Re: [PATCH 2/3] x86/amd-iommu: Work around S3 BIOS bug

["Roedel, Joerg" <Joerg.Roedel@amd.com>]

On Thu, Sep 30, 2010 at 04:21:04PM -0400, Matthew Garrett wrote:
> On Thu, Sep 23, 2010 at 04:44:48PM +0200, Joerg Roedel wrote:
> > +static void iommu_apply_quirks(struct amd_iommu *iommu)
> > +{
> > +	if (is_rd890_iommu(iommu->dev)) {
> > +		pci_write_config_dword(iommu->dev, 0xf0, iommu->cache_cfg[0]);
> > +		pci_write_config_dword(iommu->dev, 0xf4, iommu->cache_cfg[1]);
> > +		pci_write_config_dword(iommu->dev, 0xf8, iommu->cache_cfg[2]);
> > +		pci_write_config_dword(iommu->dev, 0xfc, iommu->cache_cfg[3]);
> > +	}
> > +}
> > +
> 
> This doesn't look right. 0xf0-0xff are for indexed register access, so 
> what you're doing here is just restoring the last of each of those 
> registers that the BIOS programmed - and if the BIOS cleared the 
> writable flag afterwards, you're not even doing that.

With a half-fixed BIOS (a BIOS which re-enables the IOMMU on resume)
this was the necessary step to make the IOMMU execute commands again. So
it actually fixed the problem I have seen. I agree that its better to
fully restore the indirect register spaces when we workaround the fully
broken BIOSes. When we have this I remove this code, but until then it
fixes the problem.

	Joerg

-- 
AMD Operating System Research Center

Advanced Micro Devices GmbH Einsteinring 24 85609 Dornach
General Managers: Alberto Bozzo, Andrew Bowd
Registration: Dornach, Landkr. Muenchen; Registerger. Muenchen, HRB Nr. 43632