[v2,4/4] powercap/drivers/dtpm: Add CPU energy model based support
diff mbox series

Message ID 20201116152649.11482-5-daniel.lezcano@linaro.org
State New, archived
Headers show
Series
  • powercap/dtpm: Add the DTPM framework
Related show

Commit Message

Daniel Lezcano Nov. 16, 2020, 3:26 p.m. UTC
With the powercap dtpm controller, we are able to plug devices with
power limitation features in the tree.

The following patch introduces the CPU power limitation based on the
energy model and the performance states.

The power limitation is done at the performance domain level. If some
CPUs are unplugged, the corresponding power will be subtracted from
the performance domain total power.

It is up to the platform to initialize the dtpm tree and add the CPU.

Here is an example to create a simple tree with one root node called
"pkg" and the CPU's performance domains.

static int dtpm_register_pkg(struct dtpm_descr *descr)
{
	struct dtpm *pkg;
	int ret;

	pkg = dtpm_alloc();
	if (!pkg)
		return -ENOMEM;

	ret = dtpm_register_parent(descr->name, pkg, descr->parent);
	if (ret)
		return ret;

	return dtpm_register_cpu(pkg);
}

static struct dtpm_descr descr = {
	.name = "pkg",
	.init = dtpm_register_pkg,
};
DTPM_DECLARE(descr);

Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>
---
 drivers/powercap/Kconfig    |   7 +
 drivers/powercap/Makefile   |   1 +
 drivers/powercap/dtpm_cpu.c | 282 ++++++++++++++++++++++++++++++++++++
 include/linux/cpuhotplug.h  |   1 +
 include/linux/dtpm.h        |   3 +
 5 files changed, 294 insertions(+)
 create mode 100644 drivers/powercap/dtpm_cpu.c

Comments

Lukasz Luba Nov. 17, 2020, 1:15 p.m. UTC | #1
Hi Daniel,

Only one small comment regarding the setup of 'power_limit'.

