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"
39 #include "lib/memalign.h"
46 #include "io_u_queue.h"
47 #include "workqueue.h"
48 #include "steadystate.h"
49 #include "lib/nowarn_snprintf.h"
51 #ifdef CONFIG_SOLARISAIO
52 #include <sys/asynch.h>
56 #include <linux/mempolicy.h>
60 * "local" is pseudo-policy
74 * offset generator types
82 __TD_F_VER_BACKLOG = 0,
85 __TD_F_REFILL_BUFFERS,
86 __TD_F_SCRAMBLE_BUFFERS,
99 __TD_F_LAST, /* not a real bit, keep last */
103 TD_F_VER_BACKLOG = 1U << __TD_F_VER_BACKLOG,
104 TD_F_TRIM_BACKLOG = 1U << __TD_F_TRIM_BACKLOG,
105 TD_F_READ_IOLOG = 1U << __TD_F_READ_IOLOG,
106 TD_F_REFILL_BUFFERS = 1U << __TD_F_REFILL_BUFFERS,
107 TD_F_SCRAMBLE_BUFFERS = 1U << __TD_F_SCRAMBLE_BUFFERS,
108 TD_F_DO_VERIFY = 1U << __TD_F_DO_VERIFY,
109 TD_F_PROFILE_OPS = 1U << __TD_F_PROFILE_OPS,
110 TD_F_COMPRESS = 1U << __TD_F_COMPRESS,
111 TD_F_COMPRESS_LOG = 1U << __TD_F_COMPRESS_LOG,
112 TD_F_VSTATE_SAVED = 1U << __TD_F_VSTATE_SAVED,
113 TD_F_NEED_LOCK = 1U << __TD_F_NEED_LOCK,
114 TD_F_CHILD = 1U << __TD_F_CHILD,
115 TD_F_NO_PROGRESS = 1U << __TD_F_NO_PROGRESS,
116 TD_F_REGROW_LOGS = 1U << __TD_F_REGROW_LOGS,
117 TD_F_MMAP_KEEP = 1U << __TD_F_MMAP_KEEP,
118 TD_F_DIRS_CREATED = 1U << __TD_F_DIRS_CREATED,
119 TD_F_CHECK_RATE = 1U << __TD_F_CHECK_RATE,
130 FIO_RAND_FILE_SIZE_OFF,
133 FIO_RAND_SEQ_RAND_READ_OFF,
134 FIO_RAND_SEQ_RAND_WRITE_OFF,
135 FIO_RAND_SEQ_RAND_TRIM_OFF,
136 FIO_RAND_START_DELAY,
138 FIO_RAND_POISSON_OFF,
140 FIO_RAND_POISSON2_OFF,
141 FIO_RAND_POISSON3_OFF,
150 RATE_PROCESS_LINEAR = 0,
151 RATE_PROCESS_POISSON = 1,
162 * Per-thread/process specific data. Only used for the network client
165 void sk_out_assign(struct sk_out *);
166 void sk_out_drop(void);
168 struct zone_split_index {
170 uint8_t size_perc_prev;
176 * This describes a single thread/process executing a fio job.
179 struct flist_head opt_list;
181 struct thread_options o;
184 unsigned int thread_number;
185 unsigned int subjob_number;
186 unsigned int groupid;
187 struct thread_stat ts __attribute__ ((aligned(8)));
191 struct io_log *slat_log;
192 struct io_log *clat_log;
193 struct io_log *clat_hist_log;
194 struct io_log *lat_log;
195 struct io_log *bw_log;
196 struct io_log *iops_log;
198 struct workqueue log_compress_wq;
200 struct thread_data *parent;
202 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
203 struct timespec bw_sample_time;
205 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
206 struct timespec iops_sample_time;
208 volatile int update_rusage;
209 struct fio_sem *rusage_sem;
210 struct rusage ru_start;
211 struct rusage ru_end;
213 struct fio_file **files;
214 unsigned char *file_locks;
215 unsigned int files_size;
216 unsigned int files_index;
217 unsigned int nr_open_files;
218 unsigned int nr_done_files;
220 unsigned int next_file;
221 struct frand_state next_file_state;
224 struct zipf_state next_file_zipf;
225 struct gauss_state next_file_gauss;
232 double random_center;
239 size_t orig_buffer_size;
240 volatile int runstate;
241 volatile bool terminate;
243 enum fio_ddir last_ddir;
250 uint64_t rand_seeds[FIO_RAND_NR_OFFS];
252 struct frand_state bsrange_state[DDIR_RWDIR_CNT];
253 struct frand_state verify_state;
254 struct frand_state trim_state;
255 struct frand_state delay_state;
257 struct frand_state buf_state;
258 struct frand_state buf_state_prev;
259 struct frand_state dedupe_state;
260 struct frand_state zone_state;
261 struct frand_state prio_state;
263 struct zone_split_index **zone_state_index;
264 unsigned int num_open_zones;
266 unsigned int verify_batch;
267 unsigned int trim_batch;
269 struct thread_io_list *vstate;
274 * IO engine hooks, contains everything needed to submit an io_u
275 * to any of the available IO engines.
