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; |
38dad62d | 44 | int busy_check; |
df415585 | 45 | |
b9c5b644 | 46 | block = (io_u->offset - f->file_offset) / (unsigned long long) min_bs; |
c685b5b2 | 47 | nr_blocks = (io_u->buflen + min_bs - 1) / min_bs; |
3e3357b1 | 48 | blocks = 0; |
38dad62d | 49 | busy_check = !(io_u->flags & IO_U_F_BUSY_OK); |
c685b5b2 | 50 | |
3e3357b1 JA |
51 | while (nr_blocks) { |
52 | unsigned int this_blocks, mask; | |
df415585 JA |
53 | unsigned int idx, bit; |
54 | ||
1e3d53ac JA |
55 | /* |
56 | * If we have a mixed random workload, we may | |
57 | * encounter blocks we already did IO to. | |
58 | */ | |
38dad62d JA |
59 | if (!busy_check) { |
60 | blocks = nr_blocks; | |
61 | break; | |
62 | } | |
5736c10d | 63 | if ((td->o.ddir_seq_nr == 1) && !random_map_free(f, block)) |
df415585 JA |
64 | break; |
65 | ||
aec2de20 JA |
66 | idx = RAND_MAP_IDX(f, block); |
67 | bit = RAND_MAP_BIT(f, block); | |
df415585 | 68 | |
0032bf9f | 69 | fio_assert(td, idx < f->num_maps); |
df415585 | 70 | |
3e3357b1 JA |
71 | this_blocks = nr_blocks; |
72 | if (this_blocks + bit > BLOCKS_PER_MAP) | |
73 | this_blocks = BLOCKS_PER_MAP - bit; | |
74 | ||
da3758bf JA |
75 | do { |
76 | if (this_blocks == BLOCKS_PER_MAP) | |
77 | mask = -1U; | |
78 | else | |
79 | mask = ((1U << this_blocks) - 1) << bit; | |
80 | ||
81 | if (!(f->file_map[idx] & mask)) | |
82 | break; | |
83 | ||
84 | this_blocks--; | |
85 | } while (this_blocks); | |
86 | ||
87 | if (!this_blocks) | |
88 | break; | |
3e3357b1 | 89 | |
3e3357b1 JA |
90 | f->file_map[idx] |= mask; |
91 | nr_blocks -= this_blocks; | |
92 | blocks += this_blocks; | |
c9dd34b2 | 93 | block += this_blocks; |
df415585 JA |
94 | } |
95 | ||
a00735e6 JA |
96 | if ((blocks * min_bs) < io_u->buflen) |
97 | io_u->buflen = blocks * min_bs; | |
df415585 JA |
98 | } |
99 | ||
3e3357b1 JA |
100 | static unsigned long long last_block(struct thread_data *td, struct fio_file *f, |
101 | enum fio_ddir ddir) | |
2ba1c290 JA |
102 | { |
103 | unsigned long long max_blocks; | |
d9dd70f7 | 104 | unsigned long long max_size; |
2ba1c290 | 105 | |
d9dd70f7 JA |
106 | /* |
107 | * Hmm, should we make sure that ->io_size <= ->real_file_size? | |
108 | */ | |
109 | max_size = f->io_size; | |
110 | if (max_size > f->real_file_size) | |
111 | max_size = f->real_file_size; | |
112 | ||
2b7a01d0 | 113 | max_blocks = max_size / (unsigned long long) td->o.ba[ddir]; |
2ba1c290 JA |
114 | if (!max_blocks) |
115 | return 0; | |
116 | ||
67778e88 | 117 | return max_blocks; |
2ba1c290 JA |
118 | } |
119 | ||
10ba535a JA |
120 | /* |
121 | * Return the next free block in the map. | |
122 | */ | |
123 | static int get_next_free_block(struct thread_data *td, struct fio_file *f, | |
4ba66134 | 124 | enum fio_ddir ddir, unsigned long long *b) |
10ba535a | 125 | { |
84422acd | 126 | unsigned long long min_bs = td->o.rw_min_bs; |
10ba535a JA |
127 | int i; |
128 | ||
c685b5b2 JA |
129 | i = f->last_free_lookup; |
130 | *b = (i * BLOCKS_PER_MAP); | |
899c29f2 RR |
131 | while ((*b) * min_bs < f->real_file_size && |
132 | (*b) * min_bs < f->io_size) { | |
de605666 | 133 | if (f->file_map[i] != (unsigned int) -1) { |
697a606c | 134 | *b += ffz(f->file_map[i]); |
4ba66134 | 135 | if (*b > last_block(td, f, ddir)) |
2ba1c290 | 136 | break; |
c685b5b2 | 137 | f->last_free_lookup = i; |
10ba535a JA |
138 | return 0; |
139 | } | |
140 | ||
141 | *b += BLOCKS_PER_MAP; | |
142 | i++; | |
143 | } | |
144 | ||
2ba1c290 | 145 | dprint(FD_IO, "failed finding a free block\n"); |
10ba535a JA |
146 | return 1; |
147 | } | |
148 | ||
ec4015da | 149 | static int get_next_rand_offset(struct thread_data *td, struct fio_file *f, |
4ba66134 | 150 | enum fio_ddir ddir, unsigned long long *b) |
ec4015da | 151 | { |
84422acd | 152 | unsigned long long r; |
ec4015da JA |
153 | int loops = 5; |
154 | ||
155 | do { | |
156 | r = os_random_long(&td->random_state); | |
84422acd | 157 | dprint(FD_RANDOM, "off rand %llu\n", r); |
5ec10eaa | 158 | *b = (last_block(td, f, ddir) - 1) |
dc873b6f | 159 | * (r / ((unsigned long long) OS_RAND_MAX + 1.0)); |
2ba1c290 | 160 | |
43c63a78 JA |
161 | /* |
162 | * if we are not maintaining a random map, we are done. | |
163 | */ | |
303032ae | 164 | if (!file_randommap(td, f)) |
43c63a78 JA |
165 | return 0; |
166 | ||
167 | /* | |
84422acd | 168 | * calculate map offset and check if it's free |
43c63a78 | 169 | */ |
aec2de20 | 170 | if (random_map_free(f, *b)) |
43c63a78 JA |
171 | return 0; |
172 | ||
84422acd JA |
173 | dprint(FD_RANDOM, "get_next_rand_offset: offset %llu busy\n", |
174 | *b); | |
43c63a78 | 175 | } while (--loops); |
ec4015da JA |
176 | |
177 | /* | |
43c63a78 JA |
178 | * we get here, if we didn't suceed in looking up a block. generate |
179 | * a random start offset into the filemap, and find the first free | |
180 | * block from there. | |
ec4015da | 181 | */ |
43c63a78 JA |
182 | loops = 10; |
183 | do { | |
dc873b6f JA |
184 | f->last_free_lookup = (f->num_maps - 1) * |
185 | (r / (OS_RAND_MAX + 1.0)); | |
4ba66134 | 186 | if (!get_next_free_block(td, f, ddir, b)) |
43c63a78 | 187 | return 0; |
ec4015da | 188 | |
43c63a78 JA |
189 | r = os_random_long(&td->random_state); |
190 | } while (--loops); | |
191 | ||
192 | /* | |
193 | * that didn't work either, try exhaustive search from the start | |
194 | */ | |
195 | f->last_free_lookup = 0; | |
4ba66134 | 196 | return get_next_free_block(td, f, ddir, b); |
ec4015da JA |
197 | } |
198 | ||
38dad62d JA |
199 | static int get_next_rand_block(struct thread_data *td, struct fio_file *f, |
200 | enum fio_ddir ddir, unsigned long long *b) | |
201 | { | |
202 | if (get_next_rand_offset(td, f, ddir, b)) { | |
203 | dprint(FD_IO, "%s: rand offset failed, last=%llu, size=%llu\n", | |
204 | f->file_name, f->last_pos, f->real_file_size); | |
205 | return 1; | |
206 | } | |
207 | ||
208 | return 0; | |
209 | } | |
210 | ||
211 | static int get_next_seq_block(struct thread_data *td, struct fio_file *f, | |
212 | enum fio_ddir ddir, unsigned long long *b) | |
213 | { | |
214 | if (f->last_pos < f->real_file_size) { | |
215 | *b = (f->last_pos - f->file_offset) / td->o.min_bs[ddir]; | |
216 | return 0; | |
217 | } | |
218 | ||
219 | return 1; | |
220 | } | |
221 | ||
222 | static int get_next_block(struct thread_data *td, struct io_u *io_u, | |
223 | enum fio_ddir ddir, int rw_seq, unsigned long long *b) | |
224 | { | |
225 | struct fio_file *f = io_u->file; | |
226 | int ret; | |
227 | ||
228 | if (rw_seq) { | |
229 | if (td_random(td)) | |
230 | ret = get_next_rand_block(td, f, ddir, b); | |
231 | else | |
232 | ret = get_next_seq_block(td, f, ddir, b); | |
233 | } else { | |
234 | io_u->flags |= IO_U_F_BUSY_OK; | |
235 | ||
236 | if (td->o.rw_seq == RW_SEQ_SEQ) { | |
237 | ret = get_next_seq_block(td, f, ddir, b); | |
238 | if (ret) | |
239 | ret = get_next_rand_block(td, f, ddir, b); | |
240 | } else if (td->o.rw_seq == RW_SEQ_IDENT) { | |
241 | if (f->last_start != -1ULL) | |
242 | *b = (f->last_start - f->file_offset) / td->o.min_bs[ddir]; | |
243 | else | |
244 | *b = 0; | |
245 | ret = 0; | |
246 | } else { | |
247 | log_err("fio: unknown rw_seq=%d\n", td->o.rw_seq); | |
248 | ret = 1; | |
249 | } | |
250 | } | |
251 | ||
252 | return ret; | |
253 | } | |
254 | ||
10ba535a JA |
255 | /* |
256 | * For random io, generate a random new block and see if it's used. Repeat | |
257 | * until we find a free one. For sequential io, just return the end of | |
258 | * the last io issued. | |
259 | */ | |
15dc1934 | 260 | static int __get_next_offset(struct thread_data *td, struct io_u *io_u) |
10ba535a | 261 | { |
9bf2061e | 262 | struct fio_file *f = io_u->file; |
ec4015da | 263 | unsigned long long b; |
4ba66134 | 264 | enum fio_ddir ddir = io_u->ddir; |
38dad62d | 265 | int rw_seq_hit = 0; |
10ba535a | 266 | |
38dad62d JA |
267 | if (td->o.ddir_seq_nr && !--td->ddir_seq_nr) { |
268 | rw_seq_hit = 1; | |
5736c10d | 269 | td->ddir_seq_nr = td->o.ddir_seq_nr; |
38dad62d | 270 | } |
211097b2 | 271 | |
38dad62d JA |
272 | if (get_next_block(td, io_u, ddir, rw_seq_hit, &b)) { |
273 | printf("fail\n"); | |
274 | return 1; | |
43063a1c | 275 | } |
10ba535a | 276 | |
2b7a01d0 | 277 | io_u->offset = b * td->o.ba[ddir]; |
009bd847 JA |
278 | if (io_u->offset >= f->io_size) { |
279 | dprint(FD_IO, "get_next_offset: offset %llu >= io_size %llu\n", | |
280 | io_u->offset, f->io_size); | |
281 | return 1; | |
282 | } | |
283 | ||
284 | io_u->offset += f->file_offset; | |
2ba1c290 JA |
285 | if (io_u->offset >= f->real_file_size) { |
286 | dprint(FD_IO, "get_next_offset: offset %llu >= size %llu\n", | |
287 | io_u->offset, f->real_file_size); | |
10ba535a | 288 | return 1; |
2ba1c290 | 289 | } |
10ba535a JA |
290 | |
291 | return 0; | |
292 | } | |
293 | ||
15dc1934 JA |
294 | static int get_next_offset(struct thread_data *td, struct io_u *io_u) |
295 | { | |
7eb36574 JA |
296 | struct prof_io_ops *ops = &td->prof_io_ops; |
297 | ||
298 | if (ops->fill_io_u_off) | |
299 | return ops->fill_io_u_off(td, io_u); | |
15dc1934 JA |
300 | |
301 | return __get_next_offset(td, io_u); | |
302 | } | |
303 | ||
304 | static unsigned int __get_next_buflen(struct thread_data *td, struct io_u *io_u) | |
10ba535a | 305 | { |
bca4ed4d | 306 | const int ddir = io_u->ddir; |
f3f552b9 | 307 | unsigned int uninitialized_var(buflen); |
f3059de1 | 308 | unsigned int minbs, maxbs; |
10ba535a JA |
309 | long r; |
310 | ||
f3059de1 JA |
311 | minbs = td->o.min_bs[ddir]; |
312 | maxbs = td->o.max_bs[ddir]; | |
313 | ||
314 | if (minbs == maxbs) | |
315 | buflen = minbs; | |
10ba535a JA |
316 | else { |
317 | r = os_random_long(&td->bsrange_state); | |
720e84ad | 318 | if (!td->o.bssplit_nr[ddir]) { |
f3059de1 JA |
319 | buflen = 1 + (unsigned int) ((double) maxbs * |
320 | (r / (OS_RAND_MAX + 1.0))); | |
321 | if (buflen < minbs) | |
322 | buflen = minbs; | |
5ec10eaa | 323 | } else { |
564ca972 JA |
324 | long perc = 0; |
325 | unsigned int i; | |
326 | ||
720e84ad JA |
327 | for (i = 0; i < td->o.bssplit_nr[ddir]; i++) { |
328 | struct bssplit *bsp = &td->o.bssplit[ddir][i]; | |
564ca972 JA |
329 | |
330 | buflen = bsp->bs; | |
331 | perc += bsp->perc; | |
f3059de1 | 332 | if (r <= ((OS_RAND_MAX / 100L) * perc)) |
564ca972 JA |
333 | break; |
334 | } | |
335 | } | |
f3059de1 JA |
336 | if (!td->o.bs_unaligned && is_power_of_2(minbs)) |
337 | buflen = (buflen + minbs - 1) & ~(minbs - 1); | |
10ba535a JA |
338 | } |
339 | ||
4ba66134 JA |
340 | if (io_u->offset + buflen > io_u->file->real_file_size) { |
341 | dprint(FD_IO, "lower buflen %u -> %u (ddir=%d)\n", buflen, | |
f3059de1 JA |
342 | minbs, ddir); |
343 | buflen = minbs; | |
4ba66134 | 344 | } |
6a5e6884 | 345 | |
10ba535a JA |
346 | return buflen; |
347 | } | |
348 | ||
15dc1934 JA |
349 | static unsigned int get_next_buflen(struct thread_data *td, struct io_u *io_u) |
350 | { | |
7eb36574 JA |
351 | struct prof_io_ops *ops = &td->prof_io_ops; |
352 | ||
353 | if (ops->fill_io_u_size) | |
354 | return ops->fill_io_u_size(td, io_u); | |
15dc1934 JA |
355 | |
356 | return __get_next_buflen(td, io_u); | |
357 | } | |
358 | ||
afe24a5a JA |
359 | static void set_rwmix_bytes(struct thread_data *td) |
360 | { | |
afe24a5a JA |
361 | unsigned int diff; |
362 | ||
363 | /* | |
364 | * we do time or byte based switch. this is needed because | |
365 | * buffered writes may issue a lot quicker than they complete, | |
366 | * whereas reads do not. | |
367 | */ | |
e47f799f | 368 | diff = td->o.rwmix[td->rwmix_ddir ^ 1]; |
04c540d9 | 369 | td->rwmix_issues = (td->io_issues[td->rwmix_ddir] * diff) / 100; |
e47f799f JA |
370 | } |
371 | ||
372 | static inline enum fio_ddir get_rand_ddir(struct thread_data *td) | |
373 | { | |
374 | unsigned int v; | |
375 | long r; | |
376 | ||
377 | r = os_random_long(&td->rwmix_state); | |
dc873b6f | 378 | v = 1 + (int) (100.0 * (r / (OS_RAND_MAX + 1.0))); |
04c540d9 | 379 | if (v <= td->o.rwmix[DDIR_READ]) |
e47f799f JA |
380 | return DDIR_READ; |
381 | ||
382 | return DDIR_WRITE; | |
afe24a5a JA |
383 | } |
384 | ||
581e7141 JA |
385 | static enum fio_ddir rate_ddir(struct thread_data *td, enum fio_ddir ddir) |
386 | { | |
387 | enum fio_ddir odir = ddir ^ 1; | |
388 | struct timeval t; | |
389 | long usec; | |
390 | ||
391 | if (td->rate_pending_usleep[ddir] <= 0) | |
392 | return ddir; | |
393 | ||
394 | /* | |
395 | * We have too much pending sleep in this direction. See if we | |
396 | * should switch. | |
397 | */ | |
398 | if (td_rw(td)) { | |
399 | /* | |
400 | * Other direction does not have too much pending, switch | |
401 | */ | |
402 | if (td->rate_pending_usleep[odir] < 100000) | |
403 | return odir; | |
404 | ||
405 | /* | |
406 | * Both directions have pending sleep. Sleep the minimum time | |
407 | * and deduct from both. | |
408 | */ | |
409 | if (td->rate_pending_usleep[ddir] <= | |
410 | td->rate_pending_usleep[odir]) { | |
411 | usec = td->rate_pending_usleep[ddir]; | |
412 | } else { | |
413 | usec = td->rate_pending_usleep[odir]; | |
414 | ddir = odir; | |
415 | } | |
416 | } else | |
417 | usec = td->rate_pending_usleep[ddir]; | |
418 | ||
419 | fio_gettime(&t, NULL); | |
420 | usec_sleep(td, usec); | |
421 | usec = utime_since_now(&t); | |
422 | ||
423 | td->rate_pending_usleep[ddir] -= usec; | |
424 | ||
425 | odir = ddir ^ 1; | |
426 | if (td_rw(td) && __should_check_rate(td, odir)) | |
427 | td->rate_pending_usleep[odir] -= usec; | |
0b9d69ec | 428 | |
581e7141 JA |
429 | return ddir; |
430 | } | |
431 | ||
10ba535a JA |
432 | /* |
433 | * Return the data direction for the next io_u. If the job is a | |
434 | * mixed read/write workload, check the rwmix cycle and switch if | |
435 | * necessary. | |
436 | */ | |
1e97cce9 | 437 | static enum fio_ddir get_rw_ddir(struct thread_data *td) |
10ba535a | 438 | { |
581e7141 JA |
439 | enum fio_ddir ddir; |
440 | ||
5f9099ea JA |
441 | /* |
442 | * see if it's time to fsync | |
443 | */ | |
444 | if (td->o.fsync_blocks && | |
445 | !(td->io_issues[DDIR_WRITE] % td->o.fsync_blocks) && | |
446 | td->io_issues[DDIR_WRITE] && should_fsync(td)) | |
447 | return DDIR_SYNC; | |
448 | ||
449 | /* | |
450 | * see if it's time to fdatasync | |
451 | */ | |
452 | if (td->o.fdatasync_blocks && | |
453 | !(td->io_issues[DDIR_WRITE] % td->o.fdatasync_blocks) && | |
454 | td->io_issues[DDIR_WRITE] && should_fsync(td)) | |
455 | return DDIR_DATASYNC; | |
456 | ||
44f29692 JA |
457 | /* |
458 | * see if it's time to sync_file_range | |
459 | */ | |
460 | if (td->sync_file_range_nr && | |
461 | !(td->io_issues[DDIR_WRITE] % td->sync_file_range_nr) && | |
462 | td->io_issues[DDIR_WRITE] && should_fsync(td)) | |
463 | return DDIR_SYNC_FILE_RANGE; | |
464 | ||
10ba535a | 465 | if (td_rw(td)) { |
10ba535a JA |
466 | /* |
467 | * Check if it's time to seed a new data direction. | |
468 | */ | |
e4928662 | 469 | if (td->io_issues[td->rwmix_ddir] >= td->rwmix_issues) { |
e47f799f JA |
470 | /* |
471 | * Put a top limit on how many bytes we do for | |
472 | * one data direction, to avoid overflowing the | |
473 | * ranges too much | |
474 | */ | |
475 | ddir = get_rand_ddir(td); | |
e47f799f JA |
476 | |
477 | if (ddir != td->rwmix_ddir) | |
478 | set_rwmix_bytes(td); | |
479 | ||
480 | td->rwmix_ddir = ddir; | |
10ba535a | 481 | } |
581e7141 | 482 | ddir = td->rwmix_ddir; |
10ba535a | 483 | } else if (td_read(td)) |
581e7141 | 484 | ddir = DDIR_READ; |
10ba535a | 485 | else |
581e7141 JA |
486 | ddir = DDIR_WRITE; |
487 | ||
488 | td->rwmix_ddir = rate_ddir(td, ddir); | |
489 | return td->rwmix_ddir; | |
10ba535a JA |
490 | } |
491 | ||
e8462bd8 | 492 | void put_file_log(struct thread_data *td, struct fio_file *f) |
60f2c658 JA |
493 | { |
494 | int ret = put_file(td, f); | |
495 | ||
496 | if (ret) | |
497 | td_verror(td, ret, "file close"); | |
498 | } | |
499 | ||
10ba535a JA |
500 | void put_io_u(struct thread_data *td, struct io_u *io_u) |
501 | { | |
e8462bd8 JA |
502 | td_io_u_lock(td); |
503 | ||
0c6e7517 | 504 | io_u->flags |= IO_U_F_FREE; |
e8462bd8 | 505 | io_u->flags &= ~IO_U_F_FREE_DEF; |
0c6e7517 | 506 | |
60f2c658 JA |
507 | if (io_u->file) |
508 | put_file_log(td, io_u->file); | |
2dbdab7e | 509 | |
10ba535a | 510 | io_u->file = NULL; |
0c41214f RR |
511 | if (io_u->flags & IO_U_F_IN_CUR_DEPTH) |
512 | td->cur_depth--; | |
e8462bd8 | 513 | flist_del_init(&io_u->list); |
01743ee1 | 514 | flist_add(&io_u->list, &td->io_u_freelist); |
e8462bd8 JA |
515 | td_io_u_unlock(td); |
516 | td_io_u_free_notify(td); | |
10ba535a JA |
517 | } |
518 | ||
f2bba182 RR |
519 | void clear_io_u(struct thread_data *td, struct io_u *io_u) |
520 | { | |
521 | io_u->flags &= ~IO_U_F_FLIGHT; | |
522 | put_io_u(td, io_u); | |
523 | } | |
524 | ||
755200a3 JA |
525 | void requeue_io_u(struct thread_data *td, struct io_u **io_u) |
526 | { | |
527 | struct io_u *__io_u = *io_u; | |
528 | ||
465221b0 JA |
529 | dprint(FD_IO, "requeue %p\n", __io_u); |
530 | ||
e8462bd8 JA |
531 | td_io_u_lock(td); |
532 | ||
4d2e0f49 | 533 | __io_u->flags |= IO_U_F_FREE; |
5f9099ea | 534 | if ((__io_u->flags & IO_U_F_FLIGHT) && !ddir_sync(__io_u->ddir)) |
e4f54adb | 535 | td->io_issues[__io_u->ddir]--; |
5ec10eaa | 536 | |
4d2e0f49 | 537 | __io_u->flags &= ~IO_U_F_FLIGHT; |
0c41214f RR |
538 | if (__io_u->flags & IO_U_F_IN_CUR_DEPTH) |
539 | td->cur_depth--; | |
01743ee1 JA |
540 | flist_del(&__io_u->list); |
541 | flist_add_tail(&__io_u->list, &td->io_u_requeues); | |
e8462bd8 | 542 | td_io_u_unlock(td); |
755200a3 JA |
543 | *io_u = NULL; |
544 | } | |
545 | ||
9bf2061e | 546 | static int fill_io_u(struct thread_data *td, struct io_u *io_u) |
10ba535a | 547 | { |
b4c5e1ac JA |
548 | if (td->io_ops->flags & FIO_NOIO) |
549 | goto out; | |
550 | ||
5f9099ea JA |
551 | io_u->ddir = get_rw_ddir(td); |
552 | ||
87dc1ab1 | 553 | /* |
5f9099ea | 554 | * fsync() or fdatasync(), we are done |
87dc1ab1 | 555 | */ |
5f9099ea | 556 | if (ddir_sync(io_u->ddir)) |
c38e9468 | 557 | goto out; |
a00735e6 | 558 | |
48f5abd3 JA |
559 | /* |
560 | * See if it's time to switch to a new zone | |
561 | */ | |
562 | if (td->zone_bytes >= td->o.zone_size) { | |
563 | td->zone_bytes = 0; | |
564 | io_u->file->last_pos += td->o.zone_skip; | |
565 | td->io_skip_bytes += td->o.zone_skip; | |
566 | } | |
567 | ||
10ba535a | 568 | /* |
c685b5b2 JA |
569 | * No log, let the seq/rand engine retrieve the next buflen and |
570 | * position. | |
10ba535a | 571 | */ |
2ba1c290 JA |
572 | if (get_next_offset(td, io_u)) { |
573 | dprint(FD_IO, "io_u %p, failed getting offset\n", io_u); | |
bca4ed4d | 574 | return 1; |
2ba1c290 | 575 | } |
10ba535a | 576 | |
9bf2061e | 577 | io_u->buflen = get_next_buflen(td, io_u); |
2ba1c290 JA |
578 | if (!io_u->buflen) { |
579 | dprint(FD_IO, "io_u %p, failed getting buflen\n", io_u); | |
bca4ed4d | 580 | return 1; |
2ba1c290 | 581 | } |
bca4ed4d | 582 | |
2ba1c290 JA |
583 | if (io_u->offset + io_u->buflen > io_u->file->real_file_size) { |
584 | dprint(FD_IO, "io_u %p, offset too large\n", io_u); | |
4ba66134 JA |
585 | dprint(FD_IO, " off=%llu/%lu > %llu\n", io_u->offset, |
586 | io_u->buflen, io_u->file->real_file_size); | |
6a5e6884 | 587 | return 1; |
2ba1c290 | 588 | } |
6a5e6884 | 589 | |
bca4ed4d JA |
590 | /* |
591 | * mark entry before potentially trimming io_u | |
592 | */ | |
303032ae | 593 | if (td_random(td) && file_randommap(td, io_u->file)) |
9bf2061e | 594 | mark_random_map(td, io_u); |
bca4ed4d JA |
595 | |
596 | /* | |
597 | * If using a write iolog, store this entry. | |
598 | */ | |
c38e9468 | 599 | out: |
2ba1c290 | 600 | dprint_io_u(io_u, "fill_io_u"); |
d9d91e39 | 601 | td->zone_bytes += io_u->buflen; |
f29b25a3 | 602 | log_io_u(td, io_u); |
bca4ed4d | 603 | return 0; |
10ba535a JA |
604 | } |
605 | ||
838bc709 JA |
606 | static void __io_u_mark_map(unsigned int *map, unsigned int nr) |
607 | { | |
608 | int index = 0; | |
609 | ||
610 | switch (nr) { | |
611 | default: | |
612 | index = 6; | |
613 | break; | |
614 | case 33 ... 64: | |
615 | index = 5; | |
616 | break; | |
617 | case 17 ... 32: | |
618 | index = 4; | |
619 | break; | |
620 | case 9 ... 16: | |
621 | index = 3; | |
622 | break; | |
623 | case 5 ... 8: | |
624 | index = 2; | |
625 | break; | |
626 | case 1 ... 4: | |
627 | index = 1; | |
628 | case 0: | |
629 | break; | |
630 | } | |
631 | ||
632 | map[index]++; | |
633 | } | |
634 | ||
635 | void io_u_mark_submit(struct thread_data *td, unsigned int nr) | |
636 | { | |
637 | __io_u_mark_map(td->ts.io_u_submit, nr); | |
638 | td->ts.total_submit++; | |
639 | } | |
640 | ||
641 | void io_u_mark_complete(struct thread_data *td, unsigned int nr) | |
642 | { | |
643 | __io_u_mark_map(td->ts.io_u_complete, nr); | |
644 | td->ts.total_complete++; | |
645 | } | |
646 | ||
d8005759 | 647 | void io_u_mark_depth(struct thread_data *td, unsigned int nr) |
71619dc2 JA |
648 | { |
649 | int index = 0; | |
650 | ||
651 | switch (td->cur_depth) { | |
652 | default: | |
a783e61a JA |
653 | index = 6; |
654 | break; | |
71619dc2 | 655 | case 32 ... 63: |
a783e61a JA |
656 | index = 5; |
657 | break; | |
71619dc2 | 658 | case 16 ... 31: |
a783e61a JA |
659 | index = 4; |
660 | break; | |
71619dc2 | 661 | case 8 ... 15: |
a783e61a JA |
662 | index = 3; |
663 | break; | |
71619dc2 | 664 | case 4 ... 7: |
a783e61a JA |
665 | index = 2; |
666 | break; | |
71619dc2 | 667 | case 2 ... 3: |
a783e61a | 668 | index = 1; |
71619dc2 JA |
669 | case 1: |
670 | break; | |
671 | } | |
672 | ||
3bec7ae0 | 673 | td->ts.io_u_map[index] += nr; |
71619dc2 JA |
674 | } |
675 | ||
04a0feae JA |
676 | static void io_u_mark_lat_usec(struct thread_data *td, unsigned long usec) |
677 | { | |
678 | int index = 0; | |
679 | ||
680 | assert(usec < 1000); | |
681 | ||
682 | switch (usec) { | |
683 | case 750 ... 999: | |
684 | index = 9; | |
685 | break; | |
686 | case 500 ... 749: | |
687 | index = 8; | |
688 | break; | |
689 | case 250 ... 499: | |
690 | index = 7; | |
691 | break; | |
692 | case 100 ... 249: | |
693 | index = 6; | |
694 | break; | |
695 | case 50 ... 99: | |
696 | index = 5; | |
697 | break; | |
698 | case 20 ... 49: | |
699 | index = 4; | |
700 | break; | |
701 | case 10 ... 19: | |
702 | index = 3; | |
703 | break; | |
704 | case 4 ... 9: | |
705 | index = 2; | |
706 | break; | |
707 | case 2 ... 3: | |
708 | index = 1; | |
709 | case 0 ... 1: | |
710 | break; | |
711 | } | |
712 | ||
713 | assert(index < FIO_IO_U_LAT_U_NR); | |
714 | td->ts.io_u_lat_u[index]++; | |
715 | } | |
716 | ||
717 | static void io_u_mark_lat_msec(struct thread_data *td, unsigned long msec) | |
ec118304 JA |
718 | { |
719 | int index = 0; | |
720 | ||
721 | switch (msec) { | |
722 | default: | |
04a0feae JA |
723 | index = 11; |
724 | break; | |
8abdce66 | 725 | case 1000 ... 1999: |
04a0feae JA |
726 | index = 10; |
727 | break; | |
8abdce66 | 728 | case 750 ... 999: |
04a0feae JA |
729 | index = 9; |
730 | break; | |
8abdce66 | 731 | case 500 ... 749: |
04a0feae JA |
732 | index = 8; |
733 | break; | |
8abdce66 | 734 | case 250 ... 499: |
04a0feae JA |
735 | index = 7; |
736 | break; | |
8abdce66 | 737 | case 100 ... 249: |
04a0feae JA |
738 | index = 6; |
739 | break; | |
8abdce66 | 740 | case 50 ... 99: |
04a0feae JA |
741 | index = 5; |
742 | break; | |
8abdce66 | 743 | case 20 ... 49: |
04a0feae JA |
744 | index = 4; |
745 | break; | |
8abdce66 | 746 | case 10 ... 19: |
04a0feae JA |
747 | index = 3; |
748 | break; | |
8abdce66 | 749 | case 4 ... 9: |
04a0feae JA |
750 | index = 2; |
751 | break; | |
ec118304 | 752 | case 2 ... 3: |
04a0feae | 753 | index = 1; |
ec118304 JA |
754 | case 0 ... 1: |
755 | break; | |
756 | } | |
757 | ||
04a0feae JA |
758 | assert(index < FIO_IO_U_LAT_M_NR); |
759 | td->ts.io_u_lat_m[index]++; | |
760 | } | |
761 | ||
762 | static void io_u_mark_latency(struct thread_data *td, unsigned long usec) | |
763 | { | |
764 | if (usec < 1000) | |
765 | io_u_mark_lat_usec(td, usec); | |
766 | else | |
767 | io_u_mark_lat_msec(td, usec / 1000); | |
ec118304 JA |
768 | } |
769 | ||
0aabe160 JA |
770 | /* |
771 | * Get next file to service by choosing one at random | |
772 | */ | |
2cc52930 JA |
773 | static struct fio_file *get_next_file_rand(struct thread_data *td, |
774 | enum fio_file_flags goodf, | |
d6aed795 | 775 | enum fio_file_flags badf) |
0aabe160 | 776 | { |
0aabe160 | 777 | struct fio_file *f; |
1c178180 | 778 | int fno; |
0aabe160 JA |
779 | |
780 | do { | |
7c83c089 | 781 | long r = os_random_long(&td->next_file_state); |
87b10676 | 782 | int opened = 0; |
7c83c089 | 783 | |
5ec10eaa | 784 | fno = (unsigned int) ((double) td->o.nr_files |
dc873b6f | 785 | * (r / (OS_RAND_MAX + 1.0))); |
126d65c6 | 786 | f = td->files[fno]; |
d6aed795 | 787 | if (fio_file_done(f)) |
059e63c0 | 788 | continue; |
1c178180 | 789 | |
d6aed795 | 790 | if (!fio_file_open(f)) { |
87b10676 JA |
791 | int err; |
792 | ||
793 | err = td_io_open_file(td, f); | |
794 | if (err) | |
795 | continue; | |
796 | opened = 1; | |
797 | } | |
798 | ||
2ba1c290 JA |
799 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) { |
800 | dprint(FD_FILE, "get_next_file_rand: %p\n", f); | |
0aabe160 | 801 | return f; |
2ba1c290 | 802 | } |
87b10676 JA |
803 | if (opened) |
804 | td_io_close_file(td, f); | |
0aabe160 JA |
805 | } while (1); |
806 | } | |
807 | ||
808 | /* | |
809 | * Get next file to service by doing round robin between all available ones | |
810 | */ | |
1c178180 JA |
811 | static struct fio_file *get_next_file_rr(struct thread_data *td, int goodf, |
812 | int badf) | |
3d7c391d JA |
813 | { |
814 | unsigned int old_next_file = td->next_file; | |
815 | struct fio_file *f; | |
816 | ||
817 | do { | |
87b10676 JA |
818 | int opened = 0; |
819 | ||
126d65c6 | 820 | f = td->files[td->next_file]; |
3d7c391d JA |
821 | |
822 | td->next_file++; | |
2dc1bbeb | 823 | if (td->next_file >= td->o.nr_files) |
3d7c391d JA |
824 | td->next_file = 0; |
825 | ||
87b10676 | 826 | dprint(FD_FILE, "trying file %s %x\n", f->file_name, f->flags); |
d6aed795 | 827 | if (fio_file_done(f)) { |
d5ed68ea | 828 | f = NULL; |
059e63c0 | 829 | continue; |
d5ed68ea | 830 | } |
059e63c0 | 831 | |
d6aed795 | 832 | if (!fio_file_open(f)) { |
87b10676 JA |
833 | int err; |
834 | ||
835 | err = td_io_open_file(td, f); | |
b5696bfc JA |
836 | if (err) { |
837 | dprint(FD_FILE, "error %d on open of %s\n", | |
838 | err, f->file_name); | |
87c27b45 | 839 | f = NULL; |
87b10676 | 840 | continue; |
b5696bfc | 841 | } |
87b10676 JA |
842 | opened = 1; |
843 | } | |
844 | ||
0b9d69ec JA |
845 | dprint(FD_FILE, "goodf=%x, badf=%x, ff=%x\n", goodf, badf, |
846 | f->flags); | |
1c178180 | 847 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) |
3d7c391d JA |
848 | break; |
849 | ||
87b10676 JA |
850 | if (opened) |
851 | td_io_close_file(td, f); | |
852 | ||
3d7c391d JA |
853 | f = NULL; |
854 | } while (td->next_file != old_next_file); | |
855 | ||
2ba1c290 | 856 | dprint(FD_FILE, "get_next_file_rr: %p\n", f); |
3d7c391d JA |
857 | return f; |
858 | } | |
859 | ||
7eb36574 | 860 | static struct fio_file *__get_next_file(struct thread_data *td) |
bdb4e2e9 | 861 | { |
1907dbc6 JA |
862 | struct fio_file *f; |
863 | ||
2dc1bbeb | 864 | assert(td->o.nr_files <= td->files_index); |
1c178180 | 865 | |
b5696bfc | 866 | if (td->nr_done_files >= td->o.nr_files) { |
5ec10eaa JA |
867 | dprint(FD_FILE, "get_next_file: nr_open=%d, nr_done=%d," |
868 | " nr_files=%d\n", td->nr_open_files, | |
869 | td->nr_done_files, | |
870 | td->o.nr_files); | |
bdb4e2e9 | 871 | return NULL; |
2ba1c290 | 872 | } |
bdb4e2e9 | 873 | |
1907dbc6 | 874 | f = td->file_service_file; |
d6aed795 | 875 | if (f && fio_file_open(f) && !fio_file_closing(f)) { |
a086c257 JA |
876 | if (td->o.file_service_type == FIO_FSERVICE_SEQ) |
877 | goto out; | |
878 | if (td->file_service_left--) | |
879 | goto out; | |
880 | } | |
1907dbc6 | 881 | |
a086c257 JA |
882 | if (td->o.file_service_type == FIO_FSERVICE_RR || |
883 | td->o.file_service_type == FIO_FSERVICE_SEQ) | |
d6aed795 | 884 | f = get_next_file_rr(td, FIO_FILE_open, FIO_FILE_closing); |
bdb4e2e9 | 885 | else |
d6aed795 | 886 | f = get_next_file_rand(td, FIO_FILE_open, FIO_FILE_closing); |
1907dbc6 JA |
887 | |
888 | td->file_service_file = f; | |
889 | td->file_service_left = td->file_service_nr - 1; | |
2ba1c290 | 890 | out: |
683023e8 | 891 | dprint(FD_FILE, "get_next_file: %p [%s]\n", f, f->file_name); |
1907dbc6 | 892 | return f; |
bdb4e2e9 JA |
893 | } |
894 | ||
7eb36574 JA |
895 | static struct fio_file *get_next_file(struct thread_data *td) |
896 | { | |
897 | struct prof_io_ops *ops = &td->prof_io_ops; | |
898 | ||
899 | if (ops->get_next_file) | |
900 | return ops->get_next_file(td); | |
901 | ||
902 | return __get_next_file(td); | |
903 | } | |
904 | ||
429f6675 JA |
905 | static int set_io_u_file(struct thread_data *td, struct io_u *io_u) |
906 | { | |
907 | struct fio_file *f; | |
908 | ||
909 | do { | |
910 | f = get_next_file(td); | |
911 | if (!f) | |
912 | return 1; | |
913 | ||
429f6675 JA |
914 | io_u->file = f; |
915 | get_file(f); | |
916 | ||
917 | if (!fill_io_u(td, io_u)) | |
918 | break; | |
919 | ||
b5696bfc | 920 | put_file_log(td, f); |
429f6675 | 921 | td_io_close_file(td, f); |
b5696bfc | 922 | io_u->file = NULL; |
d6aed795 | 923 | fio_file_set_done(f); |
429f6675 | 924 | td->nr_done_files++; |
0b9d69ec JA |
925 | dprint(FD_FILE, "%s: is done (%d of %d)\n", f->file_name, |
926 | td->nr_done_files, td->o.nr_files); | |
429f6675 JA |
927 | } while (1); |
928 | ||
929 | return 0; | |
930 | } | |
931 | ||
932 | ||
10ba535a JA |
933 | struct io_u *__get_io_u(struct thread_data *td) |
934 | { | |
935 | struct io_u *io_u = NULL; | |
936 | ||
e8462bd8 JA |
937 | td_io_u_lock(td); |
938 | ||
939 | again: | |
01743ee1 JA |
940 | if (!flist_empty(&td->io_u_requeues)) |
941 | io_u = flist_entry(td->io_u_requeues.next, struct io_u, list); | |
755200a3 | 942 | else if (!queue_full(td)) { |
01743ee1 | 943 | io_u = flist_entry(td->io_u_freelist.next, struct io_u, list); |
10ba535a | 944 | |
6040dabc | 945 | io_u->buflen = 0; |
10ba535a | 946 | io_u->resid = 0; |
755200a3 | 947 | io_u->file = NULL; |
d7762cf8 | 948 | io_u->end_io = NULL; |
755200a3 JA |
949 | } |
950 | ||
951 | if (io_u) { | |
0c6e7517 | 952 | assert(io_u->flags & IO_U_F_FREE); |
2ecc1b57 | 953 | io_u->flags &= ~(IO_U_F_FREE | IO_U_F_FREE_DEF); |
0c6e7517 | 954 | |
755200a3 | 955 | io_u->error = 0; |
01743ee1 JA |
956 | flist_del(&io_u->list); |
957 | flist_add(&io_u->list, &td->io_u_busylist); | |
10ba535a | 958 | td->cur_depth++; |
0c41214f | 959 | io_u->flags |= IO_U_F_IN_CUR_DEPTH; |
1dec3e07 JA |
960 | } else if (td->o.verify_async) { |
961 | /* | |
962 | * We ran out, wait for async verify threads to finish and | |
963 | * return one | |
964 | */ | |
965 | pthread_cond_wait(&td->free_cond, &td->io_u_lock); | |
966 | goto again; | |
10ba535a JA |
967 | } |
968 | ||
e8462bd8 | 969 | td_io_u_unlock(td); |
10ba535a JA |
970 | return io_u; |
971 | } | |
972 | ||
973 | /* | |
974 | * Return an io_u to be processed. Gets a buflen and offset, sets direction, | |
975 | * etc. The returned io_u is fully ready to be prepped and submitted. | |
976 | */ | |
3d7c391d | 977 | struct io_u *get_io_u(struct thread_data *td) |
10ba535a | 978 | { |
3d7c391d | 979 | struct fio_file *f; |
10ba535a JA |
980 | struct io_u *io_u; |
981 | ||
982 | io_u = __get_io_u(td); | |
2ba1c290 JA |
983 | if (!io_u) { |
984 | dprint(FD_IO, "__get_io_u failed\n"); | |
10ba535a | 985 | return NULL; |
2ba1c290 | 986 | } |
10ba535a | 987 | |
9e144189 JA |
988 | if (td->o.verify_backlog && td->io_hist_len) { |
989 | int get_verify = 0; | |
990 | ||
991 | if (td->verify_batch) { | |
992 | td->verify_batch--; | |
993 | get_verify = 1; | |
994 | } else if (!(td->io_hist_len % td->o.verify_backlog) && | |
995 | td->last_ddir != DDIR_READ) { | |
996 | td->verify_batch = td->o.verify_batch; | |
f8a75c99 JA |
997 | if (!td->verify_batch) |
998 | td->verify_batch = td->o.verify_backlog; | |
9e144189 JA |
999 | get_verify = 1; |
1000 | } | |
1001 | ||
1002 | if (get_verify && !get_next_verify(td, io_u)) | |
1003 | goto out; | |
1004 | } | |
1005 | ||
755200a3 JA |
1006 | /* |
1007 | * from a requeue, io_u already setup | |
1008 | */ | |
1009 | if (io_u->file) | |
77f392bf | 1010 | goto out; |
755200a3 | 1011 | |
429f6675 JA |
1012 | /* |
1013 | * If using an iolog, grab next piece if any available. | |
1014 | */ | |
1015 | if (td->o.read_iolog_file) { | |
1016 | if (read_iolog_get(td, io_u)) | |
1017 | goto err_put; | |
2ba1c290 JA |
1018 | } else if (set_io_u_file(td, io_u)) { |
1019 | dprint(FD_IO, "io_u %p, setting file failed\n", io_u); | |
