[113/131] mm: balance LRU lists based on relative thrashing

Commit Message

Andrew Morton June 3, 2020, 11:03 p.m. UTC

From: Johannes Weiner <[e-mail]>
Subject: mm: balance LRU lists based on relative thrashing

Since the LRUs were split into anon and file lists, the VM has been
balancing between page cache and anonymous pages based on per-list ratios
of scanned vs.  rotated pages.  In most cases that tips page reclaim
towards the list that is easier to reclaim and has the fewest actively
used pages, but there are a few problems with it:

1. Refaults and LRU rotations are weighted the same way, even though
   one costs IO and the other costs a bit of CPU.

2. The less we scan an LRU list based on already observed rotations,
   the more we increase the sampling interval for new references, and
   rotations become even more likely on that list. This can enter a
   death spiral in which we stop looking at one list completely until
   the other one is all but annihilated by page reclaim.

Since commit a528910e12ec ("mm: thrash detection-based file cache sizing")
we have refault detection for the page cache.  Along with swapin events,
they are good indicators of when the file or anon list, respectively, is
too small for its workingset and needs to grow.

For example, if the page cache is thrashing, the cache pages need more
time in memory, while there may be colder pages on the anonymous list. 
Likewise, if swapped pages are faulting back in, it indicates that we
reclaim anonymous pages too aggressively and should back off.

Replace LRU rotations with refaults and swapins as the basis for relative
reclaim cost of the two LRUs.  This will have the VM target list balances
that incur the least amount of IO on aggregate.

Link: http://lkml.kernel.org/r/[e-mail]
Signed-off-by: Johannes Weiner <[e-mail]>
Cc: Joonsoo Kim <[e-mail]>
Cc: Michal Hocko <[e-mail]>
Cc: Minchan Kim <[e-mail]>
Cc: Rik van Riel <[e-mail]>
Signed-off-by: Andrew Morton <[e-mail]>
---

 include/linux/swap.h |    3 +--
 mm/swap.c            |   11 +++++++----
 mm/swap_state.c      |    5 +++++
 mm/vmscan.c          |   39 ++++++++++-----------------------------
 mm/workingset.c      |    4 ++++
 5 files changed, 27 insertions(+), 35 deletions(-)

Comments

Alex Shi June 9, 2020, 9:15 a.m. UTC | #1

在 2020/6/4 上午7:03, Andrew Morton 写道:
>  
> +	/* XXX: Move to lru_cache_add() when it supports new vs putback */

Hi Hannes,

Sorry for a bit lost, would you like to explain a bit more of your idea here?

> +	spin_lock_irq(&page_pgdat(page)->lru_lock);
> +	lru_note_cost(page);
> +	spin_unlock_irq(&page_pgdat(page)->lru_lock);
> +


What could we see here w/o the lru_lock?

Thanks
Alex

Johannes Weiner June 9, 2020, 2:45 p.m. UTC | #2

On Tue, Jun 09, 2020 at 05:15:33PM +0800, Alex Shi wrote:
> 
> 
> 在 2020/6/4 上午7:03, Andrew Morton 写道:
> >  
> > +	/* XXX: Move to lru_cache_add() when it supports new vs putback */
> 
> Hi Hannes,
> 
> Sorry for a bit lost, would you like to explain a bit more of your idea here?
> 
> > +	spin_lock_irq(&page_pgdat(page)->lru_lock);
> > +	lru_note_cost(page);
> > +	spin_unlock_irq(&page_pgdat(page)->lru_lock);
> > +
> 
> 
> What could we see here w/o the lru_lock?

It'll just be part of the existing LRU locking in
pagevec_lru_move_fn(), when the new pages are added to the LRU in
batch. See this older patch for example:

https://lore.kernel.org/linux-mm/[e-mail]/

I didn't include it in this series to reduce conflict with Joonsoo's
WIP series that also operates in this area and does something similar:

https://lkml.org/lkml/2020/4/3/63

Joonsoo Kim June 10, 2020, 5:23 a.m. UTC | #3

2020년 6월 9일 (화) 오후 11:46, Johannes Weiner <[e-mail]>님이 작성:
>
> On Tue, Jun 09, 2020 at 05:15:33PM +0800, Alex Shi wrote:
> >
> >
> > 在 2020/6/4 上午7:03, Andrew Morton 写道:
> > >
> > > +   /* XXX: Move to lru_cache_add() when it supports new vs putback */
> >
> > Hi Hannes,
> >
> > Sorry for a bit lost, would you like to explain a bit more of your idea here?
> >
> > > +   spin_lock_irq(&page_pgdat(page)->lru_lock);
> > > +   lru_note_cost(page);
> > > +   spin_unlock_irq(&page_pgdat(page)->lru_lock);
> > > +
> >
> >
> > What could we see here w/o the lru_lock?
>
> It'll just be part of the existing LRU locking in
> pagevec_lru_move_fn(), when the new pages are added to the LRU in
> batch. See this older patch for example:
>
> https://lore.kernel.org/linux-mm/[e-mail]/
>
> I didn't include it in this series to reduce conflict with Joonsoo's
> WIP series that also operates in this area and does something similar:

Thanks!

