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,
100 __TD_F_LAST, /* not a real bit, keep last */
104 TD_F_VER_BACKLOG = 1U << __TD_F_VER_BACKLOG,
105 TD_F_TRIM_BACKLOG = 1U << __TD_F_TRIM_BACKLOG,
106 TD_F_READ_IOLOG = 1U << __TD_F_READ_IOLOG,
107 TD_F_REFILL_BUFFERS = 1U << __TD_F_REFILL_BUFFERS,
108 TD_F_SCRAMBLE_BUFFERS = 1U << __TD_F_SCRAMBLE_BUFFERS,
109 TD_F_DO_VERIFY = 1U << __TD_F_DO_VERIFY,
110 TD_F_PROFILE_OPS = 1U << __TD_F_PROFILE_OPS,
111 TD_F_COMPRESS = 1U << __TD_F_COMPRESS,
112 TD_F_COMPRESS_LOG = 1U << __TD_F_COMPRESS_LOG,
113 TD_F_VSTATE_SAVED = 1U << __TD_F_VSTATE_SAVED,
114 TD_F_NEED_LOCK = 1U << __TD_F_NEED_LOCK,
115 TD_F_CHILD = 1U << __TD_F_CHILD,
116 TD_F_NO_PROGRESS = 1U << __TD_F_NO_PROGRESS,
117 TD_F_REGROW_LOGS = 1U << __TD_F_REGROW_LOGS,
118 TD_F_MMAP_KEEP = 1U << __TD_F_MMAP_KEEP,
119 TD_F_DIRS_CREATED = 1U << __TD_F_DIRS_CREATED,
120 TD_F_CHECK_RATE = 1U << __TD_F_CHECK_RATE,
131 FIO_RAND_FILE_SIZE_OFF,
134 FIO_RAND_SEQ_RAND_READ_OFF,
135 FIO_RAND_SEQ_RAND_WRITE_OFF,
136 FIO_RAND_SEQ_RAND_TRIM_OFF,
137 FIO_RAND_START_DELAY,
139 FIO_RAND_POISSON_OFF,
141 FIO_RAND_POISSON2_OFF,
142 FIO_RAND_POISSON3_OFF,
144 FIO_RAND_DEDUPE_WORKING_SET_IX,
152 RATE_PROCESS_LINEAR = 0,
153 RATE_PROCESS_POISSON = 1,
155 THINKTIME_BLOCKS_TYPE_COMPLETE = 0,
156 THINKTIME_BLOCKS_TYPE_ISSUE = 1,
167 * Per-thread/process specific data. Only used for the network client
170 void sk_out_assign(struct sk_out *);
171 void sk_out_drop(void);
173 struct zone_split_index {
175 uint8_t size_perc_prev;
181 * This describes a single thread/process executing a fio job.
184 struct flist_head opt_list;
186 struct thread_options o;
189 unsigned int thread_number;
190 unsigned int subjob_number;
191 unsigned int groupid;
192 struct thread_stat ts __attribute__ ((aligned(8)));
196 struct io_log *slat_log;
197 struct io_log *clat_log;
198 struct io_log *clat_hist_log;
199 struct io_log *lat_log;
200 struct io_log *bw_log;
201 struct io_log *iops_log;
203 struct workqueue log_compress_wq;
205 struct thread_data *parent;
207 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
208 struct timespec bw_sample_time;
210 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
211 struct timespec iops_sample_time;
213 volatile int update_rusage;
214 struct fio_sem *rusage_sem;
215 struct rusage ru_start;
216 struct rusage ru_end;
218 struct fio_file **files;
219 unsigned char *file_locks;
220 unsigned int files_size;
221 unsigned int files_index;
222 unsigned int nr_open_files;
223 unsigned int nr_done_files;
225 unsigned int next_file;
226 struct frand_state next_file_state;
229 struct zipf_state next_file_zipf;
230 struct gauss_state next_file_gauss;
237 double random_center;
244 size_t orig_buffer_size;
245 volatile int runstate;
246 volatile bool terminate;
248 enum fio_ddir last_ddir;
255 uint64_t rand_seeds[FIO_RAND_NR_OFFS];
257 struct frand_state bsrange_state[DDIR_RWDIR_CNT];
258 struct frand_state verify_state;
259 struct frand_state trim_state;
260 struct frand_state delay_state;
262 struct frand_state buf_state;
263 struct frand_state buf_state_prev;
264 struct frand_state buf_state_ret;
265 struct frand_state dedupe_state;
266 struct frand_state zone_state;
267 struct frand_state prio_state;
268 struct frand_state dedupe_working_set_index_state;
269 struct frand_state *dedupe_working_set_states;
271 unsigned long long num_unique_pages;
273 struct zone_split_index **zone_state_index;
274 unsigned int num_open_zones;
276 unsigned int verify_batch;
277 unsigned int trim_batch;
279 struct thread_io_list *vstate;
284 * Job default IO priority set with prioclass and prio options.
