269 | icd.nr = 1; |
270 | icd.error = 0; | |
271 | io_completed(td, v_io_u, &icd); | |
272 | ||
273 | if (icd.error) { | |
274 | td_verror(td, icd.error); | |
275 | put_io_u(td, v_io_u); | |
276 | v_io_u = NULL; | |
277 | break; | |
278 | } | |
279 | ||
ebac4655 JA |
280 | /* |
281 | * if we can't submit more io, we need to verify now | |
282 | */ | |
283 | if (queue_full(td) && do_io_u_verify(td, &v_io_u)) | |
284 | break; | |
285 | ||
286 | } while (1); | |
287 | ||
288 | do_io_u_verify(td, &v_io_u); | |
289 | ||
290 | if (td->cur_depth) | |
291 | cleanup_pending_aio(td); | |
292 | ||
293 | td_set_runstate(td, TD_RUNNING); | |
294 | } | |
295 | ||
b990b5c0 JA |
296 | /* |
297 | * Not really an io thread, all it does is burn CPU cycles in the specified | |
298 | * manner. | |
299 | */ | |
300 | static void do_cpuio(struct thread_data *td) | |
301 | { | |
302 | struct timeval e; | |
303 | int split = 100 / td->cpuload; | |
304 | int i = 0; | |
305 | ||
306 | while (!td->terminate) { | |
307 | gettimeofday(&e, NULL); | |
308 | ||
309 | if (runtime_exceeded(td, &e)) | |
310 | break; | |
311 | ||
312 | if (!(i % split)) | |
313 | __usec_sleep(10000); | |
314 | else | |
315 | usec_sleep(td, 10000); | |
316 | ||
317 | i++; | |
318 | } | |
319 | } | |
320 | ||
32cd46a0 | 321 | /* |
906c8d75 | 322 | * Main IO worker function. It retrieves io_u's to process and queues |
32cd46a0 JA |
323 | * and reaps them, checking for rate and errors along the way. |
324 | */ | |
ebac4655 JA |
325 | static void do_io(struct thread_data *td) |
326 | { | |
327 | struct io_completion_data icd; | |
328 | struct timeval s, e; | |
329 | unsigned long usec; | |
53cdc686 | 330 | struct fio_file *f; |
84585003 | 331 | int i, ret = 0; |
ebac4655 | 332 | |
5853e5a8 JA |
333 | td_set_runstate(td, TD_RUNNING); |
334 | ||
ebac4655 JA |
335 | while (td->this_io_bytes[td->ddir] < td->io_size) { |
336 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0}; | |
337 | struct timespec *timeout; | |
84585003 | 338 | int min_evts = 0; |
ebac4655 JA |
339 | struct io_u *io_u; |
340 | ||
341 | if (td->terminate) | |
342 | break; | |
343 | ||
53cdc686 JA |
344 | f = get_next_file(td); |
345 | if (!f) | |
346 | break; | |
347 | ||
348 | io_u = get_io_u(td, f); | |
ebac4655 JA |
349 | if (!io_u) |
350 | break; | |
351 | ||
352 | memcpy(&s, &io_u->start_time, sizeof(s)); | |
353 | ||
45bee283 | 354 | ret = td_io_queue(td, io_u); |
ebac4655 JA |
355 | if (ret) { |
356 | put_io_u(td, io_u); | |
357 | td_verror(td, ret); | |
358 | break; | |
359 | } | |
360 | ||
361 | add_slat_sample(td, io_u->ddir, mtime_since(&io_u->start_time, &io_u->issue_time)); | |
362 | ||
363 | if (td->cur_depth < td->iodepth) { | |
364 | timeout = &ts; | |
365 | min_evts = 0; | |
366 | } else { | |
367 | timeout = NULL; | |
368 | min_evts = 1; | |
369 | } | |
370 | ||
45bee283 | 371 | ret = td_io_getevents(td, min_evts, td->cur_depth, timeout); |
ebac4655 | 372 | if (ret < 0) { |
87dc1ab1 | 373 | td_verror(td, ret); |
ebac4655 JA |
374 | break; |
375 | } else if (!ret) | |
376 | continue; | |
377 | ||
378 | icd.nr = ret; | |
379 | ios_completed(td, &icd); | |
