Message-ID: <560E1DF6.9010901@topic.nl>
Date:	Fri, 2 Oct 2015 08:02:30 +0200
From:	Mike Looijmans <mike.looijmans@...ic.nl>
To:	Stephen Boyd <sboyd@...eaurora.org>
CC:	<mturquette@...libre.com>, <linux-clk@...r.kernel.org>,
	<devicetree@...r.kernel.org>, <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH v2] Add driver for the si514 clock generator chip

On 02-10-15 01:34, Stephen Boyd wrote:
> On 09/17, Mike Looijmans wrote:
>> This patch adds the driver and devicetree documentation for the
>> Silicon Labs SI514 clock generator chip. This is an I2C controlled
>> oscilator capable of generating clock signals ranging from 100kHz
>
> s/oscilator/oscillator/
>
>> to 250MHz.
>>
>> Signed-off-by: Mike Looijmans <mike.looijmans@...ic.nl>
>> ---
>> diff --git a/Documentation/devicetree/bindings/clock/silabs,si514.txt b/Documentation/devicetree/bindings/clock/silabs,si514.txt
>> new file mode 100644
>> index 0000000..05964d1
>> --- /dev/null
>> +++ b/Documentation/devicetree/bindings/clock/silabs,si514.txt
>> @@ -0,0 +1,27 @@
>> +Binding for Silicon Labs 514 programmable I2C clock generator.
>> +
>> +Reference
>> +This binding uses the common clock binding[1]. Details about the device can be
>> +found in the datasheet[2].
>> +
>> +[1] Documentation/devicetree/bindings/clock/clock-bindings.txt
>> +[2] Si514 datasheet
>> +    http://www.silabs.com/Support%20Documents/TechnicalDocs/si514.pdf
>> +
>> +Required properties:
>> + - compatible: Shall be "silabs,si514"
>> + - reg: I2C device address.
>> + - #clock-cells: From common clock bindings: Shall be 0.
>> +
>> +Optional properties:
>> + - clock-output-names: From common clock bindings. Recommended to be "si514".
>> + - clock-frequency: Output frequency to generate. This defines the output
>> +		    frequency set during boot. It can be reprogrammed during
>> +		    runtime through the common clock framework.
>
> Can we use assigned clock rates instead of this property?

I'll first need to learn what 'assigned clock rates' means. But I suspect the 
answer will be yes.

>> +
>> +Example:
>> +	si514: clock-generator@55 {
>> +		reg = <0x55>;
>> +		#clock-cells = <0>;
>> +		compatible = "silabs,si514";
>> +	};
>> diff --git a/drivers/clk/Kconfig b/drivers/clk/Kconfig
>> index 42f7120..6ac7deb5 100644
>> --- a/drivers/clk/Kconfig
>> +++ b/drivers/clk/Kconfig
>> @@ -68,6 +68,16 @@ config COMMON_CLK_SI5351
>>   	  This driver supports Silicon Labs 5351A/B/C programmable clock
>>   	  generators.
>>
>> +config COMMON_CLK_SI514
>> +	tristate "Clock driver for SiLabs 514 devices"
>> +	depends on I2C
>> +	depends on OF
>
> It actually depends on this to build?

It calls various 'of_' methods unconditionally, so I'd think so.

>> +	select REGMAP_I2C
>> +	help
>> +	---help---
>> +	  This driver supports the Silicon Labs 514 programmable clock
>> +	  generator.
>> +
>>   config COMMON_CLK_SI570
>>   	tristate "Clock driver for SiLabs 570 and compatible devices"
>>   	depends on I2C
>> diff --git a/drivers/clk/clk-si514.c b/drivers/clk/clk-si514.c
>> new file mode 100644
>> index 0000000..ca70818
>> --- /dev/null
>> +++ b/drivers/clk/clk-si514.c
>> @@ -0,0 +1,393 @@
>> +
>> +#include <linux/clk-provider.h>
>
> I'd expect some sort of linux/clk.h include here if we're using
> clk APIs.

It probably got dragged in by the clk-provider.h include, including it 
specifically would be more appropriate.

>
>> +#include <linux/delay.h>
>> +#include <linux/module.h>
>> +#include <linux/i2c.h>
>> +#include <linux/regmap.h>
>> +#include <linux/slab.h>
> [...]
>> +
>> +/* Program clock settings into hardware */
>
> This comment is useless.
>
>> +static int si514_set_muldiv(struct clk_si514 *data,
>> +	struct clk_si514_muldiv *settings)
>> +{
>> +	u8 lp;
>> +	u8 reg[7];
>> +	int err;
>> +
>> +	/* Calculate LP1/LP2 according to table 13 in the datasheet */
>> +	/* 65.259980246 */
>> +	if ((settings->m_int < 65) ||
>> +		((settings->m_int == 65) && (settings->m_frac <= 139575831)))
>> +		lp = 0x22;
>> +	/* 67.859763463 */
>> +	else if ((settings->m_int < 67) ||
>> +		((settings->m_int == 67) && (settings->m_frac <= 461581994)))
>> +		lp = 0x23;
>> +	/* 72.937624981 */
>> +	else if ((settings->m_int < 72) ||
>> +		((settings->m_int == 72) && (settings->m_frac <= 503383578)))
>> +		lp = 0x33;
>> +	/* 75.843265046 */
>> +	else if ((settings->m_int < 75) ||
>> +		((settings->m_int == 75) && (settings->m_frac <= 452724474)))
>> +		lp = 0x34;
>
> Drop the extra parenthesis on these if statements.
>
>> +	else
>> +		lp = 0x44;
>> +
>> +	err = regmap_write(data->regmap, SI514_REG_LP, lp);
>> +	if (err < 0)
>> +		return err;
>> +
>> +	reg[0] = settings->m_frac & 0xff;
>> +	reg[1] = (settings->m_frac >> 8) & 0xff;
>> +	reg[2] = (settings->m_frac >> 16) & 0xff;
>> +	reg[3] = ((settings->m_frac >> 24) | (settings->m_int << 5)) & 0xff;
>> +	reg[4] = (settings->m_int >> 3) & 0xff;
>> +	reg[5] = (settings->hs_div) & 0xff;
>> +	reg[6] = (((settings->hs_div) >> 8) |
>> +			(settings->ls_div_bits << 4)) & 0xff;
>
> The & 0xff part is redundant. Assignment truncates for us.
>
>> +
>> +	err = regmap_bulk_write(data->regmap, SI514_REG_HS_DIV, reg + 5, 2);
>> +	if (err < 0)
>> +		return err;
>> +	/* Writing to SI514_REG_M_INT_FRAC triggers the clock change, so that
>> +	 * must be written last */
>
> Please fix multi-line commenting style.
>
>> +	return regmap_bulk_write(data->regmap, SI514_REG_M_FRAC1, reg, 5);
>> +}
>> +
>> +/* Calculate divider settings for a given frequency */
>> +static int si514_calc_muldiv(struct clk_si514_muldiv *settings,
>> +	unsigned long frequency)
>> +{
>> +	u64 m;
>> +	u32 ls_freq;
>> +	u32 tmp;
>> +	u8 res;
>> +
>> +	if ((frequency < SI514_MIN_FREQ) || (frequency > SI514_MAX_FREQ))
>> +		return -EINVAL;
>> +
>> +	/* Determine the minimum value of LS_DIV and resulting target freq. */
>> +	ls_freq = frequency;
>> +	if (frequency >= (FVCO_MIN / HS_DIV_MAX))
>> +		settings->ls_div_bits = 0;
>> +	else {
>> +		res = 1;
>> +		tmp = 2 * HS_DIV_MAX;
>> +		while (tmp <= (HS_DIV_MAX*32)) {
>
> Please add some space here between HS_DIV_MAX and 32.
>
>> +			if ((frequency * tmp) >= FVCO_MIN)
>> +				break; /* We're done */
>
> Yep, that's what break in a loop usually means.
>
>> +			++res;
>> +			tmp <<= 1;
>> +		}
>> +		settings->ls_div_bits = res;
>> +		ls_freq = frequency << res;
>> +	}
>> +
>> +	/* Determine minimum HS_DIV, round up to even number */
>> +	settings->hs_div = DIV_ROUND_UP(FVCO_MIN >> 1, ls_freq) << 1;
>> +
>> +	/* M = LS_DIV x HS_DIV x frequency / F_XO (in fixed-point) */
>> +	m = ((u64)(ls_freq * settings->hs_div) << 29) + (FXO/2);
>> +	do_div(m, FXO);
>> +	settings->m_frac = (u32)m & (BIT(29) - 1);
>> +	settings->m_int = (u32)(m >> 29);
>> +
>> +	return 0;
>> +}
>> +
>> +/* Calculate resulting frequency given the register settings */
>> +static unsigned long si514_calc_rate(struct clk_si514_muldiv *settings)
>> +{
>> +	u64 m = settings->m_frac | ((u64)settings->m_int << 29);
>> +
>> +	return ((u32)(((m * FXO) + (FXO/2)) >> 29)) /
>
> Please add space after /
>
>> +		(settings->hs_div * BIT(settings->ls_div_bits));
>
> And drop the extra parenthesis. It would be nice if we didn't do
> it all in one line too. Use some local variables.

