Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Sat, 14 Nov 2020 15:52:28 +0000
From:   Jonathan Cameron <jic23@kernel.org>
To:     Fabrice Gasnier <fabrice.gasnier@st.com>
Cc:     <linux-arm-kernel@lists.infradead.org>,
        <linux-kernel@vger.kernel.org>, <alexandre.torgue@st.com>,
        <olivier.moysan@st.com>, <linux-iio@vger.kernel.org>,
        <linux-stm32@st-md-mailman.stormreply.com>
Subject: Re: [PATCH] iio: adc: stm32-adc: adapt clock duty cycle for proper
 operation
Message-ID: <20201114155228.5b78b7d6@archlinux>
In-Reply-To: <0d7c4b81-3f4e-e952-892f-35296c87f987@st.com>
References: <1604681846-31234-1-git-send-email-fabrice.gasnier@st.com>
        <20201108151835.5d78ebca@archlinux>
        <0d7c4b81-3f4e-e952-892f-35296c87f987@st.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Precedence: bulk

On Fri, 13 Nov 2020 09:27:03 +0100
Fabrice Gasnier <fabrice.gasnier@st.com> wrote:

> On 11/8/20 4:18 PM, Jonathan Cameron wrote:
> > On Fri, 6 Nov 2020 17:57:26 +0100
> > Fabrice Gasnier <fabrice.gasnier@st.com> wrote:
> >   
> >> For proper operation, STM32 ADC should be used with a clock duty cycle
> >> of 50%, in the range of 49% to 51%. Depending on the clock tree, divider
> >> can be used in case clock duty cycle is out of this range.
> >> In case clk_get_scaled_duty_cycle() returns an error, kindly apply a
> >> divider by default (don't make the probe fail).
> >>
> >> Signed-off-by: Fabrice Gasnier <fabrice.gasnier@st.com>  
> > Hi Fabrice,
> > 
> > This sounds like it's a fix for a situation in which the device is not
> > currently working?  If so, please let me know a fixes tag.  
> 
> Hi Jonathan,
> 
> That's a good point. I also thought about adding a fixes tag. Currently
> I think this can't be hit upstream, as clock tree is tuned to fit duty
> cycle constraints. So far, nobody seems to complain about it. So this
> can probably go through the normal tree.
> 
Applied.  Will be interesting to see if the bot finds this one as a possible
candidate for backports.  I'll keep an eye on those coming through and suggest
this isn't backported if it does show up.

Thanks,

Jonathan
> Thanks,
> Fabrice
> 
> > 
> > Thanks,
> > 
> > Jonathan
> >   
> >> ---
> >>  drivers/iio/adc/stm32-adc-core.c | 21 ++++++++++++++++++++-
> >>  1 file changed, 20 insertions(+), 1 deletion(-)
> >>
> >> diff --git a/drivers/iio/adc/stm32-adc-core.c b/drivers/iio/adc/stm32-adc-core.c
> >> index cd870c0..d64a9e8 100644
> >> --- a/drivers/iio/adc/stm32-adc-core.c
> >> +++ b/drivers/iio/adc/stm32-adc-core.c
> >> @@ -202,7 +202,7 @@ static int stm32h7_adc_clk_sel(struct platform_device *pdev,
> >>  {
> >>  	u32 ckmode, presc, val;
> >>  	unsigned long rate;
> >> -	int i, div;
> >> +	int i, div, duty;
> >>  
> >>  	/* stm32h7 bus clock is common for all ADC instances (mandatory) */
> >>  	if (!priv->bclk) {
> >> @@ -226,6 +226,11 @@ static int stm32h7_adc_clk_sel(struct platform_device *pdev,
> >>  			return -EINVAL;
> >>  		}
> >>  
> >> +		/* If duty is an error, kindly use at least /2 divider */
> >> +		duty = clk_get_scaled_duty_cycle(priv->aclk, 100);
> >> +		if (duty < 0)
> >> +			dev_warn(&pdev->dev, "adc clock duty: %d\n", duty);
> >> +
> >>  		for (i = 0; i < ARRAY_SIZE(stm32h7_adc_ckmodes_spec); i++) {
> >>  			ckmode = stm32h7_adc_ckmodes_spec[i].ckmode;
> >>  			presc = stm32h7_adc_ckmodes_spec[i].presc;
> >> @@ -234,6 +239,13 @@ static int stm32h7_adc_clk_sel(struct platform_device *pdev,
> >>  			if (ckmode)
> >>  				continue;
> >>  
> >> +			/*
> >> +			 * For proper operation, clock duty cycle range is 49%
> >> +			 * to 51%. Apply at least /2 prescaler otherwise.
> >> +			 */
> >> +			if (div == 1 && (duty < 49 || duty > 51))
> >> +				continue;
> >> +
> >>  			if ((rate / div) <= priv->max_clk_rate)
> >>  				goto out;
> >>  		}
> >> @@ -246,6 +258,10 @@ static int stm32h7_adc_clk_sel(struct platform_device *pdev,
> >>  		return -EINVAL;
> >>  	}
> >>  
> >> +	duty = clk_get_scaled_duty_cycle(priv->bclk, 100);
> >> +	if (duty < 0)
> >> +		dev_warn(&pdev->dev, "bus clock duty: %d\n", duty);
> >> +
> >>  	for (i = 0; i < ARRAY_SIZE(stm32h7_adc_ckmodes_spec); i++) {
> >>  		ckmode = stm32h7_adc_ckmodes_spec[i].ckmode;
> >>  		presc = stm32h7_adc_ckmodes_spec[i].presc;
> >> @@ -254,6 +270,9 @@ static int stm32h7_adc_clk_sel(struct platform_device *pdev,
> >>  		if (!ckmode)
> >>  			continue;
> >>  
> >> +		if (div == 1 && (duty < 49 || duty > 51))
> >> +			continue;
> >> +
> >>  		if ((rate / div) <= priv->max_clk_rate)
> >>  			goto out;
> >>  	}  
> >