Re: [PATCH v8 2/3] clk: canaan: Add clock driver for Canaan K230

[Yao Zi <ziyao@disroot.org>]

> On Fri, Sep 05, 2025 at 11:10:23AM +0800, Xukai Wang wrote:
> This patch provides basic support for the K230 clock, which covers
> all clocks in K230 SoC.
> 
> The clock tree of the K230 SoC consists of a 24MHZ external crystal
> oscillator, PLLs and an external pulse input for timerX, and their
> derived clocks.
> 
> Co-developed-by: Troy Mitchell <TroyMitchell988@gmail.com>
> Signed-off-by: Troy Mitchell <TroyMitchell988@gmail.com>
> Signed-off-by: Xukai Wang <kingxukai@zohomail.com>
> ---
>  drivers/clk/Kconfig    |    6 +
>  drivers/clk/Makefile   |    1 +
>  drivers/clk/clk-k230.c | 2456 ++++++++++++++++++++++++++++++++++++++++++++++++
>  3 files changed, 2463 insertions(+)
> 
> diff --git a/drivers/clk/Kconfig b/drivers/clk/Kconfig
> index 299bc678ed1b9fcd9110bb8c5937a1bd1ea60e23..b597912607a6cc8eabff459a890a1e7353ef9c1d 100644
> --- a/drivers/clk/Kconfig
> +++ b/drivers/clk/Kconfig
> @@ -464,6 +464,12 @@ config COMMON_CLK_K210
>  	help
>  	  Support for the Canaan Kendryte K210 RISC-V SoC clocks.
>  
> +config COMMON_CLK_K230
> +	bool "Clock driver for the Canaan Kendryte K230 SoC"
> +	depends on ARCH_CANAAN || COMPILE_TEST
> +	help
> +	  Support for the Canaan Kendryte K230 RISC-V SoC clocks.
> +
>  config COMMON_CLK_SP7021
>  	tristate "Clock driver for Sunplus SP7021 SoC"
>  	depends on SOC_SP7021 || COMPILE_TEST
> diff --git a/drivers/clk/Makefile b/drivers/clk/Makefile
> index fb8878a5d7d93da6bec487460cdf63f1f764a431..5df50b1e14c701ed38397bfb257db26e8dd278b8 100644
> --- a/drivers/clk/Makefile
> +++ b/drivers/clk/Makefile
> @@ -51,6 +51,7 @@ obj-$(CONFIG_MACH_ASPEED_G6)		+= clk-ast2600.o
>  obj-$(CONFIG_ARCH_HIGHBANK)		+= clk-highbank.o
>  obj-$(CONFIG_CLK_HSDK)			+= clk-hsdk-pll.o
>  obj-$(CONFIG_COMMON_CLK_K210)		+= clk-k210.o
> +obj-$(CONFIG_COMMON_CLK_K230)		+= clk-k230.o
>  obj-$(CONFIG_LMK04832)			+= clk-lmk04832.o
>  obj-$(CONFIG_COMMON_CLK_LAN966X)	+= clk-lan966x.o
>  obj-$(CONFIG_COMMON_CLK_LOCHNAGAR)	+= clk-lochnagar.o
> diff --git a/drivers/clk/clk-k230.c b/drivers/clk/clk-k230.c
> new file mode 100644
> index 0000000000000000000000000000000000000000..2ba74c008b30ae3400acbd8c08550e8315dfe205
> --- /dev/null
> +++ b/drivers/clk/clk-k230.c
> @@ -0,0 +1,2456 @@
> +// SPDX-License-Identifier: GPL-2.0-only
> +/*
> + * Kendryte Canaan K230 Clock Drivers
> + *
> + * Author: Xukai Wang <kingxukai@zohomail.com>
> + * Author: Troy Mitchell <troymitchell988@gmail.com>
> + */
> +
> +#include <linux/clk.h>
> +#include <linux/clkdev.h>
> +#include <linux/clk-provider.h>
> +#include <linux/iopoll.h>
> +#include <linux/mod_devicetable.h>
> +#include <linux/platform_device.h>
> +#include <linux/spinlock.h>
> +
> +#include <dt-bindings/clock/canaan,k230-clk.h>
> +
> +/* PLL control register bits. */
> +#define K230_PLL_BYPASS_ENABLE			BIT(19)
> +#define K230_PLL_GATE_ENABLE			BIT(2)
> +#define K230_PLL_GATE_WRITE_ENABLE		BIT(18)
> +#define K230_PLL_OD_SHIFT			24
> +#define K230_PLL_OD_MASK			0xF
> +#define K230_PLL_R_SHIFT			16
> +#define K230_PLL_R_MASK				0x3F
> +#define K230_PLL_F_SHIFT			0
> +#define K230_PLL_F_MASK				0x1FFF
> +#define K230_PLL_DIV_REG_OFFSET			0x00
> +#define K230_PLL_BYPASS_REG_OFFSET		0x04
> +#define K230_PLL_GATE_REG_OFFSET		0x08
> +#define K230_PLL_LOCK_REG_OFFSET		0x0C

