Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 1 | /* |
| 2 | * Memory Migration functionality - linux/mm/migration.c |
| 3 | * |
| 4 | * Copyright (C) 2006 Silicon Graphics, Inc., Christoph Lameter |
| 5 | * |
| 6 | * Page migration was first developed in the context of the memory hotplug |
| 7 | * project. The main authors of the migration code are: |
| 8 | * |
| 9 | * IWAMOTO Toshihiro <iwamoto@valinux.co.jp> |
| 10 | * Hirokazu Takahashi <taka@valinux.co.jp> |
| 11 | * Dave Hansen <haveblue@us.ibm.com> |
Christoph Lameter | cde5353 | 2008-07-04 09:59:22 -0700 | [diff] [blame] | 12 | * Christoph Lameter |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 13 | */ |
| 14 | |
| 15 | #include <linux/migrate.h> |
Paul Gortmaker | b95f1b31 | 2011-10-16 02:01:52 -0400 | [diff] [blame] | 16 | #include <linux/export.h> |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 17 | #include <linux/swap.h> |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 18 | #include <linux/swapops.h> |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 19 | #include <linux/pagemap.h> |
Christoph Lameter | e23ca00 | 2006-04-10 22:52:57 -0700 | [diff] [blame] | 20 | #include <linux/buffer_head.h> |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 21 | #include <linux/mm_inline.h> |
Pavel Emelyanov | b488893 | 2007-10-18 23:40:14 -0700 | [diff] [blame] | 22 | #include <linux/nsproxy.h> |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 23 | #include <linux/pagevec.h> |
Hugh Dickins | e9995ef | 2009-12-14 17:59:31 -0800 | [diff] [blame] | 24 | #include <linux/ksm.h> |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 25 | #include <linux/rmap.h> |
| 26 | #include <linux/topology.h> |
| 27 | #include <linux/cpu.h> |
| 28 | #include <linux/cpuset.h> |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 29 | #include <linux/writeback.h> |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 30 | #include <linux/mempolicy.h> |
| 31 | #include <linux/vmalloc.h> |
David Quigley | 86c3a76 | 2006-06-23 02:04:02 -0700 | [diff] [blame] | 32 | #include <linux/security.h> |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 33 | #include <linux/memcontrol.h> |
Adrian Bunk | 4f5ca26 | 2008-07-23 21:27:02 -0700 | [diff] [blame] | 34 | #include <linux/syscalls.h> |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 35 | #include <linux/hugetlb.h> |
Aneesh Kumar K.V | 8e6ac7f | 2012-07-31 16:42:27 -0700 | [diff] [blame] | 36 | #include <linux/hugetlb_cgroup.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 37 | #include <linux/gfp.h> |
Rafael Aquini | bf6bddf | 2012-12-11 16:02:42 -0800 | [diff] [blame] | 38 | #include <linux/balloon_compaction.h> |
Mel Gorman | f714f4f | 2013-12-18 17:08:33 -0800 | [diff] [blame] | 39 | #include <linux/mmu_notifier.h> |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 40 | |
Michal Nazarewicz | 0d1836c | 2010-12-21 17:24:26 -0800 | [diff] [blame] | 41 | #include <asm/tlbflush.h> |
| 42 | |
Mel Gorman | 7b2a2d4 | 2012-10-19 14:07:31 +0100 | [diff] [blame] | 43 | #define CREATE_TRACE_POINTS |
| 44 | #include <trace/events/migrate.h> |
| 45 | |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 46 | #include "internal.h" |
| 47 | |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 48 | /* |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 49 | * migrate_prep() needs to be called before we start compiling a list of pages |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 50 | * to be migrated using isolate_lru_page(). If scheduling work on other CPUs is |
| 51 | * undesirable, use migrate_prep_local() |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 52 | */ |
| 53 | int migrate_prep(void) |
| 54 | { |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 55 | /* |
