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
75 * offset generator types
83 __TD_F_VER_BACKLOG = 0,
86 __TD_F_REFILL_BUFFERS,
87 __TD_F_SCRAMBLE_BUFFERS,
101 __TD_F_LAST, /* not a real bit, keep last */
105 TD_F_VER_BACKLOG = 1U << __TD_F_VER_BACKLOG,
106 TD_F_TRIM_BACKLOG = 1U << __TD_F_TRIM_BACKLOG,
107 TD_F_READ_IOLOG = 1U << __TD_F_READ_IOLOG,
108 TD_F_REFILL_BUFFERS = 1U << __TD_F_REFILL_BUFFERS,
109 TD_F_SCRAMBLE_BUFFERS = 1U << __TD_F_SCRAMBLE_BUFFERS,
110 TD_F_DO_VERIFY = 1U << __TD_F_DO_VERIFY,
111 TD_F_PROFILE_OPS = 1U << __TD_F_PROFILE_OPS,
112 TD_F_COMPRESS = 1U << __TD_F_COMPRESS,
113 TD_F_COMPRESS_LOG = 1U << __TD_F_COMPRESS_LOG,
114 TD_F_VSTATE_SAVED = 1U << __TD_F_VSTATE_SAVED,
115 TD_F_NEED_LOCK = 1U << __TD_F_NEED_LOCK,
116 TD_F_CHILD = 1U << __TD_F_CHILD,
117 TD_F_NO_PROGRESS = 1U << __TD_F_NO_PROGRESS,
118 TD_F_REGROW_LOGS = 1U << __TD_F_REGROW_LOGS,
119 TD_F_MMAP_KEEP = 1U << __TD_F_MMAP_KEEP,
120 TD_F_DIRS_CREATED = 1U << __TD_F_DIRS_CREATED,
121 TD_F_CHECK_RATE = 1U << __TD_F_CHECK_RATE,
122 TD_F_SYNCS = 1U << __TD_F_SYNCS,
133 FIO_RAND_FILE_SIZE_OFF,
136 FIO_RAND_SEQ_RAND_READ_OFF,
137 FIO_RAND_SEQ_RAND_WRITE_OFF,
138 FIO_RAND_SEQ_RAND_TRIM_OFF,
139 FIO_RAND_START_DELAY,
141 FIO_RAND_POISSON_OFF,
143 FIO_RAND_POISSON2_OFF,
144 FIO_RAND_POISSON3_OFF,
146 FIO_RAND_DEDUPE_WORKING_SET_IX,
155 RATE_PROCESS_LINEAR = 0,
156 RATE_PROCESS_POISSON = 1,
158 THINKTIME_BLOCKS_TYPE_COMPLETE = 0,
159 THINKTIME_BLOCKS_TYPE_ISSUE = 1,
171 * Per-thread/process specific data. Only used for the network client
174 void sk_out_assign(struct sk_out *);
175 void sk_out_drop(void);
177 struct zone_split_index {
179 uint8_t size_perc_prev;
185 * This describes a single thread/process executing a fio job.
188 struct flist_head opt_list;
189 unsigned long long flags;
190 struct thread_options o;
193 unsigned int thread_number;
194 unsigned int subjob_number;
195 unsigned int groupid;
196 struct thread_stat ts __attribute__ ((aligned(8)));
200 struct io_log *slat_log;
201 struct io_log *clat_log;
202 struct io_log *clat_hist_log;
203 struct io_log *lat_log;
204 struct io_log *bw_log;
205 struct io_log *iops_log;
207 struct workqueue log_compress_wq;
209 struct thread_data *parent;
211 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
212 struct timespec bw_sample_time;
214 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
215 struct timespec iops_sample_time;
217 volatile int update_rusage;
218 struct fio_sem *rusage_sem;
219 struct rusage ru_start;
220 struct rusage ru_end;
222 struct fio_file **files;
223 unsigned char *file_locks;
224 unsigned int files_size;
225 unsigned int files_index;
226 unsigned int nr_open_files;
227 unsigned int nr_done_files;
229 unsigned int next_file;
230 struct frand_state next_file_state;
233 struct zipf_state next_file_zipf;
234 struct gauss_state next_file_gauss;
241 double random_center;
248 size_t orig_buffer_size;
249 volatile int runstate;
250 volatile bool terminate;
252 enum fio_ddir last_ddir;
259 uint64_t rand_seeds[FIO_RAND_NR_OFFS];
261 struct frand_state bsrange_state[DDIR_RWDIR_CNT];
262 struct frand_state verify_state;
263 struct frand_state verify_state_last_do_io;
264 struct frand_state trim_state;
265 struct frand_state delay_state;
266 struct frand_state fdp_state;
268 struct frand_state buf_state;
269 struct frand_state buf_state_prev;
270 struct frand_state buf_state_ret;
271 struct frand_state dedupe_state;
272 struct frand_state zone_state;
273 struct frand_state prio_state;
274 struct frand_state dedupe_working_set_index_state;
275 struct frand_state *dedupe_working_set_states;
277 unsigned long long num_unique_pages;
279 struct zone_split_index **zone_state_index;
280 unsigned int num_write_zones;
282 unsigned int verify_batch;
283 unsigned int trim_batch;
285 struct thread_io_list *vstate;
290 * Job default IO priority set with prioclass and prio options.
