8 #include <sys/resource.h>
18 #include "compiler/compiler.h"
22 #include "arch/arch.h"
40 #ifdef FIO_HAVE_SOLARISAIO
41 #include <sys/asynch.h>
44 struct group_run_stats {
45 unsigned long long max_run[2], min_run[2];
46 unsigned long long max_bw[2], min_bw[2];
47 unsigned long long io_kb[2];
48 unsigned long long agg[2];
53 * What type of allocation to use for io buffers
56 MEM_MALLOC = 0, /* ordinary malloc */
57 MEM_SHM, /* use shared memory segments */
58 MEM_SHMHUGE, /* use shared memory segments with huge pages */
59 MEM_MMAP, /* use anonynomous mmap */
60 MEM_MMAPHUGE, /* memory mapped huge file */
64 * How many depth levels to log
66 #define FIO_IO_U_MAP_NR 8
67 #define FIO_IO_U_LAT_U_NR 10
68 #define FIO_IO_U_LAT_M_NR 12
70 #define MAX_PATTERN_SIZE 512
81 struct io_log *slat_log;
82 struct io_log *clat_log;
83 struct io_log *bw_log;
86 * bandwidth and latency stats
88 struct io_stat clat_stat[2]; /* completion latency */
89 struct io_stat slat_stat[2]; /* submission latency */
90 struct io_stat bw_stat[2]; /* bandwidth stats */
92 unsigned long long stat_io_bytes[2];
93 struct timeval stat_sample_time[2];
96 * fio system usage accounting
98 struct rusage ru_start;
100 unsigned long usr_time;
101 unsigned long sys_time;
103 unsigned long minf, majf;
106 * IO depth and latency stats
108 unsigned int io_u_map[FIO_IO_U_MAP_NR];
109 unsigned int io_u_submit[FIO_IO_U_MAP_NR];
110 unsigned int io_u_complete[FIO_IO_U_MAP_NR];
111 unsigned int io_u_lat_u[FIO_IO_U_LAT_U_NR];
112 unsigned int io_u_lat_m[FIO_IO_U_LAT_M_NR];
113 unsigned long total_io_u[2];
114 unsigned long short_io_u[2];
115 unsigned long total_submit;
116 unsigned long total_complete;
118 unsigned long long io_bytes[2];
119 unsigned long runtime[2];
120 unsigned long total_run_time;
123 * IO Error related stats
125 unsigned continue_on_error;
126 unsigned long total_err_count;
129 unsigned int kb_base;
137 struct thread_options {
145 enum td_ddir td_ddir;
146 unsigned int kb_base;
147 unsigned int ddir_nr;
148 unsigned int iodepth;
149 unsigned int iodepth_low;
150 unsigned int iodepth_batch;
151 unsigned int iodepth_batch_complete;
153 unsigned long long size;
154 unsigned int fill_device;
155 unsigned long long file_size_low;
156 unsigned long long file_size_high;
157 unsigned long long start_offset;
161 unsigned int min_bs[2];
162 unsigned int max_bs[2];
163 struct bssplit *bssplit[2];
164 unsigned int bssplit_nr[2];
166 unsigned int nr_files;
167 unsigned int open_files;
168 enum file_lock_mode file_lock_mode;
169 unsigned int lockfile_batch;
171 unsigned int odirect;
172 unsigned int invalidate_cache;
173 unsigned int create_serialize;
174 unsigned int create_fsync;
175 unsigned int create_on_open;
176 unsigned int end_fsync;
177 unsigned int pre_read;
178 unsigned int sync_io;
180 unsigned int do_verify;
181 unsigned int verifysort;
182 unsigned int verify_interval;
183 unsigned int verify_offset;
184 char verify_pattern[MAX_PATTERN_SIZE];
185 unsigned int verify_pattern_bytes;
186 unsigned int verify_fatal;
187 unsigned int verify_async;
188 unsigned long long verify_backlog;
189 unsigned int verify_batch;
190 unsigned int use_thread;
192 unsigned int do_disk_util;
193 unsigned int override_sync;
194 unsigned int rand_repeatable;
195 unsigned int write_lat_log;
196 unsigned int write_bw_log;
197 unsigned int norandommap;
198 unsigned int softrandommap;
199 unsigned int bs_unaligned;
200 unsigned int fsync_on_close;
202 unsigned int hugepage_size;
203 unsigned int rw_min_bs;
204 unsigned int thinktime;
205 unsigned int thinktime_spin;
206 unsigned int thinktime_blocks;
207 unsigned int fsync_blocks;
208 unsigned int fdatasync_blocks;
209 unsigned int start_delay;
210 unsigned long long timeout;
211 unsigned long long ramp_time;
212 unsigned int overwrite;
213 unsigned int bw_avg_time;
215 unsigned long long zone_size;
216 unsigned long long zone_skip;
217 enum fio_memtype mem_type;
218 unsigned int mem_align;
220 unsigned int stonewall;
221 unsigned int new_group;
222 unsigned int numjobs;
223 os_cpu_mask_t cpumask;
224 unsigned int cpumask_set;
225 os_cpu_mask_t verify_cpumask;
226 unsigned int verify_cpumask_set;
228 unsigned int rwmixcycle;
229 unsigned int rwmix[2];
231 unsigned int file_service_type;
232 unsigned int group_reporting;
233 unsigned int fadvise_hint;
234 unsigned int fallocate;
235 unsigned int zero_buffers;
236 unsigned int refill_buffers;
237 unsigned int time_based;
238 unsigned int disable_clat;
239 unsigned int disable_slat;
240 unsigned int disable_bw;
241 unsigned int gtod_reduce;
242 unsigned int gtod_cpu;
243 unsigned int gtod_offload;
244 enum fio_cs clocksource;
246 char *read_iolog_file;
247 char *write_iolog_file;
252 * Pre-run and post-run shell
257 unsigned int rate[2];
258 unsigned int ratemin[2];
259 unsigned int ratecycle;
260 unsigned int rate_iops[2];
261 unsigned int rate_iops_min[2];
266 * CPU "io" cycle burner
268 unsigned int cpuload;
269 unsigned int cpucycle;
274 unsigned int continue_on_error;
277 * Benchmark profile type
282 * blkio cgroup support
285 unsigned int cgroup_weight;
286 unsigned int cgroup_nodelete;
291 unsigned int sync_file_range;
294 #define FIO_VERROR_SIZE 128
297 * This describes a single thread/process executing a fio job.
