Re: PV-vNUMA issue: topology is misinterpreted by the guest

[Juergen Gross <jgross@suse.com>]

On 07/24/2015 06:09 PM, Konrad Rzeszutek Wilk wrote:
> On Fri, Jul 24, 2015 at 05:58:29PM +0200, Dario Faggioli wrote:
>> On Fri, 2015-07-24 at 17:24 +0200, Juergen Gross wrote:
>>> On 07/24/2015 05:14 PM, Juergen Gross wrote:
>>>> On 07/24/2015 04:44 PM, Dario Faggioli wrote:
>>
>>>>> In fact, I think that it is the topology, i.e., what comes from MSRs,
>>>>> that needs to adapt, and follow vNUMA, as much as possible. Do we agree
>>>>> on this?
>>>>
>>>> I think we have to be very careful here. I see two possible scenarios:
>>>>
>>>> 1) The vcpus are not pinned 1:1 on physical cpus. The hypervisor will
>>>>      try to schedule the vcpus according to their numa affinity. So they
>>>>      can change pcpus at any time in case of very busy guests. I don't
>>>>      think the linux kernel should treat the cpus differently in this
>>>>      case as it will be in vane regarding the Xen scheduler's activity.
>>>>      So we should use the "null" topology in this case.
>>>
>>> Sorry, the topology should reflect the vcpu<->numa-node relations, of
>>> course, but nothing else (so flat topolgy in each numa node).
>>>
>> Yeah, I was replying to this point saying something like this right
>> now... Luckily, I've seen this email! :-P
>>
>> With this semantic, I fully agree with this.
>>
>>>> 2) The vcpus of the guest are all pinned 1:1 to physical cpus. The Xen
>>>>      scheduler can't move vcpus between pcpus, so the linux kernel should
>>>>      see the real topology of the used pcpus in order to optimize for this
>>>>      picture.
>>>>
>>>
>> Mmm... I did think about this too, but I'm not sure. I see the value of
>> this of course, and the reason why it makes sense. However, pinning can
>> change on-line, via `xl vcpu-pin' and stuff. Also migration could make
>> things less certain, I think. What happens if we build on top of the
>> initial pinning, and then things change?
>>
>> To be fair, there is stuff building on top of the initial pinning
>> already, e.g., from which physical NUMA node we allocate the memory
>> relies depends exactly on that. That being said, I'm not sure I'm
>> comfortable with adding more of this...
>>
>> Perhaps introduce an 'immutable_pinning' flag, which will prevent
>> affinity to be changed, and then bind the topology to pinning only if
>> that one is set?
>>
>>>>> Maybe, there is room for "fixing" this at this level, hooking up inside
>>>>> the scheduler code... but I'm shooting in the dark, without having check
>>>>> whether and how this could be really feasible, should I?
>>>>
>>>> Uuh, I don't think a change of the scheduler on behalf of Xen is really
>>>> appreciated. :-)
>>>>
>> I'm sure it would (have been! :-)) a true and giant nightmare!! :-D
>>
>>>>> One thing I don't like about this approach is that it would potentially
>>>>> solve vNUMA and other scheduling anomalies, but...
>>>>>
>>>>>> cpuid instruction is available for user mode as well.
>>>>>>
>>>>> ...it would not do any good for other subsystems, and user level code
>>>>> and apps.
>>>>
>>>> Indeed. I think the optimal solution would be two-fold: give the
>>>> scheduler the information it is needing to react correctly via a
>>>> kernel patch not relying on cpuid values and fiddle with the cpuid
>>>> values from xen tools according to any needs of other subsystems and/or
>>>> user code (e.g. licensing).
>>>
>> So, just to check if I'm understanding is correct: you'd like to add an
>> abstraction layer, in Linux, like in generic (or, perhaps, scheduling)
>> code, to hide the direct interaction with CPUID.
>> Such layer, on baremetal, would just read CPUID while, on PV-ops, it'd
>> check with Xen/match vNUMA/whatever... Is this that you are saying?
>>
>> If yes, I think I like it...
>
> I don't think this is workable. For example there are applications
> which use 'cpuid' and figure out the core/thread and use it for its own
> scheduling purposes.

Might be, yes.

The pure cpuid solution won't work for all license related issues.

Doing it via an abstraction layer in the kernel would work in more than
90% of all cases AND would still enable a user to fiddle cpuids
according to his needs (either topology or license).

I'd rather have an out-of-the-box kernel solution with special user
requirements handling than a complex solution making some user
requirements impossible to meet.


Juergen