X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=backend.c;h=c1c8f96774850a755681dffa710d0ee1550fe16b;hp=507faa9992e7325d4c80de368e43f2a779893739;hb=49cba9b3e5fa6b30f0fdff52eaf83427ec1baa3f;hpb=44cbc6daaae3674ea5d0a113b66596ca24f372e0 diff --git a/backend.c b/backend.c index 507faa99..c1c8f967 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" @@ -50,23 +50,25 @@ #include "lib/rand.h" #include "memalign.h" #include "server.h" +#include "lib/getrusage.h" +#include "idletime.h" +#include "err.h" static pthread_t disk_util_thread; static struct fio_mutex *disk_thread_mutex; static struct fio_mutex *startup_mutex; -static struct fio_mutex *writeout_mutex; static struct flist_head *cgroup_list; static char *cgroup_mnt; static int exit_value; static volatile int fio_abort; +static unsigned int nr_process = 0; +static unsigned int nr_thread = 0; 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; int temp_stall_ts; unsigned long done_secs = 0; @@ -180,7 +182,11 @@ static int __check_min_rate(struct thread_data *td, struct timeval *now, ratemin); return 1; } else { - rate = ((bytes - td->rate_bytes[ddir]) * 1000) / spent; + if (spent) + rate = ((bytes - td->rate_bytes[ddir]) * 1000) / spent; + else + rate = 0; + if (rate < ratemin || bytes < td->rate_bytes[ddir]) { log_err("%s: min rate %u not met, got" @@ -216,7 +222,7 @@ 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; @@ -236,8 +242,6 @@ static int check_min_rate(struct thread_data *td, struct timeval *now, */ static void cleanup_pending_aio(struct thread_data *td) { - struct flist_head *entry, *n; - struct io_u *io_u; int r; /* @@ -251,18 +255,11 @@ static void cleanup_pending_aio(struct thread_data *td) * now cancel remaining active events */ if (td->io_ops->cancel) { - flist_for_each_safe(entry, n, &td->io_u_busylist) { - io_u = flist_entry(entry, struct io_u, list); + struct io_u *io_u; + int i; - /* - * if the io_u isn't in flight, then that generally - * means someone leaked an io_u. complain but fix - * it up, so we don't stall here. - */ - if ((io_u->flags & IO_U_F_FLIGHT) == 0) { - log_err("fio: non-busy IO on busy list\n"); - put_io_u(td, io_u); - } else { + io_u_qiter(&td->io_u_all, io_u, i) { + if (io_u->flags & IO_U_F_FLIGHT) { r = td->io_ops->cancel(td, io_u); if (!r) put_io_u(td, io_u); @@ -320,6 +317,21 @@ 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); @@ -337,7 +349,7 @@ static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) return 0; if (!td->o.timeout) return 0; - if (mtime_since(&td->epoch, t) >= td->o.timeout * 1000) + if (utime_since(&td->epoch, t) >= td->o.timeout) return 1; return 0; @@ -389,16 +401,25 @@ 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, 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; - uint64_t io_bytes; unsigned int i; dprint(FD_VERIFY, "starting loop\n"); @@ -416,18 +437,20 @@ static void do_verify(struct thread_data *td, uint64_t verify_bytes) break; } + check_update_rusage(td); + if (td->error) return; td_set_runstate(td, TD_VERIFYING); io_u = NULL; - io_bytes = 0; 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); @@ -455,11 +478,43 @@ static void do_verify(struct thread_data *td, uint64_t verify_bytes) break; } } else { - io_u = get_io_u(td); - if (!io_u) + if (ddir_rw_sum(bytes_done) + td->o.rw_min_bs > verify_bytes) break; - if (io_u->buflen + io_bytes > verify_bytes) + while ((io_u = get_io_u(td)) != NULL) { + if (IS_ERR(io_u)) { + io_u = NULL; + ret = FIO_Q_BUSY; + goto reap; + } + + /* + * 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; } @@ -491,7 +546,6 @@ static void do_verify(struct thread_data *td, uint64_t verify_bytes) io_u->xfer_buflen = io_u->resid; io_u->xfer_buf += bytes; io_u->offset += bytes; - io_bytes += bytes; if (ddir_rw(io_u->ddir)) td->ts.short_io_u[io_u->ddir]++; @@ -503,11 +557,10 @@ static void do_verify(struct thread_data *td, uint64_t verify_bytes) 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; } - io_bytes += io_u->xfer_buflen; continue; case FIO_Q_QUEUED: break; @@ -531,6 +584,7 @@ sync_done: * completed io_u's first. Note that we can get BUSY even * without IO queued, if the system is resource starved. */ +reap: full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth); if (full || !td->o.iodepth_batch_complete) { min_events = min(td->o.iodepth_batch_complete, @@ -542,24 +596,23 @@ sync_done: min_events = 1; do { - unsigned long bytes = 0; - /* * Reap required number of io units, if any, * and do the verification on them through * the callback handler */ - if (io_u_queued_complete(td, min_events, &bytes) < 0) { + if (io_u_queued_complete(td, min_events, bytes_done) < 0) { ret = -1; break; } - io_bytes += bytes; } while (full && (td->cur_depth > td->o.iodepth_low)); } if (ret < 0) break; } + check_update_rusage(td); + if (!td->error) { min_events = td->cur_depth; @@ -573,6 +626,19 @@ sync_done: dprint(FD_VERIFY, "exiting loop\n"); } +static unsigned int exceeds_number_ios(struct thread_data *td) +{ + unsigned long long number_ios; + + if (!td->o.number_ios) + return 0; + + number_ios = ddir_rw_sum(td->this_io_blocks); + number_ios += td->io_u_queued + td->io_u_in_flight; + + return number_ios >= td->o.number_ios; +} + static int io_bytes_exceeded(struct thread_data *td) { unsigned long long bytes; @@ -586,33 +652,50 @@ static int io_bytes_exceeded(struct thread_data *td) else bytes = td->this_io_bytes[DDIR_TRIM]; - return bytes >= td->o.size; + return bytes >= td->o.size || exceeds_number_ios(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; + uint64_t total_bytes, bytes_issued = 0; if (in_ramp_time(td)) td_set_runstate(td, TD_RAMP); else td_set_runstate(td, TD_RUNNING); + lat_target_init(td); + + /* + * If verify_backlog is enabled, we'll run the verify in this + * handler as well. For that case, we may need up to twice the + * amount of bytes. + */ + total_bytes = td->o.size; + if (td->o.verify != VERIFY_NONE && + (td_write(td) && td->o.verify_backlog)) + total_bytes += td->o.size; + while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) || (!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; @@ -629,9 +712,22 @@ static void do_io(struct thread_data *td) if (flow_threshold_exceeded(td)) continue; + if (bytes_issued >= total_bytes) + break; + io_u = get_io_u(td); - if (!io_u) + if (IS_ERR_OR_NULL(io_u)) { + int err = PTR_ERR(io_u); + + io_u = NULL; + if (err == -EBUSY) { + ret = FIO_Q_BUSY; + goto reap; + } + if (td->o.latency_target) + goto reap; break; + } ddir = io_u->ddir; @@ -642,6 +738,13 @@ static void do_io(struct thread_data *td) */ if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ && ((io_u->flags & IO_U_F_VER_LIST) || !td_rw(td))) { + + if (!td->o.verify_pattern_bytes) { + io_u->rand_seed = __rand(&td->__verify_state); + if (sizeof(int) != sizeof(long *)) + io_u->rand_seed *= __rand(&td->__verify_state); + } + if (td->o.verify_async) io_u->end_io = verify_io_u_async; else @@ -652,6 +755,17 @@ static void do_io(struct thread_data *td) else td_set_runstate(td, TD_RUNNING); + /* + * Always log IO before it's issued, so we know the specific + * order of it. The logged unit will track when the IO has + * completed. + */ + if (td_write(td) && io_u->ddir == DDIR_WRITE && + td->o.do_verify && + td->o.verify != VERIFY_NONE && + !td->o.experimental_verify) + log_io_piece(td, io_u); + ret = td_io_queue(td, io_u); switch (ret) { case FIO_Q_COMPLETED: @@ -662,6 +776,7 @@ static void do_io(struct thread_data *td) int bytes = io_u->xfer_buflen - io_u->resid; struct fio_file *f = io_u->file; + bytes_issued += bytes; /* * zero read, fail */ @@ -692,6 +807,7 @@ sync_done: ret = io_u_sync_complete(td, io_u, bytes_done); if (ret < 0) break; + bytes_issued += io_u->xfer_buflen; } break; case FIO_Q_QUEUED: @@ -702,6 +818,7 @@ sync_done: */ if (td->io_ops->commit == NULL) io_u_queued(td, io_u); + bytes_issued += io_u->xfer_buflen; break; case FIO_Q_BUSY: requeue_io_u(td, &io_u); @@ -723,6 +840,7 @@ sync_done: * can get BUSY even without IO queued, if the system is * resource starved. */ +reap: full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth); if (full || !td->o.iodepth_batch_complete) { min_evts = min(td->o.iodepth_batch_complete, @@ -759,6 +877,8 @@ sync_done: break; } } + if (!in_ramp_time(td) && td->o.latency_target) + lat_target_check(td); if (td->o.thinktime) { unsigned long long b; @@ -767,6 +887,8 @@ sync_done: if (!(b % td->o.thinktime_blocks)) { int left; + io_u_quiesce(td); + if (td->o.thinktime_spin) usec_spin(td->o.thinktime_spin); @@ -777,8 +899,10 @@ sync_done: } } + check_update_rusage(td); + if (td->trim_entries) - log_err("fio: %d trim entries leaked?\n", td->trim_entries); + log_err("fio: %lu trim entries leaked?\n", td->trim_entries); if (td->o.fill_device && td->error == ENOSPC) { td->error = 0; @@ -789,7 +913,7 @@ 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; } @@ -798,9 +922,11 @@ 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 @@ -811,17 +937,15 @@ 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) { - struct flist_head *entry, *n; struct io_u *io_u; - flist_for_each_safe(entry, n, &td->io_u_freelist) { - io_u = flist_entry(entry, struct io_u, list); - - flist_del(&io_u->list); + while ((io_u = io_u_qpop(&td->io_u_freelist)) != NULL) { if (td->io_ops->io_u_free) td->io_ops->io_u_free(td, io_u); @@ -830,6 +954,10 @@ static void cleanup_io_u(struct thread_data *td) } free_io_mem(td); + + io_u_rexit(&td->io_u_requeues); + io_u_qexit(&td->io_u_freelist); + io_u_qexit(&td->io_u_all); } static int init_io_u(struct thread_data *td) @@ -837,12 +965,11 @@ 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; + int data_xfer = 1, err; 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; @@ -850,6 +977,26 @@ static int init_io_u(struct thread_data *td) if ((td->io_ops->flags & 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 + * overflow later. this adjustment may be too much if we get + * lucky and the allocator gives us an aligned address. + */ + if (td->o.odirect || td->o.mem_align || td->o.oatomic || + (td->io_ops->flags & FIO_RAWIO)) + 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; @@ -865,7 +1012,7 @@ static int init_io_u(struct thread_data *td) if (data_xfer && allocate_io_mem(td)) return 1; - if (td->o.odirect || td->o.mem_align || + if (td->o.odirect || td->o.mem_align || td->o.oatomic || (td->io_ops->flags & FIO_RAWIO)) p = PAGE_ALIGN(td->orig_buffer) + td->o.mem_align; else @@ -887,7 +1034,7 @@ static int init_io_u(struct thread_data *td) io_u = ptr; memset(io_u, 0, sizeof(*io_u)); - INIT_FLIST_HEAD(&io_u->list); + INIT_FLIST_HEAD(&io_u->verify_list); dprint(FD_MEM, "io_u alloc %p, index %u\n", io_u, i); if (data_xfer) { @@ -901,13 +1048,19 @@ static int init_io_u(struct thread_data *td) * Fill the buffer with the pattern if we are * going to be doing writes. */ - fill_pattern(td, io_u->buf, max_bs, io_u, 0, 0); + fill_verify_pattern(td, io_u->buf, max_bs, io_u, 0, 0); } } io_u->index = i; io_u->flags = IO_U_F_FREE; - flist_add(&io_u->list, &td->io_u_freelist); + 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); @@ -961,7 +1114,8 @@ static int switch_ioscheduler(struct thread_data *td) /* * Read back and check that the selected scheduler is now the default. */ - ret = fread(tmp, 1, sizeof(tmp), f); + memset(tmp, 0, sizeof(tmp)); + ret = fread(tmp, 1, sizeof(tmp) - 1, f); if (ferror(f) || ret < 0) { td_verror(td, errno, "fread"); fclose(f); @@ -990,21 +1144,38 @@ static int keep_running(struct thread_data *td) td->o.loops--; return 1; } + if (exceeds_number_ios(td)) + return 0; + + 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; + + if (fio_files_done(td)) + return 0; - if (ddir_rw_sum(td->io_bytes) < td->o.size) return 1; + } return 0; } -static int exec_string(const char *string) +static int exec_string(struct thread_options *o, const char *string, const char *mode) { - int ret, newlen = strlen(string) + 1 + 8; + int ret, newlen = strlen(string) + strlen(o->name) + strlen(mode) + 9 + 1; char *str; str = malloc(newlen); - sprintf(str, "sh -c %s", string); + sprintf(str, "%s &> %s.%s.txt", string, o->name, mode); + log_info("%s : Saving output of %s in %s.%s.txt\n",o->name, mode, o->name, mode); ret = system(str); if (ret == -1) log_err("fio: exec of cmd <%s> failed\n", str); @@ -1013,6 +1184,50 @@ static int exec_string(const char *string) return ret; } +/* + * Dry run to compute correct state of numberio for verification. + */ +static uint64_t do_dry_run(struct thread_data *td) +{ + uint64_t bytes_done[DDIR_RWDIR_CNT] = { 0, 0, 0 }; + + td_set_runstate(td, TD_RUNNING); + + while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) || + (!flist_empty(&td->trim_list)) || !io_bytes_exceeded(td)) { + struct io_u *io_u; + int ret; + + if (td->terminate || td->done) + break; + + io_u = get_io_u(td); + if (!io_u) + break; + + io_u->flags |= IO_U_F_FLIGHT; + io_u->error = 0; + io_u->resid = 0; + if (ddir_rw(acct_ddir(io_u))) + td->io_issues[acct_ddir(io_u)]++; + if (ddir_rw(io_u->ddir)) { + io_u_mark_depth(td, 1); + td->ts.total_io_u[io_u->ddir]++; + } + + if (td_write(td) && io_u->ddir == DDIR_WRITE && + td->o.do_verify && + td->o.verify != VERIFY_NONE && + !td->o.experimental_verify) + log_io_piece(td, io_u); + + ret = io_u_sync_complete(td, io_u, bytes_done); + (void) ret; + } + + return bytes_done[DDIR_WRITE] + bytes_done[DDIR_TRIM]; +} + /* * Entry point for the thread based jobs. The process based jobs end up * here as well, after a little setup. @@ -1021,22 +1236,29 @@ static void *thread_main(void *data) { unsigned long long elapsed; struct thread_data *td = data; + struct thread_options *o = &td->o; pthread_condattr_t attr; int clear_state; + int ret; - if (!td->o.use_thread) { + if (!o->use_thread) { setsid(); td->pid = getpid(); } else td->pid = gettid(); - fio_local_clock_init(td->o.use_thread); + /* + * fio_time_init() may not have been called yet if running as a server + */ + fio_time_init(); + + fio_local_clock_init(o->use_thread); dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid); - INIT_FLIST_HEAD(&td->io_u_freelist); - INIT_FLIST_HEAD(&td->io_u_busylist); - INIT_FLIST_HEAD(&td->io_u_requeues); + if (is_backend) + fio_server_send_start(td); + INIT_FLIST_HEAD(&td->io_log_list); INIT_FLIST_HEAD(&td->io_hist_list); INIT_FLIST_HEAD(&td->verify_list); @@ -1056,21 +1278,15 @@ static void *thread_main(void *data) fio_mutex_down(td->mutex); dprint(FD_MUTEX, "done waiting on td->mutex\n"); - /* - * the ->mutex mutex is now no longer used, close it to avoid - * eating a file descriptor - */ - fio_mutex_remove(td->mutex); - /* * A new gid requires privilege, so we need to do this before setting * the uid. */ - if (td->o.gid != -1U && setgid(td->o.gid)) { + if (o->gid != -1U && setgid(o->gid)) { td_verror(td, errno, "setgid"); goto err; } - if (td->o.uid != -1U && setuid(td->o.uid)) { + if (o->uid != -1U && setuid(o->uid)) { td_verror(td, errno, "setuid"); goto err; } @@ -1079,21 +1295,33 @@ static void *thread_main(void *data) * If we have a gettimeofday() thread, make sure we exclude that * thread from this job */ - if (td->o.gtod_cpu) - fio_cpu_clear(&td->o.cpumask, td->o.gtod_cpu); + if (o->gtod_cpu) + fio_cpu_clear(&o->cpumask, o->gtod_cpu); /* * Set affinity first, in case it has an impact on the memory * allocations. */ - if (td->o.cpumask_set && fio_setaffinity(td->pid, td->o.cpumask) == -1) { - td_verror(td, errno, "cpu_set_affinity"); - goto err; + if (o->cpumask_set) { + if (o->cpus_allowed_policy == FIO_CPUS_SPLIT) { + ret = fio_cpus_split(&o->cpumask, td->thread_number - 1); + if (!ret) { + log_err("fio: no CPUs set\n"); + log_err("fio: Try increasing number of available CPUs\n"); + td_verror(td, EINVAL, "cpus_split"); + goto err; + } + } + ret = fio_setaffinity(td->pid, o->cpumask); + if (ret == -1) { + td_verror(td, errno, "cpu_set_affinity"); + goto err; + } } #ifdef CONFIG_LIBNUMA /* numa node setup */ - if (td->o.numa_cpumask_set || td->o.numa_memmask_set) { + if (o->numa_cpumask_set || o->numa_memmask_set) { int ret; if (numa_available() < 0) { @@ -1101,8 +1329,8 @@ static void *thread_main(void *data) goto err; } - if (td->o.numa_cpumask_set) { - ret = numa_run_on_node_mask(td->o.numa_cpunodesmask); + if (o->numa_cpumask_set) { + ret = numa_run_on_node_mask(o->numa_cpunodesmask); if (ret == -1) { td_verror(td, errno, \ "numa_run_on_node_mask failed\n"); @@ -1110,20 +1338,20 @@ static void *thread_main(void *data) } } - if (td->o.numa_memmask_set) { + if (o->numa_memmask_set) { - switch (td->o.numa_mem_mode) { + switch (o->numa_mem_mode) { case MPOL_INTERLEAVE: - numa_set_interleave_mask(td->o.numa_memnodesmask); + numa_set_interleave_mask(o->numa_memnodesmask); break; case MPOL_BIND: - numa_set_membind(td->o.numa_memnodesmask); + numa_set_membind(o->numa_memnodesmask); break; case MPOL_LOCAL: numa_set_localalloc(); break; case MPOL_PREFERRED: - numa_set_preferred(td->o.numa_mem_prefer_node); + numa_set_preferred(o->numa_mem_prefer_node); break; case MPOL_DEFAULT: default: @@ -1134,6 +1362,9 @@ static void *thread_main(void *data) } #endif + if (fio_pin_memory(td)) + goto err; + /* * May alter parameters that init_io_u() will use, so we need to * do this first. @@ -1144,29 +1375,30 @@ static void *thread_main(void *data) if (init_io_u(td)) goto err; - if (td->o.verify_async && verify_async_init(td)) + if (o->verify_async && verify_async_init(td)) goto err; - if (td->ioprio_set) { - if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { + if (o->ioprio) { + ret = ioprio_set(IOPRIO_WHO_PROCESS, 0, o->ioprio_class, o->ioprio); + if (ret == -1) { td_verror(td, errno, "ioprio_set"); goto err; } } - if (td->o.cgroup && cgroup_setup(td, cgroup_list, &cgroup_mnt)) + if (o->cgroup && cgroup_setup(td, cgroup_list, &cgroup_mnt)) goto err; errno = 0; - if (nice(td->o.nice) == -1 && errno != 0) { + if (nice(o->nice) == -1 && errno != 