Skip to contentSkip to conversation

Add cpufreq support for Armada 8K

PATCH Superseded Gregory CLEMENT 2 participants 6 messages Dec 11, 2018 – Dec 19, 2018

Subsystems (2)

v3 of 5 Superseded Latest is v5 →

lore
Share & tools
lore

Gregory CLEMENT

author maintainer
#0 lore raw

v3 cover Add cpufreq support for Armada 8K

Hi,

this is the third version of a the series adding the cpufreq support for the Marvell Armada 8K SoCs.

Gregory

Changelog:
v2 -> v3

- Make the driver really modular (modprobe;modprobe -r;modprobe; is working)

- Don't use anymore 100MHz as minimal frequency, but only max frequency/4

- Fix remaining warning reported by checkpatch

- Rework the registering of the opp

v1 -> v2:

- Fix typos in MAINTAINERS (reported by Baruch Siach) and in armada-8k-cpufreq.c (reported by Thomas Petazzoni)

- Fix indentation in Kconfig.arm (reported by Thomas Petazzoni)

- Use const for opps_div (reported by Thomas Petazzoni)

- Fix memory leak of opps_array (reported by Thomas Petazzoni)

Gregory CLEMENT (2):
MAINTAINERS: add new entries for Armada 8K cpufreq driver
cpufreq: ap806: add cpufreq driver for Armada 8K

 MAINTAINERS                         |   1 +
 drivers/cpufreq/Kconfig.arm         |  11 ++
 drivers/cpufreq/Makefile            |   1 +
 drivers/cpufreq/armada-8k-cpufreq.c | 186 ++++++++++++++++++++++++++++
 4 files changed, 199 insertions(+)
 create mode 100644 drivers/cpufreq/armada-8k-cpufreq.c
Signature · 1 line
2.19.2
author maintainer
#1 lore raw

PATCH v3 1/2 MAINTAINERS: add new entries for Armada 8K cpufreq driver

This new driver belongs to the mvebu family, update the MAINTAINER file to document it.

---
 MAINTAINERS | 1 +
 1 file changed, 1 insertion(+)
MAINTAINERS +1 -0
index f4855974f325..c71e4f9c1173 100644
@@ -1717,6 +1717,7 @@ F:	arch/arm/configs/mvebu_*_defconfig
 F:	arch/arm/mach-mvebu/
 F:	arch/arm64/boot/dts/marvell/armada*
 F:	drivers/cpufreq/armada-37xx-cpufreq.c
+F:	drivers/cpufreq/armada-8k-cpufreq.c
 F:	drivers/cpufreq/mvebu-cpufreq.c
 F:	drivers/irqchip/irq-armada-370-xp.c
 F:	drivers/irqchip/irq-mvebu-*
Signature · 1 line
2.19.2
author maintainer
#2 lore raw

PATCH v3 2/2 cpufreq: ap806: add cpufreq driver for Armada 8K

In reply to Gregory CLEMENT

Add cpufreq driver for Marvell AP-806 found on Aramda 8K. The AP-806 has DFS (Dynamic Frequency Scaling) with coupled clock domain for two clusters, so this driver will directly use generic cpufreq-dt driver as backend.

Based on the work of Omri Itach <[email protected]>.

---
 drivers/cpufreq/Kconfig.arm         |  11 ++
 drivers/cpufreq/Makefile            |   1 +
 drivers/cpufreq/armada-8k-cpufreq.c | 186 ++++++++++++++++++++++++++++
 3 files changed, 198 insertions(+)
 create mode 100644 drivers/cpufreq/armada-8k-cpufreq.c
drivers/cpufreq/Kconfig.arm +11 -0
index 4e1131ef85ae..7e32a241760d 100644
@@ -25,6 +25,17 @@ config ARM_ARMADA_37XX_CPUFREQ
 	  This adds the CPUFreq driver support for Marvell Armada 37xx SoCs.
 	  The Armada 37xx PMU supports 4 frequency and VDD levels.
 
