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 | |
3d60d1ed | 48 | #define should_fsync(td) ((td_write(td) || td_rw(td)) && (!(td)->odirect || (td)->override_sync)) |
ebac4655 | 49 | |
bbfd6b00 | 50 | static volatile int startup_sem; |
ebac4655 JA |
51 | |
52 | #define TERMINATE_ALL (-1) | |
75154845 | 53 | #define JOB_START_TIMEOUT (5 * 1000) |
ebac4655 JA |
54 | |
55 | static void terminate_threads(int group_id) | |
56 | { | |
34572e28 | 57 | struct thread_data *td; |
ebac4655 JA |
58 | int i; |
59 | ||
34572e28 | 60 | for_each_td(td, i) { |
ebac4655 JA |
61 | if (group_id == TERMINATE_ALL || groupid == td->groupid) { |
62 | td->terminate = 1; | |
63 | td->start_delay = 0; | |
64 | } | |
65 | } | |
66 | } | |
67 | ||
68 | static void sig_handler(int sig) | |
69 | { | |
70 | switch (sig) { | |
71 | case SIGALRM: | |
72 | update_io_ticks(); | |
73 | disk_util_timer_arm(); | |
74 | print_thread_status(); | |
75 | break; | |
76 | default: | |
77 | printf("\nfio: terminating on signal\n"); | |
78 | fflush(stdout); | |
79 | terminate_threads(TERMINATE_ALL); | |
80 | break; | |
81 | } | |
82 | } | |
83 | ||
906c8d75 JA |
84 | /* |
85 | * Check if we are above the minimum rate given. | |
86 | */ | |
ebac4655 JA |
87 | static int check_min_rate(struct thread_data *td, struct timeval *now) |
88 | { | |
89 | unsigned long spent; | |
90 | unsigned long rate; | |
91 | int ddir = td->ddir; | |
92 | ||
93 | /* | |
94 | * allow a 2 second settle period in the beginning | |
95 | */ | |
96 | if (mtime_since(&td->start, now) < 2000) | |
97 | return 0; | |
98 | ||
99 | /* | |
100 | * if rate blocks is set, sample is running | |
101 | */ | |
102 | if (td->rate_bytes) { | |
103 | spent = mtime_since(&td->lastrate, now); | |
104 | if (spent < td->ratecycle) | |
105 | return 0; | |
106 | ||
107 | rate = (td->this_io_bytes[ddir] - td->rate_bytes) / spent; | |
108 | if (rate < td->ratemin) { | |
eb8bbf48 | 109 | fprintf(f_out, "%s: min rate %d not met, got %ldKiB/sec\n", td->name, td->ratemin, rate); |
ebac4655 JA |
110 | return 1; |
111 | } | |
112 | } | |
113 | ||
114 | td->rate_bytes = td->this_io_bytes[ddir]; | |
115 | memcpy(&td->lastrate, now, sizeof(*now)); | |
116 | return 0; | |
117 | } | |
118 | ||
119 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) | |
120 | { | |
121 | if (!td->timeout) | |
122 | return 0; | |
123 | if (mtime_since(&td->epoch, t) >= td->timeout * 1000) | |
124 | return 1; | |
125 | ||
126 | return 0; | |
127 | } | |
128 | ||
ebac4655 JA |
129 | static inline void td_set_runstate(struct thread_data *td, int runstate) |
130 | { | |
ebac4655 JA |
131 | td->runstate = runstate; |
132 | } | |
133 | ||
53cdc686 JA |
134 | static struct fio_file *get_next_file(struct thread_data *td) |
135 | { | |
0ab8db89 | 136 | unsigned int old_next_file = td->next_file; |
b2a15192 JA |
137 | struct fio_file *f; |
138 | ||
139 | do { | |
140 | f = &td->files[td->next_file]; | |
141 | ||
142 | td->next_file++; | |
143 | if (td->next_file >= td->nr_files) | |
144 | td->next_file = 0; | |
145 | ||
146 | if (f->fd != -1) | |
147 | break; | |
53cdc686 | 148 | |
b2a15192 JA |
149 | f = NULL; |
150 | } while (td->next_file != old_next_file); | |
53cdc686 JA |
151 | |
152 | return f; | |
