Message-ID: <YCJUnWLlcSGoR1sT@dhcp22.suse.cz>
Date:   Tue, 9 Feb 2021 10:23:41 +0100
From:   Michal Hocko <mhocko@...e.com>
To:     zhou <xianrong_zhou@....com>, iamjoonsoo.kim@....com
Cc:     linux-mm@...ck.org, linux-kernel@...r.kernel.org,
        akpm@...ux-foundation.org, rostedt@...dmis.org, mingo@...hat.com,
        vbabka@...e.cz, rientjes@...gle.com, willy@...ux.intel.com,
        pankaj.gupta.linux@...il.com, bhe@...hat.com, ying.huang@...el.com,
        minchan@...nel.org, ruxian.feng@...nssion.com,
        kai.cheng@...nssion.com, zhao.xu@...nssion.com,
        yunfeng.lan@...nssion.com, zhouxianrong@....com,
        zhou xianrong <xianrong.zhou@...nssion.com>
Subject: Re: [PATCH] kswapd: no need reclaim cma pages triggered by unmovable
 allocation

On Tue 09-02-21 16:23:13, zhou wrote:
> From: zhou xianrong <xianrong.zhou@...nssion.com>
> 
> For purpose of better migration cma pages are allocated after
> failure movalbe allocations and are used normally for file pages
> or anonymous pages.
> 
> In reclaim path so many cma pages if configurated are reclaimed
> from lru lists in kswapd mainly or direct reclaim triggered by
> unmovable or reclaimable allocations. But these cma pages can not
> be used by original unmovable or reclaimable allocations. So the
> reclaim are unnecessary.
> 
> In a same system if the cma pages were configurated to large then
> more failture unmovable (vmalloc etc.) or reclaimable (slab etc.)
> allocations are arised and then more kswapd rounds are triggered
> and then more cma pages are reclaimed.

Could you be more specific? Do you have any numbers and an example
configuration when this is visible?

> So this maybe cause vicious cycle. It causes that when we are under
> low memory and still there are many cma pages that can not be
> allocated due to unnecessary cma reclaim and cma fallback allocations
> . So cma pages are not used sufficiently.
> 
> The modification is straightforward that skips reclaiming cma pages
> in reclaim procedure which is triggered only by unmovable or
> reclaimable allocations. This optimization can avoid ~3% unnecessary
> cma isolations (cma isolated / total isolated).

Joonsoo used to have a patch series to drop many of the hacks we have
for CMA and made it part of a movable zone. That would solve many
problems, including this one. I am not sure where the work stands now
but it would be probably better to revive that rather than adding more
special casing on top of what we have right now.

> Signed-off-by: zhou xianrong <xianrong.zhou@...nssion.com>
> ---
>  include/linux/mmzone.h        |  6 ++--
>  include/trace/events/vmscan.h | 20 +++++++----
>  mm/page_alloc.c               |  5 +--
>  mm/vmscan.c                   | 63 +++++++++++++++++++++++++++++------
>  4 files changed, 73 insertions(+), 21 deletions(-)
> 
> diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
> index b593316bff3d..7dd38d7372b9 100644
> --- a/include/linux/mmzone.h
> +++ b/include/linux/mmzone.h
> @@ -301,6 +301,8 @@ struct lruvec {
>  #define ISOLATE_ASYNC_MIGRATE	((__force isolate_mode_t)0x4)
>  /* Isolate unevictable pages */
>  #define ISOLATE_UNEVICTABLE	((__force isolate_mode_t)0x8)
> +/* Isolate none cma pages */
> +#define ISOLATE_NONCMA		((__force isolate_mode_t)0x10)
>  
>  /* LRU Isolation modes. */
>  typedef unsigned __bitwise isolate_mode_t;
> @@ -756,7 +758,7 @@ typedef struct pglist_data {
>  	wait_queue_head_t pfmemalloc_wait;
>  	struct task_struct *kswapd;	/* Protected by
>  					   mem_hotplug_begin/end() */
> -	int kswapd_order;
> +	int kswapd_order, kswapd_migratetype;
>  	enum zone_type kswapd_highest_zoneidx;
>  
>  	int kswapd_failures;		/* Number of 'reclaimed == 0' runs */
> @@ -840,7 +842,7 @@ static inline bool pgdat_is_empty(pg_data_t *pgdat)
>  
>  void build_all_zonelists(pg_data_t *pgdat);
>  void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