277 struct ioengine_ops *io_ops;
281 * IO engine private data and dlhandle.
284 void *io_ops_dlhandle;
287 * Queue depth of io_u's that fio MIGHT do
289 unsigned int cur_depth;
292 * io_u's about to be committed
294 unsigned int io_u_queued;
297 * io_u's submitted but not completed yet
299 unsigned int io_u_in_flight;
302 * List of free and busy io_u's
304 struct io_u_ring io_u_requeues;
305 struct io_u_queue io_u_freelist;
306 struct io_u_queue io_u_all;
307 pthread_mutex_t io_u_lock;
308 pthread_cond_t free_cond;
311 * async verify offload
313 struct flist_head verify_list;
314 pthread_t *verify_threads;
315 unsigned int nr_verify_threads;
316 pthread_cond_t verify_cond;
317 int verify_thread_exit;
322 uint64_t rate_bps[DDIR_RWDIR_CNT];
323 uint64_t rate_next_io_time[DDIR_RWDIR_CNT];
324 unsigned long long rate_bytes[DDIR_RWDIR_CNT];
325 unsigned long rate_blocks[DDIR_RWDIR_CNT];
326 unsigned long long rate_io_issue_bytes[DDIR_RWDIR_CNT];
327 struct timespec lastrate[DDIR_RWDIR_CNT];
328 int64_t last_usec[DDIR_RWDIR_CNT];
329 struct frand_state poisson_state[DDIR_RWDIR_CNT];
332 * Enforced rate submission/completion workqueue
334 struct workqueue io_wq;
336 uint64_t total_io_size;
337 uint64_t fill_device_size;
342 uint64_t io_issues[DDIR_RWDIR_CNT];
343 uint64_t io_issue_bytes[DDIR_RWDIR_CNT];
349 uint64_t io_blocks[DDIR_RWDIR_CNT];
350 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
351 uint64_t io_bytes[DDIR_RWDIR_CNT];
352 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
353 uint64_t io_skip_bytes;
356 uint64_t bytes_done[DDIR_RWDIR_CNT];
359 * State for random io, a bitmap of blocks done vs not done
361 struct frand_state random_state;
363 struct timespec start; /* start of this loop */
364 struct timespec epoch; /* time job was started */
365 unsigned long long unix_epoch; /* Time job was started, unix epoch based. */
366 struct timespec last_issue;
368 struct timespec ts_cache;
369 struct timespec terminate_time;
370 unsigned int ts_cache_nr;
371 unsigned int ts_cache_mask;
375 * Time since last latency_window was started
377 struct timespec latency_ts;
378 unsigned int latency_qd;
379 unsigned int latency_qd_high;
380 unsigned int latency_qd_low;
381 unsigned int latency_failed;
382 unsigned int latency_stable_count;
383 uint64_t latency_ios;
387 * read/write mixed workload state
389 struct frand_state rwmix_state;
390 unsigned long rwmix_issues;
391 enum fio_ddir rwmix_ddir;
392 unsigned int ddir_seq_nr;
395 * rand/seq mixed workload state
397 struct frand_state seq_rand_state[DDIR_RWDIR_CNT];
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;
405 unsigned long io_hist_len;
410 struct flist_head io_log_list;
412 unsigned int io_log_current;
413 unsigned int io_log_checkmark;
414 unsigned int io_log_highmark;
415 struct timespec io_log_highmark_time;
418 * For tracking/handling discards
420 struct flist_head trim_list;
421 unsigned long trim_entries;
424 * for fileservice, how often to switch to a new file
426 unsigned int file_service_nr;
427 unsigned int file_service_left;
428 struct fio_file *file_service_file;
430 unsigned int sync_file_range_nr;
433 * For generating file sizes
435 struct frand_state file_size_state;
440 unsigned int total_err_count;
443 struct fio_flow *flow;
444 unsigned long long flow_counter;
447 * Can be overloaded by profiles
449 struct prof_io_ops prof_io_ops;
454 struct steadystate_data ss;
456 char verror[FIO_VERROR_SIZE];
