8 #include <sys/resource.h>
17 #include "compiler/compiler.h"
18 #include "thread_options.h"
21 #include "arch/arch.h"
27 #include "ioengines.h"
35 #include "oslib/getopt.h"
37 #include "lib/rbtree.h"
38 #include "lib/num2str.h"
44 #include "io_u_queue.h"
45 #include "workqueue.h"
46 #include "steadystate.h"
47 #include "lib/nowarn_snprintf.h"
49 #ifdef CONFIG_SOLARISAIO
50 #include <sys/asynch.h>
54 #include <linux/mempolicy.h>
58 * "local" is pseudo-policy
72 * offset generator types
80 __TD_F_VER_BACKLOG = 0,
83 __TD_F_REFILL_BUFFERS,
84 __TD_F_SCRAMBLE_BUFFERS,
97 __TD_F_LAST, /* not a real bit, keep last */
101 TD_F_VER_BACKLOG = 1U << __TD_F_VER_BACKLOG,
102 TD_F_TRIM_BACKLOG = 1U << __TD_F_TRIM_BACKLOG,
103 TD_F_READ_IOLOG = 1U << __TD_F_READ_IOLOG,
104 TD_F_REFILL_BUFFERS = 1U << __TD_F_REFILL_BUFFERS,
105 TD_F_SCRAMBLE_BUFFERS = 1U << __TD_F_SCRAMBLE_BUFFERS,
106 TD_F_DO_VERIFY = 1U << __TD_F_DO_VERIFY,
107 TD_F_PROFILE_OPS = 1U << __TD_F_PROFILE_OPS,
108 TD_F_COMPRESS = 1U << __TD_F_COMPRESS,
109 TD_F_COMPRESS_LOG = 1U << __TD_F_COMPRESS_LOG,
110 TD_F_VSTATE_SAVED = 1U << __TD_F_VSTATE_SAVED,
111 TD_F_NEED_LOCK = 1U << __TD_F_NEED_LOCK,
112 TD_F_CHILD = 1U << __TD_F_CHILD,
113 TD_F_NO_PROGRESS = 1U << __TD_F_NO_PROGRESS,
114 TD_F_REGROW_LOGS = 1U << __TD_F_REGROW_LOGS,
115 TD_F_MMAP_KEEP = 1U << __TD_F_MMAP_KEEP,
116 TD_F_DIRS_CREATED = 1U << __TD_F_DIRS_CREATED,
117 TD_F_CHECK_RATE = 1U << __TD_F_CHECK_RATE,
128 FIO_RAND_FILE_SIZE_OFF,
131 FIO_RAND_SEQ_RAND_READ_OFF,
132 FIO_RAND_SEQ_RAND_WRITE_OFF,
133 FIO_RAND_SEQ_RAND_TRIM_OFF,
134 FIO_RAND_START_DELAY,
136 FIO_RAND_POISSON_OFF,
138 FIO_RAND_POISSON2_OFF,
139 FIO_RAND_POISSON3_OFF,
147 RATE_PROCESS_LINEAR = 0,
148 RATE_PROCESS_POISSON = 1,
164 * Per-thread/process specific data. Only used for the network client
167 void sk_out_assign(struct sk_out *);
168 void sk_out_drop(void);
170 struct zone_split_index {
172 uint8_t size_perc_prev;
177 #define FIO_MAX_OPEN_ZBD_ZONES 128
180 * This describes a single thread/process executing a fio job.
