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" |
07b3232d | 43 | #include "profile.h" |
ebac4655 | 44 | |
cfc99db7 JA |
45 | unsigned long page_mask; |
46 | unsigned long page_size; | |
d529ee19 JA |
47 | |
48 | #define PAGE_ALIGN(buf) \ | |
29d610e1 | 49 | (char *) (((unsigned long) (buf) + page_mask) & ~page_mask) |
ebac4655 JA |
50 | |
51 | int groupid = 0; | |
52 | int thread_number = 0; | |
9cedf167 JA |
53 | int nr_process = 0; |
54 | int nr_thread = 0; | |
ebac4655 | 55 | int shm_id = 0; |
53cdc686 | 56 | int temp_stall_ts; |
d3eeeabc | 57 | unsigned long done_secs = 0; |
ebac4655 | 58 | |
cdd18ad8 | 59 | static struct fio_mutex *startup_mutex; |
d2bbb8d9 | 60 | static struct fio_mutex *writeout_mutex; |
6ce15a32 | 61 | static volatile int fio_abort; |
437c9b71 | 62 | static int exit_value; |
ccb0fa24 | 63 | static struct itimerval itimer; |
be4ecfdf | 64 | static pthread_t gtod_thread; |
dae5341c | 65 | static struct flist_head *cgroup_list; |
6adb38a1 | 66 | static char *cgroup_mnt; |
ebac4655 | 67 | |
bb3884d8 JA |
68 | struct io_log *agg_io_log[2]; |
69 | ||
ebac4655 | 70 | #define TERMINATE_ALL (-1) |
75154845 | 71 | #define JOB_START_TIMEOUT (5 * 1000) |
ebac4655 | 72 | |
b29ee5b3 | 73 | void td_set_runstate(struct thread_data *td, int runstate) |
6ce15a32 | 74 | { |
d8fab1b6 JA |
75 | if (td->runstate == runstate) |
76 | return; | |
77 | ||
5921e80c JA |
78 | dprint(FD_PROCESS, "pid=%d: runstate %d -> %d\n", (int) td->pid, |
79 | td->runstate, runstate); | |
6ce15a32 JA |
80 | td->runstate = runstate; |
81 | } | |
82 | ||
390c40e2 | 83 | static void terminate_threads(int group_id) |
ebac4655 | 84 | { |
34572e28 | 85 | struct thread_data *td; |
ebac4655 JA |
86 | int i; |
87 | ||
6f88bcb0 JA |
88 | dprint(FD_PROCESS, "terminate group_id=%d\n", group_id); |
89 | ||
34572e28 | 90 | for_each_td(td, i) { |
ebac4655 | 91 | if (group_id == TERMINATE_ALL || groupid == td->groupid) { |
6f88bcb0 | 92 | dprint(FD_PROCESS, "setting terminate on %s/%d\n", |
5921e80c | 93 | td->o.name, (int) td->pid); |
ad830ed7 JA |
94 | td->terminate = 1; |
95 | td->o.start_delay = 0; | |
96 | ||
7a93ab19 JA |
97 | /* |
98 | * if the thread is running, just let it exit | |
99 | */ | |
100 | if (td->runstate < TD_RUNNING) | |
c1302d44 | 101 | kill(td->pid, SIGQUIT); |
f63f7f3b JA |
102 | else { |
103 | struct ioengine_ops *ops = td->io_ops; | |
104 | ||
105 | if (ops && (ops->flags & FIO_SIGQUIT)) | |
106 | kill(td->pid, SIGQUIT); | |
107 | } | |
ebac4655 JA |
108 | } |
109 | } | |
110 | } | |
111 | ||
ccb0fa24 JA |
112 | static void status_timer_arm(void) |
113 | { | |
114 | itimer.it_value.tv_sec = 0; | |
115 | itimer.it_value.tv_usec = DISK_UTIL_MSEC * 1000; | |
116 | setitimer(ITIMER_REAL, &itimer, NULL); | |
117 | } | |
118 | ||
02444ad1 | 119 | static void sig_alrm(int fio_unused sig) |
ebac4655 | 120 | { |
697a606c | 121 | if (threads) { |
5ec10eaa | 122 | update_io_ticks(); |
5ec10eaa | 123 | print_thread_status(); |
ccb0fa24 | 124 | status_timer_arm(); |
697a606c JA |
125 | } |
126 | } | |
127 | ||
5f8f9726 JA |
128 | /* |
129 | * Happens on thread runs with ctrl-c, ignore our own SIGQUIT | |
130 | */ | |
131 | static void sig_quit(int sig) | |
132 | { | |
133 | } | |
134 | ||
697a606c JA |
135 | static void sig_int(int sig) |
136 | { | |
137 | if (threads) { | |
4ceb30d4 | 138 | log_info("\nfio: terminating on signal %d\n", sig); |
5ec10eaa JA |
139 | fflush(stdout); |
140 | terminate_threads(TERMINATE_ALL); | |
ebac4655 JA |
141 | } |
142 | } | |
143 | ||
02444ad1 | 144 | static void sig_ill(int fio_unused sig) |
de79c915 JA |
145 | { |
146 | if (!threads) | |
147 | return; | |
148 | ||
149 | log_err("fio: illegal instruction. your cpu does not support " | |
150 | "the sse4.2 instruction for crc32c\n"); | |
151 | terminate_threads(TERMINATE_ALL); | |
152 | exit(4); | |
153 | } | |
154 | ||
697a606c JA |
155 | static void set_sig_handlers(void) |
156 | { | |
157 | struct sigaction act; | |
158 | ||
159 | memset(&act, 0, sizeof(act)); | |
160 | act.sa_handler = sig_alrm; | |
161 | act.sa_flags = SA_RESTART; | |
162 | sigaction(SIGALRM, &act, NULL); | |
163 | ||
164 | memset(&act, 0, sizeof(act)); | |
165 | act.sa_handler = sig_int; | |
166 | act.sa_flags = SA_RESTART; | |
167 | sigaction(SIGINT, &act, NULL); | |
de79c915 JA |
168 | |
169 | memset(&act, 0, sizeof(act)); | |
170 | act.sa_handler = sig_ill; | |
171 | act.sa_flags = SA_RESTART; | |
172 | sigaction(SIGILL, &act, NULL); | |
5f8f9726 JA |
173 | |
174 | memset(&act, 0, sizeof(act)); | |
175 | act.sa_handler = sig_quit; | |
176 | act.sa_flags = SA_RESTART; | |
177 | sigaction(SIGQUIT, &act, NULL); | |
697a606c JA |
178 | } |
179 | ||
906c8d75 JA |
180 | /* |
181 | * Check if we are above the minimum rate given. | |
182 | */ | |
581e7141 JA |
183 | static int __check_min_rate(struct thread_data *td, struct timeval *now, |
184 | enum td_ddir ddir) | |
ebac4655 | 185 | { |
0904200b | 186 | unsigned long long bytes = 0; |
4e991c23 | 187 | unsigned long iops = 0; |
ebac4655 JA |
188 | unsigned long spent; |
189 | unsigned long rate; | |
581e7141 JA |
190 | unsigned int ratemin = 0; |
191 | unsigned int rate_iops = 0; | |
192 | unsigned int rate_iops_min = 0; | |
ebac4655 | 193 | |
d982d873 JA |
194 | if (!td->o.ratemin[ddir] && !td->o.rate_iops_min[ddir]) |
195 | return 0; | |
196 | ||
ebac4655 JA |
197 | /* |
198 | * allow a 2 second settle period in the beginning | |
199 | */ | |
200 | if (mtime_since(&td->start, now) < 2000) | |
201 | return 0; | |
202 | ||
581e7141 JA |
203 | iops += td->io_blocks[ddir]; |
204 | bytes += td->this_io_bytes[ddir]; | |
205 | ratemin += td->o.ratemin[ddir]; | |
206 | rate_iops += td->o.rate_iops[ddir]; | |
207 | rate_iops_min += td->o.rate_iops_min[ddir]; | |
0904200b | 208 | |
ebac4655 JA |
209 | /* |
210 | * if rate blocks is set, sample is running | |
211 | */ | |
581e7141 JA |
212 | if (td->rate_bytes[ddir] || td->rate_blocks[ddir]) { |
213 | spent = mtime_since(&td->lastrate[ddir], now); | |
2dc1bbeb | 214 | if (spent < td->o.ratecycle) |
ebac4655 JA |
215 | return 0; |
216 | ||
581e7141 | 217 | if (td->o.rate[ddir]) { |
4e991c23 JA |
218 | /* |
219 | * check bandwidth specified rate | |
220 | */ | |
581e7141 | 221 | if (bytes < td->rate_bytes[ddir]) { |
5ec10eaa | 222 | log_err("%s: min rate %u not met\n", td->o.name, |
581e7141 | 223 | ratemin); |
4e991c23 JA |
224 | return 1; |
225 | } else { | |
581e7141 JA |
226 | rate = ((bytes - td->rate_bytes[ddir]) * 1000) / spent; |
227 | if (rate < ratemin || | |
228 | bytes < td->rate_bytes[ddir]) { | |
5ec10eaa | 229 | log_err("%s: min rate %u not met, got" |
b22989b9 | 230 | " %luKB/sec\n", td->o.name, |
581e7141 | 231 | ratemin, rate); |
4e991c23 JA |
232 | return 1; |
233 | } | |
234 | } | |
413dd459 | 235 | } else { |
4e991c23 JA |
236 | /* |
237 | * checks iops specified rate | |
238 | */ | |
581e7141 | 239 | if (iops < rate_iops) { |
5ec10eaa | 240 | log_err("%s: min iops rate %u not met\n", |
581e7141 | 241 | td->o.name, rate_iops); |
413dd459 | 242 | return 1; |
