8 #include <sys/resource.h>
18 #include "compiler/compiler.h"
22 #include "arch/arch.h"
37 #ifdef FIO_HAVE_SOLARISAIO
38 #include <sys/asynch.h>
41 struct 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];
50 * What type of allocation to use for io buffers
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 */
61 * How many depth levels to log
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
67 #define MAX_PATTERN_SIZE 512
78 struct io_log *slat_log;
79 struct io_log *clat_log;
80 struct io_log *bw_log;
83 * bandwidth and latency stats
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 */
89 unsigned long long stat_io_bytes[2];
90 struct timeval stat_sample_time[2];
93 * fio system usage accounting
95 struct rusage ru_start;
97 unsigned long usr_time;
98 unsigned long sys_time;
100 unsigned long minf, majf;
103 * IO depth and latency stats
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;
115 unsigned long long io_bytes[2];
116 unsigned long runtime[2];
117 unsigned long total_run_time;
120 * IO Error related stats
122 unsigned continue_on_error;
123 unsigned long total_err_count;
126 unsigned int kb_base;
134 struct thread_options {
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;
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;
158 unsigned int min_bs[2];
159 unsigned int max_bs[2];
160 struct bssplit *bssplit[2];
161 unsigned int bssplit_nr[2];
163 unsigned int nr_files;
164 unsigned int open_files;
165 enum file_lock_mode file_lock_mode;
166 unsigned int lockfile_batch;
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;
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;
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;
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;
210 unsigned long long zone_size;
211 unsigned long long zone_skip;
212 enum fio_memtype mem_type;
213 unsigned int mem_align;
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;
223 unsigned int rwmixcycle;
224 unsigned int rwmix[2];
226 unsigned int file_service_type;
227 unsigned int group_reporting;
228 unsigned int fadvise_hint;
229 unsigned int fallocate;
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
276 * blkio cgroup support
279 unsigned int cgroup_weight;
285 #define FIO_VERROR_SIZE 128
288 * This describes a single thread/process executing a fio job.
291 struct thread_options o;
292 char verror[FIO_VERROR_SIZE];
296 struct thread_stat ts;
297 struct fio_file **files;
298 unsigned int files_size;
299 unsigned int files_index;
300 unsigned int nr_open_files;
301 unsigned int nr_done_files;
302 unsigned int nr_normal_files;
304 unsigned int next_file;
305 os_random_state_t next_file_state;
311 size_t orig_buffer_size;
312 volatile int terminate;
313 volatile int runstate;
315 unsigned int ioprio_set;
316 unsigned int last_was_sync;
326 unsigned long rand_seeds[6];
328 os_random_state_t bsrange_state;
329 os_random_state_t verify_state;
334 * IO engine hooks, contains everything needed to submit an io_u
335 * to any of the available IO engines.
337 struct ioengine_ops *io_ops;
340 * Current IO depth and list of free and busy io_u's.
342 unsigned int cur_depth;
343 unsigned int io_u_queued;
344 struct flist_head io_u_freelist;
345 struct flist_head io_u_busylist;
346 struct flist_head io_u_requeues;
347 pthread_mutex_t io_u_lock;
348 pthread_cond_t free_cond;
351 * async verify offload
353 struct flist_head verify_list;
354 pthread_t *verify_threads;
355 unsigned int nr_verify_threads;
356 pthread_cond_t verify_cond;
357 int verify_thread_exit;
362 unsigned long rate_nsec_cycle[2];
363 long rate_pending_usleep[2];
364 unsigned long rate_bytes[2];
365 unsigned long rate_blocks[2];
366 struct timeval lastrate[2];
368 unsigned long long total_io_size;
370 unsigned long io_issues[2];
371 unsigned long long io_blocks[2];
372 unsigned long long io_bytes[2];
373 unsigned long long io_skip_bytes;
374 unsigned long long this_io_bytes[2];
375 unsigned long long zone_bytes;
376 struct fio_mutex *mutex;
379 * State for random io, a bitmap of blocks done vs not done
381 os_random_state_t random_state;
383 struct timeval start; /* start of this loop */
384 struct timeval epoch; /* time job was started */
385 struct timeval last_issue;
386 struct timeval tv_cache;
387 unsigned int tv_cache_nr;
388 unsigned int tv_cache_mask;
389 unsigned int ramp_time_over;
392 * read/write mixed workload state
394 os_random_state_t rwmix_state;
395 unsigned long rwmix_issues;
396 enum fio_ddir rwmix_ddir;
397 unsigned int ddir_nr;
400 * IO history logs for verification. We use a tree for sorting,
401 * if we are overwriting. Otherwise just use a fifo.
403 struct rb_root io_hist_tree;
404 struct flist_head io_hist_list;
409 struct flist_head io_log_list;
412 * for fileservice, how often to switch to a new file
414 unsigned int file_service_nr;
415 unsigned int file_service_left;
416 struct fio_file *file_service_file;
419 * For generating file sizes
421 os_random_state_t file_size_state;
426 unsigned int total_err_count;
431 * when should interactive ETA output be generated
439 #define __td_verror(td, err, msg, func) \
445 if (!(td)->first_error) \
446 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
450 #define td_clear_error(td) \
452 #define td_verror(td, err, func) \
453 __td_verror((td), (err), strerror((err)), (func))
454 #define td_vmsg(td, err, msg, func) \
455 __td_verror((td), (err), (msg), (func))
457 extern int exitall_on_terminate;
458 extern int thread_number;
459 extern int nr_process, nr_thread;
462 extern int terse_output;
463 extern int temp_stall_ts;
464 extern unsigned long long mlock_size;
465 extern unsigned long page_mask, page_size;
466 extern int read_only;
467 extern int eta_print;
468 extern unsigned long done_secs;
469 extern char *job_section;
470 extern int fio_gtod_offload;
471 extern int fio_gtod_cpu;
473 extern struct thread_data *threads;
475 static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
477 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
480 #define BLOCKS_PER_MAP (8 * sizeof(int))
481 #define TO_MAP_BLOCK(f, b) (b)
482 #define RAND_MAP_IDX(f, b) (TO_MAP_BLOCK(f, b) / BLOCKS_PER_MAP)
483 #define RAND_MAP_BIT(f, b) (TO_MAP_BLOCK(f, b) & (BLOCKS_PER_MAP - 1))
485 #define MAX_JOBS (1024)
487 #define td_non_fatal_error(e) ((e) == EIO || (e) == EILSEQ)
489 static inline void update_error_count(struct thread_data *td, int err)
491 td->total_err_count++;
492 if (td->total_err_count == 1)
493 td->first_error = err;
496 static inline int should_fsync(struct thread_data *td)
498 if (td->last_was_sync)
502 if (td_write(td) || td_rw(td) || td->o.override_sync)
511 extern unsigned long long utime_since(struct timeval *, struct timeval *);
512 extern unsigned long long utime_since_now(struct timeval *);
513 extern unsigned long mtime_since(struct timeval *, struct timeval *);
514 extern unsigned long mtime_since_now(struct timeval *);
515 extern unsigned long time_since_now(struct timeval *);
516 extern unsigned long mtime_since_genesis(void);
517 extern void usec_spin(unsigned int);
518 extern void usec_sleep(struct thread_data *, unsigned long);
519 extern void fill_start_time(struct timeval *);
520 extern void fio_gettime(struct timeval *, void *);
521 extern void fio_gtod_init(void);
522 extern void fio_gtod_update(void);
523 extern void set_genesis_time(void);
524 extern int ramp_time_over(struct thread_data *);
525 extern int in_ramp_time(struct thread_data *);
528 * Init/option functions
530 extern int __must_check parse_options(int, char **);
531 extern int fio_options_parse(struct thread_data *, char **, int);
532 extern void fio_keywords_init(void);
533 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
534 extern void fio_fill_default_options(struct thread_data *);
535 extern int fio_show_option_help(const char *);
536 extern void fio_options_dup_and_init(struct option *);
537 extern void options_mem_dupe(struct thread_data *);
538 extern void options_mem_free(struct thread_data *);
539 extern void td_fill_rand_seeds(struct thread_data *);
540 extern void add_job_opts(const char **);
541 #define FIO_GETOPT_JOB 0x89988998
542 #define FIO_NR_OPTIONS 128
547 extern void print_thread_status(void);
548 extern void print_status_init(int);
551 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
552 * will never back again. It may cycle between running/verififying/fsyncing.
553 * Once the thread reaches TD_EXITED, it is just waiting for the core to
569 extern void td_set_runstate(struct thread_data *, int);
574 extern int __must_check fio_pin_memory(void);
575 extern void fio_unpin_memory(void);
576 extern int __must_check allocate_io_mem(struct thread_data *);
577 extern void free_io_mem(struct thread_data *);
580 * Reset stats after ramp time completes
582 extern void reset_all_stats(struct thread_data *);
587 #ifdef FIO_HAVE_BLKTRACE
588 extern int is_blktrace(const char *);
589 extern int load_blktrace(struct thread_data *, const char *);
593 * Mark unused variables passed to ops functions as unused, to silence gcc
595 #define fio_unused __attribute((__unused__))
596 #define fio_init __attribute__((constructor))
597 #define fio_exit __attribute__((destructor))
599 #define for_each_td(td, i) \
600 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
601 #define for_each_file(td, f, i) \
602 if ((td)->files_index) \
603 for ((i) = 0, (f) = (td)->files[0]; \
604 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
607 #define fio_assert(td, cond) do { \
610 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
611 (td)->runstate = TD_EXITED; \
612 (td)->error = EFAULT; \
617 static inline int fio_fill_issue_time(struct thread_data *td)
619 if (td->o.read_iolog_file ||
620 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
627 * Cheesy number->string conversion, complete with carry rounding error.
629 static inline char *num2str(unsigned long num, int maxlen, int base, int pow2)
631 char postfix[] = { ' ', 'K', 'M', 'G', 'P', 'E' };
632 unsigned int thousand;
643 for (i = 0; base > 1; i++)
649 len = sprintf(buf, "%'lu", num);
652 buf[len] = postfix[i];
658 if ((num % thousand) >= (thousand / 2))
669 static inline int __should_check_rate(struct thread_data *td,
672 struct thread_options *o = &td->o;
675 * If some rate setting was given, we need to check it
677 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
678 o->rate_iops_min[ddir])
684 static inline int should_check_rate(struct thread_data *td,
685 unsigned long *bytes_done)
690 ret |= __should_check_rate(td, 0);
692 ret |= __should_check_rate(td, 1);
697 static inline int is_power_of_2(unsigned int val)
699 return (val != 0 && ((val & (val - 1)) == 0));
703 * We currently only need to do locking if we have verifier threads
704 * accessing our internal structures too
706 static inline void td_io_u_lock(struct thread_data *td)
708 if (td->o.verify_async)
709 pthread_mutex_lock(&td->io_u_lock);
712 static inline void td_io_u_unlock(struct thread_data *td)
714 if (td->o.verify_async)
715 pthread_mutex_unlock(&td->io_u_lock);
718 static inline void td_io_u_free_notify(struct thread_data *td)
720 if (td->o.verify_async)
721 pthread_cond_signal(&td->free_cond);