[linux-2.6-block.git] / kernel / power / energy_model.c

// SPDX-License-Identifier: GPL-2.0
/*
 * Energy Model of devices
 *
 * Copyright (c) 2018-2021, Arm ltd.
 * Written by: Quentin Perret, Arm ltd.
 * Improvements provided by: Lukasz Luba, Arm ltd.
 */

#define pr_fmt(fmt) "energy_model: " fmt

#include <linux/cpu.h>
#include <linux/cpufreq.h>
#include <linux/cpumask.h>
#include <linux/debugfs.h>
#include <linux/energy_model.h>
#include <linux/sched/topology.h>
#include <linux/slab.h>

/*
 * Mutex serializing the registrations of performance domains and letting
 * callbacks defined by drivers sleep.
 */
static DEFINE_MUTEX(em_pd_mutex);

static bool _is_cpu_device(struct device *dev)
{
	return (dev->bus == &cpu_subsys);
}

#ifdef CONFIG_DEBUG_FS
static struct dentry *rootdir;

static void em_debug_create_ps(struct em_perf_state *ps, struct dentry *pd)
{
	struct dentry *d;
	char name[24];

	snprintf(name, sizeof(name), "ps:%lu", ps->frequency);

	/* Create per-ps directory */
	d = debugfs_create_dir(name, pd);
	debugfs_create_ulong("frequency", 0444, d, &ps->frequency);
	debugfs_create_ulong("power", 0444, d, &ps->power);
	debugfs_create_ulong("cost", 0444, d, &ps->cost);
	debugfs_create_ulong("inefficient", 0444, d, &ps->flags);
}

static int em_debug_cpus_show(struct seq_file *s, void *unused)
{
	seq_printf(s, "%*pbl\n", cpumask_pr_args(to_cpumask(s->private)));

	return 0;
}
DEFINE_SHOW_ATTRIBUTE(em_debug_cpus);

static int em_debug_flags_show(struct seq_file *s, void *unused)
{
	struct em_perf_domain *pd = s->private;

	seq_printf(s, "%#lx\n", pd->flags);

	return 0;
}
DEFINE_SHOW_ATTRIBUTE(em_debug_flags);

static void em_debug_create_pd(struct device *dev)
{
	struct dentry *d;
	int i;

	/* Create the directory of the performance domain */
	d = debugfs_create_dir(dev_name(dev), rootdir);

	if (_is_cpu_device(dev))
		debugfs_create_file("cpus", 0444, d, dev->em_pd->cpus,
				    &em_debug_cpus_fops);

	debugfs_create_file("flags", 0444, d, dev->em_pd,
			    &em_debug_flags_fops);

	/* Create a sub-directory for each performance state */
	for (i = 0; i < dev->em_pd->nr_perf_states; i++)
		em_debug_create_ps(&dev->em_pd->table[i], d);

}

static void em_debug_remove_pd(struct device *dev)
{
	debugfs_lookup_and_remove(dev_name(dev), rootdir);
}

static int __init em_debug_init(void)
{
	/* Create /sys/kernel/debug/energy_model directory */
	rootdir = debugfs_create_dir("energy_model", NULL);

	return 0;
}
fs_initcall(em_debug_init);
#else /* CONFIG_DEBUG_FS */
static void em_debug_create_pd(struct device *dev) {}
static void em_debug_remove_pd(struct device *dev) {}
#endif

static int em_create_perf_table(struct device *dev, struct em_perf_domain *pd,
				int nr_states, struct em_data_callback *cb,
				unsigned long flags)
{
	unsigned long power, freq, prev_freq = 0, prev_cost = ULONG_MAX;
	struct em_perf_state *table;
	int i, ret;
	u64 fmax;

	table = kcalloc(nr_states, sizeof(*table), GFP_KERNEL);
	if (!table)
		return -ENOMEM;

