Re: [PATCH V2] clk: qcom: Add spmi_pmic clock divider support

[Stephen Boyd <sboyd@codeaurora.org>]

On 08/31, Tirupathi Reddy wrote:
> diff --git a/Documentation/devicetree/bindings/clock/clk-spmi-pmic-div.txt b/Documentation/devicetree/bindings/clock/clk-spmi-pmic-div.txt
> new file mode 100644
> index 0000000..48cb2f7
> --- /dev/null
> +++ b/Documentation/devicetree/bindings/clock/clk-spmi-pmic-div.txt
> @@ -0,0 +1,49 @@
> +Qualcomm Technologies, Inc. SPMI PMIC clock divider (clkdiv)
> +
> +clkdiv configures the clock frequency of a set of outputs on the PMIC.
> +These clocks are typically wired through alternate functions on
> +gpio pins.
> +
> +=======================
> +Supported Properties

Just "Properties".

> +=======================
> +
> +- compatible
> +	Usage:      required
> +	Value type: <string>
> +	Definition: should be "qcom,spmi-pmic-clkdiv".
> +
> +- reg
> +	Usage:      required
> +	Value type: <prop-encoded-array>
> +	Definition: Addresses and sizes for the memory of this CLKDIV
> +		    peripheral.
> +
> +- clocks:
> +	Usage: required
> +	Value type: <prop-encoded-array>
> +	Definition: reference to the xo clock.
> +
> +- clock-names:
> +	Usage: required
> +	Value type: <stringlist>
> +	Definition: must be "xo".
> +
> +=======
> +Example
> +=======
> +
> +pm8998_clk_divs: qcom,clkdiv@0 {

@5b00

> +	compatible = "qcom,spmi-pmic-clkdiv";
> +	reg = <0x5b00 0x300>;

Drop the size

> +	#clock-cells = <1>;
> +	clocks = <&xo_board>;
> +	clock-names = "xo";
> +
> +	assigned-clocks = <&pm8998_clk_divs CLKDIV1>,
> +			  <&pm8998_clk_divs CLKDIV2>,
> +			  <&pm8998_clk_divs CLKDIV3>;
> +	assigned-clock-rates = <9600000>,
> +			       <9600000>,
> +			       <9600000>;

Not sure we need this part in the example, but OK.

> +};
> diff --git a/drivers/clk/qcom/Kconfig b/drivers/clk/qcom/Kconfig
> index 9f6c278..af5e489 100644
> --- a/drivers/clk/qcom/Kconfig
> +++ b/drivers/clk/qcom/Kconfig
> @@ -196,3 +196,12 @@ config MSM_MMCC_8996
>  	  Support for the multimedia clock controller on msm8996 devices.
>  	  Say Y if you want to support multimedia devices such as display,
>  	  graphics, video encode/decode, camera, etc.
> +
> +config CLOCK_SPMI_PMIC_DIV

config SPMI_PMIC_CLKDIV?

