int queued;
int cq_ring_off;
unsigned iodepth;
+ bool ioprio_class_set;
+ bool ioprio_set;
struct ioring_mmap mmap[3];
};
struct ioring_options {
void *pad;
unsigned int hipri;
+ unsigned int cmdprio_percentage;
unsigned int fixedbufs;
unsigned int registerfiles;
unsigned int sqpoll_thread;
unsigned int sqpoll_set;
unsigned int sqpoll_cpu;
+ unsigned int nonvectored;
+ unsigned int uncached;
+ unsigned int nowait;
+};
+
+static const int ddir_to_op[2][2] = {
+ { IORING_OP_READV, IORING_OP_READ },
+ { IORING_OP_WRITEV, IORING_OP_WRITE }
+};
+
+static const int fixed_ddir_to_op[2] = {
+ IORING_OP_READ_FIXED,
+ IORING_OP_WRITE_FIXED
};
static int fio_ioring_sqpoll_cb(void *data, unsigned long long *val)
.category = FIO_OPT_C_ENGINE,
.group = FIO_OPT_G_IOURING,
},
+#ifdef FIO_HAVE_IOPRIO_CLASS
+ {
+ .name = "cmdprio_percentage",
+ .lname = "high priority percentage",
+ .type = FIO_OPT_INT,
+ .off1 = offsetof(struct ioring_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_IOURING,
+ },
+#else
+ {
+ .name = "cmdprio_percentage",
+ .lname = "high priority percentage",
+ .type = FIO_OPT_UNSUPPORTED,
+ .help = "Your platform does not support I/O priority classes",
+ },
+#endif
{
.name = "fixedbufs",
.lname = "Fixed (pre-mapped) IO buffers",
.category = FIO_OPT_C_ENGINE,
.group = FIO_OPT_G_IOURING,
},
+ {
+ .name = "nonvectored",
+ .lname = "Non-vectored",
+ .type = FIO_OPT_INT,
+ .off1 = offsetof(struct ioring_options, nonvectored),
+ .help = "Use non-vectored read/write commands",
+ .category = FIO_OPT_C_ENGINE,
+ .group = FIO_OPT_G_IOURING,
+ },
+ {
+ .name = "uncached",
+ .lname = "Uncached",
+ .type = FIO_OPT_INT,
+ .off1 = offsetof(struct ioring_options, uncached),
+ .help = "Use RWF_UNCACHED for buffered read/writes",
+ .category = FIO_OPT_C_ENGINE,
+ .group = FIO_OPT_G_IOURING,
+ },
+ {
+ .name = "nowait",
+ .lname = "RWF_NOWAIT",
+ .type = FIO_OPT_BOOL,
+ .off1 = offsetof(struct ioring_options, nowait),
+ .help = "Use RWF_NOWAIT for reads/writes",
+ .category = FIO_OPT_C_ENGINE,
+ .group = FIO_OPT_G_IOURING,
+ },
{
.name = NULL,
},
static int io_uring_enter(struct ioring_data *ld, unsigned int to_submit,
unsigned int min_complete, unsigned int flags)
{
- return syscall(__NR_sys_io_uring_enter, ld->ring_fd, to_submit,
+ return syscall(__NR_io_uring_enter, ld->ring_fd, to_submit,
min_complete, flags, NULL, 0);
}
struct io_uring_sqe *sqe;
sqe = &ld->sqes[io_u->index];
+
+ /* zero out fields not used in this submission */
+ memset(sqe, 0, sizeof(*sqe));
+
if (o->registerfiles) {
sqe->fd = f->engine_pos;
sqe->flags = IOSQE_FIXED_FILE;
} else {
sqe->fd = f->fd;
- sqe->flags = 0;
}
- sqe->ioprio = 0;
- sqe->buf_index = 0;
if (io_u->ddir == DDIR_READ || io_u->ddir == DDIR_WRITE) {
if (o->fixedbufs) {
- if (io_u->ddir == DDIR_READ)
- sqe->opcode = IORING_OP_READ_FIXED;
- else
- sqe->opcode = IORING_OP_WRITE_FIXED;
+ sqe->opcode = fixed_ddir_to_op[io_u->ddir];
sqe->addr = (unsigned long) io_u->xfer_buf;
sqe->len = io_u->xfer_buflen;
sqe->buf_index = io_u->index;
} else {
- if (io_u->ddir == DDIR_READ)
- sqe->opcode = IORING_OP_READV;
- else
- sqe->opcode = IORING_OP_WRITEV;
- sqe->addr = (unsigned long) &ld->iovecs[io_u->index];
- sqe->len = 1;
+ sqe->opcode = ddir_to_op[io_u->ddir][!!o->nonvectored];
+ if (o->nonvectored) {
+ sqe->addr = (unsigned long)
