On Tue, Nov 08, 2022 at 03:07:19PM -0500, Eric Farman wrote: > Patch 2 doesn't address the above symptoms, but a lot of that code is > getting reworked by the aforementioned series so I didn't spend a lot > of time studying your suggestion. And as I type this I see you just > sent a new patch, let me go try that... Patch2 is following the assumption the WARN_ON is triggered by a failure path that isn't work right. Eg trying to unpin a 0 IOVA because the failure flow is wrong. Removing all the calls to unpin on failure paths make that impossible. Jason
