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"
44 #include "io_u_queue.h"
45 #include "workqueue.h"
46 #include "steadystate.h"
47 #include "lib/nowarn_snprintf.h"
49 #ifdef CONFIG_SOLARISAIO
50 #include <sys/asynch.h>
54 #include <linux/mempolicy.h>
58 * "local" is pseudo-policy
60 #define MPOL_LOCAL MPOL_MAX
70 * offset generator types
78 __TD_F_VER_BACKLOG = 0,
81 __TD_F_REFILL_BUFFERS,
82 __TD_F_SCRAMBLE_BUFFERS,
95 __TD_F_LAST, /* not a real bit, keep last */
99 TD_F_VER_BACKLOG = 1U << __TD_F_VER_BACKLOG,
100 TD_F_TRIM_BACKLOG = 1U << __TD_F_TRIM_BACKLOG,
101 TD_F_READ_IOLOG = 1U << __TD_F_READ_IOLOG,
102 TD_F_REFILL_BUFFERS = 1U << __TD_F_REFILL_BUFFERS,
103 TD_F_SCRAMBLE_BUFFERS = 1U << __TD_F_SCRAMBLE_BUFFERS,
104 TD_F_DO_VERIFY = 1U << __TD_F_DO_VERIFY,
105 TD_F_PROFILE_OPS = 1U << __TD_F_PROFILE_OPS,
106 TD_F_COMPRESS = 1U << __TD_F_COMPRESS,
107 TD_F_COMPRESS_LOG = 1U << __TD_F_COMPRESS_LOG,
108 TD_F_VSTATE_SAVED = 1U << __TD_F_VSTATE_SAVED,
109 TD_F_NEED_LOCK = 1U << __TD_F_NEED_LOCK,
110 TD_F_CHILD = 1U << __TD_F_CHILD,
111 TD_F_NO_PROGRESS = 1U << __TD_F_NO_PROGRESS,
112 TD_F_REGROW_LOGS = 1U << __TD_F_REGROW_LOGS,
113 TD_F_MMAP_KEEP = 1U << __TD_F_MMAP_KEEP,
114 TD_F_DIRS_CREATED = 1U << __TD_F_DIRS_CREATED,
115 TD_F_CHECK_RATE = 1U << __TD_F_CHECK_RATE,
126 FIO_RAND_FILE_SIZE_OFF,
129 FIO_RAND_SEQ_RAND_READ_OFF,
130 FIO_RAND_SEQ_RAND_WRITE_OFF,
131 FIO_RAND_SEQ_RAND_TRIM_OFF,
132 FIO_RAND_START_DELAY,
134 FIO_RAND_POISSON_OFF,
136 FIO_RAND_POISSON2_OFF,
137 FIO_RAND_POISSON3_OFF,
145 RATE_PROCESS_LINEAR = 0,
146 RATE_PROCESS_POISSON = 1,
157 * Per-thread/process specific data. Only used for the network client
160 void sk_out_assign(struct sk_out *);
161 void sk_out_drop(void);
163 struct zone_split_index {
165 uint8_t size_perc_prev;
171 * This describes a single thread/process executing a fio job.
