Re: [PATCH v2 04/10] mm/cma: Allow dynamically creating CMA areas

[Maxime Ripard <mripard@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 3827 bytes --]

On Thu, Feb 12, 2026 at 03:44:11PM +0100, Thierry Reding wrote:
> On Fri, Jan 23, 2026 at 02:25:16PM +0100, Maxime Ripard wrote:
> > On Thu, Jan 22, 2026 at 05:10:03PM +0100, Thierry Reding wrote:
> > > From: Thierry Reding <treding@nvidia.com>
> > > 
> > > There is no technical reason why there should be a limited number of CMA
> > > regions, so extract some code into helpers and use them to create extra
> > > functions (cma_create() and cma_free()) that allow creating and freeing,
> > > respectively, CMA regions dynamically at runtime.
> > > 
> > > The static array of CMA areas cannot be replaced by dynamically created
> > > areas because for many of them, allocation must not fail and some cases
> > > may need to initialize them before the slab allocator is even available.
> > > To account for this, keep these "early" areas in a separate list and
> > > track the dynamic areas in a separate list.
> > > 
> > > Signed-off-by: Thierry Reding <treding@nvidia.com>
> > 
> > AFAIU, this won't create a new cma heap when registering. This goes
> > against the recent work we did to create one for every cma region.
> > 
> > I guess, since you have a driver that would explicitly handle that
> > region, we should create some kind of opt-out mechanism, but by default,
> > we should still create such a heap.
> 
> It sounds like there's a bit of a conflict between what you want to
> achieve and what this series attempts to do.

It's not ongoing really, it's part of 6.19.

> The way I see it, the CMA code is more of a helper that gives you a
> specific functionality set. Exposing each CMA area as a heap that
> userspace can allocate from seems like a bad idea to me.
>
> Without knowing anything specific about a CMA area you don't know if it
> makes sense to expose it as a heap. Given that there is very little
> information associated with a CMA area there's only so much guessing
> that you can do. I think it'd be more sensible to make CMA areas opt-in
> to have a heap created for them rather than requiring opt-out. Exposing
> a heap publicly applies only to a (potentially) small subset of all CMA
> areas, albeit at the moment it may seem that that is what it's primarily
> used for.

Do you have any specific example in mind except for that driver?

So, the reason why we did that was, mostly, to allow proper cgroup
memory accounting through dmem. In order to enable it in DRM and v4l2,
it was agreed upon that we would switch the use of dma_alloc_* to rely
on the heaps instead, where the memory accounting is greatly simplified.

So we want any reserved memory region a device can allocate from to have
a heap.

So I do think we need the call to register a heap in rmem_cma_setup.

That being said...

> In fact, for this particular driver nobody must allocate from any of the
> CMA regions associated with the heap driver outside of that heap driver,
> simply because the heap driver maintains meta data about these CMA
> regions for things to work. If we allow access to it from anywhere,
> things are eventually going to explode.

... I also agree that having it in dma_contiguous_reserve() might be
overdoing it, and I assume it would solve the issue with your driver?

> > That being said, it's not clear to me why the heap driver uses CMA in
> > the first place.
> 
> We use CMA as a way of reclaiming memory if needed. The heap that we
> create is meant to be resizable, so that when nothing uses the heap, the
> memory can be reused for other purposes. However, when memory is
> allocated from the heap, we need to reclaim that memory for the heap and
> relocate any buffers allocated from the region somewhere else. CMA does
> all of that for us, so it seemed like the logical choice for this.

Ack, thanks!
Maxime

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 273 bytes --]