Linux Virtualization - Mar 2016 - [PATCH net-next] virtio_net: replace netdev_alloc_skb_ip_align() with napi_alloc_skb()

This gives small but noticeable rx performance improvement (2-3%)
and will allow exploiting future napi improvement.

Signed-off-by: Paolo Abeni <pabeni at redhat.com>
---
 drivers/net/virtio_net.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index fb0eae4..49d84e5 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -260,7 +260,7 @@ static struct sk_buff *page_to_skb(struct virtnet_info *vi,
 	p = page_address(page) + offset;
 
 	/* copy small packet so we can reuse these pages for small data */
-	skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN);
+	skb = napi_alloc_skb(&rq->napi, GOOD_COPY_LEN);
 	if (unlikely(!skb))
 		return NULL;
 
-- 
1.8.3.1

On Thu, Mar 17, 2016 at 7:44 AM, Paolo Abeni <pabeni at redhat.com> wrote:
> This gives small but noticeable rx performance improvement (2-3%)
> and will allow exploiting future napi improvement.
>
> Signed-off-by: Paolo Abeni <pabeni at redhat.com>
>
Worked in my tests, though the performance win was in the noise (+0.6% -
+1% bandwidth).
What tests did you see a 2-3% win on?

Do you think its worth modifying add_recvbuf_small() to use
napi_alloc_skb() when called from
Rx NAPI (virtnet_receive)?

Tested-by: Venkatesh Srinivas <venkateshs at google.com>

Thanks,
-- vs;
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.linuxfoundation.org/pipermail/virtualization/attachments/20160317/0f394f02/attachment.html>

On Thu, Mar 17, 2016 at 7:44 AM, Paolo Abeni <pabeni at redhat.com>
wrote:
>
> This gives small but noticeable rx performance improvement (2-3%)
> and will allow exploiting future napi improvement.
>
> Signed-off-by: Paolo Abeni <pabeni at redhat.com>
> ---
Worked in my tests, though the performance win was in the noise (+0.6%
- +1% bandwidth).
What tests did you see a 2-3% win on?

Do you think its worth modifying add_recvbuf_small() to use
napi_alloc_skb() when called from
Rx NAPI (virtnet_receive)?

Tested-by: Venkatesh Srinivas <venkateshs at google.com>

Thanks,
-- vs;

On Thu, 2016-03-17 at 18:01 -0700, Venkatesh Srinivas
wrote:
> On Thu, Mar 17, 2016 at 7:44 AM, Paolo Abeni <pabeni at redhat.com>
wrote:
> >
> > This gives small but noticeable rx performance improvement (2-3%)
> > and will allow exploiting future napi improvement.
> >
> > Signed-off-by: Paolo Abeni <pabeni at redhat.com>
> > ---
> 
> Worked in my tests, though the performance win was in the noise (+0.6%
> - +1% bandwidth).
> What tests did you see a 2-3% win on?
I'm doing net2VM tests (i.e. the traffic generator is on a different
host) with guest tuned a bit for performance. i.e. cpu pinning, no
iptables, no dhclient (yes, that makes a big difference:
http://lists.openwall.net/netdev/2013/10/29/5)
> Do you think its worth modifying add_recvbuf_small() to use
> napi_alloc_skb() when called from
> Rx NAPI (virtnet_receive)?
Oops, I missed that invocation. Probably that path is not hit in my
test. I think it should be really worthy.

I'll send a v2.

The important thing, more than the current improvement, is allowing the
virtio_net driver to leverage future napi allocation improvement (i.e.
bulk alloc: http://www.spinics.net/lists/netdev/msg367568.html) which
should be a real win.

Paolo