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 | */ | |
da3758bf | 57 | if ((td->o.ddir_nr == 1) && !random_map_free(f, block)) |
df415585 JA |
58 | break; |
59 | ||
aec2de20 JA |
60 | idx = RAND_MAP_IDX(f, block); |
61 | bit = RAND_MAP_BIT(f, block); | |
df415585 | 62 | |
0032bf9f | 63 | fio_assert(td, idx < f->num_maps); |
df415585 | 64 | |
3e3357b1 JA |
65 | this_blocks = nr_blocks; |
66 | if (this_blocks + bit > BLOCKS_PER_MAP) | |
67 | this_blocks = BLOCKS_PER_MAP - bit; | |
68 | ||
da3758bf JA |
69 | do { |
70 | if (this_blocks == BLOCKS_PER_MAP) | |
71 | mask = -1U; | |
72 | else | |
73 | mask = ((1U << this_blocks) - 1) << bit; | |
74 | ||
75 | if (!(f->file_map[idx] & mask)) | |
76 | break; | |
77 | ||
78 | this_blocks--; | |
79 | } while (this_blocks); | |
80 | ||
81 | if (!this_blocks) | |
82 | break; | |
3e3357b1 | 83 | |
3e3357b1 JA |
84 | f->file_map[idx] |= mask; |
85 | nr_blocks -= this_blocks; | |
86 | blocks += this_blocks; | |
c9dd34b2 | 87 | block += this_blocks; |
df415585 JA |
88 | } |
89 | ||
a00735e6 JA |
90 | if ((blocks * min_bs) < io_u->buflen) |
91 | io_u->buflen = blocks * min_bs; | |
df415585 JA |
92 | } |
93 | ||
3e3357b1 JA |
94 | static unsigned long long last_block(struct thread_data *td, struct fio_file *f, |
95 | enum fio_ddir ddir) | |
2ba1c290 JA |
96 | { |
97 | unsigned long long max_blocks; | |
d9dd70f7 | 98 | unsigned long long max_size; |
2ba1c290 | 99 | |
d9dd70f7 JA |
100 | /* |
101 | * Hmm, should we make sure that ->io_size <= ->real_file_size? | |
102 | */ | |
103 | max_size = f->io_size; | |
104 | if (max_size > f->real_file_size) | |
105 | max_size = f->real_file_size; | |
106 | ||
2b7a01d0 | 107 | max_blocks = max_size / (unsigned long long) td->o.ba[ddir]; |
2ba1c290 JA |
108 | if (!max_blocks) |
109 | return 0; | |
110 | ||
67778e88 | 111 | return max_blocks; |
2ba1c290 JA |
112 | } |
113 | ||
10ba535a JA |
114 | /* |
115 | * Return the next free block in the map. | |
116 | */ | |
117 | static int get_next_free_block(struct thread_data *td, struct fio_file *f, | |
4ba66134 | 118 | enum fio_ddir ddir, unsigned long long *b) |
10ba535a | 119 | { |
84422acd | 120 | unsigned long long min_bs = td->o.rw_min_bs; |
10ba535a JA |
121 | int i; |
122 | ||
c685b5b2 JA |
123 | i = f->last_free_lookup; |
124 | *b = (i * BLOCKS_PER_MAP); | |
899c29f2 RR |
125 | while ((*b) * min_bs < f->real_file_size && |
126 | (*b) * min_bs < f->io_size) { | |
de605666 | 127 | if (f->file_map[i] != (unsigned int) -1) { |
697a606c | 128 | *b += ffz(f->file_map[i]); |
4ba66134 | 129 | if (*b > last_block(td, f, ddir)) |
2ba1c290 | 130 | break; |
c685b5b2 | 131 | f->last_free_lookup = i; |
10ba535a JA |
132 | return 0; |
133 | } | |
134 | ||
135 | *b += BLOCKS_PER_MAP; | |
136 | i++; | |
137 | } | |
138 | ||
2ba1c290 | 139 | dprint(FD_IO, "failed finding a free block\n"); |
10ba535a JA |
140 | return 1; |
141 | } | |
142 | ||
ec4015da | 143 | static int get_next_rand_offset(struct thread_data *td, struct fio_file *f, |
4ba66134 | 144 | enum fio_ddir ddir, unsigned long long *b) |
ec4015da | 145 | { |
84422acd | 146 | unsigned long long r; |
ec4015da JA |
147 | int loops = 5; |
148 | ||
149 | do { | |
150 | r = os_random_long(&td->random_state); | |
84422acd | 151 | dprint(FD_RANDOM, "off rand %llu\n", r); |
5ec10eaa | 152 | *b = (last_block(td, f, ddir) - 1) |
dc873b6f | 153 | * (r / ((unsigned long long) OS_RAND_MAX + 1.0)); |
2ba1c290 | 154 | |
43c63a78 JA |
155 | /* |
156 | * if we are not maintaining a random map, we are done. | |
157 | */ | |
303032ae | 158 | if (!file_randommap(td, f)) |
43c63a78 JA |
159 | return 0; |
160 | ||
161 | /* | |
84422acd | 162 | * calculate map offset and check if it's free |
43c63a78 | 163 | */ |
aec2de20 | 164 | if (random_map_free(f, *b)) |
43c63a78 JA |
165 | return 0; |
166 | ||
84422acd JA |
167 | dprint(FD_RANDOM, "get_next_rand_offset: offset %llu busy\n", |
168 | *b); | |
43c63a78 | 169 | } while (--loops); |
ec4015da JA |
170 | |
171 | /* | |
43c63a78 JA |
172 | * we get here, if we didn't suceed in looking up a block. generate |
173 | * a random start offset into the filemap, and find the first free | |
174 | * block from there. | |
ec4015da | 175 | */ |
43c63a78 JA |
176 | loops = 10; |
177 | do { | |
dc873b6f JA |
178 | f->last_free_lookup = (f->num_maps - 1) * |
179 | (r / (OS_RAND_MAX + 1.0)); | |
4ba66134 | 180 | if (!get_next_free_block(td, f, ddir, b)) |
43c63a78 | 181 | return 0; |
ec4015da | 182 | |
43c63a78 JA |
183 | r = os_random_long(&td->random_state); |
184 | } while (--loops); | |
185 | ||
186 | /* | |
187 | * that didn't work either, try exhaustive search from the start | |
188 | */ | |
189 | f->last_free_lookup = 0; | |
4ba66134 | 190 | return get_next_free_block(td, f, ddir, b); |
ec4015da JA |
191 | } |
192 | ||
10ba535a JA |
193 | /* |
194 | * For random io, generate a random new block and see if it's used. Repeat | |
195 | * until we find a free one. For sequential io, just return the end of | |
196 | * the last io issued. | |
197 | */ | |
15dc1934 | 198 | static int __get_next_offset(struct thread_data *td, struct io_u *io_u) |
10ba535a | 199 | { |
