[ofw] changes to ib_wc committed to svn

[Jan Bottorff]

>It seems to me that checksum offload should be a QP attribute, not a work >request option.  That way clients can do a check to see if the CA >attributes indicate the HW supports checksum offload, and then create their >QP accordingly.

Hardware send checksum generation has to be controllable on a per packet basis. 

The WHQL tests will send packets with an invalid checksum and validate the receiving end flags the checksum as wrong. It does this to assure your driver is not just always saying the offloaded receive checksum is good.

There is also the case of forwarding packets as a bridge/router. It seems inappropriate for a bridge connection to receive a packet with a bad checksum and then forward it with a new calculated checksum. If you do so, it means you just declared a bad backed as good, the receiving end will find the checksum is good. All forwarded packets need to not regenerate the checksum.

Jan
 

__________ Information from ESET NOD32 Antivirus, version of virus signature database 3267 (20080714) __________

The message was checked by ESET NOD32 Antivirus.

http://www.eset.com