RE: [PATCH] cpuidle: Add "cpuidle.use_deepest" to bypass governor and allow HW to go deep

RE: [PATCH] cpuidle: Add "cpuidle.use_deepest" to bypass governor and allow HW to go deep

[Doug Smythies]

Hi Len,

Other feedback notwithstanding, works fine for me.
Just a small typo type comment:

On 2017.11.08 23:39 Len Brown wrote:

...[snip]...

> Here we create the "cpuidle.use_deepest" modparam to provide this capability.
> 
> "cpuidle.use_deepest=Y" can be set at boot-time, and
> /sys/module/cpuidle/use_deepest can be modified (with Y/N) at run-time.

Should be:

/sys/module/cpuidle/parameters/use_deepest

...[snip]...

RE: [PATCH] cpuidle: Add "cpuidle.use_deepest" to bypass governor and allow HW to go deep

[Doug Smythies]

On 2017.11.16 08:11 Thomas Ilsche wrote:
> On 2017-11-09 08:38, Len Brown wrote:
>> From: Len Brown <len.brown@intel.com>
>> 
>> While there are several mechanisms (cmdline, sysfs, PM_QOS) to limit
>> cpuidle to shallow idle states, there is no simple mechanism
>> to give the hardware permission to enter the deeptest state permitted by PM_QOS.
>> 
>> Here we create the "cpuidle.use_deepest" modparam to provide this capability.
>> 
>> "cpuidle.use_deepest=Y" can be set at boot-time, and
>> /sys/module/cpuidle/use_deepest can be modified (with Y/N) at run-time.
>
> This is a good option to have and can conveniently prevent idle power consumption
> issues. But that wouldn't be a reasonable default, would it?
> I still think there is an inherent need for a heuristic and a corresponding
> mitigation to avoid staying in a sleep state too long.

I think so, yes. The potential energy savings, depending on work flow, are significant.
For the idle state 0 case, I have some supporting data that I will send on
the other e-mail thread.

... Doug

Re: [PATCH] cpuidle: Add "cpuidle.use_deepest" to bypass governor and allow HW to go deep

[Thomas Ilsche]

On 2017-11-09 08:38, Len Brown wrote:
> From: Len Brown <len.brown@intel.com>
> 
> While there are several mechanisms (cmdline, sysfs, PM_QOS) to limit
> cpuidle to shallow idle states, there is no simple mechanism
> to give the hardware permission to enter the deeptest state permitted by PM_QOS.
> 
> Here we create the "cpuidle.use_deepest" modparam to provide this capability.
> 
> "cpuidle.use_deepest=Y" can be set at boot-time, and
> /sys/module/cpuidle/use_deepest can be modified (with Y/N) at run-time.

This is a good option to have and can conveniently prevent idle power consumption
issues. But that wouldn't be a reasonable default, would it?
I still think there is an inherent need for a heuristic and a corresponding
mitigation to avoid staying in a sleep state too long.

Best,
Thomas

Re: [PATCH] cpuidle: Add "cpuidle.use_deepest" to bypass governor and allow HW to go deep

[Vincent Guittot]

Hi Len

On 9 November 2017 at 08:38, Len Brown <lenb@kernel.org> wrote:
> From: Len Brown <len.brown@intel.com>
>
> While there are several mechanisms (cmdline, sysfs, PM_QOS) to limit
> cpuidle to shallow idle states, there is no simple mechanism
> to give the hardware permission to enter the deeptest state permitted by PM_QOS.

