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" |
10ba535a | 13 | |
97601024 JA |
14 | struct io_completion_data { |
15 | int nr; /* input */ | |
97601024 JA |
16 | |
17 | int error; /* output */ | |
6eaf09d6 | 18 | unsigned long bytes_done[DDIR_RWDIR_CNT]; /* output */ |
97601024 JA |
19 | struct timeval time; /* output */ |
20 | }; | |
21 | ||
10ba535a JA |
22 | /* |
23 | * The ->file_map[] contains a map of blocks we have or have not done io | |
24 | * to yet. Used to make sure we cover the entire range in a fair fashion. | |
25 | */ | |
aec2de20 | 26 | static int random_map_free(struct fio_file *f, const unsigned long long block) |
10ba535a | 27 | { |
aec2de20 JA |
28 | unsigned int idx = RAND_MAP_IDX(f, block); |
29 | unsigned int bit = RAND_MAP_BIT(f, block); | |
10ba535a | 30 | |
84422acd JA |
31 | dprint(FD_RANDOM, "free: b=%llu, idx=%u, bit=%u\n", block, idx, bit); |
32 | ||
0ce8b119 | 33 | return (f->file_map[idx] & (1UL << bit)) == 0; |
10ba535a JA |
34 | } |
35 | ||
df415585 JA |
36 | /* |
37 | * Mark a given offset as used in the map. | |
38 | */ | |
9bf2061e | 39 | static void mark_random_map(struct thread_data *td, struct io_u *io_u) |
df415585 | 40 | { |
2dc1bbeb | 41 | unsigned int min_bs = td->o.rw_min_bs; |
9bf2061e | 42 | struct fio_file *f = io_u->file; |
a00735e6 | 43 | unsigned long long block; |
3e3357b1 | 44 | unsigned int blocks, nr_blocks; |
38dad62d | 45 | int busy_check; |
df415585 | 46 | |
b9c5b644 | 47 | block = (io_u->offset - f->file_offset) / (unsigned long long) min_bs; |
c685b5b2 | 48 | nr_blocks = (io_u->buflen + min_bs - 1) / min_bs; |
3e3357b1 | 49 | blocks = 0; |
38dad62d | 50 | busy_check = !(io_u->flags & IO_U_F_BUSY_OK); |
c685b5b2 | 51 | |
3e3357b1 | 52 | while (nr_blocks) { |
df415585 | 53 | unsigned int idx, bit; |
0ce8b119 | 54 | unsigned long mask, this_blocks; |
df415585 | 55 | |
1e3d53ac JA |
56 | /* |
57 | * If we have a mixed random workload, we may | |
58 | * encounter blocks we already did IO to. | |
59 | */ | |
38dad62d JA |
60 | if (!busy_check) { |
61 | blocks = nr_blocks; | |
62 | break; | |
63 | } | |
5736c10d | 64 | if ((td->o.ddir_seq_nr == 1) && !random_map_free(f, block)) |
df415585 JA |
65 | break; |
66 | ||
aec2de20 JA |
67 | idx = RAND_MAP_IDX(f, block); |
68 | bit = RAND_MAP_BIT(f, block); | |
df415585 | 69 | |
0032bf9f | 70 | fio_assert(td, idx < f->num_maps); |
df415585 | 71 | |
3e3357b1 JA |
72 | this_blocks = nr_blocks; |
73 | if (this_blocks + bit > BLOCKS_PER_MAP) | |
74 | this_blocks = BLOCKS_PER_MAP - bit; | |
75 | ||
da3758bf JA |
76 | do { |
77 | if (this_blocks == BLOCKS_PER_MAP) | |
0ce8b119 | 78 | mask = -1UL; |
da3758bf | 79 | else |
0ce8b119 | 80 | mask = ((1UL << this_blocks) - 1) << bit; |
3c3ed070 | 81 | |
da3758bf JA |
82 | if (!(f->file_map[idx] & mask)) |
83 | break; | |
84 | ||
85 | this_blocks--; | |
86 | } while (this_blocks); | |
87 | ||
88 | if (!this_blocks) | |
89 | break; | |
3e3357b1 | 90 | |
3e3357b1 JA |
91 | f->file_map[idx] |= mask; |
92 | nr_blocks -= this_blocks; | |
93 | blocks += this_blocks; | |
c9dd34b2 | 94 | block += this_blocks; |
df415585 JA |
95 | } |
96 | ||
a00735e6 JA |
97 | if ((blocks * min_bs) < io_u->buflen) |
98 | io_u->buflen = blocks * min_bs; | |
df415585 JA |
99 | } |
100 | ||
3e3357b1 JA |
101 | static unsigned long long last_block(struct thread_data *td, struct fio_file *f, |
102 | enum fio_ddir ddir) | |
2ba1c290 JA |
103 | { |
104 | unsigned long long max_blocks; | |
d9dd70f7 | 105 | unsigned long long max_size; |
2ba1c290 | 106 | |
ff58fced JA |
107 | assert(ddir_rw(ddir)); |
108 | ||
d9dd70f7 JA |
109 | /* |
110 | * Hmm, should we make sure that ->io_size <= ->real_file_size? | |
111 | */ | |
112 | max_size = f->io_size; | |
113 | if (max_size > f->real_file_size) | |
114 | max_size = f->real_file_size; | |
115 | ||
ed335855 SN |
116 | if (td->o.zone_range) |
117 | max_size = td->o.zone_range; | |
118 | ||
2b7a01d0 | 119 | max_blocks = max_size / (unsigned long long) td->o.ba[ddir]; |
2ba1c290 JA |
120 | if (!max_blocks) |
121 | return 0; | |
122 | ||
67778e88 | 123 | return max_blocks; |
2ba1c290 JA |
124 | } |
125 | ||
10ba535a JA |
126 | /* |
127 | * Return the next free block in the map. | |
128 | */ | |
129 | static int get_next_free_block(struct thread_data *td, struct fio_file *f, | |
4ba66134 | 130 | enum fio_ddir ddir, unsigned long long *b) |
10ba535a | 131 | { |
0ce8b119 | 132 | unsigned long long block, min_bs = td->o.rw_min_bs, lastb; |
10ba535a JA |
133 | int i; |
134 | ||
5e0baa7f JA |
135 | lastb = last_block(td, f, ddir); |
136 | if (!lastb) | |
137 | return 1; | |
138 | ||
c685b5b2 | 139 | i = f->last_free_lookup; |
0ce8b119 JA |
140 | block = i * BLOCKS_PER_MAP; |
141 | while (block * min_bs < f->real_file_size && | |
142 | block * min_bs < f->io_size) { | |
143 | if (f->file_map[i] != -1UL) { | |
144 | block += ffz(f->file_map[i]); | |
145 | if (block > lastb) | |
2ba1c290 | 146 | break; |
c685b5b2 | 147 | f->last_free_lookup = i; |
0ce8b119 | 148 | *b = block; |
10ba535a JA |
149 | return 0; |
150 | } | |
151 | ||
0ce8b119 | 152 | block += BLOCKS_PER_MAP; |
10ba535a JA |
153 | i++; |
154 | } | |
155 | ||
2ba1c290 | 156 | dprint(FD_IO, "failed finding a free block\n"); |
10ba535a JA |
157 | return 1; |
158 | } | |
159 | ||
e25839d4 JA |
160 | static int __get_next_rand_offset(struct thread_data *td, struct fio_file *f, |
161 | enum fio_ddir ddir, unsigned long long *b) | |
ec4015da | 162 | { |
15b87723 | 163 | unsigned long long rmax, r, lastb; |
ec4015da JA |
164 | int loops = 5; |
165 | ||
5e0baa7f JA |
166 | lastb = last_block(td, f, ddir); |
167 | if (!lastb) | |
168 | return 1; | |
169 | ||
0ce8b119 JA |
170 | if (f->failed_rands >= 200) |
171 | goto ffz; | |
172 | ||
15b87723 | 173 | rmax = td->o.use_os_rand ? OS_RAND_MAX : FRAND_MAX; |
ec4015da | 174 | do { |
15b87723 | 175 | if (td->o.use_os_rand) |
2615cc4b | 176 | r = os_random_long(&td->random_state); |
15b87723 | 177 | else |
2615cc4b | 178 | r = __rand(&td->__random_state); |
15b87723 JA |
179 | |
180 | *b = (lastb - 1) * (r / ((unsigned long long) rmax + 1.0)); | |
2615cc4b | 181 | |
84422acd | 182 | dprint(FD_RANDOM, "off rand %llu\n", r); |
2615cc4b | 183 | |
2ba1c290 | 184 | |
43c63a78 JA |
185 | /* |
186 | * if we are not maintaining a random map, we are done. | |
187 | */ | |
303032ae | 188 | if (!file_randommap(td, f)) |
0ce8b119 | 189 | goto ret_good; |
43c63a78 JA |
190 | |
191 | /* | |
84422acd | 192 | * calculate map offset and check if it's free |
43c63a78 | 193 | */ |
aec2de20 | 194 | if (random_map_free(f, *b)) |
0ce8b119 | 195 | goto ret_good; |
43c63a78 | 196 | |
84422acd JA |
197 | dprint(FD_RANDOM, "get_next_rand_offset: offset %llu busy\n", |
198 | *b); | |
43c63a78 | 199 | } while (--loops); |
ec4015da | 200 | |
0ce8b119 JA |
201 | if (!f->failed_rands++) |
202 | f->last_free_lookup = 0; | |
203 | ||
ec4015da | 204 | /* |
43c63a78 JA |
205 | * we get here, if we didn't suceed in looking up a block. generate |
206 | * a random start offset into the filemap, and find the first free | |
207 | * block from there. | |
ec4015da | 208 | */ |
43c63a78 JA |
209 | loops = 10; |
210 | do { | |
