8 #include <sys/resource.h>
17 #include "compiler/compiler.h"
21 #include "arch/arch.h"
34 #include "lib/getopt.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 * offset generator types
72 * How many depth levels to log
74 #define FIO_IO_U_MAP_NR 8
75 #define FIO_IO_U_LAT_U_NR 10
76 #define FIO_IO_U_LAT_M_NR 12
78 #define MAX_PATTERN_SIZE 512
89 struct io_log *slat_log;
90 struct io_log *clat_log;
91 struct io_log *lat_log;
92 struct io_log *bw_log;
95 * bandwidth and latency stats
97 struct io_stat clat_stat[2]; /* completion latency */
98 struct io_stat slat_stat[2]; /* submission latency */
99 struct io_stat lat_stat[2]; /* total latency */
100 struct io_stat bw_stat[2]; /* bandwidth stats */
102 unsigned long long stat_io_bytes[2];
103 struct timeval stat_sample_time[2];
106 * fio system usage accounting
108 struct rusage ru_start;
109 struct rusage ru_end;
110 unsigned long usr_time;
111 unsigned long sys_time;
113 unsigned long minf, majf;
116 * IO depth and latency stats
118 unsigned int io_u_map[FIO_IO_U_MAP_NR];
119 unsigned int io_u_submit[FIO_IO_U_MAP_NR];
120 unsigned int io_u_complete[FIO_IO_U_MAP_NR];
121 unsigned int io_u_lat_u[FIO_IO_U_LAT_U_NR];
122 unsigned int io_u_lat_m[FIO_IO_U_LAT_M_NR];
123 unsigned long total_io_u[2];
124 unsigned long short_io_u[2];
125 unsigned long total_submit;
126 unsigned long total_complete;
128 unsigned long long io_bytes[2];
129 unsigned long runtime[2];
130 unsigned long total_run_time;
133 * IO Error related stats
135 unsigned continue_on_error;
136 unsigned long total_err_count;
139 unsigned int kb_base;
147 struct thread_options {
155 enum td_ddir td_ddir;
157 unsigned int kb_base;
158 unsigned int ddir_seq_nr;
159 unsigned int iodepth;
160 unsigned int iodepth_low;
161 unsigned int iodepth_batch;
162 unsigned int iodepth_batch_complete;
164 unsigned long long size;
165 unsigned int fill_device;
166 unsigned long long file_size_low;
167 unsigned long long file_size_high;
168 unsigned long long start_offset;
172 unsigned int min_bs[2];
173 unsigned int max_bs[2];
174 struct bssplit *bssplit[2];
175 unsigned int bssplit_nr[2];
177 unsigned int nr_files;
178 unsigned int open_files;
179 enum file_lock_mode file_lock_mode;
180 unsigned int lockfile_batch;
182 unsigned int odirect;
183 unsigned int invalidate_cache;
184 unsigned int create_serialize;
185 unsigned int create_fsync;
186 unsigned int create_on_open;
187 unsigned int end_fsync;
188 unsigned int pre_read;
189 unsigned int sync_io;
191 unsigned int do_verify;
192 unsigned int verifysort;
193 unsigned int verify_interval;
194 unsigned int verify_offset;
195 char verify_pattern[MAX_PATTERN_SIZE];
196 unsigned int verify_pattern_bytes;
197 unsigned int verify_fatal;
198 unsigned int verify_async;
199 unsigned long long verify_backlog;
200 unsigned int verify_batch;
201 unsigned int use_thread;
203 unsigned int do_disk_util;
204 unsigned int override_sync;
205 unsigned int rand_repeatable;
206 unsigned int write_lat_log;
207 unsigned int write_bw_log;
208 unsigned int norandommap;
209 unsigned int softrandommap;
210 unsigned int bs_unaligned;
211 unsigned int fsync_on_close;
213 unsigned int hugepage_size;
214 unsigned int rw_min_bs;
215 unsigned int thinktime;
216 unsigned int thinktime_spin;
217 unsigned int thinktime_blocks;
218 unsigned int fsync_blocks;
219 unsigned int fdatasync_blocks;
220 unsigned long start_delay;
221 unsigned long long timeout;
222 unsigned long long ramp_time;
223 unsigned int overwrite;
224 unsigned int bw_avg_time;
226 unsigned long long zone_size;
227 unsigned long long zone_skip;
228 enum fio_memtype mem_type;
229 unsigned int mem_align;
231 unsigned int stonewall;
232 unsigned int new_group;
233 unsigned int numjobs;
234 os_cpu_mask_t cpumask;
235 unsigned int cpumask_set;
236 os_cpu_mask_t verify_cpumask;
237 unsigned int verify_cpumask_set;
239 unsigned int rwmixcycle;
240 unsigned int rwmix[2];
242 unsigned int file_service_type;
243 unsigned int group_reporting;
244 unsigned int fadvise_hint;
245 unsigned int fallocate;
246 unsigned int zero_buffers;
247 unsigned int refill_buffers;
248 unsigned int time_based;
249 unsigned int disable_lat;
250 unsigned int disable_clat;
251 unsigned int disable_slat;
252 unsigned int disable_bw;
253 unsigned int gtod_reduce;
254 unsigned int gtod_cpu;
255 unsigned int gtod_offload;
256 enum fio_cs clocksource;
257 unsigned int no_stall;
259 char *read_iolog_file;
260 char *write_iolog_file;
265 * Pre-run and post-run shell
270 unsigned int rate[2];
271 unsigned int ratemin[2];
272 unsigned int ratecycle;
273 unsigned int rate_iops[2];
274 unsigned int rate_iops_min[2];
279 * CPU "io" cycle burner
281 unsigned int cpuload;
282 unsigned int cpucycle;
287 unsigned int continue_on_error;
290 * Benchmark profile type
295 * blkio cgroup support
298 unsigned int cgroup_weight;
299 unsigned int cgroup_nodelete;
304 unsigned int sync_file_range;
307 #define FIO_VERROR_SIZE 128
310 * This describes a single thread/process executing a fio job.
