Skip to contentSkip to conversation

Dove cpufreq driver

PATCH Superseded Andrew Lunn 4 participants 31 messages Dec 4, 2013 – Dec 22, 2013

Subsystems (5)
lore
Share & tools
lore

Andrew Lunn

author maintainer
#0 lore raw

v4 cover Dove cpufreq driver

This patchset adds a cpufreq driver for Dove.

The patchset is based on Rafael's linux-next branch from yesterday. Quite likely Jason Cooper will take patches 2-3 via the mvebu tree, but i will let you decide on that.

For version 3 there was a comment that the device tree binding should be reviewed, and Viresh offered to do that. The binding is identical to the kirkwood binding, with the addition that Dove requires an interrupt handler, so the binding also lists this interrupt, using the standard DT properties for interrupts.

Andrew Lunn (4):
cpufreq: Add a cpufreq driver for Marvell Dove
mvebu: Dove: Instantiate cpufreq driver.
mvebu: Dove: Enable cpufreq driver in defconfig
mvebu: Dove: Add clocks and DFS interrupt to cpu node in DT.

 arch/arm/Kconfig                       |   1 +
 arch/arm/boot/dts/dove.dtsi            |   4 +
 arch/arm/configs/dove_defconfig        |   2 +
 arch/arm/mach-dove/board-dt.c          |   3 +
 arch/arm/mach-dove/common.c            |  36 +++++
 arch/arm/mach-dove/common.h            |   1 +
 arch/arm/mach-dove/include/mach/dove.h |   1 +
 drivers/cpufreq/Kconfig.arm            |   5 +
 drivers/cpufreq/Makefile               |   1 +
 drivers/cpufreq/dove-cpufreq.c         | 259 +++++++++++++++++++++++++++++++++
 10 files changed, 313 insertions(+)
 create mode 100644 drivers/cpufreq/dove-cpufreq.c
Signature · 1 line
1.8.5
author maintainer
#1 lore raw

PATCH v4 1/4 cpufreq: Add a cpufreq driver for Marvell Dove

The Marvell Dove SoC can run the CPU at two frequencies. The high frequencey is from a PLL, while the lower is the same as the DDR clock. Add a cpufreq driver to swap between these frequences.

---
Sort header files
Remove unneeded header files
Notify only on change
Use get_cpu_device(0), devm_clk_get()
Use platform_get_resource_byname()
Error check clk_prepare_enable()
Comment the interrupt handler

rebase onto 3.13-rc2 linux-next
use target_index

---
 drivers/cpufreq/Kconfig.arm    |   5 +
 drivers/cpufreq/Makefile       |   1 +
 drivers/cpufreq/dove-cpufreq.c | 259 +++++++++++++++++++++++++++++++++++++++++
 3 files changed, 265 insertions(+)
 create mode 100644 drivers/cpufreq/dove-cpufreq.c
drivers/cpufreq/Kconfig.arm +5 -0
index ce52ed949249..af7c4947f218 100644
@@ -8,6 +8,11 @@ config ARM_BIG_LITTLE_CPUFREQ
 	help
 	  This enables the Generic CPUfreq driver for ARM big.LITTLE platforms.
 
+config ARM_DOVE_CPUFREQ
+	def_bool ARCH_DOVE && OF
+	help
+	  This adds the CPUFreq driver for Marvell Dove SoCs.
+
 config ARM_DT_BL_CPUFREQ
 	tristate "Generic probing via DT for ARM big LITTLE CPUfreq driver"
 	depends on ARM_BIG_LITTLE_CPUFREQ && OF
drivers/cpufreq/Makefile +1 -0
index 74945652dd7a..cd72c2bbfd44 100644
@@ -49,6 +49,7 @@ obj-$(CONFIG_ARM_DT_BL_CPUFREQ)		+= arm_big_little_dt.o
 
 obj-$(CONFIG_ARCH_DAVINCI_DA850)	+= davinci-cpufreq.o
 obj-$(CONFIG_UX500_SOC_DB8500)		+= dbx500-cpufreq.o
+obj-$(CONFIG_ARM_DOVE_CPUFREQ)		+= dove-cpufreq.o
 obj-$(CONFIG_ARM_EXYNOS_CPUFREQ)	+= exynos-cpufreq.o
 obj-$(CONFIG_ARM_EXYNOS4210_CPUFREQ)	+= exynos4210-cpufreq.o
 obj-$(CONFIG_ARM_EXYNOS4X12_CPUFREQ)	+= exynos4x12-cpufreq.o
drivers/cpufreq/dove-cpufreq.c new file +259 -0

259 lines not on this page · raw message

Signature · 1 line
1.8.5
#2 lore raw

PATCH v4 1/4 cpufreq: Add a cpufreq driver for Marvell Dove

Quoted 16 lines · Andrew Lunn

The Marvell Dove SoC can run the CPU at two frequencies. The high frequencey is from a PLL, while the lower is the same as the DDR clock. Add a cpufreq driver to swap between these frequences.

---
Sort header files
Remove unneeded header files
Notify only on change
Use get_cpu_device(0), devm_clk_get()
Use platform_get_resource_byname()
Error check clk_prepare_enable()
Comment the interrupt handler

Quoted 1 line

rebase onto 3.13-rc2 linux-next

Is there something you need that prevents you from basing on v3.13-rc1?

Quoted 7 lines

use target_index

---
 drivers/cpufreq/Kconfig.arm    |   5 +
 drivers/cpufreq/Makefile       |   1 +
 drivers/cpufreq/dove-cpufreq.c | 259 +++++++++++++++++++++++++++++++++++++++++
 3 files changed, 265 insertions(+)
 create mode 100644 drivers/cpufreq/dove-cpufreq.c

Other than that,

thx,

Jason.

author maintainer
#3 lore raw

PATCH v4 1/4 cpufreq: Add a cpufreq driver for Marvell Dove

Quoted 17 lines · Jason Cooper
Quoted 16 lines · Andrew Lunn

The Marvell Dove SoC can run the CPU at two frequencies. The high frequencey is from a PLL, while the lower is the same as the DDR clock. Add a cpufreq driver to swap between these frequences.

---
Sort header files
Remove unneeded header files
Notify only on change
Use get_cpu_device(0), devm_clk_get()
Use platform_get_resource_byname()
Error check clk_prepare_enable()
Comment the interrupt handler

Quoted 3 lines
Quoted 1 line

rebase onto 3.13-rc2 linux-next

Is there something you need that prevents you from basing on v3.13-rc1?

It seems to be Rafael's policy that patches for pm need to be based on his linux-next branch. Viresh has also been making some major changes to all the cpufreq drivers to move code into the core and out of drivers. These changes are in rafaels linux-next, and i've made similar changes to the driver so it fits in.

Andrew

#4 lore raw

PATCH v4 1/4 cpufreq: Add a cpufreq driver for Marvell Dove

Quoted 20 lines · Andrew Lunn
Quoted 19 lines · Jason Cooper
Quoted 17 lines · Andrew Lunn

The Marvell Dove SoC can run the CPU at two frequencies. The high frequencey is from a PLL, while the lower is the same as the DDR clock. Add a cpufreq driver to swap between these frequences.

---
Sort header files
Remove unneeded header files
Notify only on change
Use get_cpu_device(0), devm_clk_get()
Use platform_get_resource_byname()
Error check clk_prepare_enable()
Comment the interrupt handler

rebase onto 3.13-rc2 linux-next

Quoted 7 lines
Quoted 1 line

Is there something you need that prevents you from basing on v3.13-rc1?

It seems to be Rafael's policy that patches for pm need to be based on his linux-next branch. Viresh has also been making some major changes to all the cpufreq drivers to move code into the core and out of drivers. These changes are in rafaels linux-next, and i've made similar changes to the driver so it fits in.

Ah, ok.

thx,

Jason.

#5 lore raw

PATCH v4 1/4 cpufreq: Add a cpufreq driver for Marvell Dove

In reply to Andrew Lunn
Quoted 24 lines · Andrew Lunn

The Marvell Dove SoC can run the CPU at two frequencies. The high frequencey is from a PLL, while the lower is the same as the DDR clock. Add a cpufreq driver to swap between these frequences.

---
Sort header files
Remove unneeded header files
Notify only on change
Use get_cpu_device(0), devm_clk_get()
Use platform_get_resource_byname()
Error check clk_prepare_enable()
Comment the interrupt handler

rebase onto 3.13-rc2 linux-next
use target_index

---
 drivers/cpufreq/Kconfig.arm    |   5 +
 drivers/cpufreq/Makefile       |   1 +
 drivers/cpufreq/dove-cpufreq.c | 259 +++++++++++++++++++++++++++++++++++++++++
 3 files changed, 265 insertions(+)
 create mode 100644 drivers/cpufreq/dove-cpufreq.c

[...]

