Re: [PATCH v3 2/3] s390x: KVM: Implementation of Multiprocessor Topology-Change-Report

[Pierre Morel <pmorel@linux.ibm.com>]

On 9/6/21 8:37 PM, David Hildenbrand wrote:
> On 03.08.21 10:26, Pierre Morel wrote:
>> We let the userland hypervisor know if the machine support the CPU
>> topology facility using a new KVM capability: KVM_CAP_S390_CPU_TOPOLOGY.
>>
>> The PTF instruction will report a topology change if there is any change
>> with a previous STSI_15_2 SYSIB.
>> Changes inside a STSI_15_2 SYSIB occur if CPU bits are set or clear
>> inside the CPU Topology List Entry CPU mask field, which happens with
>> changes in CPU polarization, dedication, CPU types and adding or
>> removing CPUs in a socket.
>>
>> The reporting to the guest is done using the Multiprocessor
>> Topology-Change-Report (MTCR) bit of the utility entry of the guest's
>> SCA which will be cleared during the interpretation of PTF.
>>
>> To check if the topology has been modified we use a new field of the
>> arch vCPU to save the previous real CPU ID at the end of a schedule
>> and verify on next schedule that the CPU used is in the same socket.
>>
>> We deliberatly ignore:
>> - polarization: only horizontal polarization is currently used in linux.
>> - CPU Type: only IFL Type are supported in Linux
>> - Dedication: we consider that only a complete dedicated CPU stack can
>>    take benefit of the CPU Topology.
>>
>> Signed-off-by: Pierre Morel <pmorel@linux.ibm.com>
> 
> 
>> @@ -228,7 +232,7 @@ struct kvm_s390_sie_block {
>>       __u8    icptcode;        /* 0x0050 */
>>       __u8    icptstatus;        /* 0x0051 */
>>       __u16    ihcpu;            /* 0x0052 */
>> -    __u8    reserved54;        /* 0x0054 */
>> +    __u8    mtcr;            /* 0x0054 */
>>   #define IICTL_CODE_NONE         0x00
>>   #define IICTL_CODE_MCHK         0x01
>>   #define IICTL_CODE_EXT         0x02
>> @@ -246,6 +250,7 @@ struct kvm_s390_sie_block {
>>   #define ECB_TE        0x10
>>   #define ECB_SRSI    0x04
>>   #define ECB_HOSTPROTINT    0x02
>> +#define ECB_PTF        0x01
> 
>  From below I understand, that ECB_PTF can be used with stfl(11) in the 
> hypervisor.
> 
> What is to happen if the hypervisor doesn't support stfl(11) and we 
> consequently cannot use ECB_PTF? Will QEMU be able to emulate PTF fully?

Yes.

> 
> 
>>       __u8    ecb;            /* 0x0061 */
>>   #define ECB2_CMMA    0x80
>>   #define ECB2_IEP    0x20
>> @@ -747,6 +752,7 @@ struct kvm_vcpu_arch {
>>       bool skey_enabled;
>>       struct kvm_s390_pv_vcpu pv;
>>       union diag318_info diag318_info;
>> +    int prev_cpu;
>>   };
>>   struct kvm_vm_stat {
>> diff --git a/arch/s390/kvm/kvm-s390.c b/arch/s390/kvm/kvm-s390.c
>> index b655a7d82bf0..ff6d8a2b511c 100644
>> --- a/arch/s390/kvm/kvm-s390.c
>> +++ b/arch/s390/kvm/kvm-s390.c
>> @@ -568,6 +568,7 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, 
>> long ext)
>>       case KVM_CAP_S390_VCPU_RESETS:
>>       case KVM_CAP_SET_GUEST_DEBUG:
>>       case KVM_CAP_S390_DIAG318:
>> +    case KVM_CAP_S390_CPU_TOPOLOGY:
> 
> I would have expected instead
> 
> r = test_facility(11);
> break

The idea is that QEMU will emulate both PTF and SYSIB_15 in this case.

> 
> ...
> 
>>           r = 1;
>>           break;
>>       case KVM_CAP_SET_GUEST_DEBUG2:
>> @@ -819,6 +820,23 @@ int kvm_vm_ioctl_enable_cap(struct kvm *kvm, 
>> struct kvm_enable_cap *cap)
>>           icpt_operexc_on_all_vcpus(kvm);
>>           r = 0;
>>           break;
>> +    case KVM_CAP_S390_CPU_TOPOLOGY:
>> +        mutex_lock(&kvm->lock);
>> +        if (kvm->created_vcpus) {
>> +            r = -EBUSY;
>> +        } else {
> 
> ...
> } else if (test_facility(11)) {
>      set_kvm_facility(kvm->arch.model.fac_mask, 11);
>      set_kvm_facility(kvm->arch.model.fac_list, 11);
>      r = 0;
> } else {
>      r = -EINVAL;
> }
> 
> similar to how we handle KVM_CAP_S390_VECTOR_REGISTERS.
> 
> But I assume you want to be able to support hosts without ECB_PTF, correct?

yes, this was the idea.

