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" | |
210f43b3 | 38 | #include "hash.h" |
ebac4655 | 39 | |
cfc99db7 JA |
40 | unsigned long page_mask; |
41 | unsigned long page_size; | |
29d610e1 JA |
42 | #define ALIGN(buf) \ |
43 | (char *) (((unsigned long) (buf) + page_mask) & ~page_mask) | |
ebac4655 JA |
44 | |
45 | int groupid = 0; | |
46 | int thread_number = 0; | |
9cedf167 JA |
47 | int nr_process = 0; |
48 | int nr_thread = 0; | |
ebac4655 | 49 | int shm_id = 0; |
53cdc686 | 50 | int temp_stall_ts; |
ebac4655 | 51 | |
07739b57 | 52 | static struct fio_sem *startup_sem; |
6ce15a32 | 53 | static volatile int fio_abort; |
437c9b71 | 54 | static int exit_value; |
ebac4655 | 55 | |
bb3884d8 JA |
56 | struct io_log *agg_io_log[2]; |
57 | ||
ebac4655 | 58 | #define TERMINATE_ALL (-1) |
75154845 | 59 | #define JOB_START_TIMEOUT (5 * 1000) |
ebac4655 | 60 | |
6ce15a32 JA |
61 | static inline void td_set_runstate(struct thread_data *td, int runstate) |
62 | { | |
63 | td->runstate = runstate; | |
64 | } | |
65 | ||
390c40e2 | 66 | static void terminate_threads(int group_id) |
ebac4655 | 67 | { |
34572e28 | 68 | struct thread_data *td; |
ebac4655 JA |
69 | int i; |
70 | ||
34572e28 | 71 | for_each_td(td, i) { |
ebac4655 | 72 | if (group_id == TERMINATE_ALL || groupid == td->groupid) { |
ef5fffa5 | 73 | if (td->runstate <= TD_RUNNING) |
c1302d44 | 74 | kill(td->pid, SIGQUIT); |
ebac4655 | 75 | td->terminate = 1; |
2dc1bbeb | 76 | td->o.start_delay = 0; |
ebac4655 JA |
77 | } |
78 | } | |
79 | } | |
80 | ||
81 | static void sig_handler(int sig) | |
82 | { | |
83 | switch (sig) { | |
84 | case SIGALRM: | |
85 | update_io_ticks(); | |
86 | disk_util_timer_arm(); | |
87 | print_thread_status(); | |
88 | break; | |
89 | default: | |
6ce15a32 | 90 | printf("\nfio: terminating on signal %d\n", sig); |
ebac4655 | 91 | fflush(stdout); |
390c40e2 | 92 | terminate_threads(TERMINATE_ALL); |
ebac4655 JA |
93 | break; |
94 | } | |
95 | } | |
96 | ||
906c8d75 JA |
97 | /* |
98 | * Check if we are above the minimum rate given. | |
99 | */ | |
ebac4655 JA |
100 | static int check_min_rate(struct thread_data *td, struct timeval *now) |
101 | { | |
0904200b | 102 | unsigned long long bytes = 0; |
4e991c23 | 103 | unsigned long iops = 0; |
ebac4655 JA |
104 | unsigned long spent; |
105 | unsigned long rate; | |
ebac4655 | 106 | |
780bf1a1 JA |
107 | /* |
108 | * No minimum rate set, always ok | |
109 | */ | |
2dc1bbeb | 110 | if (!td->o.ratemin && !td->o.rate_iops_min) |
780bf1a1 JA |
111 | return 0; |
112 | ||
ebac4655 JA |
113 | /* |
114 | * allow a 2 second settle period in the beginning | |
115 | */ | |
116 | if (mtime_since(&td->start, now) < 2000) | |
117 | return 0; | |
118 | ||
4e991c23 JA |
119 | if (td_read(td)) { |
120 | iops += td->io_blocks[DDIR_READ]; | |
0904200b | 121 | bytes += td->this_io_bytes[DDIR_READ]; |
4e991c23 JA |
122 | } |
123 | if (td_write(td)) { | |
124 | iops += td->io_blocks[DDIR_WRITE]; | |
0904200b | 125 | bytes += td->this_io_bytes[DDIR_WRITE]; |
4e991c23 | 126 | } |
0904200b | 127 | |
ebac4655 JA |
128 | /* |
129 | * if rate blocks is set, sample is running | |
130 | */ | |
4e991c23 | 131 | if (td->rate_bytes || td->rate_blocks) { |
ebac4655 | 132 | spent = mtime_since(&td->lastrate, now); |
2dc1bbeb | 133 | if (spent < td->o.ratecycle) |
ebac4655 JA |
134 | return 0; |
135 | ||
2dc1bbeb | 136 | if (td->o.rate) { |
4e991c23 JA |
137 | /* |
138 | * check bandwidth specified rate | |
139 | */ | |
140 | if (bytes < td->rate_bytes) { | |
2dc1bbeb | 141 | log_err("%s: min rate %u not met\n", td->o.name, td->o.ratemin); |
4e991c23 JA |
142 | return 1; |
143 | } else { | |
144 | rate = (bytes - td->rate_bytes) / spent; | |
2dc1bbeb JA |
145 | if (rate < td->o.ratemin || bytes < td->rate_bytes) { |
146 | log_err("%s: min rate %u not met, got %luKiB/sec\n", td->o.name, td->o.ratemin, rate); | |
4e991c23 JA |
147 | return 1; |
148 | } | |
149 | } | |
413dd459 | 150 | } else { |
4e991c23 JA |
151 | /* |
152 | * checks iops specified rate | |
153 | */ | |
2dc1bbeb JA |
154 | if (iops < td->o.rate_iops) { |
155 | log_err("%s: min iops rate %u not met\n", td->o.name, td->o.rate_iops); | |
413dd459 | 156 | return 1; |
4e991c23 JA |
157 | } else { |
158 | rate = (iops - td->rate_blocks) / spent; | |
2dc1bbeb JA |
159 | if (rate < td->o.rate_iops_min || iops < td->rate_blocks) { |
160 | log_err("%s: min iops rate %u not met, got %lu\n", td->o.name, td->o.rate_iops_min, rate); | |
4e991c23 | 161 | } |
413dd459 | 162 | } |
ebac4655 JA |
163 | } |
164 | } | |
165 | ||
0904200b | 166 | td->rate_bytes = bytes; |
4e991c23 | 167 | td->rate_blocks = iops; |
ebac4655 JA |
168 | memcpy(&td->lastrate, now, sizeof(*now)); |
169 | return 0; | |
170 | } | |
171 | ||
172 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) | |
173 | { | |
2dc1bbeb | 174 | if (!td->o.timeout) |
ebac4655 | 175 | return 0; |
2dc1bbeb | 176 | if (mtime_since(&td->epoch, t) >= td->o.timeout * 1000) |
ebac4655 JA |
177 | return 1; |
178 | ||
179 | return 0; | |
180 | } | |
181 | ||
906c8d75 JA |
182 | /* |
183 | * When job exits, we can cancel the in-flight IO if we are using async | |
