Re: [PATCH v2 1/2] mm/memcontrol: respect zswap.writeback setting from parent cg too

[Nhat Pham <nphamcs@gmail.com>]

On Tue, Aug 20, 2024 at 5:38 AM Mike Yuan <me@yhndnzj.com> wrote:
>
> On 2024-08-19 at 12:09 -0700, Yosry Ahmed wrote:
> > On Fri, Aug 16, 2024 at 7:44 AM Mike Yuan <me@yhndnzj.com> wrote:
> > >
> > > Currently, the behavior of zswap.writeback wrt.
> > > the cgroup hierarchy seems a bit odd. Unlike zswap.max,
> > > it doesn't honor the value from parent cgroups. This
> > > surfaced when people tried to globally disable zswap writeback,
> > > i.e. reserve physical swap space only for hibernation [1] -
> > > disabling zswap.writeback only for the root cgroup results
> > > in subcgroups with zswap.writeback=1 still performing writeback.
> > >
> > > The inconsistency became more noticeable after I introduced
> > > the MemoryZSwapWriteback= systemd unit setting [2] for
> > > controlling the knob. The patch assumed that the kernel would
> > > enforce the value of parent cgroups. It could probably be
> > > workarounded from systemd's side, by going up the slice unit
> > > tree and inheriting the value. Yet I think it's more sensible
> > > to make it behave consistently with zswap.max and friends.
> > >
> > > [1]
> > > https://wiki.archlinux.org/title/Power_management/Suspend_and_hibernate#Disable_zswap_writeback_to_use_the_swap_space_only_for_hibernation
> > > [2] https://github.com/systemd/systemd/pull/31734
> > >
> > > Changes in v2:
> > > - Actually base on latest tree (is_zswap_enabled() ->
> > > zswap_is_enabled())
> > > - Updated Documentation/admin-guide/cgroup-v2.rst to reflect the
> > > change
> > >
> > > Link to v1:
> > > https://lore.kernel.org/linux-kernel/20240814171800.23558-1-me@yhndnzj.com/
> > >
> > > Cc: Nhat Pham <nphamcs@gmail.com>
> > > Cc: Yosry Ahmed <yosryahmed@google.com>
> > > Cc: Johannes Weiner <hannes@cmpxchg.org>
> > > Cc: Andrew Morton <akpm@linux-foundation.org>
> > >
> > > Signed-off-by: Mike Yuan <me@yhndnzj.com>
> > > Reviewed-by: Nhat Pham <nphamcs@gmail.com>
> >
> > LGTM,
> > Acked-by: Yosry Ahmed <yosryahmed@google.com>
> >
> > > ---
> > >  Documentation/admin-guide/cgroup-v2.rst | 5 ++++-
> > >  mm/memcontrol.c                         | 9 ++++++++-
> > >  2 files changed, 12 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/Documentation/admin-guide/cgroup-v2.rst
> > > b/Documentation/admin-guide/cgroup-v2.rst
> > > index 86311c2907cd..80906cea4264 100644
> > > --- a/Documentation/admin-guide/cgroup-v2.rst
> > > +++ b/Documentation/admin-guide/cgroup-v2.rst
> > > @@ -1719,7 +1719,10 @@ The following nested keys are defined.
> > >    memory.zswap.writeback
> > >         A read-write single value file. The default value is "1".
> > > The
> > >         initial value of the root cgroup is 1, and when a new
> > > cgroup is
> > > -       created, it inherits the current value of its parent.
> > > +       created, it inherits the current value of its parent. Note
> > > that
> > > +       this setting is hierarchical, i.e. the writeback would be
> > > +       implicitly disabled for child cgroups if the upper
> > > hierarchy
> > > +       does so.
> > >
> > >         When this is set to 0, all swapping attempts to swapping
> > > devices
> > >         are disabled. This included both zswap writebacks, and
> > > swapping due
> > > diff --git a/mm/memcontrol.c b/mm/memcontrol.c
> > > index f29157288b7d..327b2b030639 100644
> > > --- a/mm/memcontrol.c
> > > +++ b/mm/memcontrol.c
> > > @@ -5320,7 +5320,14 @@ void obj_cgroup_uncharge_zswap(struct
> > > obj_cgroup *objcg, size_t size)
> > >  bool mem_cgroup_zswap_writeback_enabled(struct mem_cgroup *memcg)
> > >  {
> > >         /* if zswap is disabled, do not block pages going to the
> > > swapping device */
> > > -       return !zswap_is_enabled() || !memcg || READ_ONCE(memcg-
> > > >zswap_writeback);
> > > +       if (!zswap_is_enabled())
> > > +               return true;
> >
> > This is orthogonal to this patch, but I just realized that we
> > completely ignore memory.zswap_writeback if zswap is disabled. This
> > means that if a cgroup has disabled writeback, then zswap is globally
> > disabled for some reason, we stop respecting the cgroup knob. I guess
> > the rationale could be that we want to help get pages out of zswap as
> > much as possible to honor zswap's disablement? Nhat, did I get that
> > right?
>
> Hmm, I think the current behavior makes more sense. If zswap is
> completely
> disabled, it seems intuitive that zswap-related knobs lose their
> effect.

Mike is right here. It's less of a behavioral decision, but more of a
this-can-confuse-users kind of thing :) At least that's my rationale
when I wrote this.

If users want to disable swap still, they can still do that with
memory.swap.max = 0, which is probably better because it would fail
earlier at the swap slot allocation step.

>
> > I feel like it's a little bit odd to be honest, but I don't have a
> > strong opinion on it. Maybe we should document this behavior better.
>
> But clarify this in the documentation certainly sounds good :)

But yes, better documentation == happy Nhat :)