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:   Tue, 6 Apr 2021 22:16:48 +0000
From:   Sean Christopherson <seanjc@google.com>
To:     Wanpeng Li <kernellwp@gmail.com>
Cc:     LKML <linux-kernel@vger.kernel.org>, kvm <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>
Subject: Re: [PATCH] KVM: X86: Properly account for guest CPU time when
 considering context tracking
Message-ID: <YGzd0PndH3ovXD+H@google.com>
References: <1617011036-11734-1-git-send-email-wanpengli@tencent.com>
 <YGILHM7CHpjXtxaH@google.com>
 <CANRm+CxXAt7z5H1v_Zpjg44Ka09eWc7gaJ7HRq9USUurjqrG3A@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <CANRm+CxXAt7z5H1v_Zpjg44Ka09eWc7gaJ7HRq9USUurjqrG3A@mail.gmail.com>
Precedence: bulk

On Tue, Mar 30, 2021, Wanpeng Li wrote:
> On Tue, 30 Mar 2021 at 01:15, Sean Christopherson <seanjc@google.com> wrote:
> >
> > +Thomas
> >
> > On Mon, Mar 29, 2021, Wanpeng Li wrote:
> > > diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
> > > index 32cf828..85695b3 100644
> > > --- a/arch/x86/kvm/vmx/vmx.c
> > > +++ b/arch/x86/kvm/vmx/vmx.c
> > > @@ -6689,7 +6689,8 @@ static noinstr void vmx_vcpu_enter_exit(struct kvm_vcpu *vcpu,
> > >        * into world and some more.
> > >        */
> > >       lockdep_hardirqs_off(CALLER_ADDR0);
> > > -     guest_exit_irqoff();
> > > +     if (vtime_accounting_enabled_this_cpu())
> > > +             guest_exit_irqoff();
> >
> > This looks ok, as CONFIG_CONTEXT_TRACKING and CONFIG_VIRT_CPU_ACCOUNTING_GEN are
> > selected by CONFIG_NO_HZ_FULL=y, and can't be enabled independently, e.g. the
> > rcu_user_exit() call won't be delayed because it will never be called in the
> > !vtime case.  But it still feels wrong poking into those details, e.g. it'll
> > be weird and/or wrong guest_exit_irqoff() gains stuff that isn't vtime specific.
> 
> Could you elaborate what's the meaning of "it'll be weird and/or wrong
> guest_exit_irqoff() gains stuff that isn't vtime specific."?

For example, if RCU logic is added to guest_exit_irqoff() that is needed
irrespective of vtime, then KVM will end up with different RCU logic depending
on whether or not vtime is enabled.  RCU is just an example.  My point is that
it doesn't seem impossible that there would be something in the future that
wants to tap into the guest->host transition.

Maybe that never happens and the vtime check is perfectly ok, but for me, the
name guest_exit_irqoff() doesn't sound like something that should hinge on
time accounting being enabled.