RE: [net-2.6 v2 5/7] ixgbe: DDP last buffer size work around

David,

Sorry I missed your mail from last night.

You are right, it should be bufflen, which is 4096 and not PAGE_SIZE.

I will fix that and resubmit.
-Amir


-----Original Message-----
From: David Miller [mailto:davem@davemloft.net] 
Sent: Friday, January 28, 2011 11:51 AM
To: Kirsher, Jeffrey T
Cc: Hanania, Amir; netdev@vger.kernel.org; gospo@redhat.com; bphilips@novell.com
Subject: Re: [net-2.6 v2 5/7] ixgbe: DDP last buffer size work around

From: Jeff Kirsher <redacted>
Date: Fri, 28 Jan 2011 04:29:01 -0800

From: Amir Hanania <redacted>

We found a hardware erratum on 82599 hardware that can lead to buffer
overwriting if the last buffer in FCoE DDP is exactly PAGE_SIZE.
If this is the case, we will make sure that there is no HW access to
this buffer.

Please see the 82599 Specification Update for more information.

Signed-off-by: Amir Hanania <redacted>
Tested-by: Ross Brattain <redacted>
Signed-off-by: Jeff Kirsher <redacted>

Jeff, I still think this change is bogus.

PAGE_SIZE is variable, so the chip can't possibly only BUG on that
specific value.

Maybe the condition is "any power-of-2 larger than or equal to 4096"?
Or something like that?

I pointed this out last night, and I was really hoping I'd get actual
feedback on this issue before you're respin your tree.

I'm not pulling in these changes until I get a human being telling me
what the situation is here with this bug.