295 * IO engine hooks, contains everything needed to submit an io_u
296 * to any of the available IO engines.
298 struct ioengine_ops *io_ops;
302 * IO engine private data and dlhandle.
307 * Queue depth of io_u's that fio MIGHT do
309 unsigned int cur_depth;
312 * io_u's about to be committed
314 unsigned int io_u_queued;
317 * io_u's submitted but not completed yet
319 unsigned int io_u_in_flight;
322 * List of free and busy io_u's
324 struct io_u_ring io_u_requeues;
325 struct io_u_queue io_u_freelist;
326 struct io_u_queue io_u_all;
327 pthread_mutex_t io_u_lock;
328 pthread_cond_t free_cond;
331 * async verify offload
333 struct flist_head verify_list;
334 pthread_t *verify_threads;
335 unsigned int nr_verify_threads;
336 pthread_cond_t verify_cond;
337 int verify_thread_exit;
342 uint64_t rate_bps[DDIR_RWDIR_CNT];
343 uint64_t rate_next_io_time[DDIR_RWDIR_CNT];
344 unsigned long long last_rate_check_bytes[DDIR_RWDIR_CNT];
345 unsigned long last_rate_check_blocks[DDIR_RWDIR_CNT];
346 unsigned long long rate_io_issue_bytes[DDIR_RWDIR_CNT];
347 struct timespec last_rate_check_time[DDIR_RWDIR_CNT];
348 int64_t last_usec[DDIR_RWDIR_CNT];
349 struct frand_state poisson_state[DDIR_RWDIR_CNT];
352 * Enforced rate submission/completion workqueue
354 struct workqueue io_wq;
356 uint64_t total_io_size;
357 uint64_t fill_device_size;
362 uint64_t io_issues[DDIR_RWDIR_CNT];
363 uint64_t verify_read_issues;
364 uint64_t io_issue_bytes[DDIR_RWDIR_CNT];
370 uint64_t io_blocks[DDIR_RWDIR_CNT];
371 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
372 uint64_t io_bytes[DDIR_RWDIR_CNT];
373 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
374 uint64_t io_skip_bytes;
377 uint64_t bytes_done[DDIR_RWDIR_CNT];
378 uint64_t bytes_verified;
380 uint64_t *thinktime_blocks_counter;
381 struct timespec last_thinktime;
382 int64_t last_thinktime_blocks;
385 * State for random io, a bitmap of blocks done vs not done
387 struct frand_state random_state;
389 struct timespec start; /* start of this loop */
390 struct timespec epoch; /* time job was started */
391 unsigned long long alternate_epoch; /* Time job was started, clock_gettime's clock_id epoch based. */
392 struct timespec last_issue;
394 struct timespec ts_cache;
395 struct timespec terminate_time;
396 unsigned int ts_cache_nr;
397 unsigned int ts_cache_mask;
401 * Time since last latency_window was started
403 struct timespec latency_ts;
404 unsigned int latency_qd;
405 unsigned int latency_qd_high;
406 unsigned int latency_qd_low;
407 unsigned int latency_failed;
408 unsigned int latency_stable_count;
409 uint64_t latency_ios;
413 * read/write mixed workload state
415 struct frand_state rwmix_state;
416 unsigned long rwmix_issues;
417 enum fio_ddir rwmix_ddir;
418 unsigned int ddir_seq_nr;
421 * rand/seq mixed workload state
423 struct frand_state seq_rand_state[DDIR_RWDIR_CNT];
426 * IO history logs for verification. We use a tree for sorting,
427 * if we are overwriting. Otherwise just use a fifo.
