X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=fio.c;h=ace8c5ee09aae9e4353d9879273eaadbe1c2848d;hp=53735c6c8801c3f47568eef3fded46a136e27f2b;hb=0032bf9fdccc4491e779a0c581c9bcb714b012f7;hpb=8e9fe637622aab409101c90d26670375a50ff4f3 diff --git a/fio.c b/fio.c index 53735c6c..ace8c5ee 100644 --- a/fio.c +++ b/fio.c @@ -26,33 +26,39 @@ #include #include #include -#include +#include #include #include #include #include -#include #include #include "fio.h" #include "os.h" -#define MASK (4095) - -#define ALIGN(buf) (char *) (((unsigned long) (buf) + MASK) & ~(MASK)) +static unsigned long page_mask; +#define ALIGN(buf) \ + (char *) (((unsigned long) (buf) + page_mask) & ~page_mask) int groupid = 0; int thread_number = 0; int shm_id = 0; int temp_stall_ts; -char *fio_inst_prefix = _INST_PREFIX; static volatile int startup_sem; +static volatile int fio_abort; + +struct io_log *agg_io_log[2]; #define TERMINATE_ALL (-1) #define JOB_START_TIMEOUT (5 * 1000) -static void terminate_threads(int group_id) +static inline void td_set_runstate(struct thread_data *td, int runstate) +{ + td->runstate = runstate; +} + +static void terminate_threads(int group_id, int forced_kill) { struct thread_data *td; int i; @@ -61,6 +67,8 @@ static void terminate_threads(int group_id) if (group_id == TERMINATE_ALL || groupid == td->groupid) { td->terminate = 1; td->start_delay = 0; + if (forced_kill) + td_set_runstate(td, TD_EXITED); } } } @@ -74,9 +82,9 @@ static void sig_handler(int sig) print_thread_status(); break; default: - printf("\nfio: terminating on signal\n"); + printf("\nfio: terminating on signal %d\n", sig); fflush(stdout); - terminate_threads(TERMINATE_ALL); + terminate_threads(TERMINATE_ALL, 0); break; } } @@ -106,7 +114,7 @@ static int check_min_rate(struct thread_data *td, struct timeval *now) rate = (td->this_io_bytes[ddir] - td->rate_bytes) / spent; if (rate < td->ratemin) { - fprintf(f_out, "%s: min rate %d not met, got %ldKiB/sec\n", td->name, td->ratemin, rate); + fprintf(f_out, "%s: min rate %u not met, got %luKiB/sec\n", td->name, td->ratemin, rate); return 1; } } @@ -126,11 +134,6 @@ static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) return 0; } -static inline void td_set_runstate(struct thread_data *td, int runstate) -{ - td->runstate = runstate; -} - static struct fio_file *get_next_file(struct thread_data *td) { unsigned int old_next_file = td->next_file; @@ -218,14 +221,14 @@ static int fio_io_sync(struct thread_data *td, struct fio_file *f) ret = td_io_queue(td, io_u); if (ret) { + td_verror(td, io_u->error); put_io_u(td, io_u); - td_verror(td, ret); return 1; } ret = td_io_getevents(td, 1, td->cur_depth, NULL); if (ret < 0) { - td_verror(td, -ret); + td_verror(td, ret); return 1; } @@ -243,9 +246,8 @@ static int fio_io_sync(struct thread_data *td, struct fio_file *f) * The main verify engine. Runs over the writes we previusly submitted, * reads the blocks back in, and checks the crc/md5 of the data. */ -void do_verify(struct thread_data *td) +static void do_verify(struct thread_data *td) { - struct timeval t; struct io_u *io_u, *v_io_u = NULL; struct io_completion_data icd; struct fio_file *f; @@ -266,14 +268,15 @@ void do_verify(struct