Commit | Line | Data |
---|---|---|
10ba535a JA |
1 | #include <unistd.h> |
2 | #include <fcntl.h> | |
3 | #include <string.h> | |
4 | #include <signal.h> | |
5 | #include <time.h> | |
0c6e7517 | 6 | #include <assert.h> |
10ba535a JA |
7 | |
8 | #include "fio.h" | |
5973cafb | 9 | #include "hash.h" |
4f5af7b2 | 10 | #include "verify.h" |
0d29de83 | 11 | #include "trim.h" |
1fbbf72e | 12 | #include "lib/rand.h" |
7ebd796f | 13 | #include "lib/axmap.h" |
10ba535a | 14 | |
97601024 JA |
15 | struct io_completion_data { |
16 | int nr; /* input */ | |
97601024 JA |
17 | |
18 | int error; /* output */ | |
100f49f1 | 19 | uint64_t bytes_done[DDIR_RWDIR_CNT]; /* output */ |
97601024 JA |
20 | struct timeval time; /* output */ |
21 | }; | |
22 | ||
10ba535a | 23 | /* |
7ebd796f | 24 | * The ->io_axmap contains a map of blocks we have or have not done io |
10ba535a JA |
25 | * to yet. Used to make sure we cover the entire range in a fair fashion. |
26 | */ | |
1ae83d45 | 27 | static int random_map_free(struct fio_file *f, const uint64_t block) |
10ba535a | 28 | { |
7ebd796f | 29 | return !axmap_isset(f->io_axmap, block); |
10ba535a JA |
30 | } |
31 | ||
df415585 JA |
32 | /* |
33 | * Mark a given offset as used in the map. | |
34 | */ | |
9bf2061e | 35 | static void mark_random_map(struct thread_data *td, struct io_u *io_u) |
df415585 | 36 | { |
2dc1bbeb | 37 | unsigned int min_bs = td->o.rw_min_bs; |
9bf2061e | 38 | struct fio_file *f = io_u->file; |
51ede0b1 | 39 | unsigned int nr_blocks; |
1ae83d45 | 40 | uint64_t block; |
df415585 | 41 | |
1ae83d45 | 42 | block = (io_u->offset - f->file_offset) / (uint64_t) min_bs; |
c685b5b2 | 43 | nr_blocks = (io_u->buflen + min_bs - 1) / min_bs; |
df415585 | 44 | |
2ab9e98b | 45 | if (!(io_u->flags & IO_U_F_BUSY_OK)) |
7ebd796f | 46 | nr_blocks = axmap_set_nr(f->io_axmap, block, nr_blocks); |
df415585 | 47 | |
51ede0b1 JA |
48 | if ((nr_blocks * min_bs) < io_u->buflen) |
49 | io_u->buflen = nr_blocks * min_bs; | |
df415585 JA |
50 | } |
51 | ||
74776733 JA |
52 | static uint64_t last_block(struct thread_data *td, struct fio_file *f, |
53 | enum fio_ddir ddir) | |
2ba1c290 | 54 | { |
74776733 JA |
55 | uint64_t max_blocks; |
56 | uint64_t max_size; | |
2ba1c290 | 57 | |
ff58fced JA |
58 | assert(ddir_rw(ddir)); |
59 | ||
d9dd70f7 JA |
60 | /* |
61 | * Hmm, should we make sure that ->io_size <= ->real_file_size? | |
62 | */ | |
63 | max_size = f->io_size; | |
64 | if (max_size > f->real_file_size) | |
65 | max_size = f->real_file_size; | |
66 | ||
ed335855 SN |
67 | if (td->o.zone_range) |
68 | max_size = td->o.zone_range; | |
69 | ||
1ae83d45 | 70 | max_blocks = max_size / (uint64_t) td->o.ba[ddir]; |
2ba1c290 JA |
71 | if (!max_blocks) |
72 | return 0; | |
73 | ||
67778e88 | 74 | return max_blocks; |
2ba1c290 JA |
75 | } |
76 | ||
1ae83d45 JA |
77 | struct rand_off { |
78 | struct flist_head list; | |
79 | uint64_t off; | |
80 | }; | |
81 | ||
e25839d4 | 82 | static int __get_next_rand_offset(struct thread_data *td, struct fio_file *f, |
1ae83d45 | 83 | enum fio_ddir ddir, uint64_t *b) |
ec4015da | 84 | { |
1ae83d45 | 85 | uint64_t r, lastb; |
ec4015da | 86 | |
74776733 JA |
87 | lastb = last_block(td, f, ddir); |
88 | if (!lastb) | |
89 | return 1; | |
5e0baa7f | 90 | |
74776733 | 91 | if (td->o.random_generator == FIO_RAND_GEN_TAUSWORTHE) { |
1ae83d45 | 92 | uint64_t rmax; |
2615cc4b | 93 | |
8055e41d JA |
94 | rmax = td->o.use_os_rand ? OS_RAND_MAX : FRAND_MAX; |
95 | ||
96 | if (td->o.use_os_rand) { | |
97 | rmax = OS_RAND_MAX; | |
98 | r = os_random_long(&td->random_state); | |
99 | } else { | |
100 | rmax = FRAND_MAX; | |
101 | r = __rand(&td->__random_state); | |
102 | } | |
103 | ||
4b91ee8f | 104 | dprint(FD_RANDOM, "off rand %llu\n", (unsigned long long) r); |
8055e41d | 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 | ||
4b91ee8f JA |
128 | dprint(FD_RANDOM, "get_next_rand_offset: offset %llu busy\n", |
129 | (unsigned long long) *b); | |
51ede0b1 | 130 | |
7ebd796f | 131 | *b = axmap_next_free(f->io_axmap, *b); |
51ede0b1 JA |
132 | if (*b == (uint64_t) -1ULL) |
133 | return 1; | |
0ce8b119 JA |
134 | ret: |
135 | return 0; | |
ec4015da JA |
136 | } |
137 | ||
925fee33 JA |
138 | static int __get_next_rand_offset_zipf(struct thread_data *td, |
139 | struct fio_file *f, enum fio_ddir ddir, | |
1ae83d45 | 140 | uint64_t *b) |
e25839d4 | 141 | { |
9c6f6316 | 142 | *b = zipf_next(&f->zipf); |
e25839d4 JA |
143 | return 0; |
144 | } | |
145 | ||
925fee33 JA |
146 | static int __get_next_rand_offset_pareto(struct thread_data *td, |
147 | struct fio_file *f, enum fio_ddir ddir, | |
1ae83d45 | 148 | uint64_t *b) |
925fee33 | 149 | { |
9c6f6316 | 150 | *b = pareto_next(&f->zipf); |
925fee33 JA |
151 | return 0; |
152 | } | |
153 | ||
1ae83d45 JA |
154 | static int flist_cmp(void *data, struct flist_head *a, struct flist_head *b) |
155 | { | |
156 | struct rand_off *r1 = flist_entry(a, struct rand_off, list); | |
157 | struct rand_off *r2 = flist_entry(b, struct rand_off, list); | |
158 | ||
159 | return r1->off - r2->off; | |
160 | } | |
161 | ||
162 | static int get_off_from_method(struct thread_data *td, struct fio_file *f, | |
163 | enum fio_ddir ddir, uint64_t *b) | |
e25839d4 JA |
164 | { |
165 | if (td->o.random_distribution == FIO_RAND_DIST_RANDOM) | |
166 | return __get_next_rand_offset(td, f, ddir, b); | |
167 | else if (td->o.random_distribution == FIO_RAND_DIST_ZIPF) | |
168 | return __get_next_rand_offset_zipf(td, f, ddir, b); | |
925fee33 JA |
169 | else if (td->o.random_distribution == FIO_RAND_DIST_PARETO) |
170 | return __get_next_rand_offset_pareto(td, f, ddir, b); | |
e25839d4 JA |
171 | |
172 | log_err("fio: unknown random distribution: %d\n", td->o.random_distribution); | |
173 | return 1; | |
174 | } | |
175 | ||
bcd5abfa JA |
176 | /* |
177 | * Sort the reads for a verify phase in batches of verifysort_nr, if | |
178 | * specified. | |
179 | */ | |
180 | static inline int should_sort_io(struct thread_data *td) | |
181 | { | |
182 | if (!td->o.verifysort_nr || !td->o.do_verify) | |
183 | return 0; | |
184 | if (!td_random(td)) | |
185 | return 0; | |
186 | if (td->runstate != TD_VERIFYING) | |
187 | return 0; | |
188 | if (td->o.random_generator == FIO_RAND_GEN_TAUSWORTHE) | |
189 | return 0; | |
190 | ||
191 | return 1; | |
192 | } | |
193 | ||
d9472271 | 194 | static int should_do_random(struct thread_data *td, enum fio_ddir ddir) |
211c9b89 JA |
195 | { |
196 | unsigned int v; | |
197 | unsigned long r; | |
198 | ||
d9472271 | 199 | if (td->o.perc_rand[ddir] == 100) |
211c9b89 JA |
200 | return 1; |
201 | ||
202 | if (td->o.use_os_rand) { | |
d9472271 | 203 | r = os_random_long(&td->seq_rand_state[ddir]); |
211c9b89 JA |
204 | v = 1 + (int) (100.0 * (r / (OS_RAND_MAX + 1.0))); |
205 | } else { | |
d9472271 | 206 | r = __rand(&td->__seq_rand_state[ddir]); |
211c9b89 JA |
207 | v = 1 + (int) (100.0 * (r / (FRAND_MAX + 1.0))); |
208 | } | |
209 | ||
d9472271 | 210 | return v <= td->o.perc_rand[ddir]; |
211c9b89 JA |
211 | } |
212 | ||
1ae83d45 JA |
213 | static int get_next_rand_offset(struct thread_data *td, struct fio_file *f, |
214 | enum fio_ddir ddir, uint64_t *b) | |
215 | { | |
216 | struct rand_off *r; | |
217 | int i, ret = 1; | |
218 | ||
bcd5abfa | 219 | if (!should_sort_io(td)) |
1ae83d45 JA |
220 | return get_off_from_method(td, f, ddir, b); |
221 | ||
222 | if (!flist_empty(&td->next_rand_list)) { | |
223 | struct rand_off *r; | |
224 | fetch: | |
225 | r = flist_entry(td->next_rand_list.next, struct rand_off, list); | |
226 | flist_del(&r->list); | |
227 | *b = r->off; | |
228 | free(r); | |
229 | return 0; | |
230 | } | |
231 | ||
232 | for (i = 0; i < td->o.verifysort_nr; i++) { | |
233 | r = malloc(sizeof(*r)); | |
234 | ||
235 | ret = get_off_from_method(td, f, ddir, &r->off); | |
236 | if (ret) { | |
237 | free(r); | |
238 | break; | |
239 | } | |
240 | ||
