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 */ | |
6eaf09d6 | 19 | unsigned long 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 | ||
1ae83d45 JA |
175 | static int get_next_rand_offset(struct thread_data *td, struct fio_file *f, |
176 | enum fio_ddir ddir, uint64_t *b) | |
177 | { | |
178 | struct rand_off *r; | |
179 | int i, ret = 1; | |
180 | ||
181 | /* | |
182 | * If sort not enabled, or not a pure random read workload without | |
183 | * any stored write metadata, just return a random offset | |
184 | */ | |
6d68b997 | 185 | if (!td->o.verifysort_nr || !(ddir == DDIR_READ && td->o.do_verify && |
44cbc6da JA |
186 | td->o.verify != VERIFY_NONE && td_random(td)) || |
187 | td->o.random_generator == FIO_RAND_GEN_TAUSWORTHE) | |
1ae83d45 JA |
188 | return get_off_from_method(td, f, ddir, b); |
189 | ||
190 | if (!flist_empty(&td->next_rand_list)) { | |
191 | struct rand_off *r; | |
192 | fetch: | |
193 | r = flist_entry(td->next_rand_list.next, struct rand_off, list); | |
194 | flist_del(&r->list); | |
195 | *b = r->off; | |
196 | free(r); | |
197 | return 0; | |
198 | } | |
199 | ||
200 | for (i = 0; i < td->o.verifysort_nr; i++) { | |
201 | r = malloc(sizeof(*r)); | |
202 | ||
203 | ret = get_off_from_method(td, f, ddir, &r->off); | |
204 | if (ret) { | |
205 | free(r); | |
206 | break; | |
207 | } | |
208 | ||
209 | flist_add(&r->list, &td->next_rand_list); | |
210 | } | |
211 | ||
212 | if (ret && !i) | |
213 | return ret; | |
214 | ||
215 | assert(!flist_empty(&td->next_rand_list)); | |
216 | flist_sort(NULL, &td->next_rand_list, flist_cmp); | |
217 | goto fetch; | |
218 | } | |
219 | ||
38dad62d | 220 | static int get_next_rand_block(struct thread_data *td, struct fio_file *f, |
1ae83d45 | 221 | enum fio_ddir ddir, uint64_t *b) |
38dad62d | 222 | { |
c04e4661 DE |
223 | if (!get_next_rand_offset(td, f, ddir, b)) |
224 | return 0; | |
225 | ||
226 | if (td->o.time_based) { | |
33c48814 | 227 | fio_file_reset(td, f); |
c04e4661 DE |
228 | if (!get_next_rand_offset(td, f, ddir, b)) |
229 | return 0; | |
38dad62d JA |
230 | } |
231 | ||
c04e4661 DE |
232 | dprint(FD_IO, "%s: rand offset failed, last=%llu, size=%llu\n", |
233 | f->file_name, f->last_pos, f->real_file_size); | |
234 | return 1; | |
38dad62d JA |
235 | } |
236 | ||
37cf9e3c | 237 | static int get_next_seq_offset(struct thread_data *td, struct fio_file *f, |
1ae83d45 | 238 | enum fio_ddir ddir, uint64_t *offset) |
38dad62d | 239 | { |
ff58fced JA |
240 | assert(ddir_rw(ddir)); |
241 | ||
ce95d651 | 242 | if (f->last_pos >= f->io_size + get_start_offset(td) && td->o.time_based) |
c04e4661 DE |
243 | f->last_pos = f->last_pos - f->io_size; |
244 | ||
38dad62d | 245 | if (f->last_pos < f->real_file_size) { |
1ae83d45 | 246 | uint64_t pos; |
059b0802 | 247 | |
a66da7a2 JA |
248 | if (f->last_pos == f->file_offset && td->o.ddir_seq_add < 0) |
249 | f->last_pos = f->real_file_size; | |
250 | ||
251 | pos = f->last_pos - f->file_offset; | |
059b0802 JA |
252 | if (pos) |
253 | pos += td->o.ddir_seq_add; | |
254 | ||
37cf9e3c | 255 | *offset = pos; |
38dad62d JA |
256 | return 0; |
257 | } | |
258 | ||
259 | return 1; | |
260 | } | |
261 | ||
262 | static int get_next_block(struct thread_data *td, struct io_u *io_u, | |
37cf9e3c | 263 | enum fio_ddir ddir, int rw_seq) |
38dad62d JA |
264 | { |
265 | struct fio_file *f = io_u->file; | |
1ae83d45 | 266 | uint64_t b, offset; |
38dad62d JA |
267 | int ret; |
268 | ||
ff58fced JA |
269 | assert(ddir_rw(ddir)); |
270 | ||
37cf9e3c JA |
271 | b = offset = -1ULL; |
272 | ||
38dad62d JA |
273 | if (rw_seq) { |
274 | if (td_random(td)) | |
37cf9e3c | 275 | ret = get_next_rand_block(td, f, ddir, &b); |
38dad62d | 276 | else |
37cf9e3c | 277 | ret = get_next_seq_offset(td, f, ddir, &offset); |
38dad62d JA |
278 | } else { |
279 | io_u->flags |= IO_U_F_BUSY_OK; | |
280 | ||
281 | if (td->o.rw_seq == RW_SEQ_SEQ) { | |
37cf9e3c | 282 | ret = get_next_seq_offset(td, f, ddir, &offset); |
38dad62d | 283 | if (ret) |
37cf9e3c | 284 | ret = get_next_rand_block(td, f, ddir, &b); |
38dad62d JA |
285 | } else if (td->o.rw_seq == RW_SEQ_IDENT) { |
286 | if (f->last_start != -1ULL) | |
37cf9e3c | 287 | offset = f->last_start - f->file_offset; |
38dad62d | 288 | else |
37cf9e3c | 289 | offset = 0; |
38dad62d JA |
290 | ret = 0; |
291 | } else { | |
292 | log_err("fio: unknown rw_seq=%d\n", td->o.rw_seq); | |
293 | ret = 1; | |
294 | } | |
295 | } | |
6d68b997 | 296 | |
37cf9e3c JA |
297 | if (!ret) { |
298 | if (offset != -1ULL) | |
299 | io_u->offset = offset; | |
300 | else if (b != -1ULL) | |
301 | io_u->offset = b * td->o.ba[ddir]; | |
302 | else { | |
51ede0b1 JA |
303 | log_err("fio: bug in offset generation: offset=%llu, b=%llu\n", |
304 | offset, b); | |
37cf9e3c JA |
305 | ret = 1; |
306 | } | |
307 | } | |
308 | ||
38dad62d JA |
309 | return ret; |
310 | } | |
311 | ||
10ba535a JA |
312 | /* |
313 | * For random io, generate a random new block and see if it's used. Repeat | |
314 | * until we find a free one. For sequential io, just return the end of | |
315 | * the last io issued. | |
316 | */ | |
15dc1934 | 317 | static int __get_next_offset(struct thread_data *td, struct io_u *io_u) |
10ba535a | 318 | { |
9bf2061e | 319 | struct fio_file *f = io_u->file; |
4ba66134 | 320 | enum fio_ddir ddir = io_u->ddir; |
38dad62d | 321 | int rw_seq_hit = 0; |
10ba535a | 322 | |
ff58fced JA |
323 | assert(ddir_rw(ddir)); |
324 | ||
38dad62d JA |
325 | if (td->o.ddir_seq_nr && !--td->ddir_seq_nr) { |
326 | rw_seq_hit = 1; | |
5736c10d | 327 | td->ddir_seq_nr = td->o.ddir_seq_nr; |
38dad62d | 328 | } |
211097b2 | 329 | |
37cf9e3c | 330 | if (get_next_block(td, io_u, ddir, rw_seq_hit)) |
38dad62d | 331 | return 1; |
10ba535a | 332 | |
009bd847 JA |
333 | if (io_u->offset >= f->io_size) { |
334 | dprint(FD_IO, "get_next_offset: offset %llu >= io_size %llu\n", | |
335 | io_u->offset, f->io_size); | |
336 | return 1; | |
337 | } | |
338 | ||
339 | io_u->offset += f->file_offset; | |
2ba1c290 JA |
340 | if (io_u->offset >= f->real_file_size) { |
341 | dprint(FD_IO, "get_next_offset: offset %llu >= size %llu\n", | |
342 | io_u->offset, f->real_file_size); | |
10ba535a | 343 | return 1; |
2ba1c290 | 344 | } |
10ba535a JA |
345 | |
346 | return 0; | |
347 | } | |
348 | ||
