All of lore.kernel.org
 help / color / mirror / Atom feed
From: Mel Gorman <mgorman@suse.de>
To: David Rientjes <rientjes@google.com>
Cc: Andrew Morton <akpm@linux-foundation.org>,
	Rik van Riel <riel@redhat.com>, Vlastimil Babka <vbabka@suse.cz>,
	Joonsoo Kim <iamjoonsoo.kim@lge.com>,
	Greg Thelen <gthelen@google.com>, Hugh Dickins <hughd@google.com>,
	linux-kernel@vger.kernel.org, linux-mm@kvack.org
Subject: Re: [patch v2 4/4] mm, thp: do not perform sync compaction on pagefault
Date: Fri, 2 May 2014 12:58:34 +0100	[thread overview]
Message-ID: <20140502115834.GR23991@suse.de> (raw)
In-Reply-To: <alpine.DEB.2.02.1405020402500.19297@chino.kir.corp.google.com>

On Fri, May 02, 2014 at 04:22:03AM -0700, David Rientjes wrote:
> On Fri, 2 May 2014, Mel Gorman wrote:
> 
> > > Synchronous memory compaction can be very expensive: it can iterate an enormous 
> > > amount of memory without aborting, constantly rescheduling, waiting on page
> > > locks and lru_lock, etc, if a pageblock cannot be defragmented.
> > > 
> > > Unfortunately, it's too expensive for pagefault for transparent hugepages and 
> > > it's much better to simply fallback to pages.  On 128GB machines, we find that 
> > > synchronous memory compaction can take O(seconds) for a single thp fault.
> > > 
> > > Now that async compaction remembers where it left off without strictly relying
> > > on sync compaction, this makes thp allocations best-effort without causing
> > > egregious latency during pagefault.
> > > 
> > > Signed-off-by: David Rientjes <rientjes@google.com>
> > 
> > Compaction uses MIGRATE_SYNC_LIGHT which in the current implementation
> > avoids calling ->writepage to clean dirty page in the fallback migrate
> > case
> > 
> > static int fallback_migrate_page(struct address_space *mapping,
> >         struct page *newpage, struct page *page, enum migrate_mode mode)
> > {
> >         if (PageDirty(page)) {
> >                 /* Only writeback pages in full synchronous migration */
> >                 if (mode != MIGRATE_SYNC)
> >                         return -EBUSY;
> >                 return writeout(mapping, page);
> >         }
> > 
> > or waiting on page writeback in other cases
> > 
> >                 /*
> >                  * Only in the case of a full synchronous migration is it
> >                  * necessary to wait for PageWriteback. In the async case,
> >                  * the retry loop is too short and in the sync-light case,
> >                  * the overhead of stalling is too much
> >                  */
> >                 if (mode != MIGRATE_SYNC) {
> >                         rc = -EBUSY;
> >                         goto uncharge;
> >                 }
> > 
> > or on acquiring the page lock for unforced migrations.
> > 
> 
> The page locks I'm referring to is the lock_page() in __unmap_and_move() 
> that gets called for sync compaction after the migrate_pages() iteration 
> makes a few passes and unsuccessfully grabs it.  This becomes a forced 
> migration since __unmap_and_move() returns -EAGAIN when the trylock fails.
> 

Can that be fixed then instead of disabling it entirely?

> > However, buffers still get locked in the SYNC_LIGHT case causing stalls
> > in buffer_migrate_lock_buffers which may be undesirable or maybe you are
> > hitting some other case.
> 
> We have perf profiles from one workload in particular that shows 
> contention on i_mmap_mutex (anon isn't interesting since the vast majority 
> of memory on this workload [120GB on a 128GB machine] is has a gup pin and 
> doesn't get isolated because of 119d6d59dcc0 ("mm, compaction: avoid 
> isolating pinned pages")) between cpus all doing memory compaction trying 
> to fault thp memory.
> 

Abort SYNC_LIGHT compaction if the mutex is contended.

