mm/vmscan: remove page_queue_congested() comment
[safe/jmp/linux-2.6] / mm / page_alloc.c
index 303607f..84d9da1 100644 (file)
@@ -23,6 +23,7 @@
 #include <linux/bootmem.h>
 #include <linux/compiler.h>
 #include <linux/kernel.h>
+#include <linux/kmemcheck.h>
 #include <linux/module.h>
 #include <linux/suspend.h>
 #include <linux/pagevec.h>
@@ -47,6 +48,7 @@
 #include <linux/page_cgroup.h>
 #include <linux/debugobjects.h>
 #include <linux/kmemleak.h>
+#include <trace/events/kmem.h>
 
 #include <asm/tlbflush.h>
 #include <asm/div64.h>
@@ -72,6 +74,7 @@ unsigned long totalram_pages __read_mostly;
 unsigned long totalreserve_pages __read_mostly;
 unsigned long highest_memmap_pfn __read_mostly;
 int percpu_pagelist_fraction;
+gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
 
 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
 int pageblock_order __read_mostly;
@@ -121,8 +124,8 @@ static char * const zone_names[MAX_NR_ZONES] = {
 
 int min_free_kbytes = 1024;
 
-unsigned long __meminitdata nr_kernel_pages;
-unsigned long __meminitdata nr_all_pages;
+static unsigned long __meminitdata nr_kernel_pages;
+static unsigned long __meminitdata nr_all_pages;
 static unsigned long __meminitdata dma_reserve;
 
 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
@@ -486,7 +489,6 @@ static inline void __free_one_page(struct page *page,
  */
 static inline void free_page_mlock(struct page *page)
 {
-       __ClearPageMlocked(page);
        __dec_zone_page_state(page, NR_MLOCK);
        __count_vm_event(UNEVICTABLE_MLOCKFREED);
 }
@@ -534,6 +536,7 @@ static void free_pages_bulk(struct zone *zone, int count,
                page = list_entry(list->prev, struct page, lru);
                /* have to delete it as __free_one_page list manipulates */
                list_del(&page->lru);
+               trace_mm_page_pcpu_drain(page, order, page_private(page));
                __free_one_page(page, zone, order, page_private(page));
        }
        spin_unlock(&zone->lock);
@@ -556,7 +559,9 @@ static void __free_pages_ok(struct page *page, unsigned int order)
        unsigned long flags;
        int i;
        int bad = 0;
-       int clearMlocked = PageMlocked(page);
+       int wasMlocked = __TestClearPageMlocked(page);
+
+       kmemcheck_free_shadow(page, order);
 
        for (i = 0 ; i < (1 << order) ; ++i)
                bad += free_pages_check(page + i);
@@ -572,7 +577,7 @@ static void __free_pages_ok(struct page *page, unsigned int order)
        kernel_map_pages(page, 1 << order, 0);
 
        local_irq_save(flags);
-       if (unlikely(clearMlocked))
+       if (unlikely(wasMlocked))
                free_page_mlock(page);
        __count_vm_events(PGFREE, 1 << order);
        free_one_page(page_zone(page), page, order,
@@ -780,6 +785,17 @@ static int move_freepages_block(struct zone *zone, struct page *page,
        return move_freepages(zone, start_page, end_page, migratetype);
 }
 
+static void change_pageblock_range(struct page *pageblock_page,
+                                       int start_order, int migratetype)
+{
+       int nr_pageblocks = 1 << (start_order - pageblock_order);
+
+       while (nr_pageblocks--) {
+               set_pageblock_migratetype(pageblock_page, migratetype);
+               pageblock_page += pageblock_nr_pages;
+       }
+}
+
 /* Remove an element from the buddy allocator from the fallback list */
 static inline struct page *
 __rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
@@ -814,13 +830,15 @@ __rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
                         * agressive about taking ownership of free pages
                         */
                        if (unlikely(current_order >= (pageblock_order >> 1)) ||
-                                       start_migratetype == MIGRATE_RECLAIMABLE) {
+                                       start_migratetype == MIGRATE_RECLAIMABLE ||
+                                       page_group_by_mobility_disabled) {
                                unsigned long pages;
                                pages = move_freepages_block(zone, page,
                                                                start_migratetype);
 