15dc1934 JA |
349 | static int get_next_offset(struct thread_data *td, struct io_u *io_u) |
350 | { | |
d72be545 JA |
351 | if (td->flags & TD_F_PROFILE_OPS) { |
352 | struct prof_io_ops *ops = &td->prof_io_ops; | |
7eb36574 | 353 | |
d72be545 JA |
354 | if (ops->fill_io_u_off) |
355 | return ops->fill_io_u_off(td, io_u); | |
356 | } | |
15dc1934 JA |
357 | |
358 | return __get_next_offset(td, io_u); | |
359 | } | |
360 | ||
79944128 JA |
361 | static inline int io_u_fits(struct thread_data *td, struct io_u *io_u, |
362 | unsigned int buflen) | |
363 | { | |
364 | struct fio_file *f = io_u->file; | |
365 | ||
ce95d651 | 366 | return io_u->offset + buflen <= f->io_size + get_start_offset(td); |
79944128 JA |
367 | } |
368 | ||
15dc1934 | 369 | static unsigned int __get_next_buflen(struct thread_data *td, struct io_u *io_u) |
10ba535a | 370 | { |
bca4ed4d | 371 | const int ddir = io_u->ddir; |
24d23ca7 | 372 | unsigned int buflen = 0; |
f3059de1 | 373 | unsigned int minbs, maxbs; |
1294c3ec | 374 | unsigned long r, rand_max; |
10ba535a | 375 | |
ff58fced JA |
376 | assert(ddir_rw(ddir)); |
377 | ||
f3059de1 JA |
378 | minbs = td->o.min_bs[ddir]; |
379 | maxbs = td->o.max_bs[ddir]; | |
380 | ||
79944128 JA |
381 | if (minbs == maxbs) |
382 | return minbs; | |
383 | ||
52c58027 JA |
384 | /* |
385 | * If we can't satisfy the min block size from here, then fail | |
386 | */ | |
387 | if (!io_u_fits(td, io_u, minbs)) | |
388 | return 0; | |
389 | ||
4c07ad86 JA |
390 | if (td->o.use_os_rand) |
391 | rand_max = OS_RAND_MAX; | |
392 | else | |
393 | rand_max = FRAND_MAX; | |
394 | ||
79944128 | 395 | do { |
4c07ad86 JA |
396 | if (td->o.use_os_rand) |
397 | r = os_random_long(&td->bsrange_state); | |
398 | else | |
399 | r = __rand(&td->__bsrange_state); | |
400 | ||
720e84ad | 401 | if (!td->o.bssplit_nr[ddir]) { |
f3059de1 | 402 | buflen = 1 + (unsigned int) ((double) maxbs * |
4c07ad86 | 403 | (r / (rand_max + 1.0))); |
f3059de1 JA |
404 | if (buflen < minbs) |
405 | buflen = minbs; | |
5ec10eaa | 406 | } else { |
564ca972 JA |
407 | long perc = 0; |
408 | unsigned int i; | |
409 | ||
720e84ad JA |
410 | for (i = 0; i < td->o.bssplit_nr[ddir]; i++) { |
411 | struct bssplit *bsp = &td->o.bssplit[ddir][i]; | |
564ca972 JA |
412 | |
413 | buflen = bsp->bs; | |
414 | perc += bsp->perc; | |
79944128 JA |
415 | if ((r <= ((rand_max / 100L) * perc)) && |
416 | io_u_fits(td, io_u, buflen)) | |
564ca972 JA |
417 | break; |
418 | } | |
419 | } | |
79944128 | 420 | |
f3059de1 JA |
421 | if (!td->o.bs_unaligned && is_power_of_2(minbs)) |
422 | buflen = (buflen + minbs - 1) & ~(minbs - 1); | |
10ba535a | 423 | |
79944128 | 424 | } while (!io_u_fits(td, io_u, buflen)); |
6a5e6884 | 425 | |
10ba535a JA |
426 | return buflen; |
427 | } | |
428 | ||
15dc1934 JA |
429 | static unsigned int get_next_buflen(struct thread_data *td, struct io_u *io_u) |
430 | { | |
d72be545 JA |
431 | if (td->flags & TD_F_PROFILE_OPS) { |
432 | struct prof_io_ops *ops = &td->prof_io_ops; | |
7eb36574 | 433 | |
d72be545 JA |
434 | if (ops->fill_io_u_size) |
435 | return ops->fill_io_u_size(td, io_u); | |
436 | } | |
15dc1934 JA |
437 | |
438 | return __get_next_buflen(td, io_u); | |
439 | } | |
440 | ||
afe24a5a JA |
441 | static void set_rwmix_bytes(struct thread_data *td) |
442 | { | |
afe24a5a JA |
443 | unsigned int diff; |
444 | ||
445 | /* | |
446 | * we do time or byte based switch. this is needed because | |
447 | * buffered writes may issue a lot quicker than they complete, | |
448 | * whereas reads do not. | |
449 | */ | |
e47f799f | 450 | diff = td->o.rwmix[td->rwmix_ddir ^ 1]; |
04c540d9 | 451 | td->rwmix_issues = (td->io_issues[td->rwmix_ddir] * diff) / 100; |
e47f799f JA |
452 | } |
453 | ||
454 | static inline enum fio_ddir get_rand_ddir(struct thread_data *td) | |
455 | { | |
456 | unsigned int v; | |
1294c3ec | 457 | unsigned long r; |
e47f799f | 458 | |
4c07ad86 JA |
459 | if (td->o.use_os_rand) { |
460 | r = os_random_long(&td->rwmix_state); | |
461 | v = 1 + (int) (100.0 * (r / (OS_RAND_MAX + 1.0))); | |
462 | } else { | |
463 | r = __rand(&td->__rwmix_state); | |
464 | v = 1 + (int) (100.0 * (r / (FRAND_MAX + 1.0))); | |
465 | } | |
466 | ||
04c540d9 | 467 | if (v <= td->o.rwmix[DDIR_READ]) |
e47f799f JA |
468 | return DDIR_READ; |
469 | ||
470 | return DDIR_WRITE; | |
afe24a5a JA |
471 | } |
472 | ||
581e7141 JA |
473 | static enum fio_ddir rate_ddir(struct thread_data *td, enum fio_ddir ddir) |
474 | { | |
475 | enum fio_ddir odir = ddir ^ 1; | |
476 | struct timeval t; | |
477 | long usec; | |
478 | ||
ff58fced JA |
479 | assert(ddir_rw(ddir)); |
480 | ||
581e7141 JA |
481 | if (td->rate_pending_usleep[ddir] <= 0) |
482 | return ddir; | |
483 | ||
484 | /* | |
485 | * We have too much pending sleep in this direction. See if we | |
486 | * should switch. | |
487 | */ | |
488 | if (td_rw(td)) { | |
489 | /* | |
490 | * Other direction does not have too much pending, switch | |
491 | */ | |
492 | if (td->rate_pending_usleep[odir] < 100000) | |
493 | return odir; | |
494 | ||
495 | /* | |
496 | * Both directions have pending sleep. Sleep the minimum time | |
497 | * and deduct from both. | |
498 | */ | |
499 | if (td->rate_pending_usleep[ddir] <= | |
500 | td->rate_pending_usleep[odir]) { | |
501 | usec = td->rate_pending_usleep[ddir]; | |
502 | } else { | |
503 | usec = td->rate_pending_usleep[odir]; | |
504 | ddir = odir; | |
505 | } | |
506 | } else | |
507 | usec = td->rate_pending_usleep[ddir]; | |
508 | ||
78c1eda5 JA |
509 | /* |
510 | * We are going to sleep, ensure that we flush anything pending as | |
422f9e4b RM |
511 | * not to skew our latency numbers. |
512 | * | |
513 | * Changed to only monitor 'in flight' requests here instead of the | |
514 | * td->cur_depth, b/c td->cur_depth does not accurately represent | |
515 | * io's that have been actually submitted to an async engine, | |
516 | * and cur_depth is meaningless for sync engines. | |
78c1eda5 | 517 | */ |
422f9e4b | 518 | if (td->io_u_in_flight) { |
78c1eda5 JA |
519 | int fio_unused ret; |
520 | ||
422f9e4b | 521 | ret = io_u_queued_complete(td, td->io_u_in_flight, NULL); |
78c1eda5 JA |
522 | } |
523 | ||
581e7141 JA |
524 | fio_gettime(&t, NULL); |
525 | usec_sleep(td, usec); | |
526 | usec = utime_since_now(&t); | |
