* Re: [PATCH 3/4] PM / devfreq: Add cpu based scaling support to passive governor
2021-06-17 6:05 ` [PATCH 3/4] PM / devfreq: Add cpu based scaling support to passive governor Chanwoo Choi
@ 2021-06-17 5:51 ` Hsin-Yi Wang
2021-06-17 6:13 ` Chanwoo Choi
2021-06-22 17:42 ` Matthias Kaehlcke
` (2 subsequent siblings)
3 siblings, 1 reply; 15+ messages in thread
From: Hsin-Yi Wang @ 2021-06-17 5:51 UTC (permalink / raw)
To: Chanwoo Choi
Cc: Andrew-sh.Cheng, Sibi Sankar, Saravana Kannan, MyungJoo Ham,
Kyungmin Park, chanwoo, cwchoi00, Linux PM, lkml,
Saravana Kannan
On Thu, Jun 17, 2021 at 1:46 PM Chanwoo Choi <cw00.choi@samsung.com> wrote:
>
> From: Saravana Kannan <skannan@codeaurora.org>
>
> Many CPU architectures have caches that can scale independent of the
> CPUs. Frequency scaling of the caches is necessary to make sure that the
> cache is not a performance bottleneck that leads to poor performance and
> power. The same idea applies for RAM/DDR.
>
> To achieve this, this patch adds support for cpu based scaling to the
> passive governor. This is accomplished by taking the current frequency
> of each CPU frequency domain and then adjust the frequency of the cache
> (or any devfreq device) based on the frequency of the CPUs. It listens
> to CPU frequency transition notifiers to keep itself up to date on the
> current CPU frequency.
>
> To decide the frequency of the device, the governor does one of the
> following:
> * Derives the optimal devfreq device opp from required-opps property of
> the parent cpu opp_table.
>
> * Scales the device frequency in proportion to the CPU frequency. So, if
> the CPUs are running at their max frequency, the device runs at its
> max frequency. If the CPUs are running at their min frequency, the
> device runs at its min frequency. It is interpolated for frequencies
> in between.
>
> Signed-off-by: Saravana Kannan <skannan@codeaurora.org>
> [Sibi: Integrated cpu-freqmap governor into passive_governor]
> Signed-off-by: Sibi Sankar <sibis@codeaurora.org>
> [Chanwoo: Fix conflict with latest code and clean code up]
> Signed-off-by: Chanwoo Choi <cw00.choi@samsung.com>
> ---
> drivers/devfreq/governor.h | 22 +++
> drivers/devfreq/governor_passive.c | 264 ++++++++++++++++++++++++++++-
> include/linux/devfreq.h | 16 +-
> 3 files changed, 293 insertions(+), 9 deletions(-)
>
> diff --git a/drivers/devfreq/governor.h b/drivers/devfreq/governor.h
> index 9a9495f94ac6..3c36c92c89a9 100644
> --- a/drivers/devfreq/governor.h
> +++ b/drivers/devfreq/governor.h
> @@ -47,6 +47,28 @@
> #define DEVFREQ_GOV_ATTR_POLLING_INTERVAL BIT(0)
> #define DEVFREQ_GOV_ATTR_TIMER BIT(1)
>
> +/**
> + * struct devfreq_cpu_data - Hold the per-cpu data
> + * @dev: reference to cpu device.
> + * @first_cpu: the cpumask of the first cpu of a policy.
> + * @opp_table: reference to cpu opp table.
> + * @cur_freq: the current frequency of the cpu.
> + * @min_freq: the min frequency of the cpu.
> + * @max_freq: the max frequency of the cpu.
> + *
> + * This structure stores the required cpu_data of a cpu.
> + * This is auto-populated by the governor.
> + */
> +struct devfreq_cpu_data {
> + struct device *dev;
> + unsigned int first_cpu;
> +
> + struct opp_table *opp_table;
> + unsigned int cur_freq;
> + unsigned int min_freq;
> + unsigned int max_freq;
> +};
> +
> /**
> * struct devfreq_governor - Devfreq policy governor
> * @node: list node - contains registered devfreq governors
> diff --git a/drivers/devfreq/governor_passive.c b/drivers/devfreq/governor_passive.c
> index fc09324a03e0..07e864509b7e 100644
> --- a/drivers/devfreq/governor_passive.c
> +++ b/drivers/devfreq/governor_passive.c
> @@ -8,11 +8,84 @@
> */
>
> #include <linux/module.h>
> +#include <linux/cpu.h>
> +#include <linux/cpufreq.h>
> +#include <linux/cpumask.h>
> +#include <linux/slab.h>
> #include <linux/device.h>
> #include <linux/devfreq.h>
> #include "governor.h"
>
> -static int devfreq_passive_get_target_freq(struct devfreq *devfreq,
> +#define HZ_PER_KHZ 1000
> +
> +static unsigned long get_taget_freq_by_required_opp(struct device *p_dev,
> + struct opp_table *p_opp_table,
> + struct opp_table *opp_table,
> + unsigned long freq)
> +{
> + struct dev_pm_opp *opp = NULL, *p_opp = NULL;
> +
> + if (!p_dev || !p_opp_table || !opp_table || !freq)
> + return 0;
> +
> + p_opp = devfreq_recommended_opp(p_dev, &freq, 0);
> + if (IS_ERR(p_opp))
> + return 0;
> +
> + opp = dev_pm_opp_xlate_required_opp(p_opp_table, opp_table, p_opp);
> + dev_pm_opp_put(p_opp);
> +
> + if (IS_ERR(opp))
> + return 0;
> +
> + freq = dev_pm_opp_get_freq(opp);
> + dev_pm_opp_put(opp);
> +
> + return freq;
> +}
> +
> +static int get_target_freq_with_cpufreq(struct devfreq *devfreq,
> + unsigned long *target_freq)
> +{
> + struct devfreq_passive_data *p_data =
> + (struct devfreq_passive_data *)devfreq->data;
> + struct devfreq_cpu_data *cpu_data;
We might have to rename the cpu_data variable.
For some architectures, cpu_data is defined as macro. This results in
errors such as
include/linux/devfreq.h:331:27: note: in expansion of macro 'cpu_data'
331 | struct devfreq_cpu_data *cpu_data[NR_CPUS];
| ^~~~~~~~
In file included from include/linux/devfreq_cooling.h:13,
from drivers/devfreq/devfreq.c:14:
include/linux/devfreq.h:332:1: warning: no semicolon at end of
struct or union
> + unsigned long cpu, cpu_cur, cpu_min, cpu_max, cpu_percent;
> + unsigned long dev_min, dev_max;
> + unsigned long freq = 0;
> +
> + for_each_online_cpu(cpu) {
> + cpu_data = p_data->cpu_data[cpu];
> + if (!cpu_data || cpu_data->first_cpu != cpu)
> + continue;
> +
> + /* Get target freq via required opps */
> + cpu_cur = cpu_data->cur_freq * HZ_PER_KHZ;
> + freq = get_taget_freq_by_required_opp(cpu_data->dev,
> + cpu_data->opp_table,
> + devfreq->opp_table, cpu_cur);
> + if (freq) {
> + *target_freq = max(freq, *target_freq);
> + continue;
> + }
> +
> + /* Use Interpolation if required opps is not available */
> + devfreq_get_freq_range(devfreq, &dev_min, &dev_max);
> +
> + cpu_min = cpu_data->min_freq;
> + cpu_max = cpu_data->max_freq;
> + cpu_cur = cpu_data->cur_freq;
> +
> + cpu_percent = ((cpu_cur - cpu_min) * 100) / (cpu_max - cpu_min);
> + freq = dev_min + mult_frac(dev_max - dev_min, cpu_percent, 100);
> +
> + *target_freq = max(freq, *target_freq);
> + }
> +
> + return 0;
> +}
> +
> +static int get_target_freq_with_devfreq(struct devfreq *devfreq,
> unsigned long *freq)
> {
> struct devfreq_passive_data *p_data
> @@ -99,6 +172,172 @@ static int devfreq_passive_get_target_freq(struct devfreq *devfreq,
> return 0;
> }
>
> +static int devfreq_passive_get_target_freq(struct devfreq *devfreq,
> + unsigned long *freq)
> +{
> + struct devfreq_passive_data *p_data =
> + (struct devfreq_passive_data *)devfreq->data;
> + int ret;
> +
> + if (!p_data)
> + return -EINVAL;
> +
> + /*
> + * If the devfreq device with passive governor has the specific method
> + * to determine the next frequency, should use the get_target_freq()
> + * of struct devfreq_passive_data.
