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" |
10ba535a | 14 | |
97601024 JA |
15 | struct io_completion_data { |
16 | int nr; /* input */ | |
97601024 JA |
17 | |
18 | int error; /* output */ | |
100f49f1 | 19 | uint64_t bytes_done[DDIR_RWDIR_CNT]; /* output */ |
97601024 JA |
20 | struct timeval time; /* output */ |
21 | }; | |
22 | ||
10ba535a | 23 | /* |
7ebd796f | 24 | * The ->io_axmap contains a map of blocks we have or have not done io |
10ba535a JA |
25 | * to yet. Used to make sure we cover the entire range in a fair fashion. |
26 | */ | |
1ae83d45 | 27 | static int random_map_free(struct fio_file *f, const uint64_t block) |
10ba535a | 28 | { |
7ebd796f | 29 | return !axmap_isset(f->io_axmap, block); |
10ba535a JA |
30 | } |
31 | ||
df415585 JA |
32 | /* |
33 | * Mark a given offset as used in the map. | |
34 | */ | |
9bf2061e | 35 | static void mark_random_map(struct thread_data *td, struct io_u *io_u) |
df415585 | 36 | { |
2dc1bbeb | 37 | unsigned int min_bs = td->o.rw_min_bs; |
9bf2061e | 38 | struct fio_file *f = io_u->file; |
51ede0b1 | 39 | unsigned int nr_blocks; |
1ae83d45 | 40 | uint64_t block; |
df415585 | 41 | |
1ae83d45 | 42 | block = (io_u->offset - f->file_offset) / (uint64_t) min_bs; |
c685b5b2 | 43 | nr_blocks = (io_u->buflen + min_bs - 1) / min_bs; |
df415585 | 44 | |
2ab9e98b | 45 | if (!(io_u->flags & IO_U_F_BUSY_OK)) |
7ebd796f | 46 | nr_blocks = axmap_set_nr(f->io_axmap, block, nr_blocks); |
df415585 | 47 | |
51ede0b1 JA |
48 | if ((nr_blocks * min_bs) < io_u->buflen) |
49 | io_u->buflen = nr_blocks * min_bs; | |
df415585 JA |
50 | } |
51 | ||
74776733 JA |
52 | static uint64_t last_block(struct thread_data *td, struct fio_file *f, |
53 | enum fio_ddir ddir) | |
2ba1c290 | 54 | { |
74776733 JA |
55 | uint64_t max_blocks; |
56 | uint64_t max_size; | |
2ba1c290 | 57 | |
ff58fced JA |
58 | assert(ddir_rw(ddir)); |
59 | ||
d9dd70f7 JA |
60 | /* |
61 | * Hmm, should we make sure that ->io_size <= ->real_file_size? | |
62 | */ | |
63 | max_size = f->io_size; | |
64 | if (max_size > f->real_file_size) | |
65 | max_size = f->real_file_size; | |
66 | ||
ed335855 SN |
67 | if (td->o.zone_range) |
68 | max_size = td->o.zone_range; | |
69 | ||
1ae83d45 | 70 | max_blocks = max_size / (uint64_t) td->o.ba[ddir]; |
2ba1c290 JA |
71 | if (!max_blocks) |
72 | return 0; | |
73 | ||
67778e88 | 74 | return max_blocks; |
2ba1c290 JA |
75 | } |
76 | ||
1ae83d45 JA |
77 | struct rand_off { |
78 | struct flist_head list; | |
79 | uint64_t off; | |
80 | }; | |
81 | ||
e25839d4 | 82 | static int __get_next_rand_offset(struct thread_data *td, struct fio_file *f, |
1ae83d45 | 83 | enum fio_ddir ddir, uint64_t *b) |
ec4015da | 84 | { |
1ae83d45 | 85 | uint64_t r, lastb; |
ec4015da | 86 | |
74776733 JA |
87 | lastb = last_block(td, f, ddir); |
88 | if (!lastb) | |
89 | return 1; | |
5e0baa7f | 90 | |
74776733 | 91 | if (td->o.random_generator == FIO_RAND_GEN_TAUSWORTHE) { |
1ae83d45 | 92 | uint64_t rmax; |
2615cc4b | 93 | |
8055e41d JA |
94 | rmax = td->o.use_os_rand ? OS_RAND_MAX : FRAND_MAX; |
95 | ||
96 | if (td->o.use_os_rand) { | |
97 | rmax = OS_RAND_MAX; | |
98 | r = os_random_long(&td->random_state); | |
99 | } else { | |
100 | rmax = FRAND_MAX; | |
101 | r = __rand(&td->__random_state); | |
102 | } | |
103 | ||
104 | dprint(FD_RANDOM, "off rand %llu\n", r); | |
105 | ||
1ae83d45 | 106 | *b = (lastb - 1) * (r / ((uint64_t) rmax + 1.0)); |
51ede0b1 | 107 | } else { |
8055e41d | 108 | uint64_t off = 0; |
43c63a78 | 109 | |
74776733 | 110 | if (lfsr_next(&f->lfsr, &off, lastb)) |
8055e41d | 111 | return 1; |
ec4015da | 112 | |
8055e41d JA |
113 | *b = off; |
114 | } | |
0ce8b119 | 115 | |
ec4015da | 116 | /* |
51ede0b1 | 117 | * if we are not maintaining a random map, we are done. |
ec4015da | 118 | */ |
51ede0b1 JA |
119 | if (!file_randommap(td, f)) |
120 | goto ret; | |
43c63a78 JA |
121 | |
122 | /* | |
51ede0b1 | 123 | * calculate map offset and check if it's free |
43c63a78 | 124 | */ |
51ede0b1 JA |
125 | if (random_map_free(f, *b)) |
126 | goto ret; | |
127 | ||
128 | dprint(FD_RANDOM, "get_next_rand_offset: offset %llu busy\n", *b); | |
129 | ||
7ebd796f | 130 | *b = axmap_next_free(f->io_axmap, *b); |
51ede0b1 JA |
131 | if (*b == (uint64_t) -1ULL) |
132 | return 1; | |
0ce8b119 JA |
133 | ret: |
134 | return 0; | |
ec4015da JA |
135 | } |
136 | ||
925fee33 JA |
137 | static int __get_next_rand_offset_zipf(struct thread_data *td, |
138 | struct fio_file *f, enum fio_ddir ddir, | |
1ae83d45 | 139 | uint64_t *b) |
e25839d4 | 140 | { |
9c6f6316 | 141 | *b = zipf_next(&f->zipf); |
e25839d4 JA |
142 | return 0; |
143 | } | |
144 | ||
925fee33 JA |
145 | static int __get_next_rand_offset_pareto(struct thread_data *td, |
146 | struct fio_file *f, enum fio_ddir ddir, | |
1ae83d45 | 147 | uint64_t *b) |
925fee33 | 148 | { |
9c6f6316 | 149 | *b = pareto_next(&f->zipf); |
925fee33 JA |
150 | return 0; |
151 | } | |
152 | ||
1ae83d45 JA |
153 | static int flist_cmp(void *data, struct flist_head *a, struct flist_head *b) |
154 | { | |
155 | struct rand_off *r1 = flist_entry(a, struct rand_off, list); | |
156 | struct rand_off *r2 = flist_entry(b, struct rand_off, list); | |
157 | ||
158 | return r1->off - r2->off; | |
159 | } | |
160 | ||
161 | static int get_off_from_method(struct thread_data *td, struct fio_file *f, | |
162 | enum fio_ddir ddir, uint64_t *b) | |
e25839d4 JA |
163 | { |
164 | if (td->o.random_distribution == FIO_RAND_DIST_RANDOM) | |
165 | return __get_next_rand_offset(td, f, ddir, b); | |
166 | else if (td->o.random_distribution == FIO_RAND_DIST_ZIPF) | |
167 | return __get_next_rand_offset_zipf(td, f, ddir, b); | |
925fee33 JA |
168 | else if (td->o.random_distribution == FIO_RAND_DIST_PARETO) |
169 | return __get_next_rand_offset_pareto(td, f, ddir, b); | |
e25839d4 JA |
170 | |
171 | log_err("fio: unknown random distribution: %d\n", td->o.random_distribution); | |
172 | return 1; | |
173 | } | |
174 | ||
bcd5abfa JA |
175 | /* |
176 | * Sort the reads for a verify phase in batches of verifysort_nr, if | |
177 | * specified. | |
178 | */ | |
179 | static inline int should_sort_io(struct thread_data *td) | |
180 | { | |
181 | if (!td->o.verifysort_nr || !td->o.do_verify) | |
182 | return 0; | |
183 | if (!td_random(td)) | |
184 | return 0; | |
185 | if (td->runstate != TD_VERIFYING) | |
186 | return 0; | |
187 | if (td->o.random_generator == FIO_RAND_GEN_TAUSWORTHE) | |
188 | return 0; | |
189 | ||
190 | return 1; | |
191 | } | |
192 | ||
1ae83d45 JA |
193 | static int get_next_rand_offset(struct thread_data *td, struct fio_file *f, |
194 | enum fio_ddir ddir, uint64_t *b) | |
195 | { | |
196 | struct rand_off *r; | |
197 | int i, ret = 1; | |
198 | ||
bcd5abfa | 199 | if (!should_sort_io(td)) |
1ae83d45 JA |
200 | return get_off_from_method(td, f, ddir, b); |
201 | ||
202 | if (!flist_empty(&td->next_rand_list)) { | |
203 | struct rand_off *r; | |
204 | fetch: | |
205 | r = flist_entry(td->next_rand_list.next, struct rand_off, list); | |
206 | flist_del(&r->list); | |