527 | ||
528 | td->rate_pending_usleep[ddir] -= usec; | |
529 | ||
530 | odir = ddir ^ 1; | |
531 | if (td_rw(td) && __should_check_rate(td, odir)) | |
532 | td->rate_pending_usleep[odir] -= usec; | |
0b9d69ec | 533 | |
6eaf09d6 SL |
534 | if (ddir_trim(ddir)) |
535 | return ddir; | |
581e7141 JA |
536 | return ddir; |
537 | } | |
538 | ||
10ba535a JA |
539 | /* |
540 | * Return the data direction for the next io_u. If the job is a | |
541 | * mixed read/write workload, check the rwmix cycle and switch if | |
542 | * necessary. | |
543 | */ | |
1e97cce9 | 544 | static enum fio_ddir get_rw_ddir(struct thread_data *td) |
10ba535a | 545 | { |
581e7141 JA |
546 | enum fio_ddir ddir; |
547 | ||
44cbc6da JA |
548 | /* |
549 | * If verify phase started, it's always a READ | |
550 | */ | |
551 | if (td->runstate == TD_VERIFYING) | |
552 | return DDIR_READ; | |
553 | ||
5f9099ea JA |
554 | /* |
555 | * see if it's time to fsync | |
556 | */ | |
557 | if (td->o.fsync_blocks && | |
558 | !(td->io_issues[DDIR_WRITE] % td->o.fsync_blocks) && | |
559 | td->io_issues[DDIR_WRITE] && should_fsync(td)) | |
560 | return DDIR_SYNC; | |
561 | ||
562 | /* | |
563 | * see if it's time to fdatasync | |
564 | */ | |
565 | if (td->o.fdatasync_blocks && | |
566 | !(td->io_issues[DDIR_WRITE] % td->o.fdatasync_blocks) && | |
567 | td->io_issues[DDIR_WRITE] && should_fsync(td)) | |
568 | return DDIR_DATASYNC; | |
569 | ||
44f29692 JA |
570 | /* |
571 | * see if it's time to sync_file_range | |
572 | */ | |
573 | if (td->sync_file_range_nr && | |
574 | !(td->io_issues[DDIR_WRITE] % td->sync_file_range_nr) && | |
575 | td->io_issues[DDIR_WRITE] && should_fsync(td)) | |
576 | return DDIR_SYNC_FILE_RANGE; | |
577 | ||
10ba535a | 578 | if (td_rw(td)) { |
10ba535a JA |
579 | /* |
580 | * Check if it's time to seed a new data direction. | |
581 | */ | |
e4928662 | 582 | if (td->io_issues[td->rwmix_ddir] >= td->rwmix_issues) { |
e47f799f JA |
583 | /* |
584 | * Put a top limit on how many bytes we do for | |
585 | * one data direction, to avoid overflowing the | |
586 | * ranges too much | |
587 | */ | |
588 | ddir = get_rand_ddir(td); | |
e47f799f JA |
589 | |
590 | if (ddir != td->rwmix_ddir) | |
591 | set_rwmix_bytes(td); | |
592 | ||
593 | td->rwmix_ddir = ddir; | |
10ba535a | 594 | } |
581e7141 | 595 | ddir = td->rwmix_ddir; |
10ba535a | 596 | } else if (td_read(td)) |
581e7141 | 597 | ddir = DDIR_READ; |
6eaf09d6 | 598 | else if (td_write(td)) |
581e7141 | 599 | ddir = DDIR_WRITE; |
6eaf09d6 SL |
600 | else |
601 | ddir = DDIR_TRIM; | |
581e7141 JA |
602 | |
603 | td->rwmix_ddir = rate_ddir(td, ddir); | |
604 | return td->rwmix_ddir; | |
10ba535a JA |
605 | } |
606 | ||
1ef2b6be JA |
607 | static void set_rw_ddir(struct thread_data *td, struct io_u *io_u) |
608 | { | |
609 | io_u->ddir = get_rw_ddir(td); | |
610 | ||
611 | if (io_u->ddir == DDIR_WRITE && (td->io_ops->flags & FIO_BARRIER) && | |
612 | td->o.barrier_blocks && | |
613 | !(td->io_issues[DDIR_WRITE] % td->o.barrier_blocks) && | |
614 | td->io_issues[DDIR_WRITE]) | |
615 | io_u->flags |= IO_U_F_BARRIER; | |
616 | } | |
617 | ||
e8462bd8 | 618 | void put_file_log(struct thread_data *td, struct fio_file *f) |
60f2c658 JA |
619 | { |
620 | int ret = put_file(td, f); | |
621 | ||
622 | if (ret) | |
623 | td_verror(td, ret, "file close"); | |
624 | } | |
625 | ||
10ba535a JA |
626 | void put_io_u(struct thread_data *td, struct io_u *io_u) |
627 | { | |
e8462bd8 JA |
628 | td_io_u_lock(td); |
629 | ||
d7ee2a7d | 630 | if (io_u->file && !(io_u->flags & IO_U_F_FREE_DEF)) |
60f2c658 | 631 | put_file_log(td, io_u->file); |
10ba535a | 632 | io_u->file = NULL; |
d7ee2a7d SL |
633 | io_u->flags &= ~IO_U_F_FREE_DEF; |
634 | io_u->flags |= IO_U_F_FREE; | |
635 | ||
0c41214f RR |
636 | if (io_u->flags & IO_U_F_IN_CUR_DEPTH) |
637 | td->cur_depth--; | |
e8462bd8 | 638 | flist_del_init(&io_u->list); |
01743ee1 | 639 | flist_add(&io_u->list, &td->io_u_freelist); |
e8462bd8 JA |
640 | td_io_u_unlock(td); |
641 | td_io_u_free_notify(td); | |
10ba535a JA |
642 | } |
643 | ||
f2bba182 RR |
644 | void clear_io_u(struct thread_data *td, struct io_u *io_u) |
645 | { | |
646 | io_u->flags &= ~IO_U_F_FLIGHT; | |
647 | put_io_u(td, io_u); | |
648 | } | |
649 | ||
755200a3 JA |
650 | void requeue_io_u(struct thread_data *td, struct io_u **io_u) |
651 | { | |
652 | struct io_u *__io_u = *io_u; | |
653 | ||
465221b0 JA |
654 | dprint(FD_IO, "requeue %p\n", __io_u); |
655 | ||
e8462bd8 JA |
656 | td_io_u_lock(td); |
657 | ||
4d2e0f49 | 658 | __io_u->flags |= IO_U_F_FREE; |
ff58fced | 659 | if ((__io_u->flags & IO_U_F_FLIGHT) && ddir_rw(__io_u->ddir)) |
e4f54adb | 660 | td->io_issues[__io_u->ddir]--; |
5ec10eaa | 661 | |
4d2e0f49 | 662 | __io_u->flags &= ~IO_U_F_FLIGHT; |
0c41214f RR |
663 | if (__io_u->flags & IO_U_F_IN_CUR_DEPTH) |
664 | td->cur_depth--; | |
01743ee1 JA |
665 | flist_del(&__io_u->list); |
666 | flist_add_tail(&__io_u->list, &td->io_u_requeues); | |
e8462bd8 | 667 | td_io_u_unlock(td); |
755200a3 JA |
668 | *io_u = NULL; |
669 | } | |
670 | ||
9bf2061e | 671 | static int fill_io_u(struct thread_data *td, struct io_u *io_u) |
10ba535a | 672 | { |
b4c5e1ac JA |
673 | if (td->io_ops->flags & FIO_NOIO) |
674 | goto out; | |
675 | ||
1ef2b6be | 676 | set_rw_ddir(td, io_u); |
5f9099ea | 677 | |
87dc1ab1 | 678 | /* |
ff58fced | 679 | * fsync() or fdatasync() or trim etc, we are done |
87dc1ab1 | 680 | */ |
ff58fced | 681 | if (!ddir_rw(io_u->ddir)) |
c38e9468 | 682 | goto out; |
a00735e6 | 683 | |
48f5abd3 JA |
684 | /* |
685 | * See if it's time to switch to a new zone | |
686 | */ | |
13af05ae | 687 | if (td->zone_bytes >= td->o.zone_size && td->o.zone_skip) { |
48f5abd3 | 688 | td->zone_bytes = 0; |
ed335855 SN |
689 | io_u->file->file_offset += td->o.zone_range + td->o.zone_skip; |
690 | io_u->file->last_pos = io_u->file->file_offset; | |
48f5abd3 JA |
691 | td->io_skip_bytes += td->o.zone_skip; |
692 | } | |
693 | ||
10ba535a | 694 | /* |
c685b5b2 JA |
695 | * No log, let the seq/rand engine retrieve the next buflen and |
696 | * position. | |
10ba535a | 697 | */ |
2ba1c290 JA |
698 | if (get_next_offset(td, io_u)) { |
699 | dprint(FD_IO, "io_u %p, failed getting offset\n", io_u); | |
bca4ed4d | 700 | return 1; |
2ba1c290 | 701 | } |
10ba535a | 702 | |
9bf2061e | 703 | io_u->buflen = get_next_buflen(td, io_u); |
2ba1c290 JA |
704 | if (!io_u->buflen) { |
705 | dprint(FD_IO, "io_u %p, failed getting buflen\n", io_u); | |
bca4ed4d | 706 | return 1; |
2ba1c290 | 707 | } |
bca4ed4d | 708 | |
2ba1c290 JA |
709 | if (io_u->offset + io_u->buflen > io_u->file->real_file_size) { |
710 | dprint(FD_IO, "io_u %p, offset too large\n", io_u); | |
4ba66134 JA |
711 | dprint(FD_IO, " off=%llu/%lu > %llu\n", io_u->offset, |
712 | io_u->buflen, io_u->file->real_file_size); | |
6a5e6884 | 713 | return 1; |
2ba1c290 | 714 | } |
6a5e6884 | 715 | |
bca4ed4d JA |
716 | /* |
717 | * mark entry before potentially trimming io_u | |
718 | */ | |
303032ae | 719 | if (td_random(td) && file_randommap(td, io_u->file)) |
9bf2061e | 720 | mark_random_map(td, io_u); |
bca4ed4d JA |
721 | |
722 | /* | |
723 | * If using a write iolog, store this entry. | |
724 | */ | |
c38e9468 | 725 | out: |
2ba1c290 | 726 | dprint_io_u(io_u, "fill_io_u"); |
d9d91e39 | 727 | td->zone_bytes += io_u->buflen; |
f29b25a3 | 728 | log_io_u(td, io_u); |
bca4ed4d | 729 | return 0; |
10ba535a JA |
730 | } |
731 | ||
838bc709 JA |
732 | static void __io_u_mark_map(unsigned int *map, unsigned int nr) |
733 | { | |
2b13e716 | 734 | int idx = 0; |
838bc709 JA |
735 | |
736 | switch (nr) { | |
737 | default: | |
2b13e716 | 738 | idx = 6; |
838bc709 JA |
739 | break; |
740 | case 33 ... 64: | |
2b13e716 | 741 | idx = 5; |
838bc709 JA |
742 | break; |
743 | case 17 ... 32: | |
2b13e716 | 744 | idx = 4; |
838bc709 JA |
745 | break; |
746 | case 9 ... 16: | |
2b13e716 | 747 | idx = 3; |
838bc709 JA |
748 | break; |
749 | case 5 ... 8: | |
2b13e716 | 750 | idx = 2; |
838bc709 JA |
751 | break; |
752 | case 1 ... 4: | |
2b13e716 | 753 | idx = 1; |
838bc709 JA |
754 | case 0: |
755 | break; | |
756 | } | |
757 | ||
2b13e716 | 758 | map[idx]++; |
838bc709 JA |
759 | } |
760 | ||
761 | void io_u_mark_submit(struct thread_data *td, unsigned int nr) | |
762 | { | |
763 | __io_u_mark_map(td->ts.io_u_submit, nr); | |
764 | td->ts.total_submit++; | |
765 | } | |
766 | ||
767 | void io_u_mark_complete(struct thread_data *td, unsigned int nr) | |
768 | { | |
769 | __io_u_mark_map(td->ts.io_u_complete, nr); | |
770 | td->ts.total_complete++; | |
771 | } | |
772 | ||
d8005759 | 773 | void io_u_mark_depth(struct thread_data *td, unsigned int nr) |
71619dc2 | 774 | { |
2b13e716 | 775 | int idx = 0; |
71619dc2 JA |
776 | |
777 | switch (td->cur_depth) { | |
778 | default: | |
2b13e716 | 779 | idx = 6; |
a783e61a | 780 | break; |
71619dc2 | 781 | case 32 ... 63: |
2b13e716 | 782 | idx = 5; |
a783e61a | 783 | break; |
71619dc2 | 784 | case 16 ... 31: |
2b13e716 | 785 | idx = 4; |
a783e61a | 786 | break; |
71619dc2 | 787 | case 8 ... 15: |
2b13e716 | 788 | idx = 3; |
a783e61a | 789 | break; |
71619dc2 | 790 | case 4 ... 7: |
2b13e716 | 791 | idx = 2; |
a783e61a | 792 | break; |
71619dc2 | 793 | case 2 ... 3: |
2b13e716 | 794 | idx = 1; |
71619dc2 JA |
795 | case 1: |
796 | break; | |
797 | } | |
798 | ||
2b13e716 | 799 | td->ts.io_u_map[idx] += nr; |
71619dc2 JA |
800 | } |
801 | ||
04a0feae JA |
802 | static void io_u_mark_lat_usec(struct thread_data *td, unsigned long usec) |
803 | { | |
2b13e716 | 804 | int idx = 0; |
04a0feae JA |
805 | |
806 | assert(usec < 1000); | |
807 | ||
808 | switch (usec) { | |
809 | case 750 ... 999: | |
2b13e716 | 810 | idx = 9; |
04a0feae JA |
811 | break; |
812 | case 500 ... 749: | |
2b13e716 | 813 | idx = 8; |
04a0feae JA |
814 | break; |
815 | case 250 ... 499: | |
2b13e716 | 816 | idx = 7; |
04a0feae JA |
817 | break; |
818 | case 100 ... 249: | |
2b13e716 | 819 | idx = 6; |
04a0feae JA |
820 | break; |
821 | case 50 ... 99: | |
2b13e716 | 822 | idx = 5; |
04a0feae JA |
823 | break; |
824 | case 20 ... 49: | |
2b13e716 | 825 | idx = 4; |
04a0feae JA |
826 | break; |
827 | case 10 ... 19: | |
2b13e716 | 828 | idx = 3; |
04a0feae JA |
829 | break; |
830 | case 4 ... 9: | |
2b13e716 | 831 | idx = 2; |
04a0feae JA |
832 | break; |
833 | case 2 ... 3: | |
2b13e716 | 834 | idx = 1; |
04a0feae JA |
835 | case 0 ... 1: |
836 | break; | |
837 | } | |
838 | ||
2b13e716 JA |
839 | assert(idx < FIO_IO_U_LAT_U_NR); |
840 | td->ts.io_u_lat_u[idx]++; | |
04a0feae JA |
841 | } |
842 | ||
843 | static void io_u_mark_lat_msec(struct thread_data *td, unsigned long msec) | |
ec118304 | 844 | { |
2b13e716 | 845 | int idx = 0; |
ec118304 JA |
846 | |
847 | switch (msec) { | |
848 | default: | |
2b13e716 | 849 | idx = 11; |
04a0feae | 850 | break; |
8abdce66 | 851 | case 1000 ... 1999: |
2b13e716 | 852 | idx = 10; |
04a0feae | 853 | break; |
8abdce66 | 854 | case 750 ... 999: |
2b13e716 | 855 | idx = 9; |
04a0feae | 856 | break; |
8abdce66 | 857 | case 500 ... 749: |
2b13e716 | 858 | idx = 8; |
04a0feae | 859 | break; |
8abdce66 | 860 | case 250 ... 499: |
2b13e716 | 861 | idx = 7; |
04a0feae | 862 | break; |
8abdce66 | 863 | case 100 ... 249: |
2b13e716 | 864 | idx = 6; |
04a0feae | 865 | break; |
8abdce66 | 866 | case 50 ... 99: |
2b13e716 | 867 | idx = 5; |
04a0feae | 868 | break; |
8abdce66 | 869 | case 20 ... 49: |
2b13e716 | 870 | idx = 4; |
04a0feae | 871 | break; |
8abdce66 | 872 | case 10 ... 19: |
2b13e716 | 873 | idx = 3; |
04a0feae | 874 | break; |
8abdce66 | 875 | case 4 ... 9: |
2b13e716 | 876 | idx = 2; |
04a0feae | 877 | break; |
ec118304 | 878 | case 2 ... 3: |
2b13e716 | 879 | idx = 1; |
ec118304 JA |
880 | case 0 ... 1: |
881 | break; | |
882 | } | |
883 | ||
2b13e716 JA |
884 | assert(idx < FIO_IO_U_LAT_M_NR); |
885 | td->ts.io_u_lat_m[idx]++; | |
04a0feae JA |
886 | } |
887 | ||
888 | static void io_u_mark_latency(struct thread_data *td, unsigned long usec) | |
889 | { | |
890 | if (usec < 1000) | |
891 | io_u_mark_lat_usec(td, usec); | |
892 | else | |
893 | io_u_mark_lat_msec(td, usec / 1000); | |
ec118304 JA |
894 | } |
895 | ||
0aabe160 JA |
896 | /* |
897 | * Get next file to service by choosing one at random | |
898 | */ | |
2cc52930 JA |
899 | static struct fio_file *get_next_file_rand(struct thread_data *td, |
900 | enum fio_file_flags goodf, | |
d6aed795 | 901 | enum fio_file_flags badf) |
0aabe160 | 902 | { |
0aabe160 | 903 | struct fio_file *f; |
1c178180 | 904 | int fno; |
0aabe160 JA |
905 | |
906 | do { | |
87b10676 | 907 | int opened = 0; |
