Re: [PATCH 2/3] x86-64: Calgary IOMMU - Calgary specific bits

[Muli Ben-Yehuda <muli@il.ibm.com>]

On Thu, Mar 23, 2006 at 09:03:34PM +0200, Muli Ben-Yehuda wrote:

> > > X works :-) 
> > 
> > So it's behind a bridge that doesn't have an IOMMU?
> 
> No, it's behind a bridge that does have an IOMMU and is running with
> translation enabled (it's on PHB 0 on this machine). I guess you are
> concerned with userspace access to the graphics controller directly,
> without a kernel driver having set up mapping previously? I will look
> into it but emprirically X works so either userspace is not triggering
> DMAs or the mappings have been set up by a driver.

Turns out that X does work on my machine (SLES 9SP2) but dies with a
bad translation error on Jon's machine, which is otherwise identical
except it runs gentoo. We are thinking how to best address this (add
IOMMU aware drivers to X? *shudder*), but will disable translation by
default on PHB 0 in the mean time for a friendlier user
experience.

Thanks,
Muli