Re: [PATCH v2] net: tso: add support for IPv6

[Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>]

On 2015/10/26 17:13, Toshiaki Makita wrote:
> On 2015/10/26 16:47, Grumbach, Emmanuel wrote:
>> On 10/26/2015 06:03 AM, Toshiaki Makita wrote:
>>> On 2015/10/26 5:02, Emmanuel Grumbach wrote:
>>>> Adding IPv6 for the TSO helper API is trivial:
>>>> * Don't play with the id (which doesn't exist in IPv6)
>>>> * Correctly update the payload_len (don't include the
>>>>   length of the IP header itself)
>>> ...
>>>>  	memcpy(hdr, skb->data, hdr_len);
>>>> -	iph = (struct iphdr *)(hdr + mac_hdr_len);
>>>> -	iph->id = htons(tso->ip_id);
>>>> -	iph->tot_len = htons(size + hdr_len - mac_hdr_len);
>>>> +	if (skb->protocol == htons(ETH_P_IP)) {
>>>
>>> I guess this should be vlan_get_protocol(skb).
>>
>> I truly don't know. I guess we could have VLANs, but I'd need to check
>> how the packet would look like after it exits mac80211.
> 
> I don't know much about mac80211.
> 
> What I see is that mvneta has TSO in vlan_features and it uses
> tso_build_hdr(). When vlan device is used, we cannot access network
> protocol by skb->protocol without HW vlan acceleration.
> So it looks like this change corrupts TSO functionality on mvneta.
> 
>> If we need that, I'll likely do this check once in tso_start() and add a
>> variable to struct tso_t.
> 
> I'm not sure if an additional variable is needed.
> At least, skb_network_offset()/ip_hdr() should correctly handle (skip)
> vlan headers.

Ah, sorry, I misread your suggestion.
Additional variable would make sense to me.

Toshiaki Makita