Bug#404148: disabling hw iommu on nvidia
Hi Andi,
On Tue, Apr 03, 2007 at 12:37:09AM +0200, Andi Kleen wrote:
> On Tuesday 03 April 2007 00:29:16 dann frazier wrote:
> > hey Andi,
> > Debian is looking at patching our kernel to disable the hw iommu on
> > nvidia chipsets for the data corruption bug that's been discussed on
> > lkml[1].
> It would be better if you waited until the official solution. The
> hardware debugging people are working on it. You'll likely get more
> problems out of swiotlb because the defaults are too small for high
> loads.
Are those problems going to be comparable in severity to the data corruption
currently reported? As Dann notes, there's no immediate rush on this for us
since it won't go anywhere until the first point release of etch, but
nothing I've seen so far suggests there's any sort of ETA for an upstream
fix and I wouldn't want the point release to come and go without a
workaround while waiting for a proper fix.
> > As far as I can tell there isn't an upstream solution yet, but
> > Steve Langasek has proposed the following patch which seems to work
> > for one of our users. Would you mind taking a look at it and see if
> > you can pick out any obvious problems?
> You got at least one off by one bug. And SWIOTLB cannot be undefined
> with CONFIG_IOMMU so the nested ifdef is useless.
Thanks for the feedback. FWIW, I can only assume the off-by-one bug here is
+ if ((end_pfn > MAX_DMA32_PFN ||
which means that there's already an off-by-one bug in the pristine version
of this file, since I copied that from the VIA handling up above. :)
Cheers,
--
Steve Langasek Give me a lever long enough and a Free OS
Debian Developer to set it on, and I can move the world.
vorlon@debian.org http://www.debian.org/
Reply to: