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 | ||
858a3d47 JA |
196 | /* |
197 | * Helper to handle the final sync of a file. Works just like the normal | |
198 | * io path, just does everything sync. | |
199 | */ | |
200 | static int fio_io_sync(struct thread_data *td, struct fio_file *f) | |
201 | { | |
202 | struct io_u *io_u = __get_io_u(td); | |
203 | struct io_completion_data icd; | |
204 | int ret; | |
205 | ||
206 | if (!io_u) | |
207 | return 1; | |
208 | ||
209 | io_u->ddir = DDIR_SYNC; | |
210 | io_u->file = f; | |
211 | ||
212 | if (td_io_prep(td, io_u)) { | |
213 | put_io_u(td, io_u); | |
214 | return 1; | |
215 | } | |
216 | ||
217 | ret = td_io_queue(td, io_u); | |
218 | if (ret) { | |
219 | put_io_u(td, io_u); | |
220 | td_verror(td, ret); | |
221 | return 1; | |
222 | } | |
223 | ||
224 | ret = td_io_getevents(td, 1, td->cur_depth, NULL); | |
225 | if (ret < 0) { | |
226 | td_verror(td, -ret); | |
227 | return 1; | |
228 | } | |
229 | ||
230 | icd.nr = ret; | |
231 | ios_completed(td, &icd); | |
232 | if (icd.error) { | |
233 | td_verror(td, icd.error); | |
234 | return 1; | |
235 | } | |
236 | ||
237 | return 0; | |
238 | } | |
239 | ||
906c8d75 JA |
240 | /* |
241 | * The main verify engine. Runs over the writes we previusly submitted, | |
242 | * reads the blocks back in, and checks the crc/md5 of the data. | |
243 | */ | |
a9619d44 | 244 | void do_verify(struct thread_data *td) |
ebac4655 JA |
245 | { |
246 | struct timeval t; | |
247 | struct io_u *io_u, *v_io_u = NULL; | |
248 | struct io_completion_data icd; | |
53cdc686 | 249 | struct fio_file *f; |
e5b401d4 JA |
250 | int ret, i; |
251 | ||
252 | /* | |
253 | * sync io first and invalidate cache, to make sure we really | |
254 | * read from disk. | |
255 | */ | |
256 | for_each_file(td, f, i) { | |
858a3d47 | 257 | fio_io_sync(td, f); |
e5b401d4 JA |
258 | file_invalidate_cache(td, f); |
259 | } | |
ebac4655 JA |
260 | |
261 | td_set_runstate(td, TD_VERIFYING); | |
262 | ||
263 | do { | |
264 | if (td->terminate) | |
265 | break; | |
266 | ||
267 | gettimeofday(&t, NULL); | |
268 | if (runtime_exceeded(td, &t)) | |
269 | break; | |
270 | ||
271 | io_u = __get_io_u(td); | |
272 | if (!io_u) | |
273 | break; | |
274 | ||
aea47d44 | 275 | if (get_next_verify(td, io_u)) { |
ebac4655 JA |
276 | put_io_u(td, io_u); |
277 | break; | |
278 | } | |
279 | ||
53cdc686 JA |
280 | f = get_next_file(td); |
281 | if (!f) | |
282 | break; | |
283 | ||
284 | io_u->file = f; | |
285 | ||
aea47d44 | 286 | if (td_io_prep(td, io_u)) { |
ebac4655 JA |
287 | put_io_u(td, io_u); |
288 | break; | |
289 | } | |
290 | ||
45bee283 | 291 | ret = td_io_queue(td, io_u); |
ebac4655 JA |
292 | if (ret) { |
293 | put_io_u(td, io_u); | |
294 | td_verror(td, ret); | |
295 | break; | |
296 | } | |
297 | ||
298 | /* | |
299 | * we have one pending to verify, do that while | |
300 | * we are doing io on the next one | |
301 | */ | |
302 | if (do_io_u_verify(td, &v_io_u)) | |
303 | break; | |
304 | ||
45bee283 | 305 | ret = td_io_getevents(td, 1, 1, NULL); |
ebac4655 JA |
306 | if (ret != 1) { |
307 | if (ret < 0) | |
308 | td_verror(td, ret); | |
