On Wed, 2014-07-09 at 19:28 +0100, Will Deacon wrote: > Hello, > > This is v2 of the RFC I originally posted here: > > RFCv1: http://permalink.gmane.org/gmane.linux.kernel.iommu/5552 > > It's changed significantly since then, based on helpful feedback from > Alex. In particular, I no longer butcher the IOMMU API, instead making > use of a new iommu_attr to indicate that a domain requires support for > nesting. > > All feedback welcome (this is still an RFC after all),
Hi Will, It's a lot cleaner this way, but I'm concerned that allowing a "nesting" IOMMU to be created and used just like a TYPE1v2 IOMMU, even if the IOMMU doesn't support it, puts us in a corner for compatibility later. The safer approach might be to fail the attach_group function if we can't set the domain attribute, ie. enforce that the IOMMU must support it. I don't know how you'd handle it if only some of the domains within a container supported it, so enforcement may simplify things down the road too. We wouldn't need the vfio_domains_have_iommu_nesting() function that way either. Thanks, Alex _______________________________________________ iommu mailing list iommu@lists.linux-foundation.org https://lists.linuxfoundation.org/mailman/listinfo/iommu
