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