241 | flist_add(&r->list, &td->next_rand_list); | |
242 | } | |
243 | ||
244 | if (ret && !i) | |
245 | return ret; | |
246 | ||
247 | assert(!flist_empty(&td->next_rand_list)); | |
248 | flist_sort(NULL, &td->next_rand_list, flist_cmp); | |
249 | goto fetch; | |
250 | } | |
251 | ||
38dad62d | 252 | static int get_next_rand_block(struct thread_data *td, struct fio_file *f, |
1ae83d45 | 253 | enum fio_ddir ddir, uint64_t *b) |
38dad62d | 254 | { |
c04e4661 DE |
255 | if (!get_next_rand_offset(td, f, ddir, b)) |
256 | return 0; | |
257 | ||
258 | if (td->o.time_based) { | |
33c48814 | 259 | fio_file_reset(td, f); |
c04e4661 DE |
260 | if (!get_next_rand_offset(td, f, ddir, b)) |
261 | return 0; | |
38dad62d JA |
262 | } |
263 | ||
c04e4661 | 264 | dprint(FD_IO, "%s: rand offset failed, last=%llu, size=%llu\n", |
4b91ee8f JA |
265 | f->file_name, (unsigned long long) f->last_pos, |
266 | (unsigned long long) f->real_file_size); | |
c04e4661 | 267 | return 1; |
38dad62d JA |
268 | } |
269 | ||
37cf9e3c | 270 | static int get_next_seq_offset(struct thread_data *td, struct fio_file *f, |
1ae83d45 | 271 | enum fio_ddir ddir, uint64_t *offset) |
38dad62d | 272 | { |
ff58fced JA |
273 | assert(ddir_rw(ddir)); |
274 | ||
ce95d651 | 275 | if (f->last_pos >= f->io_size + get_start_offset(td) && td->o.time_based) |
c04e4661 DE |
276 | f->last_pos = f->last_pos - f->io_size; |
277 | ||
38dad62d | 278 | if (f->last_pos < f->real_file_size) { |
1ae83d45 | 279 | uint64_t pos; |
059b0802 | 280 | |
a66da7a2 JA |
281 | if (f->last_pos == f->file_offset && td->o.ddir_seq_add < 0) |
282 | f->last_pos = f->real_file_size; | |
283 | ||
284 | pos = f->last_pos - f->file_offset; | |
059b0802 JA |
285 | if (pos) |
286 | pos += td->o.ddir_seq_add; | |
287 | ||
37cf9e3c | 288 | *offset = pos; |
38dad62d JA |
289 | return 0; |
290 | } | |
291 | ||
292 | return 1; | |
293 | } | |
294 | ||
295 | static int get_next_block(struct thread_data *td, struct io_u *io_u, | |
6aca9b3d JA |
296 | enum fio_ddir ddir, int rw_seq, |
297 | unsigned int *is_random) | |
38dad62d JA |
298 | { |
299 | struct fio_file *f = io_u->file; | |
1ae83d45 | 300 | uint64_t b, offset; |
38dad62d JA |
301 | int ret; |
302 | ||
ff58fced JA |
303 | assert(ddir_rw(ddir)); |
304 | ||
37cf9e3c JA |
305 | b = offset = -1ULL; |
306 | ||
38dad62d | 307 | if (rw_seq) { |
211c9b89 | 308 | if (td_random(td)) { |
6aca9b3d | 309 | if (should_do_random(td, ddir)) { |
211c9b89 | 310 | ret = get_next_rand_block(td, f, ddir, &b); |
6aca9b3d JA |
311 | *is_random = 1; |
312 | } else { | |
313 | *is_random = 0; | |
211c9b89 JA |
314 | io_u->flags |= IO_U_F_BUSY_OK; |
315 | ret = get_next_seq_offset(td, f, ddir, &offset); | |
316 | if (ret) | |
317 | ret = get_next_rand_block(td, f, ddir, &b); | |
318 | } | |
6aca9b3d JA |
319 | } else { |
320 | *is_random = 0; | |
37cf9e3c | 321 | ret = get_next_seq_offset(td, f, ddir, &offset); |
6aca9b3d | 322 | } |
38dad62d JA |
323 | } else { |
324 | io_u->flags |= IO_U_F_BUSY_OK; | |
6aca9b3d | 325 | *is_random = 0; |
38dad62d JA |
326 | |
327 | if (td->o.rw_seq == RW_SEQ_SEQ) { | |
37cf9e3c | 328 | ret = get_next_seq_offset(td, f, ddir, &offset); |
6aca9b3d | 329 | if (ret) { |
37cf9e3c | 330 | ret = get_next_rand_block(td, f, ddir, &b); |
6aca9b3d JA |
331 | *is_random = 0; |
332 | } | |
38dad62d JA |
333 | } else if (td->o.rw_seq == RW_SEQ_IDENT) { |
334 | if (f->last_start != -1ULL) | |
37cf9e3c | 335 | offset = f->last_start - f->file_offset; |
38dad62d | 336 | else |
37cf9e3c | 337 | offset = 0; |
38dad62d JA |
338 | ret = 0; |
339 | } else { | |
340 | log_err("fio: unknown rw_seq=%d\n", td->o.rw_seq); | |
341 | ret = 1; | |
342 | } | |
343 | } | |
6d68b997 | 344 | |
37cf9e3c JA |
345 | if (!ret) { |
346 | if (offset != -1ULL) | |
347 | io_u->offset = offset; | |
348 | else if (b != -1ULL) | |
349 | io_u->offset = b * td->o.ba[ddir]; | |
350 | else { | |
4e0a8fa2 | 351 | log_err("fio: bug in offset generation: offset=%llu, b=%llu\n", (unsigned long long) offset, (unsigned long long) b); |
37cf9e3c JA |
352 | ret = 1; |
353 | } | |
354 | } | |
355 | ||
38dad62d JA |
356 | return ret; |
357 | } | |
358 | ||
10ba535a JA |
359 | /* |
360 | * For random io, generate a random new block and see if it's used. Repeat | |
361 | * until we find a free one. For sequential io, just return the end of | |
362 | * the last io issued. | |
363 | */ | |
6aca9b3d JA |
364 | static int __get_next_offset(struct thread_data *td, struct io_u *io_u, |
365 | unsigned int *is_random) | |
10ba535a | 366 | { |
9bf2061e | 367 | struct fio_file *f = io_u->file; |
4ba66134 | 368 | enum fio_ddir ddir = io_u->ddir; |
38dad62d | 369 | int rw_seq_hit = 0; |
10ba535a | 370 | |
ff58fced JA |
371 | assert(ddir_rw(ddir)); |
372 | ||
38dad62d JA |
373 | if (td->o.ddir_seq_nr && !--td->ddir_seq_nr) { |
374 | rw_seq_hit = 1; | |
5736c10d | 375 | td->ddir_seq_nr = td->o.ddir_seq_nr; |
38dad62d | 376 | } |
211097b2 | 377 | |
6aca9b3d | 378 | if (get_next_block(td, io_u, ddir, rw_seq_hit, is_random)) |
38dad62d | 379 | return 1; |
10ba535a | 380 | |
009bd847 JA |
381 | if (io_u->offset >= f->io_size) { |
382 | dprint(FD_IO, "get_next_offset: offset %llu >= io_size %llu\n", | |
4b91ee8f JA |
383 | (unsigned long long) io_u->offset, |
384 | (unsigned long long) f->io_size); | |
009bd847 JA |
385 | return 1; |
386 | } | |
387 | ||
388 | io_u->offset += f->file_offset; | |
2ba1c290 JA |
389 | if (io_u->offset >= f->real_file_size) { |
390 | dprint(FD_IO, "get_next_offset: offset %llu >= size %llu\n", | |
4b91ee8f JA |
391 | (unsigned long long) io_u->offset, |
392 | (unsigned long long) f->real_file_size); | |
10ba535a | 393 | return 1; |
2ba1c290 | 394 | } |
10ba535a JA |
395 | |
396 | return 0; | |
397 | } | |
398 | ||
6aca9b3d JA |
399 | static int get_next_offset(struct thread_data *td, struct io_u *io_u, |
400 | unsigned int *is_random) | |
15dc1934 | 401 | { |
d72be545 JA |
402 | if (td->flags & TD_F_PROFILE_OPS) { |
403 | struct prof_io_ops *ops = &td->prof_io_ops; | |
7eb36574 | 404 | |
d72be545 | 405 | if (ops->fill_io_u_off) |
6aca9b3d | 406 | return ops->fill_io_u_off(td, io_u, is_random); |
d72be545 | 407 | } |
15dc1934 | 408 | |
6aca9b3d | 409 | return __get_next_offset(td, io_u, is_random); |
15dc1934 JA |
410 | } |
411 | ||
79944128 JA |
412 | static inline int io_u_fits(struct thread_data *td, struct io_u *io_u, |
413 | unsigned int buflen) | |
414 | { | |
415 | struct fio_file *f = io_u->file; | |
416 | ||
ce95d651 | 417 | return io_u->offset + buflen <= f->io_size + get_start_offset(td); |
79944128 JA |
418 | } |
419 | ||
6aca9b3d JA |
420 | static unsigned int __get_next_buflen(struct thread_data *td, struct io_u *io_u, |
421 | unsigned int is_random) | |
10ba535a | 422 | { |
6aca9b3d | 423 | int ddir = io_u->ddir; |
24d23ca7 | 424 | unsigned int buflen = 0; |
f3059de1 | 425 | unsigned int minbs, maxbs; |
1294c3ec | 426 | unsigned long r, rand_max; |
10ba535a | 427 | |
6aca9b3d JA |
428 | assert(ddir_rw(io_u->ddir)); |
429 | ||
430 | if (td->o.bs_is_seq_rand) | |
431 | ddir = is_random ? DDIR_WRITE: DDIR_READ; | |
432 | else | |
433 | ddir = io_u->ddir; | |
ff58fced | 434 | |
f3059de1 JA |
435 | minbs = td->o.min_bs[ddir]; |
436 | maxbs = td->o.max_bs[ddir]; | |
437 | ||
79944128 JA |
438 | if (minbs == maxbs) |
439 | return minbs; | |
440 | ||
52c58027 JA |
441 | /* |
442 | * If we can't satisfy the min block size from here, then fail | |
443 | */ | |
444 | if (!io_u_fits(td, io_u, minbs)) | |
445 | return 0; | |
446 | ||
4c07ad86 JA |
447 | if (td->o.use_os_rand) |
448 | rand_max = OS_RAND_MAX; | |
449 | else | |
450 | rand_max = FRAND_MAX; | |
451 | ||
79944128 | 452 | do { |
4c07ad86 JA |
453 | if (td->o.use_os_rand) |
454 | r = os_random_long(&td->bsrange_state); | |
455 | else | |
456 | r = __rand(&td->__bsrange_state); | |
457 | ||
720e84ad | 458 | if (!td->o.bssplit_nr[ddir]) { |
