Re: [PATCH] mm/vmalloc: use unbound workqueue for vmap area draining

[Uladzislau Rezki]

On Wed, Mar 18, 2026 at 03:36:30AM -0400, lirongqing wrote:
> From: Li RongQing <lirongqing@xxxxxxxxx>
> 
> The workqueue watchdog currently reports that drain_vmap_area_work
> hogs the CPU for more than 10ms. This typically happens during heavy
> memory pressure or high-frequency vmap/vunmap operations where the lazy
> drain list grows large.
> 
> [ 2069.796205] workqueue: drain_vmap_area_work hogged CPU for >10000us 4 times, consider switching to WQ_UNBOUND
> [ 2192.823225] workqueue: drain_vmap_area_work hogged CPU for >10000us 5 times, consider switching to WQ_UNBOUND
> [ 3225.388966] workqueue: drain_vmap_area_work hogged CPU for >10000us 7 times, consider switching to WQ_UNBOUND
> 
> Since vmap area draining is a background housekeeping task that does
> not require strict CPU affinity or cache locality, it is a prime
> candidate for an unbound workqueue.
> 
> Switching from schedule_work() to queue_work(system_unbound_wq, ...)
> allows the scheduler to offload the draining process to any available
> CPU core. This prevents stalling the current CPU and resolves the
> "consider switching to WQ_UNBOUND" warning.
> 
> Signed-off-by: Li RongQing <lirongqing@xxxxxxxxx>
> ---
>  mm/vmalloc.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/mm/vmalloc.c b/mm/vmalloc.c
> index 61caa55..5f2218a 100644
> --- a/mm/vmalloc.c
> +++ b/mm/vmalloc.c
> @@ -2471,7 +2471,7 @@ static void free_vmap_area_noflush(struct vmap_area *va)
>  
>  	/* After this point, we may free va at any time */
>  	if (unlikely(nr_lazy > nr_lazy_max))
> -		schedule_work(&drain_vmap_work);
> +		queue_work(system_unbound_wq, &drain_vmap_work);
>  }
>  
>  /*
> -- 
> 2.9.4
> 
We free memory here. Therefore it is time to switch to our
own workqueue with WQ_MEM_RECLAIM | WQ_UNBOUND flags.

--
Uladzislau Rezki