380 | if (icd.error) { | |
381 | td_verror(td, icd.error); | |
382 | break; | |
383 | } | |
384 | ||
385 | /* | |
386 | * the rate is batched for now, it should work for batches | |
387 | * of completions except the very first one which may look | |
388 | * a little bursty | |
389 | */ | |
390 | gettimeofday(&e, NULL); | |
391 | usec = utime_since(&s, &e); | |
392 | ||
393 | rate_throttle(td, usec, icd.bytes_done[td->ddir]); | |
394 | ||
395 | if (check_min_rate(td, &e)) { | |
2f9ade3c JA |
396 | if (rate_quit) |
397 | terminate_threads(td->groupid); | |
ebac4655 JA |
398 | td_verror(td, ENOMEM); |
399 | break; | |
400 | } | |
401 | ||
402 | if (runtime_exceeded(td, &e)) | |
403 | break; | |
404 | ||
405 | if (td->thinktime) | |
406 | usec_sleep(td, td->thinktime); | |
ebac4655 JA |
407 | } |
408 | ||
84585003 JA |
409 | if (!ret) { |
410 | if (td->cur_depth) | |
411 | cleanup_pending_aio(td); | |
ebac4655 | 412 | |
84585003 JA |
413 | if (should_fsync(td) && td->end_fsync) { |
414 | td_set_runstate(td, TD_FSYNCING); | |
415 | for_each_file(td, f, i) | |
416 | td_io_sync(td, f); | |
417 | } | |
5853e5a8 | 418 | } |
ebac4655 JA |
419 | } |
420 | ||
ebac4655 JA |
421 | static void cleanup_io_u(struct thread_data *td) |
422 | { | |
423 | struct list_head *entry, *n; | |
424 | struct io_u *io_u; | |
425 | ||
426 | list_for_each_safe(entry, n, &td->io_u_freelist) { | |
427 | io_u = list_entry(entry, struct io_u, list); | |
428 | ||
429 | list_del(&io_u->list); | |
430 | free(io_u); | |
431 | } | |
432 | ||
2f9ade3c | 433 | free_io_mem(td); |
ebac4655 JA |
434 | } |
435 | ||
436 | static int init_io_u(struct thread_data *td) | |
437 | { | |
438 | struct io_u *io_u; | |
439 | int i, max_units; | |
440 | char *p; | |
441 | ||
2866c82d | 442 | if (td->io_ops->flags & FIO_CPUIO) |
b990b5c0 JA |
443 | return 0; |
444 | ||
2866c82d | 445 | if (td->io_ops->flags & FIO_SYNCIO) |
ebac4655 JA |
446 | max_units = 1; |
447 | else | |
448 | max_units = td->iodepth; | |
449 | ||
450 | td->orig_buffer_size = td->max_bs * max_units + MASK; | |
451 | ||
2f9ade3c JA |
452 | if (allocate_io_mem(td)) |
453 | return 1; | |
ebac4655 | 454 | |
ebac4655 JA |
455 | p = ALIGN(td->orig_buffer); |
456 | for (i = 0; i < max_units; i++) { | |
457 | io_u = malloc(sizeof(*io_u)); | |
458 | memset(io_u, 0, sizeof(*io_u)); | |
459 | INIT_LIST_HEAD(&io_u->list); | |
460 | ||
461 | io_u->buf = p + td->max_bs * i; | |
b1ff3403 | 462 | io_u->index = i; |
ebac4655 JA |
463 | list_add(&io_u->list, &td->io_u_freelist); |
464 | } | |
465 | ||
466 | return 0; | |
467 | } | |
468 | ||
da86774e JA |
469 | static int switch_ioscheduler(struct thread_data *td) |
470 | { | |
471 | char tmp[256], tmp2[128]; | |
472 | FILE *f; | |
473 | int ret; | |
474 | ||
475 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); | |
476 | ||
477 | f = fopen(tmp, "r+"); | |
478 | if (!f) { | |
479 | td_verror(td, errno); | |
480 | return 1; | |
481 | } | |
482 | ||
483 | /* | |
484 | * Set io scheduler. | |
485 | */ | |
486 | ret = fwrite(td->ioscheduler, strlen(td->ioscheduler), 1, f); | |
487 | if (ferror(f) || ret != 1) { | |
