[PATCH 0/4] Per CPU idle injection

[PATCH 0/4] Per CPU idle injection

[Srinivas Pandruvada]

This series introduces per CPU idle injection. In preparation for this
enhance the existing powercap/idle_inject and modify intel_powerclamp
to use this. Then add per core idle injection driver.

Srinivas Pandruvada (4):
  powercap: idle_inject: Export symbols
  powercap: idle_inject: Add begin/end callbacks
  thermal/drivers/intel_powerclamp: Use powercap idle-inject framework
  thermal/drivers/intel_cpu_idle_cooling: Introduce Intel cpu idle
    cooling driver

 drivers/powercap/idle_inject.c                |  26 +-
 drivers/thermal/cpuidle_cooling.c             |   2 +-
 drivers/thermal/intel/Kconfig                 |  11 +
 drivers/thermal/intel/Makefile                |   1 +
 .../thermal/intel/intel_cpu_idle_cooling.c    | 262 ++++++++++++++++
 drivers/thermal/intel/intel_powerclamp.c      | 293 ++++++++----------
 include/linux/idle_inject.h                   |   4 +-
 7 files changed, 428 insertions(+), 171 deletions(-)
 create mode 100644 drivers/thermal/intel/intel_cpu_idle_cooling.c

-- 
2.37.3

[PATCH 1/4] powercap: idle_inject: Export symbols

[Srinivas Pandruvada]

Export symbols for external interfaces, so that they can be used in
other loadable modules.

Export is done under name space IDLE_INJECT.

Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
---
 drivers/powercap/idle_inject.c | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/drivers/powercap/idle_inject.c b/drivers/powercap/idle_inject.c
index 999e218d7793..e73885bd9065 100644
--- a/drivers/powercap/idle_inject.c
+++ b/drivers/powercap/idle_inject.c
@@ -159,6 +159,7 @@ void idle_inject_set_duration(struct idle_inject_device *ii_dev,
 		WRITE_ONCE(ii_dev->idle_duration_us, idle_duration_us);
 	}
 }
+EXPORT_SYMBOL_NS_GPL(idle_inject_set_duration, IDLE_INJECT);
 
 /**
  * idle_inject_get_duration - idle and run duration retrieval helper
@@ -172,6 +173,7 @@ void idle_inject_get_duration(struct idle_inject_device *ii_dev,
 	*run_duration_us = READ_ONCE(ii_dev->run_duration_us);
 	*idle_duration_us = READ_ONCE(ii_dev->idle_duration_us);
 }
+EXPORT_SYMBOL_NS_GPL(idle_inject_get_duration, IDLE_INJECT);
 
 /**
  * idle_inject_set_latency - set the maximum latency allowed
@@ -182,6 +184,7 @@ void idle_inject_set_latency(struct idle_inject_device *ii_dev,
 {
 	WRITE_ONCE(ii_dev->latency_us, latency_us);
 }
+EXPORT_SYMBOL_NS_GPL(idle_inject_set_latency, IDLE_INJECT);
 
 /**
  * idle_inject_start - start idle injections
@@ -213,6 +216,7 @@ int idle_inject_start(struct idle_inject_device *ii_dev)
 
 	return 0;
 }
+EXPORT_SYMBOL_NS_GPL(idle_inject_start, IDLE_INJECT);
 
 /**
  * idle_inject_stop - stops idle injections
@@ -259,6 +263,7 @@ void idle_inject_stop(struct idle_inject_device *ii_dev)
 
 	cpu_hotplug_enable();
 }
+EXPORT_SYMBOL_NS_GPL(idle_inject_stop, IDLE_INJECT);
 
 /**
  * idle_inject_setup - prepare the current task for idle injection
@@ -334,6 +339,7 @@ struct idle_inject_device *idle_inject_register(struct cpumask *cpumask)
 
 	return NULL;
 }
+EXPORT_SYMBOL_NS_GPL(idle_inject_register, IDLE_INJECT);
 
 /**
  * idle_inject_unregister - unregister idle injection control device
@@ -354,6 +360,7 @@ void idle_inject_unregister(struct idle_inject_device *ii_dev)
 
 	kfree(ii_dev);
 }
+EXPORT_SYMBOL_NS_GPL(idle_inject_unregister, IDLE_INJECT);
 
 static struct smp_hotplug_thread idle_inject_threads = {
 	.store = &idle_inject_thread.tsk,
-- 
2.37.3

[PATCH 2/4] powercap: idle_inject: Add begin/end callbacks

[Srinivas Pandruvada]

The actual CPU Idle percent can be different than what can be observed
from the hardware. Since the objective for CPU Idle injection is for
thermal control, the idle percent observed by the hardware is more
relevant.

To account for hardware feedback the actual runtime/idle time should be
adjusted.

Add a capability to register a begin and end callback during call to
idle_inject_register(). If they are not NULL, then begin callback is
called before calling play_idle_precise() and end callback is called
after play_idle_precise().

If begin callback is present and returns non 0 value then
play_idle_precise() is not called as it means there is some over
compensation.

Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
---
 drivers/powercap/idle_inject.c    | 19 ++++++++++++++++++-
 drivers/thermal/cpuidle_cooling.c |  2 +-
 include/linux/idle_inject.h       |  4 +++-
 3 files changed, 22 insertions(+), 3 deletions(-)

diff --git a/drivers/powercap/idle_inject.c b/drivers/powercap/idle_inject.c
index e73885bd9065..14968b0ff133 100644
--- a/drivers/powercap/idle_inject.c
+++ b/drivers/powercap/idle_inject.c
@@ -70,6 +70,8 @@ struct idle_inject_device {
 	unsigned int idle_duration_us;
 	unsigned int run_duration_us;
 	unsigned int latency_us;
+	int (*idle_inject_begin)(unsigned int cpu);
+	void (*idle_inject_end)(unsigned int cpu);
 	unsigned long cpumask[];
 };
 
@@ -132,6 +134,7 @@ static void idle_inject_fn(unsigned int cpu)
 {
 	struct idle_inject_device *ii_dev;
 	struct idle_inject_thread *iit;
+	int ret;
 
 	ii_dev = per_cpu(idle_inject_device, cpu);
 	iit = per_cpu_ptr(&idle_inject_thread, cpu);
@@ -141,8 +144,18 @@ static void idle_inject_fn(unsigned int cpu)
 	 */
 	iit->should_run = 0;
 
+	if (ii_dev->idle_inject_begin) {
+		ret = ii_dev->idle_inject_begin(cpu);
+		if (ret)
+			goto skip;
+	}
+
 	play_idle_precise(READ_ONCE(ii_dev->idle_duration_us) * NSEC_PER_USEC,
 			  READ_ONCE(ii_dev->latency_us) * NSEC_PER_USEC);
+
+skip:
+	if (ii_dev->idle_inject_end)
+		ii_dev->idle_inject_end(cpu);
 }
 