                                /* Claim the whole block if over half of it is free */
-                               if (pages >= (1 << (pageblock_order-1)))
+                               if (pages >= (1 << (pageblock_order-1)) ||
+                                               page_group_by_mobility_disabled)
                                        set_pageblock_migratetype(page,
                                                                start_migratetype);
 
@@ -831,11 +849,16 @@ __rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
                        list_del(&page->lru);
                        rmv_page_order(page);
 
-                       if (current_order == pageblock_order)
-                               set_pageblock_migratetype(page,
+                       /* Take ownership for orders >= pageblock_order */
+                       if (current_order >= pageblock_order)
+                               change_pageblock_range(page, current_order,
                                                        start_migratetype);
 
                        expand(zone, page, order, current_order, area, migratetype);
+
+                       trace_mm_page_alloc_extfrag(page, order, current_order,
+                               start_migratetype, migratetype);
+
                        return page;
                }
        }
@@ -869,6 +892,7 @@ retry_reserve:
                }
        }
 
+       trace_mm_page_alloc_zone_locked(page, order, migratetype);
        return page;
 }
 
@@ -879,7 +903,7 @@ retry_reserve:
  */
 static int rmqueue_bulk(struct zone *zone, unsigned int order, 
                        unsigned long count, struct list_head *list,
-                       int migratetype)
+                       int migratetype, int cold)
 {
        int i;
        
@@ -898,7 +922,10 @@ static int rmqueue_bulk(struct zone *zone, unsigned int order,
                 * merge IO requests if the physical pages are ordered
                 * properly.
                 */
-               list_add(&page->lru, list);
+               if (likely(cold == 0))
+                       list_add(&page->lru, list);
+               else
+                       list_add_tail(&page->lru, list);
                set_page_private(page, migratetype);
                list = &page->lru;
        }
@@ -1018,7 +1045,9 @@ static void free_hot_cold_page(struct page *page, int cold)
        struct zone *zone = page_zone(page);
        struct per_cpu_pages *pcp;
        unsigned long flags;
-       int clearMlocked = PageMlocked(page);
+       int wasMlocked = __TestClearPageMlocked(page);
+
+       kmemcheck_free_shadow(page, 0);
 
        if (PageAnon(page))
                page->mapping = NULL;
@@ -1035,7 +1064,7 @@ static void free_hot_cold_page(struct page *page, int cold)
        pcp = &zone_pcp(zone, get_cpu())->pcp;
        set_page_private(page, get_pageblock_migratetype(page));
        local_irq_save(flags);
-       if (unlikely(clearMlocked))
+       if (unlikely(wasMlocked))
                free_page_mlock(page);
        __count_vm_event(PGFREE);
 
@@ -1054,14 +1083,10 @@ static void free_hot_cold_page(struct page *page, int cold)
 
 void free_hot_page(struct page *page)
 {
+       trace_mm_page_free_direct(page, 0);
        free_hot_cold_page(page, 0);
 }
        
-void free_cold_page(struct page *page)
-{
-       free_hot_cold_page(page, 1);
-}
-
 /*
  * split_page takes a non-compound higher-order page, and splits it into
  * n (1<<order) sub-pages: page[0..n]
@@ -1076,6 +1101,16 @@ void split_page(struct page *page, unsigned int order)
 
        VM_BUG_ON(PageCompound(page));
        VM_BUG_ON(!page_count(page));
+
+#ifdef CONFIG_KMEMCHECK
+       /*
+        * Split shadow pages too, because free(page[0]) would
+        * otherwise free the whole shadow.
+        */
+       if (kmemcheck_page_is_tracked(page))
+               split_page(virt_to_page(page[0].shadow), order);
+#endif
+
        for (i = 1; i < (1 << order); i++)
                set_page_refcounted(page + i);
 }
@@ -1104,7 +1139,8 @@ again:
                local_irq_save(flags);
                if (!pcp->count) {
                        pcp->count = rmqueue_bulk(zone, 0,
-                                       pcp->batch, &pcp->list, migratetype);
+                                       pcp->batch, &pcp->list,
+                                       migratetype, cold);
                        if (unlikely(!pcp->count))
                                goto failed;
                }
@@ -1122,9 +1158,20 @@ again:
 