429 struct rb_root io_hist_tree;
430 struct flist_head io_hist_list;
431 unsigned long io_hist_len;
436 struct flist_head io_log_list;
438 unsigned int io_log_blktrace;
439 unsigned int io_log_blktrace_swap;
440 unsigned long long io_log_last_ttime;
441 struct timespec io_log_start_time;
442 unsigned int io_log_current;
443 unsigned int io_log_checkmark;
444 unsigned int io_log_highmark;
445 unsigned int io_log_version;
446 struct timespec io_log_highmark_time;
449 * For tracking/handling discards
451 struct flist_head trim_list;
452 unsigned long trim_entries;
455 * for fileservice, how often to switch to a new file
457 unsigned int file_service_nr;
458 unsigned int file_service_left;
459 struct fio_file *file_service_file;
461 unsigned int sync_file_range_nr;
464 * For generating file sizes
466 struct frand_state file_size_state;
471 unsigned int total_err_count;
474 struct fio_flow *flow;
475 unsigned long long flow_counter;
478 * Can be overloaded by profiles
480 struct prof_io_ops prof_io_ops;
485 struct steadystate_data ss;
487 char verror[FIO_VERROR_SIZE];
491 * for GPU memory management
497 CUdeviceptr dev_mem_ptr;
502 struct thread_segment {
503 struct thread_data *threads;
509 * when should interactive ETA output be generated
517 #define __td_verror(td, err, msg, func) \
519 unsigned int ____e = (err); \
522 (td)->error = ____e; \
523 if (!(td)->first_error) \
524 nowarn_snprintf(td->verror, sizeof(td->verror), \
525 "file:%s:%d, func=%s, error=%s", \
526 __FILE__, __LINE__, (func), (msg)); \
530 #define td_clear_error(td) do { \
533 (td)->parent->error = 0; \
536 #define td_verror(td, err, func) do { \
537 __td_verror((td), (err), strerror((err)), (func)); \
539 __td_verror((td)->parent, (err), strerror((err)), (func)); \
542 #define td_vmsg(td, err, msg, func) do { \
543 __td_verror((td), (err), (msg), (func)); \
545 __td_verror((td)->parent, (err), (msg), (func)); \
548 #define __fio_stringify_1(x) #x
549 #define __fio_stringify(x) __fio_stringify_1(x)
551 #define REAL_MAX_JOBS 4096
552 #define JOBS_PER_SEG 8
553 #define REAL_MAX_SEG (REAL_MAX_JOBS / JOBS_PER_SEG)
555 extern bool exitall_on_terminate;
556 extern unsigned int thread_number;
557 extern unsigned int stat_number;
558 extern unsigned int nr_segments;
559 extern unsigned int cur_segment;
561 extern int output_format;
562 extern int append_terse_output;
563 extern int temp_stall_ts;
564 extern uintptr_t page_mask, page_size;
565 extern bool read_only;
566 extern int eta_print;
567 extern int eta_new_line;
568 extern unsigned int eta_interval_msec;
569 extern unsigned long done_secs;
570 extern int fio_gtod_offload;
571 extern int fio_gtod_cpu;
572 extern enum fio_cs fio_clock_source;
573 extern int fio_clock_source_set;
574 extern int warnings_fatal;
575 extern int terse_version;
576 extern bool is_backend;
577 extern bool is_local_backend;
578 extern int nr_clients;
579 extern bool log_syslog;
580 extern int status_interval;
581 extern const char fio_version_string[];
582 extern char *trigger_file;
583 extern char *trigger_cmd;
584 extern char *trigger_remote_cmd;
585 extern long long trigger_timeout;
586 extern char *aux_path;
588 extern struct thread_segment segments[REAL_MAX_SEG];
590 static inline struct thread_data *tnumber_to_td(unsigned int tnumber)
592 struct thread_segment *seg;
594 seg = &segments[tnumber / JOBS_PER_SEG];
595 return &seg->threads[tnumber & (JOBS_PER_SEG - 1)];
598 static inline bool is_running_backend(void)
600 return is_backend || is_local_backend;
603 extern bool eta_time_within_slack(unsigned int time);
605 static inline void fio_ro_check(const struct thread_data *td, struct io_u *io_u)
607 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)) &&
608 !(io_u->ddir == DDIR_TRIM && !td_trim(td)));
611 static inline bool should_fsync(struct thread_data *td)
613 if (td->last_was_sync)
