X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=backend.c;h=36bde6a587535635ee9d1dcef6c0bc0c5160f772;hp=d98e5fe4e8f8040b41a14a153ed40085756c7181;hb=dcf6ad384149ee0b3f91c5a8127160cc291f7157;hpb=f8c48f59773d0a26526472dbee116a4772aec930 diff --git a/backend.c b/backend.c index d98e5fe4..36bde6a5 100644 --- a/backend.c +++ b/backend.c @@ -18,33 +18,21 @@ * * You should have received a copy of the GNU General Public License * along with this program; if not, write to the Free Software - * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. * */ #include -#include #include -#include #include -#include -#include #include -#include #include #include #include -#include -#include #include #include "fio.h" -#ifndef FIO_NO_HAVE_SHM_H -#include -#endif -#include "hash.h" #include "smalloc.h" #include "verify.h" -#include "trim.h" #include "diskutil.h" #include "cgroup.h" #include "profile.h" @@ -58,10 +46,12 @@ #include "lib/mountcheck.h" #include "rate-submit.h" #include "helper_thread.h" +#include "pshared.h" +#include "zone-dist.h" -static struct fio_mutex *startup_mutex; +static struct fio_sem *startup_sem; static struct flist_head *cgroup_list; -static char *cgroup_mnt; +static struct cgroup_mnt *cgroup_mnt; static int exit_value; static volatile int fio_abort; static unsigned int nr_process = 0; @@ -279,7 +269,7 @@ static void cleanup_pending_aio(struct thread_data *td) static bool fio_io_sync(struct thread_data *td, struct fio_file *f) { struct io_u *io_u = __get_io_u(td); - int ret; + enum fio_q_status ret; if (!io_u) return true; @@ -294,16 +284,13 @@ static bool fio_io_sync(struct thread_data *td, struct fio_file *f) requeue: ret = td_io_queue(td, io_u); - if (ret < 0) { - td_verror(td, io_u->error, "td_io_queue"); - put_io_u(td, io_u); - return true; - } else if (ret == FIO_Q_QUEUED) { - if (td_io_commit(td)) - return true; + switch (ret) { + case FIO_Q_QUEUED: + td_io_commit(td); if (io_u_queued_complete(td, 1) < 0) return true; - } else if (ret == FIO_Q_COMPLETED) { + break; + case FIO_Q_COMPLETED: if (io_u->error) { td_verror(td, io_u->error, "td_io_queue"); return true; @@ -311,9 +298,9 @@ requeue: if (io_u_sync_complete(td, io_u) < 0) return true; - } else if (ret == FIO_Q_BUSY) { - if (td_io_commit(td)) - return true; + break; + case FIO_Q_BUSY: + td_io_commit(td); goto requeue; } @@ -426,7 +413,7 @@ static void check_update_rusage(struct thread_data *td) if (td->update_rusage) { td->update_rusage = 0; update_rusage_stat(td); - fio_mutex_up(td->rusage_sem); + fio_sem_up(td->rusage_sem); } } @@ -446,9 +433,7 @@ static int wait_for_completions(struct thread_data *td, struct timespec *time) if ((full && !min_evts) || !td->o.iodepth_batch_complete_min) min_evts = 1; - if (time && (__should_check_rate(td, DDIR_READ) || - __should_check_rate(td, DDIR_WRITE) || - __should_check_rate(td, DDIR_TRIM))) + if (time && __should_check_rate(td)) fio_gettime(time, NULL); do { @@ -464,22 +449,20 @@ int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret, enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify, struct timespec *comp_time) { - int ret2; - switch (*ret) { case FIO_Q_COMPLETED: if (io_u->error) { *ret = -io_u->error; clear_io_u(td, io_u); } else if (io_u->resid) { - int bytes = io_u->xfer_buflen - io_u->resid; + long long