Received-SPF: pass (google.com: domain of linux-kernel+bounces-61311-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45d1:ec00::1 as permitted sender) client-ip=2604:1380:45d1:ec00::1;
Precedence: bulk
MIME-Version: 1.0
References: <cover.1705965634.git.isaku.yamahata@intel.com> <cce34006a4db0e1995ce007c917f834b117b12af.1705965635.git.isaku.yamahata@intel.com>
In-Reply-To: <cce34006a4db0e1995ce007c917f834b117b12af.1705965635.git.isaku.yamahata@intel.com>
From: Paolo Bonzini <pbonzini@redhat.com>
Date: Mon, 12 Feb 2024 11:29:51 +0100
Message-ID: <CABgObfbZRO3yiXoHAoHSsBp4sKQY9r4GTLt-SRqevz2c8wOqbQ@mail.gmail.com>
Subject: Re: [PATCH v18 044/121] KVM: x86/mmu: Assume guest MMIOs are shared
To: isaku.yamahata@intel.com
Cc: kvm@vger.kernel.org, linux-kernel@vger.kernel.org, 
	isaku.yamahata@gmail.com, erdemaktas@google.com, 
	Sean Christopherson <seanjc@google.com>, Sagi Shahar <sagis@google.com>, Kai Huang <kai.huang@intel.com>, 
	chen.bo@intel.com, hang.yuan@intel.com, tina.zhang@intel.com, 
	Chao Gao <chao.gao@intel.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Tue, Jan 23, 2024 at 12:55=E2=80=AFAM <isaku.yamahata@intel.com> wrote:
>
> From: Chao Gao <chao.gao@intel.com>
>
> Guest TD doesn't necessarily invoke MAP_GPA to convert the virtual MMIO
> range to shared before accessing it.  When TD tries to access the virtual
> device's MMIO as shared, an EPT violation is raised first.
> kvm_mem_is_private() checks whether the GFN is shared or private.  If
> MAP_GPA is not called for the GPA, KVM thinks the GPA is private and
> refuses shared access, and doesn't set up shared EPT entry.  The guest
> can't make progress.
>
> Instead of requiring the guest to invoke MAP_GPA for regions of virtual
> MMIOs assume regions of virtual MMIOs are shared in KVM as well (i.e., GP=
As
> either have no kvm_memory_slot or are backed by host MMIOs). So that gues=
ts
> can access those MMIO regions.

I'm not sure how the patch below deals with host MMIOs?

> Signed-off-by: Chao Gao <chao.gao@intel.com>

Missing Signed-off-by.

Also, this patch conflicts with "[PATCH v11 09/35] KVM: x86: Determine
shared/private faults based on vm_type".  I think in general the logic
in that patch (which forces an exit to userspace if needed, to convert
the MMIO area to shared) can be applied to sw-protected and TDX guests
as well. I'm preparing a set of common patches that can be applied for
6.9 and will include something after testing with sw-protected VMs.

Paolo


> ---
>  arch/x86/kvm/mmu/mmu.c | 7 ++++++-
>  1 file changed, 6 insertions(+), 1 deletion(-)
>
> diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c
> index e93bc16a5e9b..583ae9d6bf5d 100644
> --- a/arch/x86/kvm/mmu/mmu.c
> +++ b/arch/x86/kvm/mmu/mmu.c
> @@ -4371,7 +4371,12 @@ static int __kvm_faultin_pfn(struct kvm_vcpu *vcpu=
, struct kvm_page_fault *fault
>                         return RET_PF_EMULATE;
>         }
>
> -       if (fault->is_private !=3D kvm_mem_is_private(vcpu->kvm, fault->g=
fn)) {
> +       /*
> +        * !fault->slot means MMIO.  Don't require explicit GPA conversio=
n for
> +        * MMIO because MMIO is assigned at the boot time.
> +        */
> +       if (fault->slot &&
> +           fault->is_private !=3D kvm_mem_is_private(vcpu->kvm, fault->g=
fn)) {
>                 if (vcpu->kvm->arch.vm_type =3D=3D KVM_X86_SW_PROTECTED_V=
M)
>                         return RET_PF_RETRY;
>                 kvm_mmu_prepare_memory_fault_exit(vcpu, fault);
> --
> 2.25.1
>