Re: [PATCH 1/2] KVM: x86: Fix KVM_CAP_SYNC_REGS's sync_regs() TOCTOU issues

[Michal Luczaj <mhal@xxxxxxx>]

On 8/4/23 11:53, Paolo Bonzini wrote:
> On 8/3/23 23:15, Michal Luczaj wrote:
>>>           *mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
>>>
>>> with a call to the function just before __set_sregs_common returns.
>> What about kvm_post_set_cr4() then? Should it be introduced to
>> __set_sregs_common() as well?
> 
> Yes, indeed, but it starts getting a bit unwieldy.
> 
> If we decide not to particularly optimize KVM_SYNC_X86_SREGS, however, 
> we can just chuck a KVM_REQ_TLB_FLUSH_GUEST request after __set_sregs 
> and __set_sregs2 call kvm_mmu_reset_context().

Something like this?

@@ -11562,8 +11562,10 @@ static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
        if (ret)
                return ret;

-       if (mmu_reset_needed)
+       if (mmu_reset_needed) {
                kvm_mmu_reset_context(vcpu);
+               kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
+       }

        max_bits = KVM_NR_INTERRUPTS;
        pending_vec = find_first_bit(
@@ -11604,8 +11606,10 @@ static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
                mmu_reset_needed = 1;
                vcpu->arch.pdptrs_from_userspace = true;
        }
-       if (mmu_reset_needed)
+       if (mmu_reset_needed) {
                kvm_mmu_reset_context(vcpu);
+               kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
+       }
        return 0;
 }