diff mbox

[v2,1/2] ARM: shmobile: r8a7790: implement CPU clock scaling for CPUFreq

Message ID Pine.LNX.4.64.1305211601180.2863@axis700.grange (mailing list archive)
State New, archived
Headers show

Commit Message

Guennadi Liakhovetski May 21, 2013, 2:07 p.m. UTC
This patch adds support for the Z-clock on r8a7790 SoCs, which is driving
the Cortex A15 core, and a "cpufreq-cpu0" platform device. Adding an
"operating-points" property to the CPU0 DT node and a regulator, this
patch allows platforms to use the generic cpufreq-cpu0 driver to use
SoC's DVFS capabilities.

Signed-off-by: Guennadi Liakhovetski <g.liakhovetski+renesas@gmail.com>
---

v2:
1. a typo in the r8a7790 datasheet confused me to implement kick-bit 
handling wrongly, this is fixed in this version
2. patch 3 merged with this one

Same comments, as to v1 of this series, apply.

 arch/arm/mach-shmobile/Kconfig         |    2 +
 arch/arm/mach-shmobile/clock-r8a7790.c |  129 ++++++++++++++++++++++++++++++++
 arch/arm/mach-shmobile/setup-r8a7790.c |    1 +
 3 files changed, 132 insertions(+), 0 deletions(-)

Comments

Simon Horman May 25, 2013, 1:34 a.m. UTC | #1
On Tue, May 21, 2013 at 04:07:46PM +0200, Guennadi Liakhovetski wrote:
> This patch adds support for the Z-clock on r8a7790 SoCs, which is driving
> the Cortex A15 core, and a "cpufreq-cpu0" platform device. Adding an
> "operating-points" property to the CPU0 DT node and a regulator, this
> patch allows platforms to use the generic cpufreq-cpu0 driver to use
> SoC's DVFS capabilities.
> 
> Signed-off-by: Guennadi Liakhovetski <g.liakhovetski+renesas@gmail.com>
> ---
> 
> v2:
> 1. a typo in the r8a7790 datasheet confused me to implement kick-bit 
> handling wrongly, this is fixed in this version
> 2. patch 3 merged with this one
> 
> Same comments, as to v1 of this series, apply.

Magnus,

could you please review this series?

> 
>  arch/arm/mach-shmobile/Kconfig         |    2 +
>  arch/arm/mach-shmobile/clock-r8a7790.c |  129 ++++++++++++++++++++++++++++++++
>  arch/arm/mach-shmobile/setup-r8a7790.c |    1 +
>  3 files changed, 132 insertions(+), 0 deletions(-)
> 
> diff --git a/arch/arm/mach-shmobile/Kconfig b/arch/arm/mach-shmobile/Kconfig
> index 0319cf9..97bc6cf 100644
> --- a/arch/arm/mach-shmobile/Kconfig
> +++ b/arch/arm/mach-shmobile/Kconfig
> @@ -60,6 +60,8 @@ config ARCH_R8A7790
>  	select ARM_ARCH_TIMER
>  	select SH_CLK_CPG
>  	select RENESAS_IRQC
> +	select ARCH_HAS_CPUFREQ
> +	select ARCH_HAS_OPP
>  
>  config ARCH_EMEV2
>  	bool "Emma Mobile EV2"
> diff --git a/arch/arm/mach-shmobile/clock-r8a7790.c b/arch/arm/mach-shmobile/clock-r8a7790.c
> index 53a13b4..0f2aa68 100644
> --- a/arch/arm/mach-shmobile/clock-r8a7790.c
> +++ b/arch/arm/mach-shmobile/clock-r8a7790.c
> @@ -47,6 +47,9 @@
>  #define CPG_BASE 0xe6150000
>  #define CPG_LEN 0x1000
>  
> +#define FRQCRB   0xe6150004
> +#define FRQCRC   0xe61500e0
> +
>  #define SMSTPCR2 0xe6150138
>  #define SMSTPCR3 0xe615013c
>  #define SMSTPCR7 0xe615014c
> @@ -83,11 +86,132 @@ static struct clk main_clk = {
>   * clock ratio of these clock will be updated
>   * on r8a7790_clock_init()
>   */
> +SH_FIXED_RATIO_CLK_SET(pll0_clk,		main_clk,	1, 1);
>  SH_FIXED_RATIO_CLK_SET(pll1_clk,		main_clk,	1, 1);
>  SH_FIXED_RATIO_CLK_SET(pll3_clk,		main_clk,	1, 1);
>  SH_FIXED_RATIO_CLK_SET(lb_clk,			pll1_clk,	1, 1);
>  SH_FIXED_RATIO_CLK_SET(qspi_clk,		pll1_clk,	1, 1);
>  
> +/* If more clocks need to access FRQCRB, we'll have to lock eventually */
> +static int frqcr_kick_check(struct clk *clk)
> +{
> +	return ioread32(clk->mapped_reg) & BIT(31) ? -EBUSY : 0;
> +}
> +
> +static int frqcr_kick_do(struct clk *clk)
> +{
> +	int i;
> +
> +	/* set KICK bit in FRQCRB to update hardware setting, check success */
> +	iowrite32(ioread32(clk->mapped_reg) | BIT(31), clk->mapped_reg);
> +	for (i = 1000; i; i--)
> +		if (ioread32(clk->mapped_reg) & BIT(31))
> +			cpu_relax();
> +		else
> +			return 0;
> +
> +	return -ETIMEDOUT;
> +}
> +
> +static int zclk_set_rate(struct clk *clk, unsigned long rate)
> +{
> +	void __iomem *frqcrc;
> +	int ret;
> +	unsigned long step, p_rate;
> +	u32 val;
> +
> +	if (!clk->parent || !__clk_get(clk->parent))
> +		return -ENODEV;
> +
> +	ret = frqcr_kick_check(clk);
> +	if (ret < 0)
> +		goto done;
> +
> +	frqcrc = clk->mapped_reg + (FRQCRC - (u32)clk->enable_reg);
> +
> +	p_rate = clk_get_rate(clk->parent);
> +	if (rate == p_rate) {
> +		val = 0;
> +	} else {
> +		step = DIV_ROUND_CLOSEST(p_rate, 32);
> +		val = 32 - rate / step;
> +	}
> +
> +	iowrite32((ioread32(frqcrc) & ~(clk->div_mask << clk->enable_bit)) |
> +		  (val << clk->enable_bit), frqcrc);
> +
> +	ret = frqcr_kick_do(clk);
> +
> +done:
> +	__clk_put(clk->parent);
> +	return ret;
> +}
> +
> +static long zclk_round_rate(struct clk *clk, unsigned long rate)
> +{
> +	/*
> +	 * theoretical rate = parent rate * multiplier / 32,
> +	 * where 1 <= multiplier <= 32. Therefore we should do
> +	 * multiplier = rate * 32 / parent rate
> +	 * rounded rate = parent rate * multiplier / 32.
> +	 * However, multiplication before division won't fit in 32 bits, so
> +	 * we sacrifice some precision by first dividing and then multiplying.
> +	 * To find the nearest divisor we calculate both and pick up the best
> +	 * one. This avoids 64-bit arithmetics.
> +	 */
> +	unsigned long step, mul_min, mul_max, rate_min, rate_max;
> +
> +	rate_max = clk_get_rate(clk->parent);
> +
> +	/* output freq <= parent */
> +	if (rate >= rate_max)
> +		return rate_max;
> +
> +	step = DIV_ROUND_CLOSEST(rate_max, 32);
> +	/* output freq >= parent / 32 */
> +	if (step >= rate)
> +		return step;
> +
> +	mul_min = rate / step;
> +	mul_max = DIV_ROUND_UP(rate, step);
> +	rate_min = step * mul_min;
> +	if (mul_max == mul_min)
> +		return rate_min;
> +
> +	rate_max = step * mul_max;
> +
> +	if (rate_max - rate <  rate - rate_min)
> +		return rate_max;
> +
> +	return rate_min;
> +}
> +
> +static unsigned long zclk_recalc(struct clk *clk)
> +{
> +	void __iomem *frqcrc = FRQCRC - (u32)clk->enable_reg + clk->mapped_reg;
> +	unsigned int max = clk->div_mask + 1;
> +	unsigned long val = ((ioread32(frqcrc) >> clk->enable_bit) &
> +			     clk->div_mask);
> +
> +	return DIV_ROUND_CLOSEST(clk_get_rate(clk->parent), max) *
> +		(max - val);
> +}
> +
> +static struct sh_clk_ops zclk_ops = {
> +	.recalc = zclk_recalc,
> +	.set_rate = zclk_set_rate,
> +	.round_rate = zclk_round_rate,
> +};
> +
> +static struct clk z_clk = {
> +	.parent = &pll0_clk,
> +	.div_mask = 0x1f,
> +	.enable_bit = 8,
> +	/* We'll need to access FRQCRB and FRQCRC */
> +	.enable_reg = (void __iomem *)FRQCRB,
> +	.ops = &zclk_ops,
> +};
> +
>  /* fixed ratio clock */
>  SH_FIXED_RATIO_CLK_SET(extal_div2_clk,		extal_clk,	1, 2);
>  SH_FIXED_RATIO_CLK_SET(cp_clk,			extal_clk,	1, 2);
> @@ -115,11 +239,13 @@ static struct clk *main_clks[] = {
>  	&extal_clk,
>  	&extal_div2_clk,
>  	&main_clk,
> +	&pll0_clk,
>  	&pll1_clk,
>  	&pll1_div2_clk,
>  	&pll3_clk,
>  	&lb_clk,
>  	&qspi_clk,
> +	&z_clk,
>  	&zg_clk,
>  	&zx_clk,
>  	&zs_clk,
> @@ -213,6 +339,7 @@ static struct clk_lookup lookups[] = {
>  	CLKDEV_CON_ID("pll1",		&pll1_clk),
>  	CLKDEV_CON_ID("pll1_div2",	&pll1_div2_clk),
>  	CLKDEV_CON_ID("pll3",		&pll3_clk),
> +	CLKDEV_DEV_ID("cpufreq-cpu0",	&z_clk),
>  	CLKDEV_CON_ID("zg",		&zg_clk),
>  	CLKDEV_CON_ID("zx",		&zx_clk),
>  	CLKDEV_CON_ID("zs",		&zs_clk),
> @@ -263,9 +390,11 @@ static struct clk_lookup lookups[] = {
>  	CLKDEV_DEV_ID("sh_mmcif.1", &mstp_clks[MSTP305]),
>  };
>  
> +/* "/ 2" is a "fixed divider" - see the "PLL Multiplication Ratio" table */
>  #define R8A7790_CLOCK_ROOT(e, m, p0, p1, p30, p31)		\
>  	extal_clk.rate	= e * 1000 * 1000;			\
>  	main_clk.parent	= m;					\
> +	SH_CLK_SET_RATIO(&pll0_clk_ratio, p0 / 2, 1);		\
>  	SH_CLK_SET_RATIO(&pll1_clk_ratio, p1 / 2, 1);		\
>  	if (mode & MD(19))					\
>  		SH_CLK_SET_RATIO(&pll3_clk_ratio, p31, 1);	\
> diff --git a/arch/arm/mach-shmobile/setup-r8a7790.c b/arch/arm/mach-shmobile/setup-r8a7790.c
> index eeef5f6..49c3166 100644
> --- a/arch/arm/mach-shmobile/setup-r8a7790.c
> +++ b/arch/arm/mach-shmobile/setup-r8a7790.c
> @@ -166,6 +166,7 @@ void __init r8a7790_timer_init(void)
>  #ifdef CONFIG_USE_OF
>  void __init r8a7790_add_standard_devices_dt(void)
>  {
> +	platform_device_register_simple("cpufreq-cpu0", -1, NULL, 0);
>  	of_platform_populate(NULL, of_default_bus_match_table, NULL, NULL);
>  }
>  
> -- 
> 1.7.2.5
>
Simon Horman Sept. 25, 2013, 4:50 a.m. UTC | #2
On Sat, May 25, 2013 at 10:34:45AM +0900, Simon Horman wrote:
> On Tue, May 21, 2013 at 04:07:46PM +0200, Guennadi Liakhovetski wrote:
> > This patch adds support for the Z-clock on r8a7790 SoCs, which is driving
> > the Cortex A15 core, and a "cpufreq-cpu0" platform device. Adding an
> > "operating-points" property to the CPU0 DT node and a regulator, this
> > patch allows platforms to use the generic cpufreq-cpu0 driver to use
> > SoC's DVFS capabilities.
> > 
> > Signed-off-by: Guennadi Liakhovetski <g.liakhovetski+renesas@gmail.com>
> > ---
> > 
> > v2:
> > 1. a typo in the r8a7790 datasheet confused me to implement kick-bit 
> > handling wrongly, this is fixed in this version
> > 2. patch 3 merged with this one
> > 
> > Same comments, as to v1 of this series, apply.
> 
> Magnus,
> 
> could you please review this series?

I have marked this patch as deferred in the patch tracker
as it has been rotting there for several months now.

I will need the following before taking it into the tree:
1. Repost, rebased if needed.
2. An ack from Magnus or indication from him that someone else should
   review it.

> 
> > 
> >  arch/arm/mach-shmobile/Kconfig         |    2 +
> >  arch/arm/mach-shmobile/clock-r8a7790.c |  129 ++++++++++++++++++++++++++++++++
> >  arch/arm/mach-shmobile/setup-r8a7790.c |    1 +
> >  3 files changed, 132 insertions(+), 0 deletions(-)
> > 
> > diff --git a/arch/arm/mach-shmobile/Kconfig b/arch/arm/mach-shmobile/Kconfig
> > index 0319cf9..97bc6cf 100644
> > --- a/arch/arm/mach-shmobile/Kconfig
> > +++ b/arch/arm/mach-shmobile/Kconfig
> > @@ -60,6 +60,8 @@ config ARCH_R8A7790
> >  	select ARM_ARCH_TIMER
> >  	select SH_CLK_CPG
> >  	select RENESAS_IRQC
> > +	select ARCH_HAS_CPUFREQ
> > +	select ARCH_HAS_OPP
> >  
> >  config ARCH_EMEV2
> >  	bool "Emma Mobile EV2"
> > diff --git a/arch/arm/mach-shmobile/clock-r8a7790.c b/arch/arm/mach-shmobile/clock-r8a7790.c
> > index 53a13b4..0f2aa68 100644
> > --- a/arch/arm/mach-shmobile/clock-r8a7790.c
> > +++ b/arch/arm/mach-shmobile/clock-r8a7790.c
> > @@ -47,6 +47,9 @@
> >  #define CPG_BASE 0xe6150000
> >  #define CPG_LEN 0x1000
> >  
> > +#define FRQCRB   0xe6150004
> > +#define FRQCRC   0xe61500e0
> > +
> >  #define SMSTPCR2 0xe6150138
> >  #define SMSTPCR3 0xe615013c
> >  #define SMSTPCR7 0xe615014c
> > @@ -83,11 +86,132 @@ static struct clk main_clk = {
> >   * clock ratio of these clock will be updated
> >   * on r8a7790_clock_init()
> >   */
> > +SH_FIXED_RATIO_CLK_SET(pll0_clk,		main_clk,	1, 1);
> >  SH_FIXED_RATIO_CLK_SET(pll1_clk,		main_clk,	1, 1);
> >  SH_FIXED_RATIO_CLK_SET(pll3_clk,		main_clk,	1, 1);
> >  SH_FIXED_RATIO_CLK_SET(lb_clk,			pll1_clk,	1, 1);
> >  SH_FIXED_RATIO_CLK_SET(qspi_clk,		pll1_clk,	1, 1);
> >  
> > +/* If more clocks need to access FRQCRB, we'll have to lock eventually */
> > +static int frqcr_kick_check(struct clk *clk)
> > +{
> > +	return ioread32(clk->mapped_reg) & BIT(31) ? -EBUSY : 0;
> > +}
> > +
> > +static int frqcr_kick_do(struct clk *clk)
> > +{
> > +	int i;
> > +
> > +	/* set KICK bit in FRQCRB to update hardware setting, check success */
> > +	iowrite32(ioread32(clk->mapped_reg) | BIT(31), clk->mapped_reg);
> > +	for (i = 1000; i; i--)
> > +		if (ioread32(clk->mapped_reg) & BIT(31))
> > +			cpu_relax();
> > +		else
> > +			return 0;
> > +
> > +	return -ETIMEDOUT;
> > +}
> > +
> > +static int zclk_set_rate(struct clk *clk, unsigned long rate)
> > +{
> > +	void __iomem *frqcrc;
> > +	int ret;
> > +	unsigned long step, p_rate;
> > +	u32 val;
> > +
> > +	if (!clk->parent || !__clk_get(clk->parent))
> > +		return -ENODEV;
> > +
> > +	ret = frqcr_kick_check(clk);
> > +	if (ret < 0)
> > +		goto done;
> > +
> > +	frqcrc = clk->mapped_reg + (FRQCRC - (u32)clk->enable_reg);
> > +
> > +	p_rate = clk_get_rate(clk->parent);
> > +	if (rate == p_rate) {
> > +		val = 0;
> > +	} else {
> > +		step = DIV_ROUND_CLOSEST(p_rate, 32);
> > +		val = 32 - rate / step;
> > +	}
> > +
> > +	iowrite32((ioread32(frqcrc) & ~(clk->div_mask << clk->enable_bit)) |
> > +		  (val << clk->enable_bit), frqcrc);
> > +
> > +	ret = frqcr_kick_do(clk);
> > +
> > +done:
> > +	__clk_put(clk->parent);
> > +	return ret;
> > +}
> > +
> > +static long zclk_round_rate(struct clk *clk, unsigned long rate)
> > +{
> > +	/*
> > +	 * theoretical rate = parent rate * multiplier / 32,
> > +	 * where 1 <= multiplier <= 32. Therefore we should do
> > +	 * multiplier = rate * 32 / parent rate
> > +	 * rounded rate = parent rate * multiplier / 32.
> > +	 * However, multiplication before division won't fit in 32 bits, so
