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 | 6 | * |
8e9fe637 JA |
7 | * The license below covers all files distributed with fio unless otherwise |
8 | * noted in the file itself. | |
9 | * | |
ebac4655 | 10 | * This program is free software; you can redistribute it and/or modify |
8e9fe637 JA |
11 | * it under the terms of the GNU General Public License version 2 as |
12 | * published by the Free Software Foundation. | |
ebac4655 JA |
13 | * |
14 | * This program is distributed in the hope that it will be useful, | |
15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
17 | * GNU General Public License for more details. | |
18 | * | |
19 | * You should have received a copy of the GNU General Public License | |
20 | * along with this program; if not, write to the Free Software | |
21 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
22 | * | |
23 | */ | |
ebac4655 JA |
24 | #include <unistd.h> |
25 | #include <fcntl.h> | |
26 | #include <string.h> | |
ebac4655 JA |
27 | #include <signal.h> |
28 | #include <time.h> | |
dbe1125e | 29 | #include <locale.h> |
36167d82 | 30 | #include <assert.h> |
ebac4655 JA |
31 | #include <sys/stat.h> |
32 | #include <sys/wait.h> | |
33 | #include <sys/ipc.h> | |
34 | #include <sys/shm.h> | |
ebac4655 JA |
35 | #include <sys/mman.h> |
36 | ||
37 | #include "fio.h" | |
ebac4655 | 38 | |
cfc99db7 JA |
39 | unsigned long page_mask; |
40 | unsigned long page_size; | |
29d610e1 JA |
41 | #define ALIGN(buf) \ |
42 | (char *) (((unsigned long) (buf) + page_mask) & ~page_mask) | |
ebac4655 JA |
43 | |
44 | int groupid = 0; | |
45 | int thread_number = 0; | |
9cedf167 JA |
46 | int nr_process = 0; |
47 | int nr_thread = 0; | |
ebac4655 | 48 | int shm_id = 0; |
53cdc686 | 49 | int temp_stall_ts; |
ebac4655 | 50 | |
07739b57 | 51 | static struct fio_sem *startup_sem; |
6ce15a32 | 52 | static volatile int fio_abort; |
437c9b71 | 53 | static int exit_value; |
ebac4655 | 54 | |
bb3884d8 JA |
55 | struct io_log *agg_io_log[2]; |
56 | ||
ebac4655 | 57 | #define TERMINATE_ALL (-1) |
75154845 | 58 | #define JOB_START_TIMEOUT (5 * 1000) |
ebac4655 | 59 | |
6ce15a32 JA |
60 | static inline void td_set_runstate(struct thread_data *td, int runstate) |
61 | { | |
62 | td->runstate = runstate; | |
63 | } | |
64 | ||
390c40e2 | 65 | static void terminate_threads(int group_id) |
ebac4655 | 66 | { |
34572e28 | 67 | struct thread_data *td; |
ebac4655 JA |
68 | int i; |
69 | ||
34572e28 | 70 | for_each_td(td, i) { |
ebac4655 | 71 | if (group_id == TERMINATE_ALL || groupid == td->groupid) { |
c1302d44 JA |
72 | /* |
73 | * if the thread is running, just let it exit | |
74 | */ | |
75 | if (td->runstate < TD_RUNNING) | |
76 | kill(td->pid, SIGQUIT); | |
ebac4655 | 77 | td->terminate = 1; |
2dc1bbeb | 78 | td->o.start_delay = 0; |
ebac4655 JA |
79 | } |
80 | } | |
81 | } | |
82 | ||
83 | static void sig_handler(int sig) | |
84 | { | |
85 | switch (sig) { | |
86 | case SIGALRM: | |
87 | update_io_ticks(); | |
88 | disk_util_timer_arm(); | |
89 | print_thread_status(); | |
90 | break; | |
91 | default: | |
6ce15a32 | 92 | printf("\nfio: terminating on signal %d\n", sig); |
ebac4655 | 93 | fflush(stdout); |
390c40e2 | 94 | terminate_threads(TERMINATE_ALL); |
ebac4655 JA |
95 | break; |
96 | } | |
97 | } | |
98 | ||
906c8d75 JA |
99 | /* |
100 | * Check if we are above the minimum rate given. | |
101 | */ | |
ebac4655 JA |
102 | static int check_min_rate(struct thread_data *td, struct timeval *now) |
103 | { | |
0904200b | 104 | unsigned long long bytes = 0; |
4e991c23 | 105 | unsigned long iops = 0; |
ebac4655 JA |
106 | unsigned long spent; |
107 | unsigned long rate; | |
ebac4655 | 108 | |
780bf1a1 JA |
109 | /* |
110 | * No minimum rate set, always ok | |
111 | */ | |
2dc1bbeb | 112 | if (!td->o.ratemin && !td->o.rate_iops_min) |
780bf1a1 JA |
113 | return 0; |
114 | ||
ebac4655 JA |
115 | /* |
116 | * allow a 2 second settle period in the beginning | |
117 | */ | |
118 | if (mtime_since(&td->start, now) < 2000) | |
119 | return 0; | |
120 | ||
4e991c23 JA |
121 | if (td_read(td)) { |
122 | iops += td->io_blocks[DDIR_READ]; | |
0904200b | 123 | bytes += td->this_io_bytes[DDIR_READ]; |
4e991c23 JA |
124 | } |
125 | if (td_write(td)) { | |
126 | iops += td->io_blocks[DDIR_WRITE]; | |
0904200b | 127 | bytes += td->this_io_bytes[DDIR_WRITE]; |
4e991c23 | 128 | } |
0904200b | 129 | |
ebac4655 JA |
130 | /* |
131 | * if rate blocks is set, sample is running | |
132 | */ | |
4e991c23 | 133 | if (td->rate_bytes || td->rate_blocks) { |
ebac4655 | 134 | spent = mtime_since(&td->lastrate, now); |
2dc1bbeb | 135 | if (spent < td->o.ratecycle) |
ebac4655 JA |
136 | return 0; |
137 | ||
2dc1bbeb | 138 | if (td->o.rate) { |
4e991c23 JA |
139 | /* |
140 | * check bandwidth specified rate | |
141 | */ | |
142 | if (bytes < td->rate_bytes) { | |
2dc1bbeb | 143 | log_err("%s: min rate %u not met\n", td->o.name, td->o.ratemin); |
4e991c23 JA |
144 | return 1; |
145 | } else { | |
146 | rate = (bytes - td->rate_bytes) / spent; | |
2dc1bbeb JA |
147 | if (rate < td->o.ratemin || bytes < td->rate_bytes) { |
148 | log_err("%s: min rate %u not met, got %luKiB/sec\n", td->o.name, td->o.ratemin, rate); | |
4e991c23 JA |
149 | return 1; |
