Re: [PATCH v5] PM / Core: suspend_again callback for suspend_ops.

[Rafael J. Wysocki]

On Thursday, June 02, 2011, MyungJoo Ham wrote:
> A system or a device may need to control suspend/wakeup events. It may
> want to wakeup the system after a predefined amount of time or at a
> predefined event decided while entering suspend for polling or delayed
> work. Then, it may want to enter suspend again if its predefined wakeup
> condition is the only wakeup reason and there is no outstanding events;
> thus, it does not wakeup the userspace unnecessary or unnecessary
> devices and keeps suspended as long as possible (saving the power).
> 
> Enabling a system to wakeup after a specified time can be easily
> achieved by using RTC. However, to enter suspend again immediately
> without invoking userland and unrelated devices, we need additional
> features in the suspend framework.
> 
> Such need comes from:
> 
> 1. Monitoring a critical device status without interrupts that can
> wakeup the system. (in-suspend polling)
>  An example is ambient temperature monitoring that needs to shut down
> the system or a specific device function if it is too hot or cold. The
> temperature of a specific device may be needed to be monitored as well;
> e.g., a charger monitors battery temperature in order to stop charging
> if overheated.
> 
> 2. Execute critical "delayed work" at suspend.
>  A driver or a system/board may have a delayed work (or any similar
> things) that it wants to execute at the requested time.
>  For example, some chargers want to check the battery voltage some
> time (e.g., 30 seconds) after the battery is fully charged and the
> charger has stopped. Then, the charger restarts charging if the voltage
> has dropped more than a threshold, which is smaller than "restart-charger"
> voltage, which is a threshold to restart charging regardless of the
> time passed.
> 
> This patch allows to add "suspend_again" callback at struct
> platform_suspend_ops and let the "suspend_again" callback return true if
> the system is required to enter suspend again after the current instance
> of wakeup. Device-wise suspend_again implemented at dev_pm_ops or
> syscore is not done because: a) suspend_again feature is usually under
> platform-wise decision and controls the behavior of the whole platform
> and b) There are very limited devices related to the usage cases of
> suspend_again; chargers and temperature sensors are mentioned so far.
> 
> With suspend_again callback registered at struct platform_suspend_ops
> suspend_ops in kernel/power/suspend.c with suspend_set_ops by the
> platform, the suspend framework tries to enter suspend again by
> looping suspend_enter() if suspend_again has returned true and there has
> been no errors in the suspending sequence or pending wakeups (by
> pm_wakeup_pending).
> 
> Tested at Exynos4-NURI.
> 
> Signed-off-by: MyungJoo Ham <myungjoo.ham@xxxxxxxxxxx>
> Signed-off-by: Kyungmin Park <kyungmin.park@xxxxxxxxxxx>

OK, I'll queue it up for 3.1, but I'm still not sure if executing
all of the .suspend_noirq() and .resume_noirq() callbacks in each iteration
of the loop is optimal.

Arguably, we can't do anything more fine grained at the moment, but that's
something to think of in the future.

> --
> Thank you for your valuable comments, Pavel, Rafael, Greg, and others.
> 
> Changed from v4:
> 	- rebased on Linux 3.0 rc1
> 	- removed partial_suspend/resume because they are not needed, for now.
> Changed from v3:
> 	- renamed local variable "pm_wkup_pending" to "wakeup_pending".
> 	pm_wakeup_pending is not used because there is a function with the
> 	same name.
> Changed from v2:
> 	- moved (again) from dev_pm_ops to suspend_ops
> 	- settled suspend_again point at around suspend_enter().
> Changes from v1:
> 	- moved from syscore to dev_pm_ops
> 	- added generic ops for subsystems.
> 	- cleaned up suspend_again code at kernel/power/suspend.
> ---
>  include/linux/suspend.h |    8 ++++++++
>  kernel/power/suspend.c  |   13 ++++++++++---
>  2 files changed, 18 insertions(+), 3 deletions(-)
> 
> diff --git a/include/linux/suspend.h b/include/linux/suspend.h
> index 083ffea..e1e3742 100644
> --- a/include/linux/suspend.h
> +++ b/include/linux/suspend.h
> @@ -92,6 +92,13 @@ typedef int __bitwise suspend_state_t;
>   *	@enter() and @wake(), even if any of them fails.  It is executed after
>   *	a failing @prepare.
>   *
> + * @suspend_again: Returns whether the system should suspend again (true) or
> + *	not (false). If the platform wants to poll sensors or execute some
> + *	code during suspended without invoking userspace and most of devices,
> + *	suspend_again callback is the place assuming that periodic-wakeup or
> + *	alarm-wakeup is already setup. This allows to execute some codes while
> + *	being kept suspended in the view of userland and devices.
> + *
>   * @end: Called by the PM core right after resuming devices, to indicate to
>   *	the platform that the system has returned to the working state or
>   *	the transition to the sleep state has been aborted.
> @@ -113,6 +120,7 @@ struct platform_suspend_ops {
>  	int (*enter)(suspend_state_t state);
>  	void (*wake)(void);
>  	void (*finish)(void);
> +	bool (*suspend_again)(void);
>  	void (*end)(void);
>  	void (*recover)(void);
>  };
> diff --git a/kernel/power/suspend.c b/kernel/power/suspend.c
> index 1c41ba2..d8a6c51 100644
> --- a/kernel/power/suspend.c
> +++ b/kernel/power/suspend.c
> @@ -128,10 +128,12 @@ void __attribute__ ((weak)) arch_suspend_enable_irqs(void)
>  /**
>   *	suspend_enter - enter the desired system sleep state.
>   *	@state:		state to enter
> + *	@wakeup_pending:	indicates that the power transition in progress
> + *				should be aborted.

This should be just one line, otherwise kerneldoc will have problems with it.
I'll fix it up.

Thanks,
Rafael
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/