[PATCH v2 0/2] cpufreq support for the big.LITTLE switcher

[PATCH v2 0/2] cpufreq support for the big.LITTLE switcher

[Nicolas Pitre]

This is the third and final set of patches towards a fully functional
and production quality switcher solution for big.LITTLE systems,
establishing a landmark to compare against for any scheduler based
solution meant to eventually surpass the switcher's power efficiency
available in the mainline kernel.

Rationale for this code: http://lwn.net/Articles/481055/

The first and second patch sets have already been merged by RMK.  They
implement the core switcher mechanism.  This set adds the necessary code
to drive the switcher based on cpufreq governor decisions.

This set (v2) was rebased on top of the latest linux-pm tree as
requested by Rafael.

[PATCH v2 0/2] cpufreq support for the big.LITTLE switcher

[Nicolas Pitre]

This is the third and final set of patches towards a fully functional
and production quality switcher solution for big.LITTLE systems,
establishing a landmark to compare against for any scheduler based
solution meant to eventually surpass the switcher's power efficiency
available in the mainline kernel.

Rationale for this code: http://lwn.net/Articles/481055/

The first and second patch sets have already been merged by RMK.  They
implement the core switcher mechanism.  This set adds the necessary code
to drive the switcher based on cpufreq governor decisions.

This set (v2) was rebased on top of the latest linux-pm tree as
requested by Rafael.

[PATCH v2 1/2] cpufreq: arm_big_little: add in-kernel switching (IKS) support

[Nicolas Pitre]

From: Viresh Kumar <viresh.kumar@linaro.org>

This patch adds IKS (In Kernel Switcher) support to cpufreq driver.

This creates a combined freq table for A7-A15 CPU pairs. A7 frequencies
are virtualized and scaled down to half the actual frequencies to
approximate a linear scale across the combined A7+A15 range. When the
requested frequency change crosses the A7-A15 boundary a cluster switch
is invoked.

Based on Earlier Work from Sudeep.

Signed-off-by: Sudeep KarkadaNagesha <sudeep.karkadanagesha@arm.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Nicolas Pitre <nico@linaro.org>
---
 drivers/cpufreq/arm_big_little.c | 363 ++++++++++++++++++++++++++++++++++++---
 drivers/cpufreq/arm_big_little.h |   5 -
 2 files changed, 337 insertions(+), 31 deletions(-)

diff --git a/drivers/cpufreq/arm_big_little.c b/drivers/cpufreq/arm_big_little.c
index 163e3378fe..a6a8f74811 100644
--- a/drivers/cpufreq/arm_big_little.c
+++ b/drivers/cpufreq/arm_big_little.c
@@ -24,27 +24,165 @@
 #include <linux/cpufreq.h>
 #include <linux/cpumask.h>
 #include <linux/export.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>
+#include <asm/bL_switcher.h>
 
 #include "arm_big_little.h"
 
 /* Currently we support only two clusters */
+#define A15_CLUSTER	0
+#define A7_CLUSTER	1
 #define MAX_CLUSTERS	2
 
+#ifdef CONFIG_BL_SWITCHER
+#define is_bL_switching_enabled()	true
+#else
+#define is_bL_switching_enabled()	false
+#endif
+
+#define ACTUAL_FREQ(cluster, freq)  ((cluster == A7_CLUSTER) ? freq << 1 : freq)
+#define VIRT_FREQ(cluster, freq)    ((cluster == A7_CLUSTER) ? freq >> 1 : freq)
+
 static struct cpufreq_arm_bL_ops *arm_bL_ops;
 static struct clk *clk[MAX_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 cpufreq_frequency_table *freq_table[MAX_CLUSTERS + 1];
+static atomic_t cluster_usage[MAX_CLUSTERS + 1];
+
+static unsigned int clk_big_min;	/* (Big) clock frequencies */
+static unsigned int clk_little_max;	/* Maximum clock frequency (Little) */
+
+static DEFINE_PER_CPU(unsigned int, physical_cluster);
+static DEFINE_PER_CPU(unsigned int, cpu_last_req_freq);
+
+static struct mutex cluster_lock[MAX_CLUSTERS];
+
+static inline int raw_cpu_to_cluster(int cpu)
+{
+	return topology_physical_package_id(cpu);
+}
+
+static inline int cpu_to_cluster(int cpu)
+{
+	return is_bL_switching_enabled() ?
+		MAX_CLUSTERS : raw_cpu_to_cluster(cpu);
+}
+
+static unsigned int find_cluster_maxfreq(int cluster)
+{
+	int j;
+	u32 max_freq = 0, cpu_freq;
+
+	for_each_online_cpu(j) {
+		cpu_freq = per_cpu(cpu_last_req_freq, j);
+
+		if ((cluster == per_cpu(physical_cluster, j)) &&
+				(max_freq < cpu_freq))
+			max_freq = cpu_freq;
+	}
+
+	pr_debug("%s: cluster: %d, max freq: %d\n", __func__, cluster,
+			max_freq);
+
+	return max_freq;
+}
+
+static unsigned int clk_get_cpu_rate(unsigned int cpu)
+{
+	u32 cur_cluster = per_cpu(physical_cluster, cpu);
+	u32 rate = clk_get_rate(clk[cur_cluster]) / 1000;
+
+	/* For switcher we use virtual A7 clock rates */
+	if (is_bL_switching_enabled())
+		rate = VIRT_FREQ(cur_cluster, rate);
+
+	pr_debug("%s: cpu: %d, cluster: %d, freq: %u\n", __func__, cpu,
+			cur_cluster, rate);
+
+	return rate;
+}
+
+static unsigned int bL_cpufreq_get_rate(unsigned int cpu)
+{
+	if (is_bL_switching_enabled()) {
+		pr_debug("%s: freq: %d\n", __func__, per_cpu(cpu_last_req_freq,
+					cpu));
+
+		return per_cpu(cpu_last_req_freq, cpu);
+	} else {
+		return clk_get_cpu_rate(cpu);
+	}
+}
 
-static unsigned int bL_cpufreq_get(unsigned int cpu)
+static unsigned int
+bL_cpufreq_set_rate(u32 cpu, u32 old_cluster, u32 new_cluster, u32 rate)
 {
-	u32 cur_cluster = cpu_to_cluster(cpu);
+	u32 new_rate, prev_rate;
+	int ret;
+	bool bLs = is_bL_switching_enabled();
+
+	mutex_lock(&cluster_lock[new_cluster]);
+
+	if (bLs) {
+		prev_rate = per_cpu(cpu_last_req_freq, cpu);
+		per_cpu(cpu_last_req_freq, cpu) = rate;
+		per_cpu(physical_cluster, cpu) = new_cluster;
+
+		new_rate = find_cluster_maxfreq(new_cluster);
+		new_rate = ACTUAL_FREQ(new_cluster, new_rate);
+	} else {
+		new_rate = rate;
+	}
+
+	pr_debug("%s: cpu: %d, old cluster: %d, new cluster: %d, freq: %d\n",
+			__func__, cpu, old_cluster, new_cluster, new_rate);
+
+	ret = clk_set_rate(clk[new_cluster], new_rate * 1000);
+	if (WARN_ON(ret)) {
+		pr_err("clk_set_rate failed: %d, new cluster: %d\n", ret,
+				new_cluster);
+		if (bLs) {
+			per_cpu(cpu_last_req_freq, cpu) = prev_rate;
+			per_cpu(physical_cluster, cpu) = old_cluster;
+		}
+
+		mutex_unlock(&cluster_lock[new_cluster]);
+
+		return ret;
+	}
+
+	mutex_unlock(&cluster_lock[new_cluster]);
+
+	/* Recalc freq for old cluster when switching clusters */
+	if (old_cluster != new_cluster) {
+		pr_debug("%s: cpu: %d, old cluster: %d, new cluster: %d\n",
+				__func__, cpu, old_cluster, new_cluster);
+
+		/* Switch cluster */
+		bL_switch_request(cpu, new_cluster);
+
+		mutex_lock(&cluster_lock[old_cluster]);
 