> -		   enum zone_type highest_zoneidx);
> +		   int migratetype, enum zone_type highest_zoneidx);
>  bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
>  			 int highest_zoneidx, unsigned int alloc_flags,
>  			 long free_pages);
> diff --git a/include/trace/events/vmscan.h b/include/trace/events/vmscan.h
> index 2070df64958e..41bbafdfde84 100644
> --- a/include/trace/events/vmscan.h
> +++ b/include/trace/events/vmscan.h
> @@ -51,37 +51,41 @@ TRACE_EVENT(mm_vmscan_kswapd_sleep,
>  
>  TRACE_EVENT(mm_vmscan_kswapd_wake,
>  
> -	TP_PROTO(int nid, int zid, int order),
> +	TP_PROTO(int nid, int zid, int order, int mt),
>  
> -	TP_ARGS(nid, zid, order),
> +	TP_ARGS(nid, zid, order, mt),
>  
>  	TP_STRUCT__entry(
>  		__field(	int,	nid	)
>  		__field(	int,	zid	)
>  		__field(	int,	order	)
> +		__field(	int,	mt	)
>  	),
>  
>  	TP_fast_assign(
>  		__entry->nid	= nid;
>  		__entry->zid    = zid;
>  		__entry->order	= order;
> +		__entry->mt	= mt;
>  	),
>  
> -	TP_printk("nid=%d order=%d",
> +	TP_printk("nid=%d order=%d migratetype=%d",
>  		__entry->nid,
> -		__entry->order)
> +		__entry->order,
> +		__entry->mt)
>  );
>  
>  TRACE_EVENT(mm_vmscan_wakeup_kswapd,
>  
> -	TP_PROTO(int nid, int zid, int order, gfp_t gfp_flags),
> +	TP_PROTO(int nid, int zid, int order, int mt, gfp_t gfp_flags),
>  
> -	TP_ARGS(nid, zid, order, gfp_flags),
> +	TP_ARGS(nid, zid, order, mt, gfp_flags),
>  
>  	TP_STRUCT__entry(
>  		__field(	int,	nid		)
>  		__field(	int,	zid		)
>  		__field(	int,	order		)
> +		__field(	int,	mt		)
>  		__field(	gfp_t,	gfp_flags	)
>  	),
>  
> @@ -89,12 +93,14 @@ TRACE_EVENT(mm_vmscan_wakeup_kswapd,
>  		__entry->nid		= nid;
>  		__entry->zid		= zid;
>  		__entry->order		= order;
> +		__entry->mt		= mt;
>  		__entry->gfp_flags	= gfp_flags;
>  	),
>  
> -	TP_printk("nid=%d order=%d gfp_flags=%s",
> +	TP_printk("nid=%d order=%d migratetype=%d gfp_flags=%s",
>  		__entry->nid,
>  		__entry->order,
> +		__entry->mt,
>  		show_gfp_flags(__entry->gfp_flags))
>  );
>  
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index 519a60d5b6f7..45ceb15721b8 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -3517,7 +3517,7 @@ struct page *rmqueue(struct zone *preferred_zone,
>  	/* Separate test+clear to avoid unnecessary atomics */
>  	if (test_bit(ZONE_BOOSTED_WATERMARK, &zone->flags)) {
>  		clear_bit(ZONE_BOOSTED_WATERMARK, &zone->flags);
> -		wakeup_kswapd(zone, 0, 0, zone_idx(zone));
> +		wakeup_kswapd(zone, 0, 0, migratetype, zone_idx(zone));
>  	}
>  
>  	VM_BUG_ON_PAGE(page && bad_range(zone, page), page);
> @@ -4426,11 +4426,12 @@ static void wake_all_kswapds(unsigned int order, gfp_t gfp_mask,
>  	struct zone *zone;
>  	pg_data_t *last_pgdat = NULL;
>  	enum zone_type highest_zoneidx = ac->highest_zoneidx;
> +	int migratetype = ac->migratetype;
>  
>  	for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, highest_zoneidx,
>  					ac->nodemask) {
>  		if (last_pgdat != zone->zone_pgdat)
> -			wakeup_kswapd(zone, gfp_mask, order, highest_zoneidx);
> +			wakeup_kswapd(zone, gfp_mask, order, migratetype, highest_zoneidx);
>  		last_pgdat = zone->zone_pgdat;
>  	}
>  }
> diff --git a/mm/vmscan.c b/mm/vmscan.c
> index b1b574ad199d..e61ec8747a40 100644
> --- a/mm/vmscan.c
> +++ b/mm/vmscan.c
> @@ -99,6 +99,9 @@ struct scan_control {
>  	/* Can pages be swapped as part of reclaim? */
>  	unsigned int may_swap:1;
>  
> +	/* Can cma pages be reclaimed? */
> +	unsigned int may_cma:1;
> +
>  	/*
>  	 * Cgroups are not reclaimed below their configured memory.low,
>  	 * unless we threaten to OOM. If any cgroups are skipped due to
> @@ -286,6 +289,11 @@ static bool writeback_throttling_sane(struct scan_control *sc)
>  }
>  #endif
>  
> +static bool movable_reclaim(gfp_t gfp_mask)
> +{
> +	return is_migrate_movable(gfp_migratetype(gfp_mask));
> +}
> +
>  /*
>   * This misses isolated pages which are not accounted for to save counters.