dc873b6f | 211 | f->last_free_lookup = (f->num_maps - 1) * |
15b87723 | 212 | (r / ((unsigned long long) rmax + 1.0)); |
4ba66134 | 213 | if (!get_next_free_block(td, f, ddir, b)) |
0ce8b119 | 214 | goto ret; |
ec4015da | 215 | |
d1cea521 JA |
216 | if (td->o.use_os_rand) |
217 | r = os_random_long(&td->random_state); | |
218 | else | |
219 | r = __rand(&td->__random_state); | |
43c63a78 JA |
220 | } while (--loops); |
221 | ||
222 | /* | |
223 | * that didn't work either, try exhaustive search from the start | |
224 | */ | |
225 | f->last_free_lookup = 0; | |
0ce8b119 JA |
226 | ffz: |
227 | if (!get_next_free_block(td, f, ddir, b)) | |
228 | return 0; | |
229 | f->last_free_lookup = 0; | |
4ba66134 | 230 | return get_next_free_block(td, f, ddir, b); |
0ce8b119 JA |
231 | ret_good: |
232 | f->failed_rands = 0; | |
233 | ret: | |
234 | return 0; | |
ec4015da JA |
235 | } |
236 | ||
925fee33 JA |
237 | static int __get_next_rand_offset_zipf(struct thread_data *td, |
238 | struct fio_file *f, enum fio_ddir ddir, | |
239 | unsigned long long *b) | |
e25839d4 | 240 | { |
9c6f6316 | 241 | *b = zipf_next(&f->zipf); |
e25839d4 JA |
242 | return 0; |
243 | } | |
244 | ||
925fee33 JA |
245 | static int __get_next_rand_offset_pareto(struct thread_data *td, |
246 | struct fio_file *f, enum fio_ddir ddir, | |
247 | unsigned long long *b) | |
248 | { | |
9c6f6316 | 249 | *b = pareto_next(&f->zipf); |
925fee33 JA |
250 | return 0; |
251 | } | |
252 | ||
e25839d4 JA |
253 | static int get_next_rand_offset(struct thread_data *td, struct fio_file *f, |
254 | enum fio_ddir ddir, unsigned long long *b) | |
255 | { | |
256 | if (td->o.random_distribution == FIO_RAND_DIST_RANDOM) | |
257 | return __get_next_rand_offset(td, f, ddir, b); | |
258 | else if (td->o.random_distribution == FIO_RAND_DIST_ZIPF) | |
259 | return __get_next_rand_offset_zipf(td, f, ddir, b); | |
925fee33 JA |
260 | else if (td->o.random_distribution == FIO_RAND_DIST_PARETO) |
261 | return __get_next_rand_offset_pareto(td, f, ddir, b); | |
e25839d4 JA |
262 | |
263 | log_err("fio: unknown random distribution: %d\n", td->o.random_distribution); | |
264 | return 1; | |
265 | } | |
266 | ||
38dad62d JA |
267 | static int get_next_rand_block(struct thread_data *td, struct fio_file *f, |
268 | enum fio_ddir ddir, unsigned long long *b) | |
269 | { | |
c04e4661 DE |
270 | if (!get_next_rand_offset(td, f, ddir, b)) |
271 | return 0; | |
272 | ||
273 | if (td->o.time_based) { | |
274 | fio_file_reset(f); | |
275 | if (!get_next_rand_offset(td, f, ddir, b)) | |
276 | return 0; | |
38dad62d JA |
277 | } |
278 | ||
c04e4661 DE |
279 | dprint(FD_IO, "%s: rand offset failed, last=%llu, size=%llu\n", |
280 | f->file_name, f->last_pos, f->real_file_size); | |
281 | return 1; | |
38dad62d JA |
282 | } |
283 | ||
37cf9e3c JA |
284 | static int get_next_seq_offset(struct thread_data *td, struct fio_file *f, |
285 | enum fio_ddir ddir, unsigned long long *offset) | |
38dad62d | 286 | { |
ff58fced JA |
287 | assert(ddir_rw(ddir)); |
288 | ||
ce95d651 | 289 | if (f->last_pos >= f->io_size + get_start_offset(td) && td->o.time_based) |
c04e4661 DE |
290 | f->last_pos = f->last_pos - f->io_size; |
291 | ||
38dad62d | 292 | if (f->last_pos < f->real_file_size) { |
a66da7a2 | 293 | unsigned long long pos; |
059b0802 | 294 | |
a66da7a2 JA |
295 | if (f->last_pos == f->file_offset && td->o.ddir_seq_add < 0) |
296 | f->last_pos = f->real_file_size; | |
297 | ||
298 | pos = f->last_pos - f->file_offset; | |
059b0802 JA |
299 | if (pos) |
300 | pos += td->o.ddir_seq_add; | |
301 | ||
37cf9e3c | 302 | *offset = pos; |
38dad62d JA |
303 | return 0; |
304 | } | |
305 | ||
306 | return 1; | |
307 | } | |
308 | ||
309 | static int get_next_block(struct thread_data *td, struct io_u *io_u, | |
37cf9e3c | 310 | enum fio_ddir ddir, int rw_seq) |
38dad62d JA |
311 | { |
312 | struct fio_file *f = io_u->file; | |
37cf9e3c | 313 | unsigned long long b, offset; |
38dad62d JA |
314 | int ret; |
315 | ||
ff58fced JA |
316 | assert(ddir_rw(ddir)); |
317 | ||
37cf9e3c JA |
318 | b = offset = -1ULL; |
319 | ||
38dad62d JA |
320 | if (rw_seq) { |
321 | if (td_random(td)) | |
37cf9e3c | 322 | ret = get_next_rand_block(td, f, ddir, &b); |
38dad62d | 323 | else |
37cf9e3c | 324 | ret = get_next_seq_offset(td, f, ddir, &offset); |
38dad62d JA |
325 | } else { |
326 | io_u->flags |= IO_U_F_BUSY_OK; | |
327 | ||
328 | if (td->o.rw_seq == RW_SEQ_SEQ) { | |
37cf9e3c | 329 | ret = get_next_seq_offset(td, f, ddir, &offset); |
38dad62d | 330 | if (ret) |
37cf9e3c | 331 | ret = get_next_rand_block(td, f, ddir, &b); |
38dad62d JA |
332 | } else if (td->o.rw_seq == RW_SEQ_IDENT) { |
333 | if (f->last_start != -1ULL) | |
37cf9e3c | 334 | offset = f->last_start - f->file_offset; |
38dad62d | 335 | else |
37cf9e3c | 336 | offset = 0; |
38dad62d JA |
337 | ret = 0; |
338 | } else { | |
339 | log_err("fio: unknown rw_seq=%d\n", td->o.rw_seq); | |
340 | ret = 1; | |
341 | } | |
342 | } | |
3c3ed070 | 343 | |
37cf9e3c JA |
344 | if (!ret) { |
345 | if (offset != -1ULL) | |
346 | io_u->offset = offset; | |
347 | else if (b != -1ULL) | |
348 | io_u->offset = b * td->o.ba[ddir]; | |
349 | else { | |
350 | log_err("fio: bug in offset generation\n"); | |
351 | ret = 1; | |
352 | } | |
353 | } | |
354 | ||
38dad62d JA |
355 | return ret; |
356 | } | |
357 | ||
10ba535a JA |
358 | /* |
359 | * For random io, generate a random new block and see if it's used. Repeat | |
360 | * until we find a free one. For sequential io, just return the end of | |
361 | * the last io issued. | |
362 | */ | |
15dc1934 | 363 | static int __get_next_offset(struct thread_data *td, struct io_u *io_u) |
10ba535a | 364 | { |
9bf2061e | 365 | struct fio_file *f = io_u->file; |
4ba66134 | 366 | enum fio_ddir ddir = io_u->ddir; |
38dad62d | 367 | int rw_seq_hit = 0; |
10ba535a | 368 | |
ff58fced JA |
369 | assert(ddir_rw(ddir)); |
370 | ||
38dad62d JA |
371 | if (td->o.ddir_seq_nr && !--td->ddir_seq_nr) { |
372 | rw_seq_hit = 1; | |
5736c10d | 373 | td->ddir_seq_nr = td->o.ddir_seq_nr; |
38dad62d | 374 | } |
211097b2 | 375 | |
37cf9e3c | 376 | if (get_next_block(td, io_u, ddir, rw_seq_hit)) |
38dad62d | 377 | return 1; |
10ba535a | 378 | |
009bd847 JA |
379 | if (io_u->offset >= f->io_size) { |
380 | dprint(FD_IO, "get_next_offset: offset %llu >= io_size %llu\n", | |
381 | io_u->offset, f->io_size); | |
382 | return 1; | |
383 | } | |
384 | ||
385 | io_u->offset += f->file_offset; | |
2ba1c290 JA |
386 | if (io_u->offset >= f->real_file_size) { |
387 | dprint(FD_IO, "get_next_offset: offset %llu >= size %llu\n", | |
388 | io_u->offset, f->real_file_size); | |
10ba535a | 389 | return 1; |
2ba1c290 | 390 | } |
10ba535a JA |
391 | |
392 | return 0; | |
393 | } | |
394 | ||
15dc1934 JA |
395 | static int get_next_offset(struct thread_data *td, struct io_u *io_u) |
396 | { | |
7eb36574 JA |
397 | struct prof_io_ops *ops = &td->prof_io_ops; |
398 | ||
399 | if (ops->fill_io_u_off) | |
400 | return ops->fill_io_u_off(td, io_u); | |
15dc1934 JA |
401 | |
402 | return __get_next_offset(td, io_u); | |
403 | } | |
404 | ||
79944128 JA |
405 | static inline int io_u_fits(struct thread_data *td, struct io_u *io_u, |