+config ARM_ARMADA_8K_CPUFREQ
+	tristate "Armada 8K CPUFreq driver"
+	depends on ARCH_MVEBU && CPUFREQ_DT
+	help
+	  This enables the CPUFreq driver support for Marvell
+	  Armada8k SOCs.
+	  Armada8K device has the AP806 which supports scaling
+	  to any full integer divider.
+
+	  If in doubt, say N.
+
 # big LITTLE core layer and glue drivers
 config ARM_BIG_LITTLE_CPUFREQ
 	tristate "Generic ARM big LITTLE CPUfreq driver"
drivers/cpufreq/Makefile +1 -0
index d5ee4562ed06..db1564b610ac 100644
@@ -50,6 +50,7 @@ obj-$(CONFIG_X86_SFI_CPUFREQ)		+= sfi-cpufreq.o
 obj-$(CONFIG_ARM_BIG_LITTLE_CPUFREQ)	+= arm_big_little.o
 
 obj-$(CONFIG_ARM_ARMADA_37XX_CPUFREQ)	+= armada-37xx-cpufreq.o
+obj-$(CONFIG_ARM_ARMADA_8K_CPUFREQ)	+= armada-8k-cpufreq.o
 obj-$(CONFIG_ARM_BRCMSTB_AVS_CPUFREQ)	+= brcmstb-avs-cpufreq.o
 obj-$(CONFIG_ACPI_CPPC_CPUFREQ)		+= cppc_cpufreq.o
 obj-$(CONFIG_ARCH_DAVINCI)		+= davinci-cpufreq.o
drivers/cpufreq/armada-8k-cpufreq.c new file +186 -0

186 lines not on this page · raw message

Signature · 1 line
2.19.2
maintainer
#3 lore raw

Re: [PATCH v3 2/2] cpufreq: ap806: add cpufreq driver for Armada 8K

Quoted 77 lines · Gregory CLEMENT
drivers/cpufreq/Kconfig.arm +11 -0
index 4e1131ef85ae..7e32a241760d 100644
@@ -25,6 +25,17 @@ config ARM_ARMADA_37XX_CPUFREQ
 	  This adds the CPUFreq driver support for Marvell Armada 37xx SoCs.
 	  The Armada 37xx PMU supports 4 frequency and VDD levels.
 
+config ARM_ARMADA_8K_CPUFREQ
+	tristate "Armada 8K CPUFreq driver"
+	depends on ARCH_MVEBU && CPUFREQ_DT
+	help
+	  This enables the CPUFreq driver support for Marvell
+	  Armada8k SOCs.
+	  Armada8K device has the AP806 which supports scaling
+	  to any full integer divider.
+
+	  If in doubt, say N.
+
 # big LITTLE core layer and glue drivers
 config ARM_BIG_LITTLE_CPUFREQ
 	tristate "Generic ARM big LITTLE CPUfreq driver"
drivers/cpufreq/Makefile +1 -0
index d5ee4562ed06..db1564b610ac 100644
@@ -50,6 +50,7 @@ obj-$(CONFIG_X86_SFI_CPUFREQ)		+= sfi-cpufreq.o
 obj-$(CONFIG_ARM_BIG_LITTLE_CPUFREQ)	+= arm_big_little.o
 
 obj-$(CONFIG_ARM_ARMADA_37XX_CPUFREQ)	+= armada-37xx-cpufreq.o
+obj-$(CONFIG_ARM_ARMADA_8K_CPUFREQ)	+= armada-8k-cpufreq.o
 obj-$(CONFIG_ARM_BRCMSTB_AVS_CPUFREQ)	+= brcmstb-avs-cpufreq.o
 obj-$(CONFIG_ACPI_CPPC_CPUFREQ)		+= cppc_cpufreq.o
 obj-$(CONFIG_ARCH_DAVINCI)		+= davinci-cpufreq.o
drivers/cpufreq/armada-8k-cpufreq.c new file +37 -0
new file mode 100644
index 000000000000..1db1953fb43e
@@ -0,0 +1,186 @@
+// SPDX-License-Identifier: GPL-2.0+
+/*
+ * CPUFreq support for Armada 8K
+ *
+ * Copyright (C) 2018 Marvell
+ *
+ * Omri Itach <[email protected]>
+ * Gregory Clement <[email protected]>
+ */
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <linux/clk.h>
+#include <linux/cpu.h>
+#include <linux/err.h>
+#include <linux/init.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/of.h>
+#include <linux/platform_device.h>
+#include <linux/pm_opp.h>
+#include <linux/slab.h>
+
+/*
+ * Setup the opps list with the divider for the max frequency, that
+ * will be filled at runtime.
+ */
+static const int opps_div[] __initconst = {1, 2, 3, 4};
+
+static struct platform_device *armada_8k_pdev;
+static struct freq_table *freq_tables;
+static int opps_index;
+
+struct freq_table {
+	struct device *cpu_dev;
+	struct clk *clk;
+	unsigned int freq[ARRAY_SIZE(opps_div)];

