8 #include <sys/resource.h>
18 #include "compiler/compiler.h"
22 #include "arch/arch.h"
36 #ifdef FIO_HAVE_SOLARISAIO
37 #include <sys/asynch.h>
40 struct group_run_stats {
41 unsigned long long max_run[2], min_run[2];
42 unsigned long long max_bw[2], min_bw[2];
43 unsigned long long io_kb[2];
44 unsigned long long agg[2];
49 * What type of allocation to use for io buffers
52 MEM_MALLOC = 0, /* ordinary malloc */
53 MEM_SHM, /* use shared memory segments */
54 MEM_SHMHUGE, /* use shared memory segments with huge pages */
55 MEM_MMAP, /* use anonynomous mmap */
56 MEM_MMAPHUGE, /* memory mapped huge file */
60 * How many depth levels to log
62 #define FIO_IO_U_MAP_NR 8
63 #define FIO_IO_U_LAT_U_NR 10
64 #define FIO_IO_U_LAT_M_NR 12
75 struct io_log *slat_log;
76 struct io_log *clat_log;
77 struct io_log *bw_log;
80 * bandwidth and latency stats
82 struct io_stat clat_stat[2]; /* completion latency */
83 struct io_stat slat_stat[2]; /* submission latency */
84 struct io_stat bw_stat[2]; /* bandwidth stats */
86 unsigned long long stat_io_bytes[2];
87 struct timeval stat_sample_time[2];
90 * fio system usage accounting
92 struct rusage ru_start;
94 unsigned long usr_time;
95 unsigned long sys_time;
97 unsigned long minf, majf;
100 * IO depth and latency stats
102 unsigned int io_u_map[FIO_IO_U_MAP_NR];
103 unsigned int io_u_submit[FIO_IO_U_MAP_NR];
104 unsigned int io_u_complete[FIO_IO_U_MAP_NR];
105 unsigned int io_u_lat_u[FIO_IO_U_LAT_U_NR];
106 unsigned int io_u_lat_m[FIO_IO_U_LAT_M_NR];
107 unsigned long total_io_u[2];
108 unsigned long short_io_u[2];
109 unsigned long total_submit;
110 unsigned long total_complete;
112 unsigned long long io_bytes[2];
113 unsigned long runtime[2];
114 unsigned long total_run_time;
117 * IO Error related stats
119 unsigned continue_on_error;
120 unsigned long total_err_count;
123 unsigned int kb_base;
131 struct thread_options {
139 enum td_ddir td_ddir;
140 unsigned int kb_base;
141 unsigned int ddir_nr;
142 unsigned int iodepth;
143 unsigned int iodepth_low;
144 unsigned int iodepth_batch;
145 unsigned int iodepth_batch_complete;
147 unsigned long long size;
148 unsigned int fill_device;
149 unsigned long long file_size_low;
150 unsigned long long file_size_high;
151 unsigned long long start_offset;
155 unsigned int min_bs[2];
156 unsigned int max_bs[2];
157 struct bssplit *bssplit[2];
158 unsigned int bssplit_nr[2];
160 unsigned int nr_files;
161 unsigned int open_files;
162 enum file_lock_mode file_lock_mode;
163 unsigned int lockfile_batch;
165 unsigned int odirect;
166 unsigned int invalidate_cache;
167 unsigned int create_serialize;
168 unsigned int create_fsync;
169 unsigned int create_on_open;
170 unsigned int end_fsync;
171 unsigned int pre_read;
172 unsigned int sync_io;
174 unsigned int do_verify;
175 unsigned int verifysort;
176 unsigned int verify_interval;
177 unsigned int verify_offset;
178 unsigned int verify_pattern;
179 unsigned int verify_pattern_bytes;
180 unsigned int verify_fatal;
181 unsigned int verify_async;
182 unsigned int use_thread;
184 unsigned int do_disk_util;
185 unsigned int override_sync;
186 unsigned int rand_repeatable;
187 unsigned int write_lat_log;
188 unsigned int write_bw_log;
189 unsigned int norandommap;
190 unsigned int softrandommap;
191 unsigned int bs_unaligned;
192 unsigned int fsync_on_close;
194 unsigned int hugepage_size;
195 unsigned int rw_min_bs;
196 unsigned int thinktime;
197 unsigned int thinktime_spin;
198 unsigned int thinktime_blocks;
199 unsigned int fsync_blocks;
200 unsigned int fdatasync_blocks;
201 unsigned int start_delay;
202 unsigned long long timeout;
203 unsigned long long ramp_time;
204 unsigned int overwrite;
205 unsigned int bw_avg_time;
207 unsigned long long zone_size;
208 unsigned long long zone_skip;
209 enum fio_memtype mem_type;
210 unsigned int mem_align;
212 unsigned int stonewall;
213 unsigned int new_group;
214 unsigned int numjobs;
215 os_cpu_mask_t cpumask;
216 unsigned int cpumask_set;
217 os_cpu_mask_t verify_cpumask;
218 unsigned int verify_cpumask_set;
220 unsigned int rwmixcycle;
221 unsigned int rwmix[2];
223 unsigned int file_service_type;
224 unsigned int group_reporting;
225 unsigned int fadvise_hint;
226 unsigned int zero_buffers;
227 unsigned int refill_buffers;
228 unsigned int time_based;
229 unsigned int disable_clat;
230 unsigned int disable_slat;
231 unsigned int disable_bw;
232 unsigned int gtod_reduce;
233 unsigned int gtod_cpu;
234 unsigned int gtod_offload;
236 char *read_iolog_file;
237 char *write_iolog_file;
242 * Pre-run and post-run shell
247 unsigned int rate[2];
248 unsigned int ratemin[2];
249 unsigned int ratecycle;
250 unsigned int rate_iops[2];
251 unsigned int rate_iops_min[2];
256 * CPU "io" cycle burner
258 unsigned int cpuload;
259 unsigned int cpucycle;
264 unsigned int continue_on_error;
267 #define FIO_VERROR_SIZE 128
270 * This describes a single thread/process executing a fio job.