207 | *b = r->off; | |
208 | free(r); | |
209 | return 0; | |
210 | } | |
211 | ||
212 | for (i = 0; i < td->o.verifysort_nr; i++) { | |
213 | r = malloc(sizeof(*r)); | |
214 | ||
215 | ret = get_off_from_method(td, f, ddir, &r->off); | |
216 | if (ret) { | |
217 | free(r); | |
218 | break; | |
219 | } | |
220 | ||
221 | flist_add(&r->list, &td->next_rand_list); | |
222 | } | |
223 | ||
224 | if (ret && !i) | |
225 | return ret; | |
226 | ||
227 | assert(!flist_empty(&td->next_rand_list)); | |
228 | flist_sort(NULL, &td->next_rand_list, flist_cmp); | |
229 | goto fetch; | |
230 | } | |
231 | ||
38dad62d | 232 | static int get_next_rand_block(struct thread_data *td, struct fio_file *f, |
1ae83d45 | 233 | enum fio_ddir ddir, uint64_t *b) |
38dad62d | 234 | { |
c04e4661 DE |
235 | if (!get_next_rand_offset(td, f, ddir, b)) |
236 | return 0; | |
237 | ||
238 | if (td->o.time_based) { | |
33c48814 | 239 | fio_file_reset(td, f); |
c04e4661 DE |
240 | if (!get_next_rand_offset(td, f, ddir, b)) |
241 | return 0; | |
38dad62d JA |
242 | } |
243 | ||
c04e4661 DE |
244 | dprint(FD_IO, "%s: rand offset failed, last=%llu, size=%llu\n", |
245 | f->file_name, f->last_pos, f->real_file_size); | |
246 | return 1; | |
38dad62d JA |
247 | } |
248 | ||
37cf9e3c | 249 | static int get_next_seq_offset(struct thread_data *td, struct fio_file *f, |
1ae83d45 | 250 | enum fio_ddir ddir, uint64_t *offset) |
38dad62d | 251 | { |
ff58fced JA |
252 | assert(ddir_rw(ddir)); |
253 | ||
ce95d651 | 254 | if (f->last_pos >= f->io_size + get_start_offset(td) && td->o.time_based) |
c04e4661 DE |
255 | f->last_pos = f->last_pos - f->io_size; |
256 | ||
38dad62d | 257 | if (f->last_pos < f->real_file_size) { |
1ae83d45 | 258 | uint64_t pos; |
059b0802 | 259 | |
a66da7a2 JA |
260 | if (f->last_pos == f->file_offset && td->o.ddir_seq_add < 0) |
261 | f->last_pos = f->real_file_size; | |
262 | ||
263 | pos = f->last_pos - f->file_offset; | |
059b0802 JA |
264 | if (pos) |
265 | pos += td->o.ddir_seq_add; | |
266 | ||
37cf9e3c | 267 | *offset = pos; |
38dad62d JA |
268 | return 0; |
269 | } | |
270 | ||
271 | return 1; | |
272 | } | |
273 | ||
274 | static int get_next_block(struct thread_data *td, struct io_u *io_u, | |
37cf9e3c | 275 | enum fio_ddir ddir, int rw_seq) |
38dad62d JA |
276 | { |
277 | struct fio_file *f = io_u->file; | |
1ae83d45 | 278 | uint64_t b, offset; |
38dad62d JA |
279 | int ret; |
280 | ||
ff58fced JA |
281 | assert(ddir_rw(ddir)); |
282 | ||
37cf9e3c JA |
283 | b = offset = -1ULL; |
284 | ||
38dad62d JA |
285 | if (rw_seq) { |
286 | if (td_random(td)) | |
37cf9e3c | 287 | ret = get_next_rand_block(td, f, ddir, &b); |
38dad62d | 288 | else |
37cf9e3c | 289 | ret = get_next_seq_offset(td, f, ddir, &offset); |
38dad62d JA |
290 | } else { |
291 | io_u->flags |= IO_U_F_BUSY_OK; | |
292 | ||
293 | if (td->o.rw_seq == RW_SEQ_SEQ) { | |
37cf9e3c | 294 | ret = get_next_seq_offset(td, f, ddir, &offset); |
38dad62d | 295 | if (ret) |
37cf9e3c | 296 | ret = get_next_rand_block(td, f, ddir, &b); |
38dad62d JA |
297 | } else if (td->o.rw_seq == RW_SEQ_IDENT) { |
298 | if (f->last_start != -1ULL) | |
37cf9e3c | 299 | offset = f->last_start - f->file_offset; |
38dad62d | 300 | else |
37cf9e3c | 301 | offset = 0; |
38dad62d JA |
302 | ret = 0; |
303 | } else { | |
304 | log_err("fio: unknown rw_seq=%d\n", td->o.rw_seq); | |
305 | ret = 1; | |
306 | } | |
307 | } | |
6d68b997 | 308 | |
37cf9e3c JA |
309 | if (!ret) { |
310 | if (offset != -1ULL) | |
311 | io_u->offset = offset; | |
312 | else if (b != -1ULL) | |
313 | io_u->offset = b * td->o.ba[ddir]; | |
314 | else { | |
51ede0b1 JA |
315 | log_err("fio: bug in offset generation: offset=%llu, b=%llu\n", |
316 | offset, b); | |
37cf9e3c JA |
317 | ret = 1; |
318 | } | |
319 | } | |
320 | ||
38dad62d JA |
321 | return ret; |
322 | } | |
323 | ||
10ba535a JA |
324 | /* |
325 | * For random io, generate a random new block and see if it's used. Repeat | |
326 | * until we find a free one. For sequential io, just return the end of | |
327 | * the last io issued. | |
328 | */ | |
15dc1934 | 329 | static int __get_next_offset(struct thread_data *td, struct io_u *io_u) |
10ba535a | 330 | { |
9bf2061e | 331 | struct fio_file *f = io_u->file; |
4ba66134 | 332 | enum fio_ddir ddir = io_u->ddir; |
38dad62d | 333 | int rw_seq_hit = 0; |
10ba535a | 334 | |
ff58fced JA |
335 | assert(ddir_rw(ddir)); |
336 | ||
38dad62d JA |
337 | if (td->o.ddir_seq_nr && !--td->ddir_seq_nr) { |
338 | rw_seq_hit = 1; | |
5736c10d | 339 | td->ddir_seq_nr = td->o.ddir_seq_nr; |
38dad62d | 340 | } |
211097b2 | 341 | |
37cf9e3c | 342 | if (get_next_block(td, io_u, ddir, rw_seq_hit)) |
38dad62d | 343 | return 1; |
10ba535a | 344 | |
009bd847 JA |
345 | if (io_u->offset >= f->io_size) { |
346 | dprint(FD_IO, "get_next_offset: offset %llu >= io_size %llu\n", | |
347 | io_u->offset, f->io_size); | |
348 | return 1; | |
349 | } | |
350 | ||
351 | io_u->offset += f->file_offset; | |
2ba1c290 JA |
352 | if (io_u->offset >= f->real_file_size) { |
353 | dprint(FD_IO, "get_next_offset: offset %llu >= size %llu\n", | |
354 | io_u->offset, f->real_file_size); | |
10ba535a | 355 | return 1; |
2ba1c290 | 356 | } |
10ba535a JA |
357 | |
358 | return 0; | |
359 | } | |
360 | ||
15dc1934 JA |
361 | static int get_next_offset(struct thread_data *td, struct io_u *io_u) |
362 | { | |
d72be545 JA |
363 | if (td->flags & TD_F_PROFILE_OPS) { |
364 | struct prof_io_ops *ops = &td->prof_io_ops; | |
7eb36574 | 365 | |
d72be545 JA |
366 | if (ops->fill_io_u_off) |
367 | return ops->fill_io_u_off(td, io_u); | |
368 | } | |
15dc1934 JA |
369 | |
370 | return __get_next_offset(td, io_u); | |
371 | } | |
372 | ||
79944128 JA |
373 | static inline int io_u_fits(struct thread_data *td, struct io_u *io_u, |
374 | unsigned int buflen) | |
375 | { | |
376 | struct fio_file *f = io_u->file; | |
377 | ||
ce95d651 | 378 | return io_u->offset + buflen <= f->io_size + get_start_offset(td); |
79944128 JA |
379 | } |
380 | ||
15dc1934 | 381 | static unsigned int __get_next_buflen(struct thread_data *td, struct io_u *io_u) |
10ba535a | 382 | { |
bca4ed4d | 383 | const int ddir = io_u->ddir; |
24d23ca7 | 384 | unsigned int buflen = 0; |
f3059de1 | 385 | unsigned int minbs, maxbs; |
1294c3ec | 386 | unsigned long r, rand_max; |
10ba535a | 387 | |
ff58fced JA |
388 | assert(ddir_rw(ddir)); |
389 | ||
f3059de1 JA |
390 | minbs = td->o.min_bs[ddir]; |
391 | maxbs = td->o.max_bs[ddir]; | |
392 | ||
79944128 JA |
393 | if (minbs == maxbs) |
394 | return minbs; | |
395 | ||
52c58027 JA |
396 | /* |
397 | * If we can't satisfy the min block size from here, then fail | |
398 | */ | |
399 | if (!io_u_fits(td, io_u, minbs)) | |
400 | return 0; | |
401 | ||
4c07ad86 JA |
402 | if (td->o.use_os_rand) |
403 | rand_max = OS_RAND_MAX; | |
404 | else | |
405 | rand_max = FRAND_MAX; | |
406 | ||
79944128 | 407 | do { |
4c07ad86 JA |
408 | if (td->o.use_os_rand) |
409 | r = os_random_long(&td->bsrange_state); | |
410 | else | |
411 | r = __rand(&td->__bsrange_state); | |