>   * As the data only determines if reclaim or compaction continues, it is
> @@ -1499,6 +1507,7 @@ unsigned int reclaim_clean_pages_from_list(struct zone *zone,
>  		.gfp_mask = GFP_KERNEL,
>  		.priority = DEF_PRIORITY,
>  		.may_unmap = 1,
> +		.may_cma = 1,
>  	};
>  	struct reclaim_stat stat;
>  	unsigned int nr_reclaimed;
> @@ -1593,6 +1602,9 @@ int __isolate_lru_page_prepare(struct page *page, isolate_mode_t mode)
>  	if ((mode & ISOLATE_UNMAPPED) && page_mapped(page))
>  		return ret;
>  
> +	if ((mode & ISOLATE_NONCMA) && is_migrate_cma(get_pageblock_migratetype(page)))
> +		return ret;
> +
>  	return 0;
>  }
>  
> @@ -1647,7 +1659,10 @@ static unsigned long isolate_lru_pages(unsigned long nr_to_scan,
>  	unsigned long skipped = 0;
>  	unsigned long scan, total_scan, nr_pages;
>  	LIST_HEAD(pages_skipped);
> -	isolate_mode_t mode = (sc->may_unmap ? 0 : ISOLATE_UNMAPPED);
> +	isolate_mode_t mode;
> +
> +	mode = (sc->may_unmap ? 0 : ISOLATE_UNMAPPED);
> +	mode |= (sc->may_cma ? 0 : ISOLATE_NONCMA);
>  
>  	total_scan = 0;
>  	scan = 0;
> @@ -2125,6 +2140,7 @@ unsigned long reclaim_pages(struct list_head *page_list)
>  		.may_writepage = 1,
>  		.may_unmap = 1,
>  		.may_swap = 1,
> +		.may_cma = 1,
>  	};
>  
>  	while (!list_empty(page_list)) {
> @@ -3253,6 +3269,7 @@ unsigned long try_to_free_pages(struct zonelist *zonelist, int order,
>  		.may_writepage = !laptop_mode,
>  		.may_unmap = 1,
>  		.may_swap = 1,
> +		.may_cma = movable_reclaim(gfp_mask),
>  	};
>  
>  	/*
> @@ -3298,6 +3315,7 @@ unsigned long mem_cgroup_shrink_node(struct mem_cgroup *memcg,
>  		.may_unmap = 1,
>  		.reclaim_idx = MAX_NR_ZONES - 1,
>  		.may_swap = !noswap,
> +		.may_cma = 1,
>  	};
>  
>  	WARN_ON_ONCE(!current->reclaim_state);
> @@ -3341,6 +3359,7 @@ unsigned long try_to_free_mem_cgroup_pages(struct mem_cgroup *memcg,
>  		.may_writepage = !laptop_mode,
>  		.may_unmap = 1,
>  		.may_swap = may_swap,
> +		.may_cma = 1,
>  	};
>  	/*
>  	 * Traverse the ZONELIST_FALLBACK zonelist of the current node to put
> @@ -3548,7 +3567,7 @@ static bool kswapd_shrink_node(pg_data_t *pgdat,
>   * or lower is eligible for reclaim until at least one usable zone is
>   * balanced.
