diff mbox

[1/2] cpufreq: hisilicon: add acpu driver

Message ID 1424956899-8891-2-git-send-email-leo.yan@linaro.org
State New
Headers show

Commit Message

Leo Yan Feb. 26, 2015, 1:21 p.m. UTC
Add acpu driver for hisilicon SoC, acpu is application processor
subsystem. Dependent on the H/W design, the silicon may has the coupled
clock domain for all clusters, or every cluster can have the dedicated
clock domain. So this driver will support both implementations.

Signed-off-by: Leo Yan <leo.yan@linaro.org>
---
 drivers/cpufreq/Kconfig.arm         |   9 +
 drivers/cpufreq/Makefile            |   1 +
 drivers/cpufreq/hisi-acpu-cpufreq.c | 324 ++++++++++++++++++++++++++++++++++++
 3 files changed, 334 insertions(+)
 create mode 100644 drivers/cpufreq/hisi-acpu-cpufreq.c

Comments

Viresh Kumar March 2, 2015, 6:14 a.m. UTC | #1
On 26 February 2015 at 18:51, Leo Yan <leo.yan@linaro.org> wrote:
> Add acpu driver for hisilicon SoC, acpu is application processor
> subsystem. Dependent on the H/W design, the silicon may has the coupled
> clock domain for all clusters, or every cluster can have the dedicated
> clock domain. So this driver will support both implementations.
>
> Signed-off-by: Leo Yan <leo.yan@linaro.org>
> ---
>  drivers/cpufreq/Kconfig.arm         |   9 +
>  drivers/cpufreq/Makefile            |   1 +
>  drivers/cpufreq/hisi-acpu-cpufreq.c | 324 ++++++++++++++++++++++++++++++++++++
>  3 files changed, 334 insertions(+)
>  create mode 100644 drivers/cpufreq/hisi-acpu-cpufreq.c

What is stopping from reusing cpufreq-dt driver ?
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/
Leo Yan March 2, 2015, 10:50 a.m. UTC | #2
Hi Viresh,

On Mon, Mar 02, 2015 at 11:44:28AM +0530, Viresh Kumar wrote:
> On 26 February 2015 at 18:51, Leo Yan <leo.yan@linaro.org> wrote:
> > Add acpu driver for hisilicon SoC, acpu is application processor
> > subsystem. Dependent on the H/W design, the silicon may has the coupled
> > clock domain for all clusters, or every cluster can have the dedicated
> > clock domain. So this driver will support both implementations.
> >
> > Signed-off-by: Leo Yan <leo.yan@linaro.org>
> > ---
> >  drivers/cpufreq/Kconfig.arm         |   9 +
> >  drivers/cpufreq/Makefile            |   1 +
> >  drivers/cpufreq/hisi-acpu-cpufreq.c | 324 ++++++++++++++++++++++++++++++++++++
> >  3 files changed, 334 insertions(+)
> >  create mode 100644 drivers/cpufreq/hisi-acpu-cpufreq.c
> 
> What is stopping from reusing cpufreq-dt driver ?

Thanks for reviewing.

i'm glad to use more general method, let me give more input so that we
can see if can figure out a better way. ;)

1. From hardware design, during the initialization phase, it will
bind every opps with its corresponding voltage, and pass these related
info to power controller. So later, in kernel the cpufreq driver don't
need manually change the voltage, it will only change the cpu clock
frequency and power controller will automatically handle voltage
related operations. This is similar with TC's SPC implementation.

So looks likely the cpufreq-dt driver's voltage related ops are
redundant for this case.

2. For hi6220, it has two clusters but w/t coupled clock domain; after
discussion, the later series SoC will have two clusters with
dedicated clock domain, so we need support these two cases;

if support two clusters, arm_big_little.c is also good option; but it
cannot support coupled clock domain for two clusters; furthermore, the
cpufreq driver also need enable cooling cell so that it can support
thermal framework with cpu cooling device.

Do u think it's reasonable to apply upper changes to arm_big_little.c?

3. for the file hisi-acpu-cpufreq.c, actually it's common enough; all
register's related operations have been encapsulated in clk driver;
Especially thinking about now have many SoCs have multi-clusters and
only need change the frequency from clk APIs, do u think it's a good
idea to change this driver to be a common driver?