150 | } | |
151 | } | |
413dd459 | 152 | } else { |
4e991c23 JA |
153 | /* |
154 | * checks iops specified rate | |
155 | */ | |
2dc1bbeb JA |
156 | if (iops < td->o.rate_iops) { |
157 | log_err("%s: min iops rate %u not met\n", td->o.name, td->o.rate_iops); | |
413dd459 | 158 | return 1; |
4e991c23 JA |
159 | } else { |
160 | rate = (iops - td->rate_blocks) / spent; | |
2dc1bbeb JA |
161 | if (rate < td->o.rate_iops_min || iops < td->rate_blocks) { |
162 | log_err("%s: min iops rate %u not met, got %lu\n", td->o.name, td->o.rate_iops_min, rate); | |
4e991c23 | 163 | } |
413dd459 | 164 | } |
ebac4655 JA |
165 | } |
166 | } | |
167 | ||
0904200b | 168 | td->rate_bytes = bytes; |
4e991c23 | 169 | td->rate_blocks = iops; |
ebac4655 JA |
170 | memcpy(&td->lastrate, now, sizeof(*now)); |
171 | return 0; | |
172 | } | |
173 | ||
174 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) | |
175 | { | |
2dc1bbeb | 176 | if (!td->o.timeout) |
ebac4655 | 177 | return 0; |
2dc1bbeb | 178 | if (mtime_since(&td->epoch, t) >= td->o.timeout * 1000) |
ebac4655 JA |
179 | return 1; |
180 | ||
181 | return 0; | |
182 | } | |
183 | ||
906c8d75 JA |
184 | /* |
185 | * When job exits, we can cancel the in-flight IO if we are using async | |
186 | * io. Attempt to do so. | |
187 | */ | |
ebac4655 JA |
188 | static void cleanup_pending_aio(struct thread_data *td) |
189 | { | |
ebac4655 | 190 | struct list_head *entry, *n; |
ebac4655 JA |
191 | struct io_u *io_u; |
192 | int r; | |
193 | ||
194 | /* | |
195 | * get immediately available events, if any | |
196 | */ | |
d7762cf8 | 197 | r = io_u_queued_complete(td, 0); |
b2fdda43 JA |
198 | if (r < 0) |
199 | return; | |
ebac4655 JA |
200 | |
201 | /* | |
202 | * now cancel remaining active events | |
203 | */ | |
2866c82d | 204 | if (td->io_ops->cancel) { |
ebac4655 JA |
205 | list_for_each_safe(entry, n, &td->io_u_busylist) { |
206 | io_u = list_entry(entry, struct io_u, list); | |
207 | ||
0c6e7517 JA |
208 | /* |
209 | * if the io_u isn't in flight, then that generally | |
210 | * means someone leaked an io_u. complain but fix | |
211 | * it up, so we don't stall here. | |
212 | */ | |
213 | if ((io_u->flags & IO_U_F_FLIGHT) == 0) { | |
214 | log_err("fio: non-busy IO on busy list\n"); | |
ebac4655 | 215 | put_io_u(td, io_u); |
0c6e7517 JA |
216 | } else { |
217 | r = td->io_ops->cancel(td, io_u); | |
218 | if (!r) | |
219 | put_io_u(td, io_u); | |
220 | } | |
ebac4655 JA |
221 | } |
222 | } | |
223 | ||
97601024 | 224 | if (td->cur_depth) |
d7762cf8 | 225 | r = io_u_queued_complete(td, td->cur_depth); |
ebac4655 JA |
226 | } |
227 | ||
858a3d47 JA |
228 | /* |
229 | * Helper to handle the final sync of a file. Works just like the normal | |
230 | * io path, just does everything sync. | |
231 | */ | |
232 | static int fio_io_sync(struct thread_data *td, struct fio_file *f) | |
233 | { | |
234 | struct io_u *io_u = __get_io_u(td); | |
858a3d47 JA |
235 | int ret; |
236 | ||
237 | if (!io_u) | |
238 | return 1; | |
239 | ||
240 | io_u->ddir = DDIR_SYNC; | |
241 | io_u->file = f; | |
242 | ||
243 | if (td_io_prep(td, io_u)) { | |
244 | put_io_u(td, io_u); | |
245 | return 1; | |
246 | } | |
247 | ||
755200a3 | 248 | requeue: |
858a3d47 | 249 | ret = td_io_queue(td, io_u); |
36167d82 | 250 | if (ret < 0) { |
e1161c32 | 251 | td_verror(td, io_u->error, "td_io_queue"); |
858a3d47 | 252 | put_io_u(td, io_u); |
858a3d47 | 253 | return 1; |
36167d82 | 254 | } else if (ret == FIO_Q_QUEUED) { |
d7762cf8 | 255 | if (io_u_queued_complete(td, 1) < 0) |
36167d82 | 256 | return 1; |
36167d82 JA |
257 | } else if (ret == FIO_Q_COMPLETED) { |
258 | if (io_u->error) { | |
e1161c32 | 259 | td_verror(td, io_u->error, "td_io_queue"); |
36167d82 JA |
260 | return 1; |
261 | } | |
858a3d47 | 262 | |
d7762cf8 | 263 | if (io_u_sync_complete(td, io_u) < 0) |
b2fdda43 | 264 | return 1; |
755200a3 JA |
265 | } else if (ret == FIO_Q_BUSY) { |
266 | if (td_io_commit(td)) | |
267 | return 1; | |
268 | goto requeue; | |
858a3d47 JA |
269 | } |
270 | ||
271 | return 0; | |
272 | } | |
273 | ||
906c8d75 | 274 | /* |
34403fb1 | 275 | * The main verify engine. Runs over the writes we previously submitted, |
906c8d75 JA |
276 | * reads the blocks back in, and checks the crc/md5 of the data. |
277 | */ | |
1e97cce9 | 278 | static void do_verify(struct thread_data *td) |
ebac4655 | 279 | { |
53cdc686 | 280 | struct fio_file *f; |
36167d82 | 281 | struct io_u *io_u; |
af52b345 JA |
282 | int ret, min_events; |
283 | unsigned int i; | |
e5b401d4 JA |
284 | |
285 | /* | |
286 | * sync io first and invalidate cache, to make sure we really | |
287 | * read from disk. | |
288 | */ | |
289 | for_each_file(td, f, i) { | |
3d7b485f JA |
290 | if (!(f->flags & FIO_FILE_OPEN)) |
291 | continue; | |
b2fdda43 JA |
292 | if (fio_io_sync(td, f)) |
293 | break; | |
294 | if (file_invalidate_cache(td, f)) | |
295 | break; | |
e5b401d4 | 296 | } |
ebac4655 | 297 | |
b2fdda43 JA |
298 | if (td->error) |
299 | return; | |
300 | ||
ebac4655 JA |
301 | td_set_runstate(td, TD_VERIFYING); |
302 | ||
36167d82 JA |
303 | io_u = NULL; |
304 | while (!td->terminate) { | |
5451792e JA |
305 | int ret2; |
306 | ||
ebac4655 JA |
307 | io_u = __get_io_u(td); |
308 | if (!io_u) | |
