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" |
0d29de83 | 11 | #include "trim.h" |
1fbbf72e | 12 | #include "lib/rand.h" |
7ebd796f | 13 | #include "lib/axmap.h" |
002fe734 | 14 | #include "err.h" |
0f38bbef | 15 | #include "lib/pow2.h" |
4b157ac6 | 16 | #include "minmax.h" |
10ba535a | 17 | |
97601024 JA |
18 | struct io_completion_data { |
19 | int nr; /* input */ | |
97601024 JA |
20 | |
21 | int error; /* output */ | |
100f49f1 | 22 | uint64_t bytes_done[DDIR_RWDIR_CNT]; /* output */ |
8b6a404c | 23 | struct timespec time; /* output */ |
97601024 JA |
24 | }; |
25 | ||
10ba535a | 26 | /* |
7ebd796f | 27 | * The ->io_axmap contains a map of blocks we have or have not done io |
10ba535a JA |
28 | * to yet. Used to make sure we cover the entire range in a fair fashion. |
29 | */ | |
e39c0676 | 30 | static bool random_map_free(struct fio_file *f, const uint64_t block) |
10ba535a | 31 | { |
7ebd796f | 32 | return !axmap_isset(f->io_axmap, block); |
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 | { |
956e60ea | 40 | unsigned int min_bs = td->o.min_bs[io_u->ddir]; |
9bf2061e | 41 | struct fio_file *f = io_u->file; |
51ede0b1 | 42 | unsigned int nr_blocks; |
1ae83d45 | 43 | uint64_t block; |
df415585 | 44 | |
1ae83d45 | 45 | block = (io_u->offset - f->file_offset) / (uint64_t) min_bs; |
c685b5b2 | 46 | nr_blocks = (io_u->buflen + min_bs - 1) / min_bs; |
df415585 | 47 | |
2ab9e98b | 48 | if (!(io_u->flags & IO_U_F_BUSY_OK)) |
7ebd796f | 49 | nr_blocks = axmap_set_nr(f->io_axmap, block, nr_blocks); |
df415585 | 50 | |
51ede0b1 JA |
51 | if ((nr_blocks * min_bs) < io_u->buflen) |
52 | io_u->buflen = nr_blocks * min_bs; | |
df415585 JA |
53 | } |
54 | ||
74776733 JA |
55 | static uint64_t last_block(struct thread_data *td, struct fio_file *f, |
56 | enum fio_ddir ddir) | |
2ba1c290 | 57 | { |
74776733 JA |
58 | uint64_t max_blocks; |
59 | uint64_t max_size; | |
2ba1c290 | 60 | |
ff58fced JA |
61 | assert(ddir_rw(ddir)); |
62 | ||
d9dd70f7 JA |
63 | /* |
64 | * Hmm, should we make sure that ->io_size <= ->real_file_size? | |
79591fa9 | 65 | * -> not for now since there is code assuming it could go either. |
d9dd70f7 JA |
66 | */ |
67 | max_size = f->io_size; | |
68 | if (max_size > f->real_file_size) | |
69 | max_size = f->real_file_size; | |
70 | ||
ed335855 SN |
71 | if (td->o.zone_range) |
72 | max_size = td->o.zone_range; | |
73 | ||
0412d12e JE |
74 | if (td->o.min_bs[ddir] > td->o.ba[ddir]) |
75 | max_size -= td->o.min_bs[ddir] - td->o.ba[ddir]; | |
76 | ||
1ae83d45 | 77 | max_blocks = max_size / (uint64_t) td->o.ba[ddir]; |
2ba1c290 JA |
78 | if (!max_blocks) |
79 | return 0; | |
80 | ||
67778e88 | 81 | return max_blocks; |
2ba1c290 JA |
82 | } |
83 | ||
1ae83d45 JA |
84 | struct rand_off { |
85 | struct flist_head list; | |
86 | uint64_t off; | |
87 | }; | |
88 | ||
e25839d4 | 89 | static int __get_next_rand_offset(struct thread_data *td, struct fio_file *f, |
e0a04ac1 JA |
90 | enum fio_ddir ddir, uint64_t *b, |
91 | uint64_t lastb) | |
ec4015da | 92 | { |
6f49f8bc | 93 | uint64_t r; |
5e0baa7f | 94 | |
c3546b53 JA |
95 | if (td->o.random_generator == FIO_RAND_GEN_TAUSWORTHE || |
96 | td->o.random_generator == FIO_RAND_GEN_TAUSWORTHE64) { | |
6f49f8bc | 97 | |
d6b72507 | 98 | r = __rand(&td->random_state); |
8055e41d | 99 | |
4b91ee8f | 100 | dprint(FD_RANDOM, "off rand %llu\n", (unsigned long long) r); |
8055e41d | 101 | |
e0a04ac1 | 102 | *b = lastb * (r / (rand_max(&td->random_state) + 1.0)); |
51ede0b1 | 103 | } else { |
8055e41d | 104 | uint64_t off = 0; |
43c63a78 | 105 | |
967d1b63 JA |
106 | assert(fio_file_lfsr(f)); |
107 | ||
6f49f8bc | 108 | if (lfsr_next(&f->lfsr, &off)) |
8055e41d | 109 | return 1; |
ec4015da | 110 | |
8055e41d JA |
111 | *b = off; |
112 | } | |
0ce8b119 | 113 | |
ec4015da | 114 | /* |
51ede0b1 | 115 | * if we are not maintaining a random map, we are done. |
ec4015da | 116 | */ |
51ede0b1 JA |
117 | if (!file_randommap(td, f)) |
118 | goto ret; | |
43c63a78 JA |
119 | |
120 | /* | |
51ede0b1 | 121 | * calculate map offset and check if it's free |
43c63a78 | 122 | */ |
51ede0b1 JA |
123 | if (random_map_free(f, *b)) |
124 | goto ret; | |
125 | ||
4b91ee8f JA |
126 | dprint(FD_RANDOM, "get_next_rand_offset: offset %llu busy\n", |
127 | (unsigned long long) *b); | |
51ede0b1 | 128 | |
7ebd796f | 129 | *b = axmap_next_free(f->io_axmap, *b); |
51ede0b1 JA |
130 | if (*b == (uint64_t) -1ULL) |
131 | return 1; | |
0ce8b119 JA |
132 | ret: |
133 | return 0; | |
ec4015da JA |
134 | } |
135 | ||
925fee33 JA |
136 | static int __get_next_rand_offset_zipf(struct thread_data *td, |
137 | struct fio_file *f, enum fio_ddir ddir, | |
1ae83d45 | 138 | uint64_t *b) |
e25839d4 | 139 | { |
9c6f6316 | 140 | *b = zipf_next(&f->zipf); |
e25839d4 JA |
141 | return 0; |
142 | } | |
143 | ||
925fee33 JA |
144 | static int __get_next_rand_offset_pareto(struct thread_data *td, |
145 | struct fio_file *f, enum fio_ddir ddir, | |
1ae83d45 | 146 | uint64_t *b) |
925fee33 | 147 | { |
9c6f6316 | 148 | *b = pareto_next(&f->zipf); |
925fee33 JA |
149 | return 0; |
150 | } | |
151 | ||
56d9fa4b JA |
152 | static int __get_next_rand_offset_gauss(struct thread_data *td, |
153 | struct fio_file *f, enum fio_ddir ddir, | |
154 | uint64_t *b) | |
155 | { | |
156 | *b = gauss_next(&f->gauss); | |
157 | return 0; | |
158 | } | |
159 | ||
e0a04ac1 JA |
160 | static int __get_next_rand_offset_zoned(struct thread_data *td, |
161 | struct fio_file *f, enum fio_ddir ddir, | |
162 | uint64_t *b) | |
163 | { | |
164 | unsigned int v, send, stotal; | |
165 | uint64_t offset, lastb; | |
166 | static int warned; | |
167 | struct zone_split_index *zsi; | |
168 | ||
169 | lastb = last_block(td, f, ddir); | |
170 | if (!lastb) | |
171 | return 1; | |
172 | ||
173 | if (!td->o.zone_split_nr[ddir]) { | |
174 | bail: | |
175 | return __get_next_rand_offset(td, f, ddir, b, lastb); | |
176 | } | |
177 | ||
178 | /* | |
179 | * Generate a value, v, between 1 and 100, both inclusive | |
180 | */ | |
36dd3379 | 181 | v = rand32_between(&td->zone_state, 1, 100); |
e0a04ac1 JA |
182 | |
183 | zsi = &td->zone_state_index[ddir][v - 1]; | |
184 | stotal = zsi->size_perc_prev; | |
185 | send = zsi->size_perc; | |
186 | ||
187 | /* | |
188 | * Should never happen | |
189 | */ | |
190 | if (send == -1U) { | |
191 | if (!warned) { | |
192 | log_err("fio: bug in zoned generation\n"); | |
193 | warned = 1; | |
194 | } | |
195 | goto bail; | |
196 | } | |
197 | ||
198 | /* | |
199 | * 'send' is some percentage below or equal to 100 that | |
200 | * marks the end of the current IO range. 'stotal' marks | |
201 | * the start, in percent. | |
202 | */ | |
203 | if (stotal) | |
204 | offset = stotal * lastb / 100ULL; | |
205 | else | |
206 | offset = 0; | |
207 | ||
208 | lastb = lastb * (send - stotal) / 100ULL; | |
209 | ||
210 | /* | |
211 | * Generate index from 0..send-of-lastb | |
212 | */ | |
213 | if (__get_next_rand_offset(td, f, ddir, b, lastb) == 1) | |
214 | return 1; | |
215 | ||
216 | /* | |
217 | * Add our start offset, if any | |
218 | */ | |
219 | if (offset) | |
220 | *b += offset; | |
221 | ||
222 | return 0; | |
223 | } | |
56d9fa4b | 224 | |
1ae83d45 JA |
225 | static int flist_cmp(void *data, struct flist_head *a, struct flist_head *b) |
226 | { | |
227 | struct rand_off *r1 = flist_entry(a, struct rand_off, list); | |
228 | struct rand_off *r2 = flist_entry(b, struct rand_off, list); | |
229 | ||
230 | return r1->off - r2->off; | |
231 | } | |
232 | ||
233 | static int get_off_from_method(struct thread_data *td, struct fio_file *f, | |
234 | enum fio_ddir ddir, uint64_t *b) | |
e25839d4 | 235 | { |
e0a04ac1 JA |
236 | if (td->o.random_distribution == FIO_RAND_DIST_RANDOM) { |
237 | uint64_t lastb; | |
238 | ||
239 | lastb = last_block(td, f, ddir); | |
240 | if (!lastb) | |
241 | return 1; | |
242 | ||
243 | return __get_next_rand_offset(td, f, ddir, b, lastb); | |
244 | } else if (td->o.random_distribution == FIO_RAND_DIST_ZIPF) | |
e25839d4 | 245 | return __get_next_rand_offset_zipf(td, f, ddir, b); |
925fee33 JA |
246 | else if (td->o.random_distribution == FIO_RAND_DIST_PARETO) |
247 | return __get_next_rand_offset_pareto(td, f, ddir, b); | |
56d9fa4b JA |
248 | else if (td->o.random_distribution == FIO_RAND_DIST_GAUSS) |
249 | return __get_next_rand_offset_gauss(td, f, ddir, b); | |
e0a04ac1 JA |
250 | else if (td->o.random_distribution == FIO_RAND_DIST_ZONED) |
251 | return __get_next_rand_offset_zoned(td, f, ddir, b); | |
e25839d4 JA |
252 | |
253 | log_err("fio: unknown random distribution: %d\n", td->o.random_distribution); | |
254 | return 1; | |
255 | } | |
256 | ||
bcd5abfa JA |
257 | /* |
258 | * Sort the reads for a verify phase in batches of verifysort_nr, if | |
259 | * specified. | |
260 | */ | |
e39c0676 | 261 | static inline bool should_sort_io(struct thread_data *td) |
bcd5abfa JA |
262 | { |
263 | if (!td->o.verifysort_nr || !td->o.do_verify) | |
e39c0676 | 264 | return false; |
bcd5abfa | 265 | if (!td_random(td)) |
e39c0676 | 266 | return false; |
bcd5abfa | 267 | if (td->runstate != TD_VERIFYING) |
e39c0676 | 268 | return false; |
c3546b53 JA |
269 | if (td->o.random_generator == FIO_RAND_GEN_TAUSWORTHE || |
270 | td->o.random_generator == FIO_RAND_GEN_TAUSWORTHE64) | |
e39c0676 | 271 | return false; |
bcd5abfa | 272 | |
e39c0676 | 273 | return true; |
bcd5abfa JA |
274 | } |
275 | ||
e39c0676 | 276 | static bool should_do_random(struct thread_data *td, enum fio_ddir ddir) |
211c9b89 JA |
277 | { |
278 | unsigned int v; | |
211c9b89 | 279 | |
d9472271 | 280 | if (td->o.perc_rand[ddir] == 100) |
e39c0676 | 281 | return true; |
211c9b89 | 282 | |
36dd3379 | 283 | v = rand32_between(&td->seq_rand_state[ddir], 1, 100); |
211c9b89 | 284 | |
d9472271 | 285 | return v <= td->o.perc_rand[ddir]; |
211c9b89 JA |
286 | } |
287 | ||
1ae83d45 JA |
288 | static int get_next_rand_offset(struct thread_data *td, struct fio_file *f, |
289 | enum fio_ddir ddir, uint64_t *b) | |
290 | { | |
291 | struct rand_off *r; | |
292 | int i, ret = 1; | |
293 | ||
bcd5abfa | 294 | if (!should_sort_io(td)) |
1ae83d45 JA |
295 | return get_off_from_method(td, f, ddir, b); |
296 | ||
297 | if (!flist_empty(&td->next_rand_list)) { | |
1ae83d45 | 298 | fetch: |
9342d5f8 | 299 | r = flist_first_entry(&td->next_rand_list, struct rand_off, list); |
1ae83d45 JA |
300 | flist_del(&r->list); |
301 | *b = r->off; | |
302 | free(r); | |
303 | return 0; | |
304 | } | |
305 | ||
306 | for (i = 0; i < td->o.verifysort_nr; i++) { | |
307 | r = malloc(sizeof(*r)); | |
308 | ||
309 | ret = get_off_from_method(td, f, ddir, &r->off); | |
310 | if (ret) { | |
311 | free(r); | |
312 | break; | |
313 | } | |
314 | ||
315 | flist_add(&r->list, &td->next_rand_list); | |
316 | } | |
317 | ||
318 | if (ret && !i) | |
319 | return ret; | |
320 | ||
321 | assert(!flist_empty(&td->next_rand_list)); | |
322 | flist_sort(NULL, &td->next_rand_list, flist_cmp); | |
323 | goto fetch; | |
324 | } | |
325 | ||
0bcf41cd JA |
326 | static void loop_cache_invalidate(struct thread_data *td, struct fio_file *f) |
327 | { | |
328 | struct thread_options *o = &td->o; | |
329 | ||
330 | if (o->invalidate_cache && !o->odirect) { | |
331 | int fio_unused ret; | |
332 | ||
333 | ret = file_invalidate_cache(td, f); | |
334 | } | |
335 | } | |
336 | ||
38dad62d | 337 | static int get_next_rand_block(struct thread_data *td, struct fio_file *f, |
1ae83d45 | 338 | enum fio_ddir ddir, uint64_t *b) |
38dad62d | 339 | { |
c04e4661 DE |
340 | if (!get_next_rand_offset(td, f, ddir, b)) |
341 | return 0; | |
342 | ||
8c07860d JA |
343 | if (td->o.time_based || |
344 | (td->o.file_service_type & __FIO_FSERVICE_NONUNIFORM)) { | |
33c48814 | 345 | fio_file_reset(td, f); |
c04e4661 DE |
346 | if (!get_next_rand_offset(td, f, ddir, b)) |
347 | return 0; | |
0bcf41cd | 348 | loop_cache_invalidate(td, f); |
38dad62d JA |
349 | } |
350 | ||
c04e4661 | 351 | dprint(FD_IO, "%s: rand offset failed, last=%llu, size=%llu\n", |
f1dfb668 | 352 | f->file_name, (unsigned long long) f->last_pos[ddir], |
4b91ee8f | 353 | (unsigned long long) f->real_file_size); |
c04e4661 | 354 | return 1; |
38dad62d JA |
355 | } |
356 | ||
37cf9e3c | 357 | static int get_next_seq_offset(struct thread_data *td, struct fio_file *f, |
1ae83d45 | 358 | enum fio_ddir ddir, uint64_t *offset) |
38dad62d | 359 | { |
ac002339 JA |
360 | struct thread_options *o = &td->o; |
361 | ||
ff58fced JA |
362 | assert(ddir_rw(ddir)); |
363 | ||
f1dfb668 | 364 | if (f->last_pos[ddir] >= f->io_size + get_start_offset(td, f) && |
19ddc35b JA |
365 | o->time_based) { |
366 | struct thread_options *o = &td->o; | |
367 | uint64_t io_size = f->io_size + (f->io_size % o->min_bs[ddir]); | |
368 | ||
369 | if (io_size > f->last_pos[ddir]) | |
370 | f->last_pos[ddir] = 0; | |
371 | else | |
372 | f->last_pos[ddir] = f->last_pos[ddir] - io_size; | |
0bcf41cd JA |
373 | |
374 | loop_cache_invalidate(td, f); | |
19ddc35b | 375 | } |
c04e4661 | 376 | |
f1dfb668 | 377 | if (f->last_pos[ddir] < f->real_file_size) { |
1ae83d45 | 378 | uint64_t pos; |
059b0802 | 379 | |
c22825bb JA |
380 | if (f->last_pos[ddir] == f->file_offset && o->ddir_seq_add < 0) { |
381 | if (f->real_file_size > f->io_size) | |
382 | f->last_pos[ddir] = f->io_size; | |
383 | else | |
384 | f->last_pos[ddir] = f->real_file_size; | |
385 | } | |
a66da7a2 | 386 | |
f1dfb668 | 387 | pos = f->last_pos[ddir] - f->file_offset; |
ac002339 JA |
388 | if (pos && o->ddir_seq_add) { |
389 | pos += o->ddir_seq_add; | |
390 | ||
391 | /* | |
392 | * If we reach beyond the end of the file | |
393 | * with holed IO, wrap around to the | |
b0a84f48 JA |
394 | * beginning again. If we're doing backwards IO, |
395 | * wrap to the end. | |
ac002339 | 396 | */ |
b0a84f48 JA |
397 | if (pos >= f->real_file_size) { |
398 | if (o->ddir_seq_add > 0) | |
399 | pos = f->file_offset; | |
c22825bb JA |
400 | else { |
401 | if (f->real_file_size > f->io_size) | |
402 | pos = f->io_size; | |
403 | else | |
404 | pos = f->real_file_size; | |
405 | ||
406 | pos += o->ddir_seq_add; | |
407 | } | |
b0a84f48 | 408 | } |
ac002339 | 409 | } |
059b0802 | 410 | |
37cf9e3c | 411 | *offset = pos; |
38dad62d JA |
412 | return 0; |
413 | } | |
414 | ||
415 | return 1; | |
416 | } | |
417 | ||
418 | static int get_next_block(struct thread_data *td, struct io_u *io_u, | |
6aca9b3d JA |
419 | enum fio_ddir ddir, int rw_seq, |
420 | unsigned int *is_random) | |
38dad62d JA |
421 | { |
422 | struct fio_file *f = io_u->file; | |
1ae83d45 | 423 | uint64_t b, offset; |
38dad62d JA |
424 | int ret; |
425 | ||
ff58fced JA |
426 | assert(ddir_rw(ddir)); |
427 | ||
37cf9e3c JA |
428 | b = offset = -1ULL; |
429 | ||
38dad62d | 430 | if (rw_seq) { |
211c9b89 | 431 | if (td_random(td)) { |
6aca9b3d | 432 | if (should_do_random(td, ddir)) { |
211c9b89 | 433 | ret = get_next_rand_block(td, f, ddir, &b); |
6aca9b3d JA |
434 | *is_random = 1; |
435 | } else { | |
436 | *is_random = 0; | |
1651e431 | 437 | io_u_set(td, io_u, IO_U_F_BUSY_OK); |
211c9b89 JA |
438 | ret = get_next_seq_offset(td, f, ddir, &offset); |
439 | if (ret) | |
440 | ret = get_next_rand_block(td, f, ddir, &b); | |
441 | } | |
6aca9b3d JA |
442 | } else { |
443 | *is_random = 0; | |
37cf9e3c | 444 | ret = get_next_seq_offset(td, f, ddir, &offset); |
6aca9b3d | 445 | } |
38dad62d | 446 | } else { |
1651e431 | 447 | io_u_set(td, io_u, IO_U_F_BUSY_OK); |
6aca9b3d | 448 | *is_random = 0; |
38dad62d JA |
449 | |
450 | if (td->o.rw_seq == RW_SEQ_SEQ) { | |
37cf9e3c | 451 | ret = get_next_seq_offset(td, f, ddir, &offset); |
6aca9b3d | 452 | if (ret) { |
37cf9e3c | 453 | ret = get_next_rand_block(td, f, ddir, &b); |
6aca9b3d JA |
454 | *is_random = 0; |
455 | } | |
38dad62d | 456 | } else if (td->o.rw_seq == RW_SEQ_IDENT) { |
f1dfb668 JA |
457 | if (f->last_start[ddir] != -1ULL) |
458 | offset = f->last_start[ddir] - f->file_offset; | |
38dad62d | 459 | else |
37cf9e3c | 460 | offset = 0; |
38dad62d JA |
461 | ret = 0; |
462 | } else { | |
463 | log_err("fio: unknown rw_seq=%d\n", td->o.rw_seq); | |
464 | ret = 1; | |
465 | } | |
466 | } | |
6d68b997 | 467 | |
37cf9e3c JA |
468 | if (!ret) { |
469 | if (offset != -1ULL) | |
470 | io_u->offset = offset; | |
471 | else if (b != -1ULL) | |
472 | io_u->offset = b * td->o.ba[ddir]; | |
473 | else { | |
4e0a8fa2 | 474 | log_err("fio: bug in offset generation: offset=%llu, b=%llu\n", (unsigned long long) offset, (unsigned long long) b); |
37cf9e3c JA |
475 | ret = 1; |
476 | } | |
477 | } | |
478 | ||
38dad62d JA |
479 | return ret; |
480 | } | |
481 | ||
10ba535a JA |
482 | /* |
483 | * For random io, generate a random new block and see if it's used. Repeat | |
484 | * until we find a free one. For sequential io, just return the end of | |
485 | * the last io issued. | |
486 | */ | |
6aca9b3d JA |
487 | static int __get_next_offset(struct thread_data *td, struct io_u *io_u, |
488 | unsigned int *is_random) | |
10ba535a | 489 | { |
9bf2061e | 490 | struct fio_file *f = io_u->file; |
4ba66134 | 491 | enum fio_ddir ddir = io_u->ddir; |
38dad62d | 492 | int rw_seq_hit = 0; |
10ba535a | 493 | |
ff58fced JA |
494 | assert(ddir_rw(ddir)); |
495 | ||
38dad62d JA |
496 | if (td->o.ddir_seq_nr && !--td->ddir_seq_nr) { |
497 | rw_seq_hit = 1; | |
5736c10d | 498 | td->ddir_seq_nr = td->o.ddir_seq_nr; |
38dad62d | 499 | } |
211097b2 | 500 | |
6aca9b3d | 501 | if (get_next_block(td, io_u, ddir, rw_seq_hit, is_random)) |
38dad62d | 502 | return 1; |
10ba535a | 503 | |
009bd847 JA |
504 | if (io_u->offset >= f->io_size) { |
505 | dprint(FD_IO, "get_next_offset: offset %llu >= io_size %llu\n", | |
4b91ee8f JA |
506 | (unsigned long long) io_u->offset, |
507 | (unsigned long long) f->io_size); | |
009bd847 JA |
508 | return 1; |
509 | } | |
510 | ||
511 | io_u->offset += f->file_offset; | |
2ba1c290 JA |
512 | if (io_u->offset >= f->real_file_size) { |
513 | dprint(FD_IO, "get_next_offset: offset %llu >= size %llu\n", | |
4b91ee8f JA |
514 | (unsigned long long) io_u->offset, |
515 | (unsigned long long) f->real_file_size); | |
10ba535a | 516 | return 1; |
2ba1c290 | 517 | } |
10ba535a JA |
518 | |
519 | return 0; | |
520 | } | |
521 | ||
6aca9b3d JA |
522 | static int get_next_offset(struct thread_data *td, struct io_u *io_u, |
523 | unsigned int *is_random) | |
15dc1934 | 524 | { |
d72be545 JA |
525 | if (td->flags & TD_F_PROFILE_OPS) { |
526 | struct prof_io_ops *ops = &td->prof_io_ops; | |
7eb36574 | 527 | |
d72be545 | 528 | if (ops->fill_io_u_off) |
6aca9b3d | 529 | return ops->fill_io_u_off(td, io_u, is_random); |
d72be545 | 530 | } |
15dc1934 | 531 | |
6aca9b3d | 532 | return __get_next_offset(td, io_u, is_random); |
15dc1934 JA |
533 | } |
534 | ||
e39c0676 JA |
535 | static inline bool io_u_fits(struct thread_data *td, struct io_u *io_u, |
536 | unsigned int buflen) | |
79944128 JA |
537 | { |
538 | struct fio_file *f = io_u->file; | |
539 | ||
bedc9dc2 | 540 | return io_u->offset + buflen <= f->io_size + get_start_offset(td, f); |
79944128 JA |
541 | } |
542 | ||
6aca9b3d JA |
543 | static unsigned int __get_next_buflen(struct thread_data *td, struct io_u *io_u, |
544 | unsigned int is_random) | |
10ba535a | 545 | { |
6aca9b3d | 546 | int ddir = io_u->ddir; |
24d23ca7 | 547 | unsigned int buflen = 0; |
f3059de1 | 548 | unsigned int minbs, maxbs; |
3dd29f7c | 549 | uint64_t frand_max, r; |
7c961359 | 550 | bool power_2; |
10ba535a | 551 | |
9ee1c647 | 552 | assert(ddir_rw(ddir)); |
6aca9b3d JA |
553 | |
554 | if (td->o.bs_is_seq_rand) | |
555 | ddir = is_random ? DDIR_WRITE: DDIR_READ; | |
ff58fced | 556 | |
f3059de1 JA |
557 | minbs = td->o.min_bs[ddir]; |
558 | maxbs = td->o.max_bs[ddir]; | |
559 | ||
79944128 JA |
560 | if (minbs == maxbs) |
561 | return minbs; | |
562 | ||
52c58027 JA |
563 | /* |
564 | * If we can't satisfy the min block size from here, then fail | |
565 | */ | |
566 | if (!io_u_fits(td, io_u, minbs)) | |
567 | return 0; | |
568 | ||
2f282cec | 569 | frand_max = rand_max(&td->bsrange_state[ddir]); |
79944128 | 570 | do { |
2f282cec | 571 | r = __rand(&td->bsrange_state[ddir]); |
4c07ad86 | 572 | |
720e84ad | 573 | if (!td->o.bssplit_nr[ddir]) { |
f3059de1 | 574 | buflen = 1 + (unsigned int) ((double) maxbs * |
c3546b53 | 575 | (r / (frand_max + 1.0))); |
f3059de1 JA |
576 | if (buflen < minbs) |
577 | buflen = minbs; | |
5ec10eaa | 578 | } else { |
3dd29f7c | 579 | long long perc = 0; |
564ca972 JA |
580 | unsigned int i; |
581 | ||
720e84ad JA |
582 | for (i = 0; i < td->o.bssplit_nr[ddir]; i++) { |
583 | struct bssplit *bsp = &td->o.bssplit[ddir][i]; | |
564ca972 JA |
584 | |
585 | buflen = bsp->bs; | |
586 | perc += bsp->perc; | |
3dd29f7c JA |
587 | if (!perc) |
588 | break; | |
589 | if ((r / perc <= frand_max / 100ULL) && | |
79944128 | 590 | io_u_fits(td, io_u, buflen)) |
564ca972 JA |
591 | break; |
592 | } | |
593 | } | |
79944128 | 594 | |
17a6b702 PL |
595 | power_2 = is_power_of_2(minbs); |
596 | if (!td->o.bs_unaligned && power_2) | |
7c306bb1 | 597 | buflen &= ~(minbs - 1); |
17a6b702 PL |
598 | else if (!td->o.bs_unaligned && !power_2) |
599 | buflen -= buflen % minbs; | |
79944128 | 600 | } while (!io_u_fits(td, io_u, buflen)); |
6a5e6884 | 601 | |
10ba535a JA |
602 | return buflen; |
603 | } | |
604 | ||
6aca9b3d JA |
605 | static unsigned int get_next_buflen(struct thread_data *td, struct io_u *io_u, |
606 | unsigned int is_random) | |
15dc1934 | 607 | { |
d72be545 JA |
608 | if (td->flags & TD_F_PROFILE_OPS) { |
609 | struct prof_io_ops *ops = &td->prof_io_ops; | |
7eb36574 | 610 | |
d72be545 | 611 | if (ops->fill_io_u_size) |
6aca9b3d | 612 | return ops->fill_io_u_size(td, io_u, is_random); |
d72be545 | 613 | } |
15dc1934 | 614 | |
6aca9b3d | 615 | return __get_next_buflen(td, io_u, is_random); |
15dc1934 JA |
616 | } |
617 | ||
afe24a5a JA |
618 | static void set_rwmix_bytes(struct thread_data *td) |
619 | { | |
afe24a5a JA |
620 | unsigned int diff; |
621 | ||
622 | /* | |
623 | * we do time or byte based switch. this is needed because | |
624 | * buffered writes may issue a lot quicker than they complete, | |
625 | * whereas reads do not. | |
626 | */ | |
e47f799f | 627 | diff = td->o.rwmix[td->rwmix_ddir ^ 1]; |
04c540d9 | 628 | td->rwmix_issues = (td->io_issues[td->rwmix_ddir] * diff) / 100; |
e47f799f JA |
629 | } |
630 | ||
631 | static inline enum fio_ddir get_rand_ddir(struct thread_data *td) | |
632 | { | |
633 | unsigned int v; | |
e47f799f | 634 | |
36dd3379 | 635 | v = rand32_between(&td->rwmix_state, 1, 100); |
4c07ad86 | 636 | |
04c540d9 | 637 | if (v <= td->o.rwmix[DDIR_READ]) |
e47f799f JA |
638 | return DDIR_READ; |
639 | ||
640 | return DDIR_WRITE; | |
afe24a5a JA |
641 | } |
642 | ||
0d593542 | 643 | int io_u_quiesce(struct thread_data *td) |
002e7183 | 644 | { |
0d593542 JA |
645 | int completed = 0; |
646 | ||
002e7183 JA |
647 | /* |
648 | * We are going to sleep, ensure that we flush anything pending as | |
649 | * not to skew our latency numbers. | |
650 | * | |
651 | * Changed to only monitor 'in flight' requests here instead of the | |
652 | * td->cur_depth, b/c td->cur_depth does not accurately represent | |
653 | * io's that have been actually submitted to an async engine, | |
654 | * and cur_depth is meaningless for sync engines. | |
655 | */ | |
9cc80b6d JA |
656 | if (td->io_u_queued || td->cur_depth) { |
657 | int fio_unused ret; | |
658 | ||
659 | ret = td_io_commit(td); | |
660 | } | |
661 | ||
002e7183 | 662 | while (td->io_u_in_flight) { |
89fd0573 | 663 | int ret; |
002e7183 | 664 | |
55312f9f | 665 | ret = io_u_queued_complete(td, 1); |
0d593542 JA |
666 | if (ret > 0) |
667 | completed += ret; | |
002e7183 | 668 | } |
0d593542 | 669 | |
6be06c46 JA |
670 | if (td->flags & TD_F_REGROW_LOGS) |
671 | regrow_logs(td); | |
672 | ||
0d593542 | 673 | return completed; |
002e7183 JA |
674 | } |
675 | ||
581e7141 JA |
676 | static enum fio_ddir rate_ddir(struct thread_data *td, enum fio_ddir ddir) |
677 | { | |
678 | enum fio_ddir odir = ddir ^ 1; | |
b5407f8b | 679 | uint64_t usec; |
90eff1c9 | 680 | uint64_t now; |
581e7141 | 681 | |
ff58fced | 682 | assert(ddir_rw(ddir)); |
50a8ce86 | 683 | now = utime_since_now(&td->start); |
ff58fced | 684 | |
50a8ce86 D |
685 | /* |
686 | * if rate_next_io_time is in the past, need to catch up to rate | |
687 | */ | |
688 | if (td->rate_next_io_time[ddir] <= now) | |
581e7141 JA |
689 | return ddir; |
690 | ||
691 | /* | |
50a8ce86 | 692 | * We are ahead of rate in this direction. See if we |
581e7141 JA |
693 | * should switch. |
694 | */ | |
315fcfec | 695 | if (td_rw(td) && td->o.rwmix[odir]) { |
581e7141 | 696 | /* |
50a8ce86 | 697 | * Other direction is behind rate, switch |
581e7141 | 698 | */ |
50a8ce86 | 699 | if (td->rate_next_io_time[odir] <= now) |
581e7141 JA |
700 | return odir; |
701 | ||
702 | /* | |
50a8ce86 D |
703 | * Both directions are ahead of rate. sleep the min |
704 | * switch if necissary | |
581e7141 | 705 | */ |
50a8ce86 D |
706 | if (td->rate_next_io_time[ddir] <= |
707 | td->rate_next_io_time[odir]) { | |
708 | usec = td->rate_next_io_time[ddir] - now; | |
581e7141 | 709 | } else { |
50a8ce86 | 710 | usec = td->rate_next_io_time[odir] - now; |
581e7141 JA |
711 | ddir = odir; |
712 | } | |
713 | } else | |
50a8ce86 | 714 | usec = td->rate_next_io_time[ddir] - now; |
581e7141 | 715 | |
a9da8ab2 JA |
716 | if (td->o.io_submit_mode == IO_MODE_INLINE) |
717 | io_u_quiesce(td); | |
78c1eda5 | 718 | |
9cc80b6d | 719 | usec = usec_sleep(td, usec); |
581e7141 | 720 | |
581e7141 JA |
721 | return ddir; |
722 | } | |
723 | ||
10ba535a JA |
724 | /* |
725 | * Return the data direction for the next io_u. If the job is a | |
726 | * mixed read/write workload, check the rwmix cycle and switch if | |
727 | * necessary. | |
728 | */ | |
1e97cce9 | 729 | static enum fio_ddir get_rw_ddir(struct thread_data *td) |
10ba535a | 730 | { |
581e7141 JA |
731 | enum fio_ddir ddir; |
732 | ||
5f9099ea | 733 | /* |
f6860972 TK |
734 | * See if it's time to fsync/fdatasync/sync_file_range first, |
735 | * and if not then move on to check regular I/Os. | |
5f9099ea | 736 | */ |
f6860972 TK |
737 | if (should_fsync(td)) { |
738 | if (td->o.fsync_blocks && td->io_issues[DDIR_WRITE] && | |
739 | !(td->io_issues[DDIR_WRITE] % td->o.fsync_blocks)) | |
740 | return DDIR_SYNC; | |
741 | ||
742 | if (td->o.fdatasync_blocks && td->io_issues[DDIR_WRITE] && | |
743 | !(td->io_issues[DDIR_WRITE] % td->o.fdatasync_blocks)) | |
744 | return DDIR_DATASYNC; | |
745 | ||
746 | if (td->sync_file_range_nr && td->io_issues[DDIR_WRITE] && | |
747 | !(td->io_issues[DDIR_WRITE] % td->sync_file_range_nr)) | |
748 | return DDIR_SYNC_FILE_RANGE; | |
749 | } | |
44f29692 | 750 | |
10ba535a | 751 | if (td_rw(td)) { |
10ba535a JA |
752 | /* |
753 | * Check if it's time to seed a new data direction. | |
754 | */ | |
e4928662 | 755 | if (td->io_issues[td->rwmix_ddir] >= td->rwmix_issues) { |
e47f799f JA |
756 | /* |
757 | * Put a top limit on how many bytes we do for | |
758 | * one data direction, to avoid overflowing the | |
759 | * ranges too much | |
760 | */ | |
761 | ddir = get_rand_ddir(td); | |
e47f799f JA |
762 | |
763 | if (ddir != td->rwmix_ddir) | |
764 | set_rwmix_bytes(td); | |
765 | ||
766 | td->rwmix_ddir = ddir; | |
10ba535a | 767 | } |
581e7141 | 768 | ddir = td->rwmix_ddir; |
10ba535a | 769 | } else if (td_read(td)) |
581e7141 | 770 | ddir = DDIR_READ; |
6eaf09d6 | 771 | else if (td_write(td)) |
581e7141 | 772 | ddir = DDIR_WRITE; |
5c8e84ca | 773 | else if (td_trim(td)) |
6eaf09d6 | 774 | ddir = DDIR_TRIM; |
5c8e84ca TK |
775 | else |
776 | ddir = DDIR_INVAL; | |
581e7141 JA |
777 | |
778 | td->rwmix_ddir = rate_ddir(td, ddir); | |
779 | return td->rwmix_ddir; | |
10ba535a JA |
780 | } |
781 | ||
1ef2b6be JA |
782 | static void set_rw_ddir(struct thread_data *td, struct io_u *io_u) |
783 | { | |
0e4dd95c DE |
784 | enum fio_ddir ddir = get_rw_ddir(td); |
785 | ||
82a90686 | 786 | if (td_trimwrite(td)) { |
0e4dd95c DE |
787 | struct fio_file *f = io_u->file; |
788 | if (f->last_pos[DDIR_WRITE] == f->last_pos[DDIR_TRIM]) | |
789 | ddir = DDIR_TRIM; | |
790 | else | |
791 | ddir = DDIR_WRITE; | |
792 | } | |
793 | ||
794 | io_u->ddir = io_u->acct_ddir = ddir; | |
1ef2b6be | 795 | |
9b87f09b | 796 | if (io_u->ddir == DDIR_WRITE && td_ioengine_flagged(td, FIO_BARRIER) && |
1ef2b6be JA |
797 | td->o.barrier_blocks && |
798 | !(td->io_issues[DDIR_WRITE] % td->o.barrier_blocks) && | |
799 | td->io_issues[DDIR_WRITE]) | |
1651e431 | 800 | io_u_set(td, io_u, IO_U_F_BARRIER); |
1ef2b6be JA |
801 | } |
802 | ||
e8462bd8 | 803 | void put_file_log(struct thread_data *td, struct fio_file *f) |
60f2c658 | 804 | { |
71b84caa | 805 | unsigned int ret = put_file(td, f); |
60f2c658 JA |
806 | |
807 | if (ret) | |
808 | td_verror(td, ret, "file close"); | |
809 | } | |
810 | ||
10ba535a JA |
811 | void put_io_u(struct thread_data *td, struct io_u *io_u) |
812 | { | |
a9da8ab2 JA |
813 | if (td->parent) |
814 | td = td->parent; | |
815 | ||
e8462bd8 JA |
816 | td_io_u_lock(td); |
817 | ||
f8b0bd10 | 818 | if (io_u->file && !(io_u->flags & IO_U_F_NO_FILE_PUT)) |
60f2c658 | 819 | put_file_log(td, io_u->file); |
f8b0bd10 | 820 | |
10ba535a | 821 | io_u->file = NULL; |
1651e431 | 822 | io_u_set(td, io_u, IO_U_F_FREE); |
d7ee2a7d | 823 | |
a9da8ab2 | 824 | if (io_u->flags & IO_U_F_IN_CUR_DEPTH) { |
0c41214f | 825 | td->cur_depth--; |
a9da8ab2 JA |
826 | assert(!(td->flags & TD_F_CHILD)); |
827 | } | |
2ae0b204 | 828 | io_u_qpush(&td->io_u_freelist, io_u); |
e8462bd8 JA |
829 | td_io_u_unlock(td); |
830 | td_io_u_free_notify(td); | |
10ba535a JA |
831 | } |
832 | ||
f2bba182 RR |
833 | void clear_io_u(struct thread_data *td, struct io_u *io_u) |
834 | { | |
1651e431 | 835 | io_u_clear(td, io_u, IO_U_F_FLIGHT); |
f2bba182 RR |
836 | put_io_u(td, io_u); |
837 | } | |
838 | ||
755200a3 JA |
839 | void requeue_io_u(struct thread_data *td, struct io_u **io_u) |
840 | { | |
841 | struct io_u *__io_u = *io_u; | |
bcd5abfa | 842 | enum fio_ddir ddir = acct_ddir(__io_u); |
755200a3 | 843 | |
465221b0 JA |
844 | dprint(FD_IO, "requeue %p\n", __io_u); |
845 | ||
a9da8ab2 JA |
846 | if (td->parent) |
847 | td = td->parent; | |
848 | ||
e8462bd8 JA |
849 | td_io_u_lock(td); |
850 | ||
1651e431 | 851 | io_u_set(td, __io_u, IO_U_F_FREE); |
bcd5abfa JA |
852 | if ((__io_u->flags & IO_U_F_FLIGHT) && ddir_rw(ddir)) |
853 | td->io_issues[ddir]--; | |
5ec10eaa | 854 | |
1651e431 | 855 | io_u_clear(td, __io_u, IO_U_F_FLIGHT); |
a9da8ab2 | 856 | if (__io_u->flags & IO_U_F_IN_CUR_DEPTH) { |
0c41214f | 857 | td->cur_depth--; |
a9da8ab2 JA |
858 | assert(!(td->flags & TD_F_CHILD)); |
859 | } | |
2ae0b204 JA |
860 | |
861 | io_u_rpush(&td->io_u_requeues, __io_u); | |
e8462bd8 | 862 | td_io_u_unlock(td); |
a9da8ab2 | 863 | td_io_u_free_notify(td); |
755200a3 JA |
864 | *io_u = NULL; |
865 | } | |
866 | ||
9bf2061e | 867 | static int fill_io_u(struct thread_data *td, struct io_u *io_u) |
10ba535a | 868 | { |
6aca9b3d JA |
869 | unsigned int is_random; |
870 | ||
9b87f09b | 871 | if (td_ioengine_flagged(td, FIO_NOIO)) |
b4c5e1ac JA |
872 | goto out; |
873 | ||
1ef2b6be | 874 | set_rw_ddir(td, io_u); |
5f9099ea | 875 | |
87dc1ab1 | 876 | /* |
ff58fced | 877 | * fsync() or fdatasync() or trim etc, we are done |
87dc1ab1 | 878 | */ |
ff58fced | 879 | if (!ddir_rw(io_u->ddir)) |
c38e9468 | 880 | goto out; |
a00735e6 | 881 | |
48f5abd3 JA |
882 | /* |
883 | * See if it's time to switch to a new zone | |
884 | */ | |
13af05ae | 885 | if (td->zone_bytes >= td->o.zone_size && td->o.zone_skip) { |
4c8be5b1 JA |
886 | struct fio_file *f = io_u->file; |
887 | ||
48f5abd3 | 888 | td->zone_bytes = 0; |
4c8be5b1 JA |
889 | f->file_offset += td->o.zone_range + td->o.zone_skip; |
890 | ||
891 | /* | |
892 | * Wrap from the beginning, if we exceed the file size | |
893 | */ | |
894 | if (f->file_offset >= f->real_file_size) | |
895 | f->file_offset = f->real_file_size - f->file_offset; | |
f1dfb668 | 896 | f->last_pos[io_u->ddir] = f->file_offset; |
48f5abd3 JA |
897 | td->io_skip_bytes += td->o.zone_skip; |
898 | } | |
899 | ||
10ba535a | 900 | /* |
c685b5b2 JA |
901 | * No log, let the seq/rand engine retrieve the next buflen and |
902 | * position. | |
10ba535a | 903 | */ |
6aca9b3d | 904 | if (get_next_offset(td, io_u, &is_random)) { |
2ba1c290 | 905 | dprint(FD_IO, "io_u %p, failed getting offset\n", io_u); |
bca4ed4d | 906 | return 1; |
2ba1c290 | 907 | } |
10ba535a | 908 | |
6aca9b3d | 909 | io_u->buflen = get_next_buflen(td, io_u, is_random); |
2ba1c290 JA |
910 | if (!io_u->buflen) { |
911 | dprint(FD_IO, "io_u %p, failed getting buflen\n", io_u); | |
bca4ed4d | 912 | return 1; |
2ba1c290 | 913 | } |
bca4ed4d | 914 | |
2ba1c290 | 915 | if (io_u->offset + io_u->buflen > io_u->file->real_file_size) { |
48366f3a TK |
916 | dprint(FD_IO, "io_u %p, offset + buflen exceeds file size\n", |
917 | io_u); | |
918 | dprint(FD_IO, " offset=%llu/buflen=%lu > %llu\n", | |
4b91ee8f JA |
919 | (unsigned long long) io_u->offset, io_u->buflen, |
920 | (unsigned long long) io_u->file->real_file_size); | |
6a5e6884 | 921 | return 1; |
2ba1c290 | 922 | } |
6a5e6884 | 923 | |
bca4ed4d JA |
924 | /* |
925 | * mark entry before potentially trimming io_u | |
926 | */ | |
303032ae | 927 | if (td_random(td) && file_randommap(td, io_u->file)) |
9bf2061e | 928 | mark_random_map(td, io_u); |
bca4ed4d | 929 | |
c38e9468 | 930 | out: |
2ba1c290 | 931 | dprint_io_u(io_u, "fill_io_u"); |
d9d91e39 | 932 | td->zone_bytes += io_u->buflen; |
bca4ed4d | 933 | return 0; |
10ba535a JA |
934 | } |
935 | ||
838bc709 JA |
936 | static void __io_u_mark_map(unsigned int *map, unsigned int nr) |
937 | { | |
2b13e716 | 938 | int idx = 0; |
838bc709 JA |
939 | |
940 | switch (nr) { | |
941 | default: | |
2b13e716 | 942 | idx = 6; |
838bc709 JA |
943 | break; |
944 | case 33 ... 64: | |
2b13e716 | 945 | idx = 5; |
838bc709 JA |
946 | break; |
947 | case 17 ... 32: | |
2b13e716 | 948 | idx = 4; |
838bc709 JA |
949 | break; |
950 | case 9 ... 16: | |
2b13e716 | 951 | idx = 3; |
838bc709 JA |
952 | break; |
953 | case 5 ... 8: | |
2b13e716 | 954 | idx = 2; |
838bc709 JA |
955 | break; |
956 | case 1 ... 4: | |
2b13e716 | 957 | idx = 1; |
838bc709 JA |
958 | case 0: |
959 | break; | |
960 | } | |
961 | ||
2b13e716 | 962 | map[idx]++; |
838bc709 JA |
963 | } |
964 | ||
965 | void io_u_mark_submit(struct thread_data *td, unsigned int nr) | |
966 | { | |
967 | __io_u_mark_map(td->ts.io_u_submit, nr); | |
968 | td->ts.total_submit++; | |
969 | } | |
970 | ||
971 | void io_u_mark_complete(struct thread_data *td, unsigned int nr) | |
972 | { | |
973 | __io_u_mark_map(td->ts.io_u_complete, nr); | |
974 | td->ts.total_complete++; | |
975 | } | |
976 | ||
d8005759 | 977 | void io_u_mark_depth(struct thread_data *td, unsigned int nr) |
71619dc2 | 978 | { |
2b13e716 | 979 | int idx = 0; |
71619dc2 JA |
980 | |
981 | switch (td->cur_depth) { | |
982 | default: | |
2b13e716 | 983 | idx = 6; |
a783e61a | 984 | break; |
71619dc2 | 985 | case 32 ... 63: |
2b13e716 | 986 | idx = 5; |
a783e61a | 987 | break; |
71619dc2 | 988 | case 16 ... 31: |
2b13e716 | 989 | idx = 4; |
a783e61a | 990 | break; |
71619dc2 | 991 | case 8 ... 15: |
2b13e716 | 992 | idx = 3; |
a783e61a | 993 | break; |
71619dc2 | 994 | case 4 ... 7: |
2b13e716 | 995 | idx = 2; |
a783e61a | 996 | break; |
71619dc2 | 997 | case 2 ... 3: |
2b13e716 | 998 | idx = 1; |
71619dc2 JA |
999 | case 1: |
1000 | break; | |
1001 | } | |
1002 | ||
2b13e716 | 1003 | td->ts.io_u_map[idx] += nr; |
71619dc2 JA |
1004 | } |
1005 | ||
d6bb626e | 1006 | static void io_u_mark_lat_nsec(struct thread_data *td, unsigned long long nsec) |
04a0feae | 1007 | { |
2b13e716 | 1008 | int idx = 0; |
04a0feae | 1009 | |
d6bb626e VF |
1010 | assert(nsec < 1000); |
1011 | ||
1012 | switch (nsec) { | |
1013 | case 750 ... 999: | |
1014 | idx = 9; | |
1015 | break; | |
1016 | case 500 ... 749: | |
1017 | idx = 8; | |
1018 | break; | |
1019 | case 250 ... 499: | |
1020 | idx = 7; | |
1021 | break; | |
1022 | case 100 ... 249: | |
1023 | idx = 6; | |
1024 | break; | |
1025 | case 50 ... 99: | |
1026 | idx = 5; | |
1027 | break; | |
1028 | case 20 ... 49: | |
1029 | idx = 4; | |
1030 | break; | |
1031 | case 10 ... 19: | |
1032 | idx = 3; | |
1033 | break; | |
1034 | case 4 ... 9: | |
1035 | idx = 2; | |
1036 | break; | |
1037 | case 2 ... 3: | |
1038 | idx = 1; | |
1039 | case 0 ... 1: | |
1040 | break; | |
1041 | } | |
1042 | ||
1043 | assert(idx < FIO_IO_U_LAT_N_NR); | |
1044 | td->ts.io_u_lat_n[idx]++; | |
1045 | } | |
1046 | ||
1047 | static void io_u_mark_lat_usec(struct thread_data *td, unsigned long long usec) | |
1048 | { | |
1049 | int idx = 0; | |
1050 | ||
1051 | assert(usec < 1000 && usec >= 1); | |
04a0feae JA |
1052 | |
1053 | switch (usec) { | |
1054 | case 750 ... 999: | |
2b13e716 | 1055 | idx = 9; |
04a0feae JA |
1056 | break; |
1057 | case 500 ... 749: | |
2b13e716 | 1058 | idx = 8; |
04a0feae JA |
1059 | break; |
1060 | case 250 ... 499: | |
2b13e716 | 1061 | idx = 7; |
04a0feae JA |
1062 | break; |
1063 | case 100 ... 249: | |
2b13e716 | 1064 | idx = 6; |
04a0feae JA |
1065 | break; |
1066 | case 50 ... 99: | |
2b13e716 | 1067 | idx = 5; |
04a0feae JA |
1068 | break; |
1069 | case 20 ... 49: | |
2b13e716 | 1070 | idx = 4; |
04a0feae JA |
1071 | break; |
1072 | case 10 ... 19: | |
2b13e716 | 1073 | idx = 3; |
04a0feae JA |
1074 | break; |
1075 | case 4 ... 9: | |
2b13e716 | 1076 | idx = 2; |
04a0feae JA |
1077 | break; |
1078 | case 2 ... 3: | |
2b13e716 | 1079 | idx = 1; |
04a0feae JA |
1080 | case 0 ... 1: |
1081 | break; | |
1082 | } | |
1083 | ||
2b13e716 JA |
1084 | assert(idx < FIO_IO_U_LAT_U_NR); |
1085 | td->ts.io_u_lat_u[idx]++; | |
04a0feae JA |
1086 | } |
1087 | ||
d6bb626e | 1088 | static void io_u_mark_lat_msec(struct thread_data *td, unsigned long long msec) |
ec118304 | 1089 | { |
2b13e716 | 1090 | int idx = 0; |
ec118304 | 1091 | |
d6bb626e VF |
1092 | assert(msec >= 1); |
1093 | ||
ec118304 JA |
1094 | switch (msec) { |
1095 | default: | |
2b13e716 | 1096 | idx = 11; |
04a0feae | 1097 | break; |
8abdce66 | 1098 | case 1000 ... 1999: |
2b13e716 | 1099 | idx = 10; |
04a0feae | 1100 | break; |
8abdce66 | 1101 | case 750 ... 999: |
2b13e716 | 1102 | idx = 9; |
04a0feae | 1103 | break; |
8abdce66 | 1104 | case 500 ... 749: |
2b13e716 | 1105 | idx = 8; |
04a0feae | 1106 | break; |
8abdce66 | 1107 | case 250 ... 499: |
2b13e716 | 1108 | idx = 7; |
04a0feae | 1109 | break; |
8abdce66 | 1110 | case 100 ... 249: |
2b13e716 | 1111 | idx = 6; |
04a0feae | 1112 | break; |
8abdce66 | 1113 | case 50 ... 99: |
2b13e716 | 1114 | idx = 5; |
04a0feae | 1115 | break; |
8abdce66 | 1116 | case 20 ... 49: |
2b13e716 | 1117 | idx = 4; |
04a0feae | 1118 | break; |
8abdce66 | 1119 | case 10 ... 19: |
2b13e716 | 1120 | idx = 3; |
04a0feae | 1121 | break; |
8abdce66 | 1122 | case 4 ... 9: |
2b13e716 | 1123 | idx = 2; |
04a0feae | 1124 | break; |
ec118304 | 1125 | case 2 ... 3: |
2b13e716 | 1126 | idx = 1; |
ec118304 JA |
1127 | case 0 ... 1: |
1128 | break; | |
1129 | } | |
1130 | ||
2b13e716 JA |
1131 | assert(idx < FIO_IO_U_LAT_M_NR); |
1132 | td->ts.io_u_lat_m[idx]++; | |
04a0feae JA |
1133 | } |
1134 | ||
d6bb626e | 1135 | static void io_u_mark_latency(struct thread_data *td, unsigned long long nsec) |
04a0feae | 1136 | { |
d6bb626e VF |
1137 | if (nsec < 1000) |
1138 | io_u_mark_lat_nsec(td, nsec); | |
1139 | else if (nsec < 1000000) | |
1140 | io_u_mark_lat_usec(td, nsec / 1000); | |
04a0feae | 1141 | else |
d6bb626e | 1142 | io_u_mark_lat_msec(td, nsec / 1000000); |
ec118304 JA |
1143 | } |
1144 | ||
8c07860d JA |
1145 | static unsigned int __get_next_fileno_rand(struct thread_data *td) |
1146 | { | |
1147 | unsigned long fileno; | |
1148 | ||
1149 | if (td->o.file_service_type == FIO_FSERVICE_RANDOM) { | |
1150 | uint64_t frand_max = rand_max(&td->next_file_state); | |
1151 | unsigned long r; | |
1152 | ||
1153 | r = __rand(&td->next_file_state); | |
1154 | return (unsigned int) ((double) td->o.nr_files | |
1155 | * (r / (frand_max + 1.0))); | |
1156 | } | |
1157 | ||
1158 | if (td->o.file_service_type == FIO_FSERVICE_ZIPF) | |
1159 | fileno = zipf_next(&td->next_file_zipf); | |
1160 | else if (td->o.file_service_type == FIO_FSERVICE_PARETO) | |
1161 | fileno = pareto_next(&td->next_file_zipf); | |
1162 | else if (td->o.file_service_type == FIO_FSERVICE_GAUSS) | |
1163 | fileno = gauss_next(&td->next_file_gauss); | |
1164 | else { | |
1165 | log_err("fio: bad file service type: %d\n", td->o.file_service_type); | |
1166 | assert(0); | |
1167 | return 0; | |
1168 | } | |
1169 | ||
1170 | return fileno >> FIO_FSERVICE_SHIFT; | |
1171 | } | |
1172 | ||
0aabe160 JA |
1173 | /* |
1174 | * Get next file to service by choosing one at random | |
1175 | */ | |
2cc52930 JA |
1176 | static struct fio_file *get_next_file_rand(struct thread_data *td, |
1177 | enum fio_file_flags goodf, | |
d6aed795 | 1178 | enum fio_file_flags badf) |
0aabe160 | 1179 | { |
0aabe160 | 1180 | struct fio_file *f; |
1c178180 | 1181 | int fno; |
0aabe160 JA |
1182 | |
1183 | do { | |
87b10676 | 1184 | int opened = 0; |
4c07ad86 | 1185 | |
8c07860d | 1186 | fno = __get_next_fileno_rand(td); |
7c83c089 | 1187 | |
126d65c6 | 1188 | f = td->files[fno]; |
d6aed795 | 1189 | if (fio_file_done(f)) |
059e63c0 | 1190 | continue; |
1c178180 | 1191 | |
d6aed795 | 1192 | if (!fio_file_open(f)) { |
87b10676 JA |
1193 | int err; |
1194 | ||
002fe734 JA |
1195 | if (td->nr_open_files >= td->o.open_files) |
1196 | return ERR_PTR(-EBUSY); | |
1197 | ||
87b10676 JA |
1198 | err = td_io_open_file(td, f); |
1199 | if (err) | |
1200 | continue; | |
1201 | opened = 1; | |
1202 | } | |
1203 | ||
2ba1c290 JA |
1204 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) { |
1205 | dprint(FD_FILE, "get_next_file_rand: %p\n", f); | |
0aabe160 | 1206 | return f; |
2ba1c290 | 1207 | } |
87b10676 JA |
1208 | if (opened) |
1209 | td_io_close_file(td, f); | |
0aabe160 JA |
1210 | } while (1); |
1211 | } | |
1212 | ||
1213 | /* | |
1214 | * Get next file to service by doing round robin between all available ones | |
1215 | */ | |
1c178180 JA |
1216 | static struct fio_file *get_next_file_rr(struct thread_data *td, int goodf, |
1217 | int badf) | |
3d7c391d JA |
1218 | { |
1219 | unsigned int old_next_file = td->next_file; | |
1220 | struct fio_file *f; | |
1221 | ||
1222 | do { | |
87b10676 JA |
1223 | int opened = 0; |
1224 | ||
126d65c6 | 1225 | f = td->files[td->next_file]; |
3d7c391d JA |
1226 | |
1227 | td->next_file++; | |
2dc1bbeb | 1228 | if (td->next_file >= td->o.nr_files) |
3d7c391d JA |
1229 | td->next_file = 0; |
1230 | ||
87b10676 | 1231 | dprint(FD_FILE, "trying file %s %x\n", f->file_name, f->flags); |
d6aed795 | 1232 | if (fio_file_done(f)) { |
d5ed68ea | 1233 | f = NULL; |
059e63c0 | 1234 | continue; |
d5ed68ea | 1235 | } |
059e63c0 | 1236 | |
d6aed795 | 1237 | if (!fio_file_open(f)) { |
87b10676 JA |
1238 | int err; |
1239 | ||
002fe734 JA |
1240 | if (td->nr_open_files >= td->o.open_files) |
1241 | return ERR_PTR(-EBUSY); | |
1242 | ||
87b10676 | 1243 | err = td_io_open_file(td, f); |
b5696bfc JA |
1244 | if (err) { |
1245 | dprint(FD_FILE, "error %d on open of %s\n", | |
1246 | err, f->file_name); | |
87c27b45 | 1247 | f = NULL; |
87b10676 | 1248 | continue; |
b5696bfc | 1249 | } |
87b10676 JA |
1250 | opened = 1; |
1251 | } | |
1252 | ||
0b9d69ec JA |
1253 | dprint(FD_FILE, "goodf=%x, badf=%x, ff=%x\n", goodf, badf, |
1254 | f->flags); | |
1c178180 | 1255 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) |
3d7c391d JA |
1256 | break; |
1257 | ||
87b10676 JA |
1258 | if (opened) |
1259 | td_io_close_file(td, f); | |
1260 | ||
3d7c391d JA |
1261 | f = NULL; |
1262 | } while (td->next_file != old_next_file); | |
1263 | ||
2ba1c290 | 1264 | dprint(FD_FILE, "get_next_file_rr: %p\n", f); |
3d7c391d JA |
1265 | return f; |
1266 | } | |
1267 | ||
7eb36574 | 1268 | static struct fio_file *__get_next_file(struct thread_data *td) |
bdb4e2e9 | 1269 | { |
1907dbc6 JA |
1270 | struct fio_file *f; |
1271 | ||
2dc1bbeb | 1272 | assert(td->o.nr_files <= td->files_index); |
1c178180 | 1273 | |
b5696bfc | 1274 | if (td->nr_done_files >= td->o.nr_files) { |
5ec10eaa JA |
1275 | dprint(FD_FILE, "get_next_file: nr_open=%d, nr_done=%d," |
1276 | " nr_files=%d\n", td->nr_open_files, | |
1277 | td->nr_done_files, | |
1278 | td->o.nr_files); | |
bdb4e2e9 | 1279 | return NULL; |
2ba1c290 | 1280 | } |
bdb4e2e9 | 1281 | |
1907dbc6 | 1282 | f = td->file_service_file; |
d6aed795 | 1283 | if (f && fio_file_open(f) && !fio_file_closing(f)) { |
a086c257 JA |
1284 | if (td->o.file_service_type == FIO_FSERVICE_SEQ) |
1285 | goto out; | |
1286 | if (td->file_service_left--) | |
1287 | goto out; | |
1288 | } | |
1907dbc6 | 1289 | |
a086c257 JA |
1290 | if (td->o.file_service_type == FIO_FSERVICE_RR || |
1291 | td->o.file_service_type == FIO_FSERVICE_SEQ) | |
d6aed795 | 1292 | f = get_next_file_rr(td, FIO_FILE_open, FIO_FILE_closing); |
bdb4e2e9 | 1293 | else |
d6aed795 | 1294 | f = get_next_file_rand(td, FIO_FILE_open, FIO_FILE_closing); |
1907dbc6 | 1295 | |
002fe734 JA |
1296 | if (IS_ERR(f)) |
1297 | return f; | |
1298 | ||
1907dbc6 JA |
1299 | td->file_service_file = f; |
1300 | td->file_service_left = td->file_service_nr - 1; | |
2ba1c290 | 1301 | out: |
0dac421f JA |
1302 | if (f) |
1303 | dprint(FD_FILE, "get_next_file: %p [%s]\n", f, f->file_name); | |
1304 | else | |
1305 | dprint(FD_FILE, "get_next_file: NULL\n"); | |
1907dbc6 | 1306 | return f; |
bdb4e2e9 JA |
1307 | } |
1308 | ||
7eb36574 JA |
1309 | static struct fio_file *get_next_file(struct thread_data *td) |
1310 | { | |
372d8962 | 1311 | if (td->flags & TD_F_PROFILE_OPS) { |
d72be545 | 1312 | struct prof_io_ops *ops = &td->prof_io_ops; |
7eb36574 | 1313 | |
d72be545 JA |
1314 | if (ops->get_next_file) |
1315 | return ops->get_next_file(td); | |
1316 | } | |
7eb36574 JA |
1317 | |
1318 | return __get_next_file(td); | |
1319 | } | |
1320 | ||
002fe734 | 1321 | static long set_io_u_file(struct thread_data *td, struct io_u *io_u) |
429f6675 JA |
1322 | { |
1323 | struct fio_file *f; | |
1324 | ||
1325 | do { | |
1326 | f = get_next_file(td); | |
002fe734 JA |
1327 | if (IS_ERR_OR_NULL(f)) |
1328 | return PTR_ERR(f); | |
429f6675 | 1329 | |
429f6675 JA |
1330 | io_u->file = f; |
1331 | get_file(f); | |
1332 | ||
1333 | if (!fill_io_u(td, io_u)) | |
1334 | break; | |
1335 | ||
b5696bfc | 1336 | put_file_log(td, f); |
429f6675 | 1337 | td_io_close_file(td, f); |
b5696bfc | 1338 | io_u->file = NULL; |
8c07860d JA |
1339 | if (td->o.file_service_type & __FIO_FSERVICE_NONUNIFORM) |
1340 | fio_file_reset(td, f); | |
1341 | else { | |
1342 | fio_file_set_done(f); | |
1343 | td->nr_done_files++; | |
1344 | dprint(FD_FILE, "%s: is done (%d of %d)\n", f->file_name, | |
0b9d69ec | 1345 | td->nr_done_files, td->o.nr_files); |
8c07860d | 1346 | } |
429f6675 JA |
1347 | } while (1); |
1348 | ||
1349 | return 0; | |
1350 | } | |
1351 | ||
3e260a46 JA |
1352 | static void lat_fatal(struct thread_data *td, struct io_completion_data *icd, |
1353 | unsigned long tusec, unsigned long max_usec) | |
1354 | { | |
1355 | if (!td->error) | |
1356 | log_err("fio: latency of %lu usec exceeds specified max (%lu usec)\n", tusec, max_usec); | |
1357 | td_verror(td, ETIMEDOUT, "max latency exceeded"); | |
1358 | icd->error = ETIMEDOUT; | |
1359 | } | |
1360 | ||
1361 | static void lat_new_cycle(struct thread_data *td) | |
1362 | { | |
1363 | fio_gettime(&td->latency_ts, NULL); | |
1364 | td->latency_ios = ddir_rw_sum(td->io_blocks); | |
1365 | td->latency_failed = 0; | |
1366 | } | |
1367 | ||
1368 | /* | |
1369 | * We had an IO outside the latency target. Reduce the queue depth. If we | |
1370 | * are at QD=1, then it's time to give up. | |
1371 | */ | |
e39c0676 | 1372 | static bool __lat_target_failed(struct thread_data *td) |
3e260a46 JA |
1373 | { |
1374 | if (td->latency_qd == 1) | |
e39c0676 | 1375 | return true; |
3e260a46 JA |
1376 | |
1377 | td->latency_qd_high = td->latency_qd; | |
6bb58215 JA |
1378 | |
1379 | if (td->latency_qd == td->latency_qd_low) | |
1380 | td->latency_qd_low--; | |
1381 | ||
3e260a46 JA |
1382 | td->latency_qd = (td->latency_qd + td->latency_qd_low) / 2; |
1383 | ||
1384 | dprint(FD_RATE, "Ramped down: %d %d %d\n", td->latency_qd_low, td->latency_qd, td->latency_qd_high); | |
1385 | ||
1386 | /* | |
1387 | * When we ramp QD down, quiesce existing IO to prevent | |
1388 | * a storm of ramp downs due to pending higher depth. | |
1389 | */ | |
1390 | io_u_quiesce(td); | |
1391 | lat_new_cycle(td); | |
e39c0676 | 1392 | return false; |
3e260a46 JA |
1393 | } |
1394 | ||
e39c0676 | 1395 | static bool lat_target_failed(struct thread_data *td) |
3e260a46 JA |
1396 | { |
1397 | if (td->o.latency_percentile.u.f == 100.0) | |
1398 | return __lat_target_failed(td); | |
1399 | ||
1400 | td->latency_failed++; | |
e39c0676 | 1401 | return false; |
3e260a46 JA |
1402 | } |
1403 | ||
1404 | void lat_target_init(struct thread_data *td) | |
1405 | { | |
6bb58215 JA |
1406 | td->latency_end_run = 0; |
1407 | ||
3e260a46 JA |
1408 | if (td->o.latency_target) { |
1409 | dprint(FD_RATE, "Latency target=%llu\n", td->o.latency_target); | |
1410 | fio_gettime(&td->latency_ts, NULL); | |
1411 | td->latency_qd = 1; | |
1412 | td->latency_qd_high = td->o.iodepth; | |
1413 | td->latency_qd_low = 1; | |
1414 | td->latency_ios = ddir_rw_sum(td->io_blocks); | |
1415 | } else | |
1416 | td->latency_qd = td->o.iodepth; | |
1417 | } | |
1418 | ||
6bb58215 JA |
1419 | void lat_target_reset(struct thread_data *td) |
1420 | { | |
1421 | if (!td->latency_end_run) | |
1422 | lat_target_init(td); | |
1423 | } | |
1424 | ||
3e260a46 JA |
1425 | static void lat_target_success(struct thread_data *td) |
1426 | { | |
1427 | const unsigned int qd = td->latency_qd; | |
6bb58215 | 1428 | struct thread_options *o = &td->o; |
3e260a46 JA |
1429 | |
1430 | td->latency_qd_low = td->latency_qd; | |
1431 | ||
1432 | /* | |
1433 | * If we haven't failed yet, we double up to a failing value instead | |
1434 | * of bisecting from highest possible queue depth. If we have set | |
1435 | * a limit other than td->o.iodepth, bisect between that. | |
1436 | */ | |
6bb58215 | 1437 | if (td->latency_qd_high != o->iodepth) |
3e260a46 JA |
1438 | td->latency_qd = (td->latency_qd + td->latency_qd_high) / 2; |
1439 | else | |
1440 | td->latency_qd *= 2; | |
1441 | ||
6bb58215 JA |
1442 | if (td->latency_qd > o->iodepth) |
1443 | td->latency_qd = o->iodepth; | |
3e260a46 JA |
1444 | |
1445 | dprint(FD_RATE, "Ramped up: %d %d %d\n", td->latency_qd_low, td->latency_qd, td->latency_qd_high); | |
6bb58215 | 1446 | |
3e260a46 | 1447 | /* |
6bb58215 JA |
1448 | * Same as last one, we are done. Let it run a latency cycle, so |
1449 | * we get only the results from the targeted depth. | |
3e260a46 | 1450 | */ |
6bb58215 JA |
1451 | if (td->latency_qd == qd) { |
1452 | if (td->latency_end_run) { | |
1453 | dprint(FD_RATE, "We are done\n"); | |
1454 | td->done = 1; | |
1455 | } else { | |
1456 | dprint(FD_RATE, "Quiesce and final run\n"); | |
1457 | io_u_quiesce(td); | |
1458 | td->latency_end_run = 1; | |
1459 | reset_all_stats(td); | |
1460 | reset_io_stats(td); | |
1461 | } | |
1462 | } | |
3e260a46 JA |
1463 | |
1464 | lat_new_cycle(td); | |
1465 | } | |
1466 | ||
1467 | /* | |
1468 | * Check if we can bump the queue depth | |
1469 | */ | |
1470 | void lat_target_check(struct thread_data *td) | |
1471 | { | |
1472 | uint64_t usec_window; | |
1473 | uint64_t ios; | |
1474 | double success_ios; | |
1475 | ||
1476 | usec_window = utime_since_now(&td->latency_ts); | |
1477 | if (usec_window < td->o.latency_window) | |
1478 | return; | |
1479 | ||
1480 | ios = ddir_rw_sum(td->io_blocks) - td->latency_ios; | |
1481 | success_ios = (double) (ios - td->latency_failed) / (double) ios; | |
1482 | success_ios *= 100.0; | |
1483 | ||
1484 | dprint(FD_RATE, "Success rate: %.2f%% (target %.2f%%)\n", success_ios, td->o.latency_percentile.u.f); | |
1485 | ||
1486 | if (success_ios >= td->o.latency_percentile.u.f) | |
1487 | lat_target_success(td); | |
1488 | else | |
1489 | __lat_target_failed(td); | |
1490 | } | |
1491 | ||
1492 | /* | |
1493 | * If latency target is enabled, we might be ramping up or down and not | |
1494 | * using the full queue depth available. | |
1495 | */ | |
e39c0676 | 1496 | bool queue_full(const struct thread_data *td) |
3e260a46 JA |
1497 | { |
1498 | const int qempty = io_u_qempty(&td->io_u_freelist); | |
1499 | ||
1500 | if (qempty) | |
e39c0676 | 1501 | return true; |
3e260a46 | 1502 | if (!td->o.latency_target) |
e39c0676 | 1503 | return false; |
3e260a46 JA |
1504 | |
1505 | return td->cur_depth >= td->latency_qd; | |
1506 | } | |
429f6675 | 1507 | |
10ba535a JA |
1508 | struct io_u *__get_io_u(struct thread_data *td) |
1509 | { | |
0cae66f6 | 1510 | struct io_u *io_u = NULL; |
10ba535a | 1511 | |
ca09be4b JA |
1512 | if (td->stop_io) |
1513 | return NULL; | |
1514 | ||
e8462bd8 JA |
1515 | td_io_u_lock(td); |
1516 | ||
1517 | again: | |
2ae0b204 JA |
1518 | if (!io_u_rempty(&td->io_u_requeues)) |
1519 | io_u = io_u_rpop(&td->io_u_requeues); | |
3e260a46 | 1520 | else if (!queue_full(td)) { |
2ae0b204 | 1521 | io_u = io_u_qpop(&td->io_u_freelist); |
10ba535a | 1522 | |
225ba9e3 | 1523 | io_u->file = NULL; |
6040dabc | 1524 | io_u->buflen = 0; |
10ba535a | 1525 | io_u->resid = 0; |
d7762cf8 | 1526 | io_u->end_io = NULL; |
755200a3 JA |
1527 | } |
1528 | ||
1529 | if (io_u) { | |
0c6e7517 | 1530 | assert(io_u->flags & IO_U_F_FREE); |
1651e431 | 1531 | io_u_clear(td, io_u, IO_U_F_FREE | IO_U_F_NO_FILE_PUT | |
f8b0bd10 JA |
1532 | IO_U_F_TRIMMED | IO_U_F_BARRIER | |
1533 | IO_U_F_VER_LIST); | |
0c6e7517 | 1534 | |
755200a3 | 1535 | io_u->error = 0; |
bcd5abfa | 1536 | io_u->acct_ddir = -1; |
10ba535a | 1537 | td->cur_depth++; |
a9da8ab2 | 1538 | assert(!(td->flags & TD_F_CHILD)); |
1651e431 | 1539 | io_u_set(td, io_u, IO_U_F_IN_CUR_DEPTH); |
f9401285 | 1540 | io_u->ipo = NULL; |
a9da8ab2 | 1541 | } else if (td_async_processing(td)) { |
1dec3e07 JA |
1542 | /* |
1543 | * We ran out, wait for async verify threads to finish and | |
1544 | * return one | |
1545 | */ | |
a9da8ab2 JA |
1546 | assert(!(td->flags & TD_F_CHILD)); |
1547 | assert(!pthread_cond_wait(&td->free_cond, &td->io_u_lock)); | |
1dec3e07 | 1548 | goto again; |
10ba535a JA |
1549 | } |
1550 | ||
e8462bd8 | 1551 | td_io_u_unlock(td); |
10ba535a JA |
1552 | return io_u; |
1553 | } | |
1554 | ||
e39c0676 | 1555 | static bool check_get_trim(struct thread_data *td, struct io_u *io_u) |
10ba535a | 1556 | { |
d72be545 | 1557 | if (!(td->flags & TD_F_TRIM_BACKLOG)) |
e39c0676 | 1558 | return false; |
d72be545 JA |
1559 | |
1560 | if (td->trim_entries) { | |
0d29de83 | 1561 | int get_trim = 0; |
10ba535a | 1562 | |
0d29de83 JA |
1563 | if (td->trim_batch) { |
1564 | td->trim_batch--; | |
1565 | get_trim = 1; | |
1566 | } else if (!(td->io_hist_len % td->o.trim_backlog) && | |
1567 | td->last_ddir != DDIR_READ) { | |
1568 | td->trim_batch = td->o.trim_batch; | |
1569 | if (!td->trim_batch) | |
1570 | td->trim_batch = td->o.trim_backlog; | |
1571 | get_trim = 1; | |
1572 | } | |
1573 | ||
917c64b3 | 1574 | if (get_trim && get_next_trim(td, io_u)) |
e39c0676 | 1575 | return true; |
2ba1c290 | 1576 | } |
10ba535a | 1577 | |
e39c0676 | 1578 | return false; |
0d29de83 JA |
1579 | } |
1580 | ||
e39c0676 | 1581 | static bool check_get_verify(struct thread_data *td, struct io_u *io_u) |
0d29de83 | 1582 | { |
d72be545 | 1583 | if (!(td->flags & TD_F_VER_BACKLOG)) |
e39c0676 | 1584 | return false; |
d72be545 JA |
1585 | |
1586 | if (td->io_hist_len) { | |
9e144189 JA |
1587 | int get_verify = 0; |
1588 | ||
d1ece0c7 | 1589 | if (td->verify_batch) |
9e144189 | 1590 | get_verify = 1; |
d1ece0c7 | 1591 | else if (!(td->io_hist_len % td->o.verify_backlog) && |
9e144189 JA |
1592 | td->last_ddir != DDIR_READ) { |
1593 | td->verify_batch = td->o.verify_batch; | |
f8a75c99 JA |
1594 | if (!td->verify_batch) |
1595 | td->verify_batch = td->o.verify_backlog; | |
9e144189 JA |
1596 | get_verify = 1; |
1597 | } | |
1598 | ||
d1ece0c7 JA |
1599 | if (get_verify && !get_next_verify(td, io_u)) { |
1600 | td->verify_batch--; | |
e39c0676 | 1601 | return true; |
d1ece0c7 | 1602 | } |
9e144189 JA |
1603 | } |
1604 | ||
e39c0676 | 1605 | return false; |
0d29de83 JA |
1606 | } |
1607 | ||
de789769 JA |
1608 | /* |
1609 | * Fill offset and start time into the buffer content, to prevent too | |
23f394d5 JA |
1610 | * easy compressible data for simple de-dupe attempts. Do this for every |
1611 | * 512b block in the range, since that should be the smallest block size | |
1612 | * we can expect from a device. | |
de789769 JA |
1613 | */ |
1614 | static void small_content_scramble(struct io_u *io_u) | |
1615 | { | |
23f394d5 | 1616 | unsigned int i, nr_blocks = io_u->buflen / 512; |
1ae83d45 | 1617 | uint64_t boffset; |
23f394d5 | 1618 | unsigned int offset; |
fe8d7fee | 1619 | char *p, *end; |
de789769 | 1620 | |
23f394d5 JA |
1621 | if (!nr_blocks) |
1622 | return; | |
1623 | ||
1624 | p = io_u->xfer_buf; | |
fba76ee8 | 1625 | boffset = io_u->offset; |
81f0366c | 1626 | io_u->buf_filled_len = 0; |
fad82f76 | 1627 | |
23f394d5 JA |
1628 | for (i = 0; i < nr_blocks; i++) { |
1629 | /* | |
1630 | * Fill the byte offset into a "random" start offset of | |
1631 | * the buffer, given by the product of the usec time | |
1632 | * and the actual offset. | |
1633 | */ | |
8b6a404c | 1634 | offset = ((io_u->start_time.tv_nsec/1000) ^ boffset) & 511; |
1ae83d45 JA |
1635 | offset &= ~(sizeof(uint64_t) - 1); |
1636 | if (offset >= 512 - sizeof(uint64_t)) | |
1637 | offset -= sizeof(uint64_t); | |
fba76ee8 | 1638 | memcpy(p + offset, &boffset, sizeof(boffset)); |
23f394d5 JA |
1639 | |
1640 | end = p + 512 - sizeof(io_u->start_time); | |
1641 | memcpy(end, &io_u->start_time, sizeof(io_u->start_time)); | |
1642 | p += 512; | |
fad82f76 | 1643 | boffset += 512; |
23f394d5 | 1644 | } |
de789769 JA |
1645 | } |
1646 | ||
0d29de83 JA |
1647 | /* |
1648 | * Return an io_u to be processed. Gets a buflen and offset, sets direction, | |
1649 | * etc. The returned io_u is fully ready to be prepped and submitted. | |
1650 | */ | |
1651 | struct io_u *get_io_u(struct thread_data *td) | |
1652 | { | |
1653 | struct fio_file *f; | |
1654 | struct io_u *io_u; | |
de789769 | 1655 | int do_scramble = 0; |
002fe734 | 1656 | long ret = 0; |
0d29de83 JA |
1657 | |
1658 | io_u = __get_io_u(td); | |
1659 | if (!io_u) { | |
1660 | dprint(FD_IO, "__get_io_u failed\n"); | |
1661 | return NULL; | |
1662 | } | |
1663 | ||
1664 | if (check_get_verify(td, io_u)) | |
1665 | goto out; | |
1666 | if (check_get_trim(td, io_u)) | |
1667 | goto out; | |
1668 | ||
755200a3 JA |
1669 | /* |
1670 | * from a requeue, io_u already setup | |
1671 | */ | |
1672 | if (io_u->file) | |
77f392bf | 1673 | goto out; |
755200a3 | 1674 | |
429f6675 JA |
1675 | /* |
1676 | * If using an iolog, grab next piece if any available. | |
1677 | */ | |
d72be545 | 1678 | if (td->flags & TD_F_READ_IOLOG) { |
429f6675 JA |
1679 | if (read_iolog_get(td, io_u)) |
1680 | goto err_put; | |
2ba1c290 | 1681 | } else if (set_io_u_file(td, io_u)) { |
002fe734 | 1682 | ret = -EBUSY; |
2ba1c290 | 1683 | dprint(FD_IO, "io_u %p, setting file failed\n", io_u); |
429f6675 | 1684 | goto err_put; |
2ba1c290 | 1685 | } |
5ec10eaa | 1686 | |
429f6675 | 1687 | f = io_u->file; |
002fe734 JA |
1688 | if (!f) { |
1689 | dprint(FD_IO, "io_u %p, setting file failed\n", io_u); | |
1690 | goto err_put; | |
1691 | } | |
1692 | ||
d6aed795 | 1693 | assert(fio_file_open(f)); |
97af62ce | 1694 | |
ff58fced | 1695 | if (ddir_rw(io_u->ddir)) { |
9b87f09b | 1696 | if (!io_u->buflen && !td_ioengine_flagged(td, FIO_NOIO)) { |
2ba1c290 | 1697 | dprint(FD_IO, "get_io_u: zero buflen on %p\n", io_u); |
429f6675 | 1698 | goto err_put; |
2ba1c290 | 1699 | } |
10ba535a | 1700 | |
f1dfb668 JA |
1701 | f->last_start[io_u->ddir] = io_u->offset; |
1702 | f->last_pos[io_u->ddir] = io_u->offset + io_u->buflen; | |
10ba535a | 1703 | |
fd68418e | 1704 | if (io_u->ddir == DDIR_WRITE) { |
d72be545 | 1705 | if (td->flags & TD_F_REFILL_BUFFERS) { |
9c42684e | 1706 | io_u_fill_buffer(td, io_u, |
1066358a | 1707 | td->o.min_bs[DDIR_WRITE], |
9e129577 | 1708 | io_u->buflen); |
ff441ae8 | 1709 | } else if ((td->flags & TD_F_SCRAMBLE_BUFFERS) && |
bedc9dc2 | 1710 | !(td->flags & TD_F_COMPRESS)) |
fd68418e | 1711 | do_scramble = 1; |
d72be545 | 1712 | if (td->flags & TD_F_VER_NONE) { |
629f1d71 JA |
1713 | populate_verify_io_u(td, io_u); |
1714 | do_scramble = 0; | |
1715 | } | |
fd68418e | 1716 | } else if (io_u->ddir == DDIR_READ) { |
cbe8d756 RR |
1717 | /* |
1718 | * Reset the buf_filled parameters so next time if the | |
1719 | * buffer is used for writes it is refilled. | |
1720 | */ | |
cbe8d756 RR |
1721 | io_u->buf_filled_len = 0; |
1722 | } | |
87dc1ab1 | 1723 | } |
10ba535a | 1724 | |
165faf16 JA |
1725 | /* |
1726 | * Set io data pointers. | |
1727 | */ | |
cec6b55d JA |
1728 | io_u->xfer_buf = io_u->buf; |
1729 | io_u->xfer_buflen = io_u->buflen; | |
5973cafb | 1730 | |
6ac7a331 | 1731 | out: |
0d29de83 | 1732 | assert(io_u->file); |
429f6675 | 1733 | if (!td_io_prep(td, io_u)) { |
3ecc8c67 | 1734 | if (!td->o.disable_lat) |
993bf48b | 1735 | fio_gettime(&io_u->start_time, NULL); |
03553853 | 1736 | |
de789769 JA |
1737 | if (do_scramble) |
1738 | small_content_scramble(io_u); | |
03553853 | 1739 | |
429f6675 | 1740 | return io_u; |
36167d82 | 1741 | } |
429f6675 | 1742 | err_put: |
2ba1c290 | 1743 | dprint(FD_IO, "get_io_u failed\n"); |
429f6675 | 1744 | put_io_u(td, io_u); |
002fe734 | 1745 | return ERR_PTR(ret); |
10ba535a JA |
1746 | } |
1747 | ||
a9da8ab2 | 1748 | static void __io_u_log_error(struct thread_data *td, struct io_u *io_u) |
5451792e | 1749 | { |
8b28bd41 | 1750 | enum error_type_bit eb = td_error_type(io_u->ddir, io_u->error); |
825f818e | 1751 | |
8b28bd41 DM |
1752 | if (td_non_fatal_error(td, eb, io_u->error) && !td->o.error_dump) |
1753 | return; | |
5451792e | 1754 | |
709c8313 RE |
1755 | log_err("fio: io_u error%s%s: %s: %s offset=%llu, buflen=%lu\n", |
1756 | io_u->file ? " on file " : "", | |
1757 | io_u->file ? io_u->file->file_name : "", | |
1758 | strerror(io_u->error), | |
1759 | io_ddir_name(io_u->ddir), | |
1760 | io_u->offset, io_u->xfer_buflen); | |
5451792e | 1761 | |
5ad7be56 KD |
1762 | if (td->io_ops->errdetails) { |
1763 | char *err = td->io_ops->errdetails(io_u); | |
1764 | ||
1765 | log_err("fio: %s\n", err); | |
1766 | free(err); | |
1767 | } | |
1768 | ||
5451792e JA |
1769 | if (!td->error) |
1770 | td_verror(td, io_u->error, "io_u error"); | |
1771 | } | |
1772 | ||
a9da8ab2 JA |
1773 | void io_u_log_error(struct thread_data *td, struct io_u *io_u) |
1774 | { | |
1775 | __io_u_log_error(td, io_u); | |
1776 | if (td->parent) | |
094e66cb | 1777 | __io_u_log_error(td->parent, io_u); |
a9da8ab2 JA |
1778 | } |
1779 | ||
e39c0676 | 1780 | static inline bool gtod_reduce(struct thread_data *td) |
aba6c951 | 1781 | { |
3ecc8c67 JA |
1782 | return (td->o.disable_clat && td->o.disable_slat && td->o.disable_bw) |
1783 | || td->o.gtod_reduce; | |
aba6c951 JA |
1784 | } |
1785 | ||
c8eeb9df JA |
1786 | static void account_io_completion(struct thread_data *td, struct io_u *io_u, |
1787 | struct io_completion_data *icd, | |
1788 | const enum fio_ddir idx, unsigned int bytes) | |
1789 | { | |
a9da8ab2 | 1790 | const int no_reduce = !gtod_reduce(td); |
d6bb626e | 1791 | unsigned long long llnsec = 0; |
c8eeb9df | 1792 | |
75dc383e JA |
1793 | if (td->parent) |
1794 | td = td->parent; | |
1795 | ||
132b1ee4 | 1796 | if (!td->o.stats || td_ioengine_flagged(td, FIO_NOSTATS)) |
8243be59 JA |
1797 | return; |
1798 | ||
a9da8ab2 | 1799 | if (no_reduce) |
d6bb626e | 1800 | llnsec = ntime_since(&io_u->issue_time, &icd->time); |
c8eeb9df JA |
1801 | |
1802 | if (!td->o.disable_lat) { | |
d6bb626e | 1803 | unsigned long long tnsec; |
c8eeb9df | 1804 | |
d6bb626e VF |
1805 | tnsec = ntime_since(&io_u->start_time, &icd->time); |
1806 | add_lat_sample(td, idx, tnsec, bytes, io_u->offset); | |
15501535 | 1807 | |
d4afedfd JA |
1808 | if (td->flags & TD_F_PROFILE_OPS) { |
1809 | struct prof_io_ops *ops = &td->prof_io_ops; | |
1810 | ||
1811 | if (ops->io_u_lat) | |
d6bb626e | 1812 | icd->error = ops->io_u_lat(td, tnsec/1000); |
d4afedfd JA |
1813 | } |
1814 | ||
d6bb626e VF |
1815 | if (td->o.max_latency && tnsec/1000 > td->o.max_latency) |
1816 | lat_fatal(td, icd, tnsec/1000, td->o.max_latency); | |
1817 | if (td->o.latency_target && tnsec/1000 > td->o.latency_target) { | |
3e260a46 | 1818 | if (lat_target_failed(td)) |
d6bb626e | 1819 | lat_fatal(td, icd, tnsec/1000, td->o.latency_target); |
15501535 | 1820 | } |
c8eeb9df JA |
1821 | } |
1822 | ||
a47591e4 JA |
1823 | if (ddir_rw(idx)) { |
1824 | if (!td->o.disable_clat) { | |
d6bb626e VF |
1825 | add_clat_sample(td, idx, llnsec, bytes, io_u->offset); |
1826 | io_u_mark_latency(td, llnsec); | |
a47591e4 | 1827 | } |
c8eeb9df | 1828 | |
a47591e4 | 1829 | if (!td->o.disable_bw && per_unit_log(td->bw_log)) |
d6bb626e | 1830 | add_bw_sample(td, io_u, bytes, llnsec); |
c8eeb9df | 1831 | |
a47591e4 JA |
1832 | if (no_reduce && per_unit_log(td->iops_log)) |
1833 | add_iops_sample(td, io_u, bytes); | |
1834 | } | |
66347cfa DE |
1835 | |
1836 | if (td->ts.nr_block_infos && io_u->ddir == DDIR_TRIM) { | |
1837 | uint32_t *info = io_u_block_info(td, io_u); | |
1838 | if (BLOCK_INFO_STATE(*info) < BLOCK_STATE_TRIM_FAILURE) { | |
1839 | if (io_u->ddir == DDIR_TRIM) { | |
1840 | *info = BLOCK_INFO(BLOCK_STATE_TRIMMED, | |
1841 | BLOCK_INFO_TRIMS(*info) + 1); | |
1842 | } else if (io_u->ddir == DDIR_WRITE) { | |
1843 | *info = BLOCK_INFO_SET_STATE(BLOCK_STATE_WRITTEN, | |
1844 | *info); | |
1845 | } | |
1846 | } | |
1847 | } | |
c8eeb9df JA |
1848 | } |
1849 | ||
94a6e1bb JA |
1850 | static void file_log_write_comp(const struct thread_data *td, struct fio_file *f, |
1851 | uint64_t offset, unsigned int bytes) | |
1852 | { | |
1853 | int idx; | |
1854 | ||
639ad1ea JA |
1855 | if (!f) |
1856 | return; | |
1857 | ||
94a6e1bb JA |
1858 | if (f->first_write == -1ULL || offset < f->first_write) |
1859 | f->first_write = offset; | |
1860 | if (f->last_write == -1ULL || ((offset + bytes) > f->last_write)) | |
1861 | f->last_write = offset + bytes; | |
1862 | ||
1863 | if (!f->last_write_comp) | |
1864 | return; | |
1865 | ||
1866 | idx = f->last_write_idx++; | |
1867 | f->last_write_comp[idx] = offset; | |
1868 | if (f->last_write_idx == td->o.iodepth) | |
1869 | f->last_write_idx = 0; | |
1870 | } | |
1871 | ||
f8b0bd10 | 1872 | static void io_completed(struct thread_data *td, struct io_u **io_u_ptr, |
97601024 | 1873 | struct io_completion_data *icd) |
10ba535a | 1874 | { |
f8b0bd10 JA |
1875 | struct io_u *io_u = *io_u_ptr; |
1876 | enum fio_ddir ddir = io_u->ddir; | |
1877 | struct fio_file *f = io_u->file; | |
10ba535a | 1878 | |
2ba1c290 JA |
1879 | dprint_io_u(io_u, "io complete"); |
1880 | ||
0c6e7517 | 1881 | assert(io_u->flags & IO_U_F_FLIGHT); |
1651e431 | 1882 | io_u_clear(td, io_u, IO_U_F_FLIGHT | IO_U_F_BUSY_OK); |
f9401285 JA |
1883 | |
1884 | /* | |
1885 | * Mark IO ok to verify | |
1886 | */ | |
1887 | if (io_u->ipo) { | |
890b6656 JA |
1888 | /* |
1889 | * Remove errored entry from the verification list | |
1890 | */ | |
1891 | if (io_u->error) | |
1892 | unlog_io_piece(td, io_u); | |
1893 | else { | |
1894 | io_u->ipo->flags &= ~IP_F_IN_FLIGHT; | |
1895 | write_barrier(); | |
1896 | } | |
f9401285 JA |
1897 | } |
1898 | ||
f8b0bd10 | 1899 | if (ddir_sync(ddir)) { |
87dc1ab1 | 1900 | td->last_was_sync = 1; |
44f29692 JA |
1901 | if (f) { |
1902 | f->first_write = -1ULL; | |
1903 | f->last_write = -1ULL; | |
1904 | } | |
87dc1ab1 JA |
1905 | return; |
1906 | } | |
1907 | ||
1908 | td->last_was_sync = 0; | |
f8b0bd10 | 1909 | td->last_ddir = ddir; |
87dc1ab1 | 1910 | |
f8b0bd10 | 1911 | if (!io_u->error && ddir_rw(ddir)) { |
10ba535a | 1912 | unsigned int bytes = io_u->buflen - io_u->resid; |
b29ee5b3 | 1913 | int ret; |
10ba535a | 1914 | |
f8b0bd10 JA |
1915 | td->io_blocks[ddir]++; |
1916 | td->this_io_blocks[ddir]++; | |
1917 | td->io_bytes[ddir] += bytes; | |
ae2fafc8 | 1918 | |
1919 | if (!(io_u->flags & IO_U_F_VER_LIST)) | |
f8b0bd10 JA |
1920 | td->this_io_bytes[ddir] += bytes; |
1921 | ||
639ad1ea | 1922 | if (ddir == DDIR_WRITE) |
94a6e1bb | 1923 | file_log_write_comp(td, f, io_u->offset, bytes); |
44f29692 | 1924 | |
6b1190fd | 1925 | if (ramp_time_over(td) && (td->runstate == TD_RUNNING || |
50a8ce86 | 1926 | td->runstate == TD_VERIFYING)) |
f8b0bd10 | 1927 | account_io_completion(td, io_u, icd, ddir, bytes); |
40e1a6f0 | 1928 | |
f8b0bd10 | 1929 | icd->bytes_done[ddir] += bytes; |
3af6ef39 | 1930 | |
d7762cf8 | 1931 | if (io_u->end_io) { |
f8b0bd10 JA |
1932 | ret = io_u->end_io(td, io_u_ptr); |
1933 | io_u = *io_u_ptr; | |
3af6ef39 JA |
1934 | if (ret && !icd->error) |
1935 | icd->error = ret; | |
1936 | } | |
ff58fced | 1937 | } else if (io_u->error) { |
10ba535a | 1938 | icd->error = io_u->error; |
