Branch and cache miss speedups
[fio.git] / fio.h
CommitLineData
ebac4655
JA
1#ifndef FIO_H
2#define FIO_H
3
4#include <sched.h>
5#include <limits.h>
6#include <pthread.h>
7#include <sys/time.h>
8#include <sys/resource.h>
3c39a379
JA
9#include <errno.h>
10#include <stdlib.h>
11#include <stdio.h>
6d6f031f 12#include <unistd.h>
34cfcdaf 13#include <string.h>
cd14cc10 14#include <inttypes.h>
7101d9c2 15#include <assert.h>
ebac4655 16
317b95d0 17#include "compiler/compiler.h"
8062f527 18#include "thread_options.h"
01743ee1 19#include "flist.h"
e2887563 20#include "fifo.h"
317b95d0
JA
21#include "arch/arch.h"
22#include "os/os.h"
07739b57 23#include "mutex.h"
a3d741fa
JA
24#include "log.h"
25#include "debug.h"
d6aed795
JA
26#include "file.h"
27#include "io_ddir.h"
dcefb588 28#include "ioengine.h"
5995a6a4 29#include "iolog.h"
c5c8bd5c 30#include "helpers.h"
07b3232d 31#include "options.h"
7eb36574 32#include "profile.h"
c223da83 33#include "time.h"
a7346816 34#include "gettime.h"
bf2e821a 35#include "lib/getopt.h"
2615cc4b 36#include "lib/rand.h"
7fb6d453 37#include "lib/rbtree.h"
dd366728 38#include "client.h"
37db14fe 39#include "server.h"
a64e88da 40#include "stat.h"
9e684a49 41#include "flow.h"
2ae0b204 42#include "io_u_queue.h"
ebac4655 43
4700b234 44#ifdef CONFIG_SOLARISAIO
417f0068
JA
45#include <sys/asynch.h>
46#endif
47
67bf9823 48#ifdef CONFIG_LIBNUMA
d0b937ed
YR
49#include <linux/mempolicy.h>
50#include <numa.h>
51
52/*
53 * "local" is pseudo-policy
54 */
55#define MPOL_LOCAL MPOL_MAX
56#endif
57
38dad62d
JA
58/*
59 * offset generator types
60 */
61enum {
62 RW_SEQ_SEQ = 0,
63 RW_SEQ_IDENT,
64};
65
d72be545
JA
66enum {
67 TD_F_VER_BACKLOG = 1,
68 TD_F_TRIM_BACKLOG = 2,
69 TD_F_READ_IOLOG = 4,
70 TD_F_REFILL_BUFFERS = 8,
71 TD_F_SCRAMBLE_BUFFERS = 16,
72 TD_F_VER_NONE = 32,
73 TD_F_PROFILE_OPS = 64,
74};
75
d24945f0
JA
76enum {
77 FIO_RAND_BS_OFF = 0,
78 FIO_RAND_VER_OFF,
79 FIO_RAND_MIX_OFF,
80 FIO_RAND_FILE_OFF,
81 FIO_RAND_BLOCK_OFF,
82 FIO_RAND_FILE_SIZE_OFF,
83 FIO_RAND_TRIM_OFF,
84 FIO_RAND_BUF_OFF,
d9472271
JA
85 FIO_RAND_SEQ_RAND_READ_OFF,
86 FIO_RAND_SEQ_RAND_WRITE_OFF,
87 FIO_RAND_SEQ_RAND_TRIM_OFF,
23ed19b0 88 FIO_RAND_START_DELAY,
d24945f0
JA
89 FIO_RAND_NR_OFFS,
90};
91
2dc1bbeb
JA
92/*
93 * This describes a single thread/process executing a fio job.