> > +	 * we sacrifice some precision by first dividing and then multiplying.
> > +	 * To find the nearest divisor we calculate both and pick up the best
> > +	 * one. This avoids 64-bit arithmetics.
> > +	 */
> > +	unsigned long step, mul_min, mul_max, rate_min, rate_max;
> > +
> > +	rate_max = clk_get_rate(clk->parent);
> > +
> > +	/* output freq <= parent */
> > +	if (rate >= rate_max)
> > +		return rate_max;
> > +
> > +	step = DIV_ROUND_CLOSEST(rate_max, 32);
> > +	/* output freq >= parent / 32 */
> > +	if (step >= rate)
> > +		return step;
> > +
> > +	mul_min = rate / step;
> > +	mul_max = DIV_ROUND_UP(rate, step);
> > +	rate_min = step * mul_min;
> > +	if (mul_max == mul_min)
> > +		return rate_min;
> > +
> > +	rate_max = step * mul_max;
> > +
> > +	if (rate_max - rate <  rate - rate_min)
> > +		return rate_max;
> > +
> > +	return rate_min;
> > +}
> > +
> > +static unsigned long zclk_recalc(struct clk *clk)
> > +{
> > +	void __iomem *frqcrc = FRQCRC - (u32)clk->enable_reg + clk->mapped_reg;
> > +	unsigned int max = clk->div_mask + 1;
> > +	unsigned long val = ((ioread32(frqcrc) >> clk->enable_bit) &
> > +			     clk->div_mask);
> > +
> > +	return DIV_ROUND_CLOSEST(clk_get_rate(clk->parent), max) *
> > +		(max - val);
> > +}
> > +
> > +static struct sh_clk_ops zclk_ops = {
> > +	.recalc = zclk_recalc,
> > +	.set_rate = zclk_set_rate,
> > +	.round_rate = zclk_round_rate,
> > +};
> > +
> > +static struct clk z_clk = {
> > +	.parent = &pll0_clk,
> > +	.div_mask = 0x1f,
> > +	.enable_bit = 8,
> > +	/* We'll need to access FRQCRB and FRQCRC */
> > +	.enable_reg = (void __iomem *)FRQCRB,
> > +	.ops = &zclk_ops,
> > +};
> > +
> >  /* fixed ratio clock */
> >  SH_FIXED_RATIO_CLK_SET(extal_div2_clk,		extal_clk,	1, 2);
> >  SH_FIXED_RATIO_CLK_SET(cp_clk,			extal_clk,	1, 2);
> > @@ -115,11 +239,13 @@ static struct clk *main_clks[] = {
> >  	&extal_clk,
> >  	&extal_div2_clk,
> >  	&main_clk,
> > +	&pll0_clk,
> >  	&pll1_clk,
> >  	&pll1_div2_clk,
> >  	&pll3_clk,
> >  	&lb_clk,
> >  	&qspi_clk,
> > +	&z_clk,
> >  	&zg_clk,
> >  	&zx_clk,
> >  	&zs_clk,
> > @@ -213,6 +339,7 @@ static struct clk_lookup lookups[] = {
> >  	CLKDEV_CON_ID("pll1",		&pll1_clk),
> >  	CLKDEV_CON_ID("pll1_div2",	&pll1_div2_clk),
> >  	CLKDEV_CON_ID("pll3",		&pll3_clk),
> > +	CLKDEV_DEV_ID("cpufreq-cpu0",	&z_clk),
> >  	CLKDEV_CON_ID("zg",		&zg_clk),
> >  	CLKDEV_CON_ID("zx",		&zx_clk),
> >  	CLKDEV_CON_ID("zs",		&zs_clk),
> > @@ -263,9 +390,11 @@ static struct clk_lookup lookups[] = {
> >  	CLKDEV_DEV_ID("sh_mmcif.1", &mstp_clks[MSTP305]),
> >  };
> >  
> > +/* "/ 2" is a "fixed divider" - see the "PLL Multiplication Ratio" table */
> >  #define R8A7790_CLOCK_ROOT(e, m, p0, p1, p30, p31)		\
> >  	extal_clk.rate	= e * 1000 * 1000;			\
> >  	main_clk.parent	= m;					\
> > +	SH_CLK_SET_RATIO(&pll0_clk_ratio, p0 / 2, 1);		\
> >  	SH_CLK_SET_RATIO(&pll1_clk_ratio, p1 / 2, 1);		\
> >  	if (mode & MD(19))					\
> >  		SH_CLK_SET_RATIO(&pll3_clk_ratio, p31, 1);	\
> > diff --git a/arch/arm/mach-shmobile/setup-r8a7790.c b/arch/arm/mach-shmobile/setup-r8a7790.c
> > index eeef5f6..49c3166 100644
> > --- a/arch/arm/mach-shmobile/setup-r8a7790.c
> > +++ b/arch/arm/mach-shmobile/setup-r8a7790.c
> > @@ -166,6 +166,7 @@ void __init r8a7790_timer_init(void)
> >  #ifdef CONFIG_USE_OF
> >  void __init r8a7790_add_standard_devices_dt(void)
> >  {
> > +	platform_device_register_simple("cpufreq-cpu0", -1, NULL, 0);
> >  	of_platform_populate(NULL, of_default_bus_match_table, NULL, NULL);
> >  }
> >  
> > -- 
> > 1.7.2.5
> > 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-sh" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>
Guennadi Liakhovetski Sept. 26, 2013, 8:56 a.m. UTC | #3
Hi Simon

