X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=backend.c;h=e025dbffe01b4f6a667f2b7d561603d1664eef06;hp=4e3a3ed5701f8227cabddf3bf3af0a6b0750ee9a;hb=0dcebdf4a70ef0d8144b8fcba763ae87e7fc74b5;hpb=2f694507513e44e59139f57f28e08c5d0772880e diff --git a/backend.c b/backend.c index 4e3a3ed5..e025dbff 100644 --- a/backend.c +++ b/backend.c @@ -34,7 +34,9 @@ #include #include #include +#ifndef FIO_NO_HAVE_SHM_H #include +#endif #include #include "fio.h" @@ -62,6 +64,7 @@ struct io_log *agg_io_log[DDIR_RWDIR_CNT]; int groupid = 0; unsigned int thread_number = 0; +unsigned int stat_number = 0; unsigned int nr_process = 0; unsigned int nr_thread = 0; int shm_id = 0; @@ -213,7 +216,7 @@ static int __check_min_rate(struct thread_data *td, struct timeval *now, } static int check_min_rate(struct thread_data *td, struct timeval *now, - unsigned long *bytes_done) + uint64_t *bytes_done) { int ret = 0; @@ -390,8 +393,9 @@ static int break_on_this_error(struct thread_data *td, enum fio_ddir ddir, * The main verify engine. Runs over the writes we previously submitted, * reads the blocks back in, and checks the crc/md5 of the data. */ -static void do_verify(struct thread_data *td) +static void do_verify(struct thread_data *td, uint64_t verify_bytes) { + uint64_t bytes_done[DDIR_RWDIR_CNT] = { 0, 0, 0 }; struct fio_file *f; struct io_u *io_u; int ret, min_events; @@ -419,6 +423,7 @@ static void do_verify(struct thread_data *td) io_u = NULL; while (!td->terminate) { + enum fio_ddir ddir; int ret2, full; update_tv_cache(td); @@ -434,18 +439,53 @@ static void do_verify(struct thread_data *td) if (flow_threshold_exceeded(td)) continue; - io_u = __get_io_u(td); - if (!io_u) - break; + if (!td->o.experimental_verify) { + io_u = __get_io_u(td); + if (!io_u) + break; - if (get_next_verify(td, io_u)) { - put_io_u(td, io_u); - break; - } + if (get_next_verify(td, io_u)) { + put_io_u(td, io_u); + break; + } - if (td_io_prep(td, io_u)) { - put_io_u(td, io_u); - break; + if (td_io_prep(td, io_u)) { + put_io_u(td, io_u); + break; + } + } else { + if (ddir_rw_sum(bytes_done) + td->o.rw_min_bs > verify_bytes) + break; + + while ((io_u = get_io_u(td)) != NULL) { + /* + * We are only interested in the places where + * we wrote or trimmed IOs. Turn those into + * reads for verification purposes. + */ + if (io_u->ddir == DDIR_READ) { + /* + * Pretend we issued it for rwmix + * accounting + */ + td->io_issues[DDIR_READ]++; + put_io_u(td, io_u); + continue; + } else if (io_u->ddir == DDIR_TRIM) { + io_u->ddir = DDIR_READ; + io_u->flags |= IO_U_F_TRIMMED; + break; + } else if (io_u->ddir == DDIR_WRITE) { + io_u->ddir = DDIR_READ; + break; + } else { + put_io_u(td, io_u); + continue; + } + } + + if (!io_u) + break; } if (td->o.verify_async) @@ -453,6 +493,8 @@ static void do_verify(struct thread_data *td) else io_u->end_io = verify_io_u; + ddir = io_u->ddir; + ret = td_io_queue(td, io_u); switch (ret) { case FIO_Q_COMPLETED: @@ -485,7 +527,7 @@ static void do_verify(struct thread_data *td) requeue_io_u(td, &io_u); } else { sync_done: - ret = io_u_sync_complete(td, io_u, NULL); + ret = io_u_sync_complete(td, io_u, bytes_done); if (ret < 0) break; } @@ -504,7 +546,7 @@ sync_done: break; } - if (break_on_this_error(td, io_u->ddir, &ret)) + if (break_on_this_error(td, ddir, &ret)) break; /* @@ -528,7 +570,7 @@ sync_done: * and do the verification on them through * the callback handler */ - if (io_u_queued_complete(td, min_events, NULL) < 0) { + if (io_u_queued_complete(td, min_events, bytes_done) < 0) { ret = -1; break; } @@ -570,9 +612,12 @@ static int io_bytes_exceeded(struct thread_data *td) /* * Main IO worker function. It retrieves io_u's to process and queues * and reaps them, checking for rate and errors along the way. + * + * Returns number of bytes written and trimmed. */ -static void do_io(struct thread_data *td) +static