309 | break; | |
310 | ||
069c2918 JA |
311 | if (runtime_exceeded(td, &io_u->start_time)) { |
312 | put_io_u(td, io_u); | |
02bcaa8c | 313 | break; |
069c2918 | 314 | } |
02bcaa8c | 315 | |
069c2918 JA |
316 | if (get_next_verify(td, io_u)) { |
317 | put_io_u(td, io_u); | |
ebac4655 | 318 | break; |
069c2918 | 319 | } |
ebac4655 | 320 | |
069c2918 JA |
321 | if (td_io_prep(td, io_u)) { |
322 | put_io_u(td, io_u); | |
53cdc686 | 323 | break; |
069c2918 | 324 | } |
d7762cf8 JA |
325 | |
326 | io_u->end_io = verify_io_u; | |
53cdc686 | 327 | |
11786802 | 328 | ret = td_io_queue(td, io_u); |
36167d82 JA |
329 | switch (ret) { |
330 | case FIO_Q_COMPLETED: | |
331 | if (io_u->error) | |
22819ec2 | 332 | ret = -io_u->error; |
9e9d164e | 333 | else if (io_u->resid) { |
36167d82 | 334 | int bytes = io_u->xfer_buflen - io_u->resid; |
ebac4655 | 335 | |
9e9d164e JA |
336 | /* |
337 | * zero read, fail | |
338 | */ | |
339 | if (!bytes) { | |
340 | td_verror(td, ENODATA, "full resid"); | |
341 | put_io_u(td, io_u); | |
342 | break; | |
343 | } | |
8400d9b2 | 344 | |
36167d82 JA |
345 | io_u->xfer_buflen = io_u->resid; |
346 | io_u->xfer_buf += bytes; | |
8400d9b2 JA |
347 | io_u->offset += bytes; |
348 | ||
349 | if (io_u->offset == io_u->file->real_file_size) | |
350 | goto sync_done; | |
351 | ||
11786802 JA |
352 | requeue_io_u(td, &io_u); |
353 | } else { | |
8400d9b2 | 354 | sync_done: |
11786802 JA |
355 | ret = io_u_sync_complete(td, io_u); |
356 | if (ret < 0) | |
357 | break; | |
36167d82 | 358 | } |
36167d82 JA |
359 | continue; |
360 | case FIO_Q_QUEUED: | |
361 | break; | |
755200a3 JA |
362 | case FIO_Q_BUSY: |
363 | requeue_io_u(td, &io_u); | |
5451792e JA |
364 | ret2 = td_io_commit(td); |
365 | if (ret2 < 0) | |
366 | ret = ret2; | |
755200a3 | 367 | break; |
36167d82 JA |
368 | default: |
369 | assert(ret < 0); | |
e1161c32 | 370 | td_verror(td, -ret, "td_io_queue"); |
ebac4655 JA |
371 | break; |
372 | } | |
373 | ||
99784632 | 374 | if (ret < 0 || td->error) |
3af6ef39 JA |
375 | break; |
376 | ||
ebac4655 | 377 | /* |
3af6ef39 JA |
378 | * if we can queue more, do so. but check if there are |
379 | * completed io_u's first. | |
ebac4655 | 380 | */ |
97601024 | 381 | min_events = 0; |
e916b390 | 382 | if (queue_full(td) || ret == FIO_Q_BUSY) { |
3af6ef39 | 383 | min_events = 1; |
3af6ef39 | 384 | |
2dc1bbeb JA |
385 | if (td->cur_depth > td->o.iodepth_low) |
386 | min_events = td->cur_depth - td->o.iodepth_low; | |
e916b390 JA |
387 | } |
388 | ||
3af6ef39 JA |
389 | /* |
390 | * Reap required number of io units, if any, and do the | |
391 | * verification on them through the callback handler | |
392 | */ | |
d7762cf8 | 393 | if (io_u_queued_complete(td, min_events) < 0) |
ebac4655 | 394 | break; |
36167d82 | 395 | } |
ebac4655 | 396 | |
c01c0395 JA |
397 | if (!td->error) { |
398 | min_events = td->cur_depth; | |
399 | ||
400 | if (min_events) | |
401 | ret = io_u_queued_complete(td, min_events); | |
402 | } else | |
ebac4655 JA |
403 | cleanup_pending_aio(td); |
404 | ||
405 | td_set_runstate(td, TD_RUNNING); | |
406 | } | |
407 | ||
32cd46a0 | 408 | /* |
906c8d75 | 409 | * Main IO worker function. It retrieves io_u's to process and queues |
32cd46a0 JA |
410 | * and reaps them, checking for rate and errors along the way. |
411 | */ | |
ebac4655 JA |
412 | static void do_io(struct thread_data *td) |
413 | { | |
02bcaa8c | 414 | struct timeval s; |
ebac4655 | 415 | unsigned long usec; |
af52b345 JA |
416 | unsigned int i; |
417 | int ret = 0; | |
ebac4655 | 418 | |
5853e5a8 JA |
419 | td_set_runstate(td, TD_RUNNING); |
420 | ||
7bb48f84 | 421 | while ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->o.size) { |
97601024 JA |
422 | struct timeval comp_time; |
423 | long bytes_done = 0; | |
84585003 | 424 | int min_evts = 0; |
ebac4655 | 425 | struct io_u *io_u; |
5451792e | 426 | int ret2; |
ebac4655 JA |
427 | |
428 | if (td->terminate) | |
429 | break; | |
430 | ||
3d7c391d | 431 | io_u = get_io_u(td); |
ebac4655 JA |
432 | if (!io_u) |
433 | break; | |
434 | ||
435 | memcpy(&s, &io_u->start_time, sizeof(s)); | |
97601024 JA |
436 | |
437 | if (runtime_exceeded(td, &s)) { | |
438 | put_io_u(td, io_u); | |
439 | break; | |
440 | } | |
36167d82 | 441 | |
11786802 | 442 | ret = td_io_queue(td, io_u); |
36167d82 JA |
443 | switch (ret) { |
444 | case FIO_Q_COMPLETED: | |
5451792e JA |
445 | if (io_u->error) |
446 | ret = -io_u->error; | |
9e9d164e | 447 | else if (io_u->resid) { |
36167d82 JA |
448 | int bytes = io_u->xfer_buflen - io_u->resid; |
449 | ||
9e9d164e JA |
450 | /* |
451 | * zero read, fail | |
452 | */ | |
453 | if (!bytes) { | |
454 | td_verror(td, ENODATA, "full resid"); | |
455 | put_io_u(td, io_u); | |
456 | break; | |
457 | } | |
458 | ||
cec6b55d | 459 | io_u->xfer_buflen = io_u->resid; |
36167d82 | 460 | io_u->xfer_buf += bytes; |
5a7c5680 JA |
461 | io_u->offset += bytes; |
462 | ||
463 | if (io_u->offset == io_u->file->real_file_size) | |
464 | goto sync_done; | |
465 | ||
11786802 JA |
466 | requeue_io_u(td, &io_u); |
467 | } else { | |
5a7c5680 | 468 | sync_done: |
11786802 JA |
469 | fio_gettime(&comp_time, NULL); |
470 | bytes_done = io_u_sync_complete(td, io_u); | |
471 | if (bytes_done < 0) | |
