On Mon, May 2, 2016 at 9:48 AM, Alexander Duyck
<[email protected]> wrote:
> On Mon, May 2, 2016 at 9:33 AM, Tom Herbert <[email protected]> wrote:
>> On Mon, May 2, 2016 at 9:25 AM, Alexander Duyck <[email protected]> wrote:
>>> In the case of the mlx4 and mlx5 driver they do not support IPv6 checksum
>>> offload for tunnels.  With this being the case we should disable GSO in
>>> addition to the checksum offload features when we find that a device cannot
>>> perform a checksum on a given packet type.
>>>
>> I'm not sure I understand this. If device can't support checksum
>> offload for tunnels doesn't that mean we have to do the checksum on
>> host regardless of whether GSO is being done?
>
> The use of the term GSO here might be the confusing part.  Basically
> the issue is the hardware advertises it can do TSO for IPv4 on
> encapsulated frames, however it doesn't indicate it can do IPv6
> checksum offload.  So what ends up happening is that in the case of a
> v4 over v6 tunnel we were going through validate_xmit_skb which will
> check things in netif_skb_features and come out supporting the TSO but
> no checksums.  As a result we would fall through and hit
> skb_checksum_help and trigger the warn on in there because we had TSO
> requested even though we couldn't do the checksum.
>
> Basically I am just extending the kind of logic we have in
> netdev_fix_features so that if we cannot support checksumming the
> frame then we cannot support segmenting it.
>
Thanks for the explanation. We need to drive things so that all the
encapsulation combinations (v4/v4, v4/v6, v6/v4, v6/v6) are supported
by HW TSO if any of them are supported by a device. Maybe we should
still have some sort of warning message that HW is broken for some
combination (like it apparently it is for mlnx4)?

Tom

> - Alex

Reply via email to