9bf2061e | 200 | struct fio_file *f = io_u->file; |
ec4015da | 201 | unsigned long long b; |
4ba66134 | 202 | enum fio_ddir ddir = io_u->ddir; |
10ba535a | 203 | |
ec4015da JA |
204 | if (td_random(td) && (td->o.ddir_nr && !--td->ddir_nr)) { |
205 | td->ddir_nr = td->o.ddir_nr; | |
211097b2 | 206 | |
683023e8 JA |
207 | if (get_next_rand_offset(td, f, ddir, &b)) { |
208 | dprint(FD_IO, "%s: getting rand offset failed\n", | |
209 | f->file_name); | |
bca4ed4d | 210 | return 1; |
683023e8 | 211 | } |
43063a1c | 212 | } else { |
0c3d768a | 213 | if (f->last_pos >= f->real_file_size) { |
4ba66134 | 214 | if (!td_random(td) || |
683023e8 JA |
215 | get_next_rand_offset(td, f, ddir, &b)) { |
216 | dprint(FD_IO, "%s: pos %llu > size %llu\n", | |
217 | f->file_name, f->last_pos, | |
218 | f->real_file_size); | |
0c3d768a | 219 | return 1; |
683023e8 | 220 | } |
bcdedd0a | 221 | } else |
4ba66134 | 222 | b = (f->last_pos - f->file_offset) / td->o.min_bs[ddir]; |
43063a1c | 223 | } |
10ba535a | 224 | |
2b7a01d0 | 225 | io_u->offset = b * td->o.ba[ddir]; |
009bd847 JA |
226 | if (io_u->offset >= f->io_size) { |
227 | dprint(FD_IO, "get_next_offset: offset %llu >= io_size %llu\n", | |
228 | io_u->offset, f->io_size); | |
229 | return 1; | |
230 | } | |
231 | ||
232 | io_u->offset += f->file_offset; | |
2ba1c290 JA |
233 | if (io_u->offset >= f->real_file_size) { |
234 | dprint(FD_IO, "get_next_offset: offset %llu >= size %llu\n", | |
235 | io_u->offset, f->real_file_size); | |
10ba535a | 236 | return 1; |
2ba1c290 | 237 | } |
10ba535a JA |
238 | |
239 | return 0; | |
240 | } | |
241 | ||
15dc1934 JA |
242 | static int get_next_offset(struct thread_data *td, struct io_u *io_u) |
243 | { | |
7eb36574 JA |
244 | struct prof_io_ops *ops = &td->prof_io_ops; |
245 | ||
246 | if (ops->fill_io_u_off) | |
247 | return ops->fill_io_u_off(td, io_u); | |
15dc1934 JA |
248 | |
249 | return __get_next_offset(td, io_u); | |
250 | } | |
251 | ||
252 | static unsigned int __get_next_buflen(struct thread_data *td, struct io_u *io_u) | |
10ba535a | 253 | { |
bca4ed4d | 254 | const int ddir = io_u->ddir; |
f3f552b9 | 255 | unsigned int uninitialized_var(buflen); |
f3059de1 | 256 | unsigned int minbs, maxbs; |
10ba535a JA |
257 | long r; |
258 | ||
f3059de1 JA |
259 | minbs = td->o.min_bs[ddir]; |
260 | maxbs = td->o.max_bs[ddir]; | |
261 | ||
262 | if (minbs == maxbs) | |
263 | buflen = minbs; | |
10ba535a JA |
264 | else { |
265 | r = os_random_long(&td->bsrange_state); | |
720e84ad | 266 | if (!td->o.bssplit_nr[ddir]) { |
f3059de1 JA |
267 | buflen = 1 + (unsigned int) ((double) maxbs * |
268 | (r / (OS_RAND_MAX + 1.0))); | |
269 | if (buflen < minbs) | |
270 | buflen = minbs; | |
5ec10eaa | 271 | } else { |
564ca972 JA |
272 | long perc = 0; |
273 | unsigned int i; | |
274 | ||
720e84ad JA |
275 | for (i = 0; i < td->o.bssplit_nr[ddir]; i++) { |
276 | struct bssplit *bsp = &td->o.bssplit[ddir][i]; | |
564ca972 JA |
277 | |
278 | buflen = bsp->bs; | |
279 | perc += bsp->perc; | |
f3059de1 | 280 | if (r <= ((OS_RAND_MAX / 100L) * perc)) |
564ca972 JA |
281 | break; |
282 | } | |
283 | } | |
f3059de1 JA |
284 | if (!td->o.bs_unaligned && is_power_of_2(minbs)) |
285 | buflen = (buflen + minbs - 1) & ~(minbs - 1); | |
10ba535a JA |
286 | } |
287 | ||
4ba66134 JA |
288 | if (io_u->offset + buflen > io_u->file->real_file_size) { |
289 | dprint(FD_IO, "lower buflen %u -> %u (ddir=%d)\n", buflen, | |
f3059de1 JA |
290 | minbs, ddir); |
291 | buflen = minbs; | |
4ba66134 | 292 | } |
6a5e6884 | 293 | |
10ba535a JA |
294 | return buflen; |
295 | } | |
296 | ||
15dc1934 JA |
297 | static unsigned int get_next_buflen(struct thread_data *td, struct io_u *io_u) |
298 | { | |
7eb36574 JA |
299 | struct prof_io_ops *ops = &td->prof_io_ops; |
300 | ||
301 | if (ops->fill_io_u_size) | |
302 | return ops->fill_io_u_size(td, io_u); | |
15dc1934 JA |
303 | |
304 | return __get_next_buflen(td, io_u); | |
305 | } | |
306 | ||
afe24a5a JA |
307 | static void set_rwmix_bytes(struct thread_data *td) |
308 | { | |
afe24a5a JA |
309 | unsigned int diff; |
310 | ||
311 | /* | |
312 | * we do time or byte based switch. this is needed because | |
313 | * buffered writes may issue a lot quicker than they complete, | |
314 | * whereas reads do not. | |
315 | */ | |
e47f799f | 316 | diff = td->o.rwmix[td->rwmix_ddir ^ 1]; |
04c540d9 | 317 | td->rwmix_issues = (td->io_issues[td->rwmix_ddir] * diff) / 100; |
e47f799f JA |
318 | } |
319 | ||
320 | static inline enum fio_ddir get_rand_ddir(struct thread_data *td) | |
321 | { | |
322 | unsigned int v; | |
323 | long r; | |
324 | ||
325 | r = os_random_long(&td->rwmix_state); | |
dc873b6f | 326 | v = 1 + (int) (100.0 * (r / (OS_RAND_MAX + 1.0))); |
04c540d9 | 327 | if (v <= td->o.rwmix[DDIR_READ]) |
e47f799f JA |
328 | return DDIR_READ; |
329 | ||
330 | return DDIR_WRITE; | |
afe24a5a JA |
331 | } |
332 | ||
581e7141 JA |
333 | static enum fio_ddir rate_ddir(struct thread_data *td, enum fio_ddir ddir) |
334 | { | |
335 | enum fio_ddir odir = ddir ^ 1; | |
336 | struct timeval t; | |
337 | long usec; | |
338 | ||
339 | if (td->rate_pending_usleep[ddir] <= 0) | |
340 | return ddir; | |