and by per device resume latency QoS

>
> Here we create the "cpuidle.use_deepest" modparam to provide this capability.
>
> "cpuidle.use_deepest=Y" can be set at boot-time, and
> /sys/module/cpuidle/use_deepest can be modified (with Y/N) at run-time.
>
> n.b.
>
> Within the constraints of PM_QOS, this mechanism gives the hardware
> permission to choose the deeptest power savings and highest latency
> state available.  And so choice will depend on the particular hardware.
>
> Also, if PM_QOS is not informed of latency constraints, it can't help.
> In that case, using this mechanism may result in entering high-latency
> states that impact performance.
>
> Signed-off-by: Len Brown <len.brown@intel.com>
> ---
>  Documentation/admin-guide/kernel-parameters.txt |  4 ++++
>  drivers/cpuidle/cpuidle.c                       | 19 ++++++++++++++++
>  include/linux/cpuidle.h                         |  7 ++++++
>  kernel/sched/idle.c                             | 30 +++++++++++++++++++------
>  4 files changed, 53 insertions(+), 7 deletions(-)
>
> diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
> index 05496622b4ef..20f70de688bf 100644
> --- a/Documentation/admin-guide/kernel-parameters.txt
> +++ b/Documentation/admin-guide/kernel-parameters.txt
> @@ -659,6 +659,10 @@
>         cpuidle.off=1   [CPU_IDLE]
>                         disable the cpuidle sub-system
>
> +       cpuidle.use_deepest=Y   [CPU_IDLE]
> +                       Ignore cpuidle governor, always choose deepest
> +                       PM_QOS-legal CPU idle power saving state.
> +
>         cpufreq.off=1   [CPU_FREQ]
>                         disable the cpufreq sub-system
>
> diff --git a/drivers/cpuidle/cpuidle.c b/drivers/cpuidle/cpuidle.c
> index 484cc8909d5c..afee5aab7719 100644
> --- a/drivers/cpuidle/cpuidle.c
> +++ b/drivers/cpuidle/cpuidle.c
> @@ -34,6 +34,7 @@ LIST_HEAD(cpuidle_detected_devices);
>
>  static int enabled_devices;
>  static int off __read_mostly;
> +static bool use_deepest __read_mostly;
>  static int initialized __read_mostly;
>
>  int cpuidle_disabled(void)
> @@ -116,6 +117,10 @@ void cpuidle_use_deepest_state(bool enable)
>         preempt_enable();
>  }
>
> +bool cpuidle_using_deepest_state(void)
> +{
> +       return use_deepest;
> +}
>  /**
>   * cpuidle_find_deepest_state - Find the deepest available idle state.
>   * @drv: cpuidle driver for the given CPU.
> @@ -127,6 +132,19 @@ int cpuidle_find_deepest_state(struct cpuidle_driver *drv,
>         return find_deepest_state(drv, dev, UINT_MAX, 0, false);
>  }
>
> +/**
> + * cpuidle_find_deepest_state_qos - Find the deepest available idle state.
> + * @drv: cpuidle driver for the given CPU.
> + * @dev: cpuidle device for the given CPU.
> + * Honors PM_QOS
> + */
> +int cpuidle_find_deepest_state_qos(struct cpuidle_driver *drv,
> +                              struct cpuidle_device *dev)
> +{
> +       return find_deepest_state(drv, dev,
> +                       pm_qos_request(PM_QOS_CPU_DMA_LATENCY), 0, false);

You should also take into account per device latency  like in menu governor:
dev_pm_qos_raw_read_value(device);


> +}
> +
>  #ifdef CONFIG_SUSPEND
>  static void enter_s2idle_proper(struct cpuidle_driver *drv,
>                                 struct cpuidle_device *dev, int index)
> @@ -681,4 +699,5 @@ static int __init cpuidle_init(void)
>  }
>
>  module_param(off, int, 0444);
> +module_param(use_deepest, bool, 0644);
>  core_initcall(cpuidle_init);
> diff --git a/include/linux/cpuidle.h b/include/linux/cpuidle.h
> index 8f7788d23b57..e3c2c9d1898f 100644
> --- a/include/linux/cpuidle.h
> +++ b/include/linux/cpuidle.h
> @@ -198,19 +198,26 @@ static inline struct cpuidle_device *cpuidle_get_device(void) {return NULL; }
>  #ifdef CONFIG_CPU_IDLE
>  extern int cpuidle_find_deepest_state(struct cpuidle_driver *drv,
>                                       struct cpuidle_device *dev);
> +extern int cpuidle_find_deepest_state_qos(struct cpuidle_driver *drv,
> +                                     struct cpuidle_device *dev);
>  extern int cpuidle_enter_s2idle(struct cpuidle_driver *drv,
>                                 struct cpuidle_device *dev);
>  extern void cpuidle_use_deepest_state(bool enable);
> +extern bool cpuidle_using_deepest_state(void);
>  #else
>  static inline int cpuidle_find_deepest_state(struct cpuidle_driver *drv,
>                                              struct cpuidle_device *dev)
>  {return -ENODEV; }
> +static inline int cpuidle_find_deepest_state_qos(struct cpuidle_driver *drv,
> +                                            struct cpuidle_device *dev)
> +{return -ENODEV; }
>  static inline int cpuidle_enter_s2idle(struct cpuidle_driver *drv,
>                                        struct cpuidle_device *dev)
>  {return -ENODEV; }
>  static inline void cpuidle_use_deepest_state(bool enable)
>  {
>  }
> +static inline bool cpuidle_using_deepest_state(void) {return false; }
>  #endif
>
>  /* kernel/sched/idle.c */
> diff --git a/kernel/sched/idle.c b/kernel/sched/idle.c
> index 257f4f0b4532..6c7348ae28ec 100644
> --- a/kernel/sched/idle.c
> +++ b/kernel/sched/idle.c
> @@ -167,17 +167,33 @@ static void cpuidle_idle_call(void)
>          * until a proper wakeup interrupt happens.
>          */
>
> -       if (idle_should_enter_s2idle() || dev->use_deepest_state) {
> -               if (idle_should_enter_s2idle()) {
> -                       entered_state = cpuidle_enter_s2idle(drv, dev);
> -                       if (entered_state > 0) {
> -                               local_irq_enable();
> -                               goto exit_idle;
> -                       }
> +       if (idle_should_enter_s2idle()) {
> +               entered_state = cpuidle_enter_s2idle(drv, dev);
> +               if (entered_state > 0) {
> +                       local_irq_enable();
> +                       goto exit_idle;
>                 }
> +       }
>
> +       /*
> +        * cpuidle_dev->user_deepest_state is per-device, and thus per-CPU.
> +        * it is used to force power-savings, and thus does not obey PM_QOS.
> +        */
> +
> +       if (dev->use_deepest_state) {
>                 next_state = cpuidle_find_deepest_state(drv, dev);
>                 call_cpuidle(drv, dev, next_state);
> +       }
> +
> +       /*
> +        * cpuidle_using_deepest_state() is system-wide, applying to all CPUs.
> +        * When activated, Linux gives the hardware permission to go as deep
> +        * as PM_QOS allows.
> +        */
> +
> +       else if (cpuidle_using_deepest_state()) {
> +               next_state = cpuidle_find_deepest_state_qos(drv, dev);
> +               call_cpuidle(drv, dev, next_state);
>         } else {
>                 /*
>                  * Ask the cpuidle framework to choose a convenient idle state.
> --
> 2.14.0-rc0
>

