Re: [PATCH 2/2]ACPI: assign different lockdep key for acpi works

[Andrew Morton]

On Fri, 11 Dec 2009 11:04:58 +0800
Shaohua Li <shaohua.li@xxxxxxxxx> wrote:

> Differentiate works for three different workqueue in ACPI. This fixes
> a fase alarm from lockdep, as acpi hotplug workqueue waits other
> workqueues.
> 
> http://bugzilla.kernel.org/show_bug.cgi?id=14553
> 
> Signed-off-by: Shaohua Li <shaohua.li@xxxxxxxxx>
> ---
>  drivers/acpi/osl.c |   12 +++++++++++-
>  1 file changed, 11 insertions(+), 1 deletion(-)
> 
> Index: linux/drivers/acpi/osl.c
> ===================================================================
> --- linux.orig/drivers/acpi/osl.c	2009-12-10 10:40:15.000000000 +0800
> +++ linux/drivers/acpi/osl.c	2009-12-10 10:43:25.000000000 +0800
> @@ -729,6 +729,8 @@ static acpi_status __acpi_os_execute(acp
>  	struct acpi_os_dpc *dpc;
>  	struct workqueue_struct *queue;
>  	int ret;
> +	static struct lock_class_key hp_key, notify_key, acpid_key;
> +
>  	ACPI_DEBUG_PRINT((ACPI_DB_EXEC,
>  			  "Scheduling function [%p(%p)] for deferred execution.\n",
>  			  function, context));
> @@ -758,7 +760,15 @@ static acpi_status __acpi_os_execute(acp
>  	queue = hp ? kacpi_hotplug_wq :
>  		(type == OSL_NOTIFY_HANDLER ? kacpi_notify_wq : kacpid_wq);
>  	dpc->wait = hp ? 1 : 0;
> -	INIT_WORK(&dpc->work, acpi_os_execute_deferred);
> +	if (queue == kacpi_hotplug_wq)
> +		INIT_WORK_KEY(&dpc->work, acpi_os_execute_deferred,
> +				hp_key, "acpi_hp_work");
> +	else if (queue == kacpi_notify_wq)
> +		INIT_WORK_KEY(&dpc->work, acpi_os_execute_deferred,
> +				notify_key, "acpi_notify_work");
> +	else
> +		INIT_WORK_KEY(&dpc->work, acpi_os_execute_deferred,
> +				acpid_key, "acpi_acpid_work");
>  	ret = queue_work(queue, &dpc->work);
>  
>  	if (!ret) {

Yes, that's a shortcoming in the present INIT_WORK().  However I think
you can solve this more simply, without introducing INIT_WORK_KEY():

	if (queue == kacpi_hotplug_wq)
		INIT_WORK(&dpc->work, acpi_os_execute_deferred);
	else if (queue == kacpi_notify_wq)
		INIT_WORK(&dpc->work, acpi_os_execute_deferred);
	else
		INIT_WORK(&dpc->work, acpi_os_execute_deferred);

it looks silly, but I think the compiler will do the right thing with it
- you'll get a separate lock_class_key at each site.

Obviously it would need an explanatory comment.
--
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/