 /**
@@ -302,7 +315,9 @@ static int idle_inject_should_run(unsigned int cpu)
  * Return: NULL if memory allocation fails, idle injection control device
  * pointer on success.
  */
-struct idle_inject_device *idle_inject_register(struct cpumask *cpumask)
+struct idle_inject_device *idle_inject_register(struct cpumask *cpumask,
+						int (*idle_inject_begin)(unsigned int cpu),
+						void (*idle_inject_end)(unsigned int cpu))
 {
 	struct idle_inject_device *ii_dev;
 	int cpu, cpu_rb;
@@ -315,6 +330,8 @@ struct idle_inject_device *idle_inject_register(struct cpumask *cpumask)
 	hrtimer_init(&ii_dev->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 	ii_dev->timer.function = idle_inject_timer_fn;
 	ii_dev->latency_us = UINT_MAX;
+	ii_dev->idle_inject_begin = idle_inject_begin;
+	ii_dev->idle_inject_end = idle_inject_end;
 
 	for_each_cpu(cpu, to_cpumask(ii_dev->cpumask)) {
 
diff --git a/drivers/thermal/cpuidle_cooling.c b/drivers/thermal/cpuidle_cooling.c
index 4f41102e8b16..e8b35b3b5767 100644
--- a/drivers/thermal/cpuidle_cooling.c
+++ b/drivers/thermal/cpuidle_cooling.c
@@ -184,7 +184,7 @@ static int __cpuidle_cooling_register(struct device_node *np,
 		goto out;
 	}
 
-	ii_dev = idle_inject_register(drv->cpumask);
+	ii_dev = idle_inject_register(drv->cpumask, NULL, NULL);
 	if (!ii_dev) {
 		ret = -EINVAL;
 		goto out_kfree;
diff --git a/include/linux/idle_inject.h b/include/linux/idle_inject.h
index fb88e23a99d3..73f3414fafe2 100644
--- a/include/linux/idle_inject.h
+++ b/include/linux/idle_inject.h
@@ -11,7 +11,9 @@
 /* private idle injection device structure */
 struct idle_inject_device;
 
-struct idle_inject_device *idle_inject_register(struct cpumask *cpumask);
+struct idle_inject_device *idle_inject_register(struct cpumask *cpumask,
+						int (*idle_inject_begin)(unsigned int cpu),
+						void (*idle_inject_end)(unsigned int cpu));
 
 void idle_inject_unregister(struct idle_inject_device *ii_dev);
 
-- 
2.37.3

Re: [PATCH 2/4] powercap: idle_inject: Add begin/end callbacks

[Rafael J. Wysocki]

On Tue, Nov 8, 2022 at 4:04 AM Srinivas Pandruvada
<srinivas.pandruvada@linux.intel.com> wrote:
>
> The actual CPU Idle percent can be different than what can be observed
> from the hardware.

Can you expand this a bit, please?  It is not clear what the "CPU idle
percent" and "observed from the hardware" phrases mean here.

> Since the objective for CPU Idle injection is for
> thermal control, the idle percent observed by the hardware is more
> relevant.
>
> To account for hardware feedback the actual runtime/idle time should be
> adjusted.
>
> Add a capability to register a begin and end callback during call to

I would call them "prepare" and "complete" without the "idle_inject_" prefix.

> idle_inject_register(). If they are not NULL, then begin callback is
> called before calling play_idle_precise() and end callback is called
> after play_idle_precise().
>
> If begin callback is present and returns non 0 value then
> play_idle_precise() is not called as it means there is some over
> compensation.

This behavior needs to be documented somewhere other than the patch changelog.

> Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
> ---
>  drivers/powercap/idle_inject.c    | 19 ++++++++++++++++++-
>  drivers/thermal/cpuidle_cooling.c |  2 +-
>  include/linux/idle_inject.h       |  4 +++-
>  3 files changed, 22 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/powercap/idle_inject.c b/drivers/powercap/idle_inject.c
> index e73885bd9065..14968b0ff133 100644
> --- a/drivers/powercap/idle_inject.c
> +++ b/drivers/powercap/idle_inject.c
> @@ -70,6 +70,8 @@ struct idle_inject_device {
>         unsigned int idle_duration_us;
>         unsigned int run_duration_us;
>         unsigned int latency_us;
> +       int (*idle_inject_begin)(unsigned int cpu);
> +       void (*idle_inject_end)(unsigned int cpu);

The comment above needs to be updated.  Also please see the remark
above regarding callback names.

>         unsigned long cpumask[];
>  };
>
> @@ -132,6 +134,7 @@ static void idle_inject_fn(unsigned int cpu)
>  {
>         struct idle_inject_device *ii_dev;
>         struct idle_inject_thread *iit;
> +       int ret;
>
>         ii_dev = per_cpu(idle_inject_device, cpu);
>         iit = per_cpu_ptr(&idle_inject_thread, cpu);
> @@ -141,8 +144,18 @@ static void idle_inject_fn(unsigned int cpu)
>          */
>         iit->should_run = 0;
>
> +       if (ii_dev->idle_inject_begin) {
> +               ret = ii_dev->idle_inject_begin(cpu);
> +               if (ret)
> +                       goto skip;
> +       }
> +
>         play_idle_precise(READ_ONCE(ii_dev->idle_duration_us) * NSEC_PER_USEC,
>                           READ_ONCE(ii_dev->latency_us) * NSEC_PER_USEC);
> +
> +skip:
> +       if (ii_dev->idle_inject_end)
> +               ii_dev->idle_inject_end(cpu);
>  }
>
>  /**
> @@ -302,7 +315,9 @@ static int idle_inject_should_run(unsigned int cpu)
>   * Return: NULL if memory allocation fails, idle injection control device
>   * pointer on success.
>   */
> -struct idle_inject_device *idle_inject_register(struct cpumask *cpumask)
> +struct idle_inject_device *idle_inject_register(struct cpumask *cpumask,
> +                                               int (*idle_inject_begin)(unsigned int cpu),
> +                                               void (*idle_inject_end)(unsigned int cpu))

Instead of modifying this, I would add something like
idle_inject_register_full() that will take the callback arguments and
will be called internally by idle_inject_register().