289 * IO engine hooks, contains everything needed to submit an io_u
290 * to any of the available IO engines.
292 struct ioengine_ops *io_ops;
296 * IO engine private data and dlhandle.
301 * Queue depth of io_u's that fio MIGHT do
303 unsigned int cur_depth;
306 * io_u's about to be committed
308 unsigned int io_u_queued;
311 * io_u's submitted but not completed yet
313 unsigned int io_u_in_flight;
316 * List of free and busy io_u's
318 struct io_u_ring io_u_requeues;
319 struct io_u_queue io_u_freelist;
320 struct io_u_queue io_u_all;
321 pthread_mutex_t io_u_lock;
322 pthread_cond_t free_cond;
325 * async verify offload
327 struct flist_head verify_list;
328 pthread_t *verify_threads;
329 unsigned int nr_verify_threads;
330 pthread_cond_t verify_cond;
331 int verify_thread_exit;
336 uint64_t rate_bps[DDIR_RWDIR_CNT];
337 uint64_t rate_next_io_time[DDIR_RWDIR_CNT];
338 unsigned long long rate_bytes[DDIR_RWDIR_CNT];
339 unsigned long rate_blocks[DDIR_RWDIR_CNT];
340 unsigned long long rate_io_issue_bytes[DDIR_RWDIR_CNT];
341 struct timespec lastrate[DDIR_RWDIR_CNT];
342 int64_t last_usec[DDIR_RWDIR_CNT];
343 struct frand_state poisson_state[DDIR_RWDIR_CNT];
346 * Enforced rate submission/completion workqueue
348 struct workqueue io_wq;
350 uint64_t total_io_size;
351 uint64_t fill_device_size;
356 uint64_t io_issues[DDIR_RWDIR_CNT];
357 uint64_t io_issue_bytes[DDIR_RWDIR_CNT];
363 uint64_t io_blocks[DDIR_RWDIR_CNT];
364 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
365 uint64_t io_bytes[DDIR_RWDIR_CNT];
366 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
367 uint64_t io_skip_bytes;
370 uint64_t bytes_done[DDIR_RWDIR_CNT];
372 uint64_t *thinktime_blocks_counter;
373 struct timespec last_thinktime;
374 uint64_t last_thinktime_blocks;
377 * State for random io, a bitmap of blocks done vs not done
379 struct frand_state random_state;
381 struct timespec start; /* start of this loop */
382 struct timespec epoch; /* time job was started */
383 unsigned long long unix_epoch; /* Time job was started, unix epoch based. */
384 struct timespec last_issue;
386 struct timespec ts_cache;
387 struct timespec terminate_time;
388 unsigned int ts_cache_nr;
389 unsigned int ts_cache_mask;
393 * Time since last latency_window was started
395 struct timespec latency_ts;
396 unsigned int latency_qd;
397 unsigned int latency_qd_high;
398 unsigned int latency_qd_low;
399 unsigned int latency_failed;
400 unsigned int latency_stable_count;
401 uint64_t latency_ios;
405 * read/write mixed workload state
407 struct frand_state rwmix_state;
408 unsigned long rwmix_issues;
409 enum fio_ddir rwmix_ddir;
410 unsigned int ddir_seq_nr;
413 * rand/seq mixed workload state
415 struct frand_state seq_rand_state[DDIR_RWDIR_CNT];
418 * IO history logs for verification. We use a tree for sorting,
419 * if we are overwriting. Otherwise just use a fifo.
