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"
52 #ifdef CONFIG_SOLARISAIO
53 #include <sys/asynch.h>
57 #include <linux/mempolicy.h>
61 * "local" is pseudo-policy
74 #define MAX_TRIM_RANGE 256
77 * Range for trim command
80 unsigned long long start;
81 unsigned long long len;
85 * offset generator types
93 __TD_F_VER_BACKLOG = 0,
96 __TD_F_REFILL_BUFFERS,
97 __TD_F_SCRAMBLE_BUFFERS,
111 __TD_F_LAST, /* not a real bit, keep last */
115 TD_F_VER_BACKLOG = 1U << __TD_F_VER_BACKLOG,
116 TD_F_TRIM_BACKLOG = 1U << __TD_F_TRIM_BACKLOG,
117 TD_F_READ_IOLOG = 1U << __TD_F_READ_IOLOG,
118 TD_F_REFILL_BUFFERS = 1U << __TD_F_REFILL_BUFFERS,
119 TD_F_SCRAMBLE_BUFFERS = 1U << __TD_F_SCRAMBLE_BUFFERS,
120 TD_F_DO_VERIFY = 1U << __TD_F_DO_VERIFY,
121 TD_F_PROFILE_OPS = 1U << __TD_F_PROFILE_OPS,
122 TD_F_COMPRESS = 1U << __TD_F_COMPRESS,
123 TD_F_COMPRESS_LOG = 1U << __TD_F_COMPRESS_LOG,
124 TD_F_VSTATE_SAVED = 1U << __TD_F_VSTATE_SAVED,
125 TD_F_NEED_LOCK = 1U << __TD_F_NEED_LOCK,
126 TD_F_CHILD = 1U << __TD_F_CHILD,
127 TD_F_NO_PROGRESS = 1U << __TD_F_NO_PROGRESS,
128 TD_F_REGROW_LOGS = 1U << __TD_F_REGROW_LOGS,
129 TD_F_MMAP_KEEP = 1U << __TD_F_MMAP_KEEP,
130 TD_F_DIRS_CREATED = 1U << __TD_F_DIRS_CREATED,
131 TD_F_CHECK_RATE = 1U << __TD_F_CHECK_RATE,
132 TD_F_SYNCS = 1U << __TD_F_SYNCS,
143 FIO_RAND_FILE_SIZE_OFF,
146 FIO_RAND_SEQ_RAND_READ_OFF,
147 FIO_RAND_SEQ_RAND_WRITE_OFF,
148 FIO_RAND_SEQ_RAND_TRIM_OFF,
149 FIO_RAND_START_DELAY,
151 FIO_RAND_POISSON_OFF,
153 FIO_RAND_POISSON2_OFF,
154 FIO_RAND_POISSON3_OFF,
156 FIO_RAND_DEDUPE_WORKING_SET_IX,
165 RATE_PROCESS_LINEAR = 0,
166 RATE_PROCESS_POISSON = 1,
168 THINKTIME_BLOCKS_TYPE_COMPLETE = 0,
169 THINKTIME_BLOCKS_TYPE_ISSUE = 1,
181 * Per-thread/process specific data. Only used for the network client
184 void sk_out_assign(struct sk_out *);
185 void sk_out_drop(void);
187 struct zone_split_index {
189 uint8_t size_perc_prev;
195 * This describes a single thread/process executing a fio job.