>  {
>         struct idle_inject_device *ii_dev;
>         int cpu, cpu_rb;
> @@ -315,6 +330,8 @@ struct idle_inject_device *idle_inject_register(struct cpumask *cpumask)
>         hrtimer_init(&ii_dev->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
>         ii_dev->timer.function = idle_inject_timer_fn;
>         ii_dev->latency_us = UINT_MAX;
> +       ii_dev->idle_inject_begin = idle_inject_begin;
> +       ii_dev->idle_inject_end = idle_inject_end;
>
>         for_each_cpu(cpu, to_cpumask(ii_dev->cpumask)) {
>
> diff --git a/drivers/thermal/cpuidle_cooling.c b/drivers/thermal/cpuidle_cooling.c
> index 4f41102e8b16..e8b35b3b5767 100644
> --- a/drivers/thermal/cpuidle_cooling.c
> +++ b/drivers/thermal/cpuidle_cooling.c
> @@ -184,7 +184,7 @@ static int __cpuidle_cooling_register(struct device_node *np,
>                 goto out;
>         }
>
> -       ii_dev = idle_inject_register(drv->cpumask);
> +       ii_dev = idle_inject_register(drv->cpumask, NULL, NULL);

So this change would not be necessary any more.

>         if (!ii_dev) {
>                 ret = -EINVAL;
>                 goto out_kfree;
> diff --git a/include/linux/idle_inject.h b/include/linux/idle_inject.h
> index fb88e23a99d3..73f3414fafe2 100644
> --- a/include/linux/idle_inject.h
> +++ b/include/linux/idle_inject.h
> @@ -11,7 +11,9 @@
>  /* private idle injection device structure */
>  struct idle_inject_device;
>
> -struct idle_inject_device *idle_inject_register(struct cpumask *cpumask);
> +struct idle_inject_device *idle_inject_register(struct cpumask *cpumask,
> +                                               int (*idle_inject_begin)(unsigned int cpu),
> +                                               void (*idle_inject_end)(unsigned int cpu));
>
>  void idle_inject_unregister(struct idle_inject_device *ii_dev);
>
> --

Re: [PATCH 2/4] powercap: idle_inject: Add begin/end callbacks

[srinivas pandruvada]

On Wed, 2022-11-09 at 15:48 +0100, Rafael J. Wysocki wrote:
> On Tue, Nov 8, 2022 at 4:04 AM Srinivas Pandruvada
> <srinivas.pandruvada@linux.intel.com> wrote:
> > 
> > The actual CPU Idle percent can be different than what can be
> > observed
> > from the hardware.
> 
> Can you expand this a bit, please?  It is not clear what the "CPU
> idle
> percent" and "observed from the hardware" phrases mean here.
Sure.

> 
> > Since the objective for CPU Idle injection is for
> > thermal control, the idle percent observed by the hardware is more
> > relevant.
> > 
> > To account for hardware feedback the actual runtime/idle time
> > should be
> > adjusted.
> > 
> > Add a capability to register a begin and end callback during call
> > to
> 
> I would call them "prepare" and "complete" without the "idle_inject_"
> prefix.
OK

> 
> > idle_inject_register(). If they are not NULL, then begin callback
> > is
> > called before calling play_idle_precise() and end callback is
> > called
> > after play_idle_precise().
> > 
> > If begin callback is present and returns non 0 value then
> > play_idle_precise() is not called as it means there is some over
> > compensation.
> 
> This behavior needs to be documented somewhere other than the patch
> changelog.
We can add a kernel doc and add there.

> 
> > Signed-off-by: Srinivas Pandruvada <   
> > srinivas.pandruvada@linux.intel.com>
> > ---
> >  drivers/powercap/idle_inject.c    | 19 ++++++++++++++++++-
> >  drivers/thermal/cpuidle_cooling.c |  2 +-
> >  include/linux/idle_inject.h       |  4 +++-
> >  3 files changed, 22 insertions(+), 3 deletions(-)
> > 
> > diff --git a/drivers/powercap/idle_inject.c
> > b/drivers/powercap/idle_inject.c
> > index e73885bd9065..14968b0ff133 100644
> > --- a/drivers/powercap/idle_inject.c
> > +++ b/drivers/powercap/idle_inject.c
> > @@ -70,6 +70,8 @@ struct idle_inject_device {
> >         unsigned int idle_duration_us;
> >         unsigned int run_duration_us;
> >         unsigned int latency_us;
> > +       int (*idle_inject_begin)(unsigned int cpu);
> > +       void (*idle_inject_end)(unsigned int cpu);
> 
> The comment above needs to be updated.  Also please see the remark
> above regarding callback names.
> 
> >         unsigned long cpumask[];
> >  };
> > 
> > @@ -132,6 +134,7 @@ static void idle_inject_fn(unsigned int cpu)
> >  {
> >         struct idle_inject_device *ii_dev;
> >         struct idle_inject_thread *iit;
> > +       int ret;
> > 
> >         ii_dev = per_cpu(idle_inject_device, cpu);
> >         iit = per_cpu_ptr(&idle_inject_thread, cpu);
> > @@ -141,8 +144,18 @@ static void idle_inject_fn(unsigned int cpu)
> >          */
> >         iit->should_run = 0;
> > 
> > +       if (ii_dev->idle_inject_begin) {
> > +               ret = ii_dev->idle_inject_begin(cpu);
> > +               if (ret)
> > +                       goto skip;
> > +       }
> > +
> >         play_idle_precise(READ_ONCE(ii_dev->idle_duration_us) *
> > NSEC_PER_USEC,
> >                           READ_ONCE(ii_dev->latency_us) *
> > NSEC_PER_USEC);
> > +
> > +skip:
> > +       if (ii_dev->idle_inject_end)
> > +               ii_dev->idle_inject_end(cpu);
> >  }
> > 
> >  /**
> > @@ -302,7 +315,9 @@ static int idle_inject_should_run(unsigned int
> > cpu)
> >   * Return: NULL if memory allocation fails, idle injection control
> > device
> >   * pointer on success.
> >   */
> > -struct idle_inject_device *idle_inject_register(struct cpumask
> > *cpumask)
> > +struct idle_inject_device *idle_inject_register(struct cpumask
> > *cpumask,
> > +                                               int
> > (*idle_inject_begin)(unsigned int cpu),
> > +                                               void
> > (*idle_inject_end)(unsigned int cpu))
> 
> Instead of modifying this, I would add something like
> idle_inject_register_full() that will take the callback arguments and
> will be called internally by idle_inject_register().
OK

