Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751776AbdG1Blc (ORCPT ); Thu, 27 Jul 2017 21:41:32 -0400 Received: from mail-pg0-f47.google.com ([74.125.83.47]:33780 "EHLO mail-pg0-f47.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751622AbdG1Bl3 (ORCPT ); Thu, 27 Jul 2017 21:41:29 -0400 From: Alex Shi To: rafael@kernel.org, linux-kernel@vger.kernel.org (open list) Cc: Alex Shi , Sebastian Andrzej Siewior , Thomas Gleixner , Anders Roxell , Daniel Lezcano , Tony Lindgren , linux-pm@vger.kernel.org, linux-rt-users Subject: [PATCH v3] cpu_pm: replace raw_notifier to atomic_notifier Date: Fri, 28 Jul 2017 09:39:37 +0800 Message-Id: <1501205977-10774-1-git-send-email-alex.shi@linaro.org> X-Mailer: git-send-email 2.7.4 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 5866 Lines: 187 This patch replace a rwlock and raw notifier by atomic notifier which protected by spin_lock and rcu. The first to reason to have this replace is due to a 'scheduling while atomic' bug of RT kernel on arm/arm64 platform. On arm/arm64, rwlock cpu_pm_notifier_lock in cpu_pm cause a potential schedule after irq disable in idle call chain: cpu_startup_entry cpu_idle_loop local_irq_disable() cpuidle_idle_call call_cpuidle cpuidle_enter cpuidle_enter_state ->enter :arm_enter_idle_state cpu_pm_enter/exit CPU_PM_CPU_IDLE_ENTER read_lock(&cpu_pm_notifier_lock); <-- sleep in idle __rt_spin_lock(); schedule(); The kernel panic is here: [ 4.609601] BUG: scheduling while atomic: swapper/1/0/0x00000002 [ 4.609608] [] arm_enter_idle_state+0x18/0x70 [ 4.609614] Modules linked in: [ 4.609615] [] cpuidle_enter_state+0xf0/0x218 [ 4.609620] [] cpuidle_enter+0x18/0x20 [ 4.609626] Preemption disabled at: [ 4.609627] [] call_cpuidle+0x24/0x40 [ 4.609635] [] schedule_preempt_disabled+0x1c/0x28 [ 4.609639] [] cpu_startup_entry+0x154/0x1f8 [ 4.609645] [] secondary_start_kernel+0x15c/0x1a0 Daniel Lezcano said this notification is needed on arm/arm64 platforms. Sebastian suggested using atomic_notifier instead of rwlock, which is not only removing the sleeping in idle, but also getting better latency improvement. Tony Lezcano found a miss use that rcu_read_lock used after rcu_idle_enter Paul E. McKenney suggested trying RCU_NONIDLE. Thanks everyone! :) This patch passed Fengguang's 0day testing. Signed-off-by: Alex Shi Tested-by: Tony Lindgren To: linux-kernel@vger.kernel.org Cc: Sebastian Andrzej Siewior Cc: Thomas Gleixner Cc: Anders Roxell Cc: Daniel Lezcano Cc: Rafael J. Wysocki Cc: Tony Lindgren Cc: linux-pm@vger.kernel.org Cc: linux-rt-users Acked-by: Sebastian Andrzej Siewior --- kernel/cpu_pm.c | 50 +++++++++++++------------------------------------- 1 file changed, 13 insertions(+), 37 deletions(-) diff --git a/kernel/cpu_pm.c b/kernel/cpu_pm.c index 009cc9a..67b02e1 100644 --- a/kernel/cpu_pm.c +++ b/kernel/cpu_pm.c @@ -22,15 +22,21 @@ #include #include -static DEFINE_RWLOCK(cpu_pm_notifier_lock); -static RAW_NOTIFIER_HEAD(cpu_pm_notifier_chain); +static