Re: [PATCH 2.6.28-rc6] ACPICA: don't cond_resched() when irqs_disabled()

[Len Brown <lenb@kernel.org>]

On Wed, 26 Nov 2008, Wu Fengguang wrote:

> [add CC to <stable@kernel.org>, since this bug was introduced in the
> 2.6.27-rcX time frame, and should help 2.6.28 and 2.6.27.x alike]

Actually, ACPI_PREEMPTION_POINT() went in to 2.6.28-rc1,
it is not present in 2.6.27.stable so this patch
does not apply to 27.stable.

> The ACPI routines could be called from run_workqueue() with irqs disabled.
> So we should test irqs_disabled() before calling cond_resched().

I don't know how you provoked this failure.
The ACPI interpreter should always be run with interrupts enabled.
If it isn't then that is a (different) bug.

There is, however, one exception, and it is during suspend/resume;
and that _is_ a legal way to provoke this bug:

http://bugzilla.kernel.org/show_bug.cgi?id=12252

and thus the fix is actually correct, and I'll apply it.

thanks,
-- Len Brown, Intel Open Source Technology Center


> --- a/include/acpi/platform/aclinux.h
> +++ b/include/acpi/platform/aclinux.h
> @@ -141,6 +141,10 @@ static inline void *acpi_os_acquire_object(acpi_cache_t * cache)
>  /*
>   * We need to show where it is safe to preempt execution of ACPICA
>   */
> -#define ACPI_PREEMPTION_POINT()	cond_resched()
> +#define ACPI_PREEMPTION_POINT()		\
> +	do {				\
> +		if (!irqs_disabled())	\
> +			cond_resched();	\
> +	} while (0)
>  
>  #endif				/* __ACLINUX_H__ */
> --