On 12/1/20 8:13 PM, Jakub Kicinski wrote:
>> To function, the rmnet driver must also add support for this new
>> "inline" checksum offload. The changes implementing this will be
>> submitted soon.
> We don't usually merge half of a feature. Why not wait until all
> support is in place?
>
> Do I understand right that it's rmnet that will push the csum header?
> This change seems to only reserve space for it and request the feature
> at init..
You are correct. The IPA hardware needs to be programmed to
perform the computation and verify that the checksum in the
header matches what it computes (for AP RX offload), or
insert it into the header (for AP TX offload). That's what
this patch handles.
The RMNet driver is responsible for stripping the offload
header off on RX, and acting on what it says (i.e., whether
hardware is able to state the checksum is good). And on TX
it inserts an offload header that says what to checksum and
where to put it in the packet.
It's totally fine not to merge this until we have the whole
package ready, I understand. I'll see what I can do to get
that done quickly.
Thanks Jakub.
-Alex