> That's one example that we've seen, but the fact remains that at times 
> sync compaction will iterate the entire 128GB machine and not allow an 
> order-9 page to be allocated and there's nothing to preempt it like the 
> need_resched() or lock contention checks that async compaction has. 

Make compact_control->sync the same enum field and check for contention
on the async/sync_light case but leave it for sync if compacting via the
proc interface?

> (I 
> say "at times" because deferred compaction does help if it's triggered and 
> the cached pfn values help, but the worst case scenario is when the entire 
> machine is iterated.)  Would it make sense when HPAGE_PMD_ORDER == 
> pageblock_order, which is usually the case, that we don't even try to 
> migrate pages unless the entire pageblock is successfully isolated and 
> gfp_mask & __GFP_NO_KSWAPD? 

That would be ok for async and sync_light compaction but again leave it
for sync compaction if compacting via the proc interface to give us a
basis for comparison.

> That's the only other thing that I can think 
> that will avoid such a problem other than this patch.
> 
> We're struggling to find the logic that ensures sync compaction does not 
> become too expensive at pagefault and unless that can be guaranteed, it 
> seemed best to avoid it entirely.
> 

The intent of sync_light was to avoid compaction being too expensive at
page page fault time. At the time it would have been evalated against
machines with 4G of memory. You have found that it did not go far enough.

-- 
Mel Gorman
SUSE Labs

  reply	other threads:[~2014-05-02 11:58 UTC|newest]

