Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.31 as permitted sender) client-ip=23.128.96.31;
MIME-Version: 1.0
References: <20230817003029.3073210-1-rananta@google.com> <20230817003029.3073210-9-rananta@google.com>
 <6dc460d2-c7fb-e299-b0a3-55b43de31555@redhat.com> <CAJHc60whpvOHYCFueqh0Q=SbmmeRBG_x90QOvX+vOun73ttjPA@mail.gmail.com>
 <e479914b-7ba2-3a9a-2b07-9965532cbcfa@redhat.com> <ZQTAOcTsGPos/mBD@linux.dev>
In-Reply-To: <ZQTAOcTsGPos/mBD@linux.dev>
From:   Raghavendra Rao Ananta <rananta@google.com>
Date:   Mon, 18 Sep 2023 10:02:20 -0700
Message-ID: <CAJHc60zNaCOckeBdqfq1Ac8p7TTRFK6j7bMGMA49Twa763NTkA@mail.gmail.com>
Subject: Re: [PATCH v5 08/12] KVM: arm64: PMU: Allow userspace to limit
 PMCR_EL0.N for the guest
To:     Oliver Upton <oliver.upton@linux.dev>
Cc:     Shaoqin Huang <shahuang@redhat.com>, Marc Zyngier <maz@kernel.org>,
        Alexandru Elisei <alexandru.elisei@arm.com>,
        James Morse <james.morse@arm.com>,
        Suzuki K Poulose <suzuki.poulose@arm.com>,
        Paolo Bonzini <pbonzini@redhat.com>,
        Zenghui Yu <yuzenghui@huawei.com>,
        Jing Zhang <jingzhangos@google.com>,
        Reiji Watanabe <reijiw@google.com>,
        Colton Lewis <coltonlewis@google.com>,
        linux-arm-kernel@lists.infradead.org, kvmarm@lists.linux.dev,
        linux-kernel@vger.kernel.org, kvm@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Fri, Sep 15, 2023 at 1:36=E2=80=AFPM Oliver Upton <oliver.upton@linux.de=
v> wrote:
>
> On Tue, Aug 22, 2023 at 11:26:23AM +0800, Shaoqin Huang wrote:
>
> [...]
>
> > > > > +static int set_pmcr(struct kvm_vcpu *vcpu, const struct sys_reg_=
desc *r,
> > > > > +                 u64 val)
> > > > > +{
> > > > > +     struct kvm *kvm =3D vcpu->kvm;
> > > > > +     u64 new_n, mutable_mask;
> > > > > +     int ret =3D 0;
> > > > > +
> > > > > +     new_n =3D FIELD_GET(ARMV8_PMU_PMCR_N, val);
> > > > > +
> > > > > +     mutex_lock(&kvm->arch.config_lock);
> > > > > +     if (unlikely(new_n !=3D kvm->arch.pmcr_n)) {
> > > > > +             /*
> > > > > +              * The vCPU can't have more counters than the PMU
> > > > > +              * hardware implements.
> > > > > +              */
> > > > > +             if (new_n <=3D kvm->arch.pmcr_n_limit)
> > > > > +                     kvm->arch.pmcr_n =3D new_n;
> > > > > +             else
> > > > > +                     ret =3D -EINVAL;
> > > > > +     }
> > > >
> > > > Since we have set the default value of pmcr_n, if we want to set a =
new
> > > > pmcr_n, shouldn't it be a different value?
> > > >
> > > > So how about change the checking to:
> > > >
> > > > if (likely(new_n <=3D kvm->arch.pmcr_n_limit)
> > > >          kvm->arch.pmcr_n =3D new_n;
> > > > else
> > > >          ret =3D -EINVAL;
> > > >
> > > > what do you think?
> > > >
> > > Sorry, I guess I didn't fully understand your suggestion. Are you
> > > saying that it's 'likely' that userspace would configure the correct
> > > value?
> > >
> > It depends on how userspace use this api to limit the number of pmcr. I
> > think what you mean in the code is that userspace need to set every vcp=
u's
> > pmcr to the same value, so the `unlikely` here is right, only one vcpu =
can
> > change the kvm->arch.pmcr.n, it saves the cpu cycles.
> >
> > What suggest above might be wrong. Since I think when userspace want to
> > limit the number of pmcr, it may just set the new_n on one vcpu, since =
the
> > kvm->arch.pmcr_n is a VM-local value, every vcpu can see it, so it's
> > `likely` the (new_n <=3D kvm->arch.pmcr_n_limit), it can decrease one c=
hecking
> > statement.
>
> How about we just do away with branch hints in the first place? This is
> _not_ a hot path.
>
Sounds good to me.

Thank you.
Raghavendra
> --
> Thanks,
> Oliver