488 | td_verror(td, errno); | |
489 | fclose(f); | |
490 | return 1; | |
491 | } | |
492 | ||
493 | rewind(f); | |
494 | ||
495 | /* | |
496 | * Read back and check that the selected scheduler is now the default. | |
497 | */ | |
498 | ret = fread(tmp, 1, sizeof(tmp), f); | |
499 | if (ferror(f) || ret < 0) { | |
500 | td_verror(td, errno); | |
501 | fclose(f); | |
502 | return 1; | |
503 | } | |
504 | ||
505 | sprintf(tmp2, "[%s]", td->ioscheduler); | |
506 | if (!strstr(tmp, tmp2)) { | |
3b70d7e5 | 507 | log_err("fio: io scheduler %s not found\n", td->ioscheduler); |
da86774e JA |
508 | td_verror(td, EINVAL); |
509 | fclose(f); | |
510 | return 1; | |
511 | } | |
512 | ||
513 | fclose(f); | |
514 | return 0; | |
515 | } | |
516 | ||
ebac4655 JA |
517 | static void clear_io_state(struct thread_data *td) |
518 | { | |
53cdc686 JA |
519 | struct fio_file *f; |
520 | int i; | |
ebac4655 | 521 | |
ebac4655 JA |
522 | td->stat_io_bytes[0] = td->stat_io_bytes[1] = 0; |
523 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; | |
20dc95c4 | 524 | td->zone_bytes = 0; |
ebac4655 | 525 | |
53cdc686 JA |
526 | for_each_file(td, f, i) { |
527 | f->last_pos = 0; | |
528 | if (td->io_ops->flags & FIO_SYNCIO) | |
529 | lseek(f->fd, SEEK_SET, 0); | |
530 | ||
531 | if (f->file_map) | |
532 | memset(f->file_map, 0, f->num_maps * sizeof(long)); | |
533 | } | |
ebac4655 JA |
534 | } |
535 | ||
906c8d75 JA |
536 | /* |
537 | * Entry point for the thread based jobs. The process based jobs end up | |
538 | * here as well, after a little setup. | |
539 | */ | |
ebac4655 JA |
540 | static void *thread_main(void *data) |
541 | { | |
542 | struct thread_data *td = data; | |
ebac4655 JA |
543 | |
544 | if (!td->use_thread) | |
545 | setsid(); | |
546 | ||
547 | td->pid = getpid(); | |
548 | ||
aea47d44 JA |
549 | INIT_LIST_HEAD(&td->io_u_freelist); |
550 | INIT_LIST_HEAD(&td->io_u_busylist); | |
551 | INIT_LIST_HEAD(&td->io_hist_list); | |
552 | INIT_LIST_HEAD(&td->io_log_list); | |
553 | ||
ebac4655 JA |
554 | if (init_io_u(td)) |
555 | goto err; | |
556 | ||
557 | if (fio_setaffinity(td) == -1) { | |
558 | td_verror(td, errno); | |
559 | goto err; | |
560 | } | |
561 | ||
45bee283 | 562 | if (td_io_init(td)) |
ebac4655 JA |
563 | goto err; |
564 | ||
aea47d44 JA |
565 | if (init_iolog(td)) |
566 | goto err; | |
567 | ||
ebac4655 JA |
568 | if (td->ioprio) { |
569 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { | |
570 | td_verror(td, errno); | |
571 | goto err; | |
572 | } | |
573 | } | |
574 | ||
1056eaad | 575 | if (nice(td->nice) == -1) { |
b6f4d880 JA |
576 | td_verror(td, errno); |
577 | goto err; | |
578 | } | |
579 | ||
75154845 JA |
580 | if (init_random_state(td)) |
581 | goto err; | |
582 | ||
56f9498d JA |
583 | if (td->ioscheduler && switch_ioscheduler(td)) |
584 | goto err; | |
da86774e | 585 | |
75154845 | 586 | td_set_runstate(td, TD_INITIALIZED); |
bbfd6b00 JA |
587 | fio_sem_up(&startup_sem); |
588 | fio_sem_down(&td->mutex); | |
ebac4655 | 589 | |
53cdc686 | 590 | if (!td->create_serialize && setup_files(td)) |
ebac4655 JA |
591 | goto err; |
592 | ||
ebac4655 JA |
593 | gettimeofday(&td->epoch, NULL); |
