Re: [RFC][PATCH] sched: Fix a deadlock of cpu-hotplug

["Srivatsa S. Bhat" <srivatsa.bhat@linux.vnet.ibm.com>]

On 10/25/2012 09:02 AM, Michael Wang wrote:
> On 10/24/2012 05:38 PM, Peter Zijlstra wrote:
>> On Wed, 2012-10-24 at 17:25 +0800, Huacai Chen wrote:
>>> We found poweroff sometimes fails on our computers, so we have the
>>> lock debug options configured. Then, when we do poweroff or take a
>>> cpu down via cpu-hotplug, kernel complain as below. To resove this,
>>> we modify sched_ttwu_pending(), disable the local irq when acquire
>>> rq->lock.
>>>
>>> [   83.066406] =================================
>>> [   83.066406] [ INFO: inconsistent lock state ]
>>> [   83.066406] 3.5.0-3.lemote #428 Not tainted
>>> [   83.066406] ---------------------------------
>>> [   83.066406] inconsistent {IN-HARDIRQ-W} -> {HARDIRQ-ON-W} usage.
>>> [   83.066406] migration/1/7 [HC0[0]:SC0[0]:HE1:SE1] takes:
>>> [   83.066406]  (&rq->lock){?.-.-.}, at: [<ffffffff802585ac>] sched_ttwu_pending+0x64/0x98
>>> [   83.066406] {IN-HARDIRQ-W} state was registered at:
>>> [   83.066406]   [<ffffffff8027c9ac>] __lock_acquire+0x80c/0x1cc0
>>> [   83.066406]   [<ffffffff8027e3d0>] lock_acquire+0x60/0x9c
>>> [   83.066406]   [<ffffffff8074ba04>] _raw_spin_lock+0x3c/0x50
>>> [   83.066406]   [<ffffffff8025a2fc>] scheduler_tick+0x48/0x178
>>> [   83.066406]   [<ffffffff8023b334>] update_process_times+0x54/0x70
>>> [   83.066406]   [<ffffffff80277568>] tick_handle_periodic+0x2c/0x9c
>>> [   83.066406]   [<ffffffff8020a818>] c0_compare_interrupt+0x8c/0x94
>>> [   83.066406]   [<ffffffff8029ec8c>] handle_irq_event_percpu+0x7c/0x248
>>> [   83.066406]   [<ffffffff802a2774>] handle_percpu_irq+0x8c/0xc0
>>> [   83.066406]   [<ffffffff8029e2c8>] generic_handle_irq+0x48/0x58
>>> [   83.066406]   [<ffffffff80205c04>] do_IRQ+0x18/0x24
>>> [   83.066406]   [<ffffffff802016e4>] mach_irq_dispatch+0xe4/0x124
>>> [   83.066406]   [<ffffffff80203ca0>] ret_from_irq+0x0/0x4
>>> [   83.066406]   [<ffffffff8022d114>] console_unlock+0x3e8/0x4c0
>>> [   83.066406]   [<ffffffff811ff0d0>] con_init+0x370/0x398
>>> [   83.066406]   [<ffffffff811fe3e0>] console_init+0x34/0x50
>>> [   83.066406]   [<ffffffff811e4844>] start_kernel+0x2f8/0x4e0
>>> [   83.066406] irq event stamp: 971
>>> [   83.066406] hardirqs last  enabled at (971): [<ffffffff8021c384>] local_flush_tlb_all+0x134/0x17c
>>> [   83.066406] hardirqs last disabled at (970): [<ffffffff8021c298>] local_flush_tlb_all+0x48/0x17c
>>> [   83.066406] softirqs last  enabled at (0): [<ffffffff802298a4>] copy_process+0x510/0x117c
>>> [   83.066406] softirqs last disabled at (0): [<          (null)>] (null)
>>> [   83.066406]
>>> [   83.066406] other info that might help us debug this:
>>> [   83.066406]  Possible unsafe locking scenario:
>>> [   83.066406]
>>> [   83.066406]        CPU0
>>> [   83.066406]        ----
>>> [   83.066406]   lock(&rq->lock);
>>> [   83.066406]   <Interrupt>
>>> [   83.066406]     lock(&rq->lock);
>>> [   83.066406]
>>> [   83.066406]  *** DEADLOCK ***
>>> [   83.066406]
>>> [   83.066406] no locks held by migration/1/7.
>>> [   83.066406]
>>> [   83.066406] stack backtrace:
>>> [   83.066406] Call Trace:
>>> [   83.066406] [<ffffffff80747544>] dump_stack+0x8/0x34
>>> [   83.066406] [<ffffffff8027ba04>] print_usage_bug+0x2ec/0x314
>>> [   83.066406] [<ffffffff8027be28>] mark_lock+0x3fc/0x774
>>> [   83.066406] [<ffffffff8027ca48>] __lock_acquire+0x8a8/0x1cc0
>>> [   83.066406] [<ffffffff8027e3d0>] lock_acquire+0x60/0x9c
>>> [   83.066406] [<ffffffff8074ba04>] _raw_spin_lock+0x3c/0x50
>>> [   83.066406] [<ffffffff802585ac>] sched_ttwu_pending+0x64/0x98
>>> [   83.066406] [<ffffffff80745ff4>] migration_call+0x10c/0x2e0
>>> [   83.066406] [<ffffffff80253110>] notifier_call_chain+0x44/0x94
>>> [   83.066406] [<ffffffff8022eae0>] __cpu_notify+0x30/0x5c
>>> [   83.066406] [<ffffffff8072b598>] take_cpu_down+0x5c/0x70
>>> [   83.066406] [<ffffffff80299ba4>] stop_machine_cpu_stop+0x104/0x1e8
>>> [   83.066406] [<ffffffff802997cc>] cpu_stopper_thread+0x110/0x1ac
>>> [   83.066406] [<ffffffff8024c940>] kthread+0x88/0x90
>>> [   83.066406] [<ffffffff80205ee4>] kernel_thread_helper+0x10/0x18
>>
>> Weird, that's from a CPU_DYING call, I thought those were with IRQs
>> disabled. 
>>
>> Look at how __stop_machine() calls the function with IRQs disabled for !
>> stop_machine_initialized or !SMP. Also stop_machine_cpu_stop() seems to
>> disabled interrupts, so how do we end up calling take_cpu_down() with
>> IRQs enabled?
> 
> The patch is no doubt wrong...
> 
> The discuss in:
> 
> https://lkml.org/lkml/2012/7/19/164
> 
> Which also faced the issue that the timer interrupt come in after apic
> was shut down, I'm not sure whether this could do help to Huacai, just
> as a clue...
>

