Re: [PATCH v2] ASoC: fsl_sai: Add isr to deal with error flag

[Nicolin Chen <Guangyu.Chen@freescale.com>]

On Thu, Mar 27, 2014 at 01:26:27PM +0000, Mark Brown wrote:
> On Thu, Mar 27, 2014 at 07:06:59PM +0800, Nicolin Chen wrote:
> > It's quite cricial to clear error flags because SAI might hang if getting
> > FIFO underrun during playback (I haven't confirmed the same issue on Rx
> > overflow though).
> > 
> > So this patch enables those irq and adds isr() to clear the flags so as to
> > keep playback entirely safe.
> 
> So, I've applied this since we're (hopefully!) very near the merge
> window opening and it seems like it should be an improvement overall.
> However a few things below:
> 
> > +	/* Only handle those what we enabled */
> > +	mask = (FSL_SAI_FLAGS >> FSL_SAI_CSR_xIE_SHIFT) << FSL_SAI_CSR_xF_SHIFT;
> 
> The shifting here could use a comment.

Will send an extra patch to cover it.

> > +	regmap_update_bits(sai->regmap, FSL_SAI_RCSR,
> > +			   FSL_SAI_CSR_xF_W_MASK | FSL_SAI_CSR_FR, xcsr);
> 
> Using update_bits() is going to do an extra read, better to do this as:
> 
> 	if (xcsr)
> 		regmap_write(sai->regmap, FSL_SAI_RCSR, xcsr);

T/RCSR register stands for T/Rx Control (Status -- I guess) Register.
It actually contains control bits, IRQ mask bits and IRQ status bits.
Thus we can't regard it as a entire status register. That's why I try
to update it partially for status bits only at this point.

But if it's just for saving this extra read instance, I may save those
non-status bits from upper regmap_read() and then merge them into this
regmap_write() over here.

> otherwise we might be ignoring any of the bits that are actually clear
> on read (it seems like there are some?).

For all status flags bits, actually five for each, three of them are
write-1-clear and the other two are self-clearance by SAI self -- so
none of them are clear-on-read type.

> > +	return IRQ_HANDLED;
> 
> I'd expect to see IRQ_NONE if we didn't actually see an interrupt
> source.

Will add this.

Thank you,
Nicolin