406 | unsigned int buflen) | |
407 | { | |
408 | struct fio_file *f = io_u->file; | |
409 | ||
ce95d651 | 410 | return io_u->offset + buflen <= f->io_size + get_start_offset(td); |
79944128 JA |
411 | } |
412 | ||
15dc1934 | 413 | static unsigned int __get_next_buflen(struct thread_data *td, struct io_u *io_u) |
10ba535a | 414 | { |
bca4ed4d | 415 | const int ddir = io_u->ddir; |
24d23ca7 | 416 | unsigned int buflen = 0; |
f3059de1 | 417 | unsigned int minbs, maxbs; |
1294c3ec | 418 | unsigned long r, rand_max; |
10ba535a | 419 | |
ff58fced JA |
420 | assert(ddir_rw(ddir)); |
421 | ||
f3059de1 JA |
422 | minbs = td->o.min_bs[ddir]; |
423 | maxbs = td->o.max_bs[ddir]; | |
424 | ||
79944128 JA |
425 | if (minbs == maxbs) |
426 | return minbs; | |
427 | ||
52c58027 JA |
428 | /* |
429 | * If we can't satisfy the min block size from here, then fail | |
430 | */ | |
431 | if (!io_u_fits(td, io_u, minbs)) | |
432 | return 0; | |
433 | ||
4c07ad86 JA |
434 | if (td->o.use_os_rand) |
435 | rand_max = OS_RAND_MAX; | |
436 | else | |
437 | rand_max = FRAND_MAX; | |
438 | ||
79944128 | 439 | do { |
4c07ad86 JA |
440 | if (td->o.use_os_rand) |
441 | r = os_random_long(&td->bsrange_state); | |
442 | else | |
443 | r = __rand(&td->__bsrange_state); | |
444 | ||
720e84ad | 445 | if (!td->o.bssplit_nr[ddir]) { |
f3059de1 | 446 | buflen = 1 + (unsigned int) ((double) maxbs * |
4c07ad86 | 447 | (r / (rand_max + 1.0))); |
f3059de1 JA |
448 | if (buflen < minbs) |
449 | buflen = minbs; | |
5ec10eaa | 450 | } else { |
564ca972 JA |
451 | long perc = 0; |
452 | unsigned int i; | |
453 | ||
720e84ad JA |
454 | for (i = 0; i < td->o.bssplit_nr[ddir]; i++) { |
455 | struct bssplit *bsp = &td->o.bssplit[ddir][i]; | |
564ca972 JA |
456 | |
457 | buflen = bsp->bs; | |
458 | perc += bsp->perc; | |
79944128 JA |
459 | if ((r <= ((rand_max / 100L) * perc)) && |
460 | io_u_fits(td, io_u, buflen)) | |
564ca972 JA |
461 | break; |
462 | } | |
463 | } | |
79944128 | 464 | |
f3059de1 JA |
465 | if (!td->o.bs_unaligned && is_power_of_2(minbs)) |
466 | buflen = (buflen + minbs - 1) & ~(minbs - 1); | |
10ba535a | 467 | |
79944128 | 468 | } while (!io_u_fits(td, io_u, buflen)); |
6a5e6884 | 469 | |
10ba535a JA |
470 | return buflen; |
471 | } | |
472 | ||
15dc1934 JA |
473 | static unsigned int get_next_buflen(struct thread_data *td, struct io_u *io_u) |
474 | { | |
7eb36574 JA |
475 | struct prof_io_ops *ops = &td->prof_io_ops; |
476 | ||
477 | if (ops->fill_io_u_size) | |
478 | return ops->fill_io_u_size(td, io_u); | |
15dc1934 JA |
479 | |
480 | return __get_next_buflen(td, io_u); | |
481 | } | |
482 | ||
afe24a5a JA |
483 | static void set_rwmix_bytes(struct thread_data *td) |
484 | { | |
afe24a5a JA |
485 | unsigned int diff; |
486 | ||
487 | /* | |
488 | * we do time or byte based switch. this is needed because | |
489 | * buffered writes may issue a lot quicker than they complete, | |
490 | * whereas reads do not. | |
491 | */ | |
e47f799f | 492 | diff = td->o.rwmix[td->rwmix_ddir ^ 1]; |
04c540d9 | 493 | td->rwmix_issues = (td->io_issues[td->rwmix_ddir] * diff) / 100; |
e47f799f JA |
494 | } |
495 | ||
496 | static inline enum fio_ddir get_rand_ddir(struct thread_data *td) | |
497 | { | |
498 | unsigned int v; | |
1294c3ec | 499 | unsigned long r; |
e47f799f | 500 | |
4c07ad86 JA |
501 | if (td->o.use_os_rand) { |
502 | r = os_random_long(&td->rwmix_state); | |
503 | v = 1 + (int) (100.0 * (r / (OS_RAND_MAX + 1.0))); | |
504 | } else { | |
505 | r = __rand(&td->__rwmix_state); | |
506 | v = 1 + (int) (100.0 * (r / (FRAND_MAX + 1.0))); | |
507 | } | |
508 | ||
04c540d9 | 509 | if (v <= td->o.rwmix[DDIR_READ]) |
e47f799f JA |
510 | return DDIR_READ; |
511 | ||
512 | return DDIR_WRITE; | |
afe24a5a JA |
513 | } |
514 | ||
581e7141 JA |
515 | static enum fio_ddir rate_ddir(struct thread_data *td, enum fio_ddir ddir) |
516 | { | |
517 | enum fio_ddir odir = ddir ^ 1; | |
518 | struct timeval t; | |
519 | long usec; | |
520 | ||
ff58fced JA |
521 | assert(ddir_rw(ddir)); |
522 | ||
581e7141 JA |
523 | if (td->rate_pending_usleep[ddir] <= 0) |
524 | return ddir; | |
525 | ||
526 | /* | |
527 | * We have too much pending sleep in this direction. See if we | |
528 | * should switch. | |
529 | */ | |
530 | if (td_rw(td)) { | |
531 | /* | |
532 | * Other direction does not have too much pending, switch | |
533 | */ | |
534 | if (td->rate_pending_usleep[odir] < 100000) | |
535 | return odir; | |
536 | ||
537 | /* | |
538 | * Both directions have pending sleep. Sleep the minimum time | |
539 | * and deduct from both. | |
540 | */ | |
541 | if (td->rate_pending_usleep[ddir] <= | |
542 | td->rate_pending_usleep[odir]) { | |
543 | usec = td->rate_pending_usleep[ddir]; | |
544 | } else { | |
545 | usec = td->rate_pending_usleep[odir]; | |
546 | ddir = odir; | |
547 | } | |
548 | } else | |
549 | usec = td->rate_pending_usleep[ddir]; | |
550 | ||
78c1eda5 JA |
551 | /* |
552 | * We are going to sleep, ensure that we flush anything pending as | |
422f9e4b RM |
553 | * not to skew our latency numbers. |
554 | * | |
555 | * Changed to only monitor 'in flight' requests here instead of the | |
556 | * td->cur_depth, b/c td->cur_depth does not accurately represent | |
557 | * io's that have been actually submitted to an async engine, | |
558 | * and cur_depth is meaningless for sync engines. | |
78c1eda5 | 559 | */ |
422f9e4b | 560 | if (td->io_u_in_flight) { |
78c1eda5 JA |
561 | int fio_unused ret; |
562 | ||
422f9e4b | 563 | ret = io_u_queued_complete(td, td->io_u_in_flight, NULL); |
78c1eda5 JA |
564 | } |
565 | ||
581e7141 JA |
566 | fio_gettime(&t, NULL); |
567 | usec_sleep(td, usec); | |
568 | usec = utime_since_now(&t); | |
569 | ||
570 | td->rate_pending_usleep[ddir] -= usec; | |
571 | ||
572 | odir = ddir ^ 1; | |
573 | if (td_rw(td) && __should_check_rate(td, odir)) | |
574 | td->rate_pending_usleep[odir] -= usec; | |
0b9d69ec | 575 | |
6eaf09d6 SL |
576 | if (ddir_trim(ddir)) |
577 | return ddir; | |
581e7141 JA |
578 | return ddir; |
579 | } | |
580 | ||
10ba535a JA |
581 | /* |
582 | * Return the data direction for the next io_u. If the job is a | |
583 | * mixed read/write workload, check the rwmix cycle and switch if | |
584 | * necessary. | |
585 | */ | |
1e97cce9 | 586 | static enum fio_ddir get_rw_ddir(struct thread_data *td) |
10ba535a | 587 | { |
581e7141 JA |
588 | enum fio_ddir ddir; |
589 | ||
5f9099ea JA |
590 | /* |
591 | * see if it's time to fsync | |
592 | */ | |
593 | if (td->o.fsync_blocks && | |
594 | !(td->io_issues[DDIR_WRITE] % td->o.fsync_blocks) && | |
595 | td->io_issues[DDIR_WRITE] && should_fsync(td)) | |
596 | return DDIR_SYNC; | |
597 | ||
598 | /* | |
599 | * see if it's time to fdatasync | |
600 | */ | |
601 | if (td->o.fdatasync_blocks && | |
602 | !(td->io_issues[DDIR_WRITE] % td->o.fdatasync_blocks) && | |
603 | td->io_issues[DDIR_WRITE] && should_fsync(td)) | |
604 | return DDIR_DATASYNC; | |
605 | ||
44f29692 JA |
606 | /* |
607 | * see if it's time to sync_file_range | |
608 | */ | |
609 | if (td->sync_file_range_nr && | |
610 | !(td->io_issues[DDIR_WRITE] % td->sync_file_range_nr) && | |
