Re: [PATCH v2] cpu/hotplug: Set st->cpu earlier

[Steven Price <steven.price@arm.com>]

Thanks for taking a look at this.

On 22/03/2022 22:58, Thomas Gleixner wrote:
> On Tue, Mar 22 2022 at 15:59, Vincent Donnefort wrote:
>> On 22/03/2022 15:31, Thomas Gleixner wrote:
>>> On Wed, Mar 16 2022 at 15:36, Steven Price wrote:
>>>> Setting the 'cpu' member of struct cpuhp_cpu_state in cpuhp_create() is
>>>> too late as other callbacks can be made before that point.
>>>
>>> What?
>>>
>>>          CPUHP_OFFLINE = 0,
>>>          CPUHP_CREATE_THREADS,
>>>
>>> The create threads callback is the very first callback which is invoked
>>> for a to be plugged CPU on the control CPU. So which earlier callback
>>> can be invoked and fail?
>>>
>>> Thanks,
>>>
>>>          tglx
>>
>>
>> CPUHP_CREATE_THREADS itself can fail, before st->cpu is set.
> 
> Sure. But that does not explain the problem.
> 
>> Also, that value is used outside of the callbacks (cpuhp_set_state()
>> in _cpu_up()).
> 
> And why on earth is this not spelled out in the changelog?

I apologies for that, I'm not very familiar with the code and I have to
admit I have been struggling to identify exactly what is going on here.
The actual issue I saw was if the callback fails then the rollback code
leaves things in a messed up state. By the looks of things that callback
that fails is indeed the first (CPUHP_CREATE_THREADS).

>> But indeed this description could be refined a bit.
> 
> Indeed. But the description is not the only problem here:
> 
> It's completely uncomprehensible from the code in _cpu_up() _WHY_ this
> 
>      st->cpu = cpu;
>      
> assignment has to be there.
> 
> It's non-sensical if you really think about it, right?

I entirely agree, and I did ask in my v1 posting[1] if anyone could
point me to a better place to do the assignment. Vincent suggested
moving it earlier in _cpu_up() which is this v2.

But it still seems out-of-place to me. I've just had a go at simply
removing the 'cpu' member and it doesn't look too bad. I'll post that
patch as a follow up. I'm open to other suggestions for the best way to
fix this.

Thanks,

Steve

[1]
https://lore.kernel.org/all/20220225134918.105796-1-steven.price@arm.com/

> That said, I'm pretty sure you can come up with:
> 
>  - a proper one time initialization of @st which solves your problem
> 
>  - a proper changelog which explains it
> 
> Thanks,
> 
>         tglx