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;
238 size_t orig_buffer_size;
239 volatile int runstate;
240 volatile bool terminate;
242 enum fio_ddir last_ddir;
249 uint64_t rand_seeds[FIO_RAND_NR_OFFS];
251 struct frand_state bsrange_state[DDIR_RWDIR_CNT];
252 struct frand_state verify_state;
253 struct frand_state trim_state;
254 struct frand_state delay_state;
256 struct frand_state buf_state;
257 struct frand_state buf_state_prev;
258 struct frand_state dedupe_state;
259 struct frand_state zone_state;
260 struct frand_state prio_state;
262 struct zone_split_index **zone_state_index;
264 unsigned int verify_batch;
265 unsigned int trim_batch;
267 struct thread_io_list *vstate;
272 * IO engine hooks, contains everything needed to submit an io_u
273 * to any of the available IO engines.
275 struct ioengine_ops *io_ops;
279 * IO engine private data and dlhandle.
282 void *io_ops_dlhandle;
285 * Queue depth of io_u's that fio MIGHT do
287 unsigned int cur_depth;
290 * io_u's about to be committed
292 unsigned int io_u_queued;
295 * io_u's submitted but not completed yet
297 unsigned int io_u_in_flight;
300 * List of free and busy io_u's
302 struct io_u_ring io_u_requeues;
303 struct io_u_queue io_u_freelist;
304 struct io_u_queue io_u_all;
305 pthread_mutex_t io_u_lock;
306 pthread_cond_t free_cond;
309 * async verify offload
311 struct flist_head verify_list;
312 pthread_t *verify_threads;
313 unsigned int nr_verify_threads;
314 pthread_cond_t verify_cond;
315 int verify_thread_exit;
320 uint64_t rate_bps[DDIR_RWDIR_CNT];
321 uint64_t rate_next_io_time[DDIR_RWDIR_CNT];
322 unsigned long rate_bytes[DDIR_RWDIR_CNT];
323 unsigned long rate_blocks[DDIR_RWDIR_CNT];
324 unsigned long long rate_io_issue_bytes[DDIR_RWDIR_CNT];
325 struct timespec lastrate[DDIR_RWDIR_CNT];
326 int64_t last_usec[DDIR_RWDIR_CNT];
327 struct frand_state poisson_state[DDIR_RWDIR_CNT];
330 * Enforced rate submission/completion workqueue
332 struct workqueue io_wq;
334 uint64_t total_io_size;
335 uint64_t fill_device_size;
340 uint64_t io_issues[DDIR_RWDIR_CNT];
341 uint64_t io_issue_bytes[DDIR_RWDIR_CNT];
347 uint64_t io_blocks[DDIR_RWDIR_CNT];
348 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
349 uint64_t io_bytes[DDIR_RWDIR_CNT];
350 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
351 uint64_t io_skip_bytes;
354 uint64_t bytes_done[DDIR_RWDIR_CNT];
357 * State for random io, a bitmap of blocks done vs not done
359 struct frand_state random_state;
361 struct timespec start; /* start of this loop */
362 struct timespec epoch; /* time job was started */
363 unsigned long long unix_epoch; /* Time job was started, unix epoch based. */
364 struct timespec last_issue;
366 struct timespec ts_cache;
367 struct timespec terminate_time;
368 unsigned int ts_cache_nr;
369 unsigned int ts_cache_mask;
373 * Time since last latency_window was started
375 struct timespec latency_ts;
376 unsigned int latency_qd;
377 unsigned int latency_qd_high;
378 unsigned int latency_qd_low;
379 unsigned int latency_failed;
380 uint64_t latency_ios;
384 * read/write mixed workload state
386 struct frand_state rwmix_state;
387 unsigned long rwmix_issues;
388 enum fio_ddir rwmix_ddir;
389 unsigned int ddir_seq_nr;
392 * rand/seq mixed workload state
394 struct frand_state seq_rand_state[DDIR_RWDIR_CNT];
397 * IO history logs for verification. We use a tree for sorting,
398 * if we are overwriting. Otherwise just use a fifo.