Maybe FIELD_PREP() and FIELD_GET() would help for the PLL-related
rountines, and you could get avoid of writing shifts and masks by hand.

...

> +struct k230_clk_rate_self {
> +	struct clk_hw	hw;
> +	void __iomem	*reg;
> +	bool		read_only;

Isn't a read-only multiplier, divider or something capable of both a
simple fixed-factor hardware? If so please switch to the existing clock
hardware, instead of introducing a field in description of rate clocks.

It's worth noting that you've already had at least one fixed-rate clock
(shrm_sram_div2).

> +	u32		write_enable_bit;
> +	u32		mul_min;
> +	u32		mul_max;
> +	u32		mul_shift;
> +	u32		mul_mask;
> +	u32		div_min;
> +	u32		div_max;
> +	u32		div_shift;
> +	u32		div_mask;
> +	/* ensures mutual exclusion for concurrent register access. */
> +	spinlock_t	*lock;
> +};

...

> +static int k230_clk_find_approximate_mul_div(u32 mul_min, u32 mul_max,
> +					     u32 div_min, u32 div_max,
> +					     unsigned long rate,
> +					     unsigned long parent_rate,
> +					     u32 *div, u32 *mul)
> +{
> +	long abs_min;
> +	long abs_current;
> +	long perfect_divide;
> +
> +	if (!rate || !parent_rate || !mul_min)
> +		return -EINVAL;
> +
> +	perfect_divide = (long)((parent_rate * 1000) / rate);
> +	abs_min = abs(perfect_divide -
> +		     (long)(((long)div_max * 1000) / (long)mul_min));
> +
> +	*div = div_max;
> +	*mul = mul_min;
> +
> +	for (u32 i = div_max - 1; i >= div_min; i--) {
> +		for (u32 j = mul_min + 1; j <= mul_max; j++) {
> +			abs_current = abs(perfect_divide -
> +					 (long)(((long)i * 1000) / (long)j));
> +
> +			if (abs_min > abs_current) {
> +				abs_min = abs_current;
> +				*div = i;
> +				*mul = j;
> +			}
> +		}
> +	}
> +
> +	return 0;
> +}

This looks like a poor version of rational_best_approximation(). Could
you please consider switching to it?

