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,
154 RATE_PROCESS_LINEAR = 0,
155 RATE_PROCESS_POISSON = 1,
157 THINKTIME_BLOCKS_TYPE_COMPLETE = 0,
158 THINKTIME_BLOCKS_TYPE_ISSUE = 1,
170 * Per-thread/process specific data. Only used for the network client
173 void sk_out_assign(struct sk_out *);
174 void sk_out_drop(void);
176 struct zone_split_index {
178 uint8_t size_perc_prev;
184 * This describes a single thread/process executing a fio job.
187 struct flist_head opt_list;
188 unsigned long long flags;
189 struct thread_options o;
192 unsigned int thread_number;
193 unsigned int subjob_number;
194 unsigned int groupid;
195 struct thread_stat ts __attribute__ ((aligned(8)));
199 struct io_log *slat_log;
200 struct io_log *clat_log;
201 struct io_log *clat_hist_log;
202 struct io_log *lat_log;
203 struct io_log *bw_log;
204 struct io_log *iops_log;
206 struct workqueue log_compress_wq;
208 struct thread_data *parent;
210 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
211 struct timespec bw_sample_time;
213 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
214 struct timespec iops_sample_time;
216 volatile int update_rusage;
217 struct fio_sem *rusage_sem;
218 struct rusage ru_start;
219 struct rusage ru_end;
221 struct fio_file **files;
222 unsigned char *file_locks;
223 unsigned int files_size;
224 unsigned int files_index;
225 unsigned int nr_open_files;
226 unsigned int nr_done_files;
228 unsigned int next_file;
229 struct frand_state next_file_state;
232 struct zipf_state next_file_zipf;
233 struct gauss_state next_file_gauss;
240 double random_center;
247 size_t orig_buffer_size;
248 volatile int runstate;
249 volatile bool terminate;
251 enum fio_ddir last_ddir;
258 uint64_t rand_seeds[FIO_RAND_NR_OFFS];
260 struct frand_state bsrange_state[DDIR_RWDIR_CNT];
261 struct frand_state verify_state;
262 struct frand_state verify_state_last_do_io;
263 struct frand_state trim_state;
264 struct frand_state delay_state;
266 struct frand_state buf_state;
267 struct frand_state buf_state_prev;
268 struct frand_state buf_state_ret;
269 struct frand_state dedupe_state;
270 struct frand_state zone_state;
271 struct frand_state prio_state;
272 struct frand_state dedupe_working_set_index_state;
273 struct frand_state *dedupe_working_set_states;
275 unsigned long long num_unique_pages;
277 struct zone_split_index **zone_state_index;
278 unsigned int num_open_zones;
280 unsigned int verify_batch;
281 unsigned int trim_batch;
283 struct thread_io_list *vstate;
288 * Job default IO priority set with prioclass and prio options.
293 * IO engine hooks, contains everything needed to submit an io_u
294 * to any of the available IO engines.
296 struct ioengine_ops *io_ops;
300 * IO engine private data and dlhandle.
305 * Queue depth of io_u's that fio MIGHT do
307 unsigned int cur_depth;
310 * io_u's about to be committed
312 unsigned int io_u_queued;
315 * io_u's submitted but not completed yet
317 unsigned int io_u_in_flight;
320 * List of free and busy io_u's
322 struct io_u_ring io_u_requeues;
323 struct io_u_queue io_u_freelist;
324 struct io_u_queue io_u_all;
325 pthread_mutex_t io_u_lock;
326 pthread_cond_t free_cond;
329 * async verify offload
331 struct flist_head verify_list;
332 pthread_t *verify_threads;
333 unsigned int nr_verify_threads;
334 pthread_cond_t verify_cond;
335 int verify_thread_exit;
340 uint64_t rate_bps[DDIR_RWDIR_CNT];
341 uint64_t rate_next_io_time[DDIR_RWDIR_CNT];
342 unsigned long long last_rate_check_bytes[DDIR_RWDIR_CNT];
343 unsigned long last_rate_check_blocks[DDIR_RWDIR_CNT];
344 unsigned long long rate_io_issue_bytes[DDIR_RWDIR_CNT];
345 struct timespec last_rate_check_time[DDIR_RWDIR_CNT];
346 int64_t last_usec[DDIR_RWDIR_CNT];
347 struct frand_state poisson_state[DDIR_RWDIR_CNT];
350 * Enforced rate submission/completion workqueue
352 struct workqueue io_wq;
354 uint64_t total_io_size;
355 uint64_t fill_device_size;
360 uint64_t io_issues[DDIR_RWDIR_CNT];
361 uint64_t verify_read_issues;
362 uint64_t io_issue_bytes[DDIR_RWDIR_CNT];
368 uint64_t io_blocks[DDIR_RWDIR_CNT];
369 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
370 uint64_t io_bytes[DDIR_RWDIR_CNT];
371 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
372 uint64_t io_skip_bytes;
375 uint64_t bytes_done[DDIR_RWDIR_CNT];
376 uint64_t bytes_verified;
378 uint64_t *thinktime_blocks_counter;
379 struct timespec last_thinktime;
380 int64_t last_thinktime_blocks;
383 * State for random io, a bitmap of blocks done vs not done
385 struct frand_state random_state;
387 struct timespec start; /* start of this loop */
388 struct timespec epoch; /* time job was started */
389 unsigned long long alternate_epoch; /* Time job was started, clock_gettime's clock_id epoch based. */
390 struct timespec last_issue;
392 struct timespec ts_cache;
393 struct timespec terminate_time;
394 unsigned int ts_cache_nr;
395 unsigned int ts_cache_mask;
399 * Time since last latency_window was started
401 struct timespec latency_ts;
402 unsigned int latency_qd;
403 unsigned int latency_qd_high;
404 unsigned int latency_qd_low;
405 unsigned int latency_failed;
406 unsigned int latency_stable_count;
407 uint64_t latency_ios;
411 * read/write mixed workload state
413 struct frand_state rwmix_state;
414 unsigned long rwmix_issues;
415 enum fio_ddir rwmix_ddir;
416 unsigned int ddir_seq_nr;
419 * rand/seq mixed workload state
421 struct frand_state seq_rand_state[DDIR_RWDIR_CNT];
424 * IO history logs for verification. We use a tree for sorting,
425 * if we are overwriting. Otherwise just use a fifo.
427 struct rb_root io_hist_tree;
428 struct flist_head io_hist_list;
429 unsigned long io_hist_len;
434 struct flist_head io_log_list;
436 unsigned int io_log_blktrace;
437 unsigned int io_log_blktrace_swap;
438 unsigned long long io_log_last_ttime;
439 struct timespec io_log_start_time;
440 unsigned int io_log_current;
441 unsigned int io_log_checkmark;
442 unsigned int io_log_highmark;
443 unsigned int io_log_version;
444 struct timespec io_log_highmark_time;
447 * For tracking/handling discards
449 struct flist_head trim_list;
450 unsigned long trim_entries;
453 * for fileservice, how often to switch to a new file
455 unsigned int file_service_nr;
456 unsigned int file_service_left;
457 struct fio_file *file_service_file;
459 unsigned int sync_file_range_nr;
462 * For generating file sizes
464 struct frand_state file_size_state;
469 unsigned int total_err_count;
472 struct fio_flow *flow;
473 unsigned long long flow_counter;
476 * Can be overloaded by profiles
478 struct prof_io_ops prof_io_ops;
483 struct steadystate_data ss;
485 char verror[FIO_VERROR_SIZE];
489 * for GPU memory management
495 CUdeviceptr dev_mem_ptr;
500 struct thread_segment {
501 struct thread_data *threads;
507 * when should interactive ETA output be generated
515 #define __td_verror(td, err, msg, func) \
517 unsigned int ____e = (err); \
520 (td)->error = ____e; \
521 if (!(td)->first_error) \
522 nowarn_snprintf(td->verror, sizeof(td->verror), \
523 "file:%s:%d, func=%s, error=%s", \
524 __FILE__, __LINE__, (func), (msg)); \
528 #define td_clear_error(td) do { \
531 (td)->parent->error = 0; \
534 #define td_verror(td, err, func) do { \
535 __td_verror((td), (err), strerror((err)), (func)); \
537 __td_verror((td)->parent, (err), strerror((err)), (func)); \
540 #define td_vmsg(td, err, msg, func) do { \
541 __td_verror((td), (err), (msg), (func)); \
543 __td_verror((td)->parent, (err), (msg), (func)); \
546 #define __fio_stringify_1(x) #x
547 #define __fio_stringify(x) __fio_stringify_1(x)
549 #define REAL_MAX_JOBS 4096
550 #define JOBS_PER_SEG 8
551 #define REAL_MAX_SEG (REAL_MAX_JOBS / JOBS_PER_SEG)
553 extern bool exitall_on_terminate;
554 extern unsigned int thread_number;
555 extern unsigned int stat_number;
556 extern unsigned int nr_segments;
557 extern unsigned int cur_segment;
559 extern int output_format;
560 extern int append_terse_output;
561 extern int temp_stall_ts;
562 extern uintptr_t page_mask, page_size;
563 extern bool read_only;
564 extern int eta_print;
565 extern int eta_new_line;
566 extern unsigned int eta_interval_msec;
567 extern unsigned long done_secs;
568 extern int fio_gtod_offload;
569 extern int fio_gtod_cpu;
570 extern enum fio_cs fio_clock_source;
571 extern int fio_clock_source_set;
572 extern int warnings_fatal;
573 extern int terse_version;
574 extern bool is_backend;
575 extern bool is_local_backend;
576 extern int nr_clients;
577 extern bool log_syslog;
578 extern int status_interval;
579 extern const char fio_version_string[];