309 | break; | |
310 | } | |
311 | ||
2866c82d | 312 | v_io_u = td->io_ops->event(td, 0); |
ebac4655 JA |
313 | icd.nr = 1; |
314 | icd.error = 0; | |
315 | io_completed(td, v_io_u, &icd); | |
316 | ||
317 | if (icd.error) { | |
318 | td_verror(td, icd.error); | |
319 | put_io_u(td, v_io_u); | |
320 | v_io_u = NULL; | |
321 | break; | |
322 | } | |
323 | ||
ebac4655 JA |
324 | /* |
325 | * if we can't submit more io, we need to verify now | |
326 | */ | |
327 | if (queue_full(td) && do_io_u_verify(td, &v_io_u)) | |
328 | break; | |
329 | ||
330 | } while (1); | |
331 | ||
332 | do_io_u_verify(td, &v_io_u); | |
333 | ||
334 | if (td->cur_depth) | |
335 | cleanup_pending_aio(td); | |
336 | ||
337 | td_set_runstate(td, TD_RUNNING); | |
338 | } | |
339 | ||
b990b5c0 JA |
340 | /* |
341 | * Not really an io thread, all it does is burn CPU cycles in the specified | |
342 | * manner. | |
343 | */ | |
344 | static void do_cpuio(struct thread_data *td) | |
345 | { | |
346 | struct timeval e; | |
347 | int split = 100 / td->cpuload; | |
348 | int i = 0; | |
349 | ||
350 | while (!td->terminate) { | |
351 | gettimeofday(&e, NULL); | |
352 | ||
353 | if (runtime_exceeded(td, &e)) | |
354 | break; | |
355 | ||
356 | if (!(i % split)) | |
357 | __usec_sleep(10000); | |
358 | else | |
359 | usec_sleep(td, 10000); | |
360 | ||
361 | i++; | |
362 | } | |
363 | } | |
364 | ||
32cd46a0 | 365 | /* |
906c8d75 | 366 | * Main IO worker function. It retrieves io_u's to process and queues |
32cd46a0 JA |
367 | * and reaps them, checking for rate and errors along the way. |
368 | */ | |
ebac4655 JA |
369 | static void do_io(struct thread_data *td) |
370 | { | |
371 | struct io_completion_data icd; | |
372 | struct timeval s, e; | |
373 | unsigned long usec; | |
53cdc686 | 374 | struct fio_file *f; |
84585003 | 375 | int i, ret = 0; |
ebac4655 | 376 | |
5853e5a8 JA |
377 | td_set_runstate(td, TD_RUNNING); |
378 | ||
ebac4655 JA |
379 | while (td->this_io_bytes[td->ddir] < td->io_size) { |
380 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0}; | |
381 | struct timespec *timeout; | |
84585003 | 382 | int min_evts = 0; |
ebac4655 JA |
383 | struct io_u *io_u; |
384 | ||
385 | if (td->terminate) | |
386 | break; | |
387 | ||
53cdc686 JA |
388 | f = get_next_file(td); |
389 | if (!f) | |
390 | break; | |
391 | ||
392 | io_u = get_io_u(td, f); | |
ebac4655 JA |
393 | if (!io_u) |
394 | break; | |
395 | ||
396 | memcpy(&s, &io_u->start_time, sizeof(s)); | |
397 | ||
45bee283 | 398 | ret = td_io_queue(td, io_u); |
ebac4655 JA |
399 | if (ret) { |
400 | put_io_u(td, io_u); | |
401 | td_verror(td, ret); | |
402 | break; | |
403 | } | |
404 | ||
405 | add_slat_sample(td, io_u->ddir, mtime_since(&io_u->start_time, &io_u->issue_time)); | |
406 | ||
407 | if (td->cur_depth < td->iodepth) { | |
408 | timeout = &ts; | |
409 | min_evts = 0; | |
410 | } else { | |
411 | timeout = NULL; | |
412 | min_evts = 1; | |
413 | } | |
414 | ||
45bee283 | 415 | ret = td_io_getevents(td, min_evts, td->cur_depth, timeout); |
ebac4655 | 416 | if (ret < 0) { |
87dc1ab1 | 417 | td_verror(td, ret); |
ebac4655 JA |
418 | break; |
419 | } else if (!ret) | |