	/* Build the list of performance states for this performance domain */
	for (i = 0, freq = 0; i < nr_states; i++, freq++) {
		/*
		 * active_power() is a driver callback which ceils 'freq' to
		 * lowest performance state of 'dev' above 'freq' and updates
		 * 'power' and 'freq' accordingly.
		 */
		ret = cb->active_power(dev, &power, &freq);
		if (ret) {
			dev_err(dev, "EM: invalid perf. state: %d\n",
				ret);
			goto free_ps_table;
		}

		/*
		 * We expect the driver callback to increase the frequency for
		 * higher performance states.
		 */
		if (freq <= prev_freq) {
			dev_err(dev, "EM: non-increasing freq: %lu\n",
				freq);
			goto free_ps_table;
		}

		/*
		 * The power returned by active_state() is expected to be
		 * positive and be in range.
		 */
		if (!power || power > EM_MAX_POWER) {
			dev_err(dev, "EM: invalid power: %lu\n",
				power);
			goto free_ps_table;
		}

		table[i].power = power;
		table[i].frequency = prev_freq = freq;
	}

	/* Compute the cost of each performance state. */
	fmax = (u64) table[nr_states - 1].frequency;
	for (i = nr_states - 1; i >= 0; i--) {
		unsigned long power_res, cost;

		if (flags & EM_PERF_DOMAIN_ARTIFICIAL) {
			ret = cb->get_cost(dev, table[i].frequency, &cost);
			if (ret || !cost || cost > EM_MAX_POWER) {
				dev_err(dev, "EM: invalid cost %lu %d\n",
					cost, ret);
				goto free_ps_table;
			}
		} else {
			power_res = table[i].power;
			cost = div64_u64(fmax * power_res, table[i].frequency);
		}

		table[i].cost = cost;

		if (table[i].cost >= prev_cost) {
			table[i].flags = EM_PERF_STATE_INEFFICIENT;
			dev_dbg(dev, "EM: OPP:%lu is inefficient\n",
				table[i].frequency);
		} else {
			prev_cost = table[i].cost;
		}
	}

	pd->table = table;
	pd->nr_perf_states = nr_states;

	return 0;

free_ps_table:
	kfree(table);
	return -EINVAL;
}

static int em_create_pd(struct device *dev, int nr_states,
			struct em_data_callback *cb, cpumask_t *cpus,
			unsigned long flags)
{
	struct em_perf_domain *pd;
	struct device *cpu_dev;
	int cpu, ret, num_cpus;

	if (_is_cpu_device(dev)) {
		num_cpus = cpumask_weight(cpus);

		/* Prevent max possible energy calculation to not overflow */
		if (num_cpus > EM_MAX_NUM_CPUS) {
			dev_err(dev, "EM: too many CPUs, overflow possible\n");
			return -EINVAL;
		}

		pd = kzalloc(sizeof(*pd) + cpumask_size(), GFP_KERNEL);
		if (!pd)
			return -ENOMEM;

		cpumask_copy(em_span_cpus(pd), cpus);
	} else {
		pd = kzalloc(sizeof(*pd), GFP_KERNEL);
		if (!pd)
			return -ENOMEM;
	}

	ret = em_create_perf_table(dev, pd, nr_states, cb, flags);
	if (ret) {
		kfree(pd);
		return ret;
	}

	if (_is_cpu_device(dev))
		for_each_cpu(cpu, cpus) {
			cpu_dev = get_cpu_device(cpu);
			cpu_dev->em_pd = pd;
		}

	dev->em_pd = pd;

	return 0;
}

static void em_cpufreq_update_efficiencies(struct device *dev)
{
	struct em_perf_domain *pd = dev->em_pd;
	struct em_perf_state *table;
	struct cpufreq_policy *policy;
	int found = 0;
	int i;

	if (!_is_cpu_device(dev) || !pd)
		return;

	policy = cpufreq_cpu_get(cpumask_first(em_span_cpus(pd)));
	if (!policy) {
		dev_warn(dev, "EM: Access to CPUFreq policy failed");
		return;
	}

	table = pd->table;

	for (i = 0; i < pd->nr_perf_states; i++) {
		if (!(table[i].flags & EM_PERF_STATE_INEFFICIENT))
			continue;

		if (!cpufreq_table_set_inefficient(policy, table[i].frequency))
			found++;
	}

	cpufreq_cpu_put(policy);

	if (!found)
		return;

