On Sat, Jan 18, 2020 at 10:24:18PM +0530, Amol Grover wrote:
> Passing a complex lockdep condition to __list_check_rcu results
> in false positive lockdep splat due to incorrect expression
> evaluation.
>
> For example, a lockdep check condition `cond1 || cond2` is
> evaluated as `!cond1 || cond2 && !rcu_read_lock_any_held()`
> which, according to operator precedence, evaluates to
> `!cond1 || (cond2 && !rcu_read_lock_any_held())`.
> This would result in a lockdep splat when cond1 is false
> and cond2 is true which is logically incorrect.
>
> Signed-off-by: Amol Grover <frextrite@xxxxxxxxx>
Good catch!
Acked-by: Joel Fernandes (Google) <joel@xxxxxxxxxxxxxxxxx>
thanks,
- Joel
> ---
> include/linux/rculist.h | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/include/linux/rculist.h b/include/linux/rculist.h
> index 4158b7212936..dce491f0b354 100644
> --- a/include/linux/rculist.h
> +++ b/include/linux/rculist.h
> @@ -50,9 +50,9 @@ static inline void INIT_LIST_HEAD_RCU(struct list_head *list)
> #define __list_check_rcu(dummy, cond, extra...) \
> ({ \
> check_arg_count_one(extra); \
> - RCU_LOCKDEP_WARN(!cond && !rcu_read_lock_any_held(), \
> + RCU_LOCKDEP_WARN(!(cond) && !rcu_read_lock_any_held(), \
> "RCU-list traversed in non-reader section!"); \
> - })
> + })
> #else
> #define __list_check_rcu(dummy, cond, extra...) \
> ({ check_arg_count_one(extra); })
> --
> 2.24.1
>
