[PATCH v2 2/2] mm/slab: don't queue the page cache work under krcp->lock

From: Zhan Xusheng

Date: Fri Nov 20 2026 - 03:38:06 EST


From: Zhan Xusheng <zhanxusheng@xxxxxxxxxx>

Commit 56292e8609e3 ("rcu/tree: Defer kvfree_rcu() allocation to a
clean context") deferred page allocations to a workqueue. To avoid
deadlocks with the scheduler, it used hrtimer to queue the work instead
of queueing it directly.

Commit d0bfa8b3c411 ("kvfree_rcu: Release a page cache under memory
pressure") later added a backoff mechanism under memory pressure, but
did not consider that queue_delayed_work() with delay = 0 immediately
queues the work and introduced the deadlock commit 56292e8609e3 tried to
avoid.

run_page_cache_worker() runs under krcp->lock, from kvfree_call_rcu()
callers that may hold more, such as set_cpus_allowed_force() under
pi_lock, so it kicks the refill work from an hrtimer. Under memory
pressure it instead queues the work with a delay of
rcutree.rcu_delay_page_cache_fill_msec, which may be 0, and a zero delay
queues it at once, taking pool->lock under krcp->lock.

With the delay set to 0, calling kvfree_rcu() on vmalloc memory under
pi_lock after a drop_caches makes lockdep report:

Chain exists of:
&pool->lock --> &p->pi_lock --> krc.lock

A zero delay also defeats the backoff, which is there to defer the
refill under memory pressure. Clamp the delay to [1:100000] ms, so that
queue_delayed_work() always arms a timer, and update the documented
range.

Fixes: d0bfa8b3c411 ("kvfree_rcu: Release a page cache under memory pressure")
Cc: stable@xxxxxxxxxxxxxxx
Signed-off-by: Zhan Xusheng <zhanxusheng@xxxxxxxxxx>
---
v2: clamp the delay to [1:100000] ms instead of using the hrtimer for a
zero delay (Uladzislau). The changelog is Harry's version from
slab/for-next-fixes, with the last paragraph rewritten. The
Reviewed-by tags of Vlastimil and Hao on v1 are not carried over, as
the code changed.

Tested in a VM on v7.4-rc1 with lockdep: with
rcutree.rcu_delay_page_cache_fill_msec=0, patch 1 alone reports the
circular dependency above; with this patch the boot log says "Adjusting
rcutree.rcu_delay_page_cache_fill_msec to 1 ms." and lockdep stays
quiet.

v1: https://lore.kernel.org/all/20261028134647.3176080-1-zhanxusheng1024@xxxxxxxxx/

Documentation/admin-guide/kernel-parameters.txt | 2 +-
mm/slab_common.c | 6 +++---
2 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
index 0b3c140babd9..0d1d76a6af70 100644
--- a/Documentation/admin-guide/kernel-parameters.txt
+++ b/Documentation/admin-guide/kernel-parameters.txt
@@ -5763,7 +5763,7 @@ Kernel parameters
rcutree.rcu_delay_page_cache_fill_msec= [KNL]
Set the page-cache refill delay (in milliseconds)
in response to low-memory conditions. The range
- of permitted values is in the range 0:100000.
+ of permitted values is in the range 1:100000.

rcutree.rcu_divisor= [KNL]
Set the shift-right count to use to compute
diff --git a/mm/slab_common.c b/mm/slab_common.c
index 41dea01a8c1f..6fb453118641 100644
--- a/mm/slab_common.c
+++ b/mm/slab_common.c
@@ -2262,12 +2262,12 @@ void __init kvfree_rcu_init(void)
WQ_UNBOUND | WQ_MEM_RECLAIM, 0);
WARN_ON(!rcu_reclaim_wq);

- /* Clamp it to [0:100] seconds interval. */
- if (rcu_delay_page_cache_fill_msec < 0 ||
+ /* Clamp it to [1:100000] milliseconds interval. */
+ if (rcu_delay_page_cache_fill_msec < 1 ||
rcu_delay_page_cache_fill_msec > 100 * MSEC_PER_SEC) {

rcu_delay_page_cache_fill_msec =
- clamp(rcu_delay_page_cache_fill_msec, 0,
+ clamp(rcu_delay_page_cache_fill_msec, 1,
(int) (100 * MSEC_PER_SEC));

pr_info("Adjusting rcutree.rcu_delay_page_cache_fill_msec to %d ms.\n",
--
2.43.0