>   */
> -static int balance_pgdat(pg_data_t *pgdat, int order, int highest_zoneidx)
> +static int balance_pgdat(pg_data_t *pgdat, int order, int migratetype, int highest_zoneidx)
>  {
>  	int i;
>  	unsigned long nr_soft_reclaimed;
> @@ -3650,6 +3669,7 @@ static int balance_pgdat(pg_data_t *pgdat, int order, int highest_zoneidx)
>  		 */
>  		sc.may_writepage = !laptop_mode && !nr_boost_reclaim;
>  		sc.may_swap = !nr_boost_reclaim;
> +		sc.may_cma = is_migrate_movable(migratetype);
>  
>  		/*
>  		 * Do some background aging of the anon list, to give
> @@ -3771,8 +3791,15 @@ static enum zone_type kswapd_highest_zoneidx(pg_data_t *pgdat,
>  	return curr_idx == MAX_NR_ZONES ? prev_highest_zoneidx : curr_idx;
>  }
>  
> +static int kswapd_migratetype(pg_data_t *pgdat, int prev_migratetype)
> +{
> +	int curr_migratetype = READ_ONCE(pgdat->kswapd_migratetype);
> +
> +	return curr_migratetype == MIGRATE_TYPES ? prev_migratetype : curr_migratetype;
> +}
> +
>  static void kswapd_try_to_sleep(pg_data_t *pgdat, int alloc_order, int reclaim_order,
> -				unsigned int highest_zoneidx)
> +				int migratetype, unsigned int highest_zoneidx)
>  {
>  	long remaining = 0;
>  	DEFINE_WAIT(wait);
> @@ -3807,8 +3834,8 @@ static void kswapd_try_to_sleep(pg_data_t *pgdat, int alloc_order, int reclaim_o
>  		remaining = schedule_timeout(HZ/10);
>  
>  		/*
> -		 * If woken prematurely then reset kswapd_highest_zoneidx and
> -		 * order. The values will either be from a wakeup request or
> +		 * If woken prematurely then reset kswapd_highest_zoneidx, order
> +		 * and migratetype. The values will either be from a wakeup request or
>  		 * the previous request that slept prematurely.
>  		 */
>  		if (remaining) {
> @@ -3818,6 +3845,10 @@ static void kswapd_try_to_sleep(pg_data_t *pgdat, int alloc_order, int reclaim_o
>  
>  			if (READ_ONCE(pgdat->kswapd_order) < reclaim_order)
>  				WRITE_ONCE(pgdat->kswapd_order, reclaim_order);
> +
> +			if (!is_migrate_movable(READ_ONCE(pgdat->kswapd_migratetype)))
> +				WRITE_ONCE(pgdat->kswapd_migratetype,
> +						kswapd_migratetype(pgdat, migratetype));
>  		}
>  
>  		finish_wait(&pgdat->kswapd_wait, &wait);
> @@ -3870,6 +3901,7 @@ static void kswapd_try_to_sleep(pg_data_t *pgdat, int alloc_order, int reclaim_o
>   */
>  static int kswapd(void *p)
>  {
> +	int migratetype = 0;
>  	unsigned int alloc_order, reclaim_order;
>  	unsigned int highest_zoneidx = MAX_NR_ZONES - 1;
>  	pg_data_t *pgdat = (pg_data_t*)p;
> @@ -3895,23 +3927,27 @@ static int kswapd(void *p)
>  	set_freezable();
>  
>  	WRITE_ONCE(pgdat->kswapd_order, 0);
> +	WRITE_ONCE(pgdat->kswapd_migratetype, MIGRATE_TYPES);
>  	WRITE_ONCE(pgdat->kswapd_highest_zoneidx, MAX_NR_ZONES);
>  	for ( ; ; ) {
>  		bool ret;
>  
>  		alloc_order = reclaim_order = READ_ONCE(pgdat->kswapd_order);
> +		migratetype = kswapd_migratetype(pgdat, migratetype);
>  		highest_zoneidx = kswapd_highest_zoneidx(pgdat,
>  							highest_zoneidx);
>  
>  kswapd_try_sleep:
>  		kswapd_try_to_sleep(pgdat, alloc_order, reclaim_order,
> -					highest_zoneidx);
> +					migratetype, highest_zoneidx);
>  
>  		/* Read the new order and highest_zoneidx */
>  		alloc_order = READ_ONCE(pgdat->kswapd_order);
> +		migratetype = kswapd_migratetype(pgdat, migratetype);
>  		highest_zoneidx = kswapd_highest_zoneidx(pgdat,
>  							highest_zoneidx);
>  		WRITE_ONCE(pgdat->kswapd_order, 0);
> +		WRITE_ONCE(pgdat->kswapd_migratetype, MIGRATE_TYPES);
>  		WRITE_ONCE(pgdat->kswapd_highest_zoneidx, MAX_NR_ZONES);
>  
>  		ret = try_to_freeze();
> @@ -3934,8 +3970,8 @@ static int kswapd(void *p)
>  		 * request (alloc_order).
>  		 */
>  		trace_mm_vmscan_kswapd_wake(pgdat->node_id, highest_zoneidx,
> -						alloc_order);
> -		reclaim_order = balance_pgdat(pgdat, alloc_order,
> +						alloc_order, migratetype);
> +		reclaim_order = balance_pgdat(pgdat, alloc_order, migratetype,
>  						highest_zoneidx);
>  		if (reclaim_order < alloc_order)
>  			goto kswapd_try_sleep;
> @@ -3954,10 +3990,11 @@ static int kswapd(void *p)
>   * needed.
>   */
>  void wakeup_kswapd(struct zone *zone, gfp_t gfp_flags, int order,
> -		   enum zone_type highest_zoneidx)
> +		   int migratetype, enum zone_type highest_zoneidx)
>  {
>  	pg_data_t *pgdat;
>  	enum zone_type curr_idx;
> +	int curr_migratetype;
>  
>  	if (!managed_zone(zone))
>  		return;
> @@ -3967,6 +4004,7 @@ void wakeup_kswapd(struct zone *zone, gfp_t gfp_flags, int order,
>  
>  	pgdat = zone->zone_pgdat;
>  	curr_idx = READ_ONCE(pgdat->kswapd_highest_zoneidx);
> +	curr_migratetype = READ_ONCE(pgdat->kswapd_migratetype);
>  
>  	if (curr_idx == MAX_NR_ZONES || curr_idx < highest_zoneidx)
>  		WRITE_ONCE(pgdat->kswapd_highest_zoneidx, highest_zoneidx);
> @@ -3974,6 +4012,9 @@ void wakeup_kswapd(struct zone *zone, gfp_t gfp_flags, int order,
>  	if (READ_ONCE(pgdat->kswapd_order) < order)
>  		WRITE_ONCE(pgdat->kswapd_order, order);
>  
> +	if (curr_migratetype == MIGRATE_TYPES || is_migrate_movable(migratetype))
> +		WRITE_ONCE(pgdat->kswapd_migratetype, migratetype);
> +
>  	if (!waitqueue_active(&pgdat->kswapd_wait))
>  		return;
>  
> @@ -3994,7 +4035,7 @@ void wakeup_kswapd(struct zone *zone, gfp_t gfp_flags, int order,
>  	}
>  
>  	trace_mm_vmscan_wakeup_kswapd(pgdat->node_id, highest_zoneidx, order,
> -				      gfp_flags);
> +				      migratetype, gfp_flags);
>  	wake_up_interruptible(&pgdat->kswapd_wait);
>  }
>  
> @@ -4017,6 +4058,7 @@ unsigned long shrink_all_memory(unsigned long nr_to_reclaim)
>  		.may_writepage = 1,
>  		.may_unmap = 1,
>  		.may_swap = 1,
> +		.may_cma = 1,
>  		.hibernation_mode = 1,
>  	};
>  	struct zonelist *zonelist = node_zonelist(numa_node_id(), sc.gfp_mask);
> @@ -4176,6 +4218,7 @@ static int __node_reclaim(struct pglist_data *pgdat, gfp_t gfp_mask, unsigned in
>  		.may_writepage = !!(node_reclaim_mode & RECLAIM_WRITE),
>  		.may_unmap = !!(node_reclaim_mode & RECLAIM_UNMAP),
>  		.may_swap = 1,
> +		.may_cma = movable_reclaim(gfp_mask),
>  		.reclaim_idx = gfp_zone(gfp_mask),
>  	};
>  
> -- 
> 2.25.1

-- 
Michal Hocko
SUSE Labs

Powered by blists - more mailing lists