[PATCH] ACPI: fix acpi_power_off lockdep splat

[PATCH] ACPI: fix acpi_power_off lockdep splat

[Andrea Righi]

Implement acpi_os_create_lock() as a C-preprocessor macro to assign
unique lock_class_key to dynamically allocated locks and suppress wrong
lockdep warnings.

This fixes:
https://bugzilla.kernel.org/show_bug.cgi?id=38152

Reported-by: Borislav Petkov <bp@alien8.de>
Signed-off-by: Andrea Righi <andrea@betterlinux.com>
---
 drivers/acpi/osl.c      |    2 +-
 include/acpi/acpiosxf.h |   12 +++++++++++-
 2 files changed, 12 insertions(+), 2 deletions(-)

diff --git a/drivers/acpi/osl.c b/drivers/acpi/osl.c
index 52ca964..2a67602 100644
--- a/drivers/acpi/osl.c
+++ b/drivers/acpi/osl.c
@@ -1336,7 +1336,7 @@ EXPORT_SYMBOL(acpi_resources_are_enforced);
  * Create and initialize a spinlock.
  */
 acpi_status
-acpi_os_create_lock(acpi_spinlock *out_handle)
+__acpi_os_create_lock(acpi_spinlock *out_handle)
 {
 	spinlock_t *lock;
 
diff --git a/include/acpi/acpiosxf.h b/include/acpi/acpiosxf.h
index a756bc8..4a0385d 100644
--- a/include/acpi/acpiosxf.h
+++ b/include/acpi/acpiosxf.h
@@ -99,7 +99,17 @@ acpi_os_table_override(struct acpi_table_header *existing_table,
  * Spinlock primitives
  */
 acpi_status
-acpi_os_create_lock(acpi_spinlock *out_handle);
+__acpi_os_create_lock(acpi_spinlock *out_handle);
+
+#define acpi_os_create_lock(__handle)			\
+({							\
+	acpi_status ret;				\
+							\
+	ret = __acpi_os_create_lock(__handle);		\
+	if (ret == AE_OK)				\
+		spin_lock_init(*(__handle));		\
+	ret;						\
+})
 
 void acpi_os_delete_lock(acpi_spinlock handle);
 
-- 
1.7.4.1

Re: [PATCH] ACPI: fix acpi_power_off lockdep splat

[Rafael J. Wysocki]

On Saturday, July 02, 2011, Andrea Righi wrote:
> Implement acpi_os_create_lock() as a C-preprocessor macro to assign
> unique lock_class_key to dynamically allocated locks and suppress wrong
> lockdep warnings.
> 
> This fixes:
> https://bugzilla.kernel.org/show_bug.cgi?id=38152
> 
> Reported-by: Borislav Petkov <bp@alien8.de>
> Signed-off-by: Andrea Righi <andrea@betterlinux.com>

So, how exactly does it work?

> ---
>  drivers/acpi/osl.c      |    2 +-
>  include/acpi/acpiosxf.h |   12 +++++++++++-
>  2 files changed, 12 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/acpi/osl.c b/drivers/acpi/osl.c
> index 52ca964..2a67602 100644
> --- a/drivers/acpi/osl.c
> +++ b/drivers/acpi/osl.c
> @@ -1336,7 +1336,7 @@ EXPORT_SYMBOL(acpi_resources_are_enforced);
>   * Create and initialize a spinlock.
>   */
>  acpi_status
> -acpi_os_create_lock(acpi_spinlock *out_handle)
> +__acpi_os_create_lock(acpi_spinlock *out_handle)
>  {
>  	spinlock_t *lock;
>  
> diff --git a/include/acpi/acpiosxf.h b/include/acpi/acpiosxf.h
> index a756bc8..4a0385d 100644
> --- a/include/acpi/acpiosxf.h
> +++ b/include/acpi/acpiosxf.h
> @@ -99,7 +99,17 @@ acpi_os_table_override(struct acpi_table_header *existing_table,
>   * Spinlock primitives
>   */
>  acpi_status
> -acpi_os_create_lock(acpi_spinlock *out_handle);
> +__acpi_os_create_lock(acpi_spinlock *out_handle);
> +
> +#define acpi_os_create_lock(__handle)			\
> +({							\
> +	acpi_status ret;				\
> +							\
> +	ret = __acpi_os_create_lock(__handle);		\
> +	if (ret == AE_OK)				\
> +		spin_lock_init(*(__handle));		\

It seems that spin_lock_init() has already been done on *__handle, right?

