Re: [PATCH] mips: Add dma_mmap_coherent()

[James Bottomley <James.Bottomley@HansenPartnership.com>]

On Wed, 2008-08-27 at 07:42 +0200, Takashi Iwai wrote:
> At Tue, 26 Aug 2008 15:01:18 -0600,
> Grant Grundler wrote:
> > 
> > On Tue, Aug 26, 2008 at 05:25:24PM +0200, Takashi Iwai wrote:
> > ...
> > > Now updated my git tree:
> > >     http://git.kernel.org/?p=linux/kernel/git/tiwai/sound-2.6.git;a=shortlog;h=topic/dma-fix
> > > I'll post each patch again if preferred.
> > 
> > +#ifdef CONFIG_SND_COHERENT_DMA
> >  #define SNDRV_DMA_TYPE_DEV_SG          3       /* generic device SG-buffer */
> > +#else
> > +#define SNDRV_DMA_TYPE_DEV_SG  SNDRV_DMA_TYPE_DEV /* no SG-buf support */
> > +#endif
> > 
> > Hi Takashi,
> > I had to look at a previous patch to figure out CONFIG_SND_COHERENT_DMA
> > is an arch dependent flag:
> > 
> > +config SND_COHERENT_DMA
> > +       def_bool y
> > +       depends on !PPC32 || !NOT_COHERENT_CACHE
> > +       depends on !ARM
> > +       depends on !MIPS
> > +       depends on !PARISC
> > 
> > In general, I don't expect this to be a compile time option.
> 
> Right now it has to be a compile-time option because
> - dma_mmap_coherent() isn't implemented in every architecture (thus
>   fails to build), and 
> - pages allocated via dma_mmap_coherent() aren't always suitable for
>   SG-mapping.

This is trivially fixable by the usual methods, so, as Grant says, we
should employ them rather than non-standard ways of doing this

Basically, you're asking to extend the DMA API, so this should be taken
to linux-arch.  That way, it might also give visibility to the graphics
people and we can negotiate over a unified API.

James