linux/drivers/powercap/dtpm_cpu.c

298 lines
6.8 KiB
C
Raw Normal View History

// 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.
*/
#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
#include <linux/cpumask.h>
#include <linux/cpufreq.h>
#include <linux/cpuhotplug.h>
#include <linux/dtpm.h>
#include <linux/energy_model.h>
#include <linux/of.h>
#include <linux/pm_qos.h>
#include <linux/slab.h>
#include <linux/units.h>
struct dtpm_cpu {
struct dtpm dtpm;
struct freq_qos_request qos_req;
int cpu;
};
static DEFINE_PER_CPU(struct dtpm_cpu *, dtpm_per_cpu);
static struct dtpm_cpu *to_dtpm_cpu(struct dtpm *dtpm)
{
return container_of(dtpm, struct dtpm_cpu, dtpm);
}
static u64 set_pd_power_limit(struct dtpm *dtpm, u64 power_limit)
{
struct dtpm_cpu *dtpm_cpu = to_dtpm_cpu(dtpm);
struct em_perf_domain *pd = em_cpu_get(dtpm_cpu->cpu);
struct cpumask cpus;
unsigned long freq;
u64 power;
int i, nr_cpus;
cpumask_and(&cpus, cpu_online_mask, to_cpumask(pd->cpus));
nr_cpus = cpumask_weight(&cpus);
for (i = 0; i < pd->nr_perf_states; i++) {
PM: EM: convert power field to micro-Watts precision and align drivers The milli-Watts precision causes rounding errors while calculating efficiency cost for each OPP. This is especially visible in the 'simple' Energy Model (EM), where the power for each OPP is provided from OPP framework. This can cause some OPPs to be marked inefficient, while using micro-Watts precision that might not happen. Update all EM users which access 'power' field and assume the value is in milli-Watts. Solve also an issue with potential overflow in calculation of energy estimation on 32bit machine. It's needed now since the power value (thus the 'cost' as well) are higher. Example calculation which shows the rounding error and impact: power = 'dyn-power-coeff' * volt_mV * volt_mV * freq_MHz power_a_uW = (100 * 600mW * 600mW * 500MHz) / 10^6 = 18000 power_a_mW = (100 * 600mW * 600mW * 500MHz) / 10^9 = 18 power_b_uW = (100 * 605mW * 605mW * 600MHz) / 10^6 = 21961 power_b_mW = (100 * 605mW * 605mW * 600MHz) / 10^9 = 21 max_freq = 2000MHz cost_a_mW = 18 * 2000MHz/500MHz = 72 cost_a_uW = 18000 * 2000MHz/500MHz = 72000 cost_b_mW = 21 * 2000MHz/600MHz = 70 // <- artificially better cost_b_uW = 21961 * 2000MHz/600MHz = 73203 The 'cost_b_mW' (which is based on old milli-Watts) is misleadingly better that the 'cost_b_uW' (this patch uses micro-Watts) and such would have impact on the 'inefficient OPPs' information in the Cpufreq framework. This patch set removes the rounding issue. Signed-off-by: Lukasz Luba <lukasz.luba@arm.com> Acked-by: Daniel Lezcano <daniel.lezcano@linaro.org> Acked-by: Viresh Kumar <viresh.kumar@linaro.org> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2022-07-07 07:15:52 +00:00
power = pd->table[i].power * nr_cpus;
if (power > power_limit)
break;
}
freq = pd->table[i - 1].frequency;
freq_qos_update_request(&dtpm_cpu->qos_req, freq);