> https://lkml.org/lkml/2020/4/3/63

I haven't completed the rebase of my series but I guess that referenced patch
"https://lkml.org/lkml/2020/4/3/63" would be removed in the next version.

Before the I/O cost model, a new anonymous page contributes to the LRU reclaim
balance. But, now, a new anonymous page doesn't contributes to the I/O cost
so this adjusting patch would not be needed anymore.

If anyone wants to change this part,
"/* XXX: Move to lru_cache_add() when it supports new vs putback */", feel free
to do it.

Thanks.

Alex Shi June 11, 2020, 3:28 a.m. UTC | #4

在 2020/6/10 下午1:23, Joonsoo Kim 写道:
> 2020년 6월 9일 (화) 오후 11:46, Johannes Weiner <[e-mail]>님이 작성:
>>
>> On Tue, Jun 09, 2020 at 05:15:33PM +0800, Alex Shi wrote:
>>>
>>>
>>> 在 2020/6/4 上午7:03, Andrew Morton 写道:
>>>>
>>>> +   /* XXX: Move to lru_cache_add() when it supports new vs putback */
>>>
>>> Hi Hannes,
>>>
>>> Sorry for a bit lost, would you like to explain a bit more of your idea here?
>>>
>>>> +   spin_lock_irq(&page_pgdat(page)->lru_lock);
>>>> +   lru_note_cost(page);
>>>> +   spin_unlock_irq(&page_pgdat(page)->lru_lock);
>>>> +
>>>
>>>
>>> What could we see here w/o the lru_lock?

Why I want to know the lru_lock protection here is that currently we have 5 lru lists
but only guarded by one lock, that would cause much contention when different apps
active on a server.

I guess originally we have only one lru_lock, since 5 locks would cause cacheline bouncing
if we put them together, or a bit cacheline waste to separate them in cacheline. But
after we have qspinlock, each of cpu will just loop lock on their cacheline, no interfer 
to others. It would much much relief the performance drop by cacheline bounce.

And we could use page.mapping bits to store the using lru list index for the page.
As a quick thought, I guess, except the 5 locks for 5 lists, we still need 1 more lock for
common lruvec data or for others which relay on lru_lock now, like mlock, hpage_nr_pages..
That's the reason I want to know everything under lru_lock. :)

Any comments for this idea? :)

Thanks
Alex


>>
>> It'll just be part of the existing LRU locking in
>> pagevec_lru_move_fn(), when the new pages are added to the LRU in
>> batch. See this older patch for example:
>>
>> https://lore.kernel.org/linux-mm/[e-mail]/
>>
>> I didn't include it in this series to reduce conflict with Joonsoo's
>> WIP series that also operates in this area and does something similar:
> 
> Thanks!
> 
>> https://lkml.org/lkml/2020/4/3/63
> 
> I haven't completed the rebase of my series but I guess that referenced patch
> "https://lkml.org/lkml/2020/4/3/63" would be removed in the next version.

Thanks a lot for the info, Johannes&Joonsoo! A long history for a interesting idea. :)

> 
> Before the I/O cost model, a new anonymous page contributes to the LRU reclaim
> balance. But, now, a new anonymous page doesn't contributes to the I/O cost
> so this adjusting patch would not be needed anymore.
> 
> If anyone wants to change this part,
> "/* XXX: Move to lru_cache_add() when it supports new vs putback */", feel free
> to do it.

Patch

--- a/include/linux/swap.h~mm-balance-lru-lists-based-on-relative-thrashing
+++ a/include/linux/swap.h
@@ -334,8 +334,7 @@  extern unsigned long nr_free_pagecache_p
 
 
 /* linux/mm/swap.c */
-extern void lru_note_cost(struct lruvec *lruvec, bool file,
-			  unsigned int nr_pages);
+extern void lru_note_cost(struct page *);
 extern void lru_cache_add(struct page *);
 extern void lru_add_page_tail(struct page *page, struct page *page_tail,
 			 struct lruvec *lruvec, struct list_head *head);
--- a/mm/swap.c~mm-balance-lru-lists-based-on-relative-thrashing
+++ a/mm/swap.c
@@ -278,12 +278,15 @@  void rotate_reclaimable_page(struct page
 	}
 }
 
-void lru_note_cost(struct lruvec *lruvec, bool file, unsigned int nr_pages)
+void lru_note_cost(struct page *page)
 {
-	if (file)
-		lruvec->file_cost += nr_pages;
+	struct lruvec *lruvec = mem_cgroup_page_lruvec(page, page_pgdat(page));
+
+	/* Record new data point */
+	if (page_is_file_lru(page))
+		lruvec->file_cost++;
 	else
-		lruvec->anon_cost += nr_pages;
+		lruvec->anon_cost++;
 }
 
 static void __activate_page(struct page *page, struct lruvec *lruvec,
--- a/mm/swap_state.c~mm-balance-lru-lists-based-on-relative-thrashing
+++ a/mm/swap_state.c
@@ -440,6 +440,11 @@  struct page *__read_swap_cache_async(swp
 		goto fail_unlock;
 	}
 