472 | ret = bytes_done; | |
cec6b55d | 473 | } |
36167d82 JA |
474 | break; |
475 | case FIO_Q_QUEUED: | |
7e77dd02 JA |
476 | /* |
477 | * if the engine doesn't have a commit hook, | |
478 | * the io_u is really queued. if it does have such | |
479 | * a hook, it has to call io_u_queued() itself. | |
480 | */ | |
481 | if (td->io_ops->commit == NULL) | |
482 | io_u_queued(td, io_u); | |
36167d82 | 483 | break; |
755200a3 JA |
484 | case FIO_Q_BUSY: |
485 | requeue_io_u(td, &io_u); | |
5451792e JA |
486 | ret2 = td_io_commit(td); |
487 | if (ret2 < 0) | |
488 | ret = ret2; | |
755200a3 | 489 | break; |
36167d82 JA |
490 | default: |
491 | assert(ret < 0); | |
492 | put_io_u(td, io_u); | |
493 | break; | |
ebac4655 JA |
494 | } |
495 | ||
99784632 | 496 | if (ret < 0 || td->error) |
36167d82 JA |
497 | break; |
498 | ||
97601024 JA |
499 | /* |
500 | * See if we need to complete some commands | |
501 | */ | |
755200a3 | 502 | if (ret == FIO_Q_QUEUED || ret == FIO_Q_BUSY) { |
97601024 | 503 | min_evts = 0; |
e916b390 | 504 | if (queue_full(td) || ret == FIO_Q_BUSY) { |
36167d82 | 505 | min_evts = 1; |
ebac4655 | 506 | |
2dc1bbeb JA |
507 | if (td->cur_depth > td->o.iodepth_low) |
508 | min_evts = td->cur_depth - td->o.iodepth_low; | |
e916b390 JA |
509 | } |
510 | ||
97601024 | 511 | fio_gettime(&comp_time, NULL); |
d7762cf8 | 512 | bytes_done = io_u_queued_complete(td, min_evts); |
97601024 | 513 | if (bytes_done < 0) |
36167d82 | 514 | break; |
ebac4655 JA |
515 | } |
516 | ||
97601024 JA |
517 | if (!bytes_done) |
518 | continue; | |
519 | ||
ebac4655 JA |
520 | /* |
521 | * the rate is batched for now, it should work for batches | |
522 | * of completions except the very first one which may look | |
523 | * a little bursty | |
524 | */ | |
97601024 | 525 | usec = utime_since(&s, &comp_time); |
ebac4655 | 526 | |
413dd459 | 527 | rate_throttle(td, usec, bytes_done); |
ebac4655 | 528 | |
97601024 | 529 | if (check_min_rate(td, &comp_time)) { |
98aa62d8 | 530 | if (exitall_on_terminate) |
390c40e2 | 531 | terminate_threads(td->groupid); |
e1161c32 | 532 | td_verror(td, ENODATA, "check_min_rate"); |
ebac4655 JA |
533 | break; |
534 | } | |
535 | ||
2dc1bbeb | 536 | if (td->o.thinktime) { |
9c1f7434 JA |
537 | unsigned long long b; |
538 | ||
539 | b = td->io_blocks[0] + td->io_blocks[1]; | |
2dc1bbeb | 540 | if (!(b % td->o.thinktime_blocks)) { |
48097d5c JA |
541 | int left; |
542 | ||
2dc1bbeb JA |
543 | if (td->o.thinktime_spin) |
544 | __usec_sleep(td->o.thinktime_spin); | |
48097d5c | 545 | |
2dc1bbeb | 546 | left = td->o.thinktime - td->o.thinktime_spin; |
48097d5c JA |
547 | if (left) |
548 | usec_sleep(td, left); | |
549 | } | |
9c1f7434 | 550 | } |
ebac4655 JA |
551 | } |
552 | ||
4d2413c6 | 553 | if (!td->error) { |
3d7c391d JA |
554 | struct fio_file *f; |
555 | ||
c01c0395 JA |
556 | i = td->cur_depth; |
557 | if (i) | |
558 | ret = io_u_queued_complete(td, i); | |
ebac4655 | 559 | |
2dc1bbeb | 560 | if (should_fsync(td) && td->o.end_fsync) { |
84585003 | 561 | td_set_runstate(td, TD_FSYNCING); |
3d7b485f JA |
562 | |
563 | for_each_file(td, f, i) { | |
564 | if (!(f->flags & FIO_FILE_OPEN)) | |
565 | continue; | |
858a3d47 | 566 | fio_io_sync(td, f); |
3d7b485f | 567 | } |
84585003 | 568 | } |
c01c0395 JA |
569 | } else |
570 | cleanup_pending_aio(td); | |
ebac4655 JA |
571 | } |
572 | ||
ebac4655 JA |
573 | static void cleanup_io_u(struct thread_data *td) |
574 | { | |
575 | struct list_head *entry, *n; | |
576 | struct io_u *io_u; | |
577 | ||
578 | list_for_each_safe(entry, n, &td->io_u_freelist) { | |
579 | io_u = list_entry(entry, struct io_u, list); | |
580 | ||
581 | list_del(&io_u->list); | |
582 | free(io_u); | |
583 | } | |
584 | ||
2f9ade3c | 585 | free_io_mem(td); |
ebac4655 JA |
586 | } |
587 | ||
6b9cea23 JA |
588 | /* |
589 | * "randomly" fill the buffer contents | |
590 | */ | |
66eeb296 | 591 | static void fill_rand_buf(struct io_u *io_u, int max_bs) |
6b9cea23 | 592 | { |
66eeb296 | 593 | int *ptr = io_u->buf; |
6b9cea23 JA |
594 | |
595 | while ((void *) ptr - io_u->buf < max_bs) { | |
596 | *ptr = rand() * 0x9e370001; | |
597 | ptr++; | |
598 | } | |
599 | } | |
600 | ||
ebac4655 JA |
601 | static int init_io_u(struct thread_data *td) |
602 | { | |
2b4ce34f | 603 | unsigned long long buf_size; |
ebac4655 | 604 | struct io_u *io_u; |
a00735e6 | 605 | unsigned int max_bs; |
ebac4655 JA |
606 | int i, max_units; |
607 | char *p; | |
608 | ||
2866c82d | 609 | if (td->io_ops->flags & FIO_SYNCIO) |
ebac4655 JA |
610 | max_units = 1; |
611 | else | |
2dc1bbeb | 612 | max_units = td->o.iodepth; |
ebac4655 | 613 | |
2dc1bbeb | 614 | max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]); |
2b4ce34f JA |
615 | buf_size = (unsigned long long) max_bs * (unsigned long long) max_units; |
616 | buf_size += page_mask; | |
617 | if (buf_size != (size_t) buf_size) { | |
618 | log_err("fio: IO memory too large. Reduce max_bs or iodepth\n"); | |
619 | return 1; | |
620 | } | |
621 | ||
622 | td->orig_buffer_size = buf_size; | |
74b025b0 | 623 | |
2dc1bbeb JA |
624 | if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) |
625 | td->orig_buffer_size = (td->orig_buffer_size + td->o.hugepage_size - 1) & ~(td->o.hugepage_size - 1); | |
