Date:   Tue, 24 Sep 2019 09:26:37 -0700
From:   Yang Shi <yang.shi@...ux.alibaba.com>
To:     Michal Hocko <mhocko@...nel.org>, linux-kernel@...r.kernel.org
Cc:     mm-commits@...r.kernel.org, vdavydov.dev@...il.com,
        shakeelb@...gle.com, rientjes@...gle.com, ktkhai@...tuozzo.com,
        kirill.shutemov@...ux.intel.com, hughd@...gle.com,
        hannes@...xchg.org, cai@....pw
Subject: Re: + mm-thp-extract-split_queue_-into-a-struct.patch added to -mm
 tree



On 9/24/19 6:56 AM, Michal Hocko wrote:
> Do we really need this if deferred list is going to be shrunk more
> pro-actively as discussed already - I am sorry I do not have a link handy
> but in short the deferred list would be drained from a kworker context
> more pro-actively rather than wait for the memory pressure to happen.

 From our experience I really didn't see the current waiting for memory 
pressure approach is a problem, it does work well and is still a good 
compromise. And, I'm supposed we all agree the side effect incurred by 
the more proactive kworker approach is definitely a concern (i.e. may 
waste cpu cycles, break isolation, etc) according to our discussion.

And we do have other much simpler ways to shrink THPs more proactively, 
for example, waking up kswapd more aggressively via tuning 
watermark_scale_factor, and/or do shrinking harder, etc.

Even though we have to drain THPs more proactively by whatever means in 
the future, I'd prefer it is memcg aware as well.

BTW, we already had these patches running in our production environment 
for a while, they did resolve the problem we met very well.

>
> On Mon 23-09-19 15:09:02, Andrew Morton wrote:
>> ------------------------------------------------------
>> From: Yang Shi <yang.shi@...ux.alibaba.com>
>> Subject: mm: thp: extract split_queue_* into a struct
>>
>> Patch series "Make deferred split shrinker memcg aware", v6.
>>
>> Currently THP deferred split shrinker is not memcg aware, this may cause
>> premature OOM with some configuration.  For example the below test would
>> run into premature OOM easily:
>>
>> $ cgcreate -g memory:thp
>> $ echo 4G > /sys/fs/cgroup/memory/thp/memory/limit_in_bytes
>> $ cgexec -g memory:thp transhuge-stress 4000
>>
>> transhuge-stress comes from kernel selftest.
>>
>> It is easy to hit OOM, but there are still a lot THP on the deferred split
>> queue, memcg direct reclaim can't touch them since the deferred split
>> shrinker is not memcg aware.
>>
>> Convert deferred split shrinker memcg aware by introducing per memcg
>> deferred split queue.  The THP should be on either per node or per memcg
>> deferred split queue if it belongs to a memcg.  When the page is
>> immigrated to the other memcg, it will be immigrated to the target memcg's
>> deferred split queue too.
>>
>> Reuse the second tail page's deferred_list for per memcg list since the
>> same THP can't be on multiple deferred split queues.
>>
>> Make deferred split shrinker not depend on memcg kmem since it is not
>> slab.  It doesn't make sense to not shrink THP even though memcg kmem is
>> disabled.
>>
>> With the above change the test demonstrated above doesn't trigger OOM even
>> though with cgroup.memory=nokmem.
>>
>>
>> This patch (of 4):
>>
>> Put split_queue, split_queue_lock and split_queue_len into a struct in
>> order to reduce code duplication when we convert deferred_split to memcg
>> aware in the later patches.
>>
>> Link: http://lkml.kernel.org/r/1565144277-36240-2-git-send-email-yang.shi@linux.alibaba.com
>> Signed-off-by: Yang Shi <yang.shi@...ux.alibaba.com>
>> Suggested-by: "Kirill A . Shutemov" <kirill.shutemov@...ux.intel.com>
>> Acked-by: Kirill A. Shutemov <kirill.shutemov@...ux.intel.com>
>> Reviewed-by: Kirill Tkhai <ktkhai@...tuozzo.com>
>> Cc: Johannes Weiner <hannes@...xchg.org>
>> Cc: Michal Hocko <mhocko@...e.com>
>> Cc: Hugh Dickins <hughd@...gle.com>
>> Cc: Shakeel Butt <shakeelb@...gle.com>
>> Cc: David Rientjes <rientjes@...gle.com>
>> Cc: Qian Cai <cai@....pw>
>> Cc: Vladimir Davydov <vdavydov.dev@...il.com>
>> Signed-off-by: Andrew Morton <akpm@...ux-foundation.org>
>> ---
>>
>>   include/linux/mmzone.h |   12 +++++++---
>>   mm/huge_memory.c       |   45 +++++++++++++++++++++------------------
>>   mm/page_alloc.c        |    8 ++++--
>>   3 files changed, 39 insertions(+), 26 deletions(-)
>>
>> --- a/include/linux/mmzone.h~mm-thp-extract-split_queue_-into-a-struct
>> +++ a/include/linux/mmzone.h
>> @@ -679,6 +679,14 @@ struct zonelist {
>>   extern struct page *mem_map;
>>   #endif
>>   
>> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
>> +struct deferred_split {
>> +	spinlock_t split_queue_lock;
>> +	struct list_head split_queue;
>> +	unsigned long split_queue_len;
>> +};
>> +#endif
>> +
>>   /*
>>    * On NUMA machines, each NUMA node would have a pg_data_t to describe
>>    * it's memory layout. On UMA machines there is a single pglist_data which
>> @@ -758,9 +766,7 @@ typedef struct pglist_data {
>>   #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
>>   
>>   #ifdef CONFIG_TRANSPARENT_HUGEPAGE
>> -	spinlock_t split_queue_lock;
>> -	struct list_head split_queue;
>> -	unsigned long split_queue_len;
>> +	struct deferred_split deferred_split_queue;
>>   #endif
>>   
>>   	/* Fields commonly accessed by the page reclaim scanner */
>> --- a/mm/huge_memory.c~mm-thp-extract-split_queue_-into-a-struct
>> +++ a/mm/huge_memory.c
>> @@ -2691,6 +2691,7 @@ int split_huge_page_to_list(struct page
>>   {
>>   	struct page *head = compound_head(page);
>>   	struct pglist_data *pgdata = NODE_DATA(page_to_nid(head));
>> +	struct deferred_split *ds_queue = &pgdata->deferred_split_queue;
>>   	struct anon_vma *anon_vma = NULL;
>>   	struct address_space *mapping = NULL;
>>   	int count, mapcount, extra_pins, ret;
>> @@ -2777,17 +2778,17 @@ int split_huge_page_to_list(struct page
>>   	}
>>   
>>   	/* Prevent deferred_split_scan() touching ->_refcount */
>> -	spin_lock(&pgdata->split_queue_lock);
>> +	spin_lock(&ds_queue->split_queue_lock);
>>   	count = page_count(head);
>>   	mapcount = total_mapcount(head);
>>   	if (!mapcount && page_ref_freeze(head, 1 + extra_pins)) {
>>   		if (!list_empty(page_deferred_list(head))) {
>> -			pgdata->split_queue_len--;
>> +			ds_queue->split_queue_len--;
>>   			list_del(page_deferred_list(head));
>>   		}
>>   		if (mapping)
>>   			__dec_node_page_state(page, NR_SHMEM_THPS);
>> -		spin_unlock(&pgdata->split_queue_lock);
>> +		spin_unlock(&ds_queue->split_queue_lock);
>>   		__split_huge_page(page, list, end, flags);
>>   		if (PageSwapCache(head)) {
>>   			swp_entry_t entry = { .val = page_private(head) };
>> @@ -2804,7 +2805,7 @@ int split_huge_page_to_list(struct page
>>   			dump_page(page, "total_mapcount(head) > 0");
>>   			BUG();
>>   		}
>> -		spin_unlock(&pgdata->split_queue_lock);
>> +		spin_unlock(&ds_queue->split_queue_lock);
>>   fail:		if (mapping)
>>   			xa_unlock(&mapping->i_pages);
>>   		spin_unlock_irqrestore(&pgdata->lru_lock, flags);
>> @@ -2827,52 +2828,56 @@ out:
>>   void free_transhuge_page(struct page *page)
>>   {
>>   	struct pglist_data *pgdata = NODE_DATA(page_to_nid(page));
>> +	struct deferred_split *ds_queue = &pgdata->deferred_split_queue;
>>   	unsigned long flags;
>>   
>> -	spin_lock_irqsave(&pgdata->split_queue_lock, flags);
>> +	spin_lock_irqsave(&ds_queue->split_queue_lock, flags);
>>   	if (!list_empty(page_deferred_list(page))) {
>> -		pgdata->split_queue_len--;
>> +		ds_queue->split_queue_len--;
>>   		list_del(page_deferred_list(page));
>>   	}
>> -	spin_unlock_irqrestore(&pgdata->split_queue_lock, flags);
>> +	spin_unlock_irqrestore(&ds_queue->split_queue_lock, flags);
>>   	free_compound_page(page);
>>   }
>>   
>>   void deferred_split_huge_page(struct page *page)
>>   {
>>   	struct pglist_data *pgdata = NODE_DATA(page_to_nid(page));
>> +	struct deferred_split *ds_queue = &pgdata->deferred_split_queue;
>>   	unsigned long flags;
>>   
>>   	VM_BUG_ON_PAGE(!PageTransHuge(page), page);
>>   
>> -	spin_lock_irqsave(&pgdata->split_queue_lock, flags);
>> +	spin_lock_irqsave(&ds_queue->split_queue_lock, flags);
>>   	if (list_empty(page_deferred_list(page))) {
>>   		count_vm_event(THP_DEFERRED_SPLIT_PAGE);
>> -		list_add_tail(page_deferred_list(page), &pgdata->split_queue);
>> -		pgdata->split_queue_len++;
>> +		list_add_tail(page_deferred_list(page), &ds_queue->split_queue);
>> +		ds_queue->split_queue_len++;
>>   	}
>> -	spin_unlock_irqrestore(&pgdata->split_queue_lock, flags);
>> +	spin_unlock_irqrestore(&ds_queue->split_queue_lock, flags);
>>   }
>>   
>>   static unsigned long deferred_split_count(struct shrinker *shrink,
>>   		struct shrink_control *sc)
>>   {
>>   	struct pglist_data *pgdata = NODE_DATA(sc->nid);
>> -	return READ_ONCE(pgdata->split_queue_len);
>> +	struct deferred_split *ds_queue = &pgdata->deferred_split_queue;
>> +	return READ_ONCE(ds_queue->split_queue_len);
>>   }
>>   
>>   static unsigned long deferred_split_scan(struct shrinker *shrink,
>>   		struct shrink_control *sc)
>>   {
>>   	struct pglist_data *pgdata = NODE_DATA(sc->nid);
>> +	struct deferred_split *ds_queue = &pgdata->deferred_split_queue;
>>   	unsigned long flags;
>>   	LIST_HEAD(list), *pos, *next;
>>   	struct page *page;
>>   	int split = 0;
>>   
>> -	spin_lock_irqsave(&pgdata->split_queue_lock, flags);
>> +	spin_lock_irqsave(&ds_queue->split_queue_lock, flags);
>>   	/* Take pin on all head pages to avoid freeing them under us */
>> -	list_for_each_safe(pos, next, &pgdata->split_queue) {
>> +	list_for_each_safe(pos, next, &ds_queue->split_queue) {
>>   		page = list_entry((void *)pos, struct page, mapping);
>>   		page = compound_head(page);
>>   		if (get_page_unless_zero(page)) {
>> @@ -2880,12 +2885,12 @@ static unsigned long deferred_split_scan
>>   		} else {
>>   			/* We lost race with put_compound_page() */
>>   			list_del_init(page_deferred_list(page));
>> -			pgdata->split_queue_len--;
>> +			ds_queue->split_queue_len--;
>>   		}
>>   		if (!--sc->nr_to_scan)
>>   			break;
>>   	}
>> -	spin_unlock_irqrestore(&pgdata->split_queue_lock, flags);
>> +	spin_unlock_irqrestore(&ds_queue->split_queue_lock, flags);
>>   
>>   	list_for_each_safe(pos, next, &list) {
>>   		page = list_entry((void *)pos, struct page, mapping);
>> @@ -2899,15 +2904,15 @@ next:
>>   		put_page(page);
>>   	}
>>   
>> -	spin_lock_irqsave(&pgdata->split_queue_lock, flags);
>> -	list_splice_tail(&list, &pgdata->split_queue);
>> -	spin_unlock_irqrestore(&pgdata->split_queue_lock, flags);
>> +	spin_lock_irqsave(&ds_queue->split_queue_lock, flags);
>> +	list_splice_tail(&list, &ds_queue->split_queue);
>> +	spin_unlock_irqrestore(&ds_queue->split_queue_lock, flags);
>>   
>>   	/*
>>   	 * Stop shrinker if we didn't split any page, but the queue is empty.
>>   	 * This can happen if pages were freed under us.
>>   	 */
>> -	if (!split && list_empty(&pgdata->split_queue))
>> +	if (!split && list_empty(&ds_queue->split_queue))
>>   		return SHRINK_STOP;
>>   	return split;
>>   }
>> --- a/mm/page_alloc.c~mm-thp-extract-split_queue_-into-a-struct
>> +++ a/mm/page_alloc.c
>> @@ -6646,9 +6646,11 @@ static unsigned long __init calc_memmap_
>>   #ifdef CONFIG_TRANSPARENT_HUGEPAGE
>>   static void pgdat_init_split_queue(struct pglist_data *pgdat)
>>   {
>> -	spin_lock_init(&pgdat->split_queue_lock);
>> -	INIT_LIST_HEAD(&pgdat->split_queue);
>> -	pgdat->split_queue_len = 0;
>> +	struct deferred_split *ds_queue = &pgdat->deferred_split_queue;
>> +
>> +	spin_lock_init(&ds_queue->split_queue_lock);
>> +	INIT_LIST_HEAD(&ds_queue->split_queue);
>> +	ds_queue->split_queue_len = 0;
>>   }
>>   #else
>>   static void pgdat_init_split_queue(struct pglist_data *pgdat) {}
>> _
>>
>> Patches currently in -mm which might be from yang.shi@...ux.alibaba.com are
>>
>> mmthp-add-read-only-thp-support-for-non-shmem-fs.patch
>> mm-thp-extract-split_queue_-into-a-struct.patch
>> mm-move-mem_cgroup_uncharge-out-of-__page_cache_release.patch
>> mm-shrinker-make-shrinker-not-depend-on-memcg-kmem.patch
>> mm-shrinker-make-shrinker-not-depend-on-memcg-kmem-v6.patch
>> mm-thp-make-deferred-split-shrinker-memcg-aware.patch
>> mm-thp-make-deferred-split-shrinker-memcg-aware-v6.patch

Powered by blists - more mailing lists