This patch hooks AIO support into virtio-blk, allowing for faster IO.
Signed-off-by: Sasha Levin <levinsasha928@xxxxxxxxx>
---
tools/kvm/disk/core.c | 39 ++++++++++++++++++++++++++++++++++-
tools/kvm/disk/qcow.c | 4 +++
tools/kvm/disk/raw.c | 36 +++++++++++++++++++++++++++++++-
tools/kvm/include/kvm/disk-image.h | 5 ++++
tools/kvm/virtio/blk.c | 16 +++++---------
5 files changed, 86 insertions(+), 14 deletions(-)
diff --git a/tools/kvm/disk/core.c b/tools/kvm/disk/core.c
index a5e6713..cc2a1f6 100644
--- a/tools/kvm/disk/core.c
+++ b/tools/kvm/disk/core.c
@@ -1,8 +1,32 @@
#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;
+#ifdef CONFIG_HAS_AIO
+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;
+}
+#endif
+
struct disk_image *disk_image__new(int fd, u64 size, struct disk_image_operations *ops, int use_mmap)
{
struct disk_image *disk;
@@ -26,6 +50,16 @@ struct disk_image *disk_image__new(int fd, u64 size, struct disk_image_operation
}
}
+#ifdef CONFIG_HAS_AIO
+ 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 thread");
+ }
+#endif
return disk;
}
@@ -87,6 +121,7 @@ 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 +186,7 @@ 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;
@@ -183,7 +218,7 @@ 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 51e8318..e979046 100644
--- a/tools/kvm/disk/raw.c
+++ b/tools/kvm/disk/raw.c
@@ -1,11 +1,20 @@
#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)
{
u64 offset = sector << SECTOR_SHIFT;
+#ifdef CONFIG_HAS_AIO
+ struct iocb iocb;
+
+ return aio_preadv(disk->ctx, &iocb, disk->fd, iov, iovcount, offset,
+ disk->evt, param);
+#else
return preadv_in_full(disk->fd, iov, iovcount, offset);
+#endif
}
ssize_t raw_image__write_sector(struct disk_image *disk, u64 sector, const struct iovec *iov,
@@ -13,7 +22,14 @@ ssize_t raw_image__write_sector(struct disk_image *disk, u64 sector, const struc
{
u64 offset = sector << SECTOR_SHIFT;
+#ifdef CONFIG_HAS_AIO
+ struct iocb iocb;
+
+ return aio_pwritev(disk->ctx, &iocb, disk->fd, iov, iovcount, offset,
+ disk->evt, param);
+#else
return pwritev_in_full(disk->fd, iov, iovcount, offset);
+#endif
}
ssize_t raw_image__read_sector_mmap(struct disk_image *disk, u64 sector, const struct iovec *iov,
@@ -59,6 +75,12 @@ int raw_image__close(struct disk_image *disk)
if (disk->priv != MAP_FAILED)
ret = munmap(disk->priv, disk->size);
+ close(disk->evt);
+
+#ifdef CONFIG_HAS_VIRTIO
+ io_destroy(disk->ctx);
+#endif
+
return ret;
}
@@ -83,6 +105,7 @@ struct disk_image_operations ro_ops_nowrite = {
struct disk_image *raw_image__probe(int fd, struct stat *st, bool readonly)
{
+ struct disk_image *disk;
if (readonly) {
/*
@@ -94,8 +117,12 @@ struct disk_image *raw_image__probe(int fd, struct stat *st, bool readonly)
disk = disk_image__new(fd, st->st_size, &ro_ops, DISK_IMAGE_MMAP);
if (disk == NULL) {
ro_ops = raw_image_regular_ops;
- ro_ops.write_sector = NULL;
+
disk = disk_image__new(fd, st->st_size, &ro_ops_nowrite, DISK_IMAGE_REGULAR);
+#ifdef CONFIG_HAS_VIRTIO
+ if (disk)
+ disk->async = 1;
+#endif
}
return disk;
@@ -103,6 +130,11 @@ 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);
+#ifdef CONFIG_HAS_VIRTIO
+ if (disk)
+ disk->async = 1;
+#endif
+ return disk;
}
}
diff --git a/tools/kvm/include/kvm/disk-image.h b/tools/kvm/include/kvm/disk-image.h
index 632e54d..56c08da 100644
--- a/tools/kvm/include/kvm/disk-image.h
+++ b/tools/kvm/include/kvm/disk-image.h
@@ -46,6 +46,11 @@ struct disk_image {
void *priv;
void *disk_req_cb_param;
void (*disk_req_cb)(void *param, long len);
+ bool async;
+ int evt;
+#ifdef CONFIG_HAS_AIO
+ io_context_t ctx;
+#endif
};
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 5969d27..223e79f 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;
@@ -86,8 +85,8 @@ 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);
@@ -103,17 +102,14 @@ static void virtio_blk_do_io_request(struct kvm *kvm, struct blk_dev_req *req)
struct virtio_blk_outhdr *req_hdr;
ssize_t block_cnt;
struct blk_dev *bdev;
- struct virt_queue *queue;
struct iovec *iov;
- u16 out, in, head;
+ u16 out, in;
block_cnt = -1;
bdev = req->bdev;
- queue = req->vq;
iov = req->iov;
out = req->out;
in = req->in;
- head = req->head;
req_hdr = iov[0].iov_base;
switch (req_hdr->type) {
--
1.7.7.1
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
