Re: [PATCH] dmaengine: at_hdmac: add DMA slave transfers

[Dan Williams <dan.j.williams@intel.com>]

Atsushi Nemoto wrote:
> On Mon, 27 Jul 2009 14:24:26 +0100, "Sosnowski, Maciej" <maciej.sosnowski@intel.com> wrote:
>>>> Your atc_chain_complete() calls dma_unmap_xxx unless
>>>> DMA_COMPL_SKIP_XXX_UNMAP specified.  But atmel-mci driver does not set
>>>> the flag on dma_async_tx_descriptor.  I suppose one of them should be
>>>> fixed.
>>> atmel-mci should definitely set that flag.
>>>
>>> Haavard
>> I agree with Haavard.
> 
> Then, what should dma driver do when client driver did not set these
> flags?  If it should call dma_unmap_sg(), the dma driver should keep
> sg and direction somewhere...
> 
> Also, calling dma_map_sg() in its prep_slave_sg function will not fit
> for sound drivers, which use DMA buffers prepared in its framework.
> 
> For slave DMA, doing all mapping/unmapping in DMA client is better,
> isn't it?
> 

Yes it is.  The whole point of the dma driver doing its own unmapping is 
specifically for clients that use the async_tx api, i.e. where they do 
not know if the transaction is carried out in hardware and have no way 
of tracking the details necessary to perform the unmap.  DMA-slave 
clients request specific channels and know the hardware details at a low 
level, so it should not be too high an expectation to push dma mapping 
responsibility to the client.

--
Dan