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,
169 * Per-thread/process specific data. Only used for the network client
172 void sk_out_assign(struct sk_out *);
173 void sk_out_drop(void);
175 struct zone_split_index {
177 uint8_t size_perc_prev;
183 * This describes a single thread/process executing a fio job.
186 struct flist_head opt_list;
188 struct thread_options o;
191 unsigned int thread_number;
192 unsigned int subjob_number;
193 unsigned int groupid;
194 struct thread_stat ts __attribute__ ((aligned(8)));
198 struct io_log *slat_log;
199 struct io_log *clat_log;
200 struct io_log *clat_hist_log;
201 struct io_log *lat_log;
202 struct io_log *bw_log;
203 struct io_log *iops_log;
205 struct workqueue log_compress_wq;
207 struct thread_data *parent;
209 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
210 struct timespec bw_sample_time;
212 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
213 struct timespec iops_sample_time;
215 volatile int update_rusage;
216 struct fio_sem *rusage_sem;
217 struct rusage ru_start;
218 struct rusage ru_end;
220 struct fio_file **files;
221 unsigned char *file_locks;
222 unsigned int files_size;
223 unsigned int files_index;
224 unsigned int nr_open_files;
225 unsigned int nr_done_files;
227 unsigned int next_file;
228 struct frand_state next_file_state;
231 struct zipf_state next_file_zipf;
232 struct gauss_state next_file_gauss;
239 double random_center;
246 size_t orig_buffer_size;
247 volatile int runstate;
248 volatile bool terminate;
250 enum fio_ddir last_ddir;
257 uint64_t rand_seeds[FIO_RAND_NR_OFFS];
259 struct frand_state bsrange_state[DDIR_RWDIR_CNT];
260 struct frand_state verify_state;
261 struct frand_state trim_state;
262 struct frand_state delay_state;
264 struct frand_state buf_state;
265 struct frand_state buf_state_prev;
266 struct frand_state buf_state_ret;
267 struct frand_state dedupe_state;
268 struct frand_state zone_state;
269 struct frand_state prio_state;
270 struct frand_state dedupe_working_set_index_state;
271 struct frand_state *dedupe_working_set_states;
273 unsigned long long num_unique_pages;
275 struct zone_split_index **zone_state_index;
276 unsigned int num_open_zones;
278 unsigned int verify_batch;
279 unsigned int trim_batch;
281 struct thread_io_list *vstate;
286 * Job default IO priority set with prioclass and prio options.
291 * IO engine hooks, contains everything needed to submit an io_u
292 * to any of the available IO engines.
294 struct ioengine_ops *io_ops;
298 * IO engine private data and dlhandle.
303 * Queue depth of io_u's that fio MIGHT do
305 unsigned int cur_depth;
308 * io_u's about to be committed
310 unsigned int io_u_queued;
313 * io_u's submitted but not completed yet
315 unsigned int io_u_in_flight;
318 * List of free and busy io_u's
320 struct io_u_ring io_u_requeues;
321 struct io_u_queue io_u_freelist;
322 struct io_u_queue io_u_all;
323 pthread_mutex_t io_u_lock;
324 pthread_cond_t free_cond;
327 * async verify offload
329 struct flist_head verify_list;
330 pthread_t *verify_threads;
331 unsigned int nr_verify_threads;
332 pthread_cond_t verify_cond;
333 int verify_thread_exit;
338 uint64_t rate_bps[DDIR_RWDIR_CNT];
339 uint64_t rate_next_io_time[DDIR_RWDIR_CNT];
340 unsigned long long last_rate_check_bytes[DDIR_RWDIR_CNT];
341 unsigned long last_rate_check_blocks[DDIR_RWDIR_CNT];
342 unsigned long long rate_io_issue_bytes[DDIR_RWDIR_CNT];
343 struct timespec last_rate_check_time[DDIR_RWDIR_CNT];
344 int64_t last_usec[DDIR_RWDIR_CNT];
345 struct frand_state poisson_state[DDIR_RWDIR_CNT];
348 * Enforced rate submission/completion workqueue
350 struct workqueue io_wq;
352 uint64_t total_io_size;
353 uint64_t fill_device_size;
358 uint64_t io_issues[DDIR_RWDIR_CNT];
359 uint64_t io_issue_bytes[DDIR_RWDIR_CNT];
365 uint64_t io_blocks[DDIR_RWDIR_CNT];
366 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
367 uint64_t io_bytes[DDIR_RWDIR_CNT];
368 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
369 uint64_t io_skip_bytes;
372 uint64_t bytes_done[DDIR_RWDIR_CNT];
374 uint64_t *thinktime_blocks_counter;
375 struct timespec last_thinktime;
376 uint64_t last_thinktime_blocks;
379 * State for random io, a bitmap of blocks done vs not done
381 struct frand_state random_state;
383 struct timespec start; /* start of this loop */
384 struct timespec epoch; /* time job was started */
385 unsigned long long alternate_epoch; /* Time job was started, clock_gettime's clock_id epoch based. */
386 struct timespec last_issue;
388 struct timespec ts_cache;
389 struct timespec terminate_time;
390 unsigned int ts_cache_nr;
391 unsigned int ts_cache_mask;
395 * Time since last latency_window was started
397 struct timespec latency_ts;
398 unsigned int latency_qd;
399 unsigned int latency_qd_high;
400 unsigned int latency_qd_low;
401 unsigned int latency_failed;
402 unsigned int latency_stable_count;
403 uint64_t latency_ios;
407 * read/write mixed workload state
409 struct frand_state rwmix_state;
410 unsigned long rwmix_issues;
411 enum fio_ddir rwmix_ddir;
412 unsigned int ddir_seq_nr;
415 * rand/seq mixed workload state
417 struct frand_state seq_rand_state[DDIR_RWDIR_CNT];
420 * IO history logs for verification. We use a tree for sorting,
421 * if we are overwriting. Otherwise just use a fifo.