-	return clk_get_rate(clk[cur_cluster]) / 1000;
+		/* Set freq of old cluster if there are cpus left on it */
+		new_rate = find_cluster_maxfreq(old_cluster);
+		new_rate = ACTUAL_FREQ(old_cluster, new_rate);
+
+		if (new_rate) {
+			pr_debug("%s: Updating rate of old cluster: %d, to freq: %d\n",
+					__func__, old_cluster, new_rate);
+
+			if (clk_set_rate(clk[old_cluster], new_rate * 1000))
+				pr_err("%s: clk_set_rate failed: %d, old cluster: %d\n",
+						__func__, ret, old_cluster);
+		}
+		mutex_unlock(&cluster_lock[old_cluster]);
+	}
+
+	return 0;
 }
 
 /* Set clock frequency */
@@ -52,63 +190,164 @@ static int bL_cpufreq_set_target(struct cpufreq_policy *policy,
 		unsigned int index)
 {
 	struct cpufreq_freqs freqs;
-	u32 cpu = policy->cpu, cur_cluster;
+	u32 cpu = policy->cpu, cur_cluster, new_cluster, actual_cluster;
 	int ret = 0;
 
-	cur_cluster = cpu_to_cluster(policy->cpu);
+	cur_cluster = cpu_to_cluster(cpu);
+	new_cluster = actual_cluster = per_cpu(physical_cluster, cpu);
 
-	freqs.old = bL_cpufreq_get(policy->cpu);
+	freqs.old = bL_cpufreq_get_rate(cpu);
 	freqs.new = freq_table[cur_cluster][index].frequency;
 
 	pr_debug("%s: cpu: %d, cluster: %d, oldfreq: %d, target freq: %d, new freq: %d\n",
 			__func__, cpu, cur_cluster, freqs.old, freqs.new,
 			freqs.new);
 
+	if (is_bL_switching_enabled()) {
+		if ((actual_cluster == A15_CLUSTER) &&
+				(freqs.new < clk_big_min)) {
+			new_cluster = A7_CLUSTER;
+		} else if ((actual_cluster == A7_CLUSTER) &&
+				(freqs.new > clk_little_max)) {
+			new_cluster = A15_CLUSTER;
+		}
+	}
+
 	cpufreq_notify_transition(policy, &freqs, CPUFREQ_PRECHANGE);
 
-	ret = clk_set_rate(clk[cur_cluster], freqs.new * 1000);
-	if (ret) {
-		pr_err("clk_set_rate failed: %d\n", ret);
+	ret = bL_cpufreq_set_rate(cpu, actual_cluster, new_cluster, freqs.new);
+	if (ret)
 		freqs.new = freqs.old;
-	}
 
 	cpufreq_notify_transition(policy, &freqs, CPUFREQ_POSTCHANGE);
 
 	return ret;
 }
 
+static inline u32 get_table_count(struct cpufreq_frequency_table *table)
+{
+	int count;
+
+	for (count = 0; table[count].frequency != CPUFREQ_TABLE_END; count++)
+		;
+
+	return count;
+}
+
+/* get the minimum frequency in the cpufreq_frequency_table */
+static inline u32 get_table_min(struct cpufreq_frequency_table *table)
+{
+	int i;
+	uint32_t min_freq = ~0;
+	for (i = 0; (table[i].frequency != CPUFREQ_TABLE_END); i++)
+		if (table[i].frequency < min_freq)
+			min_freq = table[i].frequency;
+	return min_freq;
+}
+
+/* get the maximum frequency in the cpufreq_frequency_table */
+static inline u32 get_table_max(struct cpufreq_frequency_table *table)
+{
+	int i;
+	uint32_t max_freq = 0;
+	for (i = 0; (table[i].frequency != CPUFREQ_TABLE_END); i++)
+		if (table[i].frequency > max_freq)
+			max_freq = table[i].frequency;
+	return max_freq;
+}
+
+static int merge_cluster_tables(void)
+{
+	int i, j, k = 0, count = 1;
+	struct cpufreq_frequency_table *table;
+
+	for (i = 0; i < MAX_CLUSTERS; i++)
+		count += get_table_count(freq_table[i]);
+
+	table = kzalloc(sizeof(*table) * count, GFP_KERNEL);
+	if (!table)
+		return -ENOMEM;
+
+	freq_table[MAX_CLUSTERS] = table;
+
+	/* Add in reverse order to get freqs in increasing order */
+	for (i = MAX_CLUSTERS - 1; i >= 0; i--) {
+		for (j = 0; freq_table[i][j].frequency != CPUFREQ_TABLE_END;
+				j++) {
+			table[k].frequency = VIRT_FREQ(i,
+					freq_table[i][j].frequency);
+			pr_debug("%s: index: %d, freq: %d\n", __func__, k,
+					table[k].frequency);
+			k++;
+		}
+	}
+
+	table[k].driver_data = k;
+	table[k].frequency = CPUFREQ_TABLE_END;
+
+	pr_debug("%s: End, table: %p, count: %d\n", __func__, table, k);
+
+	return 0;
+}
+
+static void _put_cluster_clk_and_freq_table(struct device *cpu_dev)
+{
+	u32 cluster = raw_cpu_to_cluster(cpu_dev->id);
+
+	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 void put_cluster_clk_and_freq_table(struct device *cpu_dev)
 {
 	u32 cluster = cpu_to_cluster(cpu_dev->id);
+	int i;
+
+	if (atomic_dec_return(&cluster_usage[cluster]))
+		return;
+
+	if (cluster < MAX_CLUSTERS)
+		return _put_cluster_clk_and_freq_table(cpu_dev);
 
-	if (!atomic_dec_return(&cluster_usage[cluster])) {
-		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);
+	for_each_present_cpu(i) {
+		struct device *cdev = get_cpu_device(i);
+		if (!cdev) {
+			pr_err("%s: failed to get cpu%d device\n", __func__, i);
+			return;
+		}
+
+		_put_cluster_clk_and_freq_table(cdev);
 	}
+
+	/* free virtual table */
+	kfree(freq_table[cluster]);
 }
 
-static int get_cluster_clk_and_freq_table(struct device *cpu_dev)
+static int _get_cluster_clk_and_freq_table(struct device *cpu_dev)
 {
-	u32 cluster = cpu_to_cluster(cpu_dev->id);
+	u32 cluster = raw_cpu_to_cluster(cpu_dev->id);
 	char name[14] = "cpu-cluster.";
 	int ret;
 