420 | continue; | |
421 | ||
422 | icd.nr = ret; | |
423 | ios_completed(td, &icd); | |
424 | if (icd.error) { | |
425 | td_verror(td, icd.error); | |
426 | break; | |
427 | } | |
428 | ||
429 | /* | |
430 | * the rate is batched for now, it should work for batches | |
431 | * of completions except the very first one which may look | |
432 | * a little bursty | |
433 | */ | |
434 | gettimeofday(&e, NULL); | |
435 | usec = utime_since(&s, &e); | |
436 | ||
437 | rate_throttle(td, usec, icd.bytes_done[td->ddir]); | |
438 | ||
439 | if (check_min_rate(td, &e)) { | |
2f9ade3c JA |
440 | if (rate_quit) |
441 | terminate_threads(td->groupid); | |
ebac4655 JA |
442 | td_verror(td, ENOMEM); |
443 | break; | |
444 | } | |
445 | ||
446 | if (runtime_exceeded(td, &e)) | |
447 | break; | |
448 | ||
449 | if (td->thinktime) | |
450 | usec_sleep(td, td->thinktime); | |
ebac4655 JA |
451 | } |
452 | ||
84585003 JA |
453 | if (!ret) { |
454 | if (td->cur_depth) | |
455 | cleanup_pending_aio(td); | |
ebac4655 | 456 | |
84585003 JA |
457 | if (should_fsync(td) && td->end_fsync) { |
458 | td_set_runstate(td, TD_FSYNCING); | |
459 | for_each_file(td, f, i) | |
858a3d47 | 460 | fio_io_sync(td, f); |
84585003 | 461 | } |
5853e5a8 | 462 | } |
ebac4655 JA |
463 | } |
464 | ||
ebac4655 JA |
465 | static void cleanup_io_u(struct thread_data *td) |
466 | { | |
467 | struct list_head *entry, *n; | |
468 | struct io_u *io_u; | |
469 | ||
470 | list_for_each_safe(entry, n, &td->io_u_freelist) { | |
471 | io_u = list_entry(entry, struct io_u, list); | |
472 | ||
473 | list_del(&io_u->list); | |
474 | free(io_u); | |
475 | } | |
476 | ||
2f9ade3c | 477 | free_io_mem(td); |
ebac4655 JA |
478 | } |
479 | ||
480 | static int init_io_u(struct thread_data *td) | |
481 | { | |
482 | struct io_u *io_u; | |
483 | int i, max_units; | |
484 | char *p; | |
485 | ||
2866c82d | 486 | if (td->io_ops->flags & FIO_CPUIO) |
b990b5c0 JA |
487 | return 0; |
488 | ||
2866c82d | 489 | if (td->io_ops->flags & FIO_SYNCIO) |
ebac4655 JA |
490 | max_units = 1; |
491 | else | |
492 | max_units = td->iodepth; | |
493 | ||
494 | td->orig_buffer_size = td->max_bs * max_units + MASK; | |
495 | ||
2f9ade3c JA |
496 | if (allocate_io_mem(td)) |
497 | return 1; | |
ebac4655 | 498 | |
ebac4655 JA |
499 | p = ALIGN(td->orig_buffer); |
500 | for (i = 0; i < max_units; i++) { | |
501 | io_u = malloc(sizeof(*io_u)); | |
502 | memset(io_u, 0, sizeof(*io_u)); | |
503 | INIT_LIST_HEAD(&io_u->list); | |
504 | ||
505 | io_u->buf = p + td->max_bs * i; | |
b1ff3403 | 506 | io_u->index = i; |
ebac4655 JA |
507 | list_add(&io_u->list, &td->io_u_freelist); |
508 | } | |
509 | ||
510 | return 0; | |
511 | } | |
512 | ||
da86774e JA |
513 | static int switch_ioscheduler(struct thread_data *td) |
514 | { | |
515 | char tmp[256], tmp2[128]; | |
516 | FILE *f; | |
517 | int ret; | |
518 | ||
519 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); | |
520 | ||
521 | f = fopen(tmp, "r+"); | |
522 | if (!f) { | |
523 | td_verror(td, errno); | |
524 | return 1; | |
525 | } | |
526 | ||
527 | /* | |
528 | * Set io scheduler. | |
529 | */ | |
530 | ret = fwrite(td->ioscheduler, strlen(td->ioscheduler), 1, f); | |
