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" |
2e5cdb11 | 39 | #include "smalloc.h" |
4f5af7b2 | 40 | #include "verify.h" |
7c9b1bce | 41 | #include "diskutil.h" |
a696fa2a | 42 | #include "cgroup.h" |
ebac4655 | 43 | |
cfc99db7 JA |
44 | unsigned long page_mask; |
45 | unsigned long page_size; | |
d529ee19 JA |
46 | |
47 | #define PAGE_ALIGN(buf) \ | |
29d610e1 | 48 | (char *) (((unsigned long) (buf) + page_mask) & ~page_mask) |
ebac4655 JA |
49 | |
50 | int groupid = 0; | |
51 | int thread_number = 0; | |
9cedf167 JA |
52 | int nr_process = 0; |
53 | int nr_thread = 0; | |
ebac4655 | 54 | int shm_id = 0; |
53cdc686 | 55 | int temp_stall_ts; |
d3eeeabc | 56 | unsigned long done_secs = 0; |
ebac4655 | 57 | |
cdd18ad8 | 58 | static struct fio_mutex *startup_mutex; |
d2bbb8d9 | 59 | static struct fio_mutex *writeout_mutex; |
6ce15a32 | 60 | static volatile int fio_abort; |
437c9b71 | 61 | static int exit_value; |
ccb0fa24 | 62 | static struct itimerval itimer; |
be4ecfdf | 63 | static pthread_t gtod_thread; |
ebac4655 | 64 | |
bb3884d8 JA |
65 | struct io_log *agg_io_log[2]; |
66 | ||
ebac4655 | 67 | #define TERMINATE_ALL (-1) |
75154845 | 68 | #define JOB_START_TIMEOUT (5 * 1000) |
ebac4655 | 69 | |
b29ee5b3 | 70 | void td_set_runstate(struct thread_data *td, int runstate) |
6ce15a32 | 71 | { |
d8fab1b6 JA |
72 | if (td->runstate == runstate) |
73 | return; | |
74 | ||
5921e80c JA |
75 | dprint(FD_PROCESS, "pid=%d: runstate %d -> %d\n", (int) td->pid, |
76 | td->runstate, runstate); | |
6ce15a32 JA |
77 | td->runstate = runstate; |
78 | } | |
79 | ||
390c40e2 | 80 | static void terminate_threads(int group_id) |
ebac4655 | 81 | { |
34572e28 | 82 | struct thread_data *td; |
ebac4655 JA |
83 | int i; |
84 | ||
6f88bcb0 JA |
85 | dprint(FD_PROCESS, "terminate group_id=%d\n", group_id); |
86 | ||
34572e28 | 87 | for_each_td(td, i) { |
ebac4655 | 88 | if (group_id == TERMINATE_ALL || groupid == td->groupid) { |
6f88bcb0 | 89 | dprint(FD_PROCESS, "setting terminate on %s/%d\n", |
5921e80c | 90 | td->o.name, (int) td->pid); |
ad830ed7 JA |
91 | td->terminate = 1; |
92 | td->o.start_delay = 0; | |
93 | ||
7a93ab19 JA |
94 | /* |
95 | * if the thread is running, just let it exit | |
96 | */ | |
97 | if (td->runstate < TD_RUNNING) | |
c1302d44 | 98 | kill(td->pid, SIGQUIT); |
f63f7f3b JA |
99 | else { |
100 | struct ioengine_ops *ops = td->io_ops; | |
101 | ||
102 | if (ops && (ops->flags & FIO_SIGQUIT)) | |
103 | kill(td->pid, SIGQUIT); | |
104 | } | |
ebac4655 JA |
105 | } |
106 | } | |
107 | } | |
108 | ||
ccb0fa24 JA |
109 | static void status_timer_arm(void) |
110 | { | |
111 | itimer.it_value.tv_sec = 0; | |
112 | itimer.it_value.tv_usec = DISK_UTIL_MSEC * 1000; | |
113 | setitimer(ITIMER_REAL, &itimer, NULL); | |
114 | } | |
115 | ||
02444ad1 | 116 | static void sig_alrm(int fio_unused sig) |
ebac4655 | 117 | { |
697a606c | 118 | if (threads) { |
5ec10eaa | 119 | update_io_ticks(); |
5ec10eaa | 120 | print_thread_status(); |
ccb0fa24 | 121 | status_timer_arm(); |
697a606c JA |
122 | } |
123 | } | |
124 | ||
5f8f9726 JA |
125 | /* |
126 | * Happens on thread runs with ctrl-c, ignore our own SIGQUIT | |
127 | */ | |
128 | static void sig_quit(int sig) | |
129 | { | |
130 | } | |
131 | ||
697a606c JA |
132 | static void sig_int(int sig) |
133 | { | |
134 | if (threads) { | |
5ec10eaa JA |
135 | printf("\nfio: terminating on signal %d\n", sig); |
136 | fflush(stdout); | |
137 | terminate_threads(TERMINATE_ALL); | |
ebac4655 JA |
138 | } |
139 | } | |
140 | ||
02444ad1 | 141 | static void sig_ill(int fio_unused sig) |
de79c915 JA |
142 | { |
143 | if (!threads) | |
144 | return; | |
145 | ||
146 | log_err("fio: illegal instruction. your cpu does not support " | |
147 | "the sse4.2 instruction for crc32c\n"); | |
148 | terminate_threads(TERMINATE_ALL); | |
149 | exit(4); | |
150 | } | |
151 | ||
697a606c JA |
152 | static void set_sig_handlers(void) |
153 | { | |
154 | struct sigaction act; | |
155 | ||
156 | memset(&act, 0, sizeof(act)); | |
157 | act.sa_handler = sig_alrm; | |
158 | act.sa_flags = SA_RESTART; | |
159 | sigaction(SIGALRM, &act, NULL); | |
160 | ||
161 | memset(&act, 0, sizeof(act)); | |
162 | act.sa_handler = sig_int; | |
163 | act.sa_flags = SA_RESTART; | |
164 | sigaction(SIGINT, &act, NULL); | |
de79c915 JA |
165 | |
166 | memset(&act, 0, sizeof(act)); | |
167 | act.sa_handler = sig_ill; | |
168 | act.sa_flags = SA_RESTART; | |
169 | sigaction(SIGILL, &act, NULL); | |
5f8f9726 JA |
170 | |
171 | memset(&act, 0, sizeof(act)); | |
172 | act.sa_handler = sig_quit; | |
173 | act.sa_flags = SA_RESTART; | |
174 | sigaction(SIGQUIT, &act, NULL); | |
697a606c JA |
175 | } |
176 | ||
906c8d75 JA |
177 | /* |
178 | * Check if we are above the minimum rate given. | |
179 | */ | |
581e7141 JA |
180 | static int __check_min_rate(struct thread_data *td, struct timeval *now, |
181 | enum td_ddir ddir) | |
ebac4655 | 182 | { |
0904200b | 183 | unsigned long long bytes = 0; |
4e991c23 | 184 | unsigned long iops = 0; |
ebac4655 JA |
185 | unsigned long spent; |
186 | unsigned long rate; | |
581e7141 JA |
187 | unsigned int ratemin = 0; |
188 | unsigned int rate_iops = 0; | |
189 | unsigned int rate_iops_min = 0; | |
ebac4655 | 190 | |
d982d873 JA |
191 | if (!td->o.ratemin[ddir] && !td->o.rate_iops_min[ddir]) |
192 | return 0; | |
193 | ||
ebac4655 JA |
194 | /* |
195 | * allow a 2 second settle period in the beginning | |
196 | */ | |
197 | if (mtime_since(&td->start, now) < 2000) | |
198 | return 0; | |
199 | ||
581e7141 JA |
200 | iops += td->io_blocks[ddir]; |
201 | bytes += td->this_io_bytes[ddir]; | |
202 | ratemin += td->o.ratemin[ddir]; | |
203 | rate_iops += td->o.rate_iops[ddir]; | |
204 | rate_iops_min += td->o.rate_iops_min[ddir]; | |
0904200b | 205 | |
ebac4655 JA |
206 | /* |
207 | * if rate blocks is set, sample is running | |
208 | */ | |
581e7141 JA |
209 | if (td->rate_bytes[ddir] || td->rate_blocks[ddir]) { |
210 | spent = mtime_since(&td->lastrate[ddir], now); | |
2dc1bbeb | 211 | if (spent < td->o.ratecycle) |
ebac4655 JA |
212 | return 0; |
213 | ||
581e7141 | 214 | if (td->o.rate[ddir]) { |
4e991c23 JA |
215 | /* |
216 | * check bandwidth specified rate | |
217 | */ | |
581e7141 | 218 | if (bytes < td->rate_bytes[ddir]) { |
5ec10eaa | 219 | log_err("%s: min rate %u not met\n", td->o.name, |
581e7141 | 220 | ratemin); |
4e991c23 JA |
221 | return 1; |
222 | } else { | |
581e7141 JA |
223 | rate = ((bytes - td->rate_bytes[ddir]) * 1000) / spent; |
224 | if (rate < ratemin || | |
225 | bytes < td->rate_bytes[ddir]) { | |
5ec10eaa | 226 | log_err("%s: min rate %u not met, got" |
b22989b9 | 227 | " %luKB/sec\n", td->o.name, |
581e7141 | 228 | ratemin, rate); |
4e991c23 JA |
229 | return 1; |
230 | } | |
231 | } | |
413dd459 | 232 | } else { |
4e991c23 JA |
233 | /* |
234 | * checks iops specified rate | |
235 | */ | |
581e7141 | 236 | if (iops < rate_iops) { |