ATOMIC_NOTIFIER_HEAD(cpu_pm_notifier_chain); static int cpu_pm_notify(enum cpu_pm_event event, int nr_to_call, int *nr_calls) { int ret; - ret = __raw_notifier_call_chain(&cpu_pm_notifier_chain, event, NULL, + /* + * __atomic_notifier_call_chain has a RCU read critical section, which + * could be disfunctional in cpu idle. Copy RCU_NONIDLE code to let + * RCU know this. + */ + rcu_irq_enter_irqson(); + ret = __atomic_notifier_call_chain(&cpu_pm_notifier_chain, event, NULL, nr_to_call, nr_calls); + rcu_irq_exit_irqson(); return notifier_to_errno(ret); } @@ -47,14 +53,7 @@ static int cpu_pm_notify(enum cpu_pm_event event, int nr_to_call, int *nr_calls) */ int cpu_pm_register_notifier(struct notifier_block *nb) { - unsigned long flags; - int ret; - - write_lock_irqsave(&cpu_pm_notifier_lock, flags); - ret = raw_notifier_chain_register(&cpu_pm_notifier_chain, nb); - write_unlock_irqrestore(&cpu_pm_notifier_lock, flags); - - return ret; + return atomic_notifier_chain_register(&cpu_pm_notifier_chain, nb); } EXPORT_SYMBOL_GPL(cpu_pm_register_notifier); @@ -69,14 +68,7 @@ EXPORT_SYMBOL_GPL(cpu_pm_register_notifier); */ int cpu_pm_unregister_notifier(struct notifier_block *nb) { - unsigned long flags; - int ret; - - write_lock_irqsave(&cpu_pm_notifier_lock, flags); - ret = raw_notifier_chain_unregister(&cpu_pm_notifier_chain, nb); - write_unlock_irqrestore(&cpu_pm_notifier_lock, flags); - - return ret; + return atomic_notifier_chain_unregister(&cpu_pm_notifier_chain, nb); } EXPORT_SYMBOL_GPL(cpu_pm_unregister_notifier); @@ -100,7 +92,6 @@ int cpu_pm_enter(void) int nr_calls; int ret = 0; - read_lock(&cpu_pm_notifier_lock); ret = cpu_pm_notify(CPU_PM_ENTER, -1, &nr_calls); if (ret) /* @@ -108,7 +99,6 @@ int cpu_pm_enter(void) * PM entry who are notified earlier to prepare for it. */ cpu_pm_notify(CPU_PM_ENTER_FAILED, nr_calls - 1, NULL); - read_unlock(&cpu_pm_notifier_lock); return ret; } @@ -128,13 +118,7 @@ EXPORT_SYMBOL_GPL(cpu_pm_enter); */ int cpu_pm_exit(void) { - int ret; - - read_lock(&cpu_pm_notifier_lock); - ret = cpu_pm_notify(CPU_PM_EXIT, -1, NULL); - read_unlock(&cpu_pm_notifier_lock); - - return ret; + return cpu_pm_notify(CPU_PM_EXIT, -1, NULL); } EXPORT_SYMBOL_GPL(cpu_pm_exit); @@ -159,7 +143,6 @@ int cpu_cluster_pm_enter(void) int nr_calls; int ret = 0; - read_lock(&cpu_pm_notifier_lock); ret = cpu_pm_notify(CPU_CLUSTER_PM_ENTER, -1, &nr_calls); if (ret) /* @@ -167,7 +150,6 @@ int cpu_cluster_pm_enter(void) * PM entry who are notified earlier to prepare for it. */ cpu_pm_notify(CPU_CLUSTER_PM_ENTER_FAILED, nr_calls - 1, NULL); - read_unlock(&cpu_pm_notifier_lock); return ret; } @@ -190,13 +172,7 @@ EXPORT_SYMBOL_GPL(cpu_cluster_pm_enter); */ int cpu_cluster_pm_exit(void) { - int ret; - - read_lock(&cpu_pm_notifier_lock); - ret = cpu_pm_notify(CPU_CLUSTER_PM_EXIT, -1, NULL); - read_unlock(&cpu_pm_notifier_lock); - - return ret; + return cpu_pm_notify(CPU_CLUSTER_PM_EXIT, -1, NULL); } EXPORT_SYMBOL_GPL(cpu_cluster_pm_exit); -- 2.7.4