183 struct flist_head opt_list;
185 struct thread_options o;
188 unsigned int thread_number;
189 unsigned int subjob_number;
190 unsigned int groupid;
191 struct thread_stat ts __attribute__ ((aligned(8)));
195 struct io_log *slat_log;
196 struct io_log *clat_log;
197 struct io_log *clat_hist_log;
198 struct io_log *lat_log;
199 struct io_log *bw_log;
200 struct io_log *iops_log;
202 struct workqueue log_compress_wq;
204 struct thread_data *parent;
206 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
207 struct timespec bw_sample_time;
209 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
210 struct timespec iops_sample_time;
212 volatile int update_rusage;
213 struct fio_sem *rusage_sem;
214 struct rusage ru_start;
215 struct rusage ru_end;
217 struct fio_file **files;
218 unsigned char *file_locks;
219 unsigned int files_size;
220 unsigned int files_index;
221 unsigned int nr_open_files;
222 unsigned int nr_done_files;
224 unsigned int next_file;
225 struct frand_state next_file_state;
228 struct zipf_state next_file_zipf;
229 struct gauss_state next_file_gauss;
242 size_t orig_buffer_size;
243 volatile int runstate;
244 volatile bool terminate;
246 enum fio_ddir last_ddir;
253 unsigned long rand_seeds[FIO_RAND_NR_OFFS];
255 struct frand_state bsrange_state[DDIR_RWDIR_CNT];
256 struct frand_state verify_state;
257 struct frand_state trim_state;
258 struct frand_state delay_state;
260 struct frand_state buf_state;
261 struct frand_state buf_state_prev;
262 struct frand_state dedupe_state;
263 struct frand_state zone_state;
265 struct zone_split_index **zone_state_index;
267 unsigned int verify_batch;
268 unsigned int trim_batch;
270 struct thread_io_list *vstate;
275 * IO engine hooks, contains everything needed to submit an io_u
276 * to any of the available IO engines.
278 struct ioengine_ops *io_ops;
282 * IO engine private data and dlhandle.
285 void *io_ops_dlhandle;
288 * Queue depth of io_u's that fio MIGHT do
290 unsigned int cur_depth;
293 * io_u's about to be committed
295 unsigned int io_u_queued;
298 * io_u's submitted but not completed yet
300 unsigned int io_u_in_flight;
303 * List of free and busy io_u's
305 struct io_u_ring io_u_requeues;
306 struct io_u_queue io_u_freelist;
307 struct io_u_queue io_u_all;
308 pthread_mutex_t io_u_lock;
309 pthread_cond_t free_cond;
312 * async verify offload
314 struct flist_head verify_list;
315 pthread_t *verify_threads;
316 unsigned int nr_verify_threads;
317 pthread_cond_t verify_cond;
318 int verify_thread_exit;
323 uint64_t rate_bps[DDIR_RWDIR_CNT];
324 uint64_t rate_next_io_time[DDIR_RWDIR_CNT];
325 unsigned long rate_bytes[DDIR_RWDIR_CNT];
326 unsigned long rate_blocks[DDIR_RWDIR_CNT];
327 unsigned long long rate_io_issue_bytes[DDIR_RWDIR_CNT];
328 struct timespec lastrate[DDIR_RWDIR_CNT];
329 int64_t last_usec[DDIR_RWDIR_CNT];
330 struct frand_state poisson_state[DDIR_RWDIR_CNT];
333 * Enforced rate submission/completion workqueue
335 struct workqueue io_wq;
337 uint64_t total_io_size;
338 uint64_t fill_device_size;
343 uint64_t io_issues[DDIR_RWDIR_CNT];
344 uint64_t io_issue_bytes[DDIR_RWDIR_CNT];
350 uint64_t io_blocks[DDIR_RWDIR_CNT];
351 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
352 uint64_t io_bytes[DDIR_RWDIR_CNT];
353 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
354 uint64_t io_skip_bytes;
357 uint64_t bytes_done[DDIR_RWDIR_CNT];
360 * State for random io, a bitmap of blocks done vs not done
362 struct frand_state random_state;
364 struct timespec start; /* start of this loop */
365 struct timespec epoch; /* time job was started */
366 unsigned long long unix_epoch; /* Time job was started, unix epoch based. */
367 struct timespec last_issue;
369 struct timespec ts_cache;
370 struct timespec terminate_time;
371 unsigned int ts_cache_nr;
372 unsigned int ts_cache_mask;
376 * Time since last latency_window was started
378 struct timespec latency_ts;
379 unsigned int latency_qd;
380 unsigned int latency_qd_high;
381 unsigned int latency_qd_low;
382 unsigned int latency_qd_step;
383 unsigned int latency_failed;
384 unsigned int latency_state;
385 unsigned int latency_iops[DDIR_RWDIR_CNT];
386 unsigned int latency_step;
387 uint64_t latency_ios[DDIR_RWDIR_CNT];
389 unsigned int nr_lat_stats;
392 * read/write mixed workload state
394 struct frand_state rwmix_state;
395 unsigned long rwmix_issues;
396 enum fio_ddir rwmix_ddir;
397 unsigned int ddir_seq_nr;
400 * rand/seq mixed workload state
402 struct frand_state seq_rand_state[DDIR_RWDIR_CNT];
405 * IO history logs for verification. We use a tree for sorting,
406 * if we are overwriting. Otherwise just use a fifo.
