This patch hooks AIO support into virtio-blk, allowing for faster IO.
Signed-off-by: Sasha Levin <levinsasha...@gmail.com>
---
tools/kvm/disk/core.c | 35 +++++++++++++++++++++++++++++++++--
tools/kvm/disk/qcow.c | 4 ++++
tools/kvm/disk/raw.c | 19 +++++++++++++++----
tools/kvm/include/kvm/disk-image.h | 3 +++
tools/kvm/virtio/blk.c | 11 +++++------
5 files changed, 60 insertions(+), 12 deletions(-)
diff --git a/tools/kvm/disk/core.c b/tools/kvm/disk/core.c
index 557a1a2..17fe548 100644
--- a/tools/kvm/disk/core.c
+++ b/tools/kvm/disk/core.c
@@ -1,8 +1,30 @@
#include "kvm/disk-image.h"
#include "kvm/qcow.h"
+#include "kvm/virtio-blk.h"
+
+#include <sys/eventfd.h>
+#include <sys/poll.h>
+
+#define AIO_MAX 32
int debug_iodelay;
+static void *disk_image__thread(void *param)
+{
+ struct disk_image *disk = param;
+ u64 dummy;
+
+ while (read(disk->evt, &dummy, sizeof(dummy)) > 0) {
+ struct io_event event;
+ struct timespec notime = {0};
+
+ while (io_getevents(disk->ctx, 1, 1, &event, ¬ime) > 0)
+ disk->disk_req_cb(event.data, event.res);
+ }
+
+ return NULL;
+}
+
struct disk_image *disk_image__new(int fd, u64 size, struct
disk_image_operations *ops, int use_mmap)
{
struct disk_image *disk;
@@ -26,6 +48,14 @@ struct disk_image *disk_image__new(int fd, u64 size, struct
disk_image_operation
}
}
+ if (disk) {
+ pthread_t thread;
+
+ disk->evt = eventfd(0, 0);
+ io_setup(AIO_MAX, &disk->ctx);
+ if (pthread_create(&thread, NULL, disk_image__thread, disk) !=
0)
+ die("Failed starting IO řthread");
+ }
return disk;
}
@@ -87,6 +117,7 @@ struct disk_image **disk_image__open_all(const char
**filenames, bool *readonly,
goto error;
}
}
+
return disks;
error:
for (i = 0; i < count; i++)
@@ -151,7 +182,7 @@ ssize_t disk_image__read(struct disk_image *disk, u64
sector, const struct iovec
/* Do nothing */
}
- if (disk->disk_req_cb)
+ if (!disk->async && disk->disk_req_cb)
disk->disk_req_cb(param, total);
return total;
@@ -182,7 +213,7 @@ ssize_t disk_image__write(struct disk_image *disk, u64
sector, const struct iove
/* Do nothing */
}
- if (disk->disk_req_cb)
+ if (!disk->async && disk->disk_req_cb)
disk->disk_req_cb(param, total);
return total;
diff --git a/tools/kvm/disk/qcow.c b/tools/kvm/disk/qcow.c
index f9598de..87385df 100644
--- a/tools/kvm/disk/qcow.c
+++ b/tools/kvm/disk/qcow.c
@@ -1145,6 +1145,8 @@ static struct disk_image *qcow2_probe(int fd, bool
readonly)
if (!disk_image)
goto free_refcount_table;
+
+ disk_image->async = 1;
disk_image->priv = q;
return disk_image;
@@ -1276,6 +1278,8 @@ static struct disk_image *qcow1_probe(int fd, bool
readonly)
if (!disk_image)
goto free_l1_table;
+
+ disk_image->async = 1;
disk_image->priv = q;
return disk_image;
diff --git a/tools/kvm/disk/raw.c b/tools/kvm/disk/raw.c
index fa2e013..3612623 100644
--- a/tools/kvm/disk/raw.c
+++ b/tools/kvm/disk/raw.c
@@ -1,19 +1,25 @@
#include "kvm/disk-image.h"
+#include <libaio.h>
+
ssize_t raw_image__read_sector(struct disk_image *disk, u64 sector, const
struct iovec *iov,
int iovcount, void *param)
{
+ struct iocb iocb;
u64 offset = sector << SECTOR_SHIFT;
- return preadv_in_full(disk->fd, iov, iovcount, offset);
+ return aio_preadv(disk->ctx, &iocb, disk->fd, iov, iovcount, offset,
+ disk->evt, param);
}
ssize_t raw_image__write_sector(struct disk_image *disk, u64 sector, const
struct iovec *iov,
int iovcount, void *param)
{
+ struct iocb iocb;
u64 offset = sector << SECTOR_SHIFT;
- return pwritev_in_full(disk->fd, iov, iovcount, offset);
+ return aio_pwritev(disk->ctx, &iocb, disk->fd, iov, iovcount, offset,
+ disk->evt, param);
}
ssize_t raw_image__read_sector_mmap(struct disk_image *disk, u64 sector, const
struct iovec *iov,
@@ -72,13 +78,13 @@ static struct disk_image_operations raw_image_regular_ops =
{
struct disk_image *raw_image__probe(int fd, struct stat *st, bool readonly)
{
+ struct disk_image *disk;
if (readonly) {
/*
* Use mmap's MAP_PRIVATE to implement non-persistent write
* FIXME: This does not work on 32-bit host.
