Commit | Line | Data |
---|---|---|
10ba535a JA |
1 | #include <unistd.h> |
2 | #include <fcntl.h> | |
3 | #include <string.h> | |
4 | #include <signal.h> | |
5 | #include <time.h> | |
0c6e7517 | 6 | #include <assert.h> |
10ba535a JA |
7 | |
8 | #include "fio.h" | |
5973cafb | 9 | #include "hash.h" |
4f5af7b2 | 10 | #include "verify.h" |
1fbbf72e | 11 | #include "lib/rand.h" |
10ba535a | 12 | |
97601024 JA |
13 | struct io_completion_data { |
14 | int nr; /* input */ | |
97601024 JA |
15 | |
16 | int error; /* output */ | |
17 | unsigned long bytes_done[2]; /* output */ | |
18 | struct timeval time; /* output */ | |
19 | }; | |
20 | ||
10ba535a JA |
21 | /* |
22 | * The ->file_map[] contains a map of blocks we have or have not done io | |
23 | * to yet. Used to make sure we cover the entire range in a fair fashion. | |
24 | */ | |
aec2de20 | 25 | static int random_map_free(struct fio_file *f, const unsigned long long block) |
10ba535a | 26 | { |
aec2de20 JA |
27 | unsigned int idx = RAND_MAP_IDX(f, block); |
28 | unsigned int bit = RAND_MAP_BIT(f, block); | |
10ba535a | 29 | |
84422acd JA |
30 | dprint(FD_RANDOM, "free: b=%llu, idx=%u, bit=%u\n", block, idx, bit); |
31 | ||
de605666 | 32 | return (f->file_map[idx] & (1 << bit)) == 0; |
10ba535a JA |
33 | } |
34 | ||
df415585 JA |
35 | /* |
36 | * Mark a given offset as used in the map. | |
37 | */ | |
9bf2061e | 38 | static void mark_random_map(struct thread_data *td, struct io_u *io_u) |
df415585 | 39 | { |
2dc1bbeb | 40 | unsigned int min_bs = td->o.rw_min_bs; |
9bf2061e | 41 | struct fio_file *f = io_u->file; |
a00735e6 | 42 | unsigned long long block; |
3e3357b1 | 43 | unsigned int blocks, nr_blocks; |
df415585 | 44 | |
b9c5b644 | 45 | block = (io_u->offset - f->file_offset) / (unsigned long long) min_bs; |
c685b5b2 | 46 | nr_blocks = (io_u->buflen + min_bs - 1) / min_bs; |
3e3357b1 | 47 | blocks = 0; |
c685b5b2 | 48 | |
3e3357b1 JA |
49 | while (nr_blocks) { |
50 | unsigned int this_blocks, mask; | |
df415585 JA |
51 | unsigned int idx, bit; |
52 | ||
1e3d53ac JA |
53 | /* |
54 | * If we have a mixed random workload, we may | |
55 | * encounter blocks we already did IO to. | |
56 | */ | |
8b113b4f JA |
57 | if ((td->o.ddir_nr == 1) && !random_map_free(f, block)) { |
58 | if (!blocks) | |
59 | blocks = 1; | |
df415585 | 60 | break; |
8b113b4f | 61 | } |
df415585 | 62 | |
aec2de20 JA |
63 | idx = RAND_MAP_IDX(f, block); |
64 | bit = RAND_MAP_BIT(f, block); | |
df415585 | 65 | |
0032bf9f | 66 | fio_assert(td, idx < f->num_maps); |
df415585 | 67 | |
3e3357b1 JA |
68 | this_blocks = nr_blocks; |
69 | if (this_blocks + bit > BLOCKS_PER_MAP) | |
70 | this_blocks = BLOCKS_PER_MAP - bit; | |
71 | ||
72 | if (this_blocks == BLOCKS_PER_MAP) | |
73 | mask = -1U; | |
74 | else | |
75 | mask = ((1U << this_blocks) - 1) << bit; | |
76 | ||
3e3357b1 JA |
77 | f->file_map[idx] |= mask; |
78 | nr_blocks -= this_blocks; | |
79 | blocks += this_blocks; | |
c9dd34b2 | 80 | block += this_blocks; |
df415585 JA |
81 | } |
82 | ||
a00735e6 JA |
83 | if ((blocks * min_bs) < io_u->buflen) |
84 | io_u->buflen = blocks * min_bs; | |
df415585 JA |
85 | } |
86 | ||
3e3357b1 JA |
87 | static unsigned long long last_block(struct thread_data *td, struct fio_file *f, |
88 | enum fio_ddir ddir) | |
2ba1c290 JA |
89 | { |
90 | unsigned long long max_blocks; | |
d9dd70f7 | 91 | unsigned long long max_size; |
2ba1c290 | 92 | |
d9dd70f7 JA |
93 | /* |
94 | * Hmm, should we make sure that ->io_size <= ->real_file_size? | |
95 | */ | |
96 | max_size = f->io_size; | |
97 | if (max_size > f->real_file_size) | |
98 | max_size = f->real_file_size; | |
99 | ||
2b7a01d0 | 100 | max_blocks = max_size / (unsigned long long) td->o.ba[ddir]; |
2ba1c290 JA |
101 | if (!max_blocks) |
102 | return 0; | |
103 | ||
67778e88 | 104 | return max_blocks; |
2ba1c290 JA |
105 | } |
106 | ||
10ba535a JA |
107 | /* |
108 | * Return the next free block in the map. | |
109 | */ | |
110 | static int get_next_free_block(struct thread_data *td, struct fio_file *f, | |
4ba66134 | 111 | enum fio_ddir ddir, unsigned long long *b) |
10ba535a | 112 | { |
84422acd | 113 | unsigned long long min_bs = td->o.rw_min_bs; |
10ba535a JA |
114 | int i; |
115 | ||
c685b5b2 JA |
116 | i = f->last_free_lookup; |
117 | *b = (i * BLOCKS_PER_MAP); | |
899c29f2 RR |
118 | while ((*b) * min_bs < f->real_file_size && |
119 | (*b) * min_bs < f->io_size) { | |
de605666 | 120 | if (f->file_map[i] != (unsigned int) -1) { |
697a606c | 121 | *b += ffz(f->file_map[i]); |
4ba66134 | 122 | if (*b > last_block(td, f, ddir)) |
2ba1c290 | 123 | break; |
c685b5b2 | 124 | f->last_free_lookup = i; |
10ba535a JA |
125 | return 0; |
126 | } | |
127 | ||
128 | *b += BLOCKS_PER_MAP; | |
129 | i++; | |
130 | } | |
131 | ||
2ba1c290 | 132 | dprint(FD_IO, "failed finding a free block\n"); |
10ba535a JA |
133 | return 1; |
134 | } | |
135 | ||
ec4015da | 136 | static int get_next_rand_offset(struct thread_data *td, struct fio_file *f, |
4ba66134 | 137 | enum fio_ddir ddir, unsigned long long *b) |
ec4015da | 138 | { |
84422acd | 139 | unsigned long long r; |
ec4015da JA |
140 | int loops = 5; |
141 | ||
142 | do { | |
143 | r = os_random_long(&td->random_state); | |
84422acd | 144 | dprint(FD_RANDOM, "off rand %llu\n", r); |
5ec10eaa | 145 | *b = (last_block(td, f, ddir) - 1) |