184 | * io. Attempt to do so. | |
185 | */ | |
ebac4655 JA |
186 | static void cleanup_pending_aio(struct thread_data *td) |
187 | { | |
ebac4655 | 188 | struct list_head *entry, *n; |
ebac4655 JA |
189 | struct io_u *io_u; |
190 | int r; | |
191 | ||
192 | /* | |
193 | * get immediately available events, if any | |
194 | */ | |
d7762cf8 | 195 | r = io_u_queued_complete(td, 0); |
b2fdda43 JA |
196 | if (r < 0) |
197 | return; | |
ebac4655 JA |
198 | |
199 | /* | |
200 | * now cancel remaining active events | |
201 | */ | |
2866c82d | 202 | if (td->io_ops->cancel) { |
ebac4655 JA |
203 | list_for_each_safe(entry, n, &td->io_u_busylist) { |
204 | io_u = list_entry(entry, struct io_u, list); | |
205 | ||
0c6e7517 JA |
206 | /* |
207 | * if the io_u isn't in flight, then that generally | |
208 | * means someone leaked an io_u. complain but fix | |
209 | * it up, so we don't stall here. | |
210 | */ | |
211 | if ((io_u->flags & IO_U_F_FLIGHT) == 0) { | |
212 | log_err("fio: non-busy IO on busy list\n"); | |
ebac4655 | 213 | put_io_u(td, io_u); |
0c6e7517 JA |
214 | } else { |
215 | r = td->io_ops->cancel(td, io_u); | |
216 | if (!r) | |
217 | put_io_u(td, io_u); | |
218 | } | |
ebac4655 JA |
219 | } |
220 | } | |
221 | ||
97601024 | 222 | if (td->cur_depth) |
d7762cf8 | 223 | r = io_u_queued_complete(td, td->cur_depth); |
ebac4655 JA |
224 | } |
225 | ||
858a3d47 JA |
226 | /* |
227 | * Helper to handle the final sync of a file. Works just like the normal | |
228 | * io path, just does everything sync. | |
229 | */ | |
230 | static int fio_io_sync(struct thread_data *td, struct fio_file *f) | |
231 | { | |
232 | struct io_u *io_u = __get_io_u(td); | |
858a3d47 JA |
233 | int ret; |
234 | ||
235 | if (!io_u) | |
236 | return 1; | |
237 | ||
238 | io_u->ddir = DDIR_SYNC; | |
239 | io_u->file = f; | |
240 | ||
241 | if (td_io_prep(td, io_u)) { | |
242 | put_io_u(td, io_u); | |
243 | return 1; | |
244 | } | |
245 | ||
755200a3 | 246 | requeue: |
858a3d47 | 247 | ret = td_io_queue(td, io_u); |
36167d82 | 248 | if (ret < 0) { |
e1161c32 | 249 | td_verror(td, io_u->error, "td_io_queue"); |
858a3d47 | 250 | put_io_u(td, io_u); |
858a3d47 | 251 | return 1; |
36167d82 | 252 | } else if (ret == FIO_Q_QUEUED) { |
d7762cf8 | 253 | if (io_u_queued_complete(td, 1) < 0) |
36167d82 | 254 | return 1; |
36167d82 JA |
255 | } else if (ret == FIO_Q_COMPLETED) { |
256 | if (io_u->error) { | |
e1161c32 | 257 | td_verror(td, io_u->error, "td_io_queue"); |
36167d82 JA |
258 | return 1; |
259 | } | |
858a3d47 | 260 | |
d7762cf8 | 261 | if (io_u_sync_complete(td, io_u) < 0) |
b2fdda43 | 262 | return 1; |
755200a3 JA |
263 | } else if (ret == FIO_Q_BUSY) { |
264 | if (td_io_commit(td)) | |
265 | return 1; | |
266 | goto requeue; | |
858a3d47 JA |
267 | } |
268 | ||
269 | return 0; | |
270 | } | |
271 | ||
906c8d75 | 272 | /* |
34403fb1 | 273 | * The main verify engine. Runs over the writes we previously submitted, |
906c8d75 JA |
274 | * reads the blocks back in, and checks the crc/md5 of the data. |
275 | */ | |
1e97cce9 | 276 | static void do_verify(struct thread_data *td) |
ebac4655 | 277 | { |
53cdc686 | 278 | struct fio_file *f; |
36167d82 | 279 | struct io_u *io_u; |
af52b345 JA |
280 | int ret, min_events; |
281 | unsigned int i; | |
e5b401d4 JA |
282 | |
283 | /* | |
284 | * sync io first and invalidate cache, to make sure we really | |
285 | * read from disk. | |
286 | */ | |
287 | for_each_file(td, f, i) { | |
3d7b485f JA |
288 | if (!(f->flags & FIO_FILE_OPEN)) |
289 | continue; | |
b2fdda43 JA |
290 | if (fio_io_sync(td, f)) |
291 | break; | |
292 | if (file_invalidate_cache(td, f)) | |
293 | break; | |
e5b401d4 | 294 | } |
ebac4655 | 295 | |
b2fdda43 JA |
296 | if (td->error) |
297 | return; | |
298 | ||
ebac4655 JA |
299 | td_set_runstate(td, TD_VERIFYING); |
300 | ||
36167d82 JA |
301 | io_u = NULL; |
302 | while (!td->terminate) { | |
5451792e JA |
303 | int ret2; |
304 | ||
ebac4655 JA |
305 | io_u = __get_io_u(td); |
306 | if (!io_u) | |
307 | break; | |
308 | ||
069c2918 JA |
309 | if (runtime_exceeded(td, &io_u->start_time)) { |
310 | put_io_u(td, io_u); | |
90a87d4b | 311 | td->terminate = 1; |
02bcaa8c | 312 | break; |
069c2918 | 313 | } |
02bcaa8c | 314 | |
069c2918 JA |
315 | if (get_next_verify(td, io_u)) { |
316 | put_io_u(td, io_u); | |
ebac4655 | 317 | break; |
069c2918 | 318 | } |
ebac4655 | 319 | |
069c2918 JA |
320 | if (td_io_prep(td, io_u)) { |
321 | put_io_u(td, io_u); | |
53cdc686 | 322 | break; |
069c2918 | 323 | } |
d7762cf8 JA |
324 | |
325 | io_u->end_io = verify_io_u; | |
53cdc686 | 326 | |
11786802 | 327 | ret = td_io_queue(td, io_u); |
36167d82 JA |
328 | switch (ret) { |
329 | case FIO_Q_COMPLETED: | |
330 | if (io_u->error) | |
22819ec2 | 331 | ret = -io_u->error; |
9e9d164e | 332 | else if (io_u->resid) { |
36167d82 | 333 | int bytes = io_u->xfer_buflen - io_u->resid; |
d460eb31 | 334 | struct fio_file *f = io_u->file; |
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 | 347 | io_u->offset += bytes; |
d460eb31 | 348 | f->last_completed_pos = io_u->offset; |
8400d9b2 | 349 | |
