[PATCH] VM: fix zone list restart in page allocatate
authorJens Axboe <axboe@suse.de>
Thu, 17 Nov 2005 20:35:02 +0000 (21:35 +0100)
committerLinus Torvalds <torvalds@g5.osdl.org>
Thu, 17 Nov 2005 20:43:01 +0000 (12:43 -0800)
We must reassign z before looping through the zones kicking kswapd,
since it will be NULL if we hit an OOM condition and jump back to the
beginning again. 'z' is initially assigned before the restart: label. So
move the restart label up a little.

Signed-off-by: Jens Axboe <axboe@suse.de>
mm/page_alloc.c

index 104e69ca55e0117c5ed596e9c96a01a5336a3fd8..bd4de592dc238fd3939c16b276cc7e10f7d1e21d 100644 (file)
@@ -845,21 +845,22 @@ __alloc_pages(gfp_t gfp_mask, unsigned int order,
 
        might_sleep_if(wait);
 
+restart:
        z = zonelist->zones;  /* the list of zones suitable for gfp_mask */
 
        if (unlikely(*z == NULL)) {
                /* Should this ever happen?? */
                return NULL;
        }
-restart:
+
        page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
                                zonelist, ALLOC_CPUSET);
        if (page)
                goto got_pg;
 
-       do
+       do {
                wakeup_kswapd(*z, order);
-       while (*(++z));
+       while (*(++z));
 
        /*
         * OK, we're below the kswapd watermark and have kicked background