Hello,

On Fri, 3 May 2013, Paul E. McKenney wrote:

OK, after getting some sleep, I might have located the root cause of
my confusion yesterday.

The key point is that I don't understand why we cannot get the effect
we are looking for with the following in sched.h (or wherever):

static inline int cond_resched_rcu(void)
{
#if defined(CONFIG_DEBUG_ATOMIC_SLEEP) || !defined(CONFIG_PREEMPT_RCU)
	rcu_read_unlock();
	cond_resched();
	rcu_read_lock();
#endif
}

This adds absolutely no overhead in non-debug builds of CONFIG_PREEMPT_RCU,
does the checking in debug builds, and allows voluntary preemption in
!CONFIG_PREEMPT_RCU builds.  CONFIG_PROVE_RCU builds will check for an
(illegal) outer rcu_read_lock() in CONFIG_PREEMPT_RCU builds, and you
will get "scheduling while atomic" in response to an outer rcu_read_lock()
in !CONFIG_PREEMPT_RCU builds.

It also seems to me a lot simpler.

Does this work, or am I still missing something?

	It should work. It is a better version of
the 2nd variant I mentioned here:

http://marc.info/?l=linux-netdev&m=136741839021257&w=2

	I'll stick to this version, hope Peter Zijlstra
agrees. Playing with PREEMPT_ACTIVE or another bit makes
the things more complex.

	To summarize:

- CONFIG_PREEMPT_RCU:
	- no empty functions called
	- CONFIG_DEBUG_ATOMIC_SLEEP can catch errors even
	for this case

- non-CONFIG_PREEMPT_RCU:
	- rcu_read_lock and rcu_read_unlock are barrier(),
	so it expands just to cond_resched()

	I'll repeat the tests tomorrow and if there are
no problems will post official version after the merge window.

Regards

--
Julian Anastasov [off-list ref]