projects / fio.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Add total latency log
[fio.git] / fio.h
... / ...
CommitLineData
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>
9#include <errno.h>
10#include <stdlib.h>
11#include <stdio.h>
12#include <unistd.h>
13#include <string.h>
14#include <getopt.h>
15#include <inttypes.h>
16#include <assert.h>
17
18#include "compiler/compiler.h"
19#include "flist.h"
20#include "fifo.h"
21#include "rbtree.h"
22#include "arch/arch.h"
23#include "os/os.h"
24#include "mutex.h"
25#include "log.h"
26#include "debug.h"
27#include "file.h"
28#include "io_ddir.h"
29#include "ioengine.h"
30#include "iolog.h"
31#include "helpers.h"
32#include "options.h"
33#include "profile.h"
34#include "time.h"
35
36#ifdef FIO_HAVE_GUASI
37#include <guasi.h>
38#endif
39
40#ifdef FIO_HAVE_SOLARISAIO
41#include <sys/asynch.h>
42#endif
43
44struct group_run_stats {
45 unsigned long long max_run[2], min_run[2];
46 unsigned long long max_bw[2], min_bw[2];
47 unsigned long long io_kb[2];
48 unsigned long long agg[2];
49 unsigned int kb_base;
50};
51
52/*
53 * What type of allocation to use for io buffers
54 */
55enum fio_memtype {
56 MEM_MALLOC = 0, /* ordinary malloc */
57 MEM_SHM, /* use shared memory segments */
58 MEM_SHMHUGE, /* use shared memory segments with huge pages */
59 MEM_MMAP, /* use anonynomous mmap */
60 MEM_MMAPHUGE, /* memory mapped huge file */
61};
62
63/*
64 * How many depth levels to log
65 */
66#define FIO_IO_U_MAP_NR 8
67#define FIO_IO_U_LAT_U_NR 10
68#define FIO_IO_U_LAT_M_NR 12
69
70#define MAX_PATTERN_SIZE 512
71
72struct thread_stat {
73 char *name;
74 char *verror;
75 int error;
76 int groupid;
77 pid_t pid;
78 char *description;
79 int members;
80
81 struct io_log *slat_log;
82 struct io_log *clat_log;
83 struct io_log *lat_log;
84 struct io_log *bw_log;
85
86 /*
87 * bandwidth and latency stats
88 */
89 struct io_stat clat_stat[2]; /* completion latency */
90 struct io_stat slat_stat[2]; /* submission latency */
91 struct io_stat lat_stat[2]; /* total latency */
92 struct io_stat bw_stat[2]; /* bandwidth stats */
93
94 unsigned long long stat_io_bytes[2];
95 struct timeval stat_sample_time[2];
96
97 /*
98 * fio system usage accounting
99 */
100 struct rusage ru_start;
101 struct rusage ru_end;
102 unsigned long usr_time;
103 unsigned long sys_time;
104 unsigned long ctx;
105 unsigned long minf, majf;
106
107 /*
108 * IO depth and latency stats
109 */
110 unsigned int io_u_map[FIO_IO_U_MAP_NR];
111 unsigned int io_u_submit[FIO_IO_U_MAP_NR];
112 unsigned int io_u_complete[FIO_IO_U_MAP_NR];
113 unsigned int io_u_lat_u[FIO_IO_U_LAT_U_NR];
114 unsigned int io_u_lat_m[FIO_IO_U_LAT_M_NR];
115 unsigned long total_io_u[2];
116 unsigned long short_io_u[2];
117 unsigned long total_submit;
118 unsigned long total_complete;
119
120 unsigned long long io_bytes[2];
121 unsigned long runtime[2];
122 unsigned long total_run_time;
123
124 /*
125 * IO Error related stats
126 */
127 unsigned continue_on_error;
128 unsigned long total_err_count;
129 int first_error;
130
131 unsigned int kb_base;
132};
133
134struct bssplit {
135 unsigned int bs;
136 unsigned char perc;
137};
138
139struct thread_options {
140 int pad;
141 char *description;
142 char *name;
143 char *directory;
144 char *filename;
145 char *opendir;
146 char *ioengine;
147 enum td_ddir td_ddir;
148 unsigned int kb_base;
149 unsigned int ddir_nr;
150 unsigned int iodepth;
151 unsigned int iodepth_low;
152 unsigned int iodepth_batch;
153 unsigned int iodepth_batch_complete;
154
155 unsigned long long size;
156 unsigned int fill_device;
157 unsigned long long file_size_low;