0) { td_verror(td, errno, "nice"); goto err; } - if (td->o.ioscheduler && switch_ioscheduler(td)) + if (o->ioscheduler && switch_ioscheduler(td)) goto err; - if (!td->o.create_serialize && setup_files(td)) + if (!o->create_serialize && setup_files(td)) goto err; if (td_io_init(td)) @@ -1175,30 +1407,29 @@ static void *thread_main(void *data) if (init_random_map(td)) goto err; - if (td->o.exec_prerun) { - if (exec_string(td->o.exec_prerun)) - goto err; - } + if (o->exec_prerun && exec_string(o, o->exec_prerun, (const char *)"prerun")) + goto err; - if (td->o.pre_read) { + if (o->pre_read) { if (pre_read_files(td) < 0) goto err; } - fio_gettime(&td->epoch, NULL); - getrusage(RUSAGE_SELF, &td->ru_start); + fio_verify_init(td); + fio_gettime(&td->epoch, NULL); + fio_getrusage(&td->ru_start); clear_state = 0; while (keep_running(td)) { - uint64_t write_bytes; + 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)); memcpy(&td->tv_cache, &td->start, sizeof(td->start)); - if (td->o.ratemin[DDIR_READ] || td->o.ratemin[DDIR_WRITE] || - td->o.ratemin[DDIR_TRIM]) { + if (o->ratemin[DDIR_READ] || o->ratemin[DDIR_WRITE] || + o->ratemin[DDIR_TRIM]) { memcpy(&td->lastrate[DDIR_READ], &td->bw_sample_time, sizeof(td->bw_sample_time)); memcpy(&td->lastrate[DDIR_WRITE], &td->bw_sample_time, @@ -1212,9 +1443,10 @@ static void *thread_main(void *data) prune_io_piece_log(td); - write_bytes = td->io_bytes[DDIR_WRITE]; - do_io(td); - write_bytes = td->io_bytes[DDIR_WRITE] - write_bytes; + if (td->o.verify_only && (td_write(td) || td_rw(td))) + verify_bytes = do_dry_run(td); + else + verify_bytes = do_io(td); clear_state = 1; @@ -1234,8 +1466,8 @@ static void *thread_main(void *data) if (td->error || td->terminate) break; - if (!td->o.do_verify || - td->o.verify == VERIFY_NONE || + if (!o->do_verify || + o->verify == VERIFY_NONE || (td->io_ops->flags & FIO_UNIDIR)) continue; @@ -1243,7 +1475,7 @@ static void *thread_main(void *data) fio_gettime(&td->start, NULL); - do_verify(td, write_bytes); + do_verify(td, verify_bytes); td->ts.runtime[DDIR_READ] += utime_since_now(&td->start); @@ -1260,46 +1492,12 @@ static void *thread_main(void *data) td->ts.io_bytes[DDIR_WRITE] = td->io_bytes[DDIR_WRITE]; td->ts.io_bytes[DDIR_TRIM] = td->io_bytes[DDIR_TRIM]; - fio_mutex_down(writeout_mutex); - if (td->bw_log) { - if (td->o.bw_log_file) { - finish_log_named(td, td->bw_log, - td->o.bw_log_file, "bw"); - } else - finish_log(td, td->bw_log, "bw"); - } - if (td->lat_log) { - if (td->o.lat_log_file) { - finish_log_named(td, td->lat_log, - td->o.lat_log_file, "lat"); - } else - finish_log(td, td->lat_log, "lat"); - } - if (td->slat_log) { - if (td->o.lat_log_file) { - finish_log_named(td, td->slat_log, - td->o.lat_log_file, "slat"); - } else - finish_log(td, td->slat_log, "slat"); - } - if (td->clat_log) { - if (td->o.lat_log_file) { - finish_log_named(td, td->clat_log, - td->o.lat_log_file, "clat"); - } else - finish_log(td, td->clat_log, "clat"); - } - if (td->iops_log) { - if (td->o.iops_log_file) { - finish_log_named(td, td->iops_log, - td->o.iops_log_file, "iops"); - } else - finish_log(td, td->iops_log, "iops"); - } + fio_unpin_memory(td); + + fio_writeout_logs(td); - fio_mutex_up(writeout_mutex); - if (td->o.exec_postrun) - exec_string(td->o.exec_postrun); + if (o->exec_postrun) + exec_string(o, o->exec_postrun, (const char *)"postrun"); if (exitall_on_terminate) fio_terminate_threads(td->groupid); @@ -1309,7 +1507,7 @@ err: log_info("fio: pid=%d, err=%d/%s\n", (int) td->pid, td->error, td->verror); - if (td->o.verify_async) + if (o->verify_async) verify_async_exit(td); close_and_free_files(td); @@ -1317,8 +1515,8 @@ err: close_ioengine(td); cgroup_shutdown(td, &cgroup_mnt); - if (td->o.cpumask_set) { - int ret = fio_cpuset_exit(&td->o.cpumask); + if (o->cpumask_set) { + int ret = fio_cpuset_exit(&o->cpumask); td_verror(td, ret, "fio_cpuset_exit"); } @@ -1326,9 +1524,15 @@ err: /* * do this very late, it will log file closing as well */ - if (td->o.write_iolog_file) + if (o->write_iolog_file) write_iolog_close(td); + fio_mutex_remove(td->rusage_sem); + td->rusage_sem = NULL; + + fio_mutex_remove(td->mutex); + td->mutex = NULL; + td_set_runstate(td, TD_EXITED); return (void *) (uintptr_t) td->error; } @@ -1457,29 +1661,43 @@ reaped: exit_value++; done_secs += mtime_since_now(&td->epoch) / 1000; + profile_td_exit(td); } if (*nr_running == cputhreads && !pending && realthreads) fio_terminate_threads(TERMINATE_ALL); } +static void do_usleep(unsigned int usecs) +{ + check_for_running_stats(); + usleep(usecs); +} + /* * Main function for kicking off and reaping jobs, as needed. */ static void run_threads(void) { struct thread_data *td; - unsigned long spent; unsigned int i, todo, nr_running, m_rate, t_rate, nr_started; - - if (fio_pin_memory()) - return; + uint64_t spent; if (fio_gtod_offload && fio_start_gtod_thread()) return; + fio_idle_prof_init(); + set_sig_handlers(); + nr_thread = nr_process = 0; + for_each_td(td, i) { + if (td->o.use_thread) + nr_thread++; + else + nr_process++; + } + if (output_format == FIO_OUTPUT_NORMAL) { log_info("Starting "); if (nr_thread) @@ -1534,6 +1752,9 @@ static void run_threads(void) } } + /* start idle threads before io threads start to run */ + fio_idle_prof_start(); + set_genesis_time(); while (todo) { @@ -1558,9 +1779,9 @@ static void run_threads(void) } if (td->o.start_delay) { - spent = mtime_since_genesis(); + spent = utime_since_genesis(); - if (td->o.start_delay * 1000 > spent) + if (td->o.start_delay > spent) continue; } @@ -1572,6 +1793,9 @@ 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. @@ -1628,7 +1852,7 @@ static void run_threads(void) if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) break; - usleep(100000); + do_usleep(100000); for (i = 0; i < this_jobs; i++) { td = map[i]; @@ -1679,25 +1903,18 @@ static void run_threads(void) reap_threads(&nr_running, &t_rate, &m_rate); - if (todo) { - if (is_backend) - fio_server_idle_loop(); - else - usleep(100000); - } + if (todo) + do_usleep(100000); } while (nr_running) { reap_threads(&nr_running, &t_rate, &m_rate); - - if (is_backend) - fio_server_idle_loop(); - else - usleep(10000); + do_usleep(10000); } + fio_idle_prof_stop(); + update_io_ticks(); - fio_unpin_memory(); } void wait_for_disk_thread_exit(void) @@ -1775,19 +1992,17 @@ int fio_backend(void) return 0; if (write_bw_log) { - setup_log(&agg_io_log[DDIR_READ], 0); - setup_log(&agg_io_log[DDIR_WRITE], 0); - setup_log(&agg_io_log[DDIR_TRIM], 0); + setup_log(&agg_io_log[DDIR_READ], 0, IO_LOG_TYPE_BW); + setup_log(&agg_io_log[DDIR_WRITE], 0, IO_LOG_TYPE_BW); + setup_log(&agg_io_log[DDIR_TRIM], 0, IO_LOG_TYPE_BW); } startup_mutex = fio_mutex_init(FIO_MUTEX_LOCKED); if (startup_mutex == NULL) return 1; - writeout_mutex = fio_mutex_init(FIO_MUTEX_UNLOCKED); - if (writeout_mutex == NULL) - return 1; set_genesis_time(); + stat_init(); create_disk_util_thread(); cgroup_list = smalloc(sizeof(*cgroup_list)); @@ -1815,7 +2030,7 @@ int fio_backend(void) sfree(cgroup_mnt); fio_mutex_remove(startup_mutex); - fio_mutex_remove(writeout_mutex); fio_mutex_remove(disk_thread_mutex); + stat_exit(); return exit_value; }