412 | ||
720e84ad | 413 | if (!td->o.bssplit_nr[ddir]) { |
f3059de1 | 414 | buflen = 1 + (unsigned int) ((double) maxbs * |
4c07ad86 | 415 | (r / (rand_max + 1.0))); |
f3059de1 JA |
416 | if (buflen < minbs) |
417 | buflen = minbs; | |
5ec10eaa | 418 | } else { |
564ca972 JA |
419 | long perc = 0; |
420 | unsigned int i; | |
421 | ||
720e84ad JA |
422 | for (i = 0; i < td->o.bssplit_nr[ddir]; i++) { |
423 | struct bssplit *bsp = &td->o.bssplit[ddir][i]; | |
564ca972 JA |
424 | |
425 | buflen = bsp->bs; | |
426 | perc += bsp->perc; | |
79944128 JA |
427 | if ((r <= ((rand_max / 100L) * perc)) && |
428 | io_u_fits(td, io_u, buflen)) | |
564ca972 JA |
429 | break; |
430 | } | |
431 | } | |
79944128 | 432 | |
f3059de1 JA |
433 | if (!td->o.bs_unaligned && is_power_of_2(minbs)) |
434 | buflen = (buflen + minbs - 1) & ~(minbs - 1); | |
10ba535a | 435 | |
79944128 | 436 | } while (!io_u_fits(td, io_u, buflen)); |
6a5e6884 | 437 | |
10ba535a JA |
438 | return buflen; |
439 | } | |
440 | ||
15dc1934 JA |
441 | static unsigned int get_next_buflen(struct thread_data *td, struct io_u *io_u) |
442 | { | |
d72be545 JA |
443 | if (td->flags & TD_F_PROFILE_OPS) { |
444 | struct prof_io_ops *ops = &td->prof_io_ops; | |
7eb36574 | 445 | |
d72be545 JA |
446 | if (ops->fill_io_u_size) |
447 | return ops->fill_io_u_size(td, io_u); | |
448 | } | |
15dc1934 JA |
449 | |
450 | return __get_next_buflen(td, io_u); | |
451 | } | |
452 | ||
afe24a5a JA |
453 | static void set_rwmix_bytes(struct thread_data *td) |
454 | { | |
afe24a5a JA |
455 | unsigned int diff; |
456 | ||
457 | /* | |
458 | * we do time or byte based switch. this is needed because | |
459 | * buffered writes may issue a lot quicker than they complete, | |
460 | * whereas reads do not. | |
461 | */ | |
e47f799f | 462 | diff = td->o.rwmix[td->rwmix_ddir ^ 1]; |
04c540d9 | 463 | td->rwmix_issues = (td->io_issues[td->rwmix_ddir] * diff) / 100; |
e47f799f JA |
464 | } |
465 | ||
466 | static inline enum fio_ddir get_rand_ddir(struct thread_data *td) | |
467 | { | |
468 | unsigned int v; | |
1294c3ec | 469 | unsigned long r; |
e47f799f | 470 | |
4c07ad86 JA |
471 | if (td->o.use_os_rand) { |
472 | r = os_random_long(&td->rwmix_state); | |
473 | v = 1 + (int) (100.0 * (r / (OS_RAND_MAX + 1.0))); | |
474 | } else { | |
475 | r = __rand(&td->__rwmix_state); | |
476 | v = 1 + (int) (100.0 * (r / (FRAND_MAX + 1.0))); | |
477 | } | |
478 | ||
04c540d9 | 479 | if (v <= td->o.rwmix[DDIR_READ]) |
e47f799f JA |
480 | return DDIR_READ; |
481 | ||
482 | return DDIR_WRITE; | |
afe24a5a JA |
483 | } |
484 | ||
581e7141 JA |
485 | static enum fio_ddir rate_ddir(struct thread_data *td, enum fio_ddir ddir) |
486 | { | |
487 | enum fio_ddir odir = ddir ^ 1; | |
488 | struct timeval t; | |
489 | long usec; | |
490 | ||
ff58fced JA |
491 | assert(ddir_rw(ddir)); |
492 | ||
315fcfec | 493 | if (td->rate_pending_usleep[ddir] <= 0) |
581e7141 JA |
494 | return ddir; |
495 | ||
496 | /* | |
497 | * We have too much pending sleep in this direction. See if we | |
498 | * should switch. | |
499 | */ | |
315fcfec | 500 | if (td_rw(td) && td->o.rwmix[odir]) { |
581e7141 JA |
501 | /* |
502 | * Other direction does not have too much pending, switch | |
503 | */ | |
504 | if (td->rate_pending_usleep[odir] < 100000) | |
505 | return odir; | |
506 | ||
507 | /* | |
508 | * Both directions have pending sleep. Sleep the minimum time | |
509 | * and deduct from both. | |
510 | */ | |
511 | if (td->rate_pending_usleep[ddir] <= | |
512 | td->rate_pending_usleep[odir]) { | |
513 | usec = td->rate_pending_usleep[ddir]; | |
514 | } else { | |
515 | usec = td->rate_pending_usleep[odir]; | |
516 | ddir = odir; | |
517 | } | |
518 | } else | |
519 | usec = td->rate_pending_usleep[ddir]; | |
520 | ||
78c1eda5 JA |
521 | /* |
522 | * We are going to sleep, ensure that we flush anything pending as | |
422f9e4b RM |
523 | * not to skew our latency numbers. |
524 | * | |
525 | * Changed to only monitor 'in flight' requests here instead of the | |
526 | * td->cur_depth, b/c td->cur_depth does not accurately represent | |
527 | * io's that have been actually submitted to an async engine, | |
528 | * and cur_depth is meaningless for sync engines. | |
78c1eda5 | 529 | */ |
422f9e4b | 530 | if (td->io_u_in_flight) { |
78c1eda5 JA |
531 | int fio_unused ret; |
532 | ||
422f9e4b | 533 | ret = io_u_queued_complete(td, td->io_u_in_flight, NULL); |
78c1eda5 JA |
534 | } |
535 | ||
581e7141 JA |
536 | fio_gettime(&t, NULL); |
537 | usec_sleep(td, usec); | |
538 | usec = utime_since_now(&t); | |
539 | ||
540 | td->rate_pending_usleep[ddir] -= usec; | |
541 | ||
542 | odir = ddir ^ 1; | |
543 | if (td_rw(td) && __should_check_rate(td, odir)) | |
544 | td->rate_pending_usleep[odir] -= usec; | |
0b9d69ec | 545 | |
6eaf09d6 SL |
546 | if (ddir_trim(ddir)) |
547 | return ddir; | |
e0224c6b | 548 | |
581e7141 JA |
549 | return ddir; |
550 | } | |
551 | ||
10ba535a JA |
552 | /* |
553 | * Return the data direction for the next io_u. If the job is a | |
554 | * mixed read/write workload, check the rwmix cycle and switch if | |
555 | * necessary. | |
556 | */ | |
1e97cce9 | 557 | static enum fio_ddir get_rw_ddir(struct thread_data *td) |
10ba535a | 558 | { |
581e7141 JA |
559 | enum fio_ddir ddir; |
560 | ||
5f9099ea JA |
561 | /* |
562 | * see if it's time to fsync | |
563 | */ | |
564 | if (td->o.fsync_blocks && | |
565 | !(td->io_issues[DDIR_WRITE] % td->o.fsync_blocks) && | |
566 | td->io_issues[DDIR_WRITE] && should_fsync(td)) | |
567 | return DDIR_SYNC; | |
568 | ||
569 | /* | |
570 | * see if it's time to fdatasync | |
571 | */ | |
572 | if (td->o.fdatasync_blocks && | |
573 | !(td->io_issues[DDIR_WRITE] % td->o.fdatasync_blocks) && | |
574 | td->io_issues[DDIR_WRITE] && should_fsync(td)) | |
575 | return DDIR_DATASYNC; | |
576 | ||
44f29692 JA |
577 | /* |
578 | * see if it's time to sync_file_range | |
579 | */ | |
580 | if (td->sync_file_range_nr && | |
581 | !(td->io_issues[DDIR_WRITE] % td->sync_file_range_nr) && | |
582 | td->io_issues[DDIR_WRITE] && should_fsync(td)) | |
583 | return DDIR_SYNC_FILE_RANGE; | |
584 | ||
10ba535a | 585 | if (td_rw(td)) { |
10ba535a JA |
586 | /* |
587 | * Check if it's time to seed a new data direction. | |
588 | */ | |
e4928662 | 589 | if (td->io_issues[td->rwmix_ddir] >= td->rwmix_issues) { |
e47f799f JA |
590 | /* |
591 | * Put a top limit on how many bytes we do for | |
592 | * one data direction, to avoid overflowing the | |
593 | * ranges too much | |
594 | */ | |
595 | ddir = get_rand_ddir(td); | |
e47f799f JA |
596 | |
597 | if (ddir != td->rwmix_ddir) | |
598 | set_rwmix_bytes(td); | |
599 | ||
600 | td->rwmix_ddir = ddir; | |
10ba535a | 601 | } |
581e7141 | 602 | ddir = td->rwmix_ddir; |
10ba535a | 603 | } else if (td_read(td)) |
581e7141 | 604 | ddir = DDIR_READ; |
6eaf09d6 | 605 | else if (td_write(td)) |
581e7141 | 606 | ddir = DDIR_WRITE; |
6eaf09d6 SL |
607 | else |
608 | ddir = DDIR_TRIM; | |
581e7141 JA |
609 | |
610 | td->rwmix_ddir = rate_ddir(td, ddir); | |