158 unsigned long long file_size_high;
159 unsigned long long start_offset;
160
161 unsigned int bs[2];
162 unsigned int ba[2];
163 unsigned int min_bs[2];
164 unsigned int max_bs[2];
165 struct bssplit *bssplit[2];
166 unsigned int bssplit_nr[2];
167
168 unsigned int nr_files;
169 unsigned int open_files;
170 enum file_lock_mode file_lock_mode;
171 unsigned int lockfile_batch;
172
173 unsigned int odirect;
174 unsigned int invalidate_cache;
175 unsigned int create_serialize;
176 unsigned int create_fsync;
177 unsigned int create_on_open;
178 unsigned int end_fsync;
179 unsigned int pre_read;
180 unsigned int sync_io;
181 unsigned int verify;
182 unsigned int do_verify;
183 unsigned int verifysort;
184 unsigned int verify_interval;
185 unsigned int verify_offset;
186 char verify_pattern[MAX_PATTERN_SIZE];
187 unsigned int verify_pattern_bytes;
188 unsigned int verify_fatal;
189 unsigned int verify_async;
190 unsigned long long verify_backlog;
191 unsigned int verify_batch;
192 unsigned int use_thread;
193 unsigned int unlink;
194 unsigned int do_disk_util;
195 unsigned int override_sync;
196 unsigned int rand_repeatable;
197 unsigned int write_lat_log;
198 unsigned int write_bw_log;
199 unsigned int norandommap;
200 unsigned int softrandommap;
201 unsigned int bs_unaligned;
202 unsigned int fsync_on_close;
203
204 unsigned int hugepage_size;
205 unsigned int rw_min_bs;
206 unsigned int thinktime;
207 unsigned int thinktime_spin;
208 unsigned int thinktime_blocks;
209 unsigned int fsync_blocks;
210 unsigned int fdatasync_blocks;
211 unsigned int start_delay;
212 unsigned long long timeout;
213 unsigned long long ramp_time;
214 unsigned int overwrite;
215 unsigned int bw_avg_time;
216 unsigned int loops;
217 unsigned long long zone_size;
218 unsigned long long zone_skip;
219 enum fio_memtype mem_type;
220 unsigned int mem_align;
221
222 unsigned int stonewall;
223 unsigned int new_group;
224 unsigned int numjobs;
225 os_cpu_mask_t cpumask;
226 unsigned int cpumask_set;
227 os_cpu_mask_t verify_cpumask;
228 unsigned int verify_cpumask_set;
229 unsigned int iolog;
230 unsigned int rwmixcycle;
231 unsigned int rwmix[2];
232 unsigned int nice;
233 unsigned int file_service_type;
234 unsigned int group_reporting;
235 unsigned int fadvise_hint;
236 unsigned int fallocate;
237 unsigned int zero_buffers;
238 unsigned int refill_buffers;
239 unsigned int time_based;
240 unsigned int disable_lat;
241 unsigned int disable_clat;
242 unsigned int disable_slat;
243 unsigned int disable_bw;
244 unsigned int gtod_reduce;
245 unsigned int gtod_cpu;
246 unsigned int gtod_offload;
247 enum fio_cs clocksource;
248
249 char *read_iolog_file;
250 char *write_iolog_file;
251 char *bw_log_file;
252 char *lat_log_file;
253
254 /*
255 * Pre-run and post-run shell
256 */
257 char *exec_prerun;
258 char *exec_postrun;
259
260 unsigned int rate[2];
261 unsigned int ratemin[2];
262 unsigned int ratecycle;
263 unsigned int rate_iops[2];
264 unsigned int rate_iops_min[2];
265
266 char *ioscheduler;
267
268 /*
269 * CPU "io" cycle burner
270 */
271 unsigned int cpuload;
272 unsigned int cpucycle;
273
274 /*
275 * I/O Error handling
276 */
277 unsigned int continue_on_error;
278
279 /*
280 * Benchmark profile type
281 */
282 char *profile;
283
284 /*
285 * blkio cgroup support
286 */
287 char *cgroup;
288 unsigned int cgroup_weight;
289 unsigned int cgroup_nodelete;
290
291 unsigned int uid;
292 unsigned int gid;
293
294 unsigned int sync_file_range;
295};
296
297#define FIO_VERROR_SIZE 128
298
299/*
300 * This describes a single thread/process executing a fio job.