thread_data *td) if (td->terminate) break; - gettimeofday(&t, NULL); - if (runtime_exceeded(td, &t)) - break; - io_u = __get_io_u(td); if (!io_u) break; + if (runtime_exceeded(td, &io_u->start_time)) { + put_io_u(td, io_u); + break; + } + if (get_next_verify(td, io_u)) { put_io_u(td, io_u); break; @@ -292,8 +295,8 @@ void do_verify(struct thread_data *td) ret = td_io_queue(td, io_u); if (ret) { + td_verror(td, io_u->error); put_io_u(td, io_u); - td_verror(td, ret); break; } @@ -314,6 +317,7 @@ void do_verify(struct thread_data *td) v_io_u = td->io_ops->event(td, 0); icd.nr = 1; icd.error = 0; + fio_gettime(&icd.time, NULL); io_completed(td, v_io_u, &icd); if (icd.error) { @@ -350,7 +354,7 @@ static void do_cpuio(struct thread_data *td) int i = 0; while (!td->terminate) { - gettimeofday(&e, NULL); + fio_gettime(&e, NULL); if (runtime_exceeded(td, &e)) break; @@ -371,15 +375,14 @@ static void do_cpuio(struct thread_data *td) static void do_io(struct thread_data *td) { struct io_completion_data icd; - struct timeval s, e; + struct timeval s; unsigned long usec; struct fio_file *f; int i, ret = 0; td_set_runstate(td, TD_RUNNING); - while (td->this_io_bytes[td->ddir] < td->io_size) { - struct timespec ts = { .tv_sec = 0, .tv_nsec = 0}; + while ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->io_size) { struct timespec *timeout; int min_evts = 0; struct io_u *io_u; @@ -397,16 +400,28 @@ static void do_io(struct thread_data *td) memcpy(&s, &io_u->start_time, sizeof(s)); +requeue: ret = td_io_queue(td, io_u); if (ret) { - put_io_u(td, io_u); - td_verror(td, ret); - break; + if (ret > 0 && (io_u->xfer_buflen != io_u->resid) && + io_u->resid) { + /* + * short read/write. requeue. + */ + io_u->xfer_buflen = io_u->resid; + io_u->xfer_buf += ret; + goto requeue; + } else { + put_io_u(td, io_u); + break; + } } add_slat_sample(td, io_u->ddir, mtime_since(&io_u->start_time, &io_u->issue_time)); if (td->cur_depth < td->iodepth) { + struct timespec ts = { .tv_sec = 0, .tv_nsec = 0}; + timeout = &ts; min_evts = 0; } else { @@ -433,26 +448,30 @@ static void do_io(struct thread_data *td) * of completions except the very first one which may look * a little bursty */ - gettimeofday(&e, NULL); - usec = utime_since(&s, &e); + usec = utime_since(&s, &icd.time); - rate_throttle(td, usec, icd.bytes_done[td->ddir]); + rate_throttle(td, usec, icd.bytes_done[td->ddir], td->ddir); - if (check_min_rate(td, &e)) { - if (rate_quit) - terminate_threads(td->groupid); - td_verror(td, ENOMEM); + if (check_min_rate(td, &icd.time)) { + if (exitall_on_terminate) + terminate_threads(td->groupid, 0); + td_verror(td, ENODATA); break; } - if (runtime_exceeded(td, &e)) + if (runtime_exceeded(td, &icd.time)) break; - if (td->thinktime) - usec_sleep(td, td->thinktime); + if (td->thinktime) { + unsigned long long b; + + b = td->io_blocks[0] + td->io_blocks[1]; + if (!