-	if (atomic_inc_return(&cluster_usage[cluster]) != 1)
+	if (freq_table[cluster])
 		return 0;
 
 	ret = arm_bL_ops->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);
-		goto atomic_dec;
+		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 atomic_dec;
+		goto out;
 	}
 
 	name[12] = cluster + '0';
@@ -125,13 +364,72 @@ static int get_cluster_clk_and_freq_table(struct device *cpu_dev)
 	ret = PTR_ERR(clk[cluster]);
 	dev_pm_opp_free_cpufreq_table(cpu_dev, &freq_table[cluster]);
 
-atomic_dec:
-	atomic_dec(&cluster_usage[cluster]);
+out:
 	dev_err(cpu_dev, "%s: Failed to get data for cluster: %d\n", __func__,
 			cluster);
 	return ret;
 }
 
+static int get_cluster_clk_and_freq_table(struct device *cpu_dev)
+{
+	u32 cluster = cpu_to_cluster(cpu_dev->id);
+	int i, ret;
+
+	if (atomic_inc_return(&cluster_usage[cluster]) != 1)
+		return 0;
+
+	if (cluster < MAX_CLUSTERS) {
+		ret = _get_cluster_clk_and_freq_table(cpu_dev);
+		if (ret)
+			atomic_dec(&cluster_usage[cluster]);
+		return ret;
+	}
+
+	/*
+	 * Get data for all clusters and fill virtual cluster with a merge of
+	 * both
+	 */
+	for_each_present_cpu(i) {
+		struct device *cdev = get_cpu_device(i);
+		if (!cdev) {
+			pr_err("%s: failed to get cpu%d device\n", __func__, i);
+			return -ENODEV;
+		}
+
+		ret = _get_cluster_clk_and_freq_table(cdev);
+		if (ret)
+			goto put_clusters;
+	}
+
+	ret = merge_cluster_tables();
+	if (ret)
+		goto put_clusters;
+
+	/* Assuming 2 cluster, set clk_big_min and clk_little_max */
+	clk_big_min = get_table_min(freq_table[0]);
+	clk_little_max = VIRT_FREQ(1, get_table_max(freq_table[1]));
+
+	pr_debug("%s: cluster: %d, clk_big_min: %d, clk_little_max: %d\n",
+			__func__, cluster, clk_big_min, clk_little_max);
+
+	return 0;
+
+put_clusters:
+	for_each_present_cpu(i) {
+		struct device *cdev = get_cpu_device(i);
+		if (!cdev) {
+			pr_err("%s: failed to get cpu%d device\n", __func__, i);
+			return -ENODEV;
+		}
+
+		_put_cluster_clk_and_freq_table(cdev);
+	}
+
+	atomic_dec(&cluster_usage[cluster]);
+
+	return ret;
+}
+
 /* Per-CPU initialization */
 static int bL_cpufreq_init(struct cpufreq_policy *policy)
 {
@@ -158,13 +456,23 @@ static int bL_cpufreq_init(struct cpufreq_policy *policy)
 		return ret;
 	}
 
+	if (cur_cluster < MAX_CLUSTERS) {
+		cpumask_copy(policy->cpus, topology_core_cpumask(policy->cpu));
+
+		per_cpu(physical_cluster, policy->cpu) = cur_cluster;
+	} else {
+		/* Assumption: during init, we are always running on A15 */
+		per_cpu(physical_cluster, policy->cpu) = A15_CLUSTER;
+	}
+
 	if (arm_bL_ops->get_transition_latency)
 		policy->cpuinfo.transition_latency =
 			arm_bL_ops->get_transition_latency(cpu_dev);
 	else
 		policy->cpuinfo.transition_latency = CPUFREQ_ETERNAL;
 
-	cpumask_copy(policy->cpus, topology_core_cpumask(policy->cpu));
+	if (is_bL_switching_enabled())
+		per_cpu(cpu_last_req_freq, policy->cpu) = clk_get_cpu_rate;
 
 	dev_info(cpu_dev, "%s: CPU %d initialized\n", __func__, policy->cpu);
 	return 0;
@@ -194,7 +502,7 @@ static struct cpufreq_driver bL_cpufreq_driver = {
 					CPUFREQ_HAVE_GOVERNOR_PER_POLICY,
 	.verify			= cpufreq_generic_frequency_table_verify,
 	.target_index		= bL_cpufreq_set_target,
-	.get			= bL_cpufreq_get,
+	.get			= bL_cpufreq_get_rate,
 	.init			= bL_cpufreq_init,
 	.exit			= bL_cpufreq_exit,
 	.attr			= cpufreq_generic_attr,
@@ -202,7 +510,7 @@ static struct cpufreq_driver bL_cpufreq_driver = {
 
 int bL_cpufreq_register(struct cpufreq_arm_bL_ops *ops)
 {
-	int ret;
+	int ret, i;
 
 	if (arm_bL_ops) {
 		pr_debug("%s: Already registered: %s, exiting\n", __func__,
@@ -217,6 +525,9 @@ int bL_cpufreq_register(struct cpufreq_arm_bL_ops *ops)
 
 	arm_bL_ops = ops;
 
+	for (i = 0; i < MAX_CLUSTERS; i++)
+		mutex_init(&cluster_lock[i]);
+
 	ret = cpufreq_register_driver(&bL_cpufreq_driver);
 	if (ret) {
 		pr_info("%s: Failed registering platform driver: %s, err: %d\n",
diff --git a/drivers/cpufreq/arm_big_little.h b/drivers/cpufreq/arm_big_little.h
index 79b2ce1788..70f18fc12d 100644
--- a/drivers/cpufreq/arm_big_little.h
+++ b/drivers/cpufreq/arm_big_little.h
@@ -34,11 +34,6 @@ struct cpufreq_arm_bL_ops {
 	int (*init_opp_table)(struct device *cpu_dev);
 };
 
-static inline int cpu_to_cluster(int cpu)
-{
-	return topology_physical_package_id(cpu);
-}
-
 int bL_cpufreq_register(struct cpufreq_arm_bL_ops *ops);
 void bL_cpufreq_unregister(struct cpufreq_arm_bL_ops *ops);
 
-- 
1.8.4.108.g55ea5f6

[PATCH v2 1/2] cpufreq: arm_big_little: add in-kernel switching (IKS) support

[Nicolas Pitre]

From: Viresh Kumar <viresh.kumar@linaro.org>

This patch adds IKS (In Kernel Switcher) support to cpufreq driver.

This creates a combined freq table for A7-A15 CPU pairs. A7 frequencies
are virtualized and scaled down to half the actual frequencies to
approximate a linear scale across the combined A7+A15 range. When the
requested frequency change crosses the A7-A15 boundary a cluster switch
is invoked.

Based on Earlier Work from Sudeep.