174 struct flist_head opt_list;
176 struct thread_options o;
179 unsigned int thread_number;
180 unsigned int subjob_number;
181 unsigned int groupid;
182 struct thread_stat ts __attribute__ ((aligned(8)));
186 struct io_log *slat_log;
187 struct io_log *clat_log;
188 struct io_log *clat_hist_log;
189 struct io_log *lat_log;
190 struct io_log *bw_log;
191 struct io_log *iops_log;
193 struct workqueue log_compress_wq;
195 struct thread_data *parent;
197 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
198 struct timespec bw_sample_time;
200 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
201 struct timespec iops_sample_time;
203 volatile int update_rusage;
204 struct fio_sem *rusage_sem;
205 struct rusage ru_start;
206 struct rusage ru_end;
208 struct fio_file **files;
209 unsigned char *file_locks;
210 unsigned int files_size;
211 unsigned int files_index;
212 unsigned int nr_open_files;
213 unsigned int nr_done_files;
215 unsigned int next_file;
216 struct frand_state next_file_state;
219 struct zipf_state next_file_zipf;
220 struct gauss_state next_file_gauss;
233 size_t orig_buffer_size;
234 volatile int runstate;
235 volatile bool terminate;
237 enum fio_ddir last_ddir;
244 unsigned long rand_seeds[FIO_RAND_NR_OFFS];
246 struct frand_state bsrange_state[DDIR_RWDIR_CNT];
247 struct frand_state verify_state;
248 struct frand_state trim_state;
249 struct frand_state delay_state;
251 struct frand_state buf_state;
252 struct frand_state buf_state_prev;
253 struct frand_state dedupe_state;
254 struct frand_state zone_state;
256 struct zone_split_index **zone_state_index;
258 unsigned int verify_batch;
259 unsigned int trim_batch;
261 struct thread_io_list *vstate;
266 * IO engine hooks, contains everything needed to submit an io_u
267 * to any of the available IO engines.
269 struct ioengine_ops *io_ops;
273 * IO engine private data and dlhandle.
276 void *io_ops_dlhandle;
279 * Queue depth of io_u's that fio MIGHT do
281 unsigned int cur_depth;
284 * io_u's about to be committed
286 unsigned int io_u_queued;
289 * io_u's submitted but not completed yet
291 unsigned int io_u_in_flight;
294 * List of free and busy io_u's
296 struct io_u_ring io_u_requeues;
297 struct io_u_queue io_u_freelist;
298 struct io_u_queue io_u_all;
299 pthread_mutex_t io_u_lock;
300 pthread_cond_t free_cond;
303 * async verify offload
305 struct flist_head verify_list;
306 pthread_t *verify_threads;
307 unsigned int nr_verify_threads;
308 pthread_cond_t verify_cond;
309 int verify_thread_exit;
314 uint64_t rate_bps[DDIR_RWDIR_CNT];
315 uint64_t rate_next_io_time[DDIR_RWDIR_CNT];
316 unsigned long rate_bytes[DDIR_RWDIR_CNT];
317 unsigned long rate_blocks[DDIR_RWDIR_CNT];
318 unsigned long long rate_io_issue_bytes[DDIR_RWDIR_CNT];
319 struct timespec lastrate[DDIR_RWDIR_CNT];
320 int64_t last_usec[DDIR_RWDIR_CNT];
321 struct frand_state poisson_state[DDIR_RWDIR_CNT];
324 * Enforced rate submission/completion workqueue
326 struct workqueue io_wq;
328 uint64_t total_io_size;
329 uint64_t fill_device_size;
334 uint64_t io_issues[DDIR_RWDIR_CNT];
335 uint64_t io_issue_bytes[DDIR_RWDIR_CNT];
341 uint64_t io_blocks[DDIR_RWDIR_CNT];
342 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
343 uint64_t io_bytes[DDIR_RWDIR_CNT];
344 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
345 uint64_t io_skip_bytes;
348 uint64_t bytes_done[DDIR_RWDIR_CNT];
351 * State for random io, a bitmap of blocks done vs not done
353 struct frand_state random_state;
355 struct timespec start; /* start of this loop */
356 struct timespec epoch; /* time job was started */
357 unsigned long long unix_epoch; /* Time job was started, unix epoch based. */
358 struct timespec last_issue;
360 struct timespec ts_cache;
361 struct timespec terminate_time;
362 unsigned int ts_cache_nr;
363 unsigned int ts_cache_mask;
367 * Time since last latency_window was started
369 struct timespec latency_ts;
370 unsigned int latency_qd;
371 unsigned int latency_qd_high;
372 unsigned int latency_qd_low;
373 unsigned int latency_failed;
374 uint64_t latency_ios;
378 * read/write mixed workload state
380 struct frand_state rwmix_state;
381 unsigned long rwmix_issues;
382 enum fio_ddir rwmix_ddir;
383 unsigned int ddir_seq_nr;
386 * rand/seq mixed workload state
388 struct frand_state seq_rand_state[DDIR_RWDIR_CNT];
391 * IO history logs for verification. We use a tree for sorting,
392 * if we are overwriting. Otherwise just use a fifo.