341 | ||
342 | /* | |
343 | * We have too much pending sleep in this direction. See if we | |
344 | * should switch. | |
345 | */ | |
346 | if (td_rw(td)) { | |
347 | /* | |
348 | * Other direction does not have too much pending, switch | |
349 | */ | |
350 | if (td->rate_pending_usleep[odir] < 100000) | |
351 | return odir; | |
352 | ||
353 | /* | |
354 | * Both directions have pending sleep. Sleep the minimum time | |
355 | * and deduct from both. | |
356 | */ | |
357 | if (td->rate_pending_usleep[ddir] <= | |
358 | td->rate_pending_usleep[odir]) { | |
359 | usec = td->rate_pending_usleep[ddir]; | |
360 | } else { | |
361 | usec = td->rate_pending_usleep[odir]; | |
362 | ddir = odir; | |
363 | } | |
364 | } else | |
365 | usec = td->rate_pending_usleep[ddir]; | |
366 | ||
367 | fio_gettime(&t, NULL); | |
368 | usec_sleep(td, usec); | |
369 | usec = utime_since_now(&t); | |
370 | ||
371 | td->rate_pending_usleep[ddir] -= usec; | |
372 | ||
373 | odir = ddir ^ 1; | |
374 | if (td_rw(td) && __should_check_rate(td, odir)) | |
375 | td->rate_pending_usleep[odir] -= usec; | |
0b9d69ec | 376 | |
581e7141 JA |
377 | return ddir; |
378 | } | |
379 | ||
10ba535a JA |
380 | /* |
381 | * Return the data direction for the next io_u. If the job is a | |
382 | * mixed read/write workload, check the rwmix cycle and switch if | |
383 | * necessary. | |
384 | */ | |
1e97cce9 | 385 | static enum fio_ddir get_rw_ddir(struct thread_data *td) |
10ba535a | 386 | { |
581e7141 JA |
387 | enum fio_ddir ddir; |
388 | ||
5f9099ea JA |
389 | /* |
390 | * see if it's time to fsync | |
391 | */ | |
392 | if (td->o.fsync_blocks && | |
393 | !(td->io_issues[DDIR_WRITE] % td->o.fsync_blocks) && | |
394 | td->io_issues[DDIR_WRITE] && should_fsync(td)) | |
395 | return DDIR_SYNC; | |
396 | ||
397 | /* | |
398 | * see if it's time to fdatasync | |
399 | */ | |
400 | if (td->o.fdatasync_blocks && | |
401 | !(td->io_issues[DDIR_WRITE] % td->o.fdatasync_blocks) && | |
402 | td->io_issues[DDIR_WRITE] && should_fsync(td)) | |
403 | return DDIR_DATASYNC; | |
404 | ||
44f29692 JA |
405 | /* |
406 | * see if it's time to sync_file_range | |
407 | */ | |
408 | if (td->sync_file_range_nr && | |
409 | !(td->io_issues[DDIR_WRITE] % td->sync_file_range_nr) && | |
410 | td->io_issues[DDIR_WRITE] && should_fsync(td)) | |
411 | return DDIR_SYNC_FILE_RANGE; | |
412 | ||
10ba535a | 413 | if (td_rw(td)) { |
10ba535a JA |
414 | /* |
415 | * Check if it's time to seed a new data direction. | |
416 | */ | |
e4928662 | 417 | if (td->io_issues[td->rwmix_ddir] >= td->rwmix_issues) { |
e47f799f JA |
418 | /* |
419 | * Put a top limit on how many bytes we do for | |
420 | * one data direction, to avoid overflowing the | |
421 | * ranges too much | |
422 | */ | |
423 | ddir = get_rand_ddir(td); | |
e47f799f JA |
424 | |
425 | if (ddir != td->rwmix_ddir) | |
426 | set_rwmix_bytes(td); | |
427 | ||
428 | td->rwmix_ddir = ddir; | |
10ba535a | 429 | } |
581e7141 | 430 | ddir = td->rwmix_ddir; |
10ba535a | 431 | } else if (td_read(td)) |
581e7141 | 432 | ddir = DDIR_READ; |
10ba535a | 433 | else |
581e7141 JA |
434 | ddir = DDIR_WRITE; |
435 | ||
436 | td->rwmix_ddir = rate_ddir(td, ddir); | |
437 | return td->rwmix_ddir; | |
10ba535a JA |
438 | } |
439 | ||
e8462bd8 | 440 | void put_file_log(struct thread_data *td, struct fio_file *f) |
60f2c658 JA |
441 | { |
442 | int ret = put_file(td, f); | |
443 | ||
444 | if (ret) | |
445 | td_verror(td, ret, "file close"); | |
446 | } | |
447 | ||
10ba535a JA |
448 | void put_io_u(struct thread_data *td, struct io_u *io_u) |
449 | { | |
e8462bd8 JA |
450 | td_io_u_lock(td); |
451 | ||
0c6e7517 | 452 | io_u->flags |= IO_U_F_FREE; |
e8462bd8 | 453 | io_u->flags &= ~IO_U_F_FREE_DEF; |
0c6e7517 | 454 | |
60f2c658 JA |
455 | if (io_u->file) |
456 | put_file_log(td, io_u->file); | |
2dbdab7e | 457 | |
10ba535a | 458 | io_u->file = NULL; |
0c41214f RR |
459 | if (io_u->flags & IO_U_F_IN_CUR_DEPTH) |
460 | td->cur_depth--; | |
e8462bd8 | 461 | flist_del_init(&io_u->list); |
01743ee1 | 462 | flist_add(&io_u->list, &td->io_u_freelist); |
e8462bd8 JA |
463 | td_io_u_unlock(td); |
464 | td_io_u_free_notify(td); | |
10ba535a JA |
465 | } |
466 | ||
f2bba182 RR |
467 | void clear_io_u(struct thread_data *td, struct io_u *io_u) |
468 | { | |
469 | io_u->flags &= ~IO_U_F_FLIGHT; | |
470 | put_io_u(td, io_u); | |
471 | } | |
472 | ||
755200a3 JA |
473 | void requeue_io_u(struct thread_data *td, struct io_u **io_u) |
474 | { | |
475 | struct io_u *__io_u = *io_u; | |
476 | ||
465221b0 JA |
477 | dprint(FD_IO, "requeue %p\n", __io_u); |
478 | ||
e8462bd8 JA |
479 | td_io_u_lock(td); |
480 | ||
4d2e0f49 | 481 | __io_u->flags |= IO_U_F_FREE; |
5f9099ea | 482 | if ((__io_u->flags & IO_U_F_FLIGHT) && !ddir_sync(__io_u->ddir)) |
e4f54adb | 483 | td->io_issues[__io_u->ddir]--; |
5ec10eaa | 484 | |
4d2e0f49 | 485 | __io_u->flags &= ~IO_U_F_FLIGHT; |
0c41214f RR |
486 | if (__io_u->flags & IO_U_F_IN_CUR_DEPTH) |
487 | td->cur_depth--; | |
01743ee1 JA |
488 | flist_del(&__io_u->list); |
489 | flist_add_tail(&__io_u->list, &td->io_u_requeues); | |
e8462bd8 | 490 | td_io_u_unlock(td); |
755200a3 JA |