Signed-off-by: Sudeep KarkadaNagesha <sudeep.karkadanagesha@arm.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Nicolas Pitre <nico@linaro.org>
---
 drivers/cpufreq/arm_big_little.c | 363 ++++++++++++++++++++++++++++++++++++---
 drivers/cpufreq/arm_big_little.h |   5 -
 2 files changed, 337 insertions(+), 31 deletions(-)

diff --git a/drivers/cpufreq/arm_big_little.c b/drivers/cpufreq/arm_big_little.c
index 163e3378fe..a6a8f74811 100644
--- a/drivers/cpufreq/arm_big_little.c
+++ b/drivers/cpufreq/arm_big_little.c
@@ -24,27 +24,165 @@
 #include <linux/cpufreq.h>
 #include <linux/cpumask.h>
 #include <linux/export.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>
+#include <asm/bL_switcher.h>
 
 #include "arm_big_little.h"
 
 /* Currently we support only two clusters */
+#define A15_CLUSTER	0
+#define A7_CLUSTER	1
 #define MAX_CLUSTERS	2
 
+#ifdef CONFIG_BL_SWITCHER
+#define is_bL_switching_enabled()	true
+#else
+#define is_bL_switching_enabled()	false
+#endif
+
+#define ACTUAL_FREQ(cluster, freq)  ((cluster == A7_CLUSTER) ? freq << 1 : freq)
+#define VIRT_FREQ(cluster, freq)    ((cluster == A7_CLUSTER) ? freq >> 1 : freq)
+
 static struct cpufreq_arm_bL_ops *arm_bL_ops;
 static struct clk *clk[MAX_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 cpufreq_frequency_table *freq_table[MAX_CLUSTERS + 1];
+static atomic_t cluster_usage[MAX_CLUSTERS + 1];
+
+static unsigned int clk_big_min;	/* (Big) clock frequencies */
+static unsigned int clk_little_max;	/* Maximum clock frequency (Little) */
+
+static DEFINE_PER_CPU(unsigned int, physical_cluster);
+static DEFINE_PER_CPU(unsigned int, cpu_last_req_freq);
+
+static struct mutex cluster_lock[MAX_CLUSTERS];
+
+static inline int raw_cpu_to_cluster(int cpu)
+{
+	return topology_physical_package_id(cpu);
+}
+
+static inline int cpu_to_cluster(int cpu)
+{
+	return is_bL_switching_enabled() ?
+		MAX_CLUSTERS : raw_cpu_to_cluster(cpu);
+}
+
+static unsigned int find_cluster_maxfreq(int cluster)
+{
+	int j;
+	u32 max_freq = 0, cpu_freq;
+
+	for_each_online_cpu(j) {
+		cpu_freq = per_cpu(cpu_last_req_freq, j);
+
+		if ((cluster == per_cpu(physical_cluster, j)) &&
+				(max_freq < cpu_freq))
+			max_freq = cpu_freq;
+	}
+
+	pr_debug("%s: cluster: %d, max freq: %d\n", __func__, cluster,
+			max_freq);
+
+	return max_freq;
+}
+
+static unsigned int clk_get_cpu_rate(unsigned int cpu)
+{
+	u32 cur_cluster = per_cpu(physical_cluster, cpu);
+	u32 rate = clk_get_rate(clk[cur_cluster]) / 1000;
+
+	/* For switcher we use virtual A7 clock rates */
+	if (is_bL_switching_enabled())
+		rate = VIRT_FREQ(cur_cluster, rate);
+
+	pr_debug("%s: cpu: %d, cluster: %d, freq: %u\n", __func__, cpu,
+			cur_cluster, rate);
+
+	return rate;
+}
+
+static unsigned int bL_cpufreq_get_rate(unsigned int cpu)
+{
+	if (is_bL_switching_enabled()) {
+		pr_debug("%s: freq: %d\n", __func__, per_cpu(cpu_last_req_freq,
+					cpu));
+
+		return per_cpu(cpu_last_req_freq, cpu);
+	} else {
+		return clk_get_cpu_rate(cpu);
+	}
+}
 
-static unsigned int bL_cpufreq_get(unsigned int cpu)
+static unsigned int
+bL_cpufreq_set_rate(u32 cpu, u32 old_cluster, u32 new_cluster, u32 rate)
 {
-	u32 cur_cluster = cpu_to_cluster(cpu);
+	u32 new_rate, prev_rate;
+	int ret;
+	bool bLs = is_bL_switching_enabled();
+
+	mutex_lock(&cluster_lock[new_cluster]);
+
+	if (bLs) {
+		prev_rate = per_cpu(cpu_last_req_freq, cpu);
+		per_cpu(cpu_last_req_freq, cpu) = rate;
+		per_cpu(physical_cluster, cpu) = new_cluster;
+
+		new_rate = find_cluster_maxfreq(new_cluster);
+		new_rate = ACTUAL_FREQ(new_cluster, new_rate);
+	} else {
+		new_rate = rate;
+	}
+
+	pr_debug("%s: cpu: %d, old cluster: %d, new cluster: %d, freq: %d\n",
+			__func__, cpu, old_cluster, new_cluster, new_rate);
+
+	ret = clk_set_rate(clk[new_cluster], new_rate * 1000);
+	if (WARN_ON(ret)) {
+		pr_err("clk_set_rate failed: %d, new cluster: %d\n", ret,
+				new_cluster);
+		if (bLs) {
+			per_cpu(cpu_last_req_freq, cpu) = prev_rate;
+			per_cpu(physical_cluster, cpu) = old_cluster;
+		}
+
+		mutex_unlock(&cluster_lock[new_cluster]);
+
+		return ret;
+	}
+
+	mutex_unlock(&cluster_lock[new_cluster]);
+
+	/* Recalc freq for old cluster when switching clusters */
+	if (old_cluster != new_cluster) {
+		pr_debug("%s: cpu: %d, old cluster: %d, new cluster: %d\n",
+				__func__, cpu, old_cluster, new_cluster);
+
+		/* Switch cluster */
+		bL_switch_request(cpu, new_cluster);
+
+		mutex_lock(&cluster_lock[old_cluster]);
 
-	return clk_get_rate(clk[cur_cluster]) / 1000;
+		/* Set freq of old cluster if there are cpus left on it */
+		new_rate = find_cluster_maxfreq(old_cluster);
+		new_rate = ACTUAL_FREQ(old_cluster, new_rate);
+
+		if (new_rate) {
+			pr_debug("%s: Updating rate of old cluster: %d, to freq: %d\n",
+					__func__, old_cluster, new_rate);
+
+			if (clk_set_rate(clk[old_cluster], new_rate * 1000))
+				pr_err("%s: clk_set_rate failed: %d, old cluster: %d\n",
+						__func__, ret, old_cluster);
+		}
+		mutex_unlock(&cluster_lock[old_cluster]);
+	}
+
+	return 0;
 }
 
 /* Set clock frequency */
@@ -52,63 +190,164 @@ static int bL_cpufreq_set_target(struct cpufreq_policy *policy,
 		unsigned int index)
 {
 	struct cpufreq_freqs freqs;
-	u32 cpu = policy->cpu, cur_cluster;
+	u32 cpu = policy->cpu, cur_cluster, new_cluster, actual_cluster;
 	int ret = 0;
 
-	cur_cluster = cpu_to_cluster(policy->cpu);
+	cur_cluster = cpu_to_cluster(cpu);
+	new_cluster = actual_cluster = per_cpu(physical_cluster, cpu);
 
-	freqs.old = bL_cpufreq_get(policy->cpu);
+	freqs.old = bL_cpufreq_get_rate(cpu);
 	freqs.new = freq_table[cur_cluster][index].frequency;
 
 	pr_debug("%s: cpu: %d, cluster: %d, oldfreq: %d, target freq: %d, new freq: %d\n",
 			__func__, cpu, cur_cluster, freqs.old, freqs.new,
 			freqs.new);
 
+	if (is_bL_switching_enabled()) {
+		if ((actual_cluster == A15_CLUSTER) &&
+				(freqs.new < clk_big_min)) {
+			new_cluster = A7_CLUSTER;
+		} else if ((actual_cluster == A7_CLUSTER) &&
+				(freqs.new > clk_little_max)) {
+			new_cluster = A15_CLUSTER;
+		}
+	}
+
 	cpufreq_notify_transition(policy, &freqs, CPUFREQ_PRECHANGE);
 
