Re: [PATCH] rcar-csi2: Use standby mode instead of resetting

["Niklas Söderlund" <niklas.soderlund@ragnatech.se>]

Hi Jacopo,

Thanks for your feedback.

On 2019-02-17 20:41:24 +0100, Jacopo Mondi wrote:
> Hi Niklas,
>    ah, ups, this was maybe the patch the other one I just reviewed was
>    based on... sorry, I missed this one :)

:-)

> 
> On Sat, Feb 16, 2019 at 11:56:38PM +0100, Niklas Söderlund wrote:
> > Later versions of the datasheet updates the reset procedure to more
> > closely resemble the standby mode. Update the driver to enter and exit
> > the standby mode instead of resetting the hardware before and after
> > streaming is started and stopped.
> >
> > While at it break out the full start and stop procedures from
> > rcsi2_s_stream() into the existing helper functions.
> >
> > Signed-off-by: Niklas Söderlund <niklas.soderlund+renesas@ragnatech.se>
> > ---
> >  drivers/media/platform/rcar-vin/rcar-csi2.c | 69 +++++++++++++--------
> >  1 file changed, 42 insertions(+), 27 deletions(-)
> >
> > diff --git a/drivers/media/platform/rcar-vin/rcar-csi2.c b/drivers/media/platform/rcar-vin/rcar-csi2.c
> > index f64528d2be3c95dd..f3099f3e536d808a 100644
> > --- a/drivers/media/platform/rcar-vin/rcar-csi2.c
> > +++ b/drivers/media/platform/rcar-vin/rcar-csi2.c
> > @@ -14,6 +14,7 @@
> >  #include <linux/of_graph.h>
> >  #include <linux/platform_device.h>
> >  #include <linux/pm_runtime.h>
> > +#include <linux/reset.h>
> >  #include <linux/sys_soc.h>
> >
> >  #include <media/v4l2-ctrls.h>
> > @@ -350,6 +351,7 @@ struct rcar_csi2 {
> >  	struct device *dev;
> >  	void __iomem *base;
> >  	const struct rcar_csi2_info *info;
> > +	struct reset_control *rstc;
> >
> >  	struct v4l2_subdev subdev;
> >  	struct media_pad pads[NR_OF_RCAR_CSI2_PAD];
> > @@ -387,11 +389,19 @@ static void rcsi2_write(struct rcar_csi2 *priv, unsigned int reg, u32 data)
> >  	iowrite32(data, priv->base + reg);
> >  }
> >
> > -static void rcsi2_reset(struct rcar_csi2 *priv)
> > +static void rcsi2_standby_mode(struct rcar_csi2 *priv, int on)
> >  {
> > -	rcsi2_write(priv, SRST_REG, SRST_SRST);
> > +	if (!on) {
> 
> minor thing: if (!on) { "wakeup"; } is confusing. What if you call the
> variable "standby" or just "off" ?

I agree this was a bad design, I will split this function in two.

    rcsi2_enter_standby()
    rcsi2_exit_standby()

> 
> > +		pm_runtime_get_sync(priv->dev);
> > +		reset_control_deassert(priv->rstc);
> > +		return;
> > +	}
> > +
> > +	rcsi2_write(priv, PHYCNT_REG, 0);
> > +	rcsi2_write(priv, PHTC_REG, PHTC_TESTCLR);
> > +	reset_control_assert(priv->rstc);
> >  	usleep_range(100, 150);
> > -	rcsi2_write(priv, SRST_REG, 0);
> > +	pm_runtime_put(priv->dev);
> >  }
> >
> >  static int rcsi2_wait_phy_start(struct rcar_csi2 *priv)
> > @@ -462,7 +472,7 @@ static int rcsi2_calc_mbps(struct rcar_csi2 *priv, unsigned int bpp)
> >  	return mbps;
> >  }
> >
> > -static int rcsi2_start(struct rcar_csi2 *priv)
> > +static int rcsi2_start_receiver(struct rcar_csi2 *priv)
> >  {
> >  	const struct rcar_csi2_format *format;
> >  	u32 phycnt, vcdt = 0, vcdt2 = 0;
> > @@ -506,7 +516,6 @@ static int rcsi2_start(struct rcar_csi2 *priv)
> >
> >  	/* Init */
> >  	rcsi2_write(priv, TREF_REG, TREF_TREF);
> > -	rcsi2_reset(priv);
> >  	rcsi2_write(priv, PHTC_REG, 0);
> >
> >  	/* Configure */
> > @@ -564,19 +573,36 @@ static int rcsi2_start(struct rcar_csi2 *priv)
> >  	return 0;
> >  }
> >
> > +static int rcsi2_start(struct rcar_csi2 *priv)
> > +{
> > +	int ret;
> > +
> > +	rcsi2_standby_mode(priv, 0);
> > +
> > +	ret = rcsi2_start_receiver(priv);
> > +	if (ret) {
> > +		rcsi2_standby_mode(priv, 1);
> > +		return ret;
> > +	}
> > +
> > +	ret = v4l2_subdev_call(priv->remote, video, s_stream, 1);
> 
> minor thing as well, but I feel this one was better where it was, so
> that "rcsi2_start()" only handles the hardware, while s_stream handles
> the pipeline. But then _start() and _stop() becomes very short... so
> yeah, feel free to keep it the way it is.