4afbf66f JA |
626 | else if (td->orig_buffer_size & page_mask) |
627 | td->orig_buffer_size = (td->orig_buffer_size + page_mask) & ~page_mask; | |
ebac4655 | 628 | |
2f9ade3c JA |
629 | if (allocate_io_mem(td)) |
630 | return 1; | |
ebac4655 | 631 | |
ebac4655 JA |
632 | p = ALIGN(td->orig_buffer); |
633 | for (i = 0; i < max_units; i++) { | |
634 | io_u = malloc(sizeof(*io_u)); | |
635 | memset(io_u, 0, sizeof(*io_u)); | |
636 | INIT_LIST_HEAD(&io_u->list); | |
637 | ||
a00735e6 | 638 | io_u->buf = p + max_bs * i; |
6b9cea23 | 639 | if (td_write(td) || td_rw(td)) |
a00735e6 | 640 | fill_rand_buf(io_u, max_bs); |
6b9cea23 | 641 | |
b1ff3403 | 642 | io_u->index = i; |
0c6e7517 | 643 | io_u->flags = IO_U_F_FREE; |
ebac4655 JA |
644 | list_add(&io_u->list, &td->io_u_freelist); |
645 | } | |
646 | ||
433afcb4 JA |
647 | io_u_init_timeout(); |
648 | ||
ebac4655 JA |
649 | return 0; |
650 | } | |
651 | ||
da86774e JA |
652 | static int switch_ioscheduler(struct thread_data *td) |
653 | { | |
654 | char tmp[256], tmp2[128]; | |
655 | FILE *f; | |
656 | int ret; | |
657 | ||
ba0fbe10 | 658 | if (td->io_ops->flags & FIO_DISKLESSIO) |
f48b467c JA |
659 | return 0; |
660 | ||
da86774e JA |
661 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); |
662 | ||
663 | f = fopen(tmp, "r+"); | |
664 | if (!f) { | |
cbf5c1d7 JA |
665 | if (errno == ENOENT) { |
666 | log_err("fio: os or kernel doesn't support IO scheduler switching\n"); | |
667 | return 0; | |
668 | } | |
669 | td_verror(td, errno, "fopen iosched"); | |
da86774e JA |
670 | return 1; |
671 | } | |
672 | ||
673 | /* | |
674 | * Set io scheduler. | |
675 | */ | |
2dc1bbeb | 676 | ret = fwrite(td->o.ioscheduler, strlen(td->o.ioscheduler), 1, f); |
da86774e | 677 | if (ferror(f) || ret != 1) { |
e1161c32 | 678 | td_verror(td, errno, "fwrite"); |
da86774e JA |
679 | fclose(f); |
680 | return 1; | |
681 | } | |
682 | ||
683 | rewind(f); | |
684 | ||
685 | /* | |
686 | * Read back and check that the selected scheduler is now the default. | |
687 | */ | |
688 | ret = fread(tmp, 1, sizeof(tmp), f); | |
689 | if (ferror(f) || ret < 0) { | |
e1161c32 | 690 | td_verror(td, errno, "fread"); |
da86774e JA |
691 | fclose(f); |
692 | return 1; | |
693 | } | |
694 | ||
2dc1bbeb | 695 | sprintf(tmp2, "[%s]", td->o.ioscheduler); |
da86774e | 696 | if (!strstr(tmp, tmp2)) { |
2dc1bbeb | 697 | log_err("fio: io scheduler %s not found\n", td->o.ioscheduler); |
e1161c32 | 698 | td_verror(td, EINVAL, "iosched_switch"); |
da86774e JA |
699 | fclose(f); |
700 | return 1; | |
701 | } | |
702 | ||
703 | fclose(f); | |
704 | return 0; | |
705 | } | |
706 | ||
a978ba68 | 707 | static int clear_io_state(struct thread_data *td) |
ebac4655 | 708 | { |
53cdc686 | 709 | struct fio_file *f; |
af52b345 JA |
710 | unsigned int i; |
711 | int ret; | |
ebac4655 | 712 | |
756867bd | 713 | td->ts.stat_io_bytes[0] = td->ts.stat_io_bytes[1] = 0; |
ebac4655 | 714 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; |
20dc95c4 | 715 | td->zone_bytes = 0; |
d7d3b49b | 716 | td->rate_bytes = 0; |
4e991c23 | 717 | td->rate_blocks = 0; |
38d77cae | 718 | td->rw_end_set[0] = td->rw_end_set[1] = 0; |
ebac4655 | 719 | |
c1324df1 JA |
720 | td->last_was_sync = 0; |
721 | ||
a978ba68 JA |
722 | for_each_file(td, f, i) |
723 | td_io_close_file(td, f); | |
53cdc686 | 724 | |
a978ba68 JA |
725 | ret = 0; |
726 | for_each_file(td, f, i) { | |
727 | ret = td_io_open_file(td, f); | |
728 | if (ret) | |
729 | break; | |
53cdc686 | 730 | } |
a978ba68 JA |
731 | |
732 | return ret; | |
ebac4655 JA |
733 | } |
734 | ||
906c8d75 JA |
735 | /* |
736 | * Entry point for the thread based jobs. The process based jobs end up | |
737 | * here as well, after a little setup. | |
738 | */ | |
ebac4655 JA |
739 | static void *thread_main(void *data) |
740 | { | |
69008999 | 741 | unsigned long long runtime[2]; |
ebac4655 | 742 | struct thread_data *td = data; |
38d77cae | 743 | unsigned long elapsed; |
a978ba68 | 744 | int clear_state; |
ebac4655 | 745 | |
2dc1bbeb | 746 | if (!td->o.use_thread) |
ebac4655 JA |
747 | setsid(); |
748 | ||
749 | td->pid = getpid(); | |
750 | ||
aea47d44 JA |
751 | INIT_LIST_HEAD(&td->io_u_freelist); |
752 | INIT_LIST_HEAD(&td->io_u_busylist); | |
755200a3 | 753 | INIT_LIST_HEAD(&td->io_u_requeues); |
aea47d44 | 754 | INIT_LIST_HEAD(&td->io_log_list); |
8de8f047 | 755 | INIT_LIST_HEAD(&td->io_hist_list); |
bb5d7d0b | 756 | td->io_hist_tree = RB_ROOT; |
aea47d44 | 757 | |
ebac4655 | 758 | if (init_io_u(td)) |
bda4fd9e | 759 | goto err_sem; |
ebac4655 JA |
760 | |
761 | if (fio_setaffinity(td) == -1) { | |
e1161c32 | 762 | td_verror(td, errno, "cpu_set_affinity"); |
bda4fd9e | 763 | goto err_sem; |
ebac4655 JA |
764 | } |
765 | ||
aea47d44 | 766 | if (init_iolog(td)) |
bda4fd9e | 767 | goto err_sem; |
aea47d44 | 768 | |
ebac4655 JA |
769 | if (td->ioprio) { |
770 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { | |
e1161c32 | 771 | td_verror(td, errno, "ioprio_set"); |
bda4fd9e | 772 | goto err_sem; |
ebac4655 JA |
773 | } |
774 | } | |
775 | ||
2dc1bbeb | 776 | if (nice(td->o.nice) == -1) { |
e1161c32 | 777 | td_verror(td, errno, "nice"); |
bda4fd9e | 778 | goto err_sem; |
b6f4d880 JA |
779 | } |
780 | ||