1294c3ec | 908 | unsigned long r; |
4c07ad86 JA |
909 | |
910 | if (td->o.use_os_rand) { | |
911 | r = os_random_long(&td->next_file_state); | |
912 | fno = (unsigned int) ((double) td->o.nr_files | |
913 | * (r / (OS_RAND_MAX + 1.0))); | |
914 | } else { | |
915 | r = __rand(&td->__next_file_state); | |
916 | fno = (unsigned int) ((double) td->o.nr_files | |
917 | * (r / (FRAND_MAX + 1.0))); | |
918 | } | |
7c83c089 | 919 | |
126d65c6 | 920 | f = td->files[fno]; |
d6aed795 | 921 | if (fio_file_done(f)) |
059e63c0 | 922 | continue; |
1c178180 | 923 | |
d6aed795 | 924 | if (!fio_file_open(f)) { |
87b10676 JA |
925 | int err; |
926 | ||
927 | err = td_io_open_file(td, f); | |
928 | if (err) | |
929 | continue; | |
930 | opened = 1; | |
931 | } | |
932 | ||
2ba1c290 JA |
933 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) { |
934 | dprint(FD_FILE, "get_next_file_rand: %p\n", f); | |
0aabe160 | 935 | return f; |
2ba1c290 | 936 | } |
87b10676 JA |
937 | if (opened) |
938 | td_io_close_file(td, f); | |
0aabe160 JA |
939 | } while (1); |
940 | } | |
941 | ||
942 | /* | |
943 | * Get next file to service by doing round robin between all available ones | |
944 | */ | |
1c178180 JA |
945 | static struct fio_file *get_next_file_rr(struct thread_data *td, int goodf, |
946 | int badf) | |
3d7c391d JA |
947 | { |
948 | unsigned int old_next_file = td->next_file; | |
949 | struct fio_file *f; | |
950 | ||
951 | do { | |
87b10676 JA |
952 | int opened = 0; |
953 | ||
126d65c6 | 954 | f = td->files[td->next_file]; |
3d7c391d JA |
955 | |
956 | td->next_file++; | |
2dc1bbeb | 957 | if (td->next_file >= td->o.nr_files) |
3d7c391d JA |
958 | td->next_file = 0; |
959 | ||
87b10676 | 960 | dprint(FD_FILE, "trying file %s %x\n", f->file_name, f->flags); |
d6aed795 | 961 | if (fio_file_done(f)) { |
d5ed68ea | 962 | f = NULL; |
059e63c0 | 963 | continue; |
d5ed68ea | 964 | } |
059e63c0 | 965 | |
d6aed795 | 966 | if (!fio_file_open(f)) { |
87b10676 JA |
967 | int err; |
968 | ||
969 | err = td_io_open_file(td, f); | |
b5696bfc JA |
970 | if (err) { |
971 | dprint(FD_FILE, "error %d on open of %s\n", | |
972 | err, f->file_name); | |
87c27b45 | 973 | f = NULL; |
87b10676 | 974 | continue; |
b5696bfc | 975 | } |
87b10676 JA |
976 | opened = 1; |
977 | } | |
978 | ||
0b9d69ec JA |
979 | dprint(FD_FILE, "goodf=%x, badf=%x, ff=%x\n", goodf, badf, |
980 | f->flags); | |
1c178180 | 981 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) |
3d7c391d JA |
982 | break; |
983 | ||
87b10676 JA |
984 | if (opened) |
985 | td_io_close_file(td, f); | |
986 | ||
3d7c391d JA |
987 | f = NULL; |
988 | } while (td->next_file != old_next_file); | |
989 | ||
2ba1c290 | 990 | dprint(FD_FILE, "get_next_file_rr: %p\n", f); |
3d7c391d JA |
991 | return f; |
992 | } | |
993 | ||
7eb36574 | 994 | static struct fio_file *__get_next_file(struct thread_data *td) |
bdb4e2e9 | 995 | { |
1907dbc6 JA |
996 | struct fio_file *f; |
997 | ||
2dc1bbeb | 998 | assert(td->o.nr_files <= td->files_index); |
1c178180 | 999 | |
b5696bfc | 1000 | if (td->nr_done_files >= td->o.nr_files) { |
5ec10eaa JA |
1001 | dprint(FD_FILE, "get_next_file: nr_open=%d, nr_done=%d," |
1002 | " nr_files=%d\n", td->nr_open_files, | |
1003 | td->nr_done_files, | |
1004 | td->o.nr_files); | |
bdb4e2e9 | 1005 | return NULL; |
2ba1c290 | 1006 | } |
bdb4e2e9 | 1007 | |
1907dbc6 | 1008 | f = td->file_service_file; |
d6aed795 | 1009 | if (f && fio_file_open(f) && !fio_file_closing(f)) { |
a086c257 JA |
1010 | if (td->o.file_service_type == FIO_FSERVICE_SEQ) |
1011 | goto out; | |
1012 | if (td->file_service_left--) | |
1013 | goto out; | |
1014 | } | |
1907dbc6 | 1015 | |
a086c257 JA |
1016 | if (td->o.file_service_type == FIO_FSERVICE_RR || |
1017 | td->o.file_service_type == FIO_FSERVICE_SEQ) | |
d6aed795 | 1018 | f = get_next_file_rr(td, FIO_FILE_open, FIO_FILE_closing); |
bdb4e2e9 | 1019 | else |
d6aed795 | 1020 | f = get_next_file_rand(td, FIO_FILE_open, FIO_FILE_closing); |
1907dbc6 JA |
1021 | |
1022 | td->file_service_file = f; | |
1023 | td->file_service_left = td->file_service_nr - 1; | |
2ba1c290 | 1024 | out: |
683023e8 | 1025 | dprint(FD_FILE, "get_next_file: %p [%s]\n", f, f->file_name); |
1907dbc6 | 1026 | return f; |
bdb4e2e9 JA |
1027 | } |
1028 | ||
7eb36574 JA |
1029 | static struct fio_file *get_next_file(struct thread_data *td) |
1030 | { | |
d72be545 JA |
1031 | if (!(td->flags & TD_F_PROFILE_OPS)) { |
1032 | struct prof_io_ops *ops = &td->prof_io_ops; | |
7eb36574 | 1033 | |
d72be545 JA |
1034 | if (ops->get_next_file) |
1035 | return ops->get_next_file(td); | |
1036 | } | |
7eb36574 JA |
1037 | |
1038 | return __get_next_file(td); | |
1039 | } | |
1040 | ||
429f6675 JA |
1041 | static int set_io_u_file(struct thread_data *td, struct io_u *io_u) |
1042 | { | |
1043 | struct fio_file *f; | |
1044 | ||
1045 | do { | |
1046 | f = get_next_file(td); | |
1047 | if (!f) | |
1048 | return 1; | |
1049 | ||
429f6675 JA |
1050 | io_u->file = f; |
1051 | get_file(f); | |
1052 | ||
1053 | if (!fill_io_u(td, io_u)) | |
1054 | break; | |
1055 | ||
b5696bfc | 1056 | put_file_log(td, f); |
429f6675 | 1057 | td_io_close_file(td, f); |
b5696bfc | 1058 | io_u->file = NULL; |
d6aed795 | 1059 | fio_file_set_done(f); |
429f6675 | 1060 | td->nr_done_files++; |
0b9d69ec JA |
1061 | dprint(FD_FILE, "%s: is done (%d of %d)\n", f->file_name, |
1062 | td->nr_done_files, td->o.nr_files); | |
429f6675 JA |
1063 | } while (1); |
1064 | ||
1065 | return 0; | |
1066 | } | |
1067 | ||
1068 | ||
10ba535a JA |
1069 | struct io_u *__get_io_u(struct thread_data *td) |
1070 | { | |
1071 | struct io_u *io_u = NULL; | |
1072 | ||
e8462bd8 JA |
1073 | td_io_u_lock(td); |
1074 | ||
1075 | again: | |
01743ee1 JA |
1076 | if (!flist_empty(&td->io_u_requeues)) |
1077 | io_u = flist_entry(td->io_u_requeues.next, struct io_u, list); | |
755200a3 | 1078 | else if (!queue_full(td)) { |
01743ee1 | 1079 | io_u = flist_entry(td->io_u_freelist.next, struct io_u, list); |
10ba535a | 1080 | |
6040dabc | 1081 | io_u->buflen = 0; |
10ba535a | 1082 | io_u->resid = 0; |
755200a3 | 1083 | io_u->file = NULL; |
d7762cf8 | 1084 | io_u->end_io = NULL; |
755200a3 JA |
1085 | } |
1086 | ||
1087 | if (io_u) { | |
0c6e7517 | 1088 | assert(io_u->flags & IO_U_F_FREE); |