408 struct rb_root io_hist_tree;
409 struct flist_head io_hist_list;
410 unsigned long io_hist_len;
415 struct flist_head io_log_list;
417 unsigned int io_log_current;
418 unsigned int io_log_checkmark;
419 unsigned int io_log_highmark;
420 struct timespec io_log_highmark_time;
423 * For tracking/handling discards
425 struct flist_head trim_list;
426 unsigned long trim_entries;
429 * for fileservice, how often to switch to a new file
431 unsigned int file_service_nr;
432 unsigned int file_service_left;
433 struct fio_file *file_service_file;
435 unsigned int sync_file_range_nr;
438 * For generating file sizes
440 struct frand_state file_size_state;
445 unsigned int total_err_count;
448 struct fio_flow *flow;
451 * Can be overloaded by profiles
453 struct prof_io_ops prof_io_ops;
458 struct steadystate_data ss;
460 char verror[FIO_VERROR_SIZE];
464 * for GPU memory management
470 CUdeviceptr dev_mem_ptr;
476 * when should interactive ETA output be generated
484 #define __td_verror(td, err, msg, func) \
486 unsigned int ____e = (err); \
489 (td)->error = ____e; \
490 if (!(td)->first_error) \
491 nowarn_snprintf(td->verror, sizeof(td->verror), \
492 "file:%s:%d, func=%s, error=%s", \
493 __FILE__, __LINE__, (func), (msg)); \
497 #define td_clear_error(td) do { \
500 (td)->parent->error = 0; \
503 #define td_verror(td, err, func) do { \
504 __td_verror((td), (err), strerror((err)), (func)); \
506 __td_verror((td)->parent, (err), strerror((err)), (func)); \
509 #define td_vmsg(td, err, msg, func) do { \
510 __td_verror((td), (err), (msg), (func)); \
512 __td_verror((td)->parent, (err), (msg), (func)); \
515 #define __fio_stringify_1(x) #x
516 #define __fio_stringify(x) __fio_stringify_1(x)
518 extern bool exitall_on_terminate;
519 extern unsigned int thread_number;
520 extern unsigned int stat_number;
523 extern int output_format;
524 extern int append_terse_output;
525 extern int temp_stall_ts;
526 extern uintptr_t page_mask, page_size;
527 extern bool read_only;
528 extern int eta_print;
529 extern int eta_new_line;
530 extern unsigned int eta_interval_msec;
531 extern unsigned long done_secs;
532 extern int fio_gtod_offload;
533 extern int fio_gtod_cpu;
534 extern enum fio_cs fio_clock_source;
535 extern int fio_clock_source_set;
536 extern int warnings_fatal;
537 extern int terse_version;
538 extern bool is_backend;
539 extern bool is_local_backend;
540 extern int nr_clients;
541 extern bool log_syslog;
542 extern int status_interval;
543 extern const char fio_version_string[];
544 extern char *trigger_file;
545 extern char *trigger_cmd;
546 extern char *trigger_remote_cmd;
547 extern long long trigger_timeout;
548 extern char *aux_path;
550 extern struct thread_data *threads;
552 static inline bool is_running_backend(void)
554 return is_backend || is_local_backend;
557 extern bool eta_time_within_slack(unsigned int time);
559 static inline void fio_ro_check(const struct thread_data *td, struct io_u *io_u)
561 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)) &&
562 !(io_u->ddir == DDIR_TRIM && !td_trim(td)));
565 #define REAL_MAX_JOBS 4096
567 static inline bool should_fsync(struct thread_data *td)
569 if (td->last_was_sync)
571 if (td_write(td) || td->o.override_sync)
578 * Init/option functions
580 extern int __must_check fio_init_options(void);
581 extern int __must_check parse_options(int, char **);
582 extern int parse_jobs_ini(char *, int, int, int);
583 extern int parse_cmd_line(int, char **, int);