30061b97 JA |
350 | td->ts.short_io_u[io_u->ddir]++; |
351 | ||
d460eb31 | 352 | if (io_u->offset == f->real_file_size) |
8400d9b2 JA |
353 | goto sync_done; |
354 | ||
11786802 JA |
355 | requeue_io_u(td, &io_u); |
356 | } else { | |
8400d9b2 | 357 | sync_done: |
11786802 JA |
358 | ret = io_u_sync_complete(td, io_u); |
359 | if (ret < 0) | |
360 | break; | |
36167d82 | 361 | } |
36167d82 JA |
362 | continue; |
363 | case FIO_Q_QUEUED: | |
364 | break; | |
755200a3 JA |
365 | case FIO_Q_BUSY: |
366 | requeue_io_u(td, &io_u); | |
5451792e JA |
367 | ret2 = td_io_commit(td); |
368 | if (ret2 < 0) | |
369 | ret = ret2; | |
755200a3 | 370 | break; |
36167d82 JA |
371 | default: |
372 | assert(ret < 0); | |
e1161c32 | 373 | td_verror(td, -ret, "td_io_queue"); |
ebac4655 JA |
374 | break; |
375 | } | |
376 | ||
99784632 | 377 | if (ret < 0 || td->error) |
3af6ef39 JA |
378 | break; |
379 | ||
ebac4655 | 380 | /* |
3af6ef39 JA |
381 | * if we can queue more, do so. but check if there are |
382 | * completed io_u's first. | |
ebac4655 | 383 | */ |
97601024 | 384 | min_events = 0; |
e916b390 | 385 | if (queue_full(td) || ret == FIO_Q_BUSY) { |
3af6ef39 | 386 | min_events = 1; |
3af6ef39 | 387 | |
2dc1bbeb JA |
388 | if (td->cur_depth > td->o.iodepth_low) |
389 | min_events = td->cur_depth - td->o.iodepth_low; | |
e916b390 JA |
390 | } |
391 | ||
3af6ef39 JA |
392 | /* |
393 | * Reap required number of io units, if any, and do the | |
394 | * verification on them through the callback handler | |
395 | */ | |
d7762cf8 | 396 | if (io_u_queued_complete(td, min_events) < 0) |
ebac4655 | 397 | break; |
36167d82 | 398 | } |
ebac4655 | 399 | |
c01c0395 JA |
400 | if (!td->error) { |
401 | min_events = td->cur_depth; | |
402 | ||
403 | if (min_events) | |
404 | ret = io_u_queued_complete(td, min_events); | |
405 | } else | |
ebac4655 JA |
406 | cleanup_pending_aio(td); |
407 | ||
408 | td_set_runstate(td, TD_RUNNING); | |
409 | } | |
410 | ||
32cd46a0 | 411 | /* |
906c8d75 | 412 | * Main IO worker function. It retrieves io_u's to process and queues |
32cd46a0 JA |
413 | * and reaps them, checking for rate and errors along the way. |
414 | */ | |
ebac4655 JA |
415 | static void do_io(struct thread_data *td) |
416 | { | |
02bcaa8c | 417 | struct timeval s; |
ebac4655 | 418 | unsigned long usec; |
af52b345 JA |
419 | unsigned int i; |
420 | int ret = 0; | |
ebac4655 | 421 | |
5853e5a8 JA |
422 | td_set_runstate(td, TD_RUNNING); |
423 | ||
7bb48f84 | 424 | while ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->o.size) { |
97601024 JA |
425 | struct timeval comp_time; |
426 | long bytes_done = 0; | |
84585003 | 427 | int min_evts = 0; |
ebac4655 | 428 | struct io_u *io_u; |
5451792e | 429 | int ret2; |
ebac4655 JA |
430 | |
431 | if (td->terminate) | |
432 | break; | |
433 | ||
3d7c391d | 434 | io_u = get_io_u(td); |
ebac4655 JA |
435 | if (!io_u) |
436 | break; | |
437 | ||
438 | memcpy(&s, &io_u->start_time, sizeof(s)); | |
97601024 JA |
439 | |
440 | if (runtime_exceeded(td, &s)) { | |
441 | put_io_u(td, io_u); | |
90a87d4b | 442 | td->terminate = 1; |
97601024 JA |
443 | break; |
444 | } | |
36167d82 | 445 | |
11786802 | 446 | ret = td_io_queue(td, io_u); |
36167d82 JA |
447 | switch (ret) { |
448 | case FIO_Q_COMPLETED: | |
5451792e JA |
449 | if (io_u->error) |
450 | ret = -io_u->error; | |
9e9d164e | 451 | else if (io_u->resid) { |
36167d82 | 452 | int bytes = io_u->xfer_buflen - io_u->resid; |
d460eb31 | 453 | struct fio_file *f = io_u->file; |
36167d82 | 454 | |
9e9d164e JA |
455 | /* |
456 | * zero read, fail | |
457 | */ | |
458 | if (!bytes) { | |
459 | td_verror(td, ENODATA, "full resid"); | |
460 | put_io_u(td, io_u); | |
461 | break; | |
462 | } | |
463 | ||
cec6b55d | 464 | io_u->xfer_buflen = io_u->resid; |
36167d82 | 465 | io_u->xfer_buf += bytes; |
5a7c5680 | 466 | io_u->offset += bytes; |
d460eb31 | 467 | f->last_completed_pos = io_u->offset; |
5a7c5680 | 468 | |
30061b97 JA |
469 | td->ts.short_io_u[io_u->ddir]++; |
470 | ||
d460eb31 | 471 | if (io_u->offset == f->real_file_size) |
5a7c5680 JA |
472 | goto sync_done; |
473 | ||
11786802 JA |
474 | requeue_io_u(td, &io_u); |
475 | } else { | |
5a7c5680 | 476 | sync_done: |
11786802 JA |
477 | fio_gettime(&comp_time, NULL); |
478 | bytes_done = io_u_sync_complete(td, io_u); | |
479 | if (bytes_done < 0) | |
480 | ret = bytes_done; | |
cec6b55d | 481 | } |
36167d82 JA |
482 | break; |
483 | case FIO_Q_QUEUED: | |
7e77dd02 JA |
484 | /* |
485 | * if the engine doesn't have a commit hook, | |
486 | * the io_u is really queued. if it does have such | |
487 | * a hook, it has to call io_u_queued() itself. | |
488 | */ | |
489 | if (td->io_ops->commit == NULL) | |
490 | io_u_queued(td, io_u); | |
36167d82 | 491 | break; |
755200a3 JA |
492 | case FIO_Q_BUSY: |
493 | requeue_io_u(td, &io_u); | |
5451792e JA |
494 | ret2 = td_io_commit(td); |
495 | if (ret2 < 0) | |
496 | ret = ret2; | |
755200a3 | 497 | break; |
36167d82 JA |
498 | default: |
499 | assert(ret < 0); | |
500 | put_io_u(td, io_u); | |
501 | break; | |
ebac4655 JA |
502 | } |
503 | ||
99784632 | 504 | if (ret < 0 || td->error) |