Quoted 26 lines
drivers/cpufreq/dove-cpufreq.c new file +26 -0
+static int dove_cpufreq_target(struct cpufreq_policy *policy,
+			       unsigned int index)
+{
+	unsigned int state = dove_freq_table[index].driver_data;
+	unsigned long reg, cr;
+
+	local_irq_disable();
+
+	/* Mask IRQ and FIQ to CPU */
+	cr = readl(priv.pmu_cr);
+	cr |= MASK_IRQ | MASK_FIQ;
+	writel(cr, priv.pmu_cr);
+
+	/* Set/Clear the CPU_SLOW_EN bit */
+	reg = readl_relaxed(priv.dfs + DFS_CR);
+	reg &= ~L2_RATIO_MASK;
+
+	switch (state) {
+	case STATE_CPU_FREQ:
+		reg |= priv.xpratio;
+		reg &= ~CPU_SLOW_EN;
+		break;
+	case STATE_DDR_FREQ:
+		reg |= (priv.dpratio | CPU_SLOW_EN);
+		break;
+	}

drivers/cpufreq/dove-cpufreq.c line 116

Quoted 10 lines
drivers/cpufreq/dove-cpufreq.c new file +10 -0
+
+	/* Start the DFS process */
+	reg |= DFS_EN;
+
+	writel(reg, priv.dfs + DFS_CR);
+
+	/* Wait-for-Interrupt, while the hardware changes frequency */
+	cpu_do_idle();
+
+	local_irq_enable();

Just to check -- does the PMU automatically unmask IRQ and FIQ?

[...]

Quoted 16 lines
drivers/cpufreq/dove-cpufreq.c new file +16 -0
+static int dove_cpufreq_probe(struct platform_device *pdev)
+{
+	struct device *cpu_dev;
+	struct resource *res;
+	int err, irq;
+
+	memset(&priv, 0, sizeof(priv));
+	priv.dev = &pdev->dev;
+
+	res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
+					   "cpufreq: DFS");
+	priv.dfs = devm_ioremap_resource(&pdev->dev, res);
+	if (IS_ERR(priv.dfs))
+		return PTR_ERR(priv.dfs);
+
+	res = platform_get_resource_byname(pdev, IORESOURCE_MEM,

drivers/cpufreq/dove-cpufreq.c line 171

Quoted 10 lines
drivers/cpufreq/dove-cpufreq.c new file +10 -0
+					   "cpufreq: PMU CR");
+	priv.pmu_cr = devm_ioremap_resource(&pdev->dev, res);
+	if (IS_ERR(priv.pmu_cr))
+		return PTR_ERR(priv.pmu_cr);
+
+	res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
+					   "cpufreq: PMU Clk Div");
+	priv.pmu_clk_div = devm_ioremap_resource(&pdev->dev, res);
+	if (IS_ERR(priv.pmu_clk_div))
+		return PTR_ERR(priv.pmu_clk_div);

I'd very much prefer that the PMU were described in the DT, and the driver probed based on that.

drivers/cpufreq/dove-cpufreq.c line 179

Quoted 8 lines
drivers/cpufreq/dove-cpufreq.c new file +8 -0
+
+	cpu_dev = get_cpu_device(0);
+
+	priv.cpu_clk = devm_clk_get(cpu_dev, "cpu_clk");
+	if (IS_ERR(priv.cpu_clk)) {
+		err = PTR_ERR(priv.cpu_clk);
+		goto out;
+	}

I think it would be better to describe the clocks as being fed to the PMU than directly to the CPU -- the PMU selects the appropriate clock and feeds it to the CPU.

Also, the clocks could be named better with respect to their logical function. One is the high-frequency default, and the other is a low-frequency option. They're both able to feed the CPU, and the fact that one also feeds the DDR isn't relevant to the CPU. Does the PMu also control the input to the DDR?

Quoted 14 lines
drivers/cpufreq/dove-cpufreq.c new file +14 -0
+
+	err = clk_prepare_enable(priv.cpu_clk);
+	if (err)
+		goto out;
+
+	dove_freq_table[0].frequency = clk_get_rate(priv.cpu_clk) / 1000;
+
+	priv.ddr_clk = devm_clk_get(cpu_dev, "ddrclk");
+	if (IS_ERR(priv.ddr_clk)) {
+		err = PTR_ERR(priv.ddr_clk);
+		goto out;
+	}
+
+	err = clk_prepare_enable(priv.ddr_clk);

drivers/cpufreq/dove-cpufreq.c line 203

Quoted 10 lines
drivers/cpufreq/dove-cpufreq.c new file +10 -0
+	if (err)
+		goto out;
+
+	dove_freq_table[1].frequency = clk_get_rate(priv.ddr_clk) / 1000;
+
+	irq = irq_of_parse_and_map(cpu_dev->of_node, 0);
+	if (!irq) {
+		err = -ENXIO;
+		goto out;
+	}

Similarly, the interrupt is a property of the PMU.

Thanks,
Mark.

author maintainer
#6 lore raw

PATCH v4 1/4 cpufreq: Add a cpufreq driver for Marvell Dove

Quoted 30 lines
Quoted 30 lines
drivers/cpufreq/dove-cpufreq.c new file +30 -0
+static int dove_cpufreq_target(struct cpufreq_policy *policy,
+			       unsigned int index)
+{
+	unsigned int state = dove_freq_table[index].driver_data;
+	unsigned long reg, cr;
+
+	local_irq_disable();
+
+	/* Mask IRQ and FIQ to CPU */
+	cr = readl(priv.pmu_cr);
+	cr |= MASK_IRQ | MASK_FIQ;
+	writel(cr, priv.pmu_cr);
+
+	/* Set/Clear the CPU_SLOW_EN bit */
+	reg = readl_relaxed(priv.dfs + DFS_CR);
+	reg &= ~L2_RATIO_MASK;
+
+	switch (state) {
+	case STATE_CPU_FREQ:
+		reg |= priv.xpratio;
+		reg &= ~CPU_SLOW_EN;
+		break;
+	case STATE_DDR_FREQ:
+		reg |= (priv.dpratio | CPU_SLOW_EN);
+		break;
+	}
+
+	/* Start the DFS process */
+	reg |= DFS_EN;
+
Quoted 8 lines

drivers/cpufreq/dove-cpufreq.c line 116

Quoted 6 lines
drivers/cpufreq/dove-cpufreq.c new file +6 -0
+	writel(reg, priv.dfs + DFS_CR);
+
+	/* Wait-for-Interrupt, while the hardware changes frequency */
+	cpu_do_idle();
+
+	local_irq_enable();

Just to check -- does the PMU automatically unmask IRQ and FIQ?

The data sheet says so, yes.

Quoted 21 lines
Quoted 21 lines
drivers/cpufreq/dove-cpufreq.c new file +21 -0
+static int dove_cpufreq_probe(struct platform_device *pdev)
+{
+	struct device *cpu_dev;
+	struct resource *res;
+	int err, irq;
+
+	memset(&priv, 0, sizeof(priv));
+	priv.dev = &pdev->dev;
+
+	res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
+					   "cpufreq: DFS");
+	priv.dfs = devm_ioremap_resource(&pdev->dev, res);
+	if (IS_ERR(priv.dfs))
+		return PTR_ERR(priv.dfs);
+
+	res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
+					   "cpufreq: PMU CR");
+	priv.pmu_cr = devm_ioremap_resource(&pdev->dev, res);
+	if (IS_ERR(priv.pmu_cr))
+		return PTR_ERR(priv.pmu_cr);
+
Quoted 8 lines

drivers/cpufreq/dove-cpufreq.c line 171

Quoted 5 lines
drivers/cpufreq/dove-cpufreq.c new file +5 -0
+	res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
+					   "cpufreq: PMU Clk Div");
+	priv.pmu_clk_div = devm_ioremap_resource(&pdev->dev, res);
+	if (IS_ERR(priv.pmu_clk_div))
+		return PTR_ERR(priv.pmu_clk_div);

I'd very much prefer that the PMU were described in the DT, and the driver probed based on that.

Please see my other response. We can discuss that there.

Quoted 4 lines
Quoted 4 lines
drivers/cpufreq/dove-cpufreq.c new file +4 -0
+
+	cpu_dev = get_cpu_device(0);
+
+	priv.cpu_clk = devm_clk_get(cpu_dev, "cpu_clk");
Quoted 8 lines

drivers/cpufreq/dove-cpufreq.c line 179

Quoted 4 lines
drivers/cpufreq/dove-cpufreq.c new file +4 -0
+	if (IS_ERR(priv.cpu_clk)) {
+		err = PTR_ERR(priv.cpu_clk);
+		goto out;
+	}

I think it would be better to describe the clocks as being fed to the PMU than directly to the CPU -- the PMU selects the appropriate clock and feeds it to the CPU.

The wording in the data sheet is not very clear:

The PMU initiates new clock ratio values for the CPU subsystem clocks generation unit.

So to me, the PMU is not the clock consumer, it just controls the clock generation unit. There is also a block diagram which shows the "CPU Subsystem ClockGen" as being external to the PMU.

Quoted 4 lines

Also, the clocks could be named better with respect to their logical function. One is the high-frequency default, and the other is a low-frequency option. They're both able to feed the CPU, and the fact that one also feeds the DDR isn't relevant to the CPU.

The data sheet uses the terms Turbo Speed and Slow Speed modes. So i could call them turbo and slow? However, the current names follow the kirkwood cpufreq driver. It also can swap between a fast clock and the DDR clock. So i thought keeping the drivers similar would help with the maintenance burden.