273 struct thread_options o;
274 char verror[FIO_VERROR_SIZE];
278 struct thread_stat ts;
279 struct fio_file **files;
280 unsigned int files_size;
281 unsigned int files_index;
282 unsigned int nr_open_files;
283 unsigned int nr_done_files;
284 unsigned int nr_normal_files;
286 unsigned int next_file;
287 os_random_state_t next_file_state;
293 size_t orig_buffer_size;
294 volatile int terminate;
295 volatile int runstate;
297 unsigned int ioprio_set;
298 unsigned int last_was_sync;
308 unsigned long rand_seeds[6];
310 os_random_state_t bsrange_state;
311 os_random_state_t verify_state;
316 * IO engine hooks, contains everything needed to submit an io_u
317 * to any of the available IO engines.
319 struct ioengine_ops *io_ops;
322 * Current IO depth and list of free and busy io_u's.
324 unsigned int cur_depth;
325 unsigned int io_u_queued;
326 struct flist_head io_u_freelist;
327 struct flist_head io_u_busylist;
328 struct flist_head io_u_requeues;
329 pthread_mutex_t io_u_lock;
330 pthread_cond_t free_cond;
333 * async verify offload
335 struct flist_head verify_list;
336 pthread_t *verify_threads;
337 unsigned int nr_verify_threads;
338 pthread_cond_t verify_cond;
339 int verify_thread_exit;
344 unsigned long rate_usec_cycle[2];
345 long rate_pending_usleep[2];
346 unsigned long rate_bytes[2];
347 unsigned long rate_blocks[2];
348 struct timeval lastrate[2];
350 unsigned long long total_io_size;
352 unsigned long io_issues[2];
353 unsigned long long io_blocks[2];
354 unsigned long long io_bytes[2];
355 unsigned long long io_skip_bytes;
356 unsigned long long this_io_bytes[2];
357 unsigned long long zone_bytes;
358 struct fio_mutex *mutex;
361 * State for random io, a bitmap of blocks done vs not done
363 os_random_state_t random_state;
365 struct timeval start; /* start of this loop */
366 struct timeval epoch; /* time job was started */
367 struct timeval last_issue;
368 struct timeval tv_cache;
369 unsigned int tv_cache_nr;
370 unsigned int tv_cache_mask;
371 unsigned int ramp_time_over;
374 * read/write mixed workload state
376 os_random_state_t rwmix_state;
377 unsigned long rwmix_issues;
378 enum fio_ddir rwmix_ddir;
379 unsigned int ddir_nr;
382 * IO history logs for verification. We use a tree for sorting,
383 * if we are overwriting. Otherwise just use a fifo.
385 struct rb_root io_hist_tree;
386 struct flist_head io_hist_list;
391 struct flist_head io_log_list;
394 * for fileservice, how often to switch to a new file
396 unsigned int file_service_nr;
397 unsigned int file_service_left;
398 struct fio_file *file_service_file;
401 * For generating file sizes
403 os_random_state_t file_size_state;
408 unsigned int total_err_count;
413 * when should interactive ETA output be generated
421 #define __td_verror(td, err, msg, func) \
427 if (!(td)->first_error) \
428 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
432 #define td_clear_error(td) \
434 #define td_verror(td, err, func) \
435 __td_verror((td), (err), strerror((err)), (func))
436 #define td_vmsg(td, err, msg, func) \
437 __td_verror((td), (err), (msg), (func))
439 extern int exitall_on_terminate;
440 extern int thread_number;
441 extern int nr_process, nr_thread;
444 extern int terse_output;
445 extern int temp_stall_ts;
446 extern unsigned long long mlock_size;
447 extern unsigned long page_mask, page_size;
448 extern int read_only;
449 extern int eta_print;
450 extern unsigned long done_secs;
451 extern char *job_section;
452 extern int fio_gtod_offload;
453 extern int fio_gtod_cpu;
455 extern struct thread_data *threads;
457 static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
459 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
462 #define BLOCKS_PER_MAP (8 * sizeof(int))
463 #define TO_MAP_BLOCK(f, b) (b)
464 #define RAND_MAP_IDX(f, b) (TO_MAP_BLOCK(f, b) / BLOCKS_PER_MAP)
465 #define RAND_MAP_BIT(f, b) (TO_MAP_BLOCK(f, b) & (BLOCKS_PER_MAP - 1))
467 #define MAX_JOBS (1024)
469 #define td_non_fatal_error(e) ((e) == EIO || (e) == EILSEQ)
471 static inline void update_error_count(struct thread_data *td, int err)
473 td->total_err_count++;
474 if (td->total_err_count == 1)
475 td->first_error = err;
478 static inline int should_fsync(struct thread_data *td)
480 if (td->last_was_sync)
484 if (td_write(td) || td_rw(td) || td->o.override_sync)
493 extern unsigned long long utime_since(struct timeval *, struct timeval *);
494 extern unsigned long long utime_since_now(struct timeval *);
495 extern unsigned long mtime_since(struct timeval *, struct timeval *);
496 extern unsigned long mtime_since_now(struct timeval *);
497 extern unsigned long time_since_now(struct timeval *);
498 extern unsigned long mtime_since_genesis(void);
499 extern void usec_spin(unsigned int);
500 extern void usec_sleep(struct thread_data *, unsigned long);
501 extern void fill_start_time(struct timeval *);
502 extern void fio_gettime(struct timeval *, void *);
503 extern void fio_gtod_init(void);
504 extern void fio_gtod_update(void);
505 extern void set_genesis_time(void);
506 extern int ramp_time_over(struct thread_data *);
507 extern int in_ramp_time(struct thread_data *);
510 * Init/option functions
512 extern int __must_check parse_options(int, char **);
513 extern int fio_options_parse(struct thread_data *, char **, int);
514 extern void fio_keywords_init(void);
515 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
516 extern void fio_fill_default_options(struct thread_data *);
517 extern int fio_show_option_help(const char *);
518 extern void fio_options_dup_and_init(struct option *);
519 extern void options_mem_dupe(struct thread_data *);
520 extern void options_mem_free(struct thread_data *);
521 extern void td_fill_rand_seeds(struct thread_data *);
522 #define FIO_GETOPT_JOB 0x89988998
523 #define FIO_NR_OPTIONS 128
528 extern void print_thread_status(void);
529 extern void print_status_init(int);
532 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
533 * will never back again. It may cycle between running/verififying/fsyncing.
534 * Once the thread reaches TD_EXITED, it is just waiting for the core to
550 extern void td_set_runstate(struct thread_data *, int);
555 extern int __must_check fio_pin_memory(void);
556 extern void fio_unpin_memory(void);
557 extern int __must_check allocate_io_mem(struct thread_data *);
558 extern void free_io_mem(struct thread_data *);
561 * Reset stats after ramp time completes
563 extern void reset_all_stats(struct thread_data *);
568 #ifdef FIO_HAVE_BLKTRACE
569 extern int is_blktrace(const char *);
570 extern int load_blktrace(struct thread_data *, const char *);
574 * Mark unused variables passed to ops functions as unused, to silence gcc
576 #define fio_unused __attribute((__unused__))
577 #define fio_init __attribute__((constructor))
578 #define fio_exit __attribute__((destructor))
580 #define for_each_td(td, i) \
581 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
582 #define for_each_file(td, f, i) \
583 if ((td)->files_index) \
584 for ((i) = 0, (f) = (td)->files[0]; \
585 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
588 #define fio_assert(td, cond) do { \
591 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
592 (td)->runstate = TD_EXITED; \
593 (td)->error = EFAULT; \
598 static inline int fio_fill_issue_time(struct thread_data *td)
600 if (td->o.read_iolog_file ||
601 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
608 * Cheesy number->string conversion, complete with carry rounding error.
610 static inline char *num2str(unsigned long num, int maxlen, int base, int pow2)
612 char postfix[] = { ' ', 'K', 'M', 'G', 'P', 'E' };
613 unsigned int thousand;
624 for (i = 0; base > 1; i++)
630 len = sprintf(buf, "%'lu", num);
633 buf[len] = postfix[i];
639 if ((num % thousand) >= (thousand / 2))
650 static inline int __should_check_rate(struct thread_data *td,
653 struct thread_options *o = &td->o;
656 * If some rate setting was given, we need to check it
658 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
659 o->rate_iops_min[ddir])
665 static inline int should_check_rate(struct thread_data *td,
666 unsigned long *bytes_done)
671 ret |= __should_check_rate(td, 0);
673 ret |= __should_check_rate(td, 1);
678 static inline int is_power_of_2(unsigned int val)
680 return (val != 0 && ((val & (val - 1)) == 0));
684 * We currently only need to do locking if we have verifier threads
685 * accessing our internal structures too
687 static inline void td_io_u_lock(struct thread_data *td)
689 if (td->o.verify_async)
690 pthread_mutex_lock(&td->io_u_lock);
693 static inline void td_io_u_unlock(struct thread_data *td)
695 if (td->o.verify_async)
696 pthread_mutex_unlock(&td->io_u_lock);
699 static inline void td_io_u_free_notify(struct thread_data *td)
701 if (td->o.verify_async)
702 pthread_cond_signal(&td->free_cond);