4e991c23 | 243 | } else { |
581e7141 JA |
244 | rate = ((iops - td->rate_blocks[ddir]) * 1000) / spent; |
245 | if (rate < rate_iops_min || | |
246 | iops < td->rate_blocks[ddir]) { | |
5ec10eaa JA |
247 | log_err("%s: min iops rate %u not met," |
248 | " got %lu\n", td->o.name, | |
581e7141 | 249 | rate_iops_min, rate); |
4e991c23 | 250 | } |
413dd459 | 251 | } |
ebac4655 JA |
252 | } |
253 | } | |
254 | ||
581e7141 JA |
255 | td->rate_bytes[ddir] = bytes; |
256 | td->rate_blocks[ddir] = iops; | |
257 | memcpy(&td->lastrate[ddir], now, sizeof(*now)); | |
ebac4655 JA |
258 | return 0; |
259 | } | |
260 | ||
581e7141 JA |
261 | static int check_min_rate(struct thread_data *td, struct timeval *now, |
262 | unsigned long *bytes_done) | |
263 | { | |
264 | int ret = 0; | |
265 | ||
266 | if (bytes_done[0]) | |
267 | ret |= __check_min_rate(td, now, 0); | |
268 | if (bytes_done[1]) | |
269 | ret |= __check_min_rate(td, now, 1); | |
270 | ||
271 | return ret; | |
272 | } | |
273 | ||
ebac4655 JA |
274 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) |
275 | { | |
2dc1bbeb | 276 | if (!td->o.timeout) |
ebac4655 | 277 | return 0; |
2dc1bbeb | 278 | if (mtime_since(&td->epoch, t) >= td->o.timeout * 1000) |
ebac4655 JA |
279 | return 1; |
280 | ||
281 | return 0; | |
282 | } | |
283 | ||
906c8d75 JA |
284 | /* |
285 | * When job exits, we can cancel the in-flight IO if we are using async | |
286 | * io. Attempt to do so. | |
287 | */ | |
ebac4655 JA |
288 | static void cleanup_pending_aio(struct thread_data *td) |
289 | { | |
01743ee1 | 290 | struct flist_head *entry, *n; |
ebac4655 JA |
291 | struct io_u *io_u; |
292 | int r; | |
293 | ||
294 | /* | |
295 | * get immediately available events, if any | |
296 | */ | |
581e7141 | 297 | r = io_u_queued_complete(td, 0, NULL); |
b2fdda43 JA |
298 | if (r < 0) |
299 | return; | |
ebac4655 JA |
300 | |
301 | /* | |
302 | * now cancel remaining active events | |
303 | */ | |
2866c82d | 304 | if (td->io_ops->cancel) { |
01743ee1 JA |
305 | flist_for_each_safe(entry, n, &td->io_u_busylist) { |
306 | io_u = flist_entry(entry, struct io_u, list); | |
ebac4655 | 307 | |
0c6e7517 JA |
308 | /* |
309 | * if the io_u isn't in flight, then that generally | |
310 | * means someone leaked an io_u. complain but fix | |
311 | * it up, so we don't stall here. | |
312 | */ | |
313 | if ((io_u->flags & IO_U_F_FLIGHT) == 0) { | |
314 | log_err("fio: non-busy IO on busy list\n"); | |
ebac4655 | 315 | put_io_u(td, io_u); |
0c6e7517 JA |
316 | } else { |
317 | r = td->io_ops->cancel(td, io_u); | |
318 | if (!r) | |
319 | put_io_u(td, io_u); | |
320 | } | |
ebac4655 JA |
321 | } |
322 | } | |
323 | ||
97601024 | 324 | if (td->cur_depth) |
581e7141 | 325 | r = io_u_queued_complete(td, td->cur_depth, NULL); |
ebac4655 JA |
326 | } |
327 | ||
858a3d47 JA |
328 | /* |
329 | * Helper to handle the final sync of a file. Works just like the normal | |
330 | * io path, just does everything sync. | |
331 | */ | |
332 | static int fio_io_sync(struct thread_data *td, struct fio_file *f) | |
333 | { | |
334 | struct io_u *io_u = __get_io_u(td); | |
858a3d47 JA |
335 | int ret; |
336 | ||
337 | if (!io_u) | |
338 | return 1; | |
339 | ||
340 | io_u->ddir = DDIR_SYNC; | |
341 | io_u->file = f; | |
342 | ||
343 | if (td_io_prep(td, io_u)) { | |
344 | put_io_u(td, io_u); | |
345 | return 1; | |
346 | } | |
347 | ||
755200a3 | 348 | requeue: |
858a3d47 | 349 | ret = td_io_queue(td, io_u); |
36167d82 | 350 | if (ret < 0) { |
e1161c32 | 351 | td_verror(td, io_u->error, "td_io_queue"); |
858a3d47 | 352 | put_io_u(td, io_u); |
858a3d47 | 353 | return 1; |
36167d82 | 354 | } else if (ret == FIO_Q_QUEUED) { |
581e7141 | 355 | if (io_u_queued_complete(td, 1, NULL) < 0) |
36167d82 | 356 | return 1; |
36167d82 JA |
357 | } else if (ret == FIO_Q_COMPLETED) { |
358 | if (io_u->error) { | |
e1161c32 | 359 | td_verror(td, io_u->error, "td_io_queue"); |
36167d82 JA |
360 | return 1; |
361 | } | |
858a3d47 | 362 | |
581e7141 | 363 | if (io_u_sync_complete(td, io_u, NULL) < 0) |
b2fdda43 | 364 | return 1; |
755200a3 JA |
365 | } else if (ret == FIO_Q_BUSY) { |
366 | if (td_io_commit(td)) | |
367 | return 1; | |
368 | goto requeue; | |
858a3d47 JA |
369 | } |
370 | ||
371 | return 0; | |
372 | } | |
373 | ||
993bf48b JA |
374 | static inline void update_tv_cache(struct thread_data *td) |
375 | { | |
376 | if ((++td->tv_cache_nr & td->tv_cache_mask) == td->tv_cache_mask) | |
377 | fio_gettime(&td->tv_cache, NULL); | |
378 | } | |
379 | ||
f2bba182 RR |
380 | static int break_on_this_error(struct thread_data *td, int *retptr) |
381 | { | |
382 | int ret = *retptr; | |
383 | ||
384 | if (ret < 0 || td->error) { | |
385 | int err; | |
386 | ||
1ec99eea JA |
387 | if (!td->o.continue_on_error) |
388 | return 1; | |
f2bba182 RR |
389 | |
390 | if (ret < 0) | |
391 | err = -ret; | |
392 | else | |
393 | err = td->error; | |
394 | ||
f2bba182 RR |
395 | if (td_non_fatal_error(err)) { |
396 | /* | |
397 | * Continue with the I/Os in case of | |
398 | * a non fatal error. | |
399 | */ | |
8a4f8a24 | 400 | update_error_count(td, err); |
f2bba182 RR |
401 | td_clear_error(td); |
402 | *retptr = 0; | |
403 | return 0; | |
8a4f8a24 RR |
404 | } else if (td->o.fill_device && err == ENOSPC) { |
405 | /* | |
406 | * We expect to hit this error if | |
407 | * fill_device option is set. | |
408 | */ | |
409 | td_clear_error(td); | |
410 | td->terminate = 1; | |
411 | return 1; | |
f2bba182 RR |
412 | } else { |
413 | /* | |
414 | * Stop the I/O in case of a fatal | |
415 | * error. | |
416 | */ | |
8a4f8a24 | 417 | update_error_count(td, err); |
f2bba182 RR |
418 | return 1; |
419 | } | |
420 | } | |
421 | ||
422 | return 0; | |
423 | } | |
424 | ||
906c8d75 | 425 | /* |
34403fb1 | 426 | * The main verify engine. Runs over the writes we previously submitted, |
906c8d75 JA |
427 | * reads the blocks back in, and checks the crc/md5 of the data. |
428 | */ | |
1e97cce9 | 429 | static void do_verify(struct thread_data *td) |
ebac4655 | 430 | { |
53cdc686 | 431 | struct fio_file *f; |
36167d82 | 432 | struct io_u *io_u; |
af52b345 JA |
433 | int ret, min_events; |
434 | unsigned int i; | |
e5b401d4 | 435 | |
008618a8 JA |
436 | dprint(FD_VERIFY, "starting loop\n"); |
437 | ||
e5b401d4 JA |
438 | /* |
439 | * sync io first and invalidate cache, to make sure we really | |
440 | * read from disk. | |
441 | */ | |
442 | for_each_file(td, f, i) { | |
d6aed795 | 443 | if (!fio_file_open(f)) |
3d7b485f | 444 | continue; |
b2fdda43 JA |
445 | if (fio_io_sync(td, f)) |
446 | break; | |
447 | if (file_invalidate_cache(td, f)) | |
448 | break; | |
e5b401d4 | 449 | } |
ebac4655 | 450 | |
b2fdda43 JA |
451 | if (td->error) |
452 | return; | |
453 | ||
ebac4655 JA |
454 | td_set_runstate(td, TD_VERIFYING); |
455 | ||
36167d82 JA |
456 | io_u = NULL; |
457 | while (!td->terminate) { | |
4950421a | 458 | int ret2, full; |
5451792e | 459 | |
993bf48b JA |
460 | update_tv_cache(td); |
461 | ||
462 | if (runtime_exceeded(td, &td->tv_cache)) { | |
90a87d4b | 463 | td->terminate = 1; |
