Re: [PATCH mm-unstable v1 3/5] kvm/arm64: add kvm_arch_test_clear_young()

[Yu Zhao <yuzhao@google.com>]

On Fri, Feb 17, 2023 at 2:00 AM Marc Zyngier <maz@kernel.org> wrote:
>
> On Fri, 17 Feb 2023 04:21:28 +0000,
> Yu Zhao <yuzhao@google.com> wrote:
> >
> > On Thu, Feb 16, 2023 at 9:12 PM Yu Zhao <yuzhao@google.com> wrote:
> > >
> > > This patch adds kvm_arch_test_clear_young() for the vast majority of
> > > VMs that are not pKVM and run on hardware that sets the accessed bit
> > > in KVM page tables.
>
> I'm really interested in how you can back this statement. 90% of the
> HW I have access to is not FEAT_HWAFDB capable, either because it
> predates the feature or because the feature is too buggy to be useful.

This is my expericen too -- most devices are pre v8.2.

> Do you have numbers?

Let's do a quick market survey by segment. The following only applies
to ARM CPUs:

1. Phones: none of the major Android phone vendors sell phones running
VMs; no other major Linux phone vendors.
2. Laptops: only a very limited number of Chromebooks run VMs, namely
ACRVM. No other major Linux laptop vendors.
3. Desktops: no major Linux desktop vendors.
4. Embedded/IoT/Router: no major Linux vendors run VMs (Android Auto
can be a VM guest on QNX host).
5. Cloud: this is where the vast majority VMs come from. Among the
vendors available to the general public, Ampere is the biggest player.
Here [1] is a list of its customers. The A-bit works well even on its
EVT products (Neoverse cores).

[1] https://en.wikipedia.org/wiki/Ampere_Computing

> > > It relies on two techniques, RCU and cmpxchg, to safely test and clear
> > > the accessed bit without taking the MMU lock. The former protects KVM
> > > page tables from being freed while the latter clears the accessed bit
> > > atomically against both the hardware and other software page table
> > > walkers.
> > >
> > > Signed-off-by: Yu Zhao <yuzhao@google.com>
> > > ---
> > >  arch/arm64/include/asm/kvm_host.h       |  7 +++
> > >  arch/arm64/include/asm/kvm_pgtable.h    |  8 +++
> > >  arch/arm64/include/asm/stage2_pgtable.h | 43 ++++++++++++++
> > >  arch/arm64/kvm/arm.c                    |  1 +
> > >  arch/arm64/kvm/hyp/pgtable.c            | 51 ++--------------
> > >  arch/arm64/kvm/mmu.c                    | 77 ++++++++++++++++++++++++-
> > >  6 files changed, 141 insertions(+), 46 deletions(-)
> >
> > Adding Marc and Will.
> >
> > Can you please add other interested parties that I've missed?
>
> The MAINTAINERS file has it all:
>
> KERNEL VIRTUAL MACHINE FOR ARM64 (KVM/arm64)
> M:      Marc Zyngier <maz@kernel.org>
> M:      Oliver Upton <oliver.upton@linux.dev>
> R:      James Morse <james.morse@arm.com>
> R:      Suzuki K Poulose <suzuki.poulose@arm.com>
> R:      Zenghui Yu <yuzenghui@huawei.com>
> L:      kvmarm@lists.linux.dev
>
> May I suggest that you repost your patch and Cc the interested
> parties yourself? I guess most folks will want to see this in context,
> and not as a random, isolated change with no rationale.

This clarified it. Thanks. (I was hesitant to spam people with the
entire series containing changes to other architectures.)