8 #include <sys/resource.h>
18 #include "compiler/compiler.h"
22 #include "arch/arch.h"
39 #ifdef FIO_HAVE_SOLARISAIO
40 #include <sys/asynch.h>
43 struct group_run_stats {
44 unsigned long long max_run[2], min_run[2];
45 unsigned long long max_bw[2], min_bw[2];
46 unsigned long long io_kb[2];
47 unsigned long long agg[2];
52 * What type of allocation to use for io buffers
55 MEM_MALLOC = 0, /* ordinary malloc */
56 MEM_SHM, /* use shared memory segments */
57 MEM_SHMHUGE, /* use shared memory segments with huge pages */
58 MEM_MMAP, /* use anonynomous mmap */
59 MEM_MMAPHUGE, /* memory mapped huge file */
63 * How many depth levels to log
65 #define FIO_IO_U_MAP_NR 8
66 #define FIO_IO_U_LAT_U_NR 10
67 #define FIO_IO_U_LAT_M_NR 12
69 #define MAX_PATTERN_SIZE 512
80 struct io_log *slat_log;
81 struct io_log *clat_log;
82 struct io_log *bw_log;
85 * bandwidth and latency stats
87 struct io_stat clat_stat[2]; /* completion latency */
88 struct io_stat slat_stat[2]; /* submission latency */
89 struct io_stat bw_stat[2]; /* bandwidth stats */
91 unsigned long long stat_io_bytes[2];
92 struct timeval stat_sample_time[2];
95 * fio system usage accounting
97 struct rusage ru_start;
99 unsigned long usr_time;
100 unsigned long sys_time;
102 unsigned long minf, majf;
105 * IO depth and latency stats
107 unsigned int io_u_map[FIO_IO_U_MAP_NR];
108 unsigned int io_u_submit[FIO_IO_U_MAP_NR];
109 unsigned int io_u_complete[FIO_IO_U_MAP_NR];
110 unsigned int io_u_lat_u[FIO_IO_U_LAT_U_NR];
111 unsigned int io_u_lat_m[FIO_IO_U_LAT_M_NR];
112 unsigned long total_io_u[2];
113 unsigned long short_io_u[2];
114 unsigned long total_submit;
115 unsigned long total_complete;
117 unsigned long long io_bytes[2];
118 unsigned long runtime[2];
119 unsigned long total_run_time;
122 * IO Error related stats
124 unsigned continue_on_error;
125 unsigned long total_err_count;
128 unsigned int kb_base;
136 struct thread_options {
144 enum td_ddir td_ddir;
145 unsigned int kb_base;
146 unsigned int ddir_nr;
147 unsigned int iodepth;
148 unsigned int iodepth_low;
149 unsigned int iodepth_batch;
150 unsigned int iodepth_batch_complete;
152 unsigned long long size;
153 unsigned int fill_device;
154 unsigned long long file_size_low;
155 unsigned long long file_size_high;
156 unsigned long long start_offset;
160 unsigned int min_bs[2];
161 unsigned int max_bs[2];
162 struct bssplit *bssplit[2];
163 unsigned int bssplit_nr[2];
165 unsigned int nr_files;
166 unsigned int open_files;
167 enum file_lock_mode file_lock_mode;
168 unsigned int lockfile_batch;
170 unsigned int odirect;
171 unsigned int invalidate_cache;
172 unsigned int create_serialize;
173 unsigned int create_fsync;
174 unsigned int create_on_open;
175 unsigned int end_fsync;
176 unsigned int pre_read;
177 unsigned int sync_io;
179 unsigned int do_verify;
180 unsigned int verifysort;
181 unsigned int verify_interval;
182 unsigned int verify_offset;
183 char verify_pattern[MAX_PATTERN_SIZE];