421 struct rb_root io_hist_tree;
422 struct flist_head io_hist_list;
423 unsigned long io_hist_len;
428 struct flist_head io_log_list;
430 unsigned int io_log_blktrace;
431 unsigned int io_log_current;
432 unsigned int io_log_checkmark;
433 unsigned int io_log_highmark;
434 struct timespec io_log_highmark_time;
437 * For tracking/handling discards
439 struct flist_head trim_list;
440 unsigned long trim_entries;
443 * for fileservice, how often to switch to a new file
445 unsigned int file_service_nr;
446 unsigned int file_service_left;
447 struct fio_file *file_service_file;
449 unsigned int sync_file_range_nr;
452 * For generating file sizes
454 struct frand_state file_size_state;
459 unsigned int total_err_count;
462 struct fio_flow *flow;
463 unsigned long long flow_counter;
466 * Can be overloaded by profiles
468 struct prof_io_ops prof_io_ops;
473 struct steadystate_data ss;
475 char verror[FIO_VERROR_SIZE];
479 * for GPU memory management
485 CUdeviceptr dev_mem_ptr;
490 struct thread_segment {
491 struct thread_data *threads;
497 * when should interactive ETA output be generated
505 #define __td_verror(td, err, msg, func) \
507 unsigned int ____e = (err); \
510 (td)->error = ____e; \
511 if (!(td)->first_error) \
512 nowarn_snprintf(td->verror, sizeof(td->verror), \
513 "file:%s:%d, func=%s, error=%s", \
514 __FILE__, __LINE__, (func), (msg)); \
518 #define td_clear_error(td) do { \
521 (td)->parent->error = 0; \
524 #define td_verror(td, err, func) do { \
525 __td_verror((td), (err), strerror((err)), (func)); \
527 __td_verror((td)->parent, (err), strerror((err)), (func)); \
530 #define td_vmsg(td, err, msg, func) do { \
531 __td_verror((td), (err), (msg), (func)); \
533 __td_verror((td)->parent, (err), (msg), (func)); \
536 #define __fio_stringify_1(x) #x
537 #define __fio_stringify(x) __fio_stringify_1(x)
539 #define REAL_MAX_JOBS 4096
540 #define JOBS_PER_SEG 8
541 #define REAL_MAX_SEG (REAL_MAX_JOBS / JOBS_PER_SEG)
543 extern bool exitall_on_terminate;
544 extern unsigned int thread_number;
545 extern unsigned int stat_number;
546 extern unsigned int nr_segments;
547 extern unsigned int cur_segment;
549 extern int output_format;
550 extern int append_terse_output;
551 extern int temp_stall_ts;
552 extern uintptr_t page_mask, page_size;
553 extern bool read_only;
554 extern int eta_print;
555 extern int eta_new_line;
556 extern unsigned int eta_interval_msec;
557 extern unsigned long done_secs;
558 extern int fio_gtod_offload;
559 extern int fio_gtod_cpu;
560 extern enum fio_cs fio_clock_source;
561 extern int fio_clock_source_set;
562 extern int warnings_fatal;
563 extern int terse_version;
564 extern bool is_backend;
565 extern bool is_local_backend;
566 extern int nr_clients;
567 extern bool log_syslog;
568 extern int status_interval;
569 extern const char fio_version_string[];
570 extern char *trigger_file;
571 extern char *trigger_cmd;
572 extern char *trigger_remote_cmd;
573 extern long long trigger_timeout;
574 extern char *aux_path;
576 extern struct thread_segment segments[REAL_MAX_SEG];
578 static inline struct thread_data *tnumber_to_td(unsigned int tnumber)
580 struct thread_segment *seg;
582 seg = &segments[tnumber / JOBS_PER_SEG];
583 return &seg->threads[tnumber & (JOBS_PER_SEG - 1)];
586 static inline bool is_running_backend(void)
588 return is_backend || is_local_backend;
591 extern bool eta_time_within_slack(unsigned int time);
593 static inline void fio_ro_check(const struct thread_data *td, struct io_u *io_u)
595 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)) &&
596 !(io_u->ddir == DDIR_TRIM && !td_trim(td)));
599 static inline bool should_fsync(struct thread_data *td)
601 if (td->last_was_sync)
