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;
MIME-Version: 1.0
References: <20210427223635.2711774-1-bgardon@google.com> <20210427223635.2711774-6-bgardon@google.com>
 <997f9fe3-847b-8216-c629-1ad5fdd2ffae@redhat.com> <CANgfPd8RZXQ-BamwQPS66Q5hLRZaDFhi0WaA=ZvCP4BbofiUhg@mail.gmail.com>
 <d936b13b-bb00-fc93-de3b-adc59fa32a7b@redhat.com> <CANgfPd9kVJOAR_uq+oh9kE2gr00EUAGSPiJ9jMR9BdG2CAC+BA@mail.gmail.com>
 <5b4a0c30-118c-da1f-281c-130438a1c833@redhat.com>
In-Reply-To: <5b4a0c30-118c-da1f-281c-130438a1c833@redhat.com>
From:   Ben Gardon <bgardon@google.com>
Date:   Wed, 28 Apr 2021 14:46:35 -0700
Message-ID: <CANgfPd_S=LjEs+s2UzcHZKfUHf+n498eSbfidpXNFXjJT8kxzw@mail.gmail.com>
Subject: Re: [PATCH 5/6] KVM: x86/mmu: Protect kvm->memslots with a mutex
To:     Paolo Bonzini <pbonzini@redhat.com>
Cc:     LKML <linux-kernel@vger.kernel.org>, kvm <kvm@vger.kernel.org>,
        Peter Xu <peterx@redhat.com>,
        Sean Christopherson <seanjc@google.com>,
        Peter Shier <pshier@google.com>,
        Junaid Shahid <junaids@google.com>,
        Jim Mattson <jmattson@google.com>,
        Yulei Zhang <yulei.kernel@gmail.com>,
        Wanpeng Li <kernellwp@gmail.com>,
        Vitaly Kuznetsov <vkuznets@redhat.com>,
        Xiao Guangrong <xiaoguangrong.eric@gmail.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Wed, Apr 28, 2021 at 2:41 PM Paolo Bonzini <pbonzini@redhat.com> wrote:
>
> On 28/04/21 22:40, Ben Gardon wrote:
> > ... However with the locking you propose below, we might still run
> > into issues on a move or delete, which would mean we'd still need the
> > separate memory allocation for the rmaps array. Or we do some
> > shenanigans where we try to copy the rmap pointers from the other set
> > of memslots.
>
> If that's (almost) as easy as passing old to
> kvm_arch_prepare_memory_region, that would be totally okay.

Unfortunately it's not quite that easy because it's all the slots
_besides_ the one being modified where we'd need to copy the rmaps.

>
> > My only worry is the latency this could add to a nested VM launch, but
> > it seems pretty unlikely that that would be frequently coinciding with
> > a memslot change in practice.
>
> Right, memslot changes in practice occur only at boot and on hotplug.
> If that was a problem we could always make the allocation state
> off/in-progress/on, allowing to check the allocation state out of the
> lock.  This would only potentially slow down the first nested VM launch.
>
> Paolo
>