Thanks,
Srinivas

> 
> >  {
> >         struct idle_inject_device *ii_dev;
> >         int cpu, cpu_rb;
> > @@ -315,6 +330,8 @@ struct idle_inject_device
> > *idle_inject_register(struct cpumask *cpumask)
> >         hrtimer_init(&ii_dev->timer, CLOCK_MONOTONIC,
> > HRTIMER_MODE_REL);
> >         ii_dev->timer.function = idle_inject_timer_fn;
> >         ii_dev->latency_us = UINT_MAX;
> > +       ii_dev->idle_inject_begin = idle_inject_begin;
> > +       ii_dev->idle_inject_end = idle_inject_end;
> > 
> >         for_each_cpu(cpu, to_cpumask(ii_dev->cpumask)) {
> > 
> > diff --git a/drivers/thermal/cpuidle_cooling.c
> > b/drivers/thermal/cpuidle_cooling.c
> > index 4f41102e8b16..e8b35b3b5767 100644
> > --- a/drivers/thermal/cpuidle_cooling.c
> > +++ b/drivers/thermal/cpuidle_cooling.c
> > @@ -184,7 +184,7 @@ static int __cpuidle_cooling_register(struct
> > device_node *np,
> >                 goto out;
> >         }
> > 
> > -       ii_dev = idle_inject_register(drv->cpumask);
> > +       ii_dev = idle_inject_register(drv->cpumask, NULL, NULL);
> 
> So this change would not be necessary any more.
> 
> >         if (!ii_dev) {
> >                 ret = -EINVAL;
> >                 goto out_kfree;
> > diff --git a/include/linux/idle_inject.h
> > b/include/linux/idle_inject.h
> > index fb88e23a99d3..73f3414fafe2 100644
> > --- a/include/linux/idle_inject.h
> > +++ b/include/linux/idle_inject.h
> > @@ -11,7 +11,9 @@
> >  /* private idle injection device structure */
> >  struct idle_inject_device;
> > 
> > -struct idle_inject_device *idle_inject_register(struct cpumask
> > *cpumask);
> > +struct idle_inject_device *idle_inject_register(struct cpumask
> > *cpumask,
> > +                                               int
> > (*idle_inject_begin)(unsigned int cpu),
> > +                                               void
> > (*idle_inject_end)(unsigned int cpu));
> > 
> >  void idle_inject_unregister(struct idle_inject_device *ii_dev);
> > 
> > --

[PATCH 3/4] thermal/drivers/intel_powerclamp: Use powercap idle-inject framework

[Srinivas Pandruvada]

There are two idle injection implementation in the Linux kernel. One
via intel_powerclamp and the other using powercap/idle_inject. Both
implementation end up in calling play_idle* function from a FIFO
priority thread. Both can't be used at the same time.

Currently per core idle injection (cpuidle_cooling) is using
powercap/idle_inject, which is not used in platforms where
intel_powerclamp is used for system wide idle injection. So there is
no conflict. But there are some use cases where per core idle injection
is beneficial on the same system where system wide idle injection is
also used via intel_powerclamp. To avoid conflict only one of the idle
injection type must be in use at a time. This require a common framework
which both per core and system wide idle injection can use.

Here powercap/idle_inject can be used for both per-core and for system
wide idle injection. This framework has a well defined interface
which allow registry for per-core or for all CPUs (system wide). If
particular CPU is already participating in idle injection, the call
to registry fails. Here the registry can be done when user space
changes the current cooling device state.

Also one framework for idle injection is better as there is one loop
calling play_idle*, instead of multiple for better maintenance.

So, reuse powercap/idle_inject calls in intel_powerclamp. This simplifies
the code as all per CPU kthreads which calls play_idle* can be removed.

The changes include:
- Remove unneeded include files
- Remove per CPU kthread workers: balancing_work and idle_injection_work
- Reuse the compensation related code by moving from previous worker
thread to idle_injection callbacks
- Adjust the idle_duration and runtime by using powercap/idle_inject
interface
- Remove all variables, which are not required once powercap/idle_inject
is used
- Add mutex to avoid race during removal of idle injection during module
unload and user action to change idle inject percent
- Use READ_ONCE and WRITE_ONCE for data accessed from multiple CPUs

Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
---
 drivers/thermal/intel/Kconfig            |   1 +
 drivers/thermal/intel/intel_powerclamp.c | 293 ++++++++++-------------
 2 files changed, 126 insertions(+), 168 deletions(-)

diff --git a/drivers/thermal/intel/Kconfig b/drivers/thermal/intel/Kconfig
index f0c845679250..0883740bf70d 100644
--- a/drivers/thermal/intel/Kconfig
+++ b/drivers/thermal/intel/Kconfig
@@ -3,6 +3,7 @@ config INTEL_POWERCLAMP
 	tristate "Intel PowerClamp idle injection driver"
 	depends on X86
 	depends on CPU_SUP_INTEL
+	select IDLE_INJECT
 	help
 	  Enable this to enable Intel PowerClamp idle injection driver. This
 	  enforce idle time which results in more package C-state residency. The
diff --git a/drivers/thermal/intel/intel_powerclamp.c b/drivers/thermal/intel/intel_powerclamp.c
index b80e25ec1261..17154686827d 100644
--- a/drivers/thermal/intel/intel_powerclamp.c
+++ b/drivers/thermal/intel/intel_powerclamp.c
@@ -2,7 +2,7 @@
 /*
  * intel_powerclamp.c - package c-state idle injection
  *
- * Copyright (c) 2012, Intel Corporation.
+ * Copyright (c) 2022, Intel Corporation.
  *
  * Authors:
  *     Arjan van de Ven <arjan@linux.intel.com>
@@ -27,21 +27,15 @@
 #include <linux/module.h>
 #include <linux/kernel.h>
 #include <linux/delay.h>
-#include <linux/kthread.h>
 #include <linux/cpu.h>
 #include <linux/thermal.h>
-#include <linux/slab.h>
-#include <linux/tick.h>
 #include <linux/debugfs.h>
 #include <linux/seq_file.h>
-#include <linux/sched/rt.h>
-#include <uapi/linux/sched/types.h>
+#include <linux/idle_inject.h>
 
-#include <asm/nmi.h>
 #include <asm/msr.h>
 #include <asm/mwait.h>
 #include <asm/cpu_device_id.h>
-#include <asm/hardirq.h>
 
 #define MAX_TARGET_RATIO (50U)
 /* For each undisturbed clamping period (no extra wake ups during idle time),
@@ -60,6 +54,7 @@ static struct dentry *debug_dir;
 
 /* user selected target */
 static unsigned int set_target_ratio;
+static bool target_ratio_updated;
 static unsigned int current_ratio;
 static bool should_skip;
 
@@ -67,26 +62,20 @@ static unsigned int control_cpu; /* The cpu assigned to collect stat and update
 				  * control parameters. default to BSP but BSP
 				  * can be offlined.
 				  */
-static bool clamping;
-
-struct powerclamp_worker_data {
-	struct kthread_worker *worker;
-	struct kthread_work balancing_work;
-	struct kthread_delayed_work idle_injection_work;
+struct powerclamp_data {
 	unsigned int cpu;
 	unsigned int count;
 	unsigned int guard;
 	unsigned int window_size_now;
 	unsigned int target_ratio;
-	unsigned int duration_jiffies;
 	bool clamping;
 };
 
