Re: [PATCH/RFC 1/2] V4L2: soc-camera: work around unbalanced calls to .s_power()

[Guennadi Liakhovetski <g.liakhovetski@gmx.de>]

Hi Frank

On Wed, 23 Oct 2013, Frank Schäfer wrote:

> Hi Guennadi,
> 
> thanks for the patches and sorry for the delay.
> I've tested them a few minutes ago and they are working fine.
> 
> 2 minor things/questions:
> 
> 1.) Why not always balance v4l2_clk_enable/disable() calls ?

Because I consider this a work-around for legacy / buggy drivers, that we 
cannot fix properly. New drivers should balance power-on / off calls.

> 2.) For someone who reads the code it likely looks a bit odd that if the
> flag "unbalanced_power" is set, soc_camera balances
> v4l2_clk_enable/disable but not power on/off calls (ssdd->power() calls).
> So maybe "balance_clk" or "clk_balancing_needed" would be a better name
> for the flag ?

Well, the name makes sense to me - the user doesn't balnce calls to power 
enable / disable methods, so, we have to balance clock enable / disable 
ourselves. The name might not look very logical in soc-camera, but it's a 
flag for the user, and the user doesn't know about clocks. It just tells 
the subdevice - sorry, I won't be balancing power on / off calls.

> Ok, let's summarize what we need:
> - the 3 fake clock patches patches
> - em28xx patch "make sure that all subdevices are powered on when needed"
> - these 2 patches

In fact I don't even think the second entry - your em28xx patch - is 
needed strictly speaking. It might be correct and good, , but this 
specific issue should be fixed also without it. Could you confirm?

> All patches need to marked for stable (3.11).
> Can you pick up my me28xx patch and send Mauro a pull request ?

I'd first like to get a confirmation from Mauro (at least), that this 
approach is acceptable for him.

Thanks
Guennadi

> 
> Regards,
> Frank
> 
> Am 21.10.2013 11:28, schrieb Guennadi Liakhovetski:
> > Some non soc-camera drivers, e.g. em28xx, use subdevice drivers, originally
> > written for soc-camera, which use soc_camera_power_on() and
> > soc_camera_power_off() helpers to implement their .s_power() methods. Those
> > helpers in turn can enable and disable a clock, if it is supplied to them
> > as a parameter. This works well when camera host drivers balance their
> > calls to subdevices' .s_power() methods. However, some such drivers fail to
> > do that, which leads to unbalanced calls to v4l2_clk_enable() /
> > v4l2_clk_disable(), which then in turn produce kernel warnings. Such
> > behaviour is wrong and should be fixed, however, sometimes it is difficult,
> > because some of those drivers are rather old and use lots of subdevices,
> > which all should be tested after such a fix. To support such drivers this
> > patch adds a work-around, allowing host drivers or platforms to set a flag,
> > in which case soc-camera helpers will only enable the clock, if it is
> > disabled, and disable it only once on the first call to .s_power(0).
> >
> > Signed-off-by: Guennadi Liakhovetski <g.liakhovetski@gmx.de>
> > ---
> >
> > As promised yesterday, this is an alternative approach to fixing the 
> > em28xx problem, compile tested only.
> >
> >  drivers/media/platform/soc_camera/soc_camera.c |   22 ++++++++++++++++------
> >  include/media/soc_camera.h                     |   14 ++++++++++++++
> >  2 files changed, 30 insertions(+), 6 deletions(-)
> >
> > diff --git a/drivers/media/platform/soc_camera/soc_camera.c b/drivers/media/platform/soc_camera/soc_camera.c
> > index 387a232..21136a2 100644
> > --- a/drivers/media/platform/soc_camera/soc_camera.c
> > +++ b/drivers/media/platform/soc_camera/soc_camera.c
> > @@ -71,11 +71,21 @@ static int video_dev_create(struct soc_camera_device *icd);
> >  int soc_camera_power_on(struct device *dev, struct soc_camera_subdev_desc *ssdd,
> >  			struct v4l2_clk *clk)
> >  {
> > -	int ret = clk ? v4l2_clk_enable(clk) : 0;
> > -	if (ret < 0) {
> > -		dev_err(dev, "Cannot enable clock: %d\n", ret);
> > -		return ret;
> > +	int ret;
> > +	bool clock_toggle;
> > +
> > +	if (clk && (!ssdd->unbalanced_power ||
> > +		    !test_and_set_bit(0, &ssdd->clock_state))) {
> > +		ret = v4l2_clk_enable(clk);
> > +		if (ret < 0) {
> > +			dev_err(dev, "Cannot enable clock: %d\n", ret);
> > +			return ret;
> > +		}
> > +		clock_toggle = true;
> > +	} else {
> > +		clock_toggle = false;
> >  	}
> > +
> >  	ret = regulator_bulk_enable(ssdd->num_regulators,
> >  					ssdd->regulators);
> >  	if (ret < 0) {
> > @@ -98,7 +108,7 @@ epwron:
> >  	regulator_bulk_disable(ssdd->num_regulators,
> >  			       ssdd->regulators);
> >  eregenable:
> > -	if (clk)
> > +	if (clock_toggle)
> >  		v4l2_clk_disable(clk);
> >  
> >  	return ret;
> > @@ -127,7 +137,7 @@ int soc_camera_power_off(struct device *dev, struct soc_camera_subdev_desc *ssdd
> >  		ret = ret ? : err;
> >  	}
> >  
> > -	if (clk)
> > +	if (clk && (!ssdd->unbalanced_power || test_and_clear_bit(0, &ssdd->clock_state)))
> >  		v4l2_clk_disable(clk);
> >  
> >  	return ret;
> > diff --git a/include/media/soc_camera.h b/include/media/soc_camera.h
> > index 34d2414..5678a39 100644
> > --- a/include/media/soc_camera.h
> > +++ b/include/media/soc_camera.h
> > @@ -150,6 +150,15 @@ struct soc_camera_subdev_desc {
> >  	struct regulator_bulk_data *regulators;
> >  	int num_regulators;
> >  
> > +	/*
> > +	 * Set unbalanced_power to true to deal with legacy drivers, failing to
> > +	 * balance their calls to subdevice's .s_power() method. clock_state is
> > +	 * then used internally by helper functions, it shouldn't be touched by
> > +	 * drivers or the platform code.
> > +	 */
> > +	bool unbalanced_power;
> > +	unsigned long clock_state;
> > +
> >  	/* Optional callbacks to power on or off and reset the sensor */
> >  	int (*power)(struct device *, int);
> >  	int (*reset)(struct device *);
> > @@ -206,6 +215,11 @@ struct soc_camera_link {
> >  	struct regulator_bulk_data *regulators;
> >  	int num_regulators;
> >  
> > +	/* Set by platforms to handle misbehaving drivers */
> > +	bool unbalanced_power;
> > +	/* Used by soc-camera helper functions */
> > +	unsigned long clock_state;
> > +
> >  	/* Optional callbacks to power on or off and reset the sensor */
> >  	int (*power)(struct device *, int);
> >  	int (*reset)(struct device *);
> 

---
Guennadi Liakhovetski, Ph.D.
Freelance Open-Source Software Developer
http://www.open-technology.de/