184 unsigned int verify_pattern_bytes;
185 unsigned int verify_fatal;
186 unsigned int verify_async;
187 unsigned int use_thread;
189 unsigned int do_disk_util;
190 unsigned int override_sync;
191 unsigned int rand_repeatable;
192 unsigned int write_lat_log;
193 unsigned int write_bw_log;
194 unsigned int norandommap;
195 unsigned int softrandommap;
196 unsigned int bs_unaligned;
197 unsigned int fsync_on_close;
199 unsigned int hugepage_size;
200 unsigned int rw_min_bs;
201 unsigned int thinktime;
202 unsigned int thinktime_spin;
203 unsigned int thinktime_blocks;
204 unsigned int fsync_blocks;
205 unsigned int fdatasync_blocks;
206 unsigned int start_delay;
207 unsigned long long timeout;
208 unsigned long long ramp_time;
209 unsigned int overwrite;
210 unsigned int bw_avg_time;
212 unsigned long long zone_size;
213 unsigned long long zone_skip;
214 enum fio_memtype mem_type;
215 unsigned int mem_align;
217 unsigned int stonewall;
218 unsigned int new_group;
219 unsigned int numjobs;
220 os_cpu_mask_t cpumask;
221 unsigned int cpumask_set;
222 os_cpu_mask_t verify_cpumask;
223 unsigned int verify_cpumask_set;
225 unsigned int rwmixcycle;
226 unsigned int rwmix[2];
228 unsigned int file_service_type;
229 unsigned int group_reporting;
230 unsigned int fadvise_hint;
231 unsigned int fallocate;
232 unsigned int zero_buffers;
233 unsigned int refill_buffers;
234 unsigned int time_based;
235 unsigned int disable_clat;
236 unsigned int disable_slat;
237 unsigned int disable_bw;
238 unsigned int gtod_reduce;
239 unsigned int gtod_cpu;
240 unsigned int gtod_offload;
242 char *read_iolog_file;
243 char *write_iolog_file;
248 * Pre-run and post-run shell
253 unsigned int rate[2];
254 unsigned int ratemin[2];
255 unsigned int ratecycle;
256 unsigned int rate_iops[2];
257 unsigned int rate_iops_min[2];
262 * CPU "io" cycle burner
264 unsigned int cpuload;
265 unsigned int cpucycle;
270 unsigned int continue_on_error;
273 * Benchmark profile type
278 * blkio cgroup support
281 unsigned int cgroup_weight;
286 unsigned int sync_file_range;
289 #define FIO_VERROR_SIZE 128
292 * This describes a single thread/process executing a fio job.
295 struct thread_options o;
296 char verror[FIO_VERROR_SIZE];
300 struct thread_stat ts;
301 struct fio_file **files;
302 unsigned int files_size;
303 unsigned int files_index;
304 unsigned int nr_open_files;
305 unsigned int nr_done_files;
306 unsigned int nr_normal_files;
308 unsigned int next_file;
309 os_random_state_t next_file_state;
315 size_t orig_buffer_size;
316 volatile int terminate;
317 volatile int runstate;
319 unsigned int ioprio_set;
320 unsigned int last_was_sync;
330 unsigned long rand_seeds[6];
332 os_random_state_t bsrange_state;
333 os_random_state_t verify_state;
338 * IO engine hooks, contains everything needed to submit an io_u
339 * to any of the available IO engines.
341 struct ioengine_ops *io_ops;
344 * Current IO depth and list of free and busy io_u's.