02bcaa8c | 464 | break; |
069c2918 | 465 | } |
02bcaa8c | 466 | |
dcb4830d RR |
467 | io_u = __get_io_u(td); |
468 | if (!io_u) | |
469 | break; | |
470 | ||
069c2918 JA |
471 | if (get_next_verify(td, io_u)) { |
472 | put_io_u(td, io_u); | |
ebac4655 | 473 | break; |
069c2918 | 474 | } |
ebac4655 | 475 | |
069c2918 JA |
476 | if (td_io_prep(td, io_u)) { |
477 | put_io_u(td, io_u); | |
53cdc686 | 478 | break; |
069c2918 | 479 | } |
d7762cf8 | 480 | |
e8462bd8 JA |
481 | if (td->o.verify_async) |
482 | io_u->end_io = verify_io_u_async; | |
483 | else | |
484 | io_u->end_io = verify_io_u; | |
53cdc686 | 485 | |
11786802 | 486 | ret = td_io_queue(td, io_u); |
36167d82 JA |
487 | switch (ret) { |
488 | case FIO_Q_COMPLETED: | |
f2bba182 | 489 | if (io_u->error) { |
22819ec2 | 490 | ret = -io_u->error; |
f2bba182 RR |
491 | clear_io_u(td, io_u); |
492 | } else if (io_u->resid) { | |
36167d82 | 493 | int bytes = io_u->xfer_buflen - io_u->resid; |
d460eb31 | 494 | struct fio_file *f = io_u->file; |
ebac4655 | 495 | |
9e9d164e JA |
496 | /* |
497 | * zero read, fail | |
498 | */ | |
499 | if (!bytes) { | |
41d9f0fd | 500 | td_verror(td, EIO, "full resid"); |
9e9d164e JA |
501 | put_io_u(td, io_u); |
502 | break; | |
503 | } | |
8400d9b2 | 504 | |
36167d82 JA |
505 | io_u->xfer_buflen = io_u->resid; |
506 | io_u->xfer_buf += bytes; | |
8400d9b2 JA |
507 | io_u->offset += bytes; |
508 | ||
30061b97 JA |
509 | td->ts.short_io_u[io_u->ddir]++; |
510 | ||
d460eb31 | 511 | if (io_u->offset == f->real_file_size) |
8400d9b2 JA |
512 | goto sync_done; |
513 | ||
11786802 JA |
514 | requeue_io_u(td, &io_u); |
515 | } else { | |
8400d9b2 | 516 | sync_done: |
581e7141 | 517 | ret = io_u_sync_complete(td, io_u, NULL); |
11786802 JA |
518 | if (ret < 0) |
519 | break; | |
36167d82 | 520 | } |
36167d82 JA |
521 | continue; |
522 | case FIO_Q_QUEUED: | |
523 | break; | |
755200a3 JA |
524 | case FIO_Q_BUSY: |
525 | requeue_io_u(td, &io_u); | |
5451792e JA |
526 | ret2 = td_io_commit(td); |
527 | if (ret2 < 0) | |
528 | ret = ret2; | |
755200a3 | 529 | break; |
36167d82 JA |
530 | default: |
531 | assert(ret < 0); | |
e1161c32 | 532 | td_verror(td, -ret, "td_io_queue"); |
ebac4655 JA |
533 | break; |
534 | } | |
535 | ||
f2bba182 | 536 | if (break_on_this_error(td, &ret)) |
3af6ef39 JA |
537 | break; |
538 | ||
ebac4655 | 539 | /* |
3af6ef39 JA |
540 | * if we can queue more, do so. but check if there are |
541 | * completed io_u's first. | |
ebac4655 | 542 | */ |
4950421a JA |
543 | full = queue_full(td) || ret == FIO_Q_BUSY; |
544 | if (full || !td->o.iodepth_batch_complete) { | |
6a96276c RR |
545 | min_events = min(td->o.iodepth_batch_complete, |
546 | td->cur_depth); | |
4950421a | 547 | if (full && !min_events) |
838bc709 | 548 | min_events = 1; |
e916b390 | 549 | |
4950421a JA |
550 | do { |
551 | /* | |
552 | * Reap required number of io units, if any, | |
553 | * and do the verification on them through | |
554 | * the callback handler | |
555 | */ | |
581e7141 | 556 | if (io_u_queued_complete(td, min_events, NULL) < 0) { |
4950421a JA |
557 | ret = -1; |
558 | break; | |
559 | } | |
560 | } while (full && (td->cur_depth > td->o.iodepth_low)); | |
561 | } | |
562 | if (ret < 0) | |
ebac4655 | 563 | break; |
36167d82 | 564 | } |
ebac4655 | 565 | |
c01c0395 JA |
566 | if (!td->error) { |
567 | min_events = td->cur_depth; | |
568 | ||
569 | if (min_events) | |
581e7141 | 570 | ret = io_u_queued_complete(td, min_events, NULL); |
c01c0395 | 571 | } else |
ebac4655 JA |
572 | cleanup_pending_aio(td); |
573 | ||
574 | td_set_runstate(td, TD_RUNNING); | |
008618a8 JA |
575 | |
576 | dprint(FD_VERIFY, "exiting loop\n"); | |
ebac4655 JA |
577 | } |
578 | ||
32cd46a0 | 579 | /* |
906c8d75 | 580 | * Main IO worker function. It retrieves io_u's to process and queues |
32cd46a0 JA |
581 | * and reaps them, checking for rate and errors along the way. |
582 | */ | |
ebac4655 JA |
583 | static void do_io(struct thread_data *td) |
584 | { | |
af52b345 JA |
585 | unsigned int i; |
586 | int ret = 0; | |
ebac4655 | 587 | |
b29ee5b3 JA |
588 | if (in_ramp_time(td)) |
589 | td_set_runstate(td, TD_RAMP); | |
590 | else | |
591 | td_set_runstate(td, TD_RUNNING); | |
5853e5a8 | 592 | |
457bf399 GO |
593 | while ( (td->o.read_iolog_file && !flist_empty(&td->io_log_list)) || |
594 | ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->o.size) ) { | |
97601024 | 595 | struct timeval comp_time; |
581e7141 | 596 | unsigned long bytes_done[2] = { 0, 0 }; |
84585003 | 597 | int min_evts = 0; |
ebac4655 | 598 | struct io_u *io_u; |
4950421a | 599 | int ret2, full; |
ebac4655 JA |
600 | |
601 | if (td->terminate) | |
602 | break; | |
603 | ||
993bf48b | 604 | update_tv_cache(td); |
97601024 | 605 | |
993bf48b | 606 | if (runtime_exceeded(td, &td->tv_cache)) { |
90a87d4b | 607 | td->terminate = 1; |
97601024 JA |
608 | break; |
609 | } | |
36167d82 | 610 | |
dcb4830d RR |
611 | io_u = get_io_u(td); |
612 | if (!io_u) | |
613 | break; | |
614 | ||
b67740d3 JA |
615 | /* |
616 | * Add verification end_io handler, if asked to verify | |
617 | * a previously written file. | |
618 | */ | |
715d2b3e JA |
619 | if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ && |
620 | !td_rw(td)) { | |
e8462bd8 JA |
621 | if (td->o.verify_async) |
622 | io_u->end_io = verify_io_u_async; | |
623 | else | |
624 | io_u->end_io = verify_io_u; | |
d8fab1b6 | 625 | td_set_runstate(td, TD_VERIFYING); |
b29ee5b3 JA |
626 | } else if (in_ramp_time(td)) |
627 | td_set_runstate(td, TD_RAMP); | |
628 | else | |
d8fab1b6 | 629 | td_set_runstate(td, TD_RUNNING); |
b67740d3 | 630 | |
11786802 | 631 | ret = td_io_queue(td, io_u); |
36167d82 JA |
632 | switch (ret) { |
633 | case FIO_Q_COMPLETED: | |
f2bba182 | 634 | if (io_u->error) { |
5451792e | 635 | ret = -io_u->error; |
f2bba182 RR |
636 | clear_io_u(td, io_u); |
637 | } else if (io_u->resid) { | |
36167d82 | 638 | int bytes = io_u->xfer_buflen - io_u->resid; |
d460eb31 | 639 | struct fio_file *f = io_u->file; |
36167d82 | 640 | |
9e9d164e JA |
641 | /* |
642 | * zero read, fail | |
643 | */ | |
644 | if (!bytes) { | |
41d9f0fd | 645 | td_verror(td, EIO, "full resid"); |
9e9d164e JA |
646 | put_io_u(td, io_u); |
647 | break; | |
648 | } | |
649 | ||
cec6b55d | 650 | io_u->xfer_buflen = io_u->resid; |
36167d82 | 651 | io_u->xfer_buf += bytes; |
5a7c5680 JA |
652 | io_u->offset += bytes; |
653 | ||
30061b97 JA |
654 | td->ts.short_io_u[io_u->ddir]++; |
655 | ||
d460eb31 | 656 | if (io_u->offset == f->real_file_size) |
5a7c5680 JA |
657 | goto sync_done; |
658 | ||
11786802 JA |
659 | requeue_io_u(td, &io_u); |
660 | } else { | |
5a7c5680 | 661 | sync_done: |
581e7141 JA |
662 | if (__should_check_rate(td, 0) || |
663 | __should_check_rate(td, 1)) | |
993bf48b JA |
664 | fio_gettime(&comp_time, NULL); |
665 | ||
581e7141 JA |
666 | ret = io_u_sync_complete(td, io_u, bytes_done); |
667 | if (ret < 0) | |
668 | break; | |
cec6b55d | 669 | } |
36167d82 JA |
670 | break; |
671 | case FIO_Q_QUEUED: | |
7e77dd02 JA |
672 | /* |
673 | * if the engine doesn't have a commit hook, | |