drivers/cpufreq/armada-8k-cpufreq.c line 47

Quoted 10 lines
drivers/cpufreq/armada-8k-cpufreq.c new file +10 -0
+};
+
+/* If the CPUs share the same clock, then they are in the same cluster. */
+static void __init armada_8k_get_sharing_cpus(struct clk *cur_clk,
+					      struct cpumask *cpumask)
+{
+	int cpu;
+
+	for_each_possible_cpu(cpu) {
+		struct device *cpu_dev = get_cpu_device(cpu);

What if this fails ?

drivers/cpufreq/armada-8k-cpufreq.c line 55

Quoted 8 lines
drivers/cpufreq/armada-8k-cpufreq.c new file +8 -0
+		struct clk *clk = clk_get(cpu_dev, 0);
+
+		if (IS_ERR(clk))
+			dev_warn(cpu_dev, "Cannot get clock for CPU %d\n", cpu);
+		else if (clk_is_match(clk, cur_clk))
+			cpumask_set_cpu(cpu, cpumask);
+
+		clk_put(clk);

So you will do clk_put() even if clk_get() failed, that will trigger a WARN() if I am not wrong.

Quoted 48 lines
drivers/cpufreq/armada-8k-cpufreq.c new file +48 -0
+	}
+}
+
+static int armada_8k_add_opp(struct clk *clk, struct device *cpu_dev,
+			     struct freq_table *freq_tables, int opps_index)
+{
+	unsigned int cur_frequency;
+	unsigned int freq;
+	int i, ret;
+
+	/* Get nominal (current) CPU frequency. */
+	cur_frequency = clk_get_rate(clk);
+
+	if (!cur_frequency) {
+		dev_err(cpu_dev,
+			"Failed to get clock rate for this CPU\n");
+		return -EINVAL;
+	}
+
+	freq_tables[opps_index].cpu_dev = cpu_dev;
+	for (i = 0; i < ARRAY_SIZE(opps_div); i++) {
+		freq = cur_frequency / opps_div[i];
+
+		ret = dev_pm_opp_add(cpu_dev, freq, 0);
+		if (ret)
+			return ret;
+
+		freq_tables[opps_index].freq[i] = freq;
+	}
+
+	return 0;
+}
+
+static void armada_8k_cpufreq_free_table(void)
+{
+	opps_index--;
+	for (; opps_index >= 0; opps_index--) {
+		int i;
+
+		for (i = 0; i < ARRAY_SIZE(opps_div); i++)
+			dev_pm_opp_remove(freq_tables[opps_index].cpu_dev,
+					  freq_tables[opps_index].freq[i]);
+	}
+
+	kfree(freq_tables);
+}
+
+static int __init armada_8k_cpufreq_init(void)

drivers/cpufreq/armada-8k-cpufreq.c line 113

Quoted 10 lines
drivers/cpufreq/armada-8k-cpufreq.c new file +10 -0
+{
+	int ret = 0, cpu, nb_cpus;
+	struct device_node *node;
+
+	node = of_find_compatible_node(NULL, NULL, "marvell,ap806-cpu-clock");
+	if (!node || !of_device_is_available(node))
+		return -ENODEV;
+
+	nb_cpus = num_possible_cpus();
+	freq_tables = kcalloc(nb_cpus, sizeof(*freq_tables), GFP_KERNEL);

Add a blank line here.

Quoted 10 lines
drivers/cpufreq/armada-8k-cpufreq.c new file +10 -0
+	/*
+	 * For each CPU, this loop registers the operating points
+	 * supported (which are the nominal CPU frequency and full integer
+	 * divisions of it).
+	 */
+	for_each_possible_cpu(cpu) {
+		struct device *cpu_dev;
+		struct cpumask cpus;
+		struct clk *clk;
+		int i, skip;

drivers/cpufreq/armada-8k-cpufreq.c line 133

Quoted 10 lines
drivers/cpufreq/armada-8k-cpufreq.c new file +10 -0
+
+		skip = 0;
+		cpu_dev = get_cpu_device(cpu);
+
+		if (!cpu_dev) {
+			dev_err(cpu_dev, "Cannot get CPU %d\n", cpu);
+			continue;
+		}
+
+		clk = devm_clk_get(cpu_dev, 0);

I don't think you should call devm_*() helpers on the cpu_dev pointer, this is not like the platform device structure. The problem is that no driver gets attached/detached from cpu_dev and so the resources may never get free.

drivers/cpufreq/armada-8k-cpufreq.c line 143

Quoted 10 lines
drivers/cpufreq/armada-8k-cpufreq.c new file +10 -0
+
+		if (IS_ERR(clk)) {
+			dev_err(cpu_dev, "Cannot get clock for CPU %d\n", cpu);
+			ret = PTR_ERR(clk);
+			goto remove_opp;
+		}
+
+		for (i = 0; i < nb_cpus; i++) {
+			if (clk_is_match(clk, freq_tables[i].clk)) {
+				skip = 1;

Why not do devm_clk_put() from right here and "continue" after that. That way you can avoid "skip" variable as well.

Also why are you doing this clk-match at all ? This looks similar to what armada_8k_get_sharing_cpus() is doing. You can optimize this stuff better I believe.

I will do this:

Create a cpumask variable and assign possible_cpus mask to it. And then after each iteration of this loop, I clear "cpus" (returned from armada_8k_get_sharing_cpus()) from the CPUs from that cpumask. That way the loop will never try the second CPU from the same cpumask again.

drivers/cpufreq/armada-8k-cpufreq.c line 154

Quoted 11 lines
drivers/cpufreq/armada-8k-cpufreq.c new file +11 -0
+				break;
+			}
+		}
+		if (skip) {
+			devm_clk_put(cpu_dev, clk);
+			continue;
+		}
+
+		freq_tables[opps_index].clk = clk;
+
+		ret = armada_8k_add_opp(clk, cpu_dev, freq_tables, opps_index);

And this opps_index thing is not very clean. Having a global variable this way makes things very messy, which is being updated from several routines.

So if armada_8k_add_opp() fails after adding few OPPs, we will now call armada_8k_cpufreq_free_table() which will first do opps_index-- and so you may miss freeing the OPPs added during the last call to armada_8k_add_opp().

This is very messy and requires serious cleanup to be honest. Also it will be more preferred if armada_8k_add_opp() cleans up all the allocations it has done on a call if it fails in the middle of it, rather than a bigger routine freeing both successful and failed ones.

Quoted 34 lines
drivers/cpufreq/armada-8k-cpufreq.c new file +32 -0
+		if (ret)
+			goto remove_opp;
+
+		opps_index++;
+		cpumask_clear(&cpus);
+		armada_8k_get_sharing_cpus(clk, &cpus);
+		dev_pm_opp_set_sharing_cpus(cpu_dev, &cpus);
+	}
+
+	armada_8k_pdev = platform_device_register_simple("cpufreq-dt", -1,
+							 NULL, 0);
+	ret = PTR_ERR_OR_ZERO(armada_8k_pdev);
+	if (ret)
+		goto remove_opp;
+	return 0;
+
+remove_opp:
+	armada_8k_cpufreq_free_table();
+	return ret;
+}
+module_init(armada_8k_cpufreq_init);
+
+static void __exit armada_8k_cpufreq_exit(void)
+{
+	armada_8k_cpufreq_free_table();
+	platform_device_unregister(armada_8k_pdev);
+}
+module_exit(armada_8k_cpufreq_exit);
+
+MODULE_AUTHOR("Gregory Clement <[email protected]>");
+MODULE_DESCRIPTION("Armada 8K cpufreq driver");
+MODULE_LICENSE("GPL");
Signature · 1 line
2.19.2
Signature · 1 line
viresh
author maintainer
#4 lore raw

Re: [PATCH v3 2/2] cpufreq: ap806: add cpufreq driver for Armada 8K

Hi Viresh,

Quoted 82 lines · Viresh Kumar
Quoted 81 lines · Gregory CLEMENT
drivers/cpufreq/Kconfig.arm +11 -0
index 4e1131ef85ae..7e32a241760d 100644
@@ -25,6 +25,17 @@ config ARM_ARMADA_37XX_CPUFREQ
 	  This adds the CPUFreq driver support for Marvell Armada 37xx SoCs.
 	  The Armada 37xx PMU supports 4 frequency and VDD levels.
 
+config ARM_ARMADA_8K_CPUFREQ
+	tristate "Armada 8K CPUFreq driver"
+	depends on ARCH_MVEBU && CPUFREQ_DT
+	help
+	  This enables the CPUFreq driver support for Marvell
+	  Armada8k SOCs.
+	  Armada8K device has the AP806 which supports scaling
+	  to any full integer divider.
+
+	  If in doubt, say N.
+
 # big LITTLE core layer and glue drivers
 config ARM_BIG_LITTLE_CPUFREQ
 	tristate "Generic ARM big LITTLE CPUfreq driver"
drivers/cpufreq/Makefile +1 -0
index d5ee4562ed06..db1564b610ac 100644
@@ -50,6 +50,7 @@ obj-$(CONFIG_X86_SFI_CPUFREQ)		+= sfi-cpufreq.o
 obj-$(CONFIG_ARM_BIG_LITTLE_CPUFREQ)	+= arm_big_little.o
 
 obj-$(CONFIG_ARM_ARMADA_37XX_CPUFREQ)	+= armada-37xx-cpufreq.o
+obj-$(CONFIG_ARM_ARMADA_8K_CPUFREQ)	+= armada-8k-cpufreq.o
 obj-$(CONFIG_ARM_BRCMSTB_AVS_CPUFREQ)	+= brcmstb-avs-cpufreq.o
 obj-$(CONFIG_ACPI_CPPC_CPUFREQ)		+= cppc_cpufreq.o
 obj-$(CONFIG_ARCH_DAVINCI)		+= davinci-cpufreq.o
drivers/cpufreq/armada-8k-cpufreq.c new file +41 -0
new file mode 100644
index 000000000000..1db1953fb43e
@@ -0,0 +1,186 @@
+// SPDX-License-Identifier: GPL-2.0+
+/*
+ * CPUFreq support for Armada 8K
+ *
+ * Copyright (C) 2018 Marvell
+ *
+ * Omri Itach <[email protected]>
+ * Gregory Clement <[email protected]>
+ */
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <linux/clk.h>
+#include <linux/cpu.h>
+#include <linux/err.h>
+#include <linux/init.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/of.h>
+#include <linux/platform_device.h>
+#include <linux/pm_opp.h>
+#include <linux/slab.h>
+
+/*
+ * Setup the opps list with the divider for the max frequency, that
+ * will be filled at runtime.
+ */
+static const int opps_div[] __initconst = {1, 2, 3, 4};
+
+static struct platform_device *armada_8k_pdev;
+static struct freq_table *freq_tables;
+static int opps_index;
+
+struct freq_table {
+	struct device *cpu_dev;
+	struct clk *clk;
+	unsigned int freq[ARRAY_SIZE(opps_div)];
+};
+
+/* If the CPUs share the same clock, then they are in the same cluster. */
+static void __init armada_8k_get_sharing_cpus(struct clk *cur_clk,
Quoted 8 lines

drivers/cpufreq/armada-8k-cpufreq.c line 47

Quoted 6 lines
drivers/cpufreq/armada-8k-cpufreq.c new file +6 -0
+					      struct cpumask *cpumask)
+{
+	int cpu;
+
+	for_each_possible_cpu(cpu) {
+		struct device *cpu_dev = get_cpu_device(cpu);

What if this fails ?

cpu_dev will be NULL so the clk_get() will fail to find the clock which is handled just after.

But if needed we can add this extra test and differed the clk_get.

Quoted 12 lines

drivers/cpufreq/armada-8k-cpufreq.c line 55

Quoted 9 lines
drivers/cpufreq/armada-8k-cpufreq.c new file +8 -0
+		struct clk *clk = clk_get(cpu_dev, 0);
+
+		if (IS_ERR(clk))
+			dev_warn(cpu_dev, "Cannot get clock for CPU %d\n", cpu);
+		else if (clk_is_match(clk, cur_clk))
+			cpumask_set_cpu(cpu, cpumask);
+
+		clk_put(clk);

So you will do clk_put() even if clk_get() failed, that will trigger a WARN() if I am not wrong.

Yes there is a WARN_ONCE in this case, I can move the clk_put in the else part.

Quoted 53 lines
Quoted 53 lines
drivers/cpufreq/armada-8k-cpufreq.c new file +52 -0
+	}
+}
+
+static int armada_8k_add_opp(struct clk *clk, struct device *cpu_dev,
+			     struct freq_table *freq_tables, int opps_index)
+{
+	unsigned int cur_frequency;
+	unsigned int freq;
+	int i, ret;
+
+	/* Get nominal (current) CPU frequency. */
+	cur_frequency = clk_get_rate(clk);
+
+	if (!cur_frequency) {
+		dev_err(cpu_dev,
+			"Failed to get clock rate for this CPU\n");
+		return -EINVAL;
+	}
+
+	freq_tables[opps_index].cpu_dev = cpu_dev;
+	for (i = 0; i < ARRAY_SIZE(opps_div); i++) {
+		freq = cur_frequency / opps_div[i];
+
+		ret = dev_pm_opp_add(cpu_dev, freq, 0);
+		if (ret)
+			return ret;
+
+		freq_tables[opps_index].freq[i] = freq;
+	}
+
+	return 0;
+}
+
+static void armada_8k_cpufreq_free_table(void)
+{
+	opps_index--;
+	for (; opps_index >= 0; opps_index--) {
+		int i;
+
+		for (i = 0; i < ARRAY_SIZE(opps_div); i++)
+			dev_pm_opp_remove(freq_tables[opps_index].cpu_dev,
+					  freq_tables[opps_index].freq[i]);
+	}
+
+	kfree(freq_tables);
+}
+
+static int __init armada_8k_cpufreq_init(void)
+{
+	int ret = 0, cpu, nb_cpus;
+	struct device_node *node;
+
Quoted 8 lines

drivers/cpufreq/armada-8k-cpufreq.c line 113

Quoted 6 lines
drivers/cpufreq/armada-8k-cpufreq.c new file +6 -0
+	node = of_find_compatible_node(NULL, NULL, "marvell,ap806-cpu-clock");
+	if (!node || !of_device_is_available(node))
+		return -ENODEV;
+
+	nb_cpus = num_possible_cpus();
+	freq_tables = kcalloc(nb_cpus, sizeof(*freq_tables), GFP_KERNEL);

Add a blank line here.

OK

Quoted 18 lines
Quoted 18 lines
drivers/cpufreq/armada-8k-cpufreq.c new file +17 -0
+	/*
+	 * For each CPU, this loop registers the operating points
+	 * supported (which are the nominal CPU frequency and full integer
+	 * divisions of it).
+	 */
+	for_each_possible_cpu(cpu) {
+		struct device *cpu_dev;
+		struct cpumask cpus;
+		struct clk *clk;
+		int i, skip;
+
+		skip = 0;
+		cpu_dev = get_cpu_device(cpu);
+
+		if (!cpu_dev) {
+			dev_err(cpu_dev, "Cannot get CPU %d\n", cpu);
+			continue;
Quoted 8 lines

drivers/cpufreq/armada-8k-cpufreq.c line 133

Quoted 3 lines
drivers/cpufreq/armada-8k-cpufreq.c new file +3 -0
+		}
+
+		clk = devm_clk_get(cpu_dev, 0);

I don't think you should call devm_*() helpers on the cpu_dev pointer, this is not like the platform device structure. The problem is that no driver gets attached/detached from cpu_dev and so the resources may never get free.

OK

Quoted 14 lines

drivers/cpufreq/armada-8k-cpufreq.c line 143

Quoted 11 lines
drivers/cpufreq/armada-8k-cpufreq.c new file +10 -0
+
+		if (IS_ERR(clk)) {
+			dev_err(cpu_dev, "Cannot get clock for CPU %d\n", cpu);
+			ret = PTR_ERR(clk);
+			goto remove_opp;
+		}
+
+		for (i = 0; i < nb_cpus; i++) {
+			if (clk_is_match(clk, freq_tables[i].clk)) {
+				skip = 1;

Why not do devm_clk_put() from right here and "continue" after that. That way you can avoid "skip" variable as well.

We need to keep a reference to the clk when it used later in the loop.

Quoted 10 lines

Also why are you doing this clk-match at all ? This looks similar to what armada_8k_get_sharing_cpus() is doing. You can optimize this stuff better I believe.

I will do this:

Create a cpumask variable and assign possible_cpus mask to it. And then after each iteration of this loop, I clear "cpus" (returned from armada_8k_get_sharing_cpus()) from the CPUs from that cpumask. That way the loop will never try the second CPU from the same cpumask again.