*/
- struct disk_image *disk;
struct disk_image_operations ro_ops = {
.read_sector = raw_image__read_sector_mmap,
.write_sector = raw_image__write_sector_mmap,
@@ -90,6 +96,8 @@ struct disk_image *raw_image__probe(int fd, struct stat *st,
bool readonly)
ro_ops = raw_image_regular_ops;
ro_ops.write_sector = NULL;
disk = disk_image__new(fd, st->st_size, &ro_ops,
DISK_IMAGE_REGULAR);
+ if (disk)
+ disk->async = 1;
}
return disk;
@@ -97,6 +105,9 @@ struct disk_image *raw_image__probe(int fd, struct stat *st,
bool readonly)
/*
* Use read/write instead of mmap
*/
- return disk_image__new(fd, st->st_size, &raw_image_regular_ops,
DISK_IMAGE_REGULAR);
+ disk = disk_image__new(fd, st->st_size, &raw_image_regular_ops,
DISK_IMAGE_REGULAR);
+ if (disk)
+ disk->async = 1;
+ return disk;
}
}
diff --git a/tools/kvm/include/kvm/disk-image.h
b/tools/kvm/include/kvm/disk-image.h
index bcb3396..5a14eab 100644
--- a/tools/kvm/include/kvm/disk-image.h
+++ b/tools/kvm/include/kvm/disk-image.h
@@ -46,6 +46,9 @@ struct disk_image {
void *priv;
void *disk_req_cb_param;
void (*disk_req_cb)(void *param, long len);
+ bool async;
+ int evt;
+ io_context_t ctx;
};
struct disk_image *disk_image__open(const char *filename, bool readonly);
diff --git a/tools/kvm/virtio/blk.c b/tools/kvm/virtio/blk.c
index c167dc9..3d31228d 100644
--- a/tools/kvm/virtio/blk.c
+++ b/tools/kvm/virtio/blk.c
@@ -2,7 +2,6 @@
#include "kvm/virtio-pci-dev.h"
#include "kvm/disk-image.h"
-#include "kvm/virtio.h"
#include "kvm/mutex.h"
#include "kvm/util.h"
#include "kvm/kvm.h"
@@ -11,23 +10,23 @@
#include "kvm/ioeventfd.h"
#include "kvm/guest_compat.h"
#include "kvm/virtio-pci.h"
+#include "kvm/virtio.h"
#include <linux/virtio_ring.h>
#include <linux/virtio_blk.h>
-
#include <linux/kernel.h>
#include <linux/list.h>
#include <linux/types.h>
#include <pthread.h>
#define VIRTIO_BLK_MAX_DEV 4
-#define NUM_VIRT_QUEUES 1
-#define VIRTIO_BLK_QUEUE_SIZE 128
/*
* the header and status consume too entries
*/
#define DISK_SEG_MAX (VIRTIO_BLK_QUEUE_SIZE - 2)
+#define VIRTIO_BLK_QUEUE_SIZE 128
+#define NUM_VIRT_QUEUES 1
struct blk_dev_req {
struct list_head list;
@@ -84,8 +83,8 @@ void virtio_blk_complete(void *param, long len)
u8 *status;
/* status */
- status = req->iov[req->out + req->in - 1].iov_base;
- *status = (len < 0) ? VIRTIO_BLK_S_IOERR :
VIRTIO_BLK_S_OK;
+ status = req->iov[req->out + req->in - 1].iov_base;
+ *status = (len < 0) ? VIRTIO_BLK_S_IOERR : VIRTIO_BLK_S_OK;
mutex_lock(&bdev->mutex);
virt_queue__set_used_elem(req->vq, req->head, len);
--
1.7.7.1
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