2dc1bbeb | 781 | if (td->o.ioscheduler && switch_ioscheduler(td)) |
bda4fd9e | 782 | goto err_sem; |
da86774e | 783 | |
75154845 | 784 | td_set_runstate(td, TD_INITIALIZED); |
07739b57 JA |
785 | fio_sem_up(startup_sem); |
786 | fio_sem_down(td->mutex); | |
ebac4655 | 787 | |
37f56873 JA |
788 | /* |
789 | * the ->mutex semaphore is now no longer used, close it to avoid | |
790 | * eating a file descriptor | |
791 | */ | |
792 | fio_sem_remove(td->mutex); | |
793 | ||
2dc1bbeb | 794 | if (!td->o.create_serialize && setup_files(td)) |
ebac4655 JA |
795 | goto err; |
796 | ||
7d6c5283 JA |
797 | if (td_io_init(td)) |
798 | goto err; | |
799 | ||
b5af8293 JA |
800 | if (open_files(td)) |
801 | goto err; | |
802 | ||
68727076 JA |
803 | if (init_random_map(td)) |
804 | goto err; | |
805 | ||
2dc1bbeb JA |
806 | if (td->o.exec_prerun) { |
807 | if (system(td->o.exec_prerun) < 0) | |
69cfd7e0 JA |
808 | goto err; |
809 | } | |
4e0ba8af | 810 | |
69008999 | 811 | fio_gettime(&td->epoch, NULL); |
433afcb4 | 812 | memcpy(&td->timeout_end, &td->epoch, sizeof(td->epoch)); |
756867bd | 813 | getrusage(RUSAGE_SELF, &td->ts.ru_start); |
69008999 JA |
814 | |
815 | runtime[0] = runtime[1] = 0; | |
a978ba68 | 816 | clear_state = 0; |
2dc1bbeb | 817 | while (td->o.loops--) { |
02bcaa8c | 818 | fio_gettime(&td->start, NULL); |
756867bd | 819 | memcpy(&td->ts.stat_sample_time, &td->start, sizeof(td->start)); |
ebac4655 | 820 | |
2dc1bbeb | 821 | if (td->o.ratemin) |
756867bd | 822 | memcpy(&td->lastrate, &td->ts.stat_sample_time, sizeof(td->lastrate)); |
ebac4655 | 823 | |
a978ba68 JA |
824 | if (clear_state && clear_io_state(td)) |
825 | break; | |
826 | ||
ebac4655 JA |
827 | prune_io_piece_log(td); |
828 | ||
ba0fbe10 | 829 | do_io(td); |
ebac4655 | 830 | |
a978ba68 JA |
831 | clear_state = 1; |
832 | ||
38d77cae JA |
833 | if (td_read(td) && td->io_bytes[DDIR_READ]) { |
834 | if (td->rw_end_set[DDIR_READ]) | |
835 | elapsed = utime_since(&td->start, &td->rw_end[DDIR_READ]); | |
836 | else | |
837 | elapsed = utime_since_now(&td->start); | |
838 | ||
839 | runtime[DDIR_READ] += elapsed; | |
840 | } | |
841 | if (td_write(td) && td->io_bytes[DDIR_WRITE]) { | |
842 | if (td->rw_end_set[DDIR_WRITE]) | |
843 | elapsed = utime_since(&td->start, &td->rw_end[DDIR_WRITE]); | |
844 | else | |
845 | elapsed = utime_since_now(&td->start); | |
846 | ||
847 | runtime[DDIR_WRITE] += elapsed; | |
848 | } | |
413dd459 | 849 | |
ebac4655 JA |
850 | if (td->error || td->terminate) |
851 | break; | |
852 | ||
2dc1bbeb | 853 | if (td->o.verify == VERIFY_NONE) |
ebac4655 JA |
854 | continue; |
855 | ||
a978ba68 JA |
856 | if (clear_io_state(td)) |
857 | break; | |
858 | ||
02bcaa8c | 859 | fio_gettime(&td->start, NULL); |
ebac4655 JA |
860 | |
861 | do_verify(td); | |
862 | ||
69008999 | 863 | runtime[DDIR_READ] += utime_since_now(&td->start); |
ebac4655 JA |
864 | |
865 | if (td->error || td->terminate) | |
866 | break; | |
867 | } | |
868 | ||
36dff966 | 869 | update_rusage_stat(td); |
756867bd JA |
870 | td->ts.runtime[0] = runtime[0] / 1000; |
871 | td->ts.runtime[1] = runtime[1] / 1000; | |
872 | td->ts.total_run_time = mtime_since_now(&td->epoch); | |
873 | td->ts.io_bytes[0] = td->io_bytes[0]; | |
874 | td->ts.io_bytes[1] = td->io_bytes[1]; | |
875 | ||
876 | if (td->ts.bw_log) | |
877 | finish_log(td, td->ts.bw_log, "bw"); | |
878 | if (td->ts.slat_log) | |
879 | finish_log(td, td->ts.slat_log, "slat"); | |
880 | if (td->ts.clat_log) | |
881 | finish_log(td, td->ts.clat_log, "clat"); | |
2dc1bbeb | 882 | if (td->o.write_iolog_file) |
843a7413 | 883 | write_iolog_close(td); |
2dc1bbeb JA |
884 | if (td->o.exec_postrun) { |
885 | if (system(td->o.exec_postrun) < 0) | |
886 | log_err("fio: postrun %s failed\n", td->o.exec_postrun); | |
69cfd7e0 | 887 | } |
ebac4655 JA |
888 | |
889 | if (exitall_on_terminate) | |
390c40e2 | 890 | terminate_threads(td->groupid); |
ebac4655 JA |
891 | |
892 | err: | |
5bf13a5a JA |
893 | if (td->error) |
894 | printf("fio: pid=%d, err=%d/%s\n", td->pid, td->error, td->verror); | |
53cdc686 | 895 | close_files(td); |
2866c82d | 896 | close_ioengine(td); |
ebac4655 | 897 | cleanup_io_u(td); |
d23bb327 | 898 | options_mem_free(td); |
ebac4655 | 899 | td_set_runstate(td, TD_EXITED); |
43d76807 | 900 | return (void *) (unsigned long) td->error; |
bda4fd9e JA |
901 | err_sem: |
902 | fio_sem_up(startup_sem); | |
903 | goto err; | |
ebac4655 JA |
904 | } |
905 | ||
906c8d75 JA |
906 | /* |
907 | * We cannot pass the td data into a forked process, so attach the td and | |
908 | * pass it to the thread worker. | |
909 | */ | |
a6418147 | 910 | static int fork_main(int shmid, int offset) |
ebac4655 JA |
911 | { |
912 | struct thread_data *td; | |
a6418147 | 913 | void *data, *ret; |
ebac4655 JA |
914 | |
915 | data = shmat(shmid, NULL, 0); | |
916 | if (data == (void *) -1) { | |
a6418147 JA |
917 | int __err = errno; |
918 | ||
ebac4655 | 919 | perror("shmat"); |
a6418147 | 920 | return __err; |
ebac4655 JA |
921 | } |
922 | ||
923 | td = data + offset * sizeof(struct thread_data); | |
a6418147 | 924 | ret = thread_main(td); |
ebac4655 | 925 | shmdt(data); |
43d76807 | 926 | return (int) (unsigned long) ret; |
ebac4655 JA |
927 | } |
928 | ||
906c8d75 JA |