	/*
	 * Efficiencies have been installed in CPUFreq, inefficient frequencies
	 * will be skipped. The EM can do the same.
	 */
	pd->flags |= EM_PERF_DOMAIN_SKIP_INEFFICIENCIES;
}

/**
 * em_pd_get() - Return the performance domain for a device
 * @dev : Device to find the performance domain for
 *
 * Returns the performance domain to which @dev belongs, or NULL if it doesn't
 * exist.
 */
struct em_perf_domain *em_pd_get(struct device *dev)
{
	if (IS_ERR_OR_NULL(dev))
		return NULL;

	return dev->em_pd;
}
EXPORT_SYMBOL_GPL(em_pd_get);

/**
 * em_cpu_get() - Return the performance domain for a CPU
 * @cpu : CPU to find the performance domain for
 *
 * Returns the performance domain to which @cpu belongs, or NULL if it doesn't
 * exist.
 */
struct em_perf_domain *em_cpu_get(int cpu)
{
	struct device *cpu_dev;

	cpu_dev = get_cpu_device(cpu);
	if (!cpu_dev)
		return NULL;

	return em_pd_get(cpu_dev);
}
EXPORT_SYMBOL_GPL(em_cpu_get);

/**
 * em_dev_register_perf_domain() - Register the Energy Model (EM) for a device
 * @dev		: Device for which the EM is to register
 * @nr_states	: Number of performance states to register
 * @cb		: Callback functions providing the data of the Energy Model
 * @cpus	: Pointer to cpumask_t, which in case of a CPU device is
 *		obligatory. It can be taken from i.e. 'policy->cpus'. For other
 *		type of devices this should be set to NULL.
 * @microwatts	: Flag indicating that the power values are in micro-Watts or
 *		in some other scale. It must be set properly.
 *
 * Create Energy Model tables for a performance domain using the callbacks
 * defined in cb.
 *
 * The @microwatts is important to set with correct value. Some kernel
 * sub-systems might rely on this flag and check if all devices in the EM are
 * using the same scale.
 *
 * If multiple clients register the same performance domain, all but the first
 * registration will be ignored.
 *
 * Return 0 on success
 */
int em_dev_register_perf_domain(struct device *dev, unsigned int nr_states,
				struct em_data_callback *cb, cpumask_t *cpus,
				bool microwatts)
{
	unsigned long cap, prev_cap = 0;
	unsigned long flags = 0;
	int cpu, ret;

	if (!dev || !nr_states || !cb)
		return -EINVAL;

	/*
	 * Use a mutex to serialize the registration of performance domains and
	 * let the driver-defined callback functions sleep.
	 */
	mutex_lock(&em_pd_mutex);

	if (dev->em_pd) {
		ret = -EEXIST;
		goto unlock;
	}

	if (_is_cpu_device(dev)) {
		if (!cpus) {
			dev_err(dev, "EM: invalid CPU mask\n");
			ret = -EINVAL;
			goto unlock;
		}

		for_each_cpu(cpu, cpus) {
			if (em_cpu_get(cpu)) {
				dev_err(dev, "EM: exists for CPU%d\n", cpu);
				ret = -EEXIST;
				goto unlock;
			}
			/*
			 * All CPUs of a domain must have the same
			 * micro-architecture since they all share the same
			 * table.
			 */
			cap = arch_scale_cpu_capacity(cpu);
			if (prev_cap && prev_cap != cap) {
				dev_err(dev, "EM: CPUs of %*pbl must have the same capacity\n",
					cpumask_pr_args(cpus));

				ret = -EINVAL;
				goto unlock;
			}
			prev_cap = cap;
		}
	}

	if (microwatts)
		flags |= EM_PERF_DOMAIN_MICROWATTS;
	else if (cb->get_cost)
		flags |= EM_PERF_DOMAIN_ARTIFICIAL;

	ret = em_create_pd(dev, nr_states, cb, cpus, flags);
	if (ret)
		goto unlock;

	dev->em_pd->flags |= flags;

	em_cpufreq_update_efficiencies(dev);

	em_debug_create_pd(dev);
	dev_info(dev, "EM: created perf domain\n");

unlock:
	mutex_unlock(&em_pd_mutex);
	return ret;
}
EXPORT_SYMBOL_GPL(em_dev_register_perf_domain);