-	ret = clk_set_rate(clk[cur_cluster], freqs.new * 1000);
-	if (ret) {
-		pr_err("clk_set_rate failed: %d\n", ret);
+	ret = bL_cpufreq_set_rate(cpu, actual_cluster, new_cluster, freqs.new);
+	if (ret)
 		freqs.new = freqs.old;
-	}
 
 	cpufreq_notify_transition(policy, &freqs, CPUFREQ_POSTCHANGE);
 
 	return ret;
 }
 
+static inline u32 get_table_count(struct cpufreq_frequency_table *table)
+{
+	int count;
+
+	for (count = 0; table[count].frequency != CPUFREQ_TABLE_END; count++)
+		;
+
+	return count;
+}
+
+/* get the minimum frequency in the cpufreq_frequency_table */
+static inline u32 get_table_min(struct cpufreq_frequency_table *table)
+{
+	int i;
+	uint32_t min_freq = ~0;
+	for (i = 0; (table[i].frequency != CPUFREQ_TABLE_END); i++)
+		if (table[i].frequency < min_freq)
+			min_freq = table[i].frequency;
+	return min_freq;
+}
+
+/* get the maximum frequency in the cpufreq_frequency_table */
+static inline u32 get_table_max(struct cpufreq_frequency_table *table)
+{
+	int i;
+	uint32_t max_freq = 0;
+	for (i = 0; (table[i].frequency != CPUFREQ_TABLE_END); i++)
+		if (table[i].frequency > max_freq)
+			max_freq = table[i].frequency;
+	return max_freq;
+}
+
+static int merge_cluster_tables(void)
+{
+	int i, j, k = 0, count = 1;
+	struct cpufreq_frequency_table *table;
+
+	for (i = 0; i < MAX_CLUSTERS; i++)
+		count += get_table_count(freq_table[i]);
+
+	table = kzalloc(sizeof(*table) * count, GFP_KERNEL);
+	if (!table)
+		return -ENOMEM;
+
+	freq_table[MAX_CLUSTERS] = table;
+
+	/* Add in reverse order to get freqs in increasing order */
+	for (i = MAX_CLUSTERS - 1; i >= 0; i--) {
+		for (j = 0; freq_table[i][j].frequency != CPUFREQ_TABLE_END;
+				j++) {
+			table[k].frequency = VIRT_FREQ(i,
+					freq_table[i][j].frequency);
+			pr_debug("%s: index: %d, freq: %d\n", __func__, k,
+					table[k].frequency);
+			k++;
+		}
+	}
+
+	table[k].driver_data = k;
+	table[k].frequency = CPUFREQ_TABLE_END;
+
+	pr_debug("%s: End, table: %p, count: %d\n", __func__, table, k);
+
+	return 0;
+}
+
+static void _put_cluster_clk_and_freq_table(struct device *cpu_dev)
+{
+	u32 cluster = raw_cpu_to_cluster(cpu_dev->id);
+
+	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 void put_cluster_clk_and_freq_table(struct device *cpu_dev)
 {
 	u32 cluster = cpu_to_cluster(cpu_dev->id);
+	int i;
+
+	if (atomic_dec_return(&cluster_usage[cluster]))
+		return;
+
+	if (cluster < MAX_CLUSTERS)
+		return _put_cluster_clk_and_freq_table(cpu_dev);
 
-	if (!atomic_dec_return(&cluster_usage[cluster])) {
-		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);
+	for_each_present_cpu(i) {
+		struct device *cdev = get_cpu_device(i);
+		if (!cdev) {
+			pr_err("%s: failed to get cpu%d device\n", __func__, i);
+			return;
+		}
+
+		_put_cluster_clk_and_freq_table(cdev);
 	}
+
+	/* free virtual table */
+	kfree(freq_table[cluster]);
 }
 
-static int get_cluster_clk_and_freq_table(struct device *cpu_dev)
+static int _get_cluster_clk_and_freq_table(struct device *cpu_dev)
 {
-	u32 cluster = cpu_to_cluster(cpu_dev->id);
+	u32 cluster = raw_cpu_to_cluster(cpu_dev->id);
 	char name[14] = "cpu-cluster.";
 	int ret;
 
-	if (atomic_inc_return(&cluster_usage[cluster]) != 1)
+	if (freq_table[cluster])
 		return 0;
 
 	ret = arm_bL_ops->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);
-		goto atomic_dec;
+		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 atomic_dec;
+		goto out;
 	}
 
 	name[12] = cluster + '0';
@@ -125,13 +364,72 @@ static int get_cluster_clk_and_freq_table(struct device *cpu_dev)
 	ret = PTR_ERR(clk[cluster]);
 	dev_pm_opp_free_cpufreq_table(cpu_dev, &freq_table[cluster]);
 
-atomic_dec:
-	atomic_dec(&cluster_usage[cluster]);
+out:
 	dev_err(cpu_dev, "%s: Failed to get data for cluster: %d\n", __func__,
 			cluster);
 	return ret;
 }
 
+static int get_cluster_clk_and_freq_table(struct device *cpu_dev)
+{
+	u32 cluster = cpu_to_cluster(cpu_dev->id);
+	int i, ret;
+
+	if (atomic_inc_return(&cluster_usage[cluster]) != 1)
+		return 0;
+
+	if (cluster < MAX_CLUSTERS) {
+		ret = _get_cluster_clk_and_freq_table(cpu_dev);
+		if (ret)
+			atomic_dec(&cluster_usage[cluster]);
+		return ret;
+	}
+
+	/*
+	 * Get data for all clusters and fill virtual cluster with a merge of
+	 * both
+	 */
+	for_each_present_cpu(i) {
+		struct device *cdev = get_cpu_device(i);
+		if (!cdev) {
+			pr_err("%s: failed to get cpu%d device\n", __func__, i);
+			return -ENODEV;
+		}
+
+		ret = _get_cluster_clk_and_freq_table(cdev);
+		if (ret)
+			goto put_clusters;
+	}
+
+	ret = merge_cluster_tables();
+	if (ret)
+		goto put_clusters;
+
+	/* Assuming 2 cluster, set clk_big_min and clk_little_max */
+	clk_big_min = get_table_min(freq_table[0]);
+	clk_little_max = VIRT_FREQ(1, get_table_max(freq_table[1]));
+
+	pr_debug("%s: cluster: %d, clk_big_min: %d, clk_little_max: %d\n",
+			__func__, cluster, clk_big_min, clk_little_max);
+
+	return 0;
+
+put_clusters:
+	for_each_present_cpu(i) {
+		struct device *cdev = get_cpu_device(i);
+		if (!cdev) {
+			pr_err("%s: failed to get cpu%d device\n", __func__, i);
+			return -ENODEV;
+		}
+
+		_put_cluster_clk_and_freq_table(cdev);
+	}
+
+	atomic_dec(&cluster_usage[cluster]);
+
+	return ret;
+}
+
 /* Per-CPU initialization */
 static int bL_cpufreq_init(struct cpufreq_policy *policy)
 {
@@ -158,13 +456,23 @@ static int bL_cpufreq_init(struct cpufreq_policy *policy)
 		return ret;
 	}
 
