On Tue, Aug 17, 2010 at 6:46 PM, Mark Brown broonie@opensource.wolfsonmicro.com wrote:
On Tue, Aug 17, 2010 at 08:14:43AM +0900, Jassi Brar wrote:
I guess most quality conscious products can afford to attach a dedicated OSC to a good CODEC. A CODEC already specifies the clocks it support and usually comes with 'preferred' input clocks. So, just having a specified rating accurate enough OSC can take care of quality. On the other hand the clock sources on CPU side are not particularly accurate for audio-clock generation. Or so have I seen so far.
Right, this shouldn't be an issue with the change I proposed - the clocks are still rooted from the CODEC and its clock generation facilities would still be used, it's just that they get routed differently.
The issue with using most CPU clocks isn't really the dedicated oscillator so much as the ability to generate non-integer divisions of whatever input clock is available. If this can't be done then it can be hard to generate the full range of common audio frequencies.
Ok, now that I am faced with supporting latest SMDKs that use codecs other than WM8580 (like wm8994), I think we'd better break the MACHINE driver into SMDK specific (setup CPU master, CODEC slave and root-clock sourced acc to option provided vai platform_data) and CODEC specific part. That will also help code reuse.
Any opinions, before I start working on it?
Thanks, Jassi