X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=backend.c;h=fcb74dcbcfc3d39b7916c3c1f2f8fcb43127e889;hp=225d8a36ec66a427a996a8f14fe48717fba67b6f;hb=c97f1ad6d2a2fb4fe9f3b15e40158aac21e5699e;hpb=5d879392c0e4025e508cce8118f6fc13ee57cc0e diff --git a/backend.c b/backend.c index 225d8a36..fcb74dcb 100644 --- a/backend.c +++ b/backend.c @@ -34,12 +34,12 @@ #include #include #include -#ifndef FIO_NO_HAVE_SHM_H -#include -#endif #include #include "fio.h" +#ifndef FIO_NO_HAVE_SHM_H +#include +#endif #include "hash.h" #include "smalloc.h" #include "verify.h" @@ -50,6 +50,8 @@ #include "lib/rand.h" #include "memalign.h" #include "server.h" +#include "lib/getrusage.h" +#include "idletime.h" static pthread_t disk_util_thread; static struct fio_mutex *disk_thread_mutex; @@ -216,7 +218,7 @@ 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; @@ -320,6 +322,21 @@ requeue: return 0; } +static int fio_file_fsync(struct thread_data *td, struct fio_file *f) +{ + int ret; + + if (fio_file_open(f)) + return fio_io_sync(td, f); + + if (td_io_open_file(td, f)) + return 1; + + ret = fio_io_sync(td, f); + td_io_close_file(td, f); + return ret; +} + static inline void __update_tv_cache(struct thread_data *td) { fio_gettime(&td->tv_cache, NULL); @@ -389,12 +406,22 @@ static int break_on_this_error(struct thread_data *td, enum fio_ddir ddir, return 0; } +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); + } +} + /* * 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; @@ -415,6 +442,8 @@ static void do_verify(struct thread_data *td) break; } + check_update_rusage(td); + if (td->error) return; @@ -422,9 +451,11 @@ 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); + check_update_rusage(td); if (runtime_exceeded(td, &td->tv_cache)) { __update_tv_cache(td); @@ -437,18 +468,53 @@ 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) @@ -456,6 +522,8 @@ 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: @@ -488,7 +556,7 @@ 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; } @@ -507,7 +575,7 @@ sync_done: break; } - if (break_on_this_error(td, io_u->ddir, &ret)) + if (break_on_this_error(td, ddir, &ret)) break; /* @@ -531,7 +599,7 @@ 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; } @@ -541,6 +609,8 @@ sync_done: break; } + check_update_rusage(td); + if (!td->error) { min_events = td->cur_depth; @@ -573,9 +643,12 @@ 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; @@ -588,12 +661,13 @@ 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; + check_update_rusage(td); + if (td->terminate || td->done) break; @@ -758,6 +832,8 @@ sync_done: } } + check_update_rusage(td); + if (td->trim_entries) log_err("fio: %d trim entries leaked?\n", td->trim_entries); @@ -770,7 +846,7 @@ sync_done: i = td->cur_depth; if (i) { - ret = io_u_queued_complete(td, i, NULL); + ret = io_u_queued_complete(td, i, bytes_done); if (td->o.fill_device && td->error == ENOSPC) td->error = 0; } @@ -779,9 +855,11 @@ sync_done: td_set_runstate(td, TD_FSYNCING); for_each_file(td, f, i) { - if (!fio_file_open(f)) + if (!fio_file_fsync(td, f)) continue; - fio_io_sync(td, f); + + log_err("fio: end_fsync failed for file %s\n", + f->file_name); } } } else @@ -792,6 +870,8 @@ 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) @@ -822,8 +902,7 @@ static int init_io_u(struct thread_data *td) char *p; max_units = td->o.iodepth; - max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]); - max_bs = max(td->o.max_bs[DDIR_TRIM], max_bs); + max_bs = td_max_bs(td); min_write = td->o.min_bs[DDIR_WRITE]; td->orig_buffer_size = (unsigned long long) max_bs * (unsigned long long) max_units; @@ -972,8 +1051,19 @@ static int keep_running(struct thread_data *td) return 1; } - if (ddir_rw_sum(td->io_bytes) < td->o.size) + if (td->o.size != -1ULL && ddir_rw_sum(td->io_bytes) < td->o.size) { + uint64_t diff; + + /* + * If the difference is less than the minimum IO size, we + * are done. + */ + diff = td->o.size - ddir_rw_sum(td->io_bytes); + if (diff < td_max_bs(td)) + return 0; + return 1; + } return 0; } @@ -1022,6 +1112,7 @@ 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; @@ -1071,7 +1162,7 @@ static void *thread_main(void *data) goto err; } -#ifdef FIO_HAVE_LIBNUMA +#ifdef CONFIG_LIBNUMA /* numa node setup */ if (td->o.numa_cpumask_set || td->o.numa_memmask_set) { int ret; @@ -1166,10 +1257,11 @@ static void *thread_main(void *data) } fio_gettime(&td->epoch, NULL); - getrusage(RUSAGE_SELF, &td->ru_start); - + fio_getrusage(&td->ru_start); 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)); @@ -1190,7 +1282,7 @@ static void *thread_main(void *data) prune_io_piece_log(td); - do_io(td); + verify_bytes = do_io(td); clear_state = 1; @@ -1219,7 +1311,7 @@ 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); @@ -1305,6 +1397,9 @@ err: if (td->o.write_iolog_file) write_iolog_close(td); + fio_mutex_remove(td->rusage_sem); + td->rusage_sem = NULL; + td_set_runstate(td, TD_EXITED); return (void *) (uintptr_t) td->error; } @@ -1453,6 +1548,8 @@ static void run_threads(void) if (fio_gtod_offload && fio_start_gtod_thread()) return; + + fio_idle_prof_init(); set_sig_handlers(); @@ -1510,6 +1607,9 @@ static void run_threads(void) } } + /* start idle threads before io threads start to run */ + fio_idle_prof_start(); + set_genesis_time(); while (todo) { @@ -1548,6 +1648,9 @@ static void run_threads(void) init_disk_util(td); + td->rusage_sem = fio_mutex_init(FIO_MUTEX_LOCKED); + td->update_rusage = 0; + /* * Set state to created. Thread will transition * to TD_INITIALIZED when it's done setting up. @@ -1672,6 +1775,8 @@ static void run_threads(void) usleep(10000); } + fio_idle_prof_stop(); + update_io_ticks(); fio_unpin_memory(); }