Re: [PATCH 4/5] iommu/vt-d: Add support for static identity domain

[Jason Gunthorpe]

On Wed, Nov 29, 2023 at 04:26:15PM -0400, Jason Gunthorpe wrote:
> On Mon, Nov 20, 2023 at 07:29:43PM +0800, Lu Baolu wrote:
> 
> > @@ -2311,6 +2316,13 @@ static int device_def_domain_type(struct device *dev)
> >  			return IOMMU_DOMAIN_IDENTITY;
> >  	}
> >  
> > +	/*
> > +	 * Hardware does not support the passthrough translation mode.
> > +	 * Always use a dynamaic mapping domain.
> > +	 */
> > +	if (!ecap_pass_through(iommu->ecap))
> > +		return IOMMU_DOMAIN_DMA;
> > +
> 
> Doesn't this return from def_domain_type completely prevent using an
> identity domain?
> 
> I thought the point of this was to allow the identity domain but have
> it be translating?

I suppose the answer is the next patch deletes that stuff.

I would probably have structured this in the other order, first add
this hunk and say that old HW is being de-supported. Remove all the
now-dead code creating the 1:1 page table, then refactor the remainder
to create the global static.

Jason