198 struct flist_head opt_list;
199 unsigned long long flags;
200 struct thread_options o;
203 unsigned int thread_number;
204 unsigned int subjob_number;
205 unsigned int groupid;
206 struct thread_stat ts __attribute__ ((aligned(8)));
210 struct io_log *slat_log;
211 struct io_log *clat_log;
212 struct io_log *clat_hist_log;
213 struct io_log *lat_log;
214 struct io_log *bw_log;
215 struct io_log *iops_log;
217 struct workqueue log_compress_wq;
219 struct thread_data *parent;
221 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
222 struct timespec bw_sample_time;
224 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
225 struct timespec iops_sample_time;
227 volatile int update_rusage;
228 struct fio_sem *rusage_sem;
229 struct rusage ru_start;
230 struct rusage ru_end;
232 struct fio_file **files;
233 unsigned char *file_locks;
234 unsigned int files_size;
235 unsigned int files_index;
236 unsigned int nr_open_files;
237 unsigned int nr_done_files;
239 unsigned int next_file;
240 struct frand_state next_file_state;
243 struct zipf_state next_file_zipf;
244 struct gauss_state next_file_gauss;
251 double random_center;
258 size_t orig_buffer_size;
259 volatile int runstate;
260 volatile bool terminate;
262 enum fio_ddir last_ddir;
269 uint64_t rand_seeds[FIO_RAND_NR_OFFS];
271 struct frand_state bsrange_state[DDIR_RWDIR_CNT];
272 struct frand_state verify_state;
273 struct frand_state verify_state_last_do_io;
274 struct frand_state trim_state;
275 struct frand_state delay_state;
276 struct frand_state fdp_state;
278 struct frand_state buf_state;
279 struct frand_state buf_state_prev;
280 struct frand_state buf_state_ret;
281 struct frand_state dedupe_state;
282 struct frand_state zone_state;
283 struct frand_state prio_state;
284 struct frand_state dedupe_working_set_index_state;
285 struct frand_state *dedupe_working_set_states;
287 unsigned long long num_unique_pages;
289 struct zone_split_index **zone_state_index;
290 unsigned int num_write_zones;
292 unsigned int verify_batch;
293 unsigned int trim_batch;
295 struct thread_io_list *vstate;
300 * Job default IO priority set with prioclass and prio options.
305 * IO engine hooks, contains everything needed to submit an io_u
306 * to any of the available IO engines.
308 struct ioengine_ops *io_ops;
312 * IO engine private data and dlhandle.
317 * Queue depth of io_u's that fio MIGHT do
319 unsigned int cur_depth;
322 * io_u's about to be committed
324 unsigned int io_u_queued;
327 * io_u's submitted but not completed yet
329 unsigned int io_u_in_flight;
332 * List of free and busy io_u's
334 struct io_u_ring io_u_requeues;
335 struct io_u_queue io_u_freelist;
336 struct io_u_queue io_u_all;
337 pthread_mutex_t io_u_lock;
338 pthread_cond_t free_cond;
341 * async verify offload
343 struct flist_head verify_list;
344 pthread_t *verify_threads;
345 unsigned int nr_verify_threads;
346 pthread_cond_t verify_cond;
347 int verify_thread_exit;
352 uint64_t rate_bps[DDIR_RWDIR_CNT];
353 uint64_t rate_next_io_time[DDIR_RWDIR_CNT];
354 unsigned long long last_rate_check_bytes[DDIR_RWDIR_CNT];
355 unsigned long last_rate_check_blocks[DDIR_RWDIR_CNT];
356 unsigned long long rate_io_issue_bytes[DDIR_RWDIR_CNT];
357 struct timespec last_rate_check_time[DDIR_RWDIR_CNT];
358 int64_t last_usec[DDIR_RWDIR_CNT];
359 struct frand_state poisson_state[DDIR_RWDIR_CNT];
362 * Enforced rate submission/completion workqueue
364 struct workqueue io_wq;
366 uint64_t total_io_size;
367 uint64_t fill_device_size;
372 uint64_t io_issues[DDIR_RWDIR_CNT];
373 uint64_t verify_read_issues;
374 uint64_t io_issue_bytes[DDIR_RWDIR_CNT];
380 uint64_t io_blocks[DDIR_RWDIR_CNT];
381 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
382 uint64_t io_bytes[DDIR_RWDIR_CNT];
383 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
384 uint64_t io_skip_bytes;
387 uint64_t bytes_done[DDIR_RWDIR_CNT];
388 uint64_t bytes_verified;
390 uint64_t *thinktime_blocks_counter;
391 struct timespec last_thinktime;
392 int64_t last_thinktime_blocks;
395 * State for random io, a bitmap of blocks done vs not done
397 struct frand_state random_state;
399 struct timespec start; /* start of this loop */
400 struct timespec epoch; /* time job was started */
401 unsigned long long alternate_epoch; /* Time job was started, as clock_gettime(log_alternate_epoch_clock_id) */
402 unsigned long long job_start; /* Time job was started, as clock_gettime(job_start_clock_id) */
403 struct timespec last_issue;
405 struct timespec ts_cache;
406 struct timespec terminate_time;
407 unsigned int ts_cache_nr;
408 unsigned int ts_cache_mask;
412 * Time since last latency_window was started
414 struct timespec latency_ts;
415 unsigned int latency_qd;
416 unsigned int latency_qd_high;
417 unsigned int latency_qd_low;
418 unsigned int latency_failed;
419 unsigned int latency_stable_count;
420 uint64_t latency_ios;
424 * read/write mixed workload state
426 struct frand_state rwmix_state;
427 unsigned long rwmix_issues;
428 enum fio_ddir rwmix_ddir;
429 unsigned int ddir_seq_nr;
432 * rand/seq mixed workload state
434 struct frand_state seq_rand_state[DDIR_RWDIR_CNT];
437 * IO history logs for verification. We use a tree for sorting,
438 * if we are overwriting. Otherwise just use a fifo.