dc873b6f | 146 | * (r / ((unsigned long long) OS_RAND_MAX + 1.0)); |
2ba1c290 | 147 | |
43c63a78 JA |
148 | /* |
149 | * if we are not maintaining a random map, we are done. | |
150 | */ | |
303032ae | 151 | if (!file_randommap(td, f)) |
43c63a78 JA |
152 | return 0; |
153 | ||
154 | /* | |
84422acd | 155 | * calculate map offset and check if it's free |
43c63a78 | 156 | */ |
aec2de20 | 157 | if (random_map_free(f, *b)) |
43c63a78 JA |
158 | return 0; |
159 | ||
84422acd JA |
160 | dprint(FD_RANDOM, "get_next_rand_offset: offset %llu busy\n", |
161 | *b); | |
43c63a78 | 162 | } while (--loops); |
ec4015da JA |
163 | |
164 | /* | |
43c63a78 JA |
165 | * we get here, if we didn't suceed in looking up a block. generate |
166 | * a random start offset into the filemap, and find the first free | |
167 | * block from there. | |
ec4015da | 168 | */ |
43c63a78 JA |
169 | loops = 10; |
170 | do { | |
dc873b6f JA |
171 | f->last_free_lookup = (f->num_maps - 1) * |
172 | (r / (OS_RAND_MAX + 1.0)); | |
4ba66134 | 173 | if (!get_next_free_block(td, f, ddir, b)) |
43c63a78 | 174 | return 0; |
ec4015da | 175 | |
43c63a78 JA |
176 | r = os_random_long(&td->random_state); |
177 | } while (--loops); | |
178 | ||
179 | /* | |
180 | * that didn't work either, try exhaustive search from the start | |
181 | */ | |
182 | f->last_free_lookup = 0; | |
4ba66134 | 183 | return get_next_free_block(td, f, ddir, b); |
ec4015da JA |
184 | } |
185 | ||
10ba535a JA |
186 | /* |
187 | * For random io, generate a random new block and see if it's used. Repeat | |
188 | * until we find a free one. For sequential io, just return the end of | |
189 | * the last io issued. | |
190 | */ | |
15dc1934 | 191 | static int __get_next_offset(struct thread_data *td, struct io_u *io_u) |
10ba535a | 192 | { |
9bf2061e | 193 | struct fio_file *f = io_u->file; |
ec4015da | 194 | unsigned long long b; |
4ba66134 | 195 | enum fio_ddir ddir = io_u->ddir; |
10ba535a | 196 | |
ec4015da JA |
197 | if (td_random(td) && (td->o.ddir_nr && !--td->ddir_nr)) { |
198 | td->ddir_nr = td->o.ddir_nr; | |
211097b2 | 199 | |
683023e8 JA |
200 | if (get_next_rand_offset(td, f, ddir, &b)) { |
201 | dprint(FD_IO, "%s: getting rand offset failed\n", | |
202 | f->file_name); | |
bca4ed4d | 203 | return 1; |
683023e8 | 204 | } |
43063a1c | 205 | } else { |
0c3d768a | 206 | if (f->last_pos >= f->real_file_size) { |
4ba66134 | 207 | if (!td_random(td) || |
683023e8 JA |
208 | get_next_rand_offset(td, f, ddir, &b)) { |
209 | dprint(FD_IO, "%s: pos %llu > size %llu\n", | |
210 | f->file_name, f->last_pos, | |
211 | f->real_file_size); | |
0c3d768a | 212 | return 1; |
683023e8 | 213 | } |
bcdedd0a | 214 | } else |
4ba66134 | 215 | b = (f->last_pos - f->file_offset) / td->o.min_bs[ddir]; |
43063a1c | 216 | } |
10ba535a | 217 | |
2b7a01d0 | 218 | io_u->offset = b * td->o.ba[ddir]; |
009bd847 JA |
219 | if (io_u->offset >= f->io_size) { |
220 | dprint(FD_IO, "get_next_offset: offset %llu >= io_size %llu\n", | |
221 | io_u->offset, f->io_size); | |
222 | return 1; | |
223 | } | |
224 | ||
225 | io_u->offset += f->file_offset; | |
2ba1c290 JA |
226 | if (io_u->offset >= f->real_file_size) { |
227 | dprint(FD_IO, "get_next_offset: offset %llu >= size %llu\n", | |
228 | io_u->offset, f->real_file_size); | |
10ba535a | 229 | return 1; |
2ba1c290 | 230 | } |
10ba535a JA |
231 | |
232 | return 0; | |
233 | } | |
234 | ||
15dc1934 JA |
235 | static int get_next_offset(struct thread_data *td, struct io_u *io_u) |
236 | { | |
7eb36574 JA |
237 | struct prof_io_ops *ops = &td->prof_io_ops; |
238 | ||
239 | if (ops->fill_io_u_off) | |
240 | return ops->fill_io_u_off(td, io_u); | |
15dc1934 JA |
241 | |
242 | return __get_next_offset(td, io_u); | |
243 | } | |
244 | ||
245 | static unsigned int __get_next_buflen(struct thread_data *td, struct io_u *io_u) | |
10ba535a | 246 | { |
bca4ed4d | 247 | const int ddir = io_u->ddir; |
f3f552b9 | 248 | unsigned int uninitialized_var(buflen); |
f3059de1 | 249 | unsigned int minbs, maxbs; |
10ba535a JA |
250 | long r; |
251 | ||
f3059de1 JA |
252 | minbs = td->o.min_bs[ddir]; |
253 | maxbs = td->o.max_bs[ddir]; | |
254 | ||
255 | if (minbs == maxbs) | |
256 | buflen = minbs; | |
10ba535a JA |
257 | else { |
258 | r = os_random_long(&td->bsrange_state); | |
720e84ad | 259 | if (!td->o.bssplit_nr[ddir]) { |
f3059de1 JA |
260 | buflen = 1 + (unsigned int) ((double) maxbs * |
261 | (r / (OS_RAND_MAX + 1.0))); | |
262 | if (buflen < minbs) | |
263 | buflen = minbs; | |
5ec10eaa | 264 | } else { |
564ca972 JA |
265 | long perc = 0; |
266 | unsigned int i; | |
267 | ||
720e84ad JA |
268 | for (i = 0; i < td->o.bssplit_nr[ddir]; i++) { |
269 | struct bssplit *bsp = &td->o.bssplit[ddir][i]; | |
564ca972 JA |
270 | |
271 | buflen = bsp->bs; | |
272 | perc += bsp->perc; | |
f3059de1 | 273 | if (r <= ((OS_RAND_MAX / 100L) * perc)) |
564ca972 JA |
274 | break; |
275 | } | |
276 | } | |
f3059de1 JA |
277 | if (!td->o.bs_unaligned && is_power_of_2(minbs)) |
278 | buflen = (buflen + minbs - 1) & ~(minbs - 1); | |
10ba535a JA |
279 | } |
280 | ||
4ba66134 JA |
281 | if (io_u->offset + buflen > io_u->file->real_file_size) { |
282 | dprint(FD_IO, "lower buflen %u -> %u (ddir=%d)\n", buflen, | |
f3059de1 JA |
283 | minbs, ddir); |
284 | buflen = minbs; | |
4ba66134 | 285 | } |