bytes = io_u->xfer_buflen - io_u->resid; struct fio_file *f = io_u->file; if (bytes_issued) *bytes_issued += bytes; if (!from_verify) - trim_io_piece(td, io_u); + trim_io_piece(io_u); /* * zero read, fail @@ -505,9 +488,7 @@ int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret, requeue_io_u(td, &io_u); } else { sync_done: - if (comp_time && (__should_check_rate(td, DDIR_READ) || - __should_check_rate(td, DDIR_WRITE) || - __should_check_rate(td, DDIR_TRIM))) + if (comp_time && __should_check_rate(td)) fio_gettime(comp_time, NULL); *ret = io_u_sync_complete(td, io_u); @@ -541,9 +522,7 @@ sync_done: if (!from_verify) unlog_io_piece(td, io_u); requeue_io_u(td, &io_u); - ret2 = td_io_commit(td); - if (ret2 < 0) - *ret = ret2; + td_io_commit(td); break; default: assert(*ret < 0); @@ -605,7 +584,7 @@ static bool in_flight_overlap(struct io_u_queue *q, struct io_u *io_u) if (x1 < y2 && y1 < x2) { overlap = true; - dprint(FD_IO, "in-flight overlap: %llu/%lu, %llu/%lu\n", + dprint(FD_IO, "in-flight overlap: %llu/%llu, %llu/%llu\n", x1, io_u->buflen, y1, check_io_u->buflen); break; @@ -616,7 +595,7 @@ static bool in_flight_overlap(struct io_u_queue *q, struct io_u *io_u) return overlap; } -static int io_u_submit(struct thread_data *td, struct io_u *io_u) +static enum fio_q_status io_u_submit(struct thread_data *td, struct io_u *io_u) { /* * Check for overlap if the user asked us to, and we have @@ -734,6 +713,7 @@ static void do_verify(struct thread_data *td, uint64_t verify_bytes) break; } else if (io_u->ddir == DDIR_WRITE) { io_u->ddir = DDIR_READ; + populate_verify_io_u(td, io_u); break; } else { put_io_u(td, io_u); @@ -844,14 +824,13 @@ static bool io_complete_bytes_exceeded(struct thread_data *td) */ static long long usec_for_io(struct thread_data *td, enum fio_ddir ddir) { - uint64_t secs, remainder, bps, bytes, iops; + uint64_t bps = td->rate_bps[ddir]; assert(!(td->flags & TD_F_CHILD)); - bytes = td->rate_io_issue_bytes[ddir]; - bps = td->rate_bps[ddir]; if (td->o.rate_process == RATE_PROCESS_POISSON) { - uint64_t val; + uint64_t val, iops; + iops = bps / td->o.bs[ddir]; val = (int64_t) (1000000 / iops) * -logf(__rand_0_1(&td->poisson_state[ddir])); @@ -863,14 +842,58 @@ static long long usec_for_io(struct thread_data *td, enum fio_ddir ddir) td->last_usec[ddir] += val; return td->last_usec[ddir]; } else if (bps) { - secs = bytes / bps; - remainder = bytes % bps; + uint64_t bytes = td->rate_io_issue_bytes[ddir]; + uint64_t secs = bytes / bps; + uint64_t remainder = bytes % bps; + return remainder * 1000000 / bps + secs * 1000000; } return 0; } +static void handle_thinktime(struct thread_data *td, enum fio_ddir ddir) +{ + unsigned long long b; + uint64_t total; + int left; + + b = ddir_rw_sum(td->io_blocks); + if (b % td->o.thinktime_blocks) + return; + + io_u_quiesce(td); + + total = 0; + if (td->o.thinktime_spin) + total = usec_spin(td->o.thinktime_spin); + + left = td->o.thinktime - total; + if (left) + total += usec_sleep(td, left); + + /* + * If we're ignoring thinktime for the rate, add the number of bytes + * we would have done while sleeping, minus one block to ensure we + * start issuing immediately after the sleep. + */ + if (total && td->rate_bps[ddir] && td->o.rate_ign_think) { + uint64_t missed = (td->rate_bps[ddir] * total) / 