Thanks for following on this.

On Wed, 25 Sep 2013, Simon Horman wrote:

> On Sat, May 25, 2013 at 10:34:45AM +0900, Simon Horman wrote:
> > On Tue, May 21, 2013 at 04:07:46PM +0200, Guennadi Liakhovetski wrote:
> > > This patch adds support for the Z-clock on r8a7790 SoCs, which is driving
> > > the Cortex A15 core, and a "cpufreq-cpu0" platform device. Adding an
> > > "operating-points" property to the CPU0 DT node and a regulator, this
> > > patch allows platforms to use the generic cpufreq-cpu0 driver to use
> > > SoC's DVFS capabilities.
> > > 
> > > Signed-off-by: Guennadi Liakhovetski <g.liakhovetski+renesas@gmail.com>
> > > ---
> > > 
> > > v2:
> > > 1. a typo in the r8a7790 datasheet confused me to implement kick-bit 
> > > handling wrongly, this is fixed in this version
> > > 2. patch 3 merged with this one
> > > 
> > > Same comments, as to v1 of this series, apply.
> > 
> > Magnus,
> > 
> > could you please review this series?
> 
> I have marked this patch as deferred in the patch tracker
> as it has been rotting there for several months now.
> 
> I will need the following before taking it into the tree:
> 1. Repost, rebased if needed.
> 2. An ack from Magnus or indication from him that someone else should
>    review it.

This patch is superseded by my recent patches

[PATCH 3/4] ARM: shmobile: r8a7790: add CPUFreq clock support
[PATCH/RFC 4/4] ARM: shmobile: lager: add CPUFreq support

Thanks
Guennadi
---
Guennadi Liakhovetski, Ph.D.
Freelance Open-Source Software Developer
http://www.open-technology.de/
Simon Horman Sept. 26, 2013, 9:40 p.m. UTC | #4
On Thu, Sep 26, 2013 at 10:56:02AM +0200, Guennadi Liakhovetski wrote:
> Hi Simon
> 
> Thanks for following on this.
> 
> On Wed, 25 Sep 2013, Simon Horman wrote:
> 
> > On Sat, May 25, 2013 at 10:34:45AM +0900, Simon Horman wrote:
> > > On Tue, May 21, 2013 at 04:07:46PM +0200, Guennadi Liakhovetski wrote:
> > > > This patch adds support for the Z-clock on r8a7790 SoCs, which is driving
> > > > the Cortex A15 core, and a "cpufreq-cpu0" platform device. Adding an
> > > > "operating-points" property to the CPU0 DT node and a regulator, this
> > > > patch allows platforms to use the generic cpufreq-cpu0 driver to use
> > > > SoC's DVFS capabilities.
> > > > 
> > > > Signed-off-by: Guennadi Liakhovetski <g.liakhovetski+renesas@gmail.com>
> > > > ---
> > > > 
> > > > v2:
> > > > 1. a typo in the r8a7790 datasheet confused me to implement kick-bit 
> > > > handling wrongly, this is fixed in this version
> > > > 2. patch 3 merged with this one
> > > > 
> > > > Same comments, as to v1 of this series, apply.
> > > 
> > > Magnus,
> > > 
> > > could you please review this series?
> > 
> > I have marked this patch as deferred in the patch tracker
> > as it has been rotting there for several months now.
> > 
> > I will need the following before taking it into the tree:
> > 1. Repost, rebased if needed.
> > 2. An ack from Magnus or indication from him that someone else should
> >    review it.
> 
> This patch is superseded by my recent patches
> 
> [PATCH 3/4] ARM: shmobile: r8a7790: add CPUFreq clock support
> [PATCH/RFC 4/4] ARM: shmobile: lager: add CPUFreq support