On 11/16/20 3:26 PM, Daniel Lezcano wrote:
> With the powercap dtpm controller, we are able to plug devices with
> power limitation features in the tree.
> 
> The following patch introduces the CPU power limitation based on the
> energy model and the performance states.
> 
> The power limitation is done at the performance domain level. If some
> CPUs are unplugged, the corresponding power will be subtracted from
> the performance domain total power.
> 
> It is up to the platform to initialize the dtpm tree and add the CPU.
> 
> Here is an example to create a simple tree with one root node called
> "pkg" and the CPU's performance domains.
> 
> static int dtpm_register_pkg(struct dtpm_descr *descr)
> {
> 	struct dtpm *pkg;
> 	int ret;
> 
> 	pkg = dtpm_alloc();
> 	if (!pkg)
> 		return -ENOMEM;
> 
> 	ret = dtpm_register_parent(descr->name, pkg, descr->parent);
> 	if (ret)
> 		return ret;
> 
> 	return dtpm_register_cpu(pkg);
> }
> 
> static struct dtpm_descr descr = {
> 	.name = "pkg",
> 	.init = dtpm_register_pkg,
> };
> DTPM_DECLARE(descr);
> 
> Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>
> ---
>   drivers/powercap/Kconfig    |   7 +
>   drivers/powercap/Makefile   |   1 +
>   drivers/powercap/dtpm_cpu.c | 282 ++++++++++++++++++++++++++++++++++++
>   include/linux/cpuhotplug.h  |   1 +
>   include/linux/dtpm.h        |   3 +
>   5 files changed, 294 insertions(+)
>   create mode 100644 drivers/powercap/dtpm_cpu.c
> 
> diff --git a/drivers/powercap/Kconfig b/drivers/powercap/Kconfig
> index cc1953bd8bed..20b4325c6161 100644
> --- a/drivers/powercap/Kconfig
> +++ b/drivers/powercap/Kconfig
> @@ -49,4 +49,11 @@ config DTPM
>   	help
>   	  This enables support for the power capping for the dynamic
>   	  thermal power management userspace engine.
> +
> +config DTPM_CPU
> +	bool "Add CPU power capping based on the energy model"
> +	depends on DTPM && ENERGY_MODEL
> +	help
> +	  This enables support for CPU power limitation based on
> +	  energy model.
>   endif
> diff --git a/drivers/powercap/Makefile b/drivers/powercap/Makefile
> index 6482ac52054d..fabcf388a8d3 100644
> --- a/drivers/powercap/Makefile
> +++ b/drivers/powercap/Makefile
> @@ -1,5 +1,6 @@
>   # SPDX-License-Identifier: GPL-2.0-only
>   obj-$(CONFIG_DTPM) += dtpm.o
> +obj-$(CONFIG_DTPM_CPU) += dtpm_cpu.o
>   obj-$(CONFIG_POWERCAP)	+= powercap_sys.o
>   obj-$(CONFIG_INTEL_RAPL_CORE) += intel_rapl_common.o
>   obj-$(CONFIG_INTEL_RAPL) += intel_rapl_msr.o
> diff --git a/drivers/powercap/dtpm_cpu.c b/drivers/powercap/dtpm_cpu.c
> new file mode 100644
> index 000000000000..6bff5f27d891
> --- /dev/null
> +++ b/drivers/powercap/dtpm_cpu.c
> @@ -0,0 +1,282 @@
> +// SPDX-License-Identifier: GPL-2.0-only
> +/*
> + * Copyright 2020 Linaro Limited
> + *
> + * Author: Daniel Lezcano <daniel.lezcano@linaro.org>
> + *
> + * The DTPM CPU is based on the energy model. It hooks the CPU in the
> + * DTPM tree which in turns update the power number by propagating the
> + * power number from the CPU energy model information to the parents.
> + *
> + * The association between the power and the performance state, allows
> + * to set the power of the CPU at the OPP granularity.
> + *
> + * The CPU hotplug is supported and the power numbers will be updated
> + * if a CPU is hot plugged / unplugged.
> + */
> +#include <linux/cpumask.h>
> +#include <linux/cpufreq.h>
> +#include <linux/cpuhotplug.h>
> +#include <linux/dtpm.h>
> +#include <linux/energy_model.h>
> +#include <linux/pm_qos.h>
> +#include <linux/slab.h>
> +#include <linux/units.h>
> +
> +static struct dtpm *__parent;
> +
> +static DEFINE_PER_CPU(struct dtpm *, dtpm_per_cpu);
> +
> +struct dtpm_cpu {
> +	struct freq_qos_request qos_req;
> +	int cpu;
> +};
> +
> +/*
> + * When a new CPU is inserted at hotplug or boot time, add the power
> + * contribution and update the dtpm tree.
> + */
> +static int power_add(struct dtpm *dtpm, struct em_perf_domain *em)
> +{
> +	u64 power_min, power_max;
> +
> +	power_min = em->table[0].power;
> +	power_min *= MICROWATT_PER_MILLIWATT;
> +	power_min += dtpm->power_min;
> +
> +	power_max = em->table[em->nr_perf_states - 1].power;
