11 Jul
2025
11 Jul
'25
2:49 a.m.
From: Jason Gunthorpe jgg@nvidia.com Sent: Thursday, July 10, 2025 9:23 PM
On Thu, Jul 10, 2025 at 05:53:16AM -0700, Dave Hansen wrote:
- I'm not sure if the IOMMU will even create an IOTLB entry for a supervisor-permission mapping while doing a user-permission walk.
Early VT-d platforms may cache faulted translation. But they are pretty old and way before SVA was introduced in spec.
It doesn't matter if it does or doesn't, it doesn't change the argument that we don't have to invalidate on PTEs being changed.
but yes, that fact doesn't matter.