> +	tristate "spmi pmic clkdiv driver"
> +	depends on (COMMON_CLK_QCOM && SPMI) || COMPILE_TEST
> +	help
> +	  This driver supports the clkdiv functionality on the Qualcomm
> +	  Technologies, Inc. SPMI PMIC. It configures the frequency of
> +	  clkdiv outputs on the PMIC. These clocks are typically wired
> +	  through alternate functions on gpio pins.
> diff --git a/drivers/clk/qcom/Makefile b/drivers/clk/qcom/Makefile
> index 3f3aff2..ee8c91c 100644
> --- a/drivers/clk/qcom/Makefile
> +++ b/drivers/clk/qcom/Makefile
> @@ -15,6 +15,7 @@ clk-qcom-$(CONFIG_QCOM_GDSC) += gdsc.o
>  # Keep alphabetically sorted by config
>  obj-$(CONFIG_APQ_GCC_8084) += gcc-apq8084.o
>  obj-$(CONFIG_APQ_MMCC_8084) += mmcc-apq8084.o
> +obj-$(CONFIG_CLOCK_SPMI_PMIC_DIV) += clk-spmi-pmic-div.o
>  obj-$(CONFIG_IPQ_GCC_4019) += gcc-ipq4019.o
>  obj-$(CONFIG_IPQ_GCC_806X) += gcc-ipq806x.o
>  obj-$(CONFIG_IPQ_GCC_8074) += gcc-ipq8074.o
> diff --git a/drivers/clk/qcom/clk-spmi-pmic-div.c b/drivers/clk/qcom/clk-spmi-pmic-div.c
> new file mode 100644
> index 0000000..24461fb
> --- /dev/null
> +++ b/drivers/clk/qcom/clk-spmi-pmic-div.c
> @@ -0,0 +1,327 @@
> +/* Copyright (c) 2017, The Linux Foundation. All rights reserved.
> + *
> + * This program is free software; you can redistribute it and/or modify
> + * it under the terms of the GNU General Public License version 2 and
> + * only version 2 as published by the Free Software Foundation.
> + *
> + * This program is distributed in the hope that it will be useful,
> + * but WITHOUT ANY WARRANTY; without even the implied warranty of
> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
> + * GNU General Public License for more details.
> + */
> +
> +#include <linux/bitops.h>
> +#include <linux/clk.h>
> +#include <linux/clk-provider.h>
> +#include <linux/delay.h>
> +#include <linux/err.h>
> +#include <linux/log2.h>
> +#include <linux/module.h>
> +#include <linux/of.h>
> +#include <linux/of_device.h>

Needed?

> +#include <linux/platform_device.h>
> +#include <linux/regmap.h>
> +#include <linux/slab.h>
> +#include <linux/types.h>
> +
> +#include <dt-bindings/clock/spmi_pmic_clk_div.h>
> +
> +#define REG_DIV_CTL1			0x43
> +#define DIV_CTL1_DIV_FACTOR_MASK	GENMASK(2, 0)
> +
> +#define REG_EN_CTL			0x46
> +#define REG_EN_MASK			BIT(7)
> +#define REG_EN_SET			BIT(7)

Why do we need this define? Just test if the mask comes out to be
non-zero instead.

> +
> +#define ENABLE_DELAY_NS(cxo_ns, div)	((2 + 3 * div) * cxo_ns)
> +#define DISABLE_DELAY_NS(cxo_ns, div)	((3 * div) * cxo_ns)
> +
> +#define CLK_SPMI_PMIC_DIV_OFFSET	1
> +
> +#define CLKDIV_XO_DIV_1_0		0
> +#define CLKDIV_XO_DIV_1			1
> +#define CLKDIV_XO_DIV_2			2
> +#define CLKDIV_XO_DIV_4			3
> +#define CLKDIV_XO_DIV_8			4
> +#define CLKDIV_XO_DIV_16		5
> +#define CLKDIV_XO_DIV_32		6
> +#define CLKDIV_XO_DIV_64		7
> +#define CLKDIV_MAX_ALLOWED		8
> +
> +struct clkdiv {
> +	struct regmap		*regmap;
> +	u16			base;
> +
> +	/* clock properties */
> +	struct clk_hw		hw;
> +	unsigned int		div_factor;
> +	unsigned int		cxo_period_ns;
> +};
> +
> +struct spmi_pmic_div_clk_cc {
> +	struct clk_hw	**div_clks;

I don't understand this design. Shouldn't we just allocate an
array of clkdiv structure?

> +	int		nclks;
> +};
> +
> +static inline struct clkdiv *to_clkdiv(struct clk_hw *hw)
> +{
> +	return container_of(hw, struct clkdiv, hw);
> +}
> +
> +static inline unsigned int div_factor_to_div(unsigned int div_factor)
> +{
> +	if (div_factor == CLKDIV_XO_DIV_1_0)
> +		return 1;
> +
> +	return 1 << (div_factor - CLK_SPMI_PMIC_DIV_OFFSET);
> +}
> +
> +static inline unsigned int div_to_div_factor(unsigned int div)
> +{
> +	return ilog2(div) + CLK_SPMI_PMIC_DIV_OFFSET;

Maybe we should do the clamp here?

