X-Git-Url: https://git.kernel.dk/?a=blobdiff_plain;f=fio.c;h=5358af2c8a8dc7bea5234c329bdf39087679a15f;hb=36167d82e5f49dee91c6d2cd426068edee90e36f;hp=e8a86585d4bd5d4ee83cff9202eca1ef8a3a58a9;hpb=5bf13a5a5f01986c5b87db26f0cadf95af133188;p=fio.git diff --git a/fio.c b/fio.c index e8a86585..5358af2c 100644 --- a/fio.c +++ b/fio.c @@ -27,6 +27,7 @@ #include #include #include +#include #include #include #include @@ -221,23 +222,32 @@ static int fio_io_sync(struct thread_data *td, struct fio_file *f) } ret = td_io_queue(td, io_u); - if (ret) { + if (ret < 0) { td_verror(td, io_u->error); put_io_u(td, io_u); return 1; - } + } else if (ret == FIO_Q_QUEUED) { + ret = td_io_getevents(td, 1, td->cur_depth, NULL); + if (ret < 0) { + td_verror(td, ret); + 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; + } + } else if (ret == FIO_Q_COMPLETED) { + if (io_u->error) { + td_verror(td, io_u->error); + return 1; + } - icd.nr = ret; - ios_completed(td, &icd); - if (icd.error) { - td_verror(td, icd.error); - return 1; + init_icd(&icd); + io_completed(td, io_u, &icd); + put_io_u(td, io_u); } return 0; @@ -249,9 +259,8 @@ static int fio_io_sync(struct thread_data *td, struct fio_file *f) */ static void do_verify(struct thread_data *td) { - struct io_u *io_u, *v_io_u = NULL; - struct io_completion_data icd; struct fio_file *f; + struct io_u *io_u; int ret, i; /* @@ -265,78 +274,66 @@ static void do_verify(struct thread_data *td) td_set_runstate(td, TD_VERIFYING); - do { - if (td->terminate) - break; - + io_u = NULL; + while (!td->terminate) { io_u = __get_io_u(td); if (!io_u) break; - if (runtime_exceeded(td, &io_u->start_time)) { - put_io_u(td, io_u); + if (runtime_exceeded(td, &io_u->start_time)) break; - } - if (get_next_verify(td, io_u)) { - put_io_u(td, io_u); + if (get_next_verify(td, io_u)) break; - } - f = get_next_file(td); - if (!f) + if (td_io_prep(td, io_u)) break; - io_u->file = f; +requeue: + ret = td_io_queue(td, io_u); - if (td_io_prep(td, io_u)) { - put_io_u(td, io_u); - break; - } + switch (ret) { + case FIO_Q_COMPLETED: + if (io_u->error) + ret = io_u->error; + if (io_u->xfer_buflen != io_u->resid && io_u->resid) { + int bytes = io_u->xfer_buflen - io_u->resid; - ret = td_io_queue(td, io_u); - if (ret) { - td_verror(td, io_u->error); - put_io_u(td, io_u); + io_u->xfer_buflen = io_u->resid; + io_u->xfer_buf += bytes; + goto requeue; + } + if (do_io_u_verify(td, &io_u)) { + ret = -EIO; + break; + } + continue; + case FIO_Q_QUEUED: + break; + default: + assert(ret < 0); + td_verror(td, ret); break; } /* - * we have one pending to verify, do that while - * we are doing io on the next one + * We get here for a queued request, in the future we + * want to later make this take full advantage of + * keeping IO in flight while verifying others. */ - if (do_io_u_verify(td, &v_io_u)) - break; - ret = td_io_getevents(td, 1, 1, NULL); - if (ret != 1) { - if (ret < 0) - td_verror(td, ret); + if (ret < 0) break; - } - 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) { - td_verror(td, icd.error); - put_io_u(td, v_io_u); - v_io_u = NULL; - break; - } + assert(ret == 1); + io_u = td->io_ops->event(td, 0); - /* - * if we can't submit more io, we need to verify now - */ - if (queue_full(td) && do_io_u_verify(td, &v_io_u)) + if (do_io_u_verify(td, &io_u)) break; + } - } while (1); - - do_io_u_verify(td, &v_io_u); + if (io_u) + put_io_u(td, io_u); if (td->cur_depth) cleanup_pending_aio(td); @@ -402,45 +399,63 @@ 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) { - if (ret > 0 && (io_u->xfer_buflen != io_u->resid) && - io_u->resid) { - /* - * short read/write. requeue. - */ + + switch (ret) { + case FIO_Q_COMPLETED: + if (io_u->error) { + ret = io_u->error; + break; + } + if (io_u->xfer_buflen != io_u->resid && io_u->resid) { + int bytes = io_u->xfer_buflen - io_u->resid; + io_u->xfer_buflen = io_u->resid; - io_u->xfer_buf += ret; + io_u->xfer_buf += bytes; goto requeue; - } else { - put_io_u(td, io_u); - break; } + init_icd(&icd); + io_completed(td, io_u, &icd); + put_io_u(td, io_u); + break; + case FIO_Q_QUEUED: + break; + default: + assert(ret < 0); + put_io_u(td, io_u); + break; } + if (ret < 0) + 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}; + if (ret == FIO_Q_QUEUED) { + if (td->cur_depth < td->iodepth) { + struct timespec ts; - timeout = &ts; - min_evts = 0; - } else { - timeout = NULL; - min_evts = 1; - } + ts.tv_sec = 0; + ts.tv_nsec = 0; + timeout = &ts; + min_evts = 0; + } else { + timeout = NULL; + min_evts = 1; + } - ret = td_io_getevents(td, min_evts, td->cur_depth, timeout); - if (ret < 0) { - td_verror(td, ret); - break; - } else if (!ret) - continue; + ret = td_io_getevents(td, min_evts, td->cur_depth, timeout); + if (ret < 0) { + td_verror(td, ret); + break; + } else if (!ret) + continue; - icd.nr = ret; - ios_completed(td, &icd); - if (icd.error) { - td_verror(td, icd.error); - break; + icd.nr = ret; + ios_completed(td, &icd); + if (icd.error) { + td_verror(td, icd.error); + break; + } } /* @@ -466,8 +481,16 @@ requeue: unsigned long long b; b = td->io_blocks[0] + td->io_blocks[1]; - if (!(b % td->thinktime_blocks)) - usec_sleep(td, td->thinktime); + if (!(b % td->thinktime_blocks)) { + int left; + + if (td->thinktime_spin) + __usec_sleep(td->thinktime_spin); + + left = td->thinktime - td->thinktime_spin; + if (left) + usec_sleep(td, left); + } } } @@ -761,28 +784,30 @@ err: close_ioengine(td); cleanup_io_u(td); td_set_runstate(td, TD_EXITED); - return NULL; + return (void *) td->error; } /* * We cannot pass the td data into a forked process, so attach the td and * pass it to the thread worker. */ -static void *fork_main(int shmid, int offset) +static int fork_main(int shmid, int offset) { struct thread_data *td; - void *data; + void *data, *ret; data = shmat(shmid, NULL, 0); if (data == (void *) -1) { + int __err = errno; + perror("shmat"); - return NULL; + return __err; } td = data + offset * sizeof(struct thread_data); - thread_main(td); + ret = thread_main(td); shmdt(data); - return NULL; + return (int) ret; } /* @@ -791,7 +816,7 @@ static void *fork_main(int shmid, int offset) static void reap_threads(int *nr_running, int *t_rate, int *m_rate) { struct thread_data *td; - int i, cputhreads, pending; + int i, cputhreads, pending, status, ret; /* * reap exited threads (TD_EXITED -> TD_REAPED) @@ -805,6 +830,22 @@ static void reap_threads(int *nr_running, int *t_rate, int *m_rate) if (td->io_ops && td->io_ops->flags & FIO_CPUIO) cputhreads++; + if (td->runstate < TD_EXITED) { + /* + * check if someone quit or got killed in an unusual way + */ + ret = waitpid(td->pid, &status, WNOHANG); + if (ret < 0) + perror("waitpid"); + else if ((ret == td->pid) && WIFSIGNALED(status)) { + int sig = WTERMSIG(status); + + log_err("fio: pid=%d, got signal=%d\n", td->pid, sig); + td_set_runstate(td, TD_REAPED); + goto reaped; + } + } + if (td->runstate != TD_EXITED) { if (td->runstate < TD_RUNNING) pending++; @@ -822,9 +863,19 @@ static void reap_threads(int *nr_running, int *t_rate, int *m_rate) if (pthread_join(td->thread, (void *) &ret)) perror("thread_join"); - } else - waitpid(td->pid, NULL, 0); + } else { + int status; + ret = waitpid(td->pid, &status, 0); + if (ret < 0) + perror("waitpid"); + else if (WIFEXITED(status) && WEXITSTATUS(status)) { + if (!exit_value) + exit_value++; + } + } + +reaped: (*nr_running)--; (*m_rate) -= td->ratemin; (*t_rate) -= td->rate; @@ -932,8 +983,9 @@ static void run_threads(void) if (fork()) fio_sem_down(&startup_sem); else { - fork_main(shm_id, i); - exit(0); + int ret = fork_main(shm_id, i); + + exit(ret); } } }