153 | } | |
154 | ||
906c8d75 JA |
155 | /* |
156 | * When job exits, we can cancel the in-flight IO if we are using async | |
157 | * io. Attempt to do so. | |
158 | */ | |
ebac4655 JA |
159 | static void cleanup_pending_aio(struct thread_data *td) |
160 | { | |
161 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0}; | |
162 | struct list_head *entry, *n; | |
163 | struct io_completion_data icd; | |
164 | struct io_u *io_u; | |
165 | int r; | |
166 | ||
167 | /* | |
168 | * get immediately available events, if any | |
169 | */ | |
45bee283 | 170 | r = td_io_getevents(td, 0, td->cur_depth, &ts); |
ebac4655 JA |
171 | if (r > 0) { |
172 | icd.nr = r; | |
173 | ios_completed(td, &icd); | |
174 | } | |
175 | ||
176 | /* | |
177 | * now cancel remaining active events | |
178 | */ | |
2866c82d | 179 | if (td->io_ops->cancel) { |
ebac4655 JA |
180 | list_for_each_safe(entry, n, &td->io_u_busylist) { |
181 | io_u = list_entry(entry, struct io_u, list); | |
182 | ||
2866c82d | 183 | r = td->io_ops->cancel(td, io_u); |
ebac4655 JA |
184 | if (!r) |
185 | put_io_u(td, io_u); | |
186 | } | |
187 | } | |
188 | ||
189 | if (td->cur_depth) { | |
45bee283 | 190 | r = td_io_getevents(td, td->cur_depth, td->cur_depth, NULL); |
ebac4655 JA |
191 | if (r > 0) { |
192 | icd.nr = r; | |
193 | ios_completed(td, &icd); | |
194 | } | |
195 | } | |
196 | } | |
197 | ||
906c8d75 JA |
198 | /* |
199 | * The main verify engine. Runs over the writes we previusly submitted, | |
200 | * reads the blocks back in, and checks the crc/md5 of the data. | |
201 | */ | |
a9619d44 | 202 | void do_verify(struct thread_data *td) |
ebac4655 JA |
203 | { |
204 | struct timeval t; | |
205 | struct io_u *io_u, *v_io_u = NULL; | |
206 | struct io_completion_data icd; | |
53cdc686 | 207 | struct fio_file *f; |
e5b401d4 JA |
208 | int ret, i; |
209 | ||
210 | /* | |
211 | * sync io first and invalidate cache, to make sure we really | |
212 | * read from disk. | |
213 | */ | |
214 | for_each_file(td, f, i) { | |
215 | td_io_sync(td, f); | |
216 | file_invalidate_cache(td, f); | |
217 | } | |
ebac4655 JA |
218 | |
219 | td_set_runstate(td, TD_VERIFYING); | |
220 | ||
221 | do { | |
222 | if (td->terminate) | |
223 | break; | |
224 | ||
225 | gettimeofday(&t, NULL); | |
226 | if (runtime_exceeded(td, &t)) | |
227 | break; | |
228 | ||
229 | io_u = __get_io_u(td); | |
230 | if (!io_u) | |
231 | break; | |
232 | ||
aea47d44 | 233 | if (get_next_verify(td, io_u)) { |
ebac4655 JA |
234 | put_io_u(td, io_u); |
235 | break; | |
236 | } | |
237 | ||
53cdc686 JA |
238 | f = get_next_file(td); |
239 | if (!f) | |
240 | break; | |
241 | ||
242 | io_u->file = f; | |
243 | ||
aea47d44 | 244 | if (td_io_prep(td, io_u)) { |
ebac4655 JA |
245 | put_io_u(td, io_u); |
246 | break; | |
247 | } | |
248 | ||
45bee283 | 249 | ret = td_io_queue(td, io_u); |
ebac4655 JA |
250 | if (ret) { |
251 | put_io_u(td, io_u); | |
252 | td_verror(td, ret); | |
253 | break; | |
254 | } | |
255 | ||
256 | /* | |
257 | * we have one pending to verify, do that while | |
258 | * we are doing io on the next one | |
259 | */ | |
260 | if (do_io_u_verify(td, &v_io_u)) | |
261 | break; | |
262 | ||
45bee283 | 263 | ret = td_io_getevents(td, 1, 1, NULL); |
ebac4655 JA |