674 | * the io_u is really queued. if it does have such | |
675 | * a hook, it has to call io_u_queued() itself. | |
676 | */ | |
677 | if (td->io_ops->commit == NULL) | |
678 | io_u_queued(td, io_u); | |
36167d82 | 679 | break; |
755200a3 JA |
680 | case FIO_Q_BUSY: |
681 | requeue_io_u(td, &io_u); | |
5451792e JA |
682 | ret2 = td_io_commit(td); |
683 | if (ret2 < 0) | |
684 | ret = ret2; | |
755200a3 | 685 | break; |
36167d82 JA |
686 | default: |
687 | assert(ret < 0); | |
688 | put_io_u(td, io_u); | |
689 | break; | |
ebac4655 JA |
690 | } |
691 | ||
f2bba182 | 692 | if (break_on_this_error(td, &ret)) |
36167d82 JA |
693 | break; |
694 | ||
97601024 JA |
695 | /* |
696 | * See if we need to complete some commands | |
697 | */ | |
4950421a JA |
698 | full = queue_full(td) || ret == FIO_Q_BUSY; |
699 | if (full || !td->o.iodepth_batch_complete) { | |
6a96276c RR |
700 | min_evts = min(td->o.iodepth_batch_complete, |
701 | td->cur_depth); | |
4950421a | 702 | if (full && !min_evts) |
36167d82 | 703 | min_evts = 1; |
4950421a | 704 | |
581e7141 JA |
705 | if (__should_check_rate(td, 0) || |
706 | __should_check_rate(td, 1)) | |
993bf48b | 707 | fio_gettime(&comp_time, NULL); |
4950421a JA |
708 | |
709 | do { | |
581e7141 JA |
710 | ret = io_u_queued_complete(td, min_evts, bytes_done); |
711 | if (ret < 0) | |
4950421a JA |
712 | break; |
713 | ||
4950421a | 714 | } while (full && (td->cur_depth > td->o.iodepth_low)); |
ebac4655 JA |
715 | } |
716 | ||
4950421a JA |
717 | if (ret < 0) |
718 | break; | |
581e7141 | 719 | if (!(bytes_done[0] + bytes_done[1])) |
97601024 JA |
720 | continue; |
721 | ||
581e7141 JA |
722 | if (!in_ramp_time(td) && should_check_rate(td, bytes_done)) { |
723 | if (check_min_rate(td, &comp_time, bytes_done)) { | |
721938ae JA |
724 | if (exitall_on_terminate) |
725 | terminate_threads(td->groupid); | |
726 | td_verror(td, EIO, "check_min_rate"); | |
727 | break; | |
728 | } | |
ebac4655 JA |
729 | } |
730 | ||
2dc1bbeb | 731 | if (td->o.thinktime) { |
9c1f7434 JA |
732 | unsigned long long b; |
733 | ||
734 | b = td->io_blocks[0] + td->io_blocks[1]; | |
2dc1bbeb | 735 | if (!(b % td->o.thinktime_blocks)) { |
48097d5c JA |
736 | int left; |
737 | ||
2dc1bbeb | 738 | if (td->o.thinktime_spin) |
6dd6f2cd | 739 | usec_spin(td->o.thinktime_spin); |
48097d5c | 740 | |
2dc1bbeb | 741 | left = td->o.thinktime - td->o.thinktime_spin; |
48097d5c JA |
742 | if (left) |
743 | usec_sleep(td, left); | |
744 | } | |
9c1f7434 | 745 | } |
ebac4655 JA |
746 | } |
747 | ||
aa31f1f1 SL |
748 | if (td->o.fill_device && td->error == ENOSPC) { |
749 | td->error = 0; | |
750 | td->terminate = 1; | |
751 | } | |
4d2413c6 | 752 | if (!td->error) { |
3d7c391d JA |
753 | struct fio_file *f; |
754 | ||
c01c0395 JA |
755 | i = td->cur_depth; |
756 | if (i) | |
581e7141 | 757 | ret = io_u_queued_complete(td, i, NULL); |
ebac4655 | 758 | |
2dc1bbeb | 759 | if (should_fsync(td) && td->o.end_fsync) { |
84585003 | 760 | td_set_runstate(td, TD_FSYNCING); |
3d7b485f JA |
761 | |
762 | for_each_file(td, f, i) { | |
d6aed795 | 763 | if (!fio_file_open(f)) |
3d7b485f | 764 | continue; |
858a3d47 | 765 | fio_io_sync(td, f); |
3d7b485f | 766 | } |
84585003 | 767 | } |
c01c0395 JA |
768 | } else |
769 | cleanup_pending_aio(td); | |
2ba1c290 JA |
770 | |
771 | /* | |
772 | * stop job if we failed doing any IO | |
773 | */ | |
774 | if ((td->this_io_bytes[0] + td->this_io_bytes[1]) == 0) | |
775 | td->done = 1; | |
ebac4655 JA |
776 | } |
777 | ||
ebac4655 JA |
778 | static void cleanup_io_u(struct thread_data *td) |
779 | { | |
01743ee1 | 780 | struct flist_head *entry, *n; |
ebac4655 JA |
781 | struct io_u *io_u; |
782 | ||
01743ee1 JA |
783 | flist_for_each_safe(entry, n, &td->io_u_freelist) { |
784 | io_u = flist_entry(entry, struct io_u, list); | |
ebac4655 | 785 | |
01743ee1 | 786 | flist_del(&io_u->list); |
ebac4655 JA |
787 | free(io_u); |
788 | } | |
789 | ||
2f9ade3c | 790 | free_io_mem(td); |
ebac4655 JA |
791 | } |
792 | ||
793 | static int init_io_u(struct thread_data *td) | |
794 | { | |
795 | struct io_u *io_u; | |
a00735e6 | 796 | unsigned int max_bs; |
eb7ccf38 | 797 | int cl_align, i, max_units; |
ebac4655 JA |
798 | char *p; |
799 | ||
1e3c09af | 800 | max_units = td->o.iodepth; |
2dc1bbeb | 801 | max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]); |
5ec10eaa JA |
802 | td->orig_buffer_size = (unsigned long long) max_bs |
803 | * (unsigned long long) max_units; | |
804 | ||
805 | if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) { | |
806 | unsigned long bs; | |
74b025b0 | 807 | |
5ec10eaa JA |
808 | bs = td->orig_buffer_size + td->o.hugepage_size - 1; |
809 | td->orig_buffer_size = bs & ~(td->o.hugepage_size - 1); | |
810 | } | |
ebac4655 | 811 | |
c3990645 JA |
812 | if (td->orig_buffer_size != (size_t) td->orig_buffer_size) { |
813 | log_err("fio: IO memory too large. Reduce max_bs or iodepth\n"); | |
814 | return 1; | |
815 | } | |
816 | ||
2f9ade3c JA |
817 | if (allocate_io_mem(td)) |
818 | return 1; | |
ebac4655 | 819 | |
fc966bef | 820 | if (td->o.odirect || td->o.mem_align) |
d529ee19 | 821 | p = PAGE_ALIGN(td->orig_buffer) + td->o.mem_align; |
b3f378e9 JA |
822 | else |
823 | p = td->orig_buffer; | |
824 | ||
eb7ccf38 JA |
825 | cl_align = os_cache_line_size(); |
826 | ||
ebac4655 | 827 | for (i = 0; i < max_units; i++) { |
eb7ccf38 JA |
828 | void *ptr; |
829 | ||
db1cd90b JA |
830 | if (td->terminate) |
831 | return 1; | |
eb7ccf38 JA |
832 | |
833 | if (posix_memalign(&ptr, cl_align, sizeof(*io_u))) { | |
834 | log_err("fio: posix_memalign=%s\n", strerror(errno)); | |
835 | break; | |
836 | } | |
837 | ||
838 | io_u = ptr; | |
ebac4655 | 839 | memset(io_u, 0, sizeof(*io_u)); |
01743ee1 | 840 | INIT_FLIST_HEAD(&io_u->list); |
d529ee19 | 841 | dprint(FD_MEM, "io_u alloc %p, index %u\n", io_u, i); |
ebac4655 | 842 | |
b4c5e1ac JA |
843 | if (!(td->io_ops->flags & FIO_NOIO)) { |
844 | io_u->buf = p + max_bs * i; | |
d529ee19 | 845 | dprint(FD_MEM, "io_u %p, mem %p\n", io_u, io_u->buf); |
e9459e5a | 846 | |
5973cafb JA |
847 | if (td_write(td) && !td->o.refill_buffers) |
848 | io_u_fill_buffer(td, io_u, max_bs); | |
b4c5e1ac | 849 | } |
6b9cea23 | 850 | |
b1ff3403 | 851 | io_u->index = i; |
0c6e7517 | 852 | io_u->flags = IO_U_F_FREE; |
01743ee1 | 853 | flist_add(&io_u->list, &td->io_u_freelist); |
ebac4655 JA |
854 | } |
855 | ||
856 | return 0; | |
857 | } | |
858 | ||
da86774e JA |
859 | static int switch_ioscheduler(struct thread_data *td) |
860 | { | |
861 | char tmp[256], tmp2[128]; | |
862 | FILE *f; | |
863 | int ret; | |
864 | ||
ba0fbe10 | 865 | if (td->io_ops->flags & FIO_DISKLESSIO) |
f48b467c JA |
866 | return 0; |
867 | ||
da86774e JA |
868 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); |
869 | ||
870 | f = fopen(tmp, "r+"); | |
871 | if (!f) { | |
cbf5c1d7 | 872 | if (errno == ENOENT) { |
5ec10eaa JA |
873 | log_err("fio: os or kernel doesn't support IO scheduler" |
874 | " switching\n"); | |
cbf5c1d7 JA |
875 | return 0; |
876 | } | |
877 | td_verror(td, errno, "fopen iosched"); | |
da86774e JA |
878 | return 1; |
879 | } | |
880 | ||
881 | /* | |
882 | * Set io scheduler. | |
883 | */ | |
2dc1bbeb | 884 | ret = fwrite(td->o.ioscheduler, strlen(td->o.ioscheduler), 1, f); |
da86774e | 885 | if (ferror(f) || ret != 1) { |
e1161c32 | 886 | td_verror(td, errno, "fwrite"); |
da86774e JA |
887 | fclose(f); |
888 | return 1; | |
889 | } | |
890 | ||
891 | rewind(f); | |
892 | ||
893 | /* | |
894 | * Read back and check that the selected scheduler is now the default. | |
895 | */ | |
896 | ret = fread(tmp, 1, sizeof(tmp), f); | |
897 | if (ferror(f) || ret < 0) { | |
e1161c32 | 898 | td_verror(td, errno, "fread"); |
da86774e JA |
899 | fclose(f); |
900 | return 1; | |
901 | } | |
902 | ||
2dc1bbeb | 903 | sprintf(tmp2, "[%s]", td->o.ioscheduler); |
da86774e | 904 | if (!strstr(tmp, tmp2)) { |
2dc1bbeb | 905 | log_err("fio: io scheduler %s not found\n", td->o.ioscheduler); |
e1161c32 | 906 | td_verror(td, EINVAL, "iosched_switch"); |
da86774e JA |
907 | fclose(f); |
908 | return 1; | |
909 | } | |
910 | ||
911 | fclose(f); | |
912 | return 0; | |
913 | } | |
914 | ||
492158cf JA |
915 | static int keep_running(struct thread_data *td) |
916 | { | |
917 | unsigned long long io_done; | |
918 | ||
20e354ef JA |
919 | if (td->done) |
920 | return 0; | |
492158cf JA |
921 | if (td->o.time_based) |
922 | return 1; | |
923 | if (td->o.loops) { | |
924 | td->o.loops--; | |
925 | return 1; | |
926 | } | |
927 | ||
5ec10eaa JA |
928 | io_done = td->io_bytes[DDIR_READ] + td->io_bytes[DDIR_WRITE] |
929 | + td->io_skip_bytes; | |
492158cf JA |
930 | if (io_done < td->o.size) |
931 | return 1; | |
932 | ||
933 | return 0; | |
934 | } | |
935 | ||
b29ee5b3 | 936 | static void reset_io_counters(struct thread_data *td) |
ebac4655 | 937 | { |
756867bd | 938 | td->ts.stat_io_bytes[0] = td->ts.stat_io_bytes[1] = 0; |
ebac4655 | 939 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; |
20dc95c4 | 940 | td->zone_bytes = 0; |
581e7141 JA |
941 | td->rate_bytes[0] = td->rate_bytes[1] = 0; |
942 | td->rate_blocks[0] = td->rate_blocks[1] = 0; | |
ebac4655 | 943 | |
c1324df1 JA |
944 | td->last_was_sync = 0; |
945 | ||
2ba1c290 JA |
946 | /* |
947 | * reset file done count if we are to start over | |
948 | */ | |
949 | if (td->o.time_based || td->o.loops) | |
48f5abd3 | 950 | td->nr_done_files = 0; |
5bfc35d7 JA |
951 | |
952 | /* | |
953 | * Set the same seed to get repeatable runs | |
954 | */ | |
955 | td_fill_rand_seeds(td); | |
b29ee5b3 JA |
956 | } |
957 | ||
958 | void reset_all_stats(struct thread_data *td) | |
959 | { | |
960 | struct timeval tv; | |
961 | int i; | |
962 | ||
963 | reset_io_counters(td); | |
964 | ||
965 | for (i = 0; i < 2; i++) { | |
966 | td->io_bytes[i] = 0; | |
967 | td->io_blocks[i] = 0; | |
968 | td->io_issues[i] = 0; | |
969 | td->ts.total_io_u[i] = 0; | |
970 | } | |
971 | ||
972 | fio_gettime(&tv, NULL); | |
973 | memcpy(&td->epoch, &tv, sizeof(tv)); | |
974 | memcpy(&td->start, &tv, sizeof(tv)); | |
975 | } | |
976 | ||
df9c26b1 | 977 | static void clear_io_state(struct thread_data *td) |
b29ee5b3 JA |
978 | { |
979 | struct fio_file *f; | |
980 | unsigned int i; | |
b29ee5b3 JA |
981 | |
982 | reset_io_counters(td); | |
281f9b63 | 983 | |
24ffd2c2 | 984 | close_files(td); |
df9c26b1 | 985 | for_each_file(td, f, i) |
d6aed795 | 986 | fio_file_clear_done(f); |
ebac4655 JA |
987 | } |
988 | ||
398e8c4d JA |
989 | static int exec_string(const char *string) |
990 | { | |
991 | int ret, newlen = strlen(string) + 1 + 8; | |
992 | char *str; | |
993 | ||
994 | str = malloc(newlen); | |
995 | sprintf(str, "sh -c %s", string); | |
996 | ||
997 | ret = system(str); | |
998 | if (ret == -1) | |
999 | log_err("fio: exec of cmd <%s> failed\n", str); | |
1000 | ||
1001 | free(str); | |
1002 | return ret; | |
1003 | } | |
1004 | ||
906c8d75 JA |
1005 | /* |
1006 | * Entry point for the thread based jobs. The process based jobs end up | |
1007 | * here as well, after a little setup. | |
1008 | */ | |
ebac4655 JA |
1009 | static void *thread_main(void *data) |
1010 | { | |
10927316 | 1011 | unsigned long long runtime[2], elapsed; |
ebac4655 | 1012 | struct thread_data *td = data; |
e8462bd8 | 1013 | pthread_condattr_t attr; |
a978ba68 | 1014 | int clear_state; |
ebac4655 | 1015 | |
2dc1bbeb | 1016 | if (!td->o.use_thread) |
ebac4655 JA |
1017 | setsid(); |
1018 | ||
1019 | td->pid = getpid(); | |
1020 | ||
5921e80c | 1021 | dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid); |
ee56ad50 | 1022 | |
01743ee1 JA |
1023 | INIT_FLIST_HEAD(&td->io_u_freelist); |
1024 | INIT_FLIST_HEAD(&td->io_u_busylist); | |
1025 | INIT_FLIST_HEAD(&td->io_u_requeues); | |
1026 | INIT_FLIST_HEAD(&td->io_log_list); | |
1027 | INIT_FLIST_HEAD(&td->io_hist_list); | |
e8462bd8 JA |
1028 | INIT_FLIST_HEAD(&td->verify_list); |
1029 | pthread_mutex_init(&td->io_u_lock, NULL); | |
bb5d7d0b | 1030 | td->io_hist_tree = RB_ROOT; |
aea47d44 | 1031 | |
e8462bd8 JA |
1032 | pthread_condattr_init(&attr); |
1033 | pthread_cond_init(&td->verify_cond, &attr); | |
1034 | pthread_cond_init(&td->free_cond, &attr); | |
1035 | ||
db1cd90b | 1036 | td_set_runstate(td, TD_INITIALIZED); |
29adda3c | 1037 | dprint(FD_MUTEX, "up startup_mutex\n"); |
cdd18ad8 | 1038 | fio_mutex_up(startup_mutex); |
29adda3c | 1039 | dprint(FD_MUTEX, "wait on td->mutex\n"); |
cdd18ad8 | 1040 | fio_mutex_down(td->mutex); |
29adda3c | 1041 | dprint(FD_MUTEX, "done waiting on td->mutex\n"); |
db1cd90b JA |
1042 | |
1043 | /* | |
cdd18ad8 | 1044 | * the ->mutex mutex is now no longer used, close it to avoid |
db1cd90b JA |
1045 | * eating a file descriptor |
1046 | */ | |
cdd18ad8 | 1047 | fio_mutex_remove(td->mutex); |
db1cd90b | 1048 | |
e0b0d892 JA |
1049 | if (td->o.uid != -1U && setuid(td->o.uid)) { |
1050 | td_verror(td, errno, "setuid"); | |
1051 | goto err; | |
1052 | } | |
1053 | if (td->o.gid != -1U && setgid(td->o.gid)) { | |
1054 | td_verror(td, errno, "setgid"); | |
1055 | goto err; | |
1056 | } | |
1057 | ||
d84f8d49 JA |
1058 | /* |
1059 | * May alter parameters that init_io_u() will use, so we need to | |
1060 | * do this first. | |
1061 | */ | |
1062 | if (init_iolog(td)) | |
1063 | goto err; | |
1064 | ||
ebac4655 | 1065 | if (init_io_u(td)) |
db1cd90b | 1066 | goto err; |
ebac4655 | 1067 | |
e8462bd8 JA |
1068 | if (td->o.verify_async && verify_async_init(td)) |
1069 | goto err; | |
1070 | ||
1071 | if (td->o.cpumask_set && fio_setaffinity(td->pid, td->o.cpumask) == -1) { | |
e1161c32 | 1072 | td_verror(td, errno, "cpu_set_affinity"); |
db1cd90b | 1073 | goto err; |
ebac4655 JA |
1074 | } |
1075 | ||
f484470b JA |
1076 | /* |
1077 | * If we have a gettimeofday() thread, make sure we exclude that | |
1078 | * thread from this job | |
1079 | */ | |
be4ecfdf | 1080 | if (td->o.gtod_cpu) { |
be4ecfdf | 1081 | fio_cpu_clear(&td->o.cpumask, td->o.gtod_cpu); |
e8462bd8 | 1082 | if (fio_setaffinity(td->pid, td->o.cpumask) == -1) { |
be4ecfdf JA |
1083 | td_verror(td, errno, "cpu_set_affinity"); |
1084 | goto err; | |
1085 | } | |
1086 | } | |
1087 | ||
ac684785 | 1088 | if (td->ioprio_set) { |
ebac4655 | 1089 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { |
e1161c32 | 1090 | td_verror(td, errno, "ioprio_set"); |
db1cd90b | 1091 | goto err; |
ebac4655 JA |
1092 | } |
1093 | } | |
1094 | ||
6adb38a1 | 1095 | if (td->o.cgroup_weight && cgroup_setup(td, cgroup_list, &cgroup_mnt)) |
a696fa2a JA |
1096 | goto err; |
1097 | ||
2dc1bbeb | 1098 | if (nice(td->o.nice) == -1) { |
e1161c32 | 1099 | td_verror(td, errno, "nice"); |
db1cd90b | 1100 | goto err; |
b6f4d880 JA |
1101 | } |
1102 | ||
2dc1bbeb | 1103 | if (td->o.ioscheduler && switch_ioscheduler(td)) |
db1cd90b | 1104 | goto err; |
37f56873 | 1105 | |
2dc1bbeb | 1106 | if (!td->o.create_serialize && setup_files(td)) |
ebac4655 JA |
1107 | goto err; |
1108 | ||
7d6c5283 JA |
1109 | if (td_io_init(td)) |
1110 | goto err; | |
1111 | ||
68727076 JA |
1112 | if (init_random_map(td)) |
1113 | goto err; | |
1114 | ||
2dc1bbeb | 1115 | if (td->o.exec_prerun) { |
398e8c4d | 1116 | if (exec_string(td->o.exec_prerun)) |
69cfd7e0 JA |
1117 | goto err; |
1118 | } | |
4e0ba8af | 1119 | |
afad68f7 ZY |
1120 | if (td->o.pre_read) { |
1121 | if (pre_read_files(td) < 0) | |
1122 | goto err; | |
1123 | } | |
1124 | ||
69008999 | 1125 | fio_gettime(&td->epoch, NULL); |
756867bd | 1126 | getrusage(RUSAGE_SELF, &td->ts.ru_start); |
69008999 JA |
1127 | |
1128 | runtime[0] = runtime[1] = 0; | |
a978ba68 | 1129 | clear_state = 0; |
492158cf | 1130 | while (keep_running(td)) { |
02bcaa8c | 1131 | fio_gettime(&td->start, NULL); |
c5d3d790 JA |
1132 | memcpy(&td->ts.stat_sample_time[0], &td->start, |
1133 | sizeof(td->start)); | |
1134 | memcpy(&td->ts.stat_sample_time[1], &td->start, | |
1135 | sizeof(td->start)); | |
993bf48b | 1136 | memcpy(&td->tv_cache, &td->start, sizeof(td->start)); |
ebac4655 | 1137 | |
cf2dd8d6 | 1138 | if (td->o.ratemin[0] || td->o.ratemin[1]) |
5ec10eaa JA |
1139 | memcpy(&td->lastrate, &td->ts.stat_sample_time, |
1140 | sizeof(td->lastrate)); | |
ebac4655 | 1141 | |
df9c26b1 JA |
1142 | if (clear_state) |
1143 | clear_io_state(td); | |
a978ba68 | 1144 | |
ebac4655 JA |
1145 | prune_io_piece_log(td); |
1146 | ||
ba0fbe10 | 1147 | do_io(td); |
ebac4655 | 1148 | |
a978ba68 JA |
1149 | clear_state = 1; |
1150 | ||
38d77cae | 1151 | if (td_read(td) && td->io_bytes[DDIR_READ]) { |
0a4957c6 | 1152 | elapsed = utime_since_now(&td->start); |
38d77cae JA |
1153 | runtime[DDIR_READ] += elapsed; |
1154 | } | |
1155 | if (td_write(td) && td->io_bytes[DDIR_WRITE]) { | |
0a4957c6 | 1156 | elapsed = utime_since_now(&td->start); |
38d77cae JA |
1157 | runtime[DDIR_WRITE] += elapsed; |
1158 | } | |
5ec10eaa | 1159 | |
ebac4655 JA |
1160 | if (td->error || td->terminate) |
1161 | break; | |
1162 | ||
e84c73a8 SL |
1163 | if (!td->o.do_verify || |
1164 | td->o.verify == VERIFY_NONE || | |
b67740d3 | 1165 | (td->io_ops->flags & FIO_UNIDIR)) |
ebac4655 JA |
1166 | continue; |
1167 | ||
df9c26b1 | 1168 | clear_io_state(td); |
a978ba68 | 1169 | |
02bcaa8c | 1170 | fio_gettime(&td->start, NULL); |
ebac4655 JA |
1171 | |
1172 | do_verify(td); | |
1173 | ||
69008999 | 1174 | runtime[DDIR_READ] += utime_since_now(&td->start); |
ebac4655 JA |
1175 | |
1176 | if (td->error || td->terminate) | |
1177 | break; | |
1178 | } | |
1179 | ||
36dff966 | 1180 | update_rusage_stat(td); |
47f0cc48 YHJT |
1181 | td->ts.runtime[0] = (runtime[0] + 999) / 1000; |
1182 | td->ts.runtime[1] = (runtime[1] + 999) / 1000; | |
756867bd JA |
1183 | td->ts.total_run_time = mtime_since_now(&td->epoch); |
1184 | td->ts.io_bytes[0] = td->io_bytes[0]; | |
1185 | td->ts.io_bytes[1] = td->io_bytes[1]; | |
1186 | ||
d2bbb8d9 | 1187 | fio_mutex_down(writeout_mutex); |
e3cedca7 | 1188 | if (td->ts.bw_log) { |
f484470b JA |
1189 | if (td->o.bw_log_file) { |
1190 | finish_log_named(td, td->ts.bw_log, | |
1191 | td->o.bw_log_file, "bw"); | |
1192 | } else | |
e3cedca7 JA |
1193 | finish_log(td, td->ts.bw_log, "bw"); |
1194 | } | |
1195 | if (td->ts.slat_log) { | |
f484470b JA |
1196 | if (td->o.lat_log_file) { |
1197 | finish_log_named(td, td->ts.slat_log, | |
1ca95738 | 1198 | td->o.lat_log_file, "slat"); |
f484470b | 1199 | } else |
e3cedca7 JA |
1200 | finish_log(td, td->ts.slat_log, "slat"); |
1201 | } | |
1202 | if (td->ts.clat_log) { | |
f484470b JA |
1203 | if (td->o.lat_log_file) { |
1204 | finish_log_named(td, td->ts.clat_log, | |
1205 | td->o.lat_log_file, "clat"); | |
1206 | } else | |
e3cedca7 JA |
1207 | finish_log(td, td->ts.clat_log, "clat"); |
1208 | } | |
d2bbb8d9 | 1209 | fio_mutex_up(writeout_mutex); |
398e8c4d JA |
1210 | if (td->o.exec_postrun) |
1211 | exec_string(td->o.exec_postrun); | |
ebac4655 JA |
1212 | |
1213 | if (exitall_on_terminate) | |
390c40e2 | 1214 | terminate_threads(td->groupid); |
ebac4655 JA |
1215 | |
1216 | err: | |
5bf13a5a | 1217 | if (td->error) |
4ceb30d4 | 1218 | log_info("fio: pid=%d, err=%d/%s\n", (int) td->pid, td->error, |
5ec10eaa | 1219 | td->verror); |
b4707ed7 JA |
1220 | |
1221 | if (td->o.verify_async) | |
1222 | verify_async_exit(td); | |
1223 | ||
24ffd2c2 | 1224 | close_and_free_files(td); |
2866c82d | 1225 | close_ioengine(td); |
ebac4655 | 1226 | cleanup_io_u(td); |
6adb38a1 | 1227 | cgroup_shutdown(td, &cgroup_mnt); |
f29b25a3 | 1228 | |
d2ce18b5 JA |
1229 | if (td->o.cpumask_set) { |
1230 | int ret = fio_cpuset_exit(&td->o.cpumask); | |
1231 | ||
1232 | td_verror(td, ret, "fio_cpuset_exit"); | |
1233 | } | |
4e78e405 | 1234 | |
f29b25a3 JA |
1235 | /* |
1236 | * do this very late, it will log file closing as well | |
1237 | */ | |
1238 | if (td->o.write_iolog_file) | |
1239 | write_iolog_close(td); | |
1240 | ||
d23bb327 | 1241 | options_mem_free(td); |
ebac4655 | 1242 | td_set_runstate(td, TD_EXITED); |
43d76807 | 1243 | return (void *) (unsigned long) td->error; |
ebac4655 JA |
1244 | } |
1245 | ||
906c8d75 JA |
1246 | /* |
1247 | * We cannot pass the td data into a forked process, so attach the td and | |
1248 | * pass it to the thread worker. | |
1249 | */ | |
a6418147 | 1250 | static int fork_main(int shmid, int offset) |
ebac4655 JA |
1251 | { |
1252 | struct thread_data *td; | |
a6418147 | 1253 | void *data, *ret; |
ebac4655 JA |
1254 | |
1255 | data = shmat(shmid, NULL, 0); | |
1256 | if (data == (void *) -1) { | |
a6418147 JA |
1257 | int __err = errno; |
1258 | ||
ebac4655 | 1259 | perror("shmat"); |
a6418147 | 1260 | return __err; |
ebac4655 JA |
1261 | } |
1262 | ||
1263 | td = data + offset * sizeof(struct thread_data); | |
a6418147 | 1264 | ret = thread_main(td); |
ebac4655 | 1265 | shmdt(data); |
43d76807 | 1266 | return (int) (unsigned long) ret; |
ebac4655 JA |
1267 | } |
1268 | ||
906c8d75 JA |
1269 | /* |
1270 | * Run over the job map and reap the threads that have exited, if any. | |
1271 | */ | |
ebac4655 JA |
1272 | static void reap_threads(int *nr_running, int *t_rate, int *m_rate) |
1273 | { | |
34572e28 | 1274 | struct thread_data *td; |
1f809d15 | 1275 | int i, cputhreads, realthreads, pending, status, ret; |
ebac4655 JA |
1276 | |
1277 | /* | |
