Re: [PATCH 00/16] gpu: nova-core: GPU page table, vmm, and bar1 mapping

["Danilo Krummrich" <dakr@kernel.org>]

On Wed Sep 9, 2026 at 5:59 AM CEST, Eliot Courtney wrote:
> I am reposting Joel Fernandes's original series [1]. The rebase adapts
> the patches to recent changes in the rust infra, like HRT lifetimes. It
> also adapts to the changes I posted in [2]. It keeps Joel's design and
> patch split.

I left a few comments that we really need to address. However, I'm fine if we do
that with subsequent patches. Thus, a bit hesitantly:

Acked-by: Danilo Krummrich <dakr@kernel.org>

Also note that eventually we want per-PT-page reference counts, so VM_BIND
ioctls can reserve for the whole range requested by userspace. We only know at
job execution time which parts of the requested range are mapped already (either
already with or without the correct contents) and which page size is used. I.e.
we need quite some flexibility in the API to take the optimal decision at job
execution time. So, this together with a reclaim approach will probably be a
huge rework anyway.

Eliot, when you apply this, can you please fix up the nits, e.g. make the
imports use the correct kernel vertical style, etc.?

Thanks,
Danilo