uint64_t do_io(struct thread_data *td) { + uint64_t bytes_done[DDIR_RWDIR_CNT] = { 0, 0, 0 }; unsigned int i; int ret = 0; @@ -585,13 +630,12 @@ static void do_io(struct thread_data *td) (!flist_empty(&td->trim_list)) || !io_bytes_exceeded(td) || td->o.time_based) { struct timeval comp_time; - unsigned long bytes_done[DDIR_RWDIR_CNT] = { 0, 0, 0 }; int min_evts = 0; struct io_u *io_u; int ret2, full; enum fio_ddir ddir; - if (td->terminate) + if (td->terminate || td->done) break; update_tv_cache(td); @@ -726,7 +770,7 @@ sync_done: if (ret < 0) break; - if (!ddir_rw_sum(bytes_done)) + if (!ddir_rw_sum(bytes_done) && !(td->io_ops->flags & FIO_NOIO)) continue; if (!in_ramp_time(td) && should_check_rate(td, bytes_done)) { @@ -789,6 +833,8 @@ sync_done: */ if (!ddir_rw_sum(td->this_io_bytes)) td->done = 1; + + return bytes_done[DDIR_WRITE] + bytes_done[DDIR_TRIM]; } static void cleanup_io_u(struct thread_data *td) @@ -800,6 +846,10 @@ static void cleanup_io_u(struct thread_data *td) io_u = flist_entry(entry, struct io_u, list); flist_del(&io_u->list); + + if (td->io_ops->io_u_free) + td->io_ops->io_u_free(td, io_u); + fio_memfree(io_u, sizeof(*io_u)); } @@ -882,6 +932,16 @@ static int init_io_u(struct thread_data *td) io_u->index = i; io_u->flags = IO_U_F_FREE; flist_add(&io_u->list, &td->io_u_freelist); + + 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; } @@ -994,6 +1054,8 @@ static void *thread_main(void *data) } else td->pid = gettid(); + fio_local_clock_init(td->o.use_thread); + dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid); INIT_FLIST_HEAD(&td->io_u_freelist); @@ -1003,6 +1065,7 @@ 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); pthread_mutex_init(&td->io_u_lock, NULL); td->io_hist_tree = RB_ROOT; @@ -1052,6 +1115,49 @@ static void *thread_main(void *data) goto err; } +#ifdef CONFIG_LIBNUMA + /* numa node setup */ + if (td->o.numa_cpumask_set || td->o.numa_memmask_set) { + int ret; + + if (numa_available() < 0) { + td_verror(td, errno, "Does not support NUMA API\n"); + goto err; + } + + if (td->o.numa_cpumask_set) { + ret = numa_run_on_node_mask(td->o.numa_cpunodesmask); + if (ret == -1) { + td_verror(td, errno, \ + "numa_run_on_node_mask failed\n"); + goto err; + } + } + + if (td->o.numa_memmask_set) { + + switch (td->o.numa_mem_mode) { + case MPOL_INTERLEAVE: + numa_set_interleave_mask(td->o.numa_memnodesmask); + break; + case MPOL_BIND: + numa_set_membind(td->o.numa_memnodesmask); + break; + case MPOL_LOCAL: + numa_set_localalloc(); + break; + case MPOL_PREFERRED: + numa_set_preferred(td->o.numa_mem_prefer_node); + break; + case MPOL_DEFAULT: + default: + break; + } + + } + } +#endif + /* * May alter parameters that init_io_u() will use, so we need to * do this first. @@ -1108,6 +1214,8 @@ static void *thread_main(void *data) clear_state = 0; while (keep_running(td)) { + uint64_t verify_bytes; + fio_gettime(&td->start, NULL); memcpy(&td->bw_sample_time, &td->start, sizeof(td->start)); memcpy(&td->iops_sample_time, &td->start, sizeof(td->start)); @@ -1128,7 +1236,7 @@ static void *thread_main(void *data) prune_io_piece_log(td); - do_io(td); + verify_bytes = do_io(td); clear_state = 1; @@ -1157,7 +1265,7 @@ static void *thread_main(void *data) fio_gettime(&td->start, NULL); - do_verify(td); + do_verify(td, verify_bytes); td->ts.runtime[DDIR_READ] += utime_since_now(&td->start); @@ -1227,8 +1335,8 @@ err: verify_async_exit(td); close_and_free_files(td); - close_ioengine(td); cleanup_io_u(td); + close_ioengine(td); cgroup_shutdown(td, &cgroup_mnt); if (td->o.cpumask_set) { @@ -1339,7 +1447,7 @@ static void reap_threads(unsigned int *nr_running, unsigned int *t_rate, if (WIFSIGNALED(status)) { int sig = WTERMSIG(status); - if (sig != SIGTERM) + if (sig != SIGTERM && sig != SIGUSR2) log_err("fio: pid=%d, got signal=%d\n", (int) td->pid, sig); td->sig = sig;