                /* Allocate more to the pcp list if necessary */
                if (unlikely(&page->lru == &pcp->list)) {
+                       int get_one_page = 0;
+
                        pcp->count += rmqueue_bulk(zone, 0,
-                                       pcp->batch, &pcp->list, migratetype);
-                       page = list_entry(pcp->list.next, struct page, lru);
+                                       pcp->batch, &pcp->list,
+                                       migratetype, cold);
+                       list_for_each_entry(page, &pcp->list, lru) {
+                               if (get_pageblock_migratetype(page) !=
+                                           MIGRATE_ISOLATE) {
+                                       get_one_page = 1;
+                                       break;
+                               }
+                       }
+                       if (!get_one_page)
+                               goto failed;
                }
 
                list_del(&page->lru);
@@ -1138,10 +1185,10 @@ again:
                         * properly detect and handle allocation failures.
                         *
                         * We most definitely don't want callers attempting to
-                        * allocate greater than single-page units with
+                        * allocate greater than order-1 page units with
                         * __GFP_NOFAIL.
                         */
-                       WARN_ON_ONCE(order > 0);
+                       WARN_ON_ONCE(order > 1);
                }
                spin_lock_irqsave(&zone->lock, flags);
                page = __rmqueue(zone, order, migratetype);
@@ -1462,15 +1509,33 @@ zonelist_scan:
                BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
                if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
                        unsigned long mark;
+                       int ret;
+
                        mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
-                       if (!zone_watermark_ok(zone, order, mark,
-                                   classzone_idx, alloc_flags)) {
-                               if (!zone_reclaim_mode ||
-                                   !zone_reclaim(zone, gfp_mask, order))
+                       if (zone_watermark_ok(zone, order, mark,
+                                   classzone_idx, alloc_flags))
+                               goto try_this_zone;
+
+                       if (zone_reclaim_mode == 0)
+                               goto this_zone_full;
+
+                       ret = zone_reclaim(zone, gfp_mask, order);
+                       switch (ret) {
+                       case ZONE_RECLAIM_NOSCAN:
+                               /* did not scan */
+                               goto try_next_zone;
+                       case ZONE_RECLAIM_FULL:
+                               /* scanned but unreclaimable */
+                               goto this_zone_full;
+                       default:
+                               /* did we reclaim enough */
+                               if (!zone_watermark_ok(zone, order, mark,
+                                               classzone_idx, alloc_flags))
                                        goto this_zone_full;
                        }
                }
 
+try_this_zone:
                page = buffered_rmqueue(preferred_zone, zone, order,
                                                gfp_mask, migratetype);
                if (page)
@@ -1561,7 +1626,7 @@ __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
                goto out;
 
        /* The OOM killer will not help higher order allocs */
-       if (order > PAGE_ALLOC_COSTLY_ORDER)
+       if (order > PAGE_ALLOC_COSTLY_ORDER && !(gfp_mask & __GFP_NOFAIL))
                goto out;
 
        /* Exhausted what can be done so it's blamo time */
@@ -1587,10 +1652,6 @@ __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
 
        /* We now go into synchronous reclaim */
        cpuset_memory_pressure_bump();
-
-       /*
-        * The task's cpuset might have expanded its set of allowable nodes
-        */
        p->flags |= PF_MEMALLOC;
        lockdep_set_current_reclaim_state(gfp_mask);
        reclaim_state.reclaimed_slab = 0;
@@ -1633,7 +1694,7 @@ __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
                        preferred_zone, migratetype);
 
                if (!page && gfp_mask & __GFP_NOFAIL)
-                       congestion_wait(WRITE, HZ/50);
+                       congestion_wait(BLK_RW_ASYNC, HZ/50);
        } while (!page && (gfp_mask & __GFP_NOFAIL));
 
        return page;
@@ -1707,8 +1768,10 @@ __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
         * be using allocators in order of preference for an area that is
         * too large.
         */
-       if (WARN_ON_ONCE(order >= MAX_ORDER))
+       if (order >= MAX_ORDER) {
+               WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
                return NULL;
+       }
 
        /*
         * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
@@ -1723,6 +1786,7 @@ __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
 
        wake_all_kswapd(order, zonelist, high_zoneidx);
 
+restart:
        /*
         * OK, we're below the kswapd watermark and have kicked background
         * reclaim. Now things get more complex, so set up alloc_flags according
@@ -1730,7 +1794,6 @@ __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
         */
        alloc_flags = gfp_to_alloc_flags(gfp_mask);
 