5ec10eaa | 237 | log_err("%s: min iops rate %u not met\n", |
581e7141 | 238 | td->o.name, rate_iops); |
413dd459 | 239 | return 1; |
4e991c23 | 240 | } else { |
581e7141 JA |
241 | rate = ((iops - td->rate_blocks[ddir]) * 1000) / spent; |
242 | if (rate < rate_iops_min || | |
243 | iops < td->rate_blocks[ddir]) { | |
5ec10eaa JA |
244 | log_err("%s: min iops rate %u not met," |
245 | " got %lu\n", td->o.name, | |
581e7141 | 246 | rate_iops_min, rate); |
4e991c23 | 247 | } |
413dd459 | 248 | } |
ebac4655 JA |
249 | } |
250 | } | |
251 | ||
581e7141 JA |
252 | td->rate_bytes[ddir] = bytes; |
253 | td->rate_blocks[ddir] = iops; | |
254 | memcpy(&td->lastrate[ddir], now, sizeof(*now)); | |
ebac4655 JA |
255 | return 0; |
256 | } | |
257 | ||
581e7141 JA |
258 | static int check_min_rate(struct thread_data *td, struct timeval *now, |
259 | unsigned long *bytes_done) | |
260 | { | |
261 | int ret = 0; | |
262 | ||
263 | if (bytes_done[0]) | |
264 | ret |= __check_min_rate(td, now, 0); | |
265 | if (bytes_done[1]) | |
266 | ret |= __check_min_rate(td, now, 1); | |
267 | ||
268 | return ret; | |
269 | } | |
270 | ||
ebac4655 JA |
271 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) |
272 | { | |
2dc1bbeb | 273 | if (!td->o.timeout) |
ebac4655 | 274 | return 0; |
2dc1bbeb | 275 | if (mtime_since(&td->epoch, t) >= td->o.timeout * 1000) |
ebac4655 JA |
276 | return 1; |
277 | ||
278 | return 0; | |
279 | } | |
280 | ||
906c8d75 JA |
281 | /* |
282 | * When job exits, we can cancel the in-flight IO if we are using async | |
283 | * io. Attempt to do so. | |
284 | */ | |
ebac4655 JA |
285 | static void cleanup_pending_aio(struct thread_data *td) |
286 | { | |
01743ee1 | 287 | struct flist_head *entry, *n; |
ebac4655 JA |
288 | struct io_u *io_u; |
289 | int r; | |
290 | ||
291 | /* | |
292 | * get immediately available events, if any | |
293 | */ | |
581e7141 | 294 | r = io_u_queued_complete(td, 0, NULL); |
b2fdda43 JA |
295 | if (r < 0) |
296 | return; | |
ebac4655 JA |
297 | |
298 | /* | |
299 | * now cancel remaining active events | |
300 | */ | |
2866c82d | 301 | if (td->io_ops->cancel) { |
01743ee1 JA |
302 | flist_for_each_safe(entry, n, &td->io_u_busylist) { |
303 | io_u = flist_entry(entry, struct io_u, list); | |
ebac4655 | 304 | |
0c6e7517 JA |
305 | /* |
306 | * if the io_u isn't in flight, then that generally | |
307 | * means someone leaked an io_u. complain but fix | |
308 | * it up, so we don't stall here. | |
309 | */ | |
310 | if ((io_u->flags & IO_U_F_FLIGHT) == 0) { | |
311 | log_err("fio: non-busy IO on busy list\n"); | |
ebac4655 | 312 | put_io_u(td, io_u); |
0c6e7517 JA |
313 | } else { |
314 | r = td->io_ops->cancel(td, io_u); | |
315 | if (!r) | |
316 | put_io_u(td, io_u); | |
317 | } | |
ebac4655 JA |
318 | } |
319 | } | |
320 | ||
97601024 | 321 | if (td->cur_depth) |
581e7141 | 322 | r = io_u_queued_complete(td, td->cur_depth, NULL); |
ebac4655 JA |
323 | } |
324 | ||
858a3d47 JA |
325 | /* |
326 | * Helper to handle the final sync of a file. Works just like the normal | |
327 | * io path, just does everything sync. | |
328 | */ | |
329 | static int fio_io_sync(struct thread_data *td, struct fio_file *f) | |
330 | { | |
331 | struct io_u *io_u = __get_io_u(td); | |
858a3d47 JA |
332 | int ret; |
333 | ||
334 | if (!io_u) | |
335 | return 1; | |
336 | ||
337 | io_u->ddir = DDIR_SYNC; | |
338 | io_u->file = f; | |
339 | ||
340 | if (td_io_prep(td, io_u)) { | |
341 | put_io_u(td, io_u); | |
342 | return 1; | |
343 | } | |
344 | ||
755200a3 | 345 | requeue: |
858a3d47 | 346 | ret = td_io_queue(td, io_u); |
36167d82 | 347 | if (ret < 0) { |
e1161c32 | 348 | td_verror(td, io_u->error, "td_io_queue"); |
858a3d47 | 349 | put_io_u(td, io_u); |
858a3d47 | 350 | return 1; |
36167d82 | 351 | } else if (ret == FIO_Q_QUEUED) { |
581e7141 | 352 | if (io_u_queued_complete(td, 1, NULL) < 0) |
36167d82 | 353 | return 1; |
36167d82 JA |
354 | } else if (ret == FIO_Q_COMPLETED) { |
355 | if (io_u->error) { | |
e1161c32 | 356 | td_verror(td, io_u->error, "td_io_queue"); |
36167d82 JA |
357 | return 1; |
358 | } | |
858a3d47 | 359 | |
581e7141 | 360 | if (io_u_sync_complete(td, io_u, NULL) < 0) |
b2fdda43 | 361 | return 1; |
755200a3 JA |
362 | } else if (ret == FIO_Q_BUSY) { |
363 | if (td_io_commit(td)) | |
364 | return 1; | |
365 | goto requeue; | |
858a3d47 JA |
366 | } |
367 | ||
368 | return 0; | |
369 | } | |
370 | ||
993bf48b JA |
371 | static inline void update_tv_cache(struct thread_data *td) |
372 | { | |
373 | if ((++td->tv_cache_nr & td->tv_cache_mask) == td->tv_cache_mask) | |
374 | fio_gettime(&td->tv_cache, NULL); | |
375 | } | |
376 | ||
f2bba182 RR |
377 | static int break_on_this_error(struct thread_data *td, int *retptr) |
378 | { | |
379 | int ret = *retptr; | |
380 | ||
381 | if (ret < 0 || td->error) { | |
382 | int err; | |
383 | ||
1ec99eea JA |
384 | if (!td->o.continue_on_error) |
385 | return 1; | |
f2bba182 RR |
386 | |
387 | if (ret < 0) | |
388 | err = -ret; | |
389 | else | |
390 | err = td->error; | |
391 | ||
f2bba182 RR |
392 | if (td_non_fatal_error(err)) { |
393 | /* | |
394 | * Continue with the I/Os in case of | |
395 | * a non fatal error. | |
396 | */ | |
8a4f8a24 | 397 | update_error_count(td, err); |
f2bba182 RR |
398 | td_clear_error(td); |
399 | *retptr = 0; | |
400 | return 0; | |
8a4f8a24 RR |
401 | } else if (td->o.fill_device && err == ENOSPC) { |
402 | /* | |
403 | * We expect to hit this error if | |
404 | * fill_device option is set. | |
405 | */ | |
406 | td_clear_error(td); | |
407 | td->terminate = 1; | |
408 | return 1; | |
f2bba182 RR |
409 | } else { |
410 | /* | |
411 | * Stop the I/O in case of a fatal | |
412 | * error. | |
413 | */ | |
8a4f8a24 | 414 | update_error_count(td, err); |
f2bba182 RR |
415 | return 1; |
416 | } | |
417 | } | |
418 | ||
419 | return 0; | |
420 | } | |
421 | ||
906c8d75 | 422 | /* |
34403fb1 | 423 | * The main verify engine. Runs over the writes we previously submitted, |
906c8d75 JA |
424 | * reads the blocks back in, and checks the crc/md5 of the data. |
425 | */ | |
1e97cce9 | 426 | static void do_verify(struct thread_data *td) |
ebac4655 | 427 | { |
53cdc686 | 428 | struct fio_file *f; |
36167d82 | 429 | struct io_u *io_u; |
af52b345 JA |
430 | int ret, min_events; |
431 | unsigned int i; | |
e5b401d4 JA |
432 | |
433 | /* | |
434 | * sync io first and invalidate cache, to make sure we really | |
435 | * read from disk. | |
436 | */ | |
437 | for_each_file(td, f, i) { | |
d6aed795 | 438 | if (!fio_file_open(f)) |
3d7b485f | 439 | continue; |
b2fdda43 JA |
440 | if (fio_io_sync(td, f)) |
441 | break; | |
442 | if (file_invalidate_cache(td, f)) | |
443 | break; | |
e5b401d4 | 444 | } |
ebac4655 | 445 | |
b2fdda43 JA |
446 | if (td->error) |
447 | return; | |
448 | ||
ebac4655 JA |
449 | td_set_runstate(td, TD_VERIFYING); |
450 | ||
36167d82 JA |
451 | io_u = NULL; |
452 | while (!td->terminate) { | |
4950421a | 453 | int ret2, full; |
5451792e | 454 | |
993bf48b JA |
455 | update_tv_cache(td); |
456 | ||
457 | if (runtime_exceeded(td, &td->tv_cache)) { | |
90a87d4b | 458 | td->terminate = 1; |
02bcaa8c | 459 | break; |
069c2918 | 460 | } |
02bcaa8c | 461 | |
dcb4830d RR |
462 | io_u = __get_io_u(td); |
463 | if (!io_u) | |
464 | break; | |
465 | ||
069c2918 JA |
466 | if (get_next_verify(td, io_u)) { |
467 | put_io_u(td, io_u); | |
ebac4655 | 468 | break; |
069c2918 | 469 | } |
ebac4655 | 470 | |
069c2918 JA |
471 | if (td_io_prep(td, io_u)) { |
472 | put_io_u(td, io_u); | |
53cdc686 | 473 | break; |
069c2918 | 474 | } |
d7762cf8 | 475 | |
e8462bd8 JA |
476 | if (td->o.verify_async) |
477 | io_u->end_io = verify_io_u_async; | |
478 | else | |
479 | io_u->end_io = verify_io_u; | |
53cdc686 | 480 | |
11786802 | 481 | ret = td_io_queue(td, io_u); |
36167d82 JA |
482 | switch (ret) { |
483 | case FIO_Q_COMPLETED: | |
f2bba182 | 484 | if (io_u->error) { |
22819ec2 | 485 | ret = -io_u->error; |
f2bba182 RR |
486 | clear_io_u(td, io_u); |
487 | } else if (io_u->resid) { | |
36167d82 | 488 | int bytes = io_u->xfer_buflen - io_u->resid; |
d460eb31 | 489 | struct fio_file *f = io_u->file; |
ebac4655 | 490 | |
9e9d164e JA |
491 | /* |
492 | * zero read, fail | |
493 | */ | |
494 | if (!bytes) { | |
41d9f0fd | 495 | td_verror(td, EIO, "full resid"); |
9e9d164e JA |
496 | put_io_u(td, io_u); |
497 | break; | |
498 | } | |
8400d9b2 | 499 | |
36167d82 JA |
500 | io_u->xfer_buflen = io_u->resid; |
501 | io_u->xfer_buf += bytes; | |
8400d9b2 JA |
502 | io_u->offset += bytes; |
503 | ||
30061b97 JA |
504 | td->ts.short_io_u[io_u->ddir]++; |
505 | ||
d460eb31 | 506 | if (io_u->offset == f->real_file_size) |
8400d9b2 JA |
507 | goto sync_done; |
508 | ||
11786802 JA |
509 | requeue_io_u(td, &io_u); |
510 | } else { | |
8400d9b2 | 511 | sync_done: |
581e7141 | 512 | ret = io_u_sync_complete(td, io_u, NULL); |
11786802 JA |
513 | if (ret < 0) |
514 | break; | |
36167d82 | 515 | } |
36167d82 JA |
516 | continue; |
517 | case FIO_Q_QUEUED: | |
518 | break; | |
755200a3 JA |
519 | case FIO_Q_BUSY: |
520 | requeue_io_u(td, &io_u); | |
5451792e JA |
521 | ret2 = td_io_commit(td); |
522 | if (ret2 < 0) | |
523 | ret = ret2; | |
755200a3 | 524 | break; |
36167d82 JA |
525 | default: |
526 | assert(ret < 0); | |
e1161c32 | 527 | td_verror(td, -ret, "td_io_queue"); |
ebac4655 JA |
528 | break; |
529 | } | |
530 | ||
f2bba182 | 531 | if (break_on_this_error(td, &ret)) |
3af6ef39 JA |
532 | break; |
533 | ||
ebac4655 | 534 | /* |
3af6ef39 JA |
535 | * if we can queue more, do so. but check if there are |
536 | * completed io_u's first. | |
ebac4655 | 537 | */ |
4950421a JA |
538 | full = queue_full(td) || ret == FIO_Q_BUSY; |
539 | if (full || !td->o.iodepth_batch_complete) { | |
6a96276c RR |
540 | min_events = min(td->o.iodepth_batch_complete, |
541 | td->cur_depth); | |
4950421a | 542 | if (full && !min_events) |
838bc709 | 543 | min_events = 1; |
e916b390 | 544 | |
4950421a JA |
545 | do { |
546 | /* | |
547 | * Reap required number of io units, if any, | |
548 | * and do the verification on them through | |
549 | * the callback handler | |
550 | */ | |
581e7141 | 551 | if (io_u_queued_complete(td, min_events, NULL) < 0) { |
4950421a JA |
552 | ret = -1; |
553 | break; | |
554 | } | |
555 | } while (full && (td->cur_depth > td->o.iodepth_low)); | |
556 | } | |
557 | if (ret < 0) | |
ebac4655 | 558 | break; |
36167d82 | 559 | } |
ebac4655 | 560 | |
c01c0395 JA |
561 | if (!td->error) { |
562 | min_events = td->cur_depth; | |
563 | ||
564 | if (min_events) | |
581e7141 | 565 | ret = io_u_queued_complete(td, min_events, NULL); |
c01c0395 | 566 | } else |
ebac4655 JA |
567 | cleanup_pending_aio(td); |
568 | ||
569 | td_set_runstate(td, TD_RUNNING); | |
570 | } | |
571 | ||
32cd46a0 | 572 | /* |
906c8d75 | 573 | * Main IO worker function. It retrieves io_u's to process and queues |
32cd46a0 JA |
574 | * and reaps them, checking for rate and errors along the way. |
575 | */ | |
ebac4655 JA |
576 | static void do_io(struct thread_data *td) |
577 | { | |
af52b345 JA |
578 | unsigned int i; |
579 | int ret = 0; | |
ebac4655 | 580 | |
b29ee5b3 JA |
581 | if (in_ramp_time(td)) |
582 | td_set_runstate(td, TD_RAMP); | |
583 | else | |
584 | td_set_runstate(td, TD_RUNNING); | |
5853e5a8 | 585 | |
457bf399 GO |
586 | while ( (td->o.read_iolog_file && !flist_empty(&td->io_log_list)) || |
587 | ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->o.size) ) { | |
97601024 | 588 | struct timeval comp_time; |
581e7141 | 589 | unsigned long bytes_done[2] = { 0, 0 }; |
84585003 | 590 | int min_evts = 0; |
ebac4655 | 591 | struct io_u *io_u; |
4950421a | 592 | int ret2, full; |
ebac4655 JA |
593 | |
594 | if (td->terminate) | |
595 | break; | |
596 | ||
993bf48b | 597 | update_tv_cache(td); |
97601024 | 598 | |
993bf48b | 599 | if (runtime_exceeded(td, &td->tv_cache)) { |
90a87d4b | 600 | td->terminate = 1; |
97601024 JA |
601 | break; |
602 | } | |
36167d82 | 603 | |
dcb4830d RR |
604 | io_u = get_io_u(td); |
605 | if (!io_u) | |
606 | break; | |
607 | ||
b67740d3 JA |
608 | /* |
609 | * Add verification end_io handler, if asked to verify | |
610 | * a previously written file. | |
611 | */ | |
715d2b3e JA |
612 | if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ && |
613 | !td_rw(td)) { | |
e8462bd8 JA |
614 | if (td->o.verify_async) |
615 | io_u->end_io = verify_io_u_async; | |
616 | else | |
617 | io_u->end_io = verify_io_u; | |
d8fab1b6 | 618 | td_set_runstate(td, TD_VERIFYING); |
b29ee5b3 JA |
619 | } else if (in_ramp_time(td)) |
620 | td_set_runstate(td, TD_RAMP); | |
621 | else | |
d8fab1b6 | 622 | td_set_runstate(td, TD_RUNNING); |
b67740d3 | 623 | |
11786802 | 624 | ret = td_io_queue(td, io_u); |
36167d82 JA |
625 | switch (ret) { |
626 | case FIO_Q_COMPLETED: | |
f2bba182 | 627 | if (io_u->error) { |
5451792e | 628 | ret = -io_u->error; |
f2bba182 RR |
629 | clear_io_u(td, io_u); |
630 | } else if (io_u->resid) { | |
36167d82 | 631 | int bytes = io_u->xfer_buflen - io_u->resid; |
d460eb31 | 632 | struct fio_file *f = io_u->file; |
36167d82 | 633 | |
9e9d164e JA |
634 | /* |
635 | * zero read, fail | |
636 | */ | |
637 | if (!bytes) { | |
41d9f0fd | 638 | td_verror(td, EIO, "full resid"); |
9e9d164e JA |
639 | put_io_u(td, io_u); |
640 | break; | |
641 | } | |
642 | ||
cec6b55d | 643 | io_u->xfer_buflen = io_u->resid; |
36167d82 | 644 | io_u->xfer_buf += bytes; |
5a7c5680 JA |
645 | io_u->offset += bytes; |
646 | ||
30061b97 JA |
647 | td->ts.short_io_u[io_u->ddir]++; |
648 | ||
d460eb31 | 649 | if (io_u->offset == f->real_file_size) |
5a7c5680 JA |
650 | goto sync_done; |
651 | ||
11786802 JA |
652 | requeue_io_u(td, &io_u); |
653 | } else { | |
5a7c5680 | 654 | sync_done: |
581e7141 JA |
655 | if (__should_check_rate(td, 0) || |
656 | __should_check_rate(td, 1)) | |
993bf48b JA |
657 | fio_gettime(&comp_time, NULL); |
658 | ||
581e7141 JA |
659 | ret = io_u_sync_complete(td, io_u, bytes_done); |
660 | if (ret < 0) | |
661 | break; | |
cec6b55d | 662 | } |
36167d82 JA |
663 | break; |
664 | case FIO_Q_QUEUED: | |
7e77dd02 JA |
665 | /* |
666 | * if the engine doesn't have a commit hook, | |
667 | * the io_u is really queued. if it does have such | |
668 | * a hook, it has to call io_u_queued() itself. | |
669 | */ | |
670 | if (td->io_ops->commit == NULL) | |
671 | io_u_queued(td, io_u); | |
36167d82 | 672 | break; |
755200a3 JA |
673 | case FIO_Q_BUSY: |
674 | requeue_io_u(td, &io_u); | |
5451792e JA |
675 | ret2 = td_io_commit(td); |
676 | if (ret2 < 0) | |
677 | ret = ret2; | |
755200a3 | 678 | break; |
36167d82 JA |
679 | default: |
680 | assert(ret < 0); | |
681 | put_io_u(td, io_u); | |
682 | break; | |
ebac4655 JA |
683 | } |
684 | ||
f2bba182 | 685 | if (break_on_this_error(td, &ret)) |
36167d82 JA |
686 | break; |
687 | ||
97601024 JA |
688 | /* |
689 | * See if we need to complete some commands | |
690 | */ | |
4950421a JA |
691 | full = queue_full(td) || ret == FIO_Q_BUSY; |
692 | if (full || !td->o.iodepth_batch_complete) { | |
6a96276c RR |
693 | min_evts = min(td->o.iodepth_batch_complete, |
694 | td->cur_depth); | |
4950421a | 695 | if (full && !min_evts) |
36167d82 | 696 | min_evts = 1; |
4950421a | 697 | |
581e7141 JA |
698 | if (__should_check_rate(td, 0) || |
699 | __should_check_rate(td, 1)) | |
993bf48b | 700 | fio_gettime(&comp_time, NULL); |
4950421a JA |
701 | |
702 | do { | |
581e7141 JA |
703 | ret = io_u_queued_complete(td, min_evts, bytes_done); |
704 | if (ret < 0) | |
4950421a JA |
705 | break; |
706 | ||
4950421a | 707 | } while (full && (td->cur_depth > td->o.iodepth_low)); |
ebac4655 JA |
708 | } |
709 | ||
4950421a JA |
710 | if (ret < 0) |
711 | break; | |
581e7141 | 712 | if (!(bytes_done[0] + bytes_done[1])) |
97601024 JA |
713 | continue; |
714 | ||
581e7141 JA |
715 | if (!in_ramp_time(td) && should_check_rate(td, bytes_done)) { |
716 | if (check_min_rate(td, &comp_time, bytes_done)) { | |
721938ae JA |
717 | if (exitall_on_terminate) |
718 | terminate_threads(td->groupid); | |
719 | td_verror(td, EIO, "check_min_rate"); | |
720 | break; | |
721 | } | |
ebac4655 JA |
722 | } |
723 | ||
2dc1bbeb | 724 | if (td->o.thinktime) { |
9c1f7434 JA |
725 | unsigned long long b; |
726 | ||
727 | b = td->io_blocks[0] + td->io_blocks[1]; | |
2dc1bbeb | 728 | if (!(b % td->o.thinktime_blocks)) { |
48097d5c JA |
729 | int left; |
730 | ||
2dc1bbeb | 731 | if (td->o.thinktime_spin) |
6dd6f2cd | 732 | usec_spin(td->o.thinktime_spin); |
48097d5c | 733 | |
2dc1bbeb | 734 | left = td->o.thinktime - td->o.thinktime_spin; |
48097d5c JA |
735 | if (left) |
736 | usec_sleep(td, left); | |
737 | } | |
9c1f7434 | 738 | } |
ebac4655 JA |
739 | } |
740 | ||
aa31f1f1 SL |
741 | if (td->o.fill_device && td->error == ENOSPC) { |
742 | td->error = 0; | |
743 | td->terminate = 1; | |
744 | } | |
4d2413c6 | 745 | if (!td->error) { |
3d7c391d JA |
746 | struct fio_file *f; |
747 | ||
c01c0395 JA |
748 | i = td->cur_depth; |
749 | if (i) | |
581e7141 | 750 | ret = io_u_queued_complete(td, i, NULL); |
ebac4655 | 751 | |
2dc1bbeb | 752 | if (should_fsync(td) && td->o.end_fsync) { |
84585003 | 753 | td_set_runstate(td, TD_FSYNCING); |
3d7b485f JA |
754 | |
755 | for_each_file(td, f, i) { | |
d6aed795 | 756 | if (!fio_file_open(f)) |
3d7b485f | 757 | continue; |
858a3d47 | 758 | fio_io_sync(td, f); |
3d7b485f | 759 | } |
84585003 | 760 | } |
c01c0395 JA |
761 | } else |
762 | cleanup_pending_aio(td); | |
2ba1c290 JA |
763 | |
764 | /* | |
765 | * stop job if we failed doing any IO | |
766 | */ | |
767 | if ((td->this_io_bytes[0] + td->this_io_bytes[1]) == 0) | |
768 | td->done = 1; | |
ebac4655 JA |
769 | } |
770 | ||
ebac4655 JA |
771 | static void cleanup_io_u(struct thread_data *td) |
772 | { | |
01743ee1 | 773 | struct flist_head *entry, *n; |
ebac4655 JA |
774 | struct io_u *io_u; |
775 | ||
01743ee1 JA |
776 | flist_for_each_safe(entry, n, &td->io_u_freelist) { |
777 | io_u = flist_entry(entry, struct io_u, list); | |
ebac4655 | 778 | |
01743ee1 | 779 | flist_del(&io_u->list); |
ebac4655 JA |
780 | free(io_u); |
781 | } | |
782 | ||
2f9ade3c | 783 | free_io_mem(td); |
ebac4655 JA |
784 | } |
785 | ||
786 | static int init_io_u(struct thread_data *td) | |
787 | { | |
788 | struct io_u *io_u; | |
a00735e6 | 789 | unsigned int max_bs; |
eb7ccf38 | 790 | int cl_align, i, max_units; |
ebac4655 JA |
791 | char *p; |
792 | ||
1e3c09af | 793 | max_units = td->o.iodepth; |
2dc1bbeb | 794 | max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]); |
5ec10eaa JA |
795 | td->orig_buffer_size = (unsigned long long) max_bs |
796 | * (unsigned long long) max_units; | |
797 | ||
798 | if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) { | |
799 | unsigned long bs; | |
74b025b0 | 800 | |
5ec10eaa JA |
801 | bs = td->orig_buffer_size + td->o.hugepage_size - 1; |
802 | td->orig_buffer_size = bs & ~(td->o.hugepage_size - 1); | |
803 | } | |
ebac4655 | 804 | |
c3990645 JA |
805 | if (td->orig_buffer_size != (size_t) td->orig_buffer_size) { |
806 | log_err("fio: IO memory too large. Reduce max_bs or iodepth\n"); | |
807 | return 1; | |
808 | } | |
809 | ||
2f9ade3c JA |
810 | if (allocate_io_mem(td)) |
811 | return 1; | |
ebac4655 | 812 | |
fc966bef | 813 | if (td->o.odirect || td->o.mem_align) |
d529ee19 | 814 | p = PAGE_ALIGN(td->orig_buffer) + td->o.mem_align; |
b3f378e9 JA |
815 | else |
816 | p = td->orig_buffer; | |
817 | ||
eb7ccf38 JA |
818 | cl_align = os_cache_line_size(); |
819 | ||
ebac4655 | 820 | for (i = 0; i < max_units; i++) { |
eb7ccf38 JA |
821 | void *ptr; |
822 | ||
db1cd90b JA |
823 | if (td->terminate) |
824 | return 1; | |
eb7ccf38 JA |
825 | |
826 | if (posix_memalign(&ptr, cl_align, sizeof(*io_u))) { | |
827 | log_err("fio: posix_memalign=%s\n", strerror(errno)); | |
828 | break; | |
829 | } | |
830 | ||
831 | io_u = ptr; | |
ebac4655 | 832 | memset(io_u, 0, sizeof(*io_u)); |
01743ee1 | 833 | INIT_FLIST_HEAD(&io_u->list); |
d529ee19 | 834 | dprint(FD_MEM, "io_u alloc %p, index %u\n", io_u, i); |
ebac4655 | 835 | |
b4c5e1ac JA |
836 | if (!(td->io_ops->flags & FIO_NOIO)) { |
837 | io_u->buf = p + max_bs * i; | |
d529ee19 | 838 | dprint(FD_MEM, "io_u %p, mem %p\n", io_u, io_u->buf); |
e9459e5a | 839 | |
5973cafb JA |
840 | if (td_write(td) && !td->o.refill_buffers) |
841 | io_u_fill_buffer(td, io_u, max_bs); | |
b4c5e1ac | 842 | } |
6b9cea23 | 843 | |
b1ff3403 | 844 | io_u->index = i; |
0c6e7517 | 845 | io_u->flags = IO_U_F_FREE; |
01743ee1 | 846 | flist_add(&io_u->list, &td->io_u_freelist); |
ebac4655 JA |
847 | } |
848 | ||
849 | return 0; | |
850 | } | |
851 | ||
da86774e JA |
852 | static int switch_ioscheduler(struct thread_data *td) |
853 | { | |
854 | char tmp[256], tmp2[128]; | |
855 | FILE *f; | |
856 | int ret; | |
857 | ||
ba0fbe10 | 858 | if (td->io_ops->flags & FIO_DISKLESSIO) |
f48b467c JA |
859 | return 0; |
860 | ||
da86774e JA |
