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;
MIME-Version: 1.0
References: <20210218100450.2157308-1-david.edmondson@oracle.com>
 <708f2956-fa0f-b008-d3d2-93067f95783c@redhat.com> <cuntuq9ilg4.fsf@dme.org>
 <8f9d4ef7-ddad-160b-2d94-69f4370e8702@redhat.com> <YC6XVrWPRQJ7V6Nd@google.com>
In-Reply-To: <YC6XVrWPRQJ7V6Nd@google.com>
From:   Jim Mattson <jmattson@google.com>
Date:   Thu, 18 Feb 2021 09:55:00 -0800
Message-ID: <CALMp9eTX4Na2VTY2aU=-SUrGhst5aExdCB3f=4krKj1mFPgcqQ@mail.gmail.com>
Subject: Re: [PATCH] KVM: x86: dump_vmcs should not assume GUEST_IA32_EFER is valid
To:     Sean Christopherson <seanjc@google.com>
Cc:     Paolo Bonzini <pbonzini@redhat.com>, David Edmondson <dme@dme.org>,
        LKML <linux-kernel@vger.kernel.org>,
        Borislav Petkov <bp@alien8.de>,
        Wanpeng Li <wanpengli@tencent.com>,
        Thomas Gleixner <tglx@linutronix.de>,
        Ingo Molnar <mingo@redhat.com>,
        Vitaly Kuznetsov <vkuznets@redhat.com>,
        "the arch/x86 maintainers" <x86@kernel.org>,
        "H. Peter Anvin" <hpa@zytor.com>, kvm list <kvm@vger.kernel.org>,
        Joerg Roedel <joro@8bytes.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Thu, Feb 18, 2021 at 8:35 AM Sean Christopherson <seanjc@google.com> wrote:
>
> On Thu, Feb 18, 2021, Paolo Bonzini wrote:
> > On 18/02/21 13:56, David Edmondson wrote:
> > > On Thursday, 2021-02-18 at 12:54:52 +01, Paolo Bonzini wrote:
> > >
> > > > On 18/02/21 11:04, David Edmondson wrote:
> > > > > When dumping the VMCS, retrieve the current guest value of EFER from
> > > > > the kvm_vcpu structure if neither VM_EXIT_SAVE_IA32_EFER or
> > > > > VM_ENTRY_LOAD_IA32_EFER is set, which can occur if the processor does
> > > > > not support the relevant VM-exit/entry controls.
> > > >
> > > > Printing vcpu->arch.efer is not the best choice however.  Could we dump
> > > > the whole MSR load/store area instead?
> > >
> > > I'm happy to do that, and think that it would be useful, but it won't
> > > help with the original problem (which I should have explained more).
> > >
> > > If the guest has EFER_LMA set but we aren't using the entry/exit
> > > controls, vm_read64(GUEST_IA32_EFER) returns 0, causing dump_vmcs() to
> > > erroneously dump the PDPTRs.
> >
> > Got it now.  It would sort of help, because while dumping the MSR load/store
> > area you could get hold of the real EFER, and use it to decide whether to
> > dump the PDPTRs.
>
> EFER isn't guaranteed to be in the load list, either, e.g. if guest and host
> have the same desired value.
>
> The proper way to retrieve the effective EFER is to reuse the logic in
> nested_vmx_calc_efer(), i.e. look at VM_ENTRY_IA32E_MODE if EFER isn't being
> loaded via VMCS.

Shouldn't dump_vmcs() simply dump the contents of the VMCS, in its
entirety? What does it matter what the value of EFER is?