-restart:
        /* This is the last chance, in general, before the goto nopage. */
        page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
                        high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
@@ -1756,6 +1819,10 @@ rebalance:
        if (p->flags & PF_MEMALLOC)
                goto nopage;
 
+       /* Avoid allocations with no watermarks from looping endlessly */
+       if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
+               goto nopage;
+
        /* Try direct reclaim and then allocating */
        page = __alloc_pages_direct_reclaim(gfp_mask, order,
                                        zonelist, high_zoneidx,
@@ -1781,11 +1848,13 @@ rebalance:
                                goto got_pg;
 
                        /*
-                        * The OOM killer does not trigger for high-order allocations
-                        * but if no progress is being made, there are no other
-                        * options and retrying is unlikely to help
+                        * The OOM killer does not trigger for high-order
+                        * ~__GFP_NOFAIL allocations so if no progress is being
+                        * made, there are no other options and retrying is
+                        * unlikely to help.
                         */
-                       if (order > PAGE_ALLOC_COSTLY_ORDER)
+                       if (order > PAGE_ALLOC_COSTLY_ORDER &&
+                                               !(gfp_mask & __GFP_NOFAIL))
                                goto nopage;
 
                        goto restart;
@@ -1796,7 +1865,7 @@ rebalance:
        pages_reclaimed += did_some_progress;
        if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) {
                /* Wait for some write requests to complete then retry */
-               congestion_wait(WRITE, HZ/50);
+               congestion_wait(BLK_RW_ASYNC, HZ/50);
                goto rebalance;
        }
 
@@ -1808,7 +1877,10 @@ nopage:
                dump_stack();
                show_mem();
        }
+       return page;
 got_pg:
+       if (kmemcheck_enabled)
+               kmemcheck_pagealloc_alloc(page, order, gfp_mask);
        return page;
 
 }
@@ -1825,6 +1897,8 @@ __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
        struct page *page;
        int migratetype = allocflags_to_migratetype(gfp_mask);
 
+       gfp_mask &= gfp_allowed_mask;
+
        lockdep_trace_alloc(gfp_mask);
 
        might_sleep_if(gfp_mask & __GFP_WAIT);
@@ -1854,6 +1928,7 @@ __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
                                zonelist, high_zoneidx, nodemask,
                                preferred_zone, migratetype);
 
+       trace_mm_page_alloc(page, order, gfp_mask, migratetype);
        return page;
 }
 EXPORT_SYMBOL(__alloc_pages_nodemask);
@@ -1863,44 +1938,41 @@ EXPORT_SYMBOL(__alloc_pages_nodemask);
  */
 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
 {
-       struct page * page;
+       struct page *page;
+
+       /*
+        * __get_free_pages() returns a 32-bit address, which cannot represent
+        * a highmem page
+        */
+       VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
+
        page = alloc_pages(gfp_mask, order);
        if (!page)
                return 0;
        return (unsigned long) page_address(page);
 }
-
 EXPORT_SYMBOL(__get_free_pages);
 
 unsigned long get_zeroed_page(gfp_t gfp_mask)
 {
-       struct page * page;
-
-       /*
-        * get_zeroed_page() returns a 32-bit address, which cannot represent
-        * a highmem page
-        */
-       VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
-
-       page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
-       if (page)
-               return (unsigned long) page_address(page);
-       return 0;
+       return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
 }
-
 EXPORT_SYMBOL(get_zeroed_page);
 
 void __pagevec_free(struct pagevec *pvec)
 {
        int i = pagevec_count(pvec);
 
-       while (--i >= 0)
+       while (--i >= 0) {
+               trace_mm_pagevec_free(pvec->pages[i], pvec->cold);
                free_hot_cold_page(pvec->pages[i], pvec->cold);
+       }
 }
 
 void __free_pages(struct page *page, unsigned int order)
 {
        if (put_page_testzero(page)) {
+               trace_mm_page_free_direct(page, order);
                if (order == 0)
                        free_hot_page(page);
                else
@@ -1943,7 +2015,7 @@ void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
                unsigned long alloc_end = addr + (PAGE_SIZE << order);
                unsigned long used = addr + PAGE_ALIGN(size);
 
-               split_page(virt_to_page(addr), order);
+               split_page(virt_to_page((void *)addr), order);
                while (used < alloc_end) {
                        free_page(used);
                        used += PAGE_SIZE;
@@ -2075,28 +2147,28 @@ void show_free_areas(void)
                }
        }
 
