[PATCH v3 0/4] Per CPU idle injection

[PATCH v3 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.

v3
- Change callback from per CPU to per device and use in intel_powerclamp
- Remove unused var in per cpu idle injection module

v2
- Update based on feedback from Rafael on patch 2/4
- Kconfig dependency issue
Reported-by: kernel test robot <lkp@intel.com>

Srinivas Pandruvada (4):
  powercap: idle_inject: Export symbols
  powercap: idle_inject: Add update callback
  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                |  57 ++-
 drivers/thermal/intel/Kconfig                 |  12 +
 drivers/thermal/intel/Makefile                |   1 +
 .../thermal/intel/intel_cpu_idle_cooling.c    | 259 ++++++++++++++
 drivers/thermal/intel/intel_powerclamp.c      | 328 +++++++-----------
 include/linux/idle_inject.h                   |   3 +
 6 files changed, 443 insertions(+), 217 deletions(-)
 create mode 100644 drivers/thermal/intel/intel_cpu_idle_cooling.c

-- 
2.31.1

[PATCH v3 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>
---
v2/v3:
	No change

 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 fe86a09e3b67..dfa989182e71 100644
--- a/drivers/powercap/idle_inject.c
+++ b/drivers/powercap/idle_inject.c
@@ -160,6 +160,7 @@ 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
@@ -174,6 +175,7 @@ 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
@@ -185,6 +187,7 @@ 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
@@ -216,6 +219,7 @@ 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
@@ -262,6 +266,7 @@ 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
@@ -337,6 +342,7 @@ 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
@@ -357,6 +363,7 @@ 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,
-- 
2.31.1

Re: [PATCH v3 1/4] powercap: idle_inject: Export symbols

[Daniel Lezcano]

On 17/01/2023 03:07, Srinivas Pandruvada wrote:
> 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>
> ---

Acked-by: Daniel Lezcano <daniel.lezcano@linaro.org>


-- 
<http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs

Follow Linaro:  <http://www.facebook.com/pages/Linaro> Facebook |
<http://twitter.com/#!/linaroorg> Twitter |
<http://www.linaro.org/linaro-blog/> Blog

[PATCH v3 2/4] powercap: idle_inject: Add update callback

[Srinivas Pandruvada]

The powercap/idle_inject core uses play_idle_precise() to inject idle
time. But play_idle_precise() can't ensure that the CPU is fully idle
for the specified duration because of wakeups due to interrupts. To
compensate for the reduced idle time due to these wakes, the caller
can adjust requested idle time for the next cycle.

The goal of idle injection is to keep system at some idle percent on
average, so this is fine to overshoot or undershoot instantaneous idle
times.

The idle inject core provides an interface idle_inject_set_duration()
to set idle and runtime duration.

Some architectures provide interface to get actual idle time observed
by the hardware. So, the effective idle percent can be adjusted using
the hardware feedback. For example, Intel CPUs provides package idle
counters, which is currently used by Intel powerclamp driver to
readjust runtime duration.

When the caller's desired idle time over a period is less or greater
than the actual CPU idle time observed by the hardware, caller can
readjust idle and runtime duration for the next cycle.

The only way this can be done currently is by monitoring hardware idle
time from a different software thread and readjust idle and runtime
duration using idle_inject_set_duration().

This can be avoided by adding a callback which callers can register and
readjust from this callback function.

Add a capability to register an optional update() callback, which can be
called from the idle inject core before waking up CPUs for idle injection.
This callback can be registered via a new interface:
idle_inject_register_full().

During this process of constantly adjusting idle and runtime duration
there can be some cases where actual idle time is more than the desired.
In this case idle inject can be skipped for a cycle. If update() callback
returns false, then the idle inject core skips waking up CPUs for the
idle injection.

Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
---
v3:
- Replace prepare/complete callback with update callback

v2
- Replace begin/end with prepare/complete
- Add new interface idle_inject_register_full with callbacks
- Update kernel doc
- Update commit description

 drivers/powercap/idle_inject.c | 50 ++++++++++++++++++++++++++++++----
 include/linux/idle_inject.h    |  3 ++
 2 files changed, 47 insertions(+), 6 deletions(-)