264 | if (ret != 1) { |
265 | if (ret < 0) | |
266 | td_verror(td, ret); | |
267 | break; | |
268 | } | |
269 | ||
2866c82d | 270 | v_io_u = td->io_ops->event(td, 0); |
ebac4655 JA |
271 | icd.nr = 1; |
272 | icd.error = 0; | |
273 | io_completed(td, v_io_u, &icd); | |
274 | ||
275 | if (icd.error) { | |
276 | td_verror(td, icd.error); | |
277 | put_io_u(td, v_io_u); | |
278 | v_io_u = NULL; | |
279 | break; | |
280 | } | |
281 | ||
ebac4655 JA |
282 | /* |
283 | * if we can't submit more io, we need to verify now | |
284 | */ | |
285 | if (queue_full(td) && do_io_u_verify(td, &v_io_u)) | |
286 | break; | |
287 | ||
288 | } while (1); | |
289 | ||
290 | do_io_u_verify(td, &v_io_u); | |
291 | ||
292 | if (td->cur_depth) | |
293 | cleanup_pending_aio(td); | |
294 | ||
295 | td_set_runstate(td, TD_RUNNING); | |
296 | } | |
297 | ||
b990b5c0 JA |
298 | /* |
299 | * Not really an io thread, all it does is burn CPU cycles in the specified | |
300 | * manner. | |
301 | */ | |
302 | static void do_cpuio(struct thread_data *td) | |
303 | { | |
304 | struct timeval e; | |
305 | int split = 100 / td->cpuload; | |
306 | int i = 0; | |
307 | ||
308 | while (!td->terminate) { | |
309 | gettimeofday(&e, NULL); | |
310 | ||
311 | if (runtime_exceeded(td, &e)) | |
312 | break; | |
313 | ||
314 | if (!(i % split)) | |
315 | __usec_sleep(10000); | |
316 | else | |
317 | usec_sleep(td, 10000); | |
318 | ||
319 | i++; | |
320 | } | |
321 | } | |
322 | ||
32cd46a0 | 323 | /* |
906c8d75 | 324 | * Main IO worker function. It retrieves io_u's to process and queues |
32cd46a0 JA |
325 | * and reaps them, checking for rate and errors along the way. |
326 | */ | |
ebac4655 JA |
327 | static void do_io(struct thread_data *td) |
328 | { | |
329 | struct io_completion_data icd; | |
330 | struct timeval s, e; | |
331 | unsigned long usec; | |
53cdc686 | 332 | struct fio_file *f; |
84585003 | 333 | int i, ret = 0; |
ebac4655 | 334 | |
5853e5a8 JA |
335 | td_set_runstate(td, TD_RUNNING); |
336 | ||
ebac4655 JA |
337 | while (td->this_io_bytes[td->ddir] < td->io_size) { |
338 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0}; | |
339 | struct timespec *timeout; | |
84585003 | 340 | int min_evts = 0; |
ebac4655 JA |
341 | struct io_u *io_u; |
342 | ||
343 | if (td->terminate) | |
344 | break; | |
345 | ||
53cdc686 JA |
346 | f = get_next_file(td); |
347 | if (!f) | |
348 | break; | |
349 | ||
350 | io_u = get_io_u(td, f); | |
ebac4655 JA |
351 | if (!io_u) |
352 | break; | |
353 | ||
354 | memcpy(&s, &io_u->start_time, sizeof(s)); | |
355 | ||
45bee283 | 356 | ret = td_io_queue(td, io_u); |
ebac4655 JA |
357 | if (ret) { |
358 | put_io_u(td, io_u); | |
359 | td_verror(td, ret); | |
360 | break; | |
361 | } | |
362 | ||
363 | add_slat_sample(td, io_u->ddir, mtime_since(&io_u->start_time, &io_u->issue_time)); | |
364 | ||
365 | if (td->cur_depth < td->iodepth) { | |
366 | timeout = &ts; | |
367 | min_evts = 0; | |
368 | } else { | |
369 | timeout = NULL; | |
370 | min_evts = 1; | |
371 | } | |
372 | ||
45bee283 | 373 | ret = td_io_getevents(td, min_evts, td->cur_depth, timeout); |
ebac4655 | 374 | if (ret < 0) { |
84585003 | 375 | td_verror(td, -ret); |
ebac4655 JA |
376 | break; |
377 | } else if (!ret) | |
