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;
263 unsigned int num_open_zones;
265 unsigned int verify_batch;
266 unsigned int trim_batch;
268 struct thread_io_list *vstate;
273 * IO engine hooks, contains everything needed to submit an io_u
274 * to any of the available IO engines.
276 struct ioengine_ops *io_ops;
280 * IO engine private data and dlhandle.
283 void *io_ops_dlhandle;
286 * Queue depth of io_u's that fio MIGHT do
288 unsigned int cur_depth;
291 * io_u's about to be committed
293 unsigned int io_u_queued;
296 * io_u's submitted but not completed yet
298 unsigned int io_u_in_flight;
301 * List of free and busy io_u's
303 struct io_u_ring io_u_requeues;
304 struct io_u_queue io_u_freelist;
305 struct io_u_queue io_u_all;
306 pthread_mutex_t io_u_lock;
307 pthread_cond_t free_cond;
310 * async verify offload
312 struct flist_head verify_list;
313 pthread_t *verify_threads;
314 unsigned int nr_verify_threads;
315 pthread_cond_t verify_cond;
316 int verify_thread_exit;
321 uint64_t rate_bps[DDIR_RWDIR_CNT];
322 uint64_t rate_next_io_time[DDIR_RWDIR_CNT];
323 unsigned long long rate_bytes[DDIR_RWDIR_CNT];
324 unsigned long rate_blocks[DDIR_RWDIR_CNT];
325 unsigned long long rate_io_issue_bytes[DDIR_RWDIR_CNT];
326 struct timespec lastrate[DDIR_RWDIR_CNT];
327 int64_t last_usec[DDIR_RWDIR_CNT];
328 struct frand_state poisson_state[DDIR_RWDIR_CNT];
331 * Enforced rate submission/completion workqueue
333 struct workqueue io_wq;
335 uint64_t total_io_size;
336 uint64_t fill_device_size;
341 uint64_t io_issues[DDIR_RWDIR_CNT];
342 uint64_t io_issue_bytes[DDIR_RWDIR_CNT];
348 uint64_t io_blocks[DDIR_RWDIR_CNT];
349 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
350 uint64_t io_bytes[DDIR_RWDIR_CNT];
351 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
352 uint64_t io_skip_bytes;
355 uint64_t bytes_done[DDIR_RWDIR_CNT];
358 * State for random io, a bitmap of blocks done vs not done
360 struct frand_state random_state;
362 struct timespec start; /* start of this loop */
363 struct timespec epoch; /* time job was started */
364 unsigned long long unix_epoch; /* Time job was started, unix epoch based. */
365 struct timespec last_issue;
367 struct timespec ts_cache;
368 struct timespec terminate_time;
369 unsigned int ts_cache_nr;
370 unsigned int ts_cache_mask;
374 * Time since last latency_window was started
376 struct timespec latency_ts;
377 unsigned int latency_qd;
378 unsigned int latency_qd_high;
379 unsigned int latency_qd_low;
380 unsigned int latency_failed;
381 unsigned int latency_stable_count;
382 uint64_t latency_ios;
386 * read/write mixed workload state
388 struct frand_state rwmix_state;
389 unsigned long rwmix_issues;
390 enum fio_ddir rwmix_ddir;
391 unsigned int ddir_seq_nr;
394 * rand/seq mixed workload state
396 struct frand_state seq_rand_state[DDIR_RWDIR_CNT];
399 * IO history logs for verification. We use a tree for sorting,
400 * if we are overwriting. Otherwise just use a fifo.
402 struct rb_root io_hist_tree;
403 struct flist_head io_hist_list;
404 unsigned long io_hist_len;
409 struct flist_head io_log_list;
411 unsigned int io_log_current;
412 unsigned int io_log_checkmark;
413 unsigned int io_log_highmark;
414 struct timespec io_log_highmark_time;
417 * For tracking/handling discards
419 struct flist_head trim_list;
420 unsigned long trim_entries;
423 * for fileservice, how often to switch to a new file
425 unsigned int file_service_nr;
426 unsigned int file_service_left;
427 struct fio_file *file_service_file;
429 unsigned int sync_file_range_nr;
432 * For generating file sizes
434 struct frand_state file_size_state;
439 unsigned int total_err_count;
442 struct fio_flow *flow;
445 * Can be overloaded by profiles
447 struct prof_io_ops prof_io_ops;
452 struct steadystate_data ss;
454 char verror[FIO_VERROR_SIZE];
458 * for GPU memory management
464 CUdeviceptr dev_mem_ptr;
470 * when should interactive ETA output be generated
478 #define __td_verror(td, err, msg, func) \
480 unsigned int ____e = (err); \
483 (td)->error = ____e; \
484 if (!(td)->first_error) \
485 nowarn_snprintf(td->verror, sizeof(td->verror), \
