8 #include <sys/resource.h>
19 #include "compiler/compiler.h"
23 #include "arch/arch.h"
36 #include "lib/getopt.h"
42 #ifdef FIO_HAVE_SOLARISAIO
43 #include <sys/asynch.h>
46 struct group_run_stats {
47 unsigned long long max_run[2], min_run[2];
48 unsigned long long max_bw[2], min_bw[2];
49 unsigned long long io_kb[2];
50 unsigned long long agg[2];
55 * What type of allocation to use for io buffers
58 MEM_MALLOC = 0, /* ordinary malloc */
59 MEM_SHM, /* use shared memory segments */
60 MEM_SHMHUGE, /* use shared memory segments with huge pages */
61 MEM_MMAP, /* use anonynomous mmap */
62 MEM_MMAPHUGE, /* memory mapped huge file */
66 * offset generator types
74 * How many depth levels to log
76 #define FIO_IO_U_MAP_NR 7
77 #define FIO_IO_U_LAT_U_NR 10
78 #define FIO_IO_U_LAT_M_NR 12
80 #define MAX_PATTERN_SIZE 512
91 struct io_log *slat_log;
92 struct io_log *clat_log;
93 struct io_log *lat_log;
94 struct io_log *bw_log;
97 * bandwidth and latency stats
99 struct io_stat clat_stat[2]; /* completion latency */
100 struct io_stat slat_stat[2]; /* submission latency */
101 struct io_stat lat_stat[2]; /* total latency */
102 struct io_stat bw_stat[2]; /* bandwidth stats */
104 unsigned long long stat_io_bytes[2];
105 struct timeval stat_sample_time[2];
108 * fio system usage accounting
110 struct rusage ru_start;
111 struct rusage ru_end;
112 unsigned long usr_time;
113 unsigned long sys_time;
115 unsigned long minf, majf;
118 * IO depth and latency stats
120 unsigned int io_u_map[FIO_IO_U_MAP_NR];
121 unsigned int io_u_submit[FIO_IO_U_MAP_NR];
122 unsigned int io_u_complete[FIO_IO_U_MAP_NR];
123 unsigned int io_u_lat_u[FIO_IO_U_LAT_U_NR];
124 unsigned int io_u_lat_m[FIO_IO_U_LAT_M_NR];
125 unsigned long total_io_u[3];
126 unsigned long short_io_u[3];
127 unsigned long total_submit;
128 unsigned long total_complete;
130 unsigned long long io_bytes[2];
131 unsigned long long runtime[2];
132 unsigned long total_run_time;
135 * IO Error related stats
137 unsigned continue_on_error;
138 unsigned long total_err_count;
141 unsigned int kb_base;
149 struct thread_options {
157 enum td_ddir td_ddir;
159 unsigned int kb_base;
160 unsigned int ddir_seq_nr;
161 unsigned int iodepth;
162 unsigned int iodepth_low;
163 unsigned int iodepth_batch;
164 unsigned int iodepth_batch_complete;
166 unsigned long long size;
167 unsigned int fill_device;
168 unsigned long long file_size_low;
169 unsigned long long file_size_high;
170 unsigned long long start_offset;
174 unsigned int min_bs[2];
175 unsigned int max_bs[2];
176 struct bssplit *bssplit[2];
177 unsigned int bssplit_nr[2];
179 unsigned int nr_files;
180 unsigned int open_files;
181 enum file_lock_mode file_lock_mode;
182 unsigned int lockfile_batch;
184 unsigned int odirect;
185 unsigned int invalidate_cache;
186 unsigned int create_serialize;
187 unsigned int create_fsync;
188 unsigned int create_on_open;
189 unsigned int end_fsync;
190 unsigned int pre_read;
191 unsigned int sync_io;
193 unsigned int do_verify;
194 unsigned int verifysort;
195 unsigned int verify_interval;
196 unsigned int verify_offset;
197 char verify_pattern[MAX_PATTERN_SIZE];
198 unsigned int verify_pattern_bytes;
199 unsigned int verify_fatal;
200 unsigned int verify_dump;
201 unsigned int verify_async;