394 struct rb_root io_hist_tree;
395 struct flist_head io_hist_list;
396 unsigned long io_hist_len;
401 struct flist_head io_log_list;
403 unsigned int io_log_current;
404 unsigned int io_log_checkmark;
405 unsigned int io_log_highmark;
406 struct timespec io_log_highmark_time;
409 * For tracking/handling discards
411 struct flist_head trim_list;
412 unsigned long trim_entries;
415 * for fileservice, how often to switch to a new file
417 unsigned int file_service_nr;
418 unsigned int file_service_left;
419 struct fio_file *file_service_file;
421 unsigned int sync_file_range_nr;
424 * For generating file sizes
426 struct frand_state file_size_state;
431 unsigned int total_err_count;
434 struct fio_flow *flow;
437 * Can be overloaded by profiles
439 struct prof_io_ops prof_io_ops;
444 struct steadystate_data ss;
446 char verror[FIO_VERROR_SIZE];
450 * for GPU memory management
456 CUdeviceptr dev_mem_ptr;
462 * when should interactive ETA output be generated
470 #define __td_verror(td, err, msg, func) \
472 unsigned int ____e = (err); \
475 (td)->error = ____e; \
476 if (!(td)->first_error) \
477 nowarn_snprintf(td->verror, sizeof(td->verror), \
478 "file:%s:%d, func=%s, error=%s", \
479 __FILE__, __LINE__, (func), (msg)); \
483 #define td_clear_error(td) do { \
486 (td)->parent->error = 0; \
489 #define td_verror(td, err, func) do { \
490 __td_verror((td), (err), strerror((err)), (func)); \
492 __td_verror((td)->parent, (err), strerror((err)), (func)); \
495 #define td_vmsg(td, err, msg, func) do { \
496 __td_verror((td), (err), (msg), (func)); \
498 __td_verror((td)->parent, (err), (msg), (func)); \
501 #define __fio_stringify_1(x) #x
502 #define __fio_stringify(x) __fio_stringify_1(x)
504 extern int exitall_on_terminate;
505 extern unsigned int thread_number;
506 extern unsigned int stat_number;
509 extern int output_format;
510 extern int append_terse_output;
511 extern int temp_stall_ts;
512 extern uintptr_t page_mask, page_size;
513 extern int read_only;
514 extern int eta_print;
515 extern int eta_new_line;
516 extern unsigned int eta_interval_msec;
517 extern unsigned long done_secs;
518 extern int fio_gtod_offload;
519 extern int fio_gtod_cpu;
520 extern enum fio_cs fio_clock_source;
521 extern int fio_clock_source_set;
522 extern int warnings_fatal;
523 extern int terse_version;
524 extern int is_backend;
525 extern int nr_clients;
526 extern int log_syslog;
527 extern int status_interval;
528 extern const char fio_version_string[];
529 extern char *trigger_file;
530 extern char *trigger_cmd;
531 extern char *trigger_remote_cmd;
532 extern long long trigger_timeout;
533 extern char *aux_path;
535 extern struct thread_data *threads;
537 extern bool eta_time_within_slack(unsigned int time);
539 static inline void fio_ro_check(const struct thread_data *td, struct io_u *io_u)
541 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)) &&
542 !(io_u->ddir == DDIR_TRIM && !td_trim(td)));
545 #define REAL_MAX_JOBS 4096
547 static inline bool should_fsync(struct thread_data *td)
549 if (td->last_was_sync)
551 if (td_write(td) || td->o.override_sync)
558 * Init/option functions
560 extern int __must_check fio_init_options(void);
561 extern int __must_check parse_options(int, char **);
562 extern int parse_jobs_ini(char *, int, int, int);
563 extern int parse_cmd_line(int, char **, int);
564 extern int fio_backend(struct sk_out *);
