X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=engines%2Fio_uring.c;h=e5edfcd25c60cd59062290e751fc7f5e877c064d;hp=5fda2fc96fea8f88914083f5cb9e5d9ff962d580;hb=34d6090ea96ddaa4e2a0f914939e2b8ff7008bec;hpb=919850d217e139b4f82f9798893cc816a7435e44

diff --git a/engines/io_uring.c b/engines/io_uring.c
index 5fda2fc9..e5edfcd2 100644
--- a/engines/io_uring.c
+++ b/engines/io_uring.c
@@ -50,6 +50,8 @@ struct ioring_data {
 
 	struct io_u **io_u_index;
 
+	int *fds;
+
 	struct io_sq_ring sq_ring;
 	struct io_uring_sqe *sqes;
 	struct iovec *iovecs;
@@ -62,9 +64,6 @@ struct ioring_data {
 	int cq_ring_off;
 	unsigned iodepth;
 
-	uint64_t cachehit;
-	uint64_t cachemiss;
-
 	struct ioring_mmap mmap[3];
 };
 
@@ -72,6 +71,7 @@ struct ioring_options {
 	void *pad;
 	unsigned int hipri;
 	unsigned int fixedbufs;
+	unsigned int registerfiles;
 	unsigned int sqpoll_thread;
 	unsigned int sqpoll_set;
 	unsigned int sqpoll_cpu;
@@ -94,7 +94,7 @@ static struct fio_option options[] = {
 		.off1	= offsetof(struct ioring_options, hipri),
 		.help	= "Use polled IO completions",
 		.category = FIO_OPT_C_ENGINE,
-		.group	= FIO_OPT_G_LIBAIO,
+		.group	= FIO_OPT_G_IOURING,
 	},
 	{
 		.name	= "fixedbufs",
@@ -103,7 +103,16 @@ static struct fio_option options[] = {
 		.off1	= offsetof(struct ioring_options, fixedbufs),
 		.help	= "Pre map IO buffers",
 		.category = FIO_OPT_C_ENGINE,
-		.group	= FIO_OPT_G_LIBAIO,
+		.group	= FIO_OPT_G_IOURING,
+	},
+	{
+		.name	= "registerfiles",
+		.lname	= "Register file set",
+		.type	= FIO_OPT_STR_SET,
+		.off1	= offsetof(struct ioring_options, registerfiles),
+		.help	= "Pre-open/register files",
+		.category = FIO_OPT_C_ENGINE,
+		.group	= FIO_OPT_G_IOURING,
 	},
 	{
 		.name	= "sqthread_poll",
@@ -112,7 +121,7 @@ static struct fio_option options[] = {
 		.off1	= offsetof(struct ioring_options, sqpoll_thread),
 		.help	= "Offload submission/completion to kernel thread",
 		.category = FIO_OPT_C_ENGINE,
-		.group	= FIO_OPT_G_LIBAIO,
+		.group	= FIO_OPT_G_IOURING,
 	},
 	{
 		.name	= "sqthread_poll_cpu",
@@ -121,7 +130,7 @@ static struct fio_option options[] = {
 		.cb	= fio_ioring_sqpoll_cb,
 		.help	= "What CPU to run SQ thread polling on",
 		.category = FIO_OPT_C_ENGINE,
-		.group	= FIO_OPT_G_LIBAIO,
+		.group	= FIO_OPT_G_IOURING,
 	},
 	{
 		.name	= NULL,
@@ -132,7 +141,7 @@ 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,
-			min_complete, flags);
+			min_complete, flags, NULL, 0);
 }
 
 static int fio_ioring_prep(struct thread_data *td, struct io_u *io_u)
@@ -143,10 +152,16 @@ static int fio_ioring_prep(struct thread_data *td, struct io_u *io_u)
 	struct io_uring_sqe *sqe;
 
 	sqe = &ld->sqes[io_u->index];
-	sqe->fd = f->fd;
-	sqe->flags = 0;
-	sqe->ioprio = 0;
-	sqe->buf_index = 0;
+
+	/* 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;
+	}
 