> +}
> +
> +static bool is_spmi_pmic_clkdiv_enabled(struct clkdiv *clkdiv)
> +{
> +	unsigned int val = 0;
> +
> +	regmap_read(clkdiv->regmap, clkdiv->base + REG_EN_CTL,
> +			 &val);
> +	return ((val & REG_EN_MASK) == REG_EN_SET) ? true : false;
> +}
> +
> +static int spmi_pmic_clkdiv_set_enable_state(struct clkdiv *clkdiv,
> +			bool enable_state)
> +{
> +	int rc;
> +
> +	rc = regmap_update_bits(clkdiv->regmap, clkdiv->base + REG_EN_CTL,
> +				REG_EN_MASK,
> +				(enable_state == true) ? REG_EN_SET : 0);
> +	if (rc)
> +		return rc;
> +
> +	if (enable_state == true)
> +		ndelay(ENABLE_DELAY_NS(clkdiv->cxo_period_ns,
> +				div_factor_to_div(clkdiv->div_factor)));
> +	else
> +		ndelay(DISABLE_DELAY_NS(clkdiv->cxo_period_ns,
> +				div_factor_to_div(clkdiv->div_factor)));
> +
> +	return rc;
> +}
> +
> +static int spmi_pmic_clkdiv_config_freq_div(struct clkdiv *clkdiv,
> +			unsigned int div)
> +{
> +	unsigned int div_factor;
> +	bool enabled;
> +	int rc;
> +
> +	div_factor = div_to_div_factor(div);
> +	if (div_factor <= 0 || div_factor >= CLKDIV_MAX_ALLOWED)

div_factor is unsigned. It can't be less than 0.

> +		return -EINVAL;
> +
> +	enabled = is_spmi_pmic_clkdiv_enabled(clkdiv);
> +	if (enabled) {
> +		rc = spmi_pmic_clkdiv_set_enable_state(clkdiv, false);
> +		if (rc)
> +			return rc;

This races with clk_enable() on the same clk. You'll need some
sort of local lock to make sure set_rate() and enable don't race
with each other.

> +	}
> +
> +	rc = regmap_update_bits(clkdiv->regmap,
> +				clkdiv->base + REG_DIV_CTL1,
> +				DIV_CTL1_DIV_FACTOR_MASK, div_factor);
> +	if (rc)
> +		return rc;
> +
> +	clkdiv->div_factor = div_factor;
> +
> +	if (enabled)
> +		rc = spmi_pmic_clkdiv_set_enable_state(clkdiv, true);
> +
> +	return rc;
> +}
> +
> +static int clk_spmi_pmic_div_enable(struct clk_hw *hw)
> +{
> +	struct clkdiv *clkdiv = to_clkdiv(hw);
> +
> +	return spmi_pmic_clkdiv_set_enable_state(clkdiv, true);
> +}
> +
> +static void clk_spmi_pmic_div_disable(struct clk_hw *hw)
> +{
> +	struct clkdiv *clkdiv = to_clkdiv(hw);
> +
> +	spmi_pmic_clkdiv_set_enable_state(clkdiv, false);
> +}
> +
> +static long clk_spmi_pmic_div_round_rate(struct clk_hw *hw, unsigned long rate,
> +			unsigned long *parent_rate)
> +{
> +	unsigned long new_rate;
> +	unsigned int div, div_factor;
> +
> +	div = DIV_ROUND_UP(*parent_rate, rate);
> +	div_factor = div_to_div_factor(div);
> +	if (div_factor >= CLKDIV_MAX_ALLOWED)
> +		div_factor = CLKDIV_MAX_ALLOWED - 1;

min()?

> +	new_rate = *parent_rate / div_factor_to_div(div_factor);
> +
> +	return new_rate;
> +}
> +
> +static unsigned long clk_spmi_pmic_div_recalc_rate(struct clk_hw *hw,
> +			unsigned long parent_rate)
> +{
> +	struct clkdiv *clkdiv = to_clkdiv(hw);
> +	unsigned long rate;
> +
> +	rate = parent_rate / div_factor_to_div(clkdiv->div_factor);
> +
> +	return rate;
> +}
> +
> +static int clk_spmi_pmic_div_set_rate(struct clk_hw *hw, unsigned long rate,
> +			unsigned long parent_rate)
> +{
> +	struct clkdiv *clkdiv = to_clkdiv(hw);
> +	int rc = 0;

Please don't assign variables that are then reassigned right
after.

