2023-09-12 23:30:17

by Liao, Chang

[permalink] [raw]
Subject: [PATCH 1/2] cpufreq: userspace: Use fine-grained mutex in userspace governor

The userspace governor currently uses a big global mutex to avoid the
race condition on the governor_data field of cpufreq_policy structure.
This leads to a low concurrency if multiple userspace applications are
trying to set the speed of different policies at the same time. This
patch introduces a per-policy mutex to allow the updating of different
policies to be performed concurrently, improving overall concurrency.

Signed-off-by: Liao Chang <[email protected]>
---
drivers/cpufreq/cpufreq_userspace.c | 69 +++++++++++++++++------------
1 file changed, 40 insertions(+), 29 deletions(-)

diff --git a/drivers/cpufreq/cpufreq_userspace.c b/drivers/cpufreq/cpufreq_userspace.c
index 50a4d7846580..442e31060d62 100644
--- a/drivers/cpufreq/cpufreq_userspace.c
+++ b/drivers/cpufreq/cpufreq_userspace.c
@@ -16,7 +16,11 @@
#include <linux/slab.h>

static DEFINE_PER_CPU(unsigned int, cpu_is_managed);
-static DEFINE_MUTEX(userspace_mutex);
+
+struct userspace_policy {
+ unsigned int setspeed;
+ struct mutex mutex;
+};

/**
* cpufreq_set - set the CPU frequency
@@ -28,19 +32,19 @@ static DEFINE_MUTEX(userspace_mutex);
static int cpufreq_set(struct cpufreq_policy *policy, unsigned int freq)
{
int ret = -EINVAL;
- unsigned int *setspeed = policy->governor_data;
+ struct userspace_policy *userspace = policy->governor_data;

pr_debug("cpufreq_set for cpu %u, freq %u kHz\n", policy->cpu, freq);

- mutex_lock(&userspace_mutex);
+ mutex_lock(&userspace->mutex);
if (!per_cpu(cpu_is_managed, policy->cpu))
goto err;

- *setspeed = freq;
+ userspace->setspeed = freq;

ret = __cpufreq_driver_target(policy, freq, CPUFREQ_RELATION_L);
err:
- mutex_unlock(&userspace_mutex);
+ mutex_unlock(&userspace->mutex);
return ret;
}

@@ -51,67 +55,74 @@ static ssize_t show_speed(struct cpufreq_policy *policy, char *buf)

static int cpufreq_userspace_policy_init(struct cpufreq_policy *policy)
{
- unsigned int *setspeed;
+ struct userspace_policy *userspace;

- setspeed = kzalloc(sizeof(*setspeed), GFP_KERNEL);
- if (!setspeed)
+ userspace = kzalloc(sizeof(*userspace), GFP_KERNEL);
+ if (!userspace)
return -ENOMEM;

- policy->governor_data = setspeed;
+ mutex_init(&userspace->mutex);
+
+ policy->governor_data = userspace;
return 0;
}

+/*
+ * Any routine that writes to the policy struct will hold the "rwsem" of
+ * policy struct that means it is free to free "governor_data" here.
+ */
static void cpufreq_userspace_policy_exit(struct cpufreq_policy *policy)
{
- mutex_lock(&userspace_mutex);
kfree(policy->governor_data);
policy->governor_data = NULL;
- mutex_unlock(&userspace_mutex);
}

static int cpufreq_userspace_policy_start(struct cpufreq_policy *policy)
{
- unsigned int *setspeed = policy->governor_data;
+ struct userspace_policy *userspace = policy->governor_data;

BUG_ON(!policy->cur);
pr_debug("started managing cpu %u\n", policy->cpu);

- mutex_lock(&userspace_mutex);
+ mutex_lock(&userspace->mutex);
per_cpu(cpu_is_managed, policy->cpu) = 1;
- *setspeed = policy->cur;
- mutex_unlock(&userspace_mutex);
+ userspace->setspeed = policy->cur;
+ mutex_unlock(&userspace->mutex);
return 0;
}

static void cpufreq_userspace_policy_stop(struct cpufreq_policy *policy)
{
- unsigned int *setspeed = policy->governor_data;
+ struct userspace_policy *userspace = policy->governor_data;

pr_debug("managing cpu %u stopped\n", policy->cpu);

