X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=fio.c;h=9c970333bb4090644fb32fec4ef9010a45214628;hp=5432e383d439319eb918b3dd4162c2df774b0337;hb=2eaa41cee91925db713b19ea2975dcc511648b48;hpb=e5b401d4bf67a1704f28872d3abe09eaf65cdabe diff --git a/fio.c b/fio.c index 5432e383..9c970333 100644 --- a/fio.c +++ b/fio.c @@ -4,10 +4,12 @@ * Copyright (C) 2005 Jens Axboe * Copyright (C) 2006 Jens Axboe * + * The license below covers all files distributed with fio unless otherwise + * noted in the file itself. + * * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 2 of the License, or - * (at your option) any later version. + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. * * This program is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of @@ -24,46 +26,49 @@ #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; - -extern unsigned long long mlock_size; - -#define should_fsync(td) ((td_write(td) || td_rw(td)) && (!(td)->odirect || (td)->override_sync)) 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) { - int i; + td->runstate = runstate; +} - for (i = 0; i < thread_number; i++) { - struct thread_data *td = &threads[i]; +static void terminate_threads(int group_id, int forced_kill) +{ + struct thread_data *td; + int i; + for_each_td(td, i) { if (group_id == TERMINATE_ALL || groupid == td->groupid) { td->terminate = 1; td->start_delay = 0; + if (forced_kill) + td_set_runstate(td, TD_EXITED); } } } @@ -77,141 +82,13 @@ 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; } } -/* - * The ->file_map[] contains a map of blocks we have or have not done io - * to yet. Used to make sure we cover the entire range in a fair fashion. - */ -static int random_map_free(struct thread_data *td, struct fio_file *f, - unsigned long long block) -{ - unsigned int idx = RAND_MAP_IDX(td, f, block); - unsigned int bit = RAND_MAP_BIT(td, f, block); - - return (f->file_map[idx] & (1UL << bit)) == 0; -} - -/* - * Return the next free block in the map. - */ -static int get_next_free_block(struct thread_data *td, struct fio_file *f, - unsigned long long *b) -{ - int i; - - *b = 0; - i = 0; - while ((*b) * td->min_bs < f->file_size) { - if (f->file_map[i] != -1UL) { - *b += ffz(f->file_map[i]); - return 0; - } - - *b += BLOCKS_PER_MAP; - i++; - } - - return 1; -} - -/* - * Mark a given offset as used in the map. - */ -static void mark_random_map(struct thread_data *td, struct fio_file *f, - struct io_u *io_u) -{ - unsigned long long block = io_u->offset / (unsigned long long) td->min_bs; - unsigned int blocks = 0; - - while (blocks < (io_u->buflen / td->min_bs)) { - unsigned int idx, bit; - - if (!random_map_free(td, f, block)) - break; - - idx = RAND_MAP_IDX(td, f, block); - bit = RAND_MAP_BIT(td, f, block); - - assert(idx < f->num_maps); - - f->file_map[idx] |= (1UL << bit); - block++; - blocks++; - } - - if ((blocks * td->min_bs) < io_u->buflen) - io_u->buflen = blocks * td->min_bs; -} - -/* - * For random io, generate a random new block and see if it's used. Repeat - * until we find a free one. For sequential io, just return the end of - * the last io issued. - */ -static int get_next_offset(struct thread_data *td, struct fio_file *f, - unsigned long long *offset) -{ - unsigned long long b, rb; - long r; - - if (!td->sequential) { - unsigned long long max_blocks = td->io_size / td->min_bs; - int loops = 50; - - do { - r = os_random_long(&td->random_state); - b = ((max_blocks - 1) * r / (unsigned long long) (RAND_MAX+1.0)); - rb = b + (f->file_offset / td->min_bs); - loops--; - } while (!random_map_free(td, f, rb) && loops); - - if (!loops) { - if (get_next_free_block(td, f, &b)) - return 1; - } - } else - b = f->last_pos / td->min_bs; - - *offset = (b * td->min_bs) + f->file_offset; - if (*offset > f->file_size) - return 1; - - return 0; -} - -static unsigned int get_next_buflen(struct thread_data *td) -{ - unsigned int buflen; - long r; - - if (td->min_bs == td->max_bs) - buflen = td->min_bs; - else { - r = os_random_long(&td->bsrange_state); - buflen = (1 + (double) (td->max_bs - 1) * r / (RAND_MAX + 1.0)); - buflen = (buflen + td->min_bs - 1) & ~(td->min_bs - 1); - } - - if (buflen > td->io_size - td->this_io_bytes[td->ddir]) { - /* - * if using direct/raw io, we may not be able to - * shrink the size. so just fail it. - */ - if (td->io_ops->flags & FIO_RAWIO) - return 0; - - buflen = td->io_size - td->this_io_bytes[td->ddir]; - } - - return buflen; -} - /* * Check if we are above the minimum rate given. */ @@ -237,9 +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); - if (rate_quit) - terminate_threads(td->groupid); + fprintf(f_out, "%s: min rate %u not met, got %luKiB/sec\n", td->name, td->ratemin, rate); return 1; } } @@ -259,170 +134,9 @@ static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) return 0; } -/* - * Return the data direction for the next io_u. If the job is a - * mixed read/write workload, check the rwmix cycle and switch if - * necessary. - */ -static int get_rw_ddir(struct thread_data *td) -{ - if (td_rw(td)) { - struct timeval now; - unsigned long elapsed; - - gettimeofday(&now, NULL); - elapsed = mtime_since_now(&td->rwmix_switch); - - /* - * Check if it's time to seed a new data direction. - */ - if (elapsed >= td->rwmixcycle) { - int v; - long r; - - r = os_random_long(&td->rwmix_state); - v = 1 + (int) (100.0 * (r / (RAND_MAX + 1.0))); - if (v < td->rwmixread) - td->rwmix_ddir = DDIR_READ; - else - td->rwmix_ddir = DDIR_WRITE; - memcpy(&td->rwmix_switch, &now, sizeof(now)); - } - return td->rwmix_ddir; - } else if (td_read(td)) - return DDIR_READ; - else - return DDIR_WRITE; -} - -static int td_io_prep(struct thread_data *td, struct io_u *io_u) -{ - if (td->io_ops->prep && td->io_ops->prep(td, io_u)) - return 1; - - return 0; -} - -void put_io_u(struct thread_data *td, struct io_u *io_u) -{ - io_u->file = NULL; - list_del(&io_u->list); - list_add(&io_u->list, &td->io_u_freelist); - td->cur_depth--; -} - -static int fill_io_u(struct thread_data *td, struct fio_file *f, - struct io_u *io_u) -{ - /* - * If using an iolog, grab next piece if any available. - */ - if (td->read_iolog) - return read_iolog_get(td, io_u); - - /* - * No log, let the seq/rand engine retrieve the next position. - */ - if (!get_next_offset(td, f, &io_u->offset)) { - io_u->buflen = get_next_buflen(td); - - if (io_u->buflen) { - io_u->ddir = get_rw_ddir(td); - - /* - * If using a write iolog, store this entry. - */ - if (td->write_iolog) - write_iolog_put(td, io_u); - - io_u->file = f; - return 0; - } - } - - return 1; -} - -#define queue_full(td) list_empty(&(td)->io_u_freelist) - -struct io_u *__get_io_u(struct thread_data *td) -{ - struct io_u *io_u = NULL; - - if (!queue_full(td)) { - io_u = list_entry(td->io_u_freelist.next, struct io_u, list); - - io_u->error = 0; - io_u->resid = 0; - list_del(&io_u->list); - list_add(&io_u->list, &td->io_u_busylist); - td->cur_depth++; - } - - return io_u; -} - -/* - * Return an io_u to be processed. Gets a buflen and offset, sets direction, - * etc. The returned io_u is fully ready to be prepped and submitted. - */ -static struct io_u *get_io_u(struct