Re: [PATCH] KVM: X86: fix tlb_flush_guest()

From: Sean Christopherson <seanjc@google.com>
To: Paolo Bonzini <pbonzini@redhat.com>
Cc: Lai Jiangshan <jiangshanlai@gmail.com>,
	linux-kernel@vger.kernel.org,
	Lai Jiangshan <laijs@linux.alibaba.com>,
	Vitaly Kuznetsov <vkuznets@redhat.com>,
	Wanpeng Li <wanpengli@tencent.com>,
	Jim Mattson <jmattson@google.com>, Joerg Roedel <joro@8bytes.org>,
	Thomas Gleixner <tglx@linutronix.de>,
	Ingo Molnar <mingo@redhat.com>, Borislav Petkov <bp@alien8.de>,
	x86@kernel.org, "H. Peter Anvin" <hpa@zytor.com>,
	Maxim Levitsky <mlevitsk@redhat.com>,
	kvm@vger.kernel.org
Subject: Re: [PATCH] KVM: X86: fix tlb_flush_guest()
Date: Thu, 27 May 2021 16:14:36 +0000	[thread overview]
Message-ID: <YK/FbFzKhZEmI40C@google.com> (raw)
In-Reply-To: <YK/FGYejaIu6EzSn@google.com>

+Maxim for real this time...

On Thu, May 27, 2021, Sean Christopherson wrote:
> +Maxim - A proper fix for this bug might fix your shadow paging + win10 boot
>          issue, this also affects the KVM_REQ_HV_TLB_FLUSH used for HyperV PV
> 	 flushing.
> 
> On Thu, May 27, 2021, Paolo Bonzini wrote:
> > On 27/05/21 04:39, Lai Jiangshan wrote:
> > > From: Lai Jiangshan <laijs@linux.alibaba.com>
> > > 
> > > For KVM_VCPU_FLUSH_TLB used in kvm_flush_tlb_multi(), the guest expects
> > > the hypervisor do the operation that equals to native_flush_tlb_global()
> > > or invpcid_flush_all() in the specified guest CPU.
> > > 
> > > When TDP is enabled, there is no problem to just flush the hardware
> > > TLB of the specified guest CPU.
> > > 
> > > But when using shadowpaging, the hypervisor should have to sync the
> > > shadow pagetable at first before flushing the hardware TLB so that
> > > it can truely emulate the operation of invpcid_flush_all() in guest.
> > 
> > Can you explain why?
> 
> KVM's unsync logic hinges on guest TLB flushes.  For page permission modifications
> that require a TLB flush to take effect, e.g. making a writable page read-only,
> KVM waits until the guest explicitly does said flush to propagate the changes to
> the shadow page tables.  E.g. failure to sync PTEs could result in a read-only 4k
> page being writable when the guest expects it to be read-only.
> 
> > Also it is simpler to handle this in kvm_vcpu_flush_tlb_guest, using "if
> > (tdp_enabled).  This provides also a single, good place to add a comment
> > with the explanation of what invalid entries KVM_REQ_RELOAD is presenting.
> 
> Ya.  
> 
> KVM_REQ_MMU_RELOAD is overkill, nuking the shadow page tables will completely
> offset the performance gains of the paravirtualized flush.
> 
> And making a request won't work without revamping the order of request handling
> in vcpu_enter_guest(), e.g. KVM_REQ_MMU_RELOAD and KVM_REQ_MMU_SYNC are both
> serviced before KVM_REQ_STEAL_UPDATE.
> 
> Cleaning up and documenting the MMU related requests is on my todo list, but the
> immediate fix should be tiny and I can do my cleanups on top.
> 
> I believe the minimal fix is:
> 
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index 81ab3b8f22e5..b0072063f9bf 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -3072,6 +3072,9 @@ static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
>  static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
>  {
>         ++vcpu->stat.tlb_flush;
> +
> +       if (!tdp_enabled)
> +               kvm_mmu_sync_roots(vcpu);
>         static_call(kvm_x86_tlb_flush_guest)(vcpu);
>  }
>  
>