400 struct rb_root io_hist_tree;
401 struct flist_head io_hist_list;
402 unsigned long io_hist_len;
407 struct flist_head io_log_list;
409 unsigned int io_log_current;
410 unsigned int io_log_checkmark;
411 unsigned int io_log_highmark;
412 struct timespec io_log_highmark_time;
415 * For tracking/handling discards
417 struct flist_head trim_list;
418 unsigned long trim_entries;
421 * for fileservice, how often to switch to a new file
423 unsigned int file_service_nr;
424 unsigned int file_service_left;
425 struct fio_file *file_service_file;
427 unsigned int sync_file_range_nr;
430 * For generating file sizes
432 struct frand_state file_size_state;
437 unsigned int total_err_count;
440 struct fio_flow *flow;
443 * Can be overloaded by profiles
445 struct prof_io_ops prof_io_ops;
450 struct steadystate_data ss;
452 char verror[FIO_VERROR_SIZE];
456 * for GPU memory management
462 CUdeviceptr dev_mem_ptr;
468 * when should interactive ETA output be generated
476 #define __td_verror(td, err, msg, func) \
478 unsigned int ____e = (err); \
481 (td)->error = ____e; \
482 if (!(td)->first_error) \
483 nowarn_snprintf(td->verror, sizeof(td->verror), \
484 "file:%s:%d, func=%s, error=%s", \
485 __FILE__, __LINE__, (func), (msg)); \
489 #define td_clear_error(td) do { \
492 (td)->parent->error = 0; \
495 #define td_verror(td, err, func) do { \
496 __td_verror((td), (err), strerror((err)), (func)); \
498 __td_verror((td)->parent, (err), strerror((err)), (func)); \
501 #define td_vmsg(td, err, msg, func) do { \
502 __td_verror((td), (err), (msg), (func)); \
504 __td_verror((td)->parent, (err), (msg), (func)); \
507 #define __fio_stringify_1(x) #x
508 #define __fio_stringify(x) __fio_stringify_1(x)
510 extern bool exitall_on_terminate;
511 extern unsigned int thread_number;
512 extern unsigned int stat_number;
515 extern int output_format;
516 extern int append_terse_output;
517 extern int temp_stall_ts;
518 extern uintptr_t page_mask, page_size;
519 extern bool read_only;
520 extern int eta_print;
521 extern int eta_new_line;
522 extern unsigned int eta_interval_msec;
523 extern unsigned long done_secs;
524 extern int fio_gtod_offload;
525 extern int fio_gtod_cpu;
526 extern enum fio_cs fio_clock_source;
527 extern int fio_clock_source_set;
528 extern int warnings_fatal;
529 extern int terse_version;
530 extern bool is_backend;
531 extern bool is_local_backend;
532 extern int nr_clients;
533 extern bool log_syslog;
534 extern int status_interval;
535 extern const char fio_version_string[];
536 extern char *trigger_file;
537 extern char *trigger_cmd;
538 extern char *trigger_remote_cmd;
539 extern long long trigger_timeout;
540 extern char *aux_path;
542 extern struct thread_data *threads;
544 static inline bool is_running_backend(void)
546 return is_backend || is_local_backend;
549 extern bool eta_time_within_slack(unsigned int time);
551 static inline void fio_ro_check(const struct thread_data *td, struct io_u *io_u)
553 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)) &&
554 !(io_u->ddir == DDIR_TRIM && !td_trim(td)));
557 #define REAL_MAX_JOBS 4096
559 static inline bool should_fsync(struct thread_data *td)
561 if (td->last_was_sync)
563 if (td_write(td) || td->o.override_sync)
570 * Init/option functions
572 extern int __must_check fio_init_options(void);
573 extern int __must_check parse_options(int, char **);
574 extern int parse_jobs_ini(char *, int, int, int);
575 extern int parse_cmd_line(int, char **, int);
576 extern int fio_backend(struct sk_out *);
577 extern void reset_fio_state(void);