f3059de1 | 459 | buflen = 1 + (unsigned int) ((double) maxbs * |
4c07ad86 | 460 | (r / (rand_max + 1.0))); |
f3059de1 JA |
461 | if (buflen < minbs) |
462 | buflen = minbs; | |
5ec10eaa | 463 | } else { |
564ca972 JA |
464 | long perc = 0; |
465 | unsigned int i; | |
466 | ||
720e84ad JA |
467 | for (i = 0; i < td->o.bssplit_nr[ddir]; i++) { |
468 | struct bssplit *bsp = &td->o.bssplit[ddir][i]; | |
564ca972 JA |
469 | |
470 | buflen = bsp->bs; | |
471 | perc += bsp->perc; | |
79944128 JA |
472 | if ((r <= ((rand_max / 100L) * perc)) && |
473 | io_u_fits(td, io_u, buflen)) | |
564ca972 JA |
474 | break; |
475 | } | |
476 | } | |
79944128 | 477 | |
a9f70b1f JB |
478 | if (td->o.do_verify && td->o.verify != VERIFY_NONE) |
479 | buflen = (buflen + td->o.verify_interval - 1) & | |
480 | ~(td->o.verify_interval - 1); | |
481 | ||
f3059de1 JA |
482 | if (!td->o.bs_unaligned && is_power_of_2(minbs)) |
483 | buflen = (buflen + minbs - 1) & ~(minbs - 1); | |
10ba535a | 484 | |
79944128 | 485 | } while (!io_u_fits(td, io_u, buflen)); |
6a5e6884 | 486 | |
10ba535a JA |
487 | return buflen; |
488 | } | |
489 | ||
6aca9b3d JA |
490 | static unsigned int get_next_buflen(struct thread_data *td, struct io_u *io_u, |
491 | unsigned int is_random) | |
15dc1934 | 492 | { |
d72be545 JA |
493 | if (td->flags & TD_F_PROFILE_OPS) { |
494 | struct prof_io_ops *ops = &td->prof_io_ops; | |
7eb36574 | 495 | |
d72be545 | 496 | if (ops->fill_io_u_size) |
6aca9b3d | 497 | return ops->fill_io_u_size(td, io_u, is_random); |
d72be545 | 498 | } |
15dc1934 | 499 | |
6aca9b3d | 500 | return __get_next_buflen(td, io_u, is_random); |
15dc1934 JA |
501 | } |
502 | ||
afe24a5a JA |
503 | static void set_rwmix_bytes(struct thread_data *td) |
504 | { | |
afe24a5a JA |
505 | unsigned int diff; |
506 | ||
507 | /* | |
508 | * we do time or byte based switch. this is needed because | |
509 | * buffered writes may issue a lot quicker than they complete, | |
510 | * whereas reads do not. | |
511 | */ | |
e47f799f | 512 | diff = td->o.rwmix[td->rwmix_ddir ^ 1]; |
04c540d9 | 513 | td->rwmix_issues = (td->io_issues[td->rwmix_ddir] * diff) / 100; |
e47f799f JA |
514 | } |
515 | ||
516 | static inline enum fio_ddir get_rand_ddir(struct thread_data *td) | |
517 | { | |
518 | unsigned int v; | |
1294c3ec | 519 | unsigned long r; |
e47f799f | 520 | |
4c07ad86 JA |
521 | if (td->o.use_os_rand) { |
522 | r = os_random_long(&td->rwmix_state); | |
523 | v = 1 + (int) (100.0 * (r / (OS_RAND_MAX + 1.0))); | |
524 | } else { | |
525 | r = __rand(&td->__rwmix_state); | |
526 | v = 1 + (int) (100.0 * (r / (FRAND_MAX + 1.0))); | |
527 | } | |
528 | ||
04c540d9 | 529 | if (v <= td->o.rwmix[DDIR_READ]) |
e47f799f JA |
530 | return DDIR_READ; |
531 | ||
532 | return DDIR_WRITE; | |
afe24a5a JA |
533 | } |
534 | ||
002e7183 JA |
535 | void io_u_quiesce(struct thread_data *td) |
536 | { | |
537 | /* | |
538 | * We are going to sleep, ensure that we flush anything pending as | |
539 | * not to skew our latency numbers. | |
540 | * | |
541 | * Changed to only monitor 'in flight' requests here instead of the | |
542 | * td->cur_depth, b/c td->cur_depth does not accurately represent | |
543 | * io's that have been actually submitted to an async engine, | |
544 | * and cur_depth is meaningless for sync engines. | |
545 | */ | |
546 | while (td->io_u_in_flight) { | |
547 | int fio_unused ret; | |
548 | ||
549 | ret = io_u_queued_complete(td, 1, NULL); | |
550 | } | |
551 | } | |
552 | ||
581e7141 JA |
553 | static enum fio_ddir rate_ddir(struct thread_data *td, enum fio_ddir ddir) |
554 | { | |
555 | enum fio_ddir odir = ddir ^ 1; | |
556 | struct timeval t; | |
557 | long usec; | |
558 | ||
ff58fced JA |
559 | assert(ddir_rw(ddir)); |
560 | ||
315fcfec | 561 | if (td->rate_pending_usleep[ddir] <= 0) |
581e7141 JA |
562 | return ddir; |
563 | ||
564 | /* | |
565 | * We have too much pending sleep in this direction. See if we | |
566 | * should switch. | |
567 | */ | |
315fcfec | 568 | if (td_rw(td) && td->o.rwmix[odir]) { |
581e7141 JA |
569 | /* |
570 | * Other direction does not have too much pending, switch | |
571 | */ | |
572 | if (td->rate_pending_usleep[odir] < 100000) | |
573 | return odir; | |
574 | ||
575 | /* | |
576 | * Both directions have pending sleep. Sleep the minimum time | |
577 | * and deduct from both. | |
578 | */ | |
579 | if (td->rate_pending_usleep[ddir] <= | |
580 | td->rate_pending_usleep[odir]) { | |
581 | usec = td->rate_pending_usleep[ddir]; | |
582 | } else { | |
583 | usec = td->rate_pending_usleep[odir]; | |
584 | ddir = odir; | |
585 | } | |
586 | } else | |
587 | usec = td->rate_pending_usleep[ddir]; | |
588 | ||
002e7183 | 589 | io_u_quiesce(td); |
78c1eda5 | 590 | |
581e7141 JA |
591 | fio_gettime(&t, NULL); |
592 | usec_sleep(td, usec); | |
593 | usec = utime_since_now(&t); | |
594 | ||
595 | td->rate_pending_usleep[ddir] -= usec; | |
596 | ||
597 | odir = ddir ^ 1; | |
598 | if (td_rw(td) && __should_check_rate(td, odir)) | |
599 | td->rate_pending_usleep[odir] -= usec; | |
0b9d69ec | 600 | |
6eaf09d6 SL |
601 | if (ddir_trim(ddir)) |
602 | return ddir; | |
e0224c6b | 603 | |
581e7141 JA |
604 | return ddir; |
605 | } | |
606 | ||
10ba535a JA |
607 | /* |
608 | * Return the data direction for the next io_u. If the job is a | |
609 | * mixed read/write workload, check the rwmix cycle and switch if | |
610 | * necessary. | |
611 | */ | |
1e97cce9 | 612 | static enum fio_ddir get_rw_ddir(struct thread_data *td) |
10ba535a | 613 | { |
581e7141 JA |
614 | enum fio_ddir ddir; |
615 | ||
5f9099ea JA |
616 | /* |
617 | * see if it's time to fsync | |
618 | */ | |
619 | if (td->o.fsync_blocks && | |
620 | !(td->io_issues[DDIR_WRITE] % td->o.fsync_blocks) && | |
621 | td->io_issues[DDIR_WRITE] && should_fsync(td)) | |
622 | return DDIR_SYNC; | |
623 | ||
624 | /* | |
625 | * see if it's time to fdatasync | |
626 | */ | |
627 | if (td->o.fdatasync_blocks && | |
628 | !(td->io_issues[DDIR_WRITE] % td->o.fdatasync_blocks) && | |
629 | td->io_issues[DDIR_WRITE] && should_fsync(td)) | |
630 | return DDIR_DATASYNC; | |
631 | ||
44f29692 JA |
632 | /* |
633 | * see if it's time to sync_file_range | |
634 | */ | |
635 | if (td->sync_file_range_nr && | |
636 | !(td->io_issues[DDIR_WRITE] % td->sync_file_range_nr) && | |
637 | td->io_issues[DDIR_WRITE] && should_fsync(td)) | |
638 | return DDIR_SYNC_FILE_RANGE; | |
639 | ||
10ba535a | 640 | if (td_rw(td)) { |
10ba535a JA |
641 | /* |
642 | * Check if it's time to seed a new data direction. | |
643 | */ | |
e4928662 | 644 | if (td->io_issues[td->rwmix_ddir] >= td->rwmix_issues) { |
e47f799f JA |
645 | /* |
646 | * Put a top limit on how many bytes we do for | |
647 | * one data direction, to avoid overflowing the | |
648 | * ranges too much | |
649 | */ | |
650 | ddir = get_rand_ddir(td); | |
e47f799f JA |
651 | |
652 | if (ddir != td->rwmix_ddir) | |
653 | set_rwmix_bytes(td); | |
654 | ||
655 | td->rwmix_ddir = ddir; | |
10ba535a | 656 | } |
581e7141 | 657 | ddir = td->rwmix_ddir; |
10ba535a | 658 | } else if (td_read(td)) |
581e7141 | 659 | ddir = DDIR_READ; |
6eaf09d6 | 660 | else if (td_write(td)) |
581e7141 | 661 | ddir = DDIR_WRITE; |
6eaf09d6 SL |
662 | else |
663 | ddir = DDIR_TRIM; | |
581e7141 JA |
664 | |
665 | td->rwmix_ddir = rate_ddir(td, ddir); | |
666 | return td->rwmix_ddir; | |
10ba535a JA |
667 | } |
668 | ||
1ef2b6be JA |
669 | static void set_rw_ddir(struct thread_data *td, struct io_u *io_u) |
670 | { | |
bcd5abfa | 671 | io_u->ddir = io_u->acct_ddir = get_rw_ddir(td); |
1ef2b6be JA |
672 | |
673 | if (io_u->ddir == DDIR_WRITE && (td->io_ops->flags & FIO_BARRIER) && | |
674 | td->o.barrier_blocks && | |
675 | !(td->io_issues[DDIR_WRITE] % td->o.barrier_blocks) && | |
676 | td->io_issues[DDIR_WRITE]) | |
677 | io_u->flags |= IO_U_F_BARRIER; | |
678 | } | |
679 | ||
e8462bd8 | 680 | void put_file_log(struct thread_data *td, struct fio_file *f) |
60f2c658 JA |
