linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Daniel Lezcano <daniel.lezcano@linaro.org>
To: Ulf Hansson <ulf.hansson@linaro.org>
Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net>,
	Mathieu Poirier <mathieu.poirier@linaro.org>,
	Ingo Molnar <mingo@redhat.com>,
	Peter Zijlstra <peterz@infradead.org>,
	"open list:CPU IDLE TIME MANAGEMENT FRAMEWORK" 
	<linux-pm@vger.kernel.org>,
	open list <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH V4 1/3] cpuidle: play_idle: Make play_idle more flexible
Date: Wed, 16 Oct 2019 18:41:25 +0200	[thread overview]
Message-ID: <762e7966-a276-cca2-7baa-fa22c26f25f3@linaro.org> (raw)
In-Reply-To: <CAPDyKFo4WE4trWhxcxDcp1S+uPb6OJBY=-+fYht9k_4DpuQrNQ@mail.gmail.com>

On 15/10/2019 15:06, Ulf Hansson wrote:
> On Fri, 4 Oct 2019 at 10:32, Daniel Lezcano <daniel.lezcano@linaro.org> 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>
>> ---
>>  drivers/cpuidle/cpuidle.c | 21 +++++++++++----------
>>  include/linux/cpuidle.h   | 13 ++++++-------
>>  kernel/sched/idle.c       | 10 +++++-----
>>  3 files changed, 22 insertions(+), 22 deletions(-)
>>
>> diff --git a/drivers/cpuidle/cpuidle.c b/drivers/cpuidle/cpuidle.c
>> index 0895b988fa92..f8b54f277589 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 state)
>>  {
>>         struct cpuidle_device *dev;
>>
>>         preempt_disable();
>>         dev = cpuidle_get_device();
>>         if (dev)
>> -               dev->use_deepest_state = enable;
>> +               dev->use_state = state;
>>         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);
>> +
>>         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 = CPUIDLE_STATE_NOUSE;
>>         dev->last_residency = 0;
>>         dev->next_hrtimer = 0;
>>  }
>> diff --git a/include/linux/cpuidle.h b/include/linux/cpuidle.h
>> index 2dc4c6b19c25..ba0751b26e37 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;
>>         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 state);
>>  #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 state)
>>  {
>>  }
>>  #endif
>> diff --git a/kernel/sched/idle.c b/kernel/sched/idle.c
>> index b98283fc6914..17da9cb309e1 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 != CPUIDLE_STATE_NOUSE) {
>>                 if (idle_should_enter_s2idle()) {
>>                         rcu_idle_enter();
>>
>> @@ -181,8 +182,7 @@ static void cpuidle_idle_call(void)
>>                 tick_nohz_idle_stop_tick();
>>                 rcu_idle_enter();
>>
>> -               next_state = cpuidle_find_deepest_state(drv, dev);
>> -               call_cpuidle(drv, dev, next_state);
>> +               call_cpuidle(drv, dev, dev->use_state);
> 
> This doesn't look correct to me.
> 
> More precisely, in the s2idle case, we always pick the deepest state.
> But from the change above, we would invoke call_cpuidle() with
> CPUIDLE_STATE_NOUSE, right!?

There is the condition 'if (idle_should_enter_s2idle())' which gives
another path.

But that is true if the call fails with a negative or zero value for the
entered state we fallback to call_cpuidle with CPUIDLE_STATE_NOUSE. This
context does not appear in the patch.

Perhaps, something like (on top of this patch):

diff --git a/kernel/sched/idle.c b/kernel/sched/idle.c
index 6bcea3c06ebe..fc7f5216b579 100644
--- a/kernel/sched/idle.c
+++ b/kernel/sched/idle.c
@@ -177,12 +177,15 @@ static void cpuidle_idle_call(void)
                        }

                        rcu_idle_exit();
+                       next_state = cpuidle_find_deepest_state();
+               } else {
+                       next_state = dev->use_state;
                }

                tick_nohz_idle_stop_tick();
                rcu_idle_enter();

-               call_cpuidle(drv, dev, dev->use_state);
+               call_cpuidle(drv, dev, next_state);
        } else {
                bool stop_tick = true;



> 
>>         } else {
>>                 bool stop_tick = true;
>>
>> @@ -328,7 +328,7 @@ 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());
>>
>>         it.done = 0;
>>         hrtimer_init_on_stack(&it.timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
>> @@ -339,7 +339,7 @@ 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();
>> --
>> 2.17.1
>>
> 
> Kind regards
> Uffe
> 


-- 
 <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


      reply	other threads:[~2019-10-16 16:41 UTC|newest]

Thread overview: 6+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2019-10-04  8:32 [PATCH V4 1/3] cpuidle: play_idle: Make play_idle more flexible Daniel Lezcano
2019-10-04  8:32 ` [PATCH V4 2/3] cpuidle: play_idle: Specify play_idle with an idle state Daniel Lezcano
2019-10-04  8:32 ` [PATCH V4 3/3] powercap/drivers/idle_inject: Specify the idle state to inject Daniel Lezcano
2019-10-15 10:15 ` [PATCH V4 1/3] cpuidle: play_idle: Make play_idle more flexible Daniel Lezcano
2019-10-15 13:06 ` Ulf Hansson
2019-10-16 16:41   ` Daniel Lezcano [this message]

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=762e7966-a276-cca2-7baa-fa22c26f25f3@linaro.org \
    --to=daniel.lezcano@linaro.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-pm@vger.kernel.org \
    --cc=mathieu.poirier@linaro.org \
    --cc=mingo@redhat.com \
    --cc=peterz@infradead.org \
    --cc=rjw@rjwysocki.net \
    --cc=ulf.hansson@linaro.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).