429f6675 | 1020 | goto err_put; |
2ba1c290 | 1021 | } |
5ec10eaa | 1022 | |
429f6675 | 1023 | f = io_u->file; |
d6aed795 | 1024 | assert(fio_file_open(f)); |
97af62ce | 1025 | |
5f9099ea | 1026 | if (!ddir_sync(io_u->ddir)) { |
d0656a93 | 1027 | if (!io_u->buflen && !(td->io_ops->flags & FIO_NOIO)) { |
2ba1c290 | 1028 | dprint(FD_IO, "get_io_u: zero buflen on %p\n", io_u); |
429f6675 | 1029 | goto err_put; |
2ba1c290 | 1030 | } |
10ba535a | 1031 | |
38dad62d | 1032 | f->last_start = io_u->offset; |
36167d82 | 1033 | f->last_pos = io_u->offset + io_u->buflen; |
10ba535a | 1034 | |
c311cd2a | 1035 | if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_WRITE) |
87dc1ab1 | 1036 | populate_verify_io_u(td, io_u); |
e4dad9c6 JA |
1037 | else if (td->o.refill_buffers && io_u->ddir == DDIR_WRITE) |
1038 | io_u_fill_buffer(td, io_u, io_u->xfer_buflen); | |
cbe8d756 RR |
1039 | else if (io_u->ddir == DDIR_READ) { |
1040 | /* | |
1041 | * Reset the buf_filled parameters so next time if the | |
1042 | * buffer is used for writes it is refilled. | |
1043 | */ | |
cbe8d756 RR |
1044 | io_u->buf_filled_len = 0; |
1045 | } | |
87dc1ab1 | 1046 | } |
10ba535a | 1047 | |
165faf16 JA |
1048 | /* |
1049 | * Set io data pointers. | |
1050 | */ | |
cec6b55d JA |
1051 | io_u->xfer_buf = io_u->buf; |
1052 | io_u->xfer_buflen = io_u->buflen; | |
5973cafb | 1053 | |
6ac7a331 | 1054 | out: |
429f6675 | 1055 | if (!td_io_prep(td, io_u)) { |
993bf48b JA |
1056 | if (!td->o.disable_slat) |
1057 | fio_gettime(&io_u->start_time, NULL); | |
429f6675 | 1058 | return io_u; |
36167d82 | 1059 | } |
429f6675 | 1060 | err_put: |
2ba1c290 | 1061 | dprint(FD_IO, "get_io_u failed\n"); |
429f6675 JA |
1062 | put_io_u(td, io_u); |
1063 | return NULL; | |
10ba535a JA |
1064 | } |
1065 | ||
5451792e JA |
1066 | void io_u_log_error(struct thread_data *td, struct io_u *io_u) |
1067 | { | |
1068 | const char *msg[] = { "read", "write", "sync" }; | |
1069 | ||
1070 | log_err("fio: io_u error"); | |
1071 | ||
1072 | if (io_u->file) | |
1073 | log_err(" on file %s", io_u->file->file_name); | |
1074 | ||
1075 | log_err(": %s\n", strerror(io_u->error)); | |
1076 | ||
5ec10eaa JA |
1077 | log_err(" %s offset=%llu, buflen=%lu\n", msg[io_u->ddir], |
1078 | io_u->offset, io_u->xfer_buflen); | |
5451792e JA |
1079 | |
1080 | if (!td->error) | |
1081 | td_verror(td, io_u->error, "io_u error"); | |
1082 | } | |
1083 | ||
97601024 JA |
1084 | static void io_completed(struct thread_data *td, struct io_u *io_u, |
1085 | struct io_completion_data *icd) | |
10ba535a | 1086 | { |
dbad30b0 JA |
1087 | /* |
1088 | * Older gcc's are too dumb to realize that usec is always used | |
1089 | * initialized, silence that warning. | |
1090 | */ | |
1091 | unsigned long uninitialized_var(usec); | |
44f29692 | 1092 | struct fio_file *f; |
10ba535a | 1093 | |
2ba1c290 JA |
1094 | dprint_io_u(io_u, "io complete"); |
1095 | ||
2ecc1b57 | 1096 | td_io_u_lock(td); |
0c6e7517 | 1097 | assert(io_u->flags & IO_U_F_FLIGHT); |
38dad62d | 1098 | io_u->flags &= ~(IO_U_F_FLIGHT | IO_U_F_BUSY_OK); |
2ecc1b57 | 1099 | td_io_u_unlock(td); |
0c6e7517 | 1100 | |
5f9099ea | 1101 | if (ddir_sync(io_u->ddir)) { |
87dc1ab1 | 1102 | td->last_was_sync = 1; |
44f29692 JA |
1103 | f = io_u->file; |
1104 | if (f) { | |
1105 | f->first_write = -1ULL; | |
1106 | f->last_write = -1ULL; | |
1107 | } | |
87dc1ab1 JA |
1108 | return; |
1109 | } | |
1110 | ||
1111 | td->last_was_sync = 0; | |
9e144189 | 1112 | td->last_ddir = io_u->ddir; |
87dc1ab1 | 1113 | |
10ba535a JA |
1114 | if (!io_u->error) { |
1115 | unsigned int bytes = io_u->buflen - io_u->resid; | |
1e97cce9 | 1116 | const enum fio_ddir idx = io_u->ddir; |
ba3e4e0c | 1117 | const enum fio_ddir odx = io_u->ddir ^ 1; |
b29ee5b3 | 1118 | int ret; |
10ba535a | 1119 | |
b29ee5b3 JA |
1120 | td->io_blocks[idx]++; |
1121 | td->io_bytes[idx] += bytes; | |
1122 | td->this_io_bytes[idx] += bytes; | |
10ba535a | 1123 | |
44f29692 JA |
1124 | if (idx == DDIR_WRITE) { |
1125 | f = io_u->file; | |
1126 | if (f) { | |
1127 | if (f->first_write == -1ULL || | |
1128 | io_u->offset < f->first_write) | |
1129 | f->first_write = io_u->offset; | |
1130 | if (f->last_write == -1ULL || | |
1131 | ((io_u->offset + bytes) > f->last_write)) | |
1132 | f->last_write = io_u->offset + bytes; | |
1133 | } | |
1134 | } | |
1135 | ||
b29ee5b3 | 1136 | if (ramp_time_over(td)) { |
40e1a6f0 | 1137 | unsigned long uninitialized_var(lusec); |
40e1a6f0 JA |
1138 | |
1139 | if (!td->o.disable_clat || !td->o.disable_bw) | |
1140 | lusec = utime_since(&io_u->issue_time, | |
1141 | &icd->time); | |
02af0988 JA |
1142 | if (!td->o.disable_lat) { |
1143 | unsigned long tusec; | |
721938ae | 1144 | |
02af0988 JA |
1145 | tusec = utime_since(&io_u->start_time, |
1146 | &icd->time); | |
1147 | add_lat_sample(td, idx, tusec, bytes); | |
1148 | } | |
9520ebb9 | 1149 | if (!td->o.disable_clat) { |
01a1caa5 | 1150 | add_clat_sample(td, idx, lusec, bytes); |
40e1a6f0 | 1151 | io_u_mark_latency(td, lusec); |
9520ebb9 JA |
1152 | } |
1153 | if (!td->o.disable_bw) | |
306ddc97 | 1154 | add_bw_sample(td, idx, bytes, &icd->time); |
b23b6a2f | 1155 | if (__should_check_rate(td, idx)) { |
ba3e4e0c RR |
1156 | td->rate_pending_usleep[idx] = |
1157 | ((td->this_io_bytes[idx] * | |
1158 | td->rate_nsec_cycle[idx]) / 1000 - | |
1159 | utime_since_now(&td->start)); | |
b23b6a2f | 1160 | } |
