[PATCH v2 1/1] VSOCK: remove more space available check filling TX vq

From: Gerard Garcia <redacted>

Remove unnecessary use of enable/disable callback notifications
and the incorrect more space available check.

The virtio_transport_tx_work handles when the TX virtqueue
has more buffers available.

Signed-off-by: Gerard Garcia <redacted>
Acked-by: Stefan Hajnoczi <stefanha@redhat.com>
---

v2:
 * Comment style.

 net/vmw_vsock/virtio_transport.c | 10 +++-------
 1 file changed, 3 insertions(+), 7 deletions(-)

diff --git a/net/vmw_vsock/virtio_transport.c b/net/vmw_vsock/virtio_transport.c
index 699dfab..936d7ee 100644
--- a/net/vmw_vsock/virtio_transport.c
+++ b/net/vmw_vsock/virtio_transport.c

@@ -87,9 +87,6 @@ virtio_transport_send_pkt_work(struct work_struct *work)
 
 	vq = vsock->vqs[VSOCK_VQ_TX];
 
-	/* Avoid unnecessary interrupts while we're processing the ring */
-	virtqueue_disable_cb(vq);
-
 	for (;;) {
 		struct virtio_vsock_pkt *pkt;
 		struct scatterlist hdr, buf, *sgs[2];

@@ -99,7 +96,6 @@ virtio_transport_send_pkt_work(struct work_struct *work)
 		spin_lock_bh(&vsock->send_pkt_list_lock);
 		if (list_empty(&vsock->send_pkt_list)) {
 			spin_unlock_bh(&vsock->send_pkt_list_lock);
-			virtqueue_enable_cb(vq);
 			break;
 		}
 

@@ -118,13 +114,13 @@ virtio_transport_send_pkt_work(struct work_struct *work)
 		}
 
 		ret = virtqueue_add_sgs(vq, sgs, out_sg, in_sg, pkt, GFP_KERNEL);
+		/* Usually this means that there is no more space available in
+		 * the vq
+		 */
 		if (ret < 0) {
 			spin_lock_bh(&vsock->send_pkt_list_lock);
 			list_add(&pkt->list, &vsock->send_pkt_list);
 			spin_unlock_bh(&vsock->send_pkt_list_lock);
-
-			if (!virtqueue_enable_cb(vq) && ret == -ENOSPC)
-				continue; /* retry now that we have more space */
 			break;
 		}
 

-- 
2.9.1

From: ggarcia@abra.uab.cat
Date: Wed, 10 Aug 2016 17:24:34 +0200

From: Gerard Garcia <redacted>

Remove unnecessary use of enable/disable callback notifications
and the incorrect more space available check.

The virtio_transport_tx_work handles when the TX virtqueue
has more buffers available.

Signed-off-by: Gerard Garcia <redacted>
Acked-by: Stefan Hajnoczi <stefanha@redhat.com>

This does not apply cleanly to the current net GIT tree.

On 08/13/2016 02:31 AM, David Miller wrote:

From: ggarcia@abra.uab.cat
Date: Wed, 10 Aug 2016 17:24:34 +0200

quoted

From: Gerard Garcia <redacted>

Remove unnecessary use of enable/disable callback notifications
and the incorrect more space available check.

The virtio_transport_tx_work handles when the TX virtqueue
has more buffers available.

Signed-off-by: Gerard Garcia <redacted>
Acked-by: Stefan Hajnoczi <stefanha@redhat.com>

This does not apply cleanly to the current net GIT tree.

I'm sorry, I should have said that it applies over the mst vhost tree.

On Wed, Aug 10, 2016 at 05:24:34PM +0200, ggarcia@abra.uab.cat wrote:

From: Gerard Garcia <redacted>

Remove unnecessary use of enable/disable callback notifications
and the incorrect more space available check.

The virtio_transport_tx_work handles when the TX virtqueue
has more buffers available.

Signed-off-by: Gerard Garcia <redacted>
Acked-by: Stefan Hajnoczi <stefanha@redhat.com>

Generally enable/disable is a worthwhile optimization,
but if Stefan wants to keep code simple for now,
I don't have a problem with that.

On Mon, Aug 15, 2016 at 02:16:51AM +0300, Michael S. Tsirkin wrote:

On Wed, Aug 10, 2016 at 05:24:34PM +0200, ggarcia@abra.uab.cat wrote:

quoted

From: Gerard Garcia <redacted>

Remove unnecessary use of enable/disable callback notifications
and the incorrect more space available check.

The virtio_transport_tx_work handles when the TX virtqueue
has more buffers available.

Signed-off-by: Gerard Garcia <redacted>
Acked-by: Stefan Hajnoczi <stefanha@redhat.com>

Generally enable/disable is a worthwhile optimization,
but if Stefan wants to keep code simple for now,
I don't have a problem with that.

Enabling/disabling callbacks is inappropriate in this function because
its job is *submitting* buffers rather than *reclaiming* completed
buffers.  Oops, I'm not sure how this mistake crept in :).

The callback suppression should only be done around the loop that
reclaims buffers.  Doing it in the wrong place leads to hangs if
interrupts are lost.

Stefan