On Sun, 13 Nov 2011 23:03:14 +0200, "Michael S. Tsirkin" [off-list ref] wrote:

On Thu, Nov 03, 2011 at 06:12:53PM +1030, Rusty Russell wrote:

quoted

A virtio driver does virtqueue_add_buf() multiple times before finally
calling virtqueue_kick(); previously we only exposed the added buffers
in the virtqueue_kick() call.  This means we don't need a memory
barrier in virtqueue_add_buf(), but it reduces concurrency as the
device (ie. host) can't see the buffers until the kick.

Signed-off-by: Rusty Russell <redacted>

In the past I played with a patch like this, but I didn't see a
performance gain either way. Do you see any gain?

No, but I haven't run it on real hardware.  lguest may see a win with
this in theory, since the virtqueue processing is fully async, so I'll
run some tests.

I'm a bit concerned that with this patch, a buggy driver that
adds more than 2^16 descriptors without a kick
would seem to work sometimes. Let's add WARN_ON(vq->num_added > (1 << 16))?

Hmm, I guess it could wait for the add to fail before doing a kick, but
noone does that at the moment, so I've added a slight variant:

	WARN_ON(vq->num_added > vq->vring.num);

Thanks,
Rusty.