> +	power_max *= MICROWATT_PER_MILLIWATT;
> +	power_max += dtpm->power_max;
> +
> +	return dtpm_update_power(dtpm, power_min, power_max);
> +}
> +
> +/*
> + * When a CPU is unplugged, remove its power contribution from the
> + * dtpm tree.
> + */
> +static int power_sub(struct dtpm *dtpm, struct em_perf_domain *em)
> +{
> +	u64 power_min, power_max;
> +
> +	power_min = em->table[0].power;
> +	power_min *= MICROWATT_PER_MILLIWATT;
> +	power_min = dtpm->power_min - power_min;
> +
> +	power_max = em->table[em->nr_perf_states - 1].power;
> +	power_max *= MICROWATT_PER_MILLIWATT;
> +	power_max = dtpm->power_max - power_max;
> +
> +	return dtpm_update_power(dtpm, power_min, power_max);
> +}
> +
> +static int set_pd_power_limit(struct powercap_zone *pcz, int cid,
> +			      u64 power_limit)
> +{
> +	struct dtpm *dtpm = to_dtpm(pcz);
> +	struct dtpm_cpu *dtpm_cpu = dtpm->private;
> +	struct em_perf_domain *pd;
> +	struct cpumask cpus;
> +	unsigned long freq;
> +	u64 power;
> +	int i, nr_cpus;
> +
> +	spin_lock(&dtpm->lock);
> +
> +	power_limit = clamp_val(power_limit, dtpm->power_min, dtpm->power_max);
> +
> +	pd = em_cpu_get(dtpm_cpu->cpu);
> +
> +	cpumask_and(&cpus, cpu_online_mask, to_cpumask(pd->cpus));
> +
> +	nr_cpus = cpumask_weight(&cpus);
> +
> +	for (i = 0; i < pd->nr_perf_states; i++) {
> +
> +		power = pd->table[i].power * MICROWATT_PER_MILLIWATT * nr_cpus;
> +
> +		if (power > power_limit)
> +			break;
> +	}
> +
> +	freq = pd->table[i - 1].frequency;
> +
> +	dtpm->power_limit = pd->table[i - 1].power *
> +		MICROWATT_PER_MILLIWATT * nr_cpus;
> +
> +	spin_unlock(&dtpm->lock);
> +
> +	freq_qos_update_request(&dtpm_cpu->qos_req, freq);
> +
> +	return 0;
> +}
> +
> +static int get_pd_power_limit(struct powercap_zone *pcz, int cid, u64 *data)
> +{
> +	struct dtpm *dtpm = to_dtpm(pcz);
> +
> +	spin_lock(&dtpm->lock);
> +	*data = dtpm->power_limit;
> +	spin_unlock(&dtpm->lock);
> +
> +	return 0;
> +}
> +
> +static int get_pd_power_uw(struct powercap_zone *pcz, u64 *power_uw)
> +{
> +	struct dtpm *dtpm = to_dtpm(pcz);
> +	struct dtpm_cpu *dtpm_cpu = dtpm->private;
> +	struct em_perf_domain *pd;
> +	unsigned long freq;
> +	int i, nr_cpus;
> +
> +	freq = cpufreq_quick_get(dtpm_cpu->cpu);
> +	pd = em_cpu_get(dtpm_cpu->cpu);
> +	nr_cpus = cpumask_weight(to_cpumask(pd->cpus));
> +
> +	for (i = 0; i < pd->nr_perf_states; i++) {
> +
> +		if (pd->table[i].frequency < freq)
> +			continue;
> +
> +		*power_uw = pd->table[i].power *
> +			MICROWATT_PER_MILLIWATT * nr_cpus;
> +
> +		return 0;
> +	}
> +
> +	return -EINVAL;
> +}
> +
> +static int cpu_release_zone(struct powercap_zone *pcz)
> +{
> +	struct dtpm *dtpm = to_dtpm(pcz);
> +	struct dtpm_cpu *dtpm_cpu = dtpm->private;
> +
> +	freq_qos_remove_request(&dtpm_cpu->qos_req);
> +	kfree(dtpm_cpu);
> +
> +	return dtpm_release_zone(pcz);
> +}
> +
> +static struct powercap_zone_constraint_ops pd_constraint_ops = {
> +	.set_power_limit_uw = set_pd_power_limit,
> +	.get_power_limit_uw = get_pd_power_limit,
> +};
> +
> +static struct powercap_zone_ops pd_zone_ops = {
> +	.get_power_uw = get_pd_power_uw,
> +	.release = cpu_release_zone,
> +};
> +
> +static int cpuhp_dtpm_cpu_offline(unsigned int cpu)
> +{
> +	struct cpufreq_policy *policy;
> +	struct em_perf_domain *pd;
> +	struct dtpm *dtpm;
> +
> +	policy = cpufreq_cpu_get(cpu);
> +
> +	if (!policy)
> +		return 0;
> +
> +	pd = em_cpu_get(cpu);
> +	if (!pd)
> +		return -EINVAL;
> +
> +	dtpm = per_cpu(dtpm_per_cpu, cpu);
> +
> +	power_sub(dtpm, pd);
> +
> +	if (cpumask_weight(policy->cpus) != 1)
> +		return 0;
> +
> +	for_each_cpu(cpu, policy->related_cpus)
> +		per_cpu(dtpm_per_cpu, cpu) = NULL;
> +
> +	dtpm_unregister(dtpm);
> +
> +	return 0;
> +}
> +
> +static int cpuhp_dtpm_cpu_online(unsigned int cpu)
> +{
> +        struct dtpm *dtpm;
> +	struct dtpm_cpu *dtpm_cpu;
> +	struct cpufreq_policy *policy;
> +	struct em_perf_domain *pd;
> +	char name[CPUFREQ_NAME_LEN];
> +	int ret;
> +
> +	policy = cpufreq_cpu_get(cpu);
> +
> +	if (!policy)
> +		return 0;
> +
> +	pd = em_cpu_get(cpu);
> +	if (!pd)
> +		return -EINVAL;
> +
> +	dtpm = per_cpu(dtpm_per_cpu, cpu);
> +	if (dtpm)
> +		return power_add(dtpm, pd);