2ecc1b57 | 1089 | io_u->flags &= ~(IO_U_F_FREE | IO_U_F_FREE_DEF); |
1ef2b6be | 1090 | io_u->flags &= ~(IO_U_F_TRIMMED | IO_U_F_BARRIER); |
82af2a7c | 1091 | io_u->flags &= ~IO_U_F_VER_LIST; |
0c6e7517 | 1092 | |
755200a3 | 1093 | io_u->error = 0; |
01743ee1 | 1094 | flist_del(&io_u->list); |
94c6372c | 1095 | flist_add_tail(&io_u->list, &td->io_u_busylist); |
10ba535a | 1096 | td->cur_depth++; |
0c41214f | 1097 | io_u->flags |= IO_U_F_IN_CUR_DEPTH; |
1dec3e07 JA |
1098 | } else if (td->o.verify_async) { |
1099 | /* | |
1100 | * We ran out, wait for async verify threads to finish and | |
1101 | * return one | |
1102 | */ | |
1103 | pthread_cond_wait(&td->free_cond, &td->io_u_lock); | |
1104 | goto again; | |
10ba535a JA |
1105 | } |
1106 | ||
e8462bd8 | 1107 | td_io_u_unlock(td); |
10ba535a JA |
1108 | return io_u; |
1109 | } | |
1110 | ||
0d29de83 | 1111 | static int check_get_trim(struct thread_data *td, struct io_u *io_u) |
10ba535a | 1112 | { |
d72be545 JA |
1113 | if (!(td->flags & TD_F_TRIM_BACKLOG)) |
1114 | return 0; | |
1115 | ||
1116 | if (td->trim_entries) { | |
0d29de83 | 1117 | int get_trim = 0; |
10ba535a | 1118 | |
0d29de83 JA |
1119 | if (td->trim_batch) { |
1120 | td->trim_batch--; | |
1121 | get_trim = 1; | |
1122 | } else if (!(td->io_hist_len % td->o.trim_backlog) && | |
1123 | td->last_ddir != DDIR_READ) { | |
1124 | td->trim_batch = td->o.trim_batch; | |
1125 | if (!td->trim_batch) | |
1126 | td->trim_batch = td->o.trim_backlog; | |
1127 | get_trim = 1; | |
1128 | } | |
1129 | ||
1130 | if (get_trim && !get_next_trim(td, io_u)) | |
1131 | return 1; | |
2ba1c290 | 1132 | } |
10ba535a | 1133 | |
0d29de83 JA |
1134 | return 0; |
1135 | } | |
1136 | ||
1137 | static int check_get_verify(struct thread_data *td, struct io_u *io_u) | |
1138 | { | |
d72be545 JA |
1139 | if (!(td->flags & TD_F_VER_BACKLOG)) |
1140 | return 0; | |
1141 | ||
1142 | if (td->io_hist_len) { | |
9e144189 JA |
1143 | int get_verify = 0; |
1144 | ||
d1ece0c7 | 1145 | if (td->verify_batch) |
9e144189 | 1146 | get_verify = 1; |
d1ece0c7 | 1147 | else if (!(td->io_hist_len % td->o.verify_backlog) && |
9e144189 JA |
1148 | td->last_ddir != DDIR_READ) { |
1149 | td->verify_batch = td->o.verify_batch; | |
f8a75c99 JA |
1150 | if (!td->verify_batch) |
1151 | td->verify_batch = td->o.verify_backlog; | |
9e144189 JA |
1152 | get_verify = 1; |
1153 | } | |
1154 | ||
d1ece0c7 JA |
1155 | if (get_verify && !get_next_verify(td, io_u)) { |
1156 | td->verify_batch--; | |
0d29de83 | 1157 | return 1; |
d1ece0c7 | 1158 | } |
9e144189 JA |
1159 | } |
1160 | ||
0d29de83 JA |
1161 | return 0; |
1162 | } | |
1163 | ||
de789769 JA |
1164 | /* |
1165 | * Fill offset and start time into the buffer content, to prevent too | |
23f394d5 JA |
1166 | * easy compressible data for simple de-dupe attempts. Do this for every |
1167 | * 512b block in the range, since that should be the smallest block size | |
1168 | * we can expect from a device. | |
de789769 JA |
1169 | */ |
1170 | static void small_content_scramble(struct io_u *io_u) | |
1171 | { | |
23f394d5 | 1172 | unsigned int i, nr_blocks = io_u->buflen / 512; |
1ae83d45 | 1173 | uint64_t boffset; |
23f394d5 JA |
1174 | unsigned int offset; |
1175 | void *p, *end; | |
de789769 | 1176 | |
23f394d5 JA |
1177 | if (!nr_blocks) |
1178 | return; | |
1179 | ||
1180 | p = io_u->xfer_buf; | |
fba76ee8 | 1181 | boffset = io_u->offset; |
81f0366c | 1182 | io_u->buf_filled_len = 0; |
fad82f76 | 1183 | |
23f394d5 JA |
1184 | for (i = 0; i < nr_blocks; i++) { |
1185 | /* | |
1186 | * Fill the byte offset into a "random" start offset of | |
1187 | * the buffer, given by the product of the usec time | |
1188 | * and the actual offset. | |
1189 | */ | |
fad82f76 | 1190 | offset = (io_u->start_time.tv_usec ^ boffset) & 511; |
1ae83d45 JA |
1191 | offset &= ~(sizeof(uint64_t) - 1); |
1192 | if (offset >= 512 - sizeof(uint64_t)) | |
1193 | offset -= sizeof(uint64_t); | |
fba76ee8 | 1194 | memcpy(p + offset, &boffset, sizeof(boffset)); |
23f394d5 JA |
1195 | |
1196 | end = p + 512 - sizeof(io_u->start_time); | |
1197 | memcpy(end, &io_u->start_time, sizeof(io_u->start_time)); | |
1198 | p += 512; | |
fad82f76 | 1199 | boffset += 512; |
23f394d5 | 1200 | } |
de789769 JA |
1201 | } |
1202 | ||
0d29de83 JA |
1203 | /* |
1204 | * Return an io_u to be processed. Gets a buflen and offset, sets direction, | |
1205 | * etc. The returned io_u is fully ready to be prepped and submitted. | |
1206 | */ | |
1207 | struct io_u *get_io_u(struct thread_data *td) | |
1208 | { | |
1209 | struct fio_file *f; | |
1210 | struct io_u *io_u; | |
de789769 | 1211 | int do_scramble = 0; |
0d29de83 JA |
1212 | |
1213 | io_u = __get_io_u(td); | |
1214 | if (!io_u) { | |
1215 | dprint(FD_IO, "__get_io_u failed\n"); | |
1216 | return NULL; | |
1217 | } | |
1218 | ||
1219 | if (check_get_verify(td, io_u)) | |
1220 | goto out; | |
1221 | if (check_get_trim(td, io_u)) | |
1222 | goto out; | |
1223 | ||
755200a3 JA |
1224 | /* |
1225 | * from a requeue, io_u already setup | |
1226 | */ | |
1227 | if (io_u->file) | |
77f392bf | 1228 | goto out; |
755200a3 | 1229 | |
429f6675 JA |
1230 | /* |
1231 | * If using an iolog, grab next piece if any available. | |
1232 | */ | |
d72be545 | 1233 | if (td->flags & TD_F_READ_IOLOG) { |
429f6675 JA |
1234 | if (read_iolog_get(td, io_u)) |
1235 | goto err_put; | |
2ba1c290 JA |
1236 | } else if (set_io_u_file(td, io_u)) { |
1237 | dprint(FD_IO, "io_u %p, setting file failed\n", io_u); | |
429f6675 | 1238 | goto err_put; |
2ba1c290 | 1239 | } |
5ec10eaa | 1240 | |
429f6675 | 1241 | f = io_u->file; |
d6aed795 | 1242 | assert(fio_file_open(f)); |
97af62ce | 1243 | |
ff58fced | 1244 | if (ddir_rw(io_u->ddir)) { |
d0656a93 | 1245 | if (!io_u->buflen && !(td->io_ops->flags & FIO_NOIO)) { |
2ba1c290 | 1246 | dprint(FD_IO, "get_io_u: zero buflen on %p\n", io_u); |
429f6675 | 1247 | goto err_put; |
2ba1c290 | 1248 | } |
10ba535a | 1249 | |
38dad62d | 1250 | f->last_start = io_u->offset; |
36167d82 | 1251 | f->last_pos = io_u->offset + io_u->buflen; |
10ba535a | 1252 | |
fd68418e | 1253 | if (io_u->ddir == DDIR_WRITE) { |
d72be545 | 1254 | if (td->flags & TD_F_REFILL_BUFFERS) { |
9c42684e JA |
1255 | io_u_fill_buffer(td, io_u, |
1256 | io_u->xfer_buflen, io_u->xfer_buflen); | |
