RE: [PATCH] phy: xilinx: phy-zynqmp: dynamic clock support for power-save

["Mehta, Piyush" <piyush.mehta@amd.com>]

> -----Original Message-----
> From: Vinod Koul <vkoul@kernel.org>
> Sent: Tuesday, September 20, 2022 5:20 PM
> To: Mehta, Piyush <piyush.mehta@amd.com>
> Cc: laurent.pinchart@ideasonboard.com; kishon@ti.com;
> michal.simek@xilinx.com; Simek, Michal <michal.simek@amd.com>; linux-
> kernel@vger.kernel.org; linux-phy@lists.infradead.org; linux-arm-
> kernel@lists.infradead.org; git (AMD-Xilinx) <git@amd.com>; Paladugu, Siva
> Durga Prasad <siva.durga.prasad.paladugu@amd.com>
> Subject: Re: [PATCH] phy: xilinx: phy-zynqmp: dynamic clock support for
> power-save
> 
> On 20-09-22, 10:38, Mehta, Piyush wrote:
> > Hello @Vinod Koul,
> >
> > Thanks for review comments.
> > Please find my inline comments below with tag [Piyush].
> 
> Please _do_ _not_ _top_ _post_
> 
> And learn how to reply upstream. Am sure there at people at AMD/Xilinx
> who know this..
> 
> 
> >
> > Regards,
> > Piyush Mehta
> >
> > > -----Original Message-----
> > > From: Vinod Koul <vkoul@kernel.org>
> > > Sent: Tuesday, September 20, 2022 12:11 PM
> > > To: Mehta, Piyush <piyush.mehta@amd.com>
> > > Cc: laurent.pinchart@ideasonboard.com; kishon@ti.com;
> > > michal.simek@xilinx.com; Simek, Michal <michal.simek@amd.com>;
> > > linux- kernel@vger.kernel.org; linux-phy@lists.infradead.org;
> > > linux-arm- kernel@lists.infradead.org; git (AMD-Xilinx)
> > > <git@amd.com>; Paladugu, Siva Durga Prasad
> > > <siva.durga.prasad.paladugu@amd.com>
> > > Subject: Re: [PATCH] phy: xilinx: phy-zynqmp: dynamic clock support
> > > for power-save
> > >
> > > On 09-09-22, 18:34, Piyush Mehta wrote:
> > > > Enabling clock for all the lanes, consuming power even PHY is
> > > > active or not.
> > >
> > > Pls consider revision to:
> > > Enabling clock for all the lanes consumes power even PHY is active or
> idle..
> > [Piyush]:  will address in next version of patch.
> >
> > >
> > > >
> > > > To resolve the above issue and power saving,made clock
> > > > enabled/disabled based on active PHYs on call of phy_init/phy_exit.
> > >
> > > To resolve this, enable/disable clocks in phy_init/phy_exit
> > [Piyush]:  will address in next version of patch.
> >
> > >
> > > > By default clock is disabled for all the lanes. Whenever phy_init
> > > > called from USB, SATA, SGMII, or display driver, it enabled the
> > > > required clock for requested lane. On phy_exit cycle, it disabled
> > > > clock for the active PHYs.
> > > >
> > > > During the suspend/resume cycle, each USB/ SATA/ SGMII/ display
> > > > driver called phy_exit/phy_init individually. it disabled clock on
> > > > exit, and enabled on initialization for the active PHYs.
> > > >
> > > > Active PHY configuration depends on the peripheral DT node status,
> > > > like USB DT node status 'okay' then driver enabled clock for the
> > > > USB configured lane.
> > > >
> > > > Signed-off-by: Piyush Mehta <piyush.mehta@amd.com>
> > > > ---
> > > >  drivers/phy/xilinx/phy-zynqmp.c | 59
