On Thu, Jun 06, 2013 at 12:39:20PM +0800, Nicolin Chen wrote:
On Wed, Jun 05, 2013 at 12:55:44PM +0100, Mark Brown wrote:
+- hp-det-gpios : The gpio port of Headphone detection. +- mic-det-gpios: The gpio port of Micphone detection.
I'd say a bit more about these - obviously the CODEC has its own accessory detection here, and I rather suspect that in fact the HP detection you have there is really jack detection.
hasn't been included in this patch though. So since hardware has pins here, I think it's better to put in the binding doc as well. But I guess it'd be better to put it into OPTIONAL area?
Yes, they need to be optional and you need to explain what the function of these pins is so people can tell what they are. As I said I really do expect that the pin you have labelled headphone detect is really a jack detect pin.
The sample rate checking can be done with the symmmetric_rates feature in the core. The sample_bits check can't be but it's something that a lot of systems probably ought to have so it ought to be factored out into the core too rather than open coded in the driver.
[Nic:] fsl_ssi.c uses symmmetric_rates, but last time I traced soc-pcm.c and found the core only cares about symmmetric_rates during pcm_open(), which means the startup() in dai/machine driver, while what I'm considering is something like "arecord -Dhw:0 -r 44100 | aplay -Dhw:0 -r 48000": The two startup() would run almost at the same time and each of them would be earlier than hw_param() -- the exact point we can get the sample rate from application, but before this point no sample-rate actually be set to make hw_constraint(). Then the two hw_param() would continue their code and successively call set_fll() twice with different sample rates. But I agree that it's kinda ugly to put code here. So I think maybe I need to drop this part of code and to think about another patch for the core?
Yes, you certainly need to drop this patch. There is no way of fixing this with the current ALSA APIs, they are fundamentally racy and you will always have the possibility of failures during simultaneous startup for hardware with these limits.
You're enabling and disabling the CODEC only while there's an audio stream active. This means that it's not possible to support analogue bypass paths (which the device can do) - you should probably also support enabling the FLL via set_bias_level() and use set_bias_level() to turn it off (which works in all cases).
[Nic] I've seen and tested set_bias_level() way as samsung/tobermory.c does. But I find a flaw to the method that if I play two and more wav files like 'aplay -Dhw:0 audio44Khz.wav audio48Khz.wav audio22Khz.wav', which needs to set_fll() before each playback and to keep SYSCLK valid by changing its source to MCLK after the playback, but I only found that hw_param() and hw_free() are symmetrically called before/after each playback in this case.
Two things here. One is that of course there's no reference counting on FLL enables, they just happen, and the other is that as you'd expect the bias level callbacks just won't happen for the second stream as the device is powered up. You only need to set the clocking up for the first stream as the clocking can't be changed when the device is active.