PM: EM: convert power field to micro-Watts precision and align drivers The milli-Watts precision causes rounding errors while calculating efficiency cost for each OPP. This is especially visible in the 'simple' Energy Model (EM), where the power for each OPP is provided from OPP framework. This can cause some OPPs to be marked inefficient, while using micro-Watts precision that might not happen. Update all EM users which access 'power' field and assume the value is in milli-Watts. Solve also an issue with potential overflow in calculation of energy estimation on 32bit machine. It's needed now since the power value (thus the 'cost' as well) are higher. Example calculation which shows the rounding error and impact: power = 'dyn-power-coeff' * volt_mV * volt_mV * freq_MHz power_a_uW = (100 * 600mW * 600mW * 500MHz) / 10^6 = 18000 power_a_mW = (100 * 600mW * 600mW * 500MHz) / 10^9 = 18 power_b_uW = (100 * 605mW * 605mW * 600MHz) / 10^6 = 21961 power_b_mW = (100 * 605mW * 605mW * 600MHz) / 10^9 = 21 max_freq = 2000MHz cost_a_mW = 18 * 2000MHz/500MHz = 72 cost_a_uW = 18000 * 2000MHz/500MHz = 72000 cost_b_mW = 21 * 2000MHz/600MHz = 70 // <- artificially better cost_b_uW = 21961 * 2000MHz/600MHz = 73203 The 'cost_b_mW' (which is based on old milli-Watts) is misleadingly better that the 'cost_b_uW' (this patch uses micro-Watts) and such would have impact on the 'inefficient OPPs' information in the Cpufreq framework. This patch set removes the rounding issue. Signed-off-by: Lukasz Luba <lukasz.luba@arm.com> Acked-by: Daniel Lezcano <daniel.lezcano@linaro.org> Acked-by: Viresh Kumar <viresh.kumar@linaro.org> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2022-07-07 07:15:52 +00:00
power_limit = pd->table[i - 1].power * nr_cpus;
return power_limit;
}
static u64 scale_pd_power_uw(struct cpumask *pd_mask, u64 power)
{
unsigned long max, sum_util = 0;
int cpu;
/*
* The capacity is the same for all CPUs belonging to
* the same perf domain.
*/
max = arch_scale_cpu_capacity(cpumask_first(pd_mask));
for_each_cpu_and(cpu, pd_mask, cpu_online_mask)
sum_util += sched_cpu_util(cpu);
return (power * ((sum_util << 10) / max)) >> 10;
}
static u64 get_pd_power_uw(struct dtpm *dtpm)
{
struct dtpm_cpu *dtpm_cpu = to_dtpm_cpu(dtpm);
struct em_perf_domain *pd;
struct cpumask *pd_mask;
unsigned long freq;
int i;
pd = em_cpu_get(dtpm_cpu->cpu);
pd_mask = em_span_cpus(pd);
freq = cpufreq_quick_get(dtpm_cpu->cpu);
for (i = 0; i < pd->nr_perf_states; i++) {
if (pd->table[i].frequency < freq)
continue;
return scale_pd_power_uw(pd_mask, pd->table[i].power *
MICROWATT_PER_MILLIWATT);
}
return 0;
}
static int update_pd_power_uw(struct dtpm *dtpm)
{
struct dtpm_cpu *dtpm_cpu = to_dtpm_cpu(dtpm);
struct em_perf_domain *em = em_cpu_get(dtpm_cpu->cpu);
struct cpumask cpus;
int nr_cpus;
cpumask_and(&cpus, cpu_online_mask, to_cpumask(em->cpus));
nr_cpus = cpumask_weight(&cpus);
dtpm->power_min = em->table[0].power;
dtpm->power_min *= MICROWATT_PER_MILLIWATT;
dtpm->power_min *= nr_cpus;
dtpm->power_max = em->table[em->nr_perf_states - 1].power;
dtpm->power_max *= MICROWATT_PER_MILLIWATT;
dtpm->power_max *= nr_cpus;
return 0;
}
static void pd_release(struct dtpm *dtpm)
{
struct dtpm_cpu *dtpm_cpu = to_dtpm_cpu(dtpm);
struct cpufreq_policy *policy;
if (freq_qos_request_active(&dtpm_cpu->qos_req))
freq_qos_remove_request(&dtpm_cpu->qos_req);
policy = cpufreq_cpu_get(dtpm_cpu->cpu);
if (policy) {