346 unsigned int cur_depth;
347 unsigned int io_u_queued;
348 struct flist_head io_u_freelist;
349 struct flist_head io_u_busylist;
350 struct flist_head io_u_requeues;
351 pthread_mutex_t io_u_lock;
352 pthread_cond_t free_cond;
355 * async verify offload
357 struct flist_head verify_list;
358 pthread_t *verify_threads;
359 unsigned int nr_verify_threads;
360 pthread_cond_t verify_cond;
361 int verify_thread_exit;
366 unsigned long rate_nsec_cycle[2];
367 long rate_pending_usleep[2];
368 unsigned long rate_bytes[2];
369 unsigned long rate_blocks[2];
370 struct timeval lastrate[2];
372 unsigned long long total_io_size;
374 unsigned long io_issues[2];
375 unsigned long long io_blocks[2];
376 unsigned long long io_bytes[2];
377 unsigned long long io_skip_bytes;
378 unsigned long long this_io_bytes[2];
379 unsigned long long zone_bytes;
380 struct fio_mutex *mutex;
383 * State for random io, a bitmap of blocks done vs not done
385 os_random_state_t random_state;
387 struct timeval start; /* start of this loop */
388 struct timeval epoch; /* time job was started */
389 struct timeval last_issue;
390 struct timeval tv_cache;
391 unsigned int tv_cache_nr;
392 unsigned int tv_cache_mask;
393 unsigned int ramp_time_over;
396 * read/write mixed workload state
398 os_random_state_t rwmix_state;
399 unsigned long rwmix_issues;
400 enum fio_ddir rwmix_ddir;
401 unsigned int ddir_nr;
404 * IO history logs for verification. We use a tree for sorting,
405 * if we are overwriting. Otherwise just use a fifo.
407 struct rb_root io_hist_tree;
408 struct flist_head io_hist_list;
413 struct flist_head io_log_list;
416 * for fileservice, how often to switch to a new file
418 unsigned int file_service_nr;
419 unsigned int file_service_left;
420 struct fio_file *file_service_file;
422 unsigned int sync_file_range_nr;
425 * For generating file sizes
427 os_random_state_t file_size_state;
432 unsigned int total_err_count;
436 * Can be overloaded by profiles
438 struct prof_io_ops prof_io_ops;
443 * when should interactive ETA output be generated
451 #define __td_verror(td, err, msg, func) \
457 if (!(td)->first_error) \
458 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
462 #define td_clear_error(td) \
464 #define td_verror(td, err, func) \
465 __td_verror((td), (err), strerror((err)), (func))
466 #define td_vmsg(td, err, msg, func) \
467 __td_verror((td), (err), (msg), (func))
469 extern int exitall_on_terminate;
470 extern int thread_number;
471 extern int nr_process, nr_thread;
474 extern int terse_output;
475 extern int temp_stall_ts;
476 extern unsigned long long mlock_size;
477 extern unsigned long page_mask, page_size;
478 extern int read_only;
479 extern int eta_print;
480 extern unsigned long done_secs;
481 extern char *job_section;
482 extern int fio_gtod_offload;
483 extern int fio_gtod_cpu;
485 extern struct thread_data *threads;
487 static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
489 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
492 #define BLOCKS_PER_MAP (8 * sizeof(int))
493 #define TO_MAP_BLOCK(f, b) (b)
494 #define RAND_MAP_IDX(f, b) (TO_MAP_BLOCK(f, b) / BLOCKS_PER_MAP)
495 #define RAND_MAP_BIT(f, b) (TO_MAP_BLOCK(f, b) & (BLOCKS_PER_MAP - 1))
497 #define MAX_JOBS (1024)
499 #define td_non_fatal_error(e) ((e) == EIO || (e) == EILSEQ)
501 static inline void update_error_count(struct thread_data *td, int err)
503 td->total_err_count++;
504 if (td->total_err_count == 1)
505 td->first_error = err;
508 static inline int should_fsync(struct thread_data *td)
510 if (td->last_was_sync)
514 if (td_write(td) || td_rw(td) || td->o.override_sync)
523 extern unsigned long long utime_since(struct timeval *, struct timeval *);
524 extern unsigned long long utime_since_now(struct timeval *);
525 extern unsigned long mtime_since(struct timeval *, struct timeval *);
526 extern unsigned long mtime_since_now(struct timeval *);
527 extern unsigned long time_since_now(struct timeval *);
528 extern unsigned long mtime_since_genesis(void);
529 extern void usec_spin(unsigned int);
530 extern void usec_sleep(struct thread_data *, unsigned long);
531 extern void fill_start_time(struct timeval *);
532 extern void fio_gettime(struct timeval *, void *);
533 extern void fio_gtod_init(void);
534 extern void fio_gtod_update(void);
535 extern void set_genesis_time(void);
536 extern int ramp_time_over(struct thread_data *);
537 extern int in_ramp_time(struct thread_data *);
540 * Init/option functions
542 extern int __must_check parse_options(int, char **);
543 extern int fio_options_parse(struct thread_data *, char **, int);
544 extern void fio_keywords_init(void);
545 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
546 extern void fio_fill_default_options(struct thread_data *);
547 extern int fio_show_option_help(const char *);
548 extern void fio_options_dup_and_init(struct option *);
549 extern void options_mem_dupe(struct thread_data *);
550 extern void options_mem_free(struct thread_data *);
551 extern void td_fill_rand_seeds(struct thread_data *);
552 extern void add_job_opts(const char **);
553 #define FIO_GETOPT_JOB 0x89988998
554 #define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
559 extern void print_thread_status(void);
560 extern void print_status_init(int);
563 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
564 * will never back again. It may cycle between running/verififying/fsyncing.