300 struct thread_options o;
301 char verror[FIO_VERROR_SIZE];
305 struct thread_stat ts;
306 struct fio_file **files;
307 unsigned int files_size;
308 unsigned int files_index;
309 unsigned int nr_open_files;
310 unsigned int nr_done_files;
311 unsigned int nr_normal_files;
313 unsigned int next_file;
314 os_random_state_t next_file_state;
320 size_t orig_buffer_size;
321 volatile int terminate;
322 volatile int runstate;
324 unsigned int ioprio_set;
325 unsigned int last_was_sync;
326 enum fio_ddir last_ddir;
336 unsigned long rand_seeds[6];
338 os_random_state_t bsrange_state;
339 os_random_state_t verify_state;
341 unsigned int verify_batch;
346 * IO engine hooks, contains everything needed to submit an io_u
347 * to any of the available IO engines.
349 struct ioengine_ops *io_ops;
352 * Current IO depth and list of free and busy io_u's.
354 unsigned int cur_depth;
355 unsigned int io_u_queued;
356 struct flist_head io_u_freelist;
357 struct flist_head io_u_busylist;
358 struct flist_head io_u_requeues;
359 pthread_mutex_t io_u_lock;
360 pthread_cond_t free_cond;
363 * async verify offload
365 struct flist_head verify_list;
366 pthread_t *verify_threads;
367 unsigned int nr_verify_threads;
368 pthread_cond_t verify_cond;
369 int verify_thread_exit;
374 unsigned long rate_nsec_cycle[2];
375 long rate_pending_usleep[2];
376 unsigned long rate_bytes[2];
377 unsigned long rate_blocks[2];
378 struct timeval lastrate[2];
380 unsigned long long total_io_size;
381 unsigned long long fill_device_size;
383 unsigned long io_issues[2];
384 unsigned long long io_blocks[2];
385 unsigned long long io_bytes[2];
386 unsigned long long io_skip_bytes;
387 unsigned long long this_io_bytes[2];
388 unsigned long long zone_bytes;
389 struct fio_mutex *mutex;
392 * State for random io, a bitmap of blocks done vs not done
394 os_random_state_t random_state;
396 struct timeval start; /* start of this loop */
397 struct timeval epoch; /* time job was started */
398 struct timeval last_issue;
399 struct timeval tv_cache;
400 unsigned int tv_cache_nr;
401 unsigned int tv_cache_mask;
402 unsigned int ramp_time_over;
405 * read/write mixed workload state
407 os_random_state_t rwmix_state;
408 unsigned long rwmix_issues;
409 enum fio_ddir rwmix_ddir;
410 unsigned int ddir_nr;
413 * IO history logs for verification. We use a tree for sorting,
414 * if we are overwriting. Otherwise just use a fifo.
