X-Git-Url: https://git.kernel.dk/?a=blobdiff_plain;f=engines%2Flibaio.c;h=b909b79e9c7169f7898e2aa32be37f895cafe4a0;hb=86bff16ee75cf3cd3fb679421a46347ce5acdb47;hp=da7f0c3ee7555bc4ef08bee90e405e0c4dd90acf;hpb=ebec344dd336784fe8c15fa042649a5e768af63d;p=fio.git diff --git a/engines/libaio.c b/engines/libaio.c index da7f0c3e..b909b79e 100644 --- a/engines/libaio.c +++ b/engines/libaio.c @@ -8,19 +8,26 @@ #include #include #include +#include +#include #include "../fio.h" #include "../lib/pow2.h" #include "../optgroup.h" +#include "../lib/memalign.h" -#ifndef IOCB_FLAG_HIPRI -#define IOCB_FLAG_HIPRI (1 << 2) +/* Should be defined in newest aio_abi.h */ +#ifndef IOCB_FLAG_IOPRIO +#define IOCB_FLAG_IOPRIO (1 << 1) #endif -#ifndef IOCTX_FLAG_IOPOLL -#define IOCTX_FLAG_IOPOLL (1 << 0) + +/* Hack for libaio < 0.3.111 */ +#ifndef CONFIG_LIBAIO_RW_FLAGS +#define aio_rw_flags __pad2 #endif static int fio_libaio_commit(struct thread_data *td); +static int fio_libaio_init(struct thread_data *td); struct libaio_data { io_context_t aio_ctx; @@ -28,6 +35,8 @@ struct libaio_data { struct iocb **iocbs; struct io_u **io_us; + struct io_u **io_u_index; + /* * Basic ring buffer. 'head' is incremented in _queue(), and * 'tail' is incremented in _commit(). We keep 'queued' so @@ -46,7 +55,8 @@ struct libaio_data { struct libaio_options { void *pad; unsigned int userspace_reap; - unsigned int hipri; + unsigned int cmdprio_percentage; + unsigned int nowait; }; static struct fio_option options[] = { @@ -59,12 +69,32 @@ static struct fio_option options[] = { .category = FIO_OPT_C_ENGINE, .group = FIO_OPT_G_LIBAIO, }, +#ifdef FIO_HAVE_IOPRIO_CLASS { - .name = "hipri", - .lname = "High Priority", - .type = FIO_OPT_STR_SET, - .off1 = offsetof(struct libaio_options, hipri), - .help = "Use polled IO completions", + .name = "cmdprio_percentage", + .lname = "high priority percentage", + .type = FIO_OPT_INT, + .off1 = offsetof(struct libaio_options, cmdprio_percentage), + .minval = 1, + .maxval = 100, + .help = "Send high priority I/O this percentage of the time", + .category = FIO_OPT_C_ENGINE, + .group = FIO_OPT_G_LIBAIO, + }, +#else + { + .name = "cmdprio_percentage", + .lname = "high priority percentage", + .type = FIO_OPT_UNSUPPORTED, + .help = "Your platform does not support I/O priority classes", + }, +#endif + { + .name = "nowait", + .lname = "RWF_NOWAIT", + .type = FIO_OPT_BOOL, + .off1 = offsetof(struct libaio_options, nowait), + .help = "Set RWF_NOWAIT for reads/writes", .category = FIO_OPT_C_ENGINE, .group = FIO_OPT_G_LIBAIO, }, @@ -82,25 +112,37 @@ static inline void ring_inc(struct libaio_data *ld, unsigned int *val, *val = (*val + add) % ld->entries; } -static int fio_libaio_prep(struct thread_data fio_unused *td, struct io_u *io_u) +static int fio_libaio_prep(struct thread_data *td, struct io_u *io_u) { - struct fio_file *f = io_u->file; struct libaio_options *o = td->eo; + struct fio_file *f = io_u->file; + struct iocb *iocb = &io_u->iocb; if (io_u->ddir == DDIR_READ) { - io_prep_pread(&io_u->iocb, f->fd, io_u->xfer_buf, io_u->xfer_buflen, io_u->offset); - if (o->hipri) - io_u->iocb.u.c.flags |= IOCB_FLAG_HIPRI; + io_prep_pread(iocb, f->fd, io_u->xfer_buf, io_u->xfer_buflen, io_u->offset); + if (o->nowait) + iocb->aio_rw_flags |= RWF_NOWAIT; } else if (io_u->ddir == DDIR_WRITE) { - io_prep_pwrite(&io_u->iocb, f->fd, io_u->xfer_buf, io_u->xfer_buflen, io_u->offset); - if (o->hipri) - io_u->iocb.u.c.flags |= IOCB_FLAG_HIPRI; + io_prep_pwrite(iocb, f->fd, io_u->xfer_buf, io_u->xfer_buflen, io_u->offset); + if (o->nowait) + iocb->aio_rw_flags |= RWF_NOWAIT; } else if (ddir_sync(io_u->ddir)) - io_prep_fsync(&io_u->iocb, f->fd); + io_prep_fsync(iocb, f->fd); return 