[RFC 03/34] dmaengine: virt-dma: add support for cyclic DMA periodic callbacks

[Russell King - ARM Linux]

On Thu, May 31, 2012 at 11:45:54AM +0800, zhangfei gao wrote:
> On Wed, May 30, 2012 at 4:02 PM, Russell King - ARM Linux
> <linux at arm.linux.org.uk> wrote:
> > On Wed, May 30, 2012 at 03:52:53PM +0800, Linus Walleij wrote:
> > No, I don't think we should do that - ALSA doesn't require it, and ALSA
> > is about the only user of this.  As I've said, DMA engine drivers should
> > implement a correct tx_status function, which returns the remaining bytes
> > to be transferred to the end of the buffer so it's possible to find out
> > how many periods have completed when the callback is called.
> 
> Could we directly use pos, but unmatch with dma_tx_state.residue.
> For cyclic buffer, when pos reach to buffer size, pos will change to 0,
> remaining bytes also 0, we may have to distinguish them.

What pos?

The way I said above, you work out how many periods have expired by
using dma_tx_state.residue.  This tells you how many bytes are left to
transfer until the _end_ of the looped buffer.

>From that, you can tell directly how many periods have expired.

Really, ALSA is not interested in "how many periods have expired" but
is really after a regular update of where DMA is _now_, so it knows
how many bytes it can write into or read from the circular buffer.  It
really doesn't need to know that N periods have expired _if_ you're
capable of telling it exactly where the DMA is in the buffer.

And with a working dma_tx_state.residue, you can do exactly that.