[PATCH] DMA: let filter functions of of_dma_simple_xlate possible check of_node

[Richard Zhao]

On Mon, Aug 26, 2013 at 02:17:43PM +0200, Laurent Pinchart wrote:
> On Friday 23 August 2013 09:57:43 Stephen Warren wrote:
> > On 08/22/2013 07:29 PM, Richard Zhao wrote:
> > > On Fri, Aug 23, 2013 at 04:18:27AM +0800, Stephen Warren wrote:
> > >> On 08/21/2013 11:19 PM, Richard Zhao wrote:
> > >>> On Fri, Aug 02, 2013 at 10:00:00AM +0800, Richard Zhao wrote:
> > >>>> pass of_phandle_args dma_spec to dma_request_channel in
> > >>>> of_dma_simple_xlate, so the filter function could access of_node in
> > >>>> of_phandle_args.
> > >>>> 
> > >>>> It also remove restriction of #dma-cells has to be one.
> > >>>> 
> > >>>> Signed-off-by: Richard Zhao <rizhao at nvidia.com>
> > >>>> ---
> > >>>> 
> > >>>>  drivers/dma/edma.c     |  7 +++++--
> > >>>>  drivers/dma/of-dma.c   | 10 ++++------
> > >>>>  drivers/dma/omap-dma.c |  6 ++++--
> > >>>>  3 files changed, 13 insertions(+), 10 deletions(-)
> > >>> 
> > >>> Hi Vinod,
> > >>> 
> > >>> Can you please pick up this change?
> > >>> 
> > >>> Hi Stephen,
> > >>> 
> > >>> Can you please give a ack or reviewed-by etc?
> > >> 
> > >> Hmm. Looking at the patch, I'm not sure it's right.
> > >> 
> > >> This patch simply passes all the specfier args to the filter function,
> > >> and the code to check the equality of the of_node to the filter args is
> > >> still duplicated in each DMA driver. Instead, the DMA core should be
> > >> implementing the equality check, and only even calling the
> > >> driver-specific filter function for devices where the client's phandle
> > >> matches the DMA providing device's of_node handle.
> > > 
> > > Filter function is called in dmaengine core code, independent of dt.
> > 
> > The core code can still check if a dmaengine's driver was instantiated
> > from DT and take additional actions in that case.
> > 
> > > And the reason why the driver has to write its own filter function is
> > > it has to store slave id there in its own way.
> > 
> > I'm not saying don't call the driver's filter function, but rather that
> > the dmaengine core should perform the common checks before doing so.
> 
> And it looks to me like the common case could even get rid of the driver's 
> filter function:
> 
> https://lkml.org/lkml/2013/5/15/270
> https://lkml.org/lkml/2013/3/25/250
For general case, the slave id is not staticly bind to a specific channel.

Thanks
Richard