423 struct rb_root io_hist_tree;
424 struct flist_head io_hist_list;
425 unsigned long io_hist_len;
430 struct flist_head io_log_list;
432 unsigned int io_log_blktrace;
433 unsigned int io_log_blktrace_swap;
434 unsigned long long io_log_blktrace_last_ttime;
435 unsigned int io_log_current;
436 unsigned int io_log_checkmark;
437 unsigned int io_log_highmark;
438 struct timespec io_log_highmark_time;
441 * For tracking/handling discards
443 struct flist_head trim_list;
444 unsigned long trim_entries;
447 * for fileservice, how often to switch to a new file
449 unsigned int file_service_nr;
450 unsigned int file_service_left;
451 struct fio_file *file_service_file;
453 unsigned int sync_file_range_nr;
456 * For generating file sizes
458 struct frand_state file_size_state;
463 unsigned int total_err_count;
466 struct fio_flow *flow;
467 unsigned long long flow_counter;
470 * Can be overloaded by profiles
472 struct prof_io_ops prof_io_ops;
477 struct steadystate_data ss;
479 char verror[FIO_VERROR_SIZE];
483 * for GPU memory management
489 CUdeviceptr dev_mem_ptr;
494 struct thread_segment {
495 struct thread_data *threads;
501 * when should interactive ETA output be generated
509 #define __td_verror(td, err, msg, func) \
511 unsigned int ____e = (err); \
514 (td)->error = ____e; \
515 if (!(td)->first_error) \
516 nowarn_snprintf(td->verror, sizeof(td->verror), \
517 "file:%s:%d, func=%s, error=%s", \
518 __FILE__, __LINE__, (func), (msg)); \
522 #define td_clear_error(td) do { \
525 (td)->parent->error = 0; \
528 #define td_verror(td, err, func) do { \
529 __td_verror((td), (err), strerror((err)), (func)); \
531 __td_verror((td)->parent, (err), strerror((err)), (func)); \
534 #define td_vmsg(td, err, msg, func) do { \
535 __td_verror((td), (err), (msg), (func)); \
537 __td_verror((td)->parent, (err), (msg), (func)); \
540 #define __fio_stringify_1(x) #x
541 #define __fio_stringify(x) __fio_stringify_1(x)
543 #define REAL_MAX_JOBS 4096
544 #define JOBS_PER_SEG 8
545 #define REAL_MAX_SEG (REAL_MAX_JOBS / JOBS_PER_SEG)
547 extern bool exitall_on_terminate;
548 extern unsigned int thread_number;
549 extern unsigned int stat_number;
550 extern unsigned int nr_segments;
551 extern unsigned int cur_segment;
553 extern int output_format;
554 extern int append_terse_output;
555 extern int temp_stall_ts;
556 extern uintptr_t page_mask, page_size;
557 extern bool read_only;
558 extern int eta_print;
559 extern int eta_new_line;
560 extern unsigned int eta_interval_msec;
561 extern unsigned long done_secs;
562 extern int fio_gtod_offload;
563 extern int fio_gtod_cpu;
564 extern enum fio_cs fio_clock_source;
565 extern int fio_clock_source_set;
566 extern int warnings_fatal;
567 extern int terse_version;
568 extern bool is_backend;
569 extern bool is_local_backend;
570 extern int nr_clients;
571 extern bool log_syslog;
572 extern int status_interval;
573 extern const char fio_version_string[];
574 extern char *trigger_file;
575 extern char *trigger_cmd;
576 extern char *trigger_remote_cmd;
577 extern long long trigger_timeout;
578 extern char *aux_path;
580 extern struct thread_segment segments[REAL_MAX_SEG];
582 static inline struct thread_data *tnumber_to_td(unsigned int tnumber)
584 struct thread_segment *seg;
586 seg = &segments[tnumber / JOBS_PER_SEG];
587 return &seg->threads[tnumber & (JOBS_PER_SEG - 1)];
590 static inline bool is_running_backend(void)
592 return is_backend || is_local_backend;
595 extern bool eta_time_within_slack(unsigned int time);
597 static inline void fio_ro_check(const struct thread_data *td, struct io_u *io_u)
599 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)) &&
600 !(io_u->ddir == DDIR_TRIM && !td_trim(td)));