> +
> +	rc = spmi_pmic_clkdiv_config_freq_div(clkdiv, parent_rate / rate);
> +

And also just return things when you can instead of assigning
them to local variables to return after.

> +	return rc;
> +}
> +
> +static const struct clk_ops clk_spmi_pmic_div_ops = {
> +	.enable = clk_spmi_pmic_div_enable,
> +	.disable = clk_spmi_pmic_div_disable,
> +	.set_rate = clk_spmi_pmic_div_set_rate,
> +	.recalc_rate = clk_spmi_pmic_div_recalc_rate,
> +	.round_rate = clk_spmi_pmic_div_round_rate,
> +};
> +
> +static struct clk_hw *spmi_pmic_div_clk_hw_get(struct of_phandle_args *clkspec,
> +				      void *data)
> +{
> +	struct spmi_pmic_div_clk_cc *clk_cc = data;
> +	unsigned int idx = clkspec->args[0];
> +
> +	if (idx >= clk_cc->nclks) {
> +		pr_err("%s: invalid index %u\n", __func__, idx);
> +		return ERR_PTR(-EINVAL);
> +	}
> +
> +	return clk_cc->div_clks[idx];
> +}
> +
> +#define SPMI_PMIC_DIV_CLK_ADDRESS_RANGE	0x100

SPMI_PMIC_DIV_CLK_SIZE?

> +#define SPMI_PMIC_DIV_CLK_ID_OFFSET	1
> +
> +static int spmi_pmic_clkdiv_probe(struct platform_device *pdev)
> +{
> +	struct spmi_pmic_div_clk_cc *clk_cc;
> +	struct clk_init_data init = {0};

Drop the 0 part.

> +	struct clkdiv *clkdiv;
> +	struct clk *cxo;
> +	struct regmap *regmap;
> +	struct device *dev = &pdev->dev;
> +	const char *parent_name;
> +	int nclks, i, rc, cxo_hz;
> +	u32 reg[2], start, size;
> +
> +	rc = of_property_read_u32_array(dev->of_node, "reg", reg, 2);

Oh hm... I think I suggested we count the size, but that may not
work. The SPMI node has #address-cells = 1 and #size-cells = 0.
So the reg property will only have one cell and it will be the
offset. We could count the interrupts. Or we could count
clock-output-names. Or we can match a certain PMIC compatible
string and then know the size from some hardcoded number in this
driver.

> +	if (rc < 0) {
> +		dev_err(dev, "reg property reading failed\n");
> +		return rc;
> +	}
> +	start = reg[0];
> +	size = reg[1];
> +
> +	nclks = size / SPMI_PMIC_DIV_CLK_ADDRESS_RANGE;
> +	if (nclks == 0) {
> +		dev_err(dev, "no div clocks assigned\n");

This should never happen?

> +		return -ENODEV;
> +	}
> +
> +	regmap = dev_get_regmap(dev->parent, NULL);
> +	if (!regmap) {
> +		dev_err(dev, "Couldn't get parent's regmap\n");
> +		return -EINVAL;
> +	}
> +
> +	clkdiv = devm_kcalloc(dev, nclks, sizeof(*clkdiv), GFP_KERNEL);
> +	if (!clkdiv)
> +		return -ENOMEM;
> +
> +	clk_cc = devm_kzalloc(&pdev->dev, sizeof(*clk_cc), GFP_KERNEL);
> +	if (!clk_cc)
> +		return -ENOMEM;
> +
> +	clk_cc->div_clks = devm_kcalloc(&pdev->dev, nclks,
> +					sizeof(*clk_cc->div_clks), GFP_KERNEL);
> +	if (!clk_cc->div_clks)
> +		return -ENOMEM;
> +
> +	/* Read parent clock rate */

That's obvious.

> +	cxo = of_clk_get(dev->of_node, 0);

Just use clk_get() instead? This is a platform driver after all.