+	/* XXX: Move to lru_cache_add() when it supports new vs putback */
+	spin_lock_irq(&page_pgdat(page)->lru_lock);
+	lru_note_cost(page);
+	spin_unlock_irq(&page_pgdat(page)->lru_lock);
+
 	/* Caller will initiate read into locked page */
 	SetPageWorkingset(page);
 	lru_cache_add(page);
--- a/mm/vmscan.c~mm-balance-lru-lists-based-on-relative-thrashing
+++ a/mm/vmscan.c
@@ -1958,12 +1958,6 @@  shrink_inactive_list(unsigned long nr_to
 	move_pages_to_lru(lruvec, &page_list);
 
 	__mod_node_page_state(pgdat, NR_ISOLATED_ANON + file, -nr_taken);
-	/*
-	 * Rotating pages costs CPU without actually
-	 * progressing toward the reclaim goal.
-	 */
-	lru_note_cost(lruvec, 0, stat.nr_activate[0]);
-	lru_note_cost(lruvec, 1, stat.nr_activate[1]);
 	item = current_is_kswapd() ? PGSTEAL_KSWAPD : PGSTEAL_DIRECT;
 	if (!cgroup_reclaim(sc))
 		__count_vm_events(item, nr_reclaimed);
@@ -2079,11 +2073,6 @@  static void shrink_active_list(unsigned
 	 * Move pages back to the lru list.
 	 */
 	spin_lock_irq(&pgdat->lru_lock);
-	/*
-	 * Rotating pages costs CPU without actually
-	 * progressing toward the reclaim goal.
-	 */
-	lru_note_cost(lruvec, file, nr_rotated);
 
 	nr_activate = move_pages_to_lru(lruvec, &l_active);
 	nr_deactivate = move_pages_to_lru(lruvec, &l_inactive);
@@ -2298,22 +2287,23 @@  static void get_scan_count(struct lruvec
 	scan_balance = SCAN_FRACT;
 
 	/*
-	 * With swappiness at 100, anonymous and file have the same priority.
-	 * This scanning priority is essentially the inverse of IO cost.
+	 * Calculate the pressure balance between anon and file pages.
+	 *
+	 * The amount of pressure we put on each LRU is inversely
+	 * proportional to the cost of reclaiming each list, as
+	 * determined by the share of pages that are refaulting, times
+	 * the relative IO cost of bringing back a swapped out
+	 * anonymous page vs reloading a filesystem page (swappiness).
+	 *
+	 * With swappiness at 100, anon and file have equal IO cost.
 	 */
 	anon_prio = swappiness;
 	file_prio = 200 - anon_prio;
 
 	/*
-	 * OK, so we have swap space and a fair amount of page cache
-	 * pages.  We use the recently rotated / recently scanned
-	 * ratios to determine how valuable each cache is.
-	 *
 	 * Because workloads change over time (and to avoid overflow)
 	 * we keep these statistics as a floating average, which ends
-	 * up weighing recent references more than old ones.
-	 *
-	 * anon in [0], file in [1]
+	 * up weighing recent refaults more than old ones.
 	 */
 
 	anon  = lruvec_lru_size(lruvec, LRU_ACTIVE_ANON, MAX_NR_ZONES) +
@@ -2328,15 +2318,6 @@  static void get_scan_count(struct lruvec
 		lruvec->file_cost /= 2;
 		totalcost /= 2;
 	}
-
-	/*
-	 * The amount of pressure on anon vs file pages is inversely
-	 * proportional to the assumed cost of reclaiming each list,
-	 * as determined by the share of pages that are likely going
-	 * to refault or rotate on each list (recently referenced),
-	 * times the relative IO cost of bringing back a swapped out
-	 * anonymous page vs reloading a filesystem page (swappiness).
-	 */
 	ap = anon_prio * (totalcost + 1);
 	ap /= lruvec->anon_cost + 1;
 
--- a/mm/workingset.c~mm-balance-lru-lists-based-on-relative-thrashing
+++ a/mm/workingset.c
@@ -365,6 +365,10 @@  void workingset_refault(struct page *pag
 	/* Page was active prior to eviction */
 	if (workingset) {
 		SetPageWorkingset(page);
+		/* XXX: Move to lru_cache_add() when it supports new vs putback */
+		spin_lock_irq(&page_pgdat(page)->lru_lock);
+		lru_note_cost(page);
+		spin_unlock_irq(&page_pgdat(page)->lru_lock);
 		inc_lruvec_state(lruvec, WORKINGSET_RESTORE);
 	}
 out: