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 JA |
6 | * |
7 | * This program is free software; you can redistribute it and/or modify | |
8 | * it under the terms of the GNU General Public License as published by | |
9 | * the Free Software Foundation; either version 2 of the License, or | |
10 | * (at your option) any later version. | |
11 | * | |
12 | * This program is distributed in the hope that it will be useful, | |
13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
15 | * GNU General Public License for more details. | |
16 | * | |
17 | * You should have received a copy of the GNU General Public License | |
18 | * along with this program; if not, write to the Free Software | |
19 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
20 | * | |
21 | */ | |
ebac4655 JA |
22 | #include <unistd.h> |
23 | #include <fcntl.h> | |
24 | #include <string.h> | |
ebac4655 JA |
25 | #include <signal.h> |
26 | #include <time.h> | |
ebac4655 | 27 | #include <assert.h> |
ebac4655 JA |
28 | #include <sys/stat.h> |
29 | #include <sys/wait.h> | |
30 | #include <sys/ipc.h> | |
31 | #include <sys/shm.h> | |
32 | #include <sys/ioctl.h> | |
33 | #include <sys/mman.h> | |
34 | ||
35 | #include "fio.h" | |
36 | #include "os.h" | |
37 | ||
38 | #define MASK (4095) | |
39 | ||
40 | #define ALIGN(buf) (char *) (((unsigned long) (buf) + MASK) & ~(MASK)) | |
41 | ||
42 | int groupid = 0; | |
43 | int thread_number = 0; | |
ebac4655 | 44 | int shm_id = 0; |
53cdc686 | 45 | int temp_stall_ts; |
c1d5725e | 46 | char *fio_inst_prefix = _INST_PREFIX; |
ebac4655 | 47 | |
c04f7ec3 JA |
48 | extern unsigned long long mlock_size; |
49 | ||
3d60d1ed | 50 | #define should_fsync(td) ((td_write(td) || td_rw(td)) && (!(td)->odirect || (td)->override_sync)) |
ebac4655 | 51 | |
bbfd6b00 | 52 | static volatile int startup_sem; |
ebac4655 JA |
53 | |
54 | #define TERMINATE_ALL (-1) | |
75154845 | 55 | #define JOB_START_TIMEOUT (5 * 1000) |
ebac4655 JA |
56 | |
57 | static void terminate_threads(int group_id) | |
58 | { | |
59 | int i; | |
60 | ||
61 | for (i = 0; i < thread_number; i++) { | |
62 | struct thread_data *td = &threads[i]; | |
63 | ||
64 | if (group_id == TERMINATE_ALL || groupid == td->groupid) { | |
65 | td->terminate = 1; | |
66 | td->start_delay = 0; | |
67 | } | |
68 | } | |
69 | } | |
70 | ||
71 | static void sig_handler(int sig) | |
72 | { | |
73 | switch (sig) { | |
74 | case SIGALRM: | |
75 | update_io_ticks(); | |
76 | disk_util_timer_arm(); | |
77 | print_thread_status(); | |
78 | break; | |
79 | default: | |
80 | printf("\nfio: terminating on signal\n"); | |
81 | fflush(stdout); | |
82 | terminate_threads(TERMINATE_ALL); | |
83 | break; | |
84 | } | |
85 | } | |
86 | ||
906c8d75 JA |
87 | /* |
88 | * The ->file_map[] contains a map of blocks we have or have not done io | |
89 | * to yet. Used to make sure we cover the entire range in a fair fashion. | |
90 | */ | |
53cdc686 JA |
91 | static int random_map_free(struct thread_data *td, struct fio_file *f, |
92 | unsigned long long block) | |
ebac4655 | 93 | { |
53cdc686 JA |
94 | unsigned int idx = RAND_MAP_IDX(td, f, block); |
95 | unsigned int bit = RAND_MAP_BIT(td, f, block); | |
ebac4655 | 96 | |
53cdc686 | 97 | return (f->file_map[idx] & (1UL << bit)) == 0; |
ebac4655 JA |
98 | } |
99 | ||
906c8d75 JA |
100 | /* |
101 | * Return the next free block in the map. | |
102 | */ | |
53cdc686 JA |
103 | static int get_next_free_block(struct thread_data *td, struct fio_file *f, |
104 | unsigned long long *b) | |
ebac4655 JA |
105 | { |
106 | int i; | |
107 | ||
108 | *b = 0; | |
109 | i = 0; | |
53cdc686 JA |
110 | while ((*b) * td->min_bs < f->file_size) { |
111 | if (f->file_map[i] != -1UL) { | |
112 | *b += ffz(f->file_map[i]); | |
ebac4655 JA |
113 | return 0; |
114 | } | |
115 | ||
116 | *b += BLOCKS_PER_MAP; | |
117 | i++; | |
118 | } | |
119 | ||
120 | return 1; | |
121 | } | |
122 | ||
906c8d75 JA |
123 | /* |
124 | * Mark a given offset as used in the map. | |
125 | */ | |
53cdc686 JA |
126 | static void mark_random_map(struct thread_data *td, struct fio_file *f, |
127 | struct io_u *io_u) | |
ebac4655 | 128 | { |
200bc855 | 129 | unsigned long long block = io_u->offset / (unsigned long long) td->min_bs; |
ebac4655 JA |
130 | unsigned int blocks = 0; |
131 | ||
132 | while (blocks < (io_u->buflen / td->min_bs)) { | |
133 | unsigned int idx, bit; | |
134 | ||
53cdc686 | 135 | if (!random_map_free(td, f, block)) |
ebac4655 JA |
136 | break; |
137 | ||
53cdc686 JA |
138 | idx = RAND_MAP_IDX(td, f, block); |
139 | bit = RAND_MAP_BIT(td, f, block); | |
ebac4655 | 140 | |
53cdc686 | 141 | assert(idx < f->num_maps); |
ebac4655 | 142 | |
53cdc686 | 143 | f->file_map[idx] |= (1UL << bit); |
ebac4655 JA |
144 | block++; |
145 | blocks++; | |
146 | } | |
147 | ||
148 | if ((blocks * td->min_bs) < io_u->buflen) | |
149 | io_u->buflen = blocks * td->min_bs; | |
150 | } | |
151 | ||
906c8d75 JA |
152 | /* |
153 | * For random io, generate a random new block and see if it's used. Repeat | |
154 | * until we find a free one. For sequential io, just return the end of | |
155 | * the last io issued. | |
156 | */ | |
53cdc686 JA |
157 | static int get_next_offset(struct thread_data *td, struct fio_file *f, |
158 | unsigned long long *offset) | |