-       printk("Active_anon:%lu active_file:%lu inactive_anon:%lu\n"
-               " inactive_file:%lu"
-//TODO:  check/adjust line lengths
-#ifdef CONFIG_UNEVICTABLE_LRU
+       printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
+               " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
                " unevictable:%lu"
-#endif
-               " dirty:%lu writeback:%lu unstable:%lu\n"
-               " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
+               " dirty:%lu writeback:%lu unstable:%lu buffer:%lu\n"
+               " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
+               " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n",
                global_page_state(NR_ACTIVE_ANON),
-               global_page_state(NR_ACTIVE_FILE),
                global_page_state(NR_INACTIVE_ANON),
+               global_page_state(NR_ISOLATED_ANON),
+               global_page_state(NR_ACTIVE_FILE),
                global_page_state(NR_INACTIVE_FILE),
-#ifdef CONFIG_UNEVICTABLE_LRU
+               global_page_state(NR_ISOLATED_FILE),
                global_page_state(NR_UNEVICTABLE),
-#endif
                global_page_state(NR_FILE_DIRTY),
                global_page_state(NR_WRITEBACK),
                global_page_state(NR_UNSTABLE_NFS),
+               nr_blockdev_pages(),
                global_page_state(NR_FREE_PAGES),
-               global_page_state(NR_SLAB_RECLAIMABLE) +
-                       global_page_state(NR_SLAB_UNRECLAIMABLE),
+               global_page_state(NR_SLAB_RECLAIMABLE),
+               global_page_state(NR_SLAB_UNRECLAIMABLE),
                global_page_state(NR_FILE_MAPPED),
+               global_page_state(NR_SHMEM),
                global_page_state(NR_PAGETABLE),
                global_page_state(NR_BOUNCE));
 
@@ -2113,10 +2185,22 @@ void show_free_areas(void)
                        " inactive_anon:%lukB"
                        " active_file:%lukB"
                        " inactive_file:%lukB"
-#ifdef CONFIG_UNEVICTABLE_LRU
                        " unevictable:%lukB"
-#endif
+                       " isolated(anon):%lukB"
+                       " isolated(file):%lukB"
                        " present:%lukB"
+                       " mlocked:%lukB"
+                       " dirty:%lukB"
+                       " writeback:%lukB"
+                       " mapped:%lukB"
+                       " shmem:%lukB"
+                       " slab_reclaimable:%lukB"
+                       " slab_unreclaimable:%lukB"
+                       " kernel_stack:%lukB"
+                       " pagetables:%lukB"
+                       " unstable:%lukB"
+                       " bounce:%lukB"
+                       " writeback_tmp:%lukB"
                        " pages_scanned:%lu"
                        " all_unreclaimable? %s"
                        "\n",
@@ -2129,10 +2213,23 @@ void show_free_areas(void)
                        K(zone_page_state(zone, NR_INACTIVE_ANON)),
                        K(zone_page_state(zone, NR_ACTIVE_FILE)),
                        K(zone_page_state(zone, NR_INACTIVE_FILE)),
-#ifdef CONFIG_UNEVICTABLE_LRU
                        K(zone_page_state(zone, NR_UNEVICTABLE)),
-#endif
+                       K(zone_page_state(zone, NR_ISOLATED_ANON)),
+                       K(zone_page_state(zone, NR_ISOLATED_FILE)),
                        K(zone->present_pages),
+                       K(zone_page_state(zone, NR_MLOCK)),
+                       K(zone_page_state(zone, NR_FILE_DIRTY)),
+                       K(zone_page_state(zone, NR_WRITEBACK)),
+                       K(zone_page_state(zone, NR_FILE_MAPPED)),
+                       K(zone_page_state(zone, NR_SHMEM)),
+                       K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
+                       K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
+                       zone_page_state(zone, NR_KERNEL_STACK) *
+                               THREAD_SIZE / 1024,
+                       K(zone_page_state(zone, NR_PAGETABLE)),
+                       K(zone_page_state(zone, NR_UNSTABLE_NFS)),
+                       K(zone_page_state(zone, NR_BOUNCE)),
+                       K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
                        zone->pages_scanned,
                        (zone_is_all_unreclaimable(zone) ? "yes" : "no")
                        );
@@ -2502,7 +2599,6 @@ static void build_zonelists(pg_data_t *pgdat)
        prev_node = local_node;
        nodes_clear(used_mask);
 
