Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
From:   "Rafael J. Wysocki" <rjw@rjwysocki.net>
To:     Daniel Lezcano <daniel.lezcano@linaro.org>
Cc:     mathieu.poirier@linaro.org, mingo@redhat.com, peterz@infradead.org,
        linux-pm@vger.kernel.org, linux-kernel@vger.kernel.org,
        ulf.hansson@linaro.org
Subject: Re: [PATCH V6 1/3] cpuidle: play_idle: Make play_idle more flexible
Date:   Fri, 08 Nov 2019 02:03:28 +0100
Message-ID: <2789916.f0jx01YxDY@kreacher>
In-Reply-To: <20191030075141.1039-1-daniel.lezcano@linaro.org>
References: <20191030075141.1039-1-daniel.lezcano@linaro.org>
MIME-Version: 1.0
Content-Transfer-Encoding: 7Bit
Content-Type: text/plain; charset="us-ascii"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Wednesday, October 30, 2019 8:51:39 AM CET Daniel Lezcano wrote:
> The play_idle function has two users, the intel powerclamp and the
> idle_injection.
> 
> The idle injection cooling device uses the function via the
> idle_injection powercap's APIs. Unfortunately, play_idle is currently
> limited by the idle state depth: by default the deepest idle state is
> selected. On the ARM[64] platforms, most of the time it is the cluster
> idle state, the exit latency and the residency can be very high. That
> reduces the scope of the idle injection usage because the impact on
> the performances can be very significant.
> 
> If the idle injection cycles can be done with a shallow state like a
> retention state, the cooling effect would eventually give similar
> results than the cpufreq cooling device.
> 
> In order to prepare the function to receive an idle state parameter,
> let's replace the 'use_deepest_state' boolean field with 'use_state'
> and use this value to enter the specific idle state.
> 
> The current code keeps the default behavior which is go to the deepest
> idle state.
> 
> Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>
> Acked-by: Mathieu Poirier <mathieu.poirier@linaro.org>
> Reviewed-by: Ulf Hansson <ulf.hansson@linaro.org>
> ---
>   V6:
>    - Change use_state variable name by use_state_idx:
>      https://lkml.org/lkml/2019/10/28/874
>   V5:
>    - Fix s2idle default idle state value:
>      https://lkml.org/lkml/2019/10/15/522
> ---
>  drivers/cpuidle/cpuidle.c | 21 +++++++++++----------
>  include/linux/cpuidle.h   | 13 ++++++-------
>  kernel/sched/idle.c       | 11 +++++++----
>  3 files changed, 24 insertions(+), 21 deletions(-)
> 
> diff --git a/drivers/cpuidle/cpuidle.c b/drivers/cpuidle/cpuidle.c
> index 0895b988fa92..18523ea6b11b 100644
> --- a/drivers/cpuidle/cpuidle.c
> +++ b/drivers/cpuidle/cpuidle.c
> @@ -99,31 +99,31 @@ static int find_deepest_state(struct cpuidle_driver *drv,
>  }
>  
>  /**
> - * cpuidle_use_deepest_state - Set/clear governor override flag.
> - * @enable: New value of the flag.
> + * cpuidle_use_state - Force the cpuidle framework to enter an idle state.
> + * @state: An integer for an idle state
>   *
> - * Set/unset the current CPU to use the deepest idle state (override governors
> - * going forward if set).
> + * Specify an idle state the cpuidle framework must step in and bypass
> + * the idle state selection process.
>   */
> -void cpuidle_use_deepest_state(bool enable)
> +void cpuidle_use_state(int index)
>  {
>  	struct cpuidle_device *dev;
>  
>  	preempt_disable();
>  	dev = cpuidle_get_device();
>  	if (dev)
> -		dev->use_deepest_state = enable;
> +		dev->use_state_idx = index;
>  	preempt_enable();
>  }
>  
>  /**
>   * cpuidle_find_deepest_state - Find the deepest available idle state.
> - * @drv: cpuidle driver for the given CPU.
> - * @dev: cpuidle device for the given CPU.
>   */
> -int cpuidle_find_deepest_state(struct cpuidle_driver *drv,
> -			       struct cpuidle_device *dev)
> +int cpuidle_find_deepest_state(void)
>  {
> +	struct cpuidle_device *dev = cpuidle_get_device();
> +	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);

This is totally wasteful for the cpuidle_idle_call() use case.

