Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Thu, 9 Sep 2021 17:09:11 +0000
From:   Sean Christopherson <seanjc@google.com>
To:     Yu Zhang <yu.c.zhang@linux.intel.com>
Cc:     Hou Wenlong <houwenlong93@linux.alibaba.com>, kvm@vger.kernel.org,
        Paolo Bonzini <pbonzini@redhat.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>,
        "maintainer:X86 ARCHITECTURE (32-BIT AND 64-BIT)" <x86@kernel.org>,
        "H. Peter Anvin" <hpa@zytor.com>,
        Jan Kiszka <jan.kiszka@siemens.com>,
        Avi Kivity <avi@redhat.com>,
        "open list:X86 ARCHITECTURE (32-BIT AND 64-BIT)" 
        <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH v2 1/3] kvm: x86: Introduce hypercall x86 ops for
 handling hypercall not in cpl0
Message-ID: <YTo/t4G1iI28oDmk@google.com>
References: <cover.1631188011.git.houwenlong93@linux.alibaba.com>
 <04a337801ad5aaa54144dc57df8ee2fc32bc9c4e.1631188011.git.houwenlong93@linux.alibaba.com>
 <20210909163901.2vvozmkuxjcgabs5@linux.intel.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20210909163901.2vvozmkuxjcgabs5@linux.intel.com>
Precedence: bulk

On Fri, Sep 10, 2021, Yu Zhang wrote:
> On Thu, Sep 09, 2021 at 07:55:23PM +0800, Hou Wenlong wrote:
> > Per Intel's SDM, use vmcall instruction in non VMX operation for cpl3
> > it should trigger a #UD. And in VMX root operation, it should
> 
> Are you sure? IIRC, vmcall will always cause VM exit as long as CPU
> is in non-root mode(regardless the CPL).

Correct, VMCALL unconditionally causes VM-Exit in non-root mode, but Hou is
referring to the first fault condition of "non VMX operation".  The intent of the
patch is to emulate hardware behavior for CPL>0: if L1 is not in VMX operation,
a.k.a. not post-VMXON, then #UD, else #GP (because VMCALL #GPs at CPL>0 in VMX
root).

On one hand, I agree with Hou's logic; injecting #UD/#GP is architecturally
correct if KVM is emulating a bare metal environment for the guest.  On the
other hand, that contradicts with KVM _not_ injecting #UD for guest CPL0, i.e.
KVM is clearly not emulating a bare metal environment.

In the end, this would represent an ABI change for guest CPL>0.  While it's highly
unlikely that such a change would cause problems, maintaining the current behavior
is the safe option unless there's strong motivation for changing the guest ABI.

And injecting #UD/#GP would also mean KVM would again have to change its ABI if
there is a future hypercall KVM wants to allow at CPL>0.  Again, that's unlikely,
but again I don't see sufficient justification.