On Thu, Nov 2, 2023 at 6:13 PM Yosry Ahmed <yosryahmed@xxxxxxxxxx> wrote:
>
> On Thu, Nov 2, 2023 at 4:42 PM Nhat Pham <nphamcs@xxxxxxxxx> wrote:
> >
> > During our experiment with zswap, we sometimes observe swap IOs due to
> > occasional zswap store failures and writebacks-to-swap. These swapping
> > IOs prevent many users who cannot tolerate swapping from adopting zswap
> > to save memory and improve performance where possible.
> >
> > This patch adds the option to disable this behavior entirely: do not
> > writeback to backing swapping device when a zswap store attempt fail,
> > and do not write pages in the zswap pool back to the backing swap
> > device (both when the pool is full, and when the new zswap shrinker is
> > called).
> >
> > This new behavior can be opted-in/out on a per-cgroup basis via a new
> > cgroup file. By default, writebacks to swap device is enabled, which is
> > the previous behavior.
> >
> > Note that this is subtly different from setting memory.swap.max to 0, as
> > it still allows for pages to be stored in the zswap pool (which itself
> > consumes swap space in its current form).
> >
> > Suggested-by: Johannes Weiner <hannes@xxxxxxxxxxx>
> > Signed-off-by: Nhat Pham <nphamcs@xxxxxxxxx>
> > ---
> > Documentation/admin-guide/cgroup-v2.rst | 11 +++++++
> > Documentation/admin-guide/mm/zswap.rst | 6 ++++
> > include/linux/memcontrol.h | 12 ++++++++
> > include/linux/zswap.h | 6 ++++
> > mm/memcontrol.c | 38 +++++++++++++++++++++++++
> > mm/page_io.c | 6 ++++
> > mm/shmem.c | 3 +-
> > mm/zswap.c | 14 +++++++++
> > 8 files changed, 94 insertions(+), 2 deletions(-)
> >
> > diff --git a/Documentation/admin-guide/cgroup-v2.rst b/Documentation/admin-guide/cgroup-v2.rst
> > index 606b2e0eac4b..18c4171392ea 100644
> > --- a/Documentation/admin-guide/cgroup-v2.rst
> > +++ b/Documentation/admin-guide/cgroup-v2.rst
> > @@ -1672,6 +1672,17 @@ PAGE_SIZE multiple when read back.
> > limit, it will refuse to take any more stores before existing
> > entries fault back in or are written out to disk.
> >
> > + memory.zswap.writeback
> > + A read-write single value file which exists on non-root
> > + cgroups. The default value is "1".
> > +
> > + When this is set to 0, all swapping attempts to swapping devices
> > + are disabled. This included both zswap writebacks, and swapping due
> > + to zswap store failure.
> > +
> > + Note that this is subtly different from setting memory.swap.max to
> > + 0, as it still allows for pages to be written to the zswap pool.
> > +
> > memory.pressure
> > A read-only nested-keyed file.
> >
> > diff --git a/Documentation/admin-guide/mm/zswap.rst b/Documentation/admin-guide/mm/zswap.rst
> > index 522ae22ccb84..b987e58edb70 100644
> > --- a/Documentation/admin-guide/mm/zswap.rst
> > +++ b/Documentation/admin-guide/mm/zswap.rst
> > @@ -153,6 +153,12 @@ attribute, e. g.::
> >
> > Setting this parameter to 100 will disable the hysteresis.
> >
> > +Some users cannot tolerate the swapping that comes with zswap store failures
> > +and zswap writebacks. Swapping can be disabled entirely (without disabling
> > +zswap itself) on a cgroup-basis as follows:
> > +
> > + echo 0 > /sys/fs/cgroup/<cgroup-name>/memory.zswap.writeback
> > +
> > When there is a sizable amount of cold memory residing in the zswap pool, it
> > can be advantageous to proactively write these cold pages to swap and reclaim
> > the memory for other use cases. By default, the zswap shrinker is disabled.
> > diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h
> > index 95f6c9e60ed1..e51eafdf2a15 100644
> > --- a/include/linux/memcontrol.h
> > +++ b/include/linux/memcontrol.h
> > @@ -219,6 +219,12 @@ struct mem_cgroup {
> >
> > #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP)
> > unsigned long zswap_max;
> > +
> > + /*
> > + * Prevent pages from this memcg from being written back from zswap to
> > + * swap, and from being swapped out on zswap store failures.
> > + */
> > + bool zswap_writeback;
> > #endif
> >
> > unsigned long soft_limit;
> > @@ -1931,6 +1937,7 @@ static inline void count_objcg_event(struct obj_cgroup *objcg,
> > bool obj_cgroup_may_zswap(struct obj_cgroup *objcg);
> > void obj_cgroup_charge_zswap(struct obj_cgroup *objcg, size_t size);
> > void obj_cgroup_uncharge_zswap(struct obj_cgroup *objcg, size_t size);
> > +bool mem_cgroup_zswap_writeback_enabled(struct mem_cgroup *memcg);
> > #else
> > static inline bool obj_cgroup_may_zswap(struct obj_cgroup *objcg)
> > {
> > @@ -1944,6 +1951,11 @@ static inline void obj_cgroup_uncharge_zswap(struct obj_cgroup *objcg,
> > size_t size)
> > {
> > }
> > +static inline bool mem_cgroup_zswap_writeback_enabled(struct mem_cgroup *memcg)
> > +{
> > + /* if zswap is disabled, do not block pages going to the swapping device */
> > + return true;
> > +}
> > #endif
> >
> > #endif /* _LINUX_MEMCONTROL_H */
> > diff --git a/include/linux/zswap.h b/include/linux/zswap.h
> > index cbd373ba88d2..b4997e27a74b 100644
> > --- a/include/linux/zswap.h
> > +++ b/include/linux/zswap.h
> > @@ -35,6 +35,7 @@ void zswap_swapoff(int type);
> > void zswap_memcg_offline_cleanup(struct mem_cgroup *memcg);
> > void zswap_lruvec_state_init(struct lruvec *lruvec);
> > void zswap_lruvec_swapin(struct page *page);
> > +bool is_zswap_enabled(void);
> > #else
> >
> > struct zswap_lruvec_state {};
> > @@ -55,6 +56,11 @@ static inline void zswap_swapoff(int type) {}
> > static inline void zswap_memcg_offline_cleanup(struct mem_cgroup *memcg) {}
> > static inline void zswap_lruvec_init(struct lruvec *lruvec) {}
> > static inline void zswap_lruvec_swapin(struct page *page) {}
> > +
> > +static inline bool is_zswap_enabled(void)
> > +{
> > + return false;
> > +}
> > #endif
> >
> > #endif /* _LINUX_ZSWAP_H */
> > diff --git a/mm/memcontrol.c b/mm/memcontrol.c
> > index e43b5aba8efc..8a6aadcc103c 100644
> > --- a/mm/memcontrol.c
> > +++ b/mm/memcontrol.c
> > @@ -5545,6 +5545,7 @@ mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
> > WRITE_ONCE(memcg->soft_limit, PAGE_COUNTER_MAX);
> > #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP)
> > memcg->zswap_max = PAGE_COUNTER_MAX;
> > + WRITE_ONCE(memcg->zswap_writeback, true);
>
> Generally LGTM, just one question.
>
> Would it be more convenient if the initial value is inherited from the
> parent (the root starts with true)?
>
> I can see this being useful if we want to set it to false on the
> entire machine or one a parent cgroup, we can set it before creating
> any children instead of setting it to 0 every time we create a new
> cgroup.
I'm not 100% sure about the benefit or have a strong opinion one way
or another, but this sounds like a nice-to-have detail to me, and a relatively
low cost one (both in effort and at runtime) at that too.
Propagating the change everytime we modify the memory.zswap.writeback
value of the ancestor might be data race-prone (and costly, depending on
how big the cgroup subtree is), but this is just a one-time-per-cgroup
propagation (at the new cgroup creation time).
Can anyone come up with a failure case for this change, or why it might be
a bad idea?
Thanks for the suggestion, Yosry!