Andrew

#7 lore raw

PATCH v4 1/4 cpufreq: Add a cpufreq driver for Marvell Dove

On Wed, Dec 04, 2013 at 03:27:26PM +0000, Andrew Lunn wrote:

[...]

Quoted 13 lines
Quoted 4 lines
Quoted 4 lines
drivers/cpufreq/dove-cpufreq.c new file +4 -0
+
+	cpu_dev = get_cpu_device(0);
+
+	priv.cpu_clk = devm_clk_get(cpu_dev, "cpu_clk");
Quoted 8 lines

drivers/cpufreq/dove-cpufreq.c line 179

Quoted 4 lines
drivers/cpufreq/dove-cpufreq.c new file +4 -0
+	if (IS_ERR(priv.cpu_clk)) {
+		err = PTR_ERR(priv.cpu_clk);
+		goto out;
+	}

I think it would be better to describe the clocks as being fed to the PMU than directly to the CPU -- the PMU selects the appropriate clock and feeds it to the CPU.

Quoted 8 lines

The wording in the data sheet is not very clear:

The PMU initiates new clock ratio values for the CPU subsystem clocks generation unit.

So to me, the PMU is not the clock consumer, it just controls the clock generation unit. There is also a block diagram which shows the "CPU Subsystem ClockGen" as being external to the PMU.

Hmm. That's somewhat arguable.

Quoted 1 line

From a block diagram I've found, the situation seems more like the

ClockGen takes two clocks as input, and feeds the appropriate clock to the CPU. Logically, the CPU itself is a consumer of one clock.

The ClockGen unit is controlled by a clock manager. The clock manager appears to be a component of the PMU itself, but I may have misunderstood. If that is the case, and the CPU Subsytem ClockGen is not controlled by anything else, then I don't see the problem with grouping that together logically with the clock manager (and therefore the PMU). That would imply describing the clock inputs on the PMU is fine.

Have I missed something?

One thing I'm worried about is adding lots of SoC-specific variations on data to generic nodes rather than doing that in a uniform fashion. I believe describing the clocks in the way you propose clashes with the cpufreq-cpu0 way of doing things (whereby the CPU has one clock input).

If the ClockGen were modelled as a mux that allows switching between two rates, then that would make the description more uniform.

Quoted 11 lines
Quoted 5 lines

Also, the clocks could be named better with respect to their logical function. One is the high-frequency default, and the other is a low-frequency option. They're both able to feed the CPU, and the fact that one also feeds the DDR isn't relevant to the CPU.

The data sheet uses the terms Turbo Speed and Slow Speed modes. So i could call them turbo and slow? However, the current names follow the kirkwood cpufreq driver. It also can swap between a fast clock and the DDR clock. So i thought keeping the drivers similar would help with the maintenance burden.

I think given the above, the changes I'm asking for would already involve sufficient changes to make this concern irrelevant. However, I don't want to force an unmaintainable change.

Mark.

author maintainer
#8 lore raw

v5 cover Dove Cpufreq driver

In reply to Mark Rutland

This is an RFC where i attempt to address Mark Rutland's request for a PMU node in DT, and all clocks and interrupts required by the cpufreq driver are within this node. I would like to receive comments about this binding.

This version does not address any other comments made to the code. I will fix those issues once the binding is mostly decided.

Thanks
Andrew

Andrew Lunn (4):
cpufreq: Add a cpufreq driver for Marvell Dove
mvebu: Dove: Indicate the architecture supports cpufreq.
mvebu: Dove: Enable cpufreq driver in defconfig
mvebu: Dove: Add PMU node for cpufreq driver

 .../devicetree/bindings/cpufreq/cpufreq-dove.txt   |  23 ++
 arch/arm/Kconfig                                   |   1 +
 arch/arm/boot/dts/dove.dtsi                        |  10 +
 arch/arm/configs/dove_defconfig                    |   2 +
 drivers/cpufreq/Kconfig.arm                        |   5 +
 drivers/cpufreq/Makefile                           |   1 +
 drivers/cpufreq/dove-cpufreq.c                     | 265 +++++++++++++++++++++
 7 files changed, 307 insertions(+)
 create mode 100644 Documentation/devicetree/bindings/cpufreq/cpufreq-dove.txt
 create mode 100644 drivers/cpufreq/dove-cpufreq.c
Signature · 1 line
1.8.5.1
author maintainer
#9 lore raw

RFC PATCH v5 1/4 cpufreq: Add a cpufreq driver for Marvell Dove

The Marvell Dove SoC can run the CPU at two frequencies. The high frequencey is from a PLL, while the lower is the same as the DDR clock. Add a cpufreq driver to swap between these frequences.

---
Sort header files
Remove unneeded header files
Notify only on change
Use get_cpu_device(0), devm_clk_get()
Use platform_get_resource_byname()
Error check clk_prepare_enable()
Comment the interrupt handler

rebase onto 3.13-rc2 linux-next
use target_index

---
 .../devicetree/bindings/cpufreq/cpufreq-dove.txt   |  23 ++
 drivers/cpufreq/Kconfig.arm                        |   5 +
 drivers/cpufreq/Makefile                           |   1 +
 drivers/cpufreq/dove-cpufreq.c                     | 265 +++++++++++++++++++++
 4 files changed, 294 insertions(+)
 create mode 100644 Documentation/devicetree/bindings/cpufreq/cpufreq-dove.txt
 create mode 100644 drivers/cpufreq/dove-cpufreq.c
Documentation/devicetree/bindings/cpufreq/cpufreq-dove.txt new file +23 -0
new file mode 100644
index 000000000000..f95cc2234fc1
@@ -0,0 +1,23 @@
+Dove cpufreq driver
+-------------------
+
+The Dove cpufreq driver makes use of the pmu node in the device tree.
+
+Required properties:
+- compatibility: Must be "marvell,dove-pmu"
+- interrupts:	 Interrupt to know the completion of cpu frequency change.
+- clocks:	 phandles and clock specifiers pairs for CPU and DDR clock
+- clock-names: 	 "cpu" and "ddr"
+- reg: 		 Address ranges of the PMU. Two ranges are expected.
+
+Example:
+
+	pmu: pmu at d0000 {
+		compatible = "marvell,dove-pmu";
+		reg = <0xd0000 0x0700>,
+		      <0xd8000 0x0140>;
+		clocks = <&core_clk 1>, <&core_clk 3>;
+		clock-names = "cpu", "ddr";
+		interrupt-parent = <&pmu_intc>;
+		interrupts = <0>;
+	};
drivers/cpufreq/Kconfig.arm +5 -0
index ce52ed949249..af7c4947f218 100644
@@ -8,6 +8,11 @@ config ARM_BIG_LITTLE_CPUFREQ
 	help
 	  This enables the Generic CPUfreq driver for ARM big.LITTLE platforms.
 
+config ARM_DOVE_CPUFREQ
+	def_bool ARCH_DOVE && OF
+	help
+	  This adds the CPUFreq driver for Marvell Dove SoCs.
+
 config ARM_DT_BL_CPUFREQ
 	tristate "Generic probing via DT for ARM big LITTLE CPUfreq driver"
 	depends on ARM_BIG_LITTLE_CPUFREQ && OF
drivers/cpufreq/Makefile +1 -0
index 74945652dd7a..cd72c2bbfd44 100644
@@ -49,6 +49,7 @@ obj-$(CONFIG_ARM_DT_BL_CPUFREQ)		+= arm_big_little_dt.o
 
 obj-$(CONFIG_ARCH_DAVINCI_DA850)	+= davinci-cpufreq.o
 obj-$(CONFIG_UX500_SOC_DB8500)		+= dbx500-cpufreq.o
+obj-$(CONFIG_ARM_DOVE_CPUFREQ)		+= dove-cpufreq.o
 obj-$(CONFIG_ARM_EXYNOS_CPUFREQ)	+= exynos-cpufreq.o
 obj-$(CONFIG_ARM_EXYNOS4210_CPUFREQ)	+= exynos4210-cpufreq.o
 obj-$(CONFIG_ARM_EXYNOS4X12_CPUFREQ)	+= exynos4x12-cpufreq.o
drivers/cpufreq/dove-cpufreq.c new file +265 -0

265 lines not on this page · raw message

Signature · 1 line
1.8.5.1
#10 lore raw

RFC PATCH v5 1/4 cpufreq: Add a cpufreq driver for Marvell Dove

Quoted 33 lines · Andrew Lunn

The Marvell Dove SoC can run the CPU at two frequencies. The high frequencey is from a PLL, while the lower is the same as the DDR clock. Add a cpufreq driver to swap between these frequences.

---
Sort header files
Remove unneeded header files
Notify only on change
Use get_cpu_device(0), devm_clk_get()
Use platform_get_resource_byname()
Error check clk_prepare_enable()
Comment the interrupt handler

rebase onto 3.13-rc2 linux-next
use target_index

---
 .../devicetree/bindings/cpufreq/cpufreq-dove.txt   |  23 ++
 drivers/cpufreq/Kconfig.arm                        |   5 +
 drivers/cpufreq/Makefile                           |   1 +
 drivers/cpufreq/dove-cpufreq.c                     | 265 +++++++++++++++++++++
 4 files changed, 294 insertions(+)
 create mode 100644 Documentation/devicetree/bindings/cpufreq/cpufreq-dove.txt
 create mode 100644 drivers/cpufreq/dove-cpufreq.c
Documentation/devicetree/bindings/cpufreq/cpufreq-dove.txt new file +0 -0
new file mode 100644
index 000000000000..f95cc2234fc1

Documentation/devicetree/bindings/cpufreq/cpufreq-dove.txt line 4

Quoted 4 lines
Documentation/devicetree/bindings/cpufreq/cpufreq-dove.txt new file +4 -0
@@ -0,0 +1,23 @@
+Dove cpufreq driver
+-------------------
+
+The Dove cpufreq driver makes use of the pmu node in the device tree.

I would prefer to reword this to make it OS-agnostic. The cpufreq driver is Linux-specific.

Documentation/devicetree/bindings/cpufreq/cpufreq-dove.txt line 11

Quoted 7 lines
Documentation/devicetree/bindings/cpufreq/cpufreq-dove.txt new file +7 -0
+
+Required properties:
+- compatibility: Must be "marvell,dove-pmu"
+- interrupts:	 Interrupt to know the completion of cpu frequency change.
+- clocks:	 phandles and clock specifiers pairs for CPU and DDR clock
+- clock-names: 	 "cpu" and "ddr"
+- reg: 		 Address ranges of the PMU. Two ranges are expected.

Shall we define the two ranges?

Documentation/devicetree/bindings/cpufreq/cpufreq-dove.txt line 23

Quoted 12 lines
Documentation/devicetree/bindings/cpufreq/cpufreq-dove.txt new file +12 -0
+
+Example:
+
+	pmu: pmu at d0000 {
+		compatible = "marvell,dove-pmu";
+		reg = <0xd0000 0x0700>,
+		      <0xd8000 0x0140>;
+		clocks = <&core_clk 1>, <&core_clk 3>;
+		clock-names = "cpu", "ddr";
+		interrupt-parent = <&pmu_intc>;
+		interrupts = <0>;
+	};

thx,

Jason.

author maintainer
#11 lore raw

RFC PATCH v5 1/4 cpufreq: Add a cpufreq driver for Marvell Dove

Quoted 7 lines

Documentation/devicetree/bindings/cpufreq/cpufreq-dove.txt line 4

Quoted 4 lines
Documentation/devicetree/bindings/cpufreq/cpufreq-dove.txt new file +4 -0
+Dove cpufreq driver
+-------------------
+
+The Dove cpufreq driver makes use of the pmu node in the device tree.

I would prefer to reword this to make it OS-agnostic. The cpufreq driver is Linux-specific.

Hi Jason

Why is it Linux specific? Why for example would the FreeBSD cpu freq driver not be able to use this?

Quoted 9 lines

Documentation/devicetree/bindings/cpufreq/cpufreq-dove.txt line 11

Quoted 7 lines
Documentation/devicetree/bindings/cpufreq/cpufreq-dove.txt new file +7 -0
+
+Required properties:
+- compatibility: Must be "marvell,dove-pmu"
+- interrupts:	 Interrupt to know the completion of cpu frequency change.
+- clocks:	 phandles and clock specifiers pairs for CPU and DDR clock
+- clock-names: 	 "cpu" and "ddr"
+- reg: 		 Address ranges of the PMU. Two ranges are expected.

Shall we define the two ranges?

Yes, i will extend this, once Mark agrees to the basic idea.

Thanks
Andrew

#12 lore raw

RFC PATCH v5 1/4 cpufreq: Add a cpufreq driver for Marvell Dove

Andrew,

Quoted 9 lines · Andrew Lunn
Quoted 7 lines

Documentation/devicetree/bindings/cpufreq/cpufreq-dove.txt line 4

Quoted 4 lines
Documentation/devicetree/bindings/cpufreq/cpufreq-dove.txt new file +4 -0
+Dove cpufreq driver
+-------------------
+
+The Dove cpufreq driver makes use of the pmu node in the device tree.

I would prefer to reword this to make it OS-agnostic. The cpufreq driver is Linux-specific.

Why is it Linux specific?

I'll admit I may be bike-shedding here, and I'm more seeking opinion/clarification than mandating anything. The above sentence, imho, is describing how Linux/FreeBSD uses the node, not describing the hardware.

thx,

Jason.

author maintainer
#13 lore raw

RFC PATCH v5 1/4 cpufreq: Add a cpufreq driver for Marvell Dove

Quoted 11 lines · Jason Cooper

Andrew,

Quoted 8 lines · Andrew Lunn
Quoted 7 lines

Documentation/devicetree/bindings/cpufreq/cpufreq-dove.txt line 4

Quoted 4 lines
Documentation/devicetree/bindings/cpufreq/cpufreq-dove.txt new file +4 -0
+Dove cpufreq driver
+-------------------
+
+The Dove cpufreq driver makes use of the pmu node in the device tree.

I would prefer to reword this to make it OS-agnostic. The cpufreq driver is Linux-specific.

Quoted 6 lines
Quoted 1 line

Why is it Linux specific?

I'll admit I may be bike-shedding here, and I'm more seeking opinion/clarification than mandating anything. The above sentence, imho, is describing how Linux/FreeBSD uses the node, not describing the hardware.

How about:

Dove has a PMU, which contains DFS, DVS, RTC, clock gates, and thermal management,

Andrew

#14 lore raw

RFC PATCH v5 1/4 cpufreq: Add a cpufreq driver for Marvell Dove

Quoted 19 lines · Andrew Lunn
Quoted 11 lines · Jason Cooper

Andrew,

Quoted 8 lines · Andrew Lunn
Quoted 7 lines

Documentation/devicetree/bindings/cpufreq/cpufreq-dove.txt line 4

Quoted 4 lines
Documentation/devicetree/bindings/cpufreq/cpufreq-dove.txt new file +4 -0
+Dove cpufreq driver
+-------------------
+
+The Dove cpufreq driver makes use of the pmu node in the device tree.

I would prefer to reword this to make it OS-agnostic. The cpufreq driver is Linux-specific.

Quoted 6 lines
Quoted 1 line

Why is it Linux specific?

I'll admit I may be bike-shedding here, and I'm more seeking opinion/clarification than mandating anything. The above sentence, imho, is describing how Linux/FreeBSD uses the node, not describing the hardware.

Quoted 4 lines

How about:

Dove has a PMU, which contains DFS, DVS, RTC, clock gates, and thermal management,

Yep, much better.

thx,

Jason.

author maintainer
#15 lore raw

RFC PATCH v5 2/4 mvebu: Dove: Indicate the architecture supports cpufreq.

In reply to Andrew Lunn

Enable ARCH_HAS_CPUFREQ so allowing the cpufreq framework to be enabled.

---
 arch/arm/Kconfig | 1 +
 1 file changed, 1 insertion(+)
arch/arm/Kconfig +1 -0
index c1f1a7eee953..908b02f8d901 100644
@@ -500,6 +500,7 @@ config ARCH_IXP4XX
 
 config ARCH_DOVE
 	bool "Marvell Dove"
+	select ARCH_HAS_CPUFREQ
 	select ARCH_REQUIRE_GPIOLIB
 	select CPU_PJ4
 	select GENERIC_CLOCKEVENTS
Signature · 1 line
1.8.5.1
author maintainer
#16 lore raw

RFC PATCH v5 3/4 mvebu: Dove: Enable cpufreq driver in defconfig

In reply to Andrew Lunn

Add cpufreq to dove_defconfig, so it is built by default.

---
 arch/arm/configs/dove_defconfig | 2 ++
 1 file changed, 2 insertions(+)
arch/arm/configs/dove_defconfig +2 -0
index 110105476848..3d9235017e89 100644
@@ -22,6 +22,8 @@ CONFIG_ZBOOT_ROM_TEXT=0x0
 CONFIG_ZBOOT_ROM_BSS=0x0
 CONFIG_ARM_APPENDED_DTB=y
 CONFIG_ARM_ATAG_DTB_COMPAT=y
+CONFIG_CPU_FREQ=y
+CONFIG_CPU_FREQ_GOV_ONDEMAND=y
 CONFIG_VFP=y
 CONFIG_NET=y
 CONFIG_PACKET=y
Signature · 1 line
1.8.5.1
author maintainer
#17 lore raw

RFC PATCH v5 4/4 mvebu: Dove: Add PMU node for cpufreq driver

In reply to Andrew Lunn

The cpufreq driver uses the PMU node in the device tree, since it is part of the PMU device. The dove-cpufreq driver needs access to the DDR and CPU clock, which are the two clocks the CPU can use. There is also an interrupt generated when the DFS hardware completes a change of frequencey. Add these to the DT.

