Eli Cohen wrote:
On Thu, 2008-01-17 at 15:18 +0200, Or Gerlitz wrote:
IB_DEVICE_IP_CSUM is used by IPOIB to mark the devices which are capable of verifying checksum. This flag is reflected in both IPOIB_FLAG_CSUM and NETIF_F_IP_CSUM. is your suggestion to use NETIF_F_IP_CSUM instead of IPOIB_FLAG_CSUM? Because I don't see how we can omit it unless we require the low lwvel drivers to set cqe.csum_ok only if they are sure that checksum is OK.
lets close this after setting the exact semantics of the csum_ok flag of the work completion.
Second, the csum_ok bit is not well defined, etc as I commented on patch #3
I will make it clearer by stating that Mellanox devices require one more condition but the semantics of this field is that it is set only if checksum is known to be good. Another option is to use the condition I put in ipoib in the low level driver and that will remove the confusion. What do you think?
I suggest that we first work on patch #3 (add checksum support to the core) to agree on the conventions and then implement what needed below (in the hw drivers) and above (in ipoib).
Or _______________________________________________ general mailing list [email protected] http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