+	if (cur_cluster < MAX_CLUSTERS) {
+		cpumask_copy(policy->cpus, topology_core_cpumask(policy->cpu));
+
+		per_cpu(physical_cluster, policy->cpu) = cur_cluster;
+	} else {
+		/* Assumption: during init, we are always running on A15 */
+		per_cpu(physical_cluster, policy->cpu) = A15_CLUSTER;
+	}
+
 	if (arm_bL_ops->get_transition_latency)
 		policy->cpuinfo.transition_latency =
 			arm_bL_ops->get_transition_latency(cpu_dev);
 	else
 		policy->cpuinfo.transition_latency = CPUFREQ_ETERNAL;
 
-	cpumask_copy(policy->cpus, topology_core_cpumask(policy->cpu));
+	if (is_bL_switching_enabled())
+		per_cpu(cpu_last_req_freq, policy->cpu) = clk_get_cpu_rate;
 
 	dev_info(cpu_dev, "%s: CPU %d initialized\n", __func__, policy->cpu);
 	return 0;
@@ -194,7 +502,7 @@ static struct cpufreq_driver bL_cpufreq_driver = {
 					CPUFREQ_HAVE_GOVERNOR_PER_POLICY,
 	.verify			= cpufreq_generic_frequency_table_verify,
 	.target_index		= bL_cpufreq_set_target,
-	.get			= bL_cpufreq_get,
+	.get			= bL_cpufreq_get_rate,
 	.init			= bL_cpufreq_init,
 	.exit			= bL_cpufreq_exit,
 	.attr			= cpufreq_generic_attr,
@@ -202,7 +510,7 @@ static struct cpufreq_driver bL_cpufreq_driver = {
 
 int bL_cpufreq_register(struct cpufreq_arm_bL_ops *ops)
 {
-	int ret;
+	int ret, i;
 
 	if (arm_bL_ops) {
 		pr_debug("%s: Already registered: %s, exiting\n", __func__,
@@ -217,6 +525,9 @@ int bL_cpufreq_register(struct cpufreq_arm_bL_ops *ops)
 
 	arm_bL_ops = ops;
 
+	for (i = 0; i < MAX_CLUSTERS; i++)
+		mutex_init(&cluster_lock[i]);
+
 	ret = cpufreq_register_driver(&bL_cpufreq_driver);
 	if (ret) {
 		pr_info("%s: Failed registering platform driver: %s, err: %d\n",
diff --git a/drivers/cpufreq/arm_big_little.h b/drivers/cpufreq/arm_big_little.h
index 79b2ce1788..70f18fc12d 100644
--- a/drivers/cpufreq/arm_big_little.h
+++ b/drivers/cpufreq/arm_big_little.h
@@ -34,11 +34,6 @@ struct cpufreq_arm_bL_ops {
 	int (*init_opp_table)(struct device *cpu_dev);
 };
 
-static inline int cpu_to_cluster(int cpu)
-{
-	return topology_physical_package_id(cpu);
-}
-
 int bL_cpufreq_register(struct cpufreq_arm_bL_ops *ops);
 void bL_cpufreq_unregister(struct cpufreq_arm_bL_ops *ops);
 
-- 
1.8.4.108.g55ea5f6

[PATCH v2 2/2] cpufreq: arm_big_little: reconfigure switcher behavior at run time

[Nicolas Pitre]

The b.L switcher can be turned on/off at run time.  It is therefore
necessary to change the cpufreq driver behavior accordingly.

The driver must be unregistered/registered with the cpufreq core
to reconfigure freq tables for the virtual or actual CPUs. This is
accomplished via the b.L switcher notifier callback.

Signed-off-by: Nicolas Pitre <nicolas.pitre@linaro.org>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
---
 drivers/cpufreq/arm_big_little.c | 54 +++++++++++++++++++++++++++++++++++++---
 1 file changed, 51 insertions(+), 3 deletions(-)

diff --git a/drivers/cpufreq/arm_big_little.c b/drivers/cpufreq/arm_big_little.c
index a6a8f74811..f63dc8dd80 100644
--- a/drivers/cpufreq/arm_big_little.c
+++ b/drivers/cpufreq/arm_big_little.c
@@ -40,9 +40,12 @@
 #define MAX_CLUSTERS	2
 
 #ifdef CONFIG_BL_SWITCHER
-#define is_bL_switching_enabled()	true
+static bool bL_switching_enabled;
+#define is_bL_switching_enabled()	bL_switching_enabled
+#define set_switching_enabled(x)	(bL_switching_enabled = (x))
 #else
 #define is_bL_switching_enabled()	false
+#define set_switching_enabled(x)	do { } while (0)
 #endif
 
 #define ACTUAL_FREQ(cluster, freq)  ((cluster == A7_CLUSTER) ? freq << 1 : freq)
@@ -508,6 +511,38 @@ static struct cpufreq_driver bL_cpufreq_driver = {
 	.attr			= cpufreq_generic_attr,
 };
 
+static int bL_cpufreq_switcher_notifier(struct notifier_block *nfb,
+					unsigned long action, void *_arg)
+{
+	pr_debug("%s: action: %ld\n", __func__, action);
+
+	switch (action) {
+	case BL_NOTIFY_PRE_ENABLE:
+	case BL_NOTIFY_PRE_DISABLE:
+		cpufreq_unregister_driver(&bL_cpufreq_driver);
+		break;
+
+	case BL_NOTIFY_POST_ENABLE:
+		set_switching_enabled(true);
+		cpufreq_register_driver(&bL_cpufreq_driver);
+		break;
+
+	case BL_NOTIFY_POST_DISABLE:
+		set_switching_enabled(false);
+		cpufreq_register_driver(&bL_cpufreq_driver);
+		break;
+
+	default:
+		return NOTIFY_DONE;
+	}
+
+	return NOTIFY_OK;
+}
+
+static struct notifier_block bL_switcher_notifier = {
+	.notifier_call = bL_cpufreq_switcher_notifier,
+};
+
 int bL_cpufreq_register(struct cpufreq_arm_bL_ops *ops)
 {
 	int ret, i;
@@ -525,6 +560,9 @@ int bL_cpufreq_register(struct cpufreq_arm_bL_ops *ops)
 
 	arm_bL_ops = ops;
 
+	ret = bL_switcher_get_enabled();
+	set_switching_enabled(ret);
+
 	for (i = 0; i < MAX_CLUSTERS; i++)
 		mutex_init(&cluster_lock[i]);
 
@@ -534,10 +572,17 @@ int bL_cpufreq_register(struct cpufreq_arm_bL_ops *ops)
 				__func__, ops->name, ret);
 		arm_bL_ops = NULL;
 	} else {
-		pr_info("%s: Registered platform driver: %s\n", __func__,
-				ops->name);
+		ret = bL_switcher_register_notifier(&bL_switcher_notifier);
+		if (ret) {
+			cpufreq_unregister_driver(&bL_cpufreq_driver);
+			arm_bL_ops = NULL;
+		} else {
+			pr_info("%s: Registered platform driver: %s\n",
+					__func__, ops->name);
+		}
 	}
 
+	bL_switcher_put_enabled();
 	return ret;
 }
 EXPORT_SYMBOL_GPL(bL_cpufreq_register);
@@ -550,7 +595,10 @@ void bL_cpufreq_unregister(struct cpufreq_arm_bL_ops *ops)
 		return;
 	}
 
+	bL_switcher_get_enabled();
+	bL_switcher_unregister_notifier(&bL_switcher_notifier);
 	cpufreq_unregister_driver(&bL_cpufreq_driver);
+	bL_switcher_put_enabled();
 	pr_info("%s: Un-registered platform driver: %s\n", __func__,
 			arm_bL_ops->name);
 	arm_bL_ops = NULL;
-- 
1.8.4.108.g55ea5f6

[PATCH v2 2/2] cpufreq: arm_big_little: reconfigure switcher behavior at run time

[Nicolas Pitre]

The b.L switcher can be turned on/off at run time.  It is therefore
necessary to change the cpufreq driver behavior accordingly.