Thread overview: 267+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2014-05-01  0:45 [patch 1/2] mm, migration: add destination page freeing callback David Rientjes
2014-05-01  0:45 ` David Rientjes
2014-05-01  0:45 ` [patch 2/2] mm, compaction: return failed migration target pages back to freelist David Rientjes
2014-05-01  0:45   ` David Rientjes
2014-05-01  5:10   ` Naoya Horiguchi
2014-05-01 21:02     ` David Rientjes
2014-05-01 21:02       ` David Rientjes
2014-05-01  5:08 ` [patch 1/2] mm, migration: add destination page freeing callback Naoya Horiguchi
     [not found] ` <5361d71e.236ec20a.1b3d.ffffc8aeSMTPIN_ADDED_BROKEN@mx.google.com>
2014-05-01 21:02   ` David Rientjes
2014-05-01 21:02     ` David Rientjes
2014-05-01 21:35 ` [patch v2 1/4] " David Rientjes
2014-05-01 21:35   ` David Rientjes
2014-05-01 21:35   ` [patch v2 2/4] mm, compaction: return failed migration target pages back to freelist David Rientjes
2014-05-01 21:35     ` David Rientjes
2014-05-02 10:11     ` Mel Gorman
2014-05-02 10:11       ` Mel Gorman
2014-05-02 15:23     ` Vlastimil Babka
2014-05-02 15:23       ` Vlastimil Babka
2014-05-02 15:26       ` [PATCH] mm/compaction: do not count migratepages when unnecessary Vlastimil Babka
2014-05-06 21:18         ` Naoya Horiguchi
     [not found]         ` <1399411134-k43fsr0p@n-horiguchi@ah.jp.nec.com>
2014-05-07  9:33           ` Vlastimil Babka
2014-05-07  9:33             ` Vlastimil Babka
2014-05-02 15:27       ` [PATCH 2/2] mm/compaction: avoid rescanning pageblocks in isolate_freepages Vlastimil Babka
2014-05-02 15:27         ` Vlastimil Babka
2014-05-06 22:19         ` Naoya Horiguchi
     [not found]         ` <1399414778-xakujfb3@n-horiguchi@ah.jp.nec.com>
2014-05-07  9:22           ` Vlastimil Babka
2014-05-07  9:22             ` Vlastimil Babka
2014-05-02 15:29       ` [PATCH 1/2] mm/compaction: do not count migratepages when unnecessary Vlastimil Babka
2014-05-02 15:29         ` Vlastimil Babka
2014-05-01 21:35   ` [patch v2 3/4] mm, compaction: add per-zone migration pfn cache for async compaction David Rientjes
2014-05-01 21:35     ` David Rientjes
2014-05-05  9:34     ` Vlastimil Babka
2014-05-05  9:34       ` Vlastimil Babka
2014-05-05  9:51       ` David Rientjes
2014-05-05  9:51         ` David Rientjes
2014-05-05 14:24         ` Vlastimil Babka
2014-05-05 14:24           ` Vlastimil Babka
2014-05-06  0:29           ` David Rientjes
2014-05-06  0:29             ` David Rientjes
2014-05-06 11:52             ` Vlastimil Babka
2014-05-06 11:52               ` Vlastimil Babka
2014-05-01 21:35   ` [patch v2 4/4] mm, thp: do not perform sync compaction on pagefault David Rientjes
2014-05-01 21:35     ` David Rientjes
2014-05-02 10:22     ` Mel Gorman
2014-05-02 10:22       ` Mel Gorman
2014-05-02 11:22       ` David Rientjes
2014-05-02 11:22         ` David Rientjes
2014-05-02 11:58         ` Mel Gorman [this message]
2014-05-02 20:29           ` David Rientjes
2014-05-02 20:29             ` David Rientjes
2014-05-05 14:48             ` Vlastimil Babka
2014-05-05 14:48               ` Vlastimil Babka
2014-05-06  8:55             ` Mel Gorman
2014-05-06  8:55               ` Mel Gorman
2014-05-06 15:05               ` Vlastimil Babka
2014-05-06 15:05                 ` Vlastimil Babka
2014-05-02 10:10   ` [patch v2 1/4] mm, migration: add destination page freeing callback Mel Gorman
2014-05-07  2:22   ` [patch v3 1/6] " David Rientjes
2014-05-07  2:22     ` David Rientjes
2014-05-07  2:22     ` [patch v3 2/6] mm, compaction: return failed migration target pages back to freelist David Rientjes
2014-05-07  2:22       ` David Rientjes
2014-05-07 14:14       ` Naoya Horiguchi
2014-05-07 21:15       ` Andrew Morton
2014-05-07 21:15         ` Andrew Morton
2014-05-07 21:21         ` David Rientjes
2014-05-07 21:21           ` David Rientjes
2014-05-12  8:35           ` Vlastimil Babka
2014-05-12  8:35             ` Vlastimil Babka
2014-05-07 21:39         ` Greg Thelen
2014-05-07 21:39           ` Greg Thelen
2014-05-12  8:37           ` Vlastimil Babka
2014-05-12  8:37             ` Vlastimil Babka
