Re: [PATCH v2 2/7] mm: introduce local state for lazy_mmu sections

On 12.09.25 14:37, Alexander Gordeev wrote:

On Fri, Sep 12, 2025 at 10:55:50AM +0200, David Hildenbrand wrote:

Hi David, Kevin,

quoted

Great, looking forward to seeing this all getting cleaned up and done
properly for good.

I am currently working on lazy mmu for s390 and this nesting
initiative kind of interferres. Well, in fact it looks like
it does not, but I am bit lost in last couple of iterations ;)

The prerequisite for s390 would be something like the change
below. With that change I can store the context in a per-cpu
structure and use it later in arch-specific ptep_* primitives.

Moreover, with a further (experimental) rework we could use
a custom kasan sanitizer to spot false directly compiled
PTE accesses, as opposed to set_pte()/ptep_get() accessors.

I am not quite sure see whether this could be derailed by
the new lazy mmu API. At least I do not immediately see any
obvious problem. But may be you do?

It would just be passing more context down to the architecture, right?

-- 
Cheers

David / dhildenb