861 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); |
862 | ||
863 | f = fopen(tmp, "r+"); | |
864 | if (!f) { | |
cbf5c1d7 | 865 | if (errno == ENOENT) { |
5ec10eaa JA |
866 | log_err("fio: os or kernel doesn't support IO scheduler" |
867 | " switching\n"); | |
cbf5c1d7 JA |
868 | return 0; |
869 | } | |
870 | td_verror(td, errno, "fopen iosched"); | |
da86774e JA |
871 | return 1; |
872 | } | |
873 | ||
874 | /* | |
875 | * Set io scheduler. | |
876 | */ | |
2dc1bbeb | 877 | ret = fwrite(td->o.ioscheduler, strlen(td->o.ioscheduler), 1, f); |
da86774e | 878 | if (ferror(f) || ret != 1) { |
e1161c32 | 879 | td_verror(td, errno, "fwrite"); |
da86774e JA |
880 | fclose(f); |
881 | return 1; | |
882 | } | |
883 | ||
884 | rewind(f); | |
885 | ||
886 | /* | |
887 | * Read back and check that the selected scheduler is now the default. | |
888 | */ | |
889 | ret = fread(tmp, 1, sizeof(tmp), f); | |
890 | if (ferror(f) || ret < 0) { | |
e1161c32 | 891 | td_verror(td, errno, "fread"); |
da86774e JA |
892 | fclose(f); |
893 | return 1; | |
894 | } | |
895 | ||
2dc1bbeb | 896 | sprintf(tmp2, "[%s]", td->o.ioscheduler); |
da86774e | 897 | if (!strstr(tmp, tmp2)) { |
2dc1bbeb | 898 | log_err("fio: io scheduler %s not found\n", td->o.ioscheduler); |
e1161c32 | 899 | td_verror(td, EINVAL, "iosched_switch"); |
da86774e JA |
900 | fclose(f); |
901 | return 1; | |
902 | } | |
903 | ||
904 | fclose(f); | |
905 | return 0; | |
906 | } | |
907 | ||
492158cf JA |
908 | static int keep_running(struct thread_data *td) |
909 | { | |
910 | unsigned long long io_done; | |
911 | ||
20e354ef JA |
912 | if (td->done) |
913 | return 0; | |
492158cf JA |
914 | if (td->o.time_based) |
915 | return 1; | |
916 | if (td->o.loops) { | |
917 | td->o.loops--; | |
918 | return 1; | |
919 | } | |
920 | ||
5ec10eaa JA |
921 | io_done = td->io_bytes[DDIR_READ] + td->io_bytes[DDIR_WRITE] |
922 | + td->io_skip_bytes; | |
492158cf JA |
923 | if (io_done < td->o.size) |
924 | return 1; | |
925 | ||
926 | return 0; | |
927 | } | |
928 | ||
b29ee5b3 | 929 | static void reset_io_counters(struct thread_data *td) |
ebac4655 | 930 | { |
756867bd | 931 | td->ts.stat_io_bytes[0] = td->ts.stat_io_bytes[1] = 0; |
ebac4655 | 932 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; |
20dc95c4 | 933 | td->zone_bytes = 0; |
581e7141 JA |
934 | td->rate_bytes[0] = td->rate_bytes[1] = 0; |
935 | td->rate_blocks[0] = td->rate_blocks[1] = 0; | |
ebac4655 | 936 | |
c1324df1 JA |
937 | td->last_was_sync = 0; |
938 | ||
2ba1c290 JA |
939 | /* |
940 | * reset file done count if we are to start over | |
941 | */ | |
942 | if (td->o.time_based || td->o.loops) | |
48f5abd3 | 943 | td->nr_done_files = 0; |
5bfc35d7 JA |
944 | |
945 | /* | |
946 | * Set the same seed to get repeatable runs | |
947 | */ | |
948 | td_fill_rand_seeds(td); | |
b29ee5b3 JA |
949 | } |
950 | ||
951 | void reset_all_stats(struct thread_data *td) | |
952 | { | |
953 | struct timeval tv; | |
954 | int i; | |
955 | ||
956 | reset_io_counters(td); | |
957 | ||
958 | for (i = 0; i < 2; i++) { | |
959 | td->io_bytes[i] = 0; | |
960 | td->io_blocks[i] = 0; | |
961 | td->io_issues[i] = 0; | |
962 | td->ts.total_io_u[i] = 0; | |
963 | } | |
964 | ||
965 | fio_gettime(&tv, NULL); | |
966 | memcpy(&td->epoch, &tv, sizeof(tv)); | |
967 | memcpy(&td->start, &tv, sizeof(tv)); | |
968 | } | |
969 | ||
df9c26b1 | 970 | static void clear_io_state(struct thread_data *td) |
b29ee5b3 JA |
971 | { |
972 | struct fio_file *f; | |
973 | unsigned int i; | |
b29ee5b3 JA |
974 | |
975 | reset_io_counters(td); | |
281f9b63 | 976 | |
24ffd2c2 | 977 | close_files(td); |
df9c26b1 | 978 | for_each_file(td, f, i) |
d6aed795 | 979 | fio_file_clear_done(f); |
ebac4655 JA |
980 | } |
981 | ||
398e8c4d JA |
982 | static int exec_string(const char *string) |
983 | { | |
984 | int ret, newlen = strlen(string) + 1 + 8; | |
985 | char *str; | |
986 | ||
987 | str = malloc(newlen); | |
988 | sprintf(str, "sh -c %s", string); | |
989 | ||
990 | ret = system(str); | |
991 | if (ret == -1) | |
992 | log_err("fio: exec of cmd <%s> failed\n", str); | |
993 | ||
994 | free(str); | |
995 | return ret; | |
996 | } | |
997 | ||
906c8d75 JA |
998 | /* |
999 | * Entry point for the thread based jobs. The process based jobs end up | |
1000 | * here as well, after a little setup. | |
1001 | */ | |
ebac4655 JA |
1002 | static void *thread_main(void *data) |
1003 | { | |
10927316 | 1004 | unsigned long long runtime[2], elapsed; |
ebac4655 | 1005 | struct thread_data *td = data; |
e8462bd8 | 1006 | pthread_condattr_t attr; |
a978ba68 | 1007 | int clear_state; |
ebac4655 | 1008 | |
2dc1bbeb | 1009 | if (!td->o.use_thread) |
ebac4655 JA |
1010 | setsid(); |
1011 | ||
1012 | td->pid = getpid(); | |
1013 | ||
5921e80c | 1014 | dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid); |
ee56ad50 | 1015 | |
01743ee1 JA |
1016 | INIT_FLIST_HEAD(&td->io_u_freelist); |
1017 | INIT_FLIST_HEAD(&td->io_u_busylist); | |
1018 | INIT_FLIST_HEAD(&td->io_u_requeues); | |
1019 | INIT_FLIST_HEAD(&td->io_log_list); | |
1020 | INIT_FLIST_HEAD(&td->io_hist_list); | |
e8462bd8 JA |
1021 | INIT_FLIST_HEAD(&td->verify_list); |
1022 | pthread_mutex_init(&td->io_u_lock, NULL); | |
bb5d7d0b | 1023 | td->io_hist_tree = RB_ROOT; |
aea47d44 | 1024 | |
e8462bd8 JA |
1025 | pthread_condattr_init(&attr); |
1026 | pthread_cond_init(&td->verify_cond, &attr); | |
1027 | pthread_cond_init(&td->free_cond, &attr); | |
1028 | ||
db1cd90b | 1029 | td_set_runstate(td, TD_INITIALIZED); |
29adda3c | 1030 | dprint(FD_MUTEX, "up startup_mutex\n"); |
cdd18ad8 | 1031 | fio_mutex_up(startup_mutex); |
29adda3c | 1032 | dprint(FD_MUTEX, "wait on td->mutex\n"); |
cdd18ad8 | 1033 | fio_mutex_down(td->mutex); |
29adda3c | 1034 | dprint(FD_MUTEX, "done waiting on td->mutex\n"); |
db1cd90b JA |
1035 | |
1036 | /* | |
cdd18ad8 | 1037 | * the ->mutex mutex is now no longer used, close it to avoid |
db1cd90b JA |
1038 | * eating a file descriptor |
1039 | */ | |
cdd18ad8 | 1040 | fio_mutex_remove(td->mutex); |
db1cd90b | 1041 | |
d84f8d49 JA |
1042 | /* |
1043 | * May alter parameters that init_io_u() will use, so we need to | |
1044 | * do this first. | |
1045 | */ | |
1046 | if (init_iolog(td)) | |
1047 | goto err; | |
1048 | ||
ebac4655 | 1049 | if (init_io_u(td)) |
db1cd90b | 1050 | goto err; |
ebac4655 | 1051 | |
e8462bd8 JA |
1052 | if (td->o.verify_async && verify_async_init(td)) |
1053 | goto err; | |
1054 | ||
1055 | if (td->o.cpumask_set && fio_setaffinity(td->pid, td->o.cpumask) == -1) { | |
e1161c32 | 1056 | td_verror(td, errno, "cpu_set_affinity"); |
db1cd90b | 1057 | goto err; |
ebac4655 JA |
1058 | } |
1059 | ||
f484470b JA |
1060 | /* |
1061 | * If we have a gettimeofday() thread, make sure we exclude that | |
1062 | * thread from this job | |
1063 | */ | |
be4ecfdf | 1064 | if (td->o.gtod_cpu) { |
be4ecfdf | 1065 | fio_cpu_clear(&td->o.cpumask, td->o.gtod_cpu); |
e8462bd8 | 1066 | if (fio_setaffinity(td->pid, td->o.cpumask) == -1) { |