611 | td->io_issues[DDIR_WRITE] && should_fsync(td)) | |
612 | return DDIR_SYNC_FILE_RANGE; | |
613 | ||
10ba535a | 614 | if (td_rw(td)) { |
10ba535a JA |
615 | /* |
616 | * Check if it's time to seed a new data direction. | |
617 | */ | |
e4928662 | 618 | if (td->io_issues[td->rwmix_ddir] >= td->rwmix_issues) { |
e47f799f JA |
619 | /* |
620 | * Put a top limit on how many bytes we do for | |
621 | * one data direction, to avoid overflowing the | |
622 | * ranges too much | |
623 | */ | |
624 | ddir = get_rand_ddir(td); | |
e47f799f JA |
625 | |
626 | if (ddir != td->rwmix_ddir) | |
627 | set_rwmix_bytes(td); | |
628 | ||
629 | td->rwmix_ddir = ddir; | |
10ba535a | 630 | } |
581e7141 | 631 | ddir = td->rwmix_ddir; |
10ba535a | 632 | } else if (td_read(td)) |
581e7141 | 633 | ddir = DDIR_READ; |
6eaf09d6 | 634 | else if (td_write(td)) |
581e7141 | 635 | ddir = DDIR_WRITE; |
6eaf09d6 SL |
636 | else |
637 | ddir = DDIR_TRIM; | |
581e7141 JA |
638 | |
639 | td->rwmix_ddir = rate_ddir(td, ddir); | |
640 | return td->rwmix_ddir; | |
10ba535a JA |
641 | } |
642 | ||
1ef2b6be JA |
643 | static void set_rw_ddir(struct thread_data *td, struct io_u *io_u) |
644 | { | |
645 | io_u->ddir = get_rw_ddir(td); | |
646 | ||
647 | if (io_u->ddir == DDIR_WRITE && (td->io_ops->flags & FIO_BARRIER) && | |
648 | td->o.barrier_blocks && | |
649 | !(td->io_issues[DDIR_WRITE] % td->o.barrier_blocks) && | |
650 | td->io_issues[DDIR_WRITE]) | |
651 | io_u->flags |= IO_U_F_BARRIER; | |
652 | } | |
653 | ||
e8462bd8 | 654 | void put_file_log(struct thread_data *td, struct fio_file *f) |
60f2c658 JA |
655 | { |
656 | int ret = put_file(td, f); | |
657 | ||
658 | if (ret) | |
659 | td_verror(td, ret, "file close"); | |
660 | } | |
661 | ||
10ba535a JA |
662 | void put_io_u(struct thread_data *td, struct io_u *io_u) |
663 | { | |
e8462bd8 JA |
664 | td_io_u_lock(td); |
665 | ||
d7ee2a7d | 666 | if (io_u->file && !(io_u->flags & IO_U_F_FREE_DEF)) |
60f2c658 | 667 | put_file_log(td, io_u->file); |
10ba535a | 668 | io_u->file = NULL; |
d7ee2a7d SL |
669 | io_u->flags &= ~IO_U_F_FREE_DEF; |
670 | io_u->flags |= IO_U_F_FREE; | |
671 | ||
0c41214f RR |
672 | if (io_u->flags & IO_U_F_IN_CUR_DEPTH) |
673 | td->cur_depth--; | |
e8462bd8 | 674 | flist_del_init(&io_u->list); |
01743ee1 | 675 | flist_add(&io_u->list, &td->io_u_freelist); |
e8462bd8 JA |
676 | td_io_u_unlock(td); |
677 | td_io_u_free_notify(td); | |
10ba535a JA |
678 | } |
679 | ||
f2bba182 RR |
680 | void clear_io_u(struct thread_data *td, struct io_u *io_u) |
681 | { | |
682 | io_u->flags &= ~IO_U_F_FLIGHT; | |
683 | put_io_u(td, io_u); | |
684 | } | |
685 | ||
755200a3 JA |
686 | void requeue_io_u(struct thread_data *td, struct io_u **io_u) |
687 | { | |
688 | struct io_u *__io_u = *io_u; | |
689 | ||
465221b0 JA |
690 | dprint(FD_IO, "requeue %p\n", __io_u); |
691 | ||
e8462bd8 JA |
692 | td_io_u_lock(td); |
693 | ||
4d2e0f49 | 694 | __io_u->flags |= IO_U_F_FREE; |
ff58fced | 695 | if ((__io_u->flags & IO_U_F_FLIGHT) && ddir_rw(__io_u->ddir)) |
e4f54adb | 696 | td->io_issues[__io_u->ddir]--; |
5ec10eaa | 697 | |
4d2e0f49 | 698 | __io_u->flags &= ~IO_U_F_FLIGHT; |
0c41214f RR |
699 | if (__io_u->flags & IO_U_F_IN_CUR_DEPTH) |
700 | td->cur_depth--; | |
01743ee1 JA |
701 | flist_del(&__io_u->list); |
702 | flist_add_tail(&__io_u->list, &td->io_u_requeues); | |
e8462bd8 | 703 | td_io_u_unlock(td); |
755200a3 JA |
704 | *io_u = NULL; |
705 | } | |
706 | ||
9bf2061e | 707 | static int fill_io_u(struct thread_data *td, struct io_u *io_u) |
10ba535a | 708 | { |
b4c5e1ac JA |
709 | if (td->io_ops->flags & FIO_NOIO) |
710 | goto out; | |
711 | ||
1ef2b6be | 712 | set_rw_ddir(td, io_u); |
5f9099ea | 713 | |
87dc1ab1 | 714 | /* |
ff58fced | 715 | * fsync() or fdatasync() or trim etc, we are done |
87dc1ab1 | 716 | */ |
ff58fced | 717 | if (!ddir_rw(io_u->ddir)) |
c38e9468 | 718 | goto out; |
a00735e6 | 719 | |
48f5abd3 JA |
720 | /* |
721 | * See if it's time to switch to a new zone | |
722 | */ | |
13af05ae | 723 | if (td->zone_bytes >= td->o.zone_size && td->o.zone_skip) { |
48f5abd3 | 724 | td->zone_bytes = 0; |
ed335855 SN |
725 | io_u->file->file_offset += td->o.zone_range + td->o.zone_skip; |
726 | io_u->file->last_pos = io_u->file->file_offset; | |
48f5abd3 JA |
727 | td->io_skip_bytes += td->o.zone_skip; |
728 | } | |
729 | ||
10ba535a | 730 | /* |
c685b5b2 JA |
731 | * No log, let the seq/rand engine retrieve the next buflen and |
732 | * position. | |
10ba535a | 733 | */ |
2ba1c290 JA |
734 | if (get_next_offset(td, io_u)) { |
735 | dprint(FD_IO, "io_u %p, failed getting offset\n", io_u); | |
bca4ed4d | 736 | return 1; |
2ba1c290 | 737 | } |
10ba535a | 738 | |
9bf2061e | 739 | io_u->buflen = get_next_buflen(td, io_u); |
2ba1c290 JA |
740 | if (!io_u->buflen) { |
741 | dprint(FD_IO, "io_u %p, failed getting buflen\n", io_u); | |
bca4ed4d | 742 | return 1; |
2ba1c290 | 743 | } |
bca4ed4d | 744 | |
2ba1c290 JA |
745 | if (io_u->offset + io_u->buflen > io_u->file->real_file_size) { |
746 | dprint(FD_IO, "io_u %p, offset too large\n", io_u); | |
4ba66134 JA |
747 | dprint(FD_IO, " off=%llu/%lu > %llu\n", io_u->offset, |
748 | io_u->buflen, io_u->file->real_file_size); | |
6a5e6884 | 749 | return 1; |
2ba1c290 | 750 | } |
6a5e6884 | 751 | |
bca4ed4d JA |
752 | /* |
753 | * mark entry before potentially trimming io_u | |
754 | */ | |
303032ae | 755 | if (td_random(td) && file_randommap(td, io_u->file)) |
9bf2061e | 756 | mark_random_map(td, io_u); |
bca4ed4d JA |
757 | |
758 | /* | |
759 | * If using a write iolog, store this entry. | |
760 | */ | |
c38e9468 | 761 | out: |
2ba1c290 | 762 | dprint_io_u(io_u, "fill_io_u"); |
d9d91e39 | 763 | td->zone_bytes += io_u->buflen; |
f29b25a3 | 764 | log_io_u(td, io_u); |
bca4ed4d | 765 | return 0; |
10ba535a JA |
766 | } |
767 | ||
838bc709 JA |
768 | static void __io_u_mark_map(unsigned int *map, unsigned int nr) |
769 | { | |
2b13e716 | 770 | int idx = 0; |
838bc709 JA |
771 | |
772 | switch (nr) { | |
773 | default: | |
2b13e716 | 774 | idx = 6; |
838bc709 JA |
775 | break; |
776 | case 33 ... 64: | |
2b13e716 | 777 | idx = 5; |
838bc709 JA |
778 | break; |
779 | case 17 ... 32: | |
2b13e716 | 780 | idx = 4; |
838bc709 JA |
781 | break; |
782 | case 9 ... 16: | |
2b13e716 | 783 | idx = 3; |
838bc709 JA |
784 | break; |
785 | case 5 ... 8: | |
2b13e716 | 786 | idx = 2; |
838bc709 JA |
787 | break; |
788 | case 1 ... 4: | |
2b13e716 | 789 | idx = 1; |
838bc709 JA |
790 | case 0: |
791 | break; | |
792 | } | |
793 | ||
2b13e716 | 794 | map[idx]++; |
838bc709 JA |
795 | } |
796 | ||
797 | void io_u_mark_submit(struct thread_data *td, unsigned int nr) | |
798 | { | |
799 | __io_u_mark_map(td->ts.io_u_submit, nr); | |
800 | td->ts.total_submit++; | |
801 | } | |
802 | ||
803 | void io_u_mark_complete(struct thread_data *td, unsigned int nr) | |
804 | { | |
805 | __io_u_mark_map(td->ts.io_u_complete, nr); | |
806 | td->ts.total_complete++; | |
807 | } | |
808 | ||
d8005759 | 809 | void io_u_mark_depth(struct thread_data *td, unsigned int nr) |
71619dc2 | 810 | { |
2b13e716 | 811 | int idx = 0; |
71619dc2 JA |
812 | |
813 | switch (td->cur_depth) { | |
814 | default: | |
