mm/page_isolation.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * linux/mm/page_isolation.c
   4  */
   5
   6 #include <linux/mm.h>
   7 #include <linux/page-isolation.h>
   8 #include <linux/pageblock-flags.h>
   9 #include <linux/memory.h>
  10 #include <linux/hugetlb.h>
  11 #include <linux/page_owner.h>
  12 #include <linux/migrate.h>
  13 #include "internal.h"
  14
  15 #define CREATE_TRACE_POINTS
  16 #include <trace/events/page_isolation.h>
  17
  18 static int set_migratetype_isolate(struct page *page, int migratetype, int isol_flags)
  19 {
  20         struct page *unmovable = NULL;
  21         struct zone *zone;
  22         unsigned long flags;
  23         int ret = -EBUSY;
  24
  25         zone = page_zone(page);
  26
  27         spin_lock_irqsave(&zone->lock, flags);
  28
  29         /*
  30          * We assume the caller intended to SET migrate type to isolate.
  31          * If it is already set, then someone else must have raced and
  32          * set it before us.  Return -EBUSY
  33          */
  34         if (is_migrate_isolate_page(page))
  35                 goto out;
  36
  37         /*
  38          * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
  39          * We just check MOVABLE pages.
  40          */
  41         unmovable = has_unmovable_pages(zone, page, migratetype, isol_flags);
  42         if (!unmovable) {
  43                 unsigned long nr_pages;
  44                 int mt = get_pageblock_migratetype(page);
  45
  46                 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
  47                 zone->nr_isolate_pageblock++;
  48                 nr_pages = move_freepages_block(zone, page, MIGRATE_ISOLATE,
  49                                                                         NULL);
  50
  51                 __mod_zone_freepage_state(zone, -nr_pages, mt);
  52                 ret = 0;
  53         }
  54
  55 out:
  56         spin_unlock_irqrestore(&zone->lock, flags);
  57         if (!ret) {
  58                 drain_all_pages(zone);
  59         } else {
  60                 WARN_ON_ONCE(zone_idx(zone) == ZONE_MOVABLE);
  61
  62                 if ((isol_flags & REPORT_FAILURE) && unmovable)
  63                         /*
  64                          * printk() with zone->lock held will likely trigger a
  65                          * lockdep splat, so defer it here.
  66                          */
  67                         dump_page(unmovable, "unmovable page");
  68         }
  69
  70         return ret;
  71 }
  72
  73 static void unset_migratetype_isolate(struct page *page, unsigned migratetype)
  74 {
  75         struct zone *zone;
  76         unsigned long flags, nr_pages;
  77         bool isolated_page = false;
  78         unsigned int order;
  79         unsigned long pfn, buddy_pfn;
  80         struct page *buddy;
  81
  82         zone = page_zone(page);
  83         spin_lock_irqsave(&zone->lock, flags);
  84         if (!is_migrate_isolate_page(page))
  85                 goto out;
  86
  87         /*
  88          * Because freepage with more than pageblock_order on isolated
  89          * pageblock is restricted to merge due to freepage counting problem,
  90          * it is possible that there is free buddy page.
  91          * move_freepages_block() doesn't care of merge so we need other
  92          * approach in order to merge them. Isolation and free will make
  93          * these pages to be merged.
  94          */
  95         if (PageBuddy(page)) {
  96                 order = page_order(page);
  97                 if (order >= pageblock_order) {
  98                         pfn = page_to_pfn(page);
  99                         buddy_pfn = __find_buddy_pfn(pfn, order);
 100                         buddy = page + (buddy_pfn - pfn);
 101
 102                         if (pfn_valid_within(buddy_pfn) &&
 103                             !is_migrate_isolate_page(buddy)) {
 104                                 __isolate_free_page(page, order);
 105                                 isolated_page = true;
 106                         }
 107                 }
 108         }
 109
 110         /*
 111          * If we isolate freepage with more than pageblock_order, there
 112          * should be no freepage in the range, so we could avoid costly
 113          * pageblock scanning for freepage moving.
 114          */
 115         if (!isolated_page) {
 116                 nr_pages = move_freepages_block(zone, page, migratetype, NULL);
 117                 __mod_zone_freepage_state(zone, nr_pages, migratetype);
 118         }
 119         set_pageblock_migratetype(page, migratetype);
 120         if (isolated_page)
 121                 __putback_isolated_page(page, order, migratetype);
 122         zone->nr_isolate_pageblock--;
 123 out:
 124         spin_unlock_irqrestore(&zone->lock, flags);
 125 }
 126
 127 static inline struct page *
 128 __first_valid_page(unsigned long pfn, unsigned long nr_pages)
 129 {
 130         int i;
 131
 132         for (i = 0; i < nr_pages; i++) {
 133                 struct page *page;
 134
 135                 page = pfn_to_online_page(pfn + i);
 136                 if (!page)
 137                         continue;
 138                 return page;
 139         }
 140         return NULL;
 141 }
 142
 143 /**
 144  * start_isolate_page_range() - make page-allocation-type of range of pages to
 145  * be MIGRATE_ISOLATE.
 146  * @start_pfn:          The lower PFN of the range to be isolated.
 147  * @end_pfn:            The upper PFN of the range to be isolated.
 148  *                      start_pfn/end_pfn must be aligned to pageblock_order.
 149  * @migratetype:        Migrate type to set in error recovery.
 150  * @flags:              The following flags are allowed (they can be combined in
 151  *                      a bit mask)
 152  *                      MEMORY_OFFLINE - isolate to offline (!allocate) memory
 153  *                                       e.g., skip over PageHWPoison() pages
 154  *                                       and PageOffline() pages.
 155  *                      REPORT_FAILURE - report details about the failure to
 156  *                      isolate the range
 157  *
 158  * Making page-allocation-type to be MIGRATE_ISOLATE means free pages in
 159  * the range will never be allocated. Any free pages and pages freed in the
 160  * future will not be allocated again. If specified range includes migrate types
 161  * other than MOVABLE or CMA, this will fail with -EBUSY. For isolating all
 162  * pages in the range finally, the caller have to free all pages in the range.
 163  * test_page_isolated() can be used for test it.
 164  *
 165  * There is no high level synchronization mechanism that prevents two threads
 166  * from trying to isolate overlapping ranges. If this happens, one thread
 167  * will notice pageblocks in the overlapping range already set to isolate.
 168  * This happens in set_migratetype_isolate, and set_migratetype_isolate
 169  * returns an error. We then clean up by restoring the migration type on
 170  * pageblocks we may have modified and return -EBUSY to caller. This
 171  * prevents two threads from simultaneously working on overlapping ranges.
 172  *
 173  * Return: the number of isolated pageblocks on success and -EBUSY if any part
 174  * of range cannot be isolated.
 175  */
 176 int start_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
 177                              unsigned migratetype, int flags)
 178 {
 179         unsigned long pfn;
 180         unsigned long undo_pfn;
 181         struct page *page;
 182         int nr_isolate_pageblock = 0;
 183
 184         BUG_ON(!IS_ALIGNED(start_pfn, pageblock_nr_pages));
 185         BUG_ON(!IS_ALIGNED(end_pfn, pageblock_nr_pages));
 186
 187         for (pfn = start_pfn;
 188              pfn < end_pfn;
 189              pfn += pageblock_nr_pages) {
 190                 page = __first_valid_page(pfn, pageblock_nr_pages);
 191                 if (page) {
 192                         if (set_migratetype_isolate(page, migratetype, flags)) {
 193                                 undo_pfn = pfn;
 194                                 goto undo;
 195                         }
 196                         nr_isolate_pageblock++;
 197                 }
 198         }
 199         return nr_isolate_pageblock;
 200 undo:
 201         for (pfn = start_pfn;
 202              pfn < undo_pfn;
 203              pfn += pageblock_nr_pages) {
 204                 struct page *page = pfn_to_online_page(pfn);
 205                 if (!page)
 206                         continue;
 207                 unset_migratetype_isolate(page, migratetype);
 208         }
 209
 210         return -EBUSY;
 211 }
 212
 213 /*
 214  * Make isolated pages available again.