-       memset(node_load, 0, sizeof(node_load));
        memset(node_order, 0, sizeof(node_order));
        j = 0;
 
@@ -2611,6 +2707,9 @@ static int __build_all_zonelists(void *dummy)
 {
        int nid;
 
+#ifdef CONFIG_NUMA
+       memset(node_load, 0, sizeof(node_load));
+#endif
        for_each_online_node(nid) {
                pg_data_t *pgdat = NODE_DATA(nid);
 
@@ -2737,7 +2836,8 @@ static void setup_zone_migrate_reserve(struct zone *zone)
 {
        unsigned long start_pfn, pfn, end_pfn;
        struct page *page;
-       unsigned long reserve, block_migratetype;
+       unsigned long block_migratetype;
+       int reserve;
 
        /* Get the start pfn, end pfn and the number of blocks to reserve */
        start_pfn = zone->zone_start_pfn;
@@ -2745,6 +2845,15 @@ static void setup_zone_migrate_reserve(struct zone *zone)
        reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
                                                        pageblock_order;
 
+       /*
+        * Reserve blocks are generally in place to help high-order atomic
+        * allocations that are short-lived. A min_free_kbytes value that
+        * would result in more than 2 reserve blocks for atomic allocations
+        * is assumed to be in place to help anti-fragmentation for the
+        * future allocation of hugepages at runtime.
+        */
+       reserve = min(2, reserve);
+
        for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
                if (!pfn_valid(pfn))
                        continue;
@@ -2995,7 +3104,7 @@ bad:
                if (dzone == zone)
                        break;
                kfree(zone_pcp(dzone, cpu));
-               zone_pcp(dzone, cpu) = NULL;
+               zone_pcp(dzone, cpu) = &boot_pageset[cpu];
        }
        return -ENOMEM;
 }
@@ -3010,7 +3119,7 @@ static inline void free_zone_pagesets(int cpu)
                /* Free per_cpu_pageset if it is slab allocated */
                if (pset != &boot_pageset[cpu])
                        kfree(pset);
-               zone_pcp(zone, cpu) = NULL;
+               zone_pcp(zone, cpu) = &boot_pageset[cpu];
        }
 }
 
@@ -3100,6 +3209,32 @@ int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
        return 0;
 }
 
+static int __zone_pcp_update(void *data)
+{
+       struct zone *zone = data;
+       int cpu;
+       unsigned long batch = zone_batchsize(zone), flags;
+
+       for (cpu = 0; cpu < NR_CPUS; cpu++) {
+               struct per_cpu_pageset *pset;
+               struct per_cpu_pages *pcp;
+
+               pset = zone_pcp(zone, cpu);
+               pcp = &pset->pcp;
+
+               local_irq_save(flags);
+               free_pages_bulk(zone, pcp->count, &pcp->list, 0);
+               setup_pageset(pset, batch);
+               local_irq_restore(flags);
+       }
+       return 0;
+}
+
+void zone_pcp_update(struct zone *zone)
+{
+       stop_machine(__zone_pcp_update, zone, NULL);
+}
+
 static __meminit void zone_pcp_init(struct zone *zone)
 {
        int cpu;
@@ -3674,7 +3809,7 @@ static void __paginginit free_area_init_core(struct pglist_data *pgdat,
                zone_pcp_init(zone);
                for_each_lru(l) {
                        INIT_LIST_HEAD(&zone->lru[l].list);
-                       zone->lru[l].nr_saved_scan = 0;
+                       zone->reclaim_stat.nr_saved_scan[l] = 0;
                }
                zone->reclaim_stat.recent_rotated[0] = 0;
                zone->reclaim_stat.recent_rotated[1] = 0;
@@ -4001,6 +4136,8 @@ static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
        int i, nid;
        unsigned long usable_startpfn;
        unsigned long kernelcore_node, kernelcore_remaining;
+       /* save the state before borrow the nodemask */
+       nodemask_t saved_node_state = node_states[N_HIGH_MEMORY];
        unsigned long totalpages = early_calculate_totalpages();
        int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
 
@@ -4028,7 +4165,7 @@ static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
 
        /* If kernelcore was not specified, there is no ZONE_MOVABLE */
        if (!required_kernelcore)
-               return;
+               goto out;
 
        /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
        find_usable_zone_for_movable();
@@ -4127,6 +4264,10 @@ restart:
        for (nid = 0; nid < MAX_NUMNODES; nid++)
                zone_movable_pfn[nid] =
                        roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
+
+out:
+       /* restore the node_state */
+       node_states[N_HIGH_MEMORY] = saved_node_state;
 }
 
 /* Any regular memory on that node ? */
@@ -4397,7 +4538,7 @@ static void setup_per_zone_lowmem_reserve(void)
 
 /**
  * setup_per_zone_wmarks - called when min_free_kbytes changes
- * or when memory is hot-added
+ * or when memory is hot-{added|removed}
  *
  * Ensures that the watermark[min,low,high] values for each zone are set
  * correctly with respect to min_free_kbytes.
@@ -4457,7 +4598,7 @@ void setup_per_zone_wmarks(void)
        calculate_totalreserve_pages();
 }
 
-/**
+/*
  * The inactive anon list should be small enough that the VM never has to
  * do too much work, but large enough that each inactive page has a chance
  * to be referenced again before it is swapped out.
@@ -4623,7 +4764,7 @@ int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
        ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
        if (!write || (ret == -EINVAL))
                return ret;
-       for_each_zone(zone) {
+       for_each_populated_zone(zone) {
                for_each_online_cpu(cpu) {
                        unsigned long  high;
                        high = zone->present_pages / percpu_pagelist_fraction;
@@ -4680,7 +4821,14 @@ void *__init alloc_large_system_hash(const char *tablename,
                        numentries <<= (PAGE_SHIFT - scale);
 
                /* Make sure we've got at least a 0-order allocation.. */
-               if (unlikely((numentries * bucketsize) < PAGE_SIZE))
+               if (unlikely(flags & HASH_SMALL)) {
+                       /* Makes no sense without HASH_EARLY */
+                       WARN_ON(!(flags & HASH_EARLY));
+                       if (!(numentries >> *_hash_shift)) {
+                               numentries = 1UL << *_hash_shift;
+                               BUG_ON(!numentries);
+                       }
+               } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
                        numentries = PAGE_SIZE / bucketsize;
        }
        numentries = roundup_pow_of_two(numentries);
@@ -4708,8 +4856,10 @@ void *__init alloc_large_system_hash(const char *tablename,
                         * some pages at the end of hash table which
                         * alloc_pages_exact() automatically does
                         */
-                       if (get_order(size) < MAX_ORDER)
+                       if (get_order(size) < MAX_ORDER) {
                                table = alloc_pages_exact(size, GFP_ATOMIC);
+                               kmemleak_alloc(table, size, 1, GFP_ATOMIC);
+                       }
                }
        } while (!table && size > PAGE_SIZE && --log2qty);
 
@@ -4727,16 +4877,6 @@ void *__init alloc_large_system_hash(const char *tablename,
        if (_hash_mask)
                *_hash_mask = (1 << log2qty) - 1;
 
-       /*
-        * If hashdist is set, the table allocation is done with __vmalloc()
-        * which invokes the kmemleak_alloc() callback. This function may also
-        * be called before the slab and kmemleak are initialised when
-        * kmemleak simply buffers the request to be executed later
-        * (GFP_ATOMIC flag ignored in this case).
-        */
-       if (!hashdist)
-               kmemleak_alloc(table, size, 1, GFP_ATOMIC);
-
        return table;
 }
 
@@ -4830,13 +4970,16 @@ int set_migratetype_isolate(struct page *page)
        struct zone *zone;
        unsigned long flags;
        int ret = -EBUSY;
+       int zone_idx;
 
        zone = page_zone(page);
+       zone_idx = zone_idx(zone);
        spin_lock_irqsave(&zone->lock, flags);
        /*
         * In future, more migrate types will be able to be isolation target.
         */
-       if (get_pageblock_migratetype(page) != MIGRATE_MOVABLE)
+       if (get_pageblock_migratetype(page) != MIGRATE_MOVABLE &&
+           zone_idx != ZONE_MOVABLE)
                goto out;
        set_pageblock_migratetype(page, MIGRATE_ISOLATE);
        move_freepages_block(zone, page, MIGRATE_ISOLATE);