| 56 | * Clear the LRU lists so pages can be isolated. |
| 57 | * Note that pages may be moved off the LRU after we have |
| 58 | * drained them. Those pages will fail to migrate like other |
| 59 | * pages that may be busy. |
| 60 | */ |
| 61 | lru_add_drain_all(); |
| 62 | |
| 63 | return 0; |
| 64 | } |
| 65 | |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 66 | /* Do the necessary work of migrate_prep but not if it involves other CPUs */ |
| 67 | int migrate_prep_local(void) |
| 68 | { |
| 69 | lru_add_drain(); |
| 70 | |
| 71 | return 0; |
| 72 | } |
| 73 | |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 74 | /* |
Rafael Aquini | 5733c7d | 2012-12-11 16:02:47 -0800 | [diff] [blame] | 75 | * Put previously isolated pages back onto the appropriate lists |
| 76 | * from where they were once taken off for compaction/migration. |
| 77 | * |
Joonsoo Kim | 59c82b7 | 2014-01-21 15:51:17 -0800 | [diff] [blame] | 78 | * This function shall be used whenever the isolated pageset has been |
| 79 | * built from lru, balloon, hugetlbfs page. See isolate_migratepages_range() |
| 80 | * and isolate_huge_page(). |
Rafael Aquini | 5733c7d | 2012-12-11 16:02:47 -0800 | [diff] [blame] | 81 | */ |
| 82 | void putback_movable_pages(struct list_head *l) |
| 83 | { |
| 84 | struct page *page; |
| 85 | struct page *page2; |
| 86 | |
| 87 | list_for_each_entry_safe(page, page2, l, lru) { |
Naoya Horiguchi | 31caf66 | 2013-09-11 14:21:59 -0700 | [diff] [blame] | 88 | if (unlikely(PageHuge(page))) { |
| 89 | putback_active_hugepage(page); |
| 90 | continue; |
| 91 | } |
Rafael Aquini | 5733c7d | 2012-12-11 16:02:47 -0800 | [diff] [blame] | 92 | list_del(&page->lru); |
| 93 | dec_zone_page_state(page, NR_ISOLATED_ANON + |
| 94 | page_is_file_cache(page)); |
Rafael Aquini | 117aad1 | 2013-09-30 13:45:16 -0700 | [diff] [blame] | 95 | if (unlikely(isolated_balloon_page(page))) |
Rafael Aquini | bf6bddf | 2012-12-11 16:02:42 -0800 | [diff] [blame] | 96 | balloon_page_putback(page); |
| 97 | else |
| 98 | putback_lru_page(page); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 99 | } |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 100 | } |
| 101 | |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 102 | /* |
| 103 | * Restore a potential migration pte to a working pte entry |
| 104 | */ |
Hugh Dickins | e9995ef | 2009-12-14 17:59:31 -0800 | [diff] [blame] | 105 | static int remove_migration_pte(struct page *new, struct vm_area_struct *vma, |
| 106 | unsigned long addr, void *old) |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 107 | { |
| 108 | struct mm_struct *mm = vma->vm_mm; |
| 109 | swp_entry_t entry; |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 110 | pmd_t *pmd; |
| 111 | pte_t *ptep, pte; |
| 112 | spinlock_t *ptl; |
| 113 | |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 114 | if (unlikely(PageHuge(new))) { |
| 115 | ptep = huge_pte_offset(mm, addr); |
| 116 | if (!ptep) |
| 117 | goto out; |
Kirill A. Shutemov | cb900f4 | 2013-11-14 14:31:02 -0800 | [diff] [blame] | 118 | ptl = huge_pte_lockptr(hstate_vma(vma), mm, ptep); |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 119 | } else { |
Bob Liu | 6219049 | 2012-12-11 16:00:37 -0800 | [diff] [blame] | 120 | pmd = mm_find_pmd(mm, addr); |
| 121 | if (!pmd) |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 122 | goto out; |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 123 | |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 124 | ptep = pte_offset_map(pmd, addr); |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 125 | |
Hugh Dickins | 486cf46 | 2011-10-19 12:50:35 -0700 | [diff] [blame] | 126 | /* |
| 127 | * Peek to check is_swap_pte() before taking ptlock? No, we |
| 128 | * can race mremap's move_ptes(), which skips anon_vma lock. |