594 | ||
56f9498d | 595 | if (td->exec_prerun) |
4e0ba8af JA |
596 | system(td->exec_prerun); |
597 | ||
ebac4655 JA |
598 | while (td->loops--) { |
599 | getrusage(RUSAGE_SELF, &td->ru_start); | |
600 | gettimeofday(&td->start, NULL); | |
601 | memcpy(&td->stat_sample_time, &td->start, sizeof(td->start)); | |
602 | ||
603 | if (td->ratemin) | |
604 | memcpy(&td->lastrate, &td->stat_sample_time, sizeof(td->lastrate)); | |
605 | ||
606 | clear_io_state(td); | |
607 | prune_io_piece_log(td); | |
608 | ||
2866c82d | 609 | if (td->io_ops->flags & FIO_CPUIO) |
b990b5c0 JA |
610 | do_cpuio(td); |
611 | else | |
612 | do_io(td); | |
ebac4655 JA |
613 | |
614 | td->runtime[td->ddir] += mtime_since_now(&td->start); | |
aea47d44 | 615 | if (td_rw(td) && td->io_bytes[td->ddir ^ 1]) |
3d60d1ed JA |
616 | td->runtime[td->ddir ^ 1] = td->runtime[td->ddir]; |
617 | ||
ebac4655 JA |
618 | update_rusage_stat(td); |
619 | ||
620 | if (td->error || td->terminate) | |
621 | break; | |
622 | ||
623 | if (td->verify == VERIFY_NONE) | |
624 | continue; | |
625 | ||
626 | clear_io_state(td); | |
627 | gettimeofday(&td->start, NULL); | |
628 | ||
629 | do_verify(td); | |
630 | ||
631 | td->runtime[DDIR_READ] += mtime_since_now(&td->start); | |
632 | ||
633 | if (td->error || td->terminate) | |
634 | break; | |
635 | } | |
636 | ||
ebac4655 JA |
637 | if (td->bw_log) |
638 | finish_log(td, td->bw_log, "bw"); | |
639 | if (td->slat_log) | |
640 | finish_log(td, td->slat_log, "slat"); | |
641 | if (td->clat_log) | |
642 | finish_log(td, td->clat_log, "clat"); | |
843a7413 JA |
643 | if (td->write_iolog) |
644 | write_iolog_close(td); | |
56f9498d | 645 | if (td->exec_postrun) |
4e0ba8af | 646 | system(td->exec_postrun); |
ebac4655 JA |
647 | |
648 | if (exitall_on_terminate) | |
649 | terminate_threads(td->groupid); | |
650 | ||
651 | err: | |
53cdc686 | 652 | close_files(td); |
2866c82d | 653 | close_ioengine(td); |
ebac4655 | 654 | cleanup_io_u(td); |
ebac4655 JA |
655 | td_set_runstate(td, TD_EXITED); |
656 | return NULL; | |
657 | ||
658 | } | |
659 | ||
906c8d75 JA |
660 | /* |
661 | * We cannot pass the td data into a forked process, so attach the td and | |
662 | * pass it to the thread worker. | |
663 | */ | |
ebac4655 JA |
664 | static void *fork_main(int shmid, int offset) |
665 | { | |
666 | struct thread_data *td; | |
667 | void *data; | |
668 | ||
669 | data = shmat(shmid, NULL, 0); | |
670 | if (data == (void *) -1) { | |
671 | perror("shmat"); | |
672 | return NULL; | |
673 | } | |
674 | ||
675 | td = data + offset * sizeof(struct thread_data); | |
676 | thread_main(td); | |
677 | shmdt(data); | |
678 | return NULL; | |
679 | } | |
680 | ||
906c8d75 JA |
681 | /* |
682 | * Run over the job map and reap the threads that have exited, if any. | |
683 | */ | |
ebac4655 JA |
684 | static void reap_threads(int *nr_running, int *t_rate, int *m_rate) |
685 | { | |
34572e28 | 686 | struct thread_data *td; |
b990b5c0 | 687 | int i, cputhreads; |
ebac4655 JA |
688 | |
689 | /* | |
690 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
691 | */ | |
34572e28 JA |
692 | cputhreads = 0; |
