On Fri, Dec 4, 2015, at 21:43, Tom Herbert wrote:

On Fri, Dec 4, 2015 at 12:26 PM, Hannes Frederic Sowa
[off-list ref] wrote:

quoted

Hi Dave,

On Fri, Dec 4, 2015, at 21:06, David Miller wrote:

quoted

From: Hannes Frederic Sowa <redacted>
Date: Fri, 04 Dec 2015 20:59:05 +0100

quoted

Yes, I agree, I am totally with you here. If generic offloading can be
realized by NICs I am totally with you that this should be the way to
go. I don't see that coming in the next (small number of) years, so I
don't see a reason to stop this patchset.

If I just apply this and say "yeah ok", the message is completely lost
and your prediction about "small number of years" indeed will occur.

However if I push back hard on this, as I will, then the message has
some chance of seeping back to the people designing these chips.

So that's what I'm going to do, like it or not.

Or can someone convince me that someone who understand this stuff
is telling the hardware guys to universally put 2's complement
checksums into the descriptors?

Who is doing that at each and every prominent ethernet hardware
verndor?

Who?

If I get silence, or some vague non-specific response, then I'm going
to hold my ground and keep pushing back on this stuff.

This is not only about 1's checksumming but also about TSO (and to some
smaller degree about RSS, as I tried to explain): if we attach a geneve
header in front of a skb we expect the hardware to recognize it and
duplicate it while doing the hardware segmentation. The hardware can
only do so if it is in knowledge of the specific port (in this case UDP
port used for geneve) which is in use for this particular tunneling
transport protocol. We currently cannot describe this in a generic way,
thus this patchset. (Please correct me if I am wrong!)

Yes, you are wrong. Port numbers are not used in transmit path to
signal offload. To perform TSO on UDP encapsulated packets the skb is
marked with SKB_GSO_UDP_TUNNEL or SKB_GSO_UDP_TUNNEL_CSUM and
SKB_GSO_TCP, etc. The driver can use information along with the
offsets of the inner and outer headers in the packet to set up the
operation in the device.  Some devices only support TSO for VXLAN, but
regardless SKB_GSO_UDP_TUNNEL is generic for all known UDP
encapsulations. Protocol specific offload is not needed. Please start
looking at
http://people.netfilter.org/pablo/netdev0.1/papers/UDP-Encapsulation-in-Linux.pdf
and the kernel code to see how things _actually_ work.

I am sorry. Of course we have _inner_ header pointers and the
corresponding gso_types in skb_shinfo to signal that already. Probably I
got confused by some driver sources and commit descriptions I looked
into lately. Thanks for the correction! :)

Bye,
Hannes