36167d82 JA |
505 | break; |
506 | ||
97601024 JA |
507 | /* |
508 | * See if we need to complete some commands | |
509 | */ | |
755200a3 | 510 | if (ret == FIO_Q_QUEUED || ret == FIO_Q_BUSY) { |
97601024 | 511 | min_evts = 0; |
e916b390 | 512 | if (queue_full(td) || ret == FIO_Q_BUSY) { |
36167d82 | 513 | min_evts = 1; |
ebac4655 | 514 | |
2dc1bbeb JA |
515 | if (td->cur_depth > td->o.iodepth_low) |
516 | min_evts = td->cur_depth - td->o.iodepth_low; | |
e916b390 JA |
517 | } |
518 | ||
97601024 | 519 | fio_gettime(&comp_time, NULL); |
d7762cf8 | 520 | bytes_done = io_u_queued_complete(td, min_evts); |
97601024 | 521 | if (bytes_done < 0) |
36167d82 | 522 | break; |
ebac4655 JA |
523 | } |
524 | ||
97601024 JA |
525 | if (!bytes_done) |
526 | continue; | |
527 | ||
ebac4655 JA |
528 | /* |
529 | * the rate is batched for now, it should work for batches | |
530 | * of completions except the very first one which may look | |
531 | * a little bursty | |
532 | */ | |
97601024 | 533 | usec = utime_since(&s, &comp_time); |
ebac4655 | 534 | |
413dd459 | 535 | rate_throttle(td, usec, bytes_done); |
ebac4655 | 536 | |
97601024 | 537 | if (check_min_rate(td, &comp_time)) { |
98aa62d8 | 538 | if (exitall_on_terminate) |
390c40e2 | 539 | terminate_threads(td->groupid); |
e1161c32 | 540 | td_verror(td, ENODATA, "check_min_rate"); |
ebac4655 JA |
541 | break; |
542 | } | |
543 | ||
2dc1bbeb | 544 | if (td->o.thinktime) { |
9c1f7434 JA |
545 | unsigned long long b; |
546 | ||
547 | b = td->io_blocks[0] + td->io_blocks[1]; | |
2dc1bbeb | 548 | if (!(b % td->o.thinktime_blocks)) { |
48097d5c JA |
549 | int left; |
550 | ||
2dc1bbeb JA |
551 | if (td->o.thinktime_spin) |
552 | __usec_sleep(td->o.thinktime_spin); | |
48097d5c | 553 | |
2dc1bbeb | 554 | left = td->o.thinktime - td->o.thinktime_spin; |
48097d5c JA |
555 | if (left) |
556 | usec_sleep(td, left); | |
557 | } | |
9c1f7434 | 558 | } |
ebac4655 JA |
559 | } |
560 | ||
4d2413c6 | 561 | if (!td->error) { |
3d7c391d JA |
562 | struct fio_file *f; |
563 | ||
c01c0395 JA |
564 | i = td->cur_depth; |
565 | if (i) | |
566 | ret = io_u_queued_complete(td, i); | |
ebac4655 | 567 | |
2dc1bbeb | 568 | if (should_fsync(td) && td->o.end_fsync) { |
84585003 | 569 | td_set_runstate(td, TD_FSYNCING); |
3d7b485f JA |
570 | |
571 | for_each_file(td, f, i) { | |
572 | if (!(f->flags & FIO_FILE_OPEN)) | |
573 | continue; | |
858a3d47 | 574 | fio_io_sync(td, f); |
3d7b485f | 575 | } |
84585003 | 576 | } |
c01c0395 JA |
577 | } else |
578 | cleanup_pending_aio(td); | |
ebac4655 JA |
579 | } |
580 | ||
ebac4655 JA |
581 | static void cleanup_io_u(struct thread_data *td) |
582 | { | |
583 | struct list_head *entry, *n; | |
584 | struct io_u *io_u; | |
585 | ||
586 | list_for_each_safe(entry, n, &td->io_u_freelist) { | |
587 | io_u = list_entry(entry, struct io_u, list); | |
588 | ||
589 | list_del(&io_u->list); | |
590 | free(io_u); | |
591 | } | |
592 | ||
2f9ade3c | 593 | free_io_mem(td); |
ebac4655 JA |
594 | } |
595 | ||
6b9cea23 JA |
596 | /* |
597 | * "randomly" fill the buffer contents | |
598 | */ | |
e9459e5a | 599 | static void fill_io_buf(struct thread_data *td, struct io_u *io_u, int max_bs) |
6b9cea23 | 600 | { |
210f43b3 | 601 | long *ptr = io_u->buf; |
6b9cea23 | 602 | |
e9459e5a JA |
603 | if (!td->o.zero_buffers) { |
604 | while ((void *) ptr - io_u->buf < max_bs) { | |
210f43b3 | 605 | *ptr = rand() * GOLDEN_RATIO_PRIME; |
e9459e5a JA |
606 | ptr++; |
607 | } | |
608 | } else | |
609 | memset(ptr, 0, max_bs); | |
6b9cea23 JA |
610 | } |
611 | ||
ebac4655 JA |
612 | static int init_io_u(struct thread_data *td) |
613 | { | |
2b4ce34f | 614 | unsigned long long buf_size; |
ebac4655 | 615 | struct io_u *io_u; |
a00735e6 | 616 | unsigned int max_bs; |
ebac4655 JA |
617 | int i, max_units; |
618 | char *p; | |
619 | ||
2866c82d | 620 | if (td->io_ops->flags & FIO_SYNCIO) |
ebac4655 JA |
621 | max_units = 1; |
622 | else | |
2dc1bbeb | 623 | max_units = td->o.iodepth; |
ebac4655 | 624 | |
2dc1bbeb | 625 | max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]); |
2b4ce34f JA |
626 | buf_size = (unsigned long long) max_bs * (unsigned long long) max_units; |
627 | buf_size += page_mask; | |
628 | if (buf_size != (size_t) buf_size) { | |
629 | log_err("fio: IO memory too large. Reduce max_bs or iodepth\n"); | |
630 | return 1; | |
631 | } | |
632 | ||
633 | td->orig_buffer_size = buf_size; | |
74b025b0 | 634 | |
2dc1bbeb JA |
635 | if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) |
636 | td->orig_buffer_size = (td->orig_buffer_size + td->o.hugepage_size - 1) & ~(td->o.hugepage_size - 1); | |
4afbf66f JA |
637 | else if (td->orig_buffer_size & page_mask) |
638 | td->orig_buffer_size = (td->orig_buffer_size + page_mask) & ~page_mask; | |
ebac4655 | 639 | |
2f9ade3c JA |
640 | if (allocate_io_mem(td)) |
641 | return 1; | |
ebac4655 | 642 | |
ebac4655 JA |
643 | p = ALIGN(td->orig_buffer); |
644 | for (i = 0; i < max_units; i++) { | |
645 | io_u = malloc(sizeof(*io_u)); | |
646 | memset(io_u, 0, sizeof(*io_u)); | |
647 | INIT_LIST_HEAD(&io_u->list); | |
648 | ||