584 extern int fio_backend(struct sk_out *);
585 extern void reset_fio_state(void);
586 extern void clear_io_state(struct thread_data *, int);
587 extern int fio_options_parse(struct thread_data *, char **, int);
588 extern void fio_keywords_init(void);
589 extern void fio_keywords_exit(void);
590 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
591 extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
592 extern void fio_fill_default_options(struct thread_data *);
593 extern int fio_show_option_help(const char *);
594 extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
595 extern void fio_options_dup_and_init(struct option *);
596 extern char *fio_option_dup_subs(const char *);
597 extern void fio_options_mem_dupe(struct thread_data *);
598 extern void td_fill_rand_seeds(struct thread_data *);
599 extern void td_fill_verify_state_seed(struct thread_data *);
600 extern void add_job_opts(const char **, int);
601 extern int ioengine_load(struct thread_data *);
602 extern bool parse_dryrun(void);
603 extern int fio_running_or_pending_io_threads(void);
604 extern int fio_set_fd_nonblocking(int, const char *);
605 extern void sig_show_status(int sig);
606 extern struct thread_data *get_global_options(void);
608 extern uintptr_t page_mask;
609 extern uintptr_t page_size;
610 extern int initialize_fio(char *envp[]);
611 extern void deinitialize_fio(void);
613 #define FIO_GETOPT_JOB 0x89000000
614 #define FIO_GETOPT_IOENGINE 0x98000000
615 #define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
620 extern void print_thread_status(void);
621 extern void print_status_init(int);
622 extern char *fio_uint_to_kmg(unsigned int val);
625 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
626 * will never back again. It may cycle between running/verififying/fsyncing.
627 * Once the thread reaches TD_EXITED, it is just waiting for the core to
647 #define TD_ENG_FLAG_SHIFT 17
648 #define TD_ENG_FLAG_MASK ((1U << 17) - 1)
650 static inline void td_set_ioengine_flags(struct thread_data *td)
652 td->flags = (~(TD_ENG_FLAG_MASK << TD_ENG_FLAG_SHIFT) & td->flags) |
653 (td->io_ops->flags << TD_ENG_FLAG_SHIFT);
656 static inline bool td_ioengine_flagged(struct thread_data *td,
657 enum fio_ioengine_flags flags)
659 return ((td->flags >> TD_ENG_FLAG_SHIFT) & flags) != 0;
662 extern void td_set_runstate(struct thread_data *, int);
663 extern int td_bump_runstate(struct thread_data *, int);
664 extern void td_restore_runstate(struct thread_data *, int);
665 extern const char *runstate_to_name(int runstate);
668 * Allow 60 seconds for a job to quit on its own, otherwise reap with
671 #define FIO_REAP_TIMEOUT 300
673 #define TERMINATE_ALL (-1U)
674 extern void fio_terminate_threads(unsigned int);
675 extern void fio_mark_td_terminate(struct thread_data *);
680 extern int __must_check fio_pin_memory(struct thread_data *);
681 extern void fio_unpin_memory(struct thread_data *);
682 extern int __must_check allocate_io_mem(struct thread_data *);
683 extern void free_io_mem(struct thread_data *);
684 extern void free_threads_shm(void);
687 #define PTR_ALIGN(ptr, mask) \
688 (char *) (((uintptr_t) (ptr) + (mask)) & ~(mask))
692 * Reset stats after ramp time completes
694 extern void reset_all_stats(struct thread_data *);
696 extern int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret,
697 enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify,
698 struct timespec *comp_time);