603 if (td_write(td) || td->o.override_sync)
610 * Init/option functions
612 extern int __must_check fio_init_options(void);
613 extern int __must_check parse_options(int, char **);
614 extern int parse_jobs_ini(char *, int, int, int);
615 extern int parse_cmd_line(int, char **, int);
616 extern int fio_backend(struct sk_out *);
617 extern void reset_fio_state(void);
618 extern void clear_io_state(struct thread_data *, int);
619 extern int fio_options_parse(struct thread_data *, char **, int);
620 extern void fio_keywords_init(void);
621 extern void fio_keywords_exit(void);
622 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
623 extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
624 extern void fio_fill_default_options(struct thread_data *);
625 extern int fio_show_option_help(const char *);
626 extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
627 extern void fio_options_dup_and_init(struct option *);
628 extern char *fio_option_dup_subs(const char *);
629 extern void fio_options_mem_dupe(struct thread_data *);
630 extern void td_fill_rand_seeds(struct thread_data *);
631 extern void td_fill_verify_state_seed(struct thread_data *);
632 extern void add_job_opts(const char **, int);
633 extern int ioengine_load(struct thread_data *);
634 extern bool parse_dryrun(void);
635 extern int fio_running_or_pending_io_threads(void);
636 extern int fio_set_fd_nonblocking(int, const char *);
637 extern void sig_show_status(int sig);
638 extern struct thread_data *get_global_options(void);
640 extern uintptr_t page_mask;
641 extern uintptr_t page_size;
642 extern int initialize_fio(char *envp[]);
643 extern void deinitialize_fio(void);
645 #define FIO_GETOPT_JOB 0x89000000
646 #define FIO_GETOPT_IOENGINE 0x98000000
647 #define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
652 extern void print_thread_status(void);
653 extern void print_status_init(int);
654 extern char *fio_uint_to_kmg(unsigned int val);
657 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
658 * will never back again. It may cycle between running/verififying/fsyncing.
659 * Once the thread reaches TD_EXITED, it is just waiting for the core to
679 #define TD_ENG_FLAG_SHIFT 17
680 #define TD_ENG_FLAG_MASK ((1U << 17) - 1)
682 static inline void td_set_ioengine_flags(struct thread_data *td)
684 td->flags = (~(TD_ENG_FLAG_MASK << TD_ENG_FLAG_SHIFT) & td->flags) |
685 (td->io_ops->flags << TD_ENG_FLAG_SHIFT);
688 static inline bool td_ioengine_flagged(struct thread_data *td,
689 enum fio_ioengine_flags flags)
691 return ((td->flags >> TD_ENG_FLAG_SHIFT) & flags) != 0;
694 extern void td_set_runstate(struct thread_data *, int);
695 extern int td_bump_runstate(struct thread_data *, int);
696 extern void td_restore_runstate(struct thread_data *, int);
697 extern const char *runstate_to_name(int runstate);
700 * Allow 60 seconds for a job to quit on its own, otherwise reap with
703 #define FIO_REAP_TIMEOUT 300
708 TERMINATE_STONEWALL = 2,
712 extern void fio_terminate_threads(unsigned int, unsigned int);
713 extern void fio_mark_td_terminate(struct thread_data *);
718 extern int __must_check fio_pin_memory(struct thread_data *);
719 extern void fio_unpin_memory(struct thread_data *);
720 extern int __must_check allocate_io_mem(struct thread_data *);
721 extern void free_io_mem(struct thread_data *);
722 extern void free_threads_shm(void);
725 #define PTR_ALIGN(ptr, mask) \
726 (char *) (((uintptr_t) (ptr) + (mask)) & ~(mask))
730 * Reset stats after ramp time completes
732 extern void reset_all_stats(struct thread_data *);
734 extern int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret,
735 enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify,