| 129 | */ |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 130 | |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 131 | ptl = pte_lockptr(mm, pmd); |
| 132 | } |
| 133 | |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 134 | spin_lock(ptl); |
| 135 | pte = *ptep; |
| 136 | if (!is_swap_pte(pte)) |
Hugh Dickins | e9995ef | 2009-12-14 17:59:31 -0800 | [diff] [blame] | 137 | goto unlock; |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 138 | |
| 139 | entry = pte_to_swp_entry(pte); |
| 140 | |
Hugh Dickins | e9995ef | 2009-12-14 17:59:31 -0800 | [diff] [blame] | 141 | if (!is_migration_entry(entry) || |
| 142 | migration_entry_to_page(entry) != old) |
| 143 | goto unlock; |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 144 | |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 145 | get_page(new); |
| 146 | pte = pte_mkold(mk_pte(new, vma->vm_page_prot)); |
Cyrill Gorcunov | c3d16e1 | 2013-10-16 13:46:51 -0700 | [diff] [blame] | 147 | if (pte_swp_soft_dirty(*ptep)) |
| 148 | pte = pte_mksoft_dirty(pte); |
Mel Gorman | d3cb8bf | 2014-10-02 19:47:41 +0100 | [diff] [blame] | 149 | |
| 150 | /* Recheck VMA as permissions can change since migration started */ |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 151 | if (is_write_migration_entry(entry)) |
Mel Gorman | d3cb8bf | 2014-10-02 19:47:41 +0100 | [diff] [blame] | 152 | pte = maybe_mkwrite(pte, vma); |
| 153 | |
Andi Kleen | 3ef8fd7 | 2010-10-11 16:03:21 +0200 | [diff] [blame] | 154 | #ifdef CONFIG_HUGETLB_PAGE |
Tony Lu | be7517d | 2013-02-04 14:28:46 -0800 | [diff] [blame] | 155 | if (PageHuge(new)) { |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 156 | pte = pte_mkhuge(pte); |
Tony Lu | be7517d | 2013-02-04 14:28:46 -0800 | [diff] [blame] | 157 | pte = arch_make_huge_pte(pte, vma, new, 0); |
| 158 | } |
Andi Kleen | 3ef8fd7 | 2010-10-11 16:03:21 +0200 | [diff] [blame] | 159 | #endif |
Leonid Yegoshin | c2cc499 | 2013-05-24 15:55:18 -0700 | [diff] [blame] | 160 | flush_dcache_page(new); |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 161 | set_pte_at(mm, addr, ptep, pte); |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 162 | |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 163 | if (PageHuge(new)) { |
| 164 | if (PageAnon(new)) |
| 165 | hugepage_add_anon_rmap(new, vma, addr); |
| 166 | else |
| 167 | page_dup_rmap(new); |
| 168 | } else if (PageAnon(new)) |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 169 | page_add_anon_rmap(new, vma, addr); |
| 170 | else |
| 171 | page_add_file_rmap(new); |
| 172 | |
| 173 | /* No need to invalidate - it was non-present before */ |
Russell King | 4b3073e | 2009-12-18 16:40:18 +0000 | [diff] [blame] | 174 | update_mmu_cache(vma, addr, ptep); |
Hugh Dickins | e9995ef | 2009-12-14 17:59:31 -0800 | [diff] [blame] | 175 | unlock: |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 176 | pte_unmap_unlock(ptep, ptl); |
Hugh Dickins | e9995ef | 2009-12-14 17:59:31 -0800 | [diff] [blame] | 177 | out: |
| 178 | return SWAP_AGAIN; |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 179 | } |
| 180 | |
| 181 | /* |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 182 | * Get rid of all migration entries and replace them by |
| 183 | * references to the indicated page. |
| 184 | */ |
| 185 | static void remove_migration_ptes(struct page *old, struct page *new) |
| 186 | { |
Joonsoo Kim | 051ac83 | 2014-01-21 15:49:48 -0800 | [diff] [blame] | 187 | struct rmap_walk_control rwc = { |
| 188 | .rmap_one = remove_migration_pte, |
| 189 | .arg = old, |
| 190 | }; |
| 191 | |
| 192 | rmap_walk(new, &rwc); |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 193 | } |
| 194 | |
| 195 | /* |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 196 | * Something used the pte of a page under migration. We need to |