94 */
95struct thread_data {
96 struct thread_options o;
d72be545 97 unsigned long flags;
de890a1e 98 void *eo;
e4e33258 99 char verror[FIO_VERROR_SIZE];
2dc1bbeb 100 pthread_t thread;
2f122b13
JA
101 unsigned int thread_number;
102 unsigned int groupid;
2dc1bbeb 103 struct thread_stat ts;
7b9f733a 104
46bcd498
JA
105 int client_type;
106
7b9f733a
JA
107 struct io_log *slat_log;
108 struct io_log *clat_log;
109 struct io_log *lat_log;
110 struct io_log *bw_log;
c8eeb9df 111 struct io_log *iops_log;
7b9f733a 112
6eaf09d6 113 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
c8eeb9df
JA
114 struct timeval bw_sample_time;
115
6eaf09d6 116 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
c8eeb9df 117 struct timeval iops_sample_time;
f0505a14 118
c97f1ad6
JA
119 volatile int update_rusage;
120 struct fio_mutex *rusage_sem;
c8aaba19
JA
121 struct rusage ru_start;
122 struct rusage ru_end;
123
126d65c6 124 struct fio_file **files;
d7df1d13 125 unsigned char *file_locks;
dd87b2c9 126 unsigned int files_size;
2dc1bbeb
JA
127 unsigned int files_index;
128 unsigned int nr_open_files;
1020a139 129 unsigned int nr_done_files;
2dc1bbeb
JA
130 unsigned int nr_normal_files;
131 union {
132 unsigned int next_file;
133 os_random_state_t next_file_state;
4c07ad86 134 struct frand_state __next_file_state;
2dc1bbeb
JA
135 };
136 int error;
a5e371a6 137 int sig;
20e354ef 138 int done;
2dc1bbeb
JA
139 pid_t pid;
140 char *orig_buffer;
141 size_t orig_buffer_size;
142 volatile int terminate;
143 volatile int runstate;
2dc1bbeb 144 unsigned int last_was_sync;
9e144189 145 enum fio_ddir last_ddir;
2dc1bbeb 146
2dc1bbeb
JA
147 int mmapfd;
148
843a7413
JA
149 void *iolog_buf;
150 FILE *iolog_f;
ebac4655 151
da86774e 152 char *sysfs_root;
da86774e 153
d24945f0 154 unsigned long rand_seeds[FIO_RAND_NR_OFFS];
5bfc35d7 155
4c07ad86
JA
156 union {
157 os_random_state_t bsrange_state;
158 struct frand_state __bsrange_state;
159 };
160 union {
161 os_random_state_t verify_state;
162 struct frand_state __verify_state;
163 };
164 union {
165 os_random_state_t trim_state;
166 struct frand_state __trim_state;
167 };
23ed19b0
CE
168 union {
169 os_random_state_t delay_state;
170 struct frand_state __delay_state;
171 };
ebac4655 172
3545a109
JA
173 struct frand_state buf_state;
174
9e144189 175 unsigned int verify_batch;
0d29de83 176 unsigned int trim_batch;
9e144189 177
ebac4655
JA
178 int shm_id;
179
e9c047a0
JA
180 /*
181 * IO engine hooks, contains everything needed to submit an io_u
182 * to any of the available IO engines.