460 * for GPU memory management
466 CUdeviceptr dev_mem_ptr;
471 struct thread_segment {
472 struct thread_data *threads;
478 * when should interactive ETA output be generated
486 #define __td_verror(td, err, msg, func) \
488 unsigned int ____e = (err); \
491 (td)->error = ____e; \
492 if (!(td)->first_error) \
493 nowarn_snprintf(td->verror, sizeof(td->verror), \
494 "file:%s:%d, func=%s, error=%s", \
495 __FILE__, __LINE__, (func), (msg)); \
499 #define td_clear_error(td) do { \
502 (td)->parent->error = 0; \
505 #define td_verror(td, err, func) do { \
506 __td_verror((td), (err), strerror((err)), (func)); \
508 __td_verror((td)->parent, (err), strerror((err)), (func)); \
511 #define td_vmsg(td, err, msg, func) do { \
512 __td_verror((td), (err), (msg), (func)); \
514 __td_verror((td)->parent, (err), (msg), (func)); \
517 #define __fio_stringify_1(x) #x
518 #define __fio_stringify(x) __fio_stringify_1(x)
520 #define REAL_MAX_JOBS 4096
521 #define JOBS_PER_SEG 8
522 #define REAL_MAX_SEG (REAL_MAX_JOBS / JOBS_PER_SEG)
524 extern bool exitall_on_terminate;
525 extern unsigned int thread_number;
526 extern unsigned int stat_number;
527 extern unsigned int nr_segments;
528 extern unsigned int cur_segment;
530 extern int output_format;
531 extern int append_terse_output;
532 extern int temp_stall_ts;
533 extern uintptr_t page_mask, page_size;
534 extern bool read_only;
535 extern int eta_print;
536 extern int eta_new_line;
537 extern unsigned int eta_interval_msec;
538 extern unsigned long done_secs;
539 extern int fio_gtod_offload;
540 extern int fio_gtod_cpu;
541 extern enum fio_cs fio_clock_source;
542 extern int fio_clock_source_set;
543 extern int warnings_fatal;
544 extern int terse_version;
545 extern bool is_backend;
546 extern bool is_local_backend;
547 extern int nr_clients;
548 extern bool log_syslog;
549 extern int status_interval;
550 extern const char fio_version_string[];
551 extern char *trigger_file;
552 extern char *trigger_cmd;
553 extern char *trigger_remote_cmd;
554 extern long long trigger_timeout;
555 extern char *aux_path;
557 extern struct thread_segment segments[REAL_MAX_SEG];
559 static inline struct thread_data *tnumber_to_td(unsigned int tnumber)
561 struct thread_segment *seg;
563 seg = &segments[tnumber / JOBS_PER_SEG];
564 return &seg->threads[tnumber & (JOBS_PER_SEG - 1)];
567 static inline bool is_running_backend(void)
569 return is_backend || is_local_backend;
572 extern bool eta_time_within_slack(unsigned int time);
574 static inline void fio_ro_check(const struct thread_data *td, struct io_u *io_u)
576 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)) &&
577 !(io_u->ddir == DDIR_TRIM && !td_trim(td)));
580 static inline bool should_fsync(struct thread_data *td)
582 if (td->last_was_sync)
584 if (td_write(td) || td->o.override_sync)
591 * Init/option functions
593 extern int __must_check fio_init_options(void);
594 extern int __must_check parse_options(int, char **);
595 extern int parse_jobs_ini(char *, int, int, int);
596 extern int parse_cmd_line(int, char **, int);
597 extern int fio_backend(struct sk_out *);
598 extern void reset_fio_state(void);
599 extern void clear_io_state(struct thread_data *, int);
600 extern int fio_options_parse(struct thread_data *, char **, int);
601 extern void fio_keywords_init(void);
602 extern void fio_keywords_exit(void);
603 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
604 extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
