On Tue, Jul 08, 2008 at 05:40:32AM -0400, Rajat Jain wrote:
> 
> Hi,
> 
> I found out answer to my question 2, but my other two queries persist:
> 
> Thanks,
> 
> Rajat
> 
> 
> > Hi,
> > 
> > I'm trying to add TCP / IP checksum support in my network driver and am a 
> > little confused about its interaction with the higher layers. Can some one 
> > please clarify my doubts:
> > 
> > 1) Firstly, should a driver be ready for a situation where the user first 
> > enables the HW checksum support and then disables it? Looking at the help for 
> > ifconfig, this surely looks possible:
> > 
> > # ifconfig en0 inet ip4csum-rx
> > # ifconfig en0 inet -ip4csum-rx
> > 
> > But when I try this experimentally, I find that my driver's SIOCSIFCAP devctl 
> > dutifully called with correct ifcr_capenable_rx set for the first command. But
> >  my devctl() is never called again for the second command. What am I missing 
> > here?


This is a bug in ifconfig. -ip4csum-[rt]x was trying to enable,
not disable.  Since it was already enabled the stack didn't bother
calling your devctl. -ip4csum was doing the right thing. New x86
io-net version of ifconfig attached.

> > 
> > 
> > 3) Just to make sure I understand it correctly, can somebody please confirm my
> >  understanding:
> > 
> > Taking IPv4 as example, in Rx side checksum offloading, the driver should set 
> > the flags in npkt->csum_flags like this:
> > 
> > - If no checksum done, do not set any flags

If it's not enabled, the stack ignores your flags on rx so do whatever's
easiest / most efficient.

> > - If checksum verification done, and it was Correct - Just set M_CSUM_IPv4
> > - If checksum verification done, and it was bad -set  M_CSUM_IPv4 and 
> > M_CSUM_IPv4_BAD
> >  
> > Is this right?

Yes.