Re: [PATCH 5/9] dmaengine: provide a common function for completing a dma descriptor

On Mon, 2012-04-23 at 10:50 +0100, Russell King - ARM Linux wrote:

On Mon, Apr 23, 2012 at 06:40:06PM +0900, Boojin Kim wrote:

quoted

I met a problem on DMA cyclic mode (DMA_CYCLIC) for sound playback.
Kernel BUG occurs during DMA transfer with DMA cyclic mode.
This patch makes the cookies into zero. But, cookies should be kept
during cyclic mode because cyclic mode re-uses the cookies.

The protection is there to prevent cookies being accidentally re-used.
If you're running a cyclic transfer, even then you shouldn't be completing
the same cookie time and time again - I think Vinod also concurs with this.

Right :)
I recently committed patch for imx-dma which doesn't mark the cyclic
descriptor as complete. Descriptor represents a transaction and makes no
sense to complete t if the transaction is still continuing.

I think our preference is for cyclic transfers to entire remain uncompleted,
or to get a new cookie each time they allegedly "complete".

No it is not complete. Cyclic never completes, it aborts when user
wants. The "notification" interrupt is for updating the
counters/notifying (timestamp/periods elapsed in sound), and shouldn't
be used for anything else

-- 
~Vinod