---
 arch/arm/boot/dts/dove.dtsi | 10 ++++++++++
 1 file changed, 10 insertions(+)
arch/arm/boot/dts/dove.dtsi +10 -0
index 4c8028513133..42b9810e96c1 100644
@@ -136,6 +136,16 @@
 				marvell,#interrupts = <5>;
 			};
 
+			pmu: pmu at d0000 {
+				compatible = "marvell,dove-pmu";
+				reg = <0xd0000 0x0700>,
+				      <0xd8000 0x0140>;
+				clocks = <&core_clk 1>, <&core_clk 3>;
+				clock-names = "cpu", "ddr";
+				interrupt-parent = <&pmu_intc>;
+				interrupts = <0>;
+			};
+
 			pmu_intc: pmu-interrupt-ctrl at d0050 {
 				compatible = "marvell,dove-pmu-intc";
 				interrupt-controller;
Signature · 1 line
1.8.5.1
maintainer
#18 lore raw

PATCH v4 1/4 cpufreq: Add a cpufreq driver for Marvell Dove

In reply to Andrew Lunn
Quoted 7 lines · Andrew Lunn

The Marvell Dove SoC can run the CPU at two frequencies. The high frequencey is from a PLL, while the lower is the same as the DDR clock. Add a cpufreq driver to swap between these frequences.

I was sure about it and checked it again. I haven't Acked this patch until now. :)

drivers/cpufreq/Kconfig.arm config ARM_BIG_LITTLE_CPUFREQ · line 14

Quoted 12 lines
drivers/cpufreq/Kconfig.arm +4 -0
index ce52ed949249..af7c4947f218 100644
@@ -8,6 +8,11 @@ config ARM_BIG_LITTLE_CPUFREQ
        help
          This enables the Generic CPUfreq driver for ARM big.LITTLE platforms.
 
+config ARM_DOVE_CPUFREQ
+       def_bool ARCH_DOVE && OF
+       help
+         This adds the CPUFreq driver for Marvell Dove SoCs.

Please add this one after DT_BL one.. I know you wanted to keep them in alphabetical order (probably we need to rename DT_BL), but keeping both BIG LITTLE options makes more sense.

I will do the renaming later.

Quoted 29 lines
drivers/cpufreq/Kconfig.arm +0 -0
 config ARM_DT_BL_CPUFREQ
        tristate "Generic probing via DT for ARM big LITTLE CPUfreq driver"
        depends on ARM_BIG_LITTLE_CPUFREQ && OF
drivers/cpufreq/dove-cpufreq.c +24 -0
+static int dove_cpufreq_probe(struct platform_device *pdev)
+{
+       struct device *cpu_dev;
+       struct resource *res;
+       int err, irq;
+
+       memset(&priv, 0, sizeof(priv));
+       priv.dev = &pdev->dev;
+
+       res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
+                                          "cpufreq: DFS");
+       priv.dfs = devm_ioremap_resource(&pdev->dev, res);
+       if (IS_ERR(priv.dfs))
+               return PTR_ERR(priv.dfs);
+
+       res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
+                                          "cpufreq: PMU CR");
+       priv.pmu_cr = devm_ioremap_resource(&pdev->dev, res);
+       if (IS_ERR(priv.pmu_cr))
+               return PTR_ERR(priv.pmu_cr);
+
+       res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
+                                          "cpufreq: PMU Clk Div");
+       priv.pmu_clk_div = devm_ioremap_resource(&pdev->dev, res);

drivers/cpufreq/dove-cpufreq.c line 179

Quoted 10 lines
drivers/cpufreq/dove-cpufreq.c +10 -0
+       if (IS_ERR(priv.pmu_clk_div))
+               return PTR_ERR(priv.pmu_clk_div);
+
+       cpu_dev = get_cpu_device(0);
+
+       priv.cpu_clk = devm_clk_get(cpu_dev, "cpu_clk");
+       if (IS_ERR(priv.cpu_clk)) {
+               err = PTR_ERR(priv.cpu_clk);
+               goto out;
+       }

You are doing clk_disable_unprepare in out, and you haven't enabled them yet. So, just return error from here.

drivers/cpufreq/dove-cpufreq.c line 183

Quoted 3 lines
drivers/cpufreq/dove-cpufreq.c +3 -0
+       err = clk_prepare_enable(priv.cpu_clk);
+       if (err)
+               goto out;

same here

drivers/cpufreq/dove-cpufreq.c line 190

Quoted 6 lines
drivers/cpufreq/dove-cpufreq.c +6 -0
+       dove_freq_table[0].frequency = clk_get_rate(priv.cpu_clk) / 1000;
+
+       priv.ddr_clk = devm_clk_get(cpu_dev, "ddrclk");
+       if (IS_ERR(priv.ddr_clk)) {
+               err = PTR_ERR(priv.ddr_clk);
+               goto out;

You need to do unprepare only for cpu clk here.

drivers/cpufreq/dove-cpufreq.c line 195

Quoted 5 lines
drivers/cpufreq/dove-cpufreq.c +5 -0
+       }
+
+       err = clk_prepare_enable(priv.ddr_clk);
+       if (err)
+               goto out;

same here

drivers/cpufreq/dove-cpufreq.c

Quoted 2 lines
drivers/cpufreq/dove-cpufreq.c +2 -0
+
+       dove_freq_table[1].frequency = clk_get_rate(priv.ddr_clk) / 1000;

Because you are using 0 and 1 in index for CPU and DDR freq at multiple places, probably makes sense to make Macros for them ?

Quoted 47 lines
drivers/cpufreq/dove-cpufreq.c +47 -0
+
+       irq = irq_of_parse_and_map(cpu_dev->of_node, 0);
+       if (!irq) {
+               err = -ENXIO;
+               goto out;
+       }
+
+       err = devm_request_irq(&pdev->dev, irq, dove_cpufreq_irq,
+                              0, "dove-cpufreq", NULL);
+       if (err) {
+               dev_err(&pdev->dev, "cannot assign irq %d, %d\n", irq, err);
+               goto out;
+       }
+
+       /* Read the target ratio which should be the DDR ratio */
+       priv.dpratio = readl_relaxed(priv.pmu_clk_div);
+       priv.dpratio = (priv.dpratio & DPRATIO_MASK) >> DPRATIO_OFFS;
+       priv.dpratio = priv.dpratio << L2_RATIO_OFFS;
+
+       /* Save L2 ratio at reset */
+       priv.xpratio = readl(priv.pmu_clk_div);
+       priv.xpratio = (priv.xpratio & XPRATIO_MASK) >> XPRATIO_OFFS;
+       priv.xpratio = priv.xpratio << L2_RATIO_OFFS;
+
+       err = cpufreq_register_driver(&dove_cpufreq_driver);
+       if (!err)
+               return 0;
+
+       dev_err(priv.dev, "Failed to register cpufreq driver");
+
+out:
+       clk_disable_unprepare(priv.ddr_clk);
+       clk_disable_unprepare(priv.cpu_clk);
+
+       return err;
+}
+
+static int dove_cpufreq_remove(struct platform_device *pdev)
+{
+       cpufreq_unregister_driver(&dove_cpufreq_driver);
+
+       clk_disable_unprepare(priv.ddr_clk);
+       clk_disable_unprepare(priv.cpu_clk);
+
+       return 0;
+}
+

drivers/cpufreq/dove-cpufreq.c line 254

Quoted 10 lines
drivers/cpufreq/dove-cpufreq.c +10 -0
+static struct platform_driver dove_cpufreq_platform_driver = {
+       .probe = dove_cpufreq_probe,
+       .remove = dove_cpufreq_remove,
+       .driver = {
+               .name = "dove-cpufreq",
+               .owner = THIS_MODULE,
+       },
+};
+
+module_platform_driver(dove_cpufreq_platform_driver);

Otherwise looks fine :)

author maintainer
#19 lore raw

PATCH v4 1/4 cpufreq: Add a cpufreq driver for Marvell Dove

Quoted 17 lines

drivers/cpufreq/Kconfig.arm config ARM_BIG_LITTLE_CPUFREQ · line 14

Quoted 12 lines
drivers/cpufreq/Kconfig.arm +4 -0
index ce52ed949249..af7c4947f218 100644
@@ -8,6 +8,11 @@ config ARM_BIG_LITTLE_CPUFREQ
        help
          This enables the Generic CPUfreq driver for ARM big.LITTLE platforms.
 
+config ARM_DOVE_CPUFREQ
+       def_bool ARCH_DOVE && OF
+       help
+         This adds the CPUFreq driver for Marvell Dove SoCs.

Please add this one after DT_BL one.. I know you wanted to keep them in alphabetical order (probably we need to rename DT_BL), but keeping both BIG LITTLE options makes more sense.

Quoted 1 line

I will do the renaming later.

O.K, once the binding is agreed, i will make this change.

