On Mon, 12 Aug 2019 09:03:10 -0400
Joel Fernandes <joel@xxxxxxxxxxxxxxxxx> wrote:
> > > drivers/base/core.c | 6 +++++-
> > > 1 file changed, 5 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/drivers/base/core.c b/drivers/base/core.c
> > > index 32cf83d1c744..fe25cf690562 100644
> > > --- a/drivers/base/core.c
> > > +++ b/drivers/base/core.c
> > > @@ -99,7 +99,11 @@ void device_links_read_unlock(int not_used)
> > >
> > > int device_links_read_lock_held(void)
> > > {
> > > - return lock_is_held(&device_links_lock);
> > > +#ifdef CONFIG_DEBUG_LOCK_ALLOC
> > > + return lock_is_held(&(device_links_lock.dep_map));
> > > +#else
> > > + return 1;
> > > +#endif
> >
> > return 1? So the lock is always held?
I was thinking the exact same thing.
>
> This is just the pattern of an assert that is disabled, so that
> false-positives don't happen if lockdep is disabled.
>
> So say someone writes a statement like:
> WARN_ON_ONCE(!device_links_read_lock_held());
>
> Since lockdep is disabled, we cannot check whether lock is held or not. Yet,
> we don't want false positives by reporting that the lock is not held. In this
> case, it is better to report that the lock is held to suppress
> false-positives. srcu_read_lock_held() also follows the same pattern.
>
The real answer here is to make that WARN_ON_ONCE() dependent on
lockdep. Something like:
some/header/file.h:
#ifdef CONFIG_DEBUG_LOCK_ALLOC
# define CHECK_DEVICE_LINKS_READ_LOCK_HELD() WARN_ON_ONCE(!defice_links_read_lock_held())
#else
# define CHECK_DEVICE_LINKS_READ_LOCK_HELD() do { } while (0)
#endif
And just use CHECK_DEVICE_LINK_READ_LOCK_HELD() in those places. I
agree with Greg. "device_links_read_lock_heald()" should *never*
blindly return 1. It's confusing.
-- Steve
