Re: [PATCH v5 UPDATEDv3 3/3] CPU hotplug, smp: Flush any pending IPI callbacks before CPU offline

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

On 05/20/2014 04:08 PM, Srivatsa S. Bhat wrote:
> On 05/20/2014 04:01 PM, Srivatsa S. Bhat wrote:
>> On 05/20/2014 03:55 PM, Peter Zijlstra wrote:
>>> On Tue, May 20, 2014 at 03:39:59PM +0530, Srivatsa S. Bhat wrote:
>>>>> The multi_cpu_stop() path isn't exclusive to hotplug, so your changelog
>>>>> is wrong or the patch is.
>>>>>
>>>>
>>>> Yes, I know that multi_cpu_stop() isn't exclusive to hotplug. That's why
>>>> I have explicitly referred to CPU hotplug in the comment as well as the
>>>> changelog.
>>>>
>>>> But I totally agree that code-wise this is not the best way to do it since
>>>> this affects (although harmlessly) usecases other than hotplug as well.
>>>>
>>>> Do you have any other suggestions?
>>>
>>> How about making a kernel/smp.c hotplug notifier and stuffing it in the
>>> CPU_DYING list? That's typically after we've already torn down the
>>> interrupts for that cpu, so no chance of any new ones coming in.
>>>
>>> Or is that too late?
>>>
>>
>> No, that should work just fine. Thank you for the suggestion! I'll give
>> it a shot.
>>
> 
> The only problem will be that CPU_DYING notifiers are run after marking
> the CPU offline, and hence the warning will trigger. We can avoid that by
> defining a __generic_smp_call_function_single_interrupt() that doesn't
> check for cpu_online(smp_processor_id) and call this function from the
> hotplug notifier.
> 
> 

(And that's probably something to fix while cleaning up hotplug later:
Why should we mark the CPU offline _before_ running CPU_DYING? It makes
more sense to mark it offline _after_ running CPU_DYING notifiers. I'll
audit that path as well and see what I can find).
 
Regards,
Srivatsa S. Bhat