Hi Vinod, Laurent > On Tuesday 16 December 2014 22:09:04 Vinod Koul wrote: > > On Tue, Dec 16, 2014 at 09:03:38AM +0000, Kuninori Morimoto wrote: > > > Hi Vinod > > > > > > > > This rcar-audmapp driver is very limited usage, and user is only sound > > > > > device/driver. Sound driver configures both 1st/2nd DMAC if needed > > > > > (it depends on platform). Sound driver knows all reg address / mem > > > > > address which are needed for 1st/2nd DMAC settings. 1st/2nd DMAC > > > > > needs general DMAEngine settings method, not special. Now, sound > > > > > driver + 1st/2nd DMAC works well on my local environment.> > > > > > Are you not using the sound dmaengine library then, right? > > > > > > At this point, yes. > > > But, I got request from Mark to use it > > > > Yes that is the subsytem recommendation > > > > > # current dmaengine "method" on my driver is same as library. > > > # but I need to fix non-dmaengine area > > > > > > > One more question, audio data will be in system memory and then it needs > > > > to be transfered to periphral FIFO, how will data travel thru these two > > > > DMAs? > > > > > > like this > > > > > > 1st DMA 2nd DMA > > > > > > [mem] -> [periphral IP-A] -> [periphral IP-B] -> speaker > > > > Okay and by current implementation you wont be able to use the generic > > dmaengine sound layer. That layer assumes you will have only one dmaengine > > channel to configure and will configure only one DMA and which is 1st DMA > > here. > > > > So what do we do we 2nd one here. The point that 2nd DMA is behind the 1st > > DMA, should the users know about it and configure? I am not sure... > > If not, then who should configure this. Another candidate is 1st DMA, but > > then from the point of 1st DMA should it configure 2nd DMA as well, treat it > > as its salve... sounds plausible but am not too sure... comments? > > The second DMA is not a slave of the first one, given that there's a > peripheral in-between. > > Not that I'm advocating for it, but given how simple the second DMA engine is, > and given that it is dedicated to a single function (transferring data between > sound-related IPs, nothing else is possible) and can't thus be used as a > general purpose DMA engine, I wouldn't mind handling it internally as part of > the sound drivers instead of exposing it through the DMA engine API. I need to explain more detail about data path. Our sound has SSI/SRC/DVC, and the sound data paths are pattern 1) 1st DMA [mem] -> [SSI] -> speaker pattern 2) 1st DMA 2nd DMA [mem] -> [SRC] -> [SSI] -> speaker pattern 3) 1st DMA 2nd DMA [mem] -> [SRC] -> [DVC] -> [SSI] -> speaker SRC : 9 channel DVC : 3 channel SSI : 9 channel Data path is board/platform specific at this point. (it is nice if we can select these pattern flexibility in the future) Unfortunately, this channel size is depends on SoC, and above is "playback" pattern only, we need to have "capture" pattern too. (similar path, but different direction and different ID). Now, it is controlled under sound driver, because sound knows all information and, board/platform specific data path. I thought that it is possible if sound driver uses 2 generic sound DMA layer, but is it wrong ? Best regards --- Kuninori Morimoto -- To unsubscribe from this list: send the line "unsubscribe dmaengine" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