- mutex_lock(&userspace_mutex);
+ mutex_lock(&userspace->mutex);
per_cpu(cpu_is_managed, policy->cpu) = 0;
- *setspeed = 0;
- mutex_unlock(&userspace_mutex);
+ userspace->setspeed = 0;
+ mutex_unlock(&userspace->mutex);
}

static void cpufreq_userspace_policy_limits(struct cpufreq_policy *policy)
{
- unsigned int *setspeed = policy->governor_data;
+ struct userspace_policy *userspace = policy->governor_data;

- mutex_lock(&userspace_mutex);
+ mutex_lock(&userspace->mutex);

pr_debug("limit event for cpu %u: %u - %u kHz, currently %u kHz, last set to %u kHz\n",
- policy->cpu, policy->min, policy->max, policy->cur, *setspeed);
-
- if (policy->max < *setspeed)
- __cpufreq_driver_target(policy, policy->max, CPUFREQ_RELATION_H);
- else if (policy->min > *setspeed)
- __cpufreq_driver_target(policy, policy->min, CPUFREQ_RELATION_L);
+ policy->cpu, policy->min, policy->max, policy->cur, userspace->setspeed);
+
+ if (policy->max < userspace->setspeed)
+ __cpufreq_driver_target(policy, policy->max,
+ CPUFREQ_RELATION_H);
+ else if (policy->min > userspace->setspeed)
+ __cpufreq_driver_target(policy, policy->min,
+ CPUFREQ_RELATION_L);
else
- __cpufreq_driver_target(policy, *setspeed, CPUFREQ_RELATION_L);
+ __cpufreq_driver_target(policy, userspace->setspeed,
+ CPUFREQ_RELATION_L);

- mutex_unlock(&userspace_mutex);
+ mutex_unlock(&userspace->mutex);
}

static struct cpufreq_governor cpufreq_gov_userspace = {
--
2.34.1


2023-10-05 13:56:16

by Viresh Kumar

[permalink] [raw]
Subject: Re: [PATCH 1/2] cpufreq: userspace: Use fine-grained mutex in userspace governor