681 | { |
682 | int ret = put_file(td, f); | |
683 | ||
684 | if (ret) | |
685 | td_verror(td, ret, "file close"); | |
686 | } | |
687 | ||
10ba535a JA |
688 | void put_io_u(struct thread_data *td, struct io_u *io_u) |
689 | { | |
e8462bd8 JA |
690 | td_io_u_lock(td); |
691 | ||
d7ee2a7d | 692 | if (io_u->file && !(io_u->flags & IO_U_F_FREE_DEF)) |
60f2c658 | 693 | put_file_log(td, io_u->file); |
10ba535a | 694 | io_u->file = NULL; |
d7ee2a7d SL |
695 | io_u->flags &= ~IO_U_F_FREE_DEF; |
696 | io_u->flags |= IO_U_F_FREE; | |
697 | ||
0c41214f RR |
698 | if (io_u->flags & IO_U_F_IN_CUR_DEPTH) |
699 | td->cur_depth--; | |
2ae0b204 | 700 | io_u_qpush(&td->io_u_freelist, io_u); |
e8462bd8 JA |
701 | td_io_u_unlock(td); |
702 | td_io_u_free_notify(td); | |
10ba535a JA |
703 | } |
704 | ||
f2bba182 RR |
705 | void clear_io_u(struct thread_data *td, struct io_u *io_u) |
706 | { | |
707 | io_u->flags &= ~IO_U_F_FLIGHT; | |
708 | put_io_u(td, io_u); | |
709 | } | |
710 | ||
755200a3 JA |
711 | void requeue_io_u(struct thread_data *td, struct io_u **io_u) |
712 | { | |
713 | struct io_u *__io_u = *io_u; | |
bcd5abfa | 714 | enum fio_ddir ddir = acct_ddir(__io_u); |
755200a3 | 715 | |
465221b0 JA |
716 | dprint(FD_IO, "requeue %p\n", __io_u); |
717 | ||
e8462bd8 JA |
718 | td_io_u_lock(td); |
719 | ||
4d2e0f49 | 720 | __io_u->flags |= IO_U_F_FREE; |
bcd5abfa JA |
721 | if ((__io_u->flags & IO_U_F_FLIGHT) && ddir_rw(ddir)) |
722 | td->io_issues[ddir]--; | |
5ec10eaa | 723 | |
4d2e0f49 | 724 | __io_u->flags &= ~IO_U_F_FLIGHT; |
0c41214f RR |
725 | if (__io_u->flags & IO_U_F_IN_CUR_DEPTH) |
726 | td->cur_depth--; | |
2ae0b204 JA |
727 | |
728 | io_u_rpush(&td->io_u_requeues, __io_u); | |
e8462bd8 | 729 | td_io_u_unlock(td); |
755200a3 JA |
730 | *io_u = NULL; |
731 | } | |
732 | ||
9bf2061e | 733 | static int fill_io_u(struct thread_data *td, struct io_u *io_u) |
10ba535a | 734 | { |
6aca9b3d JA |
735 | unsigned int is_random; |
736 | ||
b4c5e1ac JA |
737 | if (td->io_ops->flags & FIO_NOIO) |
738 | goto out; | |
739 | ||
1ef2b6be | 740 | set_rw_ddir(td, io_u); |
5f9099ea | 741 | |
87dc1ab1 | 742 | /* |
ff58fced | 743 | * fsync() or fdatasync() or trim etc, we are done |
87dc1ab1 | 744 | */ |
ff58fced | 745 | if (!ddir_rw(io_u->ddir)) |
c38e9468 | 746 | goto out; |
a00735e6 | 747 | |
48f5abd3 JA |
748 | /* |
749 | * See if it's time to switch to a new zone | |
750 | */ | |
13af05ae | 751 | if (td->zone_bytes >= td->o.zone_size && td->o.zone_skip) { |
48f5abd3 | 752 | td->zone_bytes = 0; |
ed335855 SN |
753 | io_u->file->file_offset += td->o.zone_range + td->o.zone_skip; |
754 | io_u->file->last_pos = io_u->file->file_offset; | |
48f5abd3 JA |
755 | td->io_skip_bytes += td->o.zone_skip; |
756 | } | |
757 | ||
10ba535a | 758 | /* |
c685b5b2 JA |
759 | * No log, let the seq/rand engine retrieve the next buflen and |
760 | * position. | |
10ba535a | 761 | */ |
6aca9b3d | 762 | if (get_next_offset(td, io_u, &is_random)) { |
2ba1c290 | 763 | dprint(FD_IO, "io_u %p, failed getting offset\n", io_u); |
bca4ed4d | 764 | return 1; |
2ba1c290 | 765 | } |
10ba535a | 766 | |
6aca9b3d | 767 | io_u->buflen = get_next_buflen(td, io_u, is_random); |
2ba1c290 JA |
768 | if (!io_u->buflen) { |
769 | dprint(FD_IO, "io_u %p, failed getting buflen\n", io_u); | |
bca4ed4d | 770 | return 1; |
2ba1c290 | 771 | } |
bca4ed4d | 772 | |
2ba1c290 JA |
773 | if (io_u->offset + io_u->buflen > io_u->file->real_file_size) { |
774 | dprint(FD_IO, "io_u %p, offset too large\n", io_u); | |
4b91ee8f JA |
775 | dprint(FD_IO, " off=%llu/%lu > %llu\n", |
776 | (unsigned long long) io_u->offset, io_u->buflen, | |
777 | (unsigned long long) io_u->file->real_file_size); | |
6a5e6884 | 778 | return 1; |
2ba1c290 | 779 | } |
6a5e6884 | 780 | |
bca4ed4d JA |
781 | /* |
782 | * mark entry before potentially trimming io_u | |
783 | */ | |
303032ae | 784 | if (td_random(td) && file_randommap(td, io_u->file)) |
9bf2061e | 785 | mark_random_map(td, io_u); |
bca4ed4d | 786 | |
c38e9468 | 787 | out: |
2ba1c290 | 788 | dprint_io_u(io_u, "fill_io_u"); |
d9d91e39 | 789 | td->zone_bytes += io_u->buflen; |
bca4ed4d | 790 | return 0; |
10ba535a JA |
791 | } |
792 | ||
838bc709 JA |
793 | static void __io_u_mark_map(unsigned int *map, unsigned int nr) |
794 | { | |
2b13e716 | 795 | int idx = 0; |
838bc709 JA |
796 | |
797 | switch (nr) { | |
798 | default: | |
2b13e716 | 799 | idx = 6; |
838bc709 JA |
800 | break; |
801 | case 33 ... 64: | |
2b13e716 | 802 | idx = 5; |
838bc709 JA |
803 | break; |
804 | case 17 ... 32: | |
2b13e716 | 805 | idx = 4; |
838bc709 JA |
806 | break; |
807 | case 9 ... 16: | |
2b13e716 | 808 | idx = 3; |
838bc709 JA |
809 | break; |
810 | case 5 ... 8: | |
2b13e716 | 811 | idx = 2; |
838bc709 JA |
812 | break; |
813 | case 1 ... 4: | |
2b13e716 | 814 | idx = 1; |
838bc709 JA |
815 | case 0: |
816 | break; | |
817 | } | |
818 | ||
2b13e716 | 819 | map[idx]++; |
838bc709 JA |
820 | } |
821 | ||
822 | void io_u_mark_submit(struct thread_data *td, unsigned int nr) | |
823 | { | |
824 | __io_u_mark_map(td->ts.io_u_submit, nr); | |
825 | td->ts.total_submit++; | |
826 | } | |
827 | ||
828 | void io_u_mark_complete(struct thread_data *td, unsigned int nr) | |
829 | { | |
830 | __io_u_mark_map(td->ts.io_u_complete, nr); | |
831 | td->ts.total_complete++; | |
832 | } | |
833 | ||
d8005759 | 834 | void io_u_mark_depth(struct thread_data *td, unsigned int nr) |
71619dc2 | 835 | { |
2b13e716 | 836 | int idx = 0; |
71619dc2 JA |
837 | |
838 | switch (td->cur_depth) { | |
839 | default: | |
2b13e716 | 840 | idx = 6; |
a783e61a | 841 | break; |
71619dc2 | 842 | case 32 ... 63: |
2b13e716 | 843 | idx = 5; |
a783e61a | 844 | break; |
71619dc2 | 845 | case 16 ... 31: |
2b13e716 | 846 | idx = 4; |
a783e61a | 847 | break; |
71619dc2 | 848 | case 8 ... 15: |
2b13e716 | 849 | idx = 3; |
a783e61a | 850 | break; |
71619dc2 | 851 | case 4 ... 7: |
2b13e716 | 852 | idx = 2; |
a783e61a | 853 | break; |
71619dc2 | 854 | case 2 ... 3: |
2b13e716 | 855 | idx = 1; |
71619dc2 JA |
856 | case 1: |
857 | break; | |
858 | } | |
859 | ||
2b13e716 | 860 | td->ts.io_u_map[idx] += nr; |
71619dc2 JA |
861 | } |
862 | ||
04a0feae JA |
863 | static void io_u_mark_lat_usec(struct thread_data *td, unsigned long usec) |
864 | { | |
2b13e716 | 865 | int idx = 0; |
04a0feae JA |
866 | |
867 | assert(usec < 1000); | |
868 | ||
869 | switch (usec) { | |
870 | case 750 ... 999: | |
2b13e716 | 871 | idx = 9; |
04a0feae JA |
872 | break; |
873 | case 500 ... 749: | |
2b13e716 | 874 | idx = 8; |
04a0feae JA |
875 | break; |
876 | case 250 ... 499: | |
2b13e716 | 877 | idx = 7; |
04a0feae JA |
878 | break; |
879 | case 100 ... 249: | |
2b13e716 | 880 | idx = 6; |
04a0feae JA |
881 | break; |
882 | case 50 ... 99: | |
2b13e716 | 883 | idx = 5; |
04a0feae JA |
884 | break; |
885 | case 20 ... 49: | |
2b13e716 | 886 | idx = 4; |
04a0feae JA |
887 | break; |
888 | case 10 ... 19: | |
2b13e716 | 889 | idx = 3; |
04a0feae JA |
890 | break; |
891 | case 4 ... 9: | |
2b13e716 | 892 | idx = 2; |
04a0feae JA |
893 | break; |
894 | case 2 ... 3: | |
2b13e716 | 895 | idx = 1; |
04a0feae JA |
896 | case 0 ... 1: |
897 | break; | |
898 | } | |
899 | ||
2b13e716 JA |
900 | assert(idx < FIO_IO_U_LAT_U_NR); |
901 | td->ts.io_u_lat_u[idx]++; | |
04a0feae JA |
902 | } |
903 | ||
904 | static void io_u_mark_lat_msec(struct thread_data *td, unsigned long msec) | |
ec118304 | 905 | { |
2b13e716 | 906 | int idx = 0; |
ec118304 JA |
907 | |
908 | switch (msec) { | |
909 | default: | |
2b13e716 | 910 | idx = 11; |
04a0feae | 911 | break; |
8abdce66 | 912 | case 1000 ... 1999: |
2b13e716 | 913 | idx = 10; |
04a0feae | 914 | break; |
8abdce66 | 915 | case 750 ... 999: |
2b13e716 | 916 | idx = 9; |
04a0feae | 917 | break; |
8abdce66 | 918 | case 500 ... 749: |
2b13e716 | 919 | idx = 8; |
04a0feae | 920 | break; |
8abdce66 | 921 | case 250 ... 499: |
2b13e716 | 922 | idx = 7; |
