To follow up on the code review comment about checksum offload:

I verified with the NIC-China team and received the following information
from Ted You:

  The bge driver hardware checksum offload itself doesn't have any
  problems. I think the purpose of this private property is for
  possible debugging in the future.

  The hardware checksum is enabled by default by the Broadcom
  chipset hardware, it cannot be disabled by the driver. And those bits
  will be set in every hw_rbd when receiving packets. If we want to
  hide the rx hardware checksum, the only way is to add the
  conditionals to check the flag and bypass the processing of those
  hw_rbd bits. So if we keep this private property, we cannot avoid
  such code in the hot code path. It's likely that the extra check does not
  have a significant impact on performance.

  Since the code has been added for debugging purposes only (to debug
  problems like data corruption, hw workaround etc.) we will remove the
  changes related to bge_chksum_flag and add them back later after verifying
  that there is no impact on performance.

One possibility is to provide support for this in DEBUG kernels only.
I've filed a place-holder RFE to track this: see CR 6629075.

--Sowmini


Reply via email to