580 extern char *trigger_file;
581 extern char *trigger_cmd;
582 extern char *trigger_remote_cmd;
583 extern long long trigger_timeout;
584 extern char *aux_path;
586 extern struct thread_segment segments[REAL_MAX_SEG];
588 static inline struct thread_data *tnumber_to_td(unsigned int tnumber)
590 struct thread_segment *seg;
592 seg = &segments[tnumber / JOBS_PER_SEG];
593 return &seg->threads[tnumber & (JOBS_PER_SEG - 1)];
596 static inline bool is_running_backend(void)
598 return is_backend || is_local_backend;
601 extern bool eta_time_within_slack(unsigned int time);
603 static inline void fio_ro_check(const struct thread_data *td, struct io_u *io_u)
605 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)) &&
606 !(io_u->ddir == DDIR_TRIM && !td_trim(td)));
609 static inline bool should_fsync(struct thread_data *td)
611 if (td->last_was_sync)
613 if (td_write(td) || td->o.override_sync)
620 * Init/option functions
622 extern int __must_check fio_init_options(void);
623 extern int __must_check parse_options(int, char **);
624 extern int parse_jobs_ini(char *, int, int, int);
625 extern int parse_cmd_line(int, char **, int);
626 extern int fio_backend(struct sk_out *);
627 extern void reset_fio_state(void);
628 extern void clear_io_state(struct thread_data *, int);
629 extern int fio_options_parse(struct thread_data *, char **, int);
630 extern void fio_keywords_init(void);
631 extern void fio_keywords_exit(void);
632 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
633 extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
634 extern void fio_fill_default_options(struct thread_data *);
635 extern int fio_show_option_help(const char *);
636 extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
637 extern void fio_options_dup_and_init(struct option *);
638 extern char *fio_option_dup_subs(const char *);
639 extern void fio_options_mem_dupe(struct thread_data *);
640 extern void td_fill_rand_seeds(struct thread_data *);
641 extern void add_job_opts(const char **, int);
642 extern int ioengine_load(struct thread_data *);
643 extern bool parse_dryrun(void);
644 extern int fio_running_or_pending_io_threads(void);
645 extern int fio_set_fd_nonblocking(int, const char *);
646 extern void sig_show_status(int sig);
647 extern struct thread_data *get_global_options(void);
649 extern uintptr_t page_mask;
650 extern uintptr_t page_size;
651 extern int initialize_fio(char *envp[]);
652 extern void deinitialize_fio(void);
654 #define FIO_GETOPT_JOB 0x89000000
655 #define FIO_GETOPT_IOENGINE 0x98000000
656 #define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
661 extern void print_thread_status(void);
662 extern void print_status_init(int);
663 extern char *fio_uint_to_kmg(unsigned int val);
666 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
667 * will never back again. It may cycle between running/verififying/fsyncing.
668 * Once the thread reaches TD_EXITED, it is just waiting for the core to
688 #define TD_ENG_FLAG_SHIFT 18
689 #define TD_ENG_FLAG_MASK ((1ULL << 18) - 1)
691 static inline void td_set_ioengine_flags(struct thread_data *td)
693 td->flags = (~(TD_ENG_FLAG_MASK << TD_ENG_FLAG_SHIFT) & td->flags) |
694 ((unsigned long long)td->io_ops->flags << TD_ENG_FLAG_SHIFT);
697 static inline bool td_ioengine_flagged(struct thread_data *td,
698 enum fio_ioengine_flags flags)
700 return ((td->flags >> TD_ENG_FLAG_SHIFT) & flags) != 0;
703 extern void td_set_runstate(struct thread_data *, int);
704 extern int td_bump_runstate(struct thread_data *, int);
705 extern void td_restore_runstate(struct thread_data *, int);
706 extern const char *runstate_to_name(int runstate);
709 * Allow 60 seconds for a job to quit on its own, otherwise reap with
712 #define FIO_REAP_TIMEOUT 300
717 TERMINATE_STONEWALL = 2,
721 extern void fio_terminate_threads(unsigned int, unsigned int);
722 extern void fio_mark_td_terminate(struct thread_data *);
727 extern int __must_check fio_pin_memory(struct thread_data *);
728 extern void fio_unpin_memory(struct thread_data *);
729 extern int __must_check allocate_io_mem(struct thread_data *);
730 extern void free_io_mem(struct thread_data *);
731 extern void free_threads_shm(void);
734 #define PTR_ALIGN(ptr, mask) \
735 (char *) (((uintptr_t) (ptr) + (mask)) & ~(mask))