(b % td->thinktime_blocks)) + usec_sleep(td, td->thinktime); + } } - if (!ret) { + if (!td->error) { if (td->cur_depth) cleanup_pending_aio(td); @@ -479,9 +498,23 @@ static void cleanup_io_u(struct thread_data *td) free_io_mem(td); } +/* + * "randomly" fill the buffer contents + */ +static void fill_rand_buf(struct io_u *io_u, int max_bs) +{ + int *ptr = io_u->buf; + + while ((void *) ptr - io_u->buf < max_bs) { + *ptr = rand() * 0x9e370001; + ptr++; + } +} + static int init_io_u(struct thread_data *td) { struct io_u *io_u; + unsigned int max_bs; int i, max_units; char *p; @@ -493,7 +526,13 @@ static int init_io_u(struct thread_data *td) else max_units = td->iodepth; - td->orig_buffer_size = td->max_bs * max_units + MASK; + max_bs = max(td->max_bs[DDIR_READ], td->max_bs[DDIR_WRITE]); + td->orig_buffer_size = max_bs * max_units; + + if (td->mem_type == MEM_SHMHUGE || td->mem_type == MEM_MMAPHUGE) + td->orig_buffer_size = (td->orig_buffer_size + td->hugepage_size - 1) & ~(td->hugepage_size - 1); + else + td->orig_buffer_size += page_mask; if (allocate_io_mem(td)) return 1; @@ -504,7 +543,10 @@ static int init_io_u(struct thread_data *td) memset(io_u, 0, sizeof(*io_u)); INIT_LIST_HEAD(&io_u->list); - io_u->buf = p + td->max_bs * i; + io_u->buf = p + max_bs * i; + if (td_write(td) || td_rw(td)) + fill_rand_buf(io_u, max_bs); + io_u->index = i; list_add(&io_u->list, &td->io_u_freelist); } @@ -518,6 +560,9 @@ static int switch_ioscheduler(struct thread_data *td) FILE *f; int ret; + if (td->io_ops->flags & FIO_CPUIO) + return 0; + sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); f = fopen(tmp, "r+"); @@ -585,6 +630,7 @@ static void clear_io_state(struct thread_data *td) */ static void *thread_main(void *data) { + unsigned long long runtime[2]; struct thread_data *td = data; if (!td->use_thread) @@ -605,9 +651,6 @@ static void *thread_main(void *data) goto err; } - if (td_io_init(td)) - goto err; - if (init_iolog(td)) goto err; @@ -635,15 +678,27 @@ static void *thread_main(void *data) if (!td->create_serialize && setup_files(td)) goto err; + if (open_files(td)) + goto err; + + /* + * Do this late, as some IO engines would like to have the + * files setup prior to initializing structures. + */ + if (td_io_init(td)) + goto err; - gettimeofday(&td->epoch, NULL); + if (td->exec_prerun) { + if (system(td->exec_prerun) < 0) + goto err; + } - if (td->exec_prerun) - system(td->exec_prerun); + fio_gettime(&td->epoch, NULL); + getrusage(RUSAGE_SELF, &td->ru_start); + runtime[0] = runtime[1] = 0; while (td->loops--) { - getrusage(RUSAGE_SELF, &td->ru_start); - gettimeofday(&td->start, NULL); + fio_gettime(&td->start, NULL); memcpy(&td->stat_sample_time, &td->start, sizeof(td->start)); if (td->ratemin) @@ -657,11 +712,9 @@ static void *thread_main(void *data) else do_io(td); - td->runtime[td->ddir] += mtime_since_now(&td->start); + runtime[td->ddir] += utime_since_now(&td->start); if (td_rw(td) && td->io_bytes[td->ddir ^ 1]) - td->runtime[td->ddir ^ 1] = td->runtime[td->ddir]; - - update_rusage_stat(td); + runtime[td->ddir ^ 1] = runtime[td->ddir]; if (td->error || td->terminate) break; @@ -670,29 +723,36 @@ static void *thread_main(void *data) continue; clear_io_state(td); - gettimeofday(&td->start, NULL); + fio_gettime(&td->start, NULL); do_verify(td); - td->runtime[DDIR_READ] += mtime_since_now(&td->start); + runtime[DDIR_READ] += utime_since_now(&td->start); if (td->error || td->terminate) break; } + update_rusage_stat(td); + fio_gettime(&td->end_time, NULL); + td->runtime[0] = runtime[0] / 