5451792e JA |
1939 | io_u_log_error(td, io_u); |
1940 | } | |
8b28bd41 | 1941 | if (icd->error) { |
f8b0bd10 JA |
1942 | enum error_type_bit eb = td_error_type(ddir, icd->error); |
1943 | ||
8b28bd41 DM |
1944 | if (!td_non_fatal_error(td, eb, icd->error)) |
1945 | return; | |
f8b0bd10 | 1946 | |
f2bba182 RR |
1947 | /* |
1948 | * If there is a non_fatal error, then add to the error count | |
1949 | * and clear all the errors. | |
1950 | */ | |
1951 | update_error_count(td, icd->error); | |
1952 | td_clear_error(td); | |
1953 | icd->error = 0; | |
f8b0bd10 JA |
1954 | if (io_u) |
1955 | io_u->error = 0; | |
f2bba182 | 1956 | } |
10ba535a JA |
1957 | } |
1958 | ||
9520ebb9 JA |
1959 | static void init_icd(struct thread_data *td, struct io_completion_data *icd, |
1960 | int nr) | |
10ba535a | 1961 | { |
6eaf09d6 | 1962 | int ddir; |
aba6c951 JA |
1963 | |
1964 | if (!gtod_reduce(td)) | |
9520ebb9 | 1965 | fio_gettime(&icd->time, NULL); |
02bcaa8c | 1966 | |
3af6ef39 JA |
1967 | icd->nr = nr; |
1968 | ||
10ba535a | 1969 | icd->error = 0; |
c1f50f76 | 1970 | for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++) |
6eaf09d6 | 1971 | icd->bytes_done[ddir] = 0; |
36167d82 JA |
1972 | } |
1973 | ||
97601024 JA |
1974 | static void ios_completed(struct thread_data *td, |
1975 | struct io_completion_data *icd) | |
36167d82 JA |
1976 | { |
1977 | struct io_u *io_u; | |
1978 | int i; | |
1979 | ||
10ba535a JA |
1980 | for (i = 0; i < icd->nr; i++) { |
1981 | io_u = td->io_ops->event(td, i); | |
1982 | ||
f8b0bd10 | 1983 | io_completed(td, &io_u, icd); |
e8462bd8 | 1984 | |
f8b0bd10 | 1985 | if (io_u) |
e8462bd8 | 1986 | put_io_u(td, io_u); |
10ba535a JA |
1987 | } |
1988 | } | |
97601024 | 1989 | |
e7e6cfb4 JA |
1990 | /* |
1991 | * Complete a single io_u for the sync engines. | |
1992 | */ | |
55312f9f | 1993 | int io_u_sync_complete(struct thread_data *td, struct io_u *io_u) |
97601024 JA |
1994 | { |
1995 | struct io_completion_data icd; | |
55312f9f | 1996 | int ddir; |
97601024 | 1997 | |
9520ebb9 | 1998 | init_icd(td, &icd, 1); |
f8b0bd10 | 1999 | io_completed(td, &io_u, &icd); |
e8462bd8 | 2000 | |
f8b0bd10 | 2001 | if (io_u) |
e8462bd8 | 2002 | put_io_u(td, io_u); |
97601024 | 2003 | |
581e7141 JA |
2004 | if (icd.error) { |
2005 | td_verror(td, icd.error, "io_u_sync_complete"); | |
2006 | return -1; | |
2007 | } | |
97601024 | 2008 | |
c1f50f76 | 2009 | for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++) |
55312f9f | 2010 | td->bytes_done[ddir] += icd.bytes_done[ddir]; |
581e7141 JA |
2011 | |
2012 | return 0; | |
97601024 JA |
2013 | } |
2014 | ||
e7e6cfb4 JA |
2015 | /* |
2016 | * Called to complete min_events number of io for the async engines. | |
2017 | */ | |
55312f9f | 2018 | int io_u_queued_complete(struct thread_data *td, int min_evts) |
97601024 | 2019 | { |
97601024 | 2020 | struct io_completion_data icd; |
00de55ef | 2021 | struct timespec *tvp = NULL; |
55312f9f | 2022 | int ret, ddir; |
4d06a338 | 2023 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0, }; |
97601024 | 2024 | |
12bb8569 | 2025 | dprint(FD_IO, "io_u_queued_complete: min=%d\n", min_evts); |
b271fe62 | 2026 | |
4950421a | 2027 | if (!min_evts) |
00de55ef | 2028 | tvp = &ts; |
5fb4b366 RE |
2029 | else if (min_evts > td->cur_depth) |
2030 | min_evts = td->cur_depth; | |
97601024 | 2031 | |
82407585 RP |
2032 | /* No worries, td_io_getevents fixes min and max if they are |
2033 | * set incorrectly */ | |
2034 | ret = td_io_getevents(td, min_evts, td->o.iodepth_batch_complete_max, tvp); | |
97601024 | 2035 | if (ret < 0) { |
e1161c32 | 2036 | td_verror(td, -ret, "td_io_getevents"); |
97601024 JA |
2037 | return ret; |
2038 | } else if (!ret) | |
2039 | return ret; | |
2040 | ||
9520ebb9 | 2041 | init_icd(td, &icd, ret); |
97601024 | 2042 | ios_completed(td, &icd); |
581e7141 JA |
2043 | if (icd.error) { |
2044 | td_verror(td, icd.error, "io_u_queued_complete"); | |
2045 | return -1; | |
2046 | } | |
97601024 | 2047 | |
c1f50f76 | 2048 | for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++) |
55312f9f | 2049 | td->bytes_done[ddir] += icd.bytes_done[ddir]; |
581e7141 | 2050 | |
0d593542 | 2051 | return ret; |
97601024 | 2052 | } |
7e77dd02 JA |
2053 | |
2054 | /* | |
2055 | * Call when io_u is really queued, to update the submission latency. | |
2056 | */ | |
2057 | void io_u_queued(struct thread_data *td, struct io_u *io_u) | |
2058 | { | |
8243be59 | 2059 | if (!td->o.disable_slat && ramp_time_over(td) && td->o.stats) { |
9520ebb9 | 2060 | unsigned long slat_time; |
7e77dd02 | 2061 | |
d6bb626e | 2062 | slat_time = ntime_since(&io_u->start_time, &io_u->issue_time); |
75dc383e JA |
2063 | |
2064 | if (td->parent) | |
2065 | td = td->parent; | |
2066 | ||
ae588852 JA |
2067 | add_slat_sample(td, io_u->ddir, slat_time, io_u->xfer_buflen, |
2068 | io_u->offset); | |
9520ebb9 | 2069 | } |
7e77dd02 | 2070 | } |
433afcb4 | 2071 | |
5c94b008 JA |
2072 | /* |
2073 | * See if we should reuse the last seed, if dedupe is enabled | |
2074 | */ | |
9451b93e | 2075 | static struct frand_state *get_buf_state(struct thread_data *td) |
5c94b008 JA |
2076 | { |
2077 | unsigned int v; | |
5c94b008 JA |
2078 | |
2079 | if (!td->o.dedupe_percentage) | |
2080 | return &td->buf_state; | |
732eedd0 | 2081 | else if (td->o.dedupe_percentage == 100) { |
9451b93e JA |
2082 | frand_copy(&td->buf_state_prev, &td->buf_state); |
2083 | return &td->buf_state; | |
732eedd0 | 2084 | } |
5c94b008 | 2085 | |
36dd3379 | 2086 | v = rand32_between(&td->dedupe_state, 1, 100); |
5c94b008 JA |
2087 | |
2088 | if (v <= td->o.dedupe_percentage) | |
2089 | return &td->buf_state_prev; | |
2090 | ||
2091 | return &td->buf_state; | |
2092 | } | |
2093 | ||
9451b93e | 2094 | static void save_buf_state(struct thread_data *td, struct frand_state *rs) |
5c94b008 | 2095 | { |
9451b93e JA |
2096 | if (td->o.dedupe_percentage == 100) |
2097 | frand_copy(rs, &td->buf_state_prev); | |
2098 | else if (rs == &td->buf_state) | |
5c94b008 JA |
2099 | frand_copy(&td->buf_state_prev, rs); |
2100 | } | |
2101 | ||
cc86c395 JA |
2102 | void fill_io_buffer(struct thread_data *td, void *buf, unsigned int min_write, |
2103 | unsigned int max_bs) | |
5973cafb | 2104 | { |
d1af2894 JA |
2105 | struct thread_options *o = &td->o; |
2106 | ||
15600335 JA |
2107 | if (o->mem_type == MEM_CUDA_MALLOC) |
2108 | return; | |
03553853 | 2109 | |
4eff3e57 | 2110 | if (o->compress_percentage || o->dedupe_percentage) { |
9c42684e | 2111 | unsigned int perc = td->o.compress_percentage; |
5c94b008 | 2112 | struct frand_state *rs; |
1066358a | 2113 | unsigned int left = max_bs; |
1e7f82e2 | 2114 | unsigned int this_write; |
5c94b008 | 2115 | |
1066358a | 2116 | do { |
9451b93e | 2117 | rs = get_buf_state(td); |
9c42684e | 2118 | |
1066358a | 2119 | min_write = min(min_write, left); |
f97a43a1 | 2120 | |
1066358a | 2121 | if (perc) { |
4b157ac6 JA |
2122 | this_write = min_not_zero(min_write, |
2123 | td->o.compress_chunk); | |
1e7f82e2 JA |
2124 | |
2125 | fill_random_buf_percentage(rs, buf, perc, | |
2126 | this_write, this_write, | |
2127 | o->buffer_pattern, | |
2128 | o->buffer_pattern_bytes); | |
2129 | } else { | |
1066358a | 2130 | fill_random_buf(rs, buf, min_write); |
1e7f82e2 JA |
2131 | this_write = min_write; |
2132 | } | |
1066358a | 2133 | |
1e7f82e2 JA |
2134 | buf += this_write; |
2135 | left -= this_write; | |
9451b93e | 2136 | save_buf_state(td, rs); |
1066358a | 2137 | } while (left); |
d1af2894 JA |
2138 | } else if (o->buffer_pattern_bytes) |
2139 | fill_buffer_pattern(td, buf, max_bs); | |
999d245e | 2140 | else if (o->zero_buffers) |
cc86c395 | 2141 | memset(buf, 0, max_bs); |
999d245e | 2142 | else |
9451b93e | 2143 | fill_random_buf(get_buf_state(td), buf, max_bs); |
cc86c395 JA |
2144 | } |
2145 | ||
2146 | /* | |
2147 | * "randomly" fill the buffer contents | |
2148 | */ | |
2149 | void io_u_fill_buffer(struct thread_data *td, struct io_u *io_u, | |
2150 | unsigned int min_write, unsigned int max_bs) | |
2151 | { | |
2152 | io_u->buf_filled_len = 0; | |
2153 | fill_io_buffer(td, io_u->buf, min_write, max_bs); | |
5973cafb | 2154 | } |
e2c75fc4 TK |
2155 | |
2156 | static int do_sync_file_range(const struct thread_data *td, | |
2157 | struct fio_file *f) | |
2158 | { | |
2159 | off64_t offset, nbytes; | |
2160 | ||
2161 | offset = f->first_write; | |
2162 | nbytes = f->last_write - f->first_write; | |
2163 | ||
2164 | if (!nbytes) | |
2165 | return 0; | |
2166 | ||
2167 | return sync_file_range(f->fd, offset, nbytes, td->o.sync_file_range); | |
2168 | } | |
2169 | ||
2170 | int do_io_u_sync(const struct thread_data *td, struct io_u *io_u) | |
2171 | { | |
2172 | int ret; | |
2173 | ||
2174 | if (io_u->ddir == DDIR_SYNC) { | |
2175 | ret = fsync(io_u->file->fd); | |
2176 | } else if (io_u->ddir == DDIR_DATASYNC) { | |
2177 | #ifdef CONFIG_FDATASYNC | |
2178 | ret = fdatasync(io_u->file->fd); | |
2179 | #else | |
2180 | ret = io_u->xfer_buflen; | |
2181 | io_u->error = EINVAL; | |
2182 | #endif | |
2183 | } else if (io_u->ddir == DDIR_SYNC_FILE_RANGE) | |
2184 | ret = do_sync_file_range(td, io_u->file); | |
2185 | else { | |
2186 | ret = io_u->xfer_buflen; | |
2187 | io_u->error = EINVAL; | |
2188 | } | |
2189 | ||
2190 | if (ret < 0) | |
2191 | io_u->error = errno; | |
2192 | ||
2193 | return ret; | |
2194 | } | |
2195 | ||
2196 | int do_io_u_trim(const struct thread_data *td, struct io_u *io_u) | |
2197 | { | |
2198 | #ifndef FIO_HAVE_TRIM | |
2199 | io_u->error = EINVAL; | |
2200 | return 0; | |
2201 | #else | |
2202 | struct fio_file *f = io_u->file; | |
2203 | int ret; | |
2204 | ||
496b1f9e | 2205 | ret = os_trim(f, io_u->offset, io_u->xfer_buflen); |
e2c75fc4 TK |
2206 | if (!ret) |
2207 | return io_u->xfer_buflen; | |
2208 | ||
2209 | io_u->error = ret; | |
2210 | return 0; | |
2211 | #endif | |
2212 | } |