The dtpm->power_limit is not incremented in this path, when a new
CPU joins the cluster.
Is it correct?

Or maybe we need something like:
------------------------------>8---------------------
         if (dtpm) {
                 ret = power_add(dtpm, pd);
                 if (!ret)
                         dtpm->power_limit = dtpm->power_max;
                 return ret;
         }
------------------------8<---------------

The power_max should be updated after successful power_add().
It would disturb user set value in power_limit, though (described
below).


> +
> +	dtpm = dtpm_alloc();
> +	if (!dtpm)
> +		return -EINVAL;
> +
> +	dtpm_cpu = kzalloc(sizeof(dtpm_cpu), GFP_KERNEL);
> +	if (!dtpm_cpu) {
> +		kfree(dtpm);
> +		return -ENOMEM;
> +	}
> +
> +	dtpm->private = dtpm_cpu;
> +	dtpm_cpu->cpu = cpu;
> +
> +	for_each_cpu(cpu, policy->related_cpus)
> +		per_cpu(dtpm_per_cpu, cpu) = dtpm;
> +
> +	ret = power_add(dtpm, pd);
> +	if (ret)
> +		goto out_kfree_dtpm_cpu;
> +
> +	dtpm->power_limit = dtpm->power_max;

Here, the power_limit will be set only once with power_max
for a single CPU. I am not sure, but maybe we can simple say:

dtpm->power_limit = dtpm->power_max * cpumask_weight(policy->related_cpus)

an avoid touching it later (?)

Because this function can be called in runtime, when the power_limit
was already set by userspace, the hotpluging in/out/in... CPU shouldn't
change this limit.

Regards,
Lukasz
Daniel Lezcano Nov. 17, 2020, 5:44 p.m. UTC | #2
On 17/11/2020 14:15, Lukasz Luba wrote:
> Hi Daniel,
> 
> Only one small comment regarding the setup of 'power_limit'.
> 
> On 11/16/20 3:26 PM, Daniel Lezcano wrote:
>> With the powercap dtpm controller, we are able to plug devices with
>> power limitation features in the tree.
>>
>> The following patch introduces the CPU power limitation based on the
>> energy model and the performance states.
>>
>> The power limitation is done at the performance domain level. If some
>> CPUs are unplugged, the corresponding power will be subtracted from
>> the performance domain total power.
>>
>> It is up to the platform to initialize the dtpm tree and add the CPU.
>>

[ ... ]

>> +
>> +    dtpm = per_cpu(dtpm_per_cpu, cpu);
>> +    if (dtpm)
>> +        return power_add(dtpm, pd);
> 
> The dtpm->power_limit is not incremented in this path, when a new
> CPU joins the cluster.
> Is it correct?

Yes, you are right, there is something missing here. It does not change
the behavior of the power capping, but the value will be inconsistent in
the tree.

> Or maybe we need something like:
> ------------------------------>8---------------------
>         if (dtpm) {
>                 ret = power_add(dtpm, pd);
>                 if (!ret)
>                         dtpm->power_limit = dtpm->power_max;
>                 return ret;
>         }
> ------------------------8<---------------
> 
> The power_max should be updated after successful power_add().
> It would disturb user set value in power_limit, though (described
> below).
> 
> 
>> +
>> +    dtpm = dtpm_alloc();
>> +    if (!dtpm)
>> +        return -EINVAL;
>> +
>> +    dtpm_cpu = kzalloc(sizeof(dtpm_cpu), GFP_KERNEL);
>> +    if (!dtpm_cpu) {
>> +        kfree(dtpm);
>> +        return -ENOMEM;
>> +    }
>> +
>> +    dtpm->private = dtpm_cpu;
>> +    dtpm_cpu->cpu = cpu;
>> +
>> +    for_each_cpu(cpu, policy->related_cpus)
>> +        per_cpu(dtpm_per_cpu, cpu) = dtpm;
>> +
>> +    ret = power_add(dtpm, pd);
>> +    if (ret)
>> +        goto out_kfree_dtpm_cpu;
>> +
>> +    dtpm->power_limit = dtpm->power_max;
> 
> Here, the power_limit will be set only once with power_max
> for a single CPU. I am not sure, but maybe we can simple say:
> 
> dtpm->power_limit = dtpm->power_max * cpumask_weight(policy->related_cpus)
> 
> an avoid touching it later (?)
> 
> Because this function can be called in runtime, when the power_limit
> was already set by userspace, the hotpluging in/out/in... CPU shouldn't
> change this limit.

