[libvirt] [PATCH] qemu: caps: use model "qemu" for s390 tcg cpu-model-expansion

[David Hildenbrand]

On 25.07.2018 01:02, Collin Walling wrote:
> On 07/24/2018 12:59 PM, Daniel P. Berrangé wrote:
>> On Tue, Jul 24, 2018 at 06:53:54PM +0200, Cornelia Huck wrote:
>>> On Tue, 24 Jul 2018 18:08:21 +0200
>>> Jiri Denemark <jdenemar at redhat.com> wrote:
>>>
>>>> On Mon, Jul 23, 2018 at 17:45:33 -0400, Collin Walling wrote:
>>>>> Use model name "qemu" instead of "max" when calling
>>>>> query-cpu-model-expansion for s390 on tcg.
>>>>>
>>>>> Signed-off-by: Collin Walling <walling at linux.ibm.com>
>>>>> ---
>>>>>  src/qemu/qemu_capabilities.c | 5 ++++-
>>>>>  1 file changed, 4 insertions(+), 1 deletion(-)
>>>>>
>>>>> diff --git a/src/qemu/qemu_capabilities.c b/src/qemu/qemu_capabilities.c
>>>>> index 23b4833..e9b44cc 100644
>>>>> --- a/src/qemu/qemu_capabilities.c
>>>>> +++ b/src/qemu/qemu_capabilities.c
>>>>> @@ -2356,7 +2356,10 @@ virQEMUCapsProbeQMPHostCPU(virQEMUCapsPtr qemuCaps,
>>>>>  
>>>>>      if (tcg || !virQEMUCapsGet(qemuCaps, QEMU_CAPS_KVM)) {
>>>>>          virtType = VIR_DOMAIN_VIRT_QEMU;
>>>>> -        model = "max";
>>>>> +        if (ARCH_IS_S390(qemuCaps->arch))
>>>>> +            model = "qemu";
>>>>> +        else
>>>>> +            model = "max";  
>>>>
>>>> I think we should also check if "max" is a supported model
>>>> (qemuCaps->tcgCPUModels is already populated at this point) and only use
>>>> "qemu" on s390 if "max" is not supported. And please, report the issue
>>>> to QEMU developers since one of the reasons behind "max" is its
>>>> universal availability on everywhere CPU model expansion is supported.
>>>
>>> Hm, can you point me to that discussion? A quick search through the
>>> QEMU log gives me the addition of the "max" model on i386 as a
>>> replacement to the "host" model for !kvm, but nothing about it being
>>> universal...
>>
>> I don't recall the link but "max" is supposed to be the standard shorthand
>> for "enable all the features supported by the virt type". IOW, 'max' should
>> work both KVM and TCG ("host" was only well defined for KVM), and across
>> all architectures.
>>
>> So having to use a different name on s390 is a bug in QEMU imho.
>>
>> Regards,
>> Daniel
>>
> 
> Thanks for expanding on what the "max" model name is suppose to be. I wonder if a 
> s/"qemu"/"max" in QEMU would suffice (I'm taking a shot in the dark here.)

Nope, it dynamically has to map to qemu/host depending on the
accelerator. But it also has to be a valid QOM object ("max-s390x-cpu").

> 
> @Connie, @David, you both are far more knowledgeable in this area than I am. What
> do either of you suggest for moving forward with this? Should we forward this
> discussion on qemu-devel?
> 

I can have a look if nobody else wants to tackle it.


-- 

Thanks,

David / dhildenb