04a0feae | 923 | break; |
8abdce66 | 924 | case 100 ... 249: |
2b13e716 | 925 | idx = 6; |
04a0feae | 926 | break; |
8abdce66 | 927 | case 50 ... 99: |
2b13e716 | 928 | idx = 5; |
04a0feae | 929 | break; |
8abdce66 | 930 | case 20 ... 49: |
2b13e716 | 931 | idx = 4; |
04a0feae | 932 | break; |
8abdce66 | 933 | case 10 ... 19: |
2b13e716 | 934 | idx = 3; |
04a0feae | 935 | break; |
8abdce66 | 936 | case 4 ... 9: |
2b13e716 | 937 | idx = 2; |
04a0feae | 938 | break; |
ec118304 | 939 | case 2 ... 3: |
2b13e716 | 940 | idx = 1; |
ec118304 JA |
941 | case 0 ... 1: |
942 | break; | |
943 | } | |
944 | ||
2b13e716 JA |
945 | assert(idx < FIO_IO_U_LAT_M_NR); |
946 | td->ts.io_u_lat_m[idx]++; | |
04a0feae JA |
947 | } |
948 | ||
949 | static void io_u_mark_latency(struct thread_data *td, unsigned long usec) | |
950 | { | |
951 | if (usec < 1000) | |
952 | io_u_mark_lat_usec(td, usec); | |
953 | else | |
954 | io_u_mark_lat_msec(td, usec / 1000); | |
ec118304 JA |
955 | } |
956 | ||
0aabe160 JA |
957 | /* |
958 | * Get next file to service by choosing one at random | |
959 | */ | |
2cc52930 JA |
960 | static struct fio_file *get_next_file_rand(struct thread_data *td, |
961 | enum fio_file_flags goodf, | |
d6aed795 | 962 | enum fio_file_flags badf) |
0aabe160 | 963 | { |
0aabe160 | 964 | struct fio_file *f; |
1c178180 | 965 | int fno; |
0aabe160 JA |
966 | |
967 | do { | |
87b10676 | 968 | int opened = 0; |
1294c3ec | 969 | unsigned long r; |
4c07ad86 JA |
970 | |
971 | if (td->o.use_os_rand) { | |
972 | r = os_random_long(&td->next_file_state); | |
973 | fno = (unsigned int) ((double) td->o.nr_files | |
974 | * (r / (OS_RAND_MAX + 1.0))); | |
975 | } else { | |
976 | r = __rand(&td->__next_file_state); | |
977 | fno = (unsigned int) ((double) td->o.nr_files | |
978 | * (r / (FRAND_MAX + 1.0))); | |
979 | } | |
7c83c089 | 980 | |
126d65c6 | 981 | f = td->files[fno]; |
d6aed795 | 982 | if (fio_file_done(f)) |
059e63c0 | 983 | continue; |
1c178180 | 984 | |
d6aed795 | 985 | if (!fio_file_open(f)) { |
87b10676 JA |
986 | int err; |
987 | ||
988 | err = td_io_open_file(td, f); | |
989 | if (err) | |
990 | continue; | |
991 | opened = 1; | |
992 | } | |
993 | ||
2ba1c290 JA |
994 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) { |
995 | dprint(FD_FILE, "get_next_file_rand: %p\n", f); | |
0aabe160 | 996 | return f; |
2ba1c290 | 997 | } |
87b10676 JA |
998 | if (opened) |
999 | td_io_close_file(td, f); | |
0aabe160 JA |
1000 | } while (1); |
1001 | } | |
1002 | ||
1003 | /* | |
1004 | * Get next file to service by doing round robin between all available ones | |
1005 | */ | |
1c178180 JA |
1006 | static struct fio_file *get_next_file_rr(struct thread_data *td, int goodf, |
1007 | int badf) | |
3d7c391d JA |
1008 | { |
1009 | unsigned int old_next_file = td->next_file; | |
1010 | struct fio_file *f; | |
1011 | ||
1012 | do { | |
87b10676 JA |
1013 | int opened = 0; |
1014 | ||
126d65c6 | 1015 | f = td->files[td->next_file]; |
3d7c391d JA |
1016 | |
1017 | td->next_file++; | |
2dc1bbeb | 1018 | if (td->next_file >= td->o.nr_files) |
3d7c391d JA |
1019 | td->next_file = 0; |
1020 | ||
87b10676 | 1021 | dprint(FD_FILE, "trying file %s %x\n", f->file_name, f->flags); |
d6aed795 | 1022 | if (fio_file_done(f)) { |
d5ed68ea | 1023 | f = NULL; |
059e63c0 | 1024 | continue; |
d5ed68ea | 1025 | } |
059e63c0 | 1026 | |
d6aed795 | 1027 | if (!fio_file_open(f)) { |
87b10676 JA |
1028 | int err; |
1029 | ||
1030 | err = td_io_open_file(td, f); | |
b5696bfc JA |
1031 | if (err) { |
1032 | dprint(FD_FILE, "error %d on open of %s\n", | |
1033 | err, f->file_name); | |
87c27b45 | 1034 | f = NULL; |
87b10676 | 1035 | continue; |
b5696bfc | 1036 | } |
87b10676 JA |
1037 | opened = 1; |
1038 | } | |
1039 | ||
0b9d69ec JA |
1040 | dprint(FD_FILE, "goodf=%x, badf=%x, ff=%x\n", goodf, badf, |
1041 | f->flags); | |
1c178180 | 1042 | if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) |
3d7c391d JA |
1043 | break; |
1044 | ||
87b10676 JA |
1045 | if (opened) |
1046 | td_io_close_file(td, f); | |
1047 | ||
3d7c391d JA |
1048 | f = NULL; |
1049 | } while (td->next_file != old_next_file); | |
1050 | ||
2ba1c290 | 1051 | dprint(FD_FILE, "get_next_file_rr: %p\n", f); |
3d7c391d JA |
1052 | return f; |
1053 | } | |
1054 | ||
7eb36574 | 1055 | static struct fio_file *__get_next_file(struct thread_data *td) |
bdb4e2e9 | 1056 | { |
1907dbc6 JA |
1057 | struct fio_file *f; |
1058 | ||
2dc1bbeb | 1059 | assert(td->o.nr_files <= td->files_index); |
1c178180 | 1060 | |
b5696bfc | 1061 | if (td->nr_done_files >= td->o.nr_files) { |
5ec10eaa JA |
1062 | dprint(FD_FILE, "get_next_file: nr_open=%d, nr_done=%d," |
1063 | " nr_files=%d\n", td->nr_open_files, | |
1064 | td->nr_done_files, | |
1065 | td->o.nr_files); | |
bdb4e2e9 | 1066 | return NULL; |
2ba1c290 | 1067 | } |
bdb4e2e9 | 1068 | |
1907dbc6 | 1069 | f = td->file_service_file; |
d6aed795 | 1070 | if (f && fio_file_open(f) && !fio_file_closing(f)) { |
a086c257 JA |
1071 | if (td->o.file_service_type == FIO_FSERVICE_SEQ) |
1072 | goto out; | |
1073 | if (td->file_service_left--) | |
1074 | goto out; | |
1075 | } | |
1907dbc6 | 1076 | |
a086c257 JA |
1077 | if (td->o.file_service_type == FIO_FSERVICE_RR || |
1078 | td->o.file_service_type == FIO_FSERVICE_SEQ) | |
d6aed795 | 1079 | f = get_next_file_rr(td, FIO_FILE_open, FIO_FILE_closing); |
bdb4e2e9 | 1080 | else |
d6aed795 | 1081 | f = get_next_file_rand(td, FIO_FILE_open, FIO_FILE_closing); |
1907dbc6 JA |
1082 | |
1083 | td->file_service_file = f; | |
1084 | td->file_service_left = td->file_service_nr - 1; | |
2ba1c290 | 1085 | out: |
683023e8 | 1086 | dprint(FD_FILE, "get_next_file: %p [%s]\n", f, f->file_name); |
1907dbc6 | 1087 | return f; |
bdb4e2e9 JA |
1088 | } |
1089 | ||
7eb36574 JA |
1090 | static struct fio_file *get_next_file(struct thread_data *td) |
1091 | { | |
d72be545 JA |
1092 | if (!(td->flags & TD_F_PROFILE_OPS)) { |
1093 | struct prof_io_ops *ops = &td->prof_io_ops; | |
7eb36574 | 1094 | |
d72be545 JA |
1095 | if (ops->get_next_file) |
1096 | return ops->get_next_file(td); | |
1097 | } | |
7eb36574 JA |
1098 | |
1099 | return __get_next_file(td); | |
1100 | } | |
1101 | ||
429f6675 JA |
1102 | static int set_io_u_file(struct thread_data *td, struct io_u *io_u) |
1103 | { | |
1104 | struct fio_file *f; | |
1105 | ||
1106 | do { | |
1107 | f = get_next_file(td); | |
1108 | if (!f) | |
1109 | return 1; | |
1110 | ||
429f6675 JA |
1111 | io_u->file = f; |
1112 | get_file(f); | |
1113 | ||
1114 | if (!fill_io_u(td, io_u)) | |
1115 | break; | |
1116 | ||
b5696bfc | 1117 | put_file_log(td, f); |
429f6675 | 1118 | td_io_close_file(td, f); |
b5696bfc | 1119 | io_u->file = NULL; |
d6aed795 | 1120 | fio_file_set_done(f); |
429f6675 | 1121 | td->nr_done_files++; |
0b9d69ec JA |
1122 | dprint(FD_FILE, "%s: is done (%d of %d)\n", f->file_name, |
1123 | td->nr_done_files, td->o.nr_files); | |
429f6675 JA |
1124 | } while (1); |
1125 | ||
1126 | return 0; | |
1127 | } | |
1128 | ||
3e260a46 JA |
1129 | static void lat_fatal(struct thread_data *td, struct io_completion_data *icd, |
1130 | unsigned long tusec, unsigned long max_usec) | |
1131 | { | |
1132 | if (!td->error) | |
1133 | log_err("fio: latency of %lu usec exceeds specified max (%lu usec)\n", tusec, max_usec); | |
1134 | td_verror(td, ETIMEDOUT, "max latency exceeded"); | |
1135 | icd->error = ETIMEDOUT; | |
1136 | } | |
1137 | ||
1138 | static void lat_new_cycle(struct thread_data *td) | |
1139 | { | |
1140 | fio_gettime(&td->latency_ts, NULL); | |
1141 | td->latency_ios = ddir_rw_sum(td->io_blocks); | |
1142 | td->latency_failed = 0; | |
1143 | } | |
1144 | ||
1145 | /* | |
1146 | * We had an IO outside the latency target. Reduce the queue depth. If we | |
1147 | * are at QD=1, then it's time to give up. | |
1148 | */ | |
1149 | static int __lat_target_failed(struct thread_data *td) | |
1150 | { | |