565 extern void reset_fio_state(void);
566 extern void clear_io_state(struct thread_data *, int);
567 extern int fio_options_parse(struct thread_data *, char **, int);
568 extern void fio_keywords_init(void);
569 extern void fio_keywords_exit(void);
570 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
571 extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
572 extern void fio_fill_default_options(struct thread_data *);
573 extern int fio_show_option_help(const char *);
574 extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
575 extern void fio_options_dup_and_init(struct option *);
576 extern char *fio_option_dup_subs(const char *);
577 extern void fio_options_mem_dupe(struct thread_data *);
578 extern void td_fill_rand_seeds(struct thread_data *);
579 extern void td_fill_verify_state_seed(struct thread_data *);
580 extern void add_job_opts(const char **, int);
581 extern int ioengine_load(struct thread_data *);
582 extern bool parse_dryrun(void);
583 extern int fio_running_or_pending_io_threads(void);
584 extern int fio_set_fd_nonblocking(int, const char *);
585 extern void sig_show_status(int sig);
586 extern struct thread_data *get_global_options(void);
588 extern uintptr_t page_mask;
589 extern uintptr_t page_size;
590 extern int initialize_fio(char *envp[]);
591 extern void deinitialize_fio(void);
593 #define FIO_GETOPT_JOB 0x89000000
594 #define FIO_GETOPT_IOENGINE 0x98000000
595 #define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
600 extern void print_thread_status(void);
601 extern void print_status_init(int);
602 extern char *fio_uint_to_kmg(unsigned int val);
605 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
606 * will never back again. It may cycle between running/verififying/fsyncing.
607 * Once the thread reaches TD_EXITED, it is just waiting for the core to
627 #define TD_ENG_FLAG_SHIFT 17
628 #define TD_ENG_FLAG_MASK ((1U << 17) - 1)
630 static inline void td_set_ioengine_flags(struct thread_data *td)
632 td->flags = (~(TD_ENG_FLAG_MASK << TD_ENG_FLAG_SHIFT) & td->flags) |
633 (td->io_ops->flags << TD_ENG_FLAG_SHIFT);
636 static inline bool td_ioengine_flagged(struct thread_data *td,
637 enum fio_ioengine_flags flags)
639 return ((td->flags >> TD_ENG_FLAG_SHIFT) & flags) != 0;
642 extern void td_set_runstate(struct thread_data *, int);
643 extern int td_bump_runstate(struct thread_data *, int);
644 extern void td_restore_runstate(struct thread_data *, int);
645 extern const char *runstate_to_name(int runstate);
648 * Allow 60 seconds for a job to quit on its own, otherwise reap with
651 #define FIO_REAP_TIMEOUT 300
653 #define TERMINATE_ALL (-1U)
654 extern void fio_terminate_threads(unsigned int);
655 extern void fio_mark_td_terminate(struct thread_data *);
660 extern int __must_check fio_pin_memory(struct thread_data *);
661 extern void fio_unpin_memory(struct thread_data *);
662 extern int __must_check allocate_io_mem(struct thread_data *);
663 extern void free_io_mem(struct thread_data *);
664 extern void free_threads_shm(void);
667 #define PTR_ALIGN(ptr, mask) \
668 (char *) (((uintptr_t) (ptr) + (mask)) & ~(mask))
672 * Reset stats after ramp time completes
674 extern void reset_all_stats(struct thread_data *);
676 extern int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret,
677 enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify,
678 struct timespec *comp_time);
681 * Latency target helpers
