8 #include <sys/resource.h>
18 #include "compiler/compiler.h"
22 #include "arch/arch.h"
38 #ifdef FIO_HAVE_SOLARISAIO
39 #include <sys/asynch.h>
42 struct group_run_stats {
43 unsigned long long max_run[2], min_run[2];
44 unsigned long long max_bw[2], min_bw[2];
45 unsigned long long io_kb[2];
46 unsigned long long agg[2];
51 * What type of allocation to use for io buffers
54 MEM_MALLOC = 0, /* ordinary malloc */
55 MEM_SHM, /* use shared memory segments */
56 MEM_SHMHUGE, /* use shared memory segments with huge pages */
57 MEM_MMAP, /* use anonynomous mmap */
58 MEM_MMAPHUGE, /* memory mapped huge file */
62 * How many depth levels to log
64 #define FIO_IO_U_MAP_NR 8
65 #define FIO_IO_U_LAT_U_NR 10
66 #define FIO_IO_U_LAT_M_NR 12
68 #define MAX_PATTERN_SIZE 512
79 struct io_log *slat_log;
80 struct io_log *clat_log;
81 struct io_log *bw_log;
84 * bandwidth and latency stats
86 struct io_stat clat_stat[2]; /* completion latency */
87 struct io_stat slat_stat[2]; /* submission latency */
88 struct io_stat bw_stat[2]; /* bandwidth stats */
90 unsigned long long stat_io_bytes[2];
91 struct timeval stat_sample_time[2];
94 * fio system usage accounting
96 struct rusage ru_start;
98 unsigned long usr_time;
99 unsigned long sys_time;
101 unsigned long minf, majf;
104 * IO depth and latency stats
106 unsigned int io_u_map[FIO_IO_U_MAP_NR];
107 unsigned int io_u_submit[FIO_IO_U_MAP_NR];
108 unsigned int io_u_complete[FIO_IO_U_MAP_NR];
109 unsigned int io_u_lat_u[FIO_IO_U_LAT_U_NR];
110 unsigned int io_u_lat_m[FIO_IO_U_LAT_M_NR];
111 unsigned long total_io_u[2];
112 unsigned long short_io_u[2];
113 unsigned long total_submit;
114 unsigned long total_complete;
116 unsigned long long io_bytes[2];
117 unsigned long runtime[2];
118 unsigned long total_run_time;
121 * IO Error related stats
123 unsigned continue_on_error;
124 unsigned long total_err_count;
127 unsigned int kb_base;
135 struct thread_options {
143 enum td_ddir td_ddir;
144 unsigned int kb_base;
145 unsigned int ddir_nr;
146 unsigned int iodepth;
147 unsigned int iodepth_low;
148 unsigned int iodepth_batch;
149 unsigned int iodepth_batch_complete;
151 unsigned long long size;
152 unsigned int fill_device;
153 unsigned long long file_size_low;
154 unsigned long long file_size_high;
155 unsigned long long start_offset;
159 unsigned int min_bs[2];
160 unsigned int max_bs[2];
161 struct bssplit *bssplit[2];
162 unsigned int bssplit_nr[2];
164 unsigned int nr_files;
165 unsigned int open_files;
166 enum file_lock_mode file_lock_mode;
167 unsigned int lockfile_batch;
169 unsigned int odirect;
170 unsigned int invalidate_cache;
171 unsigned int create_serialize;
172 unsigned int create_fsync;
173 unsigned int create_on_open;
174 unsigned int end_fsync;
175 unsigned int pre_read;
176 unsigned int sync_io;
178 unsigned int do_verify;
179 unsigned int verifysort;
180 unsigned int verify_interval;
181 unsigned int verify_offset;
182 char verify_pattern[MAX_PATTERN_SIZE];