 215  */
 216 void undo_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
 217                             unsigned migratetype)
 218 {
 219         unsigned long pfn;
 220         struct page *page;
 221
 222         BUG_ON(!IS_ALIGNED(start_pfn, pageblock_nr_pages));
 223         BUG_ON(!IS_ALIGNED(end_pfn, pageblock_nr_pages));
 224
 225         for (pfn = start_pfn;
 226              pfn < end_pfn;
 227              pfn += pageblock_nr_pages) {
 228                 page = __first_valid_page(pfn, pageblock_nr_pages);
 229                 if (!page || !is_migrate_isolate_page(page))
 230                         continue;
 231                 unset_migratetype_isolate(page, migratetype);
 232         }
 233 }
 234 /*
 235  * Test all pages in the range is free(means isolated) or not.
 236  * all pages in [start_pfn...end_pfn) must be in the same zone.
 237  * zone->lock must be held before call this.
 238  *
 239  * Returns the last tested pfn.
 240  */
 241 static unsigned long
 242 __test_page_isolated_in_pageblock(unsigned long pfn, unsigned long end_pfn,
 243                                   int flags)
 244 {
 245         struct page *page;
 246
 247         while (pfn < end_pfn) {
 248                 if (!pfn_valid_within(pfn)) {
 249                         pfn++;
 250                         continue;
 251                 }
 252                 page = pfn_to_page(pfn);
 253                 if (PageBuddy(page))
 254                         /*
 255                          * If the page is on a free list, it has to be on
 256                          * the correct MIGRATE_ISOLATE freelist. There is no
 257                          * simple way to verify that as VM_BUG_ON(), though.
 258                          */
 259                         pfn += 1 << page_order(page);
 260                 else if ((flags & MEMORY_OFFLINE) && PageHWPoison(page))
 261                         /* A HWPoisoned page cannot be also PageBuddy */
 262                         pfn++;
 263                 else if ((flags & MEMORY_OFFLINE) && PageOffline(page) &&
 264                          !page_count(page))
 265                         /*
 266                          * The responsible driver agreed to skip PageOffline()
 267                          * pages when offlining memory by dropping its
 268                          * reference in MEM_GOING_OFFLINE.
 269                          */
 270                         pfn++;
 271                 else
 272                         break;
 273         }
 274
 275         return pfn;
 276 }
 277
 278 /* Caller should ensure that requested range is in a single zone */
 279 int test_pages_isolated(unsigned long start_pfn, unsigned long end_pfn,
 280                         int isol_flags)
 281 {
 282         unsigned long pfn, flags;
 283         struct page *page;
 284         struct zone *zone;
 285
 286         /*
 287          * Note: pageblock_nr_pages != MAX_ORDER. Then, chunks of free pages
 288          * are not aligned to pageblock_nr_pages.
 289          * Then we just check migratetype first.
 290          */
 291         for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
 292                 page = __first_valid_page(pfn, pageblock_nr_pages);
 293                 if (page && !is_migrate_isolate_page(page))
 294                         break;
 295         }
 296         page = __first_valid_page(start_pfn, end_pfn - start_pfn);
 297         if ((pfn < end_pfn) || !page)
 298                 return -EBUSY;
 299         /* Check all pages are free or marked as ISOLATED */
 300         zone = page_zone(page);
 301         spin_lock_irqsave(&zone->lock, flags);
 302         pfn = __test_page_isolated_in_pageblock(start_pfn, end_pfn, isol_flags);
 303         spin_unlock_irqrestore(&zone->lock, flags);
 304
 305         trace_test_pages_isolated(start_pfn, end_pfn, pfn);
 306
 307         return pfn < end_pfn ? -EBUSY : 0;
 308 }
 309
 310 struct page *alloc_migrate_target(struct page *page, unsigned long private)
 311 {
 312         return new_page_nodemask(page, numa_node_id(), &node_states[N_MEMORY]);
 313 }