Re: [PATCH 2/6] irq_work: Implement remote queueing

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

On 06/25/2014 03:20 PM, Srivatsa S. Bhat wrote:
> On 06/25/2014 03:09 PM, Peter Zijlstra wrote:
>> On Wed, Jun 25, 2014 at 11:36:57AM +0200, Peter Zijlstra wrote:
>>> On Wed, Jun 25, 2014 at 12:07:05PM +0530, Srivatsa S. Bhat wrote:
>>>> I don't think irqs_disabled() is the problematic condition, since
>>>> hotplug_cfg() invokes irq_work_run() from CPU_DYING context (which has
>>>> irqs disabled). I guess you meant to remove the in_irq() check inside
>>>> irq_work_run() instead?
>>>
>>> Yes, clearly I should not get up at 6am.. :-) Let me go do a new one.
>>
>> ---
>> Subject: irq_work: Remove BUG_ON in irq_work_run()
>> From: Peter Zijlstra <peterz@infradead.org>
>> Date: Wed Jun 25 07:13:07 CEST 2014
>>
>> Because of a collision with 8d056c48e486 ("CPU hotplug, smp: flush any
>> pending IPI callbacks before CPU offline"), which ends up calling
>> hotplug_cfd()->flush_smp_call_function_queue()->irq_work_run(), which
>> is not from IRQ context.
>>
>> Cc: Frederic Weisbecker <fweisbec@gmail.com>
>> Reported-by: Stephen Warren <swarren@wwwdotorg.org>
>> Signed-off-by: Peter Zijlstra <peterz@infradead.org>
>> Link: http://lkml.kernel.org/n/tip-busatzs2gvz4v62258agipuf@git.kernel.org
>> ---
>>  kernel/irq_work.c |   12 +-----------
>>  1 file changed, 1 insertion(+), 11 deletions(-)
>>
>> Index: linux-2.6/kernel/irq_work.c
>> ===================================================================
>> --- linux-2.6.orig/kernel/irq_work.c
>> +++ linux-2.6/kernel/irq_work.c
>> @@ -160,21 +160,11 @@ static void irq_work_run_list(struct lli
>>  	}
>>  }
>>
>> -static void __irq_work_run(void)
> 
> Hmm, irq_work_cpu_notify() calls __irq_work_run() in the CPU_DYING
> phase, to by-pass BUG_ON(!in_irq()). How about doing the same thing
> from hotplug_cfd() as well?
> 

Wait, that was a stupid idea. hotplug_cfd() already invokes irq_work_run
indirectly via flush_smp_call_function_queue(). So irq_work_cpu_notify()
doesn't need to invoke it again, AFAIU. So perhaps we can get rid of
irq_work_cpu_notify() altogether?

Regards,
Srivatsa S. Bhat

>> +static void irq_work_run(void)
>>  {
>>  	irq_work_run_list(&__get_cpu_var(raised_list));
>>  	irq_work_run_list(&__get_cpu_var(lazy_list));
>>  }
>> -
>> -/*
>> - * Run the irq_work entries on this cpu. Requires to be ran from hardirq
>> - * context with local IRQs disabled.
>> - */
>> -void irq_work_run(void)
>> -{
>> -	BUG_ON(!in_irq());
>> -	__irq_work_run();
>> -}
>>  EXPORT_SYMBOL_GPL(irq_work_run);
>>
>>  /*
>>
>