440 struct rb_root io_hist_tree;
441 struct flist_head io_hist_list;
442 unsigned long io_hist_len;
447 struct flist_head io_log_list;
449 unsigned int io_log_blktrace;
450 unsigned int io_log_blktrace_swap;
451 unsigned long long io_log_last_ttime;
452 struct timespec io_log_start_time;
453 unsigned int io_log_current;
454 unsigned int io_log_checkmark;
455 unsigned int io_log_highmark;
456 unsigned int io_log_version;
457 struct timespec io_log_highmark_time;
460 * For tracking/handling discards
462 struct flist_head trim_list;
463 unsigned long trim_entries;
466 * for fileservice, how often to switch to a new file
468 unsigned int file_service_nr;
469 unsigned int file_service_left;
470 struct fio_file *file_service_file;
472 unsigned int sync_file_range_nr;
475 * For generating file sizes
477 struct frand_state file_size_state;
482 unsigned int total_err_count;
485 struct fio_flow *flow;
486 unsigned long long flow_counter;
489 * Can be overloaded by profiles
491 struct prof_io_ops prof_io_ops;
496 struct steadystate_data ss;
498 char verror[FIO_VERROR_SIZE];
502 * for GPU memory management
508 CUdeviceptr dev_mem_ptr;
513 struct thread_segment {
514 struct thread_data *threads;
520 * when should interactive ETA output be generated
528 #define __td_verror(td, err, msg, func) \
530 unsigned int ____e = (err); \
533 (td)->error = ____e; \
534 if (!(td)->first_error) \
535 nowarn_snprintf(td->verror, sizeof(td->verror), \
536 "file:%s:%d, func=%s, error=%s", \
537 __FILE__, __LINE__, (func), (msg)); \
541 #define td_clear_error(td) do { \
544 (td)->parent->error = 0; \
547 #define td_verror(td, err, func) do { \
548 __td_verror((td), (err), strerror((err)), (func)); \
550 __td_verror((td)->parent, (err), strerror((err)), (func)); \
553 #define td_vmsg(td, err, msg, func) do { \
554 __td_verror((td), (err), (msg), (func)); \
556 __td_verror((td)->parent, (err), (msg), (func)); \
559 #define __fio_stringify_1(x) #x
560 #define __fio_stringify(x) __fio_stringify_1(x)
562 #define REAL_MAX_JOBS 4096
563 #define JOBS_PER_SEG 8
564 #define REAL_MAX_SEG (REAL_MAX_JOBS / JOBS_PER_SEG)
566 extern bool exitall_on_terminate;
567 extern unsigned int thread_number;
568 extern unsigned int stat_number;
569 extern unsigned int nr_segments;
570 extern unsigned int cur_segment;
572 extern int output_format;
573 extern int append_terse_output;
574 extern int temp_stall_ts;
575 extern uintptr_t page_mask, page_size;
576 extern bool read_only;
577 extern int eta_print;
578 extern int eta_new_line;
579 extern unsigned int eta_interval_msec;
580 extern unsigned long done_secs;
581 extern int fio_gtod_offload;
582 extern int fio_gtod_cpu;
583 extern enum fio_cs fio_clock_source;
584 extern int fio_clock_source_set;
585 extern int warnings_fatal;
586 extern int terse_version;
587 extern bool is_backend;
588 extern bool is_local_backend;
589 extern int nr_clients;
590 extern bool log_syslog;
591 extern int status_interval;
592 extern const char fio_version_string[];
593 extern char *trigger_file;
594 extern char *trigger_cmd;
595 extern char *trigger_remote_cmd;
596 extern long long trigger_timeout;
597 extern char *aux_path;
599 extern struct thread_segment segments[REAL_MAX_SEG];
601 static inline struct thread_data *tnumber_to_td(unsigned int tnumber)
603 struct thread_segment *seg;
605 seg = &segments[tnumber / JOBS_PER_SEG];
606 return &seg->threads[tnumber & (JOBS_PER_SEG - 1)];
609 static inline bool is_running_backend(void)
611 return is_backend || is_local_backend;
614 extern bool eta_time_within_slack(unsigned int time);
616 static inline void fio_ro_check(const struct thread_data *td, struct io_u *io_u)
618 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)) &&
619 !(io_u->ddir == DDIR_TRIM && !td_trim(td)));
622 static inline bool multi_range_trim(struct thread_data *td, struct io_u *io_u)
624 if (io_u->ddir == DDIR_TRIM && td->o.num_range > 1)