thread_data *td, struct fio_file *f) -{ - struct io_u *io_u; - - io_u = __get_io_u(td); - if (!io_u) - return NULL; - - if (td->zone_bytes >= td->zone_size) { - td->zone_bytes = 0; - f->last_pos += td->zone_skip; - } - - if (fill_io_u(td, f, io_u)) { - put_io_u(td, io_u); - return NULL; - } - - if (io_u->buflen + io_u->offset > f->file_size) { - if (td->io_ops->flags & FIO_RAWIO) { - put_io_u(td, io_u); - return NULL; - } - - io_u->buflen = f->file_size - io_u->offset; - } - - if (!io_u->buflen) { - put_io_u(td, io_u); - return NULL; - } - - if (!td->read_iolog && !td->sequential) - mark_random_map(td, f, io_u); - - f->last_pos += io_u->buflen; - - if (td->verify != VERIFY_NONE) - populate_verify_io_u(td, io_u); - - if (td_io_prep(td, io_u)) { - put_io_u(td, io_u); - return NULL; - } - - gettimeofday(&io_u->start_time, NULL); - return io_u; -} - -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) { - int old_next_file = td->next_file; + unsigned int old_next_file = td->next_file; struct fio_file *f; do { @@ -441,75 +155,6 @@ static struct fio_file *get_next_file(struct thread_data *td) return f; } -static int td_io_sync(struct thread_data *td, struct fio_file *f) -{ - if (td->io_ops->sync) - return td->io_ops->sync(td, f); - - return 0; -} - -static int td_io_getevents(struct thread_data *td, int min, int max, - struct timespec *t) -{ - return td->io_ops->getevents(td, min, max, t); -} - -static int td_io_queue(struct thread_data *td, struct io_u *io_u) -{ - gettimeofday(&io_u->issue_time, NULL); - - return td->io_ops->queue(td, io_u); -} - -#define iocb_time(iocb) ((unsigned long) (iocb)->data) - -static void io_completed(struct thread_data *td, struct io_u *io_u, - struct io_completion_data *icd) -{ - struct timeval e; - unsigned long msec; - - gettimeofday(&e, NULL); - - if (!io_u->error) { - unsigned int bytes = io_u->buflen - io_u->resid; - const int idx = io_u->ddir; - - td->io_blocks[idx]++; - td->io_bytes[idx] += bytes; - td->zone_bytes += bytes; - td->this_io_bytes[idx] += bytes; - - msec = mtime_since(&io_u->issue_time, &e); - - add_clat_sample(td, idx, msec); - add_bw_sample(td, idx); - - if ((td_rw(td) || td_write(td)) && idx == DDIR_WRITE) - log_io_piece(td, io_u); - - icd->bytes_done[idx] += bytes; - } else - icd->error = io_u->error; -} - -static void ios_completed(struct thread_data *td,struct io_completion_data *icd) -{ - struct io_u *io_u; - int i; - - icd->error = 0; - icd->bytes_done[0] = icd->bytes_done[1] = 0; - - for (i = 0; i < icd->nr; i++) { - io_u = td->io_ops->event(td, i); - - io_completed(td, io_u, icd); - put_io_u(td, io_u); - } -} - /* * When job exits, we can cancel the in-flight IO if we are using async * io. Attempt to do so. @@ -553,13 +198,56 @@ static void cleanup_pending_aio(struct thread_data *td) } } +/* + * Helper to handle the final sync of a file. Works just like the normal + * io path, just does everything sync. + */ +static int fio_io_sync(struct thread_data *td, struct fio_file *f) +{ + struct io_u *io_u = __get_io_u(td); + struct io_completion_data icd; + int ret; + + if (!io_u) + return 1; + + io_u->ddir = DDIR_SYNC; + io_u->file = f; + + if (td_io_prep(td, io_u)) { + put_io_u(td, io_u); + return 1; + } + + ret = td_io_queue(td, io_u); + if (ret) { + td_verror(td, io_u->error); + put_io_u(td, io_u); + return 1; + } + + ret = td_io_getevents(td, 1, td->cur_depth, NULL); + if (ret < 0) { + td_verror(td, ret); + return 1; + } + + icd.nr = ret; + ios_completed(td, &icd); + if (icd.error) { + td_verror(td, icd.error); + return 1; + } + + return 0; +} + /* * 