Re: [PATCH 3/3] x86-64: Calgary IOMMU - hook it in

[Jon Mason <jdmason@us.ibm.com>]

On Thu, Mar 23, 2006 at 06:48:18PM +0100, Andi Kleen wrote:
> On Thursday 23 March 2006 18:30, Jon Mason wrote:
> 
> > We have a recent modification to this chunk which does both generic and
> > em64t.  Since IBM only ships this chip on em64t systems, have the option
> > on amd64 seems wrong.
> 
> Please read again what I wrote.

Oh, I agree that it should be there for the generic option, but I think
it is legitimate to have it disabled for MK8.  I'll defer to you though
:-)

> 
> > Because we need to know the size of the translation tables early, so
> > that we can alloc them from bootmem.
> 
> But __setup is parsed during the bootmem phase too. early arguments are only
> needed for things called by setup_arch

I actually tried __setup during development of this code, and it didn't
work.  The problem was that __setup, while called early, it is not
called until after setup_arch (and we need to know our table size when
detect_calgary is called at the end of setup_arch).

Thanks,
Jon

> -Andi
>