mm/block: convert rw_page users to bio op use
[cascardo/linux.git] / mm / memory_hotplug.c
index a86a66c..3894b65 100644 (file)
@@ -1047,6 +1047,37 @@ static void node_states_set_node(int node, struct memory_notify *arg)
        node_set_state(node, N_MEMORY);
 }
 
+int zone_can_shift(unsigned long pfn, unsigned long nr_pages,
+                  enum zone_type target)
+{
+       struct zone *zone = page_zone(pfn_to_page(pfn));
+       enum zone_type idx = zone_idx(zone);
+       int i;
+
+       if (idx < target) {
+               /* pages must be at end of current zone */
+               if (pfn + nr_pages != zone_end_pfn(zone))
+                       return 0;
+
+               /* no zones in use between current zone and target */
+               for (i = idx + 1; i < target; i++)
+                       if (zone_is_initialized(zone - idx + i))
+                               return 0;
+       }
+
+       if (target < idx) {
+               /* pages must be at beginning of current zone */
+               if (pfn != zone->zone_start_pfn)
+                       return 0;
+
+               /* no zones in use between current zone and target */
+               for (i = target + 1; i < idx; i++)
+                       if (zone_is_initialized(zone - idx + i))
+                               return 0;
+       }
+
+       return target - idx;
+}
 
 /* Must be protected by mem_hotplug_begin() */
 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
@@ -1072,13 +1103,10 @@ int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_typ
            !can_online_high_movable(zone))
                return -EINVAL;
 
-       if (online_type == MMOP_ONLINE_KERNEL &&
-           zone_idx(zone) == ZONE_MOVABLE)
-               zone_shift = -1;
-
-       if (online_type == MMOP_ONLINE_MOVABLE &&
-           zone_idx(zone) == ZONE_MOVABLE - 1)
-               zone_shift = 1;
+       if (online_type == MMOP_ONLINE_KERNEL)
+               zone_shift = zone_can_shift(pfn, nr_pages, ZONE_NORMAL);
+       else if (online_type == MMOP_ONLINE_MOVABLE)
+               zone_shift = zone_can_shift(pfn, nr_pages, ZONE_MOVABLE);
 
        zone = move_pfn_range(zone_shift, pfn, pfn + nr_pages);
        if (!zone)
@@ -1181,9 +1209,10 @@ static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
 
                arch_refresh_nodedata(nid, pgdat);
        } else {
-               /* Reset the nr_zones and classzone_idx to 0 before reuse */
+               /* Reset the nr_zones, order and classzone_idx before reuse */
                pgdat->nr_zones = 0;
-               pgdat->classzone_idx = 0;
+               pgdat->kswapd_order = 0;
+               pgdat->kswapd_classzone_idx = 0;
        }
 
        /* we can use NODE_DATA(nid) from here */
@@ -1519,6 +1548,37 @@ static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
        return 0;
 }
 
+static struct page *new_node_page(struct page *page, unsigned long private,
+               int **result)
+{
+       gfp_t gfp_mask = GFP_USER | __GFP_MOVABLE;
+       int nid = page_to_nid(page);
+       nodemask_t nmask = node_online_map;
+       struct page *new_page;
+
+       /*
+        * TODO: allocate a destination hugepage from a nearest neighbor node,
+        * accordance with memory policy of the user process if possible. For
+        * now as a simple work-around, we use the next node for destination.
+        */
+       if (PageHuge(page))
+               return alloc_huge_page_node(page_hstate(compound_head(page)),
+                                       next_node_in(nid, nmask));
+
+       node_clear(nid, nmask);
+       if (PageHighMem(page)
+           || (zone_idx(page_zone(page)) == ZONE_MOVABLE))
+               gfp_mask |= __GFP_HIGHMEM;
+
+       new_page = __alloc_pages_nodemask(gfp_mask, 0,
+                                       node_zonelist(nid, gfp_mask), &nmask);
+       if (!new_page)
+               new_page = __alloc_pages(gfp_mask, 0,
+                                       node_zonelist(nid, gfp_mask));
+
+       return new_page;
+}
+
 #define NR_OFFLINE_AT_ONCE_PAGES       (256)
 static int
 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
@@ -1558,7 +1618,7 @@ do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
                        put_page(page);
                        list_add_tail(&page->lru, &source);
                        move_pages--;
-                       inc_zone_page_state(page, NR_ISOLATED_ANON +
+                       inc_node_page_state(page, NR_ISOLATED_ANON +
                                            page_is_file_cache(page));
 
                } else {
@@ -1582,11 +1642,8 @@ do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
                        goto out;
                }
 
-               /*
-                * alloc_migrate_target should be improooooved!!
-                * migrate_pages returns # of failed pages.
-                */
-               ret = migrate_pages(&source, alloc_migrate_target, NULL, 0,
+               /* Allocate a new page from the nearest neighbor node */
+               ret = migrate_pages(&source, new_node_page, NULL, 0,
                                        MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
                if (ret)
                        putback_movable_pages(&source);