20dc95c4 JA |
159 | { |
160 | unsigned long long b, rb; | |
161 | long r; | |
162 | ||
163 | if (!td->sequential) { | |
085227ad | 164 | unsigned long long max_blocks = td->io_size / td->min_bs; |
20dc95c4 JA |
165 | int loops = 50; |
166 | ||
167 | do { | |
6dfd46b9 | 168 | r = os_random_long(&td->random_state); |
085227ad | 169 | b = ((max_blocks - 1) * r / (unsigned long long) (RAND_MAX+1.0)); |
53cdc686 | 170 | rb = b + (f->file_offset / td->min_bs); |
20dc95c4 | 171 | loops--; |
53cdc686 | 172 | } while (!random_map_free(td, f, rb) && loops); |
20dc95c4 JA |
173 | |
174 | if (!loops) { | |
53cdc686 | 175 | if (get_next_free_block(td, f, &b)) |
20dc95c4 JA |
176 | return 1; |
177 | } | |
178 | } else | |
53cdc686 | 179 | b = f->last_pos / td->min_bs; |
20dc95c4 | 180 | |
53cdc686 JA |
181 | *offset = (b * td->min_bs) + f->file_offset; |
182 | if (*offset > f->file_size) | |
20dc95c4 JA |
183 | return 1; |
184 | ||
185 | return 0; | |
186 | } | |
187 | ||
188 | static unsigned int get_next_buflen(struct thread_data *td) | |
189 | { | |
190 | unsigned int buflen; | |
191 | long r; | |
192 | ||
193 | if (td->min_bs == td->max_bs) | |
194 | buflen = td->min_bs; | |
195 | else { | |
6dfd46b9 | 196 | r = os_random_long(&td->bsrange_state); |
20dc95c4 JA |
197 | buflen = (1 + (double) (td->max_bs - 1) * r / (RAND_MAX + 1.0)); |
198 | buflen = (buflen + td->min_bs - 1) & ~(td->min_bs - 1); | |
199 | } | |
200 | ||
b2a15192 JA |
201 | if (buflen > td->io_size - td->this_io_bytes[td->ddir]) { |
202 | /* | |
203 | * if using direct/raw io, we may not be able to | |
204 | * shrink the size. so just fail it. | |
205 | */ | |
206 | if (td->io_ops->flags & FIO_RAWIO) | |
207 | return 0; | |
208 | ||
20dc95c4 | 209 | buflen = td->io_size - td->this_io_bytes[td->ddir]; |
b2a15192 | 210 | } |
20dc95c4 JA |
211 | |
212 | return buflen; | |
213 | } | |
214 | ||
906c8d75 JA |
215 | /* |
216 | * Check if we are above the minimum rate given. | |
217 | */ | |
ebac4655 JA |
218 | static int check_min_rate(struct thread_data *td, struct timeval *now) |
219 | { | |
220 | unsigned long spent; | |
221 | unsigned long rate; | |
222 | int ddir = td->ddir; | |
223 | ||
224 | /* | |
225 | * allow a 2 second settle period in the beginning | |
226 | */ | |
227 | if (mtime_since(&td->start, now) < 2000) | |
228 | return 0; | |
229 | ||
230 | /* | |
231 | * if rate blocks is set, sample is running | |
232 | */ | |
233 | if (td->rate_bytes) { | |
234 | spent = mtime_since(&td->lastrate, now); | |
235 | if (spent < td->ratecycle) | |
236 | return 0; | |
237 | ||
238 | rate = (td->this_io_bytes[ddir] - td->rate_bytes) / spent; | |
239 | if (rate < td->ratemin) { | |
eb8bbf48 | 240 | fprintf(f_out, "%s: min rate %d not met, got %ldKiB/sec\n", td->name, td->ratemin, rate); |
ebac4655 JA |
241 | if (rate_quit) |
242 | terminate_threads(td->groupid); | |
243 | return 1; | |
244 | } | |
245 | } | |
246 | ||
247 | td->rate_bytes = td->this_io_bytes[ddir]; | |
248 | memcpy(&td->lastrate, now, sizeof(*now)); | |
249 | return 0; | |
250 | } | |
251 | ||
252 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) | |
253 | { | |
254 | if (!td->timeout) | |
255 | return 0; | |
256 | if (mtime_since(&td->epoch, t) >= td->timeout * 1000) | |
257 | return 1; | |
258 | ||
259 | return 0; | |
260 | } | |
261 | ||
906c8d75 JA |
262 | /* |
263 | * Return the data direction for the next io_u. If the job is a | |
264 | * mixed read/write workload, check the rwmix cycle and switch if | |
265 | * necessary. | |
266 | */ | |
a6ccc7be | 267 | static int get_rw_ddir(struct thread_data *td) |
3d60d1ed | 268 | { |
a6ccc7be JA |
269 | if (td_rw(td)) { |
270 | struct timeval now; | |
271 | unsigned long elapsed; | |
3d60d1ed | 272 | |
a6ccc7be JA |
273 | gettimeofday(&now, NULL); |
274 | elapsed = mtime_since_now(&td->rwmix_switch); | |
3d60d1ed | 275 | |
a6ccc7be JA |
276 | /* |
277 | * Check if it's time to seed a new data direction. | |
278 | */ | |
279 | if (elapsed >= td->rwmixcycle) { | |
0ab8db89 | 280 | unsigned int v; |
a6ccc7be | 281 | long r; |
3d60d1ed | 282 | |
c1ee2ca4 JA |
283 | r = os_random_long(&td->rwmix_state); |
284 | v = 1 + (int) (100.0 * (r / (RAND_MAX + 1.0))); | |
a6ccc7be JA |
285 | if (v < td->rwmixread) |
286 | td->rwmix_ddir = DDIR_READ; | |
287 | else | |
288 | td->rwmix_ddir = DDIR_WRITE; | |
289 | memcpy(&td->rwmix_switch, &now, sizeof(now)); | |
290 | } | |
291 | return td->rwmix_ddir; | |
292 | } else if (td_read(td)) | |
3d60d1ed JA |
293 | return DDIR_READ; |
294 | else | |
295 | return DDIR_WRITE; | |
296 | } | |
297 | ||
aea47d44 | 298 | static int td_io_prep(struct thread_data *td, struct io_u *io_u) |
20dc95c4 | 299 | { |
2866c82d | 300 | if (td->io_ops->prep && td->io_ops->prep(td, io_u)) |
20dc95c4 JA |
301 | return 1; |
302 | ||
303 | return 0; | |
304 | } | |
305 | ||
b1ff3403 | 306 | void put_io_u(struct thread_data *td, struct io_u *io_u) |
ebac4655 | 307 | { |
53cdc686 | 308 | io_u->file = NULL; |
ebac4655 JA |
309 | list_del(&io_u->list); |
310 | list_add(&io_u->list, &td->io_u_freelist); | |
311 | td->cur_depth--; | |
312 | } | |
313 | ||
53cdc686 JA |
314 | static int fill_io_u(struct thread_data *td, struct fio_file *f, |
315 | struct io_u *io_u) | |
843a7413 JA |