> 
> 
>> +            set_kvm_facility(kvm->arch.model.fac_mask, 11);
>> +            set_kvm_facility(kvm->arch.model.fac_list, 11);
>> +            r = 0;
>> +        }
>> +        mutex_unlock(&kvm->lock);
>> +        VM_EVENT(kvm, 3, "ENABLE: CPU TOPOLOGY %s",
>> +             r ? "(not available)" : "(success)");
>> +        break;
>> +
>> +        r = -EINVAL;
>> +        break;
> 
> ^ dead code
> 

:) indeed , sorry.

> [...]
> 
>>   }
>>   void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
>>   {
>> +    vcpu->arch.prev_cpu = vcpu->cpu;
>>       vcpu->cpu = -1;
>>       if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
>>           __stop_cpu_timer_accounting(vcpu);
>> @@ -3198,6 +3239,11 @@ static int kvm_s390_vcpu_setup(struct kvm_vcpu 
>> *vcpu)
>>           vcpu->arch.sie_block->ecb |= ECB_HOSTPROTINT;
>>       if (test_kvm_facility(vcpu->kvm, 9))
>>           vcpu->arch.sie_block->ecb |= ECB_SRSI;
>> +
>> +    /* PTF needs both host and guest facilities to enable 
>> interpretation */
>> +    if (test_kvm_facility(vcpu->kvm, 11) && test_facility(11))
>> +        vcpu->arch.sie_block->ecb |= ECB_PTF;
> 
> Here you say we need both ...

Yes because for interpretation we need both.
But if PTF is not interpreted we will emulate it in QEMU.

> 
>> +
>>       if (test_kvm_facility(vcpu->kvm, 73))
>>           vcpu->arch.sie_block->ecb |= ECB_TE;
>> diff --git a/arch/s390/kvm/vsie.c b/arch/s390/kvm/vsie.c
>> index 4002a24bc43a..50d67190bf65 100644
>> --- a/arch/s390/kvm/vsie.c
>> +++ b/arch/s390/kvm/vsie.c
>> @@ -503,6 +503,9 @@ static int shadow_scb(struct kvm_vcpu *vcpu, 
>> struct vsie_page *vsie_page)
>>       /* Host-protection-interruption introduced with ESOP */
>>       if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_ESOP))
>>           scb_s->ecb |= scb_o->ecb & ECB_HOSTPROTINT;
>> +    /* CPU Topology */
>> +    if (test_kvm_facility(vcpu->kvm, 11))
>> +        scb_s->ecb |= scb_o->ecb & ECB_PTF;
> 
> but here you don't check?

Arrrg, yes, this is false, we must check both here too.

> 
>>       /* transactional execution */
>>       if (test_kvm_facility(vcpu->kvm, 73) && wants_tx) {
>>           /* remap the prefix is tx is toggled on */
>> diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h
>> index d9e4aabcb31a..081ce0cd44b9 100644
>> --- a/include/uapi/linux/kvm.h
>> +++ b/include/uapi/linux/kvm.h
>> @@ -1112,6 +1112,7 @@ struct kvm_ppc_resize_hpt {
>>   #define KVM_CAP_BINARY_STATS_FD 203
>>   #define KVM_CAP_EXIT_ON_EMULATION_FAILURE 204
>>   #define KVM_CAP_ARM_MTE 205
>> +#define KVM_CAP_S390_CPU_TOPOLOGY 206
> 
> We'll need a Documentation/virt/kvm/api.rst description.
> 
> I'm not completely confident that the way we're handling the 
> capability+facility is the right approach. It all feels a bit suboptimal.
> 
> Except stfl(74) -- STHYI --, we never enable a facility via 
> set_kvm_facility() that's not available in the host. And STHYI is 
> special such that it is never implemented in hardware.

Then we can fall back to KVM_facility + in kernel emulation but if for 
PTF it will be quite simple, for STSI_15 it will be much bigger.

> 
> I'll think about what might be cleaner once I get some more details 
> about the interaction with stfl(11) in the hypervisor.
> 

And I just saw I for an unknown reason forgot two patches in the QEMU 
series:

s390x: kvm: make topology change report pending
s390x: kvm: enable CPU Topology Function

So I will publish a new QEMU series this afternoon with the comments 
from Thomas.

thanks,
Pierre

-- 
Pierre Morel
IBM Lab Boeblingen