Quoted 34 lines
Quoted 34 lines
drivers/cpufreq/Kconfig.arm +0 -0
 config ARM_DT_BL_CPUFREQ
        tristate "Generic probing via DT for ARM big LITTLE CPUfreq driver"
        depends on ARM_BIG_LITTLE_CPUFREQ && OF
drivers/cpufreq/dove-cpufreq.c +29 -0
+static int dove_cpufreq_probe(struct platform_device *pdev)
+{
+       struct device *cpu_dev;
+       struct resource *res;
+       int err, irq;
+
+       memset(&priv, 0, sizeof(priv));
+       priv.dev = &pdev->dev;
+
+       res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
+                                          "cpufreq: DFS");
+       priv.dfs = devm_ioremap_resource(&pdev->dev, res);
+       if (IS_ERR(priv.dfs))
+               return PTR_ERR(priv.dfs);
+
+       res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
+                                          "cpufreq: PMU CR");
+       priv.pmu_cr = devm_ioremap_resource(&pdev->dev, res);
+       if (IS_ERR(priv.pmu_cr))
+               return PTR_ERR(priv.pmu_cr);
+
+       res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
+                                          "cpufreq: PMU Clk Div");
+       priv.pmu_clk_div = devm_ioremap_resource(&pdev->dev, res);
+       if (IS_ERR(priv.pmu_clk_div))
+               return PTR_ERR(priv.pmu_clk_div);
+
+       cpu_dev = get_cpu_device(0);
+
Quoted 8 lines

drivers/cpufreq/dove-cpufreq.c line 179

Quoted 5 lines
drivers/cpufreq/dove-cpufreq.c +5 -0
+       priv.cpu_clk = devm_clk_get(cpu_dev, "cpu_clk");
+       if (IS_ERR(priv.cpu_clk)) {
+               err = PTR_ERR(priv.cpu_clk);
+               goto out;
+       }

You are doing clk_disable_unprepare in out, and you haven't enabled them yet. So, just return error from here.

You are missing part of the clk API, which makes this work, and keeps the code simple. NULL is a valid clock, and enable/disable and prepare/unprepare are NOP with a NULL clock. So the code after out:
should mostly do the right thing. However what i do have wrong is not checking for IS_ERR() before clk_disable_unprepare(). I will fix that.

Quoted 5 lines

drivers/cpufreq/dove-cpufreq.c

Quoted 2 lines
drivers/cpufreq/dove-cpufreq.c +2 -0
+
+       dove_freq_table[1].frequency = clk_get_rate(priv.ddr_clk) / 1000;

Because you are using 0 and 1 in index for CPU and DDR freq at multiple places, probably makes sense to make Macros for them ?

O.K.

Thanks for the review,

Andrew

author maintainer
#20 lore raw

PATCH v4 2/4 mvebu: Dove: Instantiate cpufreq driver.

In reply to Andrew Lunn

Add a platform driver definition to instantiate the dove cpufreq driver. Also indicate the ARCH has cpufreq support, so allowing the cpufreq framework to be enabled.

---
Make resource names consistent.
Fix length of ranges

---
 arch/arm/Kconfig                       |  1 +
 arch/arm/mach-dove/board-dt.c          |  3 +++
 arch/arm/mach-dove/common.c            | 36 ++++++++++++++++++++++++++++++++++
 arch/arm/mach-dove/common.h            |  1 +
 arch/arm/mach-dove/include/mach/dove.h |  1 +
 5 files changed, 42 insertions(+)
arch/arm/Kconfig +1 -0
index c1f1a7eee953..908b02f8d901 100644
@@ -500,6 +500,7 @@ config ARCH_IXP4XX
 
 config ARCH_DOVE
 	bool "Marvell Dove"
+	select ARCH_HAS_CPUFREQ
 	select ARCH_REQUIRE_GPIOLIB
 	select CPU_PJ4
 	select GENERIC_CLOCKEVENTS
arch/arm/mach-dove/board-dt.c +3 -0
index 49fa9abd09da..845e8b328432 100644
@@ -27,6 +27,9 @@ static void __init dove_dt_init(void)
 	tauros2_init(0);
 #endif
 	BUG_ON(mvebu_mbus_dt_init());
+
+	dove_cpufreq_init();
+
 	of_platform_populate(NULL, of_default_bus_match_table, NULL, NULL);
 }
 
arch/arm/mach-dove/common.c +36 -0
index c122bcff9f7c..323fcd4b27b6 100644
@@ -344,6 +344,42 @@ void __init dove_sdio1_init(void)
 	platform_device_register(&dove_sdio1);
 }
 
+/*****************************************************************************
+ * CPU Frequency
+ ****************************************************************************/
+static struct resource dove_cpufreq_resources[] = {
+	[0] = {
+		.start  = DOVE_PMU_PHYS_BASE,
+		.end    = DOVE_PMU_PHYS_BASE + 0x7,
+		.flags  = IORESOURCE_MEM,
+		.name   = "cpufreq: DFS"
+	},
+	[1] = {
+		.start  = DOVE_PMU_PHYS_BASE + 0x8000,
+		.end    = DOVE_PMU_PHYS_BASE + 0x8003,
+		.flags  = IORESOURCE_MEM,
+		.name   = "cpufreq: PMU CR"
+	},
+	[2] = {
+		.start  = DOVE_PMU_PHYS_BASE + 0x0044,
+		.end    = DOVE_PMU_PHYS_BASE + 0x0047,
+		.flags  = IORESOURCE_MEM,
+		.name   = "cpufreq: PMU Clk Div"
+	},
+};
+
+static struct platform_device dove_cpufreq_device = {
+	.name		= "dove-cpufreq",
+	.id		= -1,
+	.num_resources	= ARRAY_SIZE(dove_cpufreq_resources),
+	.resource	= dove_cpufreq_resources,
+};
+
+void __init dove_cpufreq_init(void)
+{
+	platform_device_register(&dove_cpufreq_device);
+}
+
 void __init dove_setup_cpu_wins(void)
 {
 	/*
arch/arm/mach-dove/common.h +1 -0
index 1d725224d146..5c9a77bdd442 100644
@@ -44,6 +44,7 @@ void dove_spi1_init(void);
 void dove_i2c_init(void);
 void dove_sdio0_init(void);
 void dove_sdio1_init(void);
+void dove_cpufreq_init(void);
 void dove_restart(enum reboot_mode, const char *);
 
 #endif
arch/arm/mach-dove/include/mach/dove.h +1 -0
index 0c4b35f4ee5b..48db186ae6bc 100644
@@ -144,6 +144,7 @@
 #define  DOVE_SD0_GPIO_SEL		(1 << 0)
 
 /* Power Management */
+#define DOVE_PMU_PHYS_BASE	(DOVE_SB_REGS_PHYS_BASE + 0xd0000)
 #define DOVE_PMU_VIRT_BASE	(DOVE_SB_REGS_VIRT_BASE + 0xd0000)
 #define DOVE_PMU_SIG_CTRL	(DOVE_PMU_VIRT_BASE + 0x802c)
 
Signature · 1 line
1.8.5
#21 lore raw

PATCH v4 2/4 mvebu: Dove: Instantiate cpufreq driver.

Quoted 76 lines · Andrew Lunn

Add a platform driver definition to instantiate the dove cpufreq driver. Also indicate the ARCH has cpufreq support, so allowing the cpufreq framework to be enabled.

---
Make resource names consistent.
Fix length of ranges

---
 arch/arm/Kconfig                       |  1 +
 arch/arm/mach-dove/board-dt.c          |  3 +++
 arch/arm/mach-dove/common.c            | 36 ++++++++++++++++++++++++++++++++++
 arch/arm/mach-dove/common.h            |  1 +
 arch/arm/mach-dove/include/mach/dove.h |  1 +
 5 files changed, 42 insertions(+)
arch/arm/Kconfig +1 -0
index c1f1a7eee953..908b02f8d901 100644
@@ -500,6 +500,7 @@ config ARCH_IXP4XX
 
 config ARCH_DOVE
 	bool "Marvell Dove"
+	select ARCH_HAS_CPUFREQ
 	select ARCH_REQUIRE_GPIOLIB
 	select CPU_PJ4
 	select GENERIC_CLOCKEVENTS
arch/arm/mach-dove/board-dt.c +3 -0
index 49fa9abd09da..845e8b328432 100644
@@ -27,6 +27,9 @@ static void __init dove_dt_init(void)
 	tauros2_init(0);
 #endif
 	BUG_ON(mvebu_mbus_dt_init());
+
+	dove_cpufreq_init();
+
 	of_platform_populate(NULL, of_default_bus_match_table, NULL, NULL);
 }
 
arch/arm/mach-dove/common.c +25 -0
index c122bcff9f7c..323fcd4b27b6 100644
@@ -344,6 +344,42 @@ void __init dove_sdio1_init(void)
 	platform_device_register(&dove_sdio1);
 }
 
+/*****************************************************************************
+ * CPU Frequency
+ ****************************************************************************/
+static struct resource dove_cpufreq_resources[] = {
+	[0] = {
+		.start  = DOVE_PMU_PHYS_BASE,
+		.end    = DOVE_PMU_PHYS_BASE + 0x7,
+		.flags  = IORESOURCE_MEM,
+		.name   = "cpufreq: DFS"
+	},
+	[1] = {
+		.start  = DOVE_PMU_PHYS_BASE + 0x8000,
+		.end    = DOVE_PMU_PHYS_BASE + 0x8003,
+		.flags  = IORESOURCE_MEM,
+		.name   = "cpufreq: PMU CR"
+	},
+	[2] = {
+		.start  = DOVE_PMU_PHYS_BASE + 0x0044,
+		.end    = DOVE_PMU_PHYS_BASE + 0x0047,
+		.flags  = IORESOURCE_MEM,
+		.name   = "cpufreq: PMU Clk Div"
+	},
+};
+
+static struct platform_device dove_cpufreq_device = {

arch/arm/mach-dove/common.c void __init dove_sdio1_init(void) · line 381

Quoted 10 lines
arch/arm/mach-dove/common.c +10 -0
+	.name		= "dove-cpufreq",
+	.id		= -1,
+	.num_resources	= ARRAY_SIZE(dove_cpufreq_resources),
+	.resource	= dove_cpufreq_resources,
+};
+
+void __init dove_cpufreq_init(void)
+{
+	platform_device_register(&dove_cpufreq_device);
+}

Please describe your power management unit in the DT. Your hardcoding physical addresses that can easily be described in the DT, and you don't appear to have an early probing requirement.