693 | for_each_td(td, i) { | |
84585003 JA |
694 | /* |
695 | * ->io_ops is NULL for a thread that has closed its | |
696 | * io engine | |
697 | */ | |
698 | if (td->io_ops && td->io_ops->flags & FIO_CPUIO) | |
b990b5c0 JA |
699 | cputhreads++; |
700 | ||
ebac4655 JA |
701 | if (td->runstate != TD_EXITED) |
702 | continue; | |
703 | ||
704 | td_set_runstate(td, TD_REAPED); | |
705 | ||
706 | if (td->use_thread) { | |
707 | long ret; | |
708 | ||
709 | if (pthread_join(td->thread, (void *) &ret)) | |
710 | perror("thread_join"); | |
711 | } else | |
712 | waitpid(td->pid, NULL, 0); | |
713 | ||
714 | (*nr_running)--; | |
715 | (*m_rate) -= td->ratemin; | |
716 | (*t_rate) -= td->rate; | |
717 | } | |
b990b5c0 JA |
718 | |
719 | if (*nr_running == cputhreads) | |
720 | terminate_threads(TERMINATE_ALL); | |
ebac4655 JA |
721 | } |
722 | ||
906c8d75 JA |
723 | /* |
724 | * Main function for kicking off and reaping jobs, as needed. | |
725 | */ | |
ebac4655 JA |
726 | static void run_threads(void) |
727 | { | |
ebac4655 JA |
728 | struct thread_data *td; |
729 | unsigned long spent; | |
730 | int i, todo, nr_running, m_rate, t_rate, nr_started; | |
fcb6ade2 | 731 | |
2f9ade3c JA |
732 | if (fio_pin_memory()) |
733 | return; | |
ebac4655 | 734 | |
c6ae0a5b JA |
735 | if (!terse_output) { |
736 | printf("Starting %d thread%s\n", thread_number, thread_number > 1 ? "s" : ""); | |
737 | fflush(stdout); | |
738 | } | |
c04f7ec3 | 739 | |
4efa970e JA |
740 | signal(SIGINT, sig_handler); |
741 | signal(SIGALRM, sig_handler); | |
742 | ||
ebac4655 JA |
743 | todo = thread_number; |
744 | nr_running = 0; | |
745 | nr_started = 0; | |
746 | m_rate = t_rate = 0; | |
747 | ||
34572e28 | 748 | for_each_td(td, i) { |
263e529f | 749 | print_status_init(td->thread_number - 1); |
ebac4655 JA |
750 | |
751 | init_disk_util(td); | |
752 | ||
753 | if (!td->create_serialize) | |
754 | continue; | |
755 | ||
756 | /* | |
757 | * do file setup here so it happens sequentially, | |
758 | * we don't want X number of threads getting their | |
759 | * client data interspersed on disk | |
760 | */ | |
53cdc686 | 761 | if (setup_files(td)) { |
ebac4655 JA |
762 | td_set_runstate(td, TD_REAPED); |
763 | todo--; | |
764 | } | |
765 | } | |
766 | ||
263e529f | 767 | time_init(); |
ebac4655 JA |
768 | |
769 | while (todo) { | |
75154845 JA |
770 | struct thread_data *map[MAX_JOBS]; |
771 | struct timeval this_start; | |
772 | int this_jobs = 0, left; | |
773 | ||
ebac4655 JA |
774 | /* |
775 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
776 | */ | |
34572e28 | 777 | for_each_td(td, i) { |
ebac4655 JA |
778 | if (td->runstate != TD_NOT_CREATED) |
779 | continue; | |
780 | ||
781 | /* | |
782 | * never got a chance to start, killed by other | |
783 | * thread for some reason | |
784 | */ | |
785 | if (td->terminate) { | |
786 | todo--; | |
787 | continue; | |
788 | } | |
789 | ||
790 | if (td->start_delay) { | |
263e529f | 791 | spent = mtime_since_genesis(); |
ebac4655 JA |
792 | |
793 | if (td->start_delay * 1000 > spent) | |
794 | continue; | |
795 | } | |
796 | ||
797 | if (td->stonewall && (nr_started || nr_running)) | |