2014-05-07  2:22     ` [patch v3 3/6] mm, compaction: add per-zone migration pfn cache for async compaction David Rientjes
2014-05-07  2:22       ` David Rientjes
2014-05-07  9:34       ` Vlastimil Babka
2014-05-07  9:34         ` Vlastimil Babka
2014-05-07 20:56       ` Naoya Horiguchi
2014-05-07  2:22     ` [patch v3 4/6] mm, compaction: embed migration mode in compact_control David Rientjes
2014-05-07  2:22       ` David Rientjes
2014-05-07  9:55       ` Vlastimil Babka
2014-05-07  9:55         ` Vlastimil Babka
2014-05-07 10:36         ` [patch v4 " David Rientjes
2014-05-07 10:36           ` David Rientjes
2014-05-09 22:03           ` Andrew Morton
2014-05-09 22:03             ` Andrew Morton
2014-05-07  2:22     ` [patch v3 5/6] mm, thp: avoid excessive compaction latency during fault David Rientjes
2014-05-07  2:22       ` David Rientjes
2014-05-07  9:39       ` Mel Gorman
2014-05-07  9:39         ` Mel Gorman
2014-05-08  5:30       ` [patch -mm] mm, thp: avoid excessive compaction latency during fault fix David Rientjes
2014-05-08  5:30         ` David Rientjes
2014-05-13 10:00         ` Vlastimil Babka
2014-05-13 10:00           ` Vlastimil Babka
2014-05-22  2:49           ` David Rientjes
2014-05-22  2:49             ` David Rientjes
2014-05-22  8:43             ` Vlastimil Babka
2014-05-22  8:43               ` Vlastimil Babka
2014-05-07  2:22     ` [patch v3 6/6] mm, compaction: terminate async compaction when rescheduling David Rientjes
2014-05-07  2:22       ` David Rientjes
2014-05-07  9:41       ` Mel Gorman
2014-05-07  9:41         ` Mel Gorman
2014-05-07 12:09       ` [PATCH v2 1/2] mm/compaction: do not count migratepages when unnecessary Vlastimil Babka
2014-05-07 12:09         ` Vlastimil Babka
2014-05-07 12:09         ` [PATCH v2 2/2] mm/compaction: avoid rescanning pageblocks in isolate_freepages Vlastimil Babka
2014-05-07 12:09           ` Vlastimil Babka
2014-05-07 21:47           ` David Rientjes
2014-05-07 21:47             ` David Rientjes
2014-05-07 22:06           ` Naoya Horiguchi
2014-05-08  5:28           ` Joonsoo Kim
2014-05-08  5:28             ` Joonsoo Kim
2014-05-12  9:09             ` Vlastimil Babka
2014-05-12  9:09               ` Vlastimil Babka
2014-05-13  1:15               ` Joonsoo Kim
2014-05-13  1:15                 ` Joonsoo Kim
2014-05-09 15:49           ` Michal Nazarewicz
2014-05-09 15:49             ` Michal Nazarewicz
2014-05-19 10:14           ` Vlastimil Babka
2014-05-19 10:14             ` Vlastimil Babka
2014-05-22  2:51             ` David Rientjes
2014-05-22  2:51               ` David Rientjes
2014-05-07 21:44         ` [PATCH v2 1/2] mm/compaction: do not count migratepages when unnecessary David Rientjes
2014-05-07 21:44           ` David Rientjes
2014-05-09 15:48         ` Michal Nazarewicz
2014-05-09 15:48           ` Michal Nazarewicz
2014-05-12  9:51           ` Vlastimil Babka
2014-05-12  9:51             ` Vlastimil Babka
2014-05-07 12:10       ` [patch v3 6/6] mm, compaction: terminate async compaction when rescheduling Vlastimil Babka
2014-05-07 12:10         ` Vlastimil Babka
2014-05-07 21:20       ` Andrew Morton
2014-05-07 21:20         ` Andrew Morton
2014-05-07 21:28         ` David Rientjes
2014-05-07 21:28           ` David Rientjes
2014-05-08  5:17       ` Joonsoo Kim
2014-05-08  5:17         ` Joonsoo Kim
2014-05-12 14:15         ` [PATCH] mm, compaction: properly signal and act upon lock and need_sched() contention Vlastimil Babka
2014-05-12 14:15           ` Vlastimil Babka
2014-05-12 15:34           ` Naoya Horiguchi
     [not found]           ` <1399908847-ouuxeneo@n-horiguchi@ah.jp.nec.com>
2014-05-12 15:45             ` Vlastimil Babka
2014-05-12 15:45               ` Vlastimil Babka
2014-05-12 15:53               ` Naoya Horiguchi
2014-05-12 20:28           ` David Rientjes
2014-05-12 20:28             ` David Rientjes
2014-05-13  8:50             ` Vlastimil Babka
2014-05-13  8:50               ` Vlastimil Babka
2014-05-13  0:44           ` Joonsoo Kim
2014-05-13  0:44             ` Joonsoo Kim
2014-05-13  8:54             ` Vlastimil Babka
2014-05-13  8:54               ` Vlastimil Babka
2014-05-15  2:21               ` Joonsoo Kim