/**
 * em_dev_unregister_perf_domain() - Unregister Energy Model (EM) for a device
 * @dev		: Device for which the EM is registered
 *
 * Unregister the EM for the specified @dev (but not a CPU device).
 */
void em_dev_unregister_perf_domain(struct device *dev)
{
	if (IS_ERR_OR_NULL(dev) || !dev->em_pd)
		return;

	if (_is_cpu_device(dev))
		return;

	/*
	 * The mutex separates all register/unregister requests and protects
	 * from potential clean-up/setup issues in the debugfs directories.
	 * The debugfs directory name is the same as device's name.
	 */
	mutex_lock(&em_pd_mutex);
	em_debug_remove_pd(dev);

	kfree(dev->em_pd->table);
	kfree(dev->em_pd);
	dev->em_pd = NULL;
	mutex_unlock(&em_pd_mutex);
}
EXPORT_SYMBOL_GPL(em_dev_unregister_perf_domain);
Commit	Line	Data
27871f7a QP	1	// SPDX-License-Identifier: GPL-2.0
27871f7a QP	2	/*
1bc138c6	3	* Energy Model of devices
27871f7a	4	*
c8ed9953	5	* Copyright (c) 2018-2021, Arm ltd.
27871f7a	6	* Written by: Quentin Perret, Arm ltd.
1bc138c6	7	* Improvements provided by: Lukasz Luba, Arm ltd.
27871f7a QP	8	*/
	9
	10	#define pr_fmt(fmt) "energy_model: " fmt
	11
	12	#include <linux/cpu.h>
e458716a	13	#include <linux/cpufreq.h>
27871f7a	14	#include <linux/cpumask.h>
9cac42d0	15	#include <linux/debugfs.h>
27871f7a QP	16	#include <linux/energy_model.h>
	17	#include <linux/sched/topology.h>
	18	#include <linux/slab.h>
	19
27871f7a QP	20	/*
	21	* Mutex serializing the registrations of performance domains and letting
	22	* callbacks defined by drivers sleep.
	23	*/
	24	static DEFINE_MUTEX(em_pd_mutex);
	25
1bc138c6 LL	26	static bool _is_cpu_device(struct device *dev)
	27	{
	28	return (dev->bus == &cpu_subsys);
	29	}
	30
9cac42d0 QP	31	#ifdef CONFIG_DEBUG_FS
	32	static struct dentry *rootdir;
	33
521b512b	34	static void em_debug_create_ps(struct em_perf_state ps, struct dentry pd)
9cac42d0 QP	35	{
	36	struct dentry *d;
	37	char name[24];
	38
521b512b	39	snprintf(name, sizeof(name), "ps:%lu", ps->frequency);
9cac42d0	40
521b512b	41	/* Create per-ps directory */
9cac42d0	42	d = debugfs_create_dir(name, pd);
521b512b LL	43	debugfs_create_ulong("frequency", 0444, d, &ps->frequency);
	44	debugfs_create_ulong("power", 0444, d, &ps->power);
	45	debugfs_create_ulong("cost", 0444, d, &ps->cost);
c8ed9953	46	debugfs_create_ulong("inefficient", 0444, d, &ps->flags);
9cac42d0 QP	47	}
	48
	49	static int em_debug_cpus_show(struct seq_file s, void unused)
	50	{
	51	seq_printf(s, "%*pbl\n", cpumask_pr_args(to_cpumask(s->private)));
	52
	53	return 0;
	54	}
	55	DEFINE_SHOW_ATTRIBUTE(em_debug_cpus);
	56
16857482	57	static int em_debug_flags_show(struct seq_file s, void unused)
c250d50f LL	58	{
c250d50f LL	59	struct em_perf_domain *pd = s->private;
c250d50f	60
16857482	61	seq_printf(s, "%#lx\n", pd->flags);
c250d50f LL	62
	63	return 0;
	64	}
16857482	65	DEFINE_SHOW_ATTRIBUTE(em_debug_flags);
8354eb9e	66
1bc138c6	67	static void em_debug_create_pd(struct device *dev)