[PATCH] cpuidle: Add "cpuidle.use_deepest" to bypass governor and allow HW to go deep

[Len Brown]

From: Len Brown <len.brown@intel.com>

While there are several mechanisms (cmdline, sysfs, PM_QOS) to limit
cpuidle to shallow idle states, there is no simple mechanism
to give the hardware permission to enter the deeptest state permitted by PM_QOS.

Here we create the "cpuidle.use_deepest" modparam to provide this capability.

"cpuidle.use_deepest=Y" can be set at boot-time, and
/sys/module/cpuidle/use_deepest can be modified (with Y/N) at run-time.

n.b.

Within the constraints of PM_QOS, this mechanism gives the hardware
permission to choose the deeptest power savings and highest latency
state available.  And so choice will depend on the particular hardware.

Also, if PM_QOS is not informed of latency constraints, it can't help.
In that case, using this mechanism may result in entering high-latency
states that impact performance.

Signed-off-by: Len Brown <len.brown@intel.com>
---
 Documentation/admin-guide/kernel-parameters.txt |  4 ++++
 drivers/cpuidle/cpuidle.c                       | 19 ++++++++++++++++
 include/linux/cpuidle.h                         |  7 ++++++
 kernel/sched/idle.c                             | 30 +++++++++++++++++++------
 4 files changed, 53 insertions(+), 7 deletions(-)

diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
index 05496622b4ef..20f70de688bf 100644
--- a/Documentation/admin-guide/kernel-parameters.txt
+++ b/Documentation/admin-guide/kernel-parameters.txt
@@ -659,6 +659,10 @@
 	cpuidle.off=1	[CPU_IDLE]
 			disable the cpuidle sub-system
 
+	cpuidle.use_deepest=Y	[CPU_IDLE]
+			Ignore cpuidle governor, always choose deepest
+			PM_QOS-legal CPU idle power saving state.
+
 	cpufreq.off=1	[CPU_FREQ]
 			disable the cpufreq sub-system
 
diff --git a/drivers/cpuidle/cpuidle.c b/drivers/cpuidle/cpuidle.c
index 484cc8909d5c..afee5aab7719 100644
--- a/drivers/cpuidle/cpuidle.c
+++ b/drivers/cpuidle/cpuidle.c
@@ -34,6 +34,7 @@ LIST_HEAD(cpuidle_detected_devices);
 
 static int enabled_devices;
 static int off __read_mostly;
+static bool use_deepest __read_mostly;
 static int initialized __read_mostly;
 
 int cpuidle_disabled(void)
@@ -116,6 +117,10 @@ void cpuidle_use_deepest_state(bool enable)
 	preempt_enable();
 }
 
