]> git.karo-electronics.de Git - karo-tx-linux.git/commitdiff
mm: compaction: iron out isolate_freepages_block() and isolate_freepages_range()
authorMel Gorman <mgorman@suse.de>
Fri, 28 Sep 2012 00:19:46 +0000 (10:19 +1000)
committerStephen Rothwell <sfr@canb.auug.org.au>
Thu, 4 Oct 2012 05:03:30 +0000 (15:03 +1000)
Andrew pointed out that isolate_freepages_block() is "straggly" and
isolate_freepages_range() is making assumptions on how compact_control is
used which is delicate.  This patch straightens isolate_freepages_block()
and makes it fly straight and initialses compact_control to zeros in
isolate_freepages_range().  The code should be easier to follow and is
functionally equivalent.  The CMA failure path is now a little more
expensive but that is a marginal corner-case.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Richard Davies <richard@arachsys.com>
Cc: Shaohua Li <shli@kernel.org>
Cc: Avi Kivity <avi@redhat.com>
Cc: Rafael Aquini <aquini@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
mm/compaction.c

index 2950dc02e0e8b6cdd09a61ac75e9b3cebe76e2e6..2e68837bea60e2b4ec69f5d45b4a00225e095296 100644 (file)
@@ -96,7 +96,6 @@ static inline bool compact_trylock_irqsave(spinlock_t *lock,
 /* Returns true if the page is within a block suitable for migration to */
 static bool suitable_migration_target(struct page *page)
 {
-
        int migratetype = get_pageblock_migratetype(page);
 
        /* Don't interfere with memory hot-remove or the min_free_kbytes blocks */
@@ -188,21 +187,20 @@ static unsigned long isolate_freepages_block(struct compact_control *cc,
 
        cursor = pfn_to_page(blockpfn);
 
-       /* Isolate free pages. This assumes the block is valid */
+       /* Isolate free pages. */
        for (; blockpfn < end_pfn; blockpfn++, cursor++) {
                int isolated, i;
                struct page *page = cursor;
 
-               if (!pfn_valid_within(blockpfn))
-                       goto strict_check;
                nr_scanned++;
-
+               if (!pfn_valid_within(blockpfn))
+                       continue;
                if (!PageBuddy(page))
-                       goto strict_check;
+                       continue;
 
                /*
-                * The zone lock must be held to isolate freepages. This
-                * unfortunately this is a very coarse lock and can be
+                * The zone lock must be held to isolate freepages.
+                * Unfortunately this is a very coarse lock and can be
                 * heavily contended if there are parallel allocations
                 * or parallel compactions. For async compaction do not
                 * spin on the lock and we acquire the lock as late as
@@ -219,12 +217,12 @@ static unsigned long isolate_freepages_block(struct compact_control *cc,
 
                /* Recheck this is a buddy page under lock */
                if (!PageBuddy(page))
-                       goto strict_check;
+                       continue;
 
                /* Found a free page, break it into order-0 pages */
                isolated = split_free_page(page);
                if (!isolated && strict)
-                       goto strict_check;
+                       break;
                total_isolated += isolated;
                for (i = 0; i < isolated; i++) {
                        list_add(&page->lru, freelist);
@@ -236,20 +234,18 @@ static unsigned long isolate_freepages_block(struct compact_control *cc,
                        blockpfn += isolated - 1;
                        cursor += isolated - 1;
                }
-
-               continue;
-
-strict_check:
-               /* Abort isolation if the caller requested strict isolation */
-               if (strict) {
-                       total_isolated = 0;
-                       goto out;
-               }
        }
 
        trace_mm_compaction_isolate_freepages(nr_scanned, total_isolated);
 
-out:
+       /*
+        * If strict isolation is requested by CMA then check that all the
+        * pages scanned were isolated. If there were any failures, 0 is
+        * returned and CMA will fail.
+        */
+       if (strict && nr_scanned != total_isolated)
+               total_isolated = 0;
+
        if (locked)
                spin_unlock_irqrestore(&cc->zone->lock, flags);
 
@@ -275,14 +271,14 @@ isolate_freepages_range(unsigned long start_pfn, unsigned long end_pfn)
        unsigned long isolated, pfn, block_end_pfn;
        struct zone *zone = NULL;
        LIST_HEAD(freelist);
-       struct compact_control cc;
-
-       if (pfn_valid(start_pfn))
-               zone = page_zone(pfn_to_page(start_pfn));
 
        /* cc needed for isolate_freepages_block to acquire zone->lock */
-       cc.zone = zone;
-       cc.sync = true;
+       struct compact_control cc = {
+               .sync = true,
+       };
+
+       if (pfn_valid(start_pfn))
+               cc.zone = zone = page_zone(pfn_to_page(start_pfn));
 
        for (pfn = start_pfn; pfn < end_pfn; pfn += isolated) {
                if (!pfn_valid(pfn) || zone != page_zone(pfn_to_page(pfn)))