2014-05-15  2:21                 ` Joonsoo Kim
2014-05-16  9:47           ` [PATCH v2] " Vlastimil Babka
2014-05-16  9:47             ` Vlastimil Babka
2014-05-16 17:33             ` Michal Nazarewicz
2014-05-16 17:33               ` Michal Nazarewicz
2014-05-19 23:37             ` Andrew Morton
2014-05-19 23:37               ` Andrew Morton
2014-05-21 14:13               ` Vlastimil Babka
2014-05-21 14:13                 ` Vlastimil Babka
2014-05-21 20:11                 ` Andrew Morton
2014-05-21 20:11                   ` Andrew Morton
2014-05-22  3:20             ` compaction is still too expensive for thp (was: [PATCH v2] mm, compaction: properly signal and act upon lock and need_sched() contention) David Rientjes
2014-05-22  3:20               ` David Rientjes
2014-05-22  8:10               ` compaction is still too expensive for thp Vlastimil Babka
2014-05-22  8:10                 ` Vlastimil Babka
2014-05-22  8:55                 ` David Rientjes
2014-05-22  8:55                   ` David Rientjes
2014-05-22 12:03                   ` Vlastimil Babka
2014-05-22 12:03                     ` Vlastimil Babka
2014-06-04  0:29                     ` [patch -mm 1/3] mm: rename allocflags_to_migratetype for clarity David Rientjes
2014-06-04  0:29                       ` David Rientjes
2014-06-04  0:29                       ` [patch -mm 2/3] mm, compaction: pass gfp mask to compact_control David Rientjes
2014-06-04  0:29                         ` David Rientjes
2014-06-04  0:30                       ` [patch -mm 3/3] mm, compaction: avoid compacting memory for thp if pageblock cannot become free David Rientjes
2014-06-04  0:30                         ` David Rientjes
2014-06-04 11:04                         ` Mel Gorman
2014-06-04 11:04                           ` Mel Gorman
2014-06-04 22:02                           ` David Rientjes
2014-06-04 22:02                             ` David Rientjes
2014-06-04 16:07                         ` Vlastimil Babka
2014-06-04 16:07                           ` Vlastimil Babka
2014-06-04 16:11               ` [RFC PATCH 1/6] mm, compaction: periodically drop lock and restore IRQs in scanners Vlastimil Babka
2014-06-04 16:11                 ` Vlastimil Babka
2014-06-04 16:11                 ` [RFC PATCH 2/6] mm, compaction: skip rechecks when lock was already held Vlastimil Babka
2014-06-04 16:11                   ` Vlastimil Babka
2014-06-04 23:46                   ` David Rientjes
2014-06-04 23:46                     ` David Rientjes
2014-06-04 16:11                 ` [RFC PATCH 3/6] mm, compaction: remember position within pageblock in free pages scanner Vlastimil Babka
2014-06-04 16:11                   ` Vlastimil Babka
2014-06-04 16:11                 ` [RFC PATCH 4/6] mm, compaction: skip buddy pages by their order in the migrate scanner Vlastimil Babka
2014-06-04 16:11                   ` Vlastimil Babka
2014-06-05  0:02                   ` David Rientjes
2014-06-05  0:02                     ` David Rientjes
2014-06-05  9:24                     ` Vlastimil Babka
2014-06-05  9:24                       ` Vlastimil Babka
2014-06-05 21:30                       ` David Rientjes
2014-06-05 21:30                         ` David Rientjes
2014-06-06  7:20                         ` Vlastimil Babka
2014-06-06  7:20                           ` Vlastimil Babka
2014-06-09  9:09                           ` David Rientjes
2014-06-09  9:09                             ` David Rientjes
2014-06-09 11:35                             ` Vlastimil Babka
2014-06-09 11:35                               ` Vlastimil Babka
2014-06-09 22:25                               ` David Rientjes
2014-06-09 22:25                                 ` David Rientjes
2014-06-10  7:26                                 ` Vlastimil Babka
2014-06-10  7:26                                   ` Vlastimil Babka
2014-06-10 23:54                                   ` David Rientjes
2014-06-10 23:54                                     ` David Rientjes
2014-06-11 12:18                                     ` Vlastimil Babka
2014-06-11 12:18                                       ` Vlastimil Babka
