Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:4 as permitted sender) client-ip=2620:137:e000::3:4;
Date:   Wed, 13 Sep 2023 07:50:27 -0700
In-Reply-To: <055482bec09cae1ea56f979893c6b67e9d6b26a2.camel@infradead.org>
Mime-Version: 1.0
References: <20230801034524.64007-1-likexu@tencent.com> <ZNa9QyRmuAjNAonC@google.com>
 <055482bec09cae1ea56f979893c6b67e9d6b26a2.camel@infradead.org>
Message-ID: <ZQHMM8/7xXReZHdD@google.com>
Subject: Re: [PATCH v4] KVM: x86/tsc: Don't sync user changes to TSC with
 KVM-initiated change
From:   Sean Christopherson <seanjc@google.com>
To:     David Woodhouse <dwmw2@infradead.org>
Cc:     Like Xu <like.xu.linux@gmail.com>,
        Paolo Bonzini <pbonzini@redhat.com>,
        Oliver Upton <oliver.upton@linux.dev>, kvm@vger.kernel.org,
        linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="utf-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Wed, Sep 13, 2023, David Woodhouse wrote:
> On Fri, 2023-08-11 at 15:59 -0700, Sean Christopherson wrote:
> > On Tue, Aug 01, 2023, Like Xu wrote:
> > > diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> > > index 278dbd37dab2..eeaf4ad9174d 100644
> > > --- a/arch/x86/kvm/x86.c
> > > +++ b/arch/x86/kvm/x86.c
> > > @@ -2713,7 +2713,7 @@ static void __kvm_synchronize_tsc(struct kvm_vc=
pu *vcpu, u64 offset, u64 tsc,
> > > =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0kvm_track_tsc_matchin=
g(vcpu);
> > > =C2=A0 }
> > > =C2=A0=20
> > > -static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
> > > +static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data, boo=
l user_initiated)
> >=20
> > Rather than pass two somewhat magic values for the KVM-internal call, w=
hat about
> > making @data a pointer and passing NULL?
>=20
> Why change that at all?
>=20
> Userspace used to be able to force a sync by writing zero. You are
> removing that from the ABI without any explanation about why;

No, my suggestion did not remove that from the ABI.  A @user_value of '0' w=
ould
still force synchronization.

-static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
+static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 *user_value)
 {
+       u64 data =3D user_value ? *user_value : 0;  <=3D=3D=3D "*user_value=
" is '0'
        struct kvm *kvm =3D vcpu->kvm;
        u64 offset, ns, elapsed;
        unsigned long flags;
@@ -2712,14 +2713,17 @@ static void kvm_synchronize_tsc(struct kvm_vcpu *vc=
pu, u64 data)
        elapsed =3D ns - kvm->arch.last_tsc_nsec;

        if (vcpu->arch.virtual_tsc_khz) {
+               /*
+                * Force synchronization when creating or hotplugging a vCP=
U,
+                * i.e. when the TSC value is '0', to help keep clocks stab=
le.
+                * If this is NOT a hotplug/creation case, skip synchroniza=
tion
+                * on the first write from userspace so as not to misconstr=
ue
+                * state restoration after live migration as an attempt fro=
m
+                * userspace to synchronize.
+                */
                if (data =3D=3D 0) { <=3D=3D "data" still '0', still forces=
 synchronization
-                       /*
-                        * detection of vcpu initialization -- need to sync
-                        * with other vCPUs. This particularly helps to kee=
p
-                        * kvm_clock stable after CPU hotplug
-                        */
                        synchronizing =3D true;

> it doesn't seem necessary for fixing the original issue.

It's necessary for "user_set_tsc" to be an accurate name.  The code in v6 y=
ields
"user_set_tsc_to_non_zero_value".  And I don't think it's just a naming iss=
ue,
e.g. if userspace writes '0' immediately after creating, and then later wri=
tes a
small delta, the v6 code wouldn't trigger synchronization because "user_set=
_tsc"
would be left unseft by the write of '0'.