On Jun 16 2018, Eric Dumazet <eric.duma...@gmail.com> wrote:
> I would try something like :
>
> Basically do not bother using CHECKSUM_COMPLETE for small frames that might
> have been padded.
>
> Since we need to bring one cache line in eth_type_trans() and further header
> processing,
> computing the checksum in software will be almost free anyway.
>
> diff --git a/drivers/net/ethernet/sun/sungem.c
> b/drivers/net/ethernet/sun/sungem.c
> index
> 7a16d40a72d13cf1d522e8a3a396c826fe76f9b9..071039f211a8a33153e888bd4014314ba5e686a4
> 100644
> --- a/drivers/net/ethernet/sun/sungem.c
> +++ b/drivers/net/ethernet/sun/sungem.c
> @@ -855,9 +855,11 @@ static int gem_rx(struct gem *gp, int work_to_do)
> skb = copy_skb;
> }
>
> - csum = (__force __sum16)htons((status & RXDCTRL_TCPCSUM) ^
> 0xffff);
> - skb->csum = csum_unfold(csum);
> - skb->ip_summed = CHECKSUM_COMPLETE;
> + if (len > ETH_ZLEN) {
> + csum = (__force __sum16)htons((status &
> RXDCTRL_TCPCSUM) ^ 0xffff);
> + skb->csum = csum_unfold(csum);
> + skb->ip_summed = CHECKSUM_COMPLETE;
> + }
> skb->protocol = eth_type_trans(skb, gp->dev);
>
> napi_gro_receive(&gp->napi, skb);
That doesn't change anything.
Andreas.
--
Andreas Schwab, sch...@linux-m68k.org
GPG Key fingerprint = 7578 EB47 D4E5 4D69 2510 2552 DF73 E780 A9DA AEC1
"And now for something completely different."
