[PATCH 2/2] ASoC: dmaengine_pcm: add snd_dmaengine_generic_pcm_open()

Sebastien Guiriec s-guiriec at ti.com
Fri Mar 22 04:07:21 EDT 2013


On 03/21/2013 03:53 PM, Shawn Guo wrote:
> On Thu, Mar 21, 2013 at 03:27:18PM +0530, Vinod Koul wrote:
>> On Fri, Mar 15, 2013 at 11:36:41AM +0800, Shawn Guo wrote:
>>> With generic DMA device tree binding and helper function
>>> dma_request_slave_channel() in place, dmaengine_pcm should support
>>> that in requesting DMA channel for users that support generic DMA
>>> device tree binding.
>>>
>>> Add a new API snd_dmaengine_generic_pcm_open() as the variant of
>>> snd_dmaengine_pcm_open().  It takes DMA client struct device pointer
>>> and slave channel name as arguments and calls generic DMA helper
>>> dma_request_slave_channel() to request DMA channel from dmaengine.
>> NAK
>>
>> This is why we have dma_request_slave_channel_compat() API. You dont need to
>> write two open handlers here, just pass all the arguments and if DT is set it
>> will allocate a channel using dma_request_slave_channel() otherwise will do
>> dma_request_channel which is what you need.
>
> I do not quite follow your comment.  Let me try to understand it.  Are
> you suggesting that instead of the current call path:
>
> 	snd_dmaengine_pcm_open()
> 		dmaengine_pcm_request_channel()
> 			dma_request_channel()
>
> we should change it as below?
>
> 	snd_dmaengine_pcm_open()
> 		dmaengine_pcm_request_channel()
> 			dma_request_slave_channel_compat()
>
> If that's the case, you are fundamentally requesting to change the
> fingerprint of API snd_dmaengine_pcm_open() from the existing:
>
> int snd_dmaengine_pcm_open(struct snd_pcm_substream *substream,
> 			   dma_filter_fn filter_fn, void *filter_data)
>
> to something like:
>
> int snd_dmaengine_pcm_open(struct snd_pcm_substream *substream,
> 			   dma_filter_fn filter_fn, void *filter_data,
> 			   struct device *dev, const char *name)
>
> So every single user of snd_dmaengine_pcm_open() needs to adapt to the
> interface change.
>
> Is my understanding correct?  Or have I misunderstood your comment?

This is what has been done on OMAP for other IPs. I have tested this 
solution for Audio before you submit your first series and it is working.

Now I will wait Lars serie to check again on OMAP and let Mark comment 
on the best approach. Whereas update current API like you propose or 
wait Lars series and move on according to the different comments done on 
open function. As OMAP audio DMA binding are depending on the chosen 
solution.

Sebastien.
>
> Shawn
>
>




More information about the linux-arm-kernel mailing list