Re: [PATCH 13/16] gpu: nova-core: mm: Add multi-page mapping API to VMM

[Alistair Popple]

On 2026-09-09 at 13:59 +1000, Eliot Courtney <ecourtney@xxxxxxxxxx> wrote...
> From: Joel Fernandes <joelagnelf@xxxxxxxxxx>

[...]

> +        // TODO: Internal page table pages (PDE, PTE pages) are still kept around.
> +        // This is by design as repeated maps/unmaps will be fast. As a future TODO,
> +        // we can add a reclaimer here to reclaim if VRAM is short. For now, the PT
> +        // pages are dropped once the `Vmm` is dropped.
> +
> +        // Free the VA range regardless of PTE invalidation success, so that the VA
> +        // range is recovered even on failure (PTEs may be stale, but that is better
> +        // than leaking both PTEs and VA range).

I don't think this is the correct approach. I'm not sure what exactly can cause
PTE invalidation to fail, but I don't think we can just recover the VA range
if something might still be using it via stale PTEs. That would cause problems
if the VA was ever reused for example. TLB invalidation is also part of PTE
invalidation as written, so we could also end up with stale TLB entries.

So leaking the range might be bad, but it's the least bad option IMHO.

 - Alistair

> +        self.free_vfn(range.vfn_start);
> +
> +        // Unmap complete, safe to drop `MappedRange`.
> +        range._drop_guard.disarm();
> +        result
> +    }
>  }
> 
> -- 
> 2.55.0
>