581e7141 | 1161 | if (__should_check_rate(td, idx ^ 1)) |
ba3e4e0c RR |
1162 | td->rate_pending_usleep[odx] = |
1163 | ((td->this_io_bytes[odx] * | |
1164 | td->rate_nsec_cycle[odx]) / 1000 - | |
1165 | utime_since_now(&td->start)); | |
721938ae | 1166 | } |
10ba535a | 1167 | |
660a1cb5 | 1168 | if (td_write(td) && idx == DDIR_WRITE && |
8670579e | 1169 | td->o.do_verify && |
41128405 | 1170 | td->o.verify != VERIFY_NONE) |
10ba535a JA |
1171 | log_io_piece(td, io_u); |
1172 | ||
b29ee5b3 | 1173 | icd->bytes_done[idx] += bytes; |
3af6ef39 | 1174 | |
d7762cf8 | 1175 | if (io_u->end_io) { |
36690c9b | 1176 | ret = io_u->end_io(td, io_u); |
3af6ef39 JA |
1177 | if (ret && !icd->error) |
1178 | icd->error = ret; | |
1179 | } | |
5451792e | 1180 | } else { |
10ba535a | 1181 | icd->error = io_u->error; |
5451792e JA |
1182 | io_u_log_error(td, io_u); |
1183 | } | |
f2bba182 RR |
1184 | if (td->o.continue_on_error && icd->error && |
1185 | td_non_fatal_error(icd->error)) { | |
1186 | /* | |
1187 | * If there is a non_fatal error, then add to the error count | |
1188 | * and clear all the errors. | |
1189 | */ | |
1190 | update_error_count(td, icd->error); | |
1191 | td_clear_error(td); | |
1192 | icd->error = 0; | |
1193 | io_u->error = 0; | |
1194 | } | |
10ba535a JA |
1195 | } |
1196 | ||
9520ebb9 JA |
1197 | static void init_icd(struct thread_data *td, struct io_completion_data *icd, |
1198 | int nr) | |
10ba535a | 1199 | { |
9520ebb9 JA |
1200 | if (!td->o.disable_clat || !td->o.disable_bw) |
1201 | fio_gettime(&icd->time, NULL); | |
02bcaa8c | 1202 | |
3af6ef39 JA |
1203 | icd->nr = nr; |
1204 | ||
10ba535a JA |
1205 | icd->error = 0; |
1206 | icd->bytes_done[0] = icd->bytes_done[1] = 0; | |
36167d82 JA |
1207 | } |
1208 | ||
97601024 JA |
1209 | static void ios_completed(struct thread_data *td, |
1210 | struct io_completion_data *icd) | |
36167d82 JA |
1211 | { |
1212 | struct io_u *io_u; | |
1213 | int i; | |
1214 | ||
10ba535a JA |
1215 | for (i = 0; i < icd->nr; i++) { |
1216 | io_u = td->io_ops->event(td, i); | |
1217 | ||
1218 | io_completed(td, io_u, icd); | |
e8462bd8 JA |
1219 | |
1220 | if (!(io_u->flags & IO_U_F_FREE_DEF)) | |
1221 | put_io_u(td, io_u); | |
10ba535a JA |
1222 | } |
1223 | } | |
97601024 | 1224 | |
e7e6cfb4 JA |
1225 | /* |
1226 | * Complete a single io_u for the sync engines. | |
1227 | */ | |
581e7141 JA |
1228 | int io_u_sync_complete(struct thread_data *td, struct io_u *io_u, |
1229 | unsigned long *bytes) | |
97601024 JA |
1230 | { |
1231 | struct io_completion_data icd; | |
1232 | ||
9520ebb9 | 1233 | init_icd(td, &icd, 1); |
97601024 | 1234 | io_completed(td, io_u, &icd); |
e8462bd8 JA |
1235 | |
1236 | if (!(io_u->flags & IO_U_F_FREE_DEF)) | |
1237 | put_io_u(td, io_u); | |
97601024 | 1238 | |
581e7141 JA |
1239 | if (icd.error) { |
1240 | td_verror(td, icd.error, "io_u_sync_complete"); | |
1241 | return -1; | |
1242 | } | |
97601024 | 1243 | |
581e7141 JA |
1244 | if (bytes) { |
1245 | bytes[0] += icd.bytes_done[0]; | |
1246 | bytes[1] += icd.bytes_done[1]; | |
1247 | } | |
1248 | ||
1249 | return 0; | |
97601024 JA |
1250 | } |
1251 | ||
e7e6cfb4 JA |
1252 | /* |
1253 | * Called to complete min_events number of io for the async engines. | |
1254 | */ | |
581e7141 JA |
1255 | int io_u_queued_complete(struct thread_data *td, int min_evts, |
1256 | unsigned long *bytes) | |
97601024 | 1257 | { |
97601024 | 1258 | struct io_completion_data icd; |
00de55ef | 1259 | struct timespec *tvp = NULL; |
97601024 | 1260 | int ret; |
4d06a338 | 1261 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0, }; |
97601024 | 1262 | |
4950421a | 1263 | dprint(FD_IO, "io_u_queued_completed: min=%d\n", min_evts); |
b271fe62 | 1264 | |
4950421a | 1265 | if (!min_evts) |
00de55ef | 1266 | tvp = &ts; |
97601024 | 1267 | |
4950421a | 1268 | ret = td_io_getevents(td, min_evts, td->o.iodepth_batch_complete, tvp); |
97601024 | 1269 | if (ret < 0) { |
e1161c32 | 1270 | td_verror(td, -ret, "td_io_getevents"); |
97601024 JA |
1271 | return ret; |
1272 | } else if (!ret) | |
1273 | return ret; | |
1274 | ||
9520ebb9 | 1275 | init_icd(td, &icd, ret); |
97601024 | 1276 | ios_completed(td, &icd); |
581e7141 JA |
1277 | if (icd.error) { |
1278 | td_verror(td, icd.error, "io_u_queued_complete"); | |
1279 | return -1; | |
1280 | } | |
97601024 | 1281 | |
581e7141 JA |
1282 | if (bytes) { |
1283 | bytes[0] += icd.bytes_done[0]; | |
1284 | bytes[1] += icd.bytes_done[1]; | |
1285 | } | |
1286 | ||
1287 | return 0; | |
97601024 | 1288 | } |
7e77dd02 JA |
1289 | |
1290 | /* | |
1291 | * Call when io_u is really queued, to update the submission latency. | |
1292 | */ | |
1293 | void io_u_queued(struct thread_data *td, struct io_u *io_u) | |
1294 | { | |
9520ebb9 JA |
1295 | if (!td->o.disable_slat) { |
1296 | unsigned long slat_time; | |
7e77dd02 | 1297 | |
9520ebb9 | 1298 | slat_time = utime_since(&io_u->start_time, &io_u->issue_time); |
29a90ddb | 1299 | add_slat_sample(td, io_u->ddir, slat_time, io_u->xfer_buflen); |
9520ebb9 | 1300 | } |
7e77dd02 | 1301 | } |
433afcb4 | 1302 | |
5973cafb JA |
1303 | /* |
1304 | * "randomly" fill the buffer contents | |
1305 | */ | |
1306 | void io_u_fill_buffer(struct thread_data *td, struct io_u *io_u, | |
1307 | unsigned int max_bs) | |
1308 | { | |
637ef8d9 JA |
1309 | if (!td->o.zero_buffers) |
1310 | fill_random_buf(io_u->buf, max_bs); | |
1311 | else | |
1312 | memset(io_u->buf, 0, max_bs); | |
5973cafb | 1313 | } |