[Intel-wired-lan] [PATCH bpf-next v2 06/10] xsk: propagate napi_id to XDP socket Rx path

[Michael S. Tsirkin]

On Mon, Nov 16, 2020 at 02:24:56PM +0100, Björn Töpel wrote:
> On 2020-11-16 13:42, Michael S. Tsirkin wrote:
> > On Mon, Nov 16, 2020 at 01:01:40PM +0100, Björn Töpel wrote:
> > > 
> > > On 2020-11-16 12:55, Michael S. Tsirkin wrote:
> > > > On Mon, Nov 16, 2020 at 12:04:12PM +0100, BjÃf¶rn TÃf¶pel wrote:
> > > > > From: BjÃf¶rn TÃf¶pel <bjorn.topel at intel.com>
> > > > > 
> > > > > Add napi_id to the xdp_rxq_info structure, and make sure the XDP
> > > > > socket pick up the napi_id in the Rx path. The napi_id is used to find
> > > > > the corresponding NAPI structure for socket busy polling.
> > > > > 
> > > > > Acked-by: Ilias Apalodimas <ilias.apalodimas at linaro.org>
> > > > > Signed-off-by: BjÃf¶rn TÃf¶pel <bjorn.topel at intel.com>
> > > > 
> > > > A bunch of drivers just pass in 0. could you explain when
> > > > is that ok? how bad is it if the wrong id is used?
> > > > 
> > > 
> > > If zero is passed, which is a non-valid NAPI_ID, busy-polling will never
> > > be performed.
> > > 
> > > Depending on the structure of the driver, napi might or might not be
> > > initialized (napi_id != 0) or even available. When it wasn't obvious, I
> > > simply set it to zero.
> > > 
> > > So, short; No harm if zero, but busy-polling cannot be used in an XDP
> > > context.
> > > 
> > > 
> > > [...]
> > > > > diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> > > > > index 21b71148c532..d71fe41595b7 100644
> > > > > --- a/drivers/net/virtio_net.c
> > > > > +++ b/drivers/net/virtio_net.c
> > > > > @@ -1485,7 +1485,7 @@ static int virtnet_open(struct net_device *dev)
> > > > >    			if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL))
> > > > >    				schedule_delayed_work(&vi->refill, 0);
> > > > > -		err = xdp_rxq_info_reg(&vi->rq[i].xdp_rxq, dev, i);
> > > > > +		err = xdp_rxq_info_reg(&vi->rq[i].xdp_rxq, dev, i, 0);
> > > > >    		if (err < 0)
> > > > >    			return err;
> > > > 
> > > > Should this be rq.napi.napi_id ?
> > > > 
> > > 
> > > Yes, if rq.napi.napi_id is valid here! Is it?
> > 
> > What initializes it? netif_napi_add? Then I think yes, it's
> > initialized for all queues ...
> > Needs to be tested naturally.
> > 
> 
> Yeah, netid_napi_add does the id generation.
> 
> My idea was that driver would gradually move to a correct NAPI id (given
> that it's hard to test w/o HW. Virtio however is simpler to test. :-)
> 
> 
> Björn

tun too ;)