As of 3c3b177a9369 ("reservation: add suppport for read-only access
using rcu") linux/reservation.h uses lockdep macros:
+#define reservation_object_held(obj) lockdep_is_held(&(obj)->lock.base)
This results in build errors when lockdep is disabled as lockdep_is_held()
is only available when lockdep is enabled. This has been reported today
to break the etnaviv kernel driver, which we're hoping to submit for 4.5.
As this gets used with rcu_dereference_protected(), eg:
static inline struct reservation_object_list *
reservation_object_get_list(struct reservation_object *obj)
{
return rcu_dereference_protected(obj->fence,
reservation_object_held(obj));
}
I'm guessing that it's not going to be a simple case of making it always
return true or always return false.
Any ideas how to solve this?
--
FTTC broadband for 0.8mile line: currently at 9.6Mbps down 400kbps up
according to speedtest.net.
On Thu, Nov 26, 2015 at 01:43:40PM +0000, Russell King - ARM Linux wrote:
> As of 3c3b177a9369 ("reservation: add suppport for read-only access
> using rcu") linux/reservation.h uses lockdep macros:
>
> +#define reservation_object_held(obj) lockdep_is_held(&(obj)->lock.base)
>
> This results in build errors when lockdep is disabled as lockdep_is_held()
> is only available when lockdep is enabled. This has been reported today
> to break the etnaviv kernel driver, which we're hoping to submit for 4.5.
>
> As this gets used with rcu_dereference_protected(), eg:
>
> static inline struct reservation_object_list *
> reservation_object_get_list(struct reservation_object *obj)
> {
> return rcu_dereference_protected(obj->fence,
> reservation_object_held(obj));
> }
>
> I'm guessing that it's not going to be a simple case of making it always
> return true or always return false.
>
> Any ideas how to solve this?
The usual approach is something like this:
#ifdef CONFIG_PROVE_LOCKING
#define reservation_object_held(obj) lockdep_is_held(&(obj)->lock.base)
#else
#define reservation_object_held(obj) true
#endif
Thanx, Paul