On Tue, Mar 10, 2026 at 08:00:45PM +0000, Samiullah Khawaja wrote:

On Fri, Mar 06, 2026 at 04:26:52PM -0400, Jason Gunthorpe wrote:

quoted

On Fri, Mar 06, 2026 at 08:22:08PM +0000, Samiullah Khawaja wrote:

quoted

But do you think doing the timeout logic without fencing would be good
enough?

It is what ARM and AMD do, so I wouldn't object to it.

I think without any back pressure to the caller, a device will be able
to fill the invalidation queue with device IOTLB invalidations that get
stuck until the HW timeout occurs.

Yes, all drivers have this problem currently.

Jason