The driver must be unregistered/registered with the cpufreq core
to reconfigure freq tables for the virtual or actual CPUs. This is
accomplished via the b.L switcher notifier callback.

Signed-off-by: Nicolas Pitre <nicolas.pitre@linaro.org>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
---
 drivers/cpufreq/arm_big_little.c | 54 +++++++++++++++++++++++++++++++++++++---
 1 file changed, 51 insertions(+), 3 deletions(-)

diff --git a/drivers/cpufreq/arm_big_little.c b/drivers/cpufreq/arm_big_little.c
index a6a8f74811..f63dc8dd80 100644
--- a/drivers/cpufreq/arm_big_little.c
+++ b/drivers/cpufreq/arm_big_little.c
@@ -40,9 +40,12 @@
 #define MAX_CLUSTERS	2
 
 #ifdef CONFIG_BL_SWITCHER
-#define is_bL_switching_enabled()	true
+static bool bL_switching_enabled;
+#define is_bL_switching_enabled()	bL_switching_enabled
+#define set_switching_enabled(x)	(bL_switching_enabled = (x))
 #else
 #define is_bL_switching_enabled()	false
+#define set_switching_enabled(x)	do { } while (0)
 #endif
 
 #define ACTUAL_FREQ(cluster, freq)  ((cluster == A7_CLUSTER) ? freq << 1 : freq)
@@ -508,6 +511,38 @@ static struct cpufreq_driver bL_cpufreq_driver = {
 	.attr			= cpufreq_generic_attr,
 };
 
+static int bL_cpufreq_switcher_notifier(struct notifier_block *nfb,
+					unsigned long action, void *_arg)
+{
+	pr_debug("%s: action: %ld\n", __func__, action);
+
+	switch (action) {
+	case BL_NOTIFY_PRE_ENABLE:
+	case BL_NOTIFY_PRE_DISABLE:
+		cpufreq_unregister_driver(&bL_cpufreq_driver);
+		break;
+
+	case BL_NOTIFY_POST_ENABLE:
+		set_switching_enabled(true);
+		cpufreq_register_driver(&bL_cpufreq_driver);
+		break;
+
+	case BL_NOTIFY_POST_DISABLE:
+		set_switching_enabled(false);
+		cpufreq_register_driver(&bL_cpufreq_driver);
+		break;
+
+	default:
+		return NOTIFY_DONE;
+	}
+
+	return NOTIFY_OK;
+}
+
+static struct notifier_block bL_switcher_notifier = {
+	.notifier_call = bL_cpufreq_switcher_notifier,
+};
+
 int bL_cpufreq_register(struct cpufreq_arm_bL_ops *ops)
 {
 	int ret, i;
@@ -525,6 +560,9 @@ int bL_cpufreq_register(struct cpufreq_arm_bL_ops *ops)
 
 	arm_bL_ops = ops;
 
+	ret = bL_switcher_get_enabled();
+	set_switching_enabled(ret);
+
 	for (i = 0; i < MAX_CLUSTERS; i++)
 		mutex_init(&cluster_lock[i]);
 
@@ -534,10 +572,17 @@ int bL_cpufreq_register(struct cpufreq_arm_bL_ops *ops)
 				__func__, ops->name, ret);
 		arm_bL_ops = NULL;
 	} else {
-		pr_info("%s: Registered platform driver: %s\n", __func__,
-				ops->name);
+		ret = bL_switcher_register_notifier(&bL_switcher_notifier);
+		if (ret) {
+			cpufreq_unregister_driver(&bL_cpufreq_driver);
+			arm_bL_ops = NULL;
+		} else {
+			pr_info("%s: Registered platform driver: %s\n",
+					__func__, ops->name);
+		}
 	}
 
+	bL_switcher_put_enabled();
 	return ret;
 }
 EXPORT_SYMBOL_GPL(bL_cpufreq_register);
@@ -550,7 +595,10 @@ void bL_cpufreq_unregister(struct cpufreq_arm_bL_ops *ops)
 		return;
 	}
 
+	bL_switcher_get_enabled();
+	bL_switcher_unregister_notifier(&bL_switcher_notifier);
 	cpufreq_unregister_driver(&bL_cpufreq_driver);
+	bL_switcher_put_enabled();
 	pr_info("%s: Un-registered platform driver: %s\n", __func__,
 			arm_bL_ops->name);
 	arm_bL_ops = NULL;
-- 
1.8.4.108.g55ea5f6

Re: [PATCH v2 0/2] cpufreq support for the big.LITTLE switcher

[Rafael J. Wysocki]

On Wednesday, October 30, 2013 03:44:39 PM Nicolas Pitre wrote:
> This is the third and final set of patches towards a fully functional
> and production quality switcher solution for big.LITTLE systems,
> establishing a landmark to compare against for any scheduler based
> solution meant to eventually surpass the switcher's power efficiency
> available in the mainline kernel.
> 
> Rationale for this code: http://lwn.net/Articles/481055/
> 
> The first and second patch sets have already been merged by RMK.  They
> implement the core switcher mechanism.  This set adds the necessary code
> to drive the switcher based on cpufreq governor decisions.
> 
> This set (v2) was rebased on top of the latest linux-pm tree as
> requested by Rafael.

Queued up for 3.13, thanks!

-- 
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.

[PATCH v2 0/2] cpufreq support for the big.LITTLE switcher

[Rafael J. Wysocki]

On Wednesday, October 30, 2013 03:44:39 PM Nicolas Pitre wrote:
> This is the third and final set of patches towards a fully functional
> and production quality switcher solution for big.LITTLE systems,
> establishing a landmark to compare against for any scheduler based
> solution meant to eventually surpass the switcher's power efficiency
> available in the mainline kernel.
> 
> Rationale for this code: http://lwn.net/Articles/481055/
> 
> The first and second patch sets have already been merged by RMK.  They
> implement the core switcher mechanism.  This set adds the necessary code
> to drive the switcher based on cpufreq governor decisions.
> 
> This set (v2) was rebased on top of the latest linux-pm tree as
> requested by Rafael.

Queued up for 3.13, thanks!

-- 
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.

Re: [PATCH v2 0/2] cpufreq support for the big.LITTLE switcher

[Viresh Kumar]

[-- Attachment #1: Type: text/plain, Size: 1649 bytes --]

On 31 October 2013 03:22, Rafael J. Wysocki <rjw@rjwysocki.net> wrote:
> Queued up for 3.13, thanks!

Hmm.. thanks for applying this but there was a minor bug here which
is fixed by below commit.. You don't really need to merge this with
the original commit and so can go as a separate commit..

Attached for applying.. Tested by Nico.

