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