9cac42d0 QP	68	{
9cac42d0 QP	69	struct dentry *d;
9cac42d0 QP	70	int i;
9cac42d0 QP	71
9cac42d0	72	/* Create the directory of the performance domain */
1bc138c6	73	d = debugfs_create_dir(dev_name(dev), rootdir);
9cac42d0	74
1bc138c6 LL	75	if (_is_cpu_device(dev))
	76	debugfs_create_file("cpus", 0444, d, dev->em_pd->cpus,
	77	&em_debug_cpus_fops);
9cac42d0	78
16857482 LL	79	debugfs_create_file("flags", 0444, d, dev->em_pd,
16857482 LL	80	&em_debug_flags_fops);
c250d50f	81
521b512b	82	/* Create a sub-directory for each performance state */
1bc138c6 LL	83	for (i = 0; i < dev->em_pd->nr_perf_states; i++)
	84	em_debug_create_ps(&dev->em_pd->table[i], d);
	85
	86	}
	87
	88	static void em_debug_remove_pd(struct device *dev)
	89	{
a0e8c13c	90	debugfs_lookup_and_remove(dev_name(dev), rootdir);
9cac42d0 QP	91	}
	92
	93	static int __init em_debug_init(void)
	94	{
	95	/* Create /sys/kernel/debug/energy_model directory */
	96	rootdir = debugfs_create_dir("energy_model", NULL);
	97
	98	return 0;
	99	}
fb9d62b2	100	fs_initcall(em_debug_init);
9cac42d0	101	#else /* CONFIG_DEBUG_FS */
1bc138c6 LL	102	static void em_debug_create_pd(struct device *dev) {}
1bc138c6 LL	103	static void em_debug_remove_pd(struct device *dev) {}
9cac42d0	104	#endif
1bc138c6 LL	105
1bc138c6 LL	106	static int em_create_perf_table(struct device dev, struct em_perf_domain pd,
91362463 LL	107	int nr_states, struct em_data_callback *cb,
91362463 LL	108	unsigned long flags)
27871f7a	109	{
aa1a4326	110	unsigned long power, freq, prev_freq = 0, prev_cost = ULONG_MAX;
521b512b	111	struct em_perf_state *table;
1bc138c6	112	int i, ret;
27871f7a QP	113	u64 fmax;
27871f7a QP	114
27871f7a QP	115	table = kcalloc(nr_states, sizeof(*table), GFP_KERNEL);
27871f7a QP	116	if (!table)
1bc138c6	117	return -ENOMEM;
27871f7a	118
521b512b	119	/* Build the list of performance states for this performance domain */
27871f7a QP	120	for (i = 0, freq = 0; i < nr_states; i++, freq++) {
	121	/*
	122	* active_power() is a driver callback which ceils 'freq' to
1bc138c6	123	* lowest performance state of 'dev' above 'freq' and updates
27871f7a QP	124	* 'power' and 'freq' accordingly.
27871f7a QP	125	*/
75a3a99a	126	ret = cb->active_power(dev, &power, &freq);
27871f7a	127	if (ret) {
1bc138c6 LL	128	dev_err(dev, "EM: invalid perf. state: %d\n",
1bc138c6 LL	129	ret);
521b512b	130	goto free_ps_table;
27871f7a QP	131	}
	132
	133	/*
	134	* We expect the driver callback to increase the frequency for
521b512b	135	* higher performance states.
27871f7a QP	136	*/
27871f7a QP	137	if (freq <= prev_freq) {
1bc138c6 LL	138	dev_err(dev, "EM: non-increasing freq: %lu\n",
1bc138c6 LL	139	freq);
521b512b	140	goto free_ps_table;
27871f7a QP	141	}
	142
	143	/*
	144	* The power returned by active_state() is expected to be
ae6ccaa6	145	* positive and be in range.
27871f7a	146	*/
7d9895c7	147	if (!power \|\| power > EM_MAX_POWER) {
1bc138c6 LL	148	dev_err(dev, "EM: invalid power: %lu\n",