486 "file:%s:%d, func=%s, error=%s", \
487 __FILE__, __LINE__, (func), (msg)); \
491 #define td_clear_error(td) do { \
494 (td)->parent->error = 0; \
497 #define td_verror(td, err, func) do { \
498 __td_verror((td), (err), strerror((err)), (func)); \
500 __td_verror((td)->parent, (err), strerror((err)), (func)); \
503 #define td_vmsg(td, err, msg, func) do { \
504 __td_verror((td), (err), (msg), (func)); \
506 __td_verror((td)->parent, (err), (msg), (func)); \
509 #define __fio_stringify_1(x) #x
510 #define __fio_stringify(x) __fio_stringify_1(x)
512 extern bool exitall_on_terminate;
513 extern unsigned int thread_number;
514 extern unsigned int stat_number;
517 extern int output_format;
518 extern int append_terse_output;
519 extern int temp_stall_ts;
520 extern uintptr_t page_mask, page_size;
521 extern bool read_only;
522 extern int eta_print;
523 extern int eta_new_line;
524 extern unsigned int eta_interval_msec;
525 extern unsigned long done_secs;
526 extern int fio_gtod_offload;
527 extern int fio_gtod_cpu;
528 extern enum fio_cs fio_clock_source;
529 extern int fio_clock_source_set;
530 extern int warnings_fatal;
531 extern int terse_version;
532 extern bool is_backend;
533 extern bool is_local_backend;
534 extern int nr_clients;
535 extern bool log_syslog;
536 extern int status_interval;
537 extern const char fio_version_string[];
538 extern char *trigger_file;
539 extern char *trigger_cmd;
540 extern char *trigger_remote_cmd;
541 extern long long trigger_timeout;
542 extern char *aux_path;
544 extern struct thread_data *threads;
546 static inline bool is_running_backend(void)
548 return is_backend || is_local_backend;
551 extern bool eta_time_within_slack(unsigned int time);
553 static inline void fio_ro_check(const struct thread_data *td, struct io_u *io_u)
555 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)) &&
556 !(io_u->ddir == DDIR_TRIM && !td_trim(td)));
559 #define REAL_MAX_JOBS 4096
561 static inline bool should_fsync(struct thread_data *td)
563 if (td->last_was_sync)
565 if (td_write(td) || td->o.override_sync)
572 * Init/option functions
574 extern int __must_check fio_init_options(void);
575 extern int __must_check parse_options(int, char **);
576 extern int parse_jobs_ini(char *, int, int, int);
577 extern int parse_cmd_line(int, char **, int);
578 extern int fio_backend(struct sk_out *);
579 extern void reset_fio_state(void);
580 extern void clear_io_state(struct thread_data *, int);
581 extern int fio_options_parse(struct thread_data *, char **, int);
582 extern void fio_keywords_init(void);
583 extern void fio_keywords_exit(void);
584 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
585 extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
586 extern void fio_fill_default_options(struct thread_data *);
587 extern int fio_show_option_help(const char *);
588 extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
589 extern void fio_options_dup_and_init(struct option *);
590 extern char *fio_option_dup_subs(const char *);
591 extern void fio_options_mem_dupe(struct thread_data *);
592 extern void td_fill_rand_seeds(struct thread_data *);
593 extern void td_fill_verify_state_seed(struct thread_data *);
594 extern void add_job_opts(const char **, int);
595 extern int ioengine_load(struct thread_data *);
596 extern bool parse_dryrun(void);
597 extern int fio_running_or_pending_io_threads(void);
598 extern int fio_set_fd_nonblocking(int, const char *);
599 extern void sig_show_status(int sig);
600 extern struct thread_data *get_global_options(void);
602 extern uintptr_t page_mask;
603 extern uintptr_t page_size;
604 extern int initialize_fio(char *envp[]);
605 extern void deinitialize_fio(void);
607 #define FIO_GETOPT_JOB 0x89000000
608 #define FIO_GETOPT_IOENGINE 0x98000000
609 #define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
614 extern void print_thread_status(void);
615 extern void print_status_init(int);
616 extern char *fio_uint_to_kmg(unsigned int val);
619 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
620 * will never back again. It may cycle between running/verififying/fsyncing.