6a5e6884 | 286 | |
10ba535a JA |
287 | return buflen; |
288 | } | |
289 | ||
15dc1934 JA |
290 | static unsigned int get_next_buflen(struct thread_data *td, struct io_u *io_u) |
291 | { | |
7eb36574 JA |
292 | struct prof_io_ops *ops = &td->prof_io_ops; |
293 | ||
294 | if (ops->fill_io_u_size) | |
295 | return ops->fill_io_u_size(td, io_u); | |
15dc1934 JA |
296 | |
297 | return __get_next_buflen(td, io_u); | |
298 | } | |
299 | ||
afe24a5a JA |
300 | static void set_rwmix_bytes(struct thread_data *td) |
301 | { | |
afe24a5a JA |
302 | unsigned int diff; |
303 | ||
304 | /* | |
305 | * we do time or byte based switch. this is needed because | |
306 | * buffered writes may issue a lot quicker than they complete, | |
307 | * whereas reads do not. | |
308 | */ | |
e47f799f | 309 | diff = td->o.rwmix[td->rwmix_ddir ^ 1]; |
04c540d9 | 310 | td->rwmix_issues = (td->io_issues[td->rwmix_ddir] * diff) / 100; |
e47f799f JA |
311 | } |
312 | ||
313 | static inline enum fio_ddir get_rand_ddir(struct thread_data *td) | |
314 | { | |
315 | unsigned int v; | |
316 | long r; | |
317 | ||
318 | r = os_random_long(&td->rwmix_state); | |
dc873b6f | 319 | v = 1 + (int) (100.0 * (r / (OS_RAND_MAX + 1.0))); |
04c540d9 | 320 | if (v <= td->o.rwmix[DDIR_READ]) |
e47f799f JA |
321 | return DDIR_READ; |
322 | ||
323 | return DDIR_WRITE; | |
afe24a5a JA |
324 | } |
325 | ||
581e7141 JA |
326 | static enum fio_ddir rate_ddir(struct thread_data *td, enum fio_ddir ddir) |
327 | { | |
328 | enum fio_ddir odir = ddir ^ 1; | |
329 | struct timeval t; | |
330 | long usec; | |
331 | ||
332 | if (td->rate_pending_usleep[ddir] <= 0) | |
333 | return ddir; | |
334 | ||
335 | /* | |
336 | * We have too much pending sleep in this direction. See if we | |
337 | * should switch. | |
338 | */ | |
339 | if (td_rw(td)) { | |
340 | /* | |
341 | * Other direction does not have too much pending, switch | |
342 | */ | |
343 | if (td->rate_pending_usleep[odir] < 100000) | |
344 | return odir; | |
345 | ||
346 | /* | |
347 | * Both directions have pending sleep. Sleep the minimum time | |
348 | * and deduct from both. | |
349 | */ | |
350 | if (td->rate_pending_usleep[ddir] <= | |
351 | td->rate_pending_usleep[odir]) { | |
352 | usec = td->rate_pending_usleep[ddir]; | |
353 | } else { | |
354 | usec = td->rate_pending_usleep[odir]; | |
355 | ddir = odir; | |
356 | } | |
357 | } else | |
358 | usec = td->rate_pending_usleep[ddir]; | |
359 | ||
360 | fio_gettime(&t, NULL); | |
361 | usec_sleep(td, usec); | |
362 | usec = utime_since_now(&t); | |
363 | ||
364 | td->rate_pending_usleep[ddir] -= usec; | |
365 | ||
366 | odir = ddir ^ 1; | |
367 | if (td_rw(td) && __should_check_rate(td, odir)) | |
368 | td->rate_pending_usleep[odir] -= usec; | |
0b9d69ec | 369 | |
581e7141 JA |
370 | return ddir; |
371 | } | |
372 | ||
10ba535a JA |
373 | /* |
374 | * Return the data direction for the next io_u. If the job is a | |
375 | * mixed read/write workload, check the rwmix cycle and switch if | |
376 | * necessary. | |
377 | */ | |
1e97cce9 | 378 | static enum fio_ddir get_rw_ddir(struct thread_data *td) |
10ba535a | 379 | { |
581e7141 JA |
380 | enum fio_ddir ddir; |
381 | ||
5f9099ea JA |
382 | /* |
383 | * see if it's time to fsync | |
384 | */ | |
385 | if (td->o.fsync_blocks && | |
386 | !(td->io_issues[DDIR_WRITE] % td->o.fsync_blocks) && | |
387 | td->io_issues[DDIR_WRITE] && should_fsync(td)) | |
388 | return DDIR_SYNC; | |
389 | ||
390 | /* | |
391 | * see if it's time to fdatasync | |
392 | */ | |
393 | if (td->o.fdatasync_blocks && | |
394 | !(td->io_issues[DDIR_WRITE] % td->o.fdatasync_blocks) && | |
395 | td->io_issues[DDIR_WRITE] && should_fsync(td)) | |
396 | return DDIR_DATASYNC; | |
397 | ||
44f29692 JA |
398 | /* |
399 | * see if it's time to sync_file_range | |
400 | */ | |
401 | if (td->sync_file_range_nr && | |
402 | !(td->io_issues[DDIR_WRITE] % td->sync_file_range_nr) && | |
403 | td->io_issues[DDIR_WRITE] && should_fsync(td)) | |
404 | return DDIR_SYNC_FILE_RANGE; | |
405 | ||
10ba535a | 406 | if (td_rw(td)) { |
10ba535a JA |
407 | /* |
408 | * Check if it's time to seed a new data direction. | |
409 | */ | |
e4928662 | 410 | if (td->io_issues[td->rwmix_ddir] >= td->rwmix_issues) { |
e47f799f JA |
411 | /* |
412 | * Put a top limit on how many bytes we do for | |
413 | * one data direction, to avoid overflowing the | |
414 | * ranges too much | |
415 | */ | |
416 | ddir = get_rand_ddir(td); | |
e47f799f JA |
417 | |
418 | if (ddir != td->rwmix_ddir) | |
419 | set_rwmix_bytes(td); | |
420 | ||
421 | td->rwmix_ddir = ddir; | |
10ba535a | 422 | } |
581e7141 | 423 | ddir = td->rwmix_ddir; |
10ba535a | 424 | } else if (td_read(td)) |
581e7141 | 425 | ddir = DDIR_READ; |
10ba535a | 426 | else |
581e7141 JA |
427 | ddir = DDIR_WRITE; |
428 | ||
429 | td->rwmix_ddir = rate_ddir(td, ddir); | |
430 | return td->rwmix_ddir; | |
10ba535a JA |
431 | } |
432 | ||
e8462bd8 | 433 | void put_file_log(struct thread_data *td, struct fio_file *f) |
60f2c658 JA |
434 | { |
435 | int ret = put_file(td, f); | |
436 | ||
437 | if (ret) | |
438 | td_verror(td, ret, "file close"); | |
439 | } | |
440 | ||
10ba535a JA |
441 | void put_io_u(struct thread_data *td, struct io_u *io_u) |
442 | { | |
e8462bd8 JA |
443 | td_io_u_lock(td); |
444 | ||
0c6e7517 | 445 | io_u->flags |= IO_U_F_FREE; |
e8462bd8 | 446 | io_u->flags &= ~IO_U_F_FREE_DEF; |
0c6e7517 | 447 | |
60f2c658 JA |
448 | if (io_u->file) |
449 | put_file_log(td, io_u->file); | |
2dbdab7e | 450 | |
10ba535a | 451 | io_u->file = NULL; |
0c41214f RR |
452 | if (io_u->flags & IO_U_F_IN_CUR_DEPTH) |
453 | td->cur_depth--; | |
e8462bd8 | 454 | flist_del_init(&io_u->list); |
01743ee1 | 455 | flist_add(&io_u->list, &td->io_u_freelist); |
e8462bd8 JA |
456 | td_io_u_unlock(td); |
457 | td_io_u_free_notify(td); | |
10ba535a JA |
458 | } |
459 | ||
f2bba182 RR |
460 | void clear_io_u(struct thread_data *td, struct io_u *io_u) |
461 | { | |
462 | io_u->flags &= ~IO_U_F_FLIGHT; | |
463 | put_io_u(td, io_u); | |
464 | } | |
465 | ||
755200a3 JA |
466 | void requeue_io_u(struct thread_data *td, struct io_u **io_u) |
467 | { | |
468 | struct io_u *__io_u = *io_u; | |
469 | ||
465221b0 JA |
470 | dprint(FD_IO, "requeue %p\n", __io_u); |
471 | ||
e8462bd8 JA |
472 | td_io_u_lock(td); |
473 | ||
4d2e0f49 | 474 | __io_u->flags |= IO_U_F_FREE; |
5f9099ea | 475 | if ((__io_u->flags & IO_U_F_FLIGHT) && !ddir_sync(__io_u->ddir)) |
e4f54adb | 476 | td->io_issues[__io_u->ddir]--; |
5ec10eaa | 477 | |
4d2e0f49 | 478 | __io_u->flags &= ~IO_U_F_FLIGHT; |
0c41214f RR |
479 | if (__io_u->flags & IO_U_F_IN_CUR_DEPTH) |
480 | td->cur_depth--; | |
01743ee1 JA |
481 | flist_del(&__io_u->list); |
482 | flist_add_tail(&__io_u->list, &td->io_u_requeues); | |
e8462bd8 | 483 | td_io_u_unlock(td); |
755200a3 JA |
484 | *io_u = NULL; |
485 | } | |
486 | ||
9bf2061e | 487 | static int fill_io_u(struct thread_data *td, struct io_u *io_u) |
10ba535a | 488 | { |
b4c5e1ac JA |
489 | if (td->io_ops->flags & FIO_NOIO) |
490 | goto out; | |
491 | ||
5f9099ea JA |
492 | io_u->ddir = get_rw_ddir(td); |
493 | ||
87dc1ab1 | 494 | /* |
5f9099ea | 495 | * fsync() or fdatasync(), we are done |
87dc1ab1 | 496 | */ |
5f9099ea | 497 | if (ddir_sync(io_u->ddir)) |
c38e9468 | 498 | goto out; |
a00735e6 | 499 | |
48f5abd3 JA |
500 | /* |
501 | * See if it's time to switch to a new zone | |
502 | */ | |
503 | if (td->zone_bytes >= td->o.zone_size) { | |
504 | td->zone_bytes = 0; | |
505 | io_u->file->last_pos += td->o.zone_skip; | |
506 | td->io_skip_bytes += td->o.zone_skip; | |
507 | } | |
508 | ||
10ba535a | 509 | /* |
c685b5b2 JA |
510 | * No log, let the seq/rand engine retrieve the next buflen and |
511 | * position. | |
10ba535a | 512 | */ |
2ba1c290 JA |
513 | if (get_next_offset(td, io_u)) { |
514 | dprint(FD_IO, "io_u %p, failed getting offset\n", io_u); | |
bca4ed4d | 515 | return 1; |
2ba1c290 | 516 | } |
10ba535a | 517 | |
9bf2061e | 518 | io_u->buflen = get_next_buflen(td, io_u); |
2ba1c290 JA |
519 | if (!io_u->buflen) { |
520 | dprint(FD_IO, "io_u %p, failed getting buflen\n", io_u); | |
bca4ed4d | 521 | return 1; |
2ba1c290 | 522 | } |
bca4ed4d | 523 | |
2ba1c290 JA |
524 | if (io_u->offset + io_u->buflen > io_u->file->real_file_size) { |
525 | dprint(FD_IO, "io_u %p, offset too large\n", io_u); | |
4ba66134 JA |
526 | dprint(FD_IO, " off=%llu/%lu > %llu\n", io_u->offset, |
527 | io_u->buflen, io_u->file->real_file_size); | |
6a5e6884 | 528 | return 1; |
2ba1c290 | 529 | } |
6a5e6884 | 530 | |
bca4ed4d JA |
531 | /* |
532 | * mark entry before potentially trimming io_u | |
533 | */ | |
303032ae | 534 | if (td_random(td) && file_randommap(td, io_u->file)) |
9bf2061e | 535 | mark_random_map(td, io_u); |
bca4ed4d JA |
536 | |
537 | /* | |
538 | * If using a write iolog, store this entry. | |
539 | */ | |
c38e9468 | 540 | out: |
2ba1c290 | 541 | dprint_io_u(io_u, "fill_io_u"); |
d9d91e39 | 542 | td->zone_bytes += io_u->buflen; |
f29b25a3 | 543 | log_io_u(td, io_u); |
bca4ed4d | 544 | return 0; |
10ba535a JA |
545 | } |
546 | ||
838bc709 JA |
547 | static void __io_u_mark_map(unsigned int *map, unsigned int nr) |
548 | { | |
549 | int index = 0; | |
550 | ||
551 | switch (nr) { | |
552 | default: | |
553 | index = 6; | |
554 | break; | |
555 | case 33 ... 64: | |
556 | index = 5; | |
557 | break; | |
558 | case 17 ... 32: | |
559 | index = 4; | |
560 | break; | |
561 | case 9 ... 16: | |
562 | index = 3; | |
563 | break; | |
564 | case 5 ... 8: | |
565 | index = 2; | |
566 | break; | |
567 | case 1 ... 4: | |
568 | index = 1; | |
569 | case 0: | |
570 | break; | |
571 | } | |
572 | ||
573 | map[index]++; | |
574 | } | |
575 | ||
576 | void io_u_mark_submit(struct thread_data *td, unsigned int nr) | |
577 | { | |
578 | __io_u_mark_map(td->ts.io_u_submit, nr); | |
579 | td->ts.total_submit++; | |
580 | } | |
581 | ||
582 | void io_u_mark_complete(struct thread_data *td, unsigned int nr) | |
583 | { | |
584 | __io_u_mark_map(td->ts.io_u_complete, nr); | |
585 | td->ts.total_complete++; | |
586 | } | |
587 | ||
d8005759 | 588 | void io_u_mark_depth(struct thread_data *td, unsigned int nr) |
71619dc2 JA |
589 | { |
590 | int index = 0; | |
591 | ||
592 | switch (td->cur_depth) { | |
593 | default: | |
a783e61a JA |
594 | index = 6; |
595 | break; | |
71619dc2 | 596 | case 32 ... 63: |
a783e61a JA |
597 | index = 5; |
598 | break; | |
71619dc2 | 599 | case 16 ... 31: |
a783e61a JA |
600 | index = 4; |
601 | break; | |
71619dc2 | 602 | case 8 ... 15: |
a783e61a JA |
603 | index = 3; |
604 | break; | |
71619dc2 | 605 | case 4 ... 7: |
a783e61a JA |
606 | index = 2; |
607 | break; | |