316 | { |
317 | /* | |
318 | * If using an iolog, grab next piece if any available. | |
319 | */ | |
320 | if (td->read_iolog) | |
321 | return read_iolog_get(td, io_u); | |
322 | ||
aea47d44 JA |
323 | /* |
324 | * No log, let the seq/rand engine retrieve the next position. | |
325 | */ | |
53cdc686 | 326 | if (!get_next_offset(td, f, &io_u->offset)) { |
aea47d44 JA |
327 | io_u->buflen = get_next_buflen(td); |
328 | ||
329 | if (io_u->buflen) { | |
330 | io_u->ddir = get_rw_ddir(td); | |
843a7413 JA |
331 | |
332 | /* | |
333 | * If using a write iolog, store this entry. | |
334 | */ | |
335 | if (td->write_iolog) | |
336 | write_iolog_put(td, io_u); | |
337 | ||
53cdc686 | 338 | io_u->file = f; |
aea47d44 JA |
339 | return 0; |
340 | } | |
341 | } | |
342 | ||
343 | return 1; | |
344 | } | |
345 | ||
22f78b32 | 346 | #define queue_full(td) list_empty(&(td)->io_u_freelist) |
ebac4655 | 347 | |
b1ff3403 | 348 | struct io_u *__get_io_u(struct thread_data *td) |
ebac4655 | 349 | { |
22f78b32 | 350 | struct io_u *io_u = NULL; |
ebac4655 | 351 | |
22f78b32 JA |
352 | if (!queue_full(td)) { |
353 | io_u = list_entry(td->io_u_freelist.next, struct io_u, list); | |
354 | ||
355 | io_u->error = 0; | |
356 | io_u->resid = 0; | |
357 | list_del(&io_u->list); | |
358 | list_add(&io_u->list, &td->io_u_busylist); | |
359 | td->cur_depth++; | |
360 | } | |
ebac4655 | 361 | |
ebac4655 JA |
362 | return io_u; |
363 | } | |
364 | ||
906c8d75 JA |
365 | /* |
366 | * Return an io_u to be processed. Gets a buflen and offset, sets direction, | |
367 | * etc. The returned io_u is fully ready to be prepped and submitted. | |
368 | */ | |
53cdc686 | 369 | static struct io_u *get_io_u(struct thread_data *td, struct fio_file *f) |
ebac4655 JA |
370 | { |
371 | struct io_u *io_u; | |
372 | ||
373 | io_u = __get_io_u(td); | |
374 | if (!io_u) | |
375 | return NULL; | |
376 | ||
20dc95c4 JA |
377 | if (td->zone_bytes >= td->zone_size) { |
378 | td->zone_bytes = 0; | |
53cdc686 | 379 | f->last_pos += td->zone_skip; |
20dc95c4 JA |
380 | } |
381 | ||
53cdc686 | 382 | if (fill_io_u(td, f, io_u)) { |
ebac4655 JA |
383 | put_io_u(td, io_u); |
384 | return NULL; | |
385 | } | |
386 | ||
b2a15192 JA |
387 | if (io_u->buflen + io_u->offset > f->file_size) { |
388 | if (td->io_ops->flags & FIO_RAWIO) { | |
389 | put_io_u(td, io_u); | |
390 | return NULL; | |
391 | } | |
392 | ||
53cdc686 | 393 | io_u->buflen = f->file_size - io_u->offset; |
b2a15192 | 394 | } |
ebac4655 JA |
395 | |
396 | if (!io_u->buflen) { | |
397 | put_io_u(td, io_u); | |
398 | return NULL; | |
399 | } | |
400 | ||
843a7413 | 401 | if (!td->read_iolog && !td->sequential) |
53cdc686 | 402 | mark_random_map(td, f, io_u); |
ebac4655 | 403 | |
53cdc686 | 404 | f->last_pos += io_u->buflen; |
ebac4655 JA |
405 | |
406 | if (td->verify != VERIFY_NONE) | |
e29d1b70 | 407 | populate_verify_io_u(td, io_u); |
ebac4655 | 408 | |
aea47d44 | 409 | if (td_io_prep(td, io_u)) { |
ebac4655 JA |
410 | put_io_u(td, io_u); |
411 | return NULL; | |
412 | } | |
413 | ||
414 | gettimeofday(&io_u->start_time, NULL); | |
415 | return io_u; | |
416 | } | |
417 | ||
418 | static inline void td_set_runstate(struct thread_data *td, int runstate) | |
419 | { | |
ebac4655 JA |
420 | td->runstate = runstate; |
421 | } | |
422 | ||
53cdc686 JA |
423 | static struct fio_file *get_next_file(struct thread_data *td) |
424 | { | |
0ab8db89 | 425 | unsigned int old_next_file = td->next_file; |
b2a15192 JA |
426 | struct fio_file *f; |
427 | ||
428 | do { | |
429 | f = &td->files[td->next_file]; | |
430 | ||
431 | td->next_file++; | |
432 | if (td->next_file >= td->nr_files) | |
433 | td->next_file = 0; | |
434 | ||
435 | if (f->fd != -1) | |
436 | break; | |
53cdc686 | 437 | |
b2a15192 JA |
438 | f = NULL; |
439 | } while (td->next_file != old_next_file); | |
53cdc686 JA |
440 | |
441 | return f; | |
442 | } | |
443 | ||
444 | static int td_io_sync(struct thread_data *td, struct fio_file *f) | |
ebac4655 | 445 | { |
2866c82d | 446 | if (td->io_ops->sync) |
53cdc686 | 447 | return td->io_ops->sync(td, f); |
ebac4655 JA |
448 | |
449 | return 0; | |
450 | } | |
451 | ||
45bee283 | 452 | static int td_io_getevents(struct thread_data *td, int min, int max, |
ebac4655 JA |
453 | struct timespec *t) |
454 | { | |
2866c82d | 455 | return td->io_ops->getevents(td, min, max, t); |
ebac4655 JA |
456 | } |
457 | ||
45bee283 | 458 | static int td_io_queue(struct thread_data *td, struct io_u *io_u) |
ebac4655 JA |
459 | { |
460 | gettimeofday(&io_u->issue_time, NULL); | |
461 | ||
2866c82d | 462 | return td->io_ops->queue(td, io_u); |
ebac4655 JA |
463 | } |
464 | ||
465 | #define iocb_time(iocb) ((unsigned long) (iocb)->data) | |
466 | ||
467 | static void io_completed(struct thread_data *td, struct io_u *io_u, | |
468 | struct io_completion_data *icd) | |
469 | { | |
470 | struct timeval e; | |
471 | unsigned long msec; | |
472 | ||
473 | gettimeofday(&e, NULL); | |
474 | ||
475 | if (!io_u->error) { | |
20dc95c4 | 476 | unsigned int bytes = io_u->buflen - io_u->resid; |
3d60d1ed | 477 | const int idx = io_u->ddir; |
ebac4655 JA |
478 | |
479 | td->io_blocks[idx]++; | |
20dc95c4 JA |
480 | td->io_bytes[idx] += bytes; |
481 | td->zone_bytes += bytes; | |
482 | td->this_io_bytes[idx] += bytes; | |
ebac4655 JA |
483 | |
484 | msec = mtime_since(&io_u->issue_time, &e); | |
485 | ||
3d60d1ed JA |
486 | add_clat_sample(td, idx, msec); |
487 | add_bw_sample(td, idx); | |
ebac4655 | 488 | |
3d60d1ed | 489 | if ((td_rw(td) || td_write(td)) && idx == DDIR_WRITE) |
ebac4655 JA |
490 | log_io_piece(td, io_u); |
491 | ||
20dc95c4 | 492 | icd->bytes_done[idx] += bytes; |
ebac4655 JA |
493 | } else |
494 | icd->error = io_u->error; | |
495 | } | |
496 | ||
497 | static void ios_completed(struct thread_data *td,struct io_completion_data *icd) | |
498 | { | |
499 | struct io_u *io_u; | |
500 | int i; | |
501 | ||
502 | icd->error = 0; | |
503 | icd->bytes_done[0] = icd->bytes_done[1] = 0; | |
504 | ||
505 | for (i = 0; i < icd->nr; i++) { | |
2866c82d | 506 | io_u = td->io_ops->event(td, i); |
ebac4655 JA |
507 | |
508 | io_completed(td, io_u, icd); | |
509 | put_io_u(td, io_u); | |
510 | } | |
511 | } | |
512 | ||
906c8d75 JA |
513 | /* |
514 | * When job exits, we can cancel the in-flight IO if we are using async | |
515 | * io. Attempt to do so. | |
516 | */ | |
ebac4655 JA |
517 | static void cleanup_pending_aio(struct thread_data *td) |
518 | { | |
519 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0}; | |
520 | struct list_head *entry, *n; | |
521 | struct io_completion_data icd; | |
522 | struct io_u *io_u; | |
523 | int r; | |
524 | ||
525 | /* | |
526 | * get immediately available events, if any | |
527 | */ | |
45bee283 | 528 | r = td_io_getevents(td, 0, td->cur_depth, &ts); |
ebac4655 JA |
529 | if (r > 0) { |
530 | icd.nr = r; | |
531 | ios_completed(td, &icd); | |
532 | } | |
533 | ||
534 | /* | |
535 | * now cancel remaining active events | |
536 | */ | |
2866c82d | 537 | if (td->io_ops->cancel) { |
ebac4655 JA |
538 | list_for_each_safe(entry, n, &td->io_u_busylist) { |
539 | io_u = list_entry(entry, struct io_u, list); | |
540 | ||
2866c82d | 541 | r = td->io_ops->cancel(td, io_u); |
ebac4655 JA |
542 | if (!r) |
543 | put_io_u(td, io_u); | |
544 | } | |
545 | } | |
546 | ||
547 | if (td->cur_depth) { | |
45bee283 | 548 | r = td_io_getevents(td, td->cur_depth, td->cur_depth, NULL); |
ebac4655 JA |
549 | if (r > 0) { |
550 | icd.nr = r; | |
551 | ios_completed(td, &icd); | |
552 | } | |
553 | } | |
554 | } | |
555 | ||
906c8d75 JA |
556 | /* |
557 | * The main verify engine. Runs over the writes we previusly submitted, | |
558 | * reads the blocks back in, and checks the crc/md5 of the data. | |
559 | */ | |
a9619d44 | 560 | void do_verify(struct thread_data *td) |
ebac4655 JA |
561 | { |
562 | struct timeval t; | |
563 | struct io_u *io_u, *v_io_u = NULL; | |
564 | struct io_completion_data icd; | |
53cdc686 | 565 | struct fio_file *f; |
e5b401d4 JA |
566 | int ret, i; |
567 | ||
568 | /* | |
569 | * sync io first and invalidate cache, to make sure we really | |
570 | * read from disk. | |
571 | */ | |
572 | for_each_file(td, f, i) { | |
573 | td_io_sync(td, f); | |
574 | file_invalidate_cache(td, f); | |
575 | } | |
ebac4655 JA |
576 | |
577 | td_set_runstate(td, TD_VERIFYING); | |
578 | ||
579 | do { | |
580 | if (td->terminate) | |
581 | break; | |
582 | ||
583 | gettimeofday(&t, NULL); | |
584 | if (runtime_exceeded(td, &t)) | |
585 | break; | |
586 | ||
587 | io_u = __get_io_u(td); | |
588 | if (!io_u) | |
589 | break; | |
590 | ||
aea47d44 | 591 | if (get_next_verify(td, io_u)) { |
ebac4655 JA |
592 | put_io_u(td, io_u); |
593 | break; | |
594 | } | |
595 | ||
53cdc686 JA |
596 | f = get_next_file(td); |
597 | if (!f) | |
598 | break; | |
599 | ||
600 | io_u->file = f; | |
601 | ||
aea47d44 | 602 | if (td_io_prep(td, io_u)) { |
ebac4655 JA |
603 | put_io_u(td, io_u); |
604 | break; | |
605 | } | |
606 | ||
45bee283 | 607 | ret = td_io_queue(td, io_u); |
ebac4655 JA |
608 | if (ret) { |
609 | put_io_u(td, io_u); | |
610 | td_verror(td, ret); | |
611 | break; | |
612 | } | |
613 | ||
614 | /* | |
615 | * we have one pending to verify, do that while | |
616 | * we are doing io on the next one | |
617 | */ | |
618 | if (do_io_u_verify(td, &v_io_u)) | |
619 | break; | |
620 | ||
45bee283 | 621 | ret = td_io_getevents(td, 1, 1, NULL); |
ebac4655 JA |
622 | if (ret != 1) { |
623 | if (ret < 0) | |
624 | td_verror(td, ret); | |
625 | break; | |
626 | } | |
627 | ||
2866c82d | 628 | v_io_u = td->io_ops->event(td, 0); |
ebac4655 JA |
629 | icd.nr = 1; |
630 | icd.error = 0; | |
631 | io_completed(td, v_io_u, &icd); | |
632 | ||
633 | if (icd.error) { | |
634 | td_verror(td, icd.error); | |
635 | put_io_u(td, v_io_u); | |
636 | v_io_u = NULL; | |
637 | break; | |
638 | } | |
639 | ||
ebac4655 JA |
640 | /* |
641 | * if we can't submit more io, we need to verify now | |
642 | */ | |
643 | if (queue_full(td) && do_io_u_verify(td, &v_io_u)) | |
644 | break; | |
645 | ||
646 | } while (1); | |
647 | ||
648 | do_io_u_verify(td, &v_io_u); | |
649 | ||
650 | if (td->cur_depth) | |
651 | cleanup_pending_aio(td); | |
652 | ||
653 | td_set_runstate(td, TD_RUNNING); | |
654 | } | |
655 | ||
b990b5c0 JA |
656 | /* |
657 | * Not really an io thread, all it does is burn CPU cycles in the specified | |
658 | * manner. | |
659 | */ | |
660 | static void do_cpuio(struct thread_data *td) | |
661 | { | |
662 | struct timeval e; | |
663 | int split = 100 / td->cpuload; | |