301 */
302struct thread_data {
303 struct thread_options o;
304 char verror[FIO_VERROR_SIZE];
305 pthread_t thread;
306 int thread_number;
307 int groupid;
308 struct thread_stat ts;
309 struct fio_file **files;
310 unsigned int files_size;
311 unsigned int files_index;
312 unsigned int nr_open_files;
313 unsigned int nr_done_files;
314 unsigned int nr_normal_files;
315 union {
316 unsigned int next_file;
317 os_random_state_t next_file_state;
318 };
319 int error;
320 int done;
321 pid_t pid;
322 char *orig_buffer;
323 size_t orig_buffer_size;
324 volatile int terminate;
325 volatile int runstate;
326 unsigned int ioprio;
327 unsigned int ioprio_set;
328 unsigned int last_was_sync;
329 enum fio_ddir last_ddir;
330
331 char *mmapfile;
332 int mmapfd;
333
334 void *iolog_buf;
335 FILE *iolog_f;
336
337 char *sysfs_root;
338
339 unsigned long rand_seeds[6];
340
341 os_random_state_t bsrange_state;
342 os_random_state_t verify_state;
343
344 unsigned int verify_batch;
345
346 int shm_id;
347
348 /*
349 * IO engine hooks, contains everything needed to submit an io_u
350 * to any of the available IO engines.
351 */
352 struct ioengine_ops *io_ops;
353
354 /*
355 * Current IO depth and list of free and busy io_u's.
356 */
357 unsigned int cur_depth;
358 unsigned int io_u_queued;
359 struct flist_head io_u_freelist;
360 struct flist_head io_u_busylist;
361 struct flist_head io_u_requeues;
362 pthread_mutex_t io_u_lock;
363 pthread_cond_t free_cond;
364
365 /*
366 * async verify offload
367 */
368 struct flist_head verify_list;
369 pthread_t *verify_threads;
370 unsigned int nr_verify_threads;
371 pthread_cond_t verify_cond;
372 int verify_thread_exit;
373
374 /*
375 * Rate state
376 */
377 unsigned long rate_nsec_cycle[2];
378 long rate_pending_usleep[2];
379 unsigned long rate_bytes[2];
380 unsigned long rate_blocks[2];
381 struct timeval lastrate[2];
382
383 unsigned long long total_io_size;
384 unsigned long long fill_device_size;
385
386 unsigned long io_issues[2];
387 unsigned long long io_blocks[2];
388 unsigned long long io_bytes[2];
389 unsigned long long io_skip_bytes;
390 unsigned long long this_io_bytes[2];
391 unsigned long long zone_bytes;
392 struct fio_mutex *mutex;
393
394 /*
395 * State for random io, a bitmap of blocks done vs not done
396 */
397 os_random_state_t random_state;
398
399 struct timeval start; /* start of this loop */
400 struct timeval epoch; /* time job was started */
401 struct timeval last_issue;
402 struct timeval tv_cache;
403 unsigned int tv_cache_nr;
404 unsigned int tv_cache_mask;
405 unsigned int ramp_time_over;
406
407 /*
408 * read/write mixed workload state
409 */
410 os_random_state_t rwmix_state;
411 unsigned long rwmix_issues;
412 enum fio_ddir rwmix_ddir;
413 unsigned int ddir_nr;
414
415 /*
416 * IO history logs for verification. We use a tree for sorting,
417 * if we are overwriting. Otherwise just use a fifo.
