Re: [PATCH 2/2] dw_dmac: add cyclic API to DW DMA driver

[Hans-Christian Egtvedt <hans-christian.egtvedt@atmel.com>]

On Wed, 1 Apr 2009 10:28:30 +0200
Haavard Skinnemoen <haavard.skinnemoen@atmel.com> wrote:

> Sosnowski, Maciej wrote:
> > > +void dw_dma_cyclic_stop(struct dma_chan *chan)
> > > +{
> > > +	struct dw_dma_chan	*dwc = to_dw_dma_chan(chan);
> > > +	struct dw_dma		*dw =
> > > to_dw_dma(dwc->chan.device); +
> > > +	channel_clear_bit(dw, CH_EN, dwc->mask);
> > > +	while (dma_readl(dw, CH_EN) & dwc->mask)
> > > +		cpu_relax();
> > > +}  
> > 
> > Don't you need locks in dw_dma_cyclic_stop?
> 
> Good question. On one hand, if cyclic_start() can race with
> cyclic_stop(), the client probably has more serious issues to deal
> with. On the other hand, if something ever manages to set the CH_EN
> bit at the wrong moment, the loop may never finish.
> 
> So it's probably safest to wrap it in spin_lock_bh(). You should
> probably add a note that this function can not be called from
> interrupt context too.
> 

The problem then is when calling cyclic_stop from the trigger function
in an ALSA driver.

http://www.kernel.org/pub/linux/kernel/people/tiwai/docs/writing-an-alsa-driver/ch05s06.html#pcm-interface-operators-trigger-callback

This callback is atomic, and with interrupts disabled. So using
spin_lock_bh() is AFAICT not allowed.

I am going to brush up the interface and document that the _stop and
_start interface has to be called with interrupts disabled. Will post a
v2 shortly.

-- 
Best regards,
Hans-Christian Egtvedt