| 197 | * get to the page and wait until migration is finished. |
| 198 | * When we return from this function the fault will be retried. |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 199 | */ |
Naoya Horiguchi | 30dad30 | 2013-06-12 14:05:04 -0700 | [diff] [blame] | 200 | static void __migration_entry_wait(struct mm_struct *mm, pte_t *ptep, |
| 201 | spinlock_t *ptl) |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 202 | { |
Naoya Horiguchi | 30dad30 | 2013-06-12 14:05:04 -0700 | [diff] [blame] | 203 | pte_t pte; |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 204 | swp_entry_t entry; |
| 205 | struct page *page; |
| 206 | |
Naoya Horiguchi | 30dad30 | 2013-06-12 14:05:04 -0700 | [diff] [blame] | 207 | spin_lock(ptl); |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 208 | pte = *ptep; |
| 209 | if (!is_swap_pte(pte)) |
| 210 | goto out; |
| 211 | |
| 212 | entry = pte_to_swp_entry(pte); |
| 213 | if (!is_migration_entry(entry)) |
| 214 | goto out; |
| 215 | |
| 216 | page = migration_entry_to_page(entry); |
| 217 | |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 218 | /* |
| 219 | * Once radix-tree replacement of page migration started, page_count |
| 220 | * *must* be zero. And, we don't want to call wait_on_page_locked() |
| 221 | * against a page without get_page(). |
| 222 | * So, we use get_page_unless_zero(), here. Even failed, page fault |
| 223 | * will occur again. |
| 224 | */ |
| 225 | if (!get_page_unless_zero(page)) |
| 226 | goto out; |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 227 | pte_unmap_unlock(ptep, ptl); |
| 228 | wait_on_page_locked(page); |
| 229 | put_page(page); |
| 230 | return; |
| 231 | out: |
| 232 | pte_unmap_unlock(ptep, ptl); |
| 233 | } |
| 234 | |
Naoya Horiguchi | 30dad30 | 2013-06-12 14:05:04 -0700 | [diff] [blame] | 235 | void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd, |
| 236 | unsigned long address) |
| 237 | { |
| 238 | spinlock_t *ptl = pte_lockptr(mm, pmd); |
| 239 | pte_t *ptep = pte_offset_map(pmd, address); |
| 240 | __migration_entry_wait(mm, ptep, ptl); |
| 241 | } |
| 242 | |
Kirill A. Shutemov | cb900f4 | 2013-11-14 14:31:02 -0800 | [diff] [blame] | 243 | void migration_entry_wait_huge(struct vm_area_struct *vma, |
| 244 | struct mm_struct *mm, pte_t *pte) |
Naoya Horiguchi | 30dad30 | 2013-06-12 14:05:04 -0700 | [diff] [blame] | 245 | { |
Kirill A. Shutemov | cb900f4 | 2013-11-14 14:31:02 -0800 | [diff] [blame] | 246 | spinlock_t *ptl = huge_pte_lockptr(hstate_vma(vma), mm, pte); |
Naoya Horiguchi | 30dad30 | 2013-06-12 14:05:04 -0700 | [diff] [blame] | 247 | __migration_entry_wait(mm, pte, ptl); |
| 248 | } |
| 249 | |
Mel Gorman | b969c4a | 2012-01-12 17:19:34 -0800 | [diff] [blame] | 250 | #ifdef CONFIG_BLOCK |
| 251 | /* Returns true if all buffers are successfully locked */ |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 252 | static bool buffer_migrate_lock_buffers(struct buffer_head *head, |
| 253 | enum migrate_mode mode) |
Mel Gorman | b969c4a | 2012-01-12 17:19:34 -0800 | [diff] [blame] | 254 | { |
| 255 | struct buffer_head *bh = head; |
| 256 | |
| 257 | /* Simple case, sync compaction */ |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 258 | if (mode != MIGRATE_ASYNC) { |
Mel Gorman | b969c4a | 2012-01-12 17:19:34 -0800 | [diff] [blame] | 259 | do { |
| 260 | get_bh(bh); |
| 261 | lock_buffer(bh); |
| 262 | bh = bh->b_this_page; |
| 263 | |
| 264 | } while (bh != head); |
| 265 | |
| 266 | return true; |
| 267 | } |
| 268 | |
| 269 | /* async case, we cannot block on lock_buffer so use trylock_buffer */ |
| 270 | do { |
| 271 | get_bh(bh); |
| 272 | if (!trylock_buffer(bh)) { |
| 273 | /* |
| 274 | * We failed to lock the buffer and cannot stall in |
