Re: [PATCH 4/4] powercap/drivers/dtpm: Add CPU energy model based support

[Lukasz Luba]

Hi Daniel,


On 10/6/20 1:20 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 substracted 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.
>
> 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);
> }
>
> struct dtpm_descr descr = {
> 	.name = "pkg",
> 	.init = dtpm_register_pkg,
> };
> DTPM_DECLARE(descr);
>
> Signed-off-by: Daniel Lezcano <daniel.lezcano@xxxxxxxxxx>
> ---
>   drivers/powercap/Kconfig    |   8 ++
>   drivers/powercap/Makefile   |   1 +
>   drivers/powercap/dtpm_cpu.c | 242 ++++++++++++++++++++++++++++++++++++
>   include/linux/cpuhotplug.h  |   1 +
>   include/linux/dtpm.h        |   3 +
>   5 files changed, 255 insertions(+)
>   create mode 100644 drivers/powercap/dtpm_cpu.c
>
> diff --git a/drivers/powercap/Kconfig b/drivers/powercap/Kconfig
> index 777cf60300b8..240dc09e8dc2 100644
> --- a/drivers/powercap/Kconfig
> +++ b/drivers/powercap/Kconfig
> @@ -49,4 +49,12 @@ config DTPM
>   	help
>   	  This enables support for the power capping for the dynamic
>   	  thermal 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..23ebf704c599
> --- /dev/null
> +++ b/drivers/powercap/dtpm_cpu.c
> @@ -0,0 +1,242 @@
> +// SPDX-License-Identifier: GPL-2.0-only
> +/*
> + * Copyright 2020 Linaro Limited
> + *
> + * Author: Daniel Lezcano <daniel.lezcano@xxxxxxxxxx>
> + *
> + */
> +#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;
> +};
> +
> +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);
> +}
> +
> +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;
> +	unsigned long freq;
> +	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);
> +
> +	nr_cpus = cpumask_weight(to_cpumask(pd->cpus));
> +
> +	for (i = 0; i < pd->nr_perf_states; i++) {
> +
> +		u64 power = pd->table[i].power * MICROWATT_PER_MILLIWATT;
> +
> +		if ((power * nr_cpus) > power_limit)

We have one node in that DTPM hierarchy tree, which represents all CPUs
which are in 'related_cpus' mask. I saw below that we just remove the
node in hotplug.

I have put a comment below asking if we could change the registration,
which will affect power calculation.


> +			break;
> +	}
> +
> +	freq = pd->table[i - 1].frequency;
> +
> +	freq_qos_update_request(&dtpm_cpu->qos_req, freq);
> +
> +	dtpm->power_limit = power_limit;
> +
> +	spin_unlock(&dtpm->lock);
> +
> +	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_max;
> +	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;

Same here, we have '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);
> +
> +	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;

Hotplugging one CPU would affect others. I would keep them
all but marked somehow that CPU is offline.

> +
> +	dtpm_unregister(dtpm);

Could we keep the node in the hierarchy on CPU hotplug?

> +
> +	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)
> +		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)
> +		return ret;
> +
> +	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)
> +		return ret;
> +
> +	ret = freq_qos_add_request(&policy->constraints,
> +				   &dtpm_cpu->qos_req, FREQ_QOS_MAX,
> +				   pd->table[pd->nr_perf_states - 1].frequency);
> +	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 bf9181cef444..6792bad4a435 100644
> --- a/include/linux/cpuhotplug.h
> +++ b/include/linux/cpuhotplug.h
> @@ -190,6 +190,7 @@ 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 6696bdcfdb87..b62215a13baa 100644
> --- a/include/linux/dtpm.h
> +++ b/include/linux/dtpm.h
> @@ -70,4 +70,7 @@ 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

I have a few comments for this DTPM CPU.

1. Maybe we can register these CPUs differently. First register
the parent node as a separate dtpm based on 'policy->related_cpus. Then
register new children nodes, one for each CPU. When the CPU is up, mark
it as 'active'.

2. We don't remove the node when the CPU is hotplugged, but we mark it
'!active' Or 'offline'. The power calculation could be done in upper
node, which takes into account that flag for children.

3. We would only remove the node when it's module is unloaded (e.g. GPU)

That would make the tree more stable and also more detailed.
We would also account the power properly when one CPU went offline, but
the other are still there.

What do you think?


Regards,
Lukasz