1151 | if (td->latency_qd == 1) | |
1152 | return 1; | |
1153 | ||
1154 | td->latency_qd_high = td->latency_qd; | |
1155 | td->latency_qd = (td->latency_qd + td->latency_qd_low) / 2; | |
1156 | ||
1157 | dprint(FD_RATE, "Ramped down: %d %d %d\n", td->latency_qd_low, td->latency_qd, td->latency_qd_high); | |
1158 | ||
1159 | /* | |
1160 | * When we ramp QD down, quiesce existing IO to prevent | |
1161 | * a storm of ramp downs due to pending higher depth. | |
1162 | */ | |
1163 | io_u_quiesce(td); | |
1164 | lat_new_cycle(td); | |
1165 | return 0; | |
1166 | } | |
1167 | ||
1168 | static int lat_target_failed(struct thread_data *td) | |
1169 | { | |
1170 | if (td->o.latency_percentile.u.f == 100.0) | |
1171 | return __lat_target_failed(td); | |
1172 | ||
1173 | td->latency_failed++; | |
1174 | return 0; | |
1175 | } | |
1176 | ||
1177 | void lat_target_init(struct thread_data *td) | |
1178 | { | |
1179 | if (td->o.latency_target) { | |
1180 | dprint(FD_RATE, "Latency target=%llu\n", td->o.latency_target); | |
1181 | fio_gettime(&td->latency_ts, NULL); | |
1182 | td->latency_qd = 1; | |
1183 | td->latency_qd_high = td->o.iodepth; | |
1184 | td->latency_qd_low = 1; | |
1185 | td->latency_ios = ddir_rw_sum(td->io_blocks); | |
1186 | } else | |
1187 | td->latency_qd = td->o.iodepth; | |
1188 | } | |
1189 | ||
1190 | static void lat_target_success(struct thread_data *td) | |
1191 | { | |
1192 | const unsigned int qd = td->latency_qd; | |
1193 | ||
1194 | td->latency_qd_low = td->latency_qd; | |
1195 | ||
1196 | /* | |
1197 | * If we haven't failed yet, we double up to a failing value instead | |
1198 | * of bisecting from highest possible queue depth. If we have set | |
1199 | * a limit other than td->o.iodepth, bisect between that. | |
1200 | */ | |
1201 | if (td->latency_qd_high != td->o.iodepth) | |
1202 | td->latency_qd = (td->latency_qd + td->latency_qd_high) / 2; | |
1203 | else | |
1204 | td->latency_qd *= 2; | |
1205 | ||
1206 | if (td->latency_qd > td->o.iodepth) | |
1207 | td->latency_qd = td->o.iodepth; | |
1208 | ||
1209 | dprint(FD_RATE, "Ramped up: %d %d %d\n", td->latency_qd_low, td->latency_qd, td->latency_qd_high); | |
1210 | /* | |
1211 | * Same as last one, we are done | |
1212 | */ | |
1213 | if (td->latency_qd == qd) | |
1214 | td->done = 1; | |
1215 | ||
1216 | lat_new_cycle(td); | |
1217 | } | |
1218 | ||
1219 | /* | |
1220 | * Check if we can bump the queue depth | |
1221 | */ | |
1222 | void lat_target_check(struct thread_data *td) | |
1223 | { | |
1224 | uint64_t usec_window; | |
1225 | uint64_t ios; | |
1226 | double success_ios; | |
1227 | ||
1228 | usec_window = utime_since_now(&td->latency_ts); | |
1229 | if (usec_window < td->o.latency_window) | |
1230 | return; | |
1231 | ||
1232 | ios = ddir_rw_sum(td->io_blocks) - td->latency_ios; | |
1233 | success_ios = (double) (ios - td->latency_failed) / (double) ios; | |
1234 | success_ios *= 100.0; | |
1235 | ||
1236 | dprint(FD_RATE, "Success rate: %.2f%% (target %.2f%%)\n", success_ios, td->o.latency_percentile.u.f); | |
1237 | ||
1238 | if (success_ios >= td->o.latency_percentile.u.f) | |
1239 | lat_target_success(td); | |
1240 | else | |
1241 | __lat_target_failed(td); | |
1242 | } | |
1243 | ||
1244 | /* | |
1245 | * If latency target is enabled, we might be ramping up or down and not | |
1246 | * using the full queue depth available. | |
1247 | */ | |
1248 | int queue_full(struct thread_data *td) | |
1249 | { | |
1250 | const int qempty = io_u_qempty(&td->io_u_freelist); | |
1251 | ||
1252 | if (qempty) | |
1253 | return 1; | |
1254 | if (!td->o.latency_target) | |
1255 | return 0; | |
1256 | ||
1257 | return td->cur_depth >= td->latency_qd; | |
1258 | } | |
429f6675 | 1259 | |
10ba535a JA |
1260 | struct io_u *__get_io_u(struct thread_data *td) |
1261 | { | |
2ae0b204 | 1262 | struct io_u *io_u; |
10ba535a | 1263 | |
e8462bd8 JA |
1264 | td_io_u_lock(td); |
1265 | ||
1266 | again: | |
2ae0b204 JA |
1267 | if (!io_u_rempty(&td->io_u_requeues)) |
1268 | io_u = io_u_rpop(&td->io_u_requeues); | |
3e260a46 | 1269 | else if (!queue_full(td)) { |
2ae0b204 | 1270 | io_u = io_u_qpop(&td->io_u_freelist); |
10ba535a | 1271 | |
6040dabc | 1272 | io_u->buflen = 0; |
10ba535a | 1273 | io_u->resid = 0; |
755200a3 | 1274 | io_u->file = NULL; |
d7762cf8 | 1275 | io_u->end_io = NULL; |
755200a3 JA |
1276 | } |
1277 | ||
1278 | if (io_u) { | |
0c6e7517 | 1279 | assert(io_u->flags & IO_U_F_FREE); |
2ecc1b57 | 1280 | io_u->flags &= ~(IO_U_F_FREE | IO_U_F_FREE_DEF); |
1ef2b6be | 1281 | io_u->flags &= ~(IO_U_F_TRIMMED | IO_U_F_BARRIER); |
82af2a7c | 1282 | io_u->flags &= ~IO_U_F_VER_LIST; |
0c6e7517 | 1283 | |
755200a3 | 1284 | io_u->error = 0; |
bcd5abfa | 1285 | io_u->acct_ddir = -1; |
10ba535a | 1286 | td->cur_depth++; |
0c41214f | 1287 | io_u->flags |= IO_U_F_IN_CUR_DEPTH; |
1dec3e07 JA |
1288 | } else if (td->o.verify_async) { |
1289 | /* | |
1290 | * We ran out, wait for async verify threads to finish and | |
1291 | * return one | |
1292 | */ | |
1293 | pthread_cond_wait(&td->free_cond, &td->io_u_lock); | |
1294 | goto again; | |
10ba535a JA |
1295 | } |
1296 | ||
e8462bd8 | 1297 | td_io_u_unlock(td); |
10ba535a JA |
1298 | return io_u; |
1299 | } | |
1300 | ||
0d29de83 | 1301 | static int check_get_trim(struct thread_data *td, struct io_u *io_u) |
10ba535a | 1302 | { |
d72be545 JA |
1303 | if (!(td->flags & TD_F_TRIM_BACKLOG)) |
1304 | return 0; | |
1305 | ||
1306 | if (td->trim_entries) { | |
0d29de83 | 1307 | int get_trim = 0; |
10ba535a | 1308 | |
0d29de83 JA |
1309 | if (td->trim_batch) { |
1310 | td->trim_batch--; | |
1311 | get_trim = 1; | |
1312 | } else if (!(td->io_hist_len % td->o.trim_backlog) && | |
1313 | td->last_ddir != DDIR_READ) { | |
1314 | td->trim_batch = td->o.trim_batch; | |
1315 | if (!td->trim_batch) | |
1316 | td->trim_batch = td->o.trim_backlog; | |
1317 | get_trim = 1; | |
1318 | } | |
1319 | ||
1320 | if (get_trim && !get_next_trim(td, io_u)) | |
1321 | return 1; | |
2ba1c290 | 1322 | } |
10ba535a | 1323 | |
0d29de83 JA |
1324 | return 0; |
1325 | } | |
1326 | ||
1327 | static int check_get_verify(struct thread_data *td, struct io_u *io_u) | |
1328 | { | |
d72be545 JA |
1329 | if (!(td->flags & TD_F_VER_BACKLOG)) |
1330 | return 0; | |
1331 | ||
1332 | if (td->io_hist_len) { | |
9e144189 JA |
1333 | int get_verify = 0; |
1334 | ||
d1ece0c7 | 1335 | if (td->verify_batch) |
9e144189 | 1336 | get_verify = 1; |
d1ece0c7 | 1337 | else if (!(td->io_hist_len % td->o.verify_backlog) && |
9e144189 JA |
1338 | td->last_ddir != DDIR_READ) { |
1339 | td->verify_batch = td->o.verify_batch; | |
f8a75c99 JA |
1340 | if (!td->verify_batch) |
1341 | td->verify_batch = td->o.verify_backlog; | |
9e144189 JA |
1342 | get_verify = 1; |
1343 | } | |
1344 | ||
d1ece0c7 JA |
1345 | if (get_verify && !get_next_verify(td, io_u)) { |
1346 | td->verify_batch--; | |
0d29de83 | 1347 | return 1; |
d1ece0c7 | 1348 | } |
9e144189 JA |
1349 | } |
1350 | ||
0d29de83 JA |
1351 | return 0; |
1352 | } | |
1353 | ||
de789769 JA |
1354 | /* |
1355 | * Fill offset and start time into the buffer content, to prevent too | |
23f394d5 JA |
1356 | * easy compressible data for simple de-dupe attempts. Do this for every |
1357 | * 512b block in the range, since that should be the smallest block size | |
1358 | * we can expect from a device. | |
de789769 JA |
1359 | */ |
1360 | static void small_content_scramble(struct io_u *io_u) | |
1361 | { | |
23f394d5 | 1362 | unsigned int i, nr_blocks = io_u->buflen / 512; |
1ae83d45 | 1363 | uint64_t boffset; |
23f394d5 JA |
1364 | unsigned int offset; |
1365 | void *p, *end; | |
de789769 | 1366 | |
23f394d5 JA |
1367 | if (!nr_blocks) |
1368 | return; | |
1369 | ||
1370 | p = io_u->xfer_buf; | |
fba76ee8 | 1371 | boffset = io_u->offset; |
81f0366c | 1372 | io_u->buf_filled_len = 0; |
fad82f76 | 1373 | |
23f394d5 JA |
1374 | for (i = 0; i < nr_blocks; i++) { |
1375 | /* | |
