On Tuesday 27 January 2009 13:37:38 ext Mark Brown wrote:
On Tue, Jan 27, 2009 at 11:29:44AM +0200, Peter Ujfalusi wrote:
It seams that the analog loopback needs the DAC powered on on the channel, where the loopback is selected. The switch for the DACs has been moved from
Oh, well...
Yes, it took a while to figure out what is wrong...
After the loopback registers are changed, the event callback will update the bypass bitfield, so we will know if any of the bypass paths are enabled. If the codec is in STANDBY mode, the decision is made to keep/switch the codec off or keep/switch the codec on to allow the bypass. Also after the codec switches from ON to STANDBY the same check will be done.
The original implementation was done that way due to the lack of DAPM - it did *all* power management in the bias configuration. Now that you have DAPM support it should be possible for the driver to function more normally and avoid these checks. What most drivers do is bring all the chip wide power up when coming into standby then power on the rest under DAPM.
Well, I had the implementation like that. There were few 'issues' which resulted the current code - and that code is long gone :(. So the codec was powered at all time, except, if it would recieve the SND_SOC_BIAS_OFF event. To put things into context, let's say that the power consumption of the board when the codec turned off is 0 A. When the bypass was enabled (to external speaker) it jumped to 0.016 A, which is fine, since an external amp is powered also. Than if the bypass got disabled it jumped back to 0.01 A. After a bit of debugging I found the reason: We have outputs with dedicated gain controls (PreDriveL/R, HeadsetL/R, CarkitL/R, Earpiece), they have to be put to power-down mode (gain bits to 0). We do have volume controls for these, they are not in DAPM, but they can be modified. But now the problem is that something has to track the bypass states in userspace: if the bypass disabled, than has to mute all outputs, when enabled, it has to remember, what was the original gain for the outputs. So it's going to be a mess. Still after powering these down I have 0.006 A... Than I found another thing: PLL power. Swtiching it off APPL_CTL:APPL_EN bit resulted with 0.002 A. Now it is close enough. But than I played a bit with the registers, enabled the bypass (while the PLL was off) and the readings were: 0.020 A, in short: bypass on + PLL on = 0.016 A bypass on + PLL off = 0.020 A bypass off + PLL on = 0.006 A bypass off + PLL off = 0.002 A
Grrr, and I gave up. Oh, and we don't have control for the PLL (it is enabled all the time at the moment). I was thinking to use the SND_SOC_DAPM_PRE and SND_SOC_DAPM_POST for it but I don't know how to use those.
So I thought that I will write it in a way I have sent it and revisit later, when I (or someone else) have better idea to handle this.
Sorry for the long and most probably boring details...
Support for turning the codec completely off should really be provided by the core as an optional thing that can be done when the subsystem has been idle for a while using the regular bias level stuff like the existing delay before falling back to standby. It needs to be optional since won't be appropriate for all applications since bringing the bias levels up cleanly may take too long for some systems.
A few more specific issues in the rest of the code below. I've also got a patch I'm going to push to Takashi later today which conflicts with this one, see the for-2.6.30 branch of:
git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound-2.6.git
I can wait till it gets there and send the series on top of it.
static void twl4030_codec_enable(struct snd_soc_codec *codec, int enable) {
struct twl4030_priv *twl4030 =(struct twl4030_priv *)codec->private_data;Casts away from void aren't needed, there's quite a few of those here and in the rest of the patch code. Please fix these - they're usually a sign that something is wrong if they're needed.
codec = kzalloc(sizeof(struct snd_soc_codec), GFP_KERNEL); if (codec == NULL) return -ENOMEM;
twl4030 = kzalloc(sizeof(struct twl4030_priv), GFP_KERNEL);if (twl4030 == NULL) {Not essential but you could collapse these into a single allocation.
I'll fix these.
Thanks, Péter