2b13e716 | 815 | idx = 6; |
a783e61a | 816 | break; |
71619dc2 | 817 | case 32 ... 63: |
2b13e716 | 818 | idx = 5; |
a783e61a | 819 | break; |
71619dc2 | 820 | case 16 ... 31: |
2b13e716 | 821 | idx = 4; |
a783e61a | 822 | break; |
71619dc2 | 823 | case 8 ... 15: |
2b13e716 | 824 | idx = 3; |
a783e61a | 825 | break; |
71619dc2 | 826 | case 4 ... 7: |
2b13e716 | 827 | idx = 2; |
a783e61a | 828 | break; |
71619dc2 | 829 | case 2 ... 3: |
2b13e716 | 830 | idx = 1; |
71619dc2 JA |
831 | case 1: |
832 | break; | |
833 | } | |
834 | ||
2b13e716 | 835 | td->ts.io_u_map[idx] += nr; |
71619dc2 JA |
836 | } |
837 | ||
04a0feae JA |
838 | static void io_u_mark_lat_usec(struct thread_data *td, unsigned long usec) |
839 | { | |
2b13e716 | 840 | int idx = 0; |
04a0feae JA |
841 | |
842 | assert(usec < 1000); | |
843 | ||
844 | switch (usec) { | |
845 | case 750 ... 999: | |
2b13e716 | 846 | idx = 9; |
04a0feae JA |
847 | break; |
848 | case 500 ... 749: | |
2b13e716 | 849 | idx = 8; |
04a0feae JA |
850 | break; |
851 | case 250 ... 499: | |
2b13e716 | 852 | idx = 7; |
04a0feae JA |
853 | break; |
854 | case 100 ... 249: | |
2b13e716 | 855 | idx = 6; |
04a0feae JA |
856 | break; |
857 | case 50 ... 99: | |
2b13e716 | 858 | idx = 5; |
04a0feae JA |
859 | break; |
860 | case 20 ... 49: | |
2b13e716 | 861 | idx = 4; |
04a0feae JA |
862 | break; |
863 | case 10 ... 19: | |
2b13e716 | 864 | idx = 3; |
04a0feae JA |
865 | break; |
866 | case 4 ... 9: | |
2b13e716 | 867 | idx = 2; |
04a0feae JA |
868 | break; |
869 | case 2 ... 3: | |
2b13e716 | 870 | idx = 1; |
04a0feae JA |
871 | case 0 ... 1: |
872 | break; | |
873 | } | |
874 | ||
2b13e716 JA |
875 | assert(idx < FIO_IO_U_LAT_U_NR); |
876 | td->ts.io_u_lat_u[idx]++; | |
04a0feae JA |
877 | } |
878 | ||
879 | static void io_u_mark_lat_msec(struct thread_data *td, unsigned long msec) | |
ec118304 | 880 | { |
2b13e716 | 881 | int idx = 0; |
ec118304 JA |
882 | |
883 | switch (msec) { | |
884 | default: | |
2b13e716 | 885 | idx = 11; |
04a0feae | 886 | break; |
8abdce66 | 887 | case 1000 ... 1999: |
2b13e716 | 888 | idx = 10; |
04a0feae | 889 | break; |
8abdce66 | 890 | case 750 ... 999: |
2b13e716 | 891 | idx = 9; |
04a0feae | 892 | break; |
8abdce66 | 893 | case 500 ... 749: |
2b13e716 | 894 | idx = 8; |
04a0feae | 895 | break; |
8abdce66 | 896 | case 250 ... 499: |
2b13e716 | 897 | idx = 7; |
04a0feae | 898 | break; |
8abdce66 | 899 | case 100 ... 249: |
2b13e716 | 900 | idx = 6; |
04a0feae | 901 | break; |
8abdce66 | 902 | case 50 ... 99: |
2b13e716 | 903 | idx = 5; |
04a0feae | 904 | break; |
8abdce66 | 905 | case 20 ... 49: |
2b13e716 | 906 | idx = 4; |
04a0feae | 907 | break; |
8abdce66 | 908 | case 10 ... 19: |
2b13e716 | 909 | idx = 3; |
04a0feae | 910 | break; |
8abdce66 | 911 | case 4 ... 9: |
2b13e716 | 912 | idx = 2; |
04a0feae | 913 | break; |
ec118304 | 914 | case 2 ... 3: |
2b13e716 | 915 | idx = 1; |
ec118304 JA |
916 | case 0 ... 1: |
917 | break; | |
918 | } | |
919 | ||
2b13e716 JA |
920 | assert(idx < FIO_IO_U_LAT_M_NR); |
921 | td->ts.io_u_lat_m[idx]++; | |
04a0feae JA |
922 | } |
923 | ||
924 | static void io_u_mark_latency(struct thread_data *td, unsigned long usec) | |
925 | { | |
926 | if (usec < 1000) | |
927 | io_u_mark_lat_usec(td, usec); | |
928 | else | |
929 | io_u_mark_lat_msec(td, usec / 1000); | |
ec118304 JA |
930 | } |
931 | ||
0aabe160 JA |
932 | /* |
933 | * Get next file to service by choosing one at random | |
934 | */ | |
2cc52930 JA |
935 | static struct fio_file *get_next_file_rand(struct thread_data *td, |
936 | enum fio_file_flags goodf, | |
d6aed795 | 937 | enum fio_file_flags badf) |
0aabe160 | 938 | { |
0aabe160 | 939 | struct fio_file *f; |
1c178180 | 940 | int fno; |
0aabe160 JA |
941 | |
942 | do { | |
87b10676 | 943 | int opened = 0; |
1294c3ec | 944 | unsigned long r; |
4c07ad86 JA |
945 | |
946 | if (td->o.use_os_rand) { | |
947 | r = os_random_long(&td->next_file_state); | |
948 | fno = (unsigned int) ((double) td->o.nr_files | |
949 | * (r / (OS_RAND_MAX + 1.0))); | |
950 | } else { | |
951 | r = __rand(&td->__next_file_state); | |
952 | fno = (unsigned int) ((double) td->o.nr_files | |
953 | * (r / (FRAND_MAX + 1.0))); | |
954 | } | |
7c83c089 | 955 | |
126d65c6 | 956 | f = td->files[fno]; |
d6aed795 | 957 | if (fio_file_done(f)) |
059e63c0 | 958 | continue; |
1c178180 | 959 | |
d6aed795 | 960 | if (!fio_file_open(f)) { |
87b10676 JA |
961 | int err; |
962 | ||
963 | err = td_io_open_file(td, f); | |
964 | if (err) | |
965 | continue; | |
966 | opened = 1; | |
967 | } | |
968 | ||
2ba1c290 JA |
969 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) { |
970 | dprint(FD_FILE, "get_next_file_rand: %p\n", f); | |
0aabe160 | 971 | return f; |
2ba1c290 | 972 | } |
87b10676 JA |
973 | if (opened) |
974 | td_io_close_file(td, f); | |
0aabe160 JA |
975 | } while (1); |
976 | } | |
977 | ||
978 | /* | |
979 | * Get next file to service by doing round robin between all available ones | |
980 | */ | |
1c178180 JA |
981 | static struct fio_file *get_next_file_rr(struct thread_data *td, int goodf, |
982 | int badf) | |
3d7c391d JA |
983 | { |
984 | unsigned int old_next_file = td->next_file; | |
985 | struct fio_file *f; | |
986 | ||
987 | do { | |
87b10676 JA |
988 | int opened = 0; |
989 | ||
126d65c6 | 990 | f = td->files[td->next_file]; |
3d7c391d JA |
991 | |
992 | td->next_file++; | |
2dc1bbeb | 993 | if (td->next_file >= td->o.nr_files) |
3d7c391d JA |
994 | td->next_file = 0; |
995 | ||
87b10676 | 996 | dprint(FD_FILE, "trying file %s %x\n", f->file_name, f->flags); |
d6aed795 | 997 | if (fio_file_done(f)) { |
d5ed68ea | 998 | f = NULL; |
059e63c0 | 999 | continue; |
d5ed68ea | 1000 | } |
059e63c0 | 1001 | |
d6aed795 | 1002 | if (!fio_file_open(f)) { |
87b10676 JA |
1003 | int err; |
1004 | ||
1005 | err = td_io_open_file(td, f); | |
b5696bfc JA |
1006 | if (err) { |
1007 | dprint(FD_FILE, "error %d on open of %s\n", | |
1008 | err, f->file_name); | |
87c27b45 | 1009 | f = NULL; |
87b10676 | 1010 | continue; |
b5696bfc | 1011 | } |
87b10676 JA |
1012 | opened = 1; |
1013 | } | |
1014 | ||
0b9d69ec JA |
1015 | dprint(FD_FILE, "goodf=%x, badf=%x, ff=%x\n", goodf, badf, |
1016 | f->flags); | |
1c178180 | 1017 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) |
3d7c391d JA |
1018 | break; |
1019 | ||
87b10676 JA |
1020 | if (opened) |
1021 | td_io_close_file(td, f); | |
1022 | ||
3d7c391d JA |
1023 | f = NULL; |
1024 | } while (td->next_file != old_next_file); | |
1025 | ||
2ba1c290 | 1026 | dprint(FD_FILE, "get_next_file_rr: %p\n", f); |
3d7c391d JA |
1027 | return f; |
1028 | } | |
1029 | ||
7eb36574 | 1030 | static struct fio_file *__get_next_file(struct thread_data *td) |
bdb4e2e9 | 1031 | { |
1907dbc6 JA |
1032 | struct fio_file *f; |
1033 | ||
2dc1bbeb | 1034 | assert(td->o.nr_files <= td->files_index); |
1c178180 | 1035 | |
b5696bfc | 1036 | if (td->nr_done_files >= td->o.nr_files) { |
5ec10eaa JA |
1037 | dprint(FD_FILE, "get_next_file: nr_open=%d, nr_done=%d," |
1038 | " nr_files=%d\n", td->nr_open_files, | |
1039 | td->nr_done_files, | |
1040 | td->o.nr_files); | |
bdb4e2e9 | 1041 | return NULL; |
2ba1c290 | 1042 | } |