313 struct thread_options o;
314 char verror[FIO_VERROR_SIZE];
318 struct thread_stat ts;
319 struct fio_file **files;
320 unsigned int files_size;
321 unsigned int files_index;
322 unsigned int nr_open_files;
323 unsigned int nr_done_files;
324 unsigned int nr_normal_files;
326 unsigned int next_file;
327 os_random_state_t next_file_state;
333 size_t orig_buffer_size;
334 volatile int terminate;
335 volatile int runstate;
337 unsigned int ioprio_set;
338 unsigned int last_was_sync;
339 enum fio_ddir last_ddir;
349 unsigned long rand_seeds[6];
351 os_random_state_t bsrange_state;
352 os_random_state_t verify_state;
354 unsigned int verify_batch;
359 * IO engine hooks, contains everything needed to submit an io_u
360 * to any of the available IO engines.
362 struct ioengine_ops *io_ops;
365 * Current IO depth and list of free and busy io_u's.
367 unsigned int cur_depth;
368 unsigned int io_u_queued;
369 struct flist_head io_u_freelist;
370 struct flist_head io_u_busylist;
371 struct flist_head io_u_requeues;
372 pthread_mutex_t io_u_lock;
373 pthread_cond_t free_cond;
376 * async verify offload
378 struct flist_head verify_list;
379 pthread_t *verify_threads;
380 unsigned int nr_verify_threads;
381 pthread_cond_t verify_cond;
382 int verify_thread_exit;
387 unsigned long rate_nsec_cycle[2];
388 long rate_pending_usleep[2];
389 unsigned long rate_bytes[2];
390 unsigned long rate_blocks[2];
391 struct timeval lastrate[2];
393 unsigned long long total_io_size;
394 unsigned long long fill_device_size;
396 unsigned long io_issues[2];
397 unsigned long long io_blocks[2];
398 unsigned long long io_bytes[2];
399 unsigned long long io_skip_bytes;
400 unsigned long long this_io_bytes[2];
401 unsigned long long zone_bytes;
402 struct fio_mutex *mutex;
405 * State for random io, a bitmap of blocks done vs not done
407 os_random_state_t random_state;
409 struct timeval start; /* start of this loop */
410 struct timeval epoch; /* time job was started */
411 struct timeval last_issue;
412 struct timeval tv_cache;
413 unsigned int tv_cache_nr;
414 unsigned int tv_cache_mask;
415 unsigned int ramp_time_over;
418 * read/write mixed workload state
420 os_random_state_t rwmix_state;
421 unsigned long rwmix_issues;
422 enum fio_ddir rwmix_ddir;
423 unsigned int ddir_seq_nr;
426 * IO history logs for verification. We use a tree for sorting,
427 * if we are overwriting. Otherwise just use a fifo.