On 12-09-23, 06:10, Liao Chang wrote:
> The userspace governor currently uses a big global mutex to avoid the
> race condition on the governor_data field of cpufreq_policy structure.
> This leads to a low concurrency if multiple userspace applications are
> trying to set the speed of different policies at the same time. This
> patch introduces a per-policy mutex to allow the updating of different
> policies to be performed concurrently, improving overall concurrency.
>
> Signed-off-by: Liao Chang <[email protected]>
> ---
> drivers/cpufreq/cpufreq_userspace.c | 69 +++++++++++++++++------------
> 1 file changed, 40 insertions(+), 29 deletions(-)
>
> diff --git a/drivers/cpufreq/cpufreq_userspace.c b/drivers/cpufreq/cpufreq_userspace.c
> index 50a4d7846580..442e31060d62 100644
> --- a/drivers/cpufreq/cpufreq_userspace.c
> +++ b/drivers/cpufreq/cpufreq_userspace.c
> @@ -16,7 +16,11 @@
> #include <linux/slab.h>
>
> static DEFINE_PER_CPU(unsigned int, cpu_is_managed);
> -static DEFINE_MUTEX(userspace_mutex);
> +
> +struct userspace_policy {
> + unsigned int setspeed;
> + struct mutex mutex;
> +};
>
> /**
> * cpufreq_set - set the CPU frequency
> @@ -28,19 +32,19 @@ static DEFINE_MUTEX(userspace_mutex);
> static int cpufreq_set(struct cpufreq_policy *policy, unsigned int freq)
> {
> int ret = -EINVAL;
> - unsigned int *setspeed = policy->governor_data;
> + struct userspace_policy *userspace = policy->governor_data;
>
> pr_debug("cpufreq_set for cpu %u, freq %u kHz\n", policy->cpu, freq);
>
> - mutex_lock(&userspace_mutex);
> + mutex_lock(&userspace->mutex);
> if (!per_cpu(cpu_is_managed, policy->cpu))
> goto err;
>
> - *setspeed = freq;
> + userspace->setspeed = freq;
>
> ret = __cpufreq_driver_target(policy, freq, CPUFREQ_RELATION_L);
> err:
> - mutex_unlock(&userspace_mutex);
> + mutex_unlock(&userspace->mutex);
> return ret;
> }
>
> @@ -51,67 +55,74 @@ static ssize_t show_speed(struct cpufreq_policy *policy, char *buf)
>
> static int cpufreq_userspace_policy_init(struct cpufreq_policy *policy)
> {
> - unsigned int *setspeed;
> + struct userspace_policy *userspace;
>
> - setspeed = kzalloc(sizeof(*setspeed), GFP_KERNEL);
> - if (!setspeed)
> + userspace = kzalloc(sizeof(*userspace), GFP_KERNEL);
> + if (!userspace)
> return -ENOMEM;
>
> - policy->governor_data = setspeed;
> + mutex_init(&userspace->mutex);
> +
> + policy->governor_data = userspace;
> return 0;
> }
>
> +/*
> + * Any routine that writes to the policy struct will hold the "rwsem" of
> + * policy struct that means it is free to free "governor_data" here.
> + */
> static void cpufreq_userspace_policy_exit(struct cpufreq_policy *policy)
> {
> - mutex_lock(&userspace_mutex);
> kfree(policy->governor_data);
> policy->governor_data = NULL;
> - mutex_unlock(&userspace_mutex);
> }
>
> static int cpufreq_userspace_policy_start(struct cpufreq_policy *policy)
> {
> - unsigned int *setspeed = policy->governor_data;
> + struct userspace_policy *userspace = policy->governor_data;
>
> BUG_ON(!policy->cur);
> pr_debug("started managing cpu %u\n", policy->cpu);
>
> - mutex_lock(&userspace_mutex);
> + mutex_lock(&userspace->mutex);
> per_cpu(cpu_is_managed, policy->cpu) = 1;
> - *setspeed = policy->cur;
> - mutex_unlock(&userspace_mutex);
> + userspace->setspeed = policy->cur;
> + mutex_unlock(&userspace->mutex);
> return 0;
> }
>
> static void cpufreq_userspace_policy_stop(struct cpufreq_policy *policy)
> {
> - unsigned int *setspeed = policy->governor_data;
> + struct userspace_policy *userspace = policy->governor_data;
>
> pr_debug("managing cpu %u stopped\n", policy->cpu);
>
> - mutex_lock(&userspace_mutex);
> + mutex_lock(&userspace->mutex);
> per_cpu(cpu_is_managed, policy->cpu) = 0;
> - *setspeed = 0;
> - mutex_unlock(&userspace_mutex);
> + userspace->setspeed = 0;
> + mutex_unlock(&userspace->mutex);
> }
>
> static void cpufreq_userspace_policy_limits(struct cpufreq_policy *policy)
> {
> - unsigned int *setspeed = policy->governor_data;
> + struct userspace_policy *userspace = policy->governor_data;
>
> - mutex_lock(&userspace_mutex);
> + mutex_lock(&userspace->mutex);
>
> pr_debug("limit event for cpu %u: %u - %u kHz, currently %u kHz, last set to %u kHz\n",
> - policy->cpu, policy->min, policy->max, policy->cur, *setspeed);
> -
> - if (policy->max < *setspeed)
> - __cpufreq_driver_target(policy, policy->max, CPUFREQ_RELATION_H);
> - else if (policy->min > *setspeed)
> - __cpufreq_driver_target(policy, policy->min, CPUFREQ_RELATION_L);
> + policy->cpu, policy->min, policy->max, policy->cur, userspace->setspeed);
> +
> + if (policy->max < userspace->setspeed)
> + __cpufreq_driver_target(policy, policy->max,
> + CPUFREQ_RELATION_H);
> + else if (policy->min > userspace->setspeed)
> + __cpufreq_driver_target(policy, policy->min,
> + CPUFREQ_RELATION_L);
> else
> - __cpufreq_driver_target(policy, *setspeed, CPUFREQ_RELATION_L);
> + __cpufreq_driver_target(policy, userspace->setspeed,
> + CPUFREQ_RELATION_L);
>
> - mutex_unlock(&userspace_mutex);
> + mutex_unlock(&userspace->mutex);
> }
>
> static struct cpufreq_governor cpufreq_gov_userspace = {