bdb4e2e9 | 1043 | |
1907dbc6 | 1044 | f = td->file_service_file; |
d6aed795 | 1045 | if (f && fio_file_open(f) && !fio_file_closing(f)) { |
a086c257 JA |
1046 | if (td->o.file_service_type == FIO_FSERVICE_SEQ) |
1047 | goto out; | |
1048 | if (td->file_service_left--) | |
1049 | goto out; | |
1050 | } | |
1907dbc6 | 1051 | |
a086c257 JA |
1052 | if (td->o.file_service_type == FIO_FSERVICE_RR || |
1053 | td->o.file_service_type == FIO_FSERVICE_SEQ) | |
d6aed795 | 1054 | f = get_next_file_rr(td, FIO_FILE_open, FIO_FILE_closing); |
bdb4e2e9 | 1055 | else |
d6aed795 | 1056 | f = get_next_file_rand(td, FIO_FILE_open, FIO_FILE_closing); |
1907dbc6 JA |
1057 | |
1058 | td->file_service_file = f; | |
1059 | td->file_service_left = td->file_service_nr - 1; | |
2ba1c290 | 1060 | out: |
683023e8 | 1061 | dprint(FD_FILE, "get_next_file: %p [%s]\n", f, f->file_name); |
1907dbc6 | 1062 | return f; |
bdb4e2e9 JA |
1063 | } |
1064 | ||
7eb36574 JA |
1065 | static struct fio_file *get_next_file(struct thread_data *td) |
1066 | { | |
1067 | struct prof_io_ops *ops = &td->prof_io_ops; | |
1068 | ||
1069 | if (ops->get_next_file) | |
1070 | return ops->get_next_file(td); | |
1071 | ||
1072 | return __get_next_file(td); | |
1073 | } | |
1074 | ||
429f6675 JA |
1075 | static int set_io_u_file(struct thread_data *td, struct io_u *io_u) |
1076 | { | |
1077 | struct fio_file *f; | |
1078 | ||
1079 | do { | |
1080 | f = get_next_file(td); | |
1081 | if (!f) | |
1082 | return 1; | |
1083 | ||
429f6675 JA |
1084 | io_u->file = f; |
1085 | get_file(f); | |
1086 | ||
1087 | if (!fill_io_u(td, io_u)) | |
1088 | break; | |
1089 | ||
b5696bfc | 1090 | put_file_log(td, f); |
429f6675 | 1091 | td_io_close_file(td, f); |
b5696bfc | 1092 | io_u->file = NULL; |
d6aed795 | 1093 | fio_file_set_done(f); |
429f6675 | 1094 | td->nr_done_files++; |
0b9d69ec JA |
1095 | dprint(FD_FILE, "%s: is done (%d of %d)\n", f->file_name, |
1096 | td->nr_done_files, td->o.nr_files); | |
429f6675 JA |
1097 | } while (1); |
1098 | ||
1099 | return 0; | |
1100 | } | |
1101 | ||
1102 | ||
10ba535a JA |
1103 | struct io_u *__get_io_u(struct thread_data *td) |
1104 | { | |
1105 | struct io_u *io_u = NULL; | |
1106 | ||
e8462bd8 JA |
1107 | td_io_u_lock(td); |
1108 | ||
1109 | again: | |
01743ee1 JA |
1110 | if (!flist_empty(&td->io_u_requeues)) |
1111 | io_u = flist_entry(td->io_u_requeues.next, struct io_u, list); | |
755200a3 | 1112 | else if (!queue_full(td)) { |
01743ee1 | 1113 | io_u = flist_entry(td->io_u_freelist.next, struct io_u, list); |
10ba535a | 1114 | |
6040dabc | 1115 | io_u->buflen = 0; |
10ba535a | 1116 | io_u->resid = 0; |
755200a3 | 1117 | io_u->file = NULL; |
d7762cf8 | 1118 | io_u->end_io = NULL; |
755200a3 JA |
1119 | } |
1120 | ||
1121 | if (io_u) { | |
0c6e7517 | 1122 | assert(io_u->flags & IO_U_F_FREE); |
2ecc1b57 | 1123 | io_u->flags &= ~(IO_U_F_FREE | IO_U_F_FREE_DEF); |
1ef2b6be | 1124 | io_u->flags &= ~(IO_U_F_TRIMMED | IO_U_F_BARRIER); |
82af2a7c | 1125 | io_u->flags &= ~IO_U_F_VER_LIST; |
0c6e7517 | 1126 | |
755200a3 | 1127 | io_u->error = 0; |
01743ee1 | 1128 | flist_del(&io_u->list); |
94c6372c | 1129 | flist_add_tail(&io_u->list, &td->io_u_busylist); |
10ba535a | 1130 | td->cur_depth++; |
0c41214f | 1131 | io_u->flags |= IO_U_F_IN_CUR_DEPTH; |
1dec3e07 JA |
1132 | } else if (td->o.verify_async) { |
1133 | /* | |
1134 | * We ran out, wait for async verify threads to finish and | |
1135 | * return one | |
1136 | */ | |
1137 | pthread_cond_wait(&td->free_cond, &td->io_u_lock); | |
1138 | goto again; | |
10ba535a JA |
1139 | } |
1140 | ||
e8462bd8 | 1141 | td_io_u_unlock(td); |
10ba535a JA |
1142 | return io_u; |
1143 | } | |
1144 | ||
0d29de83 | 1145 | static int check_get_trim(struct thread_data *td, struct io_u *io_u) |
10ba535a | 1146 | { |
0d29de83 JA |
1147 | if (td->o.trim_backlog && td->trim_entries) { |
1148 | int get_trim = 0; | |
10ba535a | 1149 | |
0d29de83 JA |
1150 | if (td->trim_batch) { |
1151 | td->trim_batch--; | |
1152 | get_trim = 1; | |
1153 | } else if (!(td->io_hist_len % td->o.trim_backlog) && | |
1154 | td->last_ddir != DDIR_READ) { | |
1155 | td->trim_batch = td->o.trim_batch; | |
1156 | if (!td->trim_batch) | |
1157 | td->trim_batch = td->o.trim_backlog; | |
1158 | get_trim = 1; | |
1159 | } | |
1160 | ||
1161 | if (get_trim && !get_next_trim(td, io_u)) | |
1162 | return 1; | |
2ba1c290 | 1163 | } |
10ba535a | 1164 | |
0d29de83 JA |
1165 | return 0; |
1166 | } | |
1167 | ||
1168 | static int check_get_verify(struct thread_data *td, struct io_u *io_u) | |
1169 | { | |
9e144189 JA |
1170 | if (td->o.verify_backlog && td->io_hist_len) { |
1171 | int get_verify = 0; | |
1172 | ||
d1ece0c7 | 1173 | if (td->verify_batch) |
9e144189 | 1174 | get_verify = 1; |
d1ece0c7 | 1175 | else if (!(td->io_hist_len % td->o.verify_backlog) && |
9e144189 JA |
1176 | td->last_ddir != DDIR_READ) { |
1177 | td->verify_batch = td->o.verify_batch; | |
f8a75c99 JA |
1178 | if (!td->verify_batch) |
1179 | td->verify_batch = td->o.verify_backlog; | |
9e144189 JA |
1180 | get_verify = 1; |
1181 | } | |
1182 | ||
d1ece0c7 JA |
1183 | if (get_verify && !get_next_verify(td, io_u)) { |
1184 | td->verify_batch--; | |
0d29de83 | 1185 | return 1; |
d1ece0c7 | 1186 | } |
9e144189 JA |
1187 | } |
1188 | ||
0d29de83 JA |
1189 | return 0; |
1190 | } | |
1191 | ||
de789769 JA |
1192 | /* |
1193 | * Fill offset and start time into the buffer content, to prevent too | |
23f394d5 JA |
1194 | * easy compressible data for simple de-dupe attempts. Do this for every |
1195 | * 512b block in the range, since that should be the smallest block size | |
1196 | * we can expect from a device. | |
de789769 JA |
1197 | */ |
1198 | static void small_content_scramble(struct io_u *io_u) | |
1199 | { | |
23f394d5 | 1200 | unsigned int i, nr_blocks = io_u->buflen / 512; |
fad82f76 | 1201 | unsigned long long boffset; |
23f394d5 JA |
1202 | unsigned int offset; |
1203 | void *p, *end; | |
de789769 | 1204 | |
23f394d5 JA |
1205 | if (!nr_blocks) |
1206 | return; | |
1207 | ||
1208 | p = io_u->xfer_buf; | |
fba76ee8 | 1209 | boffset = io_u->offset; |
81f0366c | 1210 | io_u->buf_filled_len = 0; |
fad82f76 | 1211 | |
23f394d5 JA |
1212 | for (i = 0; i < nr_blocks; i++) { |
1213 | /* | |
1214 | * Fill the byte offset into a "random" start offset of | |
1215 | * the buffer, given by the product of the usec time | |
1216 | * and the actual offset. | |
1217 | */ | |
fad82f76 | 1218 | offset = (io_u->start_time.tv_usec ^ boffset) & 511; |
fba76ee8 | 1219 | offset &= ~(sizeof(unsigned long long) - 1); |
23f394d5 JA |
1220 | if (offset >= 512 - sizeof(unsigned long long)) |
1221 | offset -= sizeof(unsigned long long); | |
fba76ee8 | 1222 | memcpy(p + offset, &boffset, sizeof(boffset)); |
23f394d5 JA |
1223 | |
1224 | end = p + 512 - sizeof(io_u->start_time); | |
1225 | memcpy(end, &io_u->start_time, sizeof(io_u->start_time)); | |
1226 | p += 512; | |
fad82f76 | 1227 | boffset += 512; |
23f394d5 | 1228 | } |
de789769 JA |
1229 | } |
1230 | ||
0d29de83 JA |
1231 | /* |
1232 | * Return an io_u to be processed. Gets a buflen and offset, sets direction, | |
1233 | * etc. The returned io_u is fully ready to be prepped and submitted. | |
