Re: [PATCH 2/8] iommu/vt-d: Add entry_sync support for PASID entry updates

[Jason Gunthorpe <jgg@nvidia.com>]

On Tue, Mar 24, 2026 at 02:22:21PM +0800, Baolu Lu wrote:
> > A fencing requirement does make sense, but does it have to be done by
> > flushing the entire DID,PASID? It is ugly to have to drop the IOTLB
> > just because a context entry changed.
> 
> I believe the full [old_DID, PASID] invalidation is a functional
> necessity rather than just a fencing requirement. Even though the new
> PASID entry points to a new_DID, leaving stale translations tagged with
> [old_DID, PASID] in the IOTLB is problematic.

You don't know they are stale, the old_DID,PASID could be used by
another context entry.

The proper time to delcare an IOTLB tag as stale is when it is
returned back to the allocator.

The problem in the VT-d design is that it is complicated to manage the
DID lifecycle well - but the driver itself should have a clear idea
when tags are end of life.

Yes, it is a very reasonable simplified design for the driver to say
the DID lifecycle ends at every context entry change, but that is very
different from stating the flush as a mandatory requirement from the
HW spec.

A more complex design could have each context entry request a cache
tag for the context entry, for the specific domain, and share the tags
whenever possible, eg for SVA and S2 domains.

Jason