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];
48 * What type of allocation to use for io buffers
51 MEM_MALLOC = 0, /* ordinary malloc */
52 MEM_SHM, /* use shared memory segments */
53 MEM_SHMHUGE, /* use shared memory segments with huge pages */
54 MEM_MMAP, /* use anonynomous mmap */
55 MEM_MMAPHUGE, /* memory mapped huge file */
59 * How many depth levels to log
61 #define FIO_IO_U_MAP_NR 8
62 #define FIO_IO_U_LAT_U_NR 10
63 #define FIO_IO_U_LAT_M_NR 12
74 struct io_log *slat_log;
75 struct io_log *clat_log;
76 struct io_log *bw_log;
79 * bandwidth and latency stats
81 struct io_stat clat_stat[2]; /* completion latency */
82 struct io_stat slat_stat[2]; /* submission latency */
83 struct io_stat bw_stat[2]; /* bandwidth stats */
85 unsigned long long stat_io_bytes[2];
86 struct timeval stat_sample_time[2];
89 * fio system usage accounting
91 struct rusage ru_start;
93 unsigned long usr_time;
94 unsigned long sys_time;
96 unsigned long minf, majf;
99 * IO depth and latency stats
101 unsigned int io_u_map[FIO_IO_U_MAP_NR];
102 unsigned int io_u_submit[FIO_IO_U_MAP_NR];
103 unsigned int io_u_complete[FIO_IO_U_MAP_NR];
104 unsigned int io_u_lat_u[FIO_IO_U_LAT_U_NR];
105 unsigned int io_u_lat_m[FIO_IO_U_LAT_M_NR];
106 unsigned long total_io_u[2];
107 unsigned long short_io_u[2];
108 unsigned long total_submit;
109 unsigned long total_complete;
111 unsigned long long io_bytes[2];
112 unsigned long runtime[2];
113 unsigned long total_run_time;
116 * IO Error related stats
118 unsigned continue_on_error;
119 unsigned long total_err_count;
128 struct thread_options {
136 enum td_ddir td_ddir;
137 unsigned int ddir_nr;
138 unsigned int iodepth;
139 unsigned int iodepth_low;
140 unsigned int iodepth_batch;
141 unsigned int iodepth_batch_complete;
143 unsigned long long size;
144 unsigned int fill_device;
145 unsigned long long file_size_low;
146 unsigned long long file_size_high;
147 unsigned long long start_offset;
151 unsigned int min_bs[2];
152 unsigned int max_bs[2];
153 struct bssplit *bssplit[2];
154 unsigned int bssplit_nr[2];
156 unsigned int nr_files;
157 unsigned int open_files;
158 enum file_lock_mode file_lock_mode;
159 unsigned int lockfile_batch;
161 unsigned int odirect;
162 unsigned int invalidate_cache;
163 unsigned int create_serialize;
164 unsigned int create_fsync;
165 unsigned int create_on_open;
166 unsigned int end_fsync;
167 unsigned int pre_read;
168 unsigned int sync_io;
170 unsigned int do_verify;
171 unsigned int verifysort;
172 unsigned int verify_interval;
173 unsigned int verify_offset;
174 unsigned int verify_pattern;
175 unsigned int verify_pattern_bytes;
176 unsigned int verify_fatal;
177 unsigned int verify_async;
178 unsigned int use_thread;
180 unsigned int do_disk_util;
181 unsigned int override_sync;
182 unsigned int rand_repeatable;
183 unsigned int write_lat_log;
184 unsigned int write_bw_log;
185 unsigned int norandommap;
186 unsigned int softrandommap;
187 unsigned int bs_unaligned;
188 unsigned int fsync_on_close;
190 unsigned int hugepage_size;
191 unsigned int rw_min_bs;
192 unsigned int thinktime;
193 unsigned int thinktime_spin;
194 unsigned int thinktime_blocks;
195 unsigned int fsync_blocks;
196 unsigned int fdatasync_blocks;
197 unsigned int start_delay;
198 unsigned long long timeout;
199 unsigned long long ramp_time;
200 unsigned int overwrite;
201 unsigned int bw_avg_time;
203 unsigned long long zone_size;
204 unsigned long long zone_skip;
205 enum fio_memtype mem_type;
206 unsigned int mem_align;
208 unsigned int stonewall;
209 unsigned int new_group;
210 unsigned int numjobs;
211 os_cpu_mask_t cpumask;
212 unsigned int cpumask_set;
213 os_cpu_mask_t verify_cpumask;
214 unsigned int verify_cpumask_set;
216 unsigned int rwmixcycle;
217 unsigned int rwmix[2];
219 unsigned int file_service_type;
220 unsigned int group_reporting;
221 unsigned int fadvise_hint;
222 unsigned int zero_buffers;
223 unsigned int refill_buffers;
224 unsigned int time_based;
225 unsigned int disable_clat;
226 unsigned int disable_slat;
227 unsigned int disable_bw;
228 unsigned int gtod_reduce;
229 unsigned int gtod_cpu;
230 unsigned int gtod_offload;
232 char *read_iolog_file;
233 char *write_iolog_file;
238 * Pre-run and post-run shell
243 unsigned int rate[2];
244 unsigned int ratemin[2];
245 unsigned int ratecycle;
246 unsigned int rate_iops[2];
247 unsigned int rate_iops_min[2];
252 * CPU "io" cycle burner
254 unsigned int cpuload;
255 unsigned int cpucycle;
260 unsigned int continue_on_error;
263 #define FIO_VERROR_SIZE 128
266 * This describes a single thread/process executing a fio job.