378 | continue; | |
379 | ||
380 | icd.nr = ret; | |
381 | ios_completed(td, &icd); | |
382 | if (icd.error) { | |
383 | td_verror(td, icd.error); | |
384 | break; | |
385 | } | |
386 | ||
387 | /* | |
388 | * the rate is batched for now, it should work for batches | |
389 | * of completions except the very first one which may look | |
390 | * a little bursty | |
391 | */ | |
392 | gettimeofday(&e, NULL); | |
393 | usec = utime_since(&s, &e); | |
394 | ||
395 | rate_throttle(td, usec, icd.bytes_done[td->ddir]); | |
396 | ||
397 | if (check_min_rate(td, &e)) { | |
2f9ade3c JA |
398 | if (rate_quit) |
399 | terminate_threads(td->groupid); | |
ebac4655 JA |
400 | td_verror(td, ENOMEM); |
401 | break; | |
402 | } | |
403 | ||
404 | if (runtime_exceeded(td, &e)) | |
405 | break; | |
406 | ||
407 | if (td->thinktime) | |
408 | usec_sleep(td, td->thinktime); | |
409 | ||
410 | if (should_fsync(td) && td->fsync_blocks && | |
411 | (td->io_blocks[DDIR_WRITE] % td->fsync_blocks) == 0) | |
53cdc686 | 412 | td_io_sync(td, f); |
ebac4655 JA |
413 | } |
414 | ||
84585003 JA |
415 | if (!ret) { |
416 | if (td->cur_depth) | |
417 | cleanup_pending_aio(td); | |
ebac4655 | 418 | |
84585003 JA |
419 | if (should_fsync(td) && td->end_fsync) { |
420 | td_set_runstate(td, TD_FSYNCING); | |
421 | for_each_file(td, f, i) | |
422 | td_io_sync(td, f); | |
423 | } | |
5853e5a8 | 424 | } |
ebac4655 JA |
425 | } |
426 | ||
ebac4655 JA |
427 | static void cleanup_io_u(struct thread_data *td) |
428 | { | |
429 | struct list_head *entry, *n; | |
430 | struct io_u *io_u; | |
431 | ||
432 | list_for_each_safe(entry, n, &td->io_u_freelist) { | |
433 | io_u = list_entry(entry, struct io_u, list); | |
434 | ||
435 | list_del(&io_u->list); | |
436 | free(io_u); | |
437 | } | |
438 | ||
2f9ade3c | 439 | free_io_mem(td); |
ebac4655 JA |
440 | } |
441 | ||
442 | static int init_io_u(struct thread_data *td) | |
443 | { | |
444 | struct io_u *io_u; | |
445 | int i, max_units; | |
446 | char *p; | |
447 | ||
2866c82d | 448 | if (td->io_ops->flags & FIO_CPUIO) |
b990b5c0 JA |
449 | return 0; |
450 | ||
2866c82d | 451 | if (td->io_ops->flags & FIO_SYNCIO) |
ebac4655 JA |
452 | max_units = 1; |
453 | else | |
454 | max_units = td->iodepth; | |
455 | ||
456 | td->orig_buffer_size = td->max_bs * max_units + MASK; | |
457 | ||
2f9ade3c JA |
458 | if (allocate_io_mem(td)) |
459 | return 1; | |
ebac4655 | 460 | |
ebac4655 JA |
461 | p = ALIGN(td->orig_buffer); |
462 | for (i = 0; i < max_units; i++) { | |
463 | io_u = malloc(sizeof(*io_u)); | |
464 | memset(io_u, 0, sizeof(*io_u)); | |
465 | INIT_LIST_HEAD(&io_u->list); | |
466 | ||
467 | io_u->buf = p + td->max_bs * i; | |
b1ff3403 | 468 | io_u->index = i; |
ebac4655 JA |
469 | list_add(&io_u->list, &td->io_u_freelist); |
470 | } | |
471 | ||
472 | return 0; | |
473 | } | |
474 | ||
da86774e JA |
475 | static int switch_ioscheduler(struct thread_data *td) |
476 | { | |
477 | char tmp[256], tmp2[128]; | |
478 | FILE *f; | |
479 | int ret; | |
480 | ||
481 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); | |
482 | ||
483 | f = fopen(tmp, "r+"); | |
484 | if (!f) { | |
485 | td_verror(td, errno); | |
486 | return 1; | |
487 | } | |
488 | ||
489 | /* | |
490 | * Set io scheduler. | |
491 | */ | |
492 | ret = fwrite(td->ioscheduler, strlen(td->ioscheduler), 1, f); | |
493 | if (ferror(f) || ret != 1) { | |
494 | td_verror(td, errno); | |
495 | fclose(f); | |
496 | return 1; | |
497 | } | |
498 | ||
499 | rewind(f); | |
500 | ||
501 | /* | |
502 | * Read back and check that the selected scheduler is now the default. | |
503 | */ | |
504 | ret = fread(tmp, 1, sizeof(tmp), f); | |
505 | if (ferror(f) || ret < 0) { | |
506 | td_verror(td, errno); | |
507 | fclose(f); | |
508 | return 1; | |
509 | } | |
510 | ||
511 | sprintf(tmp2, "[%s]", td->ioscheduler); | |
512 | if (!strstr(tmp, tmp2)) { | |
3b70d7e5 | 513 | log_err("fio: io scheduler %s not found\n", td->ioscheduler); |
da86774e JA |
514 | td_verror(td, EINVAL); |
515 | fclose(f); | |
516 | return 1; | |
517 | } | |
518 | ||
519 | fclose(f); | |
520 | return 0; | |
521 | } | |
522 | ||
ebac4655 JA |
523 | static void clear_io_state(struct thread_data *td) |
524 | { | |
53cdc686 JA |
525 | struct fio_file *f; |
526 | int i; | |
ebac4655 | 527 | |
ebac4655 JA |
528 | td->stat_io_bytes[0] = td->stat_io_bytes[1] = 0; |
529 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; | |
20dc95c4 | 530 | td->zone_bytes = 0; |
ebac4655 | 531 | |
53cdc686 JA |
532 | for_each_file(td, f, i) { |
533 | f->last_pos = 0; | |
534 | if (td->io_ops->flags & FIO_SYNCIO) | |
535 | lseek(f->fd, SEEK_SET, 0); | |
536 | ||
537 | if (f->file_map) | |
538 | memset(f->file_map, 0, f->num_maps * sizeof(long)); | |
539 | } | |
ebac4655 JA |
540 | } |
541 | ||
906c8d75 JA |
542 | /* |
543 | * Entry point for the thread based jobs. The process based jobs end up | |
544 | * here as well, after a little setup. | |
545 | */ | |
ebac4655 JA |
546 | static void *thread_main(void *data) |
547 | { | |
548 | struct thread_data *td = data; | |
ebac4655 JA |
549 | |
550 | if (!td->use_thread) | |
551 | setsid(); | |
552 | ||
553 | td->pid = getpid(); | |
554 | ||
aea47d44 JA |
555 | INIT_LIST_HEAD(&td->io_u_freelist); |
556 | INIT_LIST_HEAD(&td->io_u_busylist); | |
557 | INIT_LIST_HEAD(&td->io_hist_list); | |
558 | INIT_LIST_HEAD(&td->io_log_list); | |
559 | ||
ebac4655 JA |
560 | if (init_io_u(td)) |
561 | goto err; | |
562 | ||
563 | if (fio_setaffinity(td) == -1) { | |
564 | td_verror(td, errno); | |
565 | goto err; | |
566 | } | |
567 | ||
45bee283 | 568 | if (td_io_init(td)) |
ebac4655 JA |
569 | goto err; |
570 | ||
aea47d44 JA |
571 | if (init_iolog(td)) |
572 | goto err; | |
573 | ||
ebac4655 JA |
574 | if (td->ioprio) { |
575 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { | |
576 | td_verror(td, errno); | |
577 | goto err; | |
578 | } | |
579 | } | |
580 | ||
1056eaad | 581 | if (nice(td->nice) == -1) { |
b6f4d880 JA |
582 | td_verror(td, errno); |
583 | goto err; | |
584 | } | |
585 | ||
75154845 JA |
586 | if (init_random_state(td)) |
587 | goto err; | |
588 | ||
b4a6a59a JA |
589 | if (td->ioscheduler) { |
590 | int ret = switch_ioscheduler(td); | |
591 | ||
592 | free(td->ioscheduler); | |
eecf272f | 593 | free(td->sysfs_root); |
b4a6a59a JA |
594 | if (ret) |
595 | goto err; | |
596 | } | |
da86774e | 597 | |
75154845 | 598 | td_set_runstate(td, TD_INITIALIZED); |
bbfd6b00 JA |
599 | fio_sem_up(&startup_sem); |