a00735e6 | 649 | io_u->buf = p + max_bs * i; |
e9459e5a | 650 | |
660a1cb5 | 651 | if (td_write(td)) |
e9459e5a | 652 | fill_io_buf(td, io_u, max_bs); |
6b9cea23 | 653 | |
b1ff3403 | 654 | io_u->index = i; |
0c6e7517 | 655 | io_u->flags = IO_U_F_FREE; |
ebac4655 JA |
656 | list_add(&io_u->list, &td->io_u_freelist); |
657 | } | |
658 | ||
433afcb4 JA |
659 | io_u_init_timeout(); |
660 | ||
ebac4655 JA |
661 | return 0; |
662 | } | |
663 | ||
da86774e JA |
664 | static int switch_ioscheduler(struct thread_data *td) |
665 | { | |
666 | char tmp[256], tmp2[128]; | |
667 | FILE *f; | |
668 | int ret; | |
669 | ||
ba0fbe10 | 670 | if (td->io_ops->flags & FIO_DISKLESSIO) |
f48b467c JA |
671 | return 0; |
672 | ||
da86774e JA |
673 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); |
674 | ||
675 | f = fopen(tmp, "r+"); | |
676 | if (!f) { | |
cbf5c1d7 JA |
677 | if (errno == ENOENT) { |
678 | log_err("fio: os or kernel doesn't support IO scheduler switching\n"); | |
679 | return 0; | |
680 | } | |
681 | td_verror(td, errno, "fopen iosched"); | |
da86774e JA |
682 | return 1; |
683 | } | |
684 | ||
685 | /* | |
686 | * Set io scheduler. | |
687 | */ | |
2dc1bbeb | 688 | ret = fwrite(td->o.ioscheduler, strlen(td->o.ioscheduler), 1, f); |
da86774e | 689 | if (ferror(f) || ret != 1) { |
e1161c32 | 690 | td_verror(td, errno, "fwrite"); |
da86774e JA |
691 | fclose(f); |
692 | return 1; | |
693 | } | |
694 | ||
695 | rewind(f); | |
696 | ||
697 | /* | |
698 | * Read back and check that the selected scheduler is now the default. | |
699 | */ | |
700 | ret = fread(tmp, 1, sizeof(tmp), f); | |
701 | if (ferror(f) || ret < 0) { | |
e1161c32 | 702 | td_verror(td, errno, "fread"); |
da86774e JA |
703 | fclose(f); |
704 | return 1; | |
705 | } | |
706 | ||
2dc1bbeb | 707 | sprintf(tmp2, "[%s]", td->o.ioscheduler); |
da86774e | 708 | if (!strstr(tmp, tmp2)) { |
2dc1bbeb | 709 | log_err("fio: io scheduler %s not found\n", td->o.ioscheduler); |
e1161c32 | 710 | td_verror(td, EINVAL, "iosched_switch"); |
da86774e JA |
711 | fclose(f); |
712 | return 1; | |
713 | } | |
714 | ||
715 | fclose(f); | |
716 | return 0; | |
717 | } | |
718 | ||
a978ba68 | 719 | static int clear_io_state(struct thread_data *td) |
ebac4655 | 720 | { |
53cdc686 | 721 | struct fio_file *f; |
af52b345 JA |
722 | unsigned int i; |
723 | int ret; | |
ebac4655 | 724 | |
756867bd | 725 | td->ts.stat_io_bytes[0] = td->ts.stat_io_bytes[1] = 0; |
ebac4655 | 726 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; |
20dc95c4 | 727 | td->zone_bytes = 0; |
d7d3b49b | 728 | td->rate_bytes = 0; |
4e991c23 | 729 | td->rate_blocks = 0; |
38d77cae | 730 | td->rw_end_set[0] = td->rw_end_set[1] = 0; |
ebac4655 | 731 | |
c1324df1 JA |
732 | td->last_was_sync = 0; |
733 | ||
a978ba68 JA |
734 | for_each_file(td, f, i) |
735 | td_io_close_file(td, f); | |
53cdc686 | 736 | |
a978ba68 JA |
737 | ret = 0; |
738 | for_each_file(td, f, i) { | |
739 | ret = td_io_open_file(td, f); | |
740 | if (ret) | |
741 | break; | |
53cdc686 | 742 | } |
a978ba68 JA |
743 | |
744 | return ret; | |
ebac4655 JA |
745 | } |
746 | ||
906c8d75 JA |
747 | /* |
748 | * Entry point for the thread based jobs. The process based jobs end up | |
749 | * here as well, after a little setup. | |
750 | */ | |
ebac4655 JA |
751 | static void *thread_main(void *data) |
752 | { | |
69008999 | 753 | unsigned long long runtime[2]; |
ebac4655 | 754 | struct thread_data *td = data; |
38d77cae | 755 | unsigned long elapsed; |
a978ba68 | 756 | int clear_state; |
ebac4655 | 757 | |
2dc1bbeb | 758 | if (!td->o.use_thread) |
ebac4655 JA |
759 | setsid(); |
760 | ||
761 | td->pid = getpid(); | |
762 | ||
aea47d44 JA |
763 | INIT_LIST_HEAD(&td->io_u_freelist); |
764 | INIT_LIST_HEAD(&td->io_u_busylist); | |
755200a3 | 765 | INIT_LIST_HEAD(&td->io_u_requeues); |
aea47d44 | 766 | INIT_LIST_HEAD(&td->io_log_list); |
8de8f047 | 767 | INIT_LIST_HEAD(&td->io_hist_list); |
bb5d7d0b | 768 | td->io_hist_tree = RB_ROOT; |
aea47d44 | 769 | |
ebac4655 | 770 | if (init_io_u(td)) |
bda4fd9e | 771 | goto err_sem; |
ebac4655 JA |
772 | |
773 | if (fio_setaffinity(td) == -1) { | |
e1161c32 | 774 | td_verror(td, errno, "cpu_set_affinity"); |
bda4fd9e | 775 | goto err_sem; |
ebac4655 JA |
776 | } |
777 | ||
aea47d44 | 778 | if (init_iolog(td)) |
bda4fd9e | 779 | goto err_sem; |
aea47d44 | 780 | |
ebac4655 JA |
781 | if (td->ioprio) { |
782 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { | |
e1161c32 | 783 | td_verror(td, errno, "ioprio_set"); |
bda4fd9e | 784 | goto err_sem; |
ebac4655 JA |
785 | } |
786 | } | |
787 | ||
2dc1bbeb | 788 | if (nice(td->o.nice) == -1) { |
e1161c32 | 789 | td_verror(td, errno, "nice"); |
bda4fd9e | 790 | goto err_sem; |
b6f4d880 JA |
791 | } |
792 | ||
2dc1bbeb | 793 | if (td->o.ioscheduler && switch_ioscheduler(td)) |
bda4fd9e | 794 | goto err_sem; |
da86774e | 795 | |
75154845 | 796 | td_set_runstate(td, TD_INITIALIZED); |
07739b57 JA |
797 | fio_sem_up(startup_sem); |
798 | fio_sem_down(td->mutex); | |
ebac4655 | 799 | |
37f56873 JA |
800 | /* |
801 | * the ->mutex semaphore is now no longer used, close it to avoid | |
802 | * eating a file descriptor | |
803 | */ | |
804 | fio_sem_remove(td->mutex); | |
805 | ||
2dc1bbeb | 806 | if (!td->o.create_serialize && setup_files(td)) |
ebac4655 JA |
807 | goto err; |
808 | ||
7d6c5283 JA |
809 | if (td_io_init(td)) |
810 | goto err; | |
811 | ||
b5af8293 JA |
812 | if (open_files(td)) |
813 | goto err; | |
814 | ||
68727076 JA |
815 | if (init_random_map(td)) |
816 | goto err; | |
817 | ||
2dc1bbeb JA |
818 | if (td->o.exec_prerun) { |
819 | if (system(td->o.exec_prerun) < 0) | |
69cfd7e0 JA |
820 | goto err; |
821 | } | |
4e0ba8af | 822 | |
69008999 | 823 | fio_gettime(&td->epoch, NULL); |
433afcb4 | 824 | memcpy(&td->timeout_end, &td->epoch, sizeof(td->epoch)); |
756867bd | 825 | getrusage(RUSAGE_SELF, &td->ts.ru_start); |
69008999 JA |
826 | |
827 | runtime[0] = runtime[1] = 0; | |
a978ba68 | 828 | clear_state = 0; |
cf4464ca | 829 | while (td->o.time_based || td->o.loops--) { |
02bcaa8c | 830 | fio_gettime(&td->start, NULL); |
756867bd | 831 | memcpy(&td->ts.stat_sample_time, &td->start, sizeof(td->start)); |
ebac4655 | 832 | |
2dc1bbeb | 833 | if (td->o.ratemin) |
756867bd | 834 | memcpy(&td->lastrate, &td->ts.stat_sample_time, sizeof(td->lastrate)); |
ebac4655 | 835 | |
a978ba68 JA |
836 | if (clear_state && clear_io_state(td)) |
837 | break; | |
838 | ||
ebac4655 JA |
839 | prune_io_piece_log(td); |
840 | ||
ba0fbe10 | 841 | do_io(td); |
ebac4655 | 842 | |
a978ba68 JA |
843 | clear_state = 1; |
844 | ||
38d77cae JA |
845 | if (td_read(td) && td->io_bytes[DDIR_READ]) { |
846 | if (td->rw_end_set[DDIR_READ]) | |
847 | elapsed = utime_since(&td->start, &td->rw_end[DDIR_READ]); | |
848 | else | |
849 | elapsed = utime_since_now(&td->start); | |
850 | ||
851 | runtime[DDIR_READ] += elapsed; | |
852 | } | |
853 | if (td_write(td) && td->io_bytes[DDIR_WRITE]) { | |
854 | if (td->rw_end_set[DDIR_WRITE]) | |
855 | elapsed = utime_since(&td->start, &td->rw_end[DDIR_WRITE]); | |
856 | else | |
857 | elapsed = utime_since_now(&td->start); | |
858 | ||
859 | runtime[DDIR_WRITE] += elapsed; | |
860 | } | |
413dd459 | 861 | |
ebac4655 JA |
862 | if (td->error || td->terminate) |
863 | break; | |
864 | ||
2dc1bbeb | 865 | if (td->o.verify == VERIFY_NONE) |
ebac4655 JA |
866 | continue; |
867 | ||
a978ba68 JA |
868 | if (clear_io_state(td)) |
869 | break; | |
870 | ||
02bcaa8c | 871 | fio_gettime(&td->start, NULL); |
ebac4655 JA |
872 | |
873 | do_verify(td); | |
874 | ||
69008999 | 875 | runtime[DDIR_READ] += utime_since_now(&td->start); |
ebac4655 JA |
876 | |
877 | if (td->error || td->terminate) | |
878 | break; | |
879 | } | |
880 | ||
36dff966 | 881 | update_rusage_stat(td); |
756867bd JA |
882 | td->ts.runtime[0] = runtime[0] / 1000; |
883 | td->ts.runtime[1] = runtime[1] / 1000; | |
884 | td->ts.total_run_time = mtime_since_now(&td->epoch); | |
885 | td->ts.io_bytes[0] = td->io_bytes[0]; | |
886 | td->ts.io_bytes[1] = td->io_bytes[1]; | |
887 | ||
888 | if (td->ts.bw_log) | |
889 | finish_log(td, td->ts.bw_log, "bw"); | |
890 | if (td->ts.slat_log) | |
891 | finish_log(td, td->ts.slat_log, "slat"); | |
892 | if (td->ts.clat_log) | |
893 | finish_log(td, td->ts.clat_log, "clat"); | |
2dc1bbeb | 894 | if (td->o.write_iolog_file) |
843a7413 | 895 | write_iolog_close(td); |
2dc1bbeb JA |
896 | if (td->o.exec_postrun) { |
897 | if (system(td->o.exec_postrun) < 0) | |
898 | log_err("fio: postrun %s failed\n", td->o.exec_postrun); | |
69cfd7e0 | 899 | } |
ebac4655 JA |
900 | |
901 | if (exitall_on_terminate) | |
390c40e2 | 902 | terminate_threads(td->groupid); |
ebac4655 JA |
903 | |
904 | err: | |
5bf13a5a JA |
905 | if (td->error) |
906 | printf("fio: pid=%d, err=%d/%s\n", td->pid, td->error, td->verror); | |
53cdc686 | 907 | close_files(td); |
2866c82d | 908 | close_ioengine(td); |
ebac4655 | 909 | cleanup_io_u(td); |
d23bb327 | 910 | options_mem_free(td); |
ebac4655 | 911 | td_set_runstate(td, TD_EXITED); |
43d76807 | 912 | return (void *) (unsigned long) td->error; |
bda4fd9e JA |
913 | err_sem: |
914 | fio_sem_up(startup_sem); | |
915 | goto err; | |
ebac4655 JA |
916 | } |
917 | ||
906c8d75 JA |
918 | /* |
919 | * We cannot pass the td data into a forked process, so attach the td and | |
920 | * pass it to the thread worker. | |
921 | */ | |
a6418147 | 922 | static int fork_main(int shmid, int offset) |
ebac4655 JA |
923 | { |
924 | struct thread_data *td; | |
a6418147 | 925 | void *data, *ret; |
ebac4655 JA |
926 | |
927 | data = shmat(shmid, NULL, 0); | |
928 | if (data == (void *) -1) { | |
a6418147 JA |
929 | int __err = errno; |
930 | ||
ebac4655 | 931 | perror("shmat"); |
a6418147 | 932 | return __err; |
ebac4655 JA |
933 | } |
934 | ||
935 | td = data + offset * sizeof(struct thread_data); | |
a6418147 | 936 | ret = thread_main(td); |
ebac4655 | 937 | shmdt(data); |
43d76807 | 938 | return (int) (unsigned long) ret; |
ebac4655 JA |
939 | } |
940 | ||
906c8d75 JA |
941 | /* |
942 | * Run over the job map and reap the threads that have exited, if any. | |
943 | */ | |
ebac4655 JA |
944 | static void reap_threads(int *nr_running, int *t_rate, int *m_rate) |
945 | { | |
34572e28 | 946 | struct thread_data *td; |