> > > > ++++++++-------------------------
> > > >  1 file changed, 14 insertions(+), 45 deletions(-)
> > > >
> > > > diff --git a/drivers/phy/xilinx/phy-zynqmp.c
> > > > b/drivers/phy/xilinx/phy-zynqmp.c index 9be9535ad7ab..912c4defdf8d
> > > > 100644
> > > > --- a/drivers/phy/xilinx/phy-zynqmp.c
> > > > +++ b/drivers/phy/xilinx/phy-zynqmp.c
> > > > @@ -572,6 +572,10 @@ static int xpsgtr_phy_init(struct phy *phy)
> > > >
> > > >  	mutex_lock(&gtr_dev->gtr_mutex);
> > > >
> > > > +	/* Configure and enable the clock when peripheral phy_init call
> > > > +*/
> > >
> > > when peripheral invokes phy_init
> > >
> > > > +	if (clk_prepare_enable(gtr_dev->clk[gtr_phy->lane]))
> > > > +		goto out;
> > > > +
> > > >  	/* Skip initialization if not required. */
> > > >  	if (!xpsgtr_phy_init_required(gtr_phy))
> > > >  		goto out;
> > > > @@ -616,9 +620,13 @@ static int xpsgtr_phy_init(struct phy *phy)
> > > > static int xpsgtr_phy_exit(struct phy *phy)  {
> > > >  	struct xpsgtr_phy *gtr_phy = phy_get_drvdata(phy);
> > > > +	struct xpsgtr_dev *gtr_dev = gtr_phy->dev;
> > > >
> > > >  	gtr_phy->skip_phy_init = false;
> > > >
> > > > +	/* Ensure that disable clock only, which configure for lane */
> > > > +	clk_disable_unprepare(gtr_dev->clk[gtr_phy->lane]);
> > > > +
> > > >  	return 0;
> > > >  }
> > > >
> > > > @@ -824,15 +832,11 @@ static struct phy *xpsgtr_xlate(struct
> > > > device *dev,  static int __maybe_unused xpsgtr_suspend(struct device
> *dev)  {
> > > >  	struct xpsgtr_dev *gtr_dev = dev_get_drvdata(dev);
> > > > -	unsigned int i;
> > > >
> > > >  	/* Save the snapshot ICM_CFG registers. */
> > > >  	gtr_dev->saved_icm_cfg0 = xpsgtr_read(gtr_dev, ICM_CFG0);
> > > >  	gtr_dev->saved_icm_cfg1 = xpsgtr_read(gtr_dev, ICM_CFG1);
> > > >
> > > > -	for (i = 0; i < ARRAY_SIZE(gtr_dev->clk); i++)
> > > > -		clk_disable_unprepare(gtr_dev->clk[i]);
> > > > -
> > > >  	return 0;
> > > >  }
> > > >
> > > > @@ -842,13 +846,6 @@ static int __maybe_unused
> > > > xpsgtr_resume(struct
> > > device *dev)
> > > >  	unsigned int icm_cfg0, icm_cfg1;
> > > >  	unsigned int i;
> > > >  	bool skip_phy_init;
> > > > -	int err;
> > > > -
> > > > -	for (i = 0; i < ARRAY_SIZE(gtr_dev->clk); i++) {
> > > > -		err = clk_prepare_enable(gtr_dev->clk[i]);
> > > > -		if (err)
> > > > -			goto err_clk_put;
> > > > -	}
> > >
> > > hold on, why is this being removed from suspend/resume?
> >
> >
> > [Piyush]:
> > Now, each peripheral driver calls phy_exit/phy_init in suspend/resume,
> and respective active driver will do clock enable and disable.
> >
> > In above xpsgtr resume calls, by default all the clock enabled for all the
> lanes, that consumes power even PHY is active or not.
> > So, the clock will be enabled only for active PHY in the phy_init and disabled
> in phy_exit. Clock enabling is not required at multiple times.
> 
> Learn to use a proper Linux MUA while replying on mail lists
> 
> on you comment:
> 1. leaving clock on all the time does not sound right to me 2. why not enable
> pm_runtime for this driver and do better with power