> + */
> + if (p_data->get_target_freq)
> + return p_data->get_target_freq(devfreq, freq);
> +
> + switch (p_data->parent_type) {
> + case DEVFREQ_PARENT_DEV:
> + ret = get_target_freq_with_devfreq(devfreq, freq);
> + break;
> + case CPUFREQ_PARENT_DEV:
> + ret = get_target_freq_with_cpufreq(devfreq, freq);
> + break;
> + default:
> + ret = -EINVAL;
> + dev_err(&devfreq->dev, "Invalid parent type\n");
> + break;
> + }
> +
> + return ret;
> +}
> +
> +static int cpufreq_passive_notifier_call(struct notifier_block *nb,
> + unsigned long event, void *ptr)
> +{
> + struct devfreq_passive_data *data =
> + container_of(nb, struct devfreq_passive_data, nb);
> + struct devfreq *devfreq = (struct devfreq *)data->this;
> + struct devfreq_cpu_data *cpu_data;
> + struct cpufreq_freqs *freqs = ptr;
> + unsigned int cur_freq;
> + int ret;
> +
> + if (event != CPUFREQ_POSTCHANGE || !freqs ||
> + !data->cpu_data[freqs->policy->cpu])
> + return 0;
> +
> + cpu_data = data->cpu_data[freqs->policy->cpu];
> + if (cpu_data->cur_freq == freqs->new)
> + return 0;
> +
> + cur_freq = cpu_data->cur_freq;
> + cpu_data->cur_freq = freqs->new;
> +
> + mutex_lock(&devfreq->lock);
> + ret = devfreq_update_target(devfreq, freqs->new);
> + mutex_unlock(&devfreq->lock);
> + if (ret) {
> + cpu_data->cur_freq = cur_freq;
> + dev_err(&devfreq->dev, "failed to update the frequency.\n");
> + return ret;
> + }
> +
> + return 0;
> +}
> +
> +static int cpufreq_passive_register_notifier(struct devfreq *devfreq)
> +{
> + struct devfreq_passive_data *p_data
> + = (struct devfreq_passive_data *)devfreq->data;
> + struct device *dev = devfreq->dev.parent;
> + struct opp_table *opp_table = NULL;
> + struct devfreq_cpu_data *cpu_data;
> + struct cpufreq_policy *policy;
> + struct device *cpu_dev;
> + unsigned int cpu;
> + int ret;
> +
> + get_online_cpus();
> +
> + p_data->nb.notifier_call = cpufreq_passive_notifier_call;
> + ret = cpufreq_register_notifier(&p_data->nb, CPUFREQ_TRANSITION_NOTIFIER);
> + if (ret) {
> + dev_err(dev, "failed to register cpufreq notifier\n");
> + p_data->nb.notifier_call = NULL;
> + goto out;
> + }
> +
> + for_each_online_cpu(cpu) {
> + if (p_data->cpu_data[cpu])
> + continue;
> +
> + policy = cpufreq_cpu_get(cpu);
> + if (policy) {
> + cpu_data = kzalloc(sizeof(*cpu_data), GFP_KERNEL);
> + if (!cpu_data) {
> + ret = -ENOMEM;
> + goto out;
> + }
> +
> + cpu_dev = get_cpu_device(cpu);
> + if (!cpu_dev) {
> + dev_err(dev, "failed to get cpu device\n");
> + ret = -ENODEV;
> + goto out;
> + }
> +
> + opp_table = dev_pm_opp_get_opp_table(cpu_dev);
> + if (IS_ERR(opp_table)) {
> + ret = PTR_ERR(opp_table);
> + goto out;
> + }
> +
> + cpu_data->dev = cpu_dev;
> + cpu_data->opp_table = opp_table;
> + cpu_data->first_cpu = cpumask_first(policy->related_cpus);
> + cpu_data->cur_freq = policy->cur;
> + cpu_data->min_freq = policy->cpuinfo.min_freq;
> + cpu_data->max_freq = policy->cpuinfo.max_freq;
> +
> + p_data->cpu_data[cpu] = cpu_data;
> + cpufreq_cpu_put(policy);
> + } else {
> + ret = -EPROBE_DEFER;
> + goto out;
> + }
> + }
> +out:
> + put_online_cpus();
> + if (ret)
> + return ret;
> +
> + mutex_lock(&devfreq->lock);
> + ret = devfreq_update_target(devfreq, 0L);
> + mutex_unlock(&devfreq->lock);
> + if (ret)
> + dev_err(dev, "failed to update the frequency\n");
> +
> + return ret;
> +}
> +
> +static int cpufreq_passive_unregister_notifier(struct devfreq *devfreq)
> +{
> + struct devfreq_passive_data *p_data
> + = (struct devfreq_passive_data *)devfreq->data;
> + struct devfreq_cpu_data *cpu_data;
> + int cpu;
> +
> + if (p_data->nb.notifier_call)
> + cpufreq_unregister_notifier(&p_data->nb, CPUFREQ_TRANSITION_NOTIFIER);
> +
> + for_each_possible_cpu(cpu) {
> + cpu_data = p_data->cpu_data[cpu];
> + if (cpu_data) {
> + if (cpu_data->opp_table)
> + dev_pm_opp_put_opp_table(cpu_data->opp_table);
> + kfree(cpu_data);
> + cpu_data = NULL;
> + }
> + }
> +
> + return 0;
> +}
> +
> static int devfreq_passive_notifier_call(struct notifier_block *nb,
> unsigned long event, void *ptr)
> {
> @@ -140,7 +379,7 @@ static int devfreq_passive_event_handler(struct devfreq *devfreq,
> struct notifier_block *nb = &p_data->nb;
> int ret = 0;
>
> - if (!parent)
> + if (p_data->parent_type == DEVFREQ_PARENT_DEV && !parent)
> return -EPROBE_DEFER;
>
> switch (event) {
> @@ -148,13 +387,24 @@ static int devfreq_passive_event_handler(struct devfreq *devfreq,
> if (!p_data->this)
> p_data->this = devfreq;
>
> - nb->notifier_call = devfreq_passive_notifier_call;
> - ret = devfreq_register_notifier(parent, nb,
> - DEVFREQ_TRANSITION_NOTIFIER);
> + if (p_data->parent_type == DEVFREQ_PARENT_DEV) {
> + nb->notifier_call = devfreq_passive_notifier_call;
> + ret = devfreq_register_notifier(parent, nb,
> + DEVFREQ_TRANSITION_NOTIFIER);
> + } else if (p_data->parent_type == CPUFREQ_PARENT_DEV) {
> + ret = cpufreq_passive_register_notifier(devfreq);
> + } else {
> + ret = -EINVAL;
> + }
> break;
> case DEVFREQ_GOV_STOP:
> - WARN_ON(devfreq_unregister_notifier(parent, nb,
> - DEVFREQ_TRANSITION_NOTIFIER));
> + if (p_data->parent_type == DEVFREQ_PARENT_DEV)
> + WARN_ON(devfreq_unregister_notifier(parent, nb,
> + DEVFREQ_TRANSITION_NOTIFIER));
> + else if (p_data->parent_type == CPUFREQ_PARENT_DEV)
> + WARN_ON(cpufreq_passive_unregister_notifier(devfreq));
> + else
> + ret = -EINVAL;
> break;
> default:
> break;
> diff --git a/include/linux/devfreq.h b/include/linux/devfreq.h
> index 142474b4af96..cfa0ef54841e 100644
> --- a/include/linux/devfreq.h
> +++ b/include/linux/devfreq.h
> @@ -38,6 +38,7 @@ enum devfreq_timer {
>
> struct devfreq;
> struct devfreq_governor;
> +struct devfreq_cpu_data;
> struct thermal_cooling_device;
>
> /**
> @@ -288,6 +289,11 @@ struct devfreq_simple_ondemand_data {
> #endif
>
> #if IS_ENABLED(CONFIG_DEVFREQ_GOV_PASSIVE)
> +enum devfreq_parent_dev_type {
> + DEVFREQ_PARENT_DEV,
> + CPUFREQ_PARENT_DEV,
> +};
> +
> /**
> * struct devfreq_passive_data - ``void *data`` fed to struct devfreq
> * and devfreq_add_device
> @@ -299,8 +305,10 @@ struct devfreq_simple_ondemand_data {
> * using governors except for passive governor.
> * If the devfreq device has the specific method to decide
> * the next frequency, should use this callback.
> - * @this: the devfreq instance of own device.
> - * @nb: the notifier block for DEVFREQ_TRANSITION_NOTIFIER list
> + + * @parent_type parent type of the device
> + + * @this: the devfreq instance of own device.