> +	if (IS_ERR(cxo)) {
> +		rc = PTR_ERR(cxo);
> +		if (rc != -EPROBE_DEFER)
> +			dev_err(dev, "failed to get xo clock");
> +		return rc;
> +	}
> +	cxo_hz = clk_get_rate(cxo);
> +	clk_put(cxo);
> +
> +	parent_name = of_clk_get_parent_name(dev->of_node, 0);
> +	if (!parent_name) {
> +		dev_err(dev, "missing parent clock\n");
> +		return -ENODEV;
> +	}
> +	dev_dbg(dev, "parent is: %s\n", parent_name);

Remove?

> +
> +	init.parent_names = &parent_name;
> +	init.num_parents = parent_name ? 1 : 0;
> +	init.ops = &clk_spmi_pmic_div_ops;
> +	init.flags = 0;
> +
> +	for (i = 0; i < nclks; i++) {
> +		clkdiv[i].base = start + i * SPMI_PMIC_DIV_CLK_ADDRESS_RANGE;
> +		clkdiv[i].regmap = regmap;
> +		clkdiv[i].cxo_period_ns = NSEC_PER_SEC / cxo_hz;
> +		init.name = kasprintf(GFP_KERNEL, "div_clk%d",
> +				     i + SPMI_PMIC_DIV_CLK_ID_OFFSET);

We don't need a define for 1.

> +		clkdiv[i].hw.init = &init;
> +		rc = devm_clk_hw_register(dev, &clkdiv[i].hw);
> +		if (rc)

We leak init.name here.

> +			return rc;
> +
> +		clk_cc->div_clks[i] = &clkdiv[i].hw;
> +		kfree(init.name); /* clock framework made a copy of the name */
> +	}
> +
> +	clk_cc->nclks = nclks;
> +	rc = of_clk_add_hw_provider(pdev->dev.of_node, spmi_pmic_div_clk_hw_get,
> +				    clk_cc);

Urgh, we really need this devmified. Patch coming tomorrow. For
now, you need a remove function for driver unbinding to
unregister the provider.

> +	return rc;
> +}
> +
> +static const struct of_device_id spmi_pmic_clkdiv_match_table[] = {
> +	{ .compatible = "qcom,spmi-pmic-clkdiv" },

I'm not sure we call it this in other places. How about
qcom,spmi-clkdiv and then have specific one for the pmic as well
in case we ever care in the future like qcom,pm8916-clkdiv, etc.

> +	{}
> +};
> +MODULE_DEVICE_TABLE(of, spmi_pmic_clkdiv_match_table);
> +
> +static struct platform_driver spmi_pmic_clkdiv_driver = {
> +	.driver		= {
> +		.name	= "qcom,spmi-pmic-clkdiv",
> +		.of_match_table = spmi_pmic_clkdiv_match_table,
> +	},
> +	.probe		= spmi_pmic_clkdiv_probe,
> +};
> +module_platform_driver(spmi_pmic_clkdiv_driver);
> +
> +MODULE_DESCRIPTION("spmi pmic clkdiv driver");

QCOM SPMI PMIC clkdiv driver?

> +MODULE_LICENSE("GPL v2");
> diff --git a/include/dt-bindings/clock/spmi_pmic_clk_div.h b/include/dt-bindings/clock/spmi_pmic_clk_div.h
> new file mode 100644
> index 0000000..25d035d
> --- /dev/null
> +++ b/include/dt-bindings/clock/spmi_pmic_clk_div.h
> @@ -0,0 +1,21 @@
> +/*
> + * Copyright (c) 2017, The Linux Foundation. All rights reserved.
> + *
> + * This software is licensed under the terms of the GNU General Public
> + * License version 2, as published by the Free Software Foundation, and
> + * may be copied, distributed, and modified under those terms.
> + *
> + * This program is distributed in the hope that it will be useful,
> + * but WITHOUT ANY WARRANTY; without even the implied warranty of
> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
> + * GNU General Public License for more details.
> + */
> +
> +#ifndef _DT_BINDINGS_SPMI_PMIC_CLK_DIV_H
> +#define _DT_BINDINGS_SPMI_PMIC_CLK_DIV_H
> +
> +#define CLKDIV1		0

1 is 0...

> +#define CLKDIV2		1

2 is 1...

> +#define CLKDIV3		2

How about we offset the index once and then drop the header file?

-- 
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project