71619dc2 | 608 | case 2 ... 3: |
a783e61a | 609 | index = 1; |
71619dc2 JA |
610 | case 1: |
611 | break; | |
612 | } | |
613 | ||
3bec7ae0 | 614 | td->ts.io_u_map[index] += nr; |
71619dc2 JA |
615 | } |
616 | ||
04a0feae JA |
617 | static void io_u_mark_lat_usec(struct thread_data *td, unsigned long usec) |
618 | { | |
619 | int index = 0; | |
620 | ||
621 | assert(usec < 1000); | |
622 | ||
623 | switch (usec) { | |
624 | case 750 ... 999: | |
625 | index = 9; | |
626 | break; | |
627 | case 500 ... 749: | |
628 | index = 8; | |
629 | break; | |
630 | case 250 ... 499: | |
631 | index = 7; | |
632 | break; | |
633 | case 100 ... 249: | |
634 | index = 6; | |
635 | break; | |
636 | case 50 ... 99: | |
637 | index = 5; | |
638 | break; | |
639 | case 20 ... 49: | |
640 | index = 4; | |
641 | break; | |
642 | case 10 ... 19: | |
643 | index = 3; | |
644 | break; | |
645 | case 4 ... 9: | |
646 | index = 2; | |
647 | break; | |
648 | case 2 ... 3: | |
649 | index = 1; | |
650 | case 0 ... 1: | |
651 | break; | |
652 | } | |
653 | ||
654 | assert(index < FIO_IO_U_LAT_U_NR); | |
655 | td->ts.io_u_lat_u[index]++; | |
656 | } | |
657 | ||
658 | static void io_u_mark_lat_msec(struct thread_data *td, unsigned long msec) | |
ec118304 JA |
659 | { |
660 | int index = 0; | |
661 | ||
662 | switch (msec) { | |
663 | default: | |
04a0feae JA |
664 | index = 11; |
665 | break; | |
8abdce66 | 666 | case 1000 ... 1999: |
04a0feae JA |
667 | index = 10; |
668 | break; | |
8abdce66 | 669 | case 750 ... 999: |
04a0feae JA |
670 | index = 9; |
671 | break; | |
8abdce66 | 672 | case 500 ... 749: |
04a0feae JA |
673 | index = 8; |
674 | break; | |
8abdce66 | 675 | case 250 ... 499: |
04a0feae JA |
676 | index = 7; |
677 | break; | |
8abdce66 | 678 | case 100 ... 249: |
04a0feae JA |
679 | index = 6; |
680 | break; | |
8abdce66 | 681 | case 50 ... 99: |
04a0feae JA |
682 | index = 5; |
683 | break; | |
8abdce66 | 684 | case 20 ... 49: |
04a0feae JA |
685 | index = 4; |
686 | break; | |
8abdce66 | 687 | case 10 ... 19: |
04a0feae JA |
688 | index = 3; |
689 | break; | |
8abdce66 | 690 | case 4 ... 9: |
04a0feae JA |
691 | index = 2; |
692 | break; | |
ec118304 | 693 | case 2 ... 3: |
04a0feae | 694 | index = 1; |
ec118304 JA |
695 | case 0 ... 1: |
696 | break; | |
697 | } | |
698 | ||
04a0feae JA |
699 | assert(index < FIO_IO_U_LAT_M_NR); |
700 | td->ts.io_u_lat_m[index]++; | |
701 | } | |
702 | ||
703 | static void io_u_mark_latency(struct thread_data *td, unsigned long usec) | |
704 | { | |
705 | if (usec < 1000) | |
706 | io_u_mark_lat_usec(td, usec); | |
707 | else | |
708 | io_u_mark_lat_msec(td, usec / 1000); | |
ec118304 JA |
709 | } |
710 | ||
0aabe160 JA |
711 | /* |
712 | * Get next file to service by choosing one at random | |
713 | */ | |
2cc52930 JA |
714 | static struct fio_file *get_next_file_rand(struct thread_data *td, |
715 | enum fio_file_flags goodf, | |
d6aed795 | 716 | enum fio_file_flags badf) |
0aabe160 | 717 | { |
0aabe160 | 718 | struct fio_file *f; |
1c178180 | 719 | int fno; |
0aabe160 JA |
720 | |
721 | do { | |
7c83c089 | 722 | long r = os_random_long(&td->next_file_state); |
87b10676 | 723 | int opened = 0; |
7c83c089 | 724 | |
5ec10eaa | 725 | fno = (unsigned int) ((double) td->o.nr_files |
dc873b6f | 726 | * (r / (OS_RAND_MAX + 1.0))); |
126d65c6 | 727 | f = td->files[fno]; |
d6aed795 | 728 | if (fio_file_done(f)) |
059e63c0 | 729 | continue; |
1c178180 | 730 | |
d6aed795 | 731 | if (!fio_file_open(f)) { |
87b10676 JA |
732 | int err; |
733 | ||
734 | err = td_io_open_file(td, f); | |
735 | if (err) | |
736 | continue; | |
737 | opened = 1; | |
738 | } | |
739 | ||
2ba1c290 JA |
740 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) { |
741 | dprint(FD_FILE, "get_next_file_rand: %p\n", f); | |
0aabe160 | 742 | return f; |
2ba1c290 | 743 | } |
87b10676 JA |
744 | if (opened) |
745 | td_io_close_file(td, f); | |
0aabe160 JA |
746 | } while (1); |
747 | } | |
748 | ||
749 | /* | |
750 | * Get next file to service by doing round robin between all available ones | |
751 | */ | |
1c178180 JA |
752 | static struct fio_file *get_next_file_rr(struct thread_data *td, int goodf, |
753 | int badf) | |
3d7c391d JA |
754 | { |
755 | unsigned int old_next_file = td->next_file; | |
756 | struct fio_file *f; | |
757 | ||
758 | do { | |
87b10676 JA |
759 | int opened = 0; |
760 | ||
126d65c6 | 761 | f = td->files[td->next_file]; |
3d7c391d JA |
762 | |
763 | td->next_file++; | |
2dc1bbeb | 764 | if (td->next_file >= td->o.nr_files) |
3d7c391d JA |
765 | td->next_file = 0; |
766 | ||
87b10676 | 767 | dprint(FD_FILE, "trying file %s %x\n", f->file_name, f->flags); |
d6aed795 | 768 | if (fio_file_done(f)) { |
d5ed68ea | 769 | f = NULL; |
059e63c0 | 770 | continue; |
d5ed68ea | 771 | } |
059e63c0 | 772 | |
d6aed795 | 773 | if (!fio_file_open(f)) { |
87b10676 JA |
774 | int err; |
775 | ||
776 | err = td_io_open_file(td, f); | |
b5696bfc JA |
777 | if (err) { |
778 | dprint(FD_FILE, "error %d on open of %s\n", | |
779 | err, f->file_name); | |
87c27b45 | 780 | f = NULL; |
87b10676 | 781 | continue; |
b5696bfc | 782 | } |
87b10676 JA |
783 | opened = 1; |
784 | } | |
785 | ||
0b9d69ec JA |
786 | dprint(FD_FILE, "goodf=%x, badf=%x, ff=%x\n", goodf, badf, |
787 | f->flags); | |
