Re: [PATCH v2 25/25] drm/xlnx: Compute dumb-buffer sizes with drm_mode_size_dumb()

[Laurent Pinchart <laurent.pinchart@ideasonboard.com>]

On Thu, Jan 16, 2025 at 12:07:49PM +0200, Tomi Valkeinen wrote:
> On 16/01/2025 11:38, Laurent Pinchart wrote:
> > On Thu, Jan 16, 2025 at 10:43:40AM +0200, Laurent Pinchart wrote:
> >> On Wed, Jan 15, 2025 at 02:34:26PM +0000, Daniel Stone wrote:
> >>> On Wed, 15 Jan 2025 at 14:20, Tomi Valkeinen wrote:
> >>>> No disagreement there, we need CREATE_DUMB2.
> >>>>
> >>>> My point is that we have the current UAPI, and we have userspace using
> >>>> it, but we don't have clear rules what the ioctl does with specific
> >>>> parameters, and we don't document how it has to be used.
> >>>>
> >>>> Perhaps the situation is bad, and all we can really say is that
> >>>> CREATE_DUMB only works for use with simple RGB formats, and the behavior
> >>>> for all other formats is platform specific. But I think even that would
> >>>> be valuable in the UAPI docs.
> >>>
> >>> Yeah, CREATE_DUMB only works for use with simple RGB formats in a
> >>> linear layout. Not monochrome or YUV or tiled or displayed rotated or
> >>> whatever.
> >>>
> >>> If it happens to accidentally work for other uses, that's fine, but
> >>> it's not generically reliable for anything other than simple linear
> >>> RGB. It's intended to let you do splash screens, consoles, recovery
> >>> password entries, and software-rendered compositors if you really
> >>> want. Anything more than that isn't 'dumb'.
> >>
> >> We have lots of software out there that rely on CREATE_DUMB supporting
> >> YUV linear formats, and lots of drivers (mostly on Arm I suppose) that
> >> implement YUV support in CREATE_DUMB. I'm fine replacing it with
> >> something better, but I think we need a standard ioctl that can create
> >> linear YUV buffers. I've been told many times that DRM doesn't want to
> >> standardize buffer allocation further than what CREATE_DUMB is made for.
> >> Can we reconsider this rule then ?
> > 
> > Actually... Instead of adding a CREATE_DUMB2, it would be best on trying
> > to leverage DMA heaps and deprecate allocating from the KMS device.
> 
> If we allocate from DMA heaps, I think we then need a DRM ioctl which 
> will tell you how big buffer(s) you need, based on the size and format.

We will at least a kernel API to expose constraints to userspace.
Whether that should calculate the buffer size for a given format, or
expose information to userspace to enable that calculation, I'm not
sure. But regardless of which option we pick, I agree we likely need a
new API to enable usage of DMA heaps as allocators.

-- 
Regards,

Laurent Pinchart