Thanks,
Leo Yan
--
To unsubscribe from this list: send the line "unsubscribe linux-pm" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Viresh Kumar March 2, 2015, 11:04 a.m. UTC | #3
On 2 March 2015 at 16:20, Leo Yan <leo.yan@linaro.org> wrote:
> i'm glad to use more general method, let me give more input so that we
> can see if can figure out a better way. ;)

And I am glad to hear that :)

> 1. From hardware design, during the initialization phase, it will
> bind every opps with its corresponding voltage, and pass these related
> info to power controller. So later, in kernel the cpufreq driver don't
> need manually change the voltage, it will only change the cpu clock
> frequency and power controller will automatically handle voltage
> related operations. This is similar with TC's SPC implementation.
>
> So looks likely the cpufreq-dt driver's voltage related ops are
> redundant for this case.

Its okay, they wouldn't harm. You don't have to specify any regulator
in CPUs DT node and the code will not try any fancy stuff :)

> 2. For hi6220, it has two clusters but w/t coupled clock domain; after
> discussion, the later series SoC will have two clusters with
> dedicated clock domain, so we need support these two cases;

Its okay..

> if support two clusters, arm_big_little.c is also good option; but it
> cannot support coupled clock domain for two clusters; furthermore, the
> cpufreq driver also need enable cooling cell so that it can support
> thermal framework with cpu cooling device.
>
> Do u think it's reasonable to apply upper changes to arm_big_little.c?

Yes it is, but I want you to use cpufreq-dt instead. It supports multi
clusters as well now.

> 3. for the file hisi-acpu-cpufreq.c, actually it's common enough; all
> register's related operations have been encapsulated in clk driver;
> Especially thinking about now have many SoCs have multi-clusters and
> only need change the frequency from clk APIs, do u think it's a good
> idea to change this driver to be a common driver?

Just use cpufreq-dt and you wouldn't be required to make any changes at all..
--
To unsubscribe from this list: send the line "unsubscribe linux-pm" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Leo Yan March 2, 2015, 11:14 a.m. UTC | #4
On Mon, Mar 02, 2015 at 04:34:34PM +0530, Viresh Kumar wrote:
> On 2 March 2015 at 16:20, Leo Yan <leo.yan@linaro.org> wrote:
> > i'm glad to use more general method, let me give more input so that we
> > can see if can figure out a better way. ;)
> 
> And I am glad to hear that :)
> 
> > 1. From hardware design, during the initialization phase, it will
> > bind every opps with its corresponding voltage, and pass these related
> > info to power controller. So later, in kernel the cpufreq driver don't
> > need manually change the voltage, it will only change the cpu clock
> > frequency and power controller will automatically handle voltage
> > related operations. This is similar with TC's SPC implementation.
> >
> > So looks likely the cpufreq-dt driver's voltage related ops are
> > redundant for this case.
> 
> Its okay, they wouldn't harm. You don't have to specify any regulator
> in CPUs DT node and the code will not try any fancy stuff :)

If so, it's make sense to directly use cpufreq-dt driver; i will try
it firstly. Appreciate for suggestion :)

Thanks,
Leo Yan
--
To unsubscribe from this list: send the line "unsubscribe linux-pm" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
diff mbox

Patch

diff --git a/drivers/cpufreq/Kconfig.arm b/drivers/cpufreq/Kconfig.arm
index 1b06fc4..7a61d09 100644
--- a/drivers/cpufreq/Kconfig.arm
+++ b/drivers/cpufreq/Kconfig.arm
@@ -108,6 +108,15 @@  config ARM_HIGHBANK_CPUFREQ
 
 	  If in doubt, say N.
 
+config ARM_HISI_ACPU_CPUFREQ
+	tristate "CPUfreq driver for Hisilicon ACPU"
+	depends on ARCH_HISI
+	select PM_OPP
+	help
+	  This enables the CPUfreq driver for hisilicon ACPU.
+
+	  If in doubt, say N.
+
 config ARM_IMX6Q_CPUFREQ
 	tristate "Freescale i.MX6 cpufreq support"
 	depends on ARCH_MXC
diff --git a/drivers/cpufreq/Makefile b/drivers/cpufreq/Makefile
index 82a1821..6b7a3f0 100644
--- a/drivers/cpufreq/Makefile
+++ b/drivers/cpufreq/Makefile
@@ -59,6 +59,7 @@  arm-exynos-cpufreq-$(CONFIG_ARM_EXYNOS4X12_CPUFREQ)	+= exynos4x12-cpufreq.o
 arm-exynos-cpufreq-$(CONFIG_ARM_EXYNOS5250_CPUFREQ)	+= exynos5250-cpufreq.o
 obj-$(CONFIG_ARM_EXYNOS5440_CPUFREQ)	+= exynos5440-cpufreq.o
 obj-$(CONFIG_ARM_HIGHBANK_CPUFREQ)	+= highbank-cpufreq.o
+obj-$(CONFIG_ARM_HISI_ACPU_CPUFREQ)	+= hisi-acpu-cpufreq.o
 obj-$(CONFIG_ARM_IMX6Q_CPUFREQ)		+= imx6q-cpufreq.o
 obj-$(CONFIG_ARM_INTEGRATOR)		+= integrator-cpufreq.o
 obj-$(CONFIG_ARM_KIRKWOOD_CPUFREQ)	+= kirkwood-cpufreq.o
diff --git a/drivers/cpufreq/hisi-acpu-cpufreq.c b/drivers/cpufreq/hisi-acpu-cpufreq.c
new file mode 100644
index 0000000..d8a7c65
--- /dev/null
+++ b/drivers/cpufreq/hisi-acpu-cpufreq.c
@@ -0,0 +1,324 @@ 
+/*
+ * Hisilicon Platforms Using ACPU CPUFreq support
+ *
+ * Copyright (C) 2015 Linaro.
+ * Leo Yan <leo.yan@linaro.org>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed "as is" WITHOUT ANY WARRANTY of any
+ * kind, whether express or implied; without even the implied warranty
+ * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ */
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <linux/clk.h>
+#include <linux/cpu_cooling.h>
+#include <linux/cpu.h>
+#include <linux/cpufreq.h>
+#include <linux/cpumask.h>
+#include <linux/export.h>
+#include <linux/module.h>
+#include <linux/mutex.h>
+#include <linux/of_platform.h>
+#include <linux/pm_opp.h>
+#include <linux/slab.h>
+#include <linux/topology.h>
+#include <linux/types.h>
+
+#define MAX_CLUSTERS	2
+
+static unsigned int coupled_clusters;
+
+static struct cpufreq_frequency_table *freq_table[MAX_CLUSTERS];
+static atomic_t cluster_usage[MAX_CLUSTERS] = {
+	ATOMIC_INIT(0),
+	ATOMIC_INIT(0),
+};
+
+static struct clk *clk[MAX_CLUSTERS];
+static struct mutex cluster_lock[MAX_CLUSTERS];
+static struct thermal_cooling_device *cdev;
+
+static inline int cpu_to_cluster(int cpu)
+{
+	return coupled_clusters ? 0 : topology_physical_package_id(cpu);
+}
+
+static unsigned int hisi_acpu_cpufreq_get_rate(unsigned int cpu)
+{
+	int cluster = cpu_to_cluster(cpu);
+	unsigned int freq;
+
+	mutex_lock(&cluster_lock[cluster]);
+
+	freq = clk_get_rate(clk[cluster]) / 1000;
+
+	mutex_unlock(&cluster_lock[cluster]);
+	return freq;
+}
+
+/* Set clock frequency */
+static int hisi_acpu_cpufreq_set_target(struct cpufreq_policy *policy,
+		unsigned int index)
+{
+	u32 cpu = policy->cpu, cluster;
+	unsigned int freqs_new;
+	int ret = 0;
+
+	cluster = cpu_to_cluster(cpu);
+	freqs_new = freq_table[cluster][index].frequency;
+
+	pr_debug("%s: cluster %d freq_new %d\n", __func__, cluster, freqs_new);
+
+	mutex_lock(&cluster_lock[cluster]);
+
+	ret = clk_set_rate(clk[cluster], freqs_new * 1000);
+	if (WARN_ON(ret))
+		pr_err("clk_set_rate failed: %d, cluster: %d\n", ret, cluster);
+
+	mutex_unlock(&cluster_lock[cluster]);
+	return ret;
+}
+
+static void put_cluster_clk_and_freq_table(struct device *cpu_dev)
+{
+	u32 cluster = cpu_to_cluster(cpu_dev->id);
+
+	if (atomic_dec_return(&cluster_usage[cluster]))
+		return;
+
+	if (!freq_table[cluster])
+		return;
+
+	clk_put(clk[cluster]);
+	dev_pm_opp_free_cpufreq_table(cpu_dev, &freq_table[cluster]);
+	dev_dbg(cpu_dev, "%s: cluster: %d\n", __func__, cluster);
+}
+
+static int get_cluster_clk_and_freq_table(struct device *cpu_dev)
+{
+	struct device_node *np;
+	u32 cluster;
+	char name[6];
+	int ret;
+
+	cluster = cpu_to_cluster(cpu_dev->id);
+
+	if (atomic_inc_return(&cluster_usage[cluster]) != 1)
+		return 0;
+
+	if (freq_table[cluster])
+		return 0;
+
+	np = of_node_get(cpu_dev->of_node);
+	if (!np) {
+		dev_err(cpu_dev, "%s: failed to find cpu%d node\n",
+				__func__, cpu_dev->id);
+		ret = -ENOENT;
+		goto out;
+	}
+
+	ret = of_init_opp_table(cpu_dev);
+	if (ret) {
+		dev_err(cpu_dev, "%s: init_opp_table failed, cpu: %d, err: %d\n",
+				__func__, cpu_dev->id, ret);
+		of_node_put(np);
+		goto out;
+	}
+
+	ret = dev_pm_opp_init_cpufreq_table(cpu_dev, &freq_table[cluster]);
+	if (ret) {
+		dev_err(cpu_dev, "%s: failed to init cpufreq table, cpu: %d, err: %d\n",
+				__func__, cpu_dev->id, ret);
+		goto out;
+	}
+
+	sprintf(name, "acpu%d", cluster);
+	clk[cluster] = clk_get_sys(name, NULL);
+	if (IS_ERR(clk[cluster]))
+		clk[cluster] = clk_get(cpu_dev, name);
+	if (IS_ERR(clk[cluster])) {
+		dev_err(cpu_dev, "%s: Failed to get clk for cpu: %d, cluster: %d\n",
+				__func__, cpu_dev->id, cluster);
+		ret = PTR_ERR(clk[cluster]);
+		dev_pm_opp_free_cpufreq_table(cpu_dev, &freq_table[cluster]);
+		goto out;
+	} else {
+		dev_info(cpu_dev, "%s: clk: %p & freq table: %p, cluster: %d\n",
+			 __func__, clk[cluster], freq_table[cluster], cluster);
+	}
+
+	return 0;
+
+out:
+	dev_err(cpu_dev, "%s: Failed to get data for cluster: %d\n",
+			__func__, cluster);
+	atomic_dec(&cluster_usage[cluster]);
+	return ret;
+}
+
+/* Per-CPU initialization */
+static int hisi_acpu_cpufreq_init(struct cpufreq_policy *policy)
+{
+	u32 cur_cluster = cpu_to_cluster(policy->cpu);
+	struct device *cpu_dev;
+	int ret;
+
+	cpu_dev = get_cpu_device(policy->cpu);
+	if (!cpu_dev) {
+		pr_err("%s: failed to get cpu%d device\n", __func__,
+				policy->cpu);
+		return -ENODEV;
+	}
+
+	ret = get_cluster_clk_and_freq_table(cpu_dev);
+	if (ret)
+		return ret;
+
+	/*
+	 * If system have two clusters, usually the clock has two options:
+	 * 1. two clusters share the same clock source;
+	 * 2. two clusters have dedicated clock source;
+	 *
+	 * so add the flag coupled_clusters to indicate whether two clusters
+	 * share the clock source or not. if two clusters share the clock source
+	 * then directly call the generic init flow, it will bind all cpus...
+	 */
+	if (coupled_clusters)
+		return cpufreq_generic_init(policy, freq_table[0],
+					    1000000); /* 1 ms, assumed */
+
+	/* bind the cpus within the cluster */
+	ret = cpufreq_table_validate_and_show(policy,
+					      freq_table[cur_cluster]);
+	if (ret) {
+		dev_err(cpu_dev, "CPU %d, cluster: %d invalid freq table\n",
+				policy->cpu, cur_cluster);
+		put_cluster_clk_and_freq_table(cpu_dev);
+		return ret;
+	}
+
+	cpumask_copy(policy->cpus, topology_core_cpumask(policy->cpu));
+	policy->cpuinfo.transition_latency = 1000000; /* 1 ms, assumed */
+
+	dev_info(cpu_dev, "%s: CPU %d initialized\n", __func__, policy->cpu);
+	return 0;
+}
+
+static int hisi_acpu_cpufreq_exit(struct cpufreq_policy *policy)
+{
+	struct device *cpu_dev;
+
+	cpu_dev = get_cpu_device(policy->cpu);
+	if (!cpu_dev) {
+		pr_err("%s: failed to get cpu%d device\n", __func__,
+				policy->cpu);
+		return -ENODEV;
+	}
+
+	put_cluster_clk_and_freq_table(cpu_dev);
+	dev_dbg(cpu_dev, "%s: Exited, cpu: %d\n", __func__, policy->cpu);
+	return 0;
+}
+
+static struct cpufreq_driver hisi_acpu_cpufreq_driver = {
+	.name		= "hisi-acpu",
+	.flags		= CPUFREQ_STICKY |
+			  CPUFREQ_HAVE_GOVERNOR_PER_POLICY |
+			  CPUFREQ_NEED_INITIAL_FREQ_CHECK,
+	.verify		= cpufreq_generic_frequency_table_verify,
+	.target_index	= hisi_acpu_cpufreq_set_target,
+	.get		= hisi_acpu_cpufreq_get_rate,
+	.init		= hisi_acpu_cpufreq_init,
+	.exit		= hisi_acpu_cpufreq_exit,
+	.attr		= cpufreq_generic_attr,
+};
+
+static const struct of_device_id hisi_acpu_cpufreq_match[] = {
+	{
+		.compatible = "hisilicon,hisi-acpu-cpufreq",
+	},
+	{},
+};
+MODULE_DEVICE_TABLE(of, hisi_acpu_cpufreq_match);
+
+static int hisi_acpu_cpufreq_probe(struct platform_device *pdev)
+{
+	int ret = 0, i;
+	struct device_node *np, *cpus;
+
+	np = pdev->dev.of_node;
+	if (!np) {
+		ret = -ENODEV;
+		goto out;
+	}
+
+	of_property_read_u32(np, "hisilicon,coupled-clusters",
+			     &coupled_clusters);
+	dev_dbg(&pdev->dev, "%s: coupled_clusters is %d\n",
+			__func__, coupled_clusters);
+
+	for (i = 0; i < MAX_CLUSTERS; i++)
+		mutex_init(&cluster_lock[i]);
+
+	ret = cpufreq_register_driver(&hisi_acpu_cpufreq_driver);
+	if (ret)
+		dev_err(&pdev->dev,
+			"%s: failed to register cpufreq driver\n", __func__);
+
+	cpus = of_find_node_by_path("/cpus");
+	if (!cpus) {
+		dev_err(&pdev->dev, "failed to find cpus node\n");
+		return 0;
+	}
+
+	np = of_get_next_child(cpus, NULL);
+	if (!np) {
+		dev_err(&pdev->dev, "failed to find cpus child node\n");
+		of_node_put(cpus);
+		return 0;
+	}
+
+	if (of_find_property(np, "#cooling-cells", NULL)) {
+		cdev = of_cpufreq_cooling_register(np, cpu_present_mask);
+		if (IS_ERR(cdev)) {
+			dev_err(&pdev->dev, "running cpufreq without cooling device: %ld\n",
+			       PTR_ERR(cdev));
+			cdev = NULL;
+		}
+	}
+	of_node_put(np);
+	of_node_put(cpus);
+
+	return 0;
+
+out:
+	return ret;
+}
+
+static int hisi_acpu_cpufreq_remove(struct platform_device *pdev)
+{
+	cpufreq_cooling_unregister(cdev);
+	cpufreq_unregister_driver(&hisi_acpu_cpufreq_driver);
+	return 0;
+}
+
+static struct platform_driver hisi_acpu_cpufreq_platdrv = {
+	.driver = {
+		.name	= "hisi-acpu-cpufreq",
+		.owner	= THIS_MODULE,
+		.of_match_table = hisi_acpu_cpufreq_match,
+	},
+	.probe		= hisi_acpu_cpufreq_probe,
+	.remove		= hisi_acpu_cpufreq_remove,
+};
+module_platform_driver(hisi_acpu_cpufreq_platdrv);
+
+MODULE_AUTHOR("Leo Yan <leo.yan@linaro.org>");
+MODULE_DESCRIPTION("Hisilicon acpu cpufreq driver");
+MODULE_LICENSE("GPL v2");