701 * Iterates all threads/processes within all the defined jobs
703 #define for_each_td(td, i) \
704 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
705 #define for_each_file(td, f, i) \
706 if ((td)->files_index) \
707 for ((i) = 0, (f) = (td)->files[0]; \
708 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
711 #define fio_assert(td, cond) do { \
714 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
715 td_set_runstate((td), TD_EXITED); \
716 (td)->error = EFAULT; \
721 static inline bool fio_fill_issue_time(struct thread_data *td)
723 if (td->o.read_iolog_file ||
724 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
730 static inline bool option_check_rate(struct thread_data *td, enum fio_ddir ddir)
732 struct thread_options *o = &td->o;
735 * If some rate setting was given, we need to check it
737 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
738 o->rate_iops_min[ddir])
744 static inline bool __should_check_rate(struct thread_data *td)
746 return (td->flags & TD_F_CHECK_RATE) != 0;
749 static inline bool should_check_rate(struct thread_data *td)
751 if (!__should_check_rate(td))
754 return ddir_rw_sum(td->bytes_done) != 0;
757 int setup_rate(struct thread_data *td);
759 static inline unsigned long long td_max_bs(struct thread_data *td)
761 unsigned long long max_bs;
763 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
764 return max(td->o.max_bs[DDIR_TRIM], max_bs);
767 static inline unsigned long long td_min_bs(struct thread_data *td)
769 unsigned long long min_bs;
771 min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
772 return min(td->o.min_bs[DDIR_TRIM], min_bs);
775 static inline bool td_async_processing(struct thread_data *td)
777 return (td->flags & TD_F_NEED_LOCK) != 0;
780 static inline bool td_offload_overlap(struct thread_data *td)
782 return td->o.serialize_overlap && td->o.io_submit_mode == IO_MODE_OFFLOAD;
786 * We currently only need to do locking if we have verifier threads
787 * accessing our internal structures too
789 static inline void __td_io_u_lock(struct thread_data *td)
791 pthread_mutex_lock(&td->io_u_lock);
794 static inline void __td_io_u_unlock(struct thread_data *td)
796 pthread_mutex_unlock(&td->io_u_lock);
799 static inline void td_io_u_free_notify(struct thread_data *td)
801 if (td_async_processing(td))
802 pthread_cond_signal(&td->free_cond);
805 static inline void td_flags_clear(struct thread_data *td, unsigned int *flags,
808 if (!td_async_processing(td))
811 __sync_fetch_and_and(flags, ~value);
814 static inline void td_flags_set(struct thread_data *td, unsigned int *flags,
817 if (!td_async_processing(td))
820 __sync_fetch_and_or(flags, value);
823 extern const char *fio_get_arch_string(int);
824 extern const char *fio_get_os_string(int);
827 __FIO_OUTPUT_TERSE = 0,
828 __FIO_OUTPUT_JSON = 1,
829 __FIO_OUTPUT_NORMAL = 2,
830 __FIO_OUTPUT_JSON_PLUS = 3,
833 FIO_OUTPUT_TERSE = 1U << __FIO_OUTPUT_TERSE,
834 FIO_OUTPUT_JSON = 1U << __FIO_OUTPUT_JSON,
835 FIO_OUTPUT_NORMAL = 1U << __FIO_OUTPUT_NORMAL,
836 FIO_OUTPUT_JSON_PLUS = 1U << __FIO_OUTPUT_JSON_PLUS,
840 FIO_RAND_DIST_RANDOM = 0,
842 FIO_RAND_DIST_PARETO,
845 FIO_RAND_DIST_ZONED_ABS,
848 #define FIO_DEF_ZIPF 1.1
849 #define FIO_DEF_PARETO 0.2
852 FIO_RAND_GEN_TAUSWORTHE = 0,
854 FIO_RAND_GEN_TAUSWORTHE64,
862 extern void exec_trigger(const char *);
863 extern void check_trigger_file(void);
865 extern bool in_flight_overlap(struct io_u_queue *q, struct io_u *io_u);
866 extern pthread_mutex_t overlap_check;