491 | *io_u = NULL; |
492 | } | |
493 | ||
9bf2061e | 494 | static int fill_io_u(struct thread_data *td, struct io_u *io_u) |
10ba535a | 495 | { |
b4c5e1ac JA |
496 | if (td->io_ops->flags & FIO_NOIO) |
497 | goto out; | |
498 | ||
5f9099ea JA |
499 | io_u->ddir = get_rw_ddir(td); |
500 | ||
87dc1ab1 | 501 | /* |
5f9099ea | 502 | * fsync() or fdatasync(), we are done |
87dc1ab1 | 503 | */ |
5f9099ea | 504 | if (ddir_sync(io_u->ddir)) |
c38e9468 | 505 | goto out; |
a00735e6 | 506 | |
48f5abd3 JA |
507 | /* |
508 | * See if it's time to switch to a new zone | |
509 | */ | |
510 | if (td->zone_bytes >= td->o.zone_size) { | |
511 | td->zone_bytes = 0; | |
512 | io_u->file->last_pos += td->o.zone_skip; | |
513 | td->io_skip_bytes += td->o.zone_skip; | |
514 | } | |
515 | ||
10ba535a | 516 | /* |
c685b5b2 JA |
517 | * No log, let the seq/rand engine retrieve the next buflen and |
518 | * position. | |
10ba535a | 519 | */ |
2ba1c290 JA |
520 | if (get_next_offset(td, io_u)) { |
521 | dprint(FD_IO, "io_u %p, failed getting offset\n", io_u); | |
bca4ed4d | 522 | return 1; |
2ba1c290 | 523 | } |
10ba535a | 524 | |
9bf2061e | 525 | io_u->buflen = get_next_buflen(td, io_u); |
2ba1c290 JA |
526 | if (!io_u->buflen) { |
527 | dprint(FD_IO, "io_u %p, failed getting buflen\n", io_u); | |
bca4ed4d | 528 | return 1; |
2ba1c290 | 529 | } |
bca4ed4d | 530 | |
2ba1c290 JA |
531 | if (io_u->offset + io_u->buflen > io_u->file->real_file_size) { |
532 | dprint(FD_IO, "io_u %p, offset too large\n", io_u); | |
4ba66134 JA |
533 | dprint(FD_IO, " off=%llu/%lu > %llu\n", io_u->offset, |
534 | io_u->buflen, io_u->file->real_file_size); | |
6a5e6884 | 535 | return 1; |
2ba1c290 | 536 | } |
6a5e6884 | 537 | |
bca4ed4d JA |
538 | /* |
539 | * mark entry before potentially trimming io_u | |
540 | */ | |
303032ae | 541 | if (td_random(td) && file_randommap(td, io_u->file)) |
9bf2061e | 542 | mark_random_map(td, io_u); |
bca4ed4d JA |
543 | |
544 | /* | |
545 | * If using a write iolog, store this entry. | |
546 | */ | |
c38e9468 | 547 | out: |
2ba1c290 | 548 | dprint_io_u(io_u, "fill_io_u"); |
d9d91e39 | 549 | td->zone_bytes += io_u->buflen; |
f29b25a3 | 550 | log_io_u(td, io_u); |
bca4ed4d | 551 | return 0; |
10ba535a JA |
552 | } |
553 | ||
838bc709 JA |
554 | static void __io_u_mark_map(unsigned int *map, unsigned int nr) |
555 | { | |
556 | int index = 0; | |
557 | ||
558 | switch (nr) { | |
559 | default: | |
560 | index = 6; | |
561 | break; | |
562 | case 33 ... 64: | |
563 | index = 5; | |
564 | break; | |
565 | case 17 ... 32: | |
566 | index = 4; | |
567 | break; | |
568 | case 9 ... 16: | |
569 | index = 3; | |
570 | break; | |
571 | case 5 ... 8: | |
572 | index = 2; | |
573 | break; | |
574 | case 1 ... 4: | |
575 | index = 1; | |
576 | case 0: | |
577 | break; | |
578 | } | |
579 | ||
580 | map[index]++; | |
581 | } | |
582 | ||
583 | void io_u_mark_submit(struct thread_data *td, unsigned int nr) | |
584 | { | |
585 | __io_u_mark_map(td->ts.io_u_submit, nr); | |
586 | td->ts.total_submit++; | |
587 | } | |
588 | ||
589 | void io_u_mark_complete(struct thread_data *td, unsigned int nr) | |
590 | { | |
591 | __io_u_mark_map(td->ts.io_u_complete, nr); | |
592 | td->ts.total_complete++; | |
593 | } | |
594 | ||
d8005759 | 595 | void io_u_mark_depth(struct thread_data *td, unsigned int nr) |
71619dc2 JA |
596 | { |
597 | int index = 0; | |
598 | ||
599 | switch (td->cur_depth) { | |
600 | default: | |
a783e61a JA |
601 | index = 6; |
602 | break; | |
71619dc2 | 603 | case 32 ... 63: |
a783e61a JA |
604 | index = 5; |
605 | break; | |
71619dc2 | 606 | case 16 ... 31: |
a783e61a JA |
607 | index = 4; |
608 | break; | |
71619dc2 | 609 | case 8 ... 15: |
a783e61a JA |
610 | index = 3; |
611 | break; | |
71619dc2 | 612 | case 4 ... 7: |
a783e61a JA |
613 | index = 2; |
614 | break; | |
71619dc2 | 615 | case 2 ... 3: |
a783e61a | 616 | index = 1; |
71619dc2 JA |
617 | case 1: |
618 | break; | |
619 | } | |
620 | ||
3bec7ae0 | 621 | td->ts.io_u_map[index] += nr; |
71619dc2 JA |
622 | } |
623 | ||
04a0feae JA |
624 | static void io_u_mark_lat_usec(struct thread_data *td, unsigned long usec) |
625 | { | |
626 | int index = 0; | |
627 | ||
628 | assert(usec < 1000); | |
629 | ||
630 | switch (usec) { | |
631 | case 750 ... 999: | |
632 | index = 9; | |
633 | break; | |
634 | case 500 ... 749: | |
635 | index = 8; | |
636 | break; | |
637 | case 250 ... 499: | |
638 | index = 7; | |
639 | break; | |
640 | case 100 ... 249: | |
641 | index = 6; | |
642 | break; | |
643 | case 50 ... 99: | |
644 | index = 5; | |
645 | break; | |
646 | case 20 ... 49: | |
647 | index = 4; | |
648 | break; | |
649 | case 10 ... 19: | |
650 | index = 3; | |
651 | break; | |
652 | case 4 ... 9: | |
653 | index = 2; | |
654 | break; | |
655 | case 2 ... 3: | |
656 | index = 1; | |
657 | case 0 ... 1: | |
658 | break; | |
659 | } | |
660 | ||
661 | assert(index < FIO_IO_U_LAT_U_NR); | |
662 | td->ts.io_u_lat_u[index]++; | |
663 | } | |
664 | ||
665 | static void io_u_mark_lat_msec(struct thread_data *td, unsigned long msec) | |
ec118304 JA |
666 | { |
667 | int index = 0; | |
668 | ||
669 | switch (msec) { | |
