Date:   Wed, 24 Nov 2021 18:42:22 +0000
From:   Sean Christopherson <seanjc@google.com>
To:     Ben Gardon <bgardon@google.com>
Cc:     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>, kvm@vger.kernel.org,
        linux-kernel@vger.kernel.org,
        Hou Wenlong <houwenlong93@linux.alibaba.com>
Subject: Re: [PATCH 27/28] KVM: x86/mmu: Do remote TLB flush before dropping
 RCU in TDP MMU resched
Message-ID: <YZ6HjoPKftW3QLqr@google.com>
References: <20211120045046.3940942-1-seanjc@google.com>
 <20211120045046.3940942-28-seanjc@google.com>
 <CANgfPd-MNnx0GVZCHcDYUyx5kqAQSr=s_QGr8zDyw8Wnz0devQ@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <CANgfPd-MNnx0GVZCHcDYUyx5kqAQSr=s_QGr8zDyw8Wnz0devQ@mail.gmail.com>
Precedence: bulk

On Tue, Nov 23, 2021, Ben Gardon wrote:
> On Fri, Nov 19, 2021 at 8:51 PM Sean Christopherson <seanjc@google.com> wrote:
> >
> > When yielding in the TDP MMU iterator, service any pending TLB flush
> > before  dropping RCU protections in anticipation of using the callers RCU
> > "lock" as a proxy for vCPUs in the guest.
> >
> > Signed-off-by: Sean Christopherson <seanjc@google.com>
> 
> Reviewed-by: Ben Gardon <bgardon@google.com>
> 
> > ---
> >  arch/x86/kvm/mmu/tdp_mmu.c | 4 ++--
> >  1 file changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/arch/x86/kvm/mmu/tdp_mmu.c b/arch/x86/kvm/mmu/tdp_mmu.c
> > index 79a52717916c..55c16680b927 100644
> > --- a/arch/x86/kvm/mmu/tdp_mmu.c
> > +++ b/arch/x86/kvm/mmu/tdp_mmu.c
> > @@ -732,11 +732,11 @@ static inline bool tdp_mmu_iter_cond_resched(struct kvm *kvm,
> >                 return false;
> >
> >         if (need_resched() || rwlock_needbreak(&kvm->mmu_lock)) {
> > -               rcu_read_unlock();
> > -
> >                 if (flush)
> >                         kvm_flush_remote_tlbs(kvm);
> >
> > +               rcu_read_unlock();
> > +
> 
> Just to check my understanding:
> Theoretically PT memory could be freed as soon as we release the RCU
> lock, if this is the only thread in a read critical section.In order
> to ensure that we can use RCU as a proxy for TLB flushes we need to
> flush the TLBs while still holding the RCU read lock. Without this
> change (and with the next one) we could wind up in a situation where
> we drop the RCU read lock, then the RCU callback runs and frees the
> memory, and then the guest does a lookup through the paging structure
> caches and we get a use-after-free bug. By flushing in an RCU critical
> section, we ensure that the TLBs will have been flushed by the time
> the RCU callback runs to free the memory. Clever!

Yep, exactly.