2009-12-11 03:04:59

by Shaohua Li

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

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 <[email protected]>
---
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) {

2009-12-11 21:58:09

by Andrew Morton

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

On Fri, 11 Dec 2009 11:04:58 +0800
Shaohua Li <[email protected]> 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 <[email protected]>
> ---
> 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.

2009-12-14 01:08:38

by Shaohua Li

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

On Sat, Dec 12, 2009 at 05:57:16AM +0800, Andrew Morton wrote:
> On Fri, 11 Dec 2009 11:04:58 +0800
> Shaohua Li <[email protected]> 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 <[email protected]>
> > ---
> > 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.
Yes, I tried this way before, it centainly works. I just thought maybe
others need similar API. If I'm over thinking, let's take your way.

Thanks,
Shaohua