On Thu, Jul 3, 2014 at 8:37 PM, Stefan Hajnoczi <stefa...@gmail.com> wrote:
> On Wed, Jul 02, 2014 at 08:18:48PM +0800, Ming Lei wrote:
>> diff --git a/hw/block/dataplane/virtio-blk.c
>> b/hw/block/dataplane/virtio-blk.c
>> index c10b7b7..82bb276 100644
>> --- a/hw/block/dataplane/virtio-blk.c
>> +++ b/hw/block/dataplane/virtio-blk.c
>> @@ -201,6 +201,9 @@ static void do_flush_cmd(VirtIOBlockDataPlane *s,
>> VirtQueueElement *elem,
>> req->elem = elem;
>> req->inhdr = inhdr;
>>
>> + /* flush IOs queued first */
>> + bdrv_flush_io_queue(s->blk->conf.bs);
>> +
>> bdrv_aio_flush(s->blk->conf.bs, complete_flush, req);
>> }
>>
>> @@ -289,6 +292,7 @@ static void handle_notify(EventNotifier *e)
>> int ret;
>>
>> event_notifier_test_and_clear(&s->host_notifier);
>> + bdrv_io_plug(s->blk->conf.bs);
>> for (;;) {
>> /* Disable guest->host notifies to avoid unnecessary vmexits */
>> vring_disable_notification(s->vdev, &s->vring);
>> @@ -322,6 +326,7 @@ static void handle_notify(EventNotifier *e)
>> break;
>> }
>> }
>> + bdrv_io_unplug(s->blk->conf.bs);
>> }
>
> Might as well do the same for non-dataplane in hw/block/virtio-blk.c?
It may need further optimization for non-dataplane, and I saw
the message of 'main-loop: WARNING: I/O thread spun for 1000
iterations' when I enabled for non-dataplane.
The queue depth should be decreased for non-dataplane.
Thanks,