-static struct powerclamp_worker_data __percpu *worker_data;
+static struct powerclamp_data powerclamp_data;
+
 static struct thermal_cooling_device *cooling_dev;
-static unsigned long *cpu_clamping_mask;  /* bit map for tracking per cpu
-					   * clamping kthread worker
-					   */
+
+static DEFINE_MUTEX(powerclamp_lock);
 
 static unsigned int duration;
 static unsigned int pkg_cstate_ratio_cur;
@@ -344,79 +333,33 @@ static bool powerclamp_adjust_controls(unsigned int target_ratio,
 	return set_target_ratio + guard <= current_ratio;
 }
 
-static void clamp_balancing_func(struct kthread_work *work)
+static unsigned int get_run_time(void)
 {
-	struct powerclamp_worker_data *w_data;
-	int sleeptime;
-	unsigned long target_jiffies;
 	unsigned int compensated_ratio;
-	int interval; /* jiffies to sleep for each attempt */
-
-	w_data = container_of(work, struct powerclamp_worker_data,
-			      balancing_work);
+	unsigned int runtime;
 
 	/*
 	 * make sure user selected ratio does not take effect until
 	 * the next round. adjust target_ratio if user has changed
 	 * target such that we can converge quickly.
 	 */
-	w_data->target_ratio = READ_ONCE(set_target_ratio);
-	w_data->guard = 1 + w_data->target_ratio / 20;
-	w_data->window_size_now = window_size;
-	w_data->duration_jiffies = msecs_to_jiffies(duration);
-	w_data->count++;
+	powerclamp_data.target_ratio = READ_ONCE(set_target_ratio);
+	powerclamp_data.guard = 1 + powerclamp_data.target_ratio / 20;
+	powerclamp_data.window_size_now = window_size;
 
 	/*
 	 * systems may have different ability to enter package level
 	 * c-states, thus we need to compensate the injected idle ratio
 	 * to achieve the actual target reported by the HW.
 	 */
-	compensated_ratio = w_data->target_ratio +
-		get_compensation(w_data->target_ratio);
+	compensated_ratio = powerclamp_data.target_ratio +
+		get_compensation(powerclamp_data.target_ratio);
 	if (compensated_ratio <= 0)
 		compensated_ratio = 1;
-	interval = w_data->duration_jiffies * 100 / compensated_ratio;
-
-	/* align idle time */
-	target_jiffies = roundup(jiffies, interval);
-	sleeptime = target_jiffies - jiffies;
-	if (sleeptime <= 0)
-		sleeptime = 1;
-
-	if (clamping && w_data->clamping && cpu_online(w_data->cpu))
-		kthread_queue_delayed_work(w_data->worker,
-					   &w_data->idle_injection_work,
-					   sleeptime);
-}
 
-static void clamp_idle_injection_func(struct kthread_work *work)
-{
-	struct powerclamp_worker_data *w_data;
-
-	w_data = container_of(work, struct powerclamp_worker_data,
-			      idle_injection_work.work);
+	runtime = duration * 100 / compensated_ratio - duration;
 
-	/*
-	 * only elected controlling cpu can collect stats and update
-	 * control parameters.
-	 */
-	if (w_data->cpu == control_cpu &&
-	    !(w_data->count % w_data->window_size_now)) {
-		should_skip =
-			powerclamp_adjust_controls(w_data->target_ratio,
-						   w_data->guard,
-						   w_data->window_size_now);
-		smp_mb();
-	}
-
-	if (should_skip)
-		goto balance;
-
-	play_idle(jiffies_to_usecs(w_data->duration_jiffies));
-
-balance:
-	if (clamping && w_data->clamping && cpu_online(w_data->cpu))
-		kthread_queue_work(w_data->worker, &w_data->balancing_work);
+	return runtime;
 }
 
 /*
@@ -452,104 +395,128 @@ static void poll_pkg_cstate(struct work_struct *dummy)
 	msr_last = msr_now;
 	tsc_last = tsc_now;
 
-	if (true == clamping)
+	if (powerclamp_data.clamping)
 		schedule_delayed_work(&poll_pkg_cstate_work, HZ);
 }
 
-static void start_power_clamp_worker(unsigned long cpu)
+static struct idle_inject_device *ii_dev;
+
+static int idle_inject_begin(unsigned int cpu)
 {
-	struct powerclamp_worker_data *w_data = per_cpu_ptr(worker_data, cpu);
-	struct kthread_worker *worker;
+	/*
+	 * only elected controlling cpu can collect stats and update
+	 * control parameters.
+	 */
+	if (cpu == control_cpu) {
+		bool update = READ_ONCE(target_ratio_updated);
+
+		if (!(powerclamp_data.count % powerclamp_data.window_size_now)) {
+			bool skip = powerclamp_adjust_controls(powerclamp_data.target_ratio,
+						       powerclamp_data.guard,
+						       powerclamp_data.window_size_now);
+			WRITE_ONCE(should_skip, skip);
+			update = true;
+		}
 
-	worker = kthread_create_worker_on_cpu(cpu, 0, "kidle_inj/%ld", cpu);
-	if (IS_ERR(worker))
-		return;
+		if (update) {
+			unsigned int runtime;
+
+			runtime = get_run_time();
+			idle_inject_set_duration(ii_dev, runtime, duration);
+			WRITE_ONCE(target_ratio_updated, false);
+		}
+		powerclamp_data.count++;
+	}
+
+	if (READ_ONCE(should_skip))
+		return -EAGAIN;
 
-	w_data->worker = worker;
-	w_data->count = 0;
-	w_data->cpu = cpu;
-	w_data->clamping = true;
-	set_bit(cpu, cpu_clamping_mask);
-	sched_set_fifo(worker->task);
-	kthread_init_work(&w_data->balancing_work, clamp_balancing_func);
-	kthread_init_delayed_work(&w_data->idle_injection_work,
-				  clamp_idle_injection_func);
-	kthread_queue_work(w_data->worker, &w_data->balancing_work);
+	return 0;
 }
 
-static void stop_power_clamp_worker(unsigned long cpu)
+static void trigger_idle_injection(void)
 {
-	struct powerclamp_worker_data *w_data = per_cpu_ptr(worker_data, cpu);
+	unsigned int runtime = get_run_time();
 
-	if (!w_data->worker)
-		return;
+	idle_inject_set_duration(ii_dev, runtime, duration);
+	idle_inject_start(ii_dev);
+	powerclamp_data.clamping = true;
+}
+
+static int powerclamp_idle_injection_register(void)
+{
+	static cpumask_t idle_injection_cpu_mask;
+	unsigned long cpu;
 