531 | if (ferror(f) || ret != 1) { | |
532 | td_verror(td, errno); | |
533 | fclose(f); | |
534 | return 1; | |
535 | } | |
536 | ||
537 | rewind(f); | |
538 | ||
539 | /* | |
540 | * Read back and check that the selected scheduler is now the default. | |
541 | */ | |
542 | ret = fread(tmp, 1, sizeof(tmp), f); | |
543 | if (ferror(f) || ret < 0) { | |
544 | td_verror(td, errno); | |
545 | fclose(f); | |
546 | return 1; | |
547 | } | |
548 | ||
549 | sprintf(tmp2, "[%s]", td->ioscheduler); | |
550 | if (!strstr(tmp, tmp2)) { | |
3b70d7e5 | 551 | log_err("fio: io scheduler %s not found\n", td->ioscheduler); |
da86774e JA |
552 | td_verror(td, EINVAL); |
553 | fclose(f); | |
554 | return 1; | |
555 | } | |
556 | ||
557 | fclose(f); | |
558 | return 0; | |
559 | } | |
560 | ||
ebac4655 JA |
561 | static void clear_io_state(struct thread_data *td) |
562 | { | |
53cdc686 JA |
563 | struct fio_file *f; |
564 | int i; | |
ebac4655 | 565 | |
ebac4655 JA |
566 | td->stat_io_bytes[0] = td->stat_io_bytes[1] = 0; |
567 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; | |
20dc95c4 | 568 | td->zone_bytes = 0; |
ebac4655 | 569 | |
53cdc686 JA |
570 | for_each_file(td, f, i) { |
571 | f->last_pos = 0; | |
572 | if (td->io_ops->flags & FIO_SYNCIO) | |
573 | lseek(f->fd, SEEK_SET, 0); | |
574 | ||
575 | if (f->file_map) | |
576 | memset(f->file_map, 0, f->num_maps * sizeof(long)); | |
577 | } | |
ebac4655 JA |
578 | } |
579 | ||
906c8d75 JA |
580 | /* |
581 | * Entry point for the thread based jobs. The process based jobs end up | |
582 | * here as well, after a little setup. | |
583 | */ | |
ebac4655 JA |
584 | static void *thread_main(void *data) |
585 | { | |
586 | struct thread_data *td = data; | |
ebac4655 JA |
587 | |
588 | if (!td->use_thread) | |
589 | setsid(); | |
590 | ||
591 | td->pid = getpid(); | |
592 | ||
aea47d44 JA |
593 | INIT_LIST_HEAD(&td->io_u_freelist); |
594 | INIT_LIST_HEAD(&td->io_u_busylist); | |
595 | INIT_LIST_HEAD(&td->io_hist_list); | |
596 | INIT_LIST_HEAD(&td->io_log_list); | |
597 | ||
ebac4655 JA |
598 | if (init_io_u(td)) |
599 | goto err; | |
600 | ||
601 | if (fio_setaffinity(td) == -1) { | |
602 | td_verror(td, errno); | |
603 | goto err; | |
604 | } | |
605 | ||
45bee283 | 606 | if (td_io_init(td)) |
ebac4655 JA |
607 | goto err; |
608 | ||
aea47d44 JA |
609 | if (init_iolog(td)) |
610 | goto err; | |
611 | ||
ebac4655 JA |
612 | if (td->ioprio) { |
613 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { | |
614 | td_verror(td, errno); | |
615 | goto err; | |
616 | } | |
617 | } | |
618 | ||
1056eaad | 619 | if (nice(td->nice) == -1) { |
b6f4d880 JA |
620 | td_verror(td, errno); |
621 | goto err; | |
622 | } | |
623 | ||
75154845 JA |
624 | if (init_random_state(td)) |
625 | goto err; | |
626 | ||
56f9498d JA |
627 | if (td->ioscheduler && switch_ioscheduler(td)) |
628 | goto err; | |
da86774e | 629 | |
75154845 | 630 | td_set_runstate(td, TD_INITIALIZED); |
bbfd6b00 JA |
631 | fio_sem_up(&startup_sem); |
632 | fio_sem_down(&td->mutex); | |
ebac4655 | 633 | |
53cdc686 | 634 | if (!td->create_serialize && setup_files(td)) |
ebac4655 JA |
