On Wed, 17 Sep 2008 15:52:45 -0700, Bryan Phillippe <email@example.com> wrote:
> I tested the simplified patch below (ADDC32), and it does not address
> the checksum bug. I suspect the problem is that we're still leaving
> the carry bit in the upper 16 bits of the 32 bit csum returned, and
> this is resulting in a computed checksum that is 1 greater than it
> should be. The upper 16 bits of the return value of this function
> must be 0, right?
Thank you for testing. Though this patch did not fixed your problem,
I still have a doubt on 64-bit optimization.
If your hardware could run 32-bit kernel, could you confirm the
problem can happens in 32-bit too or not?