next-20250627: IOMMU DMA warning during NVMe I/O completion after 06cae0e3f61c

[Keith Busch]

On Tue, Jul 01, 2025 at 02:05:04PM +0100, Ben Copeland wrote:
> Even though SGLs are not supported, I gave you the patch a spin and
> saw the same stack traces.

Right, that patch applies only if SGL is not supported, but it's only
fixing part of the problem, so I expected it would still fail.

The dma teardown needs to be symetric to the dma mapping, but we've
removed the sg-table now, so we don't have the original mapping context
in the completion path here. Christoph replied with a patch that tries
to infer it based on the dma_addr continuity, which I think should work
for you, but I'll have to look a bit closer to be confident it's always
the correct assumption.