> + + * @nb: the notifier block for DEVFREQ_TRANSITION_NOTIFIER list
> + + * @cpu_data: the state min/max/current frequency of all online cpu's
> *
> * The devfreq_passive_data have to set the devfreq instance of parent
> * device with governors except for the passive governor. But, don't need to
> @@ -314,9 +322,13 @@ struct devfreq_passive_data {
> /* Optional callback to decide the next frequency of passvice device */
> int (*get_target_freq)(struct devfreq *this, unsigned long *freq);
>
> + /* Should set the type of parent device */
> + enum devfreq_parent_dev_type parent_type;
> +
> /* For passive governor's internal use. Don't need to set them */
> struct devfreq *this;
> struct notifier_block nb;
> + struct devfreq_cpu_data *cpu_data[NR_CPUS];
> };
> #endif
>
> --
> 2.17.1
>
^ permalink raw reply [flat|nested] 15+ messages in thread* Re: [PATCH 3/4] PM / devfreq: Add cpu based scaling support to passive governor
2021-06-17 5:51 ` Hsin-Yi Wang
@ 2021-06-17 6:13 ` Chanwoo Choi
0 siblings, 0 replies; 15+ messages in thread
From: Chanwoo Choi @ 2021-06-17 6:13 UTC (permalink / raw)
To: Hsin-Yi Wang
Cc: Andrew-sh.Cheng, Sibi Sankar, Saravana Kannan, MyungJoo Ham,
Kyungmin Park, chanwoo, cwchoi00, Linux PM, lkml,
Saravana Kannan
On 6/17/21 2:51 PM, Hsin-Yi Wang wrote:
> On Thu, Jun 17, 2021 at 1:46 PM Chanwoo Choi <cw00.choi@samsung.com> wrote:
>>
>> From: Saravana Kannan <skannan@codeaurora.org>
>>
>> Many CPU architectures have caches that can scale independent of the
>> CPUs. Frequency scaling of the caches is necessary to make sure that the
>> cache is not a performance bottleneck that leads to poor performance and
>> power. The same idea applies for RAM/DDR.
>>
>> To achieve this, this patch adds support for cpu based scaling to the
>> passive governor. This is accomplished by taking the current frequency
>> of each CPU frequency domain and then adjust the frequency of the cache
>> (or any devfreq device) based on the frequency of the CPUs. It listens
>> to CPU frequency transition notifiers to keep itself up to date on the
>> current CPU frequency.
>>
>> To decide the frequency of the device, the governor does one of the
>> following:
>> * Derives the optimal devfreq device opp from required-opps property of
>> the parent cpu opp_table.
>>
>> * Scales the device frequency in proportion to the CPU frequency. So, if
>> the CPUs are running at their max frequency, the device runs at its
>> max frequency. If the CPUs are running at their min frequency, the
>> device runs at its min frequency. It is interpolated for frequencies
>> in between.
>>
>> Signed-off-by: Saravana Kannan <skannan@codeaurora.org>
>> [Sibi: Integrated cpu-freqmap governor into passive_governor]
>> Signed-off-by: Sibi Sankar <sibis@codeaurora.org>
>> [Chanwoo: Fix conflict with latest code and clean code up]
>> Signed-off-by: Chanwoo Choi <cw00.choi@samsung.com>
>> ---
>> drivers/devfreq/governor.h | 22 +++
>> drivers/devfreq/governor_passive.c | 264 ++++++++++++++++++++++++++++-
>> include/linux/devfreq.h | 16 +-
>> 3 files changed, 293 insertions(+), 9 deletions(-)
>>
>> diff --git a/drivers/devfreq/governor.h b/drivers/devfreq/governor.h
>> index 9a9495f94ac6..3c36c92c89a9 100644
>> --- a/drivers/devfreq/governor.h
>> +++ b/drivers/devfreq/governor.h
>> @@ -47,6 +47,28 @@
>> #define DEVFREQ_GOV_ATTR_POLLING_INTERVAL BIT(0)
>> #define DEVFREQ_GOV_ATTR_TIMER BIT(1)
>>
>> +/**
>> + * struct devfreq_cpu_data - Hold the per-cpu data
>> + * @dev: reference to cpu device.
>> + * @first_cpu: the cpumask of the first cpu of a policy.
>> + * @opp_table: reference to cpu opp table.
>> + * @cur_freq: the current frequency of the cpu.
>> + * @min_freq: the min frequency of the cpu.
>> + * @max_freq: the max frequency of the cpu.
>> + *
>> + * This structure stores the required cpu_data of a cpu.
>> + * This is auto-populated by the governor.
>> + */
>> +struct devfreq_cpu_data {
>> + struct device *dev;
>> + unsigned int first_cpu;
>> +
>> + struct opp_table *opp_table;
>> + unsigned int cur_freq;
>> + unsigned int min_freq;
>> + unsigned int max_freq;
>> +};
>> +
>> /**
>> * struct devfreq_governor - Devfreq policy governor
>> * @node: list node - contains registered devfreq governors
>> diff --git a/drivers/devfreq/governor_passive.c b/drivers/devfreq/governor_passive.c
>> index fc09324a03e0..07e864509b7e 100644
>> --- a/drivers/devfreq/governor_passive.c
>> +++ b/drivers/devfreq/governor_passive.c
>> @@ -8,11 +8,84 @@
>> */
>>
>> #include <linux/module.h>
>> +#include <linux/cpu.h>
>> +#include <linux/cpufreq.h>
>> +#include <linux/cpumask.h>
>> +#include <linux/slab.h>
>> #include <linux/device.h>
>> #include <linux/devfreq.h>
>> #include "governor.h"
>>
>> -static int devfreq_passive_get_target_freq(struct devfreq *devfreq,
>> +#define HZ_PER_KHZ 1000
>> +
>> +static unsigned long get_taget_freq_by_required_opp(struct device *p_dev,
>> + struct opp_table *p_opp_table,
>> + struct opp_table *opp_table,
>> + unsigned long freq)
>> +{
>> + struct dev_pm_opp *opp = NULL, *p_opp = NULL;
>> +
>> + if (!p_dev || !p_opp_table || !opp_table || !freq)
>> + return 0;
>> +
>> + p_opp = devfreq_recommended_opp(p_dev, &freq, 0);
>> + if (IS_ERR(p_opp))
>> + return 0;
>> +
>> + opp = dev_pm_opp_xlate_required_opp(p_opp_table, opp_table, p_opp);
>> + dev_pm_opp_put(p_opp);
>> +
>> + if (IS_ERR(opp))
>> + return 0;
>> +
>> + freq = dev_pm_opp_get_freq(opp);
>> + dev_pm_opp_put(opp);
>> +
>> + return freq;
>> +}
>> +
>> +static int get_target_freq_with_cpufreq(struct devfreq *devfreq,
>> + unsigned long *target_freq)
>> +{
>> + struct devfreq_passive_data *p_data =
>> + (struct devfreq_passive_data *)devfreq->data;
>> + struct devfreq_cpu_data *cpu_data;
> We might have to rename the cpu_data variable.
>
> For some architectures, cpu_data is defined as macro. This results in
> errors such as
>
> include/linux/devfreq.h:331:27: note: in expansion of macro 'cpu_data'
> 331 | struct devfreq_cpu_data *cpu_data[NR_CPUS];
> | ^~~~~~~~
> In file included from include/linux/devfreq_cooling.h:13,
> from drivers/devfreq/devfreq.c:14:
> include/linux/devfreq.h:332:1: warning: no semicolon at end of
> struct or union
OK. I'll rename.
>
>> + unsigned long cpu, cpu_cur, cpu_min, cpu_max, cpu_percent;
>> + unsigned long dev_min, dev_max;
>> + unsigned long freq = 0;
>> +
>> + for_each_online_cpu(cpu) {
>> + cpu_data = p_data->cpu_data[cpu];
>> + if (!cpu_data || cpu_data->first_cpu != cpu)
>> + continue;
>> +
>> + /* Get target freq via required opps */
>> + cpu_cur = cpu_data->cur_freq * HZ_PER_KHZ;
>> + freq = get_taget_freq_by_required_opp(cpu_data->dev,
>> + cpu_data->opp_table,
>> + devfreq->opp_table, cpu_cur);
>> + if (freq) {
>> + *target_freq = max(freq, *target_freq);
>> + continue;
>> + }
>> +
>> + /* Use Interpolation if required opps is not available */
>> + devfreq_get_freq_range(devfreq, &dev_min, &dev_max);
>> +
>> + cpu_min = cpu_data->min_freq;
>> + cpu_max = cpu_data->max_freq;
>> + cpu_cur = cpu_data->cur_freq;
>> +
>> + cpu_percent = ((cpu_cur - cpu_min) * 100) / (cpu_max - cpu_min);
>> + freq = dev_min + mult_frac(dev_max - dev_min, cpu_percent, 100);
>> +
>> + *target_freq = max(freq, *target_freq);
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +static int get_target_freq_with_devfreq(struct devfreq *devfreq,
>> unsigned long *freq)
>> {
>> struct devfreq_passive_data *p_data
>> @@ -99,6 +172,172 @@ static int devfreq_passive_get_target_freq(struct devfreq *devfreq,
>> return 0;
>> }
>>
>> +static int devfreq_passive_get_target_freq(struct devfreq *devfreq,
>> + unsigned long *freq)
>> +{
>> + struct devfreq_passive_data *p_data =
>> + (struct devfreq_passive_data *)devfreq->data;
>> + int ret;
>> +
>> + if (!p_data)
>> + return -EINVAL;
>> +
>> + /*
>> + * If the devfreq device with passive governor has the specific method
>> + * to determine the next frequency, should use the get_target_freq()
>> + * of struct devfreq_passive_data.
>> + */
>> + if (p_data->get_target_freq)
>> + return p_data->get_target_freq(devfreq, freq);
>> +
>> + switch (p_data->parent_type) {
>> + case DEVFREQ_PARENT_DEV:
>> + ret = get_target_freq_with_devfreq(devfreq, freq);
>> + break;
>> + case CPUFREQ_PARENT_DEV:
>> + ret = get_target_freq_with_cpufreq(devfreq, freq);
>> + break;
>> + default:
>> + ret = -EINVAL;
>> + dev_err(&devfreq->dev, "Invalid parent type\n");
>> + break;
>> + }
>> +
>> + return ret;
>> +}
>> +
>> +static int cpufreq_passive_notifier_call(struct notifier_block *nb,
>> + unsigned long event, void *ptr)
>> +{
>> + struct devfreq_passive_data *data =
>> + container_of(nb, struct devfreq_passive_data, nb);
>> + struct devfreq *devfreq = (struct devfreq *)data->this;
>> + struct devfreq_cpu_data *cpu_data;
>> + struct cpufreq_freqs *freqs = ptr;
>> + unsigned int cur_freq;
>> + int ret;
>> +
>> + if (event != CPUFREQ_POSTCHANGE || !freqs ||
>> + !data->cpu_data[freqs->policy->cpu])
>> + return 0;
>> +
>> + cpu_data = data->cpu_data[freqs->policy->cpu];
>> + if (cpu_data->cur_freq == freqs->new)
>> + return 0;
>> +
>> + cur_freq = cpu_data->cur_freq;
>> + cpu_data->cur_freq = freqs->new;
>> +
>> + mutex_lock(&devfreq->lock);
>> + ret = devfreq_update_target(devfreq, freqs->new);
>> + mutex_unlock(&devfreq->lock);
>> + if (ret) {
>> + cpu_data->cur_freq = cur_freq;
>> + dev_err(&devfreq->dev, "failed to update the frequency.\n");
>> + return ret;
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +static int cpufreq_passive_register_notifier(struct devfreq *devfreq)
>> +{
>> + struct devfreq_passive_data *p_data
>> + = (struct devfreq_passive_data *)devfreq->data;
>> + struct device *dev = devfreq->dev.parent;
>> + struct opp_table *opp_table = NULL;
>> + struct devfreq_cpu_data *cpu_data;
>> + struct cpufreq_policy *policy;
>> + struct device *cpu_dev;
>> + unsigned int cpu;
>> + int ret;
>> +
>> + get_online_cpus();
>> +
>> + p_data->nb.notifier_call = cpufreq_passive_notifier_call;
>> + ret = cpufreq_register_notifier(&p_data->nb, CPUFREQ_TRANSITION_NOTIFIER);
>> + if (ret) {
>> + dev_err(dev, "failed to register cpufreq notifier\n");
>> + p_data->nb.notifier_call = NULL;
>> + goto out;
>> + }
>> +
>> + for_each_online_cpu(cpu) {
>> + if (p_data->cpu_data[cpu])
>> + continue;
>> +
>> + policy = cpufreq_cpu_get(cpu);
>> + if (policy) {
>> + cpu_data = kzalloc(sizeof(*cpu_data), GFP_KERNEL);
>> + if (!cpu_data) {
>> + ret = -ENOMEM;
>> + goto out;
>> + }
>> +
>> + cpu_dev = get_cpu_device(cpu);
>> + if (!cpu_dev) {
>> + dev_err(dev, "failed to get cpu device\n");
>> + ret = -ENODEV;
>> + goto out;
>> + }
>> +
>> + opp_table = dev_pm_opp_get_opp_table(cpu_dev);
>> + if (IS_ERR(opp_table)) {
>> + ret = PTR_ERR(opp_table);
>> + goto out;
>> + }
>> +
>> + cpu_data->dev = cpu_dev;
>> + cpu_data->opp_table = opp_table;
>> + cpu_data->first_cpu = cpumask_first(policy->related_cpus);
>> + cpu_data->cur_freq = policy->cur;
>> + cpu_data->min_freq = policy->cpuinfo.min_freq;
>> + cpu_data->max_freq = policy->cpuinfo.max_freq;
>> +
>> + p_data->cpu_data[cpu] = cpu_data;
>> + cpufreq_cpu_put(policy);
>> + } else {
>> + ret = -EPROBE_DEFER;
>> + goto out;
>> + }
>> + }
>> +out:
>> + put_online_cpus();
>> + if (ret)
>> + return ret;
>> +
>> + mutex_lock(&devfreq->lock);
>> + ret = devfreq_update_target(devfreq, 0L);
>> + mutex_unlock(&devfreq->lock);
>> + if (ret)
>> + dev_err(dev, "failed to update the frequency\n");
>> +
>> + return ret;
>> +}
>> +
>> +static int cpufreq_passive_unregister_notifier(struct devfreq *devfreq)
>> +{
>> + struct devfreq_passive_data *p_data
>> + = (struct devfreq_passive_data *)devfreq->data;
>> + struct devfreq_cpu_data *cpu_data;
>> + int cpu;
>> +
>> + if (p_data->nb.notifier_call)
>> + cpufreq_unregister_notifier(&p_data->nb, CPUFREQ_TRANSITION_NOTIFIER);
>> +
>> + for_each_possible_cpu(cpu) {
>> + cpu_data = p_data->cpu_data[cpu];
>> + if (cpu_data) {
>> + if (cpu_data->opp_table)
>> + dev_pm_opp_put_opp_table(cpu_data->opp_table);
>> + kfree(cpu_data);
>> + cpu_data = NULL;
>> + }
>> + }
>> +
>> + return 0;
>> +}
>> +
>> static int devfreq_passive_notifier_call(struct notifier_block *nb,
>> unsigned long event, void *ptr)
>> {
>> @@ -140,7 +379,7 @@ static int devfreq_passive_event_handler(struct devfreq *devfreq,
>> struct notifier_block *nb = &p_data->nb;
>> int ret = 0;
>>
>> - if (!parent)
>> + if (p_data->parent_type == DEVFREQ_PARENT_DEV && !parent)
>> return -EPROBE_DEFER;
>>
>> switch (event) {
>> @@ -148,13 +387,24 @@ static int devfreq_passive_event_handler(struct devfreq *devfreq,
>> if (!p_data->this)
>> p_data->this = devfreq;
>>
>> - nb->notifier_call = devfreq_passive_notifier_call;
>> - ret = devfreq_register_notifier(parent, nb,
>> - DEVFREQ_TRANSITION_NOTIFIER);
>> + if (p_data->parent_type == DEVFREQ_PARENT_DEV) {
>> + nb->notifier_call = devfreq_passive_notifier_call;
>> + ret = devfreq_register_notifier(parent, nb,
>> + DEVFREQ_TRANSITION_NOTIFIER);
>> + } else if (p_data->parent_type == CPUFREQ_PARENT_DEV) {
>> + ret = cpufreq_passive_register_notifier(devfreq);
>> + } else {
>> + ret = -EINVAL;
>> + }
>> break;
>> case DEVFREQ_GOV_STOP:
>> - WARN_ON(devfreq_unregister_notifier(parent, nb,
>> - DEVFREQ_TRANSITION_NOTIFIER));
>> + if (p_data->parent_type == DEVFREQ_PARENT_DEV)
>> + WARN_ON(devfreq_unregister_notifier(parent, nb,
>> + DEVFREQ_TRANSITION_NOTIFIER));
>> + else if (p_data->parent_type == CPUFREQ_PARENT_DEV)
>> + WARN_ON(cpufreq_passive_unregister_notifier(devfreq));
>> + else
>> + ret = -EINVAL;
>> break;
>> default:
>> break;
>> diff --git a/include/linux/devfreq.h b/include/linux/devfreq.h
>> index 142474b4af96..cfa0ef54841e 100644
>> --- a/include/linux/devfreq.h
>> +++ b/include/linux/devfreq.h
>> @@ -38,6 +38,7 @@ enum devfreq_timer {
>>
>> struct devfreq;
>> struct devfreq_governor;
>> +struct devfreq_cpu_data;
>> struct thermal_cooling_device;
>>
>> /**
>> @@ -288,6 +289,11 @@ struct devfreq_simple_ondemand_data {
>> #endif
>>
>> #if IS_ENABLED(CONFIG_DEVFREQ_GOV_PASSIVE)
>> +enum devfreq_parent_dev_type {
>> + DEVFREQ_PARENT_DEV,
>> + CPUFREQ_PARENT_DEV,
>> +};
>> +
>> /**
>> * struct devfreq_passive_data - ``void *data`` fed to struct devfreq
>> * and devfreq_add_device
>> @@ -299,8 +305,10 @@ struct devfreq_simple_ondemand_data {
>> * using governors except for passive governor.
>> * If the devfreq device has the specific method to decide
>> * the next frequency, should use this callback.
>> - * @this: the devfreq instance of own device.
>> - * @nb: the notifier block for DEVFREQ_TRANSITION_NOTIFIER list
>> + + * @parent_type parent type of the device
>> + + * @this: the devfreq instance of own device.
>> + + * @nb: the notifier block for DEVFREQ_TRANSITION_NOTIFIER list
>> + + * @cpu_data: the state min/max/current frequency of all online cpu's
>> *
>> * The devfreq_passive_data have to set the devfreq instance of parent
>> * device with governors except for the passive governor. But, don't need to
>> @@ -314,9 +322,13 @@ struct devfreq_passive_data {
>> /* Optional callback to decide the next frequency of passvice device */
>> int (*get_target_freq)(struct devfreq *this, unsigned long *freq);
>>
>> + /* Should set the type of parent device */
>> + enum devfreq_parent_dev_type parent_type;
>> +
>> /* For passive governor's internal use. Don't need to set them */
>> struct devfreq *this;
>> struct notifier_block nb;
>> + struct devfreq_cpu_data *cpu_data[NR_CPUS];
>> };
>> #endif
>>
>> --
>> 2.17.1
>>
>
>
--
Best Regards,
Chanwoo Choi
Samsung Electronics
^ permalink raw reply [flat|nested] 15+ messages in thread
* Re: [PATCH 3/4] PM / devfreq: Add cpu based scaling support to passive governor
2021-06-17 6:05 ` [PATCH 3/4] PM / devfreq: Add cpu based scaling support to passive governor Chanwoo Choi
2021-06-17 5:51 ` Hsin-Yi Wang
@ 2021-06-22 17:42 ` Matthias Kaehlcke
2021-06-22 18:36 ` Matthias Kaehlcke
2021-06-23 20:50 ` kernel test robot
3 siblings, 0 replies; 15+ messages in thread
From: Matthias Kaehlcke @ 2021-06-22 17:42 UTC (permalink / raw)
To: Chanwoo Choi
Cc: andrew-sh.cheng, hsinyi, sibis, saravanak, myungjoo.ham,
kyungmin.park, chanwoo, cwchoi00, linux-pm, linux-kernel,
Saravana Kannan
On Thu, Jun 17, 2021 at 03:05:45PM +0900, Chanwoo Choi wrote:
> From: Saravana Kannan <skannan@codeaurora.org>
>
> Many CPU architectures have caches that can scale independent of the
> CPUs. Frequency scaling of the caches is necessary to make sure that the
> cache is not a performance bottleneck that leads to poor performance and
> power. The same idea applies for RAM/DDR.
>
> To achieve this, this patch adds support for cpu based scaling to the
> passive governor. This is accomplished by taking the current frequency
> of each CPU frequency domain and then adjust the frequency of the cache
> (or any devfreq device) based on the frequency of the CPUs. It listens
> to CPU frequency transition notifiers to keep itself up to date on the
> current CPU frequency.
>
> To decide the frequency of the device, the governor does one of the
> following:
> * Derives the optimal devfreq device opp from required-opps property of
> the parent cpu opp_table.
>
> * Scales the device frequency in proportion to the CPU frequency. So, if
> the CPUs are running at their max frequency, the device runs at its
> max frequency. If the CPUs are running at their min frequency, the
> device runs at its min frequency. It is interpolated for frequencies
> in between.
>
> Signed-off-by: Saravana Kannan <skannan@codeaurora.org>
> [Sibi: Integrated cpu-freqmap governor into passive_governor]
> Signed-off-by: Sibi Sankar <sibis@codeaurora.org>
> [Chanwoo: Fix conflict with latest code and clean code up]
> Signed-off-by: Chanwoo Choi <cw00.choi@samsung.com>
> ---
> drivers/devfreq/governor.h | 22 +++
> drivers/devfreq/governor_passive.c | 264 ++++++++++++++++++++++++++++-
> include/linux/devfreq.h | 16 +-
> 3 files changed, 293 insertions(+), 9 deletions(-)
>
> diff --git a/drivers/devfreq/governor.h b/drivers/devfreq/governor.h
> index 9a9495f94ac6..3c36c92c89a9 100644
> --- a/drivers/devfreq/governor.h
> +++ b/drivers/devfreq/governor.h
> @@ -47,6 +47,28 @@
> #define DEVFREQ_GOV_ATTR_POLLING_INTERVAL BIT(0)
> #define DEVFREQ_GOV_ATTR_TIMER BIT(1)
>
> +/**
> + * struct devfreq_cpu_data - Hold the per-cpu data
> + * @dev: reference to cpu device.
> + * @first_cpu: the cpumask of the first cpu of a policy.
> + * @opp_table: reference to cpu opp table.
> + * @cur_freq: the current frequency of the cpu.
> + * @min_freq: the min frequency of the cpu.
> + * @max_freq: the max frequency of the cpu.
> + *
> + * This structure stores the required cpu_data of a cpu.
> + * This is auto-populated by the governor.
> + */
> +struct devfreq_cpu_data {
> + struct device *dev;
> + unsigned int first_cpu;
> +
> + struct opp_table *opp_table;
> + unsigned int cur_freq;
> + unsigned int min_freq;
> + unsigned int max_freq;
> +};
> +
> /**
> * struct devfreq_governor - Devfreq policy governor
> * @node: list node - contains registered devfreq governors
> diff --git a/drivers/devfreq/governor_passive.c b/drivers/devfreq/governor_passive.c
> index fc09324a03e0..07e864509b7e 100644
> --- a/drivers/devfreq/governor_passive.c
> +++ b/drivers/devfreq/governor_passive.c
>
> ...
>
> +static int cpufreq_passive_register_notifier(struct devfreq *devfreq)
> +{
> + struct devfreq_passive_data *p_data
> + = (struct devfreq_passive_data *)devfreq->data;
> + struct device *dev = devfreq->dev.parent;
> + struct opp_table *opp_table = NULL;
> + struct devfreq_cpu_data *cpu_data;
> + struct cpufreq_policy *policy;
> + struct device *cpu_dev;
> + unsigned int cpu;
> + int ret;
> +
> + get_online_cpus();
> +
> + p_data->nb.notifier_call = cpufreq_passive_notifier_call;
> + ret = cpufreq_register_notifier(&p_data->nb, CPUFREQ_TRANSITION_NOTIFIER);
> + if (ret) {
> + dev_err(dev, "failed to register cpufreq notifier\n");
> + p_data->nb.notifier_call = NULL;
> + goto out;
> + }
> +
> + for_each_online_cpu(cpu) {
Is this really needed for each CPU? Wouldn't it be enough to create
a 'cpu_data' for each 'policy CPU'?
In any case should this be for_each_possible_cpu() as in _unregister_notifier()
to also support CPUs that may be offline when the notifier is registered?
> + if (p_data->cpu_data[cpu])
> + continue;
> +
> + policy = cpufreq_cpu_get(cpu);
> + if (policy) {
> + cpu_data = kzalloc(sizeof(*cpu_data), GFP_KERNEL);
> + if (!cpu_data) {
> + ret = -ENOMEM;
> + goto out;
> + }
> +
> + cpu_dev = get_cpu_device(cpu);
> + if (!cpu_dev) {
> + dev_err(dev, "failed to get cpu device\n");
> + ret = -ENODEV;
> + goto out;
Memory for 'cpu_data' is not freed in this path.
Also applies to CPUs from possible prior iterations.
> + }
> +
> + opp_table = dev_pm_opp_get_opp_table(cpu_dev);
> + if (IS_ERR(opp_table)) {
> + ret = PTR_ERR(opp_table);
> + goto out;
Ditto and cpufreq_cpu_put() is missing too.
> + }
> +
> + cpu_data->dev = cpu_dev;
> + cpu_data->opp_table = opp_table;
> + cpu_data->first_cpu = cpumask_first(policy->related_cpus);
> + cpu_data->cur_freq = policy->cur;
> + cpu_data->min_freq = policy->cpuinfo.min_freq;
> + cpu_data->max_freq = policy->cpuinfo.max_freq;
> +
> + p_data->cpu_data[cpu] = cpu_data;
> + cpufreq_cpu_put(policy);
> + } else {
> + ret = -EPROBE_DEFER;
> + goto out;
Resources from possible prior iterations aren't freed.