664 | int i = 0; | |
665 | ||
666 | while (!td->terminate) { | |
667 | gettimeofday(&e, NULL); | |
668 | ||
669 | if (runtime_exceeded(td, &e)) | |
670 | break; | |
671 | ||
672 | if (!(i % split)) | |
673 | __usec_sleep(10000); | |
674 | else | |
675 | usec_sleep(td, 10000); | |
676 | ||
677 | i++; | |
678 | } | |
679 | } | |
680 | ||
32cd46a0 | 681 | /* |
906c8d75 | 682 | * Main IO worker function. It retrieves io_u's to process and queues |
32cd46a0 JA |
683 | * and reaps them, checking for rate and errors along the way. |
684 | */ | |
ebac4655 JA |
685 | static void do_io(struct thread_data *td) |
686 | { | |
687 | struct io_completion_data icd; | |
688 | struct timeval s, e; | |
689 | unsigned long usec; | |
53cdc686 JA |
690 | struct fio_file *f; |
691 | int i; | |
ebac4655 | 692 | |
5853e5a8 JA |
693 | td_set_runstate(td, TD_RUNNING); |
694 | ||
ebac4655 JA |
695 | while (td->this_io_bytes[td->ddir] < td->io_size) { |
696 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0}; | |
697 | struct timespec *timeout; | |
698 | int ret, min_evts = 0; | |
699 | struct io_u *io_u; | |
700 | ||
701 | if (td->terminate) | |
702 | break; | |
703 | ||
53cdc686 JA |
704 | f = get_next_file(td); |
705 | if (!f) | |
706 | break; | |
707 | ||
708 | io_u = get_io_u(td, f); | |
ebac4655 JA |
709 | if (!io_u) |
710 | break; | |
711 | ||
712 | memcpy(&s, &io_u->start_time, sizeof(s)); | |
713 | ||
45bee283 | 714 | ret = td_io_queue(td, io_u); |
ebac4655 JA |
715 | if (ret) { |
716 | put_io_u(td, io_u); | |
717 | td_verror(td, ret); | |
718 | break; | |
719 | } | |
720 | ||
721 | add_slat_sample(td, io_u->ddir, mtime_since(&io_u->start_time, &io_u->issue_time)); | |
722 | ||
723 | if (td->cur_depth < td->iodepth) { | |
724 | timeout = &ts; | |
725 | min_evts = 0; | |
726 | } else { | |
727 | timeout = NULL; | |
728 | min_evts = 1; | |
729 | } | |
730 | ||
45bee283 | 731 | ret = td_io_getevents(td, min_evts, td->cur_depth, timeout); |
ebac4655 JA |
732 | if (ret < 0) { |
733 | td_verror(td, ret); | |
734 | break; | |
735 | } else if (!ret) | |
736 | continue; | |
737 | ||
738 | icd.nr = ret; | |
739 | ios_completed(td, &icd); | |
740 | if (icd.error) { | |
741 | td_verror(td, icd.error); | |
742 | break; | |
743 | } | |
744 | ||
745 | /* | |
746 | * the rate is batched for now, it should work for batches | |
747 | * of completions except the very first one which may look | |
748 | * a little bursty | |
749 | */ | |
750 | gettimeofday(&e, NULL); | |
751 | usec = utime_since(&s, &e); | |
752 | ||
753 | rate_throttle(td, usec, icd.bytes_done[td->ddir]); | |
754 | ||
755 | if (check_min_rate(td, &e)) { | |
756 | td_verror(td, ENOMEM); | |
757 | break; | |
758 | } | |
759 | ||
760 | if (runtime_exceeded(td, &e)) | |
761 | break; | |
762 | ||
763 | if (td->thinktime) | |
764 | usec_sleep(td, td->thinktime); | |
765 | ||
766 | if (should_fsync(td) && td->fsync_blocks && | |
767 | (td->io_blocks[DDIR_WRITE] % td->fsync_blocks) == 0) | |
53cdc686 | 768 | td_io_sync(td, f); |
ebac4655 JA |
769 | } |
770 | ||
771 | if (td->cur_depth) | |
772 | cleanup_pending_aio(td); | |
773 | ||
5853e5a8 JA |
774 | if (should_fsync(td) && td->end_fsync) { |
775 | td_set_runstate(td, TD_FSYNCING); | |
53cdc686 JA |
776 | for_each_file(td, f, i) |
777 | td_io_sync(td, f); | |
5853e5a8 | 778 | } |
ebac4655 JA |
779 | } |
780 | ||
45bee283 | 781 | static int td_io_init(struct thread_data *td) |
ebac4655 | 782 | { |
2866c82d JA |
783 | if (td->io_ops->init) |
784 | return td->io_ops->init(td); | |
785 | ||
786 | return 0; | |
ebac4655 JA |
787 | } |
788 | ||
789 | static void cleanup_io_u(struct thread_data *td) | |
790 | { | |
791 | struct list_head *entry, *n; | |
792 | struct io_u *io_u; | |
793 | ||
794 | list_for_each_safe(entry, n, &td->io_u_freelist) { | |
795 | io_u = list_entry(entry, struct io_u, list); | |
796 | ||
797 | list_del(&io_u->list); | |
798 | free(io_u); | |
799 | } | |
800 | ||
801 | if (td->mem_type == MEM_MALLOC) | |
802 | free(td->orig_buffer); | |
803 | else if (td->mem_type == MEM_SHM) { | |
804 | struct shmid_ds sbuf; | |
805 | ||
806 | shmdt(td->orig_buffer); | |
807 | shmctl(td->shm_id, IPC_RMID, &sbuf); | |
808 | } else if (td->mem_type == MEM_MMAP) | |
809 | munmap(td->orig_buffer, td->orig_buffer_size); | |
810 | else | |
3b70d7e5 | 811 | log_err("Bad memory type %d\n", td->mem_type); |
ebac4655 JA |
812 | |
813 | td->orig_buffer = NULL; | |
814 | } | |
815 | ||
816 | static int init_io_u(struct thread_data *td) | |
817 | { | |
818 | struct io_u *io_u; | |
819 | int i, max_units; | |
820 | char *p; | |
821 | ||
2866c82d | 822 | if (td->io_ops->flags & FIO_CPUIO) |
b990b5c0 JA |
823 | return 0; |
824 | ||
2866c82d | 825 | if (td->io_ops->flags & FIO_SYNCIO) |
ebac4655 JA |
826 | max_units = 1; |
827 | else | |
828 | max_units = td->iodepth; | |
829 | ||
830 | td->orig_buffer_size = td->max_bs * max_units + MASK; | |
831 | ||
832 | if (td->mem_type == MEM_MALLOC) | |
833 | td->orig_buffer = malloc(td->orig_buffer_size); | |
834 | else if (td->mem_type == MEM_SHM) { | |
835 | td->shm_id = shmget(IPC_PRIVATE, td->orig_buffer_size, IPC_CREAT | 0600); | |
836 | if (td->shm_id < 0) { | |
837 | td_verror(td, errno); | |
838 | perror("shmget"); | |
839 | return 1; | |