0; } +static void fio_libaio_prio_prep(struct thread_data *td, struct io_u *io_u) +{ + struct libaio_options *o = td->eo; + if (rand_between(&td->prio_state, 0, 99) < o->cmdprio_percentage) { + io_u->iocb.aio_reqprio = IOPRIO_CLASS_RT << IOPRIO_CLASS_SHIFT; + io_u->iocb.u.c.flags |= IOCB_FLAG_IOPRIO; + io_u->flags |= IO_U_F_PRIORITY; + } + return; +} + static struct io_u *fio_libaio_event(struct thread_data *td, int event) { struct libaio_data *ld = td->io_ops_data; @@ -153,8 +195,8 @@ static int user_io_getevents(io_context_t aio_ctx, unsigned int max, } else { /* There is another completion to reap */ events[i] = ring->events[head]; - read_barrier(); - ring->head = (head + 1) % ring->nr; + atomic_store_release(&ring->head, + (head + 1) % ring->nr); i++; } } @@ -204,6 +246,7 @@ static enum fio_q_status fio_libaio_queue(struct thread_data *td, struct io_u *io_u) { struct libaio_data *ld = td->io_ops_data; + struct libaio_options *o = td->eo; fio_ro_check(td, io_u); @@ -234,6 +277,9 @@ static enum fio_q_status fio_libaio_queue(struct thread_data *td, return FIO_Q_COMPLETED; } + if (o->cmdprio_percentage) + fio_libaio_prio_prep(td, io_u); + ld->iocbs[ld->head] = &io_u->iocb; ld->io_us[ld->head] = io_u; ring_inc(ld, &ld->head, 1); @@ -357,49 +403,28 @@ static void fio_libaio_cleanup(struct thread_data *td) } } -static int fio_libaio_queue_init(struct libaio_data *ld, unsigned int depth, - bool hipri) +static int fio_libaio_post_init(struct thread_data *td) { -#ifdef __NR_sys_io_setup2 - int err, flags = 0; - - if (hipri) - flags = IOCTX_FLAG_IOPOLL; + struct libaio_data *ld = td->io_ops_data; + int err; - return syscall(__NR_sys_io_setup2, depth, flags, &ld->aio_ctx); -#else - if (hipri) { - log_err("fio: polled aio not available on your platform\n"); + err = io_queue_init(td->o.iodepth, &ld->aio_ctx); + if (err) { + td_verror(td, -err, "io_queue_init"); return 1; } - return io_queue_init(depth, &ld->aio_ctx); -#endif + + return 0; } static int fio_libaio_init(struct thread_data *td) { - struct libaio_options *o = td->eo; struct libaio_data *ld; - int err = 0; + struct thread_options *to = &td->o; + struct libaio_options *o = td->eo; ld = calloc(1, sizeof(*ld)); - /* - * First try passing in 0 for queue depth, since we don't - * care about the user ring. If that fails, the kernel is too old - * and we need the right depth. - */ - if (!o->userspace_reap) - err = fio_libaio_queue_init(ld, INT_MAX, o->hipri); - if (o->userspace_reap || err == -EINVAL) - err = fio_libaio_queue_init(ld, td->o.iodepth, o->hipri); - if (err) { - td_verror(td, -err, "io_queue_init"); - log_err("fio: check /proc/sys/fs/aio-max-nr\n"); - free(ld); - return 1; - } - ld->entries = td->o.iodepth; ld->is_pow2 = is_power_of_2(ld->entries); ld->aio_events = calloc(ld->entries, sizeof(struct io_event)); @@ -407,13 +432,25 @@ static int fio_libaio_init(struct thread_data *td) ld->io_us = calloc(ld->entries, sizeof(struct io_u *)); td->io_ops_data = ld; + /* + * Check for option conflicts + */ + if ((fio_option_is_set(to, ioprio) || fio_option_is_set(to, ioprio_class)) && + o->cmdprio_percentage != 0) { + log_err("%s: cmdprio_percentage option and mutually exclusive " + "prio or prioclass option is set, exiting\n", to->name); + td_verror(td, EINVAL, "fio_libaio_init"); + return 1; + } return 0; } -static struct ioengine_ops ioengine = { +FIO_STATIC struct ioengine_ops ioengine = { .name = "libaio", .version = FIO_IOOPS_VERSION, + .flags = FIO_ASYNCIO_SYNC_TRIM, .init = fio_libaio_init, + .post_init = fio_libaio_post_init, .prep = fio_libaio_prep, .queue = fio_libaio_queue, .commit = fio_libaio_commit,