1234 | */ | |
1235 | struct io_u *get_io_u(struct thread_data *td) | |
1236 | { | |
1237 | struct fio_file *f; | |
1238 | struct io_u *io_u; | |
de789769 | 1239 | int do_scramble = 0; |
0d29de83 JA |
1240 | |
1241 | io_u = __get_io_u(td); | |
1242 | if (!io_u) { | |
1243 | dprint(FD_IO, "__get_io_u failed\n"); | |
1244 | return NULL; | |
1245 | } | |
1246 | ||
1247 | if (check_get_verify(td, io_u)) | |
1248 | goto out; | |
1249 | if (check_get_trim(td, io_u)) | |
1250 | goto out; | |
1251 | ||
755200a3 JA |
1252 | /* |
1253 | * from a requeue, io_u already setup | |
1254 | */ | |
1255 | if (io_u->file) | |
77f392bf | 1256 | goto out; |
755200a3 | 1257 | |
429f6675 JA |
1258 | /* |
1259 | * If using an iolog, grab next piece if any available. | |
1260 | */ | |
1261 | if (td->o.read_iolog_file) { | |
1262 | if (read_iolog_get(td, io_u)) | |
1263 | goto err_put; | |
2ba1c290 JA |
1264 | } else if (set_io_u_file(td, io_u)) { |
1265 | dprint(FD_IO, "io_u %p, setting file failed\n", io_u); | |
429f6675 | 1266 | goto err_put; |
2ba1c290 | 1267 | } |
5ec10eaa | 1268 | |
429f6675 | 1269 | f = io_u->file; |
d6aed795 | 1270 | assert(fio_file_open(f)); |
97af62ce | 1271 | |
ff58fced | 1272 | if (ddir_rw(io_u->ddir)) { |
d0656a93 | 1273 | if (!io_u->buflen && !(td->io_ops->flags & FIO_NOIO)) { |
2ba1c290 | 1274 | dprint(FD_IO, "get_io_u: zero buflen on %p\n", io_u); |
429f6675 | 1275 | goto err_put; |
2ba1c290 | 1276 | } |
10ba535a | 1277 | |
38dad62d | 1278 | f->last_start = io_u->offset; |
36167d82 | 1279 | f->last_pos = io_u->offset + io_u->buflen; |
10ba535a | 1280 | |
fd68418e | 1281 | if (io_u->ddir == DDIR_WRITE) { |
629f1d71 | 1282 | if (td->o.refill_buffers) { |
9c42684e JA |
1283 | io_u_fill_buffer(td, io_u, |
1284 | io_u->xfer_buflen, io_u->xfer_buflen); | |
1285 | } else if (td->o.scramble_buffers) | |
fd68418e | 1286 | do_scramble = 1; |
629f1d71 JA |
1287 | if (td->o.verify != VERIFY_NONE) { |
1288 | populate_verify_io_u(td, io_u); | |
1289 | do_scramble = 0; | |
1290 | } | |
fd68418e | 1291 | } else if (io_u->ddir == DDIR_READ) { |
cbe8d756 RR |
1292 | /* |
1293 | * Reset the buf_filled parameters so next time if the | |
1294 | * buffer is used for writes it is refilled. | |
1295 | */ | |
cbe8d756 RR |
1296 | io_u->buf_filled_len = 0; |
1297 | } | |
87dc1ab1 | 1298 | } |
10ba535a | 1299 | |
165faf16 JA |
1300 | /* |
1301 | * Set io data pointers. | |
1302 | */ | |
cec6b55d JA |
1303 | io_u->xfer_buf = io_u->buf; |
1304 | io_u->xfer_buflen = io_u->buflen; | |
5973cafb | 1305 | |
6ac7a331 | 1306 | out: |
0d29de83 | 1307 | assert(io_u->file); |
429f6675 | 1308 | if (!td_io_prep(td, io_u)) { |
993bf48b JA |
1309 | if (!td->o.disable_slat) |
1310 | fio_gettime(&io_u->start_time, NULL); | |
de789769 JA |
1311 | if (do_scramble) |
1312 | small_content_scramble(io_u); | |
429f6675 | 1313 | return io_u; |
36167d82 | 1314 | } |
429f6675 | 1315 | err_put: |
2ba1c290 | 1316 | dprint(FD_IO, "get_io_u failed\n"); |
429f6675 JA |
1317 | put_io_u(td, io_u); |
1318 | return NULL; | |
10ba535a JA |
1319 | } |
1320 | ||
5451792e JA |
1321 | void io_u_log_error(struct thread_data *td, struct io_u *io_u) |
1322 | { | |
8b28bd41 | 1323 | enum error_type_bit eb = td_error_type(io_u->ddir, io_u->error); |
825f818e JA |
1324 | const char *msg[] = { "read", "write", "sync", "datasync", |
1325 | "sync_file_range", "wait", "trim" }; | |
1326 | ||
8b28bd41 DM |
1327 | if (td_non_fatal_error(td, eb, io_u->error) && !td->o.error_dump) |
1328 | return; | |
5451792e JA |
1329 | |
1330 | log_err("fio: io_u error"); | |
1331 | ||
1332 | if (io_u->file) | |
1333 | log_err(" on file %s", io_u->file->file_name); | |
1334 | ||
1335 | log_err(": %s\n", strerror(io_u->error)); | |
1336 | ||
5ec10eaa JA |
1337 | log_err(" %s offset=%llu, buflen=%lu\n", msg[io_u->ddir], |
1338 | io_u->offset, io_u->xfer_buflen); | |
5451792e JA |
1339 | |
1340 | if (!td->error) | |
1341 | td_verror(td, io_u->error, "io_u error"); | |
1342 | } | |
1343 | ||
c8eeb9df JA |
1344 | static void account_io_completion(struct thread_data *td, struct io_u *io_u, |
1345 | struct io_completion_data *icd, | |
1346 | const enum fio_ddir idx, unsigned int bytes) | |
1347 | { | |
24d23ca7 | 1348 | unsigned long lusec = 0; |
c8eeb9df | 1349 | |
c8eeb9df JA |
1350 | if (!td->o.disable_clat || !td->o.disable_bw) |
1351 | lusec = utime_since(&io_u->issue_time, &icd->time); | |
1352 | ||
1353 | if (!td->o.disable_lat) { | |
1354 | unsigned long tusec; | |
1355 | ||
1356 | tusec = utime_since(&io_u->start_time, &icd->time); | |
1357 | add_lat_sample(td, idx, tusec, bytes); | |
15501535 JA |
1358 | |
1359 | if (td->o.max_latency && tusec > td->o.max_latency) { | |
1360 | if (!td->error) | |
1361 | log_err("fio: latency of %lu usec exceeds specified max (%u usec)\n", tusec, td->o.max_latency); | |
c5b3adc4 JA |
1362 | td_verror(td, ETIMEDOUT, "max latency exceeded"); |
1363 | icd->error = ETIMEDOUT; | |
15501535 | 1364 | } |
c8eeb9df JA |
1365 | } |
1366 | ||
1367 | if (!td->o.disable_clat) { | |
1368 | add_clat_sample(td, idx, lusec, bytes); | |
1369 | io_u_mark_latency(td, lusec); | |
1370 | } | |
1371 | ||
1372 | if (!td->o.disable_bw) | |
1373 | add_bw_sample(td, idx, bytes, &icd->time); | |
1374 | ||
1375 | add_iops_sample(td, idx, &icd->time); | |
1376 | } | |
1377 | ||
1b8dbf25 SL |
1378 | static long long usec_for_io(struct thread_data *td, enum fio_ddir ddir) |
1379 | { | |
1380 | unsigned long long secs, remainder, bps, bytes; | |
1381 | bytes = td->this_io_bytes[ddir]; | |
1382 | bps = td->rate_bps[ddir]; | |
1383 | secs = bytes / bps; | |
1384 | remainder = bytes % bps; | |
1385 | return remainder * 1000000 / bps + secs * 1000000; | |
1386 | } | |
1387 | ||
97601024 JA |
1388 | static void io_completed(struct thread_data *td, struct io_u *io_u, |
1389 | struct io_completion_data *icd) | |
10ba535a | 1390 | { |
44f29692 | 1391 | struct fio_file *f; |
10ba535a | 1392 | |
2ba1c290 JA |
1393 | dprint_io_u(io_u, "io complete"); |
1394 | ||
2ecc1b57 | 1395 | td_io_u_lock(td); |
0c6e7517 | 1396 | assert(io_u->flags & IO_U_F_FLIGHT); |
38dad62d | 1397 | io_u->flags &= ~(IO_U_F_FLIGHT | IO_U_F_BUSY_OK); |
2ecc1b57 | 1398 | td_io_u_unlock(td); |
0c6e7517 | 1399 | |
5f9099ea | 1400 | if (ddir_sync(io_u->ddir)) { |
87dc1ab1 | 1401 | td->last_was_sync = 1; |
44f29692 JA |
1402 | f = io_u->file; |
1403 | if (f) { | |
1404 | f->first_write = -1ULL; | |
1405 | f->last_write = -1ULL; | |
1406 | } | |
87dc1ab1 JA |
1407 | return; |
1408 | } | |
1409 | ||
1410 | td->last_was_sync = 0; | |
9e144189 | 1411 | td->last_ddir = io_u->ddir; |
87dc1ab1 | 1412 | |
ff58fced | 1413 | if (!io_u->error && ddir_rw(io_u->ddir)) { |
10ba535a | 1414 | unsigned int bytes = io_u->buflen - io_u->resid; |
1e97cce9 | 1415 | const enum fio_ddir idx = io_u->ddir; |
ba3e4e0c | 1416 | const enum fio_ddir odx = io_u->ddir ^ 1; |
b29ee5b3 | 1417 | int ret; |
10ba535a | 1418 | |
b29ee5b3 | 1419 | td->io_blocks[idx]++; |
c8eeb9df | 1420 | td->this_io_blocks[idx]++; |
b29ee5b3 | 1421 | td->io_bytes[idx] += bytes; |
ae2fafc8 | 1422 | |
1423 | if (!(io_u->flags & IO_U_F_VER_LIST)) | |
1424 | td->this_io_bytes[idx] += bytes; | |