diff --git a/drivers/powercap/idle_inject.c b/drivers/powercap/idle_inject.c
index dfa989182e71..c57b40477246 100644
--- a/drivers/powercap/idle_inject.c
+++ b/drivers/powercap/idle_inject.c
@@ -63,13 +63,27 @@ struct idle_inject_thread {
  * @idle_duration_us: duration of CPU idle time to inject
  * @run_duration_us: duration of CPU run time to allow
  * @latency_us: max allowed latency
+ * @update: Optional callback deciding whether or not to skip idle
+ *		injection in the given cycle.
  * @cpumask: mask of CPUs affected by idle injection
+ *
+ * This structure is used to define per instance idle inject device data. Each
+ * instance has an idle duration, a run duration and mask of CPUs to inject
+ * idle.
+ * Actual idle is injected by calling kernel scheduler interface
+ * play_idle_precise(). There is one optional callbacks which the caller can
+ * register by calling idle_inject_register_full():
+ * update() - This callback is called just before waking up CPUs to inject
+ * idle. If this callback returns false, CPUs are not woken up to inject idle
+ * for this cycle. Also gives opportunity to the caller to readjust idle
+ * and run duration by calling idle_inject_set_duration() for the next cycle.
  */
 struct idle_inject_device {
 	struct hrtimer timer;
 	unsigned int idle_duration_us;
 	unsigned int run_duration_us;
 	unsigned int latency_us;
+	bool (*update)(void);
 	unsigned long cpumask[];
 };
 
@@ -111,11 +125,12 @@ static enum hrtimer_restart idle_inject_timer_fn(struct hrtimer *timer)
 	struct idle_inject_device *ii_dev =
 		container_of(timer, struct idle_inject_device, timer);
 
+	if (!ii_dev->update || (ii_dev->update && ii_dev->update()))
+		idle_inject_wakeup(ii_dev);
+
 	duration_us = READ_ONCE(ii_dev->run_duration_us);
 	duration_us += READ_ONCE(ii_dev->idle_duration_us);
 
-	idle_inject_wakeup(ii_dev);
-
 	hrtimer_forward_now(timer, ns_to_ktime(duration_us * NSEC_PER_USEC));
 
 	return HRTIMER_RESTART;
@@ -295,17 +310,22 @@ static int idle_inject_should_run(unsigned int cpu)
 }
 
 /**
- * idle_inject_register - initialize idle injection on a set of CPUs
+ * idle_inject_register_full - initialize idle injection on a set of CPUs
  * @cpumask: CPUs to be affected by idle injection
+ * @update: This callback is called just before waking up CPUs to inject
+ * idle
  *
  * This function creates an idle injection control device structure for the
- * given set of CPUs and initializes the timer associated with it.  It does not
- * start any injection cycles.
+ * given set of CPUs and initializes the timer associated with it. This
+ * function also allows to register update()callback.
+ * It does not start any injection cycles.
  *
  * 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_full(struct cpumask *cpumask,
+						     bool (*update)(void))
 {
 	struct idle_inject_device *ii_dev;
 	int cpu, cpu_rb;
@@ -318,6 +338,7 @@ 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->update = update;
 
 	for_each_cpu(cpu, to_cpumask(ii_dev->cpumask)) {
 
@@ -342,6 +363,23 @@ struct idle_inject_device *idle_inject_register(struct cpumask *cpumask)
 
 	return NULL;
 }
+EXPORT_SYMBOL_NS_GPL(idle_inject_register_full, IDLE_INJECT);
+
+/**
+ * idle_inject_register - initialize idle injection on a set of CPUs
+ * @cpumask: CPUs to be affected by idle injection
+ *
+ * This function creates an idle injection control device structure for the
+ * given set of CPUs and initializes the timer associated with it.  It does not
+ * start any injection cycles.
+ *
+ * Return: NULL if memory allocation fails, idle injection control device
+ * pointer on success.
+ */
+struct idle_inject_device *idle_inject_register(struct cpumask *cpumask)
+{
+	return idle_inject_register_full(cpumask, NULL);
+}
 EXPORT_SYMBOL_NS_GPL(idle_inject_register, IDLE_INJECT);
 
 /**
diff --git a/include/linux/idle_inject.h b/include/linux/idle_inject.h
index fb88e23a99d3..a85d5dd40f72 100644
--- a/include/linux/idle_inject.h
+++ b/include/linux/idle_inject.h
@@ -13,6 +13,9 @@ struct idle_inject_device;
 
 struct idle_inject_device *idle_inject_register(struct cpumask *cpumask);
 
+struct idle_inject_device *idle_inject_register_full(struct cpumask *cpumask,
+						     bool (*update)(void));
+
 void idle_inject_unregister(struct idle_inject_device *ii_dev);
 
 int idle_inject_start(struct idle_inject_device *ii_dev);
-- 
2.31.1

Re: [PATCH v3 2/4] powercap: idle_inject: Add update callback

[Daniel Lezcano]

On 17/01/2023 03:07, Srinivas Pandruvada wrote:

[ ... ]

> +struct idle_inject_device *idle_inject_register(struct cpumask *cpumask)
> +{
> +	return idle_inject_register_full(cpumask, NULL);
> +}
>   EXPORT_SYMBOL_NS_GPL(idle_inject_register, IDLE_INJECT);

Why not just add the parameter to idle_inject_register() ?

There is only one user ATM


-- 
<http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs

Follow Linaro:  <http://www.facebook.com/pages/Linaro> Facebook |
<http://twitter.com/#!/linaroorg> Twitter |
<http://www.linaro.org/linaro-blog/> Blog

Re: [PATCH v3 2/4] powercap: idle_inject: Add update callback

[srinivas pandruvada]

On Tue, 2023-01-17 at 12:13 +0100, Daniel Lezcano wrote:
> On 17/01/2023 03:07, Srinivas Pandruvada wrote:
> 
> [ ... ]
> 
> > +struct idle_inject_device *idle_inject_register(struct cpumask
> > *cpumask)
> > +{
> > +       return idle_inject_register_full(cpumask, NULL);
> > +}
> >   EXPORT_SYMBOL_NS_GPL(idle_inject_register, IDLE_INJECT);
> 
> Why not just add the parameter to idle_inject_register() ?
> 
> There is only one user ATM
That was done in v1. But Rafael suggested to avoid changes to existing
kernel source.

Thanks,
Srinivas

> 
> 

Re: [PATCH v3 2/4] powercap: idle_inject: Add update callback

[Rafael J. Wysocki]

On Tue, Jan 17, 2023 at 4:08 PM srinivas pandruvada
<srinivas.pandruvada@linux.intel.com> wrote:
>
> On Tue, 2023-01-17 at 12:13 +0100, Daniel Lezcano wrote:
> > On 17/01/2023 03:07, Srinivas Pandruvada wrote:
> >
> > [ ... ]
> >
> > > +struct idle_inject_device *idle_inject_register(struct cpumask
> > > *cpumask)
> > > +{
> > > +       return idle_inject_register_full(cpumask, NULL);
> > > +}
> > >   EXPORT_SYMBOL_NS_GPL(idle_inject_register, IDLE_INJECT);
> >
> > Why not just add the parameter to idle_inject_register() ?
> >
> > There is only one user ATM
> That was done in v1. But Rafael suggested to avoid changes to existing
> kernel source.

However, it can be done if Daniel prefers that.  It is 1 callback now
only, so no big deal I suppose.

Re: [PATCH v3 2/4] powercap: idle_inject: Add update callback

[srinivas pandruvada]

On Tue, 2023-01-17 at 16:09 +0100, Rafael J. Wysocki wrote:

Hi Daniel,

> On Tue, Jan 17, 2023 at 4:08 PM srinivas pandruvada
> <srinivas.pandruvada@linux.intel.com> wrote:
> > 
> > On Tue, 2023-01-17 at 12:13 +0100, Daniel Lezcano wrote:
> > > On 17/01/2023 03:07, Srinivas Pandruvada wrote:
> > > 
> > > [ ... ]
> > > 
> > > > +struct idle_inject_device *idle_inject_register(struct cpumask
> > > > *cpumask)
> > > > +{
> > > > +       return idle_inject_register_full(cpumask, NULL);
> > > > +}
> > > >   EXPORT_SYMBOL_NS_GPL(idle_inject_register, IDLE_INJECT);
> > > 
> > > Why not just add the parameter to idle_inject_register() ?
> > > 
> > > There is only one user ATM
> > That was done in v1. But Rafael suggested to avoid changes to
> > existing
> > kernel source.
> 
> However, it can be done if Daniel prefers that.  It is 1 callback now
> only, so no big deal I suppose.
Do you have preference? I can submit an update.