1bc138c6 LL	149	power);
521b512b	150	goto free_ps_table;
27871f7a QP	151	}
	152
	153	table[i].power = power;
	154	table[i].frequency = prev_freq = freq;
27871f7a QP	155	}
27871f7a QP	156
521b512b	157	/* Compute the cost of each performance state. */
27871f7a	158	fmax = (u64) table[nr_states - 1].frequency;
aa1a4326	159	for (i = nr_states - 1; i >= 0; i--) {
91362463 LL	160	unsigned long power_res, cost;
	161
	162	if (flags & EM_PERF_DOMAIN_ARTIFICIAL) {
	163	ret = cb->get_cost(dev, table[i].frequency, &cost);
	164	if (ret \|\| !cost \|\| cost > EM_MAX_POWER) {
	165	dev_err(dev, "EM: invalid cost %lu %d\n",
	166	cost, ret);
	167	goto free_ps_table;
	168	}
	169	} else {
ae6ccaa6	170	power_res = table[i].power;
91362463 LL	171	cost = div64_u64(fmax * power_res, table[i].frequency);
	172	}
	173
	174	table[i].cost = cost;
7fcc17d0	175
aa1a4326	176	if (table[i].cost >= prev_cost) {
c8ed9953	177	table[i].flags = EM_PERF_STATE_INEFFICIENT;
aa1a4326 VD	178	dev_dbg(dev, "EM: OPP:%lu is inefficient\n",
	179	table[i].frequency);
	180	} else {
	181	prev_cost = table[i].cost;
	182	}
27871f7a QP	183	}
	184
	185	pd->table = table;
521b512b	186	pd->nr_perf_states = nr_states;
27871f7a	187
1bc138c6	188	return 0;
27871f7a	189
521b512b	190	free_ps_table:
27871f7a	191	kfree(table);
1bc138c6 LL	192	return -EINVAL;
	193	}
	194
	195	static int em_create_pd(struct device *dev, int nr_states,
91362463 LL	196	struct em_data_callback cb, cpumask_t cpus,
91362463 LL	197	unsigned long flags)
1bc138c6 LL	198	{
	199	struct em_perf_domain *pd;
	200	struct device *cpu_dev;
ae6ccaa6	201	int cpu, ret, num_cpus;
1bc138c6 LL	202
1bc138c6 LL	203	if (_is_cpu_device(dev)) {
ae6ccaa6 LL	204	num_cpus = cpumask_weight(cpus);
	205
	206	/* Prevent max possible energy calculation to not overflow */
	207	if (num_cpus > EM_MAX_NUM_CPUS) {
	208	dev_err(dev, "EM: too many CPUs, overflow possible\n");
	209	return -EINVAL;
	210	}
	211
1bc138c6 LL	212	pd = kzalloc(sizeof(*pd) + cpumask_size(), GFP_KERNEL);
	213	if (!pd)
	214	return -ENOMEM;
	215
	216	cpumask_copy(em_span_cpus(pd), cpus);
	217	} else {
	218	pd = kzalloc(sizeof(*pd), GFP_KERNEL);
	219	if (!pd)
	220	return -ENOMEM;
	221	}
	222
91362463	223	ret = em_create_perf_table(dev, pd, nr_states, cb, flags);
1bc138c6 LL	224	if (ret) {
	225	kfree(pd);
	226	return ret;
	227	}
	228
	229	if (_is_cpu_device(dev))
	230	for_each_cpu(cpu, cpus) {
	231	cpu_dev = get_cpu_device(cpu);
	232	cpu_dev->em_pd = pd;
	233	}
	234
	235	dev->em_pd = pd;
	236
	237	return 0;
	238	}
	239
e458716a VD	240	static void em_cpufreq_update_efficiencies(struct device *dev)
	241	{
	242	struct em_perf_domain *pd = dev->em_pd;
	243	struct em_perf_state *table;
	244	struct cpufreq_policy *policy;
	245	int found = 0;
	246	int i;
	247
	248	if (!_is_cpu_device(dev) \|\| !pd)
	249	return;
	250
	251	policy = cpufreq_cpu_get(cpumask_first(em_span_cpus(pd)));
	252	if (!policy) {