578 extern void clear_io_state(struct thread_data *, int);
579 extern int fio_options_parse(struct thread_data *, char **, int);
580 extern void fio_keywords_init(void);
581 extern void fio_keywords_exit(void);
582 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
583 extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
584 extern void fio_fill_default_options(struct thread_data *);
585 extern int fio_show_option_help(const char *);
586 extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
587 extern void fio_options_dup_and_init(struct option *);
588 extern char *fio_option_dup_subs(const char *);
589 extern void fio_options_mem_dupe(struct thread_data *);
590 extern void td_fill_rand_seeds(struct thread_data *);
591 extern void td_fill_verify_state_seed(struct thread_data *);
592 extern void add_job_opts(const char **, int);
593 extern int ioengine_load(struct thread_data *);
594 extern bool parse_dryrun(void);
595 extern int fio_running_or_pending_io_threads(void);
596 extern int fio_set_fd_nonblocking(int, const char *);
597 extern void sig_show_status(int sig);
598 extern struct thread_data *get_global_options(void);
600 extern uintptr_t page_mask;
601 extern uintptr_t page_size;
602 extern int initialize_fio(char *envp[]);
603 extern void deinitialize_fio(void);
605 #define FIO_GETOPT_JOB 0x89000000
606 #define FIO_GETOPT_IOENGINE 0x98000000
607 #define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
612 extern void print_thread_status(void);
613 extern void print_status_init(int);
614 extern char *fio_uint_to_kmg(unsigned int val);
617 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
618 * will never back again. It may cycle between running/verififying/fsyncing.
619 * Once the thread reaches TD_EXITED, it is just waiting for the core to
639 #define TD_ENG_FLAG_SHIFT 17
640 #define TD_ENG_FLAG_MASK ((1U << 17) - 1)
642 static inline void td_set_ioengine_flags(struct thread_data *td)
644 td->flags = (~(TD_ENG_FLAG_MASK << TD_ENG_FLAG_SHIFT) & td->flags) |
645 (td->io_ops->flags << TD_ENG_FLAG_SHIFT);
648 static inline bool td_ioengine_flagged(struct thread_data *td,
649 enum fio_ioengine_flags flags)
651 return ((td->flags >> TD_ENG_FLAG_SHIFT) & flags) != 0;
654 extern void td_set_runstate(struct thread_data *, int);
655 extern int td_bump_runstate(struct thread_data *, int);
656 extern void td_restore_runstate(struct thread_data *, int);
657 extern const char *runstate_to_name(int runstate);
660 * Allow 60 seconds for a job to quit on its own, otherwise reap with
663 #define FIO_REAP_TIMEOUT 300
668 TERMINATE_STONEWALL = 2,
672 extern void fio_terminate_threads(unsigned int, unsigned int);
673 extern void fio_mark_td_terminate(struct thread_data *);
678 extern int __must_check fio_pin_memory(struct thread_data *);
679 extern void fio_unpin_memory(struct thread_data *);
680 extern int __must_check allocate_io_mem(struct thread_data *);
681 extern void free_io_mem(struct thread_data *);
682 extern void free_threads_shm(void);
685 #define PTR_ALIGN(ptr, mask) \
686 (char *) (((uintptr_t) (ptr) + (mask)) & ~(mask))
690 * Reset stats after ramp time completes
692 extern void reset_all_stats(struct thread_data *);
694 extern int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret,
695 enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify,
696 struct timespec *comp_time);
699 * Latency target helpers
701 extern void lat_target_check(struct thread_data *);
702 extern void lat_target_init(struct thread_data *);
703 extern void lat_target_reset(struct thread_data *);