I'll refactor it into something more readable.

>> +}
>> +
> [...]
>> +
>> +	err = si514_calc_muldiv(&settings, rate);
>> +	if (err)
>> +		return err;
>> +
>> +	return si514_calc_rate(&settings);
>> +}
>> +
>> +/* Update output frequency for big frequency changes (> 1000 ppm).
>> + * The chip supports <1000ppm changes "on the fly", we haven't implemented
>> + * that here. */
>
> Please fix comment style.
>
>> +static int si514_set_rate(struct clk_hw *hw, unsigned long rate,
>> +		unsigned long parent_rate)
>> +{
>> +	struct clk_si514 *data = to_clk_si514(hw);
>> +	struct clk_si514_muldiv settings;
>> +	int err;
>> +
>> +	err = si514_calc_muldiv(&settings, rate);
>> +	if (err)
>> +		return err;
>> +
>> +	si514_enable_output(data, false);
>> +
>> +	err = si514_set_muldiv(data, &settings);
>> +	if (err < 0)
>> +		return err; /* Undefined state now, best to leave disabled */
>> +
>> +	/* Trigger calibration */
>> +	err = regmap_write(data->regmap, SI514_REG_CONTROL, SI514_CONTROL_FCAL);
>> +
>> +	/* Applying a new frequency can take up to 10ms */
>> +	usleep_range(10000, 12000);
>> +
>> +	si514_enable_output(data, true);
>> +
>
> Should we enable if regmap_write() failed?

Good question, I don't recall why I made this choice.

For a client, when set_rate returns error, it would expect either the clock 
still running at the old frequency, or not running at all. From that 
perspective, since the PLL has been reprogrammed, the better choice would be 
to leave it disabled. Expected behaviour for the client is going to be to fix 
the I2C bus problems and try again.

>> +	return err;
>> +}
>> +
>> +static const struct clk_ops si514_clk_ops = {
>> +	.recalc_rate = si514_recalc_rate,
>> +	.round_rate = si514_round_rate,
>> +	.set_rate = si514_set_rate,
>> +};
>> +
>> +static struct regmap_config si514_regmap_config = {
>
> const?
>
>> +		}
>> +	}
>> +
>> +	/* Display a message indicating that we've successfully registered */
>> +	dev_info(&client->dev, "registered, current frequency %lu Hz\n",
>> +			clk_get_rate(clk));
>
> Please remove this.
>
>> +
>> +	return 0;
>> +}
>> +
>

I'll post a v2 patch with the requested changes soon. Thank you for your review.

Mike.


Kind regards,

Mike Looijmans
System Expert

TOPIC Embedded Products
Eindhovenseweg 32-C, NL-5683 KH Best
Postbus 440, NL-5680 AK Best
Telefoon: +31 (0) 499 33 69 79
Telefax: +31 (0) 499 33 69 70
E-mail: mike.looijmans@...icproducts.com
Website: www.topicproducts.com

Please consider the environment before printing this e-mail





--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists