Re: [PATCH 5/5] dma-mapping: remove ARCH_NO_COHERENT_DMA_MMAP

[Christoph Hellwig]

On Fri, Aug 02, 2019 at 10:24:02AM +0200, Takashi Iwai wrote:
> I wasn't careful enough to look at that change, sorry.
> 
> The code there tries to check whether dma_mmap_coherent() would always
> fail on some platforms.  Then the driver clears the mmap capability
> flag at the device open time and notifies user-space to fall back to
> the dumb read/write mode.
> 
> So I'm afraid that simply dropping the check would cause the behavior
> regression, e.g. on PARISC.
> 
> Is there any simple way to test whether dma_mmap_coherent() would work
> or not in general on the target platform?  It's not necessarily in an
> ifdef at all.

This isn't really a platform, but a per-device question.  I can add a
"bool dma_can_mmap(struct device *dev)" helper to check that.  But how
do I get at a suitable struct device in hw_support_mmap()?