fab6aa71 | 947 | int i, cputhreads, pending, status, ret; |
ebac4655 JA |
948 | |
949 | /* | |
950 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
951 | */ | |
4d2413c6 | 952 | pending = cputhreads = 0; |
34572e28 | 953 | for_each_td(td, i) { |
3707f45b | 954 | int flags = 0; |
a2f77c9f | 955 | |
84585003 JA |
956 | /* |
957 | * ->io_ops is NULL for a thread that has closed its | |
958 | * io engine | |
959 | */ | |
ba0fbe10 | 960 | if (td->io_ops && !strcmp(td->io_ops->name, "cpuio")) |
b990b5c0 JA |
961 | cputhreads++; |
962 | ||
a2f77c9f JA |
963 | if (!td->pid || td->runstate == TD_REAPED) |
964 | continue; | |
2dc1bbeb | 965 | if (td->o.use_thread) { |
3707f45b JA |
966 | if (td->runstate == TD_EXITED) { |
967 | td_set_runstate(td, TD_REAPED); | |
968 | goto reaped; | |
969 | } | |
970 | continue; | |
971 | } | |
a2f77c9f JA |
972 | |
973 | flags = WNOHANG; | |
974 | if (td->runstate == TD_EXITED) | |
975 | flags = 0; | |
976 | ||
977 | /* | |
978 | * check if someone quit or got killed in an unusual way | |
979 | */ | |
980 | ret = waitpid(td->pid, &status, flags); | |
3707f45b | 981 | if (ret < 0) { |
a2f77c9f JA |
982 | if (errno == ECHILD) { |
983 | log_err("fio: pid=%d disappeared %d\n", td->pid, td->runstate); | |
984 | td_set_runstate(td, TD_REAPED); | |
985 | goto reaped; | |
986 | } | |
987 | perror("waitpid"); | |
988 | } else if (ret == td->pid) { | |
989 | if (WIFSIGNALED(status)) { | |
fab6aa71 JA |
990 | int sig = WTERMSIG(status); |
991 | ||
d9244941 JA |
992 | if (sig != SIGQUIT) |
993 | log_err("fio: pid=%d, got signal=%d\n", td->pid, sig); | |
fab6aa71 JA |
994 | td_set_runstate(td, TD_REAPED); |
995 | goto reaped; | |
996 | } | |
a2f77c9f JA |
997 | if (WIFEXITED(status)) { |
998 | if (WEXITSTATUS(status) && !td->error) | |
999 | td->error = WEXITSTATUS(status); | |
a2f77c9f | 1000 | |
a2f77c9f JA |
1001 | td_set_runstate(td, TD_REAPED); |
1002 | goto reaped; | |
a6418147 JA |
1003 | } |
1004 | } | |
ebac4655 | 1005 | |
a2f77c9f JA |
1006 | /* |
1007 | * thread is not dead, continue | |
1008 | */ | |
e48676ba | 1009 | pending++; |
a2f77c9f | 1010 | continue; |
fab6aa71 | 1011 | reaped: |
2dc1bbeb | 1012 | if (td->o.use_thread) { |
3707f45b JA |
1013 | long ret; |
1014 | ||
1015 | if (pthread_join(td->thread, (void *) &ret)) | |
1016 | perror("pthread_join"); | |
1017 | } | |
1018 | ||
ebac4655 | 1019 | (*nr_running)--; |
2dc1bbeb JA |
1020 | (*m_rate) -= td->o.ratemin; |
1021 | (*t_rate) -= td->o.rate; | |
e48676ba | 1022 | pending--; |
a2f77c9f JA |
1023 | |
1024 | if (td->error) | |
1025 | exit_value++; | |
ebac4655 | 1026 | } |
b990b5c0 | 1027 | |
4d2413c6 | 1028 | if (*nr_running == cputhreads && !pending) |
390c40e2 | 1029 | terminate_threads(TERMINATE_ALL); |
ebac4655 JA |
1030 | } |
1031 | ||
906c8d75 JA |
1032 | /* |
1033 | * Main function for kicking off and reaping jobs, as needed. | |
1034 | */ | |
ebac4655 JA |
1035 | static void run_threads(void) |
1036 | { | |
ebac4655 JA |
1037 | struct thread_data *td; |
1038 | unsigned long spent; | |
1039 | int i, todo, nr_running, m_rate, t_rate, nr_started; | |
fcb6ade2 | 1040 | |
2f9ade3c JA |
1041 | if (fio_pin_memory()) |
1042 | return; | |
ebac4655 | 1043 | |
c6ae0a5b | 1044 | if (!terse_output) { |
9cedf167 JA |
1045 | printf("Starting "); |
1046 | if (nr_thread) | |
1047 | printf("%d thread%s", nr_thread, nr_thread > 1 ? "s" : ""); | |
1048 | if (nr_process) { | |
1049 | if (nr_thread) | |
1050 | printf(" and "); | |
1051 | printf("%d process%s", nr_process, nr_process > 1 ? "es" : ""); | |
1052 | } | |
1053 | printf("\n"); | |
c6ae0a5b JA |
1054 | fflush(stdout); |
1055 | } | |
c04f7ec3 | 1056 | |
4efa970e JA |
1057 | signal(SIGINT, sig_handler); |
1058 | signal(SIGALRM, sig_handler); | |
1059 | ||
ebac4655 JA |
1060 | todo = thread_number; |
1061 | nr_running = 0; | |
1062 | nr_started = 0; | |
1063 | m_rate = t_rate = 0; | |
1064 | ||
34572e28 | 1065 | for_each_td(td, i) { |
263e529f | 1066 | print_status_init(td->thread_number - 1); |
ebac4655 | 1067 | |
2dc1bbeb | 1068 | if (!td->o.create_serialize) { |
380cf265 | 1069 | init_disk_util(td); |
ebac4655 | 1070 | continue; |
380cf265 | 1071 | } |
ebac4655 JA |
1072 | |
1073 | /* | |
1074 | * do file setup here so it happens sequentially, | |
1075 | * we don't want X number of threads getting their | |
1076 | * client data interspersed on disk | |
1077 | */ | |
53cdc686 | 1078 | if (setup_files(td)) { |
5bf13a5a JA |
1079 | exit_value++; |
1080 | if (td->error) | |
1081 | log_err("fio: pid=%d, err=%d/%s\n", td->pid, td->error, td->verror); | |
ebac4655 JA |
1082 | td_set_runstate(td, TD_REAPED); |
1083 | todo--; | |
1084 | } | |
380cf265 JA |
1085 | |
1086 | init_disk_util(td); | |
ebac4655 JA |
1087 | } |
1088 | ||
a2f77c9f JA |
1089 | set_genesis_time(); |
1090 | ||
ebac4655 | 1091 | while (todo) { |
75154845 JA |
1092 | struct thread_data *map[MAX_JOBS]; |
1093 | struct timeval this_start; | |
1094 | int this_jobs = 0, left; | |
1095 | ||
ebac4655 JA |
1096 | /* |
1097 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
1098 | */ | |
34572e28 | 1099 | for_each_td(td, i) { |
