Re: [PATCH 10/11] iommu/amd/iommu_v2: Add support to switch default domain to SVA mode

[Jason Gunthorpe <jgg@ziepe.ca>]

On Tue, Aug 08, 2023 at 10:02:31AM +0000, Vasant Hegde wrote:
> Current iommu_v2 module allocates a secondary domain every time a
> device driver calls amd_iommu_init_device(). Then it detaches all
> devices in the group from the default domain, and reattaches to the
> new domain. Finally it configures V2API mode.
> 
> Previous patch added support to switch same domain to SVA mode. Use
> these new interfaces to enable SVA mode.
> 
> Note that even with this change, PASID is managed by device driver. This
> imposes restriction where there can be only one SVA enabled device per
> IOMMU group (same as current behaviour).
> 
> Finally remove unused functions.
> 
> Co-developed-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
> Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
> Signed-off-by: Vasant Hegde <vasant.hegde@amd.com>
> ---
>  drivers/iommu/amd/amd_iommu.h |  2 --
>  drivers/iommu/amd/iommu.c     | 37 ----------------------
>  drivers/iommu/amd/iommu_v2.c  | 59 +++++++----------------------------
>  3 files changed, 12 insertions(+), 86 deletions(-)

I'm glad to see this code go, but my remarks are basically the same as
the last patch

amd_iommu_init_device() should force the kernel owned GCR3 table, not
mess with domains.

Basically all it should do is get a ref on the GCR3 table. A ref may
already exist if the GCR3 is already being used by a v2 RID domain.

Jason