8 #include <sys/resource.h>
19 #include "compiler/compiler.h"
23 #include "arch/arch.h"
36 #include "lib/getopt.h"
43 #ifdef FIO_HAVE_SOLARISAIO
44 #include <sys/asynch.h>
47 struct group_run_stats {
48 unsigned long long max_run[2], min_run[2];
49 unsigned long long max_bw[2], min_bw[2];
50 unsigned long long io_kb[2];
51 unsigned long long agg[2];
56 * What type of allocation to use for io buffers
59 MEM_MALLOC = 0, /* ordinary malloc */
60 MEM_SHM, /* use shared memory segments */
61 MEM_SHMHUGE, /* use shared memory segments with huge pages */
62 MEM_MMAP, /* use anonynomous mmap */
63 MEM_MMAPHUGE, /* memory mapped huge file */
67 * offset generator types
75 * How many depth levels to log
77 #define FIO_IO_U_MAP_NR 7
78 #define FIO_IO_U_LAT_U_NR 10
79 #define FIO_IO_U_LAT_M_NR 12
81 #define MAX_PATTERN_SIZE 512
92 struct io_log *slat_log;
93 struct io_log *clat_log;
94 struct io_log *lat_log;
95 struct io_log *bw_log;
98 * bandwidth and latency stats
100 struct io_stat clat_stat[2]; /* completion latency */
101 struct io_stat slat_stat[2]; /* submission latency */
102 struct io_stat lat_stat[2]; /* total latency */
103 struct io_stat bw_stat[2]; /* bandwidth stats */
105 unsigned long long stat_io_bytes[2];
106 struct timeval stat_sample_time[2];
109 * fio system usage accounting
111 struct rusage ru_start;
112 struct rusage ru_end;
113 unsigned long usr_time;
114 unsigned long sys_time;
116 unsigned long minf, majf;
119 * IO depth and latency stats
121 unsigned int io_u_map[FIO_IO_U_MAP_NR];
122 unsigned int io_u_submit[FIO_IO_U_MAP_NR];
123 unsigned int io_u_complete[FIO_IO_U_MAP_NR];
124 unsigned int io_u_lat_u[FIO_IO_U_LAT_U_NR];
125 unsigned int io_u_lat_m[FIO_IO_U_LAT_M_NR];
126 unsigned long total_io_u[3];
127 unsigned long short_io_u[3];
128 unsigned long total_submit;
129 unsigned long total_complete;
131 unsigned long long io_bytes[2];
132 unsigned long long runtime[2];
133 unsigned long total_run_time;
136 * IO Error related stats
138 unsigned continue_on_error;
139 unsigned long total_err_count;
142 unsigned int kb_base;
150 struct thread_options {
158 enum td_ddir td_ddir;
160 unsigned int kb_base;
161 unsigned int ddir_seq_nr;
162 unsigned int iodepth;
163 unsigned int iodepth_low;
164 unsigned int iodepth_batch;
165 unsigned int iodepth_batch_complete;
167 unsigned long long size;
168 unsigned int fill_device;
169 unsigned long long file_size_low;
170 unsigned long long file_size_high;
171 unsigned long long start_offset;
175 unsigned int min_bs[2];
176 unsigned int max_bs[2];
177 struct bssplit *bssplit[2];
178 unsigned int bssplit_nr[2];
180 unsigned int nr_files;
181 unsigned int open_files;
182 enum file_lock_mode file_lock_mode;
183 unsigned int lockfile_batch;
185 unsigned int odirect;
186 unsigned int invalidate_cache;
187 unsigned int create_serialize;
188 unsigned int create_fsync;
189 unsigned int create_on_open;
190 unsigned int end_fsync;
191 unsigned int pre_read;
192 unsigned int sync_io;
194 unsigned int do_verify;
195 unsigned int verifysort;
196 unsigned int verify_interval;
197 unsigned int verify_offset;
198 char verify_pattern[MAX_PATTERN_SIZE];
199 unsigned int verify_pattern_bytes;
200 unsigned int verify_fatal;
201 unsigned int verify_dump;
202 unsigned int verify_async;
203 unsigned long long verify_backlog;
204 unsigned int verify_batch;
205 unsigned int use_thread;
207 unsigned int do_disk_util;
208 unsigned int override_sync;
209 unsigned int rand_repeatable;
210 unsigned int use_os_rand;
211 unsigned int write_lat_log;
212 unsigned int write_bw_log;
213 unsigned int norandommap;
214 unsigned int softrandommap;
215 unsigned int bs_unaligned;
216 unsigned int fsync_on_close;
218 unsigned int hugepage_size;
219 unsigned int rw_min_bs;
220 unsigned int thinktime;
221 unsigned int thinktime_spin;
222 unsigned int thinktime_blocks;
223 unsigned int fsync_blocks;
224 unsigned int fdatasync_blocks;
225 unsigned int barrier_blocks;
226 unsigned long long start_delay;
227 unsigned long long timeout;
228 unsigned long long ramp_time;
229 unsigned int overwrite;
230 unsigned int bw_avg_time;
232 unsigned long long zone_size;
233 unsigned long long zone_skip;
234 enum fio_memtype mem_type;
235 unsigned int mem_align;
237 unsigned int stonewall;
238 unsigned int new_group;
239 unsigned int numjobs;
240 os_cpu_mask_t cpumask;
241 unsigned int cpumask_set;
242 os_cpu_mask_t verify_cpumask;
243 unsigned int verify_cpumask_set;
245 unsigned int rwmixcycle;
246 unsigned int rwmix[2];
248 unsigned int file_service_type;
249 unsigned int group_reporting;
250 unsigned int fadvise_hint;
251 enum fio_fallocate_mode fallocate_mode;
252 unsigned int zero_buffers;
253 unsigned int refill_buffers;
254 unsigned int time_based;
255 unsigned int disable_lat;
256 unsigned int disable_clat;
257 unsigned int disable_slat;
258 unsigned int disable_bw;
259 unsigned int gtod_reduce;
260 unsigned int gtod_cpu;
261 unsigned int gtod_offload;
262 enum fio_cs clocksource;
263 unsigned int no_stall;
264 unsigned int trim_percentage;
265 unsigned int trim_batch;
266 unsigned int trim_zero;
267 unsigned long long trim_backlog;
269 char *read_iolog_file;
270 char *write_iolog_file;
273 char *replay_redirect;
276 * Pre-run and post-run shell
281 unsigned int rate[2];
282 unsigned int ratemin[2];
283 unsigned int ratecycle;
284 unsigned int rate_iops[2];
285 unsigned int rate_iops_min[2];
290 * CPU "io" cycle burner
292 unsigned int cpuload;
293 unsigned int cpucycle;
298 unsigned int continue_on_error;
301 * Benchmark profile type
306 * blkio cgroup support
309 unsigned int cgroup_weight;
310 unsigned int cgroup_nodelete;
315 unsigned int sync_file_range;
318 #define FIO_VERROR_SIZE 128
321 * This describes a single thread/process executing a fio job.
324 struct thread_options o;
325 char verror[FIO_VERROR_SIZE];
329 struct thread_stat ts;
330 struct fio_file **files;
331 unsigned int files_size;
332 unsigned int files_index;
333 unsigned int nr_open_files;
334 unsigned int nr_done_files;
335 unsigned int nr_normal_files;
337 unsigned int next_file;
338 os_random_state_t next_file_state;
339 struct frand_state __next_file_state;
345 size_t orig_buffer_size;
346 volatile int terminate;
347 volatile int runstate;
349 unsigned int ioprio_set;
350 unsigned int last_was_sync;
351 enum fio_ddir last_ddir;
361 unsigned long rand_seeds[7];
364 os_random_state_t bsrange_state;
365 struct frand_state __bsrange_state;
368 os_random_state_t verify_state;
369 struct frand_state __verify_state;
372 os_random_state_t trim_state;
373 struct frand_state __trim_state;
376 unsigned int verify_batch;
377 unsigned int trim_batch;
382 * IO engine hooks, contains everything needed to submit an io_u
383 * to any of the available IO engines.