+ ld->iovecs[io_u->index].iov_base;
+ sqe->len = ld->iovecs[io_u->index].iov_len;
+ } else {
+ sqe->addr = (unsigned long) &ld->iovecs[io_u->index];
+ sqe->len = 1;
+ }
}
+ if (!td->o.odirect && o->uncached)
+ sqe->rw_flags = RWF_UNCACHED;
+ if (o->nowait)
+ sqe->rw_flags |= RWF_NOWAIT;
+ if (ld->ioprio_class_set)
+ sqe->ioprio = td->o.ioprio_class << 13;
+ if (ld->ioprio_set)
+ sqe->ioprio |= td->o.ioprio;
sqe->off = io_u->offset;
} else if (ddir_sync(io_u->ddir)) {
if (io_u->ddir == DDIR_SYNC_FILE_RANGE) {
sqe->sync_range_flags = td->o.sync_file_range;
sqe->opcode = IORING_OP_SYNC_FILE_RANGE;
} else {
- sqe->fsync_flags = 0;
if (io_u->ddir == DDIR_DATASYNC)
sqe->fsync_flags |= IORING_FSYNC_DATASYNC;
sqe->opcode = IORING_OP_FSYNC;
head = *ring->head;
do {
- read_barrier();
- if (head == *ring->tail)
+ if (head == atomic_load_acquire(ring->tail))
break;
reaped++;
head++;
} while (reaped + events < max);
- *ring->head = head;
- write_barrier();
+ atomic_store_release(ring->head, head);
return reaped;
}
return r < 0 ? r : events;
}
+static void fio_ioring_prio_prep(struct thread_data *td, struct io_u *io_u)
+{
+ struct ioring_options *o = td->eo;
+ struct ioring_data *ld = td->io_ops_data;
+ if (rand_between(&td->prio_state, 0, 99) < o->cmdprio_percentage) {
+ ld->sqes[io_u->index].ioprio = IOPRIO_CLASS_RT << IOPRIO_CLASS_SHIFT;
+ io_u->flags |= IO_U_F_PRIORITY;
+ }
+ return;
+}
+
static enum fio_q_status fio_ioring_queue(struct thread_data *td,
struct io_u *io_u)
{
struct ioring_data *ld = td->io_ops_data;
struct io_sq_ring *ring = &ld->sq_ring;
+ struct ioring_options *o = td->eo;
unsigned tail, next_tail;
fio_ro_check(td, io_u);
tail = *ring->tail;
next_tail = tail + 1;
- read_barrier();
- if (next_tail == *ring->head)
+ if (next_tail == atomic_load_acquire(ring->head))
return FIO_Q_BUSY;
- /* ensure sqe stores are ordered with tail update */
- write_barrier();
+ if (o->cmdprio_percentage)
+ fio_ioring_prio_prep(td, io_u);
ring->array[tail & ld->sq_ring_mask] = io_u->index;
- *ring->tail = next_tail;
- write_barrier();
+ atomic_store_release(ring->tail, next_tail);
ld->queued++;
return FIO_Q_QUEUED;
}
}
- ret = syscall(__NR_sys_io_uring_setup, depth, &p);
+ ret = syscall(__NR_io_uring_setup, depth, &p);
if (ret < 0)
return ret;
if (setrlimit(RLIMIT_MEMLOCK, &rlim) < 0)
return -1;
- ret = syscall(__NR_sys_io_uring_register, ld->ring_fd,
+ ret = syscall(__NR_io_uring_register, ld->ring_fd,
IORING_REGISTER_BUFFERS, ld->iovecs, depth);
if (ret < 0)
return ret;
f->engine_pos = i;
}
- ret = syscall(__NR_sys_io_uring_register, ld->ring_fd,
+ ret = syscall(__NR_io_uring_register, ld->ring_fd,
IORING_REGISTER_FILES, ld->fds, td->o.nr_files);
if (ret) {
err:
return 1;
}
- printf("files=%d\n", o->registerfiles);
if (o->registerfiles) {
err = fio_ioring_register_files(td);
if (err) {
{
struct ioring_options *o = td->eo;
struct ioring_data *ld;
+ struct thread_options *to = &td->o;
/* sqthread submission requires registered files */
if (o->sqpoll_thread)
ld->iovecs = calloc(td->o.iodepth, sizeof(struct iovec));
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_io_uring_init");
+ return 1;
+ }
+
+ if (fio_option_is_set(&td->o, ioprio_class))
+ ld->ioprio_class_set = true;
+ if (fio_option_is_set(&td->o, ioprio))
+ ld->ioprio_set = true;
+
return 0;
}