On 05/13/2018 03:33 PM, Liran Alon wrote:
>> @@ -5154,11 +5180,18 @@ void kvm_mmu_invpcid_gva(struct kvm_vcpu
>> *vcpu, gva_t gva, unsigned long pcid)
>> kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
>> }
>>
>> + if (VALID_PAGE(mmu->prev_root_hpa) &&
>> + pcid == kvm_get_pcid(vcpu, mmu->prev_cr3)) {
>> + mmu->invlpg(vcpu, gva, mmu->prev_root_hpa);
>> + kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
>> + }
>> +
>
> I think that KVM_REQ_TLB_FLUSH was needed for prev PCID as part of patch 8.
> This will solve the issue I mentioned there of breaking INVPCID handling.
>
Yes, you are right. I had refactored the patches a bit from the initial
implementation and failed to notice that this flush was needed in the earlier
patch (along with one for the SINGLE_CTXT case). Thank you for catching this.
Thanks,
Junaid