Quoted 21 lines
arch/arm/mach-dove/common.c +1 -0
+
 void __init dove_setup_cpu_wins(void)
 {
 	/*
arch/arm/mach-dove/common.h +1 -0
index 1d725224d146..5c9a77bdd442 100644
@@ -44,6 +44,7 @@ void dove_spi1_init(void);
 void dove_i2c_init(void);
 void dove_sdio0_init(void);
 void dove_sdio1_init(void);
+void dove_cpufreq_init(void);
 void dove_restart(enum reboot_mode, const char *);
 
 #endif
arch/arm/mach-dove/include/mach/dove.h +0 -0
index 0c4b35f4ee5b..48db186ae6bc 100644

arch/arm/mach-dove/include/mach/dove.h line 149

Quoted 6 lines
arch/arm/mach-dove/include/mach/dove.h +1 -0
@@ -144,6 +144,7 @@
 #define  DOVE_SD0_GPIO_SEL		(1 << 0)
 
 /* Power Management */
+#define DOVE_PMU_PHYS_BASE	(DOVE_SB_REGS_PHYS_BASE + 0xd0000)
 #define DOVE_PMU_VIRT_BASE	(DOVE_SB_REGS_VIRT_BASE + 0xd0000)
 #define DOVE_PMU_SIG_CTRL	(DOVE_PMU_VIRT_BASE + 0x802c)

If you need to describe the large block this appears to be in, do that too.

Thanks,
Mark.

author maintainer
#22 lore raw

PATCH v4 2/4 mvebu: Dove: Instantiate cpufreq driver.

Quoted 83 lines · Mark Rutland
Quoted 82 lines · Andrew Lunn

Add a platform driver definition to instantiate the dove cpufreq driver. Also indicate the ARCH has cpufreq support, so allowing the cpufreq framework to be enabled.

---
Make resource names consistent.
Fix length of ranges

---
 arch/arm/Kconfig                       |  1 +
 arch/arm/mach-dove/board-dt.c          |  3 +++
 arch/arm/mach-dove/common.c            | 36 ++++++++++++++++++++++++++++++++++
 arch/arm/mach-dove/common.h            |  1 +
 arch/arm/mach-dove/include/mach/dove.h |  1 +
 5 files changed, 42 insertions(+)
arch/arm/Kconfig +1 -0
index c1f1a7eee953..908b02f8d901 100644
@@ -500,6 +500,7 @@ config ARCH_IXP4XX
 
 config ARCH_DOVE
 	bool "Marvell Dove"
+	select ARCH_HAS_CPUFREQ
 	select ARCH_REQUIRE_GPIOLIB
 	select CPU_PJ4
 	select GENERIC_CLOCKEVENTS
arch/arm/mach-dove/board-dt.c +3 -0
index 49fa9abd09da..845e8b328432 100644
@@ -27,6 +27,9 @@ static void __init dove_dt_init(void)
 	tauros2_init(0);
 #endif
 	BUG_ON(mvebu_mbus_dt_init());
+
+	dove_cpufreq_init();
+
 	of_platform_populate(NULL, of_default_bus_match_table, NULL, NULL);
 }
 
arch/arm/mach-dove/common.c +31 -0
index c122bcff9f7c..323fcd4b27b6 100644
@@ -344,6 +344,42 @@ void __init dove_sdio1_init(void)
 	platform_device_register(&dove_sdio1);
 }
 
+/*****************************************************************************
+ * CPU Frequency
+ ****************************************************************************/
+static struct resource dove_cpufreq_resources[] = {
+	[0] = {
+		.start  = DOVE_PMU_PHYS_BASE,
+		.end    = DOVE_PMU_PHYS_BASE + 0x7,
+		.flags  = IORESOURCE_MEM,
+		.name   = "cpufreq: DFS"
+	},
+	[1] = {
+		.start  = DOVE_PMU_PHYS_BASE + 0x8000,
+		.end    = DOVE_PMU_PHYS_BASE + 0x8003,
+		.flags  = IORESOURCE_MEM,
+		.name   = "cpufreq: PMU CR"
+	},
+	[2] = {
+		.start  = DOVE_PMU_PHYS_BASE + 0x0044,
+		.end    = DOVE_PMU_PHYS_BASE + 0x0047,
+		.flags  = IORESOURCE_MEM,
+		.name   = "cpufreq: PMU Clk Div"
+	},
+};
+
+static struct platform_device dove_cpufreq_device = {
+	.name		= "dove-cpufreq",
+	.id		= -1,
+	.num_resources	= ARRAY_SIZE(dove_cpufreq_resources),
+	.resource	= dove_cpufreq_resources,
+};
+
Quoted 8 lines

arch/arm/mach-dove/common.c void __init dove_sdio1_init(void) · line 381

Quoted 4 lines
arch/arm/mach-dove/common.c +4 -0
+void __init dove_cpufreq_init(void)
+{
+	platform_device_register(&dove_cpufreq_device);
+}

Please describe your power management unit in the DT. Your hardcoding physical addresses that can easily be described in the DT, and you don't appear to have an early probing requirement.

Hi Mark

We have been here before, with the kirkwood cpufreq driver. My v1 of that driver had the cpufreq driver fully described in DT. My binding got shot down and i was told that pseudo devices, like cpufreq, cpuidle, should not be in DT, and a platform driver was the way to go. So that is how kirkwood cpufreq works and dove follows that.

Do you want to reverse that decision now?

Thanks
Andrew

#23 lore raw

PATCH v4 2/4 mvebu: Dove: Instantiate cpufreq driver.

Quoted 95 lines · Andrew Lunn
Quoted 83 lines · Mark Rutland
Quoted 82 lines · Andrew Lunn

Add a platform driver definition to instantiate the dove cpufreq driver. Also indicate the ARCH has cpufreq support, so allowing the cpufreq framework to be enabled.

---
Make resource names consistent.
Fix length of ranges

---
 arch/arm/Kconfig                       |  1 +
 arch/arm/mach-dove/board-dt.c          |  3 +++
 arch/arm/mach-dove/common.c            | 36 ++++++++++++++++++++++++++++++++++
 arch/arm/mach-dove/common.h            |  1 +
 arch/arm/mach-dove/include/mach/dove.h |  1 +
 5 files changed, 42 insertions(+)
arch/arm/Kconfig +1 -0
index c1f1a7eee953..908b02f8d901 100644
@@ -500,6 +500,7 @@ config ARCH_IXP4XX
 
 config ARCH_DOVE
 	bool "Marvell Dove"
+	select ARCH_HAS_CPUFREQ
 	select ARCH_REQUIRE_GPIOLIB
 	select CPU_PJ4
 	select GENERIC_CLOCKEVENTS
arch/arm/mach-dove/board-dt.c +3 -0
index 49fa9abd09da..845e8b328432 100644
@@ -27,6 +27,9 @@ static void __init dove_dt_init(void)
 	tauros2_init(0);
 #endif
 	BUG_ON(mvebu_mbus_dt_init());
+
+	dove_cpufreq_init();
+
 	of_platform_populate(NULL, of_default_bus_match_table, NULL, NULL);
 }
 
arch/arm/mach-dove/common.c +31 -0
index c122bcff9f7c..323fcd4b27b6 100644
@@ -344,6 +344,42 @@ void __init dove_sdio1_init(void)
 	platform_device_register(&dove_sdio1);
 }
 
+/*****************************************************************************
+ * CPU Frequency
+ ****************************************************************************/
+static struct resource dove_cpufreq_resources[] = {
+	[0] = {
+		.start  = DOVE_PMU_PHYS_BASE,
+		.end    = DOVE_PMU_PHYS_BASE + 0x7,
+		.flags  = IORESOURCE_MEM,
+		.name   = "cpufreq: DFS"
+	},
+	[1] = {
+		.start  = DOVE_PMU_PHYS_BASE + 0x8000,
+		.end    = DOVE_PMU_PHYS_BASE + 0x8003,
+		.flags  = IORESOURCE_MEM,
+		.name   = "cpufreq: PMU CR"
+	},
+	[2] = {
+		.start  = DOVE_PMU_PHYS_BASE + 0x0044,
+		.end    = DOVE_PMU_PHYS_BASE + 0x0047,
+		.flags  = IORESOURCE_MEM,
+		.name   = "cpufreq: PMU Clk Div"
+	},
+};
+
+static struct platform_device dove_cpufreq_device = {
+	.name		= "dove-cpufreq",
+	.id		= -1,
+	.num_resources	= ARRAY_SIZE(dove_cpufreq_resources),
+	.resource	= dove_cpufreq_resources,
+};
+
Quoted 8 lines

arch/arm/mach-dove/common.c void __init dove_sdio1_init(void) · line 381

Quoted 4 lines
arch/arm/mach-dove/common.c +4 -0
+void __init dove_cpufreq_init(void)
+{
+	platform_device_register(&dove_cpufreq_device);
+}

Please describe your power management unit in the DT. Your hardcoding physical addresses that can easily be described in the DT, and you don't appear to have an early probing requirement.