605 extern void fio_fill_default_options(struct thread_data *);
606 extern int fio_show_option_help(const char *);
607 extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
608 extern void fio_options_dup_and_init(struct option *);
609 extern char *fio_option_dup_subs(const char *);
610 extern void fio_options_mem_dupe(struct thread_data *);
611 extern void td_fill_rand_seeds(struct thread_data *);
612 extern void td_fill_verify_state_seed(struct thread_data *);
613 extern void add_job_opts(const char **, int);
614 extern int ioengine_load(struct thread_data *);
615 extern bool parse_dryrun(void);
616 extern int fio_running_or_pending_io_threads(void);
617 extern int fio_set_fd_nonblocking(int, const char *);
618 extern void sig_show_status(int sig);
619 extern struct thread_data *get_global_options(void);
621 extern uintptr_t page_mask;
622 extern uintptr_t page_size;
623 extern int initialize_fio(char *envp[]);
624 extern void deinitialize_fio(void);
626 #define FIO_GETOPT_JOB 0x89000000
627 #define FIO_GETOPT_IOENGINE 0x98000000
628 #define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
633 extern void print_thread_status(void);
634 extern void print_status_init(int);
635 extern char *fio_uint_to_kmg(unsigned int val);
638 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
639 * will never back again. It may cycle between running/verififying/fsyncing.
640 * Once the thread reaches TD_EXITED, it is just waiting for the core to
660 #define TD_ENG_FLAG_SHIFT 17
661 #define TD_ENG_FLAG_MASK ((1U << 17) - 1)
663 static inline void td_set_ioengine_flags(struct thread_data *td)
665 td->flags = (~(TD_ENG_FLAG_MASK << TD_ENG_FLAG_SHIFT) & td->flags) |
666 (td->io_ops->flags << TD_ENG_FLAG_SHIFT);
669 static inline bool td_ioengine_flagged(struct thread_data *td,
670 enum fio_ioengine_flags flags)
672 return ((td->flags >> TD_ENG_FLAG_SHIFT) & flags) != 0;
675 extern void td_set_runstate(struct thread_data *, int);
676 extern int td_bump_runstate(struct thread_data *, int);
677 extern void td_restore_runstate(struct thread_data *, int);
678 extern const char *runstate_to_name(int runstate);
681 * Allow 60 seconds for a job to quit on its own, otherwise reap with
684 #define FIO_REAP_TIMEOUT 300
689 TERMINATE_STONEWALL = 2,
693 extern void fio_terminate_threads(unsigned int, unsigned int);
694 extern void fio_mark_td_terminate(struct thread_data *);
699 extern int __must_check fio_pin_memory(struct thread_data *);
700 extern void fio_unpin_memory(struct thread_data *);
701 extern int __must_check allocate_io_mem(struct thread_data *);
702 extern void free_io_mem(struct thread_data *);
703 extern void free_threads_shm(void);
706 #define PTR_ALIGN(ptr, mask) \
707 (char *) (((uintptr_t) (ptr) + (mask)) & ~(mask))
711 * Reset stats after ramp time completes
713 extern void reset_all_stats(struct thread_data *);
715 extern int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret,
716 enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify,
717 struct timespec *comp_time);
720 * Latency target helpers
722 extern void lat_target_check(struct thread_data *);
723 extern void lat_target_init(struct thread_data *);
724 extern void lat_target_reset(struct thread_data *);
727 * Iterates all threads/processes within all the defined jobs
729 #define for_each_td(td, i) \
730 for ((i) = 0, (td) = &segments[0].threads[0]; (i) < (int) thread_number; (i)++, (td) = tnumber_to_td((i)))