739 * Reset stats after ramp time completes
741 extern void reset_all_stats(struct thread_data *);
743 extern int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret,
744 enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify,
745 struct timespec *comp_time);
748 * Latency target helpers
750 extern void lat_target_check(struct thread_data *);
751 extern void lat_target_init(struct thread_data *);
752 extern void lat_target_reset(struct thread_data *);
755 * Iterates all threads/processes within all the defined jobs
757 * for_each_td(var_name_for_td) {
758 * << bodoy of your loop >>
759 * Note: internally-scoped loop index availble as __td_index
760 * } end_for_each_td()
762 #define for_each_td(td) \
765 struct thread_data *(td); \
766 for (__td_index = 0, (td) = &segments[0].threads[0];\
767 __td_index < (int) thread_number; __td_index++, (td) = tnumber_to_td(__td_index))
768 #define for_each_td_index() \
771 for (__td_index = 0; __td_index < (int) thread_number; __td_index++)
772 #define end_for_each() }
774 #define for_each_file(td, f, i) \
775 if ((td)->files_index) \
776 for ((i) = 0, (f) = (td)->files[0]; \
777 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
780 static inline bool fio_fill_issue_time(struct thread_data *td)
782 if (td->o.read_iolog_file ||
783 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
789 static inline bool option_check_rate(struct thread_data *td, enum fio_ddir ddir)
791 struct thread_options *o = &td->o;
794 * If some rate setting was given, we need to check it
796 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
797 o->rate_iops_min[ddir])
803 static inline bool should_check_rate(struct thread_data *td)
805 return (td->flags & TD_F_CHECK_RATE) != 0;
808 static inline unsigned long long td_max_bs(struct thread_data *td)
810 unsigned long long max_bs;
812 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
813 return max(td->o.max_bs[DDIR_TRIM], max_bs);
816 static inline unsigned long long td_min_bs(struct thread_data *td)
818 unsigned long long min_bs;
820 min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
821 return min(td->o.min_bs[DDIR_TRIM], min_bs);
824 static inline bool td_async_processing(struct thread_data *td)
826 return (td->flags & TD_F_NEED_LOCK) != 0;
829 static inline bool td_offload_overlap(struct thread_data *td)
831 return td->o.serialize_overlap && td->o.io_submit_mode == IO_MODE_OFFLOAD;
835 * We currently only need to do locking if we have verifier threads
836 * accessing our internal structures too
838 static inline void __td_io_u_lock(struct thread_data *td)
840 pthread_mutex_lock(&td->io_u_lock);
843 static inline void __td_io_u_unlock(struct thread_data *td)
845 pthread_mutex_unlock(&td->io_u_lock);
848 static inline void td_io_u_free_notify(struct thread_data *td)
850 if (td_async_processing(td))
851 pthread_cond_signal(&td->free_cond);
854 static inline void td_flags_clear(struct thread_data *td, unsigned int *flags,
857 if (!td_async_processing(td))
860 __sync_fetch_and_and(flags, ~value);
863 static inline void td_flags_set(struct thread_data *td, unsigned int *flags,
866 if (!td_async_processing(td))
869 __sync_fetch_and_or(flags, value);
872 extern const char *fio_get_arch_string(int);
873 extern const char *fio_get_os_string(int);
876 __FIO_OUTPUT_TERSE = 0,
877 __FIO_OUTPUT_JSON = 1,
878 __FIO_OUTPUT_NORMAL = 2,
879 __FIO_OUTPUT_JSON_PLUS = 3,
882 FIO_OUTPUT_TERSE = 1U << __FIO_OUTPUT_TERSE,
883 FIO_OUTPUT_JSON = 1U << __FIO_OUTPUT_JSON,
884 FIO_OUTPUT_NORMAL = 1U << __FIO_OUTPUT_NORMAL,
885 FIO_OUTPUT_JSON_PLUS = 1U << __FIO_OUTPUT_JSON_PLUS,
889 FIO_RAND_DIST_RANDOM = 0,
891 FIO_RAND_DIST_PARETO,
894 FIO_RAND_DIST_ZONED_ABS,
897 #define FIO_DEF_ZIPF 1.1
898 #define FIO_DEF_PARETO 0.2
901 FIO_RAND_GEN_TAUSWORTHE = 0,
903 FIO_RAND_GEN_TAUSWORTHE64,
911 extern void exec_trigger(const char *);
912 extern void check_trigger_file(void);
914 extern bool in_flight_overlap(struct io_u_queue *q, struct io_u *io_u);
915 extern pthread_mutex_t overlap_check;
917 static inline void *fio_memalign(size_t alignment, size_t size, bool shared)
919 return __fio_memalign(alignment, size, shared ? smalloc : malloc);
922 static inline void fio_memfree(void *ptr, size_t size, bool shared)
924 return __fio_memfree(ptr, size, shared ? sfree : free);