418 */
419 struct rb_root io_hist_tree;
420 struct flist_head io_hist_list;
421 unsigned long io_hist_len;
422
423 /*
424 * For IO replaying
425 */
426 struct flist_head io_log_list;
427
428 /*
429 * for fileservice, how often to switch to a new file
430 */
431 unsigned int file_service_nr;
432 unsigned int file_service_left;
433 struct fio_file *file_service_file;
434
435 unsigned int sync_file_range_nr;
436
437 /*
438 * For generating file sizes
439 */
440 os_random_state_t file_size_state;
441
442 /*
443 * Error counts
444 */
445 unsigned int total_err_count;
446 int first_error;
447
448 /*
449 * Can be overloaded by profiles
450 */
451 struct prof_io_ops prof_io_ops;
452 void *prof_data;
453};
454
455/*
456 * when should interactive ETA output be generated
457 */
458enum {
459 FIO_ETA_AUTO,
460 FIO_ETA_ALWAYS,
461 FIO_ETA_NEVER,
462};
463
464#define __td_verror(td, err, msg, func) \
465 do { \
466 if ((td)->error) \
467 break; \
468 int e = (err); \
469 (td)->error = e; \
470 if (!(td)->first_error) \
471 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
472 } while (0)
473
474
475#define td_clear_error(td) \
476 (td)->error = 0;
477#define td_verror(td, err, func) \
478 __td_verror((td), (err), strerror((err)), (func))
479#define td_vmsg(td, err, msg, func) \
480 __td_verror((td), (err), (msg), (func))
481
482extern int exitall_on_terminate;
483extern int thread_number;
484extern int nr_process, nr_thread;
485extern int shm_id;
486extern int groupid;
487extern int terse_output;
488extern int temp_stall_ts;
489extern unsigned long long mlock_size;
490extern unsigned long page_mask, page_size;
491extern int read_only;
492extern int eta_print;
493extern unsigned long done_secs;
494extern char *job_section;
495extern int fio_gtod_offload;
496extern int fio_gtod_cpu;
497extern enum fio_cs fio_clock_source;
498
499extern struct thread_data *threads;
500
501static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
502{
503 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
504}
505
506#define BLOCKS_PER_MAP (8 * sizeof(int))
507#define TO_MAP_BLOCK(f, b) (b)
508#define RAND_MAP_IDX(f, b) (TO_MAP_BLOCK(f, b) / BLOCKS_PER_MAP)
509#define RAND_MAP_BIT(f, b) (TO_MAP_BLOCK(f, b) & (BLOCKS_PER_MAP - 1))
510
511#define MAX_JOBS (1024)
512
513#define td_non_fatal_error(e) ((e) == EIO || (e) == EILSEQ)
514
515static inline void update_error_count(struct thread_data *td, int err)
516{
517 td->total_err_count++;
518 if (td->total_err_count == 1)
519 td->first_error = err;
520}
521
522static inline int should_fsync(struct thread_data *td)
523{
524 if (td->last_was_sync)
525 return 0;
526 if (td->o.odirect)
527 return 0;
528 if (td_write(td) || td_rw(td) || td->o.override_sync)
529 return 1;
530
531 return 0;
532}
533
534/*
535 * Init/option functions
536 */
537extern int __must_check parse_options(int, char **);
538extern int fio_options_parse(struct thread_data *, char **, int);
539extern void fio_keywords_init(void);
540extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
541extern void fio_fill_default_options(struct thread_data *);
542extern int fio_show_option_help(const char *);
543extern void fio_options_dup_and_init(struct option *);
544extern void options_mem_dupe(struct thread_data *);
545extern void options_mem_free(struct thread_data *);
546extern void td_fill_rand_seeds(struct thread_data *);
547extern void add_job_opts(const char **);
548#define FIO_GETOPT_JOB 0x89988998
549#define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
550
551/*
552 * ETA/status stuff
553 */
554extern void print_thread_status(void);
555extern void print_status_init(int);
556
557/*
558 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
559 * will never back again. It may cycle between running/verififying/fsyncing.
560 * Once the thread reaches TD_EXITED, it is just waiting for the core to
561 * reap it.