630 static inline bool should_fsync(struct thread_data *td)
632 if (td->last_was_sync)
634 if (td_write(td) || td->o.override_sync)
641 * Init/option functions
643 extern int __must_check fio_init_options(void);
644 extern int __must_check parse_options(int, char **);
645 extern int parse_jobs_ini(char *, int, int, int);
646 extern int parse_cmd_line(int, char **, int);
647 extern int fio_backend(struct sk_out *);
648 extern void reset_fio_state(void);
649 extern void clear_io_state(struct thread_data *, int);
650 extern int fio_options_parse(struct thread_data *, char **, int);
651 extern void fio_keywords_init(void);
652 extern void fio_keywords_exit(void);
653 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
654 extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
655 extern void fio_fill_default_options(struct thread_data *);
656 extern int fio_show_option_help(const char *);
657 extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
658 extern void fio_options_dup_and_init(struct option *);
659 extern char *fio_option_dup_subs(const char *);
660 extern void fio_options_mem_dupe(struct thread_data *);
661 extern void td_fill_rand_seeds(struct thread_data *);
662 extern void add_job_opts(const char **, int);
663 extern int ioengine_load(struct thread_data *);
664 extern bool parse_dryrun(void);
665 extern int fio_running_or_pending_io_threads(void);
666 extern int fio_set_fd_nonblocking(int, const char *);
667 extern void sig_show_status(int sig);
668 extern struct thread_data *get_global_options(void);
670 extern uintptr_t page_mask;
671 extern uintptr_t page_size;
672 extern int initialize_fio(char *envp[]);
673 extern void deinitialize_fio(void);
675 #define FIO_GETOPT_JOB 0x89000000
676 #define FIO_GETOPT_IOENGINE 0x98000000
677 #define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
682 extern void print_thread_status(void);
683 extern void print_status_init(int);
684 extern char *fio_uint_to_kmg(unsigned int val);
687 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
688 * will never back again. It may cycle between running/verififying/fsyncing.
689 * Once the thread reaches TD_EXITED, it is just waiting for the core to
709 #define TD_ENG_FLAG_SHIFT 18
710 #define TD_ENG_FLAG_MASK ((1ULL << 18) - 1)
712 static inline void td_set_ioengine_flags(struct thread_data *td)
714 td->flags = (~(TD_ENG_FLAG_MASK << TD_ENG_FLAG_SHIFT) & td->flags) |
715 ((unsigned long long)td->io_ops->flags << TD_ENG_FLAG_SHIFT);
718 static inline bool td_ioengine_flagged(struct thread_data *td,
719 enum fio_ioengine_flags flags)
721 return ((td->flags >> TD_ENG_FLAG_SHIFT) & flags) != 0;
724 extern void td_set_runstate(struct thread_data *, int);
725 extern int td_bump_runstate(struct thread_data *, int);
726 extern void td_restore_runstate(struct thread_data *, int);
727 extern const char *runstate_to_name(int runstate);
730 * Allow 60 seconds for a job to quit on its own, otherwise reap with
733 #define FIO_REAP_TIMEOUT 300
738 TERMINATE_STONEWALL = 2,
742 extern void fio_terminate_threads(unsigned int, unsigned int);
743 extern void fio_mark_td_terminate(struct thread_data *);
748 extern int __must_check fio_pin_memory(struct thread_data *);
749 extern void fio_unpin_memory(struct thread_data *);
750 extern int __must_check allocate_io_mem(struct thread_data *);
751 extern void free_io_mem(struct thread_data *);
752 extern void free_threads_shm(void);
755 #define PTR_ALIGN(ptr, mask) \
756 (char *) (((uintptr_t) (ptr) + (mask)) & ~(mask))
760 * Reset stats after ramp time completes
762 extern void reset_all_stats(struct thread_data *);
764 extern int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret,
765 enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify,
766 struct timespec *comp_time);
769 * Latency target helpers
771 extern void lat_target_check(struct thread_data *);
772 extern void lat_target_init(struct thread_data *);