731 #define for_each_file(td, f, i) \
732 if ((td)->files_index) \
733 for ((i) = 0, (f) = (td)->files[0]; \
734 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
737 static inline bool fio_fill_issue_time(struct thread_data *td)
739 if (td->o.read_iolog_file ||
740 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
746 static inline bool option_check_rate(struct thread_data *td, enum fio_ddir ddir)
748 struct thread_options *o = &td->o;
751 * If some rate setting was given, we need to check it
753 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
754 o->rate_iops_min[ddir])
760 static inline bool __should_check_rate(struct thread_data *td)
762 return (td->flags & TD_F_CHECK_RATE) != 0;
765 static inline bool should_check_rate(struct thread_data *td)
767 if (!__should_check_rate(td))
770 return ddir_rw_sum(td->bytes_done) != 0;
773 static inline unsigned long long td_max_bs(struct thread_data *td)
775 unsigned long long max_bs;
777 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
778 return max(td->o.max_bs[DDIR_TRIM], max_bs);
781 static inline unsigned long long td_min_bs(struct thread_data *td)
783 unsigned long long min_bs;
785 min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
786 return min(td->o.min_bs[DDIR_TRIM], min_bs);
789 static inline bool td_async_processing(struct thread_data *td)
791 return (td->flags & TD_F_NEED_LOCK) != 0;
794 static inline bool td_offload_overlap(struct thread_data *td)
796 return td->o.serialize_overlap && td->o.io_submit_mode == IO_MODE_OFFLOAD;
800 * We currently only need to do locking if we have verifier threads
801 * accessing our internal structures too
803 static inline void __td_io_u_lock(struct thread_data *td)
805 pthread_mutex_lock(&td->io_u_lock);
808 static inline void __td_io_u_unlock(struct thread_data *td)
810 pthread_mutex_unlock(&td->io_u_lock);
813 static inline void td_io_u_free_notify(struct thread_data *td)
815 if (td_async_processing(td))
816 pthread_cond_signal(&td->free_cond);
819 static inline void td_flags_clear(struct thread_data *td, unsigned int *flags,
822 if (!td_async_processing(td))
825 __sync_fetch_and_and(flags, ~value);
828 static inline void td_flags_set(struct thread_data *td, unsigned int *flags,
831 if (!td_async_processing(td))
834 __sync_fetch_and_or(flags, value);
837 extern const char *fio_get_arch_string(int);
838 extern const char *fio_get_os_string(int);
841 __FIO_OUTPUT_TERSE = 0,
842 __FIO_OUTPUT_JSON = 1,
843 __FIO_OUTPUT_NORMAL = 2,
844 __FIO_OUTPUT_JSON_PLUS = 3,
847 FIO_OUTPUT_TERSE = 1U << __FIO_OUTPUT_TERSE,
848 FIO_OUTPUT_JSON = 1U << __FIO_OUTPUT_JSON,
849 FIO_OUTPUT_NORMAL = 1U << __FIO_OUTPUT_NORMAL,
850 FIO_OUTPUT_JSON_PLUS = 1U << __FIO_OUTPUT_JSON_PLUS,
854 FIO_RAND_DIST_RANDOM = 0,
856 FIO_RAND_DIST_PARETO,
859 FIO_RAND_DIST_ZONED_ABS,
862 #define FIO_DEF_ZIPF 1.1
863 #define FIO_DEF_PARETO 0.2
866 FIO_RAND_GEN_TAUSWORTHE = 0,
868 FIO_RAND_GEN_TAUSWORTHE64,
876 extern void exec_trigger(const char *);
877 extern void check_trigger_file(void);
879 extern bool in_flight_overlap(struct io_u_queue *q, struct io_u *io_u);
880 extern pthread_mutex_t overlap_check;
882 static inline void *fio_memalign(size_t alignment, size_t size, bool shared)
884 return __fio_memalign(alignment, size, shared ? smalloc : malloc);
887 static inline void fio_memfree(void *ptr, size_t size, bool shared)
889 return __fio_memfree(ptr, size, shared ? sfree : free);