Re: [PATCH] PM / suspend: Catch wake-up requests in suspend_again

Re: [PATCH] PM / suspend: Catch wake-up requests in suspend_again

[Pavel Machek]

On Fri 2014-09-26 17:44:24, Joseph Swantek wrote:
> It is possible during the processing of suspend_again() that a
> wake-up request (i.e. pm_stay_awake()) be made by a driver in response
> to an outside event (i.e. hw interrupt, etc.). Currently, when
> suspend_again() requests to re-enter suspend, suspend_enter() runs
> to re-enter suspend. The pm_wakeup_pending() function is called to check
> for any wake-up requests. However, this function is essentially disabled
> because the events_check_enabled flag is set to false by the previous
> run through suspend_enter(). This causes any wake-up request made
> during the processing of suspend_again() to be delayed until the next
> time the kernel resumes sometime in the future.
> 
> To resolve this issue, if suspend_again() votes to re-enter
> suspend_enter() the count of wake-up requests is re-evaluated
> and the events_check_enabled flag set correctly based on the
> presence of wake-up requests that occurred during suspend_again().
> 

> @@ -188,8 +188,25 @@ static void platform_suspend_recover(suspend_state_t
> state)
> 
>  static bool platform_suspend_again(suspend_state_t state)
>  {
> - return state != PM_SUSPEND_FREEZE && suspend_ops->suspend_again ?
> + int count;
> + bool suspend = state != PM_SUSPEND_FREEZE &&
> + suspend_ops->suspend_again ?
>   suspend_ops->suspend_again() : false;
> +
> + if (suspend) {
> + /*
> + * pm_get_wakeup_count() gets an updated count of wakeup events
> + * that have occured and will return false (i.e. abort suspend)
> + * if a wakeup event has been started during suspend_again() and
> + * is still active. pm_save_wakeup_count() stores the count
> + * and enables pm_wakeup_pending() to properly analyze wakeup
> + * events before entering suspend in suspend_enter().
> + */
> + suspend = pm_get_wakeup_count(&count, false) &&
> +  pm_save_wakeup_count(count);
> + }
> +
> + return suspend;
>  }

That seems rather whitespace damaged.

-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html

[PATCH] PM / suspend: Catch wake-up requests in suspend_again

[Joe Swantek]

It is possible during the processing of suspend_again() that a
wake-up request (i.e. pm_stay_awake()) be made by a driver in response
to an outside event (i.e. hw interrupt, etc.). Currently, when
suspend_again() requests to re-enter suspend, suspend_enter() runs
to re-enter suspend. The pm_wakeup_pending() function is called to check
for any wake-up requests. However, this function is essentially disabled
because the events_check_enabled flag is set to false by the previous
run through suspend_enter(). This causes any wake-up request made
during the processing of suspend_again() to be delayed until the next
time the kernel resumes sometime in the future.

To resolve this issue, if suspend_again() votes to re-enter
suspend_enter() the count of wake-up requests is re-evaluated
and the events_check_enabled flag set correctly based on the
presence of wake-up requests that occurred during suspend_again().

Signed-off-by: Joe Swantek <jswantek@motorola.com>
---
 kernel/power/suspend.c | 19 ++++++++++++++++++-
 1 file changed, 18 insertions(+), 1 deletion(-)

diff --git a/kernel/power/suspend.c b/kernel/power/suspend.c
index 18c6219..f619492 100644
--- a/kernel/power/suspend.c
+++ b/kernel/power/suspend.c
@@ -188,8 +188,25 @@ static void platform_suspend_recover(suspend_state_t state)
 
 static bool platform_suspend_again(suspend_state_t state)
 {
-	return state != PM_SUSPEND_FREEZE && suspend_ops->suspend_again ?
+	int count;
+	bool suspend = state != PM_SUSPEND_FREEZE &&
+		suspend_ops->suspend_again ?
 		suspend_ops->suspend_again() : false;
+
+	if (suspend) {
+		/*
+		 * pm_get_wakeup_count() gets an updated count of wakeup events
+		 * that have occured and will return false (i.e. abort suspend)
+		 * if a wakeup event has been started during suspend_again() and
+		 * is still active. pm_save_wakeup_count() stores the count
+		 * and enables pm_wakeup_pending() to properly analyze wakeup
+		 * events before entering suspend in suspend_enter().
+		 */
+		suspend = pm_get_wakeup_count(&count, false) &&
+			  pm_save_wakeup_count(count);
+	}
+
+	return suspend;
 }
 
 static int suspend_test(int level)
-- 
1.9.1