[PATCH] virtio: don't always force a notification when ring is full

[PATCH] virtio: don't always force a notification when ring is full

[Rusty Russell]

We force notification when the ring is full, even if the host has
indicated it doesn't want to know.  This seemed like a good idea at
the time: if we fill the transmit ring, we should tell the host
immediately.

Unfortunately this logic also applies to the receiving ring, which is
refilled constantly.  We should introduce real notification thesholds
to replace this logic.  Meanwhile, removing the logic altogether breaks
the heuristics which KVM uses, so we use a hack: only notify if there are
outgoing parts of the new buffer.

It even makes a difference with lguest's crappy network implementation:
	1GB guest->host before: 30.0844
	1GB guest->host after: 27.6167

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

diff -r 89bf4894cc36 drivers/virtio/virtio_ring.c
--- a/drivers/virtio/virtio_ring.c	Mon Jun 16 14:31:30 2008 +1000
+++ b/drivers/virtio/virtio_ring.c	Wed Jun 18 16:15:49 2008 +1000
@@ -87,8 +87,11 @@ static int vring_add_buf(struct virtqueu
 	if (vq->num_free < out + in) {
 		pr_debug("Can't add buf len %i - avail = %i\n",
 			 out + in, vq->num_free);
-		/* We notify *even if* VRING_USED_F_NO_NOTIFY is set here. */
-		vq->notify(&vq->vq);
+		/* FIXME: for historical reasons, we force a notify here if
+		 * there are outgoing parts to the buffer.  Presumably the
+		 * host should service the ring ASAP. */
+		if (out)
+			vq->notify(&vq->vq);
 		END_USE(vq);
 		return -ENOSPC;
 	}

Re: [PATCH] virtio: don't always force a notification when ring is full

[Anthony Liguori]

Rusty Russell wrote:
> We force notification when the ring is full, even if the host has
> indicated it doesn't want to know.  This seemed like a good idea at
> the time: if we fill the transmit ring, we should tell the host
> immediately.
>
> Unfortunately this logic also applies to the receiving ring, which is
> refilled constantly.  We should introduce real notification thesholds
> to replace this logic.  Meanwhile, removing the logic altogether breaks
> the heuristics which KVM uses, so we use a hack: only notify if there are
> outgoing parts of the new buffer.
>
> It even makes a difference with lguest's crappy network implementation:
> 	1GB guest->host before: 30.0844
> 	1GB guest->host after: 27.6167
>   

A little ugly, but no more performance regression in KVM, so

Acked-by: Anthony Liguori <aliguori@us.ibm.com>

Regards,

Anthony Liguori

> Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
>
> diff -r 89bf4894cc36 drivers/virtio/virtio_ring.c
> --- a/drivers/virtio/virtio_ring.c	Mon Jun 16 14:31:30 2008 +1000
> +++ b/drivers/virtio/virtio_ring.c	Wed Jun 18 16:15:49 2008 +1000
> @@ -87,8 +87,11 @@ static int vring_add_buf(struct virtqueu
>  	if (vq->num_free < out + in) {
>  		pr_debug("Can't add buf len %i - avail = %i\n",
>  			 out + in, vq->num_free);
> -		/* We notify *even if* VRING_USED_F_NO_NOTIFY is set here. */
> -		vq->notify(&vq->vq);
> +		/* FIXME: for historical reasons, we force a notify here if
> +		 * there are outgoing parts to the buffer.  Presumably the
> +		 * host should service the ring ASAP. */
> +		if (out)
> +			vq->notify(&vq->vq);
>  		END_USE(vq);
>  		return -ENOSPC;
>  	}
>   

Re: [PATCH] virtio: don't always force a notification when ring is full

[Christian Borntraeger]

Am Donnerstag, 19. Juni 2008 schrieb Rusty Russell:
> We force notification when the ring is full, even if the host has
> indicated it doesn't want to know.  This seemed like a good idea at
> the time: if we fill the transmit ring, we should tell the host
> immediately.
> 
> Unfortunately this logic also applies to the receiving ring, which is
> refilled constantly.  We should introduce real notification thesholds
> to replace this logic.  Meanwhile, removing the logic altogether breaks
> the heuristics which KVM uses, so we use a hack: only notify if there are
> outgoing parts of the new buffer.


I did not yet finish the VIRTIO_F_NOTIFY_ON_EMPTY implementation for s390, so 
I cant tell for sure if it makes a difference.

Nevertheless, the patch looks good to me.