Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20230322011440.2195485-1-seanjc@google.com> <20230322011440.2195485-6-seanjc@google.com>
 <373823f7-00ba-070c-53c7-384d29889e40@intel.com>
In-Reply-To: <373823f7-00ba-070c-53c7-384d29889e40@intel.com>
From:   Jim Mattson <jmattson@google.com>
Date:   Mon, 27 Mar 2023 08:37:28 -0700
Message-ID: <CALMp9eTEG5o2jQ457BTAL=srPYFbFi2Jx1YLp+a3NW3tQ19wDQ@mail.gmail.com>
Subject: Re: [PATCH 5/6] KVM: x86: Virtualize FLUSH_L1D and passthrough MSR_IA32_FLUSH_CMD
To:     Xiaoyao Li <xiaoyao.li@intel.com>
Cc:     Sean Christopherson <seanjc@google.com>,
        Paolo Bonzini <pbonzini@redhat.com>, kvm@vger.kernel.org,
        linux-kernel@vger.kernel.org,
        Nathan Chancellor <nathan@kernel.org>,
        Emanuele Giuseppe Esposito <eesposit@redhat.com>,
        Pawan Gupta <pawan.kumar.gupta@linux.intel.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Sun, Mar 26, 2023 at 8:33=E2=80=AFPM Xiaoyao Li <xiaoyao.li@intel.com> w=
rote:
>
> On 3/22/2023 9:14 AM, Sean Christopherson wrote:
> > diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> > index c83ec88da043..3c58dbae7b4c 100644
> > --- a/arch/x86/kvm/x86.c
> > +++ b/arch/x86/kvm/x86.c
> > @@ -3628,6 +3628,18 @@ int kvm_set_msr_common(struct kvm_vcpu *vcpu, st=
ruct msr_data *msr_info)
> >
> >               wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
> >               break;
> > +     case MSR_IA32_FLUSH_CMD:
> > +             if (!msr_info->host_initiated &&
> > +                 !guest_cpuid_has(vcpu, X86_FEATURE_FLUSH_L1D))
> > +                     return 1;
> > +
> > +             if (!boot_cpu_has(X86_FEATURE_FLUSH_L1D) || (data & ~L1D_=
FLUSH))
> > +                     return 1;
> > +             if (!data)
> > +                     break;
> > +
> > +             wrmsrl(MSR_IA32_FLUSH_CMD, L1D_FLUSH);
> > +             break;
>
> Then KVM provides the ability to flush the L1 data cache of host to
> userspace. Can it be exploited to degrade the host performance if
> userspace VMM keeps flushing the L1 data cache?

The L1D$ isn't very big. A guest could always flush out any previously
cached data simply by referencing its own data. Is the ability to
flush the L1D$ by WRMSR that egregious?