615 if (td_write(td) || td->o.override_sync)
622 * Init/option functions
624 extern int __must_check fio_init_options(void);
625 extern int __must_check parse_options(int, char **);
626 extern int parse_jobs_ini(char *, int, int, int);
627 extern int parse_cmd_line(int, char **, int);
628 extern int fio_backend(struct sk_out *);
629 extern void reset_fio_state(void);
630 extern void clear_io_state(struct thread_data *, int);
631 extern int fio_options_parse(struct thread_data *, char **, int);
632 extern void fio_keywords_init(void);
633 extern void fio_keywords_exit(void);
634 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
635 extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
636 extern void fio_fill_default_options(struct thread_data *);
637 extern int fio_show_option_help(const char *);
638 extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
639 extern void fio_options_dup_and_init(struct option *);
640 extern char *fio_option_dup_subs(const char *);
641 extern void fio_options_mem_dupe(struct thread_data *);
642 extern void td_fill_rand_seeds(struct thread_data *);
643 extern void add_job_opts(const char **, int);
644 extern int ioengine_load(struct thread_data *);
645 extern bool parse_dryrun(void);
646 extern int fio_running_or_pending_io_threads(void);
647 extern int fio_set_fd_nonblocking(int, const char *);
648 extern void sig_show_status(int sig);
649 extern struct thread_data *get_global_options(void);
651 extern uintptr_t page_mask;
652 extern uintptr_t page_size;
653 extern int initialize_fio(char *envp[]);
654 extern void deinitialize_fio(void);
656 #define FIO_GETOPT_JOB 0x89000000
657 #define FIO_GETOPT_IOENGINE 0x98000000
658 #define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
663 extern void print_thread_status(void);
664 extern void print_status_init(int);
665 extern char *fio_uint_to_kmg(unsigned int val);
668 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
669 * will never back again. It may cycle between running/verififying/fsyncing.
670 * Once the thread reaches TD_EXITED, it is just waiting for the core to
690 #define TD_ENG_FLAG_SHIFT 18
691 #define TD_ENG_FLAG_MASK ((1ULL << 18) - 1)
693 static inline void td_set_ioengine_flags(struct thread_data *td)
695 td->flags = (~(TD_ENG_FLAG_MASK << TD_ENG_FLAG_SHIFT) & td->flags) |
696 ((unsigned long long)td->io_ops->flags << TD_ENG_FLAG_SHIFT);
699 static inline bool td_ioengine_flagged(struct thread_data *td,
700 enum fio_ioengine_flags flags)
702 return ((td->flags >> TD_ENG_FLAG_SHIFT) & flags) != 0;
705 extern void td_set_runstate(struct thread_data *, int);
706 extern int td_bump_runstate(struct thread_data *, int);
707 extern void td_restore_runstate(struct thread_data *, int);
708 extern const char *runstate_to_name(int runstate);
711 * Allow 60 seconds for a job to quit on its own, otherwise reap with
714 #define FIO_REAP_TIMEOUT 300
719 TERMINATE_STONEWALL = 2,
723 extern void fio_terminate_threads(unsigned int, unsigned int);
724 extern void fio_mark_td_terminate(struct thread_data *);
729 extern int __must_check fio_pin_memory(struct thread_data *);
730 extern void fio_unpin_memory(struct thread_data *);
731 extern int __must_check allocate_io_mem(struct thread_data *);
732 extern void free_io_mem(struct thread_data *);
733 extern void free_threads_shm(void);
736 #define PTR_ALIGN(ptr, mask) \
737 (char *) (((uintptr_t) (ptr) + (mask)) & ~(mask))
741 * Reset stats after ramp time completes
743 extern void reset_all_stats(struct thread_data *);
745 extern int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret,
746 enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify,
747 struct timespec *comp_time);
750 * Latency target helpers
752 extern void lat_target_check(struct thread_data *);
753 extern void lat_target_init(struct thread_data *);
754 extern void lat_target_reset(struct thread_data *);
