Re: [PATCH] rcu/sync: kill rcu_sync_type/gp_type

From: Oleg Nesterov
Date: Wed Apr 24 2019 - 07:37:09 EST


On 04/23, Paul E. McKenney wrote:
>
> I wordsmithed the commit log and merged in the RCU-bh and RCU checks
> to rcu_sync_is_idle(), with the result shown below. Does that work
> OK, or did I mess something up?

Yes, thanks!

the additional RCU-bh and RCU checks matches the "or introduce rcu_read_lock_any_held()"
note from the changelog, perhaps it makes some sense...

Just one nit below,

> - * Must be invoked within an RCU read-side critical section whose
> - * flavor matches that of the rcu_sync struture.
> + * Must be invoked within an RCU-sched read-side critical section.
^^^^^^^^^

Given that the actual code:

> static inline bool rcu_sync_is_idle(struct rcu_sync *rsp)
> {
> -#ifdef CONFIG_PROVE_RCU
> - rcu_sync_lockdep_assert(rsp);
> -#endif
> + RCU_LOCKDEP_WARN(!rcu_read_lock_held() &&
> + !rcu_read_lock_bh_held() &&
> + !rcu_read_lock_sched_held(),
> + "suspicious rcu_sync_is_idle() usage");

does RCU_LOCKDEP_WARN(!rcu_read_lock_any()) the comment should say

Must be invoked within an RCU read-side critical section of
any flavor

?

Oleg.