On Wed, Sep 16, 2020 at 11:45:27AM -0700, Jakub Kicinski wrote:
> When CONFIG_LOCKDEP is not set, lock_is_held() and lockdep_is_held()
> are not declared or defined. This forces all callers to use ifdefs
> around these checks.
>
> Recent RCU changes added a lot of lockdep_is_held() calls inside
> rcu_dereference_protected(). rcu_dereference_protected() hides
> its argument on !LOCKDEP builds, but this may lead to unused variable
> warnings.
>
> Provide forward declarations of lock_is_held() and lockdep_is_held()
> but never define them. This way callers can keep them visible to
> the compiler on !LOCKDEP builds and instead depend on dead code
> elimination to remove the references before the linker barfs.
>
> We need lock_is_held() for RCU.
>
> Signed-off-by: Jakub Kicinski <kuba@xxxxxxxxxx>
> --
> CC: peterz@xxxxxxxxxxxxx
> CC: mingo@xxxxxxxxxx
> CC: will@xxxxxxxxxx
> ---
> include/linux/lockdep.h | 6 ++++++
> 1 file changed, 6 insertions(+)
>
> diff --git a/include/linux/lockdep.h b/include/linux/lockdep.h
> index 6a584b3e5c74..6b5bbc536bf6 100644
> --- a/include/linux/lockdep.h
> +++ b/include/linux/lockdep.h
> @@ -371,6 +371,12 @@ static inline void lockdep_unregister_key(struct lock_class_key *key)
>
> #define lockdep_depth(tsk) (0)
>
> +/*
> + * Dummy forward declarations, allow users to write less ifdef-y code
> + * and depend on dead code elimination.
> + */
> +int lock_is_held(const void *);
extern int lock_is_held(const struct lockdep_map *);
> +int lockdep_is_held(const void *);
extern
I suppose we can't pull the lockdep_is_held() definition out from under
CONFIG_LOCKDEP because it does the ->dep_map dereference and many types
will simply not have that member.
> #define lockdep_is_held_type(l, r) (1)
>
> #define lockdep_assert_held(l) do { (void)(l); } while (0)
