Re: [patch 3/3] clockevents: Fix resume logic - updated version

[Andrew Morton <akpm@linux-foundation.org>]

On Sat, 12 May 2007 13:44:13 +0200 Thomas Gleixner <tglx@linutronix.de> wrote:

> lockdep complains about the lock nesting of clocksource and watchdog
> lock in the resume path. Move watchdog resume out of the clocksource
> lock.
> 
> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
> 
> Index: linux-2.6.21/kernel/time/clocksource.c
> ===================================================================
> --- linux-2.6.21.orig/kernel/time/clocksource.c
> +++ linux-2.6.21/kernel/time/clocksource.c
> @@ -151,9 +151,11 @@ static void clocksource_watchdog(unsigne
>  }
>  static void clocksource_resume_watchdog(void)
>  {
> -	spin_lock(&watchdog_lock);
> +	unsigned long flags;
> +
> +	spin_lock_irqsave(&watchdog_lock, flags);
>  	watchdog_resumed = 1;
> -	spin_unlock(&watchdog_lock);
> +	spin_unlock_irqrestore(&watchdog_lock, flags);
>  }
>  
>  static void clocksource_check_watchdog(struct clocksource *cs)
> @@ -224,9 +226,9 @@ void clocksource_resume(void)
>  			cs->resume();
>  	}
>  
> -	clocksource_resume_watchdog();
> -
>  	spin_unlock_irqrestore(&clocksource_lock, flags);
> +
> +	clocksource_resume_watchdog();
>  }
>  

The locking in clocksource_resume_watchdog looks pretty pointless anyway.
Can't we just delete it?

The only thing it can race against is, conceivably,

        resumed = watchdog_resumed;
        if (unlikely(resumed))
                watchdog_resumed = 0;

which could be solved by using test_and_clear_bit().

Does anyone have any theories about my lockdep warning?