Re: [PATCH net-next v2 1/1] net: fec: Enable imx6 enet checksum acceleration.

[Jim Baxter <jim_baxter@mentor.com>]

On 16/04/13 23:41, Ben Hutchings wrote:
> On Tue, 2013-04-16 at 23:15 +0100, Jim Baxter wrote:
>> On 16/04/13 16:39, Ben Hutchings wrote:
> [...]
>>>> @@ -818,6 +890,19 @@ fec_enet_rx(struct net_device *ndev, int budget)
>>>>  				spin_unlock_irqrestore(&fep->tmreg_lock, flags);
>>>>  			}
>>>>  
>>>> +			if (fep->bufdesc_ex &&
>>>> +				(fep->csum_flags & FLAG_RX_CSUM_ENABLED)) {
>>>> +				struct bufdesc_ex *ebdp =
>>>> +					(struct bufdesc_ex *)bdp;
>>>> +				if (!(ebdp->cbd_esc & FLAG_RX_CSUM_ENABLED)) {
>>>> +					/* don't check it */
>>>> +					skb->ip_summed = CHECKSUM_UNNECESSARY;
>>>
>>> This looks very strange.  Presumably the RX_ICE and RX_PCR flags
>>> indicate checksum errors, and therefore !(ebdp->cbd_esc &
>>> FLAG_RX_CSUM_ENABLED) means the checksum(s) are good?  This would be
>>> clearer if you defined FLAG_RX_CSUM_ERROR as well (with the same numeric
>>> value).
>>>
>> Agreed, that would be clearer.
>>>> +				} else {
>>>> +					ndev->stats.rx_errors++;
>>>
>>> Layer 3 and 4 errors should not be counted in the net device stats.
>>>
>> Ok, are they just for Levls 1 and  2?
> [...]
> 
> Right, that's my understanding.  Our job as driver writers is to
> validate at layer 1 and 2 (FCS error, length error, MAC filtering).
> Checksum validation at layers 3 and 4 is an optimisation and the kernel
> stack takes care of counting checksum errors.
> 
> Ben.
> 

Thanks for that information Ben, that is good to know.

Has anyone used ping6 or iperf -V, whilst debugging I have found that
the packets being sent are of protocol 0x0800 when using those commands.

Jim