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