Thanks,
Srinivas

[PATCH v3 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 callback.
- 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. Also for
protection during dynamic adjustment of run and idle time from
update() callback.
- Remove online/offline callbacks to designate control CPU

Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
---
v3:
- Use Update callback which is per device
- Remove use of control_cpu and online/offline callback to set this

v2:
- Use idle_inject_register_full instead of idle_inject_register
- Also fix dependency issue with POWERCAP config
Reported-by: kernel test robot <lkp@intel.com>

 drivers/thermal/intel/Kconfig            |   2 +
 drivers/thermal/intel/intel_powerclamp.c | 328 ++++++++---------------
 2 files changed, 119 insertions(+), 211 deletions(-)

diff --git a/drivers/thermal/intel/Kconfig b/drivers/thermal/intel/Kconfig
index f0c845679250..6c2a95f41c81 100644
--- a/drivers/thermal/intel/Kconfig
+++ b/drivers/thermal/intel/Kconfig
@@ -3,6 +3,8 @@ config INTEL_POWERCLAMP
 	tristate "Intel PowerClamp idle injection driver"
 	depends on X86
 	depends on CPU_SUP_INTEL
+	select POWERCAP
+	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..777476e3d1f9 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) 2012-2023, 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),
@@ -60,33 +54,24 @@ 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;
 
-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 +329,33 @@ 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 = 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;
 }
 
 /*
@@ -452,126 +391,116 @@ 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 bool idle_inject_begin(void)
 {
-	struct powerclamp_worker_data *w_data = per_cpu_ptr(worker_data, cpu);
-	struct kthread_worker *worker;
+	bool update;
 
-	worker = kthread_create_worker_on_cpu(cpu, 0, "kidle_inj/%ld", cpu);
-	if (IS_ERR(worker))
-		return;
+	mutex_lock(&powerclamp_lock);
 
-	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);
-}
+	if (!(powerclamp_data.count % powerclamp_data.window_size_now)) {
 
-static void stop_power_clamp_worker(unsigned long cpu)
-{
-	struct powerclamp_worker_data *w_data = per_cpu_ptr(worker_data, cpu);
+		should_skip = powerclamp_adjust_controls(powerclamp_data.target_ratio,
+							 powerclamp_data.guard,
+							 powerclamp_data.window_size_now);
+		update = true;
+	}
 
-	if (!w_data->worker)
-		return;
+	if (update || target_ratio_updated) {
+		unsigned int runtime;
 
-	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.
-	 */
-	clear_bit(w_data->cpu, cpu_clamping_mask);
-	kthread_destroy_worker(w_data->worker);
+		runtime = get_run_time();
+		idle_inject_set_duration(ii_dev, runtime, duration);
+		target_ratio_updated = false;
+	}
+
+	powerclamp_data.count++;
+
+	mutex_unlock(&powerclamp_lock);
+
+	if (should_skip)
+		return false;
 
-	w_data->worker = NULL;
+	return true;
 }
 
-static int start_power_clamp(void)
+static void trigger_idle_injection(void)
 {
-	unsigned long cpu;
+	unsigned int runtime = get_run_time();
 
-	set_target_ratio = clamp(set_target_ratio, 0U, MAX_TARGET_RATIO - 1);
-	/* prevent cpu hotplug */
-	cpus_read_lock();
+	idle_inject_set_duration(ii_dev, runtime, duration);
+	idle_inject_start(ii_dev);
+	powerclamp_data.clamping = true;
+}
 
-	/* prefer BSP */
-	control_cpu = cpumask_first(cpu_online_mask);
+static int powerclamp_idle_injection_register(void)
+{
+	static cpumask_t idle_injection_cpu_mask;
+	unsigned long cpu;
 
-	clamping = true;
-	schedule_delayed_work(&poll_pkg_cstate_work, 0);
+	/*
+	 * 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.
+	 */
+	for_each_present_cpu(cpu) {
+		cpumask_set_cpu(cpu, &idle_injection_cpu_mask);
+	}
 
-	/* start one kthread worker per online cpu */
-	for_each_online_cpu(cpu) {
-		start_power_clamp_worker(cpu);
+	ii_dev = idle_inject_register_full(&idle_injection_cpu_mask,
+					   idle_inject_begin);
+	if (!ii_dev) {
+		pr_err("powerclamp: idle_inject_register failed\n");
+		return -EAGAIN;
 	}
-	cpus_read_unlock();
+
+	idle_inject_set_duration(ii_dev, TICK_USEC, duration);
+	idle_inject_set_latency(ii_dev, UINT_MAX);
 
 	return 0;
 }
 
-static void end_power_clamp(void)
+static void remove_idle_injection(void)
 {
-	int i;
+	if (!powerclamp_data.clamping)
+		return;
 
-	/*
-	 * 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);
-	}
+	powerclamp_data.clamping = false;
+	idle_inject_stop(ii_dev);
 }
 