1376 | * Fill the byte offset into a "random" start offset of | |
1377 | * the buffer, given by the product of the usec time | |
1378 | * and the actual offset. | |
1379 | */ | |
fad82f76 | 1380 | offset = (io_u->start_time.tv_usec ^ boffset) & 511; |
1ae83d45 JA |
1381 | offset &= ~(sizeof(uint64_t) - 1); |
1382 | if (offset >= 512 - sizeof(uint64_t)) | |
1383 | offset -= sizeof(uint64_t); | |
fba76ee8 | 1384 | memcpy(p + offset, &boffset, sizeof(boffset)); |
23f394d5 JA |
1385 | |
1386 | end = p + 512 - sizeof(io_u->start_time); | |
1387 | memcpy(end, &io_u->start_time, sizeof(io_u->start_time)); | |
1388 | p += 512; | |
fad82f76 | 1389 | boffset += 512; |
23f394d5 | 1390 | } |
de789769 JA |
1391 | } |
1392 | ||
0d29de83 JA |
1393 | /* |
1394 | * Return an io_u to be processed. Gets a buflen and offset, sets direction, | |
1395 | * etc. The returned io_u is fully ready to be prepped and submitted. | |
1396 | */ | |
1397 | struct io_u *get_io_u(struct thread_data *td) | |
1398 | { | |
1399 | struct fio_file *f; | |
1400 | struct io_u *io_u; | |
de789769 | 1401 | int do_scramble = 0; |
0d29de83 JA |
1402 | |
1403 | io_u = __get_io_u(td); | |
1404 | if (!io_u) { | |
1405 | dprint(FD_IO, "__get_io_u failed\n"); | |
1406 | return NULL; | |
1407 | } | |
1408 | ||
1409 | if (check_get_verify(td, io_u)) | |
1410 | goto out; | |
1411 | if (check_get_trim(td, io_u)) | |
1412 | goto out; | |
1413 | ||
755200a3 JA |
1414 | /* |
1415 | * from a requeue, io_u already setup | |
1416 | */ | |
1417 | if (io_u->file) | |
77f392bf | 1418 | goto out; |
755200a3 | 1419 | |
429f6675 JA |
1420 | /* |
1421 | * If using an iolog, grab next piece if any available. | |
1422 | */ | |
d72be545 | 1423 | if (td->flags & TD_F_READ_IOLOG) { |
429f6675 JA |
1424 | if (read_iolog_get(td, io_u)) |
1425 | goto err_put; | |
2ba1c290 JA |
1426 | } else if (set_io_u_file(td, io_u)) { |
1427 | dprint(FD_IO, "io_u %p, setting file failed\n", io_u); | |
429f6675 | 1428 | goto err_put; |
2ba1c290 | 1429 | } |
5ec10eaa | 1430 | |
429f6675 | 1431 | f = io_u->file; |
d6aed795 | 1432 | assert(fio_file_open(f)); |
97af62ce | 1433 | |
ff58fced | 1434 | if (ddir_rw(io_u->ddir)) { |
d0656a93 | 1435 | if (!io_u->buflen && !(td->io_ops->flags & FIO_NOIO)) { |
2ba1c290 | 1436 | dprint(FD_IO, "get_io_u: zero buflen on %p\n", io_u); |
429f6675 | 1437 | goto err_put; |
2ba1c290 | 1438 | } |
10ba535a | 1439 | |
38dad62d | 1440 | f->last_start = io_u->offset; |
36167d82 | 1441 | f->last_pos = io_u->offset + io_u->buflen; |
10ba535a | 1442 | |
fd68418e | 1443 | if (io_u->ddir == DDIR_WRITE) { |
d72be545 | 1444 | if (td->flags & TD_F_REFILL_BUFFERS) { |
9c42684e JA |
1445 | io_u_fill_buffer(td, io_u, |
1446 | io_u->xfer_buflen, io_u->xfer_buflen); | |
d72be545 | 1447 | } else if (td->flags & TD_F_SCRAMBLE_BUFFERS) |
fd68418e | 1448 | do_scramble = 1; |
d72be545 | 1449 | if (td->flags & TD_F_VER_NONE) { |
629f1d71 JA |
1450 | populate_verify_io_u(td, io_u); |
1451 | do_scramble = 0; | |
1452 | } | |
fd68418e | 1453 | } else if (io_u->ddir == DDIR_READ) { |
cbe8d756 RR |
1454 | /* |
1455 | * Reset the buf_filled parameters so next time if the | |
1456 | * buffer is used for writes it is refilled. | |
1457 | */ | |
cbe8d756 RR |
1458 | io_u->buf_filled_len = 0; |
1459 | } | |
87dc1ab1 | 1460 | } |
10ba535a | 1461 | |
165faf16 JA |
1462 | /* |
1463 | * Set io data pointers. | |
1464 | */ | |
cec6b55d JA |
1465 | io_u->xfer_buf = io_u->buf; |
1466 | io_u->xfer_buflen = io_u->buflen; | |
5973cafb | 1467 | |
6ac7a331 | 1468 | out: |
0d29de83 | 1469 | assert(io_u->file); |
429f6675 | 1470 | if (!td_io_prep(td, io_u)) { |
993bf48b JA |
1471 | if (!td->o.disable_slat) |
1472 | fio_gettime(&io_u->start_time, NULL); | |
de789769 JA |
1473 | if (do_scramble) |
1474 | small_content_scramble(io_u); | |
429f6675 | 1475 | return io_u; |
36167d82 | 1476 | } |
429f6675 | 1477 | err_put: |
2ba1c290 | 1478 | dprint(FD_IO, "get_io_u failed\n"); |
429f6675 JA |
1479 | put_io_u(td, io_u); |
1480 | return NULL; | |
10ba535a JA |
1481 | } |
1482 | ||
5451792e JA |
1483 | void io_u_log_error(struct thread_data *td, struct io_u *io_u) |
1484 | { | |
8b28bd41 | 1485 | enum error_type_bit eb = td_error_type(io_u->ddir, io_u->error); |
825f818e JA |
1486 | const char *msg[] = { "read", "write", "sync", "datasync", |
1487 | "sync_file_range", "wait", "trim" }; | |
1488 | ||
8b28bd41 DM |
1489 | if (td_non_fatal_error(td, eb, io_u->error) && !td->o.error_dump) |
1490 | return; | |
5451792e JA |
1491 | |
1492 | log_err("fio: io_u error"); | |
1493 | ||
1494 | if (io_u->file) | |
1495 | log_err(" on file %s", io_u->file->file_name); | |
1496 | ||
1497 | log_err(": %s\n", strerror(io_u->error)); | |
1498 | ||
5ec10eaa JA |
1499 | log_err(" %s offset=%llu, buflen=%lu\n", msg[io_u->ddir], |
1500 | io_u->offset, io_u->xfer_buflen); | |
5451792e JA |
1501 | |
1502 | if (!td->error) | |
1503 | td_verror(td, io_u->error, "io_u error"); | |
1504 | } | |
1505 | ||
c8eeb9df JA |
1506 | static void account_io_completion(struct thread_data *td, struct io_u *io_u, |
1507 | struct io_completion_data *icd, | |
1508 | const enum fio_ddir idx, unsigned int bytes) | |
1509 | { | |
24d23ca7 | 1510 | unsigned long lusec = 0; |
c8eeb9df | 1511 | |
c8eeb9df JA |
1512 | if (!td->o.disable_clat || !td->o.disable_bw) |
1513 | lusec = utime_since(&io_u->issue_time, &icd->time); | |
1514 | ||
1515 | if (!td->o.disable_lat) { | |
1516 | unsigned long tusec; | |
1517 | ||
1518 | tusec = utime_since(&io_u->start_time, &icd->time); | |
1519 | add_lat_sample(td, idx, tusec, bytes); | |
15501535 | 1520 | |
d4afedfd JA |
1521 | if (td->flags & TD_F_PROFILE_OPS) { |
1522 | struct prof_io_ops *ops = &td->prof_io_ops; | |
1523 | ||
1524 | if (ops->io_u_lat) | |
1525 | icd->error = ops->io_u_lat(td, tusec); | |
1526 | } | |
1527 | ||
3e260a46 JA |
1528 | if (td->o.max_latency && tusec > td->o.max_latency) |
1529 | lat_fatal(td, icd, tusec, td->o.max_latency); | |
1530 | if (td->o.latency_target && tusec > td->o.latency_target) { | |
1531 | if (lat_target_failed(td)) | |
1532 | lat_fatal(td, icd, tusec, td->o.latency_target); | |
15501535 | 1533 | } |
c8eeb9df JA |
1534 | } |
1535 | ||
1536 | if (!td->o.disable_clat) { | |
1537 | add_clat_sample(td, idx, lusec, bytes); | |
1538 | io_u_mark_latency(td, lusec); | |
1539 | } | |
1540 | ||
1541 | if (!td->o.disable_bw) | |
1542 | add_bw_sample(td, idx, bytes, &icd->time); | |
1543 | ||
9b7e600d | 1544 | add_iops_sample(td, idx, bytes, &icd->time); |
ddf24e42 JA |
1545 | |
1546 | if (td->o.number_ios && !--td->o.number_ios) | |
1547 | td->done = 1; | |
c8eeb9df JA |
1548 | } |
1549 | ||
1b8dbf25 SL |
1550 | static long long usec_for_io(struct thread_data *td, enum fio_ddir ddir) |
1551 | { | |
1ae83d45 JA |
1552 | uint64_t secs, remainder, bps, bytes; |
1553 | ||
1b8dbf25 SL |
1554 | bytes = td->this_io_bytes[ddir]; |
1555 | bps = td->rate_bps[ddir]; | |
1556 | secs = bytes / bps; | |
1557 | remainder = bytes % bps; | |
1558 | return remainder * 1000000 / bps + secs * 1000000; | |
1559 | } | |
1560 | ||
97601024 JA |
1561 | static void io_completed(struct thread_data *td, struct io_u *io_u, |
1562 | struct io_completion_data *icd) | |
10ba535a | 1563 | { |
44f29692 | 1564 | struct fio_file *f; |
10ba535a | 1565 | |
2ba1c290 JA |
1566 | dprint_io_u(io_u, "io complete"); |
1567 | ||
2ecc1b57 | 1568 | td_io_u_lock(td); |
0c6e7517 | 1569 | assert(io_u->flags & IO_U_F_FLIGHT); |
38dad62d | 1570 | io_u->flags &= ~(IO_U_F_FLIGHT | IO_U_F_BUSY_OK); |
2ecc1b57 | 1571 | td_io_u_unlock(td); |
0c6e7517 | 1572 | |
5f9099ea | 1573 | if (ddir_sync(io_u->ddir)) { |
87dc1ab1 | 1574 | td->last_was_sync = 1; |
44f29692 JA |
1575 | f = io_u->file; |
1576 | if (f) { | |
1577 | f->first_write = -1ULL; | |
1578 | f->last_write = -1ULL; | |
1579 | } | |
87dc1ab1 JA |
1580 | return; |
1581 | } | |
1582 | ||
1583 | td->last_was_sync = 0; | |