 	if (io_u->ddir == DDIR_READ || io_u->ddir == DDIR_WRITE) {
 		if (o->fixedbufs) {
@@ -167,10 +182,16 @@ static int fio_ioring_prep(struct thread_data *td, struct io_u *io_u)
 		}
 		sqe->off = io_u->offset;
 	} else if (ddir_sync(io_u->ddir)) {
-		sqe->fsync_flags = 0;
-		if (io_u->ddir == DDIR_DATASYNC)
-			sqe->fsync_flags |= IORING_FSYNC_DATASYNC;
-		sqe->opcode = IORING_OP_FSYNC;
+		if (io_u->ddir == DDIR_SYNC_FILE_RANGE) {
+			sqe->off = f->first_write;
+			sqe->len = f->last_write - f->first_write;
+			sqe->sync_range_flags = td->o.sync_file_range;
+			sqe->opcode = IORING_OP_SYNC_FILE_RANGE;
+		} else {
+			if (io_u->ddir == DDIR_DATASYNC)
+				sqe->fsync_flags |= IORING_FSYNC_DATASYNC;
+			sqe->opcode = IORING_OP_FSYNC;
+		}
 	}
 
 	sqe->user_data = (unsigned long) io_u;
@@ -197,13 +218,6 @@ static struct io_u *fio_ioring_event(struct thread_data *td, int event)
 	} else
 		io_u->error = 0;
 
-	if (io_u->ddir == DDIR_READ) {
-		if (cqe->flags & IOCQE_FLAG_CACHEHIT)
-			ld->cachehit++;
-		else
-			ld->cachemiss++;
-	}
-
 	return io_u;
 }
 
@@ -243,6 +257,8 @@ static int fio_ioring_getevents(struct thread_data *td, unsigned int min,
 		r = fio_ioring_cqring_reap(td, events, max);
 		if (r) {
 			events += r;
+			if (actual_min != 0)
+				actual_min -= r;
 			continue;
 		}
 
@@ -250,7 +266,7 @@ static int fio_ioring_getevents(struct thread_data *td, unsigned int min,
 			r = io_uring_enter(ld, 0, actual_min,
 						IORING_ENTER_GETEVENTS);
 			if (r < 0) {
-				if (errno == EAGAIN)
+				if (errno == EAGAIN || errno == EINTR)
 					continue;
 				td_verror(td, errno, "io_uring_enter");
 				break;
@@ -340,7 +356,8 @@ static int fio_ioring_commit(struct thread_data *td)
 
 		read_barrier();
 		if (*ring->flags & IORING_SQ_NEED_WAKEUP)
-			io_uring_enter(ld, ld->queued, 0, 0);
+			io_uring_enter(ld, ld->queued, 0,
+					IORING_ENTER_SQ_WAKEUP);
 		ld->queued = 0;
 		return 0;
 	}
@@ -360,7 +377,7 @@ static int fio_ioring_commit(struct thread_data *td)
 			io_u_mark_submit(td, ret);
 			continue;
 		} else {
-			if (errno == EAGAIN) {
+			if (errno == EAGAIN || errno == EINTR) {
 				ret = fio_ioring_cqring_reap(td, 0, ld->queued);
 				if (ret)
 					continue;
@@ -390,14 +407,12 @@ static void fio_ioring_cleanup(struct thread_data *td)
 	struct ioring_data *ld = td->io_ops_data;
 
 	if (ld) {
-		td->ts.cachehit += ld->cachehit;
-		td->ts.cachemiss += ld->cachemiss;
-
 		if (!(td->flags & TD_F_CHILD))
 			fio_ioring_unmap(ld);
 
 		free(ld->io_u_index);
 		free(ld->iovecs);
+		free(ld->fds);
 		free(ld);
 	}
 }
@@ -462,15 +477,6 @@ static int fio_ioring_queue_init(struct thread_data *td)
 		}
 	}
 
-	if (o->fixedbufs) {
-		struct rlimit rlim = {
-			.rlim_cur = RLIM_INFINITY,
-			.rlim_max = RLIM_INFINITY,
-		};
-
-		setrlimit(RLIMIT_MEMLOCK, &rlim);
-	}
-
 	ret = syscall(__NR_sys_io_uring_setup, depth, &p);
 	if (ret < 0)
 		return ret;
@@ -478,6 +484,14 @@ static int fio_ioring_queue_init(struct thread_data *td)
 	ld->ring_fd = ret;
 
 	if (o->fixedbufs) {
+		struct rlimit rlim = {
+			.rlim_cur = RLIM_INFINITY,
+			.rlim_max = RLIM_INFINITY,
+		};
+
+		if (setrlimit(RLIMIT_MEMLOCK, &rlim) < 0)
+			return -1;
+
 		ret = syscall(__NR_sys_io_uring_register, ld->ring_fd,
 				IORING_REGISTER_BUFFERS, ld->iovecs, depth);
 		if (ret < 0)
@@ -487,9 +501,50 @@ static int fio_ioring_queue_init(struct thread_data *td)
 	return fio_ioring_mmap(ld, &p);
 }
 
