mm/page_isolation.c

   1 /*
   2  * linux/mm/page_isolation.c
   3  */
   4
   5 #include <linux/mm.h>
   6 #include <linux/page-isolation.h>
   7 #include <linux/pageblock-flags.h>
   8 #include <linux/memory.h>
   9 #include "internal.h"
  10
  11 /* called while holding zone->lock */
  12 static void set_pageblock_isolate(struct page *page)
  13 {
  14         if (get_pageblock_migratetype(page) == MIGRATE_ISOLATE)
  15                 return;
  16
  17         set_pageblock_migratetype(page, MIGRATE_ISOLATE);
  18         page_zone(page)->nr_pageblock_isolate++;
  19 }
  20
  21 /* called while holding zone->lock */
  22 static void restore_pageblock_isolate(struct page *page, int migratetype)
  23 {
  24         struct zone *zone = page_zone(page);
  25         if (WARN_ON(get_pageblock_migratetype(page) != MIGRATE_ISOLATE))
  26                 return;
  27
  28         BUG_ON(zone->nr_pageblock_isolate <= 0);
  29         set_pageblock_migratetype(page, migratetype);
  30         zone->nr_pageblock_isolate--;
  31 }
  32
  33 int set_migratetype_isolate(struct page *page)
  34 {
  35         struct zone *zone;
  36         unsigned long flags, pfn;
  37         struct memory_isolate_notify arg;
  38         int notifier_ret;
  39         int ret = -EBUSY;
  40
  41         zone = page_zone(page);
  42
  43         spin_lock_irqsave(&zone->lock, flags);
  44
  45         pfn = page_to_pfn(page);
  46         arg.start_pfn = pfn;
  47         arg.nr_pages = pageblock_nr_pages;
  48         arg.pages_found = 0;
  49
  50         /*
  51          * It may be possible to isolate a pageblock even if the
  52          * migratetype is not MIGRATE_MOVABLE. The memory isolation
  53          * notifier chain is used by balloon drivers to return the
  54          * number of pages in a range that are held by the balloon
  55          * driver to shrink memory. If all the pages are accounted for
  56          * by balloons, are free, or on the LRU, isolation can continue.
  57          * Later, for example, when memory hotplug notifier runs, these
  58          * pages reported as "can be isolated" should be isolated(freed)
  59          * by the balloon driver through the memory notifier chain.
  60          */
  61         notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg);
  62         notifier_ret = notifier_to_errno(notifier_ret);
  63         if (notifier_ret)
  64                 goto out;
  65         /*
  66          * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
  67          * We just check MOVABLE pages.
  68          */
  69         if (!has_unmovable_pages(zone, page, arg.pages_found))
  70                 ret = 0;
  71
  72         /*
  73          * immobile means "not-on-lru" paes. If immobile is larger than
  74          * removable-by-driver pages reported by notifier, we'll fail.
  75          */
  76
  77 out:
  78         if (!ret) {
  79                 unsigned long nr_pages;
  80
  81                 set_pageblock_isolate(page);
  82                 nr_pages = move_freepages_block(zone, page, MIGRATE_ISOLATE);
  83
  84                 __mod_zone_page_state(zone, NR_FREE_PAGES, -nr_pages);
  85         }
  86
  87         spin_unlock_irqrestore(&zone->lock, flags);
  88         if (!ret)
  89                 drain_all_pages();
  90         return ret;
  91 }
  92
  93 void unset_migratetype_isolate(struct page *page, unsigned migratetype)
  94 {
  95         struct zone *zone;
  96         unsigned long flags, nr_pages;
  97
  98         zone = page_zone(page);
  99         spin_lock_irqsave(&zone->lock, flags);
 100         if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
 101                 goto out;
 102         nr_pages = move_freepages_block(zone, page, migratetype);
 103         __mod_zone_page_state(zone, NR_FREE_PAGES, nr_pages);
 104         restore_pageblock_isolate(page, migratetype);
 105 out:
 106         spin_unlock_irqrestore(&zone->lock, flags);
 107 }
 108
 109 static inline struct page *
 110 __first_valid_page(unsigned long pfn, unsigned long nr_pages)
 111 {
 112         int i;
 113         for (i = 0; i < nr_pages; i++)
 114                 if (pfn_valid_within(pfn + i))
 115                         break;
 116         if (unlikely(i == nr_pages))
 117                 return NULL;
 118         return pfn_to_page(pfn + i);
 119 }
 120
 121 /*
 122  * start_isolate_page_range() -- make page-allocation-type of range of pages
 123  * to be MIGRATE_ISOLATE.
