Re: [PATCH 3/3] KVM: x86: hyperv: implement PV IPI send hypercalls

[Vitaly Kuznetsov <vkuznets@redhat.com>]

Radim Krčmář <rkrcmar@redhat.com> writes:

> 2018-06-22 16:56+0200, Vitaly Kuznetsov:
>> +
>> +		/* We fail only when APIC is disabled */
>> +		if (!kvm_apic_set_irq(vcpu, &irq, NULL))
>> +			return HV_STATUS_INVALID_HYPERCALL_INPUT;
>
> Does Windows use this even for 1 VCPU IPI?
>

It seems that it does.

> I'm thinking we could apply the same optimization we do for LAPIC -- RCU
> protected array that maps vp_index to vcpu.

Sure, both this and PV TLB flush will benefit.

>
> Thanks.
>
>> +	}
>> +
>> +ret_success:
>> +	return HV_STATUS_SUCCESS;
>> +}
>> +
>>  bool kvm_hv_hypercall_enabled(struct kvm *kvm)
>>  {
>>  	return READ_ONCE(kvm->arch.hyperv.hv_hypercall) & HV_X64_MSR_HYPERCALL_ENABLE;
>> @@ -1526,6 +1628,20 @@ int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
>>  		}
>>  		ret = kvm_hv_flush_tlb(vcpu, ingpa, rep_cnt, true);
>>  		break;
>> +	case HVCALL_SEND_IPI:
>> +		if (unlikely(rep)) {
>> +			ret = HV_STATUS_INVALID_HYPERCALL_INPUT;
>> +			break;
>> +		}
>> +		ret = kvm_hv_send_ipi(vcpu, ingpa, outgpa, false, fast);
>> +		break;
>> +	case HVCALL_SEND_IPI_EX:
>> +		if (unlikely(fast || rep)) {
>
> Now I'm getting worried that the ex can be fast as well and we'll be
> reading the banks from XMM registers. :)

Maybe but currently we don't announce 'parameters through XMM registers'
support in KVM (and neither do we support these for Linux-on-Hyper-V as
we don't usually use FPU in kernel).

-- 
  Vitaly