Acked-by: Viresh Kumar <[email protected]>

--
viresh

2023-10-06 20:03:28

by Rafael J. Wysocki

[permalink] [raw]
Subject: Re: [PATCH 1/2] cpufreq: userspace: Use fine-grained mutex in userspace governor

On Thu, Oct 5, 2023 at 1:05 PM Viresh Kumar <[email protected]> wrote:
>
> On 12-09-23, 06:10, Liao Chang wrote:
> > The userspace governor currently uses a big global mutex to avoid the
> > race condition on the governor_data field of cpufreq_policy structure.
> > This leads to a low concurrency if multiple userspace applications are
> > trying to set the speed of different policies at the same time. This
> > patch introduces a per-policy mutex to allow the updating of different
> > policies to be performed concurrently, improving overall concurrency.
> >
> > Signed-off-by: Liao Chang <[email protected]>
> > ---
> > drivers/cpufreq/cpufreq_userspace.c | 69 +++++++++++++++++------------
> > 1 file changed, 40 insertions(+), 29 deletions(-)
> >
> > diff --git a/drivers/cpufreq/cpufreq_userspace.c b/drivers/cpufreq/cpufreq_userspace.c
> > index 50a4d7846580..442e31060d62 100644
> > --- a/drivers/cpufreq/cpufreq_userspace.c
> > +++ b/drivers/cpufreq/cpufreq_userspace.c
> > @@ -16,7 +16,11 @@
> > #include <linux/slab.h>
> >
> > static DEFINE_PER_CPU(unsigned int, cpu_is_managed);
> > -static DEFINE_MUTEX(userspace_mutex);
> > +
> > +struct userspace_policy {
> > + unsigned int setspeed;
> > + struct mutex mutex;
> > +};
> >
> > /**
> > * cpufreq_set - set the CPU frequency
> > @@ -28,19 +32,19 @@ static DEFINE_MUTEX(userspace_mutex);
> > static int cpufreq_set(struct cpufreq_policy *policy, unsigned int freq)
> > {
> > int ret = -EINVAL;
> > - unsigned int *setspeed = policy->governor_data;
> > + struct userspace_policy *userspace = policy->governor_data;
> >
> > pr_debug("cpufreq_set for cpu %u, freq %u kHz\n", policy->cpu, freq);
> >
> > - mutex_lock(&userspace_mutex);
> > + mutex_lock(&userspace->mutex);
> > if (!per_cpu(cpu_is_managed, policy->cpu))
> > goto err;
> >
> > - *setspeed = freq;
> > + userspace->setspeed = freq;
> >
> > ret = __cpufreq_driver_target(policy, freq, CPUFREQ_RELATION_L);
> > err:
> > - mutex_unlock(&userspace_mutex);
> > + mutex_unlock(&userspace->mutex);
> > return ret;
> > }
> >
> > @@ -51,67 +55,74 @@ static ssize_t show_speed(struct cpufreq_policy *policy, char *buf)
> >
> > static int cpufreq_userspace_policy_init(struct cpufreq_policy *policy)
> > {
> > - unsigned int *setspeed;
> > + struct userspace_policy *userspace;
> >
> > - setspeed = kzalloc(sizeof(*setspeed), GFP_KERNEL);
> > - if (!setspeed)
> > + userspace = kzalloc(sizeof(*userspace), GFP_KERNEL);
> > + if (!userspace)
> > return -ENOMEM;
> >
> > - policy->governor_data = setspeed;
> > + mutex_init(&userspace->mutex);
> > +
> > + policy->governor_data = userspace;
> > return 0;
> > }
> >
> > +/*
> > + * Any routine that writes to the policy struct will hold the "rwsem" of