2014-06-12  0:21                                       ` David Rientjes
2014-06-12  0:21                                         ` David Rientjes
2014-06-12 11:56                                         ` Vlastimil Babka
2014-06-12 11:56                                           ` Vlastimil Babka
2014-06-12 21:48                                           ` David Rientjes
2014-06-12 21:48                                             ` David Rientjes
2014-06-04 16:11                 ` [RFC PATCH 5/6] mm, compaction: try to capture the just-created high-order freepage Vlastimil Babka
2014-06-04 16:11                   ` Vlastimil Babka
2014-06-04 16:11                 ` [RFC PATCH 6/6] mm, compaction: don't migrate in blocks that cannot be fully compacted in async direct compaction Vlastimil Babka
2014-06-04 16:11                   ` Vlastimil Babka
2014-06-05  0:08                   ` David Rientjes
2014-06-05  0:08                     ` David Rientjes
2014-06-05 15:38                     ` Vlastimil Babka
2014-06-05 15:38                       ` Vlastimil Babka
2014-06-05 21:38                       ` David Rientjes
2014-06-05 21:38                         ` David Rientjes
2014-06-06  7:33                         ` Vlastimil Babka
2014-06-06  7:33                           ` Vlastimil Babka
2014-06-09  9:06                           ` David Rientjes
2014-06-09  9:06                             ` David Rientjes
2014-06-12 12:18                             ` Vlastimil Babka
2014-06-12 12:18                               ` Vlastimil Babka
2014-06-04 23:39                 ` [RFC PATCH 1/6] mm, compaction: periodically drop lock and restore IRQs in scanners David Rientjes
2014-06-04 23:39                   ` David Rientjes
2014-06-05  9:05                   ` Vlastimil Babka
2014-06-05  9:05                     ` Vlastimil Babka
2014-05-22 23:49             ` [PATCH v2] mm, compaction: properly signal and act upon lock and need_sched() contention Kevin Hilman
2014-05-22 23:49               ` Kevin Hilman
2014-05-23  2:48               ` Shawn Guo
2014-05-23  2:48                 ` Shawn Guo
2014-05-23  2:48                 ` Shawn Guo
2014-05-23  8:34                 ` Vlastimil Babka
2014-05-23  8:34                   ` Vlastimil Babka
2014-05-23  8:34                   ` Vlastimil Babka
2014-05-23 10:49                   ` Shawn Guo
2014-05-23 10:49                     ` Shawn Guo
2014-05-23 10:49                     ` Shawn Guo
2014-05-23 15:07                   ` Kevin Hilman
2014-05-23 15:07                     ` Kevin Hilman
2014-05-23 15:07                     ` Kevin Hilman
2014-05-30 16:59                   ` Stephen Warren
2014-05-30 16:59                     ` Stephen Warren
2014-05-30 16:59                     ` Stephen Warren
2014-06-02 13:35                   ` Fabio Estevam
2014-06-02 13:35                     ` Fabio Estevam
2014-06-02 13:35                     ` Fabio Estevam
2014-06-02 14:33                     ` [PATCH -mm] mm, compaction: properly signal and act upon lock and need_sched() contention - fix Vlastimil Babka
2014-06-02 14:33                       ` Vlastimil Babka
2014-06-02 14:33                       ` Vlastimil Babka
2014-06-02 15:18                       ` Fabio Estevam
2014-06-02 15:18                         ` Fabio Estevam
2014-06-02 15:18                         ` Fabio Estevam
2014-06-02 20:09                       ` David Rientjes
2014-06-02 20:09                         ` David Rientjes
2014-06-02 20:09                         ` David Rientjes
2014-05-02 13:16 ` [patch 1/2] mm, migration: add destination page freeing callback Vlastimil Babka
2014-05-02 13:16   ` Vlastimil Babka

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20140502115834.GR23991@suse.de \
    --to=mgorman@suse.de \
    --cc=akpm@linux-foundation.org \
    --cc=gthelen@google.com \
    --cc=hughd@google.com \
    --cc=iamjoonsoo.kim@lge.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-mm@kvack.org \
    --cc=riel@redhat.com \
    --cc=rientjes@google.com \
    --cc=vbabka@suse.cz \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.