611 | return td->rwmix_ddir; | |
10ba535a JA |
612 | } |
613 | ||
1ef2b6be JA |
614 | static void set_rw_ddir(struct thread_data *td, struct io_u *io_u) |
615 | { | |
bcd5abfa | 616 | io_u->ddir = io_u->acct_ddir = get_rw_ddir(td); |
1ef2b6be JA |
617 | |
618 | if (io_u->ddir == DDIR_WRITE && (td->io_ops->flags & FIO_BARRIER) && | |
619 | td->o.barrier_blocks && | |
620 | !(td->io_issues[DDIR_WRITE] % td->o.barrier_blocks) && | |
621 | td->io_issues[DDIR_WRITE]) | |
622 | io_u->flags |= IO_U_F_BARRIER; | |
623 | } | |
624 | ||
e8462bd8 | 625 | void put_file_log(struct thread_data *td, struct fio_file *f) |
60f2c658 JA |
626 | { |
627 | int ret = put_file(td, f); | |
628 | ||
629 | if (ret) | |
630 | td_verror(td, ret, "file close"); | |
631 | } | |
632 | ||
10ba535a JA |
633 | void put_io_u(struct thread_data *td, struct io_u *io_u) |
634 | { | |
e8462bd8 JA |
635 | td_io_u_lock(td); |
636 | ||
d7ee2a7d | 637 | if (io_u->file && !(io_u->flags & IO_U_F_FREE_DEF)) |
60f2c658 | 638 | put_file_log(td, io_u->file); |
10ba535a | 639 | io_u->file = NULL; |
d7ee2a7d SL |
640 | io_u->flags &= ~IO_U_F_FREE_DEF; |
641 | io_u->flags |= IO_U_F_FREE; | |
642 | ||
0c41214f RR |
643 | if (io_u->flags & IO_U_F_IN_CUR_DEPTH) |
644 | td->cur_depth--; | |
e8462bd8 | 645 | flist_del_init(&io_u->list); |
01743ee1 | 646 | flist_add(&io_u->list, &td->io_u_freelist); |
e8462bd8 JA |
647 | td_io_u_unlock(td); |
648 | td_io_u_free_notify(td); | |
10ba535a JA |
649 | } |
650 | ||
f2bba182 RR |
651 | void clear_io_u(struct thread_data *td, struct io_u *io_u) |
652 | { | |
653 | io_u->flags &= ~IO_U_F_FLIGHT; | |
654 | put_io_u(td, io_u); | |
655 | } | |
656 | ||
755200a3 JA |
657 | void requeue_io_u(struct thread_data *td, struct io_u **io_u) |
658 | { | |
659 | struct io_u *__io_u = *io_u; | |
bcd5abfa | 660 | enum fio_ddir ddir = acct_ddir(__io_u); |
755200a3 | 661 | |
465221b0 JA |
662 | dprint(FD_IO, "requeue %p\n", __io_u); |
663 | ||
e8462bd8 JA |
664 | td_io_u_lock(td); |
665 | ||
4d2e0f49 | 666 | __io_u->flags |= IO_U_F_FREE; |
bcd5abfa JA |
667 | if ((__io_u->flags & IO_U_F_FLIGHT) && ddir_rw(ddir)) |
668 | td->io_issues[ddir]--; | |
5ec10eaa | 669 | |
4d2e0f49 | 670 | __io_u->flags &= ~IO_U_F_FLIGHT; |
0c41214f RR |
671 | if (__io_u->flags & IO_U_F_IN_CUR_DEPTH) |
672 | td->cur_depth--; | |
01743ee1 JA |
673 | flist_del(&__io_u->list); |
674 | flist_add_tail(&__io_u->list, &td->io_u_requeues); | |
e8462bd8 | 675 | td_io_u_unlock(td); |
755200a3 JA |
676 | *io_u = NULL; |
677 | } | |
678 | ||
9bf2061e | 679 | static int fill_io_u(struct thread_data *td, struct io_u *io_u) |
10ba535a | 680 | { |
b4c5e1ac JA |
681 | if (td->io_ops->flags & FIO_NOIO) |
682 | goto out; | |
683 | ||
1ef2b6be | 684 | set_rw_ddir(td, io_u); |
5f9099ea | 685 | |
87dc1ab1 | 686 | /* |
ff58fced | 687 | * fsync() or fdatasync() or trim etc, we are done |
87dc1ab1 | 688 | */ |
ff58fced | 689 | if (!ddir_rw(io_u->ddir)) |
c38e9468 | 690 | goto out; |
a00735e6 | 691 | |
48f5abd3 JA |
692 | /* |
693 | * See if it's time to switch to a new zone | |
694 | */ | |
13af05ae | 695 | if (td->zone_bytes >= td->o.zone_size && td->o.zone_skip) { |
48f5abd3 | 696 | td->zone_bytes = 0; |
ed335855 SN |
697 | io_u->file->file_offset += td->o.zone_range + td->o.zone_skip; |
698 | io_u->file->last_pos = io_u->file->file_offset; | |
48f5abd3 JA |
699 | td->io_skip_bytes += td->o.zone_skip; |
700 | } | |
701 | ||
10ba535a | 702 | /* |
c685b5b2 JA |
703 | * No log, let the seq/rand engine retrieve the next buflen and |
704 | * position. | |
10ba535a | 705 | */ |
2ba1c290 JA |
706 | if (get_next_offset(td, io_u)) { |
707 | dprint(FD_IO, "io_u %p, failed getting offset\n", io_u); | |
bca4ed4d | 708 | return 1; |
2ba1c290 | 709 | } |
10ba535a | 710 | |
9bf2061e | 711 | io_u->buflen = get_next_buflen(td, io_u); |
2ba1c290 JA |
712 | if (!io_u->buflen) { |
713 | dprint(FD_IO, "io_u %p, failed getting buflen\n", io_u); | |
bca4ed4d | 714 | return 1; |
2ba1c290 | 715 | } |
bca4ed4d | 716 | |
2ba1c290 JA |
717 | if (io_u->offset + io_u->buflen > io_u->file->real_file_size) { |
718 | dprint(FD_IO, "io_u %p, offset too large\n", io_u); | |
4ba66134 JA |
719 | dprint(FD_IO, " off=%llu/%lu > %llu\n", io_u->offset, |
720 | io_u->buflen, io_u->file->real_file_size); | |
6a5e6884 | 721 | return 1; |
2ba1c290 | 722 | } |
6a5e6884 | 723 | |
bca4ed4d JA |
724 | /* |
725 | * mark entry before potentially trimming io_u | |
726 | */ | |
303032ae | 727 | if (td_random(td) && file_randommap(td, io_u->file)) |
9bf2061e | 728 | mark_random_map(td, io_u); |
bca4ed4d | 729 | |
c38e9468 | 730 | out: |
2ba1c290 | 731 | dprint_io_u(io_u, "fill_io_u"); |
d9d91e39 | 732 | td->zone_bytes += io_u->buflen; |
bca4ed4d | 733 | return 0; |
10ba535a JA |
734 | } |
735 | ||
838bc709 JA |
736 | static void __io_u_mark_map(unsigned int *map, unsigned int nr) |
737 | { | |
2b13e716 | 738 | int idx = 0; |
838bc709 JA |
739 | |
740 | switch (nr) { | |
741 | default: | |
2b13e716 | 742 | idx = 6; |
838bc709 JA |
743 | break; |
744 | case 33 ... 64: | |
2b13e716 | 745 | idx = 5; |
838bc709 JA |
746 | break; |
747 | case 17 ... 32: | |
2b13e716 | 748 | idx = 4; |
838bc709 JA |
749 | break; |
750 | case 9 ... 16: | |
2b13e716 | 751 | idx = 3; |
838bc709 JA |
752 | break; |
753 | case 5 ... 8: | |
2b13e716 | 754 | idx = 2; |
838bc709 JA |
755 | break; |
756 | case 1 ... 4: | |
2b13e716 | 757 | idx = 1; |
838bc709 JA |
758 | case 0: |
759 | break; | |
760 | } | |
761 | ||
2b13e716 | 762 | map[idx]++; |
838bc709 JA |
763 | } |
764 | ||
765 | void io_u_mark_submit(struct thread_data *td, unsigned int nr) | |
766 | { | |
767 | __io_u_mark_map(td->ts.io_u_submit, nr); | |
768 | td->ts.total_submit++; | |
769 | } | |
770 | ||
771 | void io_u_mark_complete(struct thread_data *td, unsigned int nr) | |
772 | { | |
773 | __io_u_mark_map(td->ts.io_u_complete, nr); | |
774 | td->ts.total_complete++; | |
775 | } | |
776 | ||
d8005759 | 777 | void io_u_mark_depth(struct thread_data *td, unsigned int nr) |
71619dc2 | 778 | { |
2b13e716 | 779 | int idx = 0; |
71619dc2 JA |
780 | |
781 | switch (td->cur_depth) { | |
782 | default: | |
2b13e716 | 783 | idx = 6; |
a783e61a | 784 | break; |
71619dc2 | 785 | case 32 ... 63: |
2b13e716 | 786 | idx = 5; |
a783e61a | 787 | break; |
71619dc2 | 788 | case 16 ... 31: |
2b13e716 | 789 | idx = 4; |
a783e61a | 790 | break; |
71619dc2 | 791 | case 8 ... 15: |
2b13e716 | 792 | idx = 3; |
a783e61a | 793 | break; |
71619dc2 | 794 | case 4 ... 7: |
2b13e716 | 795 | idx = 2; |
a783e61a | 796 | break; |
71619dc2 | 797 | case 2 ... 3: |
2b13e716 | 798 | idx = 1; |
71619dc2 JA |
799 | case 1: |
800 | break; | |
801 | } | |
802 | ||
2b13e716 | 803 | td->ts.io_u_map[idx] += nr; |
71619dc2 JA |
804 | } |
805 | ||
04a0feae JA |
806 | static void io_u_mark_lat_usec(struct thread_data *td, unsigned long usec) |
807 | { | |
2b13e716 | 808 | int idx = 0; |
04a0feae JA |
809 | |
810 | assert(usec < 1000); | |
811 | ||
812 | switch (usec) { | |
813 | case 750 ... 999: | |