603 static inline bool should_fsync(struct thread_data *td)
605 if (td->last_was_sync)
607 if (td_write(td) || td->o.override_sync)
614 * Init/option functions
616 extern int __must_check fio_init_options(void);
617 extern int __must_check parse_options(int, char **);
618 extern int parse_jobs_ini(char *, int, int, int);
619 extern int parse_cmd_line(int, char **, int);
620 extern int fio_backend(struct sk_out *);
621 extern void reset_fio_state(void);
622 extern void clear_io_state(struct thread_data *, int);
623 extern int fio_options_parse(struct thread_data *, char **, int);
624 extern void fio_keywords_init(void);
625 extern void fio_keywords_exit(void);
626 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
627 extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
628 extern void fio_fill_default_options(struct thread_data *);
629 extern int fio_show_option_help(const char *);
630 extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
631 extern void fio_options_dup_and_init(struct option *);
632 extern char *fio_option_dup_subs(const char *);
633 extern void fio_options_mem_dupe(struct thread_data *);
634 extern void td_fill_rand_seeds(struct thread_data *);
635 extern void td_fill_verify_state_seed(struct thread_data *);
636 extern void add_job_opts(const char **, int);
637 extern int ioengine_load(struct thread_data *);
638 extern bool parse_dryrun(void);
639 extern int fio_running_or_pending_io_threads(void);
640 extern int fio_set_fd_nonblocking(int, const char *);
641 extern void sig_show_status(int sig);
642 extern struct thread_data *get_global_options(void);
644 extern uintptr_t page_mask;
645 extern uintptr_t page_size;
646 extern int initialize_fio(char *envp[]);
647 extern void deinitialize_fio(void);
649 #define FIO_GETOPT_JOB 0x89000000
650 #define FIO_GETOPT_IOENGINE 0x98000000
651 #define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
656 extern void print_thread_status(void);
657 extern void print_status_init(int);
658 extern char *fio_uint_to_kmg(unsigned int val);
661 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
662 * will never back again. It may cycle between running/verififying/fsyncing.
663 * Once the thread reaches TD_EXITED, it is just waiting for the core to
683 #define TD_ENG_FLAG_SHIFT 18
684 #define TD_ENG_FLAG_MASK ((1U << 18) - 1)
686 static inline void td_set_ioengine_flags(struct thread_data *td)
688 td->flags = (~(TD_ENG_FLAG_MASK << TD_ENG_FLAG_SHIFT) & td->flags) |
689 (td->io_ops->flags << TD_ENG_FLAG_SHIFT);
692 static inline bool td_ioengine_flagged(struct thread_data *td,
693 enum fio_ioengine_flags flags)
695 return ((td->flags >> TD_ENG_FLAG_SHIFT) & flags) != 0;
698 extern void td_set_runstate(struct thread_data *, int);
699 extern int td_bump_runstate(struct thread_data *, int);
700 extern void td_restore_runstate(struct thread_data *, int);
701 extern const char *runstate_to_name(int runstate);
704 * Allow 60 seconds for a job to quit on its own, otherwise reap with
707 #define FIO_REAP_TIMEOUT 300
712 TERMINATE_STONEWALL = 2,
716 extern void fio_terminate_threads(unsigned int, unsigned int);
717 extern void fio_mark_td_terminate(struct thread_data *);
722 extern int __must_check fio_pin_memory(struct thread_data *);
723 extern void fio_unpin_memory(struct thread_data *);
724 extern int __must_check allocate_io_mem(struct thread_data *);
725 extern void free_io_mem(struct thread_data *);
726 extern void free_threads_shm(void);
729 #define PTR_ALIGN(ptr, mask) \
730 (char *) (((uintptr_t) (ptr) + (mask)) & ~(mask))
734 * Reset stats after ramp time completes
736 extern void reset_all_stats(struct thread_data *);
738 extern int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret,
739 enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify,
