Re: [PATCH rc v7 0/7] iommu/arm-smmu-v3: Fix device crash on kdump kernel

[Jason Gunthorpe]

On Tue, Jun 30, 2026 at 01:17:30PM +0000, Mostafa Saleh wrote:

> In many cases the patches assume that the CDs/STE might be corrupted,
> but still attempt to retrieve them with some validation
> (log2size/split...)
> However, the base address might be broken, TLBs state is unknown...
 
> IMO, although that might improve the status quo, there are still
> heuristics, in addition to noticeable complexity to transition the
> stream tables.

That's basically what kdump is all about, try to improve the chances
that the kdump kernel functions enough to retrieve the dump. There are
many reasons kdump can fail, but nevertheless it works well enough and
often enough to still be highly useful.

So, the cases which are frequent and problematic should be addressed.
On this HW kdump has a high failure rate because of the errors.

Given that non-disruption is exactly what the Intel and AMD drivers
both implement SMMU should also.

Jason