Re: [PATCH v4 0/5] mm: restore per-memcg reclaim for NONSLAB shrinkers under nokmem

[Qinyun Tan]

Hi, All

Gentle ping on this series.

V4 fixes the boot hang Kairui reported against v3 -- the root cause
was that the walker reverse-resolved xarray keys as private memcg IDs
after the keys were switched to cgroup IDs, causing shrink_dcache_sb()
to loop forever on the global drain. The fix passes the xarray index
directly through the internal walk path (new patch 1), so no reverse
lookup is needed.

Could someone take a look when they have a chance?

Thanks,
Qinyun Tan

On 10/13/26 11:12 AM, Qinyun Tan wrote:
> With cgroup.memory=nokmem, the THP deferred split shrinker and the
> zswap shrinker are degraded in two ways.
> 
> First, both shrinkers are missing the SHRINKER_NONSLAB flag, so
> shrinker_memcg_alloc() demotes them to non-memcg-aware shrinkers:
> limit-induced reclaim of a cgroup neither splits its partially
> unmapped THPs nor writes back its zswapped pages.  v1 of this series
> [1] restored the flag to fix that.
> 
> However, as Sashiko's review of v1 pointed out [2], the flag alone
> is not enough.  __list_lru_init() also collapses every list_lru into
> per-node lists under nokmem, so even with the flag restored, the
> objects of all cgroups share one list per node: the per-memcg
> shrinker bit is only set for whichever memcg happens to repopulate
> the empty list, so pressure in other cgroups may not even trigger
> the scan, and when it does, the scan walks everyone's objects.
> 
> Before commit fafaeceb89a5 ("mm: switch deferred split shrinker to
> list_lru"), THP had fully per-memcg deferred split queues embedded
> in struct mem_cgroup, working independently of kmem accounting.
> nokmem only opts out of kernel slab accounting; THPs and zswapped
> pages are user memory and remain charged to their cgroups, so
> per-memcg reclaim is still what these shrinkers want.
> 
> This series keeps list_lrus backed by SHRINKER_NONSLAB shrinkers
> memcg aware under nokmem:
> 
> Patch 1 passes the xarray index directly to the list_lru walker and
> drops the reverse memcg lookup that v3 still performed between the
> xarray iteration and the walk.  That lookup was the cause of the
> boot hang reported against v3 [3]: after the keys were switched to
> cgroup IDs, the walker kept resolving them as private memcg IDs, so
> the global drain kept skipping non-empty per-memcg sublists and
> shrink_dcache_sb() could loop forever.  Passing the index directly
> also removes the temporary css reference from the walk.
> 
> Patch 2 drops the kmemcg_id copy, which is only assigned when kmem
> accounting is enabled, and indexes the list_lru xarray with
> mem_cgroup_id() instead, so the index works independently of kmem
> accounting.  The list_lru plumbing passes memcg pointers now, and
> memcg_online_kmem()/memcg_offline_kmem() are inlined into the css
> online/offline hooks.  It also drops the nokmem early return from
> the offline path so these lrus are reparented on offline.
> 
> Patch 3 keeps a list_lru memcg aware under nokmem when its backing
> shrinker is registered SHRINKER_NONSLAB.
> 
> Patches 4 and 5 restore/add SHRINKER_NONSLAB on the THP deferred
> split shrinker and the zswap shrinker.
> 
> The savings of nokmem are preserved: slab-backed lrus (e.g. the
> superblock dentry/inode lrus) still fall back to per-node lists, and
> the per-memcg lists are only allocated when a memcg actually holds
> such objects.
> 
> Changes in v4:
>  - Fix the boot hang reported by Kairui against v3 [3]: the walker
>    no longer reverse-resolves the xarray keys; the index is passed
>    directly through the internal walk path (new patch 1, split out
>    as a preparation so patch 2 remains a pure key switch).
>  - Drop the kmemcg_id member from memcg_struct_check() along with
>    the field removal, as pointed out by Gregory Price [4].
>  - Otherwise unchanged from v3.
> 
> Changes in v3:
>  - Index the list_lru xarray with mem_cgroup_id() (the cgroup ID)
>    instead of the private memcg ID, whose lifetime only covers
>    online groups, and pass memcg pointers through the list_lru
>    plumbing; inline memcg_online_kmem()/memcg_offline_kmem() into
>    the css online/offline hooks (patch 1 as numbered in v3, per
>    Johannes's review of v2 [5]).
>  - Describe in the THP patch's changelog why fafaeceb89a5 dropped
>    SHRINKER_NONSLAB knowingly, per Johannes's pointer to the
>    original discussion [5].
>  - Collected review tags: Johannes's Reviewed-by on patches 2-4,
>    Nhat's and Yosry's Acked-by on patch 4 (numbering as in v3).
> 
> Changes in v2:
>  - Keep the list_lrus behind the two shrinkers per-memcg under
>    nokmem, so the shrinker bits are set for the right memcgs and
>    scans stay scoped to the target cgroup's objects (patches 1-2
>    as numbered in v2, new; addresses Sashiko's review of v1 [2]).
>  - Patches 3-4 unchanged from v1; collected the review tags.
> 
> [1] https://lore.kernel.org/lkml/20260904033503.4067283-1-qinyuntan@xxxxxxxxxxxxxxxxx/
> [2] https://sashiko.dev/#/patchset/20260904033503.4067283-1-qinyuntan@xxxxxxxxxxxxxxxxx
> [3] https://lore.kernel.org/lkml/20260910080722.3961351-1-qinyuntan@xxxxxxxxxxxxxxxxx/
> [4] https://lore.kernel.org/lkml/arNK64PxTXbVEhis@gourry-fedora-PF4VCD3F/
> [5] https://lore.kernel.org/lkml/20260907110111.2286932-1-qinyuntan@xxxxxxxxxxxxxxxxx/
> 
> Qinyun Tan (5):
>   mm: list_lru: pass the memcg index directly to the walker
>   mm: memcontrol: drop kmemcg_id and use mem_cgroup_id() for list_lru
>     indexing
>   mm: list_lru: keep per-memcg lists with nokmem for NONSLAB-backed lrus
>   mm: thp: restore SHRINKER_NONSLAB on the deferred split shrinker
>   mm: zswap: mark the zswap shrinker SHRINKER_NONSLAB
> 
>  include/linux/memcontrol.h |  16 ------
>  mm/huge_memory.c           |   3 +-
>  mm/list_lru.c             | 114 +++++++++++++++++++++----------------
>  mm/memcontrol.c            |  45 +++------------
>  mm/zswap.c                 |   4 +-
>  5 files changed, 77 insertions(+), 105 deletions(-)
>