ARM: shmobile: r8a7790: implement CPU clock scaling for CPUFreq
Subsystems (5)
PATCH v2 1/2 ARM: shmobile: r8a7790: implement CPU clock scaling for CPUFreq
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.
---
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(-)
arch/arm/mach-shmobile/Kconfig +2 -0
@@ -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"
arch/arm/mach-shmobile/clock-r8a7790.c +129 -0
172 lines not on this page · raw message
arch/arm/mach-shmobile/setup-r8a7790.c +1 -0
@@ -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 ); }
Signature · 1 line
1.7.2.5
PATCH v2 2/2 ARM: shmobile: lager: add CPUFreq support
Add DT properties, necessary to enable CPUFreq support on Lager.
---
v2: no change
Again - not to be applied, an illustration only
arch/arm/boot/dts/r8a7790-lager.dts | 8 ++++++++ 1 files changed, 8 insertions(+), 0 deletions(-)
arch/arm/boot/dts/r8a7790-lager.dts +8 -0
@@ -19,6 +19,14 @@ bootargs = "console=ttySC6,115200 ignore_loglevel"; }; + cpus { + cpu@0 { + operating-points = <1300000 1000000 + 100000 900000>; + clock-latency = <1000000>; + }; + }; + memory@40000000 { device_type = "memory"; reg = <0 0x40000000 0 0x80000000>;
Signature · 1 line
1.7.2.5
PATCH v2 1/2 ARM: shmobile: r8a7790: implement CPU clock scaling for CPUFreq
In reply to Guennadi LiakhovetskiQuoted 7 lines · Guennadi Liakhovetski
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 <[email protected]>
Quoted 8 lines
---
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 oneSame comments, as to v1 of this series, apply.
Magnus,
could you please review this series?
Quoted 215 lines
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(-)arch/arm/mach-shmobile/Kconfig +2
-0@@ -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"arch/arm/mach-shmobile/clock-r8a7790.c +129
-0@@ -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); \arch/arm/mach-shmobile/setup-r8a7790.c +1
-0@@ -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 ); }Signature · 1 line
1.7.2.5
PATCH v2 1/2 ARM: shmobile: r8a7790: implement CPU clock scaling for CPUFreq
Quoted 15 lines · Simon Horman
Quoted 14 lines · Guennadi Liakhovetski
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 <[email protected]>---
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
Quoted 5 lines
Quoted 1 line
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.
Quoted 221 lines
Quoted 216 lines
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(-)arch/arm/mach-shmobile/Kconfig +2
-0@@ -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"arch/arm/mach-shmobile/clock-r8a7790.c +129
-0@@ -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); \arch/arm/mach-shmobile/setup-r8a7790.c +1
-0@@ -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 ); }Signature · 1 line
1.7.2.5--
To unsubscribe from this list: send the line "unsubscribe linux-sh" in the body of a message to majordomo at vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
PATCH v2 1/2 ARM: shmobile: r8a7790: implement CPU clock scaling for CPUFreq
Hi Simon
Thanks for following on this.
Quoted 22 lines · Simon Horman
Quoted 15 lines · Simon Horman
Quoted 14 lines · Guennadi Liakhovetski
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 <[email protected]>---
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 oneQuoted 5 lines
Quoted 1 line
Same comments, as to v1 of this series, apply.
Magnus,
could you please review this series?
Quoted 7 lines
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/
PATCH v2 1/2 ARM: shmobile: r8a7790: implement CPU clock scaling for CPUFreq
Quoted 36 lines · Guennadi Liakhovetski
Hi Simon
Thanks for following on this.
Quoted 22 lines · Simon Horman
Quoted 15 lines · Simon Horman
Quoted 14 lines · Guennadi Liakhovetski
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 <[email protected]>---
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 oneQuoted 5 lines
Quoted 1 line
Same comments, as to v1 of this series, apply.
Magnus,
could you please review this series?
Quoted 7 lines
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.
Quoted 4 lines
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.