	253	dev_warn(dev, "EM: Access to CPUFreq policy failed");
	254	return;
	255	}
	256
	257	table = pd->table;
	258
	259	for (i = 0; i < pd->nr_perf_states; i++) {
	260	if (!(table[i].flags & EM_PERF_STATE_INEFFICIENT))
	261	continue;
	262
	263	if (!cpufreq_table_set_inefficient(policy, table[i].frequency))
	264	found++;
	265	}
	266
c9d8923b PG	267	cpufreq_cpu_put(policy);
c9d8923b PG	268
e458716a VD	269	if (!found)
	270	return;
	271
	272	/*
	273	* Efficiencies have been installed in CPUFreq, inefficient frequencies
	274	* will be skipped. The EM can do the same.
	275	*/
	276	pd->flags \|= EM_PERF_DOMAIN_SKIP_INEFFICIENCIES;
	277	}
	278
1bc138c6 LL	279	/**
	280	* em_pd_get() - Return the performance domain for a device
	281	* @dev : Device to find the performance domain for
	282	*
	283	* Returns the performance domain to which @dev belongs, or NULL if it doesn't
	284	* exist.
	285	*/
	286	struct em_perf_domain em_pd_get(struct device dev)
	287	{
	288	if (IS_ERR_OR_NULL(dev))
	289	return NULL;
27871f7a	290
1bc138c6	291	return dev->em_pd;
27871f7a	292	}
1bc138c6	293	EXPORT_SYMBOL_GPL(em_pd_get);
27871f7a QP	294
	295	/**
	296	* em_cpu_get() - Return the performance domain for a CPU
	297	* @cpu : CPU to find the performance domain for
	298	*
1bc138c6	299	* Returns the performance domain to which @cpu belongs, or NULL if it doesn't
27871f7a QP	300	* exist.
	301	*/
	302	struct em_perf_domain *em_cpu_get(int cpu)
	303	{
1bc138c6 LL	304	struct device *cpu_dev;
	305
	306	cpu_dev = get_cpu_device(cpu);
	307	if (!cpu_dev)
	308	return NULL;
	309
	310	return em_pd_get(cpu_dev);
27871f7a QP	311	}
	312	EXPORT_SYMBOL_GPL(em_cpu_get);
	313
	314	/**
7d9895c7 LL	315	* em_dev_register_perf_domain() - Register the Energy Model (EM) for a device
7d9895c7 LL	316	* @dev : Device for which the EM is to register
521b512b	317	* @nr_states : Number of performance states to register
27871f7a	318	* @cb : Callback functions providing the data of the Energy Model
1bc138c6	319	* @cpus : Pointer to cpumask_t, which in case of a CPU device is
7d9895c7 LL	320	* obligatory. It can be taken from i.e. 'policy->cpus'. For other
7d9895c7 LL	321	* type of devices this should be set to NULL.
ae6ccaa6	322	* @microwatts : Flag indicating that the power values are in micro-Watts or
c250d50f	323	* in some other scale. It must be set properly.
27871f7a QP	324	*
	325	* Create Energy Model tables for a performance domain using the callbacks
	326	* defined in cb.
	327	*
ae6ccaa6	328	* The @microwatts is important to set with correct value. Some kernel
c250d50f LL	329	* sub-systems might rely on this flag and check if all devices in the EM are
	330	* using the same scale.
	331	*
27871f7a QP	332	* If multiple clients register the same performance domain, all but the first
	333	* registration will be ignored.
	334	*
	335	* Return 0 on success
	336	*/
7d9895c7	337	int em_dev_register_perf_domain(struct device *dev, unsigned int nr_states,