740 struct timespec *comp_time);
743 * Latency target helpers
745 extern void lat_target_check(struct thread_data *);
746 extern void lat_target_init(struct thread_data *);
747 extern void lat_target_reset(struct thread_data *);
750 * Iterates all threads/processes within all the defined jobs
752 #define for_each_td(td, i) \
753 for ((i) = 0, (td) = &segments[0].threads[0]; (i) < (int) thread_number; (i)++, (td) = tnumber_to_td((i)))
754 #define for_each_file(td, f, i) \
755 if ((td)->files_index) \
756 for ((i) = 0, (f) = (td)->files[0]; \
757 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
760 static inline bool fio_fill_issue_time(struct thread_data *td)
762 if (td->o.read_iolog_file ||
763 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
769 static inline bool option_check_rate(struct thread_data *td, enum fio_ddir ddir)
771 struct thread_options *o = &td->o;
774 * If some rate setting was given, we need to check it
776 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
777 o->rate_iops_min[ddir])
783 static inline bool should_check_rate(struct thread_data *td)
785 return (td->flags & TD_F_CHECK_RATE) != 0;
788 static inline unsigned long long td_max_bs(struct thread_data *td)
790 unsigned long long max_bs;
792 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
793 return max(td->o.max_bs[DDIR_TRIM], max_bs);
796 static inline unsigned long long td_min_bs(struct thread_data *td)
798 unsigned long long min_bs;
800 min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
801 return min(td->o.min_bs[DDIR_TRIM], min_bs);
804 static inline bool td_async_processing(struct thread_data *td)
806 return (td->flags & TD_F_NEED_LOCK) != 0;
809 static inline bool td_offload_overlap(struct thread_data *td)
811 return td->o.serialize_overlap && td->o.io_submit_mode == IO_MODE_OFFLOAD;
815 * We currently only need to do locking if we have verifier threads
816 * accessing our internal structures too
818 static inline void __td_io_u_lock(struct thread_data *td)
820 pthread_mutex_lock(&td->io_u_lock);
823 static inline void __td_io_u_unlock(struct thread_data *td)
825 pthread_mutex_unlock(&td->io_u_lock);
828 static inline void td_io_u_free_notify(struct thread_data *td)
830 if (td_async_processing(td))
831 pthread_cond_signal(&td->free_cond);
834 static inline void td_flags_clear(struct thread_data *td, unsigned int *flags,
837 if (!td_async_processing(td))
840 __sync_fetch_and_and(flags, ~value);
843 static inline void td_flags_set(struct thread_data *td, unsigned int *flags,
846 if (!td_async_processing(td))
849 __sync_fetch_and_or(flags, value);
852 extern const char *fio_get_arch_string(int);
853 extern const char *fio_get_os_string(int);
856 __FIO_OUTPUT_TERSE = 0,
857 __FIO_OUTPUT_JSON = 1,
858 __FIO_OUTPUT_NORMAL = 2,
859 __FIO_OUTPUT_JSON_PLUS = 3,
862 FIO_OUTPUT_TERSE = 1U << __FIO_OUTPUT_TERSE,
863 FIO_OUTPUT_JSON = 1U << __FIO_OUTPUT_JSON,
864 FIO_OUTPUT_NORMAL = 1U << __FIO_OUTPUT_NORMAL,
865 FIO_OUTPUT_JSON_PLUS = 1U << __FIO_OUTPUT_JSON_PLUS,
869 FIO_RAND_DIST_RANDOM = 0,
871 FIO_RAND_DIST_PARETO,
874 FIO_RAND_DIST_ZONED_ABS,
877 #define FIO_DEF_ZIPF 1.1
878 #define FIO_DEF_PARETO 0.2
881 FIO_RAND_GEN_TAUSWORTHE = 0,
883 FIO_RAND_GEN_TAUSWORTHE64,
891 extern void exec_trigger(const char *);
892 extern void check_trigger_file(void);
894 extern bool in_flight_overlap(struct io_u_queue *q, struct io_u *io_u);
895 extern pthread_mutex_t overlap_check;
897 static inline void *fio_memalign(size_t alignment, size_t size, bool shared)
899 return __fio_memalign(alignment, size, shared ? smalloc : malloc);
902 static inline void fio_memfree(void *ptr, size_t size, bool shared)
904 return __fio_memfree(ptr, size, shared ? sfree : free);