> + }
> + }
> +out:
> + put_online_cpus();
> + if (ret)
> + return ret;
> +
> + mutex_lock(&devfreq->lock);
> + ret = devfreq_update_target(devfreq, 0L);
> + mutex_unlock(&devfreq->lock);
> + if (ret)
> + dev_err(dev, "failed to update the frequency\n");
> +
> + return ret;
> +}
> +
> +static int cpufreq_passive_unregister_notifier(struct devfreq *devfreq)
> +{
> + struct devfreq_passive_data *p_data
> + = (struct devfreq_passive_data *)devfreq->data;
> + struct devfreq_cpu_data *cpu_data;
> + int cpu;
> +
> + if (p_data->nb.notifier_call)
> + cpufreq_unregister_notifier(&p_data->nb, CPUFREQ_TRANSITION_NOTIFIER);
> +
> + for_each_possible_cpu(cpu) {
> + cpu_data = p_data->cpu_data[cpu];
> + if (cpu_data) {
> + if (cpu_data->opp_table)
> + dev_pm_opp_put_opp_table(cpu_data->opp_table);
> + kfree(cpu_data);
> + cpu_data = NULL;
Assignment to NULL is not needed.
> + }
> + }
> +
> + return 0;
> +}
> +
> static int devfreq_passive_notifier_call(struct notifier_block *nb,
> unsigned long event, void *ptr)
> {
> @@ -140,7 +379,7 @@ static int devfreq_passive_event_handler(struct devfreq *devfreq,
> struct notifier_block *nb = &p_data->nb;
> int ret = 0;
>
> - if (!parent)
> + if (p_data->parent_type == DEVFREQ_PARENT_DEV && !parent)
> return -EPROBE_DEFER;
>
> switch (event) {
> @@ -148,13 +387,24 @@ static int devfreq_passive_event_handler(struct devfreq *devfreq,
> if (!p_data->this)
> p_data->this = devfreq;
>
> - nb->notifier_call = devfreq_passive_notifier_call;
> - ret = devfreq_register_notifier(parent, nb,
> - DEVFREQ_TRANSITION_NOTIFIER);
> + if (p_data->parent_type == DEVFREQ_PARENT_DEV) {
> + nb->notifier_call = devfreq_passive_notifier_call;
> + ret = devfreq_register_notifier(parent, nb,
> + DEVFREQ_TRANSITION_NOTIFIER);
> + } else if (p_data->parent_type == CPUFREQ_PARENT_DEV) {
> + ret = cpufreq_passive_register_notifier(devfreq);
> + } else {
> + ret = -EINVAL;
> + }
> break;
> case DEVFREQ_GOV_STOP:
> - WARN_ON(devfreq_unregister_notifier(parent, nb,
> - DEVFREQ_TRANSITION_NOTIFIER));
> + if (p_data->parent_type == DEVFREQ_PARENT_DEV)
> + WARN_ON(devfreq_unregister_notifier(parent, nb,
> + DEVFREQ_TRANSITION_NOTIFIER));
> + else if (p_data->parent_type == CPUFREQ_PARENT_DEV)
> + WARN_ON(cpufreq_passive_unregister_notifier(devfreq));
> + else
> + ret = -EINVAL;
> break;
> default:
> break;
> diff --git a/include/linux/devfreq.h b/include/linux/devfreq.h
> index 142474b4af96..cfa0ef54841e 100644
> --- a/include/linux/devfreq.h
> +++ b/include/linux/devfreq.h
> @@ -38,6 +38,7 @@ enum devfreq_timer {
>
> struct devfreq;
> struct devfreq_governor;
> +struct devfreq_cpu_data;
> struct thermal_cooling_device;
>
> /**
> @@ -288,6 +289,11 @@ struct devfreq_simple_ondemand_data {
> #endif
>
> #if IS_ENABLED(CONFIG_DEVFREQ_GOV_PASSIVE)
> +enum devfreq_parent_dev_type {
> + DEVFREQ_PARENT_DEV,
> + CPUFREQ_PARENT_DEV,
> +};
> +
> /**
> * struct devfreq_passive_data - ``void *data`` fed to struct devfreq
> * and devfreq_add_device
> @@ -299,8 +305,10 @@ struct devfreq_simple_ondemand_data {
> * using governors except for passive governor.
> * If the devfreq device has the specific method to decide
> * the next frequency, should use this callback.
> - * @this: the devfreq instance of own device.
> - * @nb: the notifier block for DEVFREQ_TRANSITION_NOTIFIER list
> + + * @parent_type parent type of the device
> + + * @this: the devfreq instance of own device.
> + + * @nb: the notifier block for DEVFREQ_TRANSITION_NOTIFIER list
> + + * @cpu_data: the state min/max/current frequency of all online cpu's
> *
> * The devfreq_passive_data have to set the devfreq instance of parent
> * device with governors except for the passive governor. But, don't need to
> @@ -314,9 +322,13 @@ struct devfreq_passive_data {
> /* Optional callback to decide the next frequency of passvice device */
> int (*get_target_freq)(struct devfreq *this, unsigned long *freq);
>
> + /* Should set the type of parent device */
> + enum devfreq_parent_dev_type parent_type;
> +
> /* For passive governor's internal use. Don't need to set them */
> struct devfreq *this;
> struct notifier_block nb;
> + struct devfreq_cpu_data *cpu_data[NR_CPUS];
Could memory usage be a concern on systems with a really high number of CPUs
(e.g. 8k for x86 with MAXSMP)? One could argue that such systems likely have
significant amount of RAM too and a chunk of memory in the order of 100k
wouldn't make a big impact. I'm assuming that 'cpu_data' is only needed for
'policy CPUs'.
^ permalink raw reply [flat|nested] 15+ messages in thread* Re: [PATCH 3/4] PM / devfreq: Add cpu based scaling support to passive governor
2021-06-17 6:05 ` [PATCH 3/4] PM / devfreq: Add cpu based scaling support to passive governor Chanwoo Choi
2021-06-17 5:51 ` Hsin-Yi Wang
2021-06-22 17:42 ` Matthias Kaehlcke
@ 2021-06-22 18:36 ` Matthias Kaehlcke
2021-06-23 20:50 ` kernel test robot
3 siblings, 0 replies; 15+ messages in thread
From: Matthias Kaehlcke @ 2021-06-22 18:36 UTC (permalink / raw)
To: Chanwoo Choi
Cc: andrew-sh.cheng, hsinyi, sibis, saravanak, myungjoo.ham,
kyungmin.park, chanwoo, cwchoi00, linux-pm, linux-kernel,
Saravana Kannan
On Thu, Jun 17, 2021 at 03:05:45PM +0900, Chanwoo Choi wrote:
> From: Saravana Kannan <skannan@codeaurora.org>
>
> Many CPU architectures have caches that can scale independent of the
> CPUs. Frequency scaling of the caches is necessary to make sure that the
> cache is not a performance bottleneck that leads to poor performance and
> power. The same idea applies for RAM/DDR.
>
> To achieve this, this patch adds support for cpu based scaling to the
> passive governor. This is accomplished by taking the current frequency
> of each CPU frequency domain and then adjust the frequency of the cache
> (or any devfreq device) based on the frequency of the CPUs. It listens
> to CPU frequency transition notifiers to keep itself up to date on the
> current CPU frequency.
>
> To decide the frequency of the device, the governor does one of the
> following:
> * Derives the optimal devfreq device opp from required-opps property of
> the parent cpu opp_table.
>
> * Scales the device frequency in proportion to the CPU frequency. So, if
> the CPUs are running at their max frequency, the device runs at its
> max frequency. If the CPUs are running at their min frequency, the
> device runs at its min frequency. It is interpolated for frequencies
> in between.
>
> Signed-off-by: Saravana Kannan <skannan@codeaurora.org>
> [Sibi: Integrated cpu-freqmap governor into passive_governor]
> Signed-off-by: Sibi Sankar <sibis@codeaurora.org>
> [Chanwoo: Fix conflict with latest code and clean code up]
> Signed-off-by: Chanwoo Choi <cw00.choi@samsung.com>
> ---
> drivers/devfreq/governor.h | 22 +++
> drivers/devfreq/governor_passive.c | 264 ++++++++++++++++++++++++++++-
> include/linux/devfreq.h | 16 +-
> 3 files changed, 293 insertions(+), 9 deletions(-)
>
> diff --git a/drivers/devfreq/governor.h b/drivers/devfreq/governor.h
> index 9a9495f94ac6..3c36c92c89a9 100644
> --- a/drivers/devfreq/governor.h
> +++ b/drivers/devfreq/governor.h
> @@ -47,6 +47,28 @@
> #define DEVFREQ_GOV_ATTR_POLLING_INTERVAL BIT(0)
> #define DEVFREQ_GOV_ATTR_TIMER BIT(1)
>
> +/**
> + * struct devfreq_cpu_data - Hold the per-cpu data
> + * @dev: reference to cpu device.
> + * @first_cpu: the cpumask of the first cpu of a policy.
> + * @opp_table: reference to cpu opp table.