c250d50f	338	struct em_data_callback cb, cpumask_t cpus,
ae6ccaa6	339	bool microwatts)
27871f7a QP	340	{
27871f7a QP	341	unsigned long cap, prev_cap = 0;
91362463	342	unsigned long flags = 0;
1bc138c6	343	int cpu, ret;
27871f7a	344
1bc138c6	345	if (!dev \|\| !nr_states \|\| !cb)
27871f7a QP	346	return -EINVAL;
	347
	348	/*
	349	* Use a mutex to serialize the registration of performance domains and
	350	* let the driver-defined callback functions sleep.
	351	*/
	352	mutex_lock(&em_pd_mutex);
	353
1bc138c6 LL	354	if (dev->em_pd) {
	355	ret = -EEXIST;
	356	goto unlock;
	357	}
27871f7a	358
1bc138c6 LL	359	if (_is_cpu_device(dev)) {
	360	if (!cpus) {
	361	dev_err(dev, "EM: invalid CPU mask\n");
27871f7a QP	362	ret = -EINVAL;
	363	goto unlock;
	364	}
1bc138c6 LL	365
	366	for_each_cpu(cpu, cpus) {
	367	if (em_cpu_get(cpu)) {
	368	dev_err(dev, "EM: exists for CPU%d\n", cpu);
	369	ret = -EEXIST;
	370	goto unlock;
	371	}
	372	/*
	373	* All CPUs of a domain must have the same
	374	* micro-architecture since they all share the same
	375	* table.
	376	*/
	377	cap = arch_scale_cpu_capacity(cpu);
	378	if (prev_cap && prev_cap != cap) {
	379	dev_err(dev, "EM: CPUs of %*pbl must have the same capacity\n",
	380	cpumask_pr_args(cpus));
	381
	382	ret = -EINVAL;
	383	goto unlock;
	384	}
	385	prev_cap = cap;
	386	}
27871f7a QP	387	}
27871f7a QP	388
ae6ccaa6 LL	389	if (microwatts)
ae6ccaa6 LL	390	flags \|= EM_PERF_DOMAIN_MICROWATTS;
91362463 LL	391	else if (cb->get_cost)
	392	flags \|= EM_PERF_DOMAIN_ARTIFICIAL;
	393
	394	ret = em_create_pd(dev, nr_states, cb, cpus, flags);
1bc138c6	395	if (ret)
27871f7a	396	goto unlock;
27871f7a	397
91362463	398	dev->em_pd->flags \|= flags;
c250d50f	399
e458716a VD	400	em_cpufreq_update_efficiencies(dev);
e458716a VD	401
1bc138c6 LL	402	em_debug_create_pd(dev);
1bc138c6 LL	403	dev_info(dev, "EM: created perf domain\n");
27871f7a	404
27871f7a QP	405	unlock:
27871f7a QP	406	mutex_unlock(&em_pd_mutex);
27871f7a QP	407	return ret;
27871f7a QP	408	}
7d9895c7 LL	409	EXPORT_SYMBOL_GPL(em_dev_register_perf_domain);
7d9895c7 LL	410
1bc138c6 LL	411	/**
	412	* em_dev_unregister_perf_domain() - Unregister Energy Model (EM) for a device
	413	* @dev : Device for which the EM is registered
	414	*
	415	* Unregister the EM for the specified @dev (but not a CPU device).
	416	*/
	417	void em_dev_unregister_perf_domain(struct device *dev)
	418	{
	419	if (IS_ERR_OR_NULL(dev) \|\| !dev->em_pd)
	420	return;
	421
	422	if (_is_cpu_device(dev))
	423	return;
	424
	425	/*
	426	* The mutex separates all register/unregister requests and protects
	427	* from potential clean-up/setup issues in the debugfs directories.
	428	* The debugfs directory name is the same as device's name.
	429	*/
	430	mutex_lock(&em_pd_mutex);
	431	em_debug_remove_pd(dev);
	432
	433	kfree(dev->em_pd->table);
	434	kfree(dev->em_pd);
	435	dev->em_pd = NULL;
	436	mutex_unlock(&em_pd_mutex);
	437	}
	438	EXPORT_SYMBOL_GPL(em_dev_unregister_perf_domain);