ebac4655 JA |
1100 | if (td->runstate != TD_NOT_CREATED) |
1101 | continue; | |
1102 | ||
1103 | /* | |
1104 | * never got a chance to start, killed by other | |
1105 | * thread for some reason | |
1106 | */ | |
1107 | if (td->terminate) { | |
1108 | todo--; | |
1109 | continue; | |
1110 | } | |
1111 | ||
2dc1bbeb | 1112 | if (td->o.start_delay) { |
263e529f | 1113 | spent = mtime_since_genesis(); |
ebac4655 | 1114 | |
2dc1bbeb | 1115 | if (td->o.start_delay * 1000 > spent) |
ebac4655 JA |
1116 | continue; |
1117 | } | |
1118 | ||
2dc1bbeb | 1119 | if (td->o.stonewall && (nr_started || nr_running)) |
ebac4655 JA |
1120 | break; |
1121 | ||
75154845 JA |
1122 | /* |
1123 | * Set state to created. Thread will transition | |
1124 | * to TD_INITIALIZED when it's done setting up. | |
1125 | */ | |
ebac4655 | 1126 | td_set_runstate(td, TD_CREATED); |
75154845 | 1127 | map[this_jobs++] = td; |
ebac4655 JA |
1128 | nr_started++; |
1129 | ||
2dc1bbeb | 1130 | if (td->o.use_thread) { |
ebac4655 JA |
1131 | if (pthread_create(&td->thread, NULL, thread_main, td)) { |
1132 | perror("thread_create"); | |
1133 | nr_started--; | |
c8bb6faf | 1134 | break; |
ebac4655 JA |
1135 | } |
1136 | } else { | |
07739b57 | 1137 | if (!fork()) { |
a6418147 JA |
1138 | int ret = fork_main(shm_id, i); |
1139 | ||
1140 | exit(ret); | |
ebac4655 JA |
1141 | } |
1142 | } | |
07739b57 | 1143 | fio_sem_down(startup_sem); |
ebac4655 JA |
1144 | } |
1145 | ||
1146 | /* | |
75154845 JA |
1147 | * Wait for the started threads to transition to |
1148 | * TD_INITIALIZED. | |
ebac4655 | 1149 | */ |
02bcaa8c | 1150 | fio_gettime(&this_start, NULL); |
75154845 | 1151 | left = this_jobs; |
6ce15a32 | 1152 | while (left && !fio_abort) { |
75154845 JA |
1153 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) |
1154 | break; | |
1155 | ||
1156 | usleep(100000); | |
1157 | ||
1158 | for (i = 0; i < this_jobs; i++) { | |
1159 | td = map[i]; | |
1160 | if (!td) | |
1161 | continue; | |
b6f4d880 | 1162 | if (td->runstate == TD_INITIALIZED) { |
75154845 JA |
1163 | map[i] = NULL; |
1164 | left--; | |
b6f4d880 JA |
1165 | } else if (td->runstate >= TD_EXITED) { |
1166 | map[i] = NULL; | |
1167 | left--; | |
1168 | todo--; | |
1169 | nr_running++; /* work-around... */ | |
75154845 JA |
1170 | } |
1171 | } | |
1172 | } | |
1173 | ||
1174 | if (left) { | |
3b70d7e5 | 1175 | log_err("fio: %d jobs failed to start\n", left); |
75154845 JA |
1176 | for (i = 0; i < this_jobs; i++) { |
1177 | td = map[i]; | |
1178 | if (!td) | |
1179 | continue; | |
1180 | kill(td->pid, SIGTERM); | |
1181 | } | |
1182 | break; | |
1183 | } | |
1184 | ||
1185 | /* | |
b6f4d880 | 1186 | * start created threads (TD_INITIALIZED -> TD_RUNNING). |
75154845 | 1187 | */ |
34572e28 | 1188 | for_each_td(td, i) { |
75154845 | 1189 | if (td->runstate != TD_INITIALIZED) |
ebac4655 JA |
1190 | continue; |
1191 | ||
1192 | td_set_runstate(td, TD_RUNNING); | |
1193 | nr_running++; | |
1194 | nr_started--; | |
2dc1bbeb JA |
1195 | m_rate += td->o.ratemin; |
1196 | t_rate += td->o.rate; | |
75154845 | 1197 | todo--; |
07739b57 | 1198 | fio_sem_up(td->mutex); |
ebac4655 JA |
1199 | } |
1200 | ||
1201 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1202 | ||
1203 | if (todo) | |
1204 | usleep(100000); | |
1205 | } | |
1206 | ||
1207 | while (nr_running) { | |
1208 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1209 | usleep(10000); | |
1210 | } | |
1211 | ||
1212 | update_io_ticks(); | |
2f9ade3c | 1213 | fio_unpin_memory(); |
ebac4655 JA |
1214 | } |
1215 | ||
ebac4655 JA |
1216 | int main(int argc, char *argv[]) |
1217 | { | |
29d610e1 JA |
1218 | long ps; |
1219 | ||
dbe1125e JA |
1220 | /* |
1221 | * We need locale for number printing, if it isn't set then just | |
1222 | * go with the US format. | |
1223 | */ | |
1224 | if (!getenv("LC_NUMERIC")) | |
1225 | setlocale(LC_NUMERIC, "en_US"); | |
1226 | ||
ebac4655 JA |
1227 | if (parse_options(argc, argv)) |
1228 | return 1; | |
1229 | ||
4b472fa3 JA |
1230 | if (!thread_number) |
1231 | return 0; | |
ebac4655 | 1232 | |
29d610e1 JA |
1233 | ps = sysconf(_SC_PAGESIZE); |
1234 | if (ps < 0) { | |
1235 | log_err("Failed to get page size\n"); | |
1236 | return 1; | |
1237 | } | |
1238 | ||
cfc99db7 | 1239 | page_size = ps; |
29d610e1 JA |
1240 | page_mask = ps - 1; |
1241 | ||
bb3884d8 JA |
1242 | if (write_bw_log) { |
1243 | setup_log(&agg_io_log[DDIR_READ]); | |
1244 | setup_log(&agg_io_log[DDIR_WRITE]); | |
1245 | } | |
1246 | ||
07739b57 JA |
1247 | startup_sem = fio_sem_init(0); |
1248 | ||
a2f77c9f JA |
1249 | set_genesis_time(); |
1250 | ||
ebac4655 JA |
1251 | disk_util_timer_arm(); |
1252 | ||
1253 | run_threads(); | |
6ce15a32 | 1254 | |
bb3884d8 | 1255 | if (!fio_abort) { |
6ce15a32 | 1256 | show_run_stats(); |
bb3884d8 JA |
1257 | if (write_bw_log) { |
1258 | __finish_log(agg_io_log[DDIR_READ],"agg-read_bw.log"); | |
1259 | __finish_log(agg_io_log[DDIR_WRITE],"agg-write_bw.log"); | |
1260 | } | |
1261 | } | |
ebac4655 | 1262 | |
07739b57 | 1263 | fio_sem_remove(startup_sem); |
437c9b71 | 1264 | return exit_value; |
ebac4655 | 1265 | } |