840 | } | |
841 | ||
842 | td->orig_buffer = shmat(td->shm_id, NULL, 0); | |
843 | if (td->orig_buffer == (void *) -1) { | |
844 | td_verror(td, errno); | |
845 | perror("shmat"); | |
846 | td->orig_buffer = NULL; | |
847 | return 1; | |
848 | } | |
849 | } else if (td->mem_type == MEM_MMAP) { | |
850 | td->orig_buffer = mmap(NULL, td->orig_buffer_size, PROT_READ | PROT_WRITE, MAP_PRIVATE | OS_MAP_ANON, 0, 0); | |
851 | if (td->orig_buffer == MAP_FAILED) { | |
852 | td_verror(td, errno); | |
853 | perror("mmap"); | |
854 | td->orig_buffer = NULL; | |
855 | return 1; | |
856 | } | |
857 | } | |
858 | ||
ebac4655 JA |
859 | p = ALIGN(td->orig_buffer); |
860 | for (i = 0; i < max_units; i++) { | |
861 | io_u = malloc(sizeof(*io_u)); | |
862 | memset(io_u, 0, sizeof(*io_u)); | |
863 | INIT_LIST_HEAD(&io_u->list); | |
864 | ||
865 | io_u->buf = p + td->max_bs * i; | |
b1ff3403 | 866 | io_u->index = i; |
ebac4655 JA |
867 | list_add(&io_u->list, &td->io_u_freelist); |
868 | } | |
869 | ||
870 | return 0; | |
871 | } | |
872 | ||
da86774e JA |
873 | static int switch_ioscheduler(struct thread_data *td) |
874 | { | |
875 | char tmp[256], tmp2[128]; | |
876 | FILE *f; | |
877 | int ret; | |
878 | ||
879 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); | |
880 | ||
881 | f = fopen(tmp, "r+"); | |
882 | if (!f) { | |
883 | td_verror(td, errno); | |
884 | return 1; | |
885 | } | |
886 | ||
887 | /* | |
888 | * Set io scheduler. | |
889 | */ | |
890 | ret = fwrite(td->ioscheduler, strlen(td->ioscheduler), 1, f); | |
891 | if (ferror(f) || ret != 1) { | |
892 | td_verror(td, errno); | |
893 | fclose(f); | |
894 | return 1; | |
895 | } | |
896 | ||
897 | rewind(f); | |
898 | ||
899 | /* | |
900 | * Read back and check that the selected scheduler is now the default. | |
901 | */ | |
902 | ret = fread(tmp, 1, sizeof(tmp), f); | |
903 | if (ferror(f) || ret < 0) { | |
904 | td_verror(td, errno); | |
905 | fclose(f); | |
906 | return 1; | |
907 | } | |
908 | ||
909 | sprintf(tmp2, "[%s]", td->ioscheduler); | |
910 | if (!strstr(tmp, tmp2)) { | |
3b70d7e5 | 911 | log_err("fio: io scheduler %s not found\n", td->ioscheduler); |
da86774e JA |
912 | td_verror(td, EINVAL); |
913 | fclose(f); | |
914 | return 1; | |
915 | } | |
916 | ||
917 | fclose(f); | |
918 | return 0; | |
919 | } | |
920 | ||
ebac4655 JA |
921 | static void clear_io_state(struct thread_data *td) |
922 | { | |
53cdc686 JA |
923 | struct fio_file *f; |
924 | int i; | |
ebac4655 | 925 | |
ebac4655 JA |
926 | td->stat_io_bytes[0] = td->stat_io_bytes[1] = 0; |
927 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; | |
20dc95c4 | 928 | td->zone_bytes = 0; |
ebac4655 | 929 | |
53cdc686 JA |
930 | for_each_file(td, f, i) { |
931 | f->last_pos = 0; | |
932 | if (td->io_ops->flags & FIO_SYNCIO) | |
933 | lseek(f->fd, SEEK_SET, 0); | |
934 | ||
935 | if (f->file_map) | |
936 | memset(f->file_map, 0, f->num_maps * sizeof(long)); | |
937 | } | |
ebac4655 JA |
938 | } |
939 | ||
906c8d75 JA |
940 | /* |
941 | * Entry point for the thread based jobs. The process based jobs end up | |
942 | * here as well, after a little setup. | |
943 | */ | |
ebac4655 JA |
944 | static void *thread_main(void *data) |
945 | { | |
946 | struct thread_data *td = data; | |
ebac4655 JA |
947 | |
948 | if (!td->use_thread) | |
949 | setsid(); | |
950 | ||
951 | td->pid = getpid(); | |
952 | ||
aea47d44 JA |
953 | INIT_LIST_HEAD(&td->io_u_freelist); |
954 | INIT_LIST_HEAD(&td->io_u_busylist); | |
955 | INIT_LIST_HEAD(&td->io_hist_list); | |
956 | INIT_LIST_HEAD(&td->io_log_list); | |
957 | ||
ebac4655 JA |
958 | if (init_io_u(td)) |
959 | goto err; | |
960 | ||
961 | if (fio_setaffinity(td) == -1) { | |
962 | td_verror(td, errno); | |
963 | goto err; | |
964 | } | |
965 | ||
45bee283 | 966 | if (td_io_init(td)) |
ebac4655 JA |
967 | goto err; |
968 | ||
aea47d44 JA |
969 | if (init_iolog(td)) |
970 | goto err; | |
971 | ||
ebac4655 JA |
972 | if (td->ioprio) { |
973 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { | |
974 | td_verror(td, errno); | |
975 | goto err; | |
976 | } | |
977 | } | |
978 | ||
1056eaad | 979 | if (nice(td->nice) == -1) { |
b6f4d880 JA |
980 | td_verror(td, errno); |
981 | goto err; | |
982 | } | |
983 | ||
75154845 JA |
984 | if (init_random_state(td)) |
985 | goto err; | |
986 | ||
da86774e JA |
987 | if (td->ioscheduler && switch_ioscheduler(td)) |
988 | goto err; | |
989 | ||
75154845 | 990 | td_set_runstate(td, TD_INITIALIZED); |
bbfd6b00 JA |
991 | fio_sem_up(&startup_sem); |
992 | fio_sem_down(&td->mutex); | |
ebac4655 | 993 | |
53cdc686 | 994 | if (!td->create_serialize && setup_files(td)) |
ebac4655 JA |
995 | goto err; |
996 | ||
ebac4655 JA |
997 | gettimeofday(&td->epoch, NULL); |
998 | ||
4e0ba8af JA |
999 | if (td->exec_prerun) |
1000 | system(td->exec_prerun); | |
1001 | ||
ebac4655 JA |
1002 | while (td->loops--) { |
1003 | getrusage(RUSAGE_SELF, &td->ru_start); | |
1004 | gettimeofday(&td->start, NULL); | |
1005 | memcpy(&td->stat_sample_time, &td->start, sizeof(td->start)); | |
1006 | ||
1007 | if (td->ratemin) | |
1008 | memcpy(&td->lastrate, &td->stat_sample_time, sizeof(td->lastrate)); | |
1009 | ||
1010 | clear_io_state(td); | |
1011 | prune_io_piece_log(td); | |
1012 | ||