1000; + td->runtime[1] = runtime[1] / 1000; + if (td->bw_log) finish_log(td, td->bw_log, "bw"); if (td->slat_log) finish_log(td, td->slat_log, "slat"); if (td->clat_log) finish_log(td, td->clat_log, "clat"); - if (td->write_iolog) + if (td->write_iolog_file) write_iolog_close(td); - if (td->exec_postrun) - system(td->exec_postrun); + if (td->exec_postrun) { + if (system(td->exec_postrun) < 0) + log_err("fio: postrun %s failed\n", td->exec_postrun); + } if (exitall_on_terminate) - terminate_threads(td->groupid); + terminate_threads(td->groupid, 0); err: close_files(td); @@ -730,12 +790,12 @@ static void *fork_main(int shmid, int offset) static void reap_threads(int *nr_running, int *t_rate, int *m_rate) { struct thread_data *td; - int i, cputhreads; + int i, cputhreads, pending; /* * reap exited threads (TD_EXITED -> TD_REAPED) */ - cputhreads = 0; + pending = cputhreads = 0; for_each_td(td, i) { /* * ->io_ops is NULL for a thread that has closed its @@ -744,8 +804,12 @@ static void reap_threads(int *nr_running, int *t_rate, int *m_rate) if (td->io_ops && td->io_ops->flags & FIO_CPUIO) cputhreads++; - if (td->runstate != TD_EXITED) + if (td->runstate != TD_EXITED) { + if (td->runstate < TD_RUNNING) + pending++; + continue; + } td_set_runstate(td, TD_REAPED); @@ -762,8 +826,8 @@ static void reap_threads(int *nr_running, int *t_rate, int *m_rate) (*t_rate) -= td->rate; } - if (*nr_running == cputhreads) - terminate_threads(TERMINATE_ALL); + if (*nr_running == cputhreads && !pending) + terminate_threads(TERMINATE_ALL, 0); } /* @@ -794,10 +858,10 @@ static void run_threads(void) for_each_td(td, i) { print_status_init(td->thread_number - 1); - init_disk_util(td); - - if (!td->create_serialize) + if (!td->create_serialize) { + init_disk_util(td); continue; + } /* * do file setup here so it happens sequentially, @@ -808,9 +872,9 @@ static void run_threads(void) td_set_runstate(td, TD_REAPED); todo--; } - } - time_init(); + init_disk_util(td); + } while (todo) { struct thread_data *map[MAX_JOBS]; @@ -871,9 +935,9 @@ static void run_threads(void) * Wait for the started threads to transition to * TD_INITIALIZED. */ - gettimeofday(&this_start, NULL); + fio_gettime(&this_start, NULL); left = this_jobs; - while (left) { + while (left && !fio_abort) { if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) break; @@ -939,6 +1003,15 @@ static void run_threads(void) int main(int argc, char *argv[]) { + long ps; + + /* + * We need locale for number printing, if it isn't set then just + * go with the US format. + */ + if (!getenv("LC_NUMERIC")) + setlocale(LC_NUMERIC, "en_US"); + if (parse_options(argc, argv)) return 1; @@ -947,10 +1020,30 @@ int main(int argc, char *argv[]) return 1; } + ps = sysconf(_SC_PAGESIZE); + if (ps < 0) { + log_err("Failed to get page size\n"); + return 1; + } + + page_mask = ps - 1; + + if (write_bw_log) { + setup_log(&agg_io_log[DDIR_READ]); + setup_log(&agg_io_log[DDIR_WRITE]); + } + disk_util_timer_arm(); run_threads(); - show_run_stats(); + + if (!fio_abort) { + show_run_stats(); + if (write_bw_log) { + __finish_log(agg_io_log[DDIR_READ],"agg-read_bw.log"); + __finish_log(agg_io_log[DDIR_WRITE],"agg-write_bw.log"); + } + } return 0; }