183 unsigned int verify_pattern_bytes;
184 unsigned int verify_fatal;
185 unsigned int verify_async;
186 unsigned int use_thread;
188 unsigned int do_disk_util;
189 unsigned int override_sync;
190 unsigned int rand_repeatable;
191 unsigned int write_lat_log;
192 unsigned int write_bw_log;
193 unsigned int norandommap;
194 unsigned int softrandommap;
195 unsigned int bs_unaligned;
196 unsigned int fsync_on_close;
198 unsigned int hugepage_size;
199 unsigned int rw_min_bs;
200 unsigned int thinktime;
201 unsigned int thinktime_spin;
202 unsigned int thinktime_blocks;
203 unsigned int fsync_blocks;
204 unsigned int fdatasync_blocks;
205 unsigned int start_delay;
206 unsigned long long timeout;
207 unsigned long long ramp_time;
208 unsigned int overwrite;
209 unsigned int bw_avg_time;
211 unsigned long long zone_size;
212 unsigned long long zone_skip;
213 enum fio_memtype mem_type;
214 unsigned int mem_align;
216 unsigned int stonewall;
217 unsigned int new_group;
218 unsigned int numjobs;
219 os_cpu_mask_t cpumask;
220 unsigned int cpumask_set;
221 os_cpu_mask_t verify_cpumask;
222 unsigned int verify_cpumask_set;
224 unsigned int rwmixcycle;
225 unsigned int rwmix[2];
227 unsigned int file_service_type;
228 unsigned int group_reporting;
229 unsigned int fadvise_hint;
230 unsigned int zero_buffers;
231 unsigned int refill_buffers;
232 unsigned int time_based;
233 unsigned int disable_clat;
234 unsigned int disable_slat;
235 unsigned int disable_bw;
236 unsigned int gtod_reduce;
237 unsigned int gtod_cpu;
238 unsigned int gtod_offload;
240 char *read_iolog_file;
241 char *write_iolog_file;
246 * Pre-run and post-run shell
251 unsigned int rate[2];
252 unsigned int ratemin[2];
253 unsigned int ratecycle;
254 unsigned int rate_iops[2];
255 unsigned int rate_iops_min[2];
260 * CPU "io" cycle burner
262 unsigned int cpuload;
263 unsigned int cpucycle;
268 unsigned int continue_on_error;
271 * Benchmark profile type
273 unsigned int profile;
276 * blkio cgroup support
280 unsigned int cgroup_weight;
281 unsigned int cgroup_was_created;
284 #define FIO_VERROR_SIZE 128
287 * This describes a single thread/process executing a fio job.
290 struct thread_options o;
291 char verror[FIO_VERROR_SIZE];
295 struct thread_stat ts;
296 struct fio_file **files;
297 unsigned int files_size;
298 unsigned int files_index;
299 unsigned int nr_open_files;
300 unsigned int nr_done_files;
301 unsigned int nr_normal_files;
303 unsigned int next_file;
304 os_random_state_t next_file_state;
310 size_t orig_buffer_size;
311 volatile int terminate;
312 volatile int runstate;
314 unsigned int ioprio_set;
315 unsigned int last_was_sync;
325 unsigned long rand_seeds[6];
327 os_random_state_t bsrange_state;
328 os_random_state_t verify_state;
333 * IO engine hooks, contains everything needed to submit an io_u
334 * to any of the available IO engines.
336 struct ioengine_ops *io_ops;
339 * Current IO depth and list of free and busy io_u's.