Hmm, I have to think about it because the power_limit is always less or
equal to power_max.

Patch
diff mbox series

diff --git a/drivers/powercap/Kconfig b/drivers/powercap/Kconfig
index cc1953bd8bed..20b4325c6161 100644
--- a/drivers/powercap/Kconfig
+++ b/drivers/powercap/Kconfig
@@ -49,4 +49,11 @@  config DTPM
 	help
 	  This enables support for the power capping for the dynamic
 	  thermal power management userspace engine.
+
+config DTPM_CPU
+	bool "Add CPU power capping based on the energy model"
+	depends on DTPM && ENERGY_MODEL
+	help
+	  This enables support for CPU power limitation based on
+	  energy model.
 endif
diff --git a/drivers/powercap/Makefile b/drivers/powercap/Makefile
index 6482ac52054d..fabcf388a8d3 100644
--- a/drivers/powercap/Makefile
+++ b/drivers/powercap/Makefile
@@ -1,5 +1,6 @@ 
 # SPDX-License-Identifier: GPL-2.0-only
 obj-$(CONFIG_DTPM) += dtpm.o
+obj-$(CONFIG_DTPM_CPU) += dtpm_cpu.o
 obj-$(CONFIG_POWERCAP)	+= powercap_sys.o
 obj-$(CONFIG_INTEL_RAPL_CORE) += intel_rapl_common.o
 obj-$(CONFIG_INTEL_RAPL) += intel_rapl_msr.o
diff --git a/drivers/powercap/dtpm_cpu.c b/drivers/powercap/dtpm_cpu.c
new file mode 100644
index 000000000000..6bff5f27d891
--- /dev/null
+++ b/drivers/powercap/dtpm_cpu.c
@@ -0,0 +1,282 @@ 
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright 2020 Linaro Limited
+ *
+ * Author: Daniel Lezcano <daniel.lezcano@linaro.org>
+ *
+ * The DTPM CPU is based on the energy model. It hooks the CPU in the
+ * DTPM tree which in turns update the power number by propagating the
+ * power number from the CPU energy model information to the parents.
+ *
+ * The association between the power and the performance state, allows
+ * to set the power of the CPU at the OPP granularity.
+ *
+ * The CPU hotplug is supported and the power numbers will be updated
+ * if a CPU is hot plugged / unplugged.
+ */
+#include <linux/cpumask.h>
+#include <linux/cpufreq.h>
+#include <linux/cpuhotplug.h>
+#include <linux/dtpm.h>
+#include <linux/energy_model.h>
+#include <linux/pm_qos.h>
+#include <linux/slab.h>
+#include <linux/units.h>
+
+static struct dtpm *__parent;
+
+static DEFINE_PER_CPU(struct dtpm *, dtpm_per_cpu);
+
+struct dtpm_cpu {
+	struct freq_qos_request qos_req;
+	int cpu;
+};
+
+/*
+ * When a new CPU is inserted at hotplug or boot time, add the power
+ * contribution and update the dtpm tree.
+ */
+static int power_add(struct dtpm *dtpm, struct em_perf_domain *em)
+{
+	u64 power_min, power_max;
+
+	power_min = em->table[0].power;
+	power_min *= MICROWATT_PER_MILLIWATT;
+	power_min += dtpm->power_min;
+
+	power_max = em->table[em->nr_perf_states - 1].power;
+	power_max *= MICROWATT_PER_MILLIWATT;
+	power_max += dtpm->power_max;
+
+	return dtpm_update_power(dtpm, power_min, power_max);
+}
+
+/*
+ * When a CPU is unplugged, remove its power contribution from the
+ * dtpm tree.
+ */