be4ecfdf JA |
1067 | td_verror(td, errno, "cpu_set_affinity"); |
1068 | goto err; | |
1069 | } | |
1070 | } | |
1071 | ||
ac684785 | 1072 | if (td->ioprio_set) { |
ebac4655 | 1073 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { |
e1161c32 | 1074 | td_verror(td, errno, "ioprio_set"); |
db1cd90b | 1075 | goto err; |
ebac4655 JA |
1076 | } |
1077 | } | |
1078 | ||
a696fa2a JA |
1079 | if (td->o.cgroup_weight && cgroup_setup(td)) |
1080 | goto err; | |
1081 | ||
2dc1bbeb | 1082 | if (nice(td->o.nice) == -1) { |
e1161c32 | 1083 | td_verror(td, errno, "nice"); |
db1cd90b | 1084 | goto err; |
b6f4d880 JA |
1085 | } |
1086 | ||
2dc1bbeb | 1087 | if (td->o.ioscheduler && switch_ioscheduler(td)) |
db1cd90b | 1088 | goto err; |
37f56873 | 1089 | |
2dc1bbeb | 1090 | if (!td->o.create_serialize && setup_files(td)) |
ebac4655 JA |
1091 | goto err; |
1092 | ||
7d6c5283 JA |
1093 | if (td_io_init(td)) |
1094 | goto err; | |
1095 | ||
68727076 JA |
1096 | if (init_random_map(td)) |
1097 | goto err; | |
1098 | ||
2dc1bbeb | 1099 | if (td->o.exec_prerun) { |
398e8c4d | 1100 | if (exec_string(td->o.exec_prerun)) |
69cfd7e0 JA |
1101 | goto err; |
1102 | } | |
4e0ba8af | 1103 | |
afad68f7 ZY |
1104 | if (td->o.pre_read) { |
1105 | if (pre_read_files(td) < 0) | |
1106 | goto err; | |
1107 | } | |
1108 | ||
69008999 | 1109 | fio_gettime(&td->epoch, NULL); |
756867bd | 1110 | getrusage(RUSAGE_SELF, &td->ts.ru_start); |
69008999 JA |
1111 | |
1112 | runtime[0] = runtime[1] = 0; | |
a978ba68 | 1113 | clear_state = 0; |
492158cf | 1114 | while (keep_running(td)) { |
02bcaa8c | 1115 | fio_gettime(&td->start, NULL); |
c5d3d790 JA |
1116 | memcpy(&td->ts.stat_sample_time[0], &td->start, |
1117 | sizeof(td->start)); | |
1118 | memcpy(&td->ts.stat_sample_time[1], &td->start, | |
1119 | sizeof(td->start)); | |
993bf48b | 1120 | memcpy(&td->tv_cache, &td->start, sizeof(td->start)); |
ebac4655 | 1121 | |
cf2dd8d6 | 1122 | if (td->o.ratemin[0] || td->o.ratemin[1]) |
5ec10eaa JA |
1123 | memcpy(&td->lastrate, &td->ts.stat_sample_time, |
1124 | sizeof(td->lastrate)); | |
ebac4655 | 1125 | |
df9c26b1 JA |
1126 | if (clear_state) |
1127 | clear_io_state(td); | |
a978ba68 | 1128 | |
ebac4655 JA |
1129 | prune_io_piece_log(td); |
1130 | ||
ba0fbe10 | 1131 | do_io(td); |
ebac4655 | 1132 | |
a978ba68 JA |
1133 | clear_state = 1; |
1134 | ||
38d77cae | 1135 | if (td_read(td) && td->io_bytes[DDIR_READ]) { |
0a4957c6 | 1136 | elapsed = utime_since_now(&td->start); |
38d77cae JA |
1137 | runtime[DDIR_READ] += elapsed; |
1138 | } | |
1139 | if (td_write(td) && td->io_bytes[DDIR_WRITE]) { | |
0a4957c6 | 1140 | elapsed = utime_since_now(&td->start); |
38d77cae JA |
1141 | runtime[DDIR_WRITE] += elapsed; |
1142 | } | |
5ec10eaa | 1143 | |
ebac4655 JA |
1144 | if (td->error || td->terminate) |
1145 | break; | |
1146 | ||
e84c73a8 SL |
1147 | if (!td->o.do_verify || |
1148 | td->o.verify == VERIFY_NONE || | |
b67740d3 | 1149 | (td->io_ops->flags & FIO_UNIDIR)) |
ebac4655 JA |
1150 | continue; |
1151 | ||
df9c26b1 | 1152 | clear_io_state(td); |
a978ba68 | 1153 | |
02bcaa8c | 1154 | fio_gettime(&td->start, NULL); |
ebac4655 JA |
1155 | |
1156 | do_verify(td); | |
1157 | ||
69008999 | 1158 | runtime[DDIR_READ] += utime_since_now(&td->start); |
ebac4655 JA |
1159 | |
1160 | if (td->error || td->terminate) | |
1161 | break; | |
1162 | } | |
1163 | ||
36dff966 | 1164 | update_rusage_stat(td); |
47f0cc48 YHJT |
1165 | td->ts.runtime[0] = (runtime[0] + 999) / 1000; |
1166 | td->ts.runtime[1] = (runtime[1] + 999) / 1000; | |
756867bd JA |
1167 | td->ts.total_run_time = mtime_since_now(&td->epoch); |
1168 | td->ts.io_bytes[0] = td->io_bytes[0]; | |
1169 | td->ts.io_bytes[1] = td->io_bytes[1]; | |
1170 | ||
d2bbb8d9 | 1171 | fio_mutex_down(writeout_mutex); |
e3cedca7 | 1172 | if (td->ts.bw_log) { |
f484470b JA |
1173 | if (td->o.bw_log_file) { |
1174 | finish_log_named(td, td->ts.bw_log, | |
1175 | td->o.bw_log_file, "bw"); | |
1176 | } else | |
e3cedca7 JA |
1177 | finish_log(td, td->ts.bw_log, "bw"); |
1178 | } | |
1179 | if (td->ts.slat_log) { | |
f484470b JA |
1180 | if (td->o.lat_log_file) { |
1181 | finish_log_named(td, td->ts.slat_log, | |
1ca95738 | 1182 | td->o.lat_log_file, "slat"); |
f484470b | 1183 | } else |
e3cedca7 JA |
1184 | finish_log(td, td->ts.slat_log, "slat"); |
1185 | } | |
1186 | if (td->ts.clat_log) { | |
f484470b JA |
1187 | if (td->o.lat_log_file) { |
1188 | finish_log_named(td, td->ts.clat_log, | |
1189 | td->o.lat_log_file, "clat"); | |
1190 | } else | |
e3cedca7 JA |
1191 | finish_log(td, td->ts.clat_log, "clat"); |
1192 | } | |
d2bbb8d9 | 1193 | fio_mutex_up(writeout_mutex); |
398e8c4d JA |
1194 | if (td->o.exec_postrun) |
1195 | exec_string(td->o.exec_postrun); | |
ebac4655 JA |
1196 | |
1197 | if (exitall_on_terminate) | |
390c40e2 | 1198 | terminate_threads(td->groupid); |
ebac4655 JA |
1199 | |
1200 | err: | |
5bf13a5a | 1201 | if (td->error) |
5921e80c | 1202 | printf("fio: pid=%d, err=%d/%s\n", (int) td->pid, td->error, |
5ec10eaa | 1203 | td->verror); |
b4707ed7 JA |
1204 | |
1205 | if (td->o.verify_async) | |
1206 | verify_async_exit(td); | |
1207 | ||
24ffd2c2 | 1208 | close_and_free_files(td); |
2866c82d | 1209 | close_ioengine(td); |
ebac4655 | 1210 | cleanup_io_u(td); |
a696fa2a | 1211 | cgroup_shutdown(td); |
f29b25a3 | 1212 | |
d2ce18b5 JA |
1213 | if (td->o.cpumask_set) { |
1214 | int ret = fio_cpuset_exit(&td->o.cpumask); | |
1215 | ||
1216 | td_verror(td, ret, "fio_cpuset_exit"); | |
1217 | } | |
4e78e405 | 1218 | |
f29b25a3 JA |
1219 | /* |
1220 | * do this very late, it will log file closing as well | |
1221 | */ | |
1222 | if (td->o.write_iolog_file) | |
1223 | write_iolog_close(td); | |
1224 | ||
d23bb327 | 1225 | options_mem_free(td); |
ebac4655 | 1226 | td_set_runstate(td, TD_EXITED); |
43d76807 | 1227 | return (void *) (unsigned long) td->error; |
ebac4655 JA |
1228 | } |
1229 | ||
906c8d75 JA |
1230 | /* |
1231 | * We cannot pass the td data into a forked process, so attach the td and | |
1232 | * pass it to the thread worker. | |
1233 | */ | |
a6418147 | 1234 | static int fork_main(int shmid, int offset) |
ebac4655 JA |
1235 | { |
1236 | struct thread_data *td; | |
a6418147 | 1237 | void *data, *ret; |
ebac4655 JA |
1238 | |
1239 | data = shmat(shmid, NULL, 0); | |
1240 | if (data == (void *) -1) { | |
a6418147 JA |
1241 | int __err = errno; |
1242 | ||
ebac4655 | 1243 | perror("shmat"); |
a6418147 | 1244 | return __err; |
ebac4655 JA |
1245 | } |
1246 | ||
1247 | td = data + offset * sizeof(struct thread_data); | |
a6418147 | 1248 | ret = thread_main(td); |
ebac4655 | 1249 | shmdt(data); |
43d76807 | 1250 | return (int) (unsigned long) ret; |
ebac4655 JA |
1251 | } |
1252 | ||
906c8d75 JA |
1253 | /* |
1254 | * Run over the job map and reap the threads that have exited, if any. | |
1255 | */ | |
ebac4655 JA |
1256 | static void reap_threads(int *nr_running, int *t_rate, int *m_rate) |
