[PATCH v1 2/2] iommu/arm-smmu-v3: Recover ATC invalidate timeouts

[Samiullah Khawaja]

On Fri, Mar 06, 2026 at 04:26:52PM -0400, Jason Gunthorpe wrote:
>On Fri, Mar 06, 2026 at 08:22:08PM +0000, Samiullah Khawaja wrote:
>
>> 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.
>
>> Currently VT-d blocks itself, until it gets an Invalidation Timeout
>> from HW, and system ends up in a hardlockup since interrupts are
>> disabled.
>>
>> Are you concerned that if fencing is done without an RAS flow, the
>> device might not be able to detect the failure (if it really needs ATS
>> to work)?
>
>Yes, and then the device is badly locked because nothing will fix the
>IOMMU fence.
>
>VFIO might fix it if it is restarted, but other approahces like
>rmmod/insmod won't restore the broken device.
>
>So I'd rather see a more complete solution before we add fencing to
>the iommu drivers. Minimally userspace doing a rmmod, flr, insmod
>should be able to restore the device.
>
>Then auto-FLR through RAS could sit on top of that.
>
>> I am thinking, we can do translated fence and timeout change for VT-d.
>> And the device can use existing RAS mechanism to recover itself. This
>> way we atleast make sure that caller of flush can reuse the memory/IOVAs
>> without UAFs.
>
>Without a larger framework to unfence I think this will get devices
>stuck..
>
>Jason