> +static int k230_clk_set_rate_mul(struct clk_hw *hw, unsigned long rate,
> +				 unsigned long parent_rate)
> +{
> +	struct k230_clk_rate *clk = hw_to_k230_clk_rate(hw);
> +	struct k230_clk_rate_self *rate_self = &clk->clk;
> +	u32 div, mul, mul_reg;
> +
> +	if (rate > parent_rate)
> +		return -EINVAL;
> +
> +	if (rate_self->read_only)
> +		return 0;
> +
> +	if (k230_clk_find_approximate_mul(rate_self->mul_min, rate_self->mul_max,
> +					  rate_self->div_min, rate_self->div_max,
> +					  rate, parent_rate, &div, &mul))
> +		return -EINVAL;
> +
> +	guard(spinlock)(rate_self->lock);
> +
> +	mul_reg = readl(rate_self->reg + clk->mul_reg_off);
> +	mul_reg |= ((mul - 1) & rate_self->mul_mask) << (rate_self->mul_shift);
> +	mul_reg |= BIT(rate_self->write_enable_bit);
> +	writel(mul_reg, rate_self->reg + clk->mul_reg_off);
> +
> +	return 0;
> +}
> +
> +static int k230_clk_set_rate_div(struct clk_hw *hw, unsigned long rate,
> +				 unsigned long parent_rate)
> +{
> +	struct k230_clk_rate *clk = hw_to_k230_clk_rate(hw);
> +	struct k230_clk_rate_self *rate_self = &clk->clk;
> +	u32 div, mul, div_reg;
> +
> +	if (rate > parent_rate)
> +		return -EINVAL;
> +
> +	if (rate_self->read_only)
> +		return 0;
> +
> +	if (k230_clk_find_approximate_div(rate_self->mul_min, rate_self->mul_max,
> +					  rate_self->div_min, rate_self->div_max,
> +					  rate, parent_rate, &div, &mul))
> +		return -EINVAL;
> +
> +	guard(spinlock)(rate_self->lock);
> +
> +	div_reg = readl(rate_self->reg + clk->div_reg_off);
> +	div_reg |= ((div - 1) & rate_self->div_mask) << (rate_self->div_shift);
> +	div_reg |= BIT(rate_self->write_enable_bit);
> +	writel(div_reg, rate_self->reg + clk->div_reg_off);
> +
> +	return 0;
> +}
> +
> +static int k230_clk_set_rate_mul_div(struct clk_hw *hw, unsigned long rate,
> +				     unsigned long parent_rate)
> +{
> +	struct k230_clk_rate *clk = hw_to_k230_clk_rate(hw);
> +	struct k230_clk_rate_self *rate_self = &clk->clk;
> +	u32 div, mul, div_reg, mul_reg;
> +
> +	if (rate > parent_rate)
> +		return -EINVAL;
> +
> +	if (rate_self->read_only)
> +		return 0;
> +
> +	if (k230_clk_find_approximate_mul_div(rate_self->mul_min, rate_self->mul_max,
> +					      rate_self->div_min, rate_self->div_max,
> +					      rate, parent_rate, &div, &mul))
> +		return -EINVAL;
> +
> +	guard(spinlock)(rate_self->lock);
> +
> +	div_reg = readl(rate_self->reg + clk->div_reg_off);
> +	div_reg |= ((div - 1) & rate_self->div_mask) << (rate_self->div_shift);
> +	div_reg |= BIT(rate_self->write_enable_bit);
> +	writel(div_reg, rate_self->reg + clk->div_reg_off);
> +
> +	mul_reg = readl(rate_self->reg + clk->mul_reg_off);
> +	mul_reg |= ((mul - 1) & rate_self->mul_mask) << (rate_self->mul_shift);
> +	mul_reg |= BIT(rate_self->write_enable_bit);
> +	writel(mul_reg, rate_self->reg + clk->mul_reg_off);
> +
> +	return 0;
> +}

There are three variants of rate clocks, mul-only, div-only and mul-div
ones, which are similar to clk-multiplier, clk-divider,
clk-fractional-divider.

The only difference is to setup new parameters for K230's rate clocks,
a register bit, described as k230_clk_rate_self.write_enable_bit, must
be set first.

What do you think of introducing support for such "write enable bit" to
the generic implementation of multipler/divider/fractional? Then you
could reuse the generic implementation in K230's driver, avoiding code
duplication.

...

> +static const struct of_device_id k230_clk_ids[] = {
> +	{ .compatible = "canaan,k230-clk" },
> +	{ /* Sentinel */ }
> +};
> +MODULE_DEVICE_TABLE(of, k230_clk_ids);

MODULE_DEVICE_TABLE is unnecessary if your driver couldn't be built as
a module.

> +static struct platform_driver k230_clk_driver = {
> +	.driver = {
> +		.name = "k230_clock_controller",
> +		.of_match_table = k230_clk_ids,
> +	},
> +	.probe = k230_clk_probe,
> +};
> +builtin_platform_driver(k230_clk_driver);

Best regards,
Yao Zi