600 | fio_sem_down(&td->mutex); | |
ebac4655 | 601 | |
53cdc686 | 602 | if (!td->create_serialize && setup_files(td)) |
ebac4655 JA |
603 | goto err; |
604 | ||
ebac4655 JA |
605 | gettimeofday(&td->epoch, NULL); |
606 | ||
b4a6a59a | 607 | if (td->exec_prerun) { |
4e0ba8af | 608 | system(td->exec_prerun); |
b4a6a59a JA |
609 | free(td->exec_prerun); |
610 | } | |
4e0ba8af | 611 | |
ebac4655 JA |
612 | while (td->loops--) { |
613 | getrusage(RUSAGE_SELF, &td->ru_start); | |
614 | gettimeofday(&td->start, NULL); | |
615 | memcpy(&td->stat_sample_time, &td->start, sizeof(td->start)); | |
616 | ||
617 | if (td->ratemin) | |
618 | memcpy(&td->lastrate, &td->stat_sample_time, sizeof(td->lastrate)); | |
619 | ||
620 | clear_io_state(td); | |
621 | prune_io_piece_log(td); | |
622 | ||
2866c82d | 623 | if (td->io_ops->flags & FIO_CPUIO) |
b990b5c0 JA |
624 | do_cpuio(td); |
625 | else | |
626 | do_io(td); | |
ebac4655 JA |
627 | |
628 | td->runtime[td->ddir] += mtime_since_now(&td->start); | |
aea47d44 | 629 | if (td_rw(td) && td->io_bytes[td->ddir ^ 1]) |
3d60d1ed JA |
630 | td->runtime[td->ddir ^ 1] = td->runtime[td->ddir]; |
631 | ||
ebac4655 JA |
632 | update_rusage_stat(td); |
633 | ||
634 | if (td->error || td->terminate) | |
635 | break; | |
636 | ||
637 | if (td->verify == VERIFY_NONE) | |
638 | continue; | |
639 | ||
640 | clear_io_state(td); | |
641 | gettimeofday(&td->start, NULL); | |
642 | ||
643 | do_verify(td); | |
644 | ||
645 | td->runtime[DDIR_READ] += mtime_since_now(&td->start); | |
646 | ||
647 | if (td->error || td->terminate) | |
648 | break; | |
649 | } | |
650 | ||
ebac4655 JA |
651 | if (td->bw_log) |
652 | finish_log(td, td->bw_log, "bw"); | |
653 | if (td->slat_log) | |
654 | finish_log(td, td->slat_log, "slat"); | |
655 | if (td->clat_log) | |
656 | finish_log(td, td->clat_log, "clat"); | |
843a7413 JA |
657 | if (td->write_iolog) |
658 | write_iolog_close(td); | |
b4a6a59a | 659 | if (td->exec_postrun) { |
4e0ba8af | 660 | system(td->exec_postrun); |
b4a6a59a JA |
661 | free(td->exec_postrun); |
662 | } | |
ebac4655 JA |
663 | |
664 | if (exitall_on_terminate) | |
665 | terminate_threads(td->groupid); | |
666 | ||
667 | err: | |
53cdc686 | 668 | close_files(td); |
2866c82d | 669 | close_ioengine(td); |
ebac4655 | 670 | cleanup_io_u(td); |
ebac4655 JA |
671 | td_set_runstate(td, TD_EXITED); |
672 | return NULL; | |
673 | ||
674 | } | |
675 | ||
906c8d75 JA |
676 | /* |
677 | * We cannot pass the td data into a forked process, so attach the td and | |
678 | * pass it to the thread worker. | |
679 | */ | |
ebac4655 JA |
680 | static void *fork_main(int shmid, int offset) |
681 | { | |
682 | struct thread_data *td; | |
683 | void *data; | |
684 | ||
685 | data = shmat(shmid, NULL, 0); | |
686 | if (data == (void *) -1) { | |
687 | perror("shmat"); | |
688 | return NULL; | |
689 | } | |
690 | ||
691 | td = data + offset * sizeof(struct thread_data); | |
692 | thread_main(td); | |
693 | shmdt(data); | |
694 | return NULL; | |
695 | } | |
696 | ||
906c8d75 JA |
697 | /* |
698 | * Run over the job map and reap the threads that have exited, if any. | |
699 | */ | |
ebac4655 JA |
700 | static void reap_threads(int *nr_running, int *t_rate, int *m_rate) |
701 | { | |