1000000ULL; + uint64_t bs = td->o.min_bs[ddir]; + uint64_t usperop = bs * 1000000ULL / td->rate_bps[ddir]; + uint64_t over; + + if (usperop <= total) + over = bs; + else + over = (usperop - total) / usperop * -bs; + + td->rate_io_issue_bytes[ddir] += (missed - over); + /* adjust for rate_process=poisson */ + td->last_usec[ddir] += total; + } +} + /* * Main IO worker function. It retrieves io_u's to process and queues * and reaps them, checking for rate and errors along the way. @@ -944,8 +967,10 @@ static void do_io(struct thread_data *td, uint64_t *bytes_done) * Break if we exceeded the bytes. The exception is time * based runs, but we still need to break out of the loop * for those to run verification, if enabled. + * Jobs read from iolog do not use this stop condition. */ if (bytes_issued >= total_bytes && + !td->o.read_iolog_file && (!td->o.time_based || (td->o.time_based && td->o.verify != VERIFY_NONE))) break; @@ -955,6 +980,7 @@ static void do_io(struct thread_data *td, uint64_t *bytes_done) int err = PTR_ERR(io_u); io_u = NULL; + ddir = DDIR_INVAL; if (err == -EBUSY) { ret = FIO_Q_BUSY; goto reap; @@ -964,6 +990,9 @@ static void do_io(struct thread_data *td, uint64_t *bytes_done) break; } + if (io_u->ddir == DDIR_WRITE && td->flags & TD_F_DO_VERIFY) + populate_verify_io_u(td, io_u); + ddir = io_u->ddir; /* @@ -1007,8 +1036,8 @@ static void do_io(struct thread_data *td, uint64_t *bytes_done) log_io_piece(td, io_u); if (td->o.io_submit_mode == IO_MODE_OFFLOAD) { - const unsigned long blen = io_u->xfer_buflen; - const enum fio_ddir ddir = acct_ddir(io_u); + const unsigned long long blen = io_u->xfer_buflen; + const enum fio_ddir __ddir = acct_ddir(io_u); if (td->error) break; @@ -1016,14 +1045,14 @@ static void do_io(struct thread_data *td, uint64_t *bytes_done) workqueue_enqueue(&td->io_wq, &io_u->work); ret = FIO_Q_QUEUED; - if (ddir_rw(ddir)) { - td->io_issues[ddir]++; - td->io_issue_bytes[ddir] += blen; - td->rate_io_issue_bytes[ddir] += blen; + if (ddir_rw(__ddir)) { + td->io_issues[__ddir]++; + td->io_issue_bytes[__ddir] += blen; + td->rate_io_issue_bytes[__ddir] += blen; } if (should_check_rate(td)) - td->rate_next_io_time[ddir] = usec_for_io(td, ddir); + td->rate_next_io_time[__ddir] = usec_for_io(td, __ddir); } else { ret = io_u_submit(td, io_u); @@ -1062,23 +1091,8 @@ reap: if (!in_ramp_time(td) && td->o.latency_target) lat_target_check(td); - if (td->o.thinktime) { - unsigned long long b; - - b = ddir_rw_sum(td->io_blocks); - if (!(b % td->o.thinktime_blocks)) { - int left; - - io_u_quiesce(td); - - if (td->o.thinktime_spin) - usec_spin(td->o.thinktime_spin); - - left = td->o.thinktime - td->o.thinktime_spin; - if (left) - usec_sleep(td, left); - } - } + if (ddir_rw(ddir) && td->o.thinktime) + handle_thinktime(td, ddir); } check_update_rusage(td); @@ -1188,9 +1202,74 @@ static void cleanup_io_u(struct thread_data *td) static int init_io_u(struct thread_data *td) { struct io_u *io_u; - unsigned int max_bs, min_write; int cl_align, i, max_units; - int data_xfer = 1, err; + int err; + + max_units = td->o.iodepth; + + err = 0; + err += !io_u_rinit(&td->io_u_requeues, td->o.iodepth); + err += !io_u_qinit(&td->io_u_freelist, td->o.iodepth); + err += !io_u_qinit(&td->io_u_all, td->o.iodepth); + + if (err) { + log_err("fio: failed