757 * Iterates all threads/processes within all the defined jobs
759 * for_each_td(var_name_for_td) {
760 * << bodoy of your loop >>
761 * Note: internally-scoped loop index availble as __td_index
762 * } end_for_each_td()
764 #define for_each_td(td) \
767 struct thread_data *(td); \
768 for (__td_index = 0, (td) = &segments[0].threads[0];\
769 __td_index < (int) thread_number; __td_index++, (td) = tnumber_to_td(__td_index))
770 #define for_each_td_index() \
773 for (__td_index = 0; __td_index < (int) thread_number; __td_index++)
774 #define end_for_each() }
776 #define for_each_file(td, f, i) \
777 if ((td)->files_index) \
778 for ((i) = 0, (f) = (td)->files[0]; \
779 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
782 static inline bool fio_fill_issue_time(struct thread_data *td)
784 if (td->o.read_iolog_file ||
785 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
791 static inline bool option_check_rate(struct thread_data *td, enum fio_ddir ddir)
793 struct thread_options *o = &td->o;
796 * If some rate setting was given, we need to check it
798 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
799 o->rate_iops_min[ddir])
805 static inline bool should_check_rate(struct thread_data *td)
807 return (td->flags & TD_F_CHECK_RATE) != 0;
810 static inline unsigned long long td_max_bs(struct thread_data *td)
812 unsigned long long max_bs;
814 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
815 return max(td->o.max_bs[DDIR_TRIM], max_bs);
818 static inline unsigned long long td_min_bs(struct thread_data *td)
820 unsigned long long min_bs;
822 min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
823 return min(td->o.min_bs[DDIR_TRIM], min_bs);
826 static inline bool td_async_processing(struct thread_data *td)
828 return (td->flags & TD_F_NEED_LOCK) != 0;
831 static inline bool td_offload_overlap(struct thread_data *td)
833 return td->o.serialize_overlap && td->o.io_submit_mode == IO_MODE_OFFLOAD;
837 * We currently only need to do locking if we have verifier threads
838 * accessing our internal structures too
840 static inline void __td_io_u_lock(struct thread_data *td)
842 pthread_mutex_lock(&td->io_u_lock);
845 static inline void __td_io_u_unlock(struct thread_data *td)
847 pthread_mutex_unlock(&td->io_u_lock);
850 static inline void td_io_u_free_notify(struct thread_data *td)
852 if (td_async_processing(td))
853 pthread_cond_signal(&td->free_cond);
856 static inline void td_flags_clear(struct thread_data *td, unsigned int *flags,
859 if (!td_async_processing(td))
862 __sync_fetch_and_and(flags, ~value);
865 static inline void td_flags_set(struct thread_data *td, unsigned int *flags,
868 if (!td_async_processing(td))
871 __sync_fetch_and_or(flags, value);
874 extern const char *fio_get_arch_string(int);
875 extern const char *fio_get_os_string(int);
878 __FIO_OUTPUT_TERSE = 0,
879 __FIO_OUTPUT_JSON = 1,
880 __FIO_OUTPUT_NORMAL = 2,
881 __FIO_OUTPUT_JSON_PLUS = 3,
884 FIO_OUTPUT_TERSE = 1U << __FIO_OUTPUT_TERSE,
885 FIO_OUTPUT_JSON = 1U << __FIO_OUTPUT_JSON,
886 FIO_OUTPUT_NORMAL = 1U << __FIO_OUTPUT_NORMAL,
887 FIO_OUTPUT_JSON_PLUS = 1U << __FIO_OUTPUT_JSON_PLUS,
891 FIO_RAND_DIST_RANDOM = 0,
893 FIO_RAND_DIST_PARETO,
896 FIO_RAND_DIST_ZONED_ABS,
899 #define FIO_DEF_ZIPF 1.1
900 #define FIO_DEF_PARETO 0.2
903 FIO_RAND_GEN_TAUSWORTHE = 0,
905 FIO_RAND_GEN_TAUSWORTHE64,
913 extern void exec_trigger(const char *);
914 extern void check_trigger_file(void);
916 extern bool in_flight_overlap(struct io_u_queue *q, struct io_u *io_u);
917 extern pthread_mutex_t overlap_check;
919 static inline void *fio_memalign(size_t alignment, size_t size, bool shared)
921 return __fio_memalign(alignment, size, shared ? smalloc : malloc);
924 static inline void fio_memfree(void *ptr, size_t size, bool shared)
926 return __fio_memfree(ptr, size, shared ? sfree : free);