385 struct ioengine_ops *io_ops;
388 * Current IO depth and list of free and busy io_u's.
390 unsigned int cur_depth;
391 unsigned int io_u_queued;
392 struct flist_head io_u_freelist;
393 struct flist_head io_u_busylist;
394 struct flist_head io_u_requeues;
395 pthread_mutex_t io_u_lock;
396 pthread_cond_t free_cond;
399 * async verify offload
401 struct flist_head verify_list;
402 pthread_t *verify_threads;
403 unsigned int nr_verify_threads;
404 pthread_cond_t verify_cond;
405 int verify_thread_exit;
410 unsigned long rate_nsec_cycle[2];
411 long rate_pending_usleep[2];
412 unsigned long rate_bytes[2];
413 unsigned long rate_blocks[2];
414 struct timeval lastrate[2];
416 unsigned long long total_io_size;
417 unsigned long long fill_device_size;
419 unsigned long io_issues[2];
420 unsigned long long io_blocks[2];
421 unsigned long long io_bytes[2];
422 unsigned long long io_skip_bytes;
423 unsigned long long this_io_bytes[2];
424 unsigned long long zone_bytes;
425 struct fio_mutex *mutex;
428 * State for random io, a bitmap of blocks done vs not done
431 os_random_state_t random_state;
432 struct frand_state __random_state;
435 struct timeval start; /* start of this loop */
436 struct timeval epoch; /* time job was started */
437 struct timeval last_issue;
438 struct timeval tv_cache;
439 unsigned int tv_cache_nr;
440 unsigned int tv_cache_mask;
441 unsigned int ramp_time_over;
444 * read/write mixed workload state
447 os_random_state_t rwmix_state;
448 struct frand_state __rwmix_state;
450 unsigned long rwmix_issues;
451 enum fio_ddir rwmix_ddir;
452 unsigned int ddir_seq_nr;
455 * IO history logs for verification. We use a tree for sorting,
456 * if we are overwriting. Otherwise just use a fifo.
458 struct rb_root io_hist_tree;
459 struct flist_head io_hist_list;
460 unsigned long io_hist_len;
465 struct flist_head io_log_list;
468 * For tracking/handling discards
470 struct flist_head trim_list;
471 unsigned long trim_entries;
474 * for fileservice, how often to switch to a new file
476 unsigned int file_service_nr;
477 unsigned int file_service_left;
478 struct fio_file *file_service_file;
480 unsigned int sync_file_range_nr;
483 * For generating file sizes
486 os_random_state_t file_size_state;
487 struct frand_state __file_size_state;
493 unsigned int total_err_count;
497 * Can be overloaded by profiles
499 struct prof_io_ops prof_io_ops;
504 * when should interactive ETA output be generated
512 #define __td_verror(td, err, msg, func) \
518 if (!(td)->first_error) \
519 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
523 #define td_clear_error(td) \
525 #define td_verror(td, err, func) \
526 __td_verror((td), (err), strerror((err)), (func))
527 #define td_vmsg(td, err, msg, func) \
528 __td_verror((td), (err), (msg), (func))
530 extern int exitall_on_terminate;
531 extern int thread_number;
532 extern int nr_process, nr_thread;
535 extern int terse_output;
536 extern int temp_stall_ts;
537 extern unsigned long long mlock_size;
538 extern unsigned long page_mask, page_size;
539 extern int read_only;
540 extern int eta_print;
541 extern unsigned long done_secs;
542 extern char *job_section;
543 extern int fio_gtod_offload;
544 extern int fio_gtod_cpu;
545 extern enum fio_cs fio_clock_source;
546 extern int warnings_fatal;
548 extern struct thread_data *threads;
550 static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
552 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
555 #define BLOCKS_PER_MAP (8 * sizeof(unsigned long))
556 #define TO_MAP_BLOCK(f, b) (b)
557 #define RAND_MAP_IDX(f, b) (TO_MAP_BLOCK(f, b) / BLOCKS_PER_MAP)
558 #define RAND_MAP_BIT(f, b) (TO_MAP_BLOCK(f, b) & (BLOCKS_PER_MAP - 1))
560 #define REAL_MAX_JOBS 2048
562 #define td_non_fatal_error(e) ((e) == EIO || (e) == EILSEQ)