635 | goto err; |
636 | ||
ebac4655 JA |
637 | gettimeofday(&td->epoch, NULL); |
638 | ||
56f9498d | 639 | if (td->exec_prerun) |
4e0ba8af JA |
640 | system(td->exec_prerun); |
641 | ||
ebac4655 JA |
642 | while (td->loops--) { |
643 | getrusage(RUSAGE_SELF, &td->ru_start); | |
644 | gettimeofday(&td->start, NULL); | |
645 | memcpy(&td->stat_sample_time, &td->start, sizeof(td->start)); | |
646 | ||
647 | if (td->ratemin) | |
648 | memcpy(&td->lastrate, &td->stat_sample_time, sizeof(td->lastrate)); | |
649 | ||
650 | clear_io_state(td); | |
651 | prune_io_piece_log(td); | |
652 | ||
2866c82d | 653 | if (td->io_ops->flags & FIO_CPUIO) |
b990b5c0 JA |
654 | do_cpuio(td); |
655 | else | |
656 | do_io(td); | |
ebac4655 JA |
657 | |
658 | td->runtime[td->ddir] += mtime_since_now(&td->start); | |
aea47d44 | 659 | if (td_rw(td) && td->io_bytes[td->ddir ^ 1]) |
3d60d1ed JA |
660 | td->runtime[td->ddir ^ 1] = td->runtime[td->ddir]; |
661 | ||
ebac4655 JA |
662 | update_rusage_stat(td); |
663 | ||
664 | if (td->error || td->terminate) | |
665 | break; | |
666 | ||
667 | if (td->verify == VERIFY_NONE) | |
668 | continue; | |
669 | ||
670 | clear_io_state(td); | |
671 | gettimeofday(&td->start, NULL); | |
672 | ||
673 | do_verify(td); | |
674 | ||
675 | td->runtime[DDIR_READ] += mtime_since_now(&td->start); | |
676 | ||
677 | if (td->error || td->terminate) | |
678 | break; | |
679 | } | |
680 | ||
ebac4655 JA |
681 | if (td->bw_log) |
682 | finish_log(td, td->bw_log, "bw"); | |
683 | if (td->slat_log) | |
684 | finish_log(td, td->slat_log, "slat"); | |
685 | if (td->clat_log) | |
686 | finish_log(td, td->clat_log, "clat"); | |
843a7413 JA |
687 | if (td->write_iolog) |
688 | write_iolog_close(td); | |
56f9498d | 689 | if (td->exec_postrun) |
4e0ba8af | 690 | system(td->exec_postrun); |
ebac4655 JA |
691 | |
692 | if (exitall_on_terminate) | |
693 | terminate_threads(td->groupid); | |
694 | ||
695 | err: | |
53cdc686 | 696 | close_files(td); |
2866c82d | 697 | close_ioengine(td); |
ebac4655 | 698 | cleanup_io_u(td); |
ebac4655 JA |
699 | td_set_runstate(td, TD_EXITED); |
700 | return NULL; | |
701 | ||
702 | } | |
703 | ||
906c8d75 JA |
704 | /* |
705 | * We cannot pass the td data into a forked process, so attach the td and | |
706 | * pass it to the thread worker. | |
707 | */ | |
ebac4655 JA |
708 | static void *fork_main(int shmid, int offset) |
709 | { | |
710 | struct thread_data *td; | |
711 | void *data; | |
712 | ||
713 | data = shmat(shmid, NULL, 0); | |
714 | if (data == (void *) -1) { | |
715 | perror("shmat"); | |
716 | return NULL; | |
717 | } | |
718 | ||
719 | td = data + offset * sizeof(struct thread_data); | |
720 | thread_main(td); | |
721 | shmdt(data); | |
722 | return NULL; | |
723 | } | |
724 | ||
906c8d75 JA |
725 | /* |
726 | * Run over the job map and reap the threads that have exited, if any. | |
727 | */ | |
ebac4655 JA |
728 | static void reap_threads(int *nr_running, int *t_rate, int *m_rate) |
729 | { | |
34572e28 | 730 | struct thread_data *td; |
b990b5c0 | 731 | int i, cputhreads; |
ebac4655 JA |
732 | |
733 | /* | |
734 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
735 | */ | |