183 */
2866c82d 184 struct ioengine_ops *io_ops;
ebac4655 185
e9c047a0 186 /*
422f9e4b 187 * Queue depth of io_u's that fio MIGHT do
e9c047a0 188 */
ebac4655 189 unsigned int cur_depth;
422f9e4b
RM
190
191 /*
192 * io_u's about to be committed
193 */
d8005759 194 unsigned int io_u_queued;
422f9e4b
RM
195
196 /*
197 * io_u's submitted but not completed yet
198 */
199 unsigned int io_u_in_flight;
200
201 /*
202 * List of free and busy io_u's
203 */
2ae0b204
JA
204 struct io_u_ring io_u_requeues;
205 struct io_u_queue io_u_freelist;
206 struct io_u_queue io_u_all;
e8462bd8
JA
207 pthread_mutex_t io_u_lock;
208 pthread_cond_t free_cond;
209
210 /*
211 * async verify offload
212 */
213 struct flist_head verify_list;
214 pthread_t *verify_threads;
215 unsigned int nr_verify_threads;
216 pthread_cond_t verify_cond;
217 int verify_thread_exit;
ebac4655 218
e9c047a0
JA
219 /*
220 * Rate state
221 */
bdb6d20a 222 uint64_t rate_bps[DDIR_RWDIR_CNT];
6eaf09d6
SL
223 long rate_pending_usleep[DDIR_RWDIR_CNT];
224 unsigned long rate_bytes[DDIR_RWDIR_CNT];
225 unsigned long rate_blocks[DDIR_RWDIR_CNT];
226 struct timeval lastrate[DDIR_RWDIR_CNT];
ebac4655 227
bdb6d20a
JA
228 uint64_t total_io_size;
229 uint64_t fill_device_size;
ebac4655 230
6eaf09d6 231 unsigned long io_issues[DDIR_RWDIR_CNT];
bdb6d20a
JA
232 uint64_t io_blocks[DDIR_RWDIR_CNT];
233 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
234 uint64_t io_bytes[DDIR_RWDIR_CNT];
235 uint64_t io_skip_bytes;
236 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
237 uint64_t zone_bytes;
cdd18ad8 238 struct fio_mutex *mutex;
ebac4655 239
e9c047a0
JA
240 /*
241 * State for random io, a bitmap of blocks done vs not done
242 */
4c07ad86
JA
243 union {
244 os_random_state_t random_state;
245 struct frand_state __random_state;
246 };
ebac4655 247
ebac4655
JA
248 struct timeval start; /* start of this loop */
249 struct timeval epoch; /* time job was started */
a61eddec 250 struct timeval last_issue;
993bf48b
JA
251 struct timeval tv_cache;
252 unsigned int tv_cache_nr;
253 unsigned int tv_cache_mask;
721938ae 254 unsigned int ramp_time_over;
ebac4655 255
3e260a46
JA
256 /*
257 * Time since last latency_window was started
258 */
259 struct timeval latency_ts;
260 unsigned int latency_qd;
261 unsigned int latency_qd_high;
262 unsigned int latency_qd_low;
263 unsigned int latency_failed;
264 uint64_t latency_ios;
6bb58215 265 int latency_end_run;
3e260a46 266
e9c047a0
JA
267 /*
268 * read/write mixed workload state
269 */
4c07ad86
JA
270 union {
271 os_random_state_t rwmix_state;
272 struct frand_state __rwmix_state;
273 };
e4928662 274 unsigned long rwmix_issues;
e9c047a0 275 enum fio_ddir rwmix_ddir;
5736c10d 276 unsigned int ddir_seq_nr;
a6ccc7be 277
211c9b89
JA
278 /*
279 * rand/seq mixed workload state
280 */
281 union {
d9472271
JA
282 os_random_state_t seq_rand_state[DDIR_RWDIR_CNT];
283 struct frand_state __seq_rand_state[DDIR_RWDIR_CNT];
211c9b89
JA
284 };
285
e9c047a0 286 /*
8de8f047
JA
287 * IO history logs for verification. We use a tree for sorting,
288 * if we are overwriting. Otherwise just use a fifo.
e9c047a0 289 */
4b87898e 290 struct rb_root io_hist_tree;
01743ee1 291 struct flist_head io_hist_list;
9e144189 292 unsigned long io_hist_len;
8de8f047
JA
293
294 /*
295 * For IO replaying
296 */
01743ee1 297 struct flist_head io_log_list;
433afcb4 298
0d29de83
JA
299 /*
300 * For tracking/handling discards
301 */
302 struct flist_head trim_list;
303 unsigned long trim_entries;
304
1ae83d45
JA
305 struct flist_head next_rand_list;
306
1907dbc6
JA
307 /*
308 * for fileservice, how often to switch to a new file
309 */
310 unsigned int file_service_nr;
311 unsigned int file_service_left;
312 struct fio_file *file_service_file;
9c60ce64 313
44f29692
JA
314 unsigned int sync_file_range_nr;
315
9c60ce64
JA
316 /*
317 * For generating file sizes
318 */
4c07ad86
JA
319 union {
320 os_random_state_t file_size_state;
321 struct frand_state __file_size_state;
322 };
f2bba182
RR
323
324 /*
325 * Error counts
326 */
327 unsigned int total_err_count;
328 int first_error;
15dc1934 329
9e684a49
DE
330 struct fio_flow *flow;
331
15dc1934
JA
332 /*
333 * Can be overloaded by profiles
334 */
7eb36574 335 struct prof_io_ops prof_io_ops;
58c55ba0 336 void *prof_data;
9a3f1100
JA
337
338 void *pinned_mem;
ebac4655
JA
339};
340
e592a06b
AC
341/*
342 * when should interactive ETA output be generated
343 */
344enum {
345 FIO_ETA_AUTO,
346 FIO_ETA_ALWAYS,
347 FIO_ETA_NEVER,
348};
349
e1161c32 350#define __td_verror(td, err, msg, func) \
ebac4655 351 do { \
8271963e 352 int ____e = (err); \
19abcd3d
JA
353 if ((td)->error) \
354 break; \
8271963e 355 (td)->error = ____e; \
f2bba182 356 if (!(td)->first_error) \
98ffb8f3 357 snprintf(td->verror, sizeof(td->verror), "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
ebac4655
JA
358 } while (0)
359
b990b5c0 360
f2bba182
RR
361#define td_clear_error(td) \
362 (td)->error = 0;
e1161c32
JA
363#define td_verror(td, err, func) \
364 __td_verror((td), (err), strerror((err)), (func))
365#define td_vmsg(td, err, msg, func) \
366 __td_verror((td), (err), (msg), (func))
b990b5c0 367
81179eec
JA
368#define __fio_stringify_1(x) #x
369#define __fio_stringify(x) __fio_stringify_1(x)
370
ebac4655 371extern int exitall_on_terminate;
11e950bd 372extern unsigned int thread_number;
108fea77 373extern unsigned int stat_number;
ebac4655
JA
374extern int shm_id;
375extern int groupid;
f3afa57e 376extern int output_format;
2b8c71b0 377extern int append_terse_output;
53cdc686 378extern int temp_stall_ts;
7078106d 379extern uintptr_t page_mask, page_size;
4241ea8f 380extern int read_only;
e592a06b 381extern int eta_print;
e382e661 382extern int eta_new_line;
d3eeeabc 383extern unsigned long done_secs;
01f06b63 384extern char *job_section;
be4ecfdf
JA
385extern int fio_gtod_offload;
386extern int fio_gtod_cpu;
c223da83 387extern enum fio_cs fio_clock_source;
fa80feae 388extern int fio_clock_source_set;
a9523c6f 389extern int warnings_fatal;
f57a9c59 390extern int terse_version;
5c341e9a 391extern int is_backend;
a37f69b7 392extern int nr_clients;
e46d8091 393extern int log_syslog;
06464907 394extern int status_interval;
5e726d0a 395extern const char fio_version_string[];
ebac4655
JA
396
397extern struct thread_data *threads;
398
7101d9c2
JA
399static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
400{
401 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
402}
403
fca70358 404#define REAL_MAX_JOBS 2048
ebac4655 405
87dc1ab1
JA
406static inline int should_fsync(struct thread_data *td)
407{
408 if (td->last_was_sync)
409 return 0;
2dc1bbeb 410 if (td_write(td) || td_rw(td) || td->o.override_sync)
87dc1ab1
JA
411 return 1;
412
413 return 0;
414}
415
8914a9d8 416/*
214e1eca 417 * Init/option functions
8914a9d8 418 */
0420ba6a 419extern int __must_check fio_init_options(void);
72cb971b 420extern int __must_check parse_options(int, char **);
46bcd498
JA
421extern int parse_jobs_ini(char *, int, int, int);
422extern int parse_cmd_line(int, char **, int);
2e1df07d 423extern int fio_backend(void);
50d16976 424extern void reset_fio_state(void);
2e1df07d 425extern void clear_io_state(struct thread_data *);
292cc475 426extern int fio_options_parse(struct thread_data *, char **, int, int);
74929ac2 427extern void fio_keywords_init(void);
214e1eca 428extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
de890a1e 429extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
214e1eca
JA
430extern void fio_fill_default_options(struct thread_data *);
431extern int fio_show_option_help(const char *);
de890a1e 432extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
214e1eca 433extern void fio_options_dup_and_init(struct option *);
de890a1e
SL
434extern void fio_options_mem_dupe(struct thread_data *);
435extern void options_mem_dupe(void *data, struct fio_option *options);
5bfc35d7 436extern void td_fill_rand_seeds(struct thread_data *);
46bcd498 437extern void add_job_opts(const char **, int);
73798eb2 438extern char *num2str(unsigned long, int, int, int, int);
de890a1e 439extern int ioengine_load(struct thread_data *);
f9633d72 440extern int parse_dryrun(void);
1ec3d69b 441
a569b45f
JA
442extern uintptr_t page_mask;
443extern uintptr_t page_size;
444extern int initialize_fio(char *envp[]);
445
de890a1e
SL
446#define FIO_GETOPT_JOB 0x89000000
447#define FIO_GETOPT_IOENGINE 0x98000000
07b3232d 448#define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
8914a9d8 449
263e529f
JA
450/*
451 * ETA/status stuff
452 */
453extern void print_thread_status(void);
454extern void print_status_init(int);
0a428b23 455extern char *fio_uint_to_kmg(unsigned int val);
263e529f
JA
456
457/*
458 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
459 * will never back again. It may cycle between running/verififying/fsyncing.
460 * Once the thread reaches TD_EXITED, it is just waiting for the core to
461 * reap it.
462 */
463enum {
464 TD_NOT_CREATED = 0,
465 TD_CREATED,
466 TD_INITIALIZED,
b29ee5b3 467 TD_RAMP,
36d80bc7 468 TD_SETTING_UP,
263e529f 469 TD_RUNNING,
b0f65863 470 TD_PRE_READING,
263e529f
JA
471 TD_VERIFYING,
472 TD_FSYNCING,
473 TD_EXITED,
474 TD_REAPED,
475};
476
b29ee5b3 477extern void td_set_runstate(struct thread_data *, int);
cc0df00a
JA
478#define TERMINATE_ALL (-1)
479extern void fio_terminate_threads(int);
b29ee5b3 480
2f9ade3c
JA
481/*
482 * Memory helpers
483 */
9a3f1100
JA
484extern int __must_check fio_pin_memory(struct thread_data *);
485extern void fio_unpin_memory(struct thread_data *);
b2fdda43 486extern int __must_check allocate_io_mem(struct thread_data *);
2f9ade3c 487extern void free_io_mem(struct thread_data *);
2bb3f0a7 488extern void free_threads_shm(void);
2f9ade3c 489
b29ee5b3
JA
490/*
491 * Reset stats after ramp time completes
492 */
493extern void reset_all_stats(struct thread_data *);
494
fb7b71a3
JA
495/*
496 * blktrace support
497 */
5e62c22a 498#ifdef FIO_HAVE_BLKTRACE
d95b34a6
JA
499extern int is_blktrace(const char *, int *);
500extern int load_blktrace(struct thread_data *, const char *, int);
5e62c22a 501#endif
fb7b71a3 502
3e260a46
JA
503/*
504 * Latency target helpers
