Re: [PATCH v3 1/1] drm: msm: Replace dma_map_sg with dma_sync_sg*

From: Daniel Vetter
Date: Fri Nov 30 2018 - 04:46:11 EST

On Thu, Nov 29, 2018 at 07:33:34PM +0100, Christoph Hellwig wrote:
> On Thu, Nov 29, 2018 at 06:09:05PM +0100, Daniel Vetter wrote:
> > What kind of abuse do you expect? It could very well be that gpu folks
> > call that "standard use case" ... At least on x86 with the i915 driver
> > we pretty much rely on architectural guarantees for how cache flushes
> > work very much. Down to userspace doing the cache flushing for
> > mappings the kernel has set up.
> Mostly the usual bypasses of the DMA API because people know better
> (and with that I don't mean low-level IOMMU API users, but "creative"
> direct mappings).

Ah yes. I think that even gpu folks get behind :-) Best if someone
bothered to explicitly cast to dma_addr_t to shut up the tools, but not
fix the bug ...

> > > As for the buffer sharing: at least for the DMA API side I want to
> > > move the current buffer sharing users away from dma_alloc_coherent
> > > (and coherent dma_alloc_attrs users) and the remapping done in there
> > > required for non-coherent architectures. Instead I'd like to allocate
> > > plain old pages, and then just dma map them for each device separately,
> > > with DMA_ATTR_SKIP_CPU_SYNC passed for all but the first user to map
> > > or last user to unmap. On the iommu side it could probably work
> > > similar.
> >
> > I think this is what's often done. Except then there's also the issue
> > of how to get at the cma allocator if your device needs something
> > contiguous. There's a lot of that still going on in graphics/media.
> Being able to dip into CMA and mayb iommu coalescing if we want to
> get fancy is indeed the only reason for this API. If we just wanted
> to map pages we could already do that now with just a little bit
> of boilerplate code (and quite a few drivers do - just adding this
> new API will remove tons of code).

Sounds like the future is very bright indeed \o/
Daniel Vetter
Software Engineer, Intel Corporation