On 10.10.24 10:57, Vitaly Kuznetsov wrote:
> Nikolas Wipper <[email protected]> writes:
>> diff --git a/arch/x86/include/asm/kvm_host.h
>> b/arch/x86/include/asm/kvm_host.h
>> index 46e0a466d7fb..7571ac578884 100644
>> --- a/arch/x86/include/asm/kvm_host.h
>> +++ b/arch/x86/include/asm/kvm_host.h
>> @@ -695,6 +695,9 @@ struct kvm_vcpu_hv {
>> u64 vm_id;
>> u32 vp_id;
>> } nested;
>> +
>> + bool suspended;
>> + int waiting_on;
>
> I don't quite understand why we need 'suspended' at all. Isn't it always
> suspended when 'waiting_on != -1'? I can see we always update these two
> in pair.
>
This is mainly for future proofing the implementation. You are right, this
is currently not required, but it's nice to have a single flags, so that
when the suspended state is used in a different context, the whole logic
surrounding it still works.
> Also, I would suggest we use a more descriptive
> name. 'waiting_on_vcpu_id', for example.
>
Sounds good.
>> };
>>
>> struct kvm_hypervisor_cpuid {
>> diff --git a/arch/x86/kvm/hyperv.c b/arch/x86/kvm/hyperv.c
>> index 4f0a94346d00..6e7941ed25ae 100644
>> --- a/arch/x86/kvm/hyperv.c
>> +++ b/arch/x86/kvm/hyperv.c
>> @@ -971,6 +971,7 @@ int kvm_hv_vcpu_init(struct kvm_vcpu *vcpu)
>>
>> vcpu->arch.hyperv = hv_vcpu;
>> hv_vcpu->vcpu = vcpu;
>> + hv_vcpu->waiting_on = -1;
>>
>> synic_init(&hv_vcpu->synic);
>>
>> @@ -2915,3 +2916,32 @@ int kvm_get_hv_cpuid(struct kvm_vcpu *vcpu, struct
>> kvm_cpuid2 *cpuid,
>>
>> return 0;
>> }
>> +
>> +void kvm_hv_vcpu_suspend_tlb_flush(struct kvm_vcpu *vcpu, int vcpu_id)
>
> Can we make parameter's name 'waiting_on_vcpu_id' as well? Because as-is
> I'm getting confused which CPU of these two is actually getting
> suspended)
>
Yup, that would certainly help readability.
> Also, why do we need '_tlb_flush' in the name? The mechanism seems to be
> fairly generic, it's just that we use it for TLB flushes.
>
The 'waiting_on' part is TLB flushing specific.
>> +{
>> + /* waiting_on's store should happen before suspended's */
>> + WRITE_ONCE(vcpu->arch.hyperv->waiting_on, vcpu_id);
>> + WRITE_ONCE(vcpu->arch.hyperv->suspended, true);
>> +}
>> +
>> +void kvm_hv_vcpu_unsuspend_tlb_flush(struct kvm_vcpu *vcpu)
>
> And here someone may expect this means 'unsuspend vcpu' but in reality
> this means 'unsuspend all vCPUs which are waiting on 'vcpu'). I guess we
> need a rename. How about
>
> void kvm_hv_unsuspend_vcpus(struct kvm_vcpu *waiting_on_vcpu)
>
> ?
>
Also sounds good.
>> +{
>> + DECLARE_BITMAP(vcpu_mask, KVM_MAX_VCPUS);
>> + struct kvm_vcpu_hv *vcpu_hv;
>> + struct kvm_vcpu *v;
>> + unsigned long i;
>> +
>> + kvm_for_each_vcpu(i, v, vcpu->kvm) {
>> + vcpu_hv = to_hv_vcpu(v);
>> +
>> + if (kvm_hv_vcpu_suspended(v) &&
>> + READ_ONCE(vcpu_hv->waiting_on) == vcpu->vcpu_id) {
>> + /* waiting_on's store should happen before suspended's
>> */
>> + WRITE_ONCE(v->arch.hyperv->waiting_on, -1);
>> + WRITE_ONCE(v->arch.hyperv->suspended, false);
>> + __set_bit(i, vcpu_mask);
>> + }
>> + }
>> +
>> + kvm_make_vcpus_request_mask(vcpu->kvm, KVM_REQ_EVENT, vcpu_mask);
>> +}
>> diff --git a/arch/x86/kvm/hyperv.h b/arch/x86/kvm/hyperv.h
>> index 913bfc96959c..a55832cea221 100644
>> --- a/arch/x86/kvm/hyperv.h
>> +++ b/arch/x86/kvm/hyperv.h
>> @@ -265,6 +265,15 @@ static inline void
>> kvm_hv_nested_transtion_tlb_flush(struct kvm_vcpu *vcpu,
>> }
>>
>> int kvm_hv_vcpu_flush_tlb(struct kvm_vcpu *vcpu);
>> +
>> +static inline bool kvm_hv_vcpu_suspended(struct kvm_vcpu *vcpu)
>> +{
>> + return vcpu->arch.hyperv_enabled &&
>> + READ_ONCE(vcpu->arch.hyperv->suspended);
>
> I don't think READ_ONCE() means anything here, does it?
>
It does prevent compiler optimisations and is actually required[1]. Also
it makes clear that this variable is shared, and may be accessed from
remote CPUs.
[1]
https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2018/p0124r6.html#Variable%20Access
Nikolas