+static int fio_ioring_register_files(struct thread_data *td)
+{
+	struct ioring_data *ld = td->io_ops_data;
+	struct fio_file *f;
+	unsigned int i;
+	int ret;
+
+	ld->fds = calloc(td->o.nr_files, sizeof(int));
+
+	for_each_file(td, f, i) {
+		ret = generic_open_file(td, f);
+		if (ret)
+			goto err;
+		ld->fds[i] = f->fd;
+		f->engine_pos = i;
+	}
+
+	ret = syscall(__NR_sys_io_uring_register, ld->ring_fd,
+			IORING_REGISTER_FILES, ld->fds, td->o.nr_files);
+	if (ret) {
+err:
+		free(ld->fds);
+		ld->fds = NULL;
+	}
+
+	/*
+	 * Pretend the file is closed again, and really close it if we hit
+	 * an error.
+	 */
+	for_each_file(td, f, i) {
+		if (ret) {
+			int fio_unused ret2;
+			ret2 = generic_close_file(td, f);
+		} else
+			f->fd = -1;
+	}
+
+	return ret;
+}
+
 static int fio_ioring_post_init(struct thread_data *td)
 {
 	struct ioring_data *ld = td->io_ops_data;
+	struct ioring_options *o = td->eo;
 	struct io_u *io_u;
 	int err, i;
 
@@ -507,6 +562,15 @@ static int fio_ioring_post_init(struct thread_data *td)
 		return 1;
 	}
 
+	printf("files=%d\n", o->registerfiles);
+	if (o->registerfiles) {
+		err = fio_ioring_register_files(td);
+		if (err) {
+			td_verror(td, errno, "ioring_register_files");
+			return 1;
+		}
+	}
+
 	return 0;
 }
 
@@ -517,8 +581,19 @@ static unsigned roundup_pow2(unsigned depth)
 
 static int fio_ioring_init(struct thread_data *td)
 {
+	struct ioring_options *o = td->eo;
 	struct ioring_data *ld;
 
+	/* sqthread submission requires registered files */
+	if (o->sqpoll_thread)
+		o->registerfiles = 1;
+
+	if (o->registerfiles && td->o.nr_files != td->o.open_files) {
+		log_err("fio: io_uring registered files require nr_files to "
+			"be identical to open_files\n");
+		return 1;
+	}
+
 	ld = calloc(1, sizeof(*ld));
 
 	/* ring depth must be a power-of-2 */
@@ -541,9 +616,34 @@ static int fio_ioring_io_u_init(struct thread_data *td, struct io_u *io_u)
 	return 0;
 }
 
+static int fio_ioring_open_file(struct thread_data *td, struct fio_file *f)
+{
+	struct ioring_data *ld = td->io_ops_data;
+	struct ioring_options *o = td->eo;
+
+	if (!ld || !o->registerfiles)
+		return generic_open_file(td, f);
+
+	f->fd = ld->fds[f->engine_pos];
+	return 0;
+}
+
+static int fio_ioring_close_file(struct thread_data *td, struct fio_file *f)
+{
+	struct ioring_data *ld = td->io_ops_data;
+	struct ioring_options *o = td->eo;
+
+	if (!ld || !o->registerfiles)
+		return generic_close_file(td, f);
+
+	f->fd = -1;
+	return 0;
+}
+
 static struct ioengine_ops ioengine = {
 	.name			= "io_uring",
 	.version		= FIO_IOOPS_VERSION,
+	.flags			= FIO_ASYNCIO_SYNC_TRIM,
 	.init			= fio_ioring_init,
 	.post_init		= fio_ioring_post_init,
 	.io_u_init		= fio_ioring_io_u_init,
@@ -553,8 +653,8 @@ static struct ioengine_ops ioengine = {
 	.getevents		= fio_ioring_getevents,
 	.event			= fio_ioring_event,
 	.cleanup		= fio_ioring_cleanup,
-	.open_file		= generic_open_file,
-	.close_file		= generic_close_file,
+	.open_file		= fio_ioring_open_file,
+	.close_file		= fio_ioring_close_file,
 	.get_file_size		= generic_get_file_size,
 	.options		= options,
 	.option_struct_size	= sizeof(struct ioring_options),