Thanks, noted.
diff mbox

Patch

diff --git a/arch/arm/mach-shmobile/Kconfig b/arch/arm/mach-shmobile/Kconfig
index 0319cf9..97bc6cf 100644
--- a/arch/arm/mach-shmobile/Kconfig
+++ b/arch/arm/mach-shmobile/Kconfig
@@ -60,6 +60,8 @@  config ARCH_R8A7790
 	select ARM_ARCH_TIMER
 	select SH_CLK_CPG
 	select RENESAS_IRQC
+	select ARCH_HAS_CPUFREQ
+	select ARCH_HAS_OPP
 
 config ARCH_EMEV2
 	bool "Emma Mobile EV2"
diff --git a/arch/arm/mach-shmobile/clock-r8a7790.c b/arch/arm/mach-shmobile/clock-r8a7790.c
index 53a13b4..0f2aa68 100644
--- a/arch/arm/mach-shmobile/clock-r8a7790.c
+++ b/arch/arm/mach-shmobile/clock-r8a7790.c
@@ -47,6 +47,9 @@ 
 #define CPG_BASE 0xe6150000
 #define CPG_LEN 0x1000
 
+#define FRQCRB   0xe6150004
+#define FRQCRC   0xe61500e0
+
 #define SMSTPCR2 0xe6150138
 #define SMSTPCR3 0xe615013c
 #define SMSTPCR7 0xe615014c
@@ -83,11 +86,132 @@  static struct clk main_clk = {
  * clock ratio of these clock will be updated
  * on r8a7790_clock_init()
  */
+SH_FIXED_RATIO_CLK_SET(pll0_clk,		main_clk,	1, 1);
 SH_FIXED_RATIO_CLK_SET(pll1_clk,		main_clk,	1, 1);
 SH_FIXED_RATIO_CLK_SET(pll3_clk,		main_clk,	1, 1);
 SH_FIXED_RATIO_CLK_SET(lb_clk,			pll1_clk,	1, 1);
 SH_FIXED_RATIO_CLK_SET(qspi_clk,		pll1_clk,	1, 1);
 
