| Message ID | 1552658678-7816-1-git-send-email-lizhengui@huawei.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | vpc: unlock Coroutine lock to make IO submit Concurrently | expand |
On 15/03/19 15:04, Zhengui li wrote: > Concurrent IO becomes serial IO because of the qemu Coroutine lock, > which reduce IO performance severely. > > So unlock Coroutine lock before bdrv_co_pwritev and > bdrv_co_preadv to fix it. > > Signed-off-by: Zhengui li <lizhengui@huawei.com> > --- > block/vpc.c | 4 ++++ > 1 file changed, 4 insertions(+) > > diff --git a/block/vpc.c b/block/vpc.c > index 52ab717..1133855 100644 > --- a/block/vpc.c > +++ b/block/vpc.c > @@ -639,8 +639,10 @@ vpc_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes, > qemu_iovec_reset(&local_qiov); > qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes); > > + qemu_co_mutex_unlock(&s->lock); > ret = bdrv_co_preadv(bs->file, image_offset, n_bytes, > &local_qiov, 0); > + qemu_co_mutex_lock(&s->lock); > if (ret < 0) { > goto fail; > } > @@ -697,8 +699,10 @@ vpc_co_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes, > qemu_iovec_reset(&local_qiov); > qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes); > > + qemu_co_mutex_unlock(&s->lock); > ret = bdrv_co_pwritev(bs->file, image_offset, n_bytes, > &local_qiov, 0); > + qemu_co_mutex_lock(&s->lock); > if (ret < 0) { > goto fail; > } > This should be okay, because vpc.c is somewhat simple-minded and it doesn't recycle unused blocks in the middle of the file. Reviewed-by: Paolo Bonzini <pbonzini@redhat.com> Paolo
Am 15.03.2019 um 15:04 hat Zhengui li geschrieben: > Concurrent IO becomes serial IO because of the qemu Coroutine lock, > which reduce IO performance severely. > > So unlock Coroutine lock before bdrv_co_pwritev and > bdrv_co_preadv to fix it. > > Signed-off-by: Zhengui li <lizhengui@huawei.com> Thanks, applied to the block-next branch for 4.1. Kevin
diff --git a/block/vpc.c b/block/vpc.c index 52ab717..1133855 100644 --- a/block/vpc.c +++ b/block/vpc.c @@ -639,8 +639,10 @@ vpc_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes, qemu_iovec_reset(&local_qiov); qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes); + qemu_co_mutex_unlock(&s->lock); ret = bdrv_co_preadv(bs->file, image_offset, n_bytes, &local_qiov, 0); + qemu_co_mutex_lock(&s->lock); if (ret < 0) { goto fail; } @@ -697,8 +699,10 @@ vpc_co_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes, qemu_iovec_reset(&local_qiov); qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes); + qemu_co_mutex_unlock(&s->lock); ret = bdrv_co_pwritev(bs->file, image_offset, n_bytes, &local_qiov, 0); + qemu_co_mutex_lock(&s->lock); if (ret < 0) { goto fail; }
Concurrent IO becomes serial IO because of the qemu Coroutine lock, which reduce IO performance severely. So unlock Coroutine lock before bdrv_co_pwritev and bdrv_co_preadv to fix it. Signed-off-by: Zhengui li <lizhengui@huawei.com> --- block/vpc.c | 4 ++++ 1 file changed, 4 insertions(+)