341 unsigned int cur_depth;
342 unsigned int io_u_queued;
343 struct flist_head io_u_freelist;
344 struct flist_head io_u_busylist;
345 struct flist_head io_u_requeues;
346 pthread_mutex_t io_u_lock;
347 pthread_cond_t free_cond;
350 * async verify offload
352 struct flist_head verify_list;
353 pthread_t *verify_threads;
354 unsigned int nr_verify_threads;
355 pthread_cond_t verify_cond;
356 int verify_thread_exit;
361 unsigned long rate_usec_cycle[2];
362 long rate_pending_usleep[2];
363 unsigned long rate_bytes[2];
364 unsigned long rate_blocks[2];
365 struct timeval lastrate[2];
367 unsigned long long total_io_size;
369 unsigned long io_issues[2];
370 unsigned long long io_blocks[2];
371 unsigned long long io_bytes[2];
372 unsigned long long io_skip_bytes;
373 unsigned long long this_io_bytes[2];
374 unsigned long long zone_bytes;
375 struct fio_mutex *mutex;
378 * State for random io, a bitmap of blocks done vs not done
380 os_random_state_t random_state;
382 struct timeval start; /* start of this loop */
383 struct timeval epoch; /* time job was started */
384 struct timeval last_issue;
385 struct timeval tv_cache;
386 unsigned int tv_cache_nr;
387 unsigned int tv_cache_mask;
388 unsigned int ramp_time_over;
391 * read/write mixed workload state
393 os_random_state_t rwmix_state;
394 unsigned long rwmix_issues;
395 enum fio_ddir rwmix_ddir;
396 unsigned int ddir_nr;
399 * IO history logs for verification. We use a tree for sorting,
400 * if we are overwriting. Otherwise just use a fifo.
402 struct rb_root io_hist_tree;
403 struct flist_head io_hist_list;
408 struct flist_head io_log_list;
411 * for fileservice, how often to switch to a new file
413 unsigned int file_service_nr;
414 unsigned int file_service_left;
415 struct fio_file *file_service_file;
418 * For generating file sizes
420 os_random_state_t file_size_state;
425 unsigned int total_err_count;
430 * when should interactive ETA output be generated
438 #define __td_verror(td, err, msg, func) \
444 if (!(td)->first_error) \
445 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
449 #define td_clear_error(td) \
451 #define td_verror(td, err, func) \
452 __td_verror((td), (err), strerror((err)), (func))
453 #define td_vmsg(td, err, msg, func) \
454 __td_verror((td), (err), (msg), (func))
456 extern int exitall_on_terminate;
457 extern int thread_number;
458 extern int nr_process, nr_thread;
461 extern int terse_output;
462 extern int temp_stall_ts;
463 extern unsigned long long mlock_size;
464 extern unsigned long page_mask, page_size;
465 extern int read_only;
466 extern int eta_print;
467 extern unsigned long done_secs;
468 extern char *job_section;
469 extern int fio_gtod_offload;
470 extern int fio_gtod_cpu;
472 extern struct thread_data *threads;
474 static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
476 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
479 #define BLOCKS_PER_MAP (8 * sizeof(int))
480 #define TO_MAP_BLOCK(f, b) (b)
481 #define RAND_MAP_IDX(f, b) (TO_MAP_BLOCK(f, b) / BLOCKS_PER_MAP)
482 #define RAND_MAP_BIT(f, b) (TO_MAP_BLOCK(f, b) & (BLOCKS_PER_MAP - 1))
484 #define MAX_JOBS (1024)
486 #define td_non_fatal_error(e) ((e) == EIO || (e) == EILSEQ)
488 static inline void update_error_count(struct thread_data *td, int err)
490 td->total_err_count++;
491 if (td->total_err_count == 1)
492 td->first_error = err;
495 static inline int should_fsync(struct thread_data *td)
497 if (td->last_was_sync)
501 if (td_write(td) || td_rw(td) || td->o.override_sync)
510 extern unsigned long long utime_since(struct timeval *, struct timeval *);
511 extern unsigned long long utime_since_now(struct timeval *);
512 extern unsigned long mtime_since(struct timeval *, struct timeval *);
513 extern unsigned long mtime_since_now(struct timeval *);
514 extern unsigned long time_since_now(struct timeval *);
515 extern unsigned long mtime_since_genesis(void);
516 extern void usec_spin(unsigned int);
517 extern void usec_sleep(struct thread_data *, unsigned long);
518 extern void fill_start_time(struct timeval *);
519 extern void fio_gettime(struct timeval *, void *);
520 extern void fio_gtod_init(void);
521 extern void fio_gtod_update(void);
522 extern void set_genesis_time(void);
523 extern int ramp_time_over(struct thread_data *);
524 extern int in_ramp_time(struct thread_data *);
527 * Init/option functions
529 extern int __must_check parse_options(int, char **);
530 extern int fio_options_parse(struct thread_data *, char **, int);
531 extern void fio_keywords_init(void);
532 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
533 extern void fio_fill_default_options(struct thread_data *);
534 extern int fio_show_option_help(const char *);
535 extern void fio_options_dup_and_init(struct option *);
536 extern void options_mem_dupe(struct thread_data *);
537 extern void options_mem_free(struct thread_data *);
538 extern void td_fill_rand_seeds(struct thread_data *);
539 #define FIO_GETOPT_JOB 0x89988998
540 #define FIO_NR_OPTIONS 128
545 extern void print_thread_status(void);
546 extern void print_status_init(int);
549 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
550 * will never back again. It may cycle between running/verififying/fsyncing.