416 struct rb_root io_hist_tree;
417 struct flist_head io_hist_list;
418 unsigned long io_hist_len;
423 struct flist_head io_log_list;
426 * for fileservice, how often to switch to a new file
428 unsigned int file_service_nr;
429 unsigned int file_service_left;
430 struct fio_file *file_service_file;
432 unsigned int sync_file_range_nr;
435 * For generating file sizes
437 os_random_state_t file_size_state;
442 unsigned int total_err_count;
446 * Can be overloaded by profiles
448 struct prof_io_ops prof_io_ops;
453 * when should interactive ETA output be generated
461 #define __td_verror(td, err, msg, func) \
467 if (!(td)->first_error) \
468 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
472 #define td_clear_error(td) \
474 #define td_verror(td, err, func) \
475 __td_verror((td), (err), strerror((err)), (func))
476 #define td_vmsg(td, err, msg, func) \
477 __td_verror((td), (err), (msg), (func))
479 extern int exitall_on_terminate;
480 extern int thread_number;
481 extern int nr_process, nr_thread;
484 extern int terse_output;
485 extern int temp_stall_ts;
486 extern unsigned long long mlock_size;
487 extern unsigned long page_mask, page_size;
488 extern int read_only;
489 extern int eta_print;
490 extern unsigned long done_secs;
491 extern char *job_section;
492 extern int fio_gtod_offload;
493 extern int fio_gtod_cpu;
494 extern enum fio_cs fio_clock_source;
496 extern struct thread_data *threads;
498 static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
500 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
503 #define BLOCKS_PER_MAP (8 * sizeof(int))
504 #define TO_MAP_BLOCK(f, b) (b)
505 #define RAND_MAP_IDX(f, b) (TO_MAP_BLOCK(f, b) / BLOCKS_PER_MAP)
506 #define RAND_MAP_BIT(f, b) (TO_MAP_BLOCK(f, b) & (BLOCKS_PER_MAP - 1))
508 #define MAX_JOBS (1024)
510 #define td_non_fatal_error(e) ((e) == EIO || (e) == EILSEQ)
512 static inline void update_error_count(struct thread_data *td, int err)
514 td->total_err_count++;
515 if (td->total_err_count == 1)
516 td->first_error = err;
519 static inline int should_fsync(struct thread_data *td)
521 if (td->last_was_sync)
525 if (td_write(td) || td_rw(td) || td->o.override_sync)
532 * Init/option functions
534 extern int __must_check parse_options(int, char **);
535 extern int fio_options_parse(struct thread_data *, char **, int);
536 extern void fio_keywords_init(void);
537 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
538 extern void fio_fill_default_options(struct thread_data *);
539 extern int fio_show_option_help(const char *);
540 extern void fio_options_dup_and_init(struct option *);
541 extern void options_mem_dupe(struct thread_data *);
542 extern void options_mem_free(struct thread_data *);
543 extern void td_fill_rand_seeds(struct thread_data *);
544 extern void add_job_opts(const char **);
545 #define FIO_GETOPT_JOB 0x89988998
546 #define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
551 extern void print_thread_status(void);
552 extern void print_status_init(int);
555 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
556 * will never back again. It may cycle between running/verififying/fsyncing.
557 * Once the thread reaches TD_EXITED, it is just waiting for the core to
573 extern void td_set_runstate(struct thread_data *, int);
578 extern int __must_check fio_pin_memory(void);
579 extern void fio_unpin_memory(void);
580 extern int __must_check allocate_io_mem(struct thread_data *);
581 extern void free_io_mem(struct thread_data *);
584 * Reset stats after ramp time completes
586 extern void reset_all_stats(struct thread_data *);
591 #ifdef FIO_HAVE_BLKTRACE
592 extern int is_blktrace(const char *);
593 extern int load_blktrace(struct thread_data *, const char *);
597 * Mark unused variables passed to ops functions as unused, to silence gcc
599 #define fio_unused __attribute((__unused__))
600 #define fio_init __attribute__((constructor))
601 #define fio_exit __attribute__((destructor))
603 #define for_each_td(td, i) \
604 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
605 #define for_each_file(td, f, i) \
606 if ((td)->files_index) \
607 for ((i) = 0, (f) = (td)->files[0]; \
608 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
611 #define fio_assert(td, cond) do { \
614 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
615 (td)->runstate = TD_EXITED; \
616 (td)->error = EFAULT; \
621 static inline int fio_fill_issue_time(struct thread_data *td)
623 if (td->o.read_iolog_file ||
624 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
631 * Cheesy number->string conversion, complete with carry rounding error.
633 static inline char *num2str(unsigned long num, int maxlen, int base, int pow2)
635 char postfix[] = { ' ', 'K', 'M', 'G', 'P', 'E' };
636 unsigned int thousand;
647 for (i = 0; base > 1; i++)
653 len = sprintf(buf, "%'lu", num);
656 buf[len] = postfix[i];
662 if ((num % thousand) >= (thousand / 2))
673 static inline int __should_check_rate(struct thread_data *td,
676 struct thread_options *o = &td->o;
679 * If some rate setting was given, we need to check it
681 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
682 o->rate_iops_min[ddir])
688 static inline int should_check_rate(struct thread_data *td,
689 unsigned long *bytes_done)
694 ret |= __should_check_rate(td, 0);
696 ret |= __should_check_rate(td, 1);
701 static inline int is_power_of_2(unsigned int val)
703 return (val != 0 && ((val & (val - 1)) == 0));
707 * We currently only need to do locking if we have verifier threads
708 * accessing our internal structures too
710 static inline void td_io_u_lock(struct thread_data *td)
712 if (td->o.verify_async)
713 pthread_mutex_lock(&td->io_u_lock);
716 static inline void td_io_u_unlock(struct thread_data *td)
718 if (td->o.verify_async)
719 pthread_mutex_unlock(&td->io_u_lock);
722 static inline void td_io_u_free_notify(struct thread_data *td)
724 if (td->o.verify_async)
725 pthread_cond_signal(&td->free_cond);