Re: [PATCH v2 3/5] mm/mglru: cache the last memcg node, make tail insertion constant time
From: Barry Song
Date: Sun Nov 15 2026 - 03:32:38 EST
On Wed, Nov 4, 2026 at 3:55 PM Max Kellermann <max.kellermann@xxxxxxxxx> wrote:
>
> On Wed, Nov 4, 2026 at 4:40 AM Barry Song <baohua@xxxxxxxxxx> wrote:
> >
> > On Tue, Nov 3, 2026 at 7:37 PM Max Kellermann <max.kellermann@xxxxxxxxx> wrote:
> > >
> > > Rotating or onlining a memcg currently uses hlist_nulls_add_tail_rcu()
> > > which walks the destination bin to find its tail while holding
> > > `memcg_lru.lock` with interrupts disabled. With thousands of cgroups,
> > > this takes a long time and causes severe lock contention.
> >
> > This is the main cause of the soft lockups and RCU stalls,
> > right? If so, we may want to add a Fixes tag.
>
> Yes. I added (for v3): Fixes: e4dde56cd208 ("mm: multi-gen LRU:
> per-node lru_gen_folio lists")
>
> This commit added lru_gen_rotate_memcg() which calls
> hlist_nulls_add_tail_rcu() (full linked list walk under spinlock).
>
> > The fix itself may not be directly applicable to stable
> > kernels, but based on the logic of this patch, we should
> > be able to work out a backport, right?
>
> The primary copy of these patches lives in our kernel branch that is
> based on 6.18 LTS. So I already have a set of patches that can be
> backported more easily. Do you want me to submit these versions, too
> - and if yes, how?
>
> For me, it is not entirely clear if a stable backport is really
> necessary (thus I did not Cc stable). It's not a bug; just awful
> performance when you have thousands of memcgs, but few people have
> that. But that's up for you to decide.
Okay, thanks!
I guess we can only reproduce the soft lockup with thousands of
memcgs being rapidly rotated by artificial test scripts? In real-world
scenarios, would this be more likely to manifest as a performance issue
rather than a soft lockup?
If so, let's focus on upstream mainline development first.
>
> > Another option is to leave the cleanups in patches 1 and 2
> > out for now, and make patch 3 self-contained as the first
> > patch in your series, if that makes backporting easier.
>
> I don't think that makes backporting any easier - they are not really
> cleanups, but refactoring to make [3/5] as simple&small as possible.
> I wouldn't squash them; it improves nothing but makes review harder.
>
> > I'd appreciate it if you could also mention Ridong's earlier
> > work in the changelog:
> >
> > https://lore.kernel.org/all/20251119083722.1365680-1-chenridong@xxxxxxxxxxxxxxx/
>
> Aye.
>
> > > + if (at_head || !last)
> > > hlist_nulls_add_head_rcu(&lrugen->list, head);
> >
> > Do we still need at_head now that MEMCG_LRU_OLD has been removed?
>
> Yes, for the "!last" case. This is, unfortunately, a special case.
> If you prefer, we could move this special-casing to the hlist_nulls
> API so callers need not be aware of it.
I mean, you might not need `at_head` at all if you rebase your code
onto either mm-new or mm-unstable.
https://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm.git/log/?h=mm-new
https://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm.git/log/?h=mm-unstable
Thanks
Barry