34572e28 JA |
736 | cputhreads = 0; |
737 | for_each_td(td, i) { | |
84585003 JA |
738 | /* |
739 | * ->io_ops is NULL for a thread that has closed its | |
740 | * io engine | |
741 | */ | |
742 | if (td->io_ops && td->io_ops->flags & FIO_CPUIO) | |
b990b5c0 JA |
743 | cputhreads++; |
744 | ||
ebac4655 JA |
745 | if (td->runstate != TD_EXITED) |
746 | continue; | |
747 | ||
748 | td_set_runstate(td, TD_REAPED); | |
749 | ||
750 | if (td->use_thread) { | |
751 | long ret; | |
752 | ||
753 | if (pthread_join(td->thread, (void *) &ret)) | |
754 | perror("thread_join"); | |
755 | } else | |
756 | waitpid(td->pid, NULL, 0); | |
757 | ||
758 | (*nr_running)--; | |
759 | (*m_rate) -= td->ratemin; | |
760 | (*t_rate) -= td->rate; | |
761 | } | |
b990b5c0 JA |
762 | |
763 | if (*nr_running == cputhreads) | |
764 | terminate_threads(TERMINATE_ALL); | |
ebac4655 JA |
765 | } |
766 | ||
906c8d75 JA |
767 | /* |
768 | * Main function for kicking off and reaping jobs, as needed. | |
769 | */ | |
ebac4655 JA |
770 | static void run_threads(void) |
771 | { | |
ebac4655 JA |
772 | struct thread_data *td; |
773 | unsigned long spent; | |
774 | int i, todo, nr_running, m_rate, t_rate, nr_started; | |
fcb6ade2 | 775 | |
2f9ade3c JA |
776 | if (fio_pin_memory()) |
777 | return; | |
ebac4655 | 778 | |
c6ae0a5b JA |
779 | if (!terse_output) { |
780 | printf("Starting %d thread%s\n", thread_number, thread_number > 1 ? "s" : ""); | |
781 | fflush(stdout); | |
782 | } | |
c04f7ec3 | 783 | |
4efa970e JA |
784 | signal(SIGINT, sig_handler); |
785 | signal(SIGALRM, sig_handler); | |
786 | ||
ebac4655 JA |
787 | todo = thread_number; |
788 | nr_running = 0; | |
789 | nr_started = 0; | |
790 | m_rate = t_rate = 0; | |
791 | ||
34572e28 | 792 | for_each_td(td, i) { |
263e529f | 793 | print_status_init(td->thread_number - 1); |
ebac4655 JA |
794 | |
795 | init_disk_util(td); | |
796 | ||
797 | if (!td->create_serialize) | |
798 | continue; | |
799 | ||
800 | /* | |
801 | * do file setup here so it happens sequentially, | |
802 | * we don't want X number of threads getting their | |
803 | * client data interspersed on disk | |
804 | */ | |
53cdc686 | 805 | if (setup_files(td)) { |
ebac4655 JA |
806 | td_set_runstate(td, TD_REAPED); |
807 | todo--; | |
808 | } | |
809 | } | |
810 | ||
263e529f | 811 | time_init(); |
ebac4655 JA |
812 | |
813 | while (todo) { | |
75154845 JA |
814 | struct thread_data *map[MAX_JOBS]; |
815 | struct timeval this_start; | |
816 | int this_jobs = 0, left; | |
817 | ||
ebac4655 JA |
818 | /* |
819 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
820 | */ | |
34572e28 | 821 | for_each_td(td, i) { |
ebac4655 JA |
822 | if (td->runstate != TD_NOT_CREATED) |
823 | continue; | |
824 | ||
825 | /* | |
826 | * never got a chance to start, killed by other | |
827 | * thread for some reason | |
828 | */ | |
829 | if (td->terminate) { | |
830 | todo--; | |
831 | continue; | |
832 | } | |
833 | ||
834 | if (td->start_delay) { | |
263e529f | 835 | spent = mtime_since_genesis(); |
ebac4655 JA |
836 | |
837 | if (td->start_delay * 1000 > spent) | |
838 | continue; | |
839 | } | |
840 | ||