9e144189 | 1584 | td->last_ddir = io_u->ddir; |
87dc1ab1 | 1585 | |
ff58fced | 1586 | if (!io_u->error && ddir_rw(io_u->ddir)) { |
10ba535a | 1587 | unsigned int bytes = io_u->buflen - io_u->resid; |
1e97cce9 | 1588 | const enum fio_ddir idx = io_u->ddir; |
ba3e4e0c | 1589 | const enum fio_ddir odx = io_u->ddir ^ 1; |
b29ee5b3 | 1590 | int ret; |
10ba535a | 1591 | |
b29ee5b3 | 1592 | td->io_blocks[idx]++; |
c8eeb9df | 1593 | td->this_io_blocks[idx]++; |
b29ee5b3 | 1594 | td->io_bytes[idx] += bytes; |
ae2fafc8 | 1595 | |
1596 | if (!(io_u->flags & IO_U_F_VER_LIST)) | |
1597 | td->this_io_bytes[idx] += bytes; | |
10ba535a | 1598 | |
44f29692 JA |
1599 | if (idx == DDIR_WRITE) { |
1600 | f = io_u->file; | |
1601 | if (f) { | |
1602 | if (f->first_write == -1ULL || | |
1603 | io_u->offset < f->first_write) | |
1604 | f->first_write = io_u->offset; | |
1605 | if (f->last_write == -1ULL || | |
1606 | ((io_u->offset + bytes) > f->last_write)) | |
1607 | f->last_write = io_u->offset + bytes; | |
1608 | } | |
1609 | } | |
1610 | ||
6b1190fd SL |
1611 | if (ramp_time_over(td) && (td->runstate == TD_RUNNING || |
1612 | td->runstate == TD_VERIFYING)) { | |
c8eeb9df | 1613 | account_io_completion(td, io_u, icd, idx, bytes); |
40e1a6f0 | 1614 | |
b23b6a2f | 1615 | if (__should_check_rate(td, idx)) { |
ba3e4e0c | 1616 | td->rate_pending_usleep[idx] = |
1b8dbf25 | 1617 | (usec_for_io(td, idx) - |
ba3e4e0c | 1618 | utime_since_now(&td->start)); |
b23b6a2f | 1619 | } |
6eaf09d6 | 1620 | if (idx != DDIR_TRIM && __should_check_rate(td, odx)) |
ba3e4e0c | 1621 | td->rate_pending_usleep[odx] = |
1b8dbf25 | 1622 | (usec_for_io(td, odx) - |
ba3e4e0c | 1623 | utime_since_now(&td->start)); |
721938ae | 1624 | } |
10ba535a | 1625 | |
b29ee5b3 | 1626 | icd->bytes_done[idx] += bytes; |
3af6ef39 | 1627 | |
d7762cf8 | 1628 | if (io_u->end_io) { |
36690c9b | 1629 | ret = io_u->end_io(td, io_u); |
3af6ef39 JA |
1630 | if (ret && !icd->error) |
1631 | icd->error = ret; | |
1632 | } | |
ff58fced | 1633 | } else if (io_u->error) { |
10ba535a | 1634 | icd->error = io_u->error; |
5451792e JA |
1635 | io_u_log_error(td, io_u); |
1636 | } | |
8b28bd41 DM |
1637 | if (icd->error) { |
1638 | enum error_type_bit eb = td_error_type(io_u->ddir, icd->error); | |
1639 | if (!td_non_fatal_error(td, eb, icd->error)) | |
1640 | return; | |
f2bba182 RR |
1641 | /* |
1642 | * If there is a non_fatal error, then add to the error count | |
1643 | * and clear all the errors. | |
1644 | */ | |
1645 | update_error_count(td, icd->error); | |
1646 | td_clear_error(td); | |
1647 | icd->error = 0; | |
1648 | io_u->error = 0; | |
1649 | } | |
10ba535a JA |
1650 | } |
1651 | ||
9520ebb9 JA |
1652 | static void init_icd(struct thread_data *td, struct io_completion_data *icd, |
1653 | int nr) | |
10ba535a | 1654 | { |
6eaf09d6 | 1655 | int ddir; |
9520ebb9 JA |
1656 | if (!td->o.disable_clat || !td->o.disable_bw) |
1657 | fio_gettime(&icd->time, NULL); | |
02bcaa8c | 1658 | |
3af6ef39 JA |
1659 | icd->nr = nr; |
1660 | ||
10ba535a | 1661 | icd->error = 0; |
6eaf09d6 SL |
1662 | for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) |
1663 | icd->bytes_done[ddir] = 0; | |
36167d82 JA |
1664 | } |
1665 | ||
97601024 JA |
1666 | static void ios_completed(struct thread_data *td, |
1667 | struct io_completion_data *icd) | |
36167d82 JA |
1668 | { |
1669 | struct io_u *io_u; | |
1670 | int i; | |
1671 | ||
10ba535a JA |
1672 | for (i = 0; i < icd->nr; i++) { |
1673 | io_u = td->io_ops->event(td, i); | |
1674 | ||
1675 | io_completed(td, io_u, icd); | |
e8462bd8 JA |
1676 | |
1677 | if (!(io_u->flags & IO_U_F_FREE_DEF)) | |
1678 | put_io_u(td, io_u); | |
10ba535a JA |
1679 | } |
1680 | } | |
97601024 | 1681 | |
e7e6cfb4 JA |
1682 | /* |
1683 | * Complete a single io_u for the sync engines. | |
1684 | */ | |
581e7141 | 1685 | int io_u_sync_complete(struct thread_data *td, struct io_u *io_u, |
100f49f1 | 1686 | uint64_t *bytes) |
97601024 JA |
1687 | { |
1688 | struct io_completion_data icd; | |
1689 | ||
9520ebb9 | 1690 | init_icd(td, &icd, 1); |
97601024 | 1691 | io_completed(td, io_u, &icd); |
e8462bd8 JA |
1692 | |
1693 | if (!(io_u->flags & IO_U_F_FREE_DEF)) | |
1694 | put_io_u(td, io_u); | |
97601024 | 1695 | |
581e7141 JA |
1696 | if (icd.error) { |
1697 | td_verror(td, icd.error, "io_u_sync_complete"); | |
1698 | return -1; | |
1699 | } | |
97601024 | 1700 | |
581e7141 | 1701 | if (bytes) { |
6eaf09d6 SL |
1702 | int ddir; |
1703 | ||
1704 | for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) | |
1705 | bytes[ddir] += icd.bytes_done[ddir]; | |
581e7141 JA |
1706 | } |
1707 | ||
1708 | return 0; | |
97601024 JA |
1709 | } |
1710 | ||
e7e6cfb4 JA |
1711 | /* |
1712 | * Called to complete min_events number of io for the async engines. | |
1713 | */ | |
581e7141 | 1714 | int io_u_queued_complete(struct thread_data *td, int min_evts, |
100f49f1 | 1715 | uint64_t *bytes) |
97601024 | 1716 | { |
97601024 | 1717 | struct io_completion_data icd; |
00de55ef | 1718 | struct timespec *tvp = NULL; |
97601024 | 1719 | int ret; |
4d06a338 | 1720 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0, }; |
97601024 | 1721 | |
4950421a | 1722 | dprint(FD_IO, "io_u_queued_completed: min=%d\n", min_evts); |
b271fe62 | 1723 | |
4950421a | 1724 | if (!min_evts) |
00de55ef | 1725 | tvp = &ts; |
97601024 | 1726 | |
4950421a | 1727 | ret = td_io_getevents(td, min_evts, td->o.iodepth_batch_complete, tvp); |
97601024 | 1728 | if (ret < 0) { |
e1161c32 | 1729 | td_verror(td, -ret, "td_io_getevents"); |
97601024 JA |
1730 | return ret; |
1731 | } else if (!ret) | |
1732 | return ret; | |
1733 | ||
9520ebb9 | 1734 | init_icd(td, &icd, ret); |
97601024 | 1735 | ios_completed(td, &icd); |
581e7141 JA |
1736 | if (icd.error) { |
1737 | td_verror(td, icd.error, "io_u_queued_complete"); | |
1738 | return -1; | |
1739 | } | |
97601024 | 1740 | |
581e7141 | 1741 | if (bytes) { |
6eaf09d6 SL |
1742 | int ddir; |
1743 | ||
1744 | for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) | |
1745 | bytes[ddir] += icd.bytes_done[ddir]; | |
581e7141 JA |
1746 | } |
1747 | ||
1748 | return 0; | |
97601024 | 1749 | } |
7e77dd02 JA |
1750 | |
1751 | /* | |
1752 | * Call when io_u is really queued, to update the submission latency. | |
1753 | */ | |
1754 | void io_u_queued(struct thread_data *td, struct io_u *io_u) | |
1755 | { | |
9520ebb9 JA |
1756 | if (!td->o.disable_slat) { |
1757 | unsigned long slat_time; | |
7e77dd02 | 1758 | |
9520ebb9 | 1759 | slat_time = utime_since(&io_u->start_time, &io_u->issue_time); |
29a90ddb | 1760 | add_slat_sample(td, io_u->ddir, slat_time, io_u->xfer_buflen); |
9520ebb9 | 1761 | } |
7e77dd02 | 1762 | } |
433afcb4 | 1763 | |
cc86c395 JA |
1764 | void fill_io_buffer(struct thread_data *td, void *buf, unsigned int min_write, |
1765 | unsigned int max_bs) | |
5973cafb | 1766 | { |
ce35b1ec JA |
1767 | if (td->o.buffer_pattern_bytes) |
1768 | fill_buffer_pattern(td, buf, max_bs); | |
1769 | else if (!td->o.zero_buffers) { | |
9c42684e JA |
1770 | unsigned int perc = td->o.compress_percentage; |
1771 | ||
1772 | if (perc) { | |
f97a43a1 JA |
1773 | unsigned int seg = min_write; |
1774 | ||
1775 | seg = min(min_write, td->o.compress_chunk); | |
cc86c395 JA |
1776 | if (!seg) |
1777 | seg = min_write; | |
1778 | ||
1779 | fill_random_buf_percentage(&td->buf_state, buf, | |
f97a43a1 | 1780 | perc, seg, max_bs); |
9c42684e | 1781 | } else |
cc86c395 | 1782 | fill_random_buf(&td->buf_state, buf, max_bs); |
9c42684e | 1783 | } else |
cc86c395 JA |
1784 | memset(buf, 0, max_bs); |
1785 | } | |
1786 | ||
1787 | /* | |
1788 | * "randomly" fill the buffer contents | |
1789 | */ | |
1790 | void io_u_fill_buffer(struct thread_data *td, struct io_u *io_u, | |
1791 | unsigned int min_write, unsigned int max_bs) | |
1792 | { | |
1793 | io_u->buf_filled_len = 0; | |
1794 | fill_io_buffer(td, io_u->buf, min_write, max_bs); | |
5973cafb | 1795 | } |