+static int power_sub(struct dtpm *dtpm, struct em_perf_domain *em)
+{
+	u64 power_min, power_max;
+
+	power_min = em->table[0].power;
+	power_min *= MICROWATT_PER_MILLIWATT;
+	power_min = dtpm->power_min - power_min;
+
+	power_max = em->table[em->nr_perf_states - 1].power;
+	power_max *= MICROWATT_PER_MILLIWATT;
+	power_max = dtpm->power_max - power_max;
+
+	return dtpm_update_power(dtpm, power_min, power_max);
+}
+
+static int set_pd_power_limit(struct powercap_zone *pcz, int cid,
+			      u64 power_limit)
+{
+	struct dtpm *dtpm = to_dtpm(pcz);
+	struct dtpm_cpu *dtpm_cpu = dtpm->private;
+	struct em_perf_domain *pd;
+	struct cpumask cpus;
+	unsigned long freq;
+	u64 power;
+	int i, nr_cpus;
+
+	spin_lock(&dtpm->lock);
+
+	power_limit = clamp_val(power_limit, dtpm->power_min, dtpm->power_max);
+
+	pd = em_cpu_get(dtpm_cpu->cpu);
+
+	cpumask_and(&cpus, cpu_online_mask, to_cpumask(pd->cpus));
+
+	nr_cpus = cpumask_weight(&cpus);
+
+	for (i = 0; i < pd->nr_perf_states; i++) {
+
+		power = pd->table[i].power * MICROWATT_PER_MILLIWATT * nr_cpus;
+
+		if (power > power_limit)
+			break;
+	}
+
+	freq = pd->table[i - 1].frequency;
+
+	dtpm->power_limit = pd->table[i - 1].power *
+		MICROWATT_PER_MILLIWATT * nr_cpus;
+
+	spin_unlock(&dtpm->lock);
+
+	freq_qos_update_request(&dtpm_cpu->qos_req, freq);
+
+	return 0;
+}
+
+static int get_pd_power_limit(struct powercap_zone *pcz, int cid, u64 *data)
+{
+	struct dtpm *dtpm = to_dtpm(pcz);
+
+	spin_lock(&dtpm->lock);
+	*data = dtpm->power_limit;
+	spin_unlock(&dtpm->lock);
+
+	return 0;
+}
+
+static int get_pd_power_uw(struct powercap_zone *pcz, u64 *power_uw)
+{
+	struct dtpm *dtpm = to_dtpm(pcz);
+	struct dtpm_cpu *dtpm_cpu = dtpm->private;
+	struct em_perf_domain *pd;
+	unsigned long freq;
+	int i, nr_cpus;
+
+	freq = cpufreq_quick_get(dtpm_cpu->cpu);
+	pd = em_cpu_get(dtpm_cpu->cpu);
+	nr_cpus = cpumask_weight(to_cpumask(pd->cpus));
+
+	for (i = 0; i < pd->nr_perf_states; i++) {
+
+		if (pd->table[i].frequency < freq)
+			continue;
+
+		*power_uw = pd->table[i].power *
+			MICROWATT_PER_MILLIWATT * nr_cpus;
+
+		return 0;
+	}
+
+	return -EINVAL;
+}
+
+static int cpu_release_zone(struct powercap_zone *pcz)
+{
+	struct dtpm *dtpm = to_dtpm(pcz);
+	struct dtpm_cpu *dtpm_cpu = dtpm->private;
+
+	freq_qos_remove_request(&dtpm_cpu->qos_req);
+	kfree(dtpm_cpu);
+
+	return dtpm_release_zone(pcz);
+}
+
+static struct powercap_zone_constraint_ops pd_constraint_ops = {
+	.set_power_limit_uw = set_pd_power_limit,
+	.get_power_limit_uw = get_pd_power_limit,
+};
+
+static struct powercap_zone_ops pd_zone_ops = {
+	.get_power_uw = get_pd_power_uw,
+	.release = cpu_release_zone,
+};
+
+static int cpuhp_dtpm_cpu_offline(unsigned int cpu)
+{
+	struct cpufreq_policy *policy;
+	struct em_perf_domain *pd;
+	struct dtpm *dtpm;
+
+	policy = cpufreq_cpu_get(cpu);
+
+	if (!policy)
+		return 0;
+