-	w_data->clamping = false;
-	/*
-	 * Make sure that all works that get queued after this point see
-	 * the clamping disabled. The counter part is not needed because
-	 * there is an implicit memory barrier when the queued work
-	 * is proceed.
-	 */
-	smp_wmb();
-	kthread_cancel_work_sync(&w_data->balancing_work);
-	kthread_cancel_delayed_work_sync(&w_data->idle_injection_work);
 	/*
-	 * The balancing work still might be queued here because
-	 * the handling of the "clapming" variable, cancel, and queue
-	 * operations are not synchronized via a lock. But it is not
-	 * a big deal. The balancing work is fast and destroy kthread
-	 * will wait for it.
+	 * The idle inject core will only inject for online CPUs,
+	 * So we can register for all present CPUs. In this way
+	 * if some CPU goes online/offline while idle inject
+	 * is registered, nothing additional calls are required.
+	 * The same runtime and idle time is applicable for
+	 * newly onlined CPUs if any.
 	 */
-	clear_bit(w_data->cpu, cpu_clamping_mask);
-	kthread_destroy_worker(w_data->worker);
+	for_each_present_cpu(cpu) {
+		cpumask_set_cpu(cpu, &idle_injection_cpu_mask);
+	}
+
+	ii_dev = idle_inject_register(&idle_injection_cpu_mask,
+				      idle_inject_begin,
+				      NULL);
+	if (!ii_dev) {
+		pr_err("powerclamp: idle_inject_register failed\n");
+		return -EAGAIN;
+	}
 
-	w_data->worker = NULL;
+	idle_inject_set_duration(ii_dev, TICK_USEC, duration);
+	idle_inject_set_latency(ii_dev, UINT_MAX);
+
+	return 0;
+}
+
+static void remove_idle_injection(void)
+{
+	if (!powerclamp_data.clamping)
+		return;
+
+	powerclamp_data.clamping = false;
+	idle_inject_stop(ii_dev);
 }
 
 static int start_power_clamp(void)
 {
-	unsigned long cpu;
+	int ret;
 
-	set_target_ratio = clamp(set_target_ratio, 0U, MAX_TARGET_RATIO - 1);
 	/* prevent cpu hotplug */
 	cpus_read_lock();
 
 	/* prefer BSP */
 	control_cpu = cpumask_first(cpu_online_mask);
 
-	clamping = true;
-	schedule_delayed_work(&poll_pkg_cstate_work, 0);
-
-	/* start one kthread worker per online cpu */
-	for_each_online_cpu(cpu) {
-		start_power_clamp_worker(cpu);
+	ret = powerclamp_idle_injection_register();
+	if (!ret) {
+		trigger_idle_injection();
+		schedule_delayed_work(&poll_pkg_cstate_work, 0);
 	}
+
 	cpus_read_unlock();
 
-	return 0;
+	return ret;
 }
 
 static void end_power_clamp(void)
 {
-	int i;
-
-	/*
-	 * Block requeuing in all the kthread workers. They will flush and
-	 * stop faster.
-	 */
-	clamping = false;
-	for_each_set_bit(i, cpu_clamping_mask, num_possible_cpus()) {
-		pr_debug("clamping worker for cpu %d alive, destroy\n", i);
-		stop_power_clamp_worker(i);
+	if (powerclamp_data.clamping) {
+		remove_idle_injection();
+		idle_inject_unregister(ii_dev);
 	}
 }
 
 static int powerclamp_cpu_online(unsigned int cpu)
 {
-	if (clamping == false)
+	if (!powerclamp_data.clamping)
 		return 0;
-	start_power_clamp_worker(cpu);
+
 	/* prefer BSP as controlling CPU */
 	if (cpu == 0) {
 		control_cpu = 0;
@@ -560,10 +527,6 @@ static int powerclamp_cpu_online(unsigned int cpu)
 
 static int powerclamp_cpu_predown(unsigned int cpu)
 {
-	if (clamping == false)
-		return 0;
-
-	stop_power_clamp_worker(cpu);
 	if (cpu != control_cpu)
 		return 0;
 
@@ -585,7 +548,7 @@ static int powerclamp_get_max_state(struct thermal_cooling_device *cdev,
 static int powerclamp_get_cur_state(struct thermal_cooling_device *cdev,
 				 unsigned long *state)
 {
-	if (true == clamping)
+	if (powerclamp_data.clamping)
 		*state = pkg_cstate_ratio_cur;
 	else
 		/* to save power, do not poll idle ratio while not clamping */
@@ -599,24 +562,30 @@ static int powerclamp_set_cur_state(struct thermal_cooling_device *cdev,
 {
 	int ret = 0;
 
+	mutex_lock(&powerclamp_lock);
+
 	new_target_ratio = clamp(new_target_ratio, 0UL,
-				(unsigned long) (MAX_TARGET_RATIO-1));
-	if (set_target_ratio == 0 && new_target_ratio > 0) {
+				(unsigned long) (MAX_TARGET_RATIO - 1));
+	if (READ_ONCE(set_target_ratio) == 0 && new_target_ratio > 0) {
 		pr_info("Start idle injection to reduce power\n");
-		set_target_ratio = new_target_ratio;
+		WRITE_ONCE(set_target_ratio, new_target_ratio);
 		ret = start_power_clamp();
+		if (ret)
+			WRITE_ONCE(set_target_ratio, 0);
 		goto exit_set;
-	} else	if (set_target_ratio > 0 && new_target_ratio == 0) {
+	} else	if (READ_ONCE(set_target_ratio) > 0 && new_target_ratio == 0) {
 		pr_info("Stop forced idle injection\n");
 		end_power_clamp();
-		set_target_ratio = 0;
+		WRITE_ONCE(set_target_ratio, 0);
+		WRITE_ONCE(target_ratio_updated, false);
 	} else	/* adjust currently running */ {
-		set_target_ratio = new_target_ratio;
-		/* make new set_target_ratio visible to other cpus */
-		smp_mb();
+		WRITE_ONCE(set_target_ratio, new_target_ratio);
+		WRITE_ONCE(target_ratio_updated, true);
 	}
 
 exit_set:
+	mutex_unlock(&powerclamp_lock);
+
 	return ret;
 }
 
@@ -686,14 +655,10 @@ static int __init powerclamp_init(void)
 {
 	int retval;
 
-	cpu_clamping_mask = bitmap_zalloc(num_possible_cpus(), GFP_KERNEL);
-	if (!cpu_clamping_mask)
-		return -ENOMEM;
-
 	/* probe cpu features and ids here */
 	retval = powerclamp_probe();
 	if (retval)
-		goto exit_free;
+		return retval;
 
 	/* set default limit, maybe adjusted during runtime based on feedback */
 	window_size = 2;
@@ -702,53 +667,45 @@ static int __init powerclamp_init(void)
 					   powerclamp_cpu_online,
 					   powerclamp_cpu_predown);
 	if (retval < 0)
-		goto exit_free;
+		return retval;
 
 	hp_state = retval;
 
