[EARLY RFC][PATCH] dma-buf: Add dma-buf heaps framework

[Brian Starkey]

On Mon, Mar 04, 2019 at 08:53:42AM -0600, Andrew F. Davis wrote:
> On 3/1/19 6:06 AM, Brian Starkey wrote:
> > 
> > Am I right in thinking "cmd" comes from userspace? It might be a good
> > idea to not trust _IOC_SIZE(cmd) and use our own. I'm looking at
> > Documentation/ioctl/botching-up-ioctls.txt and drm_ioctl()
> > 
> 
> Yes cmd is from userspace, but we are in a switch-case that has already
> matched cmd == DMA_HEAP_IOC_ALLOC which is sized correctly.
> 

Ah yeah, sorry I missed that. It's fine until the struct gets
extended, which can be handled if/when.

[snip]

> > 
> > For me, the main thing that's missing is a way to unregister a heap.
> > If drivers/modules are registering heaps, then they also need to be
> > able to remove them again when they go away (rmmod or whatever).
> > 
> > I think that starts to be quite a can of worms, as you might end up
> > with buffers which outlive their allocator (that's true today as well
> > afaik, but is in drivers rather than something more central).
> > 
> 
> Yeah, was wanting to avoid dealing with all that, at least to start out
> with. At a high level memory shouldn't ever just disappear, we should
> block removing a heap driver until all buffers have been released.
> 

Correct me if I'm wrong, but I don't think there's a mechanism in the
kernel to do that right now is there? If my DRM driver registers a
heap, and then gets rmmod-ed, then kaboom. Just not exporting the
symbol might be OK to start with.

Thanks,
-Brian