269 struct thread_options o;
270 char verror[FIO_VERROR_SIZE];
274 struct thread_stat ts;
275 struct fio_file **files;
276 unsigned int files_size;
277 unsigned int files_index;
278 unsigned int nr_open_files;
279 unsigned int nr_done_files;
280 unsigned int nr_normal_files;
282 unsigned int next_file;
283 os_random_state_t next_file_state;
289 size_t orig_buffer_size;
290 volatile int terminate;
291 volatile int runstate;
293 unsigned int ioprio_set;
294 unsigned int last_was_sync;
304 unsigned long rand_seeds[6];
306 os_random_state_t bsrange_state;
307 os_random_state_t verify_state;
312 * IO engine hooks, contains everything needed to submit an io_u
313 * to any of the available IO engines.
315 struct ioengine_ops *io_ops;
318 * Current IO depth and list of free and busy io_u's.
320 unsigned int cur_depth;
321 unsigned int io_u_queued;
322 struct flist_head io_u_freelist;
323 struct flist_head io_u_busylist;
324 struct flist_head io_u_requeues;
325 pthread_mutex_t io_u_lock;
326 pthread_cond_t free_cond;
329 * async verify offload
331 struct flist_head verify_list;
332 pthread_t *verify_threads;
333 unsigned int nr_verify_threads;
334 pthread_cond_t verify_cond;
335 int verify_thread_exit;
340 unsigned long rate_usec_cycle[2];
341 long rate_pending_usleep[2];
342 unsigned long rate_bytes[2];
343 unsigned long rate_blocks[2];
344 struct timeval lastrate[2];
346 unsigned long long total_io_size;
348 unsigned long io_issues[2];
349 unsigned long long io_blocks[2];
350 unsigned long long io_bytes[2];
351 unsigned long long io_skip_bytes;
352 unsigned long long this_io_bytes[2];
353 unsigned long long zone_bytes;
354 struct fio_mutex *mutex;
357 * State for random io, a bitmap of blocks done vs not done
359 os_random_state_t random_state;
361 struct timeval start; /* start of this loop */
362 struct timeval epoch; /* time job was started */
363 struct timeval last_issue;
364 struct timeval tv_cache;
365 unsigned int tv_cache_nr;
366 unsigned int tv_cache_mask;
367 unsigned int ramp_time_over;
370 * read/write mixed workload state
372 os_random_state_t rwmix_state;
373 unsigned long rwmix_issues;
374 enum fio_ddir rwmix_ddir;
375 unsigned int ddir_nr;
378 * IO history logs for verification. We use a tree for sorting,
379 * if we are overwriting. Otherwise just use a fifo.
381 struct rb_root io_hist_tree;
382 struct flist_head io_hist_list;
387 struct flist_head io_log_list;
390 * for fileservice, how often to switch to a new file
392 unsigned int file_service_nr;
393 unsigned int file_service_left;
394 struct fio_file *file_service_file;
397 * For generating file sizes
399 os_random_state_t file_size_state;
404 unsigned int total_err_count;
409 * when should interactive ETA output be generated
417 #define __td_verror(td, err, msg, func) \
423 if (!(td)->first_error) \
424 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
428 #define td_clear_error(td) \
430 #define td_verror(td, err, func) \
431 __td_verror((td), (err), strerror((err)), (func))
432 #define td_vmsg(td, err, msg, func) \
433 __td_verror((td), (err), (msg), (func))
435 extern int exitall_on_terminate;
436 extern int thread_number;
437 extern int nr_process, nr_thread;
440 extern int terse_output;
441 extern int temp_stall_ts;
442 extern unsigned long long mlock_size;
443 extern unsigned long page_mask, page_size;
444 extern int read_only;
445 extern int eta_print;
446 extern unsigned long done_secs;
447 extern char *job_section;
448 extern int fio_gtod_offload;
449 extern int fio_gtod_cpu;
451 extern struct thread_data *threads;
453 static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
455 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
458 #define BLOCKS_PER_MAP (8 * sizeof(int))
459 #define TO_MAP_BLOCK(f, b) (b)
460 #define RAND_MAP_IDX(f, b) (TO_MAP_BLOCK(f, b) / BLOCKS_PER_MAP)
461 #define RAND_MAP_BIT(f, b) (TO_MAP_BLOCK(f, b) & (BLOCKS_PER_MAP - 1))
463 #define MAX_JOBS (1024)
465 #define td_non_fatal_error(e) ((e) == EIO || (e) == EILSEQ)
467 static inline void update_error_count(struct thread_data *td, int err)
469 td->total_err_count++;
470 if (td->total_err_count == 1)
471 td->first_error = err;
474 static inline int should_fsync(struct thread_data *td)
476 if (td->last_was_sync)
480 if (td_write(td) || td_rw(td) || td->o.override_sync)
489 extern unsigned long long utime_since(struct timeval *, struct timeval *);