One interesting thing that I noted in that case was that we noticed that
(stale) interrupt exactly at the call to local_irq_restore() in
stop_machine_cpu_stop().

However, as Peter pointed out, migration_call's CPU_DYING notifier runs
right in the middle of the stop machine dance, much much before the call
to local_irq_restore().. so it doesn't look like a case of a stale interrupt
being recognized.. it looks as if the sequence of local_irq_disable(),
hard_irq_disable() and __cpu_disable() somehow managed to wrongly keep the
interrupts still enabled...

Regards,
Srivatsa S. Bhat

> 
>>
>> That simply doesn't make any sense.
>>
>>> Signed-off-by: Huacai Chen <chenhc@lemote.com>
>>> ---
>>>  kernel/sched/core.c |    5 +++--
>>>  1 files changed, 3 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
>>> index 36e2666..703754a 100644
>>> --- a/kernel/sched/core.c
>>> +++ b/kernel/sched/core.c
>>> @@ -1468,9 +1468,10 @@ static void sched_ttwu_pending(void)
>>>  {
>>>  	struct rq *rq = this_rq();
>>>  	struct llist_node *llist = llist_del_all(&rq->wake_list);
>>> +	unsigned long flags;
>>>  	struct task_struct *p;
>>>  
>>> -	raw_spin_lock(&rq->lock);
>>> +	raw_spin_lock_irqsave(&rq->lock, flags);
>>>  
>>>  	while (llist) {
>>>  		p = llist_entry(llist, struct task_struct, wake_entry);
>>> @@ -1478,7 +1479,7 @@ static void sched_ttwu_pending(void)
>>>  		ttwu_do_activate(rq, p, 0);
>>>  	}
>>>  
>>> -	raw_spin_unlock(&rq->lock);
>>> +	raw_spin_unlock_irqrestore(&rq->lock, flags);
>>>  }
>>>  
>>>  void scheduler_ipi(void)
>>
>>
>> That's wrong though, you add the cost to the common case instead of the
>> hardly ever ran hotplug case.