2866c82d | 1013 | if (td->io_ops->flags & FIO_CPUIO) |
b990b5c0 JA |
1014 | do_cpuio(td); |
1015 | else | |
1016 | do_io(td); | |
ebac4655 JA |
1017 | |
1018 | td->runtime[td->ddir] += mtime_since_now(&td->start); | |
aea47d44 | 1019 | if (td_rw(td) && td->io_bytes[td->ddir ^ 1]) |
3d60d1ed JA |
1020 | td->runtime[td->ddir ^ 1] = td->runtime[td->ddir]; |
1021 | ||
ebac4655 JA |
1022 | update_rusage_stat(td); |
1023 | ||
1024 | if (td->error || td->terminate) | |
1025 | break; | |
1026 | ||
1027 | if (td->verify == VERIFY_NONE) | |
1028 | continue; | |
1029 | ||
1030 | clear_io_state(td); | |
1031 | gettimeofday(&td->start, NULL); | |
1032 | ||
1033 | do_verify(td); | |
1034 | ||
1035 | td->runtime[DDIR_READ] += mtime_since_now(&td->start); | |
1036 | ||
1037 | if (td->error || td->terminate) | |
1038 | break; | |
1039 | } | |
1040 | ||
ebac4655 JA |
1041 | if (td->bw_log) |
1042 | finish_log(td, td->bw_log, "bw"); | |
1043 | if (td->slat_log) | |
1044 | finish_log(td, td->slat_log, "slat"); | |
1045 | if (td->clat_log) | |
1046 | finish_log(td, td->clat_log, "clat"); | |
843a7413 JA |
1047 | if (td->write_iolog) |
1048 | write_iolog_close(td); | |
4e0ba8af JA |
1049 | if (td->exec_postrun) |
1050 | system(td->exec_postrun); | |
ebac4655 JA |
1051 | |
1052 | if (exitall_on_terminate) | |
1053 | terminate_threads(td->groupid); | |
1054 | ||
1055 | err: | |
53cdc686 | 1056 | close_files(td); |
2866c82d | 1057 | close_ioengine(td); |
ebac4655 | 1058 | cleanup_io_u(td); |
ebac4655 JA |
1059 | td_set_runstate(td, TD_EXITED); |
1060 | return NULL; | |
1061 | ||
1062 | } | |
1063 | ||
906c8d75 JA |
1064 | /* |
1065 | * We cannot pass the td data into a forked process, so attach the td and | |
1066 | * pass it to the thread worker. | |
1067 | */ | |
ebac4655 JA |
1068 | static void *fork_main(int shmid, int offset) |
1069 | { | |
1070 | struct thread_data *td; | |
1071 | void *data; | |
1072 | ||
1073 | data = shmat(shmid, NULL, 0); | |
1074 | if (data == (void *) -1) { | |
1075 | perror("shmat"); | |
1076 | return NULL; | |
1077 | } | |
1078 | ||
1079 | td = data + offset * sizeof(struct thread_data); | |
1080 | thread_main(td); | |
1081 | shmdt(data); | |
1082 | return NULL; | |
1083 | } | |
1084 | ||
906c8d75 JA |
1085 | /* |
1086 | * Run over the job map and reap the threads that have exited, if any. | |
1087 | */ | |
ebac4655 JA |
1088 | static void reap_threads(int *nr_running, int *t_rate, int *m_rate) |
1089 | { | |
b990b5c0 | 1090 | int i, cputhreads; |
ebac4655 JA |
1091 | |
1092 | /* | |
1093 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
1094 | */ | |
b990b5c0 | 1095 | for (i = 0, cputhreads = 0; i < thread_number; i++) { |
ebac4655 JA |
1096 | struct thread_data *td = &threads[i]; |
1097 | ||
2866c82d | 1098 | if (td->io_ops->flags & FIO_CPUIO) |
b990b5c0 JA |
1099 | cputhreads++; |
1100 | ||
ebac4655 JA |
1101 | if (td->runstate != TD_EXITED) |
1102 | continue; | |
1103 | ||
1104 | td_set_runstate(td, TD_REAPED); | |
1105 | ||
1106 | if (td->use_thread) { | |
1107 | long ret; | |
1108 | ||
1109 | if (pthread_join(td->thread, (void *) &ret)) | |
1110 | perror("thread_join"); | |
1111 | } else | |
1112 | waitpid(td->pid, NULL, 0); | |
1113 | ||
1114 | (*nr_running)--; | |
1115 | (*m_rate) -= td->ratemin; | |
1116 | (*t_rate) -= td->rate; | |
1117 | } | |
b990b5c0 JA |
1118 | |
1119 | if (*nr_running == cputhreads) | |
1120 | terminate_threads(TERMINATE_ALL); | |
ebac4655 JA |
1121 | } |
1122 | ||
fcb6ade2 JA |
1123 | static void fio_unpin_memory(void *pinned) |
1124 | { | |
1125 | if (pinned) { | |
1126 | if (munlock(pinned, mlock_size) < 0) | |
1127 | perror("munlock"); | |
1128 | munmap(pinned, mlock_size); | |
1129 | } | |
1130 | } | |
1131 | ||
1132 | static void *fio_pin_memory(void) | |
1133 | { | |
32cd46a0 | 1134 | unsigned long long phys_mem; |
fcb6ade2 JA |
1135 | void *ptr; |
1136 | ||
1137 | if (!mlock_size) | |
1138 | return NULL; | |
1139 | ||
1140 | /* | |
1141 | * Don't allow mlock of more than real_mem-128MB | |
1142 | */ | |
32cd46a0 JA |
1143 | phys_mem = os_phys_mem(); |
1144 | if (phys_mem) { | |
1145 | if ((mlock_size + 128 * 1024 * 1024) > phys_mem) { | |
1146 | mlock_size = phys_mem - 128 * 1024 * 1024; | |
eb8bbf48 | 1147 | fprintf(f_out, "fio: limiting mlocked memory to %lluMiB\n", mlock_size >> 20); |
fcb6ade2 JA |
1148 | } |
1149 | } | |
1150 | ||
1151 | ptr = mmap(NULL, mlock_size, PROT_READ | PROT_WRITE, MAP_PRIVATE | OS_MAP_ANON, 0, 0); | |
1152 | if (!ptr) { | |
1153 | perror("malloc locked mem"); | |
1154 | return NULL; | |
1155 | } | |
1156 | if (mlock(ptr, mlock_size) < 0) { | |
1157 | munmap(ptr, mlock_size); | |
1158 | perror("mlock"); | |
1159 | return NULL; | |
1160 | } | |
1161 | ||
1162 | return ptr; | |
1163 | } | |
1164 | ||
906c8d75 JA |
1165 | /* |
1166 | * Main function for kicking off and reaping jobs, as needed. | |
1167 | */ | |
ebac4655 JA |
1168 | static void run_threads(void) |
1169 | { | |
ebac4655 JA |
1170 | struct thread_data *td; |
1171 | unsigned long spent; | |
1172 | int i, todo, nr_running, m_rate, t_rate, nr_started; | |
fcb6ade2 JA |
1173 | void *mlocked_mem; |
1174 | ||
1175 | mlocked_mem = fio_pin_memory(); | |
ebac4655 | 1176 | |
c6ae0a5b JA |
1177 | if (!terse_output) { |