798 | break; | |
799 | ||
75154845 JA |
800 | /* |
801 | * Set state to created. Thread will transition | |
802 | * to TD_INITIALIZED when it's done setting up. | |
803 | */ | |
ebac4655 | 804 | td_set_runstate(td, TD_CREATED); |
75154845 | 805 | map[this_jobs++] = td; |
bbfd6b00 | 806 | fio_sem_init(&startup_sem, 1); |
ebac4655 JA |
807 | nr_started++; |
808 | ||
809 | if (td->use_thread) { | |
810 | if (pthread_create(&td->thread, NULL, thread_main, td)) { | |
811 | perror("thread_create"); | |
812 | nr_started--; | |
813 | } | |
814 | } else { | |
815 | if (fork()) | |
bbfd6b00 | 816 | fio_sem_down(&startup_sem); |
ebac4655 JA |
817 | else { |
818 | fork_main(shm_id, i); | |
819 | exit(0); | |
820 | } | |
821 | } | |
822 | } | |
823 | ||
824 | /* | |
75154845 JA |
825 | * Wait for the started threads to transition to |
826 | * TD_INITIALIZED. | |
ebac4655 | 827 | */ |
75154845 JA |
828 | gettimeofday(&this_start, NULL); |
829 | left = this_jobs; | |
830 | while (left) { | |
831 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) | |
832 | break; | |
833 | ||
834 | usleep(100000); | |
835 | ||
836 | for (i = 0; i < this_jobs; i++) { | |
837 | td = map[i]; | |
838 | if (!td) | |
839 | continue; | |
b6f4d880 | 840 | if (td->runstate == TD_INITIALIZED) { |
75154845 JA |
841 | map[i] = NULL; |
842 | left--; | |
b6f4d880 JA |
843 | } else if (td->runstate >= TD_EXITED) { |
844 | map[i] = NULL; | |
845 | left--; | |
846 | todo--; | |
847 | nr_running++; /* work-around... */ | |
75154845 JA |
848 | } |
849 | } | |
850 | } | |
851 | ||
852 | if (left) { | |
3b70d7e5 | 853 | log_err("fio: %d jobs failed to start\n", left); |
75154845 JA |
854 | for (i = 0; i < this_jobs; i++) { |
855 | td = map[i]; | |
856 | if (!td) | |
857 | continue; | |
858 | kill(td->pid, SIGTERM); | |
859 | } | |
860 | break; | |
861 | } | |
862 | ||
863 | /* | |
b6f4d880 | 864 | * start created threads (TD_INITIALIZED -> TD_RUNNING). |
75154845 | 865 | */ |
34572e28 | 866 | for_each_td(td, i) { |
75154845 | 867 | if (td->runstate != TD_INITIALIZED) |
ebac4655 JA |
868 | continue; |
869 | ||
870 | td_set_runstate(td, TD_RUNNING); | |
871 | nr_running++; | |
872 | nr_started--; | |
873 | m_rate += td->ratemin; | |
874 | t_rate += td->rate; | |
75154845 | 875 | todo--; |
bbfd6b00 | 876 | fio_sem_up(&td->mutex); |
ebac4655 JA |
877 | } |
878 | ||
879 | reap_threads(&nr_running, &t_rate, &m_rate); | |
880 | ||
881 | if (todo) | |
882 | usleep(100000); | |
883 | } | |
884 | ||
885 | while (nr_running) { | |
886 | reap_threads(&nr_running, &t_rate, &m_rate); | |
887 | usleep(10000); | |
888 | } | |
889 | ||
890 | update_io_ticks(); | |
2f9ade3c | 891 | fio_unpin_memory(); |
ebac4655 JA |
892 | } |
893 | ||
ebac4655 JA |
894 | int main(int argc, char *argv[]) |
895 | { | |
896 | if (parse_options(argc, argv)) | |
897 | return 1; | |
898 | ||
899 | if (!thread_number) { | |
3b70d7e5 | 900 | log_err("Nothing to do\n"); |
ebac4655 JA |
901 | return 1; |
902 | } | |
903 | ||
904 | disk_util_timer_arm(); | |
905 | ||
906 | run_threads(); | |
907 | show_run_stats(); | |
908 | ||
909 | return 0; | |
910 | } |