1278 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
1279 | */ | |
1f809d15 | 1280 | realthreads = pending = cputhreads = 0; |
34572e28 | 1281 | for_each_td(td, i) { |
3707f45b | 1282 | int flags = 0; |
a2f77c9f | 1283 | |
84585003 JA |
1284 | /* |
1285 | * ->io_ops is NULL for a thread that has closed its | |
1286 | * io engine | |
1287 | */ | |
ba0fbe10 | 1288 | if (td->io_ops && !strcmp(td->io_ops->name, "cpuio")) |
b990b5c0 | 1289 | cputhreads++; |
1f809d15 JA |
1290 | else |
1291 | realthreads++; | |
b990b5c0 | 1292 | |
a1dedc1f AC |
1293 | if (!td->pid) { |
1294 | pending++; | |
1295 | continue; | |
1296 | } | |
1297 | if (td->runstate == TD_REAPED) | |
a2f77c9f | 1298 | continue; |
2dc1bbeb | 1299 | if (td->o.use_thread) { |
3707f45b JA |
1300 | if (td->runstate == TD_EXITED) { |
1301 | td_set_runstate(td, TD_REAPED); | |
1302 | goto reaped; | |
1303 | } | |
1304 | continue; | |
1305 | } | |
a2f77c9f JA |
1306 | |
1307 | flags = WNOHANG; | |
1308 | if (td->runstate == TD_EXITED) | |
1309 | flags = 0; | |
1310 | ||
1311 | /* | |
1312 | * check if someone quit or got killed in an unusual way | |
1313 | */ | |
1314 | ret = waitpid(td->pid, &status, flags); | |
3707f45b | 1315 | if (ret < 0) { |
a2f77c9f | 1316 | if (errno == ECHILD) { |
5921e80c JA |
1317 | log_err("fio: pid=%d disappeared %d\n", |
1318 | (int) td->pid, td->runstate); | |
a2f77c9f JA |
1319 | td_set_runstate(td, TD_REAPED); |
1320 | goto reaped; | |
1321 | } | |
1322 | perror("waitpid"); | |
1323 | } else if (ret == td->pid) { | |
1324 | if (WIFSIGNALED(status)) { | |
fab6aa71 JA |
1325 | int sig = WTERMSIG(status); |
1326 | ||
d9244941 | 1327 | if (sig != SIGQUIT) |
5ec10eaa | 1328 | log_err("fio: pid=%d, got signal=%d\n", |
5921e80c | 1329 | (int) td->pid, sig); |
fab6aa71 JA |
1330 | td_set_runstate(td, TD_REAPED); |
1331 | goto reaped; | |
1332 | } | |
a2f77c9f JA |
1333 | if (WIFEXITED(status)) { |
1334 | if (WEXITSTATUS(status) && !td->error) | |
1335 | td->error = WEXITSTATUS(status); | |
a2f77c9f | 1336 | |
a2f77c9f JA |
1337 | td_set_runstate(td, TD_REAPED); |
1338 | goto reaped; | |
a6418147 JA |
1339 | } |
1340 | } | |
ebac4655 | 1341 | |
a2f77c9f JA |
1342 | /* |
1343 | * thread is not dead, continue | |
1344 | */ | |
e48676ba | 1345 | pending++; |
a2f77c9f | 1346 | continue; |
fab6aa71 | 1347 | reaped: |
ebac4655 | 1348 | (*nr_running)--; |
581e7141 JA |
1349 | (*m_rate) -= (td->o.ratemin[0] + td->o.ratemin[1]); |
1350 | (*t_rate) -= (td->o.rate[0] + td->o.rate[1]); | |
6f88bcb0 JA |
1351 | if (!td->pid) |
1352 | pending--; | |
a2f77c9f JA |
1353 | |
1354 | if (td->error) | |
1355 | exit_value++; | |
d3eeeabc JA |
1356 | |
1357 | done_secs += mtime_since_now(&td->epoch) / 1000; | |
ebac4655 | 1358 | } |
b990b5c0 | 1359 | |
1f809d15 | 1360 | if (*nr_running == cputhreads && !pending && realthreads) |
390c40e2 | 1361 | terminate_threads(TERMINATE_ALL); |
ebac4655 JA |
1362 | } |
1363 | ||
be4ecfdf JA |
1364 | static void *gtod_thread_main(void *data) |
1365 | { | |
1366 | fio_mutex_up(startup_mutex); | |
1367 | ||
1368 | /* | |
1369 | * As long as we have jobs around, update the clock. It would be nice | |
1370 | * to have some way of NOT hammering that CPU with gettimeofday(), | |
1371 | * but I'm not sure what to use outside of a simple CPU nop to relax | |
1372 | * it - we don't want to lose precision. | |
1373 | */ | |
1374 | while (threads) { | |
1375 | fio_gtod_update(); | |
1376 | nop; | |
1377 | } | |
1378 | ||
1379 | return NULL; | |
1380 | } | |
1381 | ||
1382 | static int fio_start_gtod_thread(void) | |
1383 | { | |
f42c153d JA |
1384 | int ret; |
1385 | ||
1386 | ret = pthread_create(>od_thread, NULL, gtod_thread_main, NULL); | |
1387 | if (ret) { | |
1388 | log_err("Can't create gtod thread: %s\n", strerror(ret)); | |
be4ecfdf JA |
1389 | return 1; |
1390 | } | |
f42c153d JA |
1391 | |
1392 | ret = pthread_detach(gtod_thread); | |
1393 | if (ret) { | |
1394 | log_err("Can't detatch gtod thread: %s\n", strerror(ret)); | |
be4ecfdf JA |
1395 | return 1; |
1396 | } | |
1397 | ||
29adda3c | 1398 | dprint(FD_MUTEX, "wait on startup_mutex\n"); |
be4ecfdf | 1399 | fio_mutex_down(startup_mutex); |
29adda3c | 1400 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); |
be4ecfdf JA |
1401 | return 0; |
1402 | } | |
1403 | ||
906c8d75 JA |
1404 | /* |
1405 | * Main function for kicking off and reaping jobs, as needed. | |
1406 | */ | |
ebac4655 JA |
1407 | static void run_threads(void) |
1408 | { | |
ebac4655 JA |
1409 | struct thread_data *td; |
1410 | unsigned long spent; | |
1411 | int i, todo, nr_running, m_rate, t_rate, nr_started; | |
fcb6ade2 | 1412 | |
2f9ade3c JA |
1413 | if (fio_pin_memory()) |
1414 | return; | |
ebac4655 | 1415 | |
be4ecfdf JA |
1416 | if (fio_gtod_offload && fio_start_gtod_thread()) |
1417 | return; | |
1418 | ||
c6ae0a5b | 1419 | if (!terse_output) { |
4ceb30d4 | 1420 | log_info("Starting "); |
9cedf167 | 1421 | if (nr_thread) |
4ceb30d4 | 1422 | log_info("%d thread%s", nr_thread, |
5ec10eaa | 1423 | nr_thread > 1 ? "s" : ""); |
9cedf167 JA |
1424 | if (nr_process) { |
1425 | if (nr_thread) | |
1426 | printf(" and "); | |
4ceb30d4 | 1427 | log_info("%d process%s", nr_process, |
5ec10eaa | 1428 | nr_process > 1 ? "es" : ""); |
9cedf167 | 1429 | } |
4ceb30d4 | 1430 | log_info("\n"); |
c6ae0a5b JA |
1431 | fflush(stdout); |
1432 | } | |
c04f7ec3 | 1433 | |
697a606c | 1434 | set_sig_handlers(); |
4efa970e | 1435 | |
ebac4655 JA |
1436 | todo = thread_number; |
1437 | nr_running = 0; | |
1438 | nr_started = 0; | |
1439 | m_rate = t_rate = 0; | |
1440 | ||
34572e28 | 1441 | for_each_td(td, i) { |
263e529f | 1442 | print_status_init(td->thread_number - 1); |
ebac4655 | 1443 | |
2dc1bbeb | 1444 | if (!td->o.create_serialize) { |
380cf265 | 1445 | init_disk_util(td); |
ebac4655 | 1446 | continue; |
380cf265 | 1447 | } |
ebac4655 JA |
1448 | |
1449 | /* | |
1450 | * do file setup here so it happens sequentially, | |
1451 | * we don't want X number of threads getting their | |
1452 | * client data interspersed on disk | |
1453 | */ | |
53cdc686 | 1454 | if (setup_files(td)) { |
5bf13a5a JA |
1455 | exit_value++; |
1456 | if (td->error) | |
5921e80c JA |
1457 | log_err("fio: pid=%d, err=%d/%s\n", |
1458 | (int) td->pid, td->error, td->verror); | |
ebac4655 JA |
1459 | td_set_runstate(td, TD_REAPED); |
1460 | todo--; | |
c69e8875 JA |
1461 | } else { |
1462 | struct fio_file *f; | |
1463 | unsigned int i; | |
1464 | ||
1465 | /* | |
1466 | * for sharing to work, each job must always open | |
1467 | * its own files. so close them, if we opened them | |
1468 | * for creation | |
1469 | */ | |
22a57ba8 JA |
1470 | for_each_file(td, f, i) { |
1471 | if (fio_file_open(f)) | |
1472 | td_io_close_file(td, f); | |
22a57ba8 | 1473 | } |
5ec10eaa | 1474 | } |
380cf265 JA |
1475 | |
1476 | init_disk_util(td); | |
ebac4655 JA |
1477 | } |
1478 | ||
a2f77c9f JA |
1479 | set_genesis_time(); |
1480 | ||
ebac4655 | 1481 | while (todo) { |
75154845 JA |
1482 | struct thread_data *map[MAX_JOBS]; |
1483 | struct timeval this_start; | |