+/* If more clocks need to access FRQCRB, we'll have to lock eventually */
+static int frqcr_kick_check(struct clk *clk)
+{
+	return ioread32(clk->mapped_reg) & BIT(31) ? -EBUSY : 0;
+}
+
+static int frqcr_kick_do(struct clk *clk)
+{
+	int i;
+
+	/* set KICK bit in FRQCRB to update hardware setting, check success */
+	iowrite32(ioread32(clk->mapped_reg) | BIT(31), clk->mapped_reg);
+	for (i = 1000; i; i--)
+		if (ioread32(clk->mapped_reg) & BIT(31))
+			cpu_relax();
+		else
+			return 0;
+
+	return -ETIMEDOUT;
+}
+
+static int zclk_set_rate(struct clk *clk, unsigned long rate)
+{
+	void __iomem *frqcrc;
+	int ret;
+	unsigned long step, p_rate;
+	u32 val;
+
+	if (!clk->parent || !__clk_get(clk->parent))
+		return -ENODEV;
+
+	ret = frqcr_kick_check(clk);
+	if (ret < 0)
+		goto done;
+
+	frqcrc = clk->mapped_reg + (FRQCRC - (u32)clk->enable_reg);
+
+	p_rate = clk_get_rate(clk->parent);
+	if (rate == p_rate) {
+		val = 0;
+	} else {
+		step = DIV_ROUND_CLOSEST(p_rate, 32);
+		val = 32 - rate / step;
+	}
+
+	iowrite32((ioread32(frqcrc) & ~(clk->div_mask << clk->enable_bit)) |
+		  (val << clk->enable_bit), frqcrc);
+
+	ret = frqcr_kick_do(clk);
+
+done:
+	__clk_put(clk->parent);
+	return ret;
+}
+
+static long zclk_round_rate(struct clk *clk, unsigned long rate)
+{
+	/*
+	 * theoretical rate = parent rate * multiplier / 32,
+	 * where 1 <= multiplier <= 32. Therefore we should do
+	 * multiplier = rate * 32 / parent rate
+	 * rounded rate = parent rate * multiplier / 32.
+	 * However, multiplication before division won't fit in 32 bits, so
+	 * we sacrifice some precision by first dividing and then multiplying.
+	 * To find the nearest divisor we calculate both and pick up the best
+	 * one. This avoids 64-bit arithmetics.
+	 */
+	unsigned long step, mul_min, mul_max, rate_min, rate_max;
+
+	rate_max = clk_get_rate(clk->parent);
+
+	/* output freq <= parent */
+	if (rate >= rate_max)
+		return rate_max;
+
+	step = DIV_ROUND_CLOSEST(rate_max, 32);
+	/* output freq >= parent / 32 */
+	if (step >= rate)
+		return step;
+
+	mul_min = rate / step;
+	mul_max = DIV_ROUND_UP(rate, step);
+	rate_min = step * mul_min;
+	if (mul_max == mul_min)
+		return rate_min;
+
+	rate_max = step * mul_max;
+
+	if (rate_max - rate <  rate - rate_min)
+		return rate_max;
+
+	return rate_min;
+}
+
+static unsigned long zclk_recalc(struct clk *clk)
+{
+	void __iomem *frqcrc = FRQCRC - (u32)clk->enable_reg + clk->mapped_reg;
+	unsigned int max = clk->div_mask + 1;
+	unsigned long val = ((ioread32(frqcrc) >> clk->enable_bit) &
+			     clk->div_mask);
+
+	return DIV_ROUND_CLOSEST(clk_get_rate(clk->parent), max) *
+		(max - val);
+}
+
+static struct sh_clk_ops zclk_ops = {
+	.recalc = zclk_recalc,
+	.set_rate = zclk_set_rate,
+	.round_rate = zclk_round_rate,
+};
+
+static struct clk z_clk = {
+	.parent = &pll0_clk,
+	.div_mask = 0x1f,
+	.enable_bit = 8,
+	/* We'll need to access FRQCRB and FRQCRC */
+	.enable_reg = (void __iomem *)FRQCRB,
+	.ops = &zclk_ops,
+};
+
 /* fixed ratio clock */
 SH_FIXED_RATIO_CLK_SET(extal_div2_clk,		extal_clk,	1, 2);
 SH_FIXED_RATIO_CLK_SET(cp_clk,			extal_clk,	1, 2);
@@ -115,11 +239,13 @@  static struct clk *main_clks[] = {
 	&extal_clk,
 	&extal_div2_clk,
 	&main_clk,
+	&pll0_clk,
 	&pll1_clk,
 	&pll1_div2_clk,
 	&pll3_clk,
 	&lb_clk,
 	&qspi_clk,
+	&z_clk,
 	&zg_clk,
 	&zx_clk,
 	&zs_clk,
@@ -213,6 +339,7 @@  static struct clk_lookup lookups[] = {
 	CLKDEV_CON_ID("pll1",		&pll1_clk),
 	CLKDEV_CON_ID("pll1_div2",	&pll1_div2_clk),
 	CLKDEV_CON_ID("pll3",		&pll3_clk),
+	CLKDEV_DEV_ID("cpufreq-cpu0",	&z_clk),
 	CLKDEV_CON_ID("zg",		&zg_clk),
 	CLKDEV_CON_ID("zx",		&zx_clk),
 	CLKDEV_CON_ID("zs",		&zs_clk),
@@ -263,9 +390,11 @@  static struct clk_lookup lookups[] = {
 	CLKDEV_DEV_ID("sh_mmcif.1", &mstp_clks[MSTP305]),
 };
 
+/* "/ 2" is a "fixed divider" - see the "PLL Multiplication Ratio" table */
 #define R8A7790_CLOCK_ROOT(e, m, p0, p1, p30, p31)		\
 	extal_clk.rate	= e * 1000 * 1000;			\
 	main_clk.parent	= m;					\
+	SH_CLK_SET_RATIO(&pll0_clk_ratio, p0 / 2, 1);		\
 	SH_CLK_SET_RATIO(&pll1_clk_ratio, p1 / 2, 1);		\
 	if (mode & MD(19))					\
 		SH_CLK_SET_RATIO(&pll3_clk_ratio, p31, 1);	\
diff --git a/arch/arm/mach-shmobile/setup-r8a7790.c b/arch/arm/mach-shmobile/setup-r8a7790.c
index eeef5f6..49c3166 100644
--- a/arch/arm/mach-shmobile/setup-r8a7790.c
+++ b/arch/arm/mach-shmobile/setup-r8a7790.c
@@ -166,6 +166,7 @@  void __init r8a7790_timer_init(void)
 #ifdef CONFIG_USE_OF
 void __init r8a7790_add_standard_devices_dt(void)
 {
+	platform_device_register_simple("cpufreq-cpu0", -1, NULL, 0);
 	of_platform_populate(NULL, of_default_bus_match_table, NULL, NULL);
 }