mm, compaction: simplify handling restart position in free pages scanner
authorVlastimil Babka <vbabka@suse.cz>
Tue, 8 Sep 2015 22:02:39 +0000 (15:02 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Tue, 8 Sep 2015 22:35:28 +0000 (15:35 -0700)
Handling the position where compaction free scanner should restart
(stored in cc->free_pfn) got more complex with commit e14c720efdd7 ("mm,
compaction: remember position within pageblock in free pages scanner").
Currently the position is updated in each loop iteration of
isolate_freepages(), although it should be enough to update it only when
breaking from the loop.  There's also an extra check outside the loop
updates the position in case we have met the migration scanner.

This can be simplified if we move the test for having isolated enough
from the for-loop header next to the test for contention, and
determining the restart position only in these cases.  We can reuse the
isolate_start_pfn variable for this instead of setting cc->free_pfn
directly.  Outside the loop, we can simply set cc->free_pfn to current
value of isolate_start_pfn without any extra check.

Also add a VM_BUG_ON to catch possible mistake in the future, in case we
later add a new condition that terminates isolate_freepages_block()
prematurely without also considering the condition in
isolate_freepages().

Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Minchan Kim <minchan@kernel.org>
Acked-by: Mel Gorman <mgorman@suse.de>
Acked-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Michal Nazarewicz <mina86@mina86.com>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/compaction.c

index 7077b81..2c1e1ff 100644 (file)
@@ -943,8 +943,7 @@ static void isolate_freepages(struct compact_control *cc)
         * pages on cc->migratepages. We stop searching if the migrate
         * and free page scanners meet or enough free pages are isolated.
         */
-       for (; block_start_pfn >= low_pfn &&
-                       cc->nr_migratepages > cc->nr_freepages;
+       for (; block_start_pfn >= low_pfn;
                                block_end_pfn = block_start_pfn,
                                block_start_pfn -= pageblock_nr_pages,
                                isolate_start_pfn = block_start_pfn) {
@@ -976,6 +975,8 @@ static void isolate_freepages(struct compact_control *cc)
                                        block_end_pfn, freelist, false);
 
                /*
+                * If we isolated enough freepages, or aborted due to async
+                * compaction being contended, terminate the loop.
                 * Remember where the free scanner should restart next time,
                 * which is where isolate_freepages_block() left off.
                 * But if it scanned the whole pageblock, isolate_start_pfn
@@ -984,27 +985,31 @@ static void isolate_freepages(struct compact_control *cc)
                 * In that case we will however want to restart at the start
                 * of the previous pageblock.
                 */
-               cc->free_pfn = (isolate_start_pfn < block_end_pfn) ?
-                               isolate_start_pfn :
-                               block_start_pfn - pageblock_nr_pages;
-
-               /*
-                * isolate_freepages_block() might have aborted due to async
-                * compaction being contended
-                */
-               if (cc->contended)
+               if ((cc->nr_freepages >= cc->nr_migratepages)
+                                                       || cc->contended) {
+                       if (isolate_start_pfn >= block_end_pfn)
+                               isolate_start_pfn =
+                                       block_start_pfn - pageblock_nr_pages;
                        break;
+               } else {
+                       /*
+                        * isolate_freepages_block() should not terminate
+                        * prematurely unless contended, or isolated enough
+                        */
+                       VM_BUG_ON(isolate_start_pfn < block_end_pfn);
+               }
        }
 
        /* split_free_page does not map the pages */
        map_pages(freelist);
 
        /*
-        * If we crossed the migrate scanner, we want to keep it that way
-        * so that compact_finished() may detect this
+        * Record where the free scanner will restart next time. Either we
+        * broke from the loop and set isolate_start_pfn based on the last
+        * call to isolate_freepages_block(), or we met the migration scanner
+        * and the loop terminated due to isolate_start_pfn < low_pfn
         */
-       if (block_start_pfn < low_pfn)
-               cc->free_pfn = cc->migrate_pfn;
+       cc->free_pfn = isolate_start_pfn;
 }
 
 /*