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; @@ -570,7 +258,7 @@ void do_verify(struct thread_data *td) * read from disk. */ for_each_file(td, f, i) { - td_io_sync(td, f); + fio_io_sync(td, f); file_invalidate_cache(td, f); } @@ -580,14 +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; @@ -606,8 +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; } @@ -628,6 +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) { @@ -664,7 +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; @@ -685,17 +375,16 @@ 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; + 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}; struct timespec *timeout; - int ret, min_evts = 0; + int min_evts = 0; struct io_u *io_u; if (td->terminate) @@ -711,16 +400,29 @@ 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 { + td_verror(td, io_u->error); + 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 { @@ -747,45 +449,41 @@ 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)) { - 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; - if (should_fsync(td) && td->fsync_blocks && - (td->io_blocks[DDIR_WRITE] % td->fsync_blocks) == 0) - td_io_sync(td, f); + b = td->io_blocks[0] + td->io_blocks[1]; + if (!(b % td->thinktime_blocks)) + usec_sleep(td, td->thinktime); + } } - if (td->cur_depth) - cleanup_pending_aio(td); + if (!td->error) { + if (td->cur_depth) + cleanup_pending_aio(td); - if (should_fsync(td) && td->end_fsync) { - td_set_runstate(td, TD_FSYNCING); - for_each_file(td, f, i) - td_io_sync(td, f); + if (should_fsync(td) && td->end_fsync) { + td_set_runstate(td, TD_FSYNCING); + for_each_file(td, f, i) + fio_io_sync(td, f); + } } } -static int td_io_init(struct thread_data *td) -{ - if (td->io_ops->init) - return td->io_ops->init(td); - - return 0; -} - static void cleanup_io_u(struct thread_data *td) { struct list_head *entry, *n; @@ -798,24 +496,26 @@ static void cleanup_io_u(struct thread_data *td) free(io_u); } - if (td->mem_type == MEM_MALLOC) - free(td->orig_buffer); - else if (td->mem_type == MEM_SHM) { - struct shmid_ds sbuf; + free_io_mem(td); +} - shmdt(td->orig_buffer); - shmctl(td->shm_id, IPC_RMID, &sbuf); - } else if (td->mem_type == MEM_MMAP) - munmap(td->orig_buffer, td->orig_buffer_size); - else - log_err("Bad memory type %d\n", td->mem_type); +/* + * "randomly" fill the buffer contents + */ +static void fill_rand_buf(struct io_u *io_u, int max_bs) +{ + int *ptr = io_u->buf; - td->orig_buffer = NULL; + 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; @@ -827,34 +527,16 @@ 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_MALLOC) - td->orig_buffer = malloc(td->orig_buffer_size); - else if (td->mem_type == MEM_SHM) { - td->shm_id = shmget(IPC_PRIVATE, td->orig_buffer_size, IPC_CREAT | 0600); - if (td->shm_id < 0) { - td_verror(td, errno); - perror("shmget"); - return 1; - } + 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; - td->orig_buffer = shmat(td->shm_id, NULL, 0); - if (td->orig_buffer == (void *) -1) { - td_verror(td, errno); - perror("shmat"); - td->orig_buffer = NULL; - return 1; - } - } else if (td->mem_type == MEM_MMAP) { - td->orig_buffer = mmap(NULL, td->orig_buffer_size, PROT_READ | PROT_WRITE, MAP_PRIVATE | OS_MAP_ANON, 0, 0); - if (td->orig_buffer == MAP_FAILED) { - td_verror(td, errno); - perror("mmap"); - td->orig_buffer = NULL; - return 1; - } - } + if (allocate_io_mem(td)) + return 1; p = ALIGN(td->orig_buffer); for (i = 0; i < max_units; i++) { @@ -862,7 +544,10 @@ 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); } @@ -876,6 +561,9 @@ 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+"); @@ -943,6 +631,7 @@ 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) @@ -963,9 +652,6 @@ static void *thread_main(void *data) goto err; } - if (td_io_init(td)) - goto err; - if (init_iolog(td)) goto err; @@ -993,15 +679,27 @@ 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) @@ -1015,11 +713,9 @@ 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; @@ -1028,29 +724,36 @@ 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); @@ -1087,19 +790,27 @@ static void *fork_main(int shmid, int offset) */ static void reap_threads(int *nr_running, int *t_rate, int *m_rate) { - int i, cputhreads; + struct thread_data *td; + int i, cputhreads, pending; /* * reap exited threads (TD_EXITED -> TD_REAPED) */ - for (i = 0, cputhreads = 0; i < thread_number; i++) { - struct thread_data *td = &threads[i]; - - if (td->io_ops->flags & FIO_CPUIO) + pending = cputhreads = 0; + for_each_td(td, i) { + /* + * ->io_ops is NULL for a thread that has closed its + * io engine + */ + 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); @@ -1116,50 +827,8 @@ 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); -} - -static void fio_unpin_memory(void *pinned) -{ - if (pinned) { - if (munlock(pinned, mlock_size) < 0) - perror("munlock"); - munmap(pinned, mlock_size); - } -} - -static void *fio_pin_memory(void) -{ - unsigned long long phys_mem; - void *ptr; - - if (!mlock_size) - return NULL; - - /* - * Don't allow mlock of more than real_mem-128MB - */ - phys_mem = os_phys_mem(); - if (phys_mem) { - if ((mlock_size + 128 * 1024 * 1024) > phys_mem) { - mlock_size = phys_mem - 128 * 1024 * 1024; - fprintf(f_out, "fio: limiting mlocked memory to %lluMiB\n", mlock_size >> 20); - } - } - - ptr = mmap(NULL, mlock_size, PROT_READ | PROT_WRITE, MAP_PRIVATE | OS_MAP_ANON, 0, 0); - if (!ptr) { - perror("malloc locked mem"); - return NULL; - } - if (mlock(ptr, mlock_size) < 0) { - munmap(ptr, mlock_size); - perror("mlock"); - return NULL; - } - - return ptr; + if (*nr_running == cputhreads && !pending) + terminate_threads(TERMINATE_ALL, 0); } /* @@ -1170,9 +839,9 @@ static void run_threads(void) struct thread_data *td; unsigned long spent; int i, todo, nr_running, m_rate, t_rate, nr_started; - void *mlocked_mem; - mlocked_mem = fio_pin_memory(); + if (fio_pin_memory()) + return; if (!terse_output) { printf("Starting %d thread%s\n", thread_number, thread_number > 1 ? "s" : ""); @@ -1187,15 +856,13 @@ static void run_threads(void) nr_started = 0; m_rate = t_rate = 0; - for (i = 0; i < thread_number; i++) { - td = &threads[i]; - + 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, @@ -1206,9 +873,9 @@ 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]; @@ -1218,9 +885,7 @@ static void run_threads(void) /* * create threads (TD_NOT_CREATED -> TD_CREATED) */ - for (i = 0; i < thread_number; i++) { - td = &threads[i]; - + for_each_td(td, i) { if (td->runstate != TD_NOT_CREATED) continue; @@ -1271,9 +936,9 @@ 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; @@ -1309,9 +974,7 @@ static void run_threads(void) /* * start created threads (TD_INITIALIZED -> TD_RUNNING). */ - for (i = 0; i < thread_number; i++) { - td = &threads[i]; - + for_each_td(td, i) { if (td->runstate != TD_INITIALIZED) continue; @@ -1336,11 +999,20 @@ static void run_threads(void) } update_io_ticks(); - fio_unpin_memory(mlocked_mem); + fio_unpin_memory(); } 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; @@ -1349,10 +1021,30 @@ 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; }