929 | /* |
930 | * Run over the job map and reap the threads that have exited, if any. | |
931 | */ | |
ebac4655 JA |
932 | static void reap_threads(int *nr_running, int *t_rate, int *m_rate) |
933 | { | |
34572e28 | 934 | struct thread_data *td; |
fab6aa71 | 935 | int i, cputhreads, pending, status, ret; |
ebac4655 JA |
936 | |
937 | /* | |
938 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
939 | */ | |
4d2413c6 | 940 | pending = cputhreads = 0; |
34572e28 | 941 | for_each_td(td, i) { |
3707f45b | 942 | int flags = 0; |
a2f77c9f | 943 | |
84585003 JA |
944 | /* |
945 | * ->io_ops is NULL for a thread that has closed its | |
946 | * io engine | |
947 | */ | |
ba0fbe10 | 948 | if (td->io_ops && !strcmp(td->io_ops->name, "cpuio")) |
b990b5c0 JA |
949 | cputhreads++; |
950 | ||
a2f77c9f JA |
951 | if (!td->pid || td->runstate == TD_REAPED) |
952 | continue; | |
2dc1bbeb | 953 | if (td->o.use_thread) { |
3707f45b JA |
954 | if (td->runstate == TD_EXITED) { |
955 | td_set_runstate(td, TD_REAPED); | |
956 | goto reaped; | |
957 | } | |
958 | continue; | |
959 | } | |
a2f77c9f JA |
960 | |
961 | flags = WNOHANG; | |
962 | if (td->runstate == TD_EXITED) | |
963 | flags = 0; | |
964 | ||
965 | /* | |
966 | * check if someone quit or got killed in an unusual way | |
967 | */ | |
968 | ret = waitpid(td->pid, &status, flags); | |
3707f45b | 969 | if (ret < 0) { |
a2f77c9f JA |
970 | if (errno == ECHILD) { |
971 | log_err("fio: pid=%d disappeared %d\n", td->pid, td->runstate); | |
972 | td_set_runstate(td, TD_REAPED); | |
973 | goto reaped; | |
974 | } | |
975 | perror("waitpid"); | |
976 | } else if (ret == td->pid) { | |
977 | if (WIFSIGNALED(status)) { | |
fab6aa71 JA |
978 | int sig = WTERMSIG(status); |
979 | ||
d9244941 JA |
980 | if (sig != SIGQUIT) |
981 | log_err("fio: pid=%d, got signal=%d\n", td->pid, sig); | |
fab6aa71 JA |
982 | td_set_runstate(td, TD_REAPED); |
983 | goto reaped; | |
984 | } | |
a2f77c9f JA |
985 | if (WIFEXITED(status)) { |
986 | if (WEXITSTATUS(status) && !td->error) | |
987 | td->error = WEXITSTATUS(status); | |
a2f77c9f | 988 | |
a2f77c9f JA |
989 | td_set_runstate(td, TD_REAPED); |
990 | goto reaped; | |
a6418147 JA |
991 | } |
992 | } | |
ebac4655 | 993 | |
a2f77c9f JA |
994 | /* |
995 | * thread is not dead, continue | |
996 | */ | |
e48676ba | 997 | pending++; |
a2f77c9f | 998 | continue; |
fab6aa71 | 999 | reaped: |
2dc1bbeb | 1000 | if (td->o.use_thread) { |
3707f45b JA |
1001 | long ret; |
1002 | ||
1003 | if (pthread_join(td->thread, (void *) &ret)) | |
1004 | perror("pthread_join"); | |
1005 | } | |
1006 | ||
ebac4655 | 1007 | (*nr_running)--; |
2dc1bbeb JA |
1008 | (*m_rate) -= td->o.ratemin; |
1009 | (*t_rate) -= td->o.rate; | |
e48676ba | 1010 | pending--; |
a2f77c9f JA |
1011 | |
1012 | if (td->error) | |
1013 | exit_value++; | |
ebac4655 | 1014 | } |
b990b5c0 | 1015 | |
4d2413c6 | 1016 | if (*nr_running == cputhreads && !pending) |
390c40e2 | 1017 | terminate_threads(TERMINATE_ALL); |
ebac4655 JA |
1018 | } |
1019 | ||
906c8d75 JA |
1020 | /* |
1021 | * Main function for kicking off and reaping jobs, as needed. | |
1022 | */ | |
ebac4655 JA |
1023 | static void run_threads(void) |
1024 | { | |
ebac4655 JA |
1025 | struct thread_data *td; |
1026 | unsigned long spent; | |
1027 | int i, todo, nr_running, m_rate, t_rate, nr_started; | |
fcb6ade2 | 1028 | |
2f9ade3c JA |
1029 | if (fio_pin_memory()) |
1030 | return; | |
ebac4655 | 1031 | |
c6ae0a5b | 1032 | if (!terse_output) { |
9cedf167 JA |
1033 | printf("Starting "); |
1034 | if (nr_thread) | |
1035 | printf("%d thread%s", nr_thread, nr_thread > 1 ? "s" : ""); | |
1036 | if (nr_process) { | |
1037 | if (nr_thread) | |
1038 | printf(" and "); | |
1039 | printf("%d process%s", nr_process, nr_process > 1 ? "es" : ""); | |
1040 | } | |
1041 | printf("\n"); | |
c6ae0a5b JA |
1042 | fflush(stdout); |
1043 | } | |
c04f7ec3 | 1044 | |
4efa970e JA |
1045 | signal(SIGINT, sig_handler); |
1046 | signal(SIGALRM, sig_handler); | |
1047 | ||
ebac4655 JA |
1048 | todo = thread_number; |
1049 | nr_running = 0; | |
1050 | nr_started = 0; | |
1051 | m_rate = t_rate = 0; | |
1052 | ||
34572e28 | 1053 | for_each_td(td, i) { |
263e529f | 1054 | print_status_init(td->thread_number - 1); |
ebac4655 | 1055 | |
2dc1bbeb | 1056 | if (!td->o.create_serialize) { |
380cf265 | 1057 | init_disk_util(td); |
ebac4655 | 1058 | continue; |
380cf265 | 1059 | } |
ebac4655 JA |
1060 | |
1061 | /* | |
1062 | * do file setup here so it happens sequentially, | |
1063 | * we don't want X number of threads getting their | |
1064 | * client data interspersed on disk | |
1065 | */ | |
53cdc686 | 1066 | if (setup_files(td)) { |
5bf13a5a JA |
1067 | exit_value++; |
1068 | if (td->error) | |
1069 | log_err("fio: pid=%d, err=%d/%s\n", td->pid, td->error, td->verror); | |
ebac4655 JA |
1070 | td_set_runstate(td, TD_REAPED); |
1071 | todo--; | |
1072 | } | |
380cf265 JA |
1073 | |
1074 | init_disk_util(td); | |
ebac4655 JA |
1075 | } |
1076 | ||
a2f77c9f JA |
1077 | set_genesis_time(); |
1078 | ||
ebac4655 | 1079 | while (todo) { |
75154845 JA |
1080 | struct thread_data *map[MAX_JOBS]; |
1081 | struct timeval this_start; | |
1082 | int this_jobs = 0, left; | |
1083 | ||
ebac4655 JA |
