Re: [PATCH 0/6] Replace private domain with per-group default

[Joerg Roedel]

On Wed, Mar 11, 2020 at 02:50:39PM +0800, Lu Baolu wrote:
> On 2020/3/10 19:15, Joerg Roedel wrote:
> > Hi Baolu,
> > 
> > On Sat, Mar 07, 2020 at 02:20:08PM +0800, Lu Baolu wrote:
> > > Lu Baolu (5):
> > >    iommu: Configure default domain with dev_def_domain_type
> > >    iommu/vt-d: Don't force 32bit devices to uses DMA domain
> > >    iommu/vt-d: Don't force PCI sub-hierarchy to use DMA domain
> > >    iommu/vt-d: Add dev_def_domain_type callback
> > >    iommu/vt-d: Apply per-device dma_ops
> > > 
> > > Sai Praneeth Prakhya (1):
> > >    iommu: Add dev_def_domain_type() callback in iommu_ops
> > 
> > I like this patch-set, but I fear some regressions from patch
> > "iommu/vt-d: Don't force 32bit devices to uses DMA domain". Have you
> > tested this series on a couple of machines, ideally even older ones from
> > the first generation of VT-d hardware?
> 
> The oldest hardware I have is Ivy Bridge. :-) Actually, The effect of
> using identity domain for 32-bit devices is the same as that of adding
> intel_iommu=off in the kernel parameter. Hence, if there is any
> regression, people should also find it with intel_iommu=off.
> intel_iommu=off support is added at the very beginning of VT-d driver.

Okay, I will also do some testing on it, one of my VT-d machines is a
Haswell. Please send a new version with the recent comments addressed.

Regards,

	Joerg