 124  * @start_pfn: The lower PFN of the range to be isolated.
 125  * @end_pfn: The upper PFN of the range to be isolated.
 126  * @migratetype: migrate type to set in error recovery.
 127  *
 128  * Making page-allocation-type to be MIGRATE_ISOLATE means free pages in
 129  * the range will never be allocated. Any free pages and pages freed in the
 130  * future will not be allocated again.
 131  *
 132  * start_pfn/end_pfn must be aligned to pageblock_order.
 133  * Returns 0 on success and -EBUSY if any part of range cannot be isolated.
 134  */
 135 int start_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
 136                              unsigned migratetype)
 137 {
 138         unsigned long pfn;
 139         unsigned long undo_pfn;
 140         struct page *page;
 141
 142         BUG_ON((start_pfn) & (pageblock_nr_pages - 1));
 143         BUG_ON((end_pfn) & (pageblock_nr_pages - 1));
 144
 145         for (pfn = start_pfn;
 146              pfn < end_pfn;
 147              pfn += pageblock_nr_pages) {
 148                 page = __first_valid_page(pfn, pageblock_nr_pages);
 149                 if (page && set_migratetype_isolate(page)) {
 150                         undo_pfn = pfn;
 151                         goto undo;
 152                 }
 153         }
 154         return 0;
 155 undo:
 156         for (pfn = start_pfn;
 157              pfn < undo_pfn;
 158              pfn += pageblock_nr_pages)
 159                 unset_migratetype_isolate(pfn_to_page(pfn), migratetype);
 160
 161         return -EBUSY;
 162 }
 163
 164 /*
 165  * Make isolated pages available again.
 166  */
 167 int undo_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
 168                             unsigned migratetype)
 169 {
 170         unsigned long pfn;
 171         struct page *page;
 172         BUG_ON((start_pfn) & (pageblock_nr_pages - 1));
 173         BUG_ON((end_pfn) & (pageblock_nr_pages - 1));
 174         for (pfn = start_pfn;
 175              pfn < end_pfn;
 176              pfn += pageblock_nr_pages) {
 177                 page = __first_valid_page(pfn, pageblock_nr_pages);
 178                 if (!page || get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
 179                         continue;
 180                 unset_migratetype_isolate(page, migratetype);
 181         }
 182         return 0;
 183 }
 184 /*
 185  * Test all pages in the range is free(means isolated) or not.
 186  * all pages in [start_pfn...end_pfn) must be in the same zone.
 187  * zone->lock must be held before call this.
 188  *
 189  * Returns 1 if all pages in the range are isolated.
 190  */
 191 static int
 192 __test_page_isolated_in_pageblock(unsigned long pfn, unsigned long end_pfn)
 193 {
 194         struct page *page;
 195
 196         while (pfn < end_pfn) {
 197                 if (!pfn_valid_within(pfn)) {
 198                         pfn++;
 199                         continue;
 200                 }
 201                 page = pfn_to_page(pfn);
 202                 if (PageBuddy(page))
 203                         pfn += 1 << page_order(page);
 204                 else if (page_count(page) == 0 &&
 205                                 page_private(page) == MIGRATE_ISOLATE)
 206                         pfn += 1;
 207                 else
 208                         break;
 209         }
 210         if (pfn < end_pfn)
 211                 return 0;
 212         return 1;
 213 }
 214
 215 int test_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
 216 {
 217         unsigned long pfn, flags;
 218         struct page *page;
 219         struct zone *zone;
 220         int ret;
 221
 222         /*
 223          * Note: pageblock_nr_page != MAX_ORDER. Then, chunks of free page
 224          * is not aligned to pageblock_nr_pages.
 225          * Then we just check pagetype fist.
 226          */
 227         for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
 228                 page = __first_valid_page(pfn, pageblock_nr_pages);
 229                 if (page && get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
 230                         break;
 231         }
 232         page = __first_valid_page(start_pfn, end_pfn - start_pfn);
 233         if ((pfn < end_pfn) || !page)
 234                 return -EBUSY;
 235         /* Check all pages are free or Marked as ISOLATED */
 236         zone = page_zone(page);
 237         spin_lock_irqsave(&zone->lock, flags);
 238         ret = __test_page_isolated_in_pageblock(start_pfn, end_pfn);
 239         spin_unlock_irqrestore(&zone->lock, flags);
 240         return ret ? 0 : -EBUSY;
 241 }