cpufreq: arm_big_little: add in-kernel switching (IKS) support

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 KarkadaNagesha.

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>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

2 files changed, 337 insertions, 31 deletions

diff --git a/drivers/cpufreq/arm_big_little.c b/drivers/cpufreq/arm_big_little.c
index 163e3378fe17..9986f7912328 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(policy->cpu);
 
 	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 79b2ce17884d..70f18fc12d4a 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);