On Wed, 23 May 2018 11:34:22 +0200 Daniel Borkmann <dan...@iogearbox.net> wrote:
> > +int dev_map_enqueue(struct bpf_dtab_netdev *dst, struct xdp_buff *xdp)
> > +{
> > + struct net_device *dev = dst->dev;
> > + struct xdp_frame *xdpf;
> > + int err;
> > +
> > + if (!dev->netdev_ops->ndo_xdp_xmit)
> > + return -EOPNOTSUPP;
> > +
> > + xdpf = convert_to_xdp_frame(xdp);
> > + if (unlikely(!xdpf))
> > + return -EOVERFLOW;
> > +
> > + /* TODO: implement a bulking/enqueue step later */
> > + err = dev->netdev_ops->ndo_xdp_xmit(dev, xdpf);
> > + if (err)
> > + return err;
> > +
> > + return 0;
>
> The 'err' is just unnecessary, lets just do:
>
> return dev->netdev_ops->ndo_xdp_xmit(dev, xdpf);
>
> Later after the other patches this becomes:
>
> return bq_enqueue(dst, xdpf, dev_rx);
I agree, I'll fix this up in V5.
After this patchset gets applied, there are also other opportunities to
do similar micro-optimizations. I have a branch (on top of this
patchset) which does such micro-optimizations (including this) plus
I've looked at the resulting asm-code layout. But my benchmarks only
show a 2 nanosec improvement for all these micro-optimizations (where
the focus is to reduce the asm-code I-cache size of xdp_do_redirect).
--
Best regards,
Jesper Dangaard Brouer
MSc.CS, Principal Kernel Engineer at Red Hat
LinkedIn: http://www.linkedin.com/in/brouer