202 unsigned long long verify_backlog;
203 unsigned int verify_batch;
204 unsigned int use_thread;
206 unsigned int do_disk_util;
207 unsigned int override_sync;
208 unsigned int rand_repeatable;
209 unsigned int write_lat_log;
210 unsigned int write_bw_log;
211 unsigned int norandommap;
212 unsigned int softrandommap;
213 unsigned int bs_unaligned;
214 unsigned int fsync_on_close;
216 unsigned int hugepage_size;
217 unsigned int rw_min_bs;
218 unsigned int thinktime;
219 unsigned int thinktime_spin;
220 unsigned int thinktime_blocks;
221 unsigned int fsync_blocks;
222 unsigned int fdatasync_blocks;
223 unsigned int barrier_blocks;
224 unsigned long start_delay;
225 unsigned long long timeout;
226 unsigned long long ramp_time;
227 unsigned int overwrite;
228 unsigned int bw_avg_time;
230 unsigned long long zone_size;
231 unsigned long long zone_skip;
232 enum fio_memtype mem_type;
233 unsigned int mem_align;
235 unsigned int stonewall;
236 unsigned int new_group;
237 unsigned int numjobs;
238 os_cpu_mask_t cpumask;
239 unsigned int cpumask_set;
240 os_cpu_mask_t verify_cpumask;
241 unsigned int verify_cpumask_set;
243 unsigned int rwmixcycle;
244 unsigned int rwmix[2];
246 unsigned int file_service_type;
247 unsigned int group_reporting;
248 unsigned int fadvise_hint;
249 unsigned int fallocate;
250 unsigned int zero_buffers;
251 unsigned int refill_buffers;
252 unsigned int time_based;
253 unsigned int disable_lat;
254 unsigned int disable_clat;
255 unsigned int disable_slat;
256 unsigned int disable_bw;
257 unsigned int gtod_reduce;
258 unsigned int gtod_cpu;
259 unsigned int gtod_offload;
260 enum fio_cs clocksource;
261 unsigned int no_stall;
262 unsigned int trim_percentage;
263 unsigned int trim_batch;
264 unsigned int trim_zero;
265 unsigned long long trim_backlog;
267 char *read_iolog_file;
268 char *write_iolog_file;
271 char *replay_redirect;
274 * Pre-run and post-run shell
279 unsigned int rate[2];
280 unsigned int ratemin[2];
281 unsigned int ratecycle;
282 unsigned int rate_iops[2];
283 unsigned int rate_iops_min[2];
288 * CPU "io" cycle burner
290 unsigned int cpuload;
291 unsigned int cpucycle;
296 unsigned int continue_on_error;
299 * Benchmark profile type
304 * blkio cgroup support
307 unsigned int cgroup_weight;
308 unsigned int cgroup_nodelete;
313 unsigned int sync_file_range;
316 #define FIO_VERROR_SIZE 128
319 * This describes a single thread/process executing a fio job.
322 struct thread_options o;
323 char verror[FIO_VERROR_SIZE];
327 struct thread_stat ts;
328 struct fio_file **files;
329 unsigned int files_size;
330 unsigned int files_index;
331 unsigned int nr_open_files;
332 unsigned int nr_done_files;
333 unsigned int nr_normal_files;
335 unsigned int next_file;
336 os_random_state_t next_file_state;
342 size_t orig_buffer_size;
343 volatile int terminate;
344 volatile int runstate;
346 unsigned int ioprio_set;
347 unsigned int last_was_sync;
348 enum fio_ddir last_ddir;
358 unsigned long rand_seeds[7];
360 os_random_state_t bsrange_state;
361 os_random_state_t verify_state;
362 os_random_state_t trim_state;
364 unsigned int verify_batch;
365 unsigned int trim_batch;
370 * IO engine hooks, contains everything needed to submit an io_u
371 * to any of the available IO engines.