1084 | /* |
1085 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
1086 | */ | |
34572e28 | 1087 | for_each_td(td, i) { |
ebac4655 JA |
1088 | if (td->runstate != TD_NOT_CREATED) |
1089 | continue; | |
1090 | ||
1091 | /* | |
1092 | * never got a chance to start, killed by other | |
1093 | * thread for some reason | |
1094 | */ | |
1095 | if (td->terminate) { | |
1096 | todo--; | |
1097 | continue; | |
1098 | } | |
1099 | ||
2dc1bbeb | 1100 | if (td->o.start_delay) { |
263e529f | 1101 | spent = mtime_since_genesis(); |
ebac4655 | 1102 | |
2dc1bbeb | 1103 | if (td->o.start_delay * 1000 > spent) |
ebac4655 JA |
1104 | continue; |
1105 | } | |
1106 | ||
2dc1bbeb | 1107 | if (td->o.stonewall && (nr_started || nr_running)) |
ebac4655 JA |
1108 | break; |
1109 | ||
75154845 JA |
1110 | /* |
1111 | * Set state to created. Thread will transition | |
1112 | * to TD_INITIALIZED when it's done setting up. | |
1113 | */ | |
ebac4655 | 1114 | td_set_runstate(td, TD_CREATED); |
75154845 | 1115 | map[this_jobs++] = td; |
ebac4655 JA |
1116 | nr_started++; |
1117 | ||
2dc1bbeb | 1118 | if (td->o.use_thread) { |
ebac4655 JA |
1119 | if (pthread_create(&td->thread, NULL, thread_main, td)) { |
1120 | perror("thread_create"); | |
1121 | nr_started--; | |
c8bb6faf | 1122 | break; |
ebac4655 JA |
1123 | } |
1124 | } else { | |
07739b57 | 1125 | if (!fork()) { |
a6418147 JA |
1126 | int ret = fork_main(shm_id, i); |
1127 | ||
1128 | exit(ret); | |
ebac4655 JA |
1129 | } |
1130 | } | |
07739b57 | 1131 | fio_sem_down(startup_sem); |
ebac4655 JA |
1132 | } |
1133 | ||
1134 | /* | |
75154845 JA |
1135 | * Wait for the started threads to transition to |
1136 | * TD_INITIALIZED. | |
ebac4655 | 1137 | */ |
02bcaa8c | 1138 | fio_gettime(&this_start, NULL); |
75154845 | 1139 | left = this_jobs; |
6ce15a32 | 1140 | while (left && !fio_abort) { |
75154845 JA |
1141 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) |
1142 | break; | |
1143 | ||
1144 | usleep(100000); | |
1145 | ||
1146 | for (i = 0; i < this_jobs; i++) { | |
1147 | td = map[i]; | |
1148 | if (!td) | |
1149 | continue; | |
b6f4d880 | 1150 | if (td->runstate == TD_INITIALIZED) { |
75154845 JA |
1151 | map[i] = NULL; |
1152 | left--; | |
b6f4d880 JA |
1153 | } else if (td->runstate >= TD_EXITED) { |
1154 | map[i] = NULL; | |
1155 | left--; | |
1156 | todo--; | |
1157 | nr_running++; /* work-around... */ | |
75154845 JA |
1158 | } |
1159 | } | |
1160 | } | |
1161 | ||
1162 | if (left) { | |
3b70d7e5 | 1163 | log_err("fio: %d jobs failed to start\n", left); |
75154845 JA |
1164 | for (i = 0; i < this_jobs; i++) { |
1165 | td = map[i]; | |
1166 | if (!td) | |
1167 | continue; | |
1168 | kill(td->pid, SIGTERM); | |
1169 | } | |
1170 | break; | |
1171 | } | |
1172 | ||
1173 | /* | |
b6f4d880 | 1174 | * start created threads (TD_INITIALIZED -> TD_RUNNING). |
75154845 | 1175 | */ |
34572e28 | 1176 | for_each_td(td, i) { |
75154845 | 1177 | if (td->runstate != TD_INITIALIZED) |
ebac4655 JA |
1178 | continue; |
1179 | ||
1180 | td_set_runstate(td, TD_RUNNING); | |
1181 | nr_running++; | |
1182 | nr_started--; | |
2dc1bbeb JA |
1183 | m_rate += td->o.ratemin; |
1184 | t_rate += td->o.rate; | |
75154845 | 1185 | todo--; |
07739b57 | 1186 | fio_sem_up(td->mutex); |
ebac4655 JA |
1187 | } |
1188 | ||
1189 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1190 | ||
1191 | if (todo) | |
1192 | usleep(100000); | |
1193 | } | |
1194 | ||
1195 | while (nr_running) { | |
1196 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1197 | usleep(10000); | |
1198 | } | |
1199 | ||
1200 | update_io_ticks(); | |
2f9ade3c | 1201 | fio_unpin_memory(); |
ebac4655 JA |
1202 | } |
1203 | ||
ebac4655 JA |
1204 | int main(int argc, char *argv[]) |
1205 | { | |
29d610e1 JA |
1206 | long ps; |
1207 | ||
dbe1125e JA |
1208 | /* |
1209 | * We need locale for number printing, if it isn't set then just | |
1210 | * go with the US format. | |
1211 | */ | |
1212 | if (!getenv("LC_NUMERIC")) | |
1213 | setlocale(LC_NUMERIC, "en_US"); | |
1214 | ||
ebac4655 JA |
1215 | if (parse_options(argc, argv)) |
1216 | return 1; | |
1217 | ||
4b472fa3 JA |
1218 | if (!thread_number) |
1219 | return 0; | |
ebac4655 | 1220 | |
29d610e1 JA |
1221 | ps = sysconf(_SC_PAGESIZE); |
1222 | if (ps < 0) { | |
1223 | log_err("Failed to get page size\n"); | |
1224 | return 1; | |
1225 | } | |
1226 | ||
cfc99db7 | 1227 | page_size = ps; |
29d610e1 JA |
1228 | page_mask = ps - 1; |
1229 | ||
bb3884d8 JA |
1230 | if (write_bw_log) { |
1231 | setup_log(&agg_io_log[DDIR_READ]); | |
1232 | setup_log(&agg_io_log[DDIR_WRITE]); | |
1233 | } | |
1234 | ||
07739b57 JA |
1235 | startup_sem = fio_sem_init(0); |
1236 | ||
a2f77c9f JA |
1237 | set_genesis_time(); |
1238 | ||
ebac4655 JA |
1239 | disk_util_timer_arm(); |
1240 | ||
1241 | run_threads(); | |
6ce15a32 | 1242 | |
bb3884d8 | 1243 | if (!fio_abort) { |
6ce15a32 | 1244 | show_run_stats(); |
bb3884d8 JA |
1245 | if (write_bw_log) { |
1246 | __finish_log(agg_io_log[DDIR_READ],"agg-read_bw.log"); | |
1247 | __finish_log(agg_io_log[DDIR_WRITE],"agg-write_bw.log"); | |
1248 | } | |
1249 | } | |
ebac4655 | 1250 | |
07739b57 | 1251 | fio_sem_remove(startup_sem); |
437c9b71 | 1252 | return exit_value; |
ebac4655 | 1253 | } |