Do not agree, I like this :-)

> 
> > +	if (ret) {
> > +		rcsi2_standby_mode(priv, 1);
> > +		return ret;
> > +	}
> > +
> > +	return 0;
> > +}
> > +
> >  static void rcsi2_stop(struct rcar_csi2 *priv)
> >  {
> > -	rcsi2_write(priv, PHYCNT_REG, 0);
> > -
> > -	rcsi2_reset(priv);
> > -
> > -	rcsi2_write(priv, PHTC_REG, PHTC_TESTCLR);
> > +	v4l2_subdev_call(priv->remote, video, s_stream, 0);
> > +	rcsi2_standby_mode(priv, 1);
> >  }
> >
> >  static int rcsi2_s_stream(struct v4l2_subdev *sd, int enable)
> >  {
> >  	struct rcar_csi2 *priv = sd_to_csi2(sd);
> > -	struct v4l2_subdev *nextsd;
> >  	int ret = 0;
> >
> >  	mutex_lock(&priv->lock);
> > @@ -586,27 +612,12 @@ static int rcsi2_s_stream(struct v4l2_subdev *sd, int enable)
> >  		goto out;
> >  	}
> >
> > -	nextsd = priv->remote;
> > -
> >  	if (enable && priv->stream_count == 0) {
> > -		pm_runtime_get_sync(priv->dev);
> > -
> >  		ret = rcsi2_start(priv);
> > -		if (ret) {
> > -			pm_runtime_put(priv->dev);
> > +		if (ret)
> >  			goto out;
> > -		}
> > -
> > -		ret = v4l2_subdev_call(nextsd, video, s_stream, 1);
> > -		if (ret) {
> > -			rcsi2_stop(priv);
> > -			pm_runtime_put(priv->dev);
> > -			goto out;
> > -		}
> >  	} else if (!enable && priv->stream_count == 1) {
> >  		rcsi2_stop(priv);
> > -		v4l2_subdev_call(nextsd, video, s_stream, 0);
> > -		pm_runtime_put(priv->dev);
> >  	}
> >
> >  	priv->stream_count += enable ? 1 : -1;
> > @@ -936,6 +947,10 @@ static int rcsi2_probe_resources(struct rcar_csi2 *priv,
> >  	if (irq < 0)
> >  		return irq;
> >
> > +	priv->rstc = devm_reset_control_get(&pdev->dev, NULL);
> > +	if (IS_ERR(priv->rstc))
> > +		return PTR_ERR(priv->rstc);
> > +
> 
> I don't see 'resets' listed as a mandatory property of the rcar-csi2
> bindings, shouldn't you fallback to software reset if not 'reset'
> is specified? True that all mainline users have a reset property specified,
> so you could also add 'resets' among the mandatory properties, could
> that break out of tree implementations in your opinion?

Nice catch! I will add a patch to this series listing it as mandatory.  
It's a good thing the resets property always have been part of the dts 
sources so it will not create any regressions.

> 
> Thanks
>    j
> 
> >  	return 0;
> >  }
> >
> > --
> > 2.20.1
> >



-- 
Regards,
Niklas Söderlund