490 extern unsigned long long utime_since_now(struct timeval *);
491 extern unsigned long mtime_since(struct timeval *, struct timeval *);
492 extern unsigned long mtime_since_now(struct timeval *);
493 extern unsigned long time_since_now(struct timeval *);
494 extern unsigned long mtime_since_genesis(void);
495 extern void usec_spin(unsigned int);
496 extern void usec_sleep(struct thread_data *, unsigned long);
497 extern void fill_start_time(struct timeval *);
498 extern void fio_gettime(struct timeval *, void *);
499 extern void fio_gtod_init(void);
500 extern void fio_gtod_update(void);
501 extern void set_genesis_time(void);
502 extern int ramp_time_over(struct thread_data *);
503 extern int in_ramp_time(struct thread_data *);
506 * Init/option functions
508 extern int __must_check parse_options(int, char **);
509 extern int fio_options_parse(struct thread_data *, char **, int);
510 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
511 extern void fio_fill_default_options(struct thread_data *);
512 extern int fio_show_option_help(const char *);
513 extern void fio_options_dup_and_init(struct option *);
514 extern void options_mem_dupe(struct thread_data *);
515 extern void options_mem_free(struct thread_data *);
516 extern void td_fill_rand_seeds(struct thread_data *);
517 #define FIO_GETOPT_JOB 0x89988998
518 #define FIO_NR_OPTIONS 128
523 extern void print_thread_status(void);
524 extern void print_status_init(int);
527 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
528 * will never back again. It may cycle between running/verififying/fsyncing.
529 * Once the thread reaches TD_EXITED, it is just waiting for the core to
545 extern void td_set_runstate(struct thread_data *, int);
550 extern int __must_check fio_pin_memory(void);
551 extern void fio_unpin_memory(void);
552 extern int __must_check allocate_io_mem(struct thread_data *);
553 extern void free_io_mem(struct thread_data *);
556 * Reset stats after ramp time completes
558 extern void reset_all_stats(struct thread_data *);
563 #ifdef FIO_HAVE_BLKTRACE
564 extern int is_blktrace(const char *);
565 extern int load_blktrace(struct thread_data *, const char *);
569 * Mark unused variables passed to ops functions as unused, to silence gcc
571 #define fio_unused __attribute((__unused__))
572 #define fio_init __attribute__((constructor))
573 #define fio_exit __attribute__((destructor))
575 #define for_each_td(td, i) \
576 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
577 #define for_each_file(td, f, i) \
578 if ((td)->files_index) \
579 for ((i) = 0, (f) = (td)->files[0]; \
580 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
583 #define fio_assert(td, cond) do { \
586 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
587 (td)->runstate = TD_EXITED; \
588 (td)->error = EFAULT; \
593 static inline int fio_fill_issue_time(struct thread_data *td)
595 if (td->o.read_iolog_file ||
596 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
603 * Cheesy number->string conversion, complete with carry rounding error.
605 static inline char *num2str(unsigned long num, int maxlen, int base, int pow2)
607 char postfix[] = { ' ', 'K', 'M', 'G', 'P', 'E' };
608 unsigned int thousand;
619 for (i = 0; base > 1; i++)
625 len = sprintf(buf, "%'lu", num);
628 buf[len] = postfix[i];
634 if ((num % thousand) >= (thousand / 2))
645 static inline int __should_check_rate(struct thread_data *td,
648 struct thread_options *o = &td->o;
651 * If some rate setting was given, we need to check it
653 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
654 o->rate_iops_min[ddir])
660 static inline int should_check_rate(struct thread_data *td,
661 unsigned long *bytes_done)
666 ret |= __should_check_rate(td, 0);
668 ret |= __should_check_rate(td, 1);
673 static inline int is_power_of_2(unsigned int val)
675 return (val != 0 && ((val & (val - 1)) == 0));
679 * We currently only need to do locking if we have verifier threads
680 * accessing our internal structures too
682 static inline void td_io_u_lock(struct thread_data *td)
684 if (td->o.verify_async)
685 pthread_mutex_lock(&td->io_u_lock);
688 static inline void td_io_u_unlock(struct thread_data *td)
690 if (td->o.verify_async)
691 pthread_mutex_unlock(&td->io_u_lock);
694 static inline void td_io_u_free_notify(struct thread_data *td)
696 if (td->o.verify_async)
697 pthread_cond_signal(&td->free_cond);