551 * Once the thread reaches TD_EXITED, it is just waiting for the core to
567 extern void td_set_runstate(struct thread_data *, int);
572 extern int __must_check fio_pin_memory(void);
573 extern void fio_unpin_memory(void);
574 extern int __must_check allocate_io_mem(struct thread_data *);
575 extern void free_io_mem(struct thread_data *);
578 * Reset stats after ramp time completes
580 extern void reset_all_stats(struct thread_data *);
585 #ifdef FIO_HAVE_BLKTRACE
586 extern int is_blktrace(const char *);
587 extern int load_blktrace(struct thread_data *, const char *);
591 * Mark unused variables passed to ops functions as unused, to silence gcc
593 #define fio_unused __attribute((__unused__))
594 #define fio_init __attribute__((constructor))
595 #define fio_exit __attribute__((destructor))
597 #define for_each_td(td, i) \
598 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
599 #define for_each_file(td, f, i) \
600 if ((td)->files_index) \
601 for ((i) = 0, (f) = (td)->files[0]; \
602 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
605 #define fio_assert(td, cond) do { \
608 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
609 (td)->runstate = TD_EXITED; \
610 (td)->error = EFAULT; \
615 static inline int fio_fill_issue_time(struct thread_data *td)
617 if (td->o.read_iolog_file ||
618 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
625 * Cheesy number->string conversion, complete with carry rounding error.
627 static inline char *num2str(unsigned long num, int maxlen, int base, int pow2)
629 char postfix[] = { ' ', 'K', 'M', 'G', 'P', 'E' };
630 unsigned int thousand;
641 for (i = 0; base > 1; i++)
647 len = sprintf(buf, "%'lu", num);
650 buf[len] = postfix[i];
656 if ((num % thousand) >= (thousand / 2))
667 static inline int __should_check_rate(struct thread_data *td,
670 struct thread_options *o = &td->o;
673 * If some rate setting was given, we need to check it
675 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
676 o->rate_iops_min[ddir])
682 static inline int should_check_rate(struct thread_data *td,
683 unsigned long *bytes_done)
688 ret |= __should_check_rate(td, 0);
690 ret |= __should_check_rate(td, 1);
695 static inline int is_power_of_2(unsigned int val)
697 return (val != 0 && ((val & (val - 1)) == 0));
701 * We currently only need to do locking if we have verifier threads
702 * accessing our internal structures too
704 static inline void td_io_u_lock(struct thread_data *td)
706 if (td->o.verify_async)
707 pthread_mutex_lock(&td->io_u_lock);
710 static inline void td_io_u_unlock(struct thread_data *td)
712 if (td->o.verify_async)
713 pthread_mutex_unlock(&td->io_u_lock);
716 static inline void td_io_u_free_notify(struct thread_data *td)
718 if (td->o.verify_async)
719 pthread_cond_signal(&td->free_cond);