1c178180 | 788 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) |
3d7c391d JA |
789 | break; |
790 | ||
87b10676 JA |
791 | if (opened) |
792 | td_io_close_file(td, f); | |
793 | ||
3d7c391d JA |
794 | f = NULL; |
795 | } while (td->next_file != old_next_file); | |
796 | ||
2ba1c290 | 797 | dprint(FD_FILE, "get_next_file_rr: %p\n", f); |
3d7c391d JA |
798 | return f; |
799 | } | |
800 | ||
7eb36574 | 801 | static struct fio_file *__get_next_file(struct thread_data *td) |
bdb4e2e9 | 802 | { |
1907dbc6 JA |
803 | struct fio_file *f; |
804 | ||
2dc1bbeb | 805 | assert(td->o.nr_files <= td->files_index); |
1c178180 | 806 | |
b5696bfc | 807 | if (td->nr_done_files >= td->o.nr_files) { |
5ec10eaa JA |
808 | dprint(FD_FILE, "get_next_file: nr_open=%d, nr_done=%d," |
809 | " nr_files=%d\n", td->nr_open_files, | |
810 | td->nr_done_files, | |
811 | td->o.nr_files); | |
bdb4e2e9 | 812 | return NULL; |
2ba1c290 | 813 | } |
bdb4e2e9 | 814 | |
1907dbc6 | 815 | f = td->file_service_file; |
d6aed795 | 816 | if (f && fio_file_open(f) && !fio_file_closing(f)) { |
a086c257 JA |
817 | if (td->o.file_service_type == FIO_FSERVICE_SEQ) |
818 | goto out; | |
819 | if (td->file_service_left--) | |
820 | goto out; | |
821 | } | |
1907dbc6 | 822 | |
a086c257 JA |
823 | if (td->o.file_service_type == FIO_FSERVICE_RR || |
824 | td->o.file_service_type == FIO_FSERVICE_SEQ) | |
d6aed795 | 825 | f = get_next_file_rr(td, FIO_FILE_open, FIO_FILE_closing); |
bdb4e2e9 | 826 | else |
d6aed795 | 827 | f = get_next_file_rand(td, FIO_FILE_open, FIO_FILE_closing); |
1907dbc6 JA |
828 | |
829 | td->file_service_file = f; | |
830 | td->file_service_left = td->file_service_nr - 1; | |
2ba1c290 | 831 | out: |
683023e8 | 832 | dprint(FD_FILE, "get_next_file: %p [%s]\n", f, f->file_name); |
1907dbc6 | 833 | return f; |
bdb4e2e9 JA |
834 | } |
835 | ||
7eb36574 JA |
836 | static struct fio_file *get_next_file(struct thread_data *td) |
837 | { | |
838 | struct prof_io_ops *ops = &td->prof_io_ops; | |
839 | ||
840 | if (ops->get_next_file) | |
841 | return ops->get_next_file(td); | |
842 | ||
843 | return __get_next_file(td); | |
844 | } | |
845 | ||
429f6675 JA |
846 | static int set_io_u_file(struct thread_data *td, struct io_u *io_u) |
847 | { | |
848 | struct fio_file *f; | |
849 | ||
850 | do { | |
851 | f = get_next_file(td); | |
852 | if (!f) | |
853 | return 1; | |
854 | ||
429f6675 JA |
855 | io_u->file = f; |
856 | get_file(f); | |
857 | ||
858 | if (!fill_io_u(td, io_u)) | |
859 | break; | |
860 | ||
b5696bfc | 861 | put_file_log(td, f); |
429f6675 | 862 | td_io_close_file(td, f); |
b5696bfc | 863 | io_u->file = NULL; |
d6aed795 | 864 | fio_file_set_done(f); |
429f6675 | 865 | td->nr_done_files++; |
0b9d69ec JA |
866 | dprint(FD_FILE, "%s: is done (%d of %d)\n", f->file_name, |
867 | td->nr_done_files, td->o.nr_files); | |
429f6675 JA |
868 | } while (1); |
869 | ||
870 | return 0; | |
871 | } | |
872 | ||
873 | ||
10ba535a JA |
874 | struct io_u *__get_io_u(struct thread_data *td) |
875 | { | |
876 | struct io_u *io_u = NULL; | |
877 | ||
e8462bd8 JA |
878 | td_io_u_lock(td); |
879 | ||
880 | again: | |
01743ee1 JA |
881 | if (!flist_empty(&td->io_u_requeues)) |
882 | io_u = flist_entry(td->io_u_requeues.next, struct io_u, list); | |
755200a3 | 883 | else if (!queue_full(td)) { |
01743ee1 | 884 | io_u = flist_entry(td->io_u_freelist.next, struct io_u, list); |
10ba535a | 885 | |
6040dabc | 886 | io_u->buflen = 0; |
10ba535a | 887 | io_u->resid = 0; |
755200a3 | 888 | io_u->file = NULL; |
d7762cf8 | 889 | io_u->end_io = NULL; |
755200a3 JA |
890 | } |
891 | ||
892 | if (io_u) { | |
0c6e7517 | 893 | assert(io_u->flags & IO_U_F_FREE); |
2ecc1b57 | 894 | io_u->flags &= ~(IO_U_F_FREE | IO_U_F_FREE_DEF); |
0c6e7517 | 895 | |
755200a3 | 896 | io_u->error = 0; |
01743ee1 JA |
897 | flist_del(&io_u->list); |
898 | flist_add(&io_u->list, &td->io_u_busylist); | |
10ba535a | 899 | td->cur_depth++; |
0c41214f | 900 | io_u->flags |= IO_U_F_IN_CUR_DEPTH; |
1dec3e07 JA |
901 | } else if (td->o.verify_async) { |
902 | /* | |
903 | * We ran out, wait for async verify threads to finish and | |
904 | * return one | |
905 | */ | |
906 | pthread_cond_wait(&td->free_cond, &td->io_u_lock); | |
907 | goto again; | |
10ba535a JA |
908 | } |
909 | ||
e8462bd8 | 910 | td_io_u_unlock(td); |
10ba535a JA |
911 | return io_u; |
912 | } | |
913 | ||
914 | /* | |
915 | * Return an io_u to be processed. Gets a buflen and offset, sets direction, | |
916 | * etc. The returned io_u is fully ready to be prepped and submitted. | |
917 | */ | |
3d7c391d | 918 | struct io_u *get_io_u(struct thread_data *td) |
10ba535a | 919 | { |
3d7c391d | 920 | struct fio_file *f; |
10ba535a JA |
921 | struct io_u *io_u; |
922 | ||
923 | io_u = __get_io_u(td); | |
2ba1c290 JA |
924 | if (!io_u) { |
925 | dprint(FD_IO, "__get_io_u failed\n"); | |
10ba535a | 926 | return NULL; |
2ba1c290 | 927 | } |
10ba535a | 928 | |
755200a3 JA |
929 | /* |
930 | * from a requeue, io_u already setup | |
931 | */ | |
932 | if (io_u->file) | |
77f392bf | 933 | goto out; |
755200a3 | 934 | |
429f6675 JA |
935 | /* |
936 | * If using an iolog, grab next piece if any available. | |
937 | */ | |
938 | if (td->o.read_iolog_file) { | |
939 | if (read_iolog_get(td, io_u)) | |
940 | goto err_put; | |
2ba1c290 JA |
941 | } else if (set_io_u_file(td, io_u)) { |
942 | dprint(FD_IO, "io_u %p, setting file failed\n", io_u); | |
429f6675 | 943 | goto err_put; |
2ba1c290 | 944 | } |
5ec10eaa | 945 | |
429f6675 | 946 | f = io_u->file; |
d6aed795 | 947 | assert(fio_file_open(f)); |
97af62ce | 948 | |
5f9099ea | 949 | if (!ddir_sync(io_u->ddir)) { |
d0656a93 | 950 | if (!io_u->buflen && !(td->io_ops->flags & FIO_NOIO)) { |
2ba1c290 | 951 | dprint(FD_IO, "get_io_u: zero buflen on %p\n", io_u); |
429f6675 | 952 | goto err_put; |
2ba1c290 | 953 | } |
10ba535a | 954 | |
36167d82 | 955 | f->last_pos = io_u->offset + io_u->buflen; |
10ba535a | 956 | |
c311cd2a | 957 | if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_WRITE) |
87dc1ab1 | 958 | populate_verify_io_u(td, io_u); |
e4dad9c6 JA |
959 | else if (td->o.refill_buffers && io_u->ddir == DDIR_WRITE) |
960 | io_u_fill_buffer(td, io_u, io_u->xfer_buflen); | |
87dc1ab1 | 961 | } |
10ba535a | 962 | |
165faf16 JA |
963 | /* |
964 | * Set io data pointers. | |
965 | */ | |
cec6b55d JA |
966 | io_u->xfer_buf = io_u->buf; |
967 | io_u->xfer_buflen = io_u->buflen; | |
5973cafb | 968 | |
6ac7a331 | 969 | out: |
429f6675 | 970 | if (!td_io_prep(td, io_u)) { |
993bf48b JA |
971 | if (!td->o.disable_slat) |
972 | fio_gettime(&io_u->start_time, NULL); | |
429f6675 | 973 | return io_u; |
36167d82 | 974 | } |
429f6675 | 975 | err_put: |
2ba1c290 | 976 | dprint(FD_IO, "get_io_u failed\n"); |
429f6675 JA |
977 | put_io_u(td, io_u); |
978 | return NULL; | |
10ba535a JA |
979 | } |
980 | ||
5451792e JA |
981 | void io_u_log_error(struct thread_data *td, struct io_u *io_u) |
982 | { | |
983 | const char *msg[] = { "read", "write", "sync" }; | |
984 | ||
985 | log_err("fio: io_u error"); | |
986 | ||
987 | if (io_u->file) | |
988 | log_err(" on file %s", io_u->file->file_name); | |
989 | ||
990 | log_err(": %s\n", strerror(io_u->error)); | |
991 | ||
5ec10eaa JA |
992 | log_err(" %s offset=%llu, buflen=%lu\n", msg[io_u->ddir], |
993 | io_u->offset, io_u->xfer_buflen); | |
5451792e JA |
994 | |
995 | if (!td->error) | |
996 | td_verror(td, io_u->error, "io_u error"); | |
997 | } | |
998 | ||
97601024 JA |
999 | static void io_completed(struct thread_data *td, struct io_u *io_u, |
1000 | struct io_completion_data *icd) | |
10ba535a | 1001 | { |
dbad30b0 JA |
1002 | /* |
1003 | * Older gcc's are too dumb to realize that usec is always used | |
1004 | * initialized, silence that warning. | |
1005 | */ | |
1006 | unsigned long uninitialized_var(usec); | |
44f29692 | 1007 | struct fio_file *f; |
10ba535a | 1008 | |
2ba1c290 JA |
1009 | dprint_io_u(io_u, "io complete"); |
1010 | ||
2ecc1b57 | 1011 | td_io_u_lock(td); |
0c6e7517 JA |
1012 | assert(io_u->flags & IO_U_F_FLIGHT); |
1013 | io_u->flags &= ~IO_U_F_FLIGHT; | |
2ecc1b57 | 1014 | td_io_u_unlock(td); |
0c6e7517 | 1015 | |
5f9099ea | 1016 | if (ddir_sync(io_u->ddir)) { |
87dc1ab1 | 1017 | td->last_was_sync = 1; |
44f29692 JA |
1018 | f = io_u->file; |
1019 | if (f) { | |
1020 | f->first_write = -1ULL; | |
1021 | f->last_write = -1ULL; | |
1022 | } | |
87dc1ab1 JA |
1023 | return; |
1024 | } | |
1025 | ||
1026 | td->last_was_sync = 0; | |
1027 | ||
10ba535a JA |
1028 | if (!io_u->error) { |
1029 | unsigned int bytes = io_u->buflen - io_u->resid; | |
1e97cce9 | 1030 | const enum fio_ddir idx = io_u->ddir; |
ba3e4e0c | 1031 | const enum fio_ddir odx = io_u->ddir ^ 1; |
b29ee5b3 | 1032 | int ret; |
10ba535a | 1033 | |
b29ee5b3 JA |
1034 | td->io_blocks[idx]++; |
1035 | td->io_bytes[idx] += bytes; | |
1036 | td->this_io_bytes[idx] += bytes; | |
10ba535a | 1037 | |
44f29692 JA |
1038 | if (idx == DDIR_WRITE) { |
1039 | f = io_u->file; | |
1040 | if (f) { | |
1041 | if (f->first_write == -1ULL || | |
1042 | io_u->offset < f->first_write) | |
1043 | f->first_write = io_u->offset; | |
1044 | if (f->last_write == -1ULL || | |
1045 | ((io_u->offset + bytes) > f->last_write)) | |
1046 | f->last_write = io_u->offset + bytes; | |
1047 | } | |
1048 | } | |
1049 | ||
b29ee5b3 | 1050 | if (ramp_time_over(td)) { |
40e1a6f0 | 1051 | unsigned long uninitialized_var(lusec); |
40e1a6f0 JA |
1052 | |
1053 | if (!td->o.disable_clat || !td->o.disable_bw) | |
1054 | lusec = utime_since(&io_u->issue_time, | |
1055 | &icd->time); | |
721938ae | 1056 | |
9520ebb9 | 1057 | if (!td->o.disable_clat) { |
01a1caa5 | 1058 | add_clat_sample(td, idx, lusec, bytes); |
40e1a6f0 | 1059 | io_u_mark_latency(td, lusec); |
9520ebb9 JA |
1060 | } |
1061 | if (!td->o.disable_bw) | |
306ddc97 | 1062 | add_bw_sample(td, idx, bytes, &icd->time); |
b23b6a2f | 1063 | if (__should_check_rate(td, idx)) { |
ba3e4e0c RR |
1064 | td->rate_pending_usleep[idx] = |
1065 | ((td->this_io_bytes[idx] * | |
1066 | td->rate_nsec_cycle[idx]) / 1000 - | |
1067 | utime_since_now(&td->start)); | |
b23b6a2f | 1068 | } |
581e7141 | 1069 | if (__should_check_rate(td, idx ^ 1)) |
ba3e4e0c RR |
1070 | td->rate_pending_usleep[odx] = |
1071 | ((td->this_io_bytes[odx] * | |
1072 | td->rate_nsec_cycle[odx]) / 1000 - | |
1073 | utime_since_now(&td->start)); | |
721938ae | 1074 | } |
10ba535a | 1075 | |
660a1cb5 | 1076 | if (td_write(td) && idx == DDIR_WRITE && |
8670579e | 1077 | td->o.do_verify && |
41128405 | 1078 | td->o.verify != VERIFY_NONE) |
10ba535a JA |
1079 | log_io_piece(td, io_u); |
1080 | ||
b29ee5b3 | 1081 | icd->bytes_done[idx] += bytes; |
3af6ef39 | 1082 | |
