On Sat, Feb 13, 2021 at 2:57 AM Shakeel Butt <shakeelb@xxxxxxxxxx> wrote:
>
> CCing more folks.
>
> On Fri, Feb 12, 2021 at 9:14 AM Muchun Song <songmuchun@xxxxxxxxxxxxx> wrote:
> >
> > The swap charges the actual number of swap entries on cgroup v2.
> > If a swap cache page is charged successful, and then we uncharge
> > the swap counter. It is wrong on cgroup v2. Because the swap
> > entry is not freed.
> >
> > Fixes: 2d1c498072de ("mm: memcontrol: make swap tracking an integral part of memory control")
> > Signed-off-by: Muchun Song <songmuchun@xxxxxxxxxxxxx>
>
> What's the user visible impact of this change?
IIUC, I think that we cannot limit the swap to memory.swap.max
on cgroup v2.
cd /sys/fs/cgroup/
mkdir test
cd test
echo 8192 > memory.max
echo 4096 > memory.swap.max
OK. Now we limit swap to 1 page and memory to 2 pages.
Firstly, we allocate 1 page from this memory cgroup and
swap this page to swap disk. We can see:
memory.current: 0
memory.swap.current: 1
Then we touch this page, we will swap in and charge
the swap cache page to the memory counter and uncharge
the swap counter.
memory.current: 1
memory.swap.current: 0 (but actually we use a swap entry)
Then we allocate another 1 page from this memory cgroup.
memory.current: 2
memory.swap.current: 0 (but actually we use a swap entry)
If we swap those 2 pages to swap disk. We can charge and swap
those 2 pages successfully. Right? Maybe I am wrong.
>
> One impact I can see is that without this patch meminfo's (SwapTotal -
> SwapFree) is larger than the sum of top level memory.swap.current.
> This change will reduce that gap.
>
> BTW what about per-cpu slots_ret cache? Should we call
> mem_cgroup_uncharge_swap() before putting in the cache after this
> change?
>
> > ---
> > mm/memcontrol.c | 2 +-
> > 1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/mm/memcontrol.c b/mm/memcontrol.c
> > index c737c8f05992..be6bc5044150 100644
> > --- a/mm/memcontrol.c
> > +++ b/mm/memcontrol.c
> > @@ -6753,7 +6753,7 @@ int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask)
> > memcg_check_events(memcg, page);
> > local_irq_enable();
> >
> > - if (PageSwapCache(page)) {
> > + if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && PageSwapCache(page)) {
> > swp_entry_t entry = { .val = page_private(page) };
> > /*
> > * The swap entry might not get freed for a long time,
> > --
> > 2.11.0
> >