-	worker_data = alloc_percpu(struct powerclamp_worker_data);
-	if (!worker_data) {
-		retval = -ENOMEM;
-		goto exit_unregister;
-	}
-
 	cooling_dev = thermal_cooling_device_register("intel_powerclamp", NULL,
-						&powerclamp_cooling_ops);
+						      &powerclamp_cooling_ops);
 	if (IS_ERR(cooling_dev)) {
 		retval = -ENODEV;
-		goto exit_free_thread;
+		goto exit_unregister;
 	}
 
 	if (!duration)
-		duration = jiffies_to_msecs(DEFAULT_DURATION_JIFFIES);
+		duration = jiffies_to_usecs(DEFAULT_DURATION_JIFFIES);
 
 	powerclamp_create_debug_files();
 
 	return 0;
 
-exit_free_thread:
-	free_percpu(worker_data);
 exit_unregister:
 	cpuhp_remove_state_nocalls(hp_state);
-exit_free:
-	bitmap_free(cpu_clamping_mask);
 	return retval;
 }
 module_init(powerclamp_init);
 
 static void __exit powerclamp_exit(void)
 {
+	mutex_lock(&powerclamp_lock);
 	end_power_clamp();
+	mutex_unlock(&powerclamp_lock);
 	cpuhp_remove_state_nocalls(hp_state);
-	free_percpu(worker_data);
 	thermal_cooling_device_unregister(cooling_dev);
-	bitmap_free(cpu_clamping_mask);
 
 	cancel_delayed_work_sync(&poll_pkg_cstate_work);
 	debugfs_remove_recursive(debug_dir);
 }
 module_exit(powerclamp_exit);
 
+MODULE_IMPORT_NS(IDLE_INJECT);
+
 MODULE_LICENSE("GPL");
 MODULE_AUTHOR("Arjan van de Ven <arjan@linux.intel.com>");
 MODULE_AUTHOR("Jacob Pan <jacob.jun.pan@linux.intel.com>");
-- 
2.37.3

[PATCH 4/4] thermal/drivers/intel_cpu_idle_cooling: Introduce Intel cpu idle cooling driver

[Srinivas Pandruvada]

The cpu idle cooling is used to cool down a CPU by injecting idle cycles
at runtime. The objective is similar to intel_powerclamp driver, which is
used for system wide cooling by injecting idle on each CPU.

This driver is modeled after drivers/thermal/cpuidle_cooling.c by reusing
powercap/idle_inject framework.

On each CPU online a thermal cooling device is registered. The minimum
state of the cooling device is 0 and maximum is 100. When user space
changes the current state to non zero, then register with idle inject
framework and start idle inject.

The default idle duration is 24 milli seconds, matching intel_powerclamp,
which doesn't change based on the current state of cooling device. The
runtime is changed based on the current state.

Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
---
 drivers/thermal/intel/Kconfig                 |  10 +
 drivers/thermal/intel/Makefile                |   1 +
 .../thermal/intel/intel_cpu_idle_cooling.c    | 262 ++++++++++++++++++
 3 files changed, 273 insertions(+)
 create mode 100644 drivers/thermal/intel/intel_cpu_idle_cooling.c