2b13e716 | 814 | idx = 9; |
04a0feae JA |
815 | break; |
816 | case 500 ... 749: | |
2b13e716 | 817 | idx = 8; |
04a0feae JA |
818 | break; |
819 | case 250 ... 499: | |
2b13e716 | 820 | idx = 7; |
04a0feae JA |
821 | break; |
822 | case 100 ... 249: | |
2b13e716 | 823 | idx = 6; |
04a0feae JA |
824 | break; |
825 | case 50 ... 99: | |
2b13e716 | 826 | idx = 5; |
04a0feae JA |
827 | break; |
828 | case 20 ... 49: | |
2b13e716 | 829 | idx = 4; |
04a0feae JA |
830 | break; |
831 | case 10 ... 19: | |
2b13e716 | 832 | idx = 3; |
04a0feae JA |
833 | break; |
834 | case 4 ... 9: | |
2b13e716 | 835 | idx = 2; |
04a0feae JA |
836 | break; |
837 | case 2 ... 3: | |
2b13e716 | 838 | idx = 1; |
04a0feae JA |
839 | case 0 ... 1: |
840 | break; | |
841 | } | |
842 | ||
2b13e716 JA |
843 | assert(idx < FIO_IO_U_LAT_U_NR); |
844 | td->ts.io_u_lat_u[idx]++; | |
04a0feae JA |
845 | } |
846 | ||
847 | static void io_u_mark_lat_msec(struct thread_data *td, unsigned long msec) | |
ec118304 | 848 | { |
2b13e716 | 849 | int idx = 0; |
ec118304 JA |
850 | |
851 | switch (msec) { | |
852 | default: | |
2b13e716 | 853 | idx = 11; |
04a0feae | 854 | break; |
8abdce66 | 855 | case 1000 ... 1999: |
2b13e716 | 856 | idx = 10; |
04a0feae | 857 | break; |
8abdce66 | 858 | case 750 ... 999: |
2b13e716 | 859 | idx = 9; |
04a0feae | 860 | break; |
8abdce66 | 861 | case 500 ... 749: |
2b13e716 | 862 | idx = 8; |
04a0feae | 863 | break; |
8abdce66 | 864 | case 250 ... 499: |
2b13e716 | 865 | idx = 7; |
04a0feae | 866 | break; |
8abdce66 | 867 | case 100 ... 249: |
2b13e716 | 868 | idx = 6; |
04a0feae | 869 | break; |
8abdce66 | 870 | case 50 ... 99: |
2b13e716 | 871 | idx = 5; |
04a0feae | 872 | break; |
8abdce66 | 873 | case 20 ... 49: |
2b13e716 | 874 | idx = 4; |
04a0feae | 875 | break; |
8abdce66 | 876 | case 10 ... 19: |
2b13e716 | 877 | idx = 3; |
04a0feae | 878 | break; |
8abdce66 | 879 | case 4 ... 9: |
2b13e716 | 880 | idx = 2; |
04a0feae | 881 | break; |
ec118304 | 882 | case 2 ... 3: |
2b13e716 | 883 | idx = 1; |
ec118304 JA |
884 | case 0 ... 1: |
885 | break; | |
886 | } | |
887 | ||
2b13e716 JA |
888 | assert(idx < FIO_IO_U_LAT_M_NR); |
889 | td->ts.io_u_lat_m[idx]++; | |
04a0feae JA |
890 | } |
891 | ||
892 | static void io_u_mark_latency(struct thread_data *td, unsigned long usec) | |
893 | { | |
894 | if (usec < 1000) | |
895 | io_u_mark_lat_usec(td, usec); | |
896 | else | |
897 | io_u_mark_lat_msec(td, usec / 1000); | |
ec118304 JA |
898 | } |
899 | ||
0aabe160 JA |
900 | /* |
901 | * Get next file to service by choosing one at random | |
902 | */ | |
2cc52930 JA |
903 | static struct fio_file *get_next_file_rand(struct thread_data *td, |
904 | enum fio_file_flags goodf, | |
d6aed795 | 905 | enum fio_file_flags badf) |
0aabe160 | 906 | { |
0aabe160 | 907 | struct fio_file *f; |
1c178180 | 908 | int fno; |
0aabe160 JA |
909 | |
910 | do { | |
87b10676 | 911 | int opened = 0; |
1294c3ec | 912 | unsigned long r; |
4c07ad86 JA |
913 | |
914 | if (td->o.use_os_rand) { | |
915 | r = os_random_long(&td->next_file_state); | |
916 | fno = (unsigned int) ((double) td->o.nr_files | |
917 | * (r / (OS_RAND_MAX + 1.0))); | |
918 | } else { | |
919 | r = __rand(&td->__next_file_state); | |
920 | fno = (unsigned int) ((double) td->o.nr_files | |
921 | * (r / (FRAND_MAX + 1.0))); | |
922 | } | |
7c83c089 | 923 | |
126d65c6 | 924 | f = td->files[fno]; |
d6aed795 | 925 | if (fio_file_done(f)) |
059e63c0 | 926 | continue; |
1c178180 | 927 | |
d6aed795 | 928 | if (!fio_file_open(f)) { |
87b10676 JA |
929 | int err; |
930 | ||
931 | err = td_io_open_file(td, f); | |
932 | if (err) | |
933 | continue; | |
934 | opened = 1; | |
935 | } | |
936 | ||
2ba1c290 JA |
937 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) { |
938 | dprint(FD_FILE, "get_next_file_rand: %p\n", f); | |
0aabe160 | 939 | return f; |
2ba1c290 | 940 | } |
87b10676 JA |
941 | if (opened) |
942 | td_io_close_file(td, f); | |
0aabe160 JA |
943 | } while (1); |
944 | } | |
945 | ||
946 | /* | |
947 | * Get next file to service by doing round robin between all available ones | |
948 | */ | |
1c178180 JA |
949 | static struct fio_file *get_next_file_rr(struct thread_data *td, int goodf, |
950 | int badf) | |
3d7c391d JA |
951 | { |
952 | unsigned int old_next_file = td->next_file; | |
953 | struct fio_file *f; | |
954 | ||
955 | do { | |
87b10676 JA |
956 | int opened = 0; |
957 | ||
126d65c6 | 958 | f = td->files[td->next_file]; |
3d7c391d JA |
959 | |
960 | td->next_file++; | |
2dc1bbeb | 961 | if (td->next_file >= td->o.nr_files) |
3d7c391d JA |
962 | td->next_file = 0; |
963 | ||
87b10676 | 964 | dprint(FD_FILE, "trying file %s %x\n", f->file_name, f->flags); |
d6aed795 | 965 | if (fio_file_done(f)) { |
d5ed68ea | 966 | f = NULL; |
059e63c0 | 967 | continue; |
d5ed68ea | 968 | } |
059e63c0 | 969 | |
d6aed795 | 970 | if (!fio_file_open(f)) { |
87b10676 JA |
971 | int err; |
972 | ||
973 | err = td_io_open_file(td, f); | |
b5696bfc JA |
974 | if (err) { |
975 | dprint(FD_FILE, "error %d on open of %s\n", | |
976 | err, f->file_name); | |
87c27b45 | 977 | f = NULL; |
87b10676 | 978 | continue; |
b5696bfc | 979 | } |
87b10676 JA |
980 | opened = 1; |
981 | } | |
982 | ||
0b9d69ec JA |
983 | dprint(FD_FILE, "goodf=%x, badf=%x, ff=%x\n", goodf, badf, |
984 | f->flags); | |
1c178180 | 985 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) |
3d7c391d JA |
986 | break; |
987 | ||
87b10676 JA |
988 | if (opened) |
989 | td_io_close_file(td, f); | |
990 | ||
3d7c391d JA |
991 | f = NULL; |
992 | } while (td->next_file != old_next_file); | |
993 | ||
2ba1c290 | 994 | dprint(FD_FILE, "get_next_file_rr: %p\n", f); |
3d7c391d JA |
995 | return f; |
996 | } | |
997 | ||
7eb36574 | 998 | static struct fio_file *__get_next_file(struct thread_data *td) |
bdb4e2e9 | 999 | { |
1907dbc6 JA |
1000 | struct fio_file *f; |
1001 | ||
2dc1bbeb | 1002 | assert(td->o.nr_files <= td->files_index); |
1c178180 | 1003 | |
b5696bfc | 1004 | if (td->nr_done_files >= td->o.nr_files) { |
5ec10eaa JA |
1005 | dprint(FD_FILE, "get_next_file: nr_open=%d, nr_done=%d," |
1006 | " nr_files=%d\n", td->nr_open_files, | |
1007 | td->nr_done_files, | |
1008 | td->o.nr_files); | |
bdb4e2e9 | 1009 | return NULL; |
2ba1c290 | 1010 | } |
bdb4e2e9 | 1011 | |
1907dbc6 | 1012 | f = td->file_service_file; |
d6aed795 | 1013 | if (f && fio_file_open(f) && !fio_file_closing(f)) { |
a086c257 JA |
1014 | if (td->o.file_service_type == FIO_FSERVICE_SEQ) |
1015 | goto out; | |
1016 | if (td->file_service_left--) | |
1017 | goto out; | |
1018 | } | |
1907dbc6 | 1019 | |
a086c257 JA |
1020 | if (td->o.file_service_type == FIO_FSERVICE_RR || |
1021 | td->o.file_service_type == FIO_FSERVICE_SEQ) | |
d6aed795 | 1022 | f = get_next_file_rr(td, FIO_FILE_open, FIO_FILE_closing); |
bdb4e2e9 | 1023 | else |
d6aed795 | 1024 | f = get_next_file_rand(td, FIO_FILE_open, FIO_FILE_closing); |
1907dbc6 JA |
1025 | |
1026 | td->file_service_file = f; | |
1027 | td->file_service_left = td->file_service_nr - 1; | |
2ba1c290 | 1028 | out: |