621 * Once the thread reaches TD_EXITED, it is just waiting for the core to
641 #define TD_ENG_FLAG_SHIFT 17
642 #define TD_ENG_FLAG_MASK ((1U << 17) - 1)
644 static inline void td_set_ioengine_flags(struct thread_data *td)
646 td->flags = (~(TD_ENG_FLAG_MASK << TD_ENG_FLAG_SHIFT) & td->flags) |
647 (td->io_ops->flags << TD_ENG_FLAG_SHIFT);
650 static inline bool td_ioengine_flagged(struct thread_data *td,
651 enum fio_ioengine_flags flags)
653 return ((td->flags >> TD_ENG_FLAG_SHIFT) & flags) != 0;
656 extern void td_set_runstate(struct thread_data *, int);
657 extern int td_bump_runstate(struct thread_data *, int);
658 extern void td_restore_runstate(struct thread_data *, int);
659 extern const char *runstate_to_name(int runstate);
662 * Allow 60 seconds for a job to quit on its own, otherwise reap with
665 #define FIO_REAP_TIMEOUT 300
670 TERMINATE_STONEWALL = 2,
674 extern void fio_terminate_threads(unsigned int, unsigned int);
675 extern void fio_mark_td_terminate(struct thread_data *);
680 extern int __must_check fio_pin_memory(struct thread_data *);
681 extern void fio_unpin_memory(struct thread_data *);
682 extern int __must_check allocate_io_mem(struct thread_data *);
683 extern void free_io_mem(struct thread_data *);
684 extern void free_threads_shm(void);
687 #define PTR_ALIGN(ptr, mask) \
688 (char *) (((uintptr_t) (ptr) + (mask)) & ~(mask))
692 * Reset stats after ramp time completes
694 extern void reset_all_stats(struct thread_data *);
696 extern int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret,
697 enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify,
698 struct timespec *comp_time);
701 * Latency target helpers
703 extern void lat_target_check(struct thread_data *);
704 extern void lat_target_init(struct thread_data *);
705 extern void lat_target_reset(struct thread_data *);
708 * Iterates all threads/processes within all the defined jobs
710 #define for_each_td(td, i) \
711 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
712 #define for_each_file(td, f, i) \
713 if ((td)->files_index) \
714 for ((i) = 0, (f) = (td)->files[0]; \
715 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
718 static inline bool fio_fill_issue_time(struct thread_data *td)
720 if (td->o.read_iolog_file ||
721 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
727 static inline bool option_check_rate(struct thread_data *td, enum fio_ddir ddir)
729 struct thread_options *o = &td->o;
732 * If some rate setting was given, we need to check it
734 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
735 o->rate_iops_min[ddir])
741 static inline bool __should_check_rate(struct thread_data *td)
743 return (td->flags & TD_F_CHECK_RATE) != 0;
746 static inline bool should_check_rate(struct thread_data *td)
748 if (!__should_check_rate(td))
751 return ddir_rw_sum(td->bytes_done) != 0;
754 static inline unsigned long long td_max_bs(struct thread_data *td)
756 unsigned long long max_bs;
758 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
759 return max(td->o.max_bs[DDIR_TRIM], max_bs);
762 static inline unsigned long long td_min_bs(struct thread_data *td)
764 unsigned long long min_bs;
766 min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
767 return min(td->o.min_bs[DDIR_TRIM], min_bs);
770 static inline bool td_async_processing(struct thread_data *td)
772 return (td->flags & TD_F_NEED_LOCK) != 0;
775 static inline bool td_offload_overlap(struct thread_data *td)
777 return td->o.serialize_overlap && td->o.io_submit_mode == IO_MODE_OFFLOAD;
781 * We currently only need to do locking if we have verifier threads
782 * accessing our internal structures too
784 static inline void __td_io_u_lock(struct thread_data *td)
786 pthread_mutex_lock(&td->io_u_lock);
789 static inline void __td_io_u_unlock(struct thread_data *td)
791 pthread_mutex_unlock(&td->io_u_lock);
794 static inline void td_io_u_free_notify(struct thread_data *td)
796 if (td_async_processing(td))
797 pthread_cond_signal(&td->free_cond);
800 static inline void td_flags_clear(struct thread_data *td, unsigned int *flags,
803 if (!td_async_processing(td))
806 __sync_fetch_and_and(flags, ~value);
809 static inline void td_flags_set(struct thread_data *td, unsigned int *flags,
812 if (!td_async_processing(td))
815 __sync_fetch_and_or(flags, value);
818 extern const char *fio_get_arch_string(int);
819 extern const char *fio_get_os_string(int);
822 __FIO_OUTPUT_TERSE = 0,
823 __FIO_OUTPUT_JSON = 1,
824 __FIO_OUTPUT_NORMAL = 2,
825 __FIO_OUTPUT_JSON_PLUS = 3,
828 FIO_OUTPUT_TERSE = 1U << __FIO_OUTPUT_TERSE,
829 FIO_OUTPUT_JSON = 1U << __FIO_OUTPUT_JSON,
830 FIO_OUTPUT_NORMAL = 1U << __FIO_OUTPUT_NORMAL,
831 FIO_OUTPUT_JSON_PLUS = 1U << __FIO_OUTPUT_JSON_PLUS,
835 FIO_RAND_DIST_RANDOM = 0,
837 FIO_RAND_DIST_PARETO,
840 FIO_RAND_DIST_ZONED_ABS,
843 #define FIO_DEF_ZIPF 1.1
844 #define FIO_DEF_PARETO 0.2
847 FIO_RAND_GEN_TAUSWORTHE = 0,
849 FIO_RAND_GEN_TAUSWORTHE64,
857 extern void exec_trigger(const char *);
858 extern void check_trigger_file(void);
860 extern bool in_flight_overlap(struct io_u_queue *q, struct io_u *io_u);
861 extern pthread_mutex_t overlap_check;
863 static inline void *fio_memalign(size_t alignment, size_t size, bool shared)
865 return __fio_memalign(alignment, size, shared ? smalloc : malloc);
868 static inline void fio_memfree(void *ptr, size_t size, bool shared)
870 return __fio_memfree(ptr, size, shared ? sfree : free);