683 extern void lat_target_check(struct thread_data *);
684 extern void lat_target_init(struct thread_data *);
685 extern void lat_target_reset(struct thread_data *);
688 * Iterates all threads/processes within all the defined jobs
690 #define for_each_td(td, i) \
691 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
692 #define for_each_file(td, f, i) \
693 if ((td)->files_index) \
694 for ((i) = 0, (f) = (td)->files[0]; \
695 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
698 #define fio_assert(td, cond) do { \
701 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
702 td_set_runstate((td), TD_EXITED); \
703 (td)->error = EFAULT; \
708 static inline bool fio_fill_issue_time(struct thread_data *td)
710 if (td->o.read_iolog_file ||
711 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
717 static inline bool option_check_rate(struct thread_data *td, enum fio_ddir ddir)
719 struct thread_options *o = &td->o;
722 * If some rate setting was given, we need to check it
724 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
725 o->rate_iops_min[ddir])
731 static inline bool __should_check_rate(struct thread_data *td)
733 return (td->flags & TD_F_CHECK_RATE) != 0;
736 static inline bool should_check_rate(struct thread_data *td)
738 if (!__should_check_rate(td))
741 return ddir_rw_sum(td->bytes_done) != 0;
744 static inline unsigned long long td_max_bs(struct thread_data *td)
746 unsigned long long max_bs;
748 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
749 return max(td->o.max_bs[DDIR_TRIM], max_bs);
752 static inline unsigned long long td_min_bs(struct thread_data *td)
754 unsigned long long min_bs;
756 min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
757 return min(td->o.min_bs[DDIR_TRIM], min_bs);
760 static inline bool td_async_processing(struct thread_data *td)
762 return (td->flags & TD_F_NEED_LOCK) != 0;
766 * We currently only need to do locking if we have verifier threads
767 * accessing our internal structures too
769 static inline void td_io_u_lock(struct thread_data *td)
771 if (td_async_processing(td))
772 pthread_mutex_lock(&td->io_u_lock);
775 static inline void td_io_u_unlock(struct thread_data *td)
777 if (td_async_processing(td))
778 pthread_mutex_unlock(&td->io_u_lock);
781 static inline void td_io_u_free_notify(struct thread_data *td)
783 if (td_async_processing(td))
784 pthread_cond_signal(&td->free_cond);
787 static inline void td_flags_clear(struct thread_data *td, unsigned int *flags,
790 if (!td_async_processing(td))
793 __sync_fetch_and_and(flags, ~value);
796 static inline void td_flags_set(struct thread_data *td, unsigned int *flags,
799 if (!td_async_processing(td))
802 __sync_fetch_and_or(flags, value);
805 extern const char *fio_get_arch_string(int);
806 extern const char *fio_get_os_string(int);
809 __FIO_OUTPUT_TERSE = 0,
810 __FIO_OUTPUT_JSON = 1,
811 __FIO_OUTPUT_NORMAL = 2,
812 __FIO_OUTPUT_JSON_PLUS = 3,
815 FIO_OUTPUT_TERSE = 1U << __FIO_OUTPUT_TERSE,
816 FIO_OUTPUT_JSON = 1U << __FIO_OUTPUT_JSON,
817 FIO_OUTPUT_NORMAL = 1U << __FIO_OUTPUT_NORMAL,
818 FIO_OUTPUT_JSON_PLUS = 1U << __FIO_OUTPUT_JSON_PLUS,
822 FIO_RAND_DIST_RANDOM = 0,
824 FIO_RAND_DIST_PARETO,
827 FIO_RAND_DIST_ZONED_ABS,
830 #define FIO_DEF_ZIPF 1.1
831 #define FIO_DEF_PARETO 0.2
834 FIO_RAND_GEN_TAUSWORTHE = 0,
836 FIO_RAND_GEN_TAUSWORTHE64,
844 extern void exec_trigger(const char *);
845 extern void check_trigger_file(void);