> + * @cur_freq: the current frequency of the cpu.
> + * @min_freq: the min frequency of the cpu.
> + * @max_freq: the max frequency of the cpu.
> + *
> + * This structure stores the required cpu_data of a cpu.
> + * This is auto-populated by the governor.
> + */
> +struct devfreq_cpu_data {
> + struct device *dev;
> + unsigned int first_cpu;
> +
> + struct opp_table *opp_table;
> + unsigned int cur_freq;
> + unsigned int min_freq;
> + unsigned int max_freq;
> +};
> +
> /**
> * struct devfreq_governor - Devfreq policy governor
> * @node: list node - contains registered devfreq governors
> diff --git a/drivers/devfreq/governor_passive.c b/drivers/devfreq/governor_passive.c
> index fc09324a03e0..07e864509b7e 100644
> --- a/drivers/devfreq/governor_passive.c
> +++ b/drivers/devfreq/governor_passive.c
> @@ -8,11 +8,84 @@
> */
>
> #include <linux/module.h>
> +#include <linux/cpu.h>
> +#include <linux/cpufreq.h>
> +#include <linux/cpumask.h>
> +#include <linux/slab.h>
> #include <linux/device.h>
> #include <linux/devfreq.h>
> #include "governor.h"
>
> -static int devfreq_passive_get_target_freq(struct devfreq *devfreq,
> +#define HZ_PER_KHZ 1000
> +
> +static unsigned long get_taget_freq_by_required_opp(struct device *p_dev,
> + struct opp_table *p_opp_table,
> + struct opp_table *opp_table,
> + unsigned long freq)
> +{
s/get_taget_freq_by_required_opp/get_target_freq_by_required_opp/
^ permalink raw reply [flat|nested] 15+ messages in thread* Re: [PATCH 3/4] PM / devfreq: Add cpu based scaling support to passive governor
2021-06-17 6:05 ` [PATCH 3/4] PM / devfreq: Add cpu based scaling support to passive governor Chanwoo Choi
` (2 preceding siblings ...)
2021-06-22 18:36 ` Matthias Kaehlcke
@ 2021-06-23 20:50 ` kernel test robot
3 siblings, 0 replies; 15+ messages in thread
From: kernel test robot @ 2021-06-23 20:50 UTC (permalink / raw)
To: kbuild-all
[-- Attachment #1: Type: text/plain, Size: 15773 bytes --]
Hi Chanwoo,
I love your patch! Yet something to improve:
[auto build test ERROR on linus/master]
[also build test ERROR on v5.13-rc7 next-20210623]
[If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch]
url: https://github.com/0day-ci/linux/commits/Chanwoo-Choi/PM-devfreq-passive-Fix-get_target_freq-when-not-using-required-opp/20210617-150556
base: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git 70585216fe7730d9fb5453d3e2804e149d0fe201
config: um-allmodconfig (attached as .config)
compiler: gcc-9 (Debian 9.3.0-22) 9.3.0
reproduce (this is a W=1 build):
# https://github.com/0day-ci/linux/commit/53908c80cda313fc07cffa4f47611bda4fe7637c
git remote add linux-review https://github.com/0day-ci/linux
git fetch --no-tags linux-review Chanwoo-Choi/PM-devfreq-passive-Fix-get_target_freq-when-not-using-required-opp/20210617-150556
git checkout 53908c80cda313fc07cffa4f47611bda4fe7637c
# save the attached .config to linux build tree
make W=1 ARCH=um
If you fix the issue, kindly add following tag as appropriate
Reported-by: kernel test robot <lkp@intel.com>
All error/warnings (new ones prefixed by >>):
In file included from arch/x86/um/asm/processor.h:41,
from include/linux/spinlock_up.h:8,
from include/linux/spinlock.h:92,
from include/linux/mmzone.h:8,
from include/linux/gfp.h:6,
from include/linux/umh.h:4,
from include/linux/kmod.h:9,
from include/linux/module.h:16,
from drivers/devfreq/governor_simpleondemand.c:10:
>> arch/um/include/asm/processor-generic.h:97:19: error: expected identifier or '(' before '&' token
97 | #define cpu_data (&boot_cpu_data)
| ^
include/linux/devfreq.h:331:27: note: in expansion of macro 'cpu_data'
331 | struct devfreq_cpu_data *cpu_data[NR_CPUS];
| ^~~~~~~~
In file included from drivers/devfreq/governor_simpleondemand.c:11:
>> include/linux/devfreq.h:332:1: warning: no semicolon at end of struct or union
332 | };
| ^
--
In file included from arch/x86/um/asm/processor.h:41,
from include/linux/spinlock_up.h:8,
from include/linux/spinlock.h:92,
from include/linux/mmzone.h:8,
from include/linux/gfp.h:6,
from include/linux/umh.h:4,
from include/linux/kmod.h:9,
from include/linux/module.h:16,
from drivers/devfreq/governor_passive.c:10:
>> arch/um/include/asm/processor-generic.h:97:19: error: expected identifier or '(' before '&' token
97 | #define cpu_data (&boot_cpu_data)
| ^
include/linux/devfreq.h:331:27: note: in expansion of macro 'cpu_data'
331 | struct devfreq_cpu_data *cpu_data[NR_CPUS];
| ^~~~~~~~
In file included from drivers/devfreq/governor_passive.c:16:
>> include/linux/devfreq.h:332:1: warning: no semicolon at end of struct or union
332 | };
| ^
In file included from arch/x86/um/asm/processor.h:41,
from include/linux/spinlock_up.h:8,
from include/linux/spinlock.h:92,
from include/linux/mmzone.h:8,
from include/linux/gfp.h:6,
from include/linux/umh.h:4,
from include/linux/kmod.h:9,
from include/linux/module.h:16,
from drivers/devfreq/governor_passive.c:10:
drivers/devfreq/governor_passive.c: In function 'get_target_freq_with_cpufreq':
>> arch/um/include/asm/processor-generic.h:97:19: error: expected identifier or '(' before '&' token
97 | #define cpu_data (&boot_cpu_data)
| ^
drivers/devfreq/governor_passive.c:52:27: note: in expansion of macro 'cpu_data'
52 | struct devfreq_cpu_data *cpu_data;
| ^~~~~~~~
>> arch/um/include/asm/processor-generic.h:97:18: error: expected identifier before '(' token
97 | #define cpu_data (&boot_cpu_data)
| ^
drivers/devfreq/governor_passive.c:58:22: note: in expansion of macro 'cpu_data'
58 | cpu_data = p_data->cpu_data[cpu];
| ^~~~~~~~
>> drivers/devfreq/governor_passive.c:59:7: warning: the address of 'boot_cpu_data' will always evaluate as 'true' [-Waddress]
59 | if (!cpu_data || cpu_data->first_cpu != cpu)
| ^
>> drivers/devfreq/governor_passive.c:59:28: error: 'struct cpuinfo_um' has no member named 'first_cpu'
59 | if (!cpu_data || cpu_data->first_cpu != cpu)
| ^~
>> drivers/devfreq/governor_passive.c:63:21: error: 'struct cpuinfo_um' has no member named 'cur_freq'
63 | cpu_cur = cpu_data->cur_freq * HZ_PER_KHZ;
| ^~
>> drivers/devfreq/governor_passive.c:64:49: error: 'struct cpuinfo_um' has no member named 'dev'
64 | freq = get_taget_freq_by_required_opp(cpu_data->dev,
| ^~
>> drivers/devfreq/governor_passive.c:65:14: error: 'struct cpuinfo_um' has no member named 'opp_table'
65 | cpu_data->opp_table,
| ^~
>> drivers/devfreq/governor_passive.c:75:21: error: 'struct cpuinfo_um' has no member named 'min_freq'
75 | cpu_min = cpu_data->min_freq;
| ^~
>> drivers/devfreq/governor_passive.c:76:21: error: 'struct cpuinfo_um' has no member named 'max_freq'
76 | cpu_max = cpu_data->max_freq;
| ^~
drivers/devfreq/governor_passive.c:77:21: error: 'struct cpuinfo_um' has no member named 'cur_freq'
77 | cpu_cur = cpu_data->cur_freq;
| ^~
>> drivers/devfreq/governor_passive.c:50:31: warning: variable 'p_data' set but not used [-Wunused-but-set-variable]
50 | struct devfreq_passive_data *p_data =
| ^~~~~~
In file included from arch/x86/um/asm/processor.h:41,
from include/linux/spinlock_up.h:8,
from include/linux/spinlock.h:92,
from include/linux/mmzone.h:8,
from include/linux/gfp.h:6,
from include/linux/umh.h:4,
from include/linux/kmod.h:9,
from include/linux/module.h:16,
from drivers/devfreq/governor_passive.c:10:
drivers/devfreq/governor_passive.c: In function 'cpufreq_passive_notifier_call':
>> arch/um/include/asm/processor-generic.h:97:19: error: expected identifier or '(' before '&' token
97 | #define cpu_data (&boot_cpu_data)
| ^
drivers/devfreq/governor_passive.c:215:27: note: in expansion of macro 'cpu_data'
215 | struct devfreq_cpu_data *cpu_data;
| ^~~~~~~~
>> arch/um/include/asm/processor-generic.h:97:18: error: expected identifier before '(' token
97 | #define cpu_data (&boot_cpu_data)
| ^
drivers/devfreq/governor_passive.c:221:10: note: in expansion of macro 'cpu_data'
221 | !data->cpu_data[freqs->policy->cpu])
| ^~~~~~~~
>> arch/um/include/asm/processor-generic.h:97:18: error: expected identifier before '(' token
97 | #define cpu_data (&boot_cpu_data)
| ^
drivers/devfreq/governor_passive.c:224:19: note: in expansion of macro 'cpu_data'
224 | cpu_data = data->cpu_data[freqs->policy->cpu];
| ^~~~~~~~
drivers/devfreq/governor_passive.c:225:14: error: 'struct cpuinfo_um' has no member named 'cur_freq'
225 | if (cpu_data->cur_freq == freqs->new)
| ^~
drivers/devfreq/governor_passive.c:228:21: error: 'struct cpuinfo_um' has no member named 'cur_freq'
228 | cur_freq = cpu_data->cur_freq;
| ^~
drivers/devfreq/governor_passive.c:229:10: error: 'struct cpuinfo_um' has no member named 'cur_freq'
229 | cpu_data->cur_freq = freqs->new;
| ^~
drivers/devfreq/governor_passive.c:235:11: error: 'struct cpuinfo_um' has no member named 'cur_freq'
235 | cpu_data->cur_freq = cur_freq;
| ^~
In file included from arch/x86/um/asm/processor.h:41,
from include/linux/spinlock_up.h:8,
from include/linux/spinlock.h:92,
from include/linux/mmzone.h:8,
from include/linux/gfp.h:6,
from include/linux/umh.h:4,
from include/linux/kmod.h:9,
from include/linux/module.h:16,
from drivers/devfreq/governor_passive.c:10:
drivers/devfreq/governor_passive.c: In function 'cpufreq_passive_register_notifier':
>> arch/um/include/asm/processor-generic.h:97:19: error: expected identifier or '(' before '&' token
97 | #define cpu_data (&boot_cpu_data)
| ^
drivers/devfreq/governor_passive.c:249:27: note: in expansion of macro 'cpu_data'
249 | struct devfreq_cpu_data *cpu_data;
| ^~~~~~~~
>> arch/um/include/asm/processor-generic.h:97:18: error: expected identifier before '(' token
97 | #define cpu_data (&boot_cpu_data)
| ^
drivers/devfreq/governor_passive.c:266:15: note: in expansion of macro 'cpu_data'
266 | if (p_data->cpu_data[cpu])
| ^~~~~~~~
>> drivers/devfreq/governor_passive.c:271:13: error: lvalue required as left operand of assignment
271 | cpu_data = kzalloc(sizeof(*cpu_data), GFP_KERNEL);
| ^
drivers/devfreq/governor_passive.c:272:8: warning: the address of 'boot_cpu_data' will always evaluate as 'true' [-Waddress]
272 | if (!cpu_data) {
| ^
drivers/devfreq/governor_passive.c:290:12: error: 'struct cpuinfo_um' has no member named 'dev'
290 | cpu_data->dev = cpu_dev;
| ^~
drivers/devfreq/governor_passive.c:291:12: error: 'struct cpuinfo_um' has no member named 'opp_table'
291 | cpu_data->opp_table = opp_table;
| ^~
drivers/devfreq/governor_passive.c:292:12: error: 'struct cpuinfo_um' has no member named 'first_cpu'
292 | cpu_data->first_cpu = cpumask_first(policy->related_cpus);
| ^~
drivers/devfreq/governor_passive.c:293:12: error: 'struct cpuinfo_um' has no member named 'cur_freq'
293 | cpu_data->cur_freq = policy->cur;
| ^~
drivers/devfreq/governor_passive.c:294:12: error: 'struct cpuinfo_um' has no member named 'min_freq'
294 | cpu_data->min_freq = policy->cpuinfo.min_freq;
| ^~
drivers/devfreq/governor_passive.c:295:12: error: 'struct cpuinfo_um' has no member named 'max_freq'
295 | cpu_data->max_freq = policy->cpuinfo.max_freq;
| ^~
In file included from arch/x86/um/asm/processor.h:41,
from include/linux/spinlock_up.h:8,
from include/linux/spinlock.h:92,
from include/linux/mmzone.h:8,
from include/linux/gfp.h:6,
from include/linux/umh.h:4,
from include/linux/kmod.h:9,
from include/linux/module.h:16,
from drivers/devfreq/governor_passive.c:10:
>> arch/um/include/asm/processor-generic.h:97:18: error: expected identifier before '(' token
97 | #define cpu_data (&boot_cpu_data)
| ^
drivers/devfreq/governor_passive.c:297:12: note: in expansion of macro 'cpu_data'
297 | p_data->cpu_data[cpu] = cpu_data;
| ^~~~~~~~
drivers/devfreq/governor_passive.c: In function 'cpufreq_passive_unregister_notifier':
>> arch/um/include/asm/processor-generic.h:97:19: error: expected identifier or '(' before '&' token
97 | #define cpu_data (&boot_cpu_data)
| ^
drivers/devfreq/governor_passive.c:322:27: note: in expansion of macro 'cpu_data'
322 | struct devfreq_cpu_data *cpu_data;
| ^~~~~~~~
arch/um/include/asm/processor-generic.h:97:18: error: expected identifier before '(' token
97 | #define cpu_data (&boot_cpu_data)
| ^
drivers/devfreq/governor_passive.c:329:22: note: in expansion of macro 'cpu_data'
329 | cpu_data = p_data->cpu_data[cpu];
| ^~~~~~~~
arch/um/include/asm/processor-generic.h:97:18: warning: the address of 'boot_cpu_data' will always evaluate as 'true' [-Waddress]
97 | #define cpu_data (&boot_cpu_data)
| ^
drivers/devfreq/governor_passive.c:330:7: note: in expansion of macro 'cpu_data'
330 | if (cpu_data) {
| ^~~~~~~~
drivers/devfreq/governor_passive.c:331:16: error: 'struct cpuinfo_um' has no member named 'opp_table'
331 | if (cpu_data->opp_table)
| ^~
drivers/devfreq/governor_passive.c:332:38: error: 'struct cpuinfo_um' has no member named 'opp_table'
332 | dev_pm_opp_put_opp_table(cpu_data->opp_table);
| ^~
drivers/devfreq/governor_passive.c:334:13: error: lvalue required as left operand of assignment
334 | cpu_data = NULL;
| ^
vim +59 drivers/devfreq/governor_passive.c
46
47 static int get_target_freq_with_cpufreq(struct devfreq *devfreq,
48 unsigned long *target_freq)
49 {
> 50 struct devfreq_passive_data *p_data =
51 (struct devfreq_passive_data *)devfreq->data;
52 struct devfreq_cpu_data *cpu_data;
53 unsigned long cpu, cpu_cur, cpu_min, cpu_max, cpu_percent;
54 unsigned long dev_min, dev_max;
55 unsigned long freq = 0;
56
57 for_each_online_cpu(cpu) {
58 cpu_data = p_data->cpu_data[cpu];
> 59 if (!cpu_data || cpu_data->first_cpu != cpu)
60 continue;
61
62 /* Get target freq via required opps */
> 63 cpu_cur = cpu_data->cur_freq * HZ_PER_KHZ;
> 64 freq = get_taget_freq_by_required_opp(cpu_data->dev,
> 65 cpu_data->opp_table,
66 devfreq->opp_table, cpu_cur);
67 if (freq) {
68 *target_freq = max(freq, *target_freq);
69 continue;
70 }
71
72 /* Use Interpolation if required opps is not available */
73 devfreq_get_freq_range(devfreq, &dev_min, &dev_max);
74
> 75 cpu_min = cpu_data->min_freq;
> 76 cpu_max = cpu_data->max_freq;
77 cpu_cur = cpu_data->cur_freq;
78
79 cpu_percent = ((cpu_cur - cpu_min) * 100) / (cpu_max - cpu_min);
80 freq = dev_min + mult_frac(dev_max - dev_min, cpu_percent, 100);
81
82 *target_freq = max(freq, *target_freq);
83 }
84
85 return 0;
86 }
87
---
0-DAY CI Kernel Test Service, Intel Corporation
https://lists.01.org/hyperkitty/list/kbuild-all(a)lists.01.org
[-- Attachment #2: config.gz --]
[-- Type: application/gzip, Size: 24520 bytes --]
^ permalink raw reply [flat|nested] 15+ messages in thread