--------x------------------x-------------

Author: Viresh Kumar <viresh.kumar@linaro.org>
Date:   Thu Oct 31 03:45:42 2013 +0530

    cpufreq: arm_big_little: Call routine instead of passing its address

    In bL_cpufreq_init() we need to fill per-cpu cpu_last_req_freq
with the current
    frequency of cpu. Instead of calling the function we passed its
address to this
    routine by mistake earlier. Lets fix it by calling the routine instead.

    Tested-by: Nicolas Pitre <nico@linaro.org>
    Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
---
 drivers/cpufreq/arm_big_little.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/cpufreq/arm_big_little.c b/drivers/cpufreq/arm_big_little.c
index 2b2a44a..ee66449 100644
--- a/drivers/cpufreq/arm_big_little.c
+++ b/drivers/cpufreq/arm_big_little.c
@@ -461,7 +461,8 @@ static int bL_cpufreq_init(struct cpufreq_policy *policy)
                policy->cpuinfo.transition_latency = CPUFREQ_ETERNAL;

        if (is_bL_switching_enabled())
-               per_cpu(cpu_last_req_freq, policy->cpu) = clk_get_cpu_rate;
+               per_cpu(cpu_last_req_freq, policy->cpu) =
+                       clk_get_cpu_rate(policy->cpu);

        dev_info(cpu_dev, "%s: CPU %d initialized\n", __func__, policy->cpu);
        return 0;

[-- Attachment #2: 0001-cpufreq-arm_big_little-Call-routine-instead-of-passi.patch --]
[-- Type: text/x-patch, Size: 1377 bytes --]

From 88c0ffaec1df0f0ec24d095159459d2554780d9d Mon Sep 17 00:00:00 2001
Message-Id: <88c0ffaec1df0f0ec24d095159459d2554780d9d.1383172083.git.viresh.kumar@linaro.org>
From: Viresh Kumar <viresh.kumar@linaro.org>
Date: Thu, 31 Oct 2013 03:45:42 +0530
Subject: [PATCH] cpufreq: arm_big_little: Call routine instead of passing its
 address

In bL_cpufreq_init() we need to fill per-cpu cpu_last_req_freq with the current
frequency of cpu. Instead of calling the function we passed its address to this
routine by mistake earlier. Lets fix it by calling the routine instead.

Tested-by: Nicolas Pitre <nico@linaro.org>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
---
 drivers/cpufreq/arm_big_little.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/cpufreq/arm_big_little.c b/drivers/cpufreq/arm_big_little.c
index 2b2a44a..ee66449 100644
--- a/drivers/cpufreq/arm_big_little.c
+++ b/drivers/cpufreq/arm_big_little.c
@@ -461,7 +461,8 @@ static int bL_cpufreq_init(struct cpufreq_policy *policy)
 		policy->cpuinfo.transition_latency = CPUFREQ_ETERNAL;
 
 	if (is_bL_switching_enabled())
-		per_cpu(cpu_last_req_freq, policy->cpu) = clk_get_cpu_rate;
+		per_cpu(cpu_last_req_freq, policy->cpu) =
+			clk_get_cpu_rate(policy->cpu);
 
 	dev_info(cpu_dev, "%s: CPU %d initialized\n", __func__, policy->cpu);
 	return 0;
-- 
1.7.12.rc2.18.g61b472e

[PATCH v2 0/2] cpufreq support for the big.LITTLE switcher

[Viresh Kumar]

On 31 October 2013 03:22, Rafael J. Wysocki <rjw@rjwysocki.net> wrote:
> Queued up for 3.13, thanks!

Hmm.. thanks for applying this but there was a minor bug here which
is fixed by below commit.. You don't really need to merge this with
the original commit and so can go as a separate commit..

Attached for applying.. Tested by Nico.

--------x------------------x-------------

Author: Viresh Kumar <viresh.kumar@linaro.org>
Date:   Thu Oct 31 03:45:42 2013 +0530

    cpufreq: arm_big_little: Call routine instead of passing its address

    In bL_cpufreq_init() we need to fill per-cpu cpu_last_req_freq
with the current
    frequency of cpu. Instead of calling the function we passed its
address to this
    routine by mistake earlier. Lets fix it by calling the routine instead.

    Tested-by: Nicolas Pitre <nico@linaro.org>
    Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
---
 drivers/cpufreq/arm_big_little.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/cpufreq/arm_big_little.c b/drivers/cpufreq/arm_big_little.c
index 2b2a44a..ee66449 100644
--- a/drivers/cpufreq/arm_big_little.c
+++ b/drivers/cpufreq/arm_big_little.c
@@ -461,7 +461,8 @@ static int bL_cpufreq_init(struct cpufreq_policy *policy)
                policy->cpuinfo.transition_latency = CPUFREQ_ETERNAL;

        if (is_bL_switching_enabled())
-               per_cpu(cpu_last_req_freq, policy->cpu) = clk_get_cpu_rate;
+               per_cpu(cpu_last_req_freq, policy->cpu) =
+                       clk_get_cpu_rate(policy->cpu);

        dev_info(cpu_dev, "%s: CPU %d initialized\n", __func__, policy->cpu);
        return 0;
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-cpufreq-arm_big_little-Call-routine-instead-of-passi.patch
Type: text/x-patch
Size: 1377 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20131031/ce1660dd/attachment.bin>

Re: [PATCH v2 0/2] cpufreq support for the big.LITTLE switcher

[Viresh Kumar]

On 31 October 2013 04:55, Rafael J. Wysocki <rjw@rjwysocki.net> wrote:
> I've just fixed it up manually, please check the result in bleeding-edge.

It crossed 80 columns, otherwise its fine :)

[PATCH v2 0/2] cpufreq support for the big.LITTLE switcher

[Viresh Kumar]

On 31 October 2013 04:55, Rafael J. Wysocki <rjw@rjwysocki.net> wrote:
> I've just fixed it up manually, please check the result in bleeding-edge.

It crossed 80 columns, otherwise its fine :)

Re: [PATCH v2 0/2] cpufreq support for the big.LITTLE switcher

[Rafael J. Wysocki]

On Thursday, October 31, 2013 04:01:13 AM Viresh Kumar wrote:
> 
> --047d7b33ce860f693404e9fce430
> Content-Type: text/plain; charset=ISO-8859-1
> 
> On 31 October 2013 03:22, Rafael J. Wysocki <rjw@rjwysocki.net> wrote:
> > Queued up for 3.13, thanks!
> 
> Hmm.. thanks for applying this but there was a minor bug here which
> is fixed by below commit.. You don't really need to merge this with
> the original commit and so can go as a separate commit..
> 
> Attached for applying.. Tested by Nico.

I've just fixed it up manually, please check the result in bleeding-edge.

Thanks!

-- 
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.

[PATCH v2 0/2] cpufreq support for the big.LITTLE switcher

[Rafael J. Wysocki]

On Thursday, October 31, 2013 04:01:13 AM Viresh Kumar wrote:
> 
> --047d7b33ce860f693404e9fce430
> Content-Type: text/plain; charset=ISO-8859-1
> 
> On 31 October 2013 03:22, Rafael J. Wysocki <rjw@rjwysocki.net> wrote:
> > Queued up for 3.13, thanks!
> 
> Hmm.. thanks for applying this but there was a minor bug here which
> is fixed by below commit.. You don't really need to merge this with
> the original commit and so can go as a separate commit..
> 
> Attached for applying.. Tested by Nico.

I've just fixed it up manually, please check the result in bleeding-edge.

Thanks!

-- 
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.