+	pd = em_cpu_get(cpu);
+	if (!pd)
+		return -EINVAL;
+
+	dtpm = per_cpu(dtpm_per_cpu, cpu);
+
+	power_sub(dtpm, pd);
+
+	if (cpumask_weight(policy->cpus) != 1)
+		return 0;
+
+	for_each_cpu(cpu, policy->related_cpus)
+		per_cpu(dtpm_per_cpu, cpu) = NULL;
+
+	dtpm_unregister(dtpm);
+
+	return 0;
+}
+
+static int cpuhp_dtpm_cpu_online(unsigned int cpu)
+{
+        struct dtpm *dtpm;
+	struct dtpm_cpu *dtpm_cpu;
+	struct cpufreq_policy *policy;
+	struct em_perf_domain *pd;
+	char name[CPUFREQ_NAME_LEN];
+	int ret;
+
+	policy = cpufreq_cpu_get(cpu);
+
+	if (!policy)
+		return 0;
+
+	pd = em_cpu_get(cpu);
+	if (!pd)
+		return -EINVAL;
+
+	dtpm = per_cpu(dtpm_per_cpu, cpu);
+	if (dtpm)
+		return power_add(dtpm, pd);
+
+	dtpm = dtpm_alloc();
+	if (!dtpm)
+		return -EINVAL;
+
+	dtpm_cpu = kzalloc(sizeof(dtpm_cpu), GFP_KERNEL);
+	if (!dtpm_cpu) {
+		kfree(dtpm);
+		return -ENOMEM;
+	}
+
+	dtpm->private = dtpm_cpu;
+	dtpm_cpu->cpu = cpu;
+
+	for_each_cpu(cpu, policy->related_cpus)
+		per_cpu(dtpm_per_cpu, cpu) = dtpm;
+
+	ret = power_add(dtpm, pd);
+	if (ret)
+		goto out_kfree_dtpm_cpu;
+
+	dtpm->power_limit = dtpm->power_max;
+
+	sprintf(name, "cpu%d", dtpm_cpu->cpu);
+
+	ret = dtpm_register(name, dtpm, __parent, &pd_zone_ops,
+			    1, &pd_constraint_ops);
+	if (ret)
+		goto out_power_sub;
+
+	ret = freq_qos_add_request(&policy->constraints,
+				   &dtpm_cpu->qos_req, FREQ_QOS_MAX,
+				   pd->table[pd->nr_perf_states - 1].frequency);
+	if (ret)
+		goto out_dtpm_unregister;
+
+	return 0;
+
+out_dtpm_unregister:
+	dtpm_unregister(dtpm);
+	dtpm_cpu = NULL; /* Already freed by the release ops */
+out_power_sub:
+	power_sub(dtpm, pd);
+out_kfree_dtpm_cpu:
+	for_each_cpu(cpu, policy->related_cpus)
+		per_cpu(dtpm_per_cpu, cpu) = NULL;
+	kfree(dtpm_cpu);
+
+	return ret;
+}
+
+int dtpm_register_cpu(struct dtpm *parent)
+{
+	__parent = parent;
+
+	return cpuhp_setup_state(CPUHP_AP_DTPM_CPU_ONLINE,
+				 "dtpm_cpu:online",
+				 cpuhp_dtpm_cpu_online,
+				 cpuhp_dtpm_cpu_offline);
+}
diff --git a/include/linux/cpuhotplug.h b/include/linux/cpuhotplug.h
index bc56287a1ed1..72fd8db62342 100644
--- a/include/linux/cpuhotplug.h
+++ b/include/linux/cpuhotplug.h
@@ -192,6 +192,7 @@  enum cpuhp_state {
 	CPUHP_AP_ONLINE_DYN_END		= CPUHP_AP_ONLINE_DYN + 30,
 	CPUHP_AP_X86_HPET_ONLINE,
 	CPUHP_AP_X86_KVM_CLK_ONLINE,
+	CPUHP_AP_DTPM_CPU_ONLINE,
 	CPUHP_AP_ACTIVE,
 	CPUHP_ONLINE,
 };
diff --git a/include/linux/dtpm.h b/include/linux/dtpm.h
index 899d9d8fbccf..fece9bede28b 100644
--- a/include/linux/dtpm.h
+++ b/include/linux/dtpm.h
@@ -69,4 +69,7 @@  int dtpm_register_parent(const char *name, struct dtpm *dtpm,
 int dtpm_register(const char *name, struct dtpm *dtpm, struct dtpm *parent,
 		  struct powercap_zone_ops *ops, int nr_constraints,
 		  struct powercap_zone_constraint_ops *const_ops);
+
+int dtpm_register_cpu(struct dtpm *parent);
+
 #endif