> +
>  	return find_deepest_state(drv, dev, UINT_MAX, 0, false);
>  }
>  
> @@ -554,6 +554,7 @@ static void __cpuidle_unregister_device(struct cpuidle_device *dev)
>  static void __cpuidle_device_init(struct cpuidle_device *dev)
>  {
>  	memset(dev->states_usage, 0, sizeof(dev->states_usage));
> +	dev->use_state_idx = CPUIDLE_STATE_NOUSE;
>  	dev->last_residency = 0;
>  	dev->next_hrtimer = 0;
>  }
> diff --git a/include/linux/cpuidle.h b/include/linux/cpuidle.h
> index 4b6b5bea8f79..d53ad36cb2de 100644
> --- a/include/linux/cpuidle.h
> +++ b/include/linux/cpuidle.h
> @@ -15,6 +15,7 @@
>  #include <linux/list.h>
>  #include <linux/hrtimer.h>
>  
> +#define CPUIDLE_STATE_NOUSE	-1
>  #define CPUIDLE_STATE_MAX	10
>  #define CPUIDLE_NAME_LEN	16
>  #define CPUIDLE_DESC_LEN	32
> @@ -80,11 +81,11 @@ struct cpuidle_driver_kobj;
>  struct cpuidle_device {
>  	unsigned int		registered:1;
>  	unsigned int		enabled:1;
> -	unsigned int		use_deepest_state:1;
>  	unsigned int		poll_time_limit:1;
>  	unsigned int		cpu;
>  	ktime_t			next_hrtimer;
>  
> +	int			use_state_idx;
>  	int			last_state_idx;
>  	int			last_residency;
>  	u64			poll_limit_ns;
> @@ -203,19 +204,17 @@ static inline struct cpuidle_device *cpuidle_get_device(void) {return NULL; }
>  #endif
>  
>  #ifdef CONFIG_CPU_IDLE
> -extern int cpuidle_find_deepest_state(struct cpuidle_driver *drv,
> -				      struct cpuidle_device *dev);
> +extern int cpuidle_find_deepest_state(void);
>  extern int cpuidle_enter_s2idle(struct cpuidle_driver *drv,
>  				struct cpuidle_device *dev);
> -extern void cpuidle_use_deepest_state(bool enable);
> +extern void cpuidle_use_state(int index);
>  #else
> -static inline int cpuidle_find_deepest_state(struct cpuidle_driver *drv,
> -					     struct cpuidle_device *dev)
> +static inline int cpuidle_find_deepest_state(void)
>  {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 void cpuidle_use_state(int index)
>  {
>  }
>  #endif
> diff --git a/kernel/sched/idle.c b/kernel/sched/idle.c
> index 8dad5aa600ea..fb9fc93f1497 100644
> --- a/kernel/sched/idle.c
> +++ b/kernel/sched/idle.c
> @@ -165,7 +165,8 @@ 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() ||
> +	    dev->use_state_idx != CPUIDLE_STATE_NOUSE) {
>  		if (idle_should_enter_s2idle()) {
>  			rcu_idle_enter();
>  
> @@ -176,12 +177,14 @@ static void cpuidle_idle_call(void)
>  			}
>  
>  			rcu_idle_exit();
> +			next_state = cpuidle_find_deepest_state();
> +		} else {
> +			next_state = dev->use_state_idx;
>  		}
>  
>  		tick_nohz_idle_stop_tick();
>  		rcu_idle_enter();
>  
> -		next_state = cpuidle_find_deepest_state(drv, dev);
>  		call_cpuidle(drv, dev, next_state);
>  	} else {
>  		bool stop_tick = true;
> @@ -328,7 +331,7 @@ void play_idle(unsigned long duration_us)
>  	rcu_sleep_check();
>  	preempt_disable();
>  	current->flags |= PF_IDLE;
> -	cpuidle_use_deepest_state(true);
> +	cpuidle_use_state(cpuidle_find_deepest_state());

And this assumes that the deepest state will not change for the whole
play_idle() duration, but what if it is disabled by user space in the
meantime?

>  
>  	it.done = 0;
>  	hrtimer_init_on_stack(&it.timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
> @@ -339,7 +342,7 @@ void play_idle(unsigned long duration_us)
>  	while (!READ_ONCE(it.done))
>  		do_idle();
>  
> -	cpuidle_use_deepest_state(false);
> +	cpuidle_use_state(CPUIDLE_STATE_NOUSE);
>  	current->flags &= ~PF_IDLE;
>  
>  	preempt_fold_need_resched();
>