Re: [PATCH v11 07/20] gpu: nova-core: mm: Add TLB flush support

[Joel Fernandes]

On 4/16/2026 5:45 PM, Danilo Krummrich wrote:
> On Thu Apr 16, 2026 at 11:23 PM CEST, Joel Fernandes wrote:
>> Btw, I changed this to doing it in 2 phases to avoid holding the RCU readlock
>> across read_poll_timeout, which can sleep. Will squash it in for v12.
> 
> Why do we need the try_access() dance in the first place? I assume this ends up
> being called from the BarAccess destructor?

BarAccess is different. The try_access() calls here are in tlb.rs and
pramin.rs for Bar0. BarAccess uses &'a Bar1 directly — its Drop impl just
calls self.vmm.unmap_pages(self.mm, mapped), without needing try_access().

> 
> If so, I think this is solvable. Gary and me are currently working on
> higher-ranked types and a chained Devres type.

Hmm, the issue here is we cannot hold revocable guard while sleeping, but
we have read the bar as a condition in the body of the poll. So I split the
try_access()'es. First phase for writing and second for reading (bar 0).

> 
> With that, such use-cases should be cleanly solvable without the need for
> try_access().
> 
> Besides that, I can't find where BarAccess is ever constructed.

BarUser::map() constructs it.
https://lore.kernel.org/all/20260415210548.3776595-18-joelagnelf@xxxxxxxxxx/

> 
> It already has a lifetime 'a for &'a Bar1, so I don't see why you can't do the
> same for Bar0.>
> But again, I don't see this being constructed and I'm not sure the whole
> construct works in the first place.

BarAccess uses &'a Bar1 because it's a short-lived scoped object. In long
lived objects I am trying to avoid this. I guess I would be Ok with
switching &'a Bar1 to Devres as well. But let me know your suggestions!

If it is easier to look at a git tree to get a complete picture, here is
the git tag of v11:
https://git.kernel.org/pub/scm/linux/kernel/git/jfern/linux.git/tag/?h=nova-mm-v11-20260415

Thanks.