d7762cf8 | 1083 | if (io_u->end_io) { |
36690c9b | 1084 | ret = io_u->end_io(td, io_u); |
3af6ef39 JA |
1085 | if (ret && !icd->error) |
1086 | icd->error = ret; | |
1087 | } | |
5451792e | 1088 | } else { |
10ba535a | 1089 | icd->error = io_u->error; |
5451792e JA |
1090 | io_u_log_error(td, io_u); |
1091 | } | |
f2bba182 RR |
1092 | if (td->o.continue_on_error && icd->error && |
1093 | td_non_fatal_error(icd->error)) { | |
1094 | /* | |
1095 | * If there is a non_fatal error, then add to the error count | |
1096 | * and clear all the errors. | |
1097 | */ | |
1098 | update_error_count(td, icd->error); | |
1099 | td_clear_error(td); | |
1100 | icd->error = 0; | |
1101 | io_u->error = 0; | |
1102 | } | |
10ba535a JA |
1103 | } |
1104 | ||
9520ebb9 JA |
1105 | static void init_icd(struct thread_data *td, struct io_completion_data *icd, |
1106 | int nr) | |
10ba535a | 1107 | { |
9520ebb9 JA |
1108 | if (!td->o.disable_clat || !td->o.disable_bw) |
1109 | fio_gettime(&icd->time, NULL); | |
02bcaa8c | 1110 | |
3af6ef39 JA |
1111 | icd->nr = nr; |
1112 | ||
10ba535a JA |
1113 | icd->error = 0; |
1114 | icd->bytes_done[0] = icd->bytes_done[1] = 0; | |
36167d82 JA |
1115 | } |
1116 | ||
97601024 JA |
1117 | static void ios_completed(struct thread_data *td, |
1118 | struct io_completion_data *icd) | |
36167d82 JA |
1119 | { |
1120 | struct io_u *io_u; | |
1121 | int i; | |
1122 | ||
10ba535a JA |
1123 | for (i = 0; i < icd->nr; i++) { |
1124 | io_u = td->io_ops->event(td, i); | |
1125 | ||
1126 | io_completed(td, io_u, icd); | |
e8462bd8 JA |
1127 | |
1128 | if (!(io_u->flags & IO_U_F_FREE_DEF)) | |
1129 | put_io_u(td, io_u); | |
10ba535a JA |
1130 | } |
1131 | } | |
97601024 | 1132 | |
e7e6cfb4 JA |
1133 | /* |
1134 | * Complete a single io_u for the sync engines. | |
1135 | */ | |
581e7141 JA |
1136 | int io_u_sync_complete(struct thread_data *td, struct io_u *io_u, |
1137 | unsigned long *bytes) | |
97601024 JA |
1138 | { |
1139 | struct io_completion_data icd; | |
1140 | ||
9520ebb9 | 1141 | init_icd(td, &icd, 1); |
97601024 | 1142 | io_completed(td, io_u, &icd); |
e8462bd8 JA |
1143 | |
1144 | if (!(io_u->flags & IO_U_F_FREE_DEF)) | |
1145 | put_io_u(td, io_u); | |
97601024 | 1146 | |
581e7141 JA |
1147 | if (icd.error) { |
1148 | td_verror(td, icd.error, "io_u_sync_complete"); | |
1149 | return -1; | |
1150 | } | |
97601024 | 1151 | |
581e7141 JA |
1152 | if (bytes) { |
1153 | bytes[0] += icd.bytes_done[0]; | |
1154 | bytes[1] += icd.bytes_done[1]; | |
1155 | } | |
1156 | ||
1157 | return 0; | |
97601024 JA |
1158 | } |
1159 | ||
e7e6cfb4 JA |
1160 | /* |
1161 | * Called to complete min_events number of io for the async engines. | |
1162 | */ | |
581e7141 JA |
1163 | int io_u_queued_complete(struct thread_data *td, int min_evts, |
1164 | unsigned long *bytes) | |
97601024 | 1165 | { |
97601024 | 1166 | struct io_completion_data icd; |
00de55ef | 1167 | struct timespec *tvp = NULL; |
97601024 | 1168 | int ret; |
4d06a338 | 1169 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0, }; |
97601024 | 1170 | |
4950421a | 1171 | dprint(FD_IO, "io_u_queued_completed: min=%d\n", min_evts); |
b271fe62 | 1172 | |
4950421a | 1173 | if (!min_evts) |
00de55ef | 1174 | tvp = &ts; |
97601024 | 1175 | |
4950421a | 1176 | ret = td_io_getevents(td, min_evts, td->o.iodepth_batch_complete, tvp); |
97601024 | 1177 | if (ret < 0) { |
e1161c32 | 1178 | td_verror(td, -ret, "td_io_getevents"); |
97601024 JA |
1179 | return ret; |
1180 | } else if (!ret) | |
1181 | return ret; | |
1182 | ||
9520ebb9 | 1183 | init_icd(td, &icd, ret); |
97601024 | 1184 | ios_completed(td, &icd); |
581e7141 JA |
1185 | if (icd.error) { |
1186 | td_verror(td, icd.error, "io_u_queued_complete"); | |
1187 | return -1; | |
1188 | } | |
97601024 | 1189 | |
581e7141 JA |
1190 | if (bytes) { |
1191 | bytes[0] += icd.bytes_done[0]; | |
1192 | bytes[1] += icd.bytes_done[1]; | |
1193 | } | |
1194 | ||
1195 | return 0; | |
97601024 | 1196 | } |
7e77dd02 JA |
1197 | |
1198 | /* | |
1199 | * Call when io_u is really queued, to update the submission latency. | |
1200 | */ | |
1201 | void io_u_queued(struct thread_data *td, struct io_u *io_u) | |
1202 | { | |
9520ebb9 JA |
1203 | if (!td->o.disable_slat) { |
1204 | unsigned long slat_time; | |
7e77dd02 | 1205 | |
9520ebb9 | 1206 | slat_time = utime_since(&io_u->start_time, &io_u->issue_time); |
29a90ddb | 1207 | add_slat_sample(td, io_u->ddir, slat_time, io_u->xfer_buflen); |
9520ebb9 | 1208 | } |
7e77dd02 | 1209 | } |
433afcb4 | 1210 | |
5973cafb JA |
1211 | /* |
1212 | * "randomly" fill the buffer contents | |
1213 | */ | |
1214 | void io_u_fill_buffer(struct thread_data *td, struct io_u *io_u, | |
1215 | unsigned int max_bs) | |
1216 | { | |
1217 | long *ptr = io_u->buf; | |
1218 | ||
1219 | if (!td->o.zero_buffers) { | |
d77431b5 | 1220 | unsigned long r = __rand(&__fio_rand_state); |
27ea68a5 | 1221 | |
d77431b5 JA |
1222 | if (sizeof(int) != sizeof(*ptr)) |
1223 | r *= (unsigned long) __rand(&__fio_rand_state); | |
27ea68a5 | 1224 | |
d77431b5 | 1225 | while ((void *) ptr - io_u->buf < max_bs) { |
27ea68a5 | 1226 | *ptr = r; |
5973cafb | 1227 | ptr++; |
d77431b5 JA |
1228 | r *= GOLDEN_RATIO_PRIME; |
1229 | r >>= 3; | |
5973cafb JA |
1230 | } |
1231 | } else | |
1232 | memset(ptr, 0, max_bs); | |
1233 | } |