-static int powerclamp_cpu_online(unsigned int cpu)
+static int start_power_clamp(void)
 {
-	if (clamping == false)
-		return 0;
-	start_power_clamp_worker(cpu);
-	/* prefer BSP as controlling CPU */
-	if (cpu == 0) {
-		control_cpu = 0;
-		smp_mb();
+	int ret;
+
+	/* prevent cpu hotplug */
+	cpus_read_lock();
+
+	ret = powerclamp_idle_injection_register();
+	if (!ret) {
+		trigger_idle_injection();
+		schedule_delayed_work(&poll_pkg_cstate_work, 0);
 	}
-	return 0;
-}
 
-static int powerclamp_cpu_predown(unsigned int cpu)
-{
-	if (clamping == false)
-		return 0;
+	cpus_read_unlock();
 
-	stop_power_clamp_worker(cpu);
-	if (cpu != control_cpu)
-		return 0;
+	return ret;
+}
 
-	control_cpu = cpumask_first(cpu_online_mask);
-	if (control_cpu == cpu)
-		control_cpu = cpumask_next(cpu, cpu_online_mask);
-	smp_mb();
-	return 0;
+static void end_power_clamp(void)
+{
+	if (powerclamp_data.clamping) {
+		remove_idle_injection();
+		idle_inject_unregister(ii_dev);
+	}
 }
 
 static int powerclamp_get_max_state(struct thermal_cooling_device *cdev,
@@ -585,7 +514,7 @@ 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 +528,30 @@ 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 (!set_target_ratio && new_target_ratio > 0) {
 		pr_info("Start idle injection to reduce power\n");
 		set_target_ratio = new_target_ratio;
 		ret = start_power_clamp();
+		if (ret)
+			set_target_ratio = 0;
 		goto exit_set;
 	} else	if (set_target_ratio > 0 && new_target_ratio == 0) {
 		pr_info("Stop forced idle injection\n");
 		end_power_clamp();
 		set_target_ratio = 0;
+		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();
+		target_ratio_updated= true;
 	}
 
 exit_set:
+	mutex_unlock(&powerclamp_lock);
+
 	return ret;
 }
 
@@ -657,7 +592,6 @@ static int powerclamp_debug_show(struct seq_file *m, void *unused)
 {
 	int i = 0;
 
-	seq_printf(m, "controlling cpu: %d\n", control_cpu);
 	seq_printf(m, "pct confidence steady dynamic (compensation)\n");
 	for (i = 0; i < MAX_TARGET_RATIO; i++) {
 		seq_printf(m, "%d\t%lu\t%lu\t%lu\n",
@@ -680,75 +614,47 @@ static inline void powerclamp_create_debug_files(void)
 			    &powerclamp_debug_fops);
 }
 
-static enum cpuhp_state hp_state;
-
 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;
-	retval = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN,
-					   "thermal/intel_powerclamp:online",
-					   powerclamp_cpu_online,
-					   powerclamp_cpu_predown);
-	if (retval < 0)
-		goto exit_free;
-
-	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);
-	if (IS_ERR(cooling_dev)) {
-		retval = -ENODEV;
-		goto exit_free_thread;
-	}
+						      &powerclamp_cooling_ops);
+	if (IS_ERR(cooling_dev))
+		return -ENODEV;
 
 	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();
-	cpuhp_remove_state_nocalls(hp_state);
-	free_percpu(worker_data);
+	mutex_unlock(&powerclamp_lock);
+
 	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>");
-- 
2.31.1

Re: [PATCH v3 3/4] thermal/drivers/intel_powerclamp: Use powercap idle-inject framework

[Daniel Lezcano]

On 17/01/2023 03:07, Srinivas Pandruvada wrote:

[ ... ]

> diff --git a/drivers/thermal/intel/Kconfig b/drivers/thermal/intel/Kconfig
> index f0c845679250..6c2a95f41c81 100644
> --- a/drivers/thermal/intel/Kconfig
> +++ b/drivers/thermal/intel/Kconfig
> @@ -3,6 +3,8 @@ config INTEL_POWERCLAMP
>   	tristate "Intel PowerClamp idle injection driver"
>   	depends on X86
>   	depends on CPU_SUP_INTEL
> +	select POWERCAP
> +	select IDLE_INJECT

Just a question, not a comment. As IDLE_INJECT depends on POWERCAP, is 
it necessary to select POWERCAP also here ?