683023e8 | 1029 | dprint(FD_FILE, "get_next_file: %p [%s]\n", f, f->file_name); |
1907dbc6 | 1030 | return f; |
bdb4e2e9 JA |
1031 | } |
1032 | ||
7eb36574 JA |
1033 | static struct fio_file *get_next_file(struct thread_data *td) |
1034 | { | |
d72be545 JA |
1035 | if (!(td->flags & TD_F_PROFILE_OPS)) { |
1036 | struct prof_io_ops *ops = &td->prof_io_ops; | |
7eb36574 | 1037 | |
d72be545 JA |
1038 | if (ops->get_next_file) |
1039 | return ops->get_next_file(td); | |
1040 | } | |
7eb36574 JA |
1041 | |
1042 | return __get_next_file(td); | |
1043 | } | |
1044 | ||
429f6675 JA |
1045 | static int set_io_u_file(struct thread_data *td, struct io_u *io_u) |
1046 | { | |
1047 | struct fio_file *f; | |
1048 | ||
1049 | do { | |
1050 | f = get_next_file(td); | |
1051 | if (!f) | |
1052 | return 1; | |
1053 | ||
429f6675 JA |
1054 | io_u->file = f; |
1055 | get_file(f); | |
1056 | ||
1057 | if (!fill_io_u(td, io_u)) | |
1058 | break; | |
1059 | ||
b5696bfc | 1060 | put_file_log(td, f); |
429f6675 | 1061 | td_io_close_file(td, f); |
b5696bfc | 1062 | io_u->file = NULL; |
d6aed795 | 1063 | fio_file_set_done(f); |
429f6675 | 1064 | td->nr_done_files++; |
0b9d69ec JA |
1065 | dprint(FD_FILE, "%s: is done (%d of %d)\n", f->file_name, |
1066 | td->nr_done_files, td->o.nr_files); | |
429f6675 JA |
1067 | } while (1); |
1068 | ||
1069 | return 0; | |
1070 | } | |
1071 | ||
1072 | ||
10ba535a JA |
1073 | struct io_u *__get_io_u(struct thread_data *td) |
1074 | { | |
1075 | struct io_u *io_u = NULL; | |
1076 | ||
e8462bd8 JA |
1077 | td_io_u_lock(td); |
1078 | ||
1079 | again: | |
01743ee1 JA |
1080 | if (!flist_empty(&td->io_u_requeues)) |
1081 | io_u = flist_entry(td->io_u_requeues.next, struct io_u, list); | |
755200a3 | 1082 | else if (!queue_full(td)) { |
01743ee1 | 1083 | io_u = flist_entry(td->io_u_freelist.next, struct io_u, list); |
10ba535a | 1084 | |
6040dabc | 1085 | io_u->buflen = 0; |
10ba535a | 1086 | io_u->resid = 0; |
755200a3 | 1087 | io_u->file = NULL; |
d7762cf8 | 1088 | io_u->end_io = NULL; |
755200a3 JA |
1089 | } |
1090 | ||
1091 | if (io_u) { | |
0c6e7517 | 1092 | assert(io_u->flags & IO_U_F_FREE); |
2ecc1b57 | 1093 | io_u->flags &= ~(IO_U_F_FREE | IO_U_F_FREE_DEF); |
1ef2b6be | 1094 | io_u->flags &= ~(IO_U_F_TRIMMED | IO_U_F_BARRIER); |
82af2a7c | 1095 | io_u->flags &= ~IO_U_F_VER_LIST; |
0c6e7517 | 1096 | |
755200a3 | 1097 | io_u->error = 0; |
bcd5abfa | 1098 | io_u->acct_ddir = -1; |
01743ee1 | 1099 | flist_del(&io_u->list); |
94c6372c | 1100 | flist_add_tail(&io_u->list, &td->io_u_busylist); |
10ba535a | 1101 | td->cur_depth++; |
0c41214f | 1102 | io_u->flags |= IO_U_F_IN_CUR_DEPTH; |
1dec3e07 JA |
1103 | } else if (td->o.verify_async) { |
1104 | /* | |
1105 | * We ran out, wait for async verify threads to finish and | |
1106 | * return one | |
1107 | */ | |
1108 | pthread_cond_wait(&td->free_cond, &td->io_u_lock); | |
1109 | goto again; | |
10ba535a JA |
1110 | } |
1111 | ||
e8462bd8 | 1112 | td_io_u_unlock(td); |
10ba535a JA |
1113 | return io_u; |
1114 | } | |
1115 | ||
0d29de83 | 1116 | static int check_get_trim(struct thread_data *td, struct io_u *io_u) |
10ba535a | 1117 | { |
d72be545 JA |
1118 | if (!(td->flags & TD_F_TRIM_BACKLOG)) |
1119 | return 0; | |
1120 | ||
1121 | if (td->trim_entries) { | |
0d29de83 | 1122 | int get_trim = 0; |
10ba535a | 1123 | |
0d29de83 JA |
1124 | if (td->trim_batch) { |
1125 | td->trim_batch--; | |
1126 | get_trim = 1; | |
1127 | } else if (!(td->io_hist_len % td->o.trim_backlog) && | |
1128 | td->last_ddir != DDIR_READ) { | |
1129 | td->trim_batch = td->o.trim_batch; | |
1130 | if (!td->trim_batch) | |
1131 | td->trim_batch = td->o.trim_backlog; | |
1132 | get_trim = 1; | |
1133 | } | |
1134 | ||
1135 | if (get_trim && !get_next_trim(td, io_u)) | |
1136 | return 1; | |
2ba1c290 | 1137 | } |
10ba535a | 1138 | |
0d29de83 JA |
1139 | return 0; |
1140 | } | |
1141 | ||
1142 | static int check_get_verify(struct thread_data *td, struct io_u *io_u) | |
1143 | { | |
d72be545 JA |
1144 | if (!(td->flags & TD_F_VER_BACKLOG)) |
1145 | return 0; | |
1146 | ||
1147 | if (td->io_hist_len) { | |
9e144189 JA |
1148 | int get_verify = 0; |
1149 | ||
d1ece0c7 | 1150 | if (td->verify_batch) |
9e144189 | 1151 | get_verify = 1; |
d1ece0c7 | 1152 | else if (!(td->io_hist_len % td->o.verify_backlog) && |
9e144189 JA |
1153 | td->last_ddir != DDIR_READ) { |
1154 | td->verify_batch = td->o.verify_batch; | |
f8a75c99 JA |
1155 | if (!td->verify_batch) |
1156 | td->verify_batch = td->o.verify_backlog; | |
9e144189 JA |
1157 | get_verify = 1; |
1158 | } | |
1159 | ||
d1ece0c7 JA |
1160 | if (get_verify && !get_next_verify(td, io_u)) { |
1161 | td->verify_batch--; | |
0d29de83 | 1162 | return 1; |
d1ece0c7 | 1163 | } |
9e144189 JA |
1164 | } |
1165 | ||
0d29de83 JA |
1166 | return 0; |
1167 | } | |
1168 | ||
de789769 JA |
1169 | /* |
1170 | * Fill offset and start time into the buffer content, to prevent too | |
23f394d5 JA |
1171 | * easy compressible data for simple de-dupe attempts. Do this for every |
1172 | * 512b block in the range, since that should be the smallest block size | |
1173 | * we can expect from a device. | |
de789769 JA |
1174 | */ |
1175 | static void small_content_scramble(struct io_u *io_u) | |
1176 | { | |
23f394d5 | 1177 | unsigned int i, nr_blocks = io_u->buflen / 512; |
1ae83d45 | 1178 | uint64_t boffset; |
23f394d5 JA |
1179 | unsigned int offset; |
1180 | void *p, *end; | |
de789769 | 1181 | |
23f394d5 JA |
1182 | if (!nr_blocks) |
1183 | return; | |
1184 | ||
1185 | p = io_u->xfer_buf; | |
fba76ee8 | 1186 | boffset = io_u->offset; |
81f0366c | 1187 | io_u->buf_filled_len = 0; |
fad82f76 | 1188 | |
23f394d5 JA |
1189 | for (i = 0; i < nr_blocks; i++) { |
1190 | /* | |
1191 | * Fill the byte offset into a "random" start offset of | |
1192 | * the buffer, given by the product of the usec time | |
1193 | * and the actual offset. | |
1194 | */ | |
fad82f76 | 1195 | offset = (io_u->start_time.tv_usec ^ boffset) & 511; |
1ae83d45 JA |
1196 | offset &= ~(sizeof(uint64_t) - 1); |
1197 | if (offset >= 512 - sizeof(uint64_t)) | |
1198 | offset -= sizeof(uint64_t); | |
fba76ee8 | 1199 | memcpy(p + offset, &boffset, sizeof(boffset)); |
23f394d5 JA |
1200 | |
1201 | end = p + 512 - sizeof(io_u->start_time); | |
1202 | memcpy(end, &io_u->start_time, sizeof(io_u->start_time)); | |
1203 | p += 512; | |
fad82f76 | 1204 | boffset += 512; |
23f394d5 | 1205 | } |
de789769 JA |
1206 | } |
1207 | ||
0d29de83 JA |
1208 | /* |
1209 | * Return an io_u to be processed. Gets a buflen and offset, sets direction, | |
1210 | * etc. The returned io_u is fully ready to be prepped and submitted. | |
1211 | */ | |
1212 | struct io_u *get_io_u(struct thread_data *td) | |
1213 | { | |
1214 | struct fio_file *f; | |
1215 | struct io_u *io_u; | |
de789769 | 1216 | int do_scramble = 0; |
0d29de83 JA |
1217 | |
1218 | io_u = __get_io_u(td); | |
1219 | if (!io_u) { | |
1220 | dprint(FD_IO, "__get_io_u failed\n"); | |
1221 | return NULL; | |
1222 | } | |
1223 | ||
1224 | if (check_get_verify(td, io_u)) | |
1225 | goto out; | |
1226 | if (check_get_trim(td, io_u)) | |
1227 | goto out; | |
1228 | ||
755200a3 JA |
1229 | /* |
1230 | * from a requeue, io_u already setup | |
1231 | */ | |
1232 | if (io_u->file) | |
77f392bf | 1233 | goto out; |
755200a3 | 1234 | |
429f6675 JA |
1235 | /* |
1236 | * If using an iolog, grab next piece if any available. | |
1237 | */ | |
d72be545 | 1238 | if (td->flags & TD_F_READ_IOLOG) { |
429f6675 JA |
1239 | if (read_iolog_get(td, io_u)) |
1240 | goto err_put; | |
2ba1c290 JA |
1241 | } else if (set_io_u_file(td, io_u)) { |
1242 | dprint(FD_IO, "io_u %p, setting file failed\n", io_u); | |
429f6675 | 1243 | goto err_put; |
2ba1c290 | 1244 | } |
5ec10eaa | 1245 | |
429f6675 | 1246 | f = io_u->file; |
d6aed795 | 1247 | assert(fio_file_open(f)); |
97af62ce | 1248 | |
ff58fced | 1249 | if (ddir_rw(io_u->ddir)) { |
d0656a93 | 1250 | if (!io_u->buflen && !(td->io_ops->flags & FIO_NOIO)) { |
2ba1c290 | 1251 | dprint(FD_IO, "get_io_u: zero buflen on %p\n", io_u); |
429f6675 | 1252 | goto err_put; |
2ba1c290 | 1253 | } |
10ba535a | 1254 | |
38dad62d | 1255 | f->last_start = io_u->offset; |
36167d82 | 1256 | f->last_pos = io_u->offset + io_u->buflen; |
10ba535a | 1257 | |
fd68418e | 1258 | if (io_u->ddir == DDIR_WRITE) { |
d72be545 | 1259 | if (td->flags & TD_F_REFILL_BUFFERS) { |
9c42684e JA |
1260 | io_u_fill_buffer(td, io_u, |
1261 | io_u->xfer_buflen, io_u->xfer_buflen); | |
d72be545 | 1262 | } else if (td->flags & TD_F_SCRAMBLE_BUFFERS) |
fd68418e | 1263 | do_scramble = 1; |
d72be545 | 1264 | if (td->flags & TD_F_VER_NONE) { |
629f1d71 JA |
1265 | populate_verify_io_u(td, io_u); |
1266 | do_scramble = 0; | |
1267 | } | |
fd68418e | 1268 | } else if (io_u->ddir == DDIR_READ) { |
cbe8d756 RR |
1269 | /* |
1270 | * Reset the buf_filled parameters so next time if the | |
1271 | * buffer is used for writes it is refilled. | |
1272 | */ | |
cbe8d756 RR |
1273 | io_u->buf_filled_len = 0; |
1274 | } | |
87dc1ab1 | 1275 | } |
10ba535a | 1276 | |
165faf16 JA |
1277 | /* |
1278 | * Set io data pointers. | |
1279 | */ | |
cec6b55d JA |
1280 | io_u->xfer_buf = io_u->buf; |
1281 | io_u->xfer_buflen = io_u->buflen; | |
5973cafb | 1282 | |
6ac7a331 | 1283 | out: |
0d29de83 | 1284 | assert(io_u->file); |
429f6675 | 1285 | if (!td_io_prep(td, io_u)) { |
993bf48b JA |
1286 | if (!td->o.disable_slat) |
1287 | fio_gettime(&io_u->start_time, NULL); | |
de789769 JA |
1288 | if (do_scramble) |
1289 | small_content_scramble(io_u); | |
429f6675 | 1290 | return io_u; |
36167d82 | 1291 | } |
429f6675 | 1292 | err_put: |
2ba1c290 | 1293 | dprint(FD_IO, "get_io_u failed\n"); |
429f6675 JA |
1294 | put_io_u(td, io_u); |
1295 | return NULL; | |
10ba535a JA |
1296 | } |
1297 | ||
5451792e JA |
1298 | void io_u_log_error(struct thread_data *td, struct io_u *io_u) |
1299 | { | |
8b28bd41 | 1300 | enum error_type_bit eb = td_error_type(io_u->ddir, io_u->error); |
825f818e JA |
1301 | const char *msg[] = { "read", "write", "sync", "datasync", |
1302 | "sync_file_range", "wait", "trim" }; | |
1303 | ||
8b28bd41 DM |
1304 | if (td_non_fatal_error(td, eb, io_u->error) && !td->o.error_dump) |
1305 | return; | |
5451792e JA |
1306 | |
1307 | log_err("fio: io_u error"); | |
1308 | ||
1309 | if (io_u->file) | |
1310 | log_err(" on file %s", io_u->file->file_name); | |
1311 | ||
1312 | log_err(": %s\n", strerror(io_u->error)); | |
1313 | ||
5ec10eaa JA |
1314 | log_err(" %s offset=%llu, buflen=%lu\n", msg[io_u->ddir], |
1315 | io_u->offset, io_u->xfer_buflen); | |
5451792e JA |
1316 | |
1317 | if (!td->error) | |
1318 | td_verror(td, io_u->error, "io_u error"); | |
1319 | } | |
1320 | ||
c8eeb9df JA |
1321 | static void account_io_completion(struct thread_data *td, struct io_u *io_u, |
1322 | struct io_completion_data *icd, | |
1323 | const enum fio_ddir idx, unsigned int bytes) | |
1324 | { | |
24d23ca7 | 1325 | unsigned long lusec = 0; |
c8eeb9df | 1326 | |
c8eeb9df JA |
1327 | if (!td->o.disable_clat || !td->o.disable_bw) |
1328 | lusec = utime_since(&io_u->issue_time, &icd->time); | |
1329 | ||
1330 | if (!td->o.disable_lat) { | |
1331 | unsigned long tusec; | |
1332 | ||
1333 | tusec = utime_since(&io_u->start_time, &icd->time); | |
1334 | add_lat_sample(td, idx, tusec, bytes); | |
15501535 JA |
1335 | |
1336 | if (td->o.max_latency && tusec > td->o.max_latency) { | |
1337 | if (!td->error) | |
1338 | log_err("fio: latency of %lu usec exceeds specified max (%u usec)\n", tusec, td->o.max_latency); | |
c5b3adc4 JA |
1339 | td_verror(td, ETIMEDOUT, "max latency exceeded"); |
1340 | icd->error = ETIMEDOUT; | |
15501535 | 1341 | } |
c8eeb9df JA |
1342 | } |
1343 | ||
1344 | if (!td->o.disable_clat) { | |
1345 | add_clat_sample(td, idx, lusec, bytes); | |
1346 | io_u_mark_latency(td, lusec); | |
1347 | } | |
1348 | ||
1349 | if (!td->o.disable_bw) | |
1350 | add_bw_sample(td, idx, bytes, &icd->time); | |
1351 | ||
1352 | add_iops_sample(td, idx, &icd->time); | |
1353 | } | |
1354 | ||
1b8dbf25 SL |
1355 | static long long usec_for_io(struct thread_data *td, enum fio_ddir ddir) |
1356 | { | |
1ae83d45 JA |
1357 | uint64_t secs, remainder, bps, bytes; |
1358 | ||
1b8dbf25 SL |
1359 | bytes = td->this_io_bytes[ddir]; |
1360 | bps = td->rate_bps[ddir]; | |
1361 | secs = bytes / bps; | |
1362 | remainder = bytes % bps; | |
1363 | return remainder * 1000000 / bps + secs * 1000000; | |
1364 | } | |
1365 | ||
97601024 JA |
1366 | static void io_completed(struct thread_data *td, struct io_u *io_u, |
1367 | struct io_completion_data *icd) | |
10ba535a | 1368 | { |
44f29692 | 1369 | struct fio_file *f; |
10ba535a | 1370 | |
2ba1c290 JA |
1371 | dprint_io_u(io_u, "io complete"); |
1372 | ||
2ecc1b57 | 1373 | td_io_u_lock(td); |
0c6e7517 | 1374 | assert(io_u->flags & IO_U_F_FLIGHT); |
38dad62d | 1375 | io_u->flags &= ~(IO_U_F_FLIGHT | IO_U_F_BUSY_OK); |
2ecc1b57 | 1376 | td_io_u_unlock(td); |
0c6e7517 | 1377 | |
5f9099ea | 1378 | if (ddir_sync(io_u->ddir)) { |
87dc1ab1 | 1379 | td->last_was_sync = 1; |
44f29692 JA |
1380 | f = io_u->file; |
1381 | if (f) { | |
1382 | f->first_write = -1ULL; | |
1383 | f->last_write = -1ULL; | |
1384 | } | |
87dc1ab1 JA |
1385 | return; |
1386 | } | |
1387 | ||
1388 | td->last_was_sync = 0; | |
9e144189 | 1389 | td->last_ddir = io_u->ddir; |
87dc1ab1 | 1390 | |
ff58fced | 1391 | if (!io_u->error && ddir_rw(io_u->ddir)) { |
10ba535a | 1392 | unsigned int bytes = io_u->buflen - io_u->resid; |
1e97cce9 | 1393 | const enum fio_ddir idx = io_u->ddir; |
ba3e4e0c | 1394 | const enum fio_ddir odx = io_u->ddir ^ 1; |
b29ee5b3 | 1395 | int ret; |
10ba535a | 1396 | |
b29ee5b3 | 1397 | td->io_blocks[idx]++; |
c8eeb9df | 1398 | td->this_io_blocks[idx]++; |
b29ee5b3 | 1399 | td->io_bytes[idx] += bytes; |
ae2fafc8 | 1400 | |
1401 | if (!(io_u->flags & IO_U_F_VER_LIST)) | |
1402 | td->this_io_bytes[idx] += bytes; | |
10ba535a | 1403 | |
44f29692 JA |
1404 | if (idx == DDIR_WRITE) { |
1405 | f = io_u->file; | |
1406 | if (f) { | |
1407 | if (f->first_write == -1ULL || | |
1408 | io_u->offset < f->first_write) | |
1409 | f->first_write = io_u->offset; | |
1410 | if (f->last_write == -1ULL || | |
1411 | ((io_u->offset + bytes) > f->last_write)) | |
1412 | f->last_write = io_u->offset + bytes; | |
1413 | } | |
1414 | } | |
1415 | ||
6b1190fd SL |
1416 | if (ramp_time_over(td) && (td->runstate == TD_RUNNING || |
1417 | td->runstate == TD_VERIFYING)) { | |
c8eeb9df | 1418 | account_io_completion(td, io_u, icd, idx, bytes); |
40e1a6f0 | 1419 | |
b23b6a2f | 1420 | if (__should_check_rate(td, idx)) { |
ba3e4e0c | 1421 | td->rate_pending_usleep[idx] = |
1b8dbf25 | 1422 | (usec_for_io(td, idx) - |
ba3e4e0c | 1423 | utime_since_now(&td->start)); |
b23b6a2f | 1424 | } |
6eaf09d6 | 1425 | if (idx != DDIR_TRIM && __should_check_rate(td, odx)) |
ba3e4e0c | 1426 | td->rate_pending_usleep[odx] = |
1b8dbf25 | 1427 | (usec_for_io(td, odx) - |
ba3e4e0c | 1428 | utime_since_now(&td->start)); |
721938ae | 1429 | } |
10ba535a | 1430 | |
660a1cb5 | 1431 | if (td_write(td) && idx == DDIR_WRITE && |
8670579e | 1432 | td->o.do_verify && |
44cbc6da JA |
1433 | td->o.verify != VERIFY_NONE && |
1434 | !td->o.experimental_verify) | |
10ba535a JA |
1435 | log_io_piece(td, io_u); |
1436 | ||
b29ee5b3 | 1437 | icd->bytes_done[idx] += bytes; |
3af6ef39 | 1438 | |
d7762cf8 | 1439 | if (io_u->end_io) { |
36690c9b | 1440 | ret = io_u->end_io(td, io_u); |
3af6ef39 JA |
1441 | if (ret && !icd->error) |
1442 | icd->error = ret; | |
1443 | } | |
ff58fced | 1444 | } else if (io_u->error) { |
10ba535a | 1445 | icd->error = io_u->error; |
5451792e JA |
1446 | io_u_log_error(td, io_u); |
1447 | } | |
8b28bd41 DM |
1448 | if (icd->error) { |
1449 | enum error_type_bit eb = td_error_type(io_u->ddir, icd->error); | |
1450 | if (!td_non_fatal_error(td, eb, icd->error)) | |
1451 | return; | |
f2bba182 RR |
1452 | /* |
1453 | * If there is a non_fatal error, then add to the error count | |
1454 | * and clear all the errors. | |
1455 | */ | |
1456 | update_error_count(td, icd->error); | |
1457 | td_clear_error(td); | |
1458 | icd->error = 0; | |
1459 | io_u->error = 0; | |
1460 | } | |
10ba535a JA |
1461 | } |
1462 | ||
9520ebb9 JA |
1463 | static void init_icd(struct thread_data *td, struct io_completion_data *icd, |
1464 | int nr) | |
10ba535a | 1465 | { |
6eaf09d6 | 1466 | int ddir; |
9520ebb9 JA |
1467 | if (!td->o.disable_clat || !td->o.disable_bw) |
1468 | fio_gettime(&icd->time, NULL); | |
02bcaa8c | 1469 | |
3af6ef39 JA |
1470 | icd->nr = nr; |
1471 | ||
10ba535a | 1472 | icd->error = 0; |
6eaf09d6 SL |
1473 | for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) |
1474 | icd->bytes_done[ddir] = 0; | |
36167d82 JA |
1475 | } |
1476 | ||
97601024 JA |
1477 | static void ios_completed(struct thread_data *td, |
1478 | struct io_completion_data *icd) | |
36167d82 JA |
1479 | { |
1480 | struct io_u *io_u; | |
1481 | int i; | |
1482 | ||
10ba535a JA |
1483 | for (i = 0; i < icd->nr; i++) { |
1484 | io_u = td->io_ops->event(td, i); | |
1485 | ||
1486 | io_completed(td, io_u, icd); | |
e8462bd8 JA |
1487 | |
1488 | if (!(io_u->flags & IO_U_F_FREE_DEF)) | |
1489 | put_io_u(td, io_u); | |
10ba535a JA |
1490 | } |
1491 | } | |
97601024 | 1492 | |
e7e6cfb4 JA |
1493 | /* |
1494 | * Complete a single io_u for the sync engines. | |
1495 | */ | |
581e7141 | 1496 | int io_u_sync_complete(struct thread_data *td, struct io_u *io_u, |
100f49f1 | 1497 | uint64_t *bytes) |
97601024 JA |
1498 | { |
1499 | struct io_completion_data icd; | |
1500 | ||
9520ebb9 | 1501 | init_icd(td, &icd, 1); |
97601024 | 1502 | io_completed(td, io_u, &icd); |
e8462bd8 JA |
1503 | |
1504 | if (!(io_u->flags & IO_U_F_FREE_DEF)) | |
1505 | put_io_u(td, io_u); | |
97601024 | 1506 | |
581e7141 JA |
1507 | if (icd.error) { |
1508 | td_verror(td, icd.error, "io_u_sync_complete"); | |
1509 | return -1; | |
1510 | } | |
97601024 | 1511 | |
581e7141 | 1512 | if (bytes) { |
6eaf09d6 SL |
1513 | int ddir; |
1514 | ||
1515 | for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) | |
1516 | bytes[ddir] += icd.bytes_done[ddir]; | |
581e7141 JA |
1517 | } |
1518 | ||
1519 | return 0; | |
97601024 JA |
1520 | } |
1521 | ||
e7e6cfb4 JA |
1522 | /* |
1523 | * Called to complete min_events number of io for the async engines. | |
1524 | */ | |
581e7141 | 1525 | int io_u_queued_complete(struct thread_data *td, int min_evts, |
100f49f1 | 1526 | uint64_t *bytes) |
97601024 | 1527 | { |
97601024 | 1528 | struct io_completion_data icd; |
00de55ef | 1529 | struct timespec *tvp = NULL; |
97601024 | 1530 | int ret; |
4d06a338 | 1531 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0, }; |
97601024 | 1532 | |
4950421a | 1533 | dprint(FD_IO, "io_u_queued_completed: min=%d\n", min_evts); |
b271fe62 | 1534 | |
4950421a | 1535 | if (!min_evts) |
00de55ef | 1536 | tvp = &ts; |
97601024 | 1537 | |
4950421a | 1538 | ret = td_io_getevents(td, min_evts, td->o.iodepth_batch_complete, tvp); |
97601024 | 1539 | if (ret < 0) { |
e1161c32 | 1540 | td_verror(td, -ret, "td_io_getevents"); |
97601024 JA |
1541 | return ret; |
1542 | } else if (!ret) | |
1543 | return ret; | |
1544 | ||
9520ebb9 | 1545 | init_icd(td, &icd, ret); |
97601024 | 1546 | ios_completed(td, &icd); |
581e7141 JA |
1547 | if (icd.error) { |
1548 | td_verror(td, icd.error, "io_u_queued_complete"); | |
1549 | return -1; | |
1550 | } | |
97601024 | 1551 | |
581e7141 | 1552 | if (bytes) { |
6eaf09d6 SL |
1553 | int ddir; |
1554 | ||
1555 | for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) | |
1556 | bytes[ddir] += icd.bytes_done[ddir]; | |
581e7141 JA |
1557 | } |
1558 | ||
1559 | return 0; | |
97601024 | 1560 | } |
7e77dd02 JA |
1561 | |
1562 | /* | |
1563 | * Call when io_u is really queued, to update the submission latency. | |
1564 | */ | |
1565 | void io_u_queued(struct thread_data *td, struct io_u *io_u) | |
1566 | { | |
9520ebb9 JA |
1567 | if (!td->o.disable_slat) { |
1568 | unsigned long slat_time; | |
7e77dd02 | 1569 | |
9520ebb9 | 1570 | slat_time = utime_since(&io_u->start_time, &io_u->issue_time); |
29a90ddb | 1571 | add_slat_sample(td, io_u->ddir, slat_time, io_u->xfer_buflen); |
9520ebb9 | 1572 | } |
7e77dd02 | 1573 | } |
433afcb4 | 1574 | |
5973cafb JA |
1575 | /* |
1576 | * "randomly" fill the buffer contents | |
1577 | */ | |
1578 | void io_u_fill_buffer(struct thread_data *td, struct io_u *io_u, | |
9c42684e | 1579 | unsigned int min_write, unsigned int max_bs) |
5973cafb | 1580 | { |
e84b371e JA |
1581 | io_u->buf_filled_len = 0; |
1582 | ||
9c42684e JA |
1583 | if (!td->o.zero_buffers) { |
1584 | unsigned int perc = td->o.compress_percentage; | |
1585 | ||
1586 | if (perc) { | |
f97a43a1 JA |
1587 | unsigned int seg = min_write; |
1588 | ||
1589 | seg = min(min_write, td->o.compress_chunk); | |
9c42684e | 1590 | fill_random_buf_percentage(&td->buf_state, io_u->buf, |
f97a43a1 | 1591 | perc, seg, max_bs); |
9c42684e JA |
1592 | } else |
1593 | fill_random_buf(&td->buf_state, io_u->buf, max_bs); | |
1594 | } else | |
637ef8d9 | 1595 | memset(io_u->buf, 0, max_bs); |
5973cafb | 1596 | } |