505 */
506extern void lat_target_check(struct thread_data *);
507extern void lat_target_init(struct thread_data *);
6bb58215 508extern void lat_target_reset(struct thread_data *);
3e260a46 509
34572e28
JA
510#define for_each_td(td, i) \
511 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
53cdc686 512#define for_each_file(td, f, i) \
691c8fb0
JA
513 if ((td)->files_index) \
514 for ((i) = 0, (f) = (td)->files[0]; \
515 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
516 (i)++)
53cdc686 517
0032bf9f
JA
518#define fio_assert(td, cond) do { \
519 if (!(cond)) { \
340fd243 520 int *__foo = NULL; \
0032bf9f 521 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
ac18ea38 522 td_set_runstate((td), TD_EXITED); \
437c9b71 523 (td)->error = EFAULT; \
340fd243 524 *__foo = 0; \
0032bf9f
JA
525 } \
526} while (0)
527
12d9d841
JA
528static inline int fio_fill_issue_time(struct thread_data *td)
529{
530 if (td->o.read_iolog_file ||
531 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
532 return 1;
533
534 return 0;
535}
536
581e7141
JA
537static inline int __should_check_rate(struct thread_data *td,
538 enum fio_ddir ddir)
539{
540 struct thread_options *o = &td->o;
541
542 /*
543 * If some rate setting was given, we need to check it
544 */
545 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
546 o->rate_iops_min[ddir])
547 return 1;
548
549 return 0;
550}
551
552static inline int should_check_rate(struct thread_data *td,
aa83321f 553 uint64_t *bytes_done)
581e7141
JA
554{
555 int ret = 0;
556
6eaf09d6
SL
557 if (bytes_done[DDIR_READ])
558 ret |= __should_check_rate(td, DDIR_READ);
559 if (bytes_done[DDIR_WRITE])
560 ret |= __should_check_rate(td, DDIR_WRITE);
561 if (bytes_done[DDIR_TRIM])
562 ret |= __should_check_rate(td, DDIR_TRIM);
581e7141
JA
563
564 return ret;
565}
566
74f4b020
JA
567static inline unsigned int td_max_bs(struct thread_data *td)
568{
569 unsigned int max_bs;
570
571 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
572 return max(td->o.max_bs[DDIR_TRIM], max_bs);
573}
574
8d916c94
JA
575static inline unsigned int td_min_bs(struct thread_data *td)
576{
577 unsigned int min_bs;
578
579 min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
580 return min(td->o.min_bs[DDIR_TRIM], min_bs);
581}
582
d529ee19
JA
583static inline int is_power_of_2(unsigned int val)
584{
585 return (val != 0 && ((val & (val - 1)) == 0));
586}
587
e8462bd8
JA
588/*
589 * We currently only need to do locking if we have verifier threads
590 * accessing our internal structures too
591 */
592static inline void td_io_u_lock(struct thread_data *td)
593{
594 if (td->o.verify_async)
595 pthread_mutex_lock(&td->io_u_lock);
596}
597
598static inline void td_io_u_unlock(struct thread_data *td)
599{
600 if (td->o.verify_async)
601 pthread_mutex_unlock(&td->io_u_lock);
602}
603
604static inline void td_io_u_free_notify(struct thread_data *td)
605{
606 if (td->o.verify_async)
607 pthread_cond_signal(&td->free_cond);
608}
609
cca84643
JA
610extern const char *fio_get_arch_string(int);
611extern const char *fio_get_os_string(int);
612
05775438
JA
613#define ARRAY_SIZE(x) (sizeof((x)) / (sizeof((x)[0])))
614
f3afa57e
JA
615enum {
616 FIO_OUTPUT_TERSE = 0,
617 FIO_OUTPUT_JSON,
618 FIO_OUTPUT_NORMAL,
619};
620
e25839d4
JA
621enum {
622 FIO_RAND_DIST_RANDOM = 0,
623 FIO_RAND_DIST_ZIPF,
925fee33 624 FIO_RAND_DIST_PARETO,
e25839d4
JA
625};
626
8055e41d
JA
627enum {
628 FIO_RAND_GEN_TAUSWORTHE = 0,
629 FIO_RAND_GEN_LFSR,
630};
631
ebac4655 632#endif