> +	ret;						\
> +})
>  
>  void acpi_os_delete_lock(acpi_spinlock handle);

Rafael

Re: [PATCH] ACPI: fix acpi_power_off lockdep splat

[Florian Mickler]

On Sun, 3 Jul 2011 08:58:41 +0200
"Rafael J. Wysocki" <rjw@sisk.pl> wrote:

> On Saturday, July 02, 2011, Andrea Righi wrote:
> > Implement acpi_os_create_lock() as a C-preprocessor macro to assign
> > unique lock_class_key to dynamically allocated locks and suppress wrong
> > lockdep warnings.
> > 
> > This fixes:
> > https://bugzilla.kernel.org/show_bug.cgi?id=38152
> > 
> > Reported-by: Borislav Petkov <bp@alien8.de>
> > Signed-off-by: Andrea Righi <andrea@betterlinux.com>
> 
> So, how exactly does it work?

The spin_lock_init macro stringifies it's argument and uses that as a
name for the lock in the debugging.

By re-executing spin_lock_init (have not checked if that harms
anything, but it should be ok) in the _macro_ the key changes from
"lock" for all three locks to the actual argument
"&acpi_gbl_global_lock_pending_lock", "&acpi_gbl_gpe_lock" or
"&acpi_gbl_hardware_lock". 

Regards,
Flo

> > ---
> >  drivers/acpi/osl.c      |    2 +-
> >  include/acpi/acpiosxf.h |   12 +++++++++++-
> >  2 files changed, 12 insertions(+), 2 deletions(-)
> > 
> > diff --git a/drivers/acpi/osl.c b/drivers/acpi/osl.c
> > index 52ca964..2a67602 100644
> > --- a/drivers/acpi/osl.c
> > +++ b/drivers/acpi/osl.c
> > @@ -1336,7 +1336,7 @@ EXPORT_SYMBOL(acpi_resources_are_enforced);
> >   * Create and initialize a spinlock.
> >   */
> >  acpi_status
> > -acpi_os_create_lock(acpi_spinlock *out_handle)
> > +__acpi_os_create_lock(acpi_spinlock *out_handle)
> >  {
> >  	spinlock_t *lock;
> >  
> > diff --git a/include/acpi/acpiosxf.h b/include/acpi/acpiosxf.h
> > index a756bc8..4a0385d 100644
> > --- a/include/acpi/acpiosxf.h
> > +++ b/include/acpi/acpiosxf.h
> > @@ -99,7 +99,17 @@ acpi_os_table_override(struct acpi_table_header *existing_table,
> >   * Spinlock primitives
> >   */
> >  acpi_status
> > -acpi_os_create_lock(acpi_spinlock *out_handle);
> > +__acpi_os_create_lock(acpi_spinlock *out_handle);
> > +
> > +#define acpi_os_create_lock(__handle)			\
> > +({							\
> > +	acpi_status ret;				\
> > +							\
> > +	ret = __acpi_os_create_lock(__handle);		\
> > +	if (ret == AE_OK)				\
> > +		spin_lock_init(*(__handle));		\
> 
> It seems that spin_lock_init() has already been done on *__handle, right?
> 
> > +	ret;						\
> > +})
> >  
> >  void acpi_os_delete_lock(acpi_spinlock handle);
> 
> Rafael

Re: [PATCH] ACPI: fix acpi_power_off lockdep splat

[Andrea Righi]

On Sun, Jul 03, 2011 at 10:21:37AM +0200, Florian Mickler wrote:
> On Sun, 3 Jul 2011 08:58:41 +0200
> "Rafael J. Wysocki" <rjw@sisk.pl> wrote:
> 
> > On Saturday, July 02, 2011, Andrea Righi wrote:
> > > Implement acpi_os_create_lock() as a C-preprocessor macro to assign
> > > unique lock_class_key to dynamically allocated locks and suppress wrong
> > > lockdep warnings.
> > > 
> > > This fixes:
> > > https://bugzilla.kernel.org/show_bug.cgi?id=38152
> > > 
> > > Reported-by: Borislav Petkov <bp@alien8.de>
> > > Signed-off-by: Andrea Righi <andrea@betterlinux.com>
> > 
> > So, how exactly does it work?
> 
> The spin_lock_init macro stringifies it's argument and uses that as a
> name for the lock in the debugging.