Enable pm_runtime support:

We have a fixed clock for the GT lanes, as earlier explained GT specific clock is enabled on phy
initialization of the active state of peripheral (node status = "Okay"), and disabled on exit.

For pm_runtime design support:

static int __maybe_unused xpsgtr_runtime_suspend(struct device *dev) {
	return 0;
}

static int __maybe_unused xpsgtr_runtime_resume(struct device *dev) {
	return 0;
}

static int __maybe_unused xpsgtr_runtime_idle(struct device *dev) {
	pm_runtime_mark_last_busy(dev);
	pm_runtime_autosuspend(dev);
	return 0;
}

static const struct dev_pm_ops xpsgtr_pm_ops = {
	SET_RUNTIME_PM_OPS(xpsgtr_runtime_suspend,
			   xpsgtr_runtime_resume, xpsgtr_runtime_idle)
 };

We don't enable/disable the clock on suspend/resume calls, so is it acceptable to have empty PM ops (xpsgtr_runtime_suspend / xpsgtr_runtime_resume)?

 3. Changes need to be
> documented in changelog, this was not explained in log so unexpected

I will rephrase the commit description to have a detailed description.

> 
> >
> > >
> > > >
> > > >  	icm_cfg0 = xpsgtr_read(gtr_dev, ICM_CFG0);
> > > >  	icm_cfg1 = xpsgtr_read(gtr_dev, ICM_CFG1); @@ -869,12 +866,6
> > > @@
> > > > static int __maybe_unused xpsgtr_resume(struct device *dev)
> > > >  		gtr_dev->phys[i].skip_phy_init = skip_phy_init;
> > > >
> > > >  	return 0;
> > > > -
> > > > -err_clk_put:
> > > > -	while (i--)
> > > > -		clk_disable_unprepare(gtr_dev->clk[i]);
> > > > -
> > > > -	return err;
> > > >  }
> > > >
> > > >  static const struct dev_pm_ops xpsgtr_pm_ops = { @@ -888,7 +879,6
> > > > @@ static const struct dev_pm_ops xpsgtr_pm_ops = {  static int
> > > > xpsgtr_get_ref_clocks(struct xpsgtr_dev *gtr_dev)  {
> > > >  	unsigned int refclk;
> > > > -	int ret;
> > > >
> > > >  	for (refclk = 0; refclk < ARRAY_SIZE(gtr_dev->refclk_sscs); ++refclk) {
> > > >  		unsigned long rate;
> > > > @@ -899,19 +889,14 @@ static int xpsgtr_get_ref_clocks(struct
> > > > xpsgtr_dev
> > > *gtr_dev)
> > > >  		snprintf(name, sizeof(name), "ref%u", refclk);
> > > >  		clk = devm_clk_get_optional(gtr_dev->dev, name);
> > > >  		if (IS_ERR(clk)) {
> > > > -			ret = dev_err_probe(gtr_dev->dev, PTR_ERR(clk),
> > > > -					    "Failed to get reference clock
> > > %u\n",
> > > > -					    refclk);
> > > > -			goto err_clk_put;
> > > > +			return dev_err_probe(gtr_dev->dev, PTR_ERR(clk),
> > > > +					     "Failed to get ref clock %u\n",
> > > > +					     refclk);
> > > >  		}
> > > >
> > > >  		if (!clk)
> > > >  			continue;
> > > >
> > > > -		ret = clk_prepare_enable(clk);
> > > > -		if (ret)
> > > > -			goto err_clk_put;
> > > > -
> > > >  		gtr_dev->clk[refclk] = clk;
> > > >
> > > >  		/*
> > > > @@ -931,18 +916,11 @@ static int xpsgtr_get_ref_clocks(struct
> > > > xpsgtr_dev
> > > *gtr_dev)
> > > >  			dev_err(gtr_dev->dev,
> > > >  				"Invalid rate %lu for reference clock %u\n",
> > > >  				rate, refclk);
> > > > -			ret = -EINVAL;
> > > > -			goto err_clk_put;
> > > > +			return -EINVAL;
> > > >  		}
> > > >  	}
> > > >
> > > >  	return 0;
> > > > -
> > > > -err_clk_put:
> > > > -	while (refclk--)
> > > > -		clk_disable_unprepare(gtr_dev->clk[refclk]);
> > > > -
> > > > -	return ret;
> > > >  }
> > > >
> > > >  static int xpsgtr_probe(struct platform_device *pdev) @@ -951,7
> > > > +929,6 @@ static int xpsgtr_probe(struct platform_device *pdev)
> > > >  	struct xpsgtr_dev *gtr_dev;
> > > >  	struct phy_provider *provider;
> > > >  	unsigned int port;
> > > > -	unsigned int i;
> > > >  	int ret;
> > > >
> > > >  	gtr_dev = devm_kzalloc(&pdev->dev, sizeof(*gtr_dev),
> > > > GFP_KERNEL);
> > > @@
> > > > -991,8 +968,7 @@ static int xpsgtr_probe(struct platform_device
> *pdev)
> > > >  		phy = devm_phy_create(&pdev->dev, np, &xpsgtr_phyops);
> > > >  		if (IS_ERR(phy)) {
> > > >  			dev_err(&pdev->dev, "failed to create PHY\n");
> > > > -			ret = PTR_ERR(phy);
> > > > -			goto err_clk_put;
> > > > +			return PTR_ERR(phy);
> > > >  		}
> > > >
> > > >  		gtr_phy->phy = phy;
> > > > @@ -1003,16 +979,9 @@ static int xpsgtr_probe(struct
> > > > platform_device
> > > *pdev)
> > > >  	provider = devm_of_phy_provider_register(&pdev->dev,
> > > xpsgtr_xlate);
> > > >  	if (IS_ERR(provider)) {
> > > >  		dev_err(&pdev->dev, "registering provider failed\n");
> > > > -		ret = PTR_ERR(provider);
> > > > -		goto err_clk_put;
> > > > +		return PTR_ERR(provider);
> > > >  	}
> > > >  	return 0;
> > > > -
> > > > -err_clk_put:
> > > > -	for (i = 0; i < ARRAY_SIZE(gtr_dev->clk); i++)
> > > > -		clk_disable_unprepare(gtr_dev->clk[i]);
> > > > -
> > > > -	return ret;
> > > >  }
> > > >
> > > >  static const struct of_device_id xpsgtr_of_match[] = {
> > > > --
> > > > 2.25.1
> > >
> > > --
> > > ~Vinod
> 
> --
> ~Vinod

Regards,
Piyush Mehta