-- 
<http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs

Follow Linaro:  <http://www.facebook.com/pages/Linaro> Facebook |
<http://twitter.com/#!/linaroorg> Twitter |
<http://www.linaro.org/linaro-blog/> Blog

Re: [PATCH v3 3/4] thermal/drivers/intel_powerclamp: Use powercap idle-inject framework

[Rafael J. Wysocki]

On Tue, Jan 17, 2023 at 12:16 PM Daniel Lezcano
<daniel.lezcano@linaro.org> wrote:
>
> On 17/01/2023 03:07, Srinivas Pandruvada wrote:
>
> [ ... ]
>
> > diff --git a/drivers/thermal/intel/Kconfig b/drivers/thermal/intel/Kconfig
> > index f0c845679250..6c2a95f41c81 100644
> > --- a/drivers/thermal/intel/Kconfig
> > +++ b/drivers/thermal/intel/Kconfig
> > @@ -3,6 +3,8 @@ config INTEL_POWERCLAMP
> >       tristate "Intel PowerClamp idle injection driver"
> >       depends on X86
> >       depends on CPU_SUP_INTEL
> > +     select POWERCAP
> > +     select IDLE_INJECT
>
> Just a question, not a comment. As IDLE_INJECT depends on POWERCAP, is
> it necessary to select POWERCAP also here ?

Yes, it is.

If IDLE_INJECT alone is selected, POWERCAP may not be and there will
be a broken dependency.

Re: [PATCH v3 3/4] thermal/drivers/intel_powerclamp: Use powercap idle-inject framework

[Daniel Lezcano]

On 17/01/2023 15:49, Rafael J. Wysocki wrote:
> On Tue, Jan 17, 2023 at 12:16 PM Daniel Lezcano
> <daniel.lezcano@linaro.org> wrote:
>>
>> On 17/01/2023 03:07, Srinivas Pandruvada wrote:
>>
>> [ ... ]
>>
>>> diff --git a/drivers/thermal/intel/Kconfig b/drivers/thermal/intel/Kconfig
>>> index f0c845679250..6c2a95f41c81 100644
>>> --- a/drivers/thermal/intel/Kconfig
>>> +++ b/drivers/thermal/intel/Kconfig
>>> @@ -3,6 +3,8 @@ config INTEL_POWERCLAMP
>>>        tristate "Intel PowerClamp idle injection driver"
>>>        depends on X86
>>>        depends on CPU_SUP_INTEL
>>> +     select POWERCAP
>>> +     select IDLE_INJECT
>>
>> Just a question, not a comment. As IDLE_INJECT depends on POWERCAP, is
>> it necessary to select POWERCAP also here ?
> 
> Yes, it is.
> 
> If IDLE_INJECT alone is selected, POWERCAP may not be and there will
> be a broken dependency.

Ok, thanks

-- 
<http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs

Follow Linaro:  <http://www.facebook.com/pages/Linaro> Facebook |
<http://twitter.com/#!/linaroorg> Twitter |
<http://www.linaro.org/linaro-blog/> Blog

[PATCH v3 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>
---
v3:
Suggestion from Daniel:
- Remove unused cpuidle_cpu_mask
- Why not register idle_inject here before calling 
thermal_cooling_device_register() so you get ride of the lock?
We can't as cooling device is registered during CPU online. But if you also register idle,
 then every CPU is registered with idle inject core. So intel_powerclamp can't
use those CPUs to register again.
Here the logic is only register when user actually changing the state of cooling
device from 0 to 1. In this way either per CPU or all cpu (intel_powerclamp)
can be used one at a time.

Reuse cpuidle_cooling_device.c. The RFC tried to do that, But you don't
save lots of reuse. The of_device separation and separate idle_injection
register will need a separate function.
To be clean implementation, added a new module. This module adds very few lines
of new code.

v2:
- Removed callback arguments for idle_inject_register

 drivers/thermal/intel/Kconfig                 |  10 +
 drivers/thermal/intel/Makefile                |   1 +
 .../thermal/intel/intel_cpu_idle_cooling.c    | 259 ++++++++++++++++++
 3 files changed, 270 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 6c2a95f41c81..8c88d6e18414 100644
--- a/drivers/thermal/intel/Kconfig
+++ b/drivers/thermal/intel/Kconfig
@@ -115,3 +115,13 @@ 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..f157018ffb63
--- /dev/null
+++ b/drivers/thermal/intel/intel_cpu_idle_cooling.c
@@ -0,0 +1,259 @@
+// 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);
+
+/* 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));
+	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;
+	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");
-- 
2.31.1