Re: [RFC PATCH v2 2/8] cpuidle: record the overhead of idle entry

[Rafael J. Wysocki]

On Saturday, September 30, 2017 9:20:28 AM CEST Aubrey Li wrote:
> Record the overhead of idle entry in micro-second
> 

What is this needed for?

> Signed-off-by: Aubrey Li <aubrey.li@xxxxxxxxxxxxxxx>
> ---
>  drivers/cpuidle/cpuidle.c | 33 +++++++++++++++++++++++++++++++++
>  include/linux/cpuidle.h   | 14 ++++++++++++++
>  kernel/sched/idle.c       |  8 +++++++-
>  3 files changed, 54 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/cpuidle/cpuidle.c b/drivers/cpuidle/cpuidle.c
> index 60bb64f..4066308 100644
> --- a/drivers/cpuidle/cpuidle.c
> +++ b/drivers/cpuidle/cpuidle.c
> @@ -302,6 +302,39 @@ void cpuidle_reflect(struct cpuidle_device *dev, int index)
>  		cpuidle_curr_governor->reflect(dev, index);
>  }
>  
> +/* cpuidle_entry_start - record idle entry start */

A proper kerneldoc, please.

> +void cpuidle_entry_start(void)
> +{
> +	struct cpuidle_device *dev = cpuidle_get_device();
> +
> +	if (dev)
> +		dev->idle_stat.entry_start = local_clock();
> +}
> +
> +/*
> + * cpuidle_entry_end - record idle entry end, and maintain
> + * the entry overhead average in micro-second
> + */

A proper kerneldoc, please.

> +void cpuidle_entry_end(void)
> +{
> +	struct cpuidle_device *dev = cpuidle_get_device();
> +	u64 overhead;
> +	s64 diff;
> +
> +	if (dev) {
> +		dev->idle_stat.entry_end = local_clock();
> +		overhead = div_u64(dev->idle_stat.entry_end -
> +				dev->idle_stat.entry_start, NSEC_PER_USEC);

Is the conversion really necessary?

If so, then why?

And if there is a good reason, what about using right shift to do
an approximate conversion to avoid the extra division here?

> +		diff = overhead - dev->idle_stat.overhead;
> +		dev->idle_stat.overhead += diff >> 3;

Can you please explain what happens in the two lines above?

> +		/*
> +		 * limit overhead to 1us
> +		 */
> +		if (dev->idle_stat.overhead == 0)
> +			dev->idle_stat.overhead = 1;
> +	}
> +}
> +
>  /**
>   * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
>   */
> diff --git a/include/linux/cpuidle.h b/include/linux/cpuidle.h
> index fc1e5d7..cad9b71 100644
> --- a/include/linux/cpuidle.h
> +++ b/include/linux/cpuidle.h
> @@ -72,6 +72,15 @@ struct cpuidle_device_kobj;
>  struct cpuidle_state_kobj;
>  struct cpuidle_driver_kobj;
>  
> +struct cpuidle_stat {
> +	u64			entry_start;	/* nanosecond */
> +	u64			entry_end;	/* nanosecond */
> +	u64			overhead;	/* nanosecond */
> +	unsigned int		predicted_us;	/* microsecond */
> +	bool			predicted;	/* ever predicted? */
> +	bool			fast_idle;	/* fast idle? */

Some of the fields here are never used in the code after this patch.

Also it would be good to add a comment describing the meaning of the
fields.

> +};
> +
>  struct cpuidle_device {
>  	unsigned int		registered:1;
>  	unsigned int		enabled:1;
> @@ -89,6 +98,7 @@ struct cpuidle_device {
>  	cpumask_t		coupled_cpus;
>  	struct cpuidle_coupled	*coupled;
>  #endif
> +	struct cpuidle_stat	idle_stat;
>  };
>  
>  DECLARE_PER_CPU(struct cpuidle_device *, cpuidle_devices);
> @@ -131,6 +141,8 @@ extern bool cpuidle_not_available(struct cpuidle_driver *drv,
>  
>  extern int cpuidle_select(struct cpuidle_driver *drv,
>  			  struct cpuidle_device *dev);
> +extern void cpuidle_entry_start(void);
> +extern void cpuidle_entry_end(void);
>  extern int cpuidle_enter(struct cpuidle_driver *drv,
>  			 struct cpuidle_device *dev, int index);
>  extern void cpuidle_reflect(struct cpuidle_device *dev, int index);
> @@ -164,6 +176,8 @@ static inline bool cpuidle_not_available(struct cpuidle_driver *drv,
>  static inline int cpuidle_select(struct cpuidle_driver *drv,
>  				 struct cpuidle_device *dev)
>  {return -ENODEV; }
> +static inline void cpuidle_entry_start(void) { }
> +static inline void cpuidle_entry_end(void) { }
>  static inline int cpuidle_enter(struct cpuidle_driver *drv,
>  				struct cpuidle_device *dev, int index)
>  {return -ENODEV; }
> diff --git a/kernel/sched/idle.c b/kernel/sched/idle.c
> index 6c23e30..0951dac 100644
> --- a/kernel/sched/idle.c
> +++ b/kernel/sched/idle.c
> @@ -210,6 +210,12 @@ static void cpuidle_idle_call(void)
>  static void do_idle(void)
>  {
>  	/*
> +	 * we record idle entry overhead now, so any deferrable items
> +	 * in idle entry path need to be placed between cpuidle_entry_start()
> +	 * and cpuidle_entry_end()
> +	 */
> +	cpuidle_entry_start();
> +	/*
>  	 * If the arch has a polling bit, we maintain an invariant:
>  	 *
>  	 * Our polling bit is clear if we're not scheduled (i.e. if rq->curr !=
> @@ -217,10 +223,10 @@ static void do_idle(void)
>  	 * then setting need_resched is guaranteed to cause the CPU to
>  	 * reschedule.
>  	 */
> -
>  	__current_set_polling();
>  	quiet_vmstat();
>  	tick_nohz_idle_enter();
> +	cpuidle_entry_end();
>  
>  	while (!need_resched()) {
>  		check_pgt_cache();
>