Received-SPF: pass (google.com: domain of linux-kernel+bounces-54104-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.80.249 as permitted sender) client-ip=147.75.80.249;
Precedence: bulk
MIME-Version: 1.0
References: <20240205173124.366901-1-avagin@google.com> <ZcErs9rPqT09nNge@google.com>
In-Reply-To: <ZcErs9rPqT09nNge@google.com>
From: Andrei Vagin <avagin@gmail.com>
Date: Mon, 5 Feb 2024 15:49:53 -0800
Message-ID: <CANaxB-yAJf4wFyRgkc+XzAdkC9JUKtx-BoZ=eCV7jRSagjsv0g@mail.gmail.com>
Subject: Re: [PATCH] kvm/x86: add capability to disable the write-track mechanism
To: Sean Christopherson <seanjc@google.com>
Cc: Andrei Vagin <avagin@google.com>, Paolo Bonzini <pbonzini@redhat.com>, linux-kernel@vger.kernel.org, 
	x86@kernel.org, kvm@vger.kernel.org, Zhenyu Wang <zhenyuw@linux.intel.com>, 
	Zhi Wang <zhi.a.wang@intel.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Mon, Feb 5, 2024 at 10:41=E2=80=AFAM Sean Christopherson <seanjc@google.=
com> wrote:
>
> On Mon, Feb 05, 2024, Andrei Vagin wrote:
> > The write-track is used externally only by the gpu/drm/i915 driver.
> > Currently, it is always enabled, if a kernel has been compiled with thi=
s
> > driver.
> >
> > Enabling the write-track mechanism adds a two-byte overhead per page ac=
ross
> > all memory slots. It isn't significant for regular VMs. However in gVis=
or,
> > where the entire process virtual address space is mapped into the VM, e=
ven
> > with a 39-bit address space, the overhead amounts to 256MB.
> >
> > This change introduces the new KVM_CAP_PAGE_WRITE_TRACKING capability,
> > allowing users to enable/disable the write-track mechanism. It is enabl=
ed
> > by default for backward compatibility.
>
> I would much prefer to allocate the write-tracking metadata on-demand in
> kvm_page_track_register_notifier(), i.e. do the same as mmu_first_shadow_=
root_alloc(),
> except for just gfn_write_track.
>
> The only potential hiccup would be if taking slots_arch_lock would deadlo=
ck, but
> it should be impossible for slots_arch_lock to be taken in any other path=
 that
> involves VFIO and/or KVMGT *and* can be coincident.  Except for kvm_arch_=
destroy_vm()
> (which deletes KVM's internal memslots), slots_arch_lock is taken only th=
rough
> KVM ioctls(), and the caller of kvm_page_track_register_notifier() *must*=
 hold
> a reference to the VM.
>
> That way there's no need for new uAPI and no need for userspace changes.

I think it is a good idea, I don't know why I didn't consider it.
Thanks. I will prepare a new patch.

Thanks,
Andrei