564 static inline void update_error_count(struct thread_data *td, int err)
566 td->total_err_count++;
567 if (td->total_err_count == 1)
568 td->first_error = err;
571 static inline int should_fsync(struct thread_data *td)
573 if (td->last_was_sync)
577 if (td_write(td) || td_rw(td) || td->o.override_sync)
584 * Init/option functions
586 extern int __must_check parse_options(int, char **);
587 extern int fio_options_parse(struct thread_data *, char **, int);
588 extern void fio_keywords_init(void);
589 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
590 extern void fio_fill_default_options(struct thread_data *);
591 extern int fio_show_option_help(const char *);
592 extern void fio_options_dup_and_init(struct option *);
593 extern void options_mem_dupe(struct thread_data *);
594 extern void options_mem_free(struct thread_data *);
595 extern void td_fill_rand_seeds(struct thread_data *);
596 extern void add_job_opts(const char **);
597 extern char *num2str(unsigned long, int, int, int);
599 #define FIO_GETOPT_JOB 0x89988998
600 #define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
605 extern void print_thread_status(void);
606 extern void print_status_init(int);
609 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
610 * will never back again. It may cycle between running/verififying/fsyncing.
611 * Once the thread reaches TD_EXITED, it is just waiting for the core to
627 extern void td_set_runstate(struct thread_data *, int);
632 extern int __must_check fio_pin_memory(void);
633 extern void fio_unpin_memory(void);
634 extern int __must_check allocate_io_mem(struct thread_data *);
635 extern void free_io_mem(struct thread_data *);
638 * Reset stats after ramp time completes
640 extern void reset_all_stats(struct thread_data *);
645 #ifdef FIO_HAVE_BLKTRACE
646 extern int is_blktrace(const char *);
647 extern int load_blktrace(struct thread_data *, const char *);
651 * Mark unused variables passed to ops functions as unused, to silence gcc
653 #define fio_unused __attribute((__unused__))
654 #define fio_init __attribute__((constructor))
655 #define fio_exit __attribute__((destructor))
657 #define for_each_td(td, i) \
658 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
659 #define for_each_file(td, f, i) \
660 if ((td)->files_index) \
661 for ((i) = 0, (f) = (td)->files[0]; \
662 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
665 #define fio_assert(td, cond) do { \
668 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
669 td_set_runstate((td), TD_EXITED); \
670 (td)->error = EFAULT; \
675 static inline int fio_fill_issue_time(struct thread_data *td)
677 if (td->o.read_iolog_file ||
678 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
684 static inline int __should_check_rate(struct thread_data *td,
687 struct thread_options *o = &td->o;
690 * If some rate setting was given, we need to check it
692 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
693 o->rate_iops_min[ddir])
699 static inline int should_check_rate(struct thread_data *td,
700 unsigned long *bytes_done)
705 ret |= __should_check_rate(td, 0);
707 ret |= __should_check_rate(td, 1);
712 static inline int is_power_of_2(unsigned int val)
714 return (val != 0 && ((val & (val - 1)) == 0));
718 * We currently only need to do locking if we have verifier threads
719 * accessing our internal structures too
721 static inline void td_io_u_lock(struct thread_data *td)
723 if (td->o.verify_async)
724 pthread_mutex_lock(&td->io_u_lock);
727 static inline void td_io_u_unlock(struct thread_data *td)
729 if (td->o.verify_async)
730 pthread_mutex_unlock(&td->io_u_lock);
733 static inline void td_io_u_free_notify(struct thread_data *td)
735 if (td->o.verify_async)
736 pthread_cond_signal(&td->free_cond);