373 struct ioengine_ops *io_ops;
376 * Current IO depth and list of free and busy io_u's.
378 unsigned int cur_depth;
379 unsigned int io_u_queued;
380 struct flist_head io_u_freelist;
381 struct flist_head io_u_busylist;
382 struct flist_head io_u_requeues;
383 pthread_mutex_t io_u_lock;
384 pthread_cond_t free_cond;
387 * async verify offload
389 struct flist_head verify_list;
390 pthread_t *verify_threads;
391 unsigned int nr_verify_threads;
392 pthread_cond_t verify_cond;
393 int verify_thread_exit;
398 unsigned long rate_nsec_cycle[2];
399 long rate_pending_usleep[2];
400 unsigned long rate_bytes[2];
401 unsigned long rate_blocks[2];
402 struct timeval lastrate[2];
404 unsigned long long total_io_size;
405 unsigned long long fill_device_size;
407 unsigned long io_issues[2];
408 unsigned long long io_blocks[2];
409 unsigned long long io_bytes[2];
410 unsigned long long io_skip_bytes;
411 unsigned long long this_io_bytes[2];
412 unsigned long long zone_bytes;
413 struct fio_mutex *mutex;
416 * State for random io, a bitmap of blocks done vs not done
418 os_random_state_t random_state;
420 struct timeval start; /* start of this loop */
421 struct timeval epoch; /* time job was started */
422 struct timeval last_issue;
423 struct timeval tv_cache;
424 unsigned int tv_cache_nr;
425 unsigned int tv_cache_mask;
426 unsigned int ramp_time_over;
429 * read/write mixed workload state
431 os_random_state_t rwmix_state;
432 unsigned long rwmix_issues;
433 enum fio_ddir rwmix_ddir;
434 unsigned int ddir_seq_nr;
437 * IO history logs for verification. We use a tree for sorting,
438 * if we are overwriting. Otherwise just use a fifo.
440 struct rb_root io_hist_tree;
441 struct flist_head io_hist_list;
442 unsigned long io_hist_len;
447 struct flist_head io_log_list;
450 * For tracking/handling discards
452 struct flist_head trim_list;
453 unsigned long trim_entries;
456 * for fileservice, how often to switch to a new file
458 unsigned int file_service_nr;
459 unsigned int file_service_left;
460 struct fio_file *file_service_file;
462 unsigned int sync_file_range_nr;
465 * For generating file sizes
467 os_random_state_t file_size_state;
472 unsigned int total_err_count;
476 * Can be overloaded by profiles
478 struct prof_io_ops prof_io_ops;
483 * when should interactive ETA output be generated
491 #define __td_verror(td, err, msg, func) \
497 if (!(td)->first_error) \
498 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
502 #define td_clear_error(td) \
504 #define td_verror(td, err, func) \
505 __td_verror((td), (err), strerror((err)), (func))
506 #define td_vmsg(td, err, msg, func) \
507 __td_verror((td), (err), (msg), (func))
509 extern int exitall_on_terminate;
510 extern int thread_number;
511 extern int nr_process, nr_thread;
514 extern int terse_output;
515 extern int temp_stall_ts;
516 extern unsigned long long mlock_size;
517 extern unsigned long page_mask, page_size;
518 extern int read_only;
519 extern int eta_print;
520 extern unsigned long done_secs;
521 extern char *job_section;
522 extern int fio_gtod_offload;
523 extern int fio_gtod_cpu;
524 extern enum fio_cs fio_clock_source;
526 extern struct thread_data *threads;
528 static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
530 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
533 #define BLOCKS_PER_MAP (8 * sizeof(int))
534 #define TO_MAP_BLOCK(f, b) (b)
535 #define RAND_MAP_IDX(f, b) (TO_MAP_BLOCK(f, b) / BLOCKS_PER_MAP)
536 #define RAND_MAP_BIT(f, b) (TO_MAP_BLOCK(f, b) & (BLOCKS_PER_MAP - 1))
538 #define MAX_JOBS (1024)
540 #define td_non_fatal_error(e) ((e) == EIO || (e) == EILSEQ)
542 static inline void update_error_count(struct thread_data *td, int err)