736 struct timespec *comp_time);
739 * Latency target helpers
741 extern void lat_target_check(struct thread_data *);
742 extern void lat_target_init(struct thread_data *);
743 extern void lat_target_reset(struct thread_data *);
746 * Iterates all threads/processes within all the defined jobs
748 #define for_each_td(td, i) \
749 for ((i) = 0, (td) = &segments[0].threads[0]; (i) < (int) thread_number; (i)++, (td) = tnumber_to_td((i)))
750 #define for_each_file(td, f, i) \
751 if ((td)->files_index) \
752 for ((i) = 0, (f) = (td)->files[0]; \
753 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
756 static inline bool fio_fill_issue_time(struct thread_data *td)
758 if (td->o.read_iolog_file ||
759 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
765 static inline bool option_check_rate(struct thread_data *td, enum fio_ddir ddir)
767 struct thread_options *o = &td->o;
770 * If some rate setting was given, we need to check it
772 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
773 o->rate_iops_min[ddir])
779 static inline bool should_check_rate(struct thread_data *td)
781 return (td->flags & TD_F_CHECK_RATE) != 0;
784 static inline unsigned long long td_max_bs(struct thread_data *td)
786 unsigned long long max_bs;
788 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
789 return max(td->o.max_bs[DDIR_TRIM], max_bs);
792 static inline unsigned long long td_min_bs(struct thread_data *td)
794 unsigned long long min_bs;
796 min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
797 return min(td->o.min_bs[DDIR_TRIM], min_bs);
800 static inline bool td_async_processing(struct thread_data *td)
802 return (td->flags & TD_F_NEED_LOCK) != 0;
805 static inline bool td_offload_overlap(struct thread_data *td)
807 return td->o.serialize_overlap && td->o.io_submit_mode == IO_MODE_OFFLOAD;
811 * We currently only need to do locking if we have verifier threads
812 * accessing our internal structures too
814 static inline void __td_io_u_lock(struct thread_data *td)
816 pthread_mutex_lock(&td->io_u_lock);
819 static inline void __td_io_u_unlock(struct thread_data *td)
821 pthread_mutex_unlock(&td->io_u_lock);
824 static inline void td_io_u_free_notify(struct thread_data *td)
826 if (td_async_processing(td))
827 pthread_cond_signal(&td->free_cond);
830 static inline void td_flags_clear(struct thread_data *td, unsigned int *flags,
833 if (!td_async_processing(td))
836 __sync_fetch_and_and(flags, ~value);
839 static inline void td_flags_set(struct thread_data *td, unsigned int *flags,
842 if (!td_async_processing(td))
845 __sync_fetch_and_or(flags, value);
848 extern const char *fio_get_arch_string(int);
849 extern const char *fio_get_os_string(int);
852 __FIO_OUTPUT_TERSE = 0,
853 __FIO_OUTPUT_JSON = 1,
854 __FIO_OUTPUT_NORMAL = 2,
855 __FIO_OUTPUT_JSON_PLUS = 3,
858 FIO_OUTPUT_TERSE = 1U << __FIO_OUTPUT_TERSE,
859 FIO_OUTPUT_JSON = 1U << __FIO_OUTPUT_JSON,
860 FIO_OUTPUT_NORMAL = 1U << __FIO_OUTPUT_NORMAL,
861 FIO_OUTPUT_JSON_PLUS = 1U << __FIO_OUTPUT_JSON_PLUS,
865 FIO_RAND_DIST_RANDOM = 0,
867 FIO_RAND_DIST_PARETO,
870 FIO_RAND_DIST_ZONED_ABS,
873 #define FIO_DEF_ZIPF 1.1
874 #define FIO_DEF_PARETO 0.2
877 FIO_RAND_GEN_TAUSWORTHE = 0,
879 FIO_RAND_GEN_TAUSWORTHE64,
887 extern void exec_trigger(const char *);
888 extern void check_trigger_file(void);
890 extern bool in_flight_overlap(struct io_u_queue *q, struct io_u *io_u);
891 extern pthread_mutex_t overlap_check;
893 static inline void *fio_memalign(size_t alignment, size_t size, bool shared)
895 return __fio_memalign(alignment, size, shared ? smalloc : malloc);
898 static inline void fio_memfree(void *ptr, size_t size, bool shared)
900 return __fio_memfree(ptr, size, shared ? sfree : free);