[alsa-devel] [PATCH v2] ASoC: atmel_ssc_dai: Allow more rates

[Peter Rosin]

Bo Shen wrote:
> Hi Peter,
> 
> On 02/09/2015 03:35 PM, Peter Rosin wrote:
> > Bo Shen wrote:
> >> Hi Peter,
> >
> > Hi!
> >
> >> On 02/07/2015 06:51 PM, Peter Rosin wrote:
> >>> Mark Brown wrote:
> >>>> On Wed, Feb 04, 2015 at 12:52:25PM +0100, Peter Rosin wrote:
> >>>>
> >>>>> One thing remains a bit unclear, and that is the 500ppm deduction.
> >>>>> Is that really warranted? The number was just pulled out of my hat...
> >>>>
> >>>> I don't really get what this is supposed to be protecting against.
> >>>>
> >>>>> +	case SND_SOC_DAIFMT_CBM_CFS:
> >>>>> +	case SND_SOC_DAIFMT_CBM_CFM:
> >>>>> +		t.min = 8000;
> >>>>> +		t.max = ssc_p->mck_rate / mck_div / frame_size;
> >>>>> +		/* Take away 500ppm, just to be on the safe side. */
> >>>>> +		t.max -= t.max / 2000;
> >>>>> +		t.openmin = t.openmax = 0;
> >>>>> +		t.integer = 0;
> >>>>> +		ret = snd_interval_refine(i, &t);
> >>>>
> >>>> As I understand it this is a straight divider rather than something
> >>>> that's doing dithering or anything else more fancy.  Given that it
> >>>> seems as well just to trust the clock rate we've got - we don't do
> >>>> any error tracking with the clock API (perhaps we should) and for
> >>>> many applications some degree of divergence from the nominal rate
> >>>> is not
> >>>> *too* bad for audio systems (for application specific values of "some"
> >>>> and "too" of course).  If it is just dividers I'm not sure the
> >>>> situation is really improved materially by knocking off the top frequency.
> >>>>
> >>>> If we are doing something more fancy than divididing my analysis is
> >>>> off base of course.
> >>>
> >>> I'm thinking that the SSC samples the selected BCK pin using the
> >>> (possibly
> >>> divided) peripheral clock. Getting too near the theoretical rate
> >>> limit would be bad, if these two independent clocks drift the wrong
> >>> way. At least that is my take on it, but I don't know the internal workings of the SSC, so...
> >>>
> >>> I was hoping that someone from Atmel could chime in? Maybe I'm
> >>> totally
> >>
> >> Sorry for late response.
> >
> > No problem!
> >
> >>> off base, and the SSC is doing this completely differently?
> >>
> >> What you mean here? I am not sure I fully understand.
> >
> > The SSC spec list a maximum rate (which varies with the direction of
> > various signals, ignoring that for the sake of this explanation). Lets
> > assume that this maximum rate is 11MHz, derived from the peripheral
> > clock which might be 66MHz. If you then try to input an 11MHz signal
> > derived from some unrelated xtal you might think it should work. My
> > theory was that the rate limit would be broken if the peripheral clock
> > wasn't really 66MHz, but instead a few ppm lower than nominal, and the
> > unrelated xtal was a few ppm higher than nominal.
> >
> > If this matters or not depends on how the SSC is implemented.
> 
> This is to let the user to know the clock limitation, am I right?

Yes, sort of, to prevent the user from even attempting to go too
near the nominal limit.

> And at the same time deal with the un-precise clock which come to SSC?
> If this case, I think we should trust the clock come to SSC.

Ok, I'll just kill the 500ppm thing for the next round. I'll wait a bit
for the discussion in the other branch to fade out though. :-)

Cheers,
Peter

> > There might be other reasons for not caring all that much about this
> > fringe case, and just trust the nominal rates and limits.
> >
> >>> In our application, we're not near the limit. Therefore, it really
> >>> doesn't matter much to us.
> >>>
> >>> Should I resend w/o the 500ppm deduction?
> >>>
> >>> Cheers,
> >>> Peter
> >>>
> >>
> >> Best Regards,
> >> Bo Shen
> 
> Best Regards,
> Bo Shen