34572e28 | 702 | struct thread_data *td; |
b990b5c0 | 703 | int i, cputhreads; |
ebac4655 JA |
704 | |
705 | /* | |
706 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
707 | */ | |
34572e28 JA |
708 | cputhreads = 0; |
709 | for_each_td(td, i) { | |
84585003 JA |
710 | /* |
711 | * ->io_ops is NULL for a thread that has closed its | |
712 | * io engine | |
713 | */ | |
714 | if (td->io_ops && td->io_ops->flags & FIO_CPUIO) | |
b990b5c0 JA |
715 | cputhreads++; |
716 | ||
ebac4655 JA |
717 | if (td->runstate != TD_EXITED) |
718 | continue; | |
719 | ||
720 | td_set_runstate(td, TD_REAPED); | |
721 | ||
722 | if (td->use_thread) { | |
723 | long ret; | |
724 | ||
725 | if (pthread_join(td->thread, (void *) &ret)) | |
726 | perror("thread_join"); | |
727 | } else | |
728 | waitpid(td->pid, NULL, 0); | |
729 | ||
730 | (*nr_running)--; | |
731 | (*m_rate) -= td->ratemin; | |
732 | (*t_rate) -= td->rate; | |
733 | } | |
b990b5c0 JA |
734 | |
735 | if (*nr_running == cputhreads) | |
736 | terminate_threads(TERMINATE_ALL); | |
ebac4655 JA |
737 | } |
738 | ||
906c8d75 JA |
739 | /* |
740 | * Main function for kicking off and reaping jobs, as needed. | |
741 | */ | |
ebac4655 JA |
742 | static void run_threads(void) |
743 | { | |
ebac4655 JA |
744 | struct thread_data *td; |
745 | unsigned long spent; | |
746 | int i, todo, nr_running, m_rate, t_rate, nr_started; | |
fcb6ade2 | 747 | |
2f9ade3c JA |
748 | if (fio_pin_memory()) |
749 | return; | |
ebac4655 | 750 | |
c6ae0a5b JA |
751 | if (!terse_output) { |
752 | printf("Starting %d thread%s\n", thread_number, thread_number > 1 ? "s" : ""); | |
753 | fflush(stdout); | |
754 | } | |
c04f7ec3 | 755 | |
4efa970e JA |
756 | signal(SIGINT, sig_handler); |
757 | signal(SIGALRM, sig_handler); | |
758 | ||
ebac4655 JA |
759 | todo = thread_number; |
760 | nr_running = 0; | |
761 | nr_started = 0; | |
762 | m_rate = t_rate = 0; | |
763 | ||
34572e28 | 764 | for_each_td(td, i) { |
263e529f | 765 | print_status_init(td->thread_number - 1); |
ebac4655 JA |
766 | |
767 | init_disk_util(td); | |
768 | ||
769 | if (!td->create_serialize) | |
770 | continue; | |
771 | ||
772 | /* | |
773 | * do file setup here so it happens sequentially, | |
774 | * we don't want X number of threads getting their | |
775 | * client data interspersed on disk | |
776 | */ | |
53cdc686 | 777 | if (setup_files(td)) { |
ebac4655 JA |
778 | td_set_runstate(td, TD_REAPED); |
779 | todo--; | |
780 | } | |
781 | } | |
782 | ||
263e529f | 783 | time_init(); |
ebac4655 JA |
784 | |
785 | while (todo) { | |
75154845 JA |
786 | struct thread_data *map[MAX_JOBS]; |
787 | struct timeval this_start; | |
788 | int this_jobs = 0, left; | |
789 | ||
ebac4655 JA |
790 | /* |
791 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
792 | */ | |
34572e28 | 793 | for_each_td(td, i) { |
ebac4655 JA |
794 | if (td->runstate != TD_NOT_CREATED) |
795 | continue; | |
796 | ||
797 | /* | |
798 | * never got a chance to start, killed by other | |
799 | * thread for some reason | |
800 | */ | |
801 | if (td->terminate) { | |
802 | todo--; | |
803 | continue; | |
804 | } | |
805 | ||
806 | if (td->start_delay) { | |
263e529f | 807 | spent = mtime_since_genesis(); |
ebac4655 JA |
808 | |
809 | if (td->start_delay * 1000 > spent) | |