+bool cpuidle_using_deepest_state(void)
+{
+	return use_deepest;
+}
 /**
  * cpuidle_find_deepest_state - Find the deepest available idle state.
  * @drv: cpuidle driver for the given CPU.
@@ -127,6 +132,19 @@ int cpuidle_find_deepest_state(struct cpuidle_driver *drv,
 	return find_deepest_state(drv, dev, UINT_MAX, 0, false);
 }
 
+/**
+ * cpuidle_find_deepest_state_qos - Find the deepest available idle state.
+ * @drv: cpuidle driver for the given CPU.
+ * @dev: cpuidle device for the given CPU.
+ * Honors PM_QOS
+ */
+int cpuidle_find_deepest_state_qos(struct cpuidle_driver *drv,
+			       struct cpuidle_device *dev)
+{
+	return find_deepest_state(drv, dev,
+			pm_qos_request(PM_QOS_CPU_DMA_LATENCY), 0, false);
+}
+
 #ifdef CONFIG_SUSPEND
 static void enter_s2idle_proper(struct cpuidle_driver *drv,
 				struct cpuidle_device *dev, int index)
@@ -681,4 +699,5 @@ static int __init cpuidle_init(void)
 }
 
 module_param(off, int, 0444);
+module_param(use_deepest, bool, 0644);
 core_initcall(cpuidle_init);
diff --git a/include/linux/cpuidle.h b/include/linux/cpuidle.h
index 8f7788d23b57..e3c2c9d1898f 100644
--- a/include/linux/cpuidle.h
+++ b/include/linux/cpuidle.h
@@ -198,19 +198,26 @@ static inline struct cpuidle_device *cpuidle_get_device(void) {return NULL; }
 #ifdef CONFIG_CPU_IDLE
 extern int cpuidle_find_deepest_state(struct cpuidle_driver *drv,
 				      struct cpuidle_device *dev);
+extern int cpuidle_find_deepest_state_qos(struct cpuidle_driver *drv,
+				      struct cpuidle_device *dev);
 extern int cpuidle_enter_s2idle(struct cpuidle_driver *drv,
 				struct cpuidle_device *dev);
 extern void cpuidle_use_deepest_state(bool enable);
+extern bool cpuidle_using_deepest_state(void);
 #else
 static inline int cpuidle_find_deepest_state(struct cpuidle_driver *drv,
 					     struct cpuidle_device *dev)
 {return -ENODEV; }
+static inline int cpuidle_find_deepest_state_qos(struct cpuidle_driver *drv,
+					     struct cpuidle_device *dev)
+{return -ENODEV; }
 static inline int cpuidle_enter_s2idle(struct cpuidle_driver *drv,
 				       struct cpuidle_device *dev)
 {return -ENODEV; }
 static inline void cpuidle_use_deepest_state(bool enable)
 {
 }
+static inline bool cpuidle_using_deepest_state(void) {return false; }
 #endif
 
 /* kernel/sched/idle.c */
diff --git a/kernel/sched/idle.c b/kernel/sched/idle.c
index 257f4f0b4532..6c7348ae28ec 100644
--- a/kernel/sched/idle.c
+++ b/kernel/sched/idle.c
@@ -167,17 +167,33 @@ static void cpuidle_idle_call(void)
 	 * until a proper wakeup interrupt happens.
 	 */
 
-	if (idle_should_enter_s2idle() || dev->use_deepest_state) {
-		if (idle_should_enter_s2idle()) {
-			entered_state = cpuidle_enter_s2idle(drv, dev);
-			if (entered_state > 0) {
-				local_irq_enable();
-				goto exit_idle;
-			}
+	if (idle_should_enter_s2idle()) {
+		entered_state = cpuidle_enter_s2idle(drv, dev);
+		if (entered_state > 0) {
+			local_irq_enable();
+			goto exit_idle;
 		}
+	}
 
+	/*
+	 * cpuidle_dev->user_deepest_state is per-device, and thus per-CPU.
+	 * it is used to force power-savings, and thus does not obey PM_QOS.
+	 */
+
+	if (dev->use_deepest_state) {
 		next_state = cpuidle_find_deepest_state(drv, dev);
 		call_cpuidle(drv, dev, next_state);
+	}
+
+	/*
+	 * cpuidle_using_deepest_state() is system-wide, applying to all CPUs.
+	 * When activated, Linux gives the hardware permission to go as deep
+	 * as PM_QOS allows.
+	 */
+
+	else if (cpuidle_using_deepest_state()) {
+		next_state = cpuidle_find_deepest_state_qos(drv, dev);
+		call_cpuidle(drv, dev, next_state);
 	} else {
 		/*
 		 * Ask the cpuidle framework to choose a convenient idle state.
-- 
2.14.0-rc0