diff --git a/drivers/thermal/intel/Kconfig b/drivers/thermal/intel/Kconfig
index 0883740bf70d..c93daa7c83eb 100644
--- a/drivers/thermal/intel/Kconfig
+++ b/drivers/thermal/intel/Kconfig
@@ -114,3 +114,13 @@ config INTEL_HFI_THERMAL
 	  These capabilities may change as a result of changes in the operating
 	  conditions of the system such power and thermal limits. If selected,
 	  the kernel relays updates in CPUs' capabilities to userspace.
+
+config INTEL_CPU_IDLE_COOLING
+	tristate "Intel CPU idle cooling device"
+	depends on IDLE_INJECT
+	help
+	  This implements the CPU cooling mechanism through
+	  idle injection. This will throttle the CPU by injecting
+	  idle cycle.
+	  Unlike Intel Power clamp driver, this driver provides
+	  idle injection for each CPU.
diff --git a/drivers/thermal/intel/Makefile b/drivers/thermal/intel/Makefile
index 9a8d8054f316..8d5f7b5cf9b7 100644
--- a/drivers/thermal/intel/Makefile
+++ b/drivers/thermal/intel/Makefile
@@ -14,3 +14,4 @@ obj-$(CONFIG_INTEL_TCC_COOLING)	+= intel_tcc_cooling.o
 obj-$(CONFIG_X86_THERMAL_VECTOR) += therm_throt.o
 obj-$(CONFIG_INTEL_MENLOW)	+= intel_menlow.o
 obj-$(CONFIG_INTEL_HFI_THERMAL) += intel_hfi.o
+obj-$(CONFIG_INTEL_CPU_IDLE_COOLING) += intel_cpu_idle_cooling.o
diff --git a/drivers/thermal/intel/intel_cpu_idle_cooling.c b/drivers/thermal/intel/intel_cpu_idle_cooling.c
new file mode 100644
index 000000000000..5df79f38d9fb
--- /dev/null
+++ b/drivers/thermal/intel/intel_cpu_idle_cooling.c
@@ -0,0 +1,262 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Per CPU Idle injection cooling device implementation
+ *
+ * Copyright (c) 2022, Intel Corporation.
+ * All rights reserved.
+ *
+ */
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <linux/cpufeature.h>
+#include <linux/cpuhotplug.h>
+#include <linux/idle_inject.h>
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/slab.h>
+#include <linux/thermal.h>
+#include <linux/topology.h>
+
+#include <asm/cpu_device_id.h>
+
+/* Duration match with intel_powerclamp driver */
+#define IDLE_DURATION		24000
+#define IDLE_LATENCY		UINT_MAX
+
+static int idle_duration_us = IDLE_DURATION;
+static int idle_latency_us = IDLE_LATENCY;
+
+module_param(idle_duration_us, int, 0644);
+MODULE_PARM_DESC(idle_duration_us,
+		 "Idle duration in us.");
+
+module_param(idle_latency_us, int, 0644);
+MODULE_PARM_DESC(idle_latency_us,
+		 "Idle latency in us.");
+
+/**
+ * struct cpuidle_cooling - Per instance data for cooling device
+ * @cpu: CPU number for this cooling device
+ * @ii_dev: Idle inject core instance pointer
+ * @cdev: Thermal core cooling device instance
+ * @state:  Current cooling device state
+ *
+ * Stores per instance cooling device state.
+ */
+struct cpuidle_cooling {
+	int cpu;
+	struct idle_inject_device *ii_dev;
+	struct thermal_cooling_device *cdev;
+	unsigned long state;
+};
+
+static DEFINE_PER_CPU(struct cpuidle_cooling, cooling_devs);
+static cpumask_t cpuidle_cpu_mask;
+
+/* Used for module unload protection with idle injection operations */
+static DEFINE_MUTEX(idle_cooling_lock);
+
+static unsigned int cpuidle_cooling_runtime(unsigned int idle_duration_us,
+					    unsigned long state)
+{
+	if (!state)
+		return 0;
+
+	return ((idle_duration_us * 100) / state) - idle_duration_us;
+}
+
+static int cpuidle_idle_injection_register(struct cpuidle_cooling *cooling_dev)
+{
+	struct idle_inject_device *ii_dev;
+
+	ii_dev = idle_inject_register((struct cpumask *)cpumask_of(cooling_dev->cpu),
+				      NULL, NULL);
+	if (!ii_dev) {
+		/*
+		 * It is busy as some other device claimed idle injection for this CPU
+		 * Also it is possible that memory allocation failure.
+		 */
+		pr_err("idle_inject_register failed for cpu:%d\n", cooling_dev->cpu);
+		return -EAGAIN;
+	}
+
+	idle_inject_set_duration(ii_dev, TICK_USEC, idle_duration_us);
+	idle_inject_set_latency(ii_dev, idle_latency_us);
+
+	cooling_dev->ii_dev = ii_dev;
+
+	return 0;
+}
+
+static void cpuidle_idle_injection_unregister(struct cpuidle_cooling *cooling_dev)
+{
+	idle_inject_unregister(cooling_dev->ii_dev);
+}
+
+static int cpuidle_cooling_get_max_state(struct thermal_cooling_device *cdev,
+					 unsigned long *state)
+{
+	*state = 100;
+
+	return 0;
+}
+
+static int cpuidle_cooling_get_cur_state(struct thermal_cooling_device *cdev,
+					 unsigned long *state)
+{
+	struct cpuidle_cooling *cooling_dev = cdev->devdata;
+
+	*state = READ_ONCE(cooling_dev->state);
+
+	return 0;
+}
+
+static int cpuidle_cooling_set_cur_state(struct thermal_cooling_device *cdev,
+					 unsigned long state)
+{
+	struct cpuidle_cooling *cooling_dev = cdev->devdata;
+	unsigned int runtime_us;
+	unsigned long curr_state;
+	int ret = 0;
+
+	mutex_lock(&idle_cooling_lock);
+
+	curr_state = READ_ONCE(cooling_dev->state);
+
+	if (!curr_state && state > 0) {
+		/*
+		 * This is the first time to start cooling, so register with
+		 * idle injection framework.
+		 */
+		if (!cooling_dev->ii_dev) {
+			ret = cpuidle_idle_injection_register(cooling_dev);
+			if (ret)
+				goto unlock_set_state;
+		}
+
+		runtime_us = cpuidle_cooling_runtime(idle_duration_us, state);
+
+		idle_inject_set_duration(cooling_dev->ii_dev, runtime_us, idle_duration_us);
+		idle_inject_start(cooling_dev->ii_dev);
+	} else if (curr_state > 0 && state) {
+		/* Simply update runtime */
+		runtime_us = cpuidle_cooling_runtime(idle_duration_us, state);
+		idle_inject_set_duration(cooling_dev->ii_dev, runtime_us, idle_duration_us);
+	} else if (curr_state > 0 && !state) {
+		idle_inject_stop(cooling_dev->ii_dev);
+		cpuidle_idle_injection_unregister(cooling_dev);
+		cooling_dev->ii_dev = NULL;
+	}
+
+	WRITE_ONCE(cooling_dev->state, state);
+
+unlock_set_state:
+	mutex_unlock(&idle_cooling_lock);
+
+	return ret;
+}
+
+/**
+ * cpuidle_cooling_ops - thermal cooling device ops
+ */
+static struct thermal_cooling_device_ops cpuidle_cooling_ops = {
+	.get_max_state = cpuidle_cooling_get_max_state,
+	.get_cur_state = cpuidle_cooling_get_cur_state,
+	.set_cur_state = cpuidle_cooling_set_cur_state,
+};
+
+static int cpuidle_cooling_register(int cpu)
+{
+	struct cpuidle_cooling *cooling_dev = &per_cpu(cooling_devs, cpu);
+	struct thermal_cooling_device *cdev;
+	char name[14]; /* storage for cpuidle-XXXX */
+	int ret = 0;
+
+	mutex_lock(&idle_cooling_lock);
+
+	snprintf(name, sizeof(name), "cpuidle-%d", cpu);
+	cdev = thermal_cooling_device_register(name, cooling_dev, &cpuidle_cooling_ops);
+	if (IS_ERR(cdev)) {
+		ret = PTR_ERR(cdev);
+		goto unlock_register;
+	}
+
+	cooling_dev->cdev = cdev;
+	cpumask_set_cpu(cpu, &cpuidle_cpu_mask);
+	cooling_dev->cpu = cpu;
+
+unlock_register:
+	mutex_unlock(&idle_cooling_lock);
+
+	return ret;
+}
+
+static void cpuidle_cooling_unregister(int cpu)
+{
+	struct cpuidle_cooling *cooling_dev = &per_cpu(cooling_devs, cpu);
+
+	mutex_lock(&idle_cooling_lock);
+
+	if (cooling_dev->state) {
+		idle_inject_stop(cooling_dev->ii_dev);
+		cpuidle_idle_injection_unregister(cooling_dev);
+	}
+
+	thermal_cooling_device_unregister(cooling_dev->cdev);
+	cooling_dev->state = 0;
+
+	mutex_unlock(&idle_cooling_lock);
+}
+
+static int cpuidle_cooling_cpu_online(unsigned int cpu)
+{
+	cpuidle_cooling_register(cpu);
+
+	return 0;
+}
+
+static int cpuidle_cooling_cpu_offline(unsigned int cpu)
+{
+	cpuidle_cooling_unregister(cpu);
+
+	return 0;
+}
+
+static enum cpuhp_state cpuidle_cooling_hp_state __read_mostly;
+
+static const struct x86_cpu_id intel_cpuidle_cooling_ids[] __initconst = {
+	X86_MATCH_VENDOR_FEATURE(INTEL, X86_FEATURE_MWAIT, NULL),
+	{}
+};
+MODULE_DEVICE_TABLE(x86cpu, intel_cpuidle_cooling_ids);
+
+static int __init cpuidle_cooling_init(void)
+{
+	int ret;
+
+	if (!x86_match_cpu(intel_cpuidle_cooling_ids))
+		return -ENODEV;
+
+	ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN,
+				"thermal/cpuidle_cooling:online",
+				cpuidle_cooling_cpu_online,
+				cpuidle_cooling_cpu_offline);
+	if (ret < 0)
+		return ret;
+
+	cpuidle_cooling_hp_state = ret;
+
+	return 0;
+}
+module_init(cpuidle_cooling_init)
+
+static void __exit cpuidle_cooling_exit(void)
+{
+	cpuhp_remove_state(cpuidle_cooling_hp_state);
+}
+module_exit(cpuidle_cooling_exit)
+
+MODULE_IMPORT_NS(IDLE_INJECT);
+
+MODULE_LICENSE("GPL");
-- 
2.37.3