Re: [PATCH] KVM: s390: Enable specification exception interpretation

[Janis Schoetterl-Glausch <scgl@linux.vnet.ibm.com>]

On 7/6/21 5:16 PM, David Hildenbrand wrote:
> On 06.07.21 14:02, Christian Borntraeger wrote:
>>
>>
>> On 06.07.21 13:59, David Hildenbrand wrote:
>>> On 06.07.21 13:56, Christian Borntraeger wrote:
>>>>
>>>>
>>>> On 06.07.21 13:52, Cornelia Huck wrote:
>>>>> On Tue, Jul 06 2021, Janis Schoetterl-Glausch <scgl@linux.ibm.com> wrote:
>>>>>
>>>>>> When this feature is enabled the hardware is free to interpret
>>>>>> specification exceptions generated by the guest, instead of causing
>>>>>> program interruption interceptions.
>>>>>>
>>>>>> This benefits (test) programs that generate a lot of specification
>>>>>> exceptions (roughly 4x increase in exceptions/sec).
>>>>>>
>>>>>> Interceptions will occur as before if ICTL_PINT is set,
>>>>>> i.e. if guest debug is enabled.
>>>>>>
>>>>>> Signed-off-by: Janis Schoetterl-Glausch <scgl@linux.ibm.com>
>>>>>> ---
>>>>>> I'll additionally send kvm-unit-tests for testing this feature.
>>>>>>
>>>>>>     arch/s390/include/asm/kvm_host.h | 1 +
>>>>>>     arch/s390/kvm/kvm-s390.c         | 2 ++
>>>>>>     arch/s390/kvm/vsie.c             | 2 ++
>>>>>>     3 files changed, 5 insertions(+)
>>>>>
>>>>> (...)
>>>>>
>>>>>> diff --git a/arch/s390/kvm/kvm-s390.c b/arch/s390/kvm/kvm-s390.c
>>>>>> index b655a7d82bf0..aadd589a3755 100644
>>>>>> --- a/arch/s390/kvm/kvm-s390.c
>>>>>> +++ b/arch/s390/kvm/kvm-s390.c
>>>>>> @@ -3200,6 +3200,8 @@ static int kvm_s390_vcpu_setup(struct kvm_vcpu *vcpu)
>>>>>>             vcpu->arch.sie_block->ecb |= ECB_SRSI;
>>>>>>         if (test_kvm_facility(vcpu->kvm, 73))
>>>>>>             vcpu->arch.sie_block->ecb |= ECB_TE;
>>>>>> +    if (!kvm_is_ucontrol(vcpu->kvm))
>>>>>> +        vcpu->arch.sie_block->ecb |= ECB_SPECI;
>>>>>
>>>>> Does this exist for any hardware version (i.e. not guarded by a cpu
>>>>> feature?)
>>>>
>>>> Not for all hardware versions, but also no indication. The architecture
>>>> says that the HW is free to do this or not. (which makes the vsie code
>>>> simpler).
>>>
>>> I remember the architecture said at some point to never set undefined bits - and this bit is undefined on older HW generations. I might be wrong, though.
>>
>> I can confirm that this bit will be ignored on older machines. The notion of
>> never setting undefined bits comes from "you never know what this bit will
>> change in future machines". Now we know :-)
> 
> Well, okay then :)
> 
> So the plan for vSIE is to always keep it disabled? IIUC, one could similarly always forward the bit of set.

The bit does get copied for vSIE.