810 | continue; | |
811 | } | |
812 | ||
813 | if (td->stonewall && (nr_started || nr_running)) | |
814 | break; | |
815 | ||
75154845 JA |
816 | /* |
817 | * Set state to created. Thread will transition | |
818 | * to TD_INITIALIZED when it's done setting up. | |
819 | */ | |
ebac4655 | 820 | td_set_runstate(td, TD_CREATED); |
75154845 | 821 | map[this_jobs++] = td; |
bbfd6b00 | 822 | fio_sem_init(&startup_sem, 1); |
ebac4655 JA |
823 | nr_started++; |
824 | ||
825 | if (td->use_thread) { | |
826 | if (pthread_create(&td->thread, NULL, thread_main, td)) { | |
827 | perror("thread_create"); | |
828 | nr_started--; | |
829 | } | |
830 | } else { | |
831 | if (fork()) | |
bbfd6b00 | 832 | fio_sem_down(&startup_sem); |
ebac4655 JA |
833 | else { |
834 | fork_main(shm_id, i); | |
835 | exit(0); | |
836 | } | |
837 | } | |
838 | } | |
839 | ||
840 | /* | |
75154845 JA |
841 | * Wait for the started threads to transition to |
842 | * TD_INITIALIZED. | |
ebac4655 | 843 | */ |
75154845 JA |
844 | gettimeofday(&this_start, NULL); |
845 | left = this_jobs; | |
846 | while (left) { | |
847 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) | |
848 | break; | |
849 | ||
850 | usleep(100000); | |
851 | ||
852 | for (i = 0; i < this_jobs; i++) { | |
853 | td = map[i]; | |
854 | if (!td) | |
855 | continue; | |
b6f4d880 | 856 | if (td->runstate == TD_INITIALIZED) { |
75154845 JA |
857 | map[i] = NULL; |
858 | left--; | |
b6f4d880 JA |
859 | } else if (td->runstate >= TD_EXITED) { |
860 | map[i] = NULL; | |
861 | left--; | |
862 | todo--; | |
863 | nr_running++; /* work-around... */ | |
75154845 JA |
864 | } |
865 | } | |
866 | } | |
867 | ||
868 | if (left) { | |
3b70d7e5 | 869 | log_err("fio: %d jobs failed to start\n", left); |
75154845 JA |
870 | for (i = 0; i < this_jobs; i++) { |
871 | td = map[i]; | |
872 | if (!td) | |
873 | continue; | |
874 | kill(td->pid, SIGTERM); | |
875 | } | |
876 | break; | |
877 | } | |
878 | ||
879 | /* | |
b6f4d880 | 880 | * start created threads (TD_INITIALIZED -> TD_RUNNING). |
75154845 | 881 | */ |
34572e28 | 882 | for_each_td(td, i) { |
75154845 | 883 | if (td->runstate != TD_INITIALIZED) |
ebac4655 JA |
884 | continue; |
885 | ||
886 | td_set_runstate(td, TD_RUNNING); | |
887 | nr_running++; | |
888 | nr_started--; | |
889 | m_rate += td->ratemin; | |
890 | t_rate += td->rate; | |
75154845 | 891 | todo--; |
bbfd6b00 | 892 | fio_sem_up(&td->mutex); |
ebac4655 JA |
893 | } |
894 | ||
895 | reap_threads(&nr_running, &t_rate, &m_rate); | |
896 | ||
897 | if (todo) | |
898 | usleep(100000); | |
899 | } | |
900 | ||
901 | while (nr_running) { | |
902 | reap_threads(&nr_running, &t_rate, &m_rate); | |
903 | usleep(10000); | |
904 | } | |
905 | ||
906 | update_io_ticks(); | |
2f9ade3c | 907 | fio_unpin_memory(); |
ebac4655 JA |
908 | } |
909 | ||
ebac4655 JA |
910 | int main(int argc, char *argv[]) |
911 | { | |
912 | if (parse_options(argc, argv)) | |
913 | return 1; | |
914 | ||
915 | if (!thread_number) { | |
3b70d7e5 | 916 | log_err("Nothing to do\n"); |
ebac4655 JA |
917 | return 1; |
918 | } | |
919 | ||
920 | disk_util_timer_arm(); | |
921 | ||
922 | run_threads(); | |
923 | show_run_stats(); | |
924 | ||
925 | return 0; | |
926 | } |