for_each_cpu(dtpm_cpu->cpu, policy->related_cpus)
per_cpu(dtpm_per_cpu, dtpm_cpu->cpu) = NULL;
}
kfree(dtpm_cpu);
}
static struct dtpm_ops dtpm_ops = {
.set_power_uw = set_pd_power_limit,
.get_power_uw = get_pd_power_uw,
.update_power_uw = update_pd_power_uw,
.release = pd_release,
};
static int cpuhp_dtpm_cpu_offline(unsigned int cpu)
{
struct dtpm_cpu *dtpm_cpu;
dtpm_cpu = per_cpu(dtpm_per_cpu, cpu);
if (dtpm_cpu)
dtpm_update_power(&dtpm_cpu->dtpm);
return 0;
}
static int cpuhp_dtpm_cpu_online(unsigned int cpu)
{
struct dtpm_cpu *dtpm_cpu;
dtpm_cpu = per_cpu(dtpm_per_cpu, cpu);
if (dtpm_cpu)
return dtpm_update_power(&dtpm_cpu->dtpm);
return 0;
}
static int __dtpm_cpu_setup(int cpu, struct dtpm *parent)
{
struct dtpm_cpu *dtpm_cpu;
struct cpufreq_policy *policy;
struct em_perf_domain *pd;
char name[CPUFREQ_NAME_LEN];
int ret = -ENOMEM;
dtpm_cpu = per_cpu(dtpm_per_cpu, cpu);
if (dtpm_cpu)
return 0;
policy = cpufreq_cpu_get(cpu);
if (!policy)
return 0;
pd = em_cpu_get(cpu);
if (!pd || em_is_artificial(pd))
return -EINVAL;
dtpm_cpu = kzalloc(sizeof(*dtpm_cpu), GFP_KERNEL);
if (!dtpm_cpu)
return -ENOMEM;
dtpm_init(&dtpm_cpu->dtpm, &dtpm_ops);
dtpm_cpu->cpu = cpu;
for_each_cpu(cpu, policy->related_cpus)
per_cpu(dtpm_per_cpu, cpu) = dtpm_cpu;
snprintf(name, sizeof(name), "cpu%d-cpufreq", dtpm_cpu->cpu);
ret = dtpm_register(name, &dtpm_cpu->dtpm, parent);
if (ret)
goto out_kfree_dtpm_cpu;
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_cpu->dtpm);
dtpm_cpu = NULL;
out_kfree_dtpm_cpu:
for_each_cpu(cpu, policy->related_cpus)
per_cpu(dtpm_per_cpu, cpu) = NULL;
kfree(dtpm_cpu);
return ret;
}
static int dtpm_cpu_setup(struct dtpm *dtpm, struct device_node *np)
{
int cpu;
cpu = of_cpu_node_to_id(np);
if (cpu < 0)
return 0;
return __dtpm_cpu_setup(cpu, dtpm);
}
static int dtpm_cpu_init(void)
{
int ret;
/*
* The callbacks at CPU hotplug time are calling
* dtpm_update_power() which in turns calls update_pd_power().
*
* The function update_pd_power() uses the online mask to
* figure out the power consumption limits.
*
* At CPUHP_AP_ONLINE_DYN, the CPU is present in the CPU
* online mask when the cpuhp_dtpm_cpu_online function is
* called, but the CPU is still in the online mask for the
* tear down callback. So the power can not be updated when
* the CPU is unplugged.
*
* At CPUHP_AP_DTPM_CPU_DEAD, the situation is the opposite as
* above. The CPU online mask is not up to date when the CPU
* is plugged in.
*
* For this reason, we need to call the online and offline
* callbacks at different moments when the CPU online mask is
* consistent with the power numbers we want to update.
*/
ret = cpuhp_setup_state(CPUHP_AP_DTPM_CPU_DEAD, "dtpm_cpu:offline",
NULL, cpuhp_dtpm_cpu_offline);
if (ret < 0)
return ret;
ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "dtpm_cpu:online",
cpuhp_dtpm_cpu_online, NULL);
if (ret < 0)
return ret;
return 0;
}
static void dtpm_cpu_exit(void)
{
cpuhp_remove_state_nocalls(CPUHP_AP_ONLINE_DYN);
cpuhp_remove_state_nocalls(CPUHP_AP_DTPM_CPU_DEAD);
}
struct dtpm_subsys_ops dtpm_cpu_ops = {
.name = KBUILD_MODNAME,
.init = dtpm_cpu_init,
.exit = dtpm_cpu_exit,
.setup = dtpm_cpu_setup,
};