d72be545 | 1257 | } else if (td->flags & TD_F_SCRAMBLE_BUFFERS) |
fd68418e | 1258 | do_scramble = 1; |
d72be545 | 1259 | if (td->flags & TD_F_VER_NONE) { |
629f1d71 JA |
1260 | populate_verify_io_u(td, io_u); |
1261 | do_scramble = 0; | |
1262 | } | |
fd68418e | 1263 | } else if (io_u->ddir == DDIR_READ) { |
cbe8d756 RR |
1264 | /* |
1265 | * Reset the buf_filled parameters so next time if the | |
1266 | * buffer is used for writes it is refilled. | |
1267 | */ | |
cbe8d756 RR |
1268 | io_u->buf_filled_len = 0; |
1269 | } | |
87dc1ab1 | 1270 | } |
10ba535a | 1271 | |
165faf16 JA |
1272 | /* |
1273 | * Set io data pointers. | |
1274 | */ | |
cec6b55d JA |
1275 | io_u->xfer_buf = io_u->buf; |
1276 | io_u->xfer_buflen = io_u->buflen; | |
5973cafb | 1277 | |
6ac7a331 | 1278 | out: |
0d29de83 | 1279 | assert(io_u->file); |
429f6675 | 1280 | if (!td_io_prep(td, io_u)) { |
993bf48b JA |
1281 | if (!td->o.disable_slat) |
1282 | fio_gettime(&io_u->start_time, NULL); | |
de789769 JA |
1283 | if (do_scramble) |
1284 | small_content_scramble(io_u); | |
429f6675 | 1285 | return io_u; |
36167d82 | 1286 | } |
429f6675 | 1287 | err_put: |
2ba1c290 | 1288 | dprint(FD_IO, "get_io_u failed\n"); |
429f6675 JA |
1289 | put_io_u(td, io_u); |
1290 | return NULL; | |
10ba535a JA |
1291 | } |
1292 | ||
5451792e JA |
1293 | void io_u_log_error(struct thread_data *td, struct io_u *io_u) |
1294 | { | |
8b28bd41 | 1295 | enum error_type_bit eb = td_error_type(io_u->ddir, io_u->error); |
825f818e JA |
1296 | const char *msg[] = { "read", "write", "sync", "datasync", |
1297 | "sync_file_range", "wait", "trim" }; | |
1298 | ||
8b28bd41 DM |
1299 | if (td_non_fatal_error(td, eb, io_u->error) && !td->o.error_dump) |
1300 | return; | |
5451792e JA |
1301 | |
1302 | log_err("fio: io_u error"); | |
1303 | ||
1304 | if (io_u->file) | |
1305 | log_err(" on file %s", io_u->file->file_name); | |
1306 | ||
1307 | log_err(": %s\n", strerror(io_u->error)); | |
1308 | ||
5ec10eaa JA |
1309 | log_err(" %s offset=%llu, buflen=%lu\n", msg[io_u->ddir], |
1310 | io_u->offset, io_u->xfer_buflen); | |
5451792e JA |
1311 | |
1312 | if (!td->error) | |
1313 | td_verror(td, io_u->error, "io_u error"); | |
1314 | } | |
1315 | ||
c8eeb9df JA |
1316 | static void account_io_completion(struct thread_data *td, struct io_u *io_u, |
1317 | struct io_completion_data *icd, | |
1318 | const enum fio_ddir idx, unsigned int bytes) | |
1319 | { | |
24d23ca7 | 1320 | unsigned long lusec = 0; |
c8eeb9df | 1321 | |
c8eeb9df JA |
1322 | if (!td->o.disable_clat || !td->o.disable_bw) |
1323 | lusec = utime_since(&io_u->issue_time, &icd->time); | |
1324 | ||
1325 | if (!td->o.disable_lat) { | |
1326 | unsigned long tusec; | |
1327 | ||
1328 | tusec = utime_since(&io_u->start_time, &icd->time); | |
1329 | add_lat_sample(td, idx, tusec, bytes); | |
15501535 JA |
1330 | |
1331 | if (td->o.max_latency && tusec > td->o.max_latency) { | |
1332 | if (!td->error) | |
1333 | log_err("fio: latency of %lu usec exceeds specified max (%u usec)\n", tusec, td->o.max_latency); | |
c5b3adc4 JA |
1334 | td_verror(td, ETIMEDOUT, "max latency exceeded"); |
1335 | icd->error = ETIMEDOUT; | |
15501535 | 1336 | } |
c8eeb9df JA |
1337 | } |
1338 | ||
1339 | if (!td->o.disable_clat) { | |
1340 | add_clat_sample(td, idx, lusec, bytes); | |
1341 | io_u_mark_latency(td, lusec); | |
1342 | } | |
1343 | ||
1344 | if (!td->o.disable_bw) | |
1345 | add_bw_sample(td, idx, bytes, &icd->time); | |
1346 | ||
1347 | add_iops_sample(td, idx, &icd->time); | |
1348 | } | |
1349 | ||
1b8dbf25 SL |
1350 | static long long usec_for_io(struct thread_data *td, enum fio_ddir ddir) |
1351 | { | |
1ae83d45 JA |
1352 | uint64_t secs, remainder, bps, bytes; |
1353 | ||
1b8dbf25 SL |
1354 | bytes = td->this_io_bytes[ddir]; |
1355 | bps = td->rate_bps[ddir]; | |
1356 | secs = bytes / bps; | |
1357 | remainder = bytes % bps; | |
1358 | return remainder * 1000000 / bps + secs * 1000000; | |
1359 | } | |
1360 | ||
97601024 JA |
1361 | static void io_completed(struct thread_data *td, struct io_u *io_u, |
1362 | struct io_completion_data *icd) | |
10ba535a | 1363 | { |
44f29692 | 1364 | struct fio_file *f; |
10ba535a | 1365 | |
2ba1c290 JA |
1366 | dprint_io_u(io_u, "io complete"); |
1367 | ||
2ecc1b57 | 1368 | td_io_u_lock(td); |
0c6e7517 | 1369 | assert(io_u->flags & IO_U_F_FLIGHT); |
38dad62d | 1370 | io_u->flags &= ~(IO_U_F_FLIGHT | IO_U_F_BUSY_OK); |
2ecc1b57 | 1371 | td_io_u_unlock(td); |
0c6e7517 | 1372 | |
5f9099ea | 1373 | if (ddir_sync(io_u->ddir)) { |
87dc1ab1 | 1374 | td->last_was_sync = 1; |
44f29692 JA |
1375 | f = io_u->file; |
1376 | if (f) { | |
1377 | f->first_write = -1ULL; | |
1378 | f->last_write = -1ULL; | |
1379 | } | |
87dc1ab1 JA |
1380 | return; |
1381 | } | |
1382 | ||
1383 | td->last_was_sync = 0; | |
9e144189 | 1384 | td->last_ddir = io_u->ddir; |
87dc1ab1 | 1385 | |
ff58fced | 1386 | if (!io_u->error && ddir_rw(io_u->ddir)) { |
10ba535a | 1387 | unsigned int bytes = io_u->buflen - io_u->resid; |
1e97cce9 | 1388 | const enum fio_ddir idx = io_u->ddir; |
ba3e4e0c | 1389 | const enum fio_ddir odx = io_u->ddir ^ 1; |
b29ee5b3 | 1390 | int ret; |
10ba535a | 1391 | |
b29ee5b3 | 1392 | td->io_blocks[idx]++; |
c8eeb9df | 1393 | td->this_io_blocks[idx]++; |
b29ee5b3 | 1394 | td->io_bytes[idx] += bytes; |
ae2fafc8 | 1395 | |
1396 | if (!(io_u->flags & IO_U_F_VER_LIST)) | |
1397 | td->this_io_bytes[idx] += bytes; | |
10ba535a | 1398 | |
44f29692 JA |
1399 | if (idx == DDIR_WRITE) { |
1400 | f = io_u->file; | |
1401 | if (f) { | |
1402 | if (f->first_write == -1ULL || | |
1403 | io_u->offset < f->first_write) | |
1404 | f->first_write = io_u->offset; | |
1405 | if (f->last_write == -1ULL || | |
1406 | ((io_u->offset + bytes) > f->last_write)) | |
1407 | f->last_write = io_u->offset + bytes; | |
1408 | } | |
1409 | } | |
1410 | ||
6b1190fd SL |
1411 | if (ramp_time_over(td) && (td->runstate == TD_RUNNING || |
1412 | td->runstate == TD_VERIFYING)) { | |
c8eeb9df | 1413 | account_io_completion(td, io_u, icd, idx, bytes); |
40e1a6f0 | 1414 | |
b23b6a2f | 1415 | if (__should_check_rate(td, idx)) { |
ba3e4e0c | 1416 | td->rate_pending_usleep[idx] = |
1b8dbf25 | 1417 | (usec_for_io(td, idx) - |
ba3e4e0c | 1418 | utime_since_now(&td->start)); |
b23b6a2f | 1419 | } |
6eaf09d6 | 1420 | if (idx != DDIR_TRIM && __should_check_rate(td, odx)) |