1257 | { | |
34572e28 | 1258 | struct thread_data *td; |
1f809d15 | 1259 | int i, cputhreads, realthreads, pending, status, ret; |
ebac4655 JA |
1260 | |
1261 | /* | |
1262 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
1263 | */ | |
1f809d15 | 1264 | realthreads = pending = cputhreads = 0; |
34572e28 | 1265 | for_each_td(td, i) { |
3707f45b | 1266 | int flags = 0; |
a2f77c9f | 1267 | |
84585003 JA |
1268 | /* |
1269 | * ->io_ops is NULL for a thread that has closed its | |
1270 | * io engine | |
1271 | */ | |
ba0fbe10 | 1272 | if (td->io_ops && !strcmp(td->io_ops->name, "cpuio")) |
b990b5c0 | 1273 | cputhreads++; |
1f809d15 JA |
1274 | else |
1275 | realthreads++; | |
b990b5c0 | 1276 | |
a1dedc1f AC |
1277 | if (!td->pid) { |
1278 | pending++; | |
1279 | continue; | |
1280 | } | |
1281 | if (td->runstate == TD_REAPED) | |
a2f77c9f | 1282 | continue; |
2dc1bbeb | 1283 | if (td->o.use_thread) { |
3707f45b JA |
1284 | if (td->runstate == TD_EXITED) { |
1285 | td_set_runstate(td, TD_REAPED); | |
1286 | goto reaped; | |
1287 | } | |
1288 | continue; | |
1289 | } | |
a2f77c9f JA |
1290 | |
1291 | flags = WNOHANG; | |
1292 | if (td->runstate == TD_EXITED) | |
1293 | flags = 0; | |
1294 | ||
1295 | /* | |
1296 | * check if someone quit or got killed in an unusual way | |
1297 | */ | |
1298 | ret = waitpid(td->pid, &status, flags); | |
3707f45b | 1299 | if (ret < 0) { |
a2f77c9f | 1300 | if (errno == ECHILD) { |
5921e80c JA |
1301 | log_err("fio: pid=%d disappeared %d\n", |
1302 | (int) td->pid, td->runstate); | |
a2f77c9f JA |
1303 | td_set_runstate(td, TD_REAPED); |
1304 | goto reaped; | |
1305 | } | |
1306 | perror("waitpid"); | |
1307 | } else if (ret == td->pid) { | |
1308 | if (WIFSIGNALED(status)) { | |
fab6aa71 JA |
1309 | int sig = WTERMSIG(status); |
1310 | ||
d9244941 | 1311 | if (sig != SIGQUIT) |
5ec10eaa | 1312 | log_err("fio: pid=%d, got signal=%d\n", |
5921e80c | 1313 | (int) td->pid, sig); |
fab6aa71 JA |
1314 | td_set_runstate(td, TD_REAPED); |
1315 | goto reaped; | |
1316 | } | |
a2f77c9f JA |
1317 | if (WIFEXITED(status)) { |
1318 | if (WEXITSTATUS(status) && !td->error) | |
1319 | td->error = WEXITSTATUS(status); | |
a2f77c9f | 1320 | |
a2f77c9f JA |
1321 | td_set_runstate(td, TD_REAPED); |
1322 | goto reaped; | |
a6418147 JA |
1323 | } |
1324 | } | |
ebac4655 | 1325 | |
a2f77c9f JA |
1326 | /* |
1327 | * thread is not dead, continue | |
1328 | */ | |
e48676ba | 1329 | pending++; |
a2f77c9f | 1330 | continue; |
fab6aa71 | 1331 | reaped: |
ebac4655 | 1332 | (*nr_running)--; |
581e7141 JA |
1333 | (*m_rate) -= (td->o.ratemin[0] + td->o.ratemin[1]); |
1334 | (*t_rate) -= (td->o.rate[0] + td->o.rate[1]); | |
6f88bcb0 JA |
1335 | if (!td->pid) |
1336 | pending--; | |
a2f77c9f JA |
1337 | |
1338 | if (td->error) | |
1339 | exit_value++; | |
d3eeeabc JA |
1340 | |
1341 | done_secs += mtime_since_now(&td->epoch) / 1000; | |
ebac4655 | 1342 | } |
b990b5c0 | 1343 | |
1f809d15 | 1344 | if (*nr_running == cputhreads && !pending && realthreads) |
390c40e2 | 1345 | terminate_threads(TERMINATE_ALL); |
ebac4655 JA |
1346 | } |
1347 | ||
be4ecfdf JA |
1348 | static void *gtod_thread_main(void *data) |
1349 | { | |
1350 | fio_mutex_up(startup_mutex); | |
1351 | ||
1352 | /* | |
1353 | * As long as we have jobs around, update the clock. It would be nice | |
1354 | * to have some way of NOT hammering that CPU with gettimeofday(), | |
1355 | * but I'm not sure what to use outside of a simple CPU nop to relax | |
1356 | * it - we don't want to lose precision. | |
1357 | */ | |
1358 | while (threads) { | |
1359 | fio_gtod_update(); | |
1360 | nop; | |
1361 | } | |
1362 | ||
1363 | return NULL; | |
1364 | } | |
1365 | ||
1366 | static int fio_start_gtod_thread(void) | |
1367 | { | |
f42c153d JA |
1368 | int ret; |
1369 | ||
1370 | ret = pthread_create(>od_thread, NULL, gtod_thread_main, NULL); | |
1371 | if (ret) { | |
1372 | log_err("Can't create gtod thread: %s\n", strerror(ret)); | |
be4ecfdf JA |
1373 | return 1; |
1374 | } | |
f42c153d JA |
1375 | |
1376 | ret = pthread_detach(gtod_thread); | |
1377 | if (ret) { | |
1378 | log_err("Can't detatch gtod thread: %s\n", strerror(ret)); | |
be4ecfdf JA |
1379 | return 1; |
1380 | } | |
1381 | ||
29adda3c | 1382 | dprint(FD_MUTEX, "wait on startup_mutex\n"); |
be4ecfdf | 1383 | fio_mutex_down(startup_mutex); |
29adda3c | 1384 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); |
be4ecfdf JA |
1385 | return 0; |
1386 | } | |
1387 | ||
906c8d75 JA |
1388 | /* |
1389 | * Main function for kicking off and reaping jobs, as needed. | |
1390 | */ | |
ebac4655 JA |
1391 | static void run_threads(void) |
1392 | { | |
ebac4655 JA |
1393 | struct thread_data *td; |
1394 | unsigned long spent; | |
1395 | int i, todo, nr_running, m_rate, t_rate, nr_started; | |
fcb6ade2 | 1396 | |
2f9ade3c JA |
1397 | if (fio_pin_memory()) |
1398 | return; | |
ebac4655 | 1399 | |
be4ecfdf JA |
1400 | if (fio_gtod_offload && fio_start_gtod_thread()) |
1401 | return; | |
1402 | ||
c6ae0a5b | 1403 | if (!terse_output) { |
9cedf167 JA |
1404 | printf("Starting "); |
1405 | if (nr_thread) | |
5ec10eaa JA |
1406 | printf("%d thread%s", nr_thread, |
1407 | nr_thread > 1 ? "s" : ""); | |
9cedf167 JA |
1408 | if (nr_process) { |
1409 | if (nr_thread) | |
1410 | printf(" and "); | |
5ec10eaa JA |
1411 | printf("%d process%s", nr_process, |
1412 | nr_process > 1 ? "es" : ""); | |
9cedf167 JA |
1413 | } |
1414 | printf("\n"); | |
c6ae0a5b JA |
1415 | fflush(stdout); |
1416 | } | |
c04f7ec3 | 1417 | |
697a606c | 1418 | set_sig_handlers(); |
4efa970e | 1419 | |
ebac4655 JA |
1420 | todo = thread_number; |
1421 | nr_running = 0; | |
1422 | nr_started = 0; | |
1423 | m_rate = t_rate = 0; | |
1424 | ||
34572e28 | 1425 | for_each_td(td, i) { |
263e529f | 1426 | print_status_init(td->thread_number - 1); |
ebac4655 | 1427 | |
2dc1bbeb | 1428 | if (!td->o.create_serialize) { |
380cf265 | 1429 | init_disk_util(td); |
ebac4655 | 1430 | continue; |
380cf265 | 1431 | } |
ebac4655 JA |
1432 | |
1433 | /* | |
1434 | * do file setup here so it happens sequentially, | |
1435 | * we don't want X number of threads getting their | |
1436 | * client data interspersed on disk | |
1437 | */ | |
53cdc686 | 1438 | if (setup_files(td)) { |
5bf13a5a JA |
1439 | exit_value++; |
1440 | if (td->error) | |
5921e80c JA |
1441 | log_err("fio: pid=%d, err=%d/%s\n", |
1442 | (int) td->pid, td->error, td->verror); | |
ebac4655 JA |
1443 | td_set_runstate(td, TD_REAPED); |
1444 | todo--; | |
c69e8875 JA |
1445 | } else { |
1446 | struct fio_file *f; | |
1447 | unsigned int i; | |
1448 | ||
1449 | /* | |
1450 | * for sharing to work, each job must always open | |
1451 | * its own files. so close them, if we opened them | |
1452 | * for creation | |
1453 | */ | |
22a57ba8 JA |
1454 | for_each_file(td, f, i) { |
1455 | if (fio_file_open(f)) | |
1456 | td_io_close_file(td, f); | |
22a57ba8 | 1457 | } |
5ec10eaa | 1458 | } |
