Re: [PATCH 1/8] iio: adc: stm32-dfsdm: make spi_master_freq more accurate

From: Fabrice Gasnier <fabrice.gasnier@st.com>
To: Jonathan Cameron <jic23@kernel.org>
Cc: <linux-arm-kernel@lists.infradead.org>,
	<linux-kernel@vger.kernel.org>, <mcoquelin.stm32@gmail.com>,
	<alexandre.torgue@st.com>, <linux-iio@vger.kernel.org>,
	<lars@metafoo.de>, <knaack.h@gmx.de>, <pmeerw@pmeerw.net>,
	<linux-stm32@st-md-mailman.stormreply.com>,
	<arnaud.pouliquen@st.com>, <olivier.moysan@st.com>
Subject: Re: [PATCH 1/8] iio: adc: stm32-dfsdm: make spi_master_freq more accurate
Date: Thu, 21 Mar 2019 14:41:13 +0100	[thread overview]
Message-ID: <62742e3f-3860-3020-33a0-dafc9ae66dd1@st.com> (raw)
In-Reply-To: <20190310100932.71fd1af1@archlinux>

On 3/10/19 11:09 AM, Jonathan Cameron wrote:
> On Wed, 6 Mar 2019 09:55:17 +0100
> Fabrice Gasnier <fabrice.gasnier@st.com> wrote:
> 
>> When SPI clock isn't accurate, 'spi_master_freq' is filled in with
>> expected frequency. Use computed value instead:
>> - e.g. source clock / (CKOUTDIV + 1)
>> Also, current divider may be set to value that makes CKOUT to exceed
>> spi-max-frequency. Rather use lower value (e.g. round up divider when
>> ckout isn't accurate).
>>
>> Signed-off-by: Fabrice Gasnier <fabrice.gasnier@st.com>
>> ---
>>  drivers/iio/adc/stm32-dfsdm-core.c | 9 +++++++--
>>  1 file changed, 7 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/iio/adc/stm32-dfsdm-core.c b/drivers/iio/adc/stm32-dfsdm-core.c
>> index bf089f5..65b7556 100644
>> --- a/drivers/iio/adc/stm32-dfsdm-core.c
>> +++ b/drivers/iio/adc/stm32-dfsdm-core.c
>> @@ -243,13 +243,18 @@ static int stm32_dfsdm_parse_of(struct platform_device *pdev,
>>  		return 0;
>>  	}
>>  
>> -	priv->spi_clk_out_div = div_u64_rem(clk_freq, spi_freq, &rem) - 1;
>> +	priv->spi_clk_out_div = div_u64_rem(clk_freq, spi_freq, &rem);
>> +
>> +	/* round up divider when clkout isn't accurate (e.g. !rem) */
>> +	if (priv->spi_clk_out_div && !rem)
>> +		priv->spi_clk_out_div--;
>> +
> This comment perhaps needs adjusting because at the moment it looks
> like it decrements when it is accurate.   With the old code in place
> in the patch it's obvious that's because you actually want one less.
> 
> Might even be worth the dance of
> 
> /* round up if not precise */
> if (priv->spi_clk_out_div && rem)
> 	priv->spi_clk_out_div++;
> 
> /* subtract one because....  */
> priv->spi_clk_out_div--;

Hi Jonathan,

I'll rework this patch in v2, make it easier to understand & read.

In a few words: the clock output is divider = ckoutdiv + 1.
ckoutdiv range can be from 1-255 to provide divider of 2-256.
So I'll introduce this divider as a variable, and use it to setup
spi_clk_out_div.
This should address all your remarks here.

Thanks for reviewing,
Best Regards,
Fabrice
> 
>>  	if (!priv->spi_clk_out_div) {
>>  		/* spi_clk_out_div == 0 means ckout is OFF */
>>  		dev_err(&pdev->dev, "spi-max-frequency not achievable\n");
>>  		return -EINVAL;
>>  	}
>> -	priv->dfsdm.spi_master_freq = spi_freq;
>> +	priv->dfsdm.spi_master_freq = clk_freq / (priv->spi_clk_out_div + 1);
> And we increment it again here? That needs an explanation as well.
>>  
>>  	if (rem) {
>>  		dev_warn(&pdev->dev, "SPI clock not accurate\n");
> 