ba3e4e0c | 1421 | td->rate_pending_usleep[odx] = |
1b8dbf25 | 1422 | (usec_for_io(td, odx) - |
ba3e4e0c | 1423 | utime_since_now(&td->start)); |
721938ae | 1424 | } |
10ba535a | 1425 | |
660a1cb5 | 1426 | if (td_write(td) && idx == DDIR_WRITE && |
8670579e | 1427 | td->o.do_verify && |
44cbc6da JA |
1428 | td->o.verify != VERIFY_NONE && |
1429 | !td->o.experimental_verify) | |
10ba535a JA |
1430 | log_io_piece(td, io_u); |
1431 | ||
b29ee5b3 | 1432 | icd->bytes_done[idx] += bytes; |
3af6ef39 | 1433 | |
d7762cf8 | 1434 | if (io_u->end_io) { |
36690c9b | 1435 | ret = io_u->end_io(td, io_u); |
3af6ef39 JA |
1436 | if (ret && !icd->error) |
1437 | icd->error = ret; | |
1438 | } | |
ff58fced | 1439 | } else if (io_u->error) { |
10ba535a | 1440 | icd->error = io_u->error; |
5451792e JA |
1441 | io_u_log_error(td, io_u); |
1442 | } | |
8b28bd41 DM |
1443 | if (icd->error) { |
1444 | enum error_type_bit eb = td_error_type(io_u->ddir, icd->error); | |
1445 | if (!td_non_fatal_error(td, eb, icd->error)) | |
1446 | return; | |
f2bba182 RR |
1447 | /* |
1448 | * If there is a non_fatal error, then add to the error count | |
1449 | * and clear all the errors. | |
1450 | */ | |
1451 | update_error_count(td, icd->error); | |
1452 | td_clear_error(td); | |
1453 | icd->error = 0; | |
1454 | io_u->error = 0; | |
1455 | } | |
10ba535a JA |
1456 | } |
1457 | ||
9520ebb9 JA |
1458 | static void init_icd(struct thread_data *td, struct io_completion_data *icd, |
1459 | int nr) | |
10ba535a | 1460 | { |
6eaf09d6 | 1461 | int ddir; |
9520ebb9 JA |
1462 | if (!td->o.disable_clat || !td->o.disable_bw) |
1463 | fio_gettime(&icd->time, NULL); | |
02bcaa8c | 1464 | |
3af6ef39 JA |
1465 | icd->nr = nr; |
1466 | ||
10ba535a | 1467 | icd->error = 0; |
6eaf09d6 SL |
1468 | for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) |
1469 | icd->bytes_done[ddir] = 0; | |
36167d82 JA |
1470 | } |
1471 | ||
97601024 JA |
1472 | static void ios_completed(struct thread_data *td, |
1473 | struct io_completion_data *icd) | |
36167d82 JA |
1474 | { |
1475 | struct io_u *io_u; | |
1476 | int i; | |
1477 | ||
10ba535a JA |
1478 | for (i = 0; i < icd->nr; i++) { |
1479 | io_u = td->io_ops->event(td, i); | |
1480 | ||
1481 | io_completed(td, io_u, icd); | |
e8462bd8 JA |
1482 | |
1483 | if (!(io_u->flags & IO_U_F_FREE_DEF)) | |
1484 | put_io_u(td, io_u); | |
10ba535a JA |
1485 | } |
1486 | } | |
97601024 | 1487 | |
e7e6cfb4 JA |
1488 | /* |
1489 | * Complete a single io_u for the sync engines. | |
1490 | */ | |
581e7141 JA |
1491 | int io_u_sync_complete(struct thread_data *td, struct io_u *io_u, |
1492 | unsigned long *bytes) | |
97601024 JA |
1493 | { |
1494 | struct io_completion_data icd; | |
1495 | ||
9520ebb9 | 1496 | init_icd(td, &icd, 1); |
97601024 | 1497 | io_completed(td, io_u, &icd); |
e8462bd8 JA |
1498 | |
1499 | if (!(io_u->flags & IO_U_F_FREE_DEF)) | |
1500 | put_io_u(td, io_u); | |
97601024 | 1501 | |
581e7141 JA |
1502 | if (icd.error) { |
1503 | td_verror(td, icd.error, "io_u_sync_complete"); | |
1504 | return -1; | |
1505 | } | |
97601024 | 1506 | |
581e7141 | 1507 | if (bytes) { |
6eaf09d6 SL |
1508 | int ddir; |
1509 | ||
1510 | for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) | |
1511 | bytes[ddir] += icd.bytes_done[ddir]; | |
581e7141 JA |
1512 | } |
1513 | ||
1514 | return 0; | |
97601024 JA |
1515 | } |
1516 | ||
e7e6cfb4 JA |
1517 | /* |
1518 | * Called to complete min_events number of io for the async engines. | |
1519 | */ | |
581e7141 JA |
1520 | int io_u_queued_complete(struct thread_data *td, int min_evts, |
1521 | unsigned long *bytes) | |
97601024 | 1522 | { |
97601024 | 1523 | struct io_completion_data icd; |
00de55ef | 1524 | struct timespec *tvp = NULL; |
97601024 | 1525 | int ret; |
4d06a338 | 1526 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0, }; |
97601024 | 1527 | |
4950421a | 1528 | dprint(FD_IO, "io_u_queued_completed: min=%d\n", min_evts); |
b271fe62 | 1529 | |
4950421a | 1530 | if (!min_evts) |
00de55ef | 1531 | tvp = &ts; |
97601024 | 1532 | |
4950421a | 1533 | ret = td_io_getevents(td, min_evts, td->o.iodepth_batch_complete, tvp); |
97601024 | 1534 | if (ret < 0) { |
e1161c32 | 1535 | td_verror(td, -ret, "td_io_getevents"); |
97601024 JA |
1536 | return ret; |
1537 | } else if (!ret) | |
1538 | return ret; | |
1539 | ||
9520ebb9 | 1540 | init_icd(td, &icd, ret); |
97601024 | 1541 | ios_completed(td, &icd); |
581e7141 JA |
1542 | if (icd.error) { |
1543 | td_verror(td, icd.error, "io_u_queued_complete"); | |
1544 | return -1; | |
1545 | } | |
97601024 | 1546 | |
581e7141 | 1547 | if (bytes) { |
6eaf09d6 SL |
1548 | int ddir; |
1549 | ||
1550 | for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) | |
1551 | bytes[ddir] += icd.bytes_done[ddir]; | |
581e7141 JA |
1552 | } |
1553 | ||
1554 | return 0; | |
97601024 | 1555 | } |
7e77dd02 JA |
1556 | |
1557 | /* | |
1558 | * Call when io_u is really queued, to update the submission latency. | |
1559 | */ | |
1560 | void io_u_queued(struct thread_data *td, struct io_u *io_u) | |
1561 | { | |
9520ebb9 JA |
1562 | if (!td->o.disable_slat) { |
1563 | unsigned long slat_time; | |
7e77dd02 | 1564 | |
9520ebb9 | 1565 | slat_time = utime_since(&io_u->start_time, &io_u->issue_time); |
29a90ddb | 1566 | add_slat_sample(td, io_u->ddir, slat_time, io_u->xfer_buflen); |
9520ebb9 | 1567 | } |
7e77dd02 | 1568 | } |
433afcb4 | 1569 | |
5973cafb JA |
1570 | /* |
1571 | * "randomly" fill the buffer contents | |
1572 | */ | |
1573 | void io_u_fill_buffer(struct thread_data *td, struct io_u *io_u, | |
9c42684e | 1574 | unsigned int min_write, unsigned int max_bs) |
5973cafb | 1575 | { |
e84b371e JA |
1576 | io_u->buf_filled_len = 0; |
1577 | ||
9c42684e JA |
1578 | if (!td->o.zero_buffers) { |
1579 | unsigned int perc = td->o.compress_percentage; | |
1580 | ||
1581 | if (perc) { | |
f97a43a1 JA |
1582 | unsigned int seg = min_write; |
1583 | ||
1584 | seg = min(min_write, td->o.compress_chunk); | |
9c42684e | 1585 | fill_random_buf_percentage(&td->buf_state, io_u->buf, |
f97a43a1 | 1586 | perc, seg, max_bs); |
9c42684e JA |
1587 | } else |
1588 | fill_random_buf(&td->buf_state, io_u->buf, max_bs); | |
1589 | } else | |
637ef8d9 | 1590 | memset(io_u->buf, 0, max_bs); |
5973cafb | 1591 | } |