380cf265 JA |
1459 | |
1460 | init_disk_util(td); | |
ebac4655 JA |
1461 | } |
1462 | ||
a2f77c9f JA |
1463 | set_genesis_time(); |
1464 | ||
ebac4655 | 1465 | while (todo) { |
75154845 JA |
1466 | struct thread_data *map[MAX_JOBS]; |
1467 | struct timeval this_start; | |
1468 | int this_jobs = 0, left; | |
1469 | ||
ebac4655 JA |
1470 | /* |
1471 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
1472 | */ | |
34572e28 | 1473 | for_each_td(td, i) { |
ebac4655 JA |
1474 | if (td->runstate != TD_NOT_CREATED) |
1475 | continue; | |
1476 | ||
1477 | /* | |
1478 | * never got a chance to start, killed by other | |
1479 | * thread for some reason | |
1480 | */ | |
1481 | if (td->terminate) { | |
1482 | todo--; | |
1483 | continue; | |
1484 | } | |
1485 | ||
2dc1bbeb | 1486 | if (td->o.start_delay) { |
263e529f | 1487 | spent = mtime_since_genesis(); |
ebac4655 | 1488 | |
2dc1bbeb | 1489 | if (td->o.start_delay * 1000 > spent) |
ebac4655 JA |
1490 | continue; |
1491 | } | |
1492 | ||
6f88bcb0 JA |
1493 | if (td->o.stonewall && (nr_started || nr_running)) { |
1494 | dprint(FD_PROCESS, "%s: stonewall wait\n", | |
1495 | td->o.name); | |
ebac4655 | 1496 | break; |
6f88bcb0 | 1497 | } |
ebac4655 | 1498 | |
75154845 JA |
1499 | /* |
1500 | * Set state to created. Thread will transition | |
1501 | * to TD_INITIALIZED when it's done setting up. | |
1502 | */ | |
ebac4655 | 1503 | td_set_runstate(td, TD_CREATED); |
75154845 | 1504 | map[this_jobs++] = td; |
ebac4655 JA |
1505 | nr_started++; |
1506 | ||
2dc1bbeb | 1507 | if (td->o.use_thread) { |
f42c153d JA |
1508 | int ret; |
1509 | ||
ee56ad50 | 1510 | dprint(FD_PROCESS, "will pthread_create\n"); |
f42c153d JA |
1511 | ret = pthread_create(&td->thread, NULL, |
1512 | thread_main, td); | |
1513 | if (ret) { | |
1514 | log_err("pthread_create: %s\n", | |
1515 | strerror(ret)); | |
ebac4655 | 1516 | nr_started--; |
c8bb6faf | 1517 | break; |
ebac4655 | 1518 | } |
f42c153d JA |
1519 | ret = pthread_detach(td->thread); |
1520 | if (ret) | |
1521 | log_err("pthread_detach: %s", | |
1522 | strerror(ret)); | |
ebac4655 | 1523 | } else { |
5e1d306e | 1524 | pid_t pid; |
ee56ad50 | 1525 | dprint(FD_PROCESS, "will fork\n"); |
5e1d306e JA |
1526 | pid = fork(); |
1527 | if (!pid) { | |
a6418147 JA |
1528 | int ret = fork_main(shm_id, i); |
1529 | ||
5e1d306e JA |
1530 | _exit(ret); |
1531 | } else if (i == fio_debug_jobno) | |
1532 | *fio_debug_jobp = pid; | |
ebac4655 | 1533 | } |
29adda3c | 1534 | dprint(FD_MUTEX, "wait on startup_mutex\n"); |
656b1393 JA |
1535 | if (fio_mutex_down_timeout(startup_mutex, 10)) { |
1536 | log_err("fio: job startup hung? exiting.\n"); | |
1537 | terminate_threads(TERMINATE_ALL); | |
1538 | fio_abort = 1; | |
1539 | nr_started--; | |
1540 | break; | |
1541 | } | |
29adda3c | 1542 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); |
ebac4655 JA |
1543 | } |
1544 | ||
1545 | /* | |
75154845 JA |
1546 | * Wait for the started threads to transition to |
1547 | * TD_INITIALIZED. | |
ebac4655 | 1548 | */ |
02bcaa8c | 1549 | fio_gettime(&this_start, NULL); |
75154845 | 1550 | left = this_jobs; |
6ce15a32 | 1551 | while (left && !fio_abort) { |
75154845 JA |
1552 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) |
1553 | break; | |
1554 | ||
1555 | usleep(100000); | |
1556 | ||
1557 | for (i = 0; i < this_jobs; i++) { | |
1558 | td = map[i]; | |
1559 | if (!td) | |
1560 | continue; | |
b6f4d880 | 1561 | if (td->runstate == TD_INITIALIZED) { |
75154845 JA |
1562 | map[i] = NULL; |
1563 | left--; | |
b6f4d880 JA |
1564 | } else if (td->runstate >= TD_EXITED) { |
1565 | map[i] = NULL; | |
1566 | left--; | |
1567 | todo--; | |
1568 | nr_running++; /* work-around... */ | |
75154845 JA |
1569 | } |
1570 | } | |
1571 | } | |
1572 | ||
1573 | if (left) { | |
3b70d7e5 | 1574 | log_err("fio: %d jobs failed to start\n", left); |
75154845 JA |
1575 | for (i = 0; i < this_jobs; i++) { |
1576 | td = map[i]; | |
1577 | if (!td) | |
1578 | continue; | |
1579 | kill(td->pid, SIGTERM); | |
1580 | } | |
1581 | break; | |
1582 | } | |
1583 | ||
1584 | /* | |
b6f4d880 | 1585 | * start created threads (TD_INITIALIZED -> TD_RUNNING). |
75154845 | 1586 | */ |
34572e28 | 1587 | for_each_td(td, i) { |
75154845 | 1588 | if (td->runstate != TD_INITIALIZED) |
ebac4655 JA |
1589 | continue; |
1590 | ||
b29ee5b3 JA |
1591 | if (in_ramp_time(td)) |
1592 | td_set_runstate(td, TD_RAMP); | |
1593 | else | |
1594 | td_set_runstate(td, TD_RUNNING); | |
ebac4655 JA |
1595 | nr_running++; |
1596 | nr_started--; | |
581e7141 JA |
1597 | m_rate += td->o.ratemin[0] + td->o.ratemin[1]; |
1598 | t_rate += td->o.rate[0] + td->o.rate[1]; | |
75154845 | 1599 | todo--; |
cdd18ad8 | 1600 | fio_mutex_up(td->mutex); |
ebac4655 JA |
1601 | } |
1602 | ||
1603 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1604 | ||
1605 | if (todo) | |
1606 | usleep(100000); | |
1607 | } | |
1608 | ||
1609 | while (nr_running) { | |
1610 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1611 | usleep(10000); | |
1612 | } | |
1613 | ||
1614 | update_io_ticks(); | |
2f9ade3c | 1615 | fio_unpin_memory(); |
ebac4655 JA |
1616 | } |
1617 | ||
ebac4655 JA |
1618 | int main(int argc, char *argv[]) |
1619 | { | |
29d610e1 JA |
1620 | long ps; |
1621 | ||
2e5cdb11 JA |
1622 | sinit(); |
1623 | ||
dbe1125e JA |
1624 | /* |
1625 | * We need locale for number printing, if it isn't set then just | |
1626 | * go with the US format. | |
1627 | */ | |
1628 | if (!getenv("LC_NUMERIC")) | |
1629 | setlocale(LC_NUMERIC, "en_US"); | |
1630 | ||
29d610e1 JA |
1631 | ps = sysconf(_SC_PAGESIZE); |
1632 | if (ps < 0) { | |
1633 | log_err("Failed to get page size\n"); | |
1634 | return 1; | |
1635 | } | |
1636 | ||
cfc99db7 | 1637 | page_size = ps; |
29d610e1 JA |
1638 | page_mask = ps - 1; |
1639 | ||
74929ac2 JA |
1640 | fio_keywords_init(); |
1641 | ||
1642 | if (parse_options(argc, argv)) | |
1643 | return 1; | |
1644 | ||
1645 | if (!thread_number) | |
1646 | return 0; | |
1647 | ||
bb3884d8 JA |
1648 | if (write_bw_log) { |
1649 | setup_log(&agg_io_log[DDIR_READ]); | |
1650 | setup_log(&agg_io_log[DDIR_WRITE]); | |
1651 | } | |
1652 | ||
cdd18ad8 | 1653 | startup_mutex = fio_mutex_init(0); |
d2bbb8d9 | 1654 | writeout_mutex = fio_mutex_init(1); |
07739b57 | 1655 | |
a2f77c9f JA |
1656 | set_genesis_time(); |
1657 | ||
ccb0fa24 | 1658 | status_timer_arm(); |
ebac4655 JA |
1659 | |
1660 | run_threads(); | |
6ce15a32 | 1661 | |
bb3884d8 | 1662 | if (!fio_abort) { |
6ce15a32 | 1663 | show_run_stats(); |
bb3884d8 | 1664 | if (write_bw_log) { |
5ec10eaa JA |
1665 | __finish_log(agg_io_log[DDIR_READ], "agg-read_bw.log"); |
1666 | __finish_log(agg_io_log[DDIR_WRITE], | |
1667 | "agg-write_bw.log"); | |
bb3884d8 JA |
1668 | } |
1669 | } | |
ebac4655 | 1670 | |
cdd18ad8 | 1671 | fio_mutex_remove(startup_mutex); |
d2bbb8d9 | 1672 | fio_mutex_remove(writeout_mutex); |
437c9b71 | 1673 | return exit_value; |
ebac4655 | 1674 | } |