565 * Once the thread reaches TD_EXITED, it is just waiting for the core to
581 extern void td_set_runstate(struct thread_data *, int);
586 extern int __must_check fio_pin_memory(void);
587 extern void fio_unpin_memory(void);
588 extern int __must_check allocate_io_mem(struct thread_data *);
589 extern void free_io_mem(struct thread_data *);
592 * Reset stats after ramp time completes
594 extern void reset_all_stats(struct thread_data *);
599 #ifdef FIO_HAVE_BLKTRACE
600 extern int is_blktrace(const char *);
601 extern int load_blktrace(struct thread_data *, const char *);
605 * Mark unused variables passed to ops functions as unused, to silence gcc
607 #define fio_unused __attribute((__unused__))
608 #define fio_init __attribute__((constructor))
609 #define fio_exit __attribute__((destructor))
611 #define for_each_td(td, i) \
612 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
613 #define for_each_file(td, f, i) \
614 if ((td)->files_index) \
615 for ((i) = 0, (f) = (td)->files[0]; \
616 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
619 #define fio_assert(td, cond) do { \
622 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
623 (td)->runstate = TD_EXITED; \
624 (td)->error = EFAULT; \
629 static inline int fio_fill_issue_time(struct thread_data *td)
631 if (td->o.read_iolog_file ||
632 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
639 * Cheesy number->string conversion, complete with carry rounding error.
641 static inline char *num2str(unsigned long num, int maxlen, int base, int pow2)
643 char postfix[] = { ' ', 'K', 'M', 'G', 'P', 'E' };
644 unsigned int thousand;
655 for (i = 0; base > 1; i++)
661 len = sprintf(buf, "%'lu", num);
664 buf[len] = postfix[i];
670 if ((num % thousand) >= (thousand / 2))
681 static inline int __should_check_rate(struct thread_data *td,
684 struct thread_options *o = &td->o;
687 * If some rate setting was given, we need to check it
689 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
690 o->rate_iops_min[ddir])
696 static inline int should_check_rate(struct thread_data *td,
697 unsigned long *bytes_done)
702 ret |= __should_check_rate(td, 0);
704 ret |= __should_check_rate(td, 1);
709 static inline int is_power_of_2(unsigned int val)
711 return (val != 0 && ((val & (val - 1)) == 0));
715 * We currently only need to do locking if we have verifier threads
716 * accessing our internal structures too
718 static inline void td_io_u_lock(struct thread_data *td)
720 if (td->o.verify_async)
721 pthread_mutex_lock(&td->io_u_lock);
724 static inline void td_io_u_unlock(struct thread_data *td)
726 if (td->o.verify_async)
727 pthread_mutex_unlock(&td->io_u_lock);
730 static inline void td_io_u_free_notify(struct thread_data *td)
732 if (td->o.verify_async)
733 pthread_cond_signal(&td->free_cond);