[ofw] RE: [patch] fix ipoib performance problem in rev 1430

[Fab Tillier]

Hi Anatoly,

> You're right although some may say chksum on receive path is not needed
> at all. Anyway, mthcd does not update csum_ok at all, therefore ndis
> checksum flags have random value in each packet. Fix is attached.

I'm sorry to make you jump through these hoops.  If the csum_ok field was embedded in the recv_opt field (reserve 8 bytes in there, there are plenty available) like we had discussed/agreed, this issue would probably not have come up as drivers already set recv_opt and would likely have unknown bits cleared.

Can you make your patch put the csum information into the recv_opt as was originally agreed upon, and restore the vendor_specific field to a single uint64?  I know it's extra work for something you didn't break, but it's the right fix.  If you don't want to that's fine, we can ask Tzachi to.

-Fab