429 struct rb_root io_hist_tree;
430 struct flist_head io_hist_list;
431 unsigned long io_hist_len;
436 struct flist_head io_log_list;
439 * for fileservice, how often to switch to a new file
441 unsigned int file_service_nr;
442 unsigned int file_service_left;
443 struct fio_file *file_service_file;
445 unsigned int sync_file_range_nr;
448 * For generating file sizes
450 os_random_state_t file_size_state;
455 unsigned int total_err_count;
459 * Can be overloaded by profiles
461 struct prof_io_ops prof_io_ops;
466 * when should interactive ETA output be generated
474 #define __td_verror(td, err, msg, func) \
480 if (!(td)->first_error) \
481 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
485 #define td_clear_error(td) \
487 #define td_verror(td, err, func) \
488 __td_verror((td), (err), strerror((err)), (func))
489 #define td_vmsg(td, err, msg, func) \
490 __td_verror((td), (err), (msg), (func))
492 extern int exitall_on_terminate;
493 extern int thread_number;
494 extern int nr_process, nr_thread;
497 extern int terse_output;
498 extern int temp_stall_ts;
499 extern unsigned long long mlock_size;
500 extern unsigned long page_mask, page_size;
501 extern int read_only;
502 extern int eta_print;
503 extern unsigned long done_secs;
504 extern char *job_section;
505 extern int fio_gtod_offload;
506 extern int fio_gtod_cpu;
507 extern enum fio_cs fio_clock_source;
509 extern struct thread_data *threads;
511 static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
513 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
516 #define BLOCKS_PER_MAP (8 * sizeof(int))
517 #define TO_MAP_BLOCK(f, b) (b)
518 #define RAND_MAP_IDX(f, b) (TO_MAP_BLOCK(f, b) / BLOCKS_PER_MAP)
519 #define RAND_MAP_BIT(f, b) (TO_MAP_BLOCK(f, b) & (BLOCKS_PER_MAP - 1))
521 #define MAX_JOBS (1024)
523 #define td_non_fatal_error(e) ((e) == EIO || (e) == EILSEQ)
525 static inline void update_error_count(struct thread_data *td, int err)
527 td->total_err_count++;
528 if (td->total_err_count == 1)
529 td->first_error = err;
532 static inline int should_fsync(struct thread_data *td)
534 if (td->last_was_sync)
538 if (td_write(td) || td_rw(td) || td->o.override_sync)
545 * Init/option functions
547 extern int __must_check parse_options(int, char **);
548 extern int fio_options_parse(struct thread_data *, char **, int);
549 extern void fio_keywords_init(void);
550 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
551 extern void fio_fill_default_options(struct thread_data *);
552 extern int fio_show_option_help(const char *);
553 extern void fio_options_dup_and_init(struct option *);
554 extern void options_mem_dupe(struct thread_data *);
555 extern void options_mem_free(struct thread_data *);
556 extern void td_fill_rand_seeds(struct thread_data *);
557 extern void add_job_opts(const char **);
558 #define FIO_GETOPT_JOB 0x89988998
559 #define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
564 extern void print_thread_status(void);
565 extern void print_status_init(int);
568 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
569 * will never back again. It may cycle between running/verififying/fsyncing.
570 * Once the thread reaches TD_EXITED, it is just waiting for the core to
586 extern void td_set_runstate(struct thread_data *, int);
591 extern int __must_check fio_pin_memory(void);
592 extern void fio_unpin_memory(void);
593 extern int __must_check allocate_io_mem(struct thread_data *);
594 extern void free_io_mem(struct thread_data *);
597 * Reset stats after ramp time completes
599 extern void reset_all_stats(struct thread_data *);
604 #ifdef FIO_HAVE_BLKTRACE
605 extern int is_blktrace(const char *);
606 extern int load_blktrace(struct thread_data *, const char *);
610 * Mark unused variables passed to ops functions as unused, to silence gcc
612 #define fio_unused __attribute((__unused__))
613 #define fio_init __attribute__((constructor))
614 #define fio_exit __attribute__((destructor))
616 #define for_each_td(td, i) \
617 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
618 #define for_each_file(td, f, i) \
619 if ((td)->files_index) \
620 for ((i) = 0, (f) = (td)->files[0]; \
621 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
624 #define fio_assert(td, cond) do { \
627 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
628 (td)->runstate = TD_EXITED; \
629 (td)->error = EFAULT; \
634 static inline int fio_fill_issue_time(struct thread_data *td)
636 if (td->o.read_iolog_file ||
637 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
644 * Cheesy number->string conversion, complete with carry rounding error.
646 static inline char *num2str(unsigned long num, int maxlen, int base, int pow2)
648 char postfix[] = { ' ', 'K', 'M', 'G', 'P', 'E' };
649 unsigned int thousand;
660 for (i = 0; base > 1; i++)
666 len = sprintf(buf, "%'lu", num);
669 buf[len] = postfix[i];
675 if ((num % thousand) >= (thousand / 2))
686 static inline int __should_check_rate(struct thread_data *td,
689 struct thread_options *o = &td->o;
692 * If some rate setting was given, we need to check it
694 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
695 o->rate_iops_min[ddir])
701 static inline int should_check_rate(struct thread_data *td,
702 unsigned long *bytes_done)
707 ret |= __should_check_rate(td, 0);
709 ret |= __should_check_rate(td, 1);
714 static inline int is_power_of_2(unsigned int val)
716 return (val != 0 && ((val & (val - 1)) == 0));
720 * We currently only need to do locking if we have verifier threads
721 * accessing our internal structures too
723 static inline void td_io_u_lock(struct thread_data *td)
725 if (td->o.verify_async)
726 pthread_mutex_lock(&td->io_u_lock);
729 static inline void td_io_u_unlock(struct thread_data *td)
731 if (td->o.verify_async)
732 pthread_mutex_unlock(&td->io_u_lock);
735 static inline void td_io_u_free_notify(struct thread_data *td)
737 if (td->o.verify_async)
738 pthread_cond_signal(&td->free_cond);