544 td->total_err_count++;
545 if (td->total_err_count == 1)
546 td->first_error = err;
549 static inline int should_fsync(struct thread_data *td)
551 if (td->last_was_sync)
555 if (td_write(td) || td_rw(td) || td->o.override_sync)
562 * Init/option functions
564 extern int __must_check parse_options(int, char **);
565 extern int fio_options_parse(struct thread_data *, char **, int);
566 extern void fio_keywords_init(void);
567 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
568 extern void fio_fill_default_options(struct thread_data *);
569 extern int fio_show_option_help(const char *);
570 extern void fio_options_dup_and_init(struct option *);
571 extern void options_mem_dupe(struct thread_data *);
572 extern void options_mem_free(struct thread_data *);
573 extern void td_fill_rand_seeds(struct thread_data *);
574 extern void add_job_opts(const char **);
575 extern char *num2str(unsigned long, int, int, int);
577 #define FIO_GETOPT_JOB 0x89988998
578 #define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
583 extern void print_thread_status(void);
584 extern void print_status_init(int);
587 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
588 * will never back again. It may cycle between running/verififying/fsyncing.
589 * Once the thread reaches TD_EXITED, it is just waiting for the core to
605 extern void td_set_runstate(struct thread_data *, int);
610 extern int __must_check fio_pin_memory(void);
611 extern void fio_unpin_memory(void);
612 extern int __must_check allocate_io_mem(struct thread_data *);
613 extern void free_io_mem(struct thread_data *);
616 * Reset stats after ramp time completes
618 extern void reset_all_stats(struct thread_data *);
623 #ifdef FIO_HAVE_BLKTRACE
624 extern int is_blktrace(const char *);
625 extern int load_blktrace(struct thread_data *, const char *);
629 * Mark unused variables passed to ops functions as unused, to silence gcc
631 #define fio_unused __attribute((__unused__))
632 #define fio_init __attribute__((constructor))
633 #define fio_exit __attribute__((destructor))
635 #define for_each_td(td, i) \
636 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
637 #define for_each_file(td, f, i) \
638 if ((td)->files_index) \
639 for ((i) = 0, (f) = (td)->files[0]; \
640 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
643 #define fio_assert(td, cond) do { \
646 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
647 (td)->runstate = TD_EXITED; \
648 (td)->error = EFAULT; \
653 static inline int fio_fill_issue_time(struct thread_data *td)
655 if (td->o.read_iolog_file ||
656 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
662 static inline int __should_check_rate(struct thread_data *td,
665 struct thread_options *o = &td->o;
668 * If some rate setting was given, we need to check it
670 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
671 o->rate_iops_min[ddir])
677 static inline int should_check_rate(struct thread_data *td,
678 unsigned long *bytes_done)
683 ret |= __should_check_rate(td, 0);
685 ret |= __should_check_rate(td, 1);
690 static inline int is_power_of_2(unsigned int val)
692 return (val != 0 && ((val & (val - 1)) == 0));
696 * We currently only need to do locking if we have verifier threads
697 * accessing our internal structures too
699 static inline void td_io_u_lock(struct thread_data *td)
701 if (td->o.verify_async)
702 pthread_mutex_lock(&td->io_u_lock);
705 static inline void td_io_u_unlock(struct thread_data *td)
707 if (td->o.verify_async)
708 pthread_mutex_unlock(&td->io_u_lock);
711 static inline void td_io_u_free_notify(struct thread_data *td)
713 if (td->o.verify_async)
714 pthread_cond_signal(&td->free_cond);