[PATCH 3/4] x86: Don't call idle notifier inside rcu extended QS

[Frederic Weisbecker]

The idle notifier in enter_idle / __exit_idle is called
inside the RCU extended quiescent state.

This results in the following warning:

[    2.081278] ===================================================
[    2.081281] [ INFO: suspicious rcu_dereference_check() usage. ]
[    2.081282] ---------------------------------------------------
[    2.081284] kernel/notifier.c:81 invoked rcu_dereference_check() while in RCU extended quiescent state!
[    2.081286]
[    2.081287] other info that might help us debug this:
[    2.081288]
[    2.081289]
[    2.081289] rcu_scheduler_active = 1, debug_locks = 0
[    2.081292] 1 lock held by kworker/0:0/0:
[    2.081293]  #0:  (rcu_read_lock){.+.+..}, at: [<ffffffff81785c60>] __atomic_notifier_call_chain+0x0/0xa0
[    2.081301]
[    2.081302] stack backtrace:
[    2.081305] Pid: 0, comm: kworker/0:0 Not tainted 3.0.0-rc1+ #103
[    2.081307] Call Trace:
[    2.081313]  [<ffffffff8108e0e4>] lockdep_rcu_dereference+0xd4/0x100
[    2.081316]  [<ffffffff81785c4a>] notifier_call_chain+0x13a/0x150
[    2.081319]  [<ffffffff81785cc7>] __atomic_notifier_call_chain+0x67/0xa0
[    2.081322]  [<ffffffff81785c60>] ? notifier_call_chain+0x150/0x150
[    2.081325]  [<ffffffff81785d11>] atomic_notifier_call_chain+0x11/0x20
[    2.081329]  [<ffffffff8100afa0>] enter_idle+0x20/0x30
[    2.081332]  [<ffffffff8100b05e>] cpu_idle+0xae/0x100
[    2.081335]  [<ffffffff8177951f>] start_secondary+0x1cf/0x1d6

Fix this by entering RCU extended QS later, after we call the
notifier. Exit it also sooner to call the exit idle notifier.

Signed-off-by: Frederic Weisbecker <fweisbec@xxxxxxxxx>
Cc: Paul E. McKenney <paulmck@xxxxxxxxxxxxxxxxxx>
Cc: Ingo Molnar <mingo@xxxxxxx>
Cc: Peter Zijlstra <a.p.zijlstra@xxxxxxxxx>
Cc: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
Cc: H. Peter Anvin <hpa@xxxxxxxxx>
---
 arch/x86/kernel/process_64.c |   11 ++++++++---
 1 files changed, 8 insertions(+), 3 deletions(-)

diff --git a/arch/x86/kernel/process_64.c b/arch/x86/kernel/process_64.c
index 3fe0883..a8bd222 100644
--- a/arch/x86/kernel/process_64.c
+++ b/arch/x86/kernel/process_64.c
@@ -120,7 +120,7 @@ void cpu_idle(void)
 
 	/* endless idle loop with no priority at all */
 	while (1) {
-		tick_nohz_enter_idle();
+		tick_nohz_stop_sched_tick(1);
 		while (!need_resched()) {
 
 			rmb();
@@ -134,18 +134,23 @@ void cpu_idle(void)
 			 */
 			local_irq_disable();
 			enter_idle();
+			/*
+			 * We can't enter extended QS before due to idle
+			 * notifier that uses RCU.
+			 */
+			rcu_enter_nohz();
 			/* Don't trace irqs off for idle */
 			stop_critical_timings();
 			pm_idle();
 			start_critical_timings();
-
+			rcu_exit_nohz();
 			/* In many cases the interrupt that ended idle
 			   has already called exit_idle. But some idle
 			   loops can be woken up without interrupt. */
 			__exit_idle();
 		}
 
-		tick_nohz_exit_idle();
+		tick_nohz_restart_sched_tick();
 		preempt_enable_no_resched();
 		schedule();
 		preempt_disable();
-- 
1.7.5.2

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/