I will try it

Quoted 15 lines

drivers/cpufreq/armada-8k-cpufreq.c line 154

Quoted 12 lines
drivers/cpufreq/armada-8k-cpufreq.c new file +11 -0
+				break;
+			}
+		}
+		if (skip) {
+			devm_clk_put(cpu_dev, clk);
+			continue;
+		}
+
+		freq_tables[opps_index].clk = clk;
+
+		ret = armada_8k_add_opp(clk, cpu_dev, freq_tables, opps_index);

And this opps_index thing is not very clean. Having a global variable this way makes things very messy, which is being updated from several routines.

I think you missed the fact that no variable are passed to the exit function but we still need to have a reference on the OPP to free, so we cant do without this global variable. However I could reduce the amount by hiding them in a platform data associated to the platform_device.

Quoted 5 lines

So if armada_8k_add_opp() fails after adding few OPPs, we will now call armada_8k_cpufreq_free_table() which will first do opps_index-- and so you may miss freeing the OPPs added during the last call to armada_8k_add_opp().

Actually opps_index is only modify after adding succefully an opp, so an OPP can't be missed.

Quoted 4 lines

This is very messy and requires serious cleanup to be honest. Also it will be more preferred if armada_8k_add_opp() cleans up all the allocations it has done on a call if it fails in the middle of it, rather than a bigger routine freeing both successful and failed ones.

OK we can duplicate code for this if you prefer.

Thanks,

Gregory

Quoted 38 lines
Quoted 35 lines
drivers/cpufreq/armada-8k-cpufreq.c new file +32 -0
+		if (ret)
+			goto remove_opp;
+
+		opps_index++;
+		cpumask_clear(&cpus);
+		armada_8k_get_sharing_cpus(clk, &cpus);
+		dev_pm_opp_set_sharing_cpus(cpu_dev, &cpus);
+	}
+
+	armada_8k_pdev = platform_device_register_simple("cpufreq-dt", -1,
+							 NULL, 0);
+	ret = PTR_ERR_OR_ZERO(armada_8k_pdev);
+	if (ret)
+		goto remove_opp;
+	return 0;
+
+remove_opp:
+	armada_8k_cpufreq_free_table();
+	return ret;
+}
+module_init(armada_8k_cpufreq_init);
+
+static void __exit armada_8k_cpufreq_exit(void)
+{
+	armada_8k_cpufreq_free_table();
+	platform_device_unregister(armada_8k_pdev);
+}
+module_exit(armada_8k_cpufreq_exit);
+
+MODULE_AUTHOR("Gregory Clement <[email protected]>");
+MODULE_DESCRIPTION("Armada 8K cpufreq driver");
+MODULE_LICENSE("GPL");
Signature · 1 line
2.19.2
Signature · 1 line
viresh
Signature · 8 lines
Gregory Clement, Bootlin
Embedded Linux and Kernel engineering
http://bootlin.com

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
maintainer
#5 lore raw

Re: [PATCH v3 2/2] cpufreq: ap806: add cpufreq driver for Armada 8K

Quoted 4 lines · Gregory CLEMENT
Quoted 2 lines · Viresh Kumar
Quoted 1 line · Gregory CLEMENT

+++ b/drivers/cpufreq/armada-8k-cpufreq.c

Quoted 8 lines
Quoted 6 lines

drivers/cpufreq/armada-8k-cpufreq.c line 143

Quoted 3 lines
drivers/cpufreq/armada-8k-cpufreq.c new file +3 -0
+		for (i = 0; i < nb_cpus; i++) {
+			if (clk_is_match(clk, freq_tables[i].clk)) {
+				skip = 1;

Why not do devm_clk_put() from right here and "continue" after that. That way you can avoid "skip" variable as well.

We need to keep a reference to the clk when it used later in the loop.

Maybe I wasn't able to clarify what I was suggesting. This code can be written as:

		for (i = 0; i < nb_cpus; i++) {
			if (clk_is_match(clk, freq_tables[i].clk)) {
                                devm_clk_put(cpu_dev, clk);
                                continue;
                        }
                }

I don't see how the current code is different than this.

Quoted 4 lines

I think you missed the fact that no variable are passed to the exit function but we still need to have a reference on the OPP to free, so we cant do without this global variable. However I could reduce the amount by hiding them in a platform data associated to the platform_device.

Maybe, lets see how the simplified form of the driver looks like. And then see if something is still left to be done.

Signature · 1 line
viresh