670 | default: | |
04a0feae JA |
671 | index = 11; |
672 | break; | |
8abdce66 | 673 | case 1000 ... 1999: |
04a0feae JA |
674 | index = 10; |
675 | break; | |
8abdce66 | 676 | case 750 ... 999: |
04a0feae JA |
677 | index = 9; |
678 | break; | |
8abdce66 | 679 | case 500 ... 749: |
04a0feae JA |
680 | index = 8; |
681 | break; | |
8abdce66 | 682 | case 250 ... 499: |
04a0feae JA |
683 | index = 7; |
684 | break; | |
8abdce66 | 685 | case 100 ... 249: |
04a0feae JA |
686 | index = 6; |
687 | break; | |
8abdce66 | 688 | case 50 ... 99: |
04a0feae JA |
689 | index = 5; |
690 | break; | |
8abdce66 | 691 | case 20 ... 49: |
04a0feae JA |
692 | index = 4; |
693 | break; | |
8abdce66 | 694 | case 10 ... 19: |
04a0feae JA |
695 | index = 3; |
696 | break; | |
8abdce66 | 697 | case 4 ... 9: |
04a0feae JA |
698 | index = 2; |
699 | break; | |
ec118304 | 700 | case 2 ... 3: |
04a0feae | 701 | index = 1; |
ec118304 JA |
702 | case 0 ... 1: |
703 | break; | |
704 | } | |
705 | ||
04a0feae JA |
706 | assert(index < FIO_IO_U_LAT_M_NR); |
707 | td->ts.io_u_lat_m[index]++; | |
708 | } | |
709 | ||
710 | static void io_u_mark_latency(struct thread_data *td, unsigned long usec) | |
711 | { | |
712 | if (usec < 1000) | |
713 | io_u_mark_lat_usec(td, usec); | |
714 | else | |
715 | io_u_mark_lat_msec(td, usec / 1000); | |
ec118304 JA |
716 | } |
717 | ||
0aabe160 JA |
718 | /* |
719 | * Get next file to service by choosing one at random | |
720 | */ | |
2cc52930 JA |
721 | static struct fio_file *get_next_file_rand(struct thread_data *td, |
722 | enum fio_file_flags goodf, | |
d6aed795 | 723 | enum fio_file_flags badf) |
0aabe160 | 724 | { |
0aabe160 | 725 | struct fio_file *f; |
1c178180 | 726 | int fno; |
0aabe160 JA |
727 | |
728 | do { | |
7c83c089 | 729 | long r = os_random_long(&td->next_file_state); |
87b10676 | 730 | int opened = 0; |
7c83c089 | 731 | |
5ec10eaa | 732 | fno = (unsigned int) ((double) td->o.nr_files |
dc873b6f | 733 | * (r / (OS_RAND_MAX + 1.0))); |
126d65c6 | 734 | f = td->files[fno]; |
d6aed795 | 735 | if (fio_file_done(f)) |
059e63c0 | 736 | continue; |
1c178180 | 737 | |
d6aed795 | 738 | if (!fio_file_open(f)) { |
87b10676 JA |
739 | int err; |
740 | ||
741 | err = td_io_open_file(td, f); | |
742 | if (err) | |
743 | continue; | |
744 | opened = 1; | |
745 | } | |
746 | ||
2ba1c290 JA |
747 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) { |
748 | dprint(FD_FILE, "get_next_file_rand: %p\n", f); | |
0aabe160 | 749 | return f; |
2ba1c290 | 750 | } |
87b10676 JA |
751 | if (opened) |
752 | td_io_close_file(td, f); | |
0aabe160 JA |
753 | } while (1); |
754 | } | |
755 | ||
756 | /* | |
757 | * Get next file to service by doing round robin between all available ones | |
758 | */ | |
1c178180 JA |
759 | static struct fio_file *get_next_file_rr(struct thread_data *td, int goodf, |
760 | int badf) | |
3d7c391d JA |
761 | { |
762 | unsigned int old_next_file = td->next_file; | |
763 | struct fio_file *f; | |
764 | ||
765 | do { | |
87b10676 JA |
766 | int opened = 0; |
767 | ||
126d65c6 | 768 | f = td->files[td->next_file]; |
3d7c391d JA |
769 | |
770 | td->next_file++; | |
2dc1bbeb | 771 | if (td->next_file >= td->o.nr_files) |
3d7c391d JA |
772 | td->next_file = 0; |
773 | ||
87b10676 | 774 | dprint(FD_FILE, "trying file %s %x\n", f->file_name, f->flags); |
d6aed795 | 775 | if (fio_file_done(f)) { |
d5ed68ea | 776 | f = NULL; |
059e63c0 | 777 | continue; |
d5ed68ea | 778 | } |
059e63c0 | 779 | |
d6aed795 | 780 | if (!fio_file_open(f)) { |
87b10676 JA |
781 | int err; |
782 | ||
783 | err = td_io_open_file(td, f); | |
b5696bfc JA |
784 | if (err) { |
785 | dprint(FD_FILE, "error %d on open of %s\n", | |
786 | err, f->file_name); | |
87c27b45 | 787 | f = NULL; |
87b10676 | 788 | continue; |
b5696bfc | 789 | } |
87b10676 JA |
790 | opened = 1; |
791 | } | |
792 | ||
0b9d69ec JA |
793 | dprint(FD_FILE, "goodf=%x, badf=%x, ff=%x\n", goodf, badf, |
794 | f->flags); | |
1c178180 | 795 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) |
3d7c391d JA |
796 | break; |
797 | ||
87b10676 JA |
798 | if (opened) |
799 | td_io_close_file(td, f); | |
800 | ||
3d7c391d JA |
801 | f = NULL; |
802 | } while (td->next_file != old_next_file); | |
803 | ||
2ba1c290 | 804 | dprint(FD_FILE, "get_next_file_rr: %p\n", f); |
3d7c391d JA |
805 | return f; |
806 | } | |
807 | ||
7eb36574 | 808 | static struct fio_file *__get_next_file(struct thread_data *td) |
bdb4e2e9 | 809 | { |
1907dbc6 JA |
810 | struct fio_file *f; |
811 | ||
2dc1bbeb | 812 | assert(td->o.nr_files <= td->files_index); |
1c178180 | 813 | |
b5696bfc | 814 | if (td->nr_done_files >= td->o.nr_files) { |
5ec10eaa JA |
815 | dprint(FD_FILE, "get_next_file: nr_open=%d, nr_done=%d," |
816 | " nr_files=%d\n", td->nr_open_files, | |
817 | td->nr_done_files, | |
818 | td->o.nr_files); | |
bdb4e2e9 | 819 | return NULL; |
2ba1c290 | 820 | } |
bdb4e2e9 | 821 | |
1907dbc6 | 822 | f = td->file_service_file; |
d6aed795 | 823 | if (f && fio_file_open(f) && !fio_file_closing(f)) { |
a086c257 JA |
824 | if (td->o.file_service_type == FIO_FSERVICE_SEQ) |
825 | goto out; | |
826 | if (td->file_service_left--) | |
827 | goto out; | |
828 | } | |
1907dbc6 | 829 | |
a086c257 JA |
830 | if (td->o.file_service_type == FIO_FSERVICE_RR || |
831 | td->o.file_service_type == FIO_FSERVICE_SEQ) | |
d6aed795 | 832 | f = get_next_file_rr(td, FIO_FILE_open, FIO_FILE_closing); |
bdb4e2e9 | 833 | else |
d6aed795 | 834 | f = get_next_file_rand(td, FIO_FILE_open, FIO_FILE_closing); |
1907dbc6 JA |
835 | |
836 | td->file_service_file = f; | |
837 | td->file_service_left = td->file_service_nr - 1; | |
2ba1c290 | 838 | out: |
683023e8 | 839 | dprint(FD_FILE, "get_next_file: %p [%s]\n", f, f->file_name); |
1907dbc6 | 840 | return f; |
bdb4e2e9 JA |
841 | } |
842 | ||
7eb36574 JA |
843 | static struct fio_file *get_next_file(struct thread_data *td) |
844 | { | |
845 | struct prof_io_ops *ops = &td->prof_io_ops; | |
846 | ||
847 | if (ops->get_next_file) | |
848 | return ops->get_next_file(td); | |
849 | ||
850 | return __get_next_file(td); | |
851 | } | |
852 | ||
429f6675 JA |
853 | static int set_io_u_file(struct thread_data *td, struct io_u *io_u) |
854 | { | |
855 | struct fio_file *f; | |
856 | ||
857 | do { | |
858 | f = get_next_file(td); | |
859 | if (!f) | |
860 | return 1; | |
861 | ||
429f6675 JA |
862 | io_u->file = f; |
863 | get_file(f); | |
864 | ||
865 | if (!fill_io_u(td, io_u)) | |
866 | break; | |
867 | ||
b5696bfc | 868 | put_file_log(td, f); |
429f6675 | 869 | td_io_close_file(td, f); |
b5696bfc | 870 | io_u->file = NULL; |
d6aed795 | 871 | fio_file_set_done(f); |
429f6675 | 872 | td->nr_done_files++; |
0b9d69ec JA |
873 | dprint(FD_FILE, "%s: is done (%d of %d)\n", f->file_name, |
874 | td->nr_done_files, td->o.nr_files); | |
429f6675 JA |
875 | } while (1); |
876 | ||
877 | return 0; | |
878 | } | |
879 | ||
880 | ||
10ba535a JA |
881 | struct io_u *__get_io_u(struct thread_data *td) |
882 | { | |
883 | struct io_u *io_u = NULL; | |
884 | ||
e8462bd8 JA |
885 | td_io_u_lock(td); |
886 | ||
887 | again: | |
01743ee1 JA |
888 | if (!flist_empty(&td->io_u_requeues)) |
889 | io_u = flist_entry(td->io_u_requeues.next, struct io_u, list); | |
755200a3 | 890 | else if (!queue_full(td)) { |
01743ee1 | 891 | io_u = flist_entry(td->io_u_freelist.next, struct io_u, list); |
10ba535a | 892 | |
6040dabc | 893 | io_u->buflen = 0; |
10ba535a | 894 | io_u->resid = 0; |
755200a3 | 895 | io_u->file = NULL; |
d7762cf8 | 896 | io_u->end_io = NULL; |
755200a3 JA |
897 | } |
898 | ||
899 | if (io_u) { | |
0c6e7517 | 900 | assert(io_u->flags & IO_U_F_FREE); |
2ecc1b57 | 901 | io_u->flags &= ~(IO_U_F_FREE | IO_U_F_FREE_DEF); |
0c6e7517 | 902 | |
755200a3 | 903 | io_u->error = 0; |
01743ee1 JA |
904 | flist_del(&io_u->list); |
905 | flist_add(&io_u->list, &td->io_u_busylist); | |
10ba535a | 906 | td->cur_depth++; |
0c41214f | 907 | io_u->flags |= IO_U_F_IN_CUR_DEPTH; |
1dec3e07 JA |
908 | } else if (td->o.verify_async) { |
909 | /* | |
910 | * We ran out, wait for async verify threads to finish and | |
911 | * return one | |
912 | */ | |
913 | pthread_cond_wait(&td->free_cond, &td->io_u_lock); | |
914 | goto again; | |
10ba535a JA |
915 | } |
916 | ||
e8462bd8 | 917 | td_io_u_unlock(td); |
10ba535a JA |
918 | return io_u; |
919 | } | |
920 | ||
921 | /* | |
922 | * Return an io_u to be processed. Gets a buflen and offset, sets direction, | |
923 | * etc. The returned io_u is fully ready to be prepped and submitted. | |
924 | */ | |
3d7c391d | 925 | struct io_u *get_io_u(struct thread_data *td) |
10ba535a | 926 | { |
3d7c391d | 927 | struct fio_file *f; |
10ba535a JA |
928 | struct io_u *io_u; |
929 | ||
930 | io_u = __get_io_u(td); | |
2ba1c290 JA |
931 | if (!io_u) { |
932 | dprint(FD_IO, "__get_io_u failed\n"); | |
10ba535a | 933 | return NULL; |
2ba1c290 | 934 | } |
10ba535a | 935 | |
9e144189 JA |
936 | if (td->o.verify_backlog && td->io_hist_len) { |
937 | int get_verify = 0; | |
938 | ||
939 | if (td->verify_batch) { | |
940 | td->verify_batch--; | |
941 | get_verify = 1; | |
942 | } else if (!(td->io_hist_len % td->o.verify_backlog) && | |
943 | td->last_ddir != DDIR_READ) { | |
944 | td->verify_batch = td->o.verify_batch; | |
f8a75c99 JA |
945 | if (!td->verify_batch) |
946 | td->verify_batch = td->o.verify_backlog; | |
9e144189 JA |
947 | get_verify = 1; |
948 | } | |
949 | ||
950 | if (get_verify && !get_next_verify(td, io_u)) | |
951 | goto out; | |
952 | } | |
953 | ||
755200a3 JA |
954 | /* |
955 | * from a requeue, io_u already setup | |
956 | */ | |
957 | if (io_u->file) | |
77f392bf | 958 | goto out; |
755200a3 | 959 | |
429f6675 JA |
960 | /* |
961 | * If using an iolog, grab next piece if any available. | |
962 | */ | |
963 | if (td->o.read_iolog_file) { | |
964 | if (read_iolog_get(td, io_u)) | |
965 | goto err_put; | |
2ba1c290 JA |
966 | } else if (set_io_u_file(td, io_u)) { |
967 | dprint(FD_IO, "io_u %p, setting file failed\n", io_u); | |
429f6675 | 968 | goto err_put; |
2ba1c290 | 969 | } |
5ec10eaa | 970 | |
429f6675 | 971 | f = io_u->file; |
d6aed795 | 972 | assert(fio_file_open(f)); |
97af62ce | 973 | |
5f9099ea | 974 | if (!ddir_sync(io_u->ddir)) { |
d0656a93 | 975 | if (!io_u->buflen && !(td->io_ops->flags & FIO_NOIO)) { |
2ba1c290 | 976 | dprint(FD_IO, "get_io_u: zero buflen on %p\n", io_u); |
