Re: [PATCH 2/3] ARM: edma: Rename header file for dmaengine filter function definition

[Peter Ujfalusi <peter.ujfalusi@ti.com>]

On 11/28/2014 12:51 PM, Arnd Bergmann wrote:
> On Friday 28 November 2014 09:16:24 Peter Ujfalusi wrote:
>> On 11/27/2014 11:52 PM, Arnd Bergmann wrote:
>>> On Thursday 27 November 2014 20:46:12 Peter Ujfalusi wrote:
>>>>
>>>> I see. With this series I did not planed to fix all edma related issues, just
>>>> as a start clean up the related header files. I would rather not add fixes to
>>>> mmc, spi, etc drivers since while you have valid point it is not in the scope
>>>> of this series.
>>>> Can we do the changes you are suggesting in an incremental manner?
>>>
>>> Sure, but I'd leave the existing filter function declaration alone then
>>> and not move it, since we wouldn't want to keep it in the long run.
>>
>> but if you want to reference the filter function (which is in
>> drivers/dma/edma.c) in arch/arm/mach-davinci/ directory, we will need it.
>> Don't we?
> 
> Yes, unless you move the definition of the filter function into
> arch/arm/common/edma.c or arch/arm/mach-davinci/devices.c, but that
> would require other changes.

At the end the aim is to get rid of the edma code form arch/arm and have only
dmaengine API towards eDMA. The ASoC davinci-pcm is the only user of the
legacy API AFAIK. It has a mode called ping-pong which is not possible with
the dmaeingine at all. This is to overcome underflow situations on parts where
the audio IP does not have FIFO.
My edma-pcm (which is using dmaengine) should be able to handle this
situation, but I need to verify it before I can remove the davinci-pcm and
then we can get rid of the direct eDMA API and code.

>> If I leave the header as it is, then how would we clean up the edma headers? I
>> would not put the API definitions for the arch code into the same file as we
>> have the filter definition.
> 
> Ok, just go ahead with your current patch then, we can always follow up.
> The most important cleanup for edma is elsewhere anyway, so once the asoc
> drivers can use the dmaengine interface, this should be easier.
> 
> 	Arnd
> 


-- 
Péter