706 * Iterates all threads/processes within all the defined jobs
708 #define for_each_td(td, i) \
709 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
710 #define for_each_file(td, f, i) \
711 if ((td)->files_index) \
712 for ((i) = 0, (f) = (td)->files[0]; \
713 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
716 static inline bool fio_fill_issue_time(struct thread_data *td)
718 if (td->o.read_iolog_file ||
719 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
725 static inline bool option_check_rate(struct thread_data *td, enum fio_ddir ddir)
727 struct thread_options *o = &td->o;
730 * If some rate setting was given, we need to check it
732 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
733 o->rate_iops_min[ddir])
739 static inline bool __should_check_rate(struct thread_data *td)
741 return (td->flags & TD_F_CHECK_RATE) != 0;
744 static inline bool should_check_rate(struct thread_data *td)
746 if (!__should_check_rate(td))
749 return ddir_rw_sum(td->bytes_done) != 0;
752 static inline unsigned long long td_max_bs(struct thread_data *td)
754 unsigned long long max_bs;
756 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
757 return max(td->o.max_bs[DDIR_TRIM], max_bs);
760 static inline unsigned long long td_min_bs(struct thread_data *td)
762 unsigned long long min_bs;
764 min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
765 return min(td->o.min_bs[DDIR_TRIM], min_bs);
768 static inline bool td_async_processing(struct thread_data *td)
770 return (td->flags & TD_F_NEED_LOCK) != 0;
773 static inline bool td_offload_overlap(struct thread_data *td)
775 return td->o.serialize_overlap && td->o.io_submit_mode == IO_MODE_OFFLOAD;
779 * We currently only need to do locking if we have verifier threads
780 * accessing our internal structures too
782 static inline void __td_io_u_lock(struct thread_data *td)
784 pthread_mutex_lock(&td->io_u_lock);
787 static inline void __td_io_u_unlock(struct thread_data *td)
789 pthread_mutex_unlock(&td->io_u_lock);
792 static inline void td_io_u_free_notify(struct thread_data *td)
794 if (td_async_processing(td))
795 pthread_cond_signal(&td->free_cond);
798 static inline void td_flags_clear(struct thread_data *td, unsigned int *flags,
801 if (!td_async_processing(td))
804 __sync_fetch_and_and(flags, ~value);
807 static inline void td_flags_set(struct thread_data *td, unsigned int *flags,
810 if (!td_async_processing(td))
813 __sync_fetch_and_or(flags, value);
816 extern const char *fio_get_arch_string(int);
817 extern const char *fio_get_os_string(int);
820 __FIO_OUTPUT_TERSE = 0,
821 __FIO_OUTPUT_JSON = 1,
822 __FIO_OUTPUT_NORMAL = 2,
823 __FIO_OUTPUT_JSON_PLUS = 3,
826 FIO_OUTPUT_TERSE = 1U << __FIO_OUTPUT_TERSE,
827 FIO_OUTPUT_JSON = 1U << __FIO_OUTPUT_JSON,
828 FIO_OUTPUT_NORMAL = 1U << __FIO_OUTPUT_NORMAL,
829 FIO_OUTPUT_JSON_PLUS = 1U << __FIO_OUTPUT_JSON_PLUS,
833 FIO_RAND_DIST_RANDOM = 0,
835 FIO_RAND_DIST_PARETO,
838 FIO_RAND_DIST_ZONED_ABS,
841 #define FIO_DEF_ZIPF 1.1
842 #define FIO_DEF_PARETO 0.2
845 FIO_RAND_GEN_TAUSWORTHE = 0,
847 FIO_RAND_GEN_TAUSWORTHE64,
855 extern void exec_trigger(const char *);
856 extern void check_trigger_file(void);
858 extern bool in_flight_overlap(struct io_u_queue *q, struct io_u *io_u);
859 extern pthread_mutex_t overlap_check;
861 static inline void *fio_memalign(size_t alignment, size_t size, bool shared)
863 return __fio_memalign(alignment, size, shared ? smalloc : malloc);
866 static inline void fio_memfree(void *ptr, size_t size, bool shared)
868 return __fio_memfree(ptr, size, shared ? sfree : free);