429f6675 | 977 | goto err_put; |
2ba1c290 | 978 | } |
10ba535a | 979 | |
36167d82 | 980 | f->last_pos = io_u->offset + io_u->buflen; |
10ba535a | 981 | |
c311cd2a | 982 | if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_WRITE) |
87dc1ab1 | 983 | populate_verify_io_u(td, io_u); |
e4dad9c6 JA |
984 | else if (td->o.refill_buffers && io_u->ddir == DDIR_WRITE) |
985 | io_u_fill_buffer(td, io_u, io_u->xfer_buflen); | |
87dc1ab1 | 986 | } |
10ba535a | 987 | |
165faf16 JA |
988 | /* |
989 | * Set io data pointers. | |
990 | */ | |
cec6b55d JA |
991 | io_u->xfer_buf = io_u->buf; |
992 | io_u->xfer_buflen = io_u->buflen; | |
5973cafb | 993 | |
6ac7a331 | 994 | out: |
429f6675 | 995 | if (!td_io_prep(td, io_u)) { |
993bf48b JA |
996 | if (!td->o.disable_slat) |
997 | fio_gettime(&io_u->start_time, NULL); | |
429f6675 | 998 | return io_u; |
36167d82 | 999 | } |
429f6675 | 1000 | err_put: |
2ba1c290 | 1001 | dprint(FD_IO, "get_io_u failed\n"); |
429f6675 JA |
1002 | put_io_u(td, io_u); |
1003 | return NULL; | |
10ba535a JA |
1004 | } |
1005 | ||
5451792e JA |
1006 | void io_u_log_error(struct thread_data *td, struct io_u *io_u) |
1007 | { | |
1008 | const char *msg[] = { "read", "write", "sync" }; | |
1009 | ||
1010 | log_err("fio: io_u error"); | |
1011 | ||
1012 | if (io_u->file) | |
1013 | log_err(" on file %s", io_u->file->file_name); | |
1014 | ||
1015 | log_err(": %s\n", strerror(io_u->error)); | |
1016 | ||
5ec10eaa JA |
1017 | log_err(" %s offset=%llu, buflen=%lu\n", msg[io_u->ddir], |
1018 | io_u->offset, io_u->xfer_buflen); | |
5451792e JA |
1019 | |
1020 | if (!td->error) | |
1021 | td_verror(td, io_u->error, "io_u error"); | |
1022 | } | |
1023 | ||
97601024 JA |
1024 | static void io_completed(struct thread_data *td, struct io_u *io_u, |
1025 | struct io_completion_data *icd) | |
10ba535a | 1026 | { |
dbad30b0 JA |
1027 | /* |
1028 | * Older gcc's are too dumb to realize that usec is always used | |
1029 | * initialized, silence that warning. | |
1030 | */ | |
1031 | unsigned long uninitialized_var(usec); | |
44f29692 | 1032 | struct fio_file *f; |
10ba535a | 1033 | |
2ba1c290 JA |
1034 | dprint_io_u(io_u, "io complete"); |
1035 | ||
2ecc1b57 | 1036 | td_io_u_lock(td); |
0c6e7517 JA |
1037 | assert(io_u->flags & IO_U_F_FLIGHT); |
1038 | io_u->flags &= ~IO_U_F_FLIGHT; | |
2ecc1b57 | 1039 | td_io_u_unlock(td); |
0c6e7517 | 1040 | |
5f9099ea | 1041 | if (ddir_sync(io_u->ddir)) { |
87dc1ab1 | 1042 | td->last_was_sync = 1; |
44f29692 JA |
1043 | f = io_u->file; |
1044 | if (f) { | |
1045 | f->first_write = -1ULL; | |
1046 | f->last_write = -1ULL; | |
1047 | } | |
87dc1ab1 JA |
1048 | return; |
1049 | } | |
1050 | ||
1051 | td->last_was_sync = 0; | |
9e144189 | 1052 | td->last_ddir = io_u->ddir; |
87dc1ab1 | 1053 | |
10ba535a JA |
1054 | if (!io_u->error) { |
1055 | unsigned int bytes = io_u->buflen - io_u->resid; | |
1e97cce9 | 1056 | const enum fio_ddir idx = io_u->ddir; |
ba3e4e0c | 1057 | const enum fio_ddir odx = io_u->ddir ^ 1; |
b29ee5b3 | 1058 | int ret; |
10ba535a | 1059 | |
b29ee5b3 JA |
1060 | td->io_blocks[idx]++; |
1061 | td->io_bytes[idx] += bytes; | |
1062 | td->this_io_bytes[idx] += bytes; | |
10ba535a | 1063 | |
44f29692 JA |
1064 | if (idx == DDIR_WRITE) { |
1065 | f = io_u->file; | |
1066 | if (f) { | |
1067 | if (f->first_write == -1ULL || | |
1068 | io_u->offset < f->first_write) | |
1069 | f->first_write = io_u->offset; | |
1070 | if (f->last_write == -1ULL || | |
1071 | ((io_u->offset + bytes) > f->last_write)) | |
1072 | f->last_write = io_u->offset + bytes; | |
1073 | } | |
1074 | } | |
1075 | ||
b29ee5b3 | 1076 | if (ramp_time_over(td)) { |
40e1a6f0 | 1077 | unsigned long uninitialized_var(lusec); |
40e1a6f0 JA |
1078 | |
1079 | if (!td->o.disable_clat || !td->o.disable_bw) | |
1080 | lusec = utime_since(&io_u->issue_time, | |
1081 | &icd->time); | |
721938ae | 1082 | |
9520ebb9 | 1083 | if (!td->o.disable_clat) { |
01a1caa5 | 1084 | add_clat_sample(td, idx, lusec, bytes); |
40e1a6f0 | 1085 | io_u_mark_latency(td, lusec); |
9520ebb9 JA |
1086 | } |
1087 | if (!td->o.disable_bw) | |
306ddc97 | 1088 | add_bw_sample(td, idx, bytes, &icd->time); |
b23b6a2f | 1089 | if (__should_check_rate(td, idx)) { |
ba3e4e0c RR |
1090 | td->rate_pending_usleep[idx] = |
1091 | ((td->this_io_bytes[idx] * | |
1092 | td->rate_nsec_cycle[idx]) / 1000 - | |
1093 | utime_since_now(&td->start)); | |
b23b6a2f | 1094 | } |
581e7141 | 1095 | if (__should_check_rate(td, idx ^ 1)) |
ba3e4e0c RR |
1096 | td->rate_pending_usleep[odx] = |
1097 | ((td->this_io_bytes[odx] * | |
1098 | td->rate_nsec_cycle[odx]) / 1000 - | |
1099 | utime_since_now(&td->start)); | |
721938ae | 1100 | } |
10ba535a | 1101 | |
660a1cb5 | 1102 | if (td_write(td) && idx == DDIR_WRITE && |
8670579e | 1103 | td->o.do_verify && |
41128405 | 1104 | td->o.verify != VERIFY_NONE) |
10ba535a JA |
1105 | log_io_piece(td, io_u); |
1106 | ||
b29ee5b3 | 1107 | icd->bytes_done[idx] += bytes; |
3af6ef39 | 1108 | |
d7762cf8 | 1109 | if (io_u->end_io) { |
36690c9b | 1110 | ret = io_u->end_io(td, io_u); |
3af6ef39 JA |
1111 | if (ret && !icd->error) |
1112 | icd->error = ret; | |
1113 | } | |
5451792e | 1114 | } else { |
10ba535a | 1115 | icd->error = io_u->error; |
5451792e JA |
1116 | io_u_log_error(td, io_u); |
1117 | } | |
f2bba182 RR |
1118 | if (td->o.continue_on_error && icd->error && |
1119 | td_non_fatal_error(icd->error)) { | |
1120 | /* | |
1121 | * If there is a non_fatal error, then add to the error count | |
1122 | * and clear all the errors. | |
1123 | */ | |
1124 | update_error_count(td, icd->error); | |
1125 | td_clear_error(td); | |
1126 | icd->error = 0; | |
1127 | io_u->error = 0; | |
1128 | } | |
10ba535a JA |
1129 | } |
1130 | ||
9520ebb9 JA |
1131 | static void init_icd(struct thread_data *td, struct io_completion_data *icd, |
1132 | int nr) | |
10ba535a | 1133 | { |
9520ebb9 JA |
1134 | if (!td->o.disable_clat || !td->o.disable_bw) |
1135 | fio_gettime(&icd->time, NULL); | |
02bcaa8c | 1136 | |
3af6ef39 JA |
1137 | icd->nr = nr; |
1138 | ||
10ba535a JA |
1139 | icd->error = 0; |
1140 | icd->bytes_done[0] = icd->bytes_done[1] = 0; | |
36167d82 JA |
1141 | } |
1142 | ||
97601024 JA |
1143 | static void ios_completed(struct thread_data *td, |
1144 | struct io_completion_data *icd) | |
36167d82 JA |
1145 | { |
1146 | struct io_u *io_u; | |
1147 | int i; | |
1148 | ||
10ba535a JA |
1149 | for (i = 0; i < icd->nr; i++) { |
1150 | io_u = td->io_ops->event(td, i); | |
1151 | ||
1152 | io_completed(td, io_u, icd); | |
e8462bd8 JA |
1153 | |
1154 | if (!(io_u->flags & IO_U_F_FREE_DEF)) | |
1155 | put_io_u(td, io_u); | |
10ba535a JA |
1156 | } |
1157 | } | |
97601024 | 1158 | |
e7e6cfb4 JA |
1159 | /* |
1160 | * Complete a single io_u for the sync engines. | |
1161 | */ | |
581e7141 JA |
1162 | int io_u_sync_complete(struct thread_data *td, struct io_u *io_u, |
1163 | unsigned long *bytes) | |
97601024 JA |
1164 | { |
1165 | struct io_completion_data icd; | |
1166 | ||
9520ebb9 | 1167 | init_icd(td, &icd, 1); |
97601024 | 1168 | io_completed(td, io_u, &icd); |
e8462bd8 JA |
1169 | |
1170 | if (!(io_u->flags & IO_U_F_FREE_DEF)) | |
1171 | put_io_u(td, io_u); | |
97601024 | 1172 | |
581e7141 JA |
1173 | if (icd.error) { |
1174 | td_verror(td, icd.error, "io_u_sync_complete"); | |
1175 | return -1; | |
1176 | } | |
97601024 | 1177 | |
581e7141 JA |
1178 | if (bytes) { |
1179 | bytes[0] += icd.bytes_done[0]; | |
1180 | bytes[1] += icd.bytes_done[1]; | |
1181 | } | |
1182 | ||
1183 | return 0; | |
97601024 JA |
1184 | } |
1185 | ||
e7e6cfb4 JA |
1186 | /* |
1187 | * Called to complete min_events number of io for the async engines. | |
1188 | */ | |
581e7141 JA |
1189 | int io_u_queued_complete(struct thread_data *td, int min_evts, |
1190 | unsigned long *bytes) | |
97601024 | 1191 | { |
97601024 | 1192 | struct io_completion_data icd; |
00de55ef | 1193 | struct timespec *tvp = NULL; |
97601024 | 1194 | int ret; |
4d06a338 | 1195 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0, }; |
97601024 | 1196 | |
4950421a | 1197 | dprint(FD_IO, "io_u_queued_completed: min=%d\n", min_evts); |
b271fe62 | 1198 | |
4950421a | 1199 | if (!min_evts) |
00de55ef | 1200 | tvp = &ts; |
97601024 | 1201 | |
4950421a | 1202 | ret = td_io_getevents(td, min_evts, td->o.iodepth_batch_complete, tvp); |
97601024 | 1203 | if (ret < 0) { |
e1161c32 | 1204 | td_verror(td, -ret, "td_io_getevents"); |
97601024 JA |
1205 | return ret; |
1206 | } else if (!ret) | |
1207 | return ret; | |
1208 | ||
9520ebb9 | 1209 | init_icd(td, &icd, ret); |
97601024 | 1210 | ios_completed(td, &icd); |
581e7141 JA |
1211 | if (icd.error) { |
1212 | td_verror(td, icd.error, "io_u_queued_complete"); | |
1213 | return -1; | |
1214 | } | |
97601024 | 1215 | |
581e7141 JA |
1216 | if (bytes) { |
1217 | bytes[0] += icd.bytes_done[0]; | |
1218 | bytes[1] += icd.bytes_done[1]; | |
1219 | } | |
1220 | ||
1221 | return 0; | |
97601024 | 1222 | } |
7e77dd02 JA |
1223 | |
1224 | /* | |
1225 | * Call when io_u is really queued, to update the submission latency. | |
1226 | */ | |
1227 | void io_u_queued(struct thread_data *td, struct io_u *io_u) | |
1228 | { | |
9520ebb9 JA |
1229 | if (!td->o.disable_slat) { |
1230 | unsigned long slat_time; | |
7e77dd02 | 1231 | |
9520ebb9 | 1232 | slat_time = utime_since(&io_u->start_time, &io_u->issue_time); |
29a90ddb | 1233 | add_slat_sample(td, io_u->ddir, slat_time, io_u->xfer_buflen); |
9520ebb9 | 1234 | } |
7e77dd02 | 1235 | } |
433afcb4 | 1236 | |
5973cafb JA |
1237 | /* |
1238 | * "randomly" fill the buffer contents | |
1239 | */ | |
1240 | void io_u_fill_buffer(struct thread_data *td, struct io_u *io_u, | |
1241 | unsigned int max_bs) | |
1242 | { | |
1243 | long *ptr = io_u->buf; | |
1244 | ||
1245 | if (!td->o.zero_buffers) { | |
d77431b5 | 1246 | unsigned long r = __rand(&__fio_rand_state); |
27ea68a5 | 1247 | |
d77431b5 JA |
1248 | if (sizeof(int) != sizeof(*ptr)) |
1249 | r *= (unsigned long) __rand(&__fio_rand_state); | |
27ea68a5 | 1250 | |
d77431b5 | 1251 | while ((void *) ptr - io_u->buf < max_bs) { |
27ea68a5 | 1252 | *ptr = r; |
5973cafb | 1253 | ptr++; |
d77431b5 JA |
1254 | r *= GOLDEN_RATIO_PRIME; |
1255 | r >>= 3; | |
5973cafb JA |
1256 | } |
1257 | } else | |
1258 | memset(ptr, 0, max_bs); | |
1259 | } |