Re: Drivers: net: Hyperv

[Ben Hutchings <ben@decadent.org.uk>]

[-- Attachment #1: Type: text/plain, Size: 1569 bytes --]

On Tue, 2014-04-22 at 11:22 -0700, K. Y. Srinivasan wrote:
> I am trying to port the various offloads (checksum, TSO etc.) that I recently 
> implemented for the Hyper-V network driver to a 2.6.32 kernel and while this
> code works on the upstream bits, TCP checksum offload does not seem to work
> on the 2.6.32 kernel. Would you know if there is a different way to turn on
> checksum offload on 2.6.32 kernels?  I am setting the relevant feature flags:
> 
> 
>         net->features =  NETIF_F_SG | NETIF_F_RXCSUM |
>                         NETIF_F_IP_CSUM | NETIF_F_TSO;

There is no NETIF_F_RXCSUM in 2.6.32, but otherwise this should work.

I notice that NETIF_F_HIGHDMA is not there, and it seems like you could
add it as netvsc_start_xmit() doesn't assume page fragments are in
lowmem.

> What I am observing is that when the skb is presented to the driver for transmission,
> The ip_summed field in the skb is set to 0.

Maybe you tested on a 32-bit system and the packet had to be copied to
lowmem?  The allocation behaviour may have changed since 2.6.32.

> On the kernel tip, I see this value set to CHECKSUM_PARTIAL and the offloads
> work correctly on the kernel tip. The other difference I see is that on entry into the driver,
> the eth_hdr state is not correctly set in the skb.
> I need to invoke skb_reset_mac_header() to get this state correctly set.
> Any help would be greatly appreciated.

I don't recall that problem.

Ben.

-- 
Ben Hutchings
Beware of programmers who carry screwdrivers. - Leonard Brandwein

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 811 bytes --]