1484 | int this_jobs = 0, left; | |
1485 | ||
ebac4655 JA |
1486 | /* |
1487 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
1488 | */ | |
34572e28 | 1489 | for_each_td(td, i) { |
ebac4655 JA |
1490 | if (td->runstate != TD_NOT_CREATED) |
1491 | continue; | |
1492 | ||
1493 | /* | |
1494 | * never got a chance to start, killed by other | |
1495 | * thread for some reason | |
1496 | */ | |
1497 | if (td->terminate) { | |
1498 | todo--; | |
1499 | continue; | |
1500 | } | |
1501 | ||
2dc1bbeb | 1502 | if (td->o.start_delay) { |
263e529f | 1503 | spent = mtime_since_genesis(); |
ebac4655 | 1504 | |
2dc1bbeb | 1505 | if (td->o.start_delay * 1000 > spent) |
ebac4655 JA |
1506 | continue; |
1507 | } | |
1508 | ||
6f88bcb0 JA |
1509 | if (td->o.stonewall && (nr_started || nr_running)) { |
1510 | dprint(FD_PROCESS, "%s: stonewall wait\n", | |
1511 | td->o.name); | |
ebac4655 | 1512 | break; |
6f88bcb0 | 1513 | } |
ebac4655 | 1514 | |
75154845 JA |
1515 | /* |
1516 | * Set state to created. Thread will transition | |
1517 | * to TD_INITIALIZED when it's done setting up. | |
1518 | */ | |
ebac4655 | 1519 | td_set_runstate(td, TD_CREATED); |
75154845 | 1520 | map[this_jobs++] = td; |
ebac4655 JA |
1521 | nr_started++; |
1522 | ||
2dc1bbeb | 1523 | if (td->o.use_thread) { |
f42c153d JA |
1524 | int ret; |
1525 | ||
ee56ad50 | 1526 | dprint(FD_PROCESS, "will pthread_create\n"); |
f42c153d JA |
1527 | ret = pthread_create(&td->thread, NULL, |
1528 | thread_main, td); | |
1529 | if (ret) { | |
1530 | log_err("pthread_create: %s\n", | |
1531 | strerror(ret)); | |
ebac4655 | 1532 | nr_started--; |
c8bb6faf | 1533 | break; |
ebac4655 | 1534 | } |
f42c153d JA |
1535 | ret = pthread_detach(td->thread); |
1536 | if (ret) | |
1537 | log_err("pthread_detach: %s", | |
1538 | strerror(ret)); | |
ebac4655 | 1539 | } else { |
5e1d306e | 1540 | pid_t pid; |
ee56ad50 | 1541 | dprint(FD_PROCESS, "will fork\n"); |
5e1d306e JA |
1542 | pid = fork(); |
1543 | if (!pid) { | |
a6418147 JA |
1544 | int ret = fork_main(shm_id, i); |
1545 | ||
5e1d306e JA |
1546 | _exit(ret); |
1547 | } else if (i == fio_debug_jobno) | |
1548 | *fio_debug_jobp = pid; | |
ebac4655 | 1549 | } |
29adda3c | 1550 | dprint(FD_MUTEX, "wait on startup_mutex\n"); |
656b1393 JA |
1551 | if (fio_mutex_down_timeout(startup_mutex, 10)) { |
1552 | log_err("fio: job startup hung? exiting.\n"); | |
1553 | terminate_threads(TERMINATE_ALL); | |
1554 | fio_abort = 1; | |
1555 | nr_started--; | |
1556 | break; | |
1557 | } | |
29adda3c | 1558 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); |
ebac4655 JA |
1559 | } |
1560 | ||
1561 | /* | |
75154845 JA |
1562 | * Wait for the started threads to transition to |
1563 | * TD_INITIALIZED. | |
ebac4655 | 1564 | */ |
02bcaa8c | 1565 | fio_gettime(&this_start, NULL); |
75154845 | 1566 | left = this_jobs; |
6ce15a32 | 1567 | while (left && !fio_abort) { |
75154845 JA |
1568 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) |
1569 | break; | |
1570 | ||
1571 | usleep(100000); | |
1572 | ||
1573 | for (i = 0; i < this_jobs; i++) { | |
1574 | td = map[i]; | |
1575 | if (!td) | |
1576 | continue; | |
b6f4d880 | 1577 | if (td->runstate == TD_INITIALIZED) { |
75154845 JA |
1578 | map[i] = NULL; |
1579 | left--; | |
b6f4d880 JA |
1580 | } else if (td->runstate >= TD_EXITED) { |
1581 | map[i] = NULL; | |
1582 | left--; | |
1583 | todo--; | |
1584 | nr_running++; /* work-around... */ | |
75154845 JA |
1585 | } |
1586 | } | |
1587 | } | |
1588 | ||
1589 | if (left) { | |
3b70d7e5 | 1590 | log_err("fio: %d jobs failed to start\n", left); |
75154845 JA |
1591 | for (i = 0; i < this_jobs; i++) { |
1592 | td = map[i]; | |
1593 | if (!td) | |
1594 | continue; | |
1595 | kill(td->pid, SIGTERM); | |
1596 | } | |
1597 | break; | |
1598 | } | |
1599 | ||
1600 | /* | |
b6f4d880 | 1601 | * start created threads (TD_INITIALIZED -> TD_RUNNING). |
75154845 | 1602 | */ |
34572e28 | 1603 | for_each_td(td, i) { |
75154845 | 1604 | if (td->runstate != TD_INITIALIZED) |
ebac4655 JA |
1605 | continue; |
1606 | ||
b29ee5b3 JA |
1607 | if (in_ramp_time(td)) |
1608 | td_set_runstate(td, TD_RAMP); | |
1609 | else | |
1610 | td_set_runstate(td, TD_RUNNING); | |
ebac4655 JA |
1611 | nr_running++; |
1612 | nr_started--; | |
581e7141 JA |
1613 | m_rate += td->o.ratemin[0] + td->o.ratemin[1]; |
1614 | t_rate += td->o.rate[0] + td->o.rate[1]; | |
75154845 | 1615 | todo--; |
cdd18ad8 | 1616 | fio_mutex_up(td->mutex); |
ebac4655 JA |
1617 | } |
1618 | ||
1619 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1620 | ||
1621 | if (todo) | |
1622 | usleep(100000); | |
1623 | } | |
1624 | ||
1625 | while (nr_running) { | |
1626 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1627 | usleep(10000); | |
1628 | } | |
1629 | ||
1630 | update_io_ticks(); | |
2f9ade3c | 1631 | fio_unpin_memory(); |
ebac4655 JA |
1632 | } |
1633 | ||
ebac4655 JA |
1634 | int main(int argc, char *argv[]) |
1635 | { | |
29d610e1 JA |
1636 | long ps; |
1637 | ||
2e5cdb11 JA |
1638 | sinit(); |
1639 | ||
dbe1125e JA |
1640 | /* |
1641 | * We need locale for number printing, if it isn't set then just | |
1642 | * go with the US format. | |
1643 | */ | |
1644 | if (!getenv("LC_NUMERIC")) | |
1645 | setlocale(LC_NUMERIC, "en_US"); | |
1646 | ||
29d610e1 JA |
1647 | ps = sysconf(_SC_PAGESIZE); |
1648 | if (ps < 0) { | |
1649 | log_err("Failed to get page size\n"); | |
1650 | return 1; | |
1651 | } | |
1652 | ||
cfc99db7 | 1653 | page_size = ps; |
29d610e1 JA |
1654 | page_mask = ps - 1; |
1655 | ||
74929ac2 JA |
1656 | fio_keywords_init(); |
1657 | ||
1658 | if (parse_options(argc, argv)) | |
1659 | return 1; | |
1660 | ||
07b3232d JA |
1661 | if (exec_profile && load_profile(exec_profile)) |
1662 | return 1; | |
1663 | ||
74929ac2 JA |
1664 | if (!thread_number) |
1665 | return 0; | |
1666 | ||
bb3884d8 JA |
1667 | if (write_bw_log) { |
1668 | setup_log(&agg_io_log[DDIR_READ]); | |
1669 | setup_log(&agg_io_log[DDIR_WRITE]); | |
1670 | } | |
1671 | ||
cdd18ad8 | 1672 | startup_mutex = fio_mutex_init(0); |
d2bbb8d9 | 1673 | writeout_mutex = fio_mutex_init(1); |
07739b57 | 1674 | |
a2f77c9f JA |
1675 | set_genesis_time(); |
1676 | ||
ccb0fa24 | 1677 | status_timer_arm(); |
ebac4655 | 1678 | |
dae5341c JA |
1679 | cgroup_list = smalloc(sizeof(*cgroup_list)); |
1680 | INIT_FLIST_HEAD(cgroup_list); | |
1681 | ||
ebac4655 | 1682 | run_threads(); |
6ce15a32 | 1683 | |
bb3884d8 | 1684 | if (!fio_abort) { |
6ce15a32 | 1685 | show_run_stats(); |
bb3884d8 | 1686 | if (write_bw_log) { |
5ec10eaa JA |
1687 | __finish_log(agg_io_log[DDIR_READ], "agg-read_bw.log"); |
1688 | __finish_log(agg_io_log[DDIR_WRITE], | |
1689 | "agg-write_bw.log"); | |
bb3884d8 JA |
1690 | } |
1691 | } | |
ebac4655 | 1692 | |
dae5341c JA |
1693 | cgroup_kill(cgroup_list); |
1694 | sfree(cgroup_list); | |
6adb38a1 JA |
1695 | if (cgroup_mnt) |
1696 | sfree(cgroup_mnt); | |
39f22027 | 1697 | |
cdd18ad8 | 1698 | fio_mutex_remove(startup_mutex); |
d2bbb8d9 | 1699 | fio_mutex_remove(writeout_mutex); |
437c9b71 | 1700 | return exit_value; |
ebac4655 | 1701 | } |