> > + * policy struct that means it is free to free "governor_data" here.
> > + */
> > static void cpufreq_userspace_policy_exit(struct cpufreq_policy *policy)
> > {
> > - mutex_lock(&userspace_mutex);
> > kfree(policy->governor_data);
> > policy->governor_data = NULL;
> > - mutex_unlock(&userspace_mutex);
> > }
> >
> > static int cpufreq_userspace_policy_start(struct cpufreq_policy *policy)
> > {
> > - unsigned int *setspeed = policy->governor_data;
> > + struct userspace_policy *userspace = policy->governor_data;
> >
> > BUG_ON(!policy->cur);
> > pr_debug("started managing cpu %u\n", policy->cpu);
> >
> > - mutex_lock(&userspace_mutex);
> > + mutex_lock(&userspace->mutex);
> > per_cpu(cpu_is_managed, policy->cpu) = 1;
> > - *setspeed = policy->cur;
> > - mutex_unlock(&userspace_mutex);
> > + userspace->setspeed = policy->cur;
> > + mutex_unlock(&userspace->mutex);
> > return 0;
> > }
> >
> > static void cpufreq_userspace_policy_stop(struct cpufreq_policy *policy)
> > {
> > - unsigned int *setspeed = policy->governor_data;
> > + struct userspace_policy *userspace = policy->governor_data;
> >
> > pr_debug("managing cpu %u stopped\n", policy->cpu);
> >
> > - mutex_lock(&userspace_mutex);
> > + mutex_lock(&userspace->mutex);
> > per_cpu(cpu_is_managed, policy->cpu) = 0;
> > - *setspeed = 0;
> > - mutex_unlock(&userspace_mutex);
> > + userspace->setspeed = 0;
> > + mutex_unlock(&userspace->mutex);
> > }
> >
> > static void cpufreq_userspace_policy_limits(struct cpufreq_policy *policy)
> > {
> > - unsigned int *setspeed = policy->governor_data;
> > + struct userspace_policy *userspace = policy->governor_data;
> >
> > - mutex_lock(&userspace_mutex);
> > + mutex_lock(&userspace->mutex);
> >
> > pr_debug("limit event for cpu %u: %u - %u kHz, currently %u kHz, last set to %u kHz\n",
> > - policy->cpu, policy->min, policy->max, policy->cur, *setspeed);
> > -
> > - if (policy->max < *setspeed)
> > - __cpufreq_driver_target(policy, policy->max, CPUFREQ_RELATION_H);
> > - else if (policy->min > *setspeed)
> > - __cpufreq_driver_target(policy, policy->min, CPUFREQ_RELATION_L);
> > + policy->cpu, policy->min, policy->max, policy->cur, userspace->setspeed);
> > +
> > + if (policy->max < userspace->setspeed)
> > + __cpufreq_driver_target(policy, policy->max,
> > + CPUFREQ_RELATION_H);
> > + else if (policy->min > userspace->setspeed)
> > + __cpufreq_driver_target(policy, policy->min,
> > + CPUFREQ_RELATION_L);
> > else
> > - __cpufreq_driver_target(policy, *setspeed, CPUFREQ_RELATION_L);
> > + __cpufreq_driver_target(policy, userspace->setspeed,
> > + CPUFREQ_RELATION_L);
> >
> > - mutex_unlock(&userspace_mutex);
> > + mutex_unlock(&userspace->mutex);
> > }
> >
> > static struct cpufreq_governor cpufreq_gov_userspace = {
>
> Acked-by: Viresh Kumar <[email protected]>

Applied as 6.7 material along with the [2/2], thanks!