Re: [PATCH] mm: re-enable kswapd when memory pressure subsides or demotion is toggled

[Chanwon Park <flyinrm@gmail.com>]

On Mon, Sep 08, 2025 at 05:06:50PM -0700, Andrew Morton wrote:
> On Mon, 8 Sep 2025 19:04:10 +0900 Chanwon Park <flyinrm@gmail.com> wrote:
> 
> > If kswapd fails to reclaim pages from a node MAX_RECLAIM_RETRIES in a
> > row, kswapd on that node gets disabled. That is, the system won't wakeup
> > kswapd for that node until page reclamation is observed at least once.
> > That reclamation is mostly done by direct reclaim, which in turn enables
> > kswapd back.
> > 
> > However, on systems with CXL memory nodes, workloads with high anon page
> > usage can disable kswapd indefinitely, without triggering direct
> > reclaim. This can be reproduced with following steps:
> > 
> >    numa node 0   (32GB memory, 48 CPUs)
> >    numa node 2~5 (512GB CXL memory, 128GB each)
> >    (numa node 1 is disabled)
> >    swap space 8GB
> > 
> >    1) Set /sys/kernel/mm/demotion_enabled to 0.
> >    2) Set /proc/sys/kernel/numa_balancing to 0.
> >    3) Run a process that allocates and random accesses 500GB of anon
> >       pages.
> >    4) Let the process exit normally.
> 
> hm, OK, I guess this is longstanding misbehavior?
> 

Yes, unless there's any application forced to allocate pages on node 0
running, kswapd stays disabled until reboot.

> > 
> > Since kswapd_failures resets may be missed by ++ operation, it is
> > changed from int to atomic_t.
> 
> Possibly this should have been a separate (earlier) patch.  But I
> assume the need for this conversion was inroduced by this patch, so
> it's debatable.
> 

May be I should've done that, but I wasn't sure if it was the right
thing to do... It seemed that atomic_t was not needed before, and
changing the type alone meant it just adds overhead without any gain
(for that patch). But I also think splitting them is a logical thing to
do. Should I split and reupload the patch (with changes you made)?

> > --- a/include/linux/mmzone.h
> > +++ b/include/linux/mmzone.h
> > @@ -1411,7 +1411,7 @@ typedef struct pglist_data {
> >  	int kswapd_order;
> >  	enum zone_type kswapd_highest_zoneidx;
> >  
> > -	int kswapd_failures;		/* Number of 'reclaimed == 0' runs */
> > +	atomic_t kswapd_failures;	/* Number of 'reclaimed == 0' runs */
> 
> This caused a number of 80-column horrors!  I had a fiddle, what do you
> think?
> 

The changes you made look good to me! Sorry for the noise.

Sorry, my previous reply missed the mailing lists.
Resending with proper Cc.

--
Best regards,
Chanwon Park

> --- a/mm/page_alloc.c~mm-re-enable-kswapd-when-memory-pressure-subsides-or-demotion-is-toggled-fix
> +++ a/mm/page_alloc.c
> @@ -2860,29 +2860,29 @@ static void free_frozen_page_commit(stru
>  		 */
>  		return;
>  	}
> +
>  	high = nr_pcp_high(pcp, zone, batch, free_high);
> -	if (pcp->count >= high) {
> -		free_pcppages_bulk(zone, nr_pcp_free(pcp, batch, high, free_high),
> -				   pcp, pindex);
> -		if (test_bit(ZONE_BELOW_HIGH, &zone->flags) &&
> -		    zone_watermark_ok(zone, 0, high_wmark_pages(zone),
> -				      ZONE_MOVABLE, 0)) {
> -			struct pglist_data *pgdat = zone->zone_pgdat;
> -			clear_bit(ZONE_BELOW_HIGH, &zone->flags);
> +	if (pcp->count < high)
> +		return;
>  
> -			/*
> -			 * Assume that memory pressure on this node is gone
> -			 * and may be in a reclaimable state. If a memory
> -			 * fallback node exists, direct reclaim may not have
> -			 * been triggered, leaving 'hopeless node' stay in
> -			 * that state for a while. Let kswapd work again by
> -			 * resetting kswapd_failures.
> -			 */
> -			if (atomic_read(&pgdat->kswapd_failures)
> -			    >= MAX_RECLAIM_RETRIES &&
> -			    next_memory_node(pgdat->node_id) < MAX_NUMNODES)
> -				atomic_set(&pgdat->kswapd_failures, 0);
> -		}
> +	free_pcppages_bulk(zone, nr_pcp_free(pcp, batch, high, free_high),
> +			   pcp, pindex);
> +	if (test_bit(ZONE_BELOW_HIGH, &zone->flags) &&
> +	    zone_watermark_ok(zone, 0, high_wmark_pages(zone),
> +			      ZONE_MOVABLE, 0)) {
> +		struct pglist_data *pgdat = zone->zone_pgdat;
> +		clear_bit(ZONE_BELOW_HIGH, &zone->flags);
> +
> +		/*
> +		 * Assume that memory pressure on this node is gone and may be
> +		 * in a reclaimable state. If a memory fallback node exists,
> +		 * direct reclaim may not have been triggered, causing a
> +		 * 'hopeless node' to stay in that state for a while.  Let
> +		 * kswapd work again by resetting kswapd_failures.
> +		 */
> +		if (atomic_read(&pgdat->kswapd_failures) >= MAX_RECLAIM_RETRIES &&
> +		    next_memory_node(pgdat->node_id) < MAX_NUMNODES)
> +			atomic_set(&pgdat->kswapd_failures, 0);
>  	}
>  }
>  
> --- a/mm/show_mem.c~mm-re-enable-kswapd-when-memory-pressure-subsides-or-demotion-is-toggled-fix
> +++ a/mm/show_mem.c
> @@ -278,8 +278,8 @@ static void show_free_areas(unsigned int
>  #endif
>  			K(node_page_state(pgdat, NR_PAGETABLE)),
>  			K(node_page_state(pgdat, NR_SECONDARY_PAGETABLE)),
> -			str_yes_no(atomic_read(&pgdat->kswapd_failures)
> -				   >= MAX_RECLAIM_RETRIES),
> +			str_yes_no(atomic_read(&pgdat->kswapd_failures) >=
> +				   MAX_RECLAIM_RETRIES),
>  			K(node_page_state(pgdat, NR_BALLOON_PAGES)));
>  	}
>  
> _
>