841 | if (td->stonewall && (nr_started || nr_running)) | |
842 | break; | |
843 | ||
75154845 JA |
844 | /* |
845 | * Set state to created. Thread will transition | |
846 | * to TD_INITIALIZED when it's done setting up. | |
847 | */ | |
ebac4655 | 848 | td_set_runstate(td, TD_CREATED); |
75154845 | 849 | map[this_jobs++] = td; |
bbfd6b00 | 850 | fio_sem_init(&startup_sem, 1); |
ebac4655 JA |
851 | nr_started++; |
852 | ||
853 | if (td->use_thread) { | |
854 | if (pthread_create(&td->thread, NULL, thread_main, td)) { | |
855 | perror("thread_create"); | |
856 | nr_started--; | |
857 | } | |
858 | } else { | |
859 | if (fork()) | |
bbfd6b00 | 860 | fio_sem_down(&startup_sem); |
ebac4655 JA |
861 | else { |
862 | fork_main(shm_id, i); | |
863 | exit(0); | |
864 | } | |
865 | } | |
866 | } | |
867 | ||
868 | /* | |
75154845 JA |
869 | * Wait for the started threads to transition to |
870 | * TD_INITIALIZED. | |
ebac4655 | 871 | */ |
75154845 JA |
872 | gettimeofday(&this_start, NULL); |
873 | left = this_jobs; | |
874 | while (left) { | |
875 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) | |
876 | break; | |
877 | ||
878 | usleep(100000); | |
879 | ||
880 | for (i = 0; i < this_jobs; i++) { | |
881 | td = map[i]; | |
882 | if (!td) | |
883 | continue; | |
b6f4d880 | 884 | if (td->runstate == TD_INITIALIZED) { |
75154845 JA |
885 | map[i] = NULL; |
886 | left--; | |
b6f4d880 JA |
887 | } else if (td->runstate >= TD_EXITED) { |
888 | map[i] = NULL; | |
889 | left--; | |
890 | todo--; | |
891 | nr_running++; /* work-around... */ | |
75154845 JA |
892 | } |
893 | } | |
894 | } | |
895 | ||
896 | if (left) { | |
3b70d7e5 | 897 | log_err("fio: %d jobs failed to start\n", left); |
75154845 JA |
898 | for (i = 0; i < this_jobs; i++) { |
899 | td = map[i]; | |
900 | if (!td) | |
901 | continue; | |
902 | kill(td->pid, SIGTERM); | |
903 | } | |
904 | break; | |
905 | } | |
906 | ||
907 | /* | |
b6f4d880 | 908 | * start created threads (TD_INITIALIZED -> TD_RUNNING). |
75154845 | 909 | */ |
34572e28 | 910 | for_each_td(td, i) { |
75154845 | 911 | if (td->runstate != TD_INITIALIZED) |
ebac4655 JA |
912 | continue; |
913 | ||
914 | td_set_runstate(td, TD_RUNNING); | |
915 | nr_running++; | |
916 | nr_started--; | |
917 | m_rate += td->ratemin; | |
918 | t_rate += td->rate; | |
75154845 | 919 | todo--; |
bbfd6b00 | 920 | fio_sem_up(&td->mutex); |
ebac4655 JA |
921 | } |
922 | ||
923 | reap_threads(&nr_running, &t_rate, &m_rate); | |
924 | ||
925 | if (todo) | |
926 | usleep(100000); | |
927 | } | |
928 | ||
929 | while (nr_running) { | |
930 | reap_threads(&nr_running, &t_rate, &m_rate); | |
931 | usleep(10000); | |
932 | } | |
933 | ||
934 | update_io_ticks(); | |
2f9ade3c | 935 | fio_unpin_memory(); |
ebac4655 JA |
936 | } |
937 | ||
ebac4655 JA |
938 | int main(int argc, char *argv[]) |
939 | { | |
940 | if (parse_options(argc, argv)) | |
941 | return 1; | |
942 | ||
943 | if (!thread_number) { | |
3b70d7e5 | 944 | log_err("Nothing to do\n"); |
ebac4655 JA |
945 | return 1; |
946 | } | |
947 | ||
948 | disk_util_timer_arm(); | |
949 | ||
950 | run_threads(); | |
951 | show_run_stats(); | |
952 | ||
953 | return 0; | |
954 | } |