On Wed, Aug 26, 2026 at 04:21:50PM +0800, Longlong Xia wrote:
> From: Longlong Xia <[email protected]>
>
> The lazy RCU shrinker decremented sc->nr_to_scan (unsigned long)
> and then tested the result with <= 0. When a single CPU's lazy
> callback count exceeds the remaining budget, the subtraction wraps
> to a large positive value and the <= 0 comparison, which is
> equivalent to == 0 for an unsigned type, never fires again. The
> scan loop then iterates through every nocb CPU instead of honouring
> the reclaim budget.
>
> Accumulate into count and stop once count >= nr_to_scan.
>
> Fixes: c945b4da7a448 ("rcu: Shrinker for lazy rcu")
> Assisted-by: Zcode:GLM-5.2
> Signed-off-by: Longlong Xia <[email protected]>
OK, I should have read ahead.
This one looks even more plausible, and quite a bit simpler. I have
replaced your v1 with this patch.
Frederic, thoughts?
Thanx, Paul
> ---
> Changes in v2:
> - Rework to accumulate into count directly and compare count >= nr_to_scan
> instead of decrementing nr_to_scan, matching the kfree_rcu_shrink_scan()
> fix reworked per Hao Li's suggestion.
>
> Link:
> https://lore.kernel.org/all/[email protected]/
> ---
> kernel/rcu/tree_nocb.h | 3 +--
> 1 file changed, 1 insertion(+), 2 deletions(-)
>
> diff --git a/kernel/rcu/tree_nocb.h b/kernel/rcu/tree_nocb.h
> index 19bb42672baf..b6cfee4e5039 100644
> --- a/kernel/rcu/tree_nocb.h
> +++ b/kernel/rcu/tree_nocb.h
> @@ -1332,9 +1332,8 @@ lazy_rcu_shrink_scan(struct shrinker *shrink, struct
> shrink_control *sc)
> rcu_nocb_try_flush_bypass(rdp, jiffies);
> rcu_nocb_unlock_irqrestore(rdp, flags);
> wake_nocb_gp(rdp);
> - sc->nr_to_scan -= _count;
> count += _count;
> - if (sc->nr_to_scan <= 0)
> + if (count >= sc->nr_to_scan)
> break;
> }
>
> --
> 2.43.0
>