setting up IO queues\n"); + return 1; + } + + cl_align = os_cache_line_size(); + + for (i = 0; i < max_units; i++) { + void *ptr; + + if (td->terminate) + return 1; + + ptr = fio_memalign(cl_align, sizeof(*io_u)); + if (!ptr) { + log_err("fio: unable to allocate aligned memory\n"); + break; + } + + io_u = ptr; + memset(io_u, 0, sizeof(*io_u)); + INIT_FLIST_HEAD(&io_u->verify_list); + dprint(FD_MEM, "io_u alloc %p, index %u\n", io_u, i); + + io_u->index = i; + io_u->flags = IO_U_F_FREE; + io_u_qpush(&td->io_u_freelist, io_u); + + /* + * io_u never leaves this stack, used for iteration of all + * io_u buffers. + */ + io_u_qpush(&td->io_u_all, io_u); + + if (td->io_ops->io_u_init) { + int ret = td->io_ops->io_u_init(td, io_u); + + if (ret) { + log_err("fio: failed to init engine data: %d\n", ret); + return 1; + } + } + } + + init_io_u_buffers(td); + + if (init_file_completion_logging(td, max_units)) + return 1; + + return 0; +} + +int init_io_u_buffers(struct thread_data *td) +{ + struct io_u *io_u; + unsigned long long max_bs, min_write; + int i, max_units; + int data_xfer = 1; char *p; max_units = td->o.iodepth; @@ -1202,16 +1281,6 @@ static int init_io_u(struct thread_data *td) if (td_ioengine_flagged(td, FIO_NOIO) || !(td_read(td) || td_write(td))) data_xfer = 0; - err = 0; - err += io_u_rinit(&td->io_u_requeues, td->o.iodepth); - err += io_u_qinit(&td->io_u_freelist, td->o.iodepth); - err += io_u_qinit(&td->io_u_all, td->o.iodepth); - - if (err) { - log_err("fio: failed setting up IO queues\n"); - return 1; - } - /* * if we may later need to do address alignment, then add any * possible adjustment here so that we don't cause a buffer @@ -1223,7 +1292,7 @@ static int init_io_u(struct thread_data *td) td->orig_buffer_size += page_mask + td->o.mem_align; if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) { - unsigned long bs; + unsigned long long bs; bs = td->orig_buffer_size + td->o.hugepage_size - 1; td->orig_buffer_size = bs & ~(td->o.hugepage_size - 1); @@ -1243,23 +1312,8 @@ static int init_io_u(struct thread_data *td) else p = td->orig_buffer; - cl_align = os_cache_line_size(); - for (i = 0; i < max_units; i++) { - void *ptr; - - if (td->terminate) - return 1; - - ptr = fio_memalign(cl_align, sizeof(*io_u)); - if (!ptr) { - log_err("fio: unable to allocate aligned memory\n"); - break; - } - - io_u = ptr; - memset(io_u, 0, sizeof(*io_u)); - INIT_FLIST_HEAD(&io_u->verify_list); + io_u = td->io_u_all.io_us[i]; dprint(FD_MEM, "io_u alloc %p, index %u\n", io_u, i); if (data_xfer) { @@ -1276,32 +1330,9 @@ static int init_io_u(struct thread_data *td) fill_verify_pattern(td, io_u->buf, max_bs, io_u, 0, 0); } } - - io_u->index = i; - io_u->flags = IO_U_F_FREE; - io_u_qpush(&td->io_u_freelist, io_u); - - /* - * io_u never leaves this stack, used for iteration of all - * io_u buffers. - */ - io_u_qpush(&td->io_u_all, io_u); - - if (td->io_ops->io_u_init) { - int ret = td->io_ops->io_u_init(td, io_u); - - if (ret) { - log_err("fio: failed to init engine data: %d\n", ret); - return 1; - } - } - p += max_bs; } - if (init_file_completion_logging(td, max_units)) - return 1; - return 0; } @@ -1312,7 +1343,7 @@ static int init_io_u(struct thread_data *td) static int switch_ioscheduler(struct thread_data *td) { #ifdef FIO_HAVE_IOSCHED_SWITCH - char tmp[256], tmp2[128]; + char tmp[256], tmp2[128], *p; FILE *f; int ret; @@ -1348,17 +1379,19 @@ static int switch_ioscheduler(struct thread_data *td) /* * Read back and check that the selected scheduler is now the default. */ - memset(tmp, 0, sizeof(tmp)); - ret = fread(tmp, sizeof(tmp), 1, f); + ret = fread(tmp, 1, sizeof(tmp) - 1, f); if (ferror(f) || ret < 0) { td_verror(td, errno, "fread"); fclose(f); return 1; } + tmp[ret] = '\0'; /* - * either a list of io schedulers or "none\n" is expected. + * either a list of io schedulers or "none\n" is expected. Strip the + * trailing newline. */ - tmp[strlen(tmp) - 1] = '\0'; + p = tmp; + strsep(&p, "\n"); /* * Write to "none" entry doesn't fail, so check the result here. @@ -1516,7 +1549,7 @@ static void *thread_main(void *data) } else td->pid = gettid(); - fio_local_clock_init(o->use_thread); + fio_local_clock_init(); dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid); @@ -1527,7 +1560,6 @@ static void *thread_main(void *data) INIT_FLIST_HEAD(&td->io_hist_list); INIT_FLIST_HEAD(&td->verify_list); INIT_FLIST_HEAD(&td->trim_list); - INIT_FLIST_HEAD(&td->next_rand_list); td->io_hist_tree = RB_ROOT; ret = mutex_cond_init_pshared(&td->io_u_lock, &td->free_cond); @@ -1542,11 +1574,11 @@ static void *thread_main(void *data) } td_set_runstate(td, TD_INITIALIZED); - dprint(FD_MUTEX, "up startup_mutex\n"); - fio_mutex_up(startup_mutex); - dprint(FD_MUTEX, "wait on td->mutex\n"); - fio_mutex_down(td->mutex); - dprint(FD_MUTEX, "done waiting on td->mutex\n"); + dprint(FD_MUTEX, "up startup_sem\n"); + fio_sem_up(startup_sem); + dprint(FD_MUTEX, "wait on td->sem\n"); + fio_sem_down(td->sem); + dprint(FD_MUTEX, "done waiting on td->sem\n"); /* * A new gid requires privilege, so we need to do this before setting @@ -1561,6 +1593,8 @@ static void *thread_main(void *data) goto err; } + td_zone_gen_index(td); + /* * Do this early, we don't want the compress threads to be limited * to the same CPUs as the IO workers. So do this before we set @@ -1656,7 +1690,7 @@ static void *thread_main(void *data) * May alter parameters that init_io_u() will use, so we need to * do this first. */ - if (init_iolog(td)) + if (!init_iolog(td)) goto err; if (init_io_u(td)) @@ -1692,16 +1726,14 @@ static void *thread_main(void *data) if (td_io_init(td)) goto err; - if (init_random_map(td)) + if (!init_random_map(td)) goto err; if (o->exec_prerun && exec_string(o, o->exec_prerun, (const char *)"prerun")) goto err; - if (o->pre_read) { - if (pre_read_files(td) < 0) - goto err; - } + if (o->pre_read && !pre_read_files(td)) + goto err; fio_verify_init(td); @@ -1777,11 +1809,11 @@ static void *thread_main(void *data) deadlock_loop_cnt = 0; do { check_update_rusage(td); - if (!fio_mutex_down_trylock(stat_mutex)) + if (!fio_sem_down_trylock(stat_sem)) break; usleep(1000); if (deadlock_loop_cnt++ > 5000) { - log_err("fio seems to be stuck grabbing stat_mutex, forcibly exiting\n"); + log_err("fio seems to be stuck grabbing stat_sem, forcibly exiting\n"); td->error = EDEADLK; goto err; } @@ -1794,7 +1826,7 @@ static void *thread_main(void *data) if (td_trim(td) && td->io_bytes[DDIR_TRIM]) update_runtime(td, elapsed_us, DDIR_TRIM); fio_gettime(&td->start, NULL); - fio_mutex_up(stat_mutex); + fio_sem_up(stat_sem); if (td->error || td->terminate) break; @@ -1818,10 +1850,10 @@ static void *thread_main(void *data) */ check_update_rusage(td); - fio_mutex_down(stat_mutex); + fio_sem_down(stat_sem); update_runtime(td, elapsed_us, DDIR_READ); fio_gettime(&td->start, NULL); - fio_mutex_up(stat_mutex); + fio_sem_up(stat_sem); if (td->error || td->terminate) break; @@ -1876,17 +1908,9 @@ err: close_and_free_files(td); cleanup_io_u(td); close_ioengine(td); - cgroup_shutdown(td, &cgroup_mnt); + cgroup_shutdown(td, cgroup_mnt); verify_free_state(td); - - if (td->zone_state_index) { - int i; - - for (i = 0; i < DDIR_RWDIR_CNT; i++) - free(td->zone_state_index[i]); - free(td->zone_state_index); - td->zone_state_index = NULL; - } + td_zone_free_index(td); if (fio_option_is_set(o, cpumask)) { ret = fio_cpuset_exit(&o->cpumask); @@ -1899,6 +1923,8 @@ err: */ if (o->write_iolog_file) write_iolog_close(td); + if (td->io_log_rfile) + fclose(td->io_log_rfile); td_set_runstate(td, TD_EXITED); @@ -2292,7 +2318,7 @@ reap: init_disk_util(td); - td->rusage_sem = fio_mutex_init(FIO_MUTEX_LOCKED); + td->rusage_sem = fio_sem_init(FIO_SEM_LOCKED); td->update_rusage = 0; /* @@ -2320,6 +2346,7 @@ reap: nr_started--; break; } + fd = NULL; ret = pthread_detach(td->thread); if (ret) log_err("pthread_detach: %s", @@ -2336,8 +2363,8 @@ reap: } else if (i == fio_debug_jobno) *fio_debug_jobp = pid; } - dprint(FD_MUTEX, "wait on startup_mutex\n"); - if (fio_mutex_down_timeout(startup_mutex, 10000)) { + dprint(FD_MUTEX, "wait on startup_sem\n"); + if (fio_sem_down_timeout(startup_sem, 10000)) { log_err("fio: job startup hung? exiting.\n"); fio_terminate_threads(TERMINATE_ALL); fio_abort = 1; @@ -2345,7 +2372,7 @@ reap: free(fd); break; } - dprint(FD_MUTEX, "done waiting on startup_mutex\n"); + dprint(FD_MUTEX, "done waiting on startup_sem\n"); } /* @@ -2404,7 +2431,7 @@ reap: m_rate += ddir_rw_sum(td->o.ratemin); t_rate += ddir_rw_sum(td->o.rate); todo--; - fio_mutex_up(td->mutex); + fio_sem_up(td->sem); } reap_threads(&nr_running, &t_rate, &m_rate); @@ -2453,16 +2480,17 @@ int fio_backend(struct sk_out *sk_out) setup_log(&agg_io_log[DDIR_TRIM], &p, "agg-trim_bw.log"); } - startup_mutex = fio_mutex_init(FIO_MUTEX_LOCKED); - if (startup_mutex == NULL) + startup_sem = fio_sem_init(FIO_SEM_LOCKED); + if (startup_sem == NULL) return 1; set_genesis_time(); stat_init(); - helper_thread_create(startup_mutex, sk_out); + helper_thread_create(startup_sem, sk_out); cgroup_list = smalloc(sizeof(*cgroup_list)); - INIT_FLIST_HEAD(cgroup_list); + if (cgroup_list) + INIT_FLIST_HEAD(cgroup_list); run_threads(sk_out); @@ -2481,27 +2509,23 @@ int fio_backend(struct sk_out *sk_out) } for_each_td(td, i) { - if (td->ss.dur) { - if (td->ss.iops_data != NULL) { - free(td->ss.iops_data); - free(td->ss.bw_data); - } - } + steadystate_free(td); fio_options_free(td); if (td->rusage_sem) { - fio_mutex_remove(td->rusage_sem); + fio_sem_remove(td->rusage_sem); td->rusage_sem = NULL; } - fio_mutex_remove(td->mutex); - td->mutex = NULL; + fio_sem_remove(td->sem); + td->sem = NULL; } free_disk_util(); - cgroup_kill(cgroup_list); - sfree(cgroup_list); - sfree(cgroup_mnt); + if (cgroup_list) { + cgroup_kill(cgroup_list); + sfree(cgroup_list); + } - fio_mutex_remove(startup_mutex); + fio_sem_remove(startup_sem); stat_exit(); return exit_value; }