Correct.

> 
> By re-executing spin_lock_init (have not checked if that harms
> anything, but it should be ok) in the _macro_ the key changes from
> "lock" for all three locks to the actual argument
> "&acpi_gbl_global_lock_pending_lock", "&acpi_gbl_gpe_lock" or
> "&acpi_gbl_hardware_lock". 

We can simply move spin_lock_init() out of __acpi_os_create_lock(), so
that we have a single spin_lock_init() in the acpi_os_create_lock macro.

I'll post another patch soon (and I'll also add your description on
how it works).

Thanks!
-Andrea

> 
> Regards,
> Flo
> 
> > > ---
> > >  drivers/acpi/osl.c      |    2 +-
> > >  include/acpi/acpiosxf.h |   12 +++++++++++-
> > >  2 files changed, 12 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/drivers/acpi/osl.c b/drivers/acpi/osl.c
> > > index 52ca964..2a67602 100644
> > > --- a/drivers/acpi/osl.c
> > > +++ b/drivers/acpi/osl.c
> > > @@ -1336,7 +1336,7 @@ EXPORT_SYMBOL(acpi_resources_are_enforced);
> > >   * Create and initialize a spinlock.
> > >   */
> > >  acpi_status
> > > -acpi_os_create_lock(acpi_spinlock *out_handle)
> > > +__acpi_os_create_lock(acpi_spinlock *out_handle)
> > >  {
> > >  	spinlock_t *lock;
> > >  
> > > diff --git a/include/acpi/acpiosxf.h b/include/acpi/acpiosxf.h
> > > index a756bc8..4a0385d 100644
> > > --- a/include/acpi/acpiosxf.h
> > > +++ b/include/acpi/acpiosxf.h
> > > @@ -99,7 +99,17 @@ acpi_os_table_override(struct acpi_table_header *existing_table,
> > >   * Spinlock primitives
> > >   */
> > >  acpi_status
> > > -acpi_os_create_lock(acpi_spinlock *out_handle);
> > > +__acpi_os_create_lock(acpi_spinlock *out_handle);
> > > +
> > > +#define acpi_os_create_lock(__handle)			\
> > > +({							\
> > > +	acpi_status ret;				\
> > > +							\
> > > +	ret = __acpi_os_create_lock(__handle);		\
> > > +	if (ret == AE_OK)				\
> > > +		spin_lock_init(*(__handle));		\
> > 
> > It seems that spin_lock_init() has already been done on *__handle, right?
> > 
> > > +	ret;						\
> > > +})
> > >  
> > >  void acpi_os_delete_lock(acpi_spinlock handle);
> > 
> > Rafael

Re: [PATCH] ACPI: fix acpi_power_off lockdep splat

[Rafael J. Wysocki]

On Sunday, July 03, 2011, Florian Mickler wrote:
> On Sun, 3 Jul 2011 08:58:41 +0200
> "Rafael J. Wysocki" <rjw@sisk.pl> wrote:
> 
> > On Saturday, July 02, 2011, Andrea Righi wrote:
> > > Implement acpi_os_create_lock() as a C-preprocessor macro to assign
> > > unique lock_class_key to dynamically allocated locks and suppress wrong
> > > lockdep warnings.
> > > 
> > > This fixes:
> > > https://bugzilla.kernel.org/show_bug.cgi?id=38152
> > > 
> > > Reported-by: Borislav Petkov <bp@alien8.de>
> > > Signed-off-by: Andrea Righi <andrea@betterlinux.com>
> > 
> > So, how exactly does it work?
> 
> The spin_lock_init macro stringifies it's argument and uses that as a
> name for the lock in the debugging.
> 
> By re-executing spin_lock_init (have not checked if that harms
> anything, but it should be ok) in the _macro_ the key changes from
> "lock" for all three locks to the actual argument
> "&acpi_gbl_global_lock_pending_lock", "&acpi_gbl_gpe_lock" or
> "&acpi_gbl_hardware_lock". 

OK, thanks for the explanation.

The proposed solution doesn't seem to be extremely straightforward,
but perhaps it's the simplest one.  I need to think about it a bit more.

Thanks,
Rafael