1178 | printf("Starting %d thread%s\n", thread_number, thread_number > 1 ? "s" : ""); | |
1179 | fflush(stdout); | |
1180 | } | |
c04f7ec3 | 1181 | |
4efa970e JA |
1182 | signal(SIGINT, sig_handler); |
1183 | signal(SIGALRM, sig_handler); | |
1184 | ||
ebac4655 JA |
1185 | todo = thread_number; |
1186 | nr_running = 0; | |
1187 | nr_started = 0; | |
1188 | m_rate = t_rate = 0; | |
1189 | ||
1190 | for (i = 0; i < thread_number; i++) { | |
1191 | td = &threads[i]; | |
1192 | ||
263e529f | 1193 | print_status_init(td->thread_number - 1); |
ebac4655 JA |
1194 | |
1195 | init_disk_util(td); | |
1196 | ||
1197 | if (!td->create_serialize) | |
1198 | continue; | |
1199 | ||
1200 | /* | |
1201 | * do file setup here so it happens sequentially, | |
1202 | * we don't want X number of threads getting their | |
1203 | * client data interspersed on disk | |
1204 | */ | |
53cdc686 | 1205 | if (setup_files(td)) { |
ebac4655 JA |
1206 | td_set_runstate(td, TD_REAPED); |
1207 | todo--; | |
1208 | } | |
1209 | } | |
1210 | ||
263e529f | 1211 | time_init(); |
ebac4655 JA |
1212 | |
1213 | while (todo) { | |
75154845 JA |
1214 | struct thread_data *map[MAX_JOBS]; |
1215 | struct timeval this_start; | |
1216 | int this_jobs = 0, left; | |
1217 | ||
ebac4655 JA |
1218 | /* |
1219 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
1220 | */ | |
1221 | for (i = 0; i < thread_number; i++) { | |
1222 | td = &threads[i]; | |
1223 | ||
1224 | if (td->runstate != TD_NOT_CREATED) | |
1225 | continue; | |
1226 | ||
1227 | /* | |
1228 | * never got a chance to start, killed by other | |
1229 | * thread for some reason | |
1230 | */ | |
1231 | if (td->terminate) { | |
1232 | todo--; | |
1233 | continue; | |
1234 | } | |
1235 | ||
1236 | if (td->start_delay) { | |
263e529f | 1237 | spent = mtime_since_genesis(); |
ebac4655 JA |
1238 | |
1239 | if (td->start_delay * 1000 > spent) | |
1240 | continue; | |
1241 | } | |
1242 | ||
1243 | if (td->stonewall && (nr_started || nr_running)) | |
1244 | break; | |
1245 | ||
75154845 JA |
1246 | /* |
1247 | * Set state to created. Thread will transition | |
1248 | * to TD_INITIALIZED when it's done setting up. | |
1249 | */ | |
ebac4655 | 1250 | td_set_runstate(td, TD_CREATED); |
75154845 | 1251 | map[this_jobs++] = td; |
bbfd6b00 | 1252 | fio_sem_init(&startup_sem, 1); |
ebac4655 JA |
1253 | nr_started++; |
1254 | ||
1255 | if (td->use_thread) { | |
1256 | if (pthread_create(&td->thread, NULL, thread_main, td)) { | |
1257 | perror("thread_create"); | |
1258 | nr_started--; | |
1259 | } | |
1260 | } else { | |
1261 | if (fork()) | |
bbfd6b00 | 1262 | fio_sem_down(&startup_sem); |
ebac4655 JA |
1263 | else { |
1264 | fork_main(shm_id, i); | |
1265 | exit(0); | |
1266 | } | |
1267 | } | |
1268 | } | |
1269 | ||
1270 | /* | |
75154845 JA |
1271 | * Wait for the started threads to transition to |
1272 | * TD_INITIALIZED. | |
ebac4655 | 1273 | */ |
75154845 JA |
1274 | gettimeofday(&this_start, NULL); |
1275 | left = this_jobs; | |
1276 | while (left) { | |
1277 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) | |
1278 | break; | |
1279 | ||
1280 | usleep(100000); | |
1281 | ||
1282 | for (i = 0; i < this_jobs; i++) { | |
1283 | td = map[i]; | |
1284 | if (!td) | |
1285 | continue; | |
b6f4d880 | 1286 | if (td->runstate == TD_INITIALIZED) { |
75154845 JA |
1287 | map[i] = NULL; |
1288 | left--; | |
b6f4d880 JA |
1289 | } else if (td->runstate >= TD_EXITED) { |
1290 | map[i] = NULL; | |
1291 | left--; | |
1292 | todo--; | |
1293 | nr_running++; /* work-around... */ | |
75154845 JA |
1294 | } |
1295 | } | |
1296 | } | |
1297 | ||
1298 | if (left) { | |
3b70d7e5 | 1299 | log_err("fio: %d jobs failed to start\n", left); |
75154845 JA |
1300 | for (i = 0; i < this_jobs; i++) { |
1301 | td = map[i]; | |
1302 | if (!td) | |
1303 | continue; | |
1304 | kill(td->pid, SIGTERM); | |
1305 | } | |
1306 | break; | |
1307 | } | |
1308 | ||
1309 | /* | |
b6f4d880 | 1310 | * start created threads (TD_INITIALIZED -> TD_RUNNING). |
75154845 | 1311 | */ |
ebac4655 JA |
1312 | for (i = 0; i < thread_number; i++) { |
1313 | td = &threads[i]; | |
1314 | ||
75154845 | 1315 | if (td->runstate != TD_INITIALIZED) |
ebac4655 JA |
1316 | continue; |
1317 | ||
1318 | td_set_runstate(td, TD_RUNNING); | |
1319 | nr_running++; | |
1320 | nr_started--; | |
1321 | m_rate += td->ratemin; | |
1322 | t_rate += td->rate; | |
75154845 | 1323 | todo--; |
bbfd6b00 | 1324 | fio_sem_up(&td->mutex); |
ebac4655 JA |
1325 | } |
1326 | ||
1327 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1328 | ||
1329 | if (todo) | |
1330 | usleep(100000); | |
1331 | } | |
1332 | ||
1333 | while (nr_running) { | |
1334 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1335 | usleep(10000); | |
1336 | } | |
1337 | ||
1338 | update_io_ticks(); | |
fcb6ade2 | 1339 | fio_unpin_memory(mlocked_mem); |
ebac4655 JA |
1340 | } |
1341 | ||
ebac4655 JA |
1342 | int main(int argc, char *argv[]) |
1343 | { | |
1344 | if (parse_options(argc, argv)) | |
1345 | return 1; | |
1346 | ||
1347 | if (!thread_number) { | |
3b70d7e5 | 1348 | log_err("Nothing to do\n"); |
ebac4655 JA |
1349 | return 1; |
1350 | } | |
1351 | ||
1352 | disk_util_timer_arm(); | |
1353 | ||
1354 | run_threads(); | |
1355 | show_run_stats(); | |
1356 | ||
1357 | return 0; | |
1358 | } |