Hi Mark

Quoted 7 lines

We have been here before, with the kirkwood cpufreq driver. My v1 of that driver had the cpufreq driver fully described in DT. My binding got shot down and i was told that pseudo devices, like cpufreq, cpuidle, should not be in DT, and a platform driver was the way to go. So that is how kirkwood cpufreq works and dove follows that.

Do you want to reverse that decision now?

I'm not arguing for describing a pseudo-device. The power management unit is a real device. If we used that to register a cpufreq driver, it doesn't mean that we've registered a pseudo-device, but rather that we made a decision to register the cpufreq portions based on the necessary real devices being present.

This code registers a platform device that has PMU register fields. We can easily describe the PMU in dt. The PMU _could_ be used by frameworks other than cpufreq, so describing it as a cpufreq device is wrong. Perhaps this was the problem with the proposed kirkwood binding?

Thanks,
Mark.

author maintainer
#24 lore raw

PATCH v4 2/4 mvebu: Dove: Instantiate cpufreq driver.

Quoted 5 lines

I'm not arguing for describing a pseudo-device. The power management unit is a real device. If we used that to register a cpufreq driver, it doesn't mean that we've registered a pseudo-device, but rather that we made a decision to register the cpufreq portions based on the necessary real devices being present.

Hi Mark

I need a more concrete explanation to understand what you mean.

Are you suggesting that i add a drivers/pmu/dove.c, which gets instantiated by DT? This driver would then instantiate the cpufreq driver? If so, it should also instantiate the gated clocks in drivers/clk/mvebu/dove.c, since they are also part of the PMU, according to the data sheet. The RTC and the thermal management unit are also part of the PMU. Should these drivers also be instantiated by the PMU somehow?

Quoted 1 line

Perhaps this was the problem with the proposed kirkwood binding?

Kirkwood does not have a PMU, at least not according to the data sheet. There are registers to control cpuidle, cpufreq, gating some clocks, the exact same RTC and a somewhat similar thermal sensor. However the data sheet makes no reference as to calling the collection the "PMU".

Thanks
Andrew

#25 lore raw

PATCH v4 2/4 mvebu: Dove: Instantiate cpufreq driver.

Quoted 10 lines · Andrew Lunn
Quoted 5 lines

I'm not arguing for describing a pseudo-device. The power management unit is a real device. If we used that to register a cpufreq driver, it doesn't mean that we've registered a pseudo-device, but rather that we made a decision to register the cpufreq portions based on the necessary real devices being present.

Hi Mark

I need a more concrete explanation to understand what you mean.

Quoted 7 lines

Are you suggesting that i add a drivers/pmu/dove.c, which gets instantiated by DT? This driver would then instantiate the cpufreq driver? If so, it should also instantiate the gated clocks in drivers/clk/mvebu/dove.c, since they are also part of the PMU, according to the data sheet. The RTC and the thermal management unit are also part of the PMU. Should these drivers also be instantiated by the PMU somehow?

No, I don't think that is where Mark is heading. Separate the DT from Linux. Describe the PMU in DT. It's then up to Linux to figure out how to use (or not) that device.

I don't know if we've encountered this scenario yet, but is it possible for Linux to have two drivers (cpufreq and cpuidle) answer to the same compatible string? Not that I'm advocating for that solution here...

The only problem I see with that is locking on any resources both drivers use. I think we had a similar issue with watchdog/rtc.

Quoted 7 lines
Quoted 1 line

Perhaps this was the problem with the proposed kirkwood binding?

Kirkwood does not have a PMU, at least not according to the data sheet. There are registers to control cpuidle, cpufreq, gating some clocks, the exact same RTC and a somewhat similar thermal sensor. However the data sheet makes no reference as to calling the collection the "PMU".

If it quacks like a duck... :)

thx,

Jason.

author maintainer
#26 lore raw

PATCH v4 3/4 mvebu: Dove: Enable cpufreq driver in defconfig

In reply to Andrew Lunn

Add cpufreq to dove_defconfig, so it is built by default.

---
 arch/arm/configs/dove_defconfig | 2 ++
 1 file changed, 2 insertions(+)
arch/arm/configs/dove_defconfig +2 -0
index 110105476848..3d9235017e89 100644
@@ -22,6 +22,8 @@ CONFIG_ZBOOT_ROM_TEXT=0x0
 CONFIG_ZBOOT_ROM_BSS=0x0
 CONFIG_ARM_APPENDED_DTB=y
 CONFIG_ARM_ATAG_DTB_COMPAT=y
+CONFIG_CPU_FREQ=y
+CONFIG_CPU_FREQ_GOV_ONDEMAND=y
 CONFIG_VFP=y
 CONFIG_NET=y
 CONFIG_PACKET=y
Signature · 1 line
1.8.5
author maintainer
#27 lore raw

PATCH v4 4/4 mvebu: Dove: Add clocks and DFS interrupt to cpu node in DT.

In reply to Andrew Lunn

The dove-cpufreq driver needs access to the DDR and CPU clock. There is also an interrupt generated when the DFS hardware completes a change of frequencey. Add these to the cpu node in DT.

---
 arch/arm/boot/dts/dove.dtsi | 4 ++++
 1 file changed, 4 insertions(+)
arch/arm/boot/dts/dove.dtsi +4 -0
index 4c8028513133..3842ba02dddf 100644
@@ -22,6 +22,10 @@
 			device_type = "cpu";
 			next-level-cache = <&l2>;
 			reg = <0>;
+			clocks = <&core_clk 1>, <&core_clk 3>;
+			clock-names = "cpu_clk", "ddrclk";
+			interrupt-parent = <&pmu_intc>;
+			interrupts = <0>;
 		};
 	};
 
Signature · 1 line
1.8.5
#28 lore raw

PATCH v4 4/4 mvebu: Dove: Add clocks and DFS interrupt to cpu node in DT.

Quoted 15 lines · Andrew Lunn

The dove-cpufreq driver needs access to the DDR and CPU clock. There is also an interrupt generated when the DFS hardware completes a change of frequencey. Add these to the cpu node in DT.

---
 arch/arm/boot/dts/dove.dtsi | 4 ++++
 1 file changed, 4 insertions(+)
arch/arm/boot/dts/dove.dtsi +0 -0
index 4c8028513133..3842ba02dddf 100644

arch/arm/boot/dts/dove.dtsi line 28

Quoted 7 lines
arch/arm/boot/dts/dove.dtsi +4 -0
@@ -22,6 +22,10 @@
 			device_type = "cpu";
 			next-level-cache = <&l2>;
 			reg = <0>;
+			clocks = <&core_clk 1>, <&core_clk 3>;
+			clock-names = "cpu_clk", "ddrclk";
+			interrupt-parent = <&pmu_intc>;
+			interrupts = <0>;

Is this interrupt actually generated by the CPU, or by the power management unit?

Is ddrclk actually fed into the CPU? Given that you seem to have an external L2 I don't see why the CPU would need the DDR clock.

Thanks,
Mark.

#29 lore raw

PATCH v4 4/4 mvebu: Dove: Add clocks and DFS interrupt to cpu node in DT.

Quoted 2 lines · Mark Rutland

Is ddrclk actually fed into the CPU? Given that you seem to have an external L2 I don't see why the CPU would need the DDR clock.

I see from the other patches that it's simply a misleading name. Sorry for the noise.

Mark.

#30 lore raw

PATCH v4 4/4 mvebu: Dove: Add clocks and DFS interrupt to cpu node in DT.

In reply to Andrew Lunn
Quoted 15 lines · Andrew Lunn

The dove-cpufreq driver needs access to the DDR and CPU clock. There is also an interrupt generated when the DFS hardware completes a change of frequencey. Add these to the cpu node in DT.

---
 arch/arm/boot/dts/dove.dtsi | 4 ++++
 1 file changed, 4 insertions(+)
arch/arm/boot/dts/dove.dtsi +0 -0
index 4c8028513133..3842ba02dddf 100644

arch/arm/boot/dts/dove.dtsi line 25

Quoted 4 lines
arch/arm/boot/dts/dove.dtsi +1 -0
@@ -22,6 +22,10 @@
 			device_type = "cpu";
 			next-level-cache = <&l2>;
 			reg = <0>;
+			clocks = <&core_clk 1>, <&core_clk 3>;

This isn't a blocker by any means, but does anyone have on their TODO list a DT include file of macros for the clocks? So this would become:

			clocks = <&core_clk DOVE_CPU_CLK>, <&core_clk DOVE_DDRCLK>;

or similar?

If not, would it be considered useful?

thx,

Jason.

Quoted 9 lines
arch/arm/boot/dts/dove.dtsi +3 -0
+			clock-names = "cpu_clk", "ddrclk";
+			interrupt-parent = <&pmu_intc>;
+			interrupts = <0>;
 		};
 	};
 
Signature · 1 line
1.8.5