10ba535a | 1425 | |
44f29692 JA |
1426 | if (idx == DDIR_WRITE) { |
1427 | f = io_u->file; | |
1428 | if (f) { | |
1429 | if (f->first_write == -1ULL || | |
1430 | io_u->offset < f->first_write) | |
1431 | f->first_write = io_u->offset; | |
1432 | if (f->last_write == -1ULL || | |
1433 | ((io_u->offset + bytes) > f->last_write)) | |
1434 | f->last_write = io_u->offset + bytes; | |
1435 | } | |
1436 | } | |
1437 | ||
6b1190fd SL |
1438 | if (ramp_time_over(td) && (td->runstate == TD_RUNNING || |
1439 | td->runstate == TD_VERIFYING)) { | |
c8eeb9df | 1440 | account_io_completion(td, io_u, icd, idx, bytes); |
40e1a6f0 | 1441 | |
b23b6a2f | 1442 | if (__should_check_rate(td, idx)) { |
ba3e4e0c | 1443 | td->rate_pending_usleep[idx] = |
1b8dbf25 | 1444 | (usec_for_io(td, idx) - |
ba3e4e0c | 1445 | utime_since_now(&td->start)); |
b23b6a2f | 1446 | } |
6eaf09d6 | 1447 | if (idx != DDIR_TRIM && __should_check_rate(td, odx)) |
ba3e4e0c | 1448 | td->rate_pending_usleep[odx] = |
1b8dbf25 | 1449 | (usec_for_io(td, odx) - |
ba3e4e0c | 1450 | utime_since_now(&td->start)); |
721938ae | 1451 | } |
10ba535a | 1452 | |
660a1cb5 | 1453 | if (td_write(td) && idx == DDIR_WRITE && |
8670579e | 1454 | td->o.do_verify && |
41128405 | 1455 | td->o.verify != VERIFY_NONE) |
10ba535a JA |
1456 | log_io_piece(td, io_u); |
1457 | ||
b29ee5b3 | 1458 | icd->bytes_done[idx] += bytes; |
3af6ef39 | 1459 | |
d7762cf8 | 1460 | if (io_u->end_io) { |
36690c9b | 1461 | ret = io_u->end_io(td, io_u); |
3af6ef39 JA |
1462 | if (ret && !icd->error) |
1463 | icd->error = ret; | |
1464 | } | |
ff58fced | 1465 | } else if (io_u->error) { |
10ba535a | 1466 | icd->error = io_u->error; |
5451792e JA |
1467 | io_u_log_error(td, io_u); |
1468 | } | |
8b28bd41 DM |
1469 | if (icd->error) { |
1470 | enum error_type_bit eb = td_error_type(io_u->ddir, icd->error); | |
1471 | if (!td_non_fatal_error(td, eb, icd->error)) | |
1472 | return; | |
f2bba182 RR |
1473 | /* |
1474 | * If there is a non_fatal error, then add to the error count | |
1475 | * and clear all the errors. | |
1476 | */ | |
1477 | update_error_count(td, icd->error); | |
1478 | td_clear_error(td); | |
1479 | icd->error = 0; | |
1480 | io_u->error = 0; | |
1481 | } | |
10ba535a JA |
1482 | } |
1483 | ||
9520ebb9 JA |
1484 | static void init_icd(struct thread_data *td, struct io_completion_data *icd, |
1485 | int nr) | |
10ba535a | 1486 | { |
6eaf09d6 | 1487 | int ddir; |
9520ebb9 JA |
1488 | if (!td->o.disable_clat || !td->o.disable_bw) |
1489 | fio_gettime(&icd->time, NULL); | |
02bcaa8c | 1490 | |
3af6ef39 JA |
1491 | icd->nr = nr; |
1492 | ||
10ba535a | 1493 | icd->error = 0; |
6eaf09d6 SL |
1494 | for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) |
1495 | icd->bytes_done[ddir] = 0; | |
36167d82 JA |
1496 | } |
1497 | ||
97601024 JA |
1498 | static void ios_completed(struct thread_data *td, |
1499 | struct io_completion_data *icd) | |
36167d82 JA |
1500 | { |
1501 | struct io_u *io_u; | |
1502 | int i; | |
1503 | ||
10ba535a JA |
1504 | for (i = 0; i < icd->nr; i++) { |
1505 | io_u = td->io_ops->event(td, i); | |
1506 | ||
1507 | io_completed(td, io_u, icd); | |
e8462bd8 JA |
1508 | |
1509 | if (!(io_u->flags & IO_U_F_FREE_DEF)) | |
1510 | put_io_u(td, io_u); | |
10ba535a JA |
1511 | } |
1512 | } | |
97601024 | 1513 | |
e7e6cfb4 JA |
1514 | /* |
1515 | * Complete a single io_u for the sync engines. | |
1516 | */ | |
581e7141 JA |
1517 | int io_u_sync_complete(struct thread_data *td, struct io_u *io_u, |
1518 | unsigned long *bytes) | |
97601024 JA |
1519 | { |
1520 | struct io_completion_data icd; | |
1521 | ||
9520ebb9 | 1522 | init_icd(td, &icd, 1); |
97601024 | 1523 | io_completed(td, io_u, &icd); |
e8462bd8 JA |
1524 | |
1525 | if (!(io_u->flags & IO_U_F_FREE_DEF)) | |
1526 | put_io_u(td, io_u); | |
97601024 | 1527 | |
581e7141 JA |
1528 | if (icd.error) { |
1529 | td_verror(td, icd.error, "io_u_sync_complete"); | |
1530 | return -1; | |
1531 | } | |
97601024 | 1532 | |
581e7141 | 1533 | if (bytes) { |
6eaf09d6 SL |
1534 | int ddir; |
1535 | ||
1536 | for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) | |
1537 | bytes[ddir] += icd.bytes_done[ddir]; | |
581e7141 JA |
1538 | } |
1539 | ||
1540 | return 0; | |
97601024 JA |
1541 | } |
1542 | ||
e7e6cfb4 JA |
1543 | /* |
1544 | * Called to complete min_events number of io for the async engines. | |
1545 | */ | |
581e7141 JA |
1546 | int io_u_queued_complete(struct thread_data *td, int min_evts, |
1547 | unsigned long *bytes) | |
97601024 | 1548 | { |
97601024 | 1549 | struct io_completion_data icd; |
00de55ef | 1550 | struct timespec *tvp = NULL; |
97601024 | 1551 | int ret; |
4d06a338 | 1552 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0, }; |
97601024 | 1553 | |
4950421a | 1554 | dprint(FD_IO, "io_u_queued_completed: min=%d\n", min_evts); |
b271fe62 | 1555 | |
4950421a | 1556 | if (!min_evts) |
00de55ef | 1557 | tvp = &ts; |
97601024 | 1558 | |
4950421a | 1559 | ret = td_io_getevents(td, min_evts, td->o.iodepth_batch_complete, tvp); |
97601024 | 1560 | if (ret < 0) { |
e1161c32 | 1561 | td_verror(td, -ret, "td_io_getevents"); |
97601024 JA |
1562 | return ret; |
1563 | } else if (!ret) | |
1564 | return ret; | |
1565 | ||
9520ebb9 | 1566 | init_icd(td, &icd, ret); |
97601024 | 1567 | ios_completed(td, &icd); |
581e7141 JA |
1568 | if (icd.error) { |
1569 | td_verror(td, icd.error, "io_u_queued_complete"); | |
1570 | return -1; | |
1571 | } | |
97601024 | 1572 | |
581e7141 | 1573 | if (bytes) { |
6eaf09d6 SL |
1574 | int ddir; |
1575 | ||
1576 | for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) | |
1577 | bytes[ddir] += icd.bytes_done[ddir]; | |
581e7141 JA |
1578 | } |
1579 | ||
1580 | return 0; | |
97601024 | 1581 | } |
7e77dd02 JA |
1582 | |
1583 | /* | |
1584 | * Call when io_u is really queued, to update the submission latency. | |
1585 | */ | |
1586 | void io_u_queued(struct thread_data *td, struct io_u *io_u) | |
1587 | { | |
9520ebb9 JA |
1588 | if (!td->o.disable_slat) { |
1589 | unsigned long slat_time; | |
7e77dd02 | 1590 | |
9520ebb9 | 1591 | slat_time = utime_since(&io_u->start_time, &io_u->issue_time); |
29a90ddb | 1592 | add_slat_sample(td, io_u->ddir, slat_time, io_u->xfer_buflen); |
9520ebb9 | 1593 | } |
7e77dd02 | 1594 | } |
433afcb4 | 1595 | |
5973cafb JA |
1596 | /* |
1597 | * "randomly" fill the buffer contents | |
1598 | */ | |
1599 | void io_u_fill_buffer(struct thread_data *td, struct io_u *io_u, | |
9c42684e | 1600 | unsigned int min_write, unsigned int max_bs) |
5973cafb | 1601 | { |
e84b371e JA |
1602 | io_u->buf_filled_len = 0; |
1603 | ||
9c42684e JA |
1604 | if (!td->o.zero_buffers) { |
1605 | unsigned int perc = td->o.compress_percentage; | |
1606 | ||
1607 | if (perc) { | |
f97a43a1 JA |
1608 | unsigned int seg = min_write; |
1609 | ||
1610 | seg = min(min_write, td->o.compress_chunk); | |
9c42684e | 1611 | fill_random_buf_percentage(&td->buf_state, io_u->buf, |
f97a43a1 | 1612 | perc, seg, max_bs); |
9c42684e JA |
1613 | } else |
1614 | fill_random_buf(&td->buf_state, io_u->buf, max_bs); | |
1615 | } else | |
637ef8d9 | 1616 | memset(io_u->buf, 0, max_bs); |
5973cafb | 1617 | } |