562 */
563enum {
564 TD_NOT_CREATED = 0,
565 TD_CREATED,
566 TD_INITIALIZED,
567 TD_RAMP,
568 TD_RUNNING,
569 TD_PRE_READING,
570 TD_VERIFYING,
571 TD_FSYNCING,
572 TD_EXITED,
573 TD_REAPED,
574};
575
576extern void td_set_runstate(struct thread_data *, int);
577
578/*
579 * Memory helpers
580 */
581extern int __must_check fio_pin_memory(void);
582extern void fio_unpin_memory(void);
583extern int __must_check allocate_io_mem(struct thread_data *);
584extern void free_io_mem(struct thread_data *);
585
586/*
587 * Reset stats after ramp time completes
588 */
589extern void reset_all_stats(struct thread_data *);
590
591/*
592 * blktrace support
593 */
594#ifdef FIO_HAVE_BLKTRACE
595extern int is_blktrace(const char *);
596extern int load_blktrace(struct thread_data *, const char *);
597#endif
598
599/*
600 * Mark unused variables passed to ops functions as unused, to silence gcc
601 */
602#define fio_unused __attribute((__unused__))
603#define fio_init __attribute__((constructor))
604#define fio_exit __attribute__((destructor))
605
606#define for_each_td(td, i) \
607 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
608#define for_each_file(td, f, i) \
609 if ((td)->files_index) \
610 for ((i) = 0, (f) = (td)->files[0]; \
611 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
612 (i)++)
613
614#define fio_assert(td, cond) do { \
615 if (!(cond)) { \
616 int *__foo = NULL; \
617 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
618 (td)->runstate = TD_EXITED; \
619 (td)->error = EFAULT; \
620 *__foo = 0; \
621 } \
622} while (0)
623
624static inline int fio_fill_issue_time(struct thread_data *td)
625{
626 if (td->o.read_iolog_file ||
627 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
628 return 1;
629
630 return 0;
631}
632
633/*
634 * Cheesy number->string conversion, complete with carry rounding error.
635 */
636static inline char *num2str(unsigned long num, int maxlen, int base, int pow2)
637{
638 char postfix[] = { ' ', 'K', 'M', 'G', 'P', 'E' };
639 unsigned int thousand;
640 char *buf;
641 int i;
642
643 if (pow2)
644 thousand = 1024;
645 else
646 thousand = 1000;
647
648 buf = malloc(128);
649
650 for (i = 0; base > 1; i++)
651 base /= thousand;
652
653 do {
654 int len, carry = 0;
655
656 len = sprintf(buf, "%'lu", num);
657 if (len <= maxlen) {
658 if (i >= 1) {
659 buf[len] = postfix[i];
660 buf[len + 1] = '\0';
661 }
662 return buf;
663 }
664
665 if ((num % thousand) >= (thousand / 2))
666 carry = 1;
667
668 num /= thousand;
669 num += carry;
670 i++;
671 } while (i <= 5);
672
673 return buf;
674}
675
676static inline int __should_check_rate(struct thread_data *td,
677 enum fio_ddir ddir)
678{
679 struct thread_options *o = &td->o;
680
681 /*
682 * If some rate setting was given, we need to check it
683 */
684 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
685 o->rate_iops_min[ddir])
686 return 1;
687
688 return 0;
689}
690
691static inline int should_check_rate(struct thread_data *td,
692 unsigned long *bytes_done)
693{
694 int ret = 0;
695
696 if (bytes_done[0])
697 ret |= __should_check_rate(td, 0);
698 if (bytes_done[1])
699 ret |= __should_check_rate(td, 1);
700
701 return ret;
702}
703
704static inline int is_power_of_2(unsigned int val)
705{
706 return (val != 0 && ((val & (val - 1)) == 0));
707}
708
709/*
710 * We currently only need to do locking if we have verifier threads
711 * accessing our internal structures too
712 */
713static inline void td_io_u_lock(struct thread_data *td)
714{
715 if (td->o.verify_async)
716 pthread_mutex_lock(&td->io_u_lock);
717}
718
719static inline void td_io_u_unlock(struct thread_data *td)
720{
721 if (td->o.verify_async)
722 pthread_mutex_unlock(&td->io_u_lock);
723}
724
725static inline void td_io_u_free_notify(struct thread_data *td)
726{
727 if (td->o.verify_async)
728 pthread_cond_signal(&td->free_cond);
729}
730
731#endif
fio - Flexible IO Tester