773 extern void lat_target_reset(struct thread_data *);
776 * Iterates all threads/processes within all the defined jobs
778 * for_each_td(var_name_for_td) {
779 * << bodoy of your loop >>
780 * Note: internally-scoped loop index availble as __td_index
781 * } end_for_each_td()
783 #define for_each_td(td) \
786 struct thread_data *(td); \
787 for (__td_index = 0, (td) = &segments[0].threads[0];\
788 __td_index < (int) thread_number; __td_index++, (td) = tnumber_to_td(__td_index))
789 #define for_each_td_index() \
792 for (__td_index = 0; __td_index < (int) thread_number; __td_index++)
793 #define end_for_each() }
795 #define for_each_file(td, f, i) \
796 if ((td)->files_index) \
797 for ((i) = 0, (f) = (td)->files[0]; \
798 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
801 static inline bool fio_fill_issue_time(struct thread_data *td)
803 if (td->o.read_iolog_file ||
804 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
810 static inline bool option_check_rate(struct thread_data *td, enum fio_ddir ddir)
812 struct thread_options *o = &td->o;
815 * If some rate setting was given, we need to check it
817 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
818 o->rate_iops_min[ddir])
824 static inline bool should_check_rate(struct thread_data *td)
826 return (td->flags & TD_F_CHECK_RATE) != 0;
829 static inline unsigned long long td_max_bs(struct thread_data *td)
831 unsigned long long max_bs;
833 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
834 return max(td->o.max_bs[DDIR_TRIM], max_bs);
837 static inline unsigned long long td_min_bs(struct thread_data *td)
839 unsigned long long min_bs;
841 min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
842 return min(td->o.min_bs[DDIR_TRIM], min_bs);
845 static inline bool td_async_processing(struct thread_data *td)
847 return (td->flags & TD_F_NEED_LOCK) != 0;
850 static inline bool td_offload_overlap(struct thread_data *td)
852 return td->o.serialize_overlap && td->o.io_submit_mode == IO_MODE_OFFLOAD;
856 * We currently only need to do locking if we have verifier threads
857 * accessing our internal structures too
859 static inline void __td_io_u_lock(struct thread_data *td)
861 pthread_mutex_lock(&td->io_u_lock);
864 static inline void __td_io_u_unlock(struct thread_data *td)
866 pthread_mutex_unlock(&td->io_u_lock);
869 static inline void td_io_u_free_notify(struct thread_data *td)
871 if (td_async_processing(td))
872 pthread_cond_signal(&td->free_cond);
875 static inline void td_flags_clear(struct thread_data *td, unsigned int *flags,
878 if (!td_async_processing(td))
881 __sync_fetch_and_and(flags, ~value);
884 static inline void td_flags_set(struct thread_data *td, unsigned int *flags,
887 if (!td_async_processing(td))
890 __sync_fetch_and_or(flags, value);
893 extern const char *fio_get_arch_string(int);
894 extern const char *fio_get_os_string(int);
897 __FIO_OUTPUT_TERSE = 0,
898 __FIO_OUTPUT_JSON = 1,
899 __FIO_OUTPUT_NORMAL = 2,
900 __FIO_OUTPUT_JSON_PLUS = 3,
903 FIO_OUTPUT_TERSE = 1U << __FIO_OUTPUT_TERSE,
904 FIO_OUTPUT_JSON = 1U << __FIO_OUTPUT_JSON,
905 FIO_OUTPUT_NORMAL = 1U << __FIO_OUTPUT_NORMAL,
906 FIO_OUTPUT_JSON_PLUS = 1U << __FIO_OUTPUT_JSON_PLUS,
910 FIO_RAND_DIST_RANDOM = 0,
912 FIO_RAND_DIST_PARETO,
915 FIO_RAND_DIST_ZONED_ABS,
918 #define FIO_DEF_ZIPF 1.1
919 #define FIO_DEF_PARETO 0.2
922 FIO_RAND_GEN_TAUSWORTHE = 0,
924 FIO_RAND_GEN_TAUSWORTHE64,
932 extern void exec_trigger(const char *);
933 extern void check_trigger_file(void);
935 extern bool in_flight_overlap(struct io_u_queue *q, struct io_u *io_u);
936 extern pthread_mutex_t overlap_check;
938 static inline void *fio_memalign(size_t alignment, size_t size, bool shared)
940 return __fio_memalign(alignment, size, shared ? smalloc : malloc);
943 static inline void fio_memfree(void *ptr, size_t size, bool shared)
945 return __fio_memfree(ptr, size, shared ? sfree : free);