| 275 | * async migration. Release the taken locks |
| 276 | */ |
| 277 | struct buffer_head *failed_bh = bh; |
| 278 | put_bh(failed_bh); |
| 279 | bh = head; |
| 280 | while (bh != failed_bh) { |
| 281 | unlock_buffer(bh); |
| 282 | put_bh(bh); |
| 283 | bh = bh->b_this_page; |
| 284 | } |
| 285 | return false; |
| 286 | } |
| 287 | |
| 288 | bh = bh->b_this_page; |
| 289 | } while (bh != head); |
| 290 | return true; |
| 291 | } |
| 292 | #else |
| 293 | static inline bool buffer_migrate_lock_buffers(struct buffer_head *head, |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 294 | enum migrate_mode mode) |
Mel Gorman | b969c4a | 2012-01-12 17:19:34 -0800 | [diff] [blame] | 295 | { |
| 296 | return true; |
| 297 | } |
| 298 | #endif /* CONFIG_BLOCK */ |
| 299 | |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 300 | /* |
Christoph Lameter | c3fcf8a | 2006-06-23 02:03:32 -0700 | [diff] [blame] | 301 | * Replace the page in the mapping. |
Christoph Lameter | 5b5c712 | 2006-06-23 02:03:29 -0700 | [diff] [blame] | 302 | * |
| 303 | * The number of remaining references must be: |
| 304 | * 1 for anonymous pages without a mapping |
| 305 | * 2 for pages with a mapping |
David Howells | 266cf65 | 2009-04-03 16:42:36 +0100 | [diff] [blame] | 306 | * 3 for pages with a mapping and PagePrivate/PagePrivate2 set. |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 307 | */ |
Gu Zheng | 36bc08c | 2013-07-16 17:56:16 +0800 | [diff] [blame] | 308 | int migrate_page_move_mapping(struct address_space *mapping, |
Mel Gorman | b969c4a | 2012-01-12 17:19:34 -0800 | [diff] [blame] | 309 | struct page *newpage, struct page *page, |
Benjamin LaHaise | 8e321fe | 2013-12-21 17:56:08 -0500 | [diff] [blame] | 310 | struct buffer_head *head, enum migrate_mode mode, |
| 311 | int extra_count) |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 312 | { |
Benjamin LaHaise | 8e321fe | 2013-12-21 17:56:08 -0500 | [diff] [blame] | 313 | int expected_count = 1 + extra_count; |
Nick Piggin | 7cf9c2c | 2006-12-06 20:33:44 -0800 | [diff] [blame] | 314 | void **pslot; |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 315 | |
Christoph Lameter | 6c5240a | 2006-06-23 02:03:37 -0700 | [diff] [blame] | 316 | if (!mapping) { |
Christoph Lameter | 0e8c7d0 | 2007-04-23 14:41:09 -0700 | [diff] [blame] | 317 | /* Anonymous page without mapping */ |
Benjamin LaHaise | 8e321fe | 2013-12-21 17:56:08 -0500 | [diff] [blame] | 318 | if (page_count(page) != expected_count) |
Christoph Lameter | 6c5240a | 2006-06-23 02:03:37 -0700 | [diff] [blame] | 319 | return -EAGAIN; |
Rafael Aquini | 78bd520 | 2012-12-11 16:02:31 -0800 | [diff] [blame] | 320 | return MIGRATEPAGE_SUCCESS; |
Christoph Lameter | 6c5240a | 2006-06-23 02:03:37 -0700 | [diff] [blame] | 321 | } |
| 322 | |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 323 | spin_lock_irq(&mapping->tree_lock); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 324 | |
Nick Piggin | 7cf9c2c | 2006-12-06 20:33:44 -0800 | [diff] [blame] | 325 | pslot = radix_tree_lookup_slot(&mapping->page_tree, |
| 326 | page_index(page)); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 327 | |
Benjamin LaHaise | 8e321fe | 2013-12-21 17:56:08 -0500 | [diff] [blame] | 328 | expected_count += 1 + page_has_private(page); |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 329 | if (page_count(page) != expected_count || |
Mel Gorman | 29c1f67 | 2011-01-13 15:47:21 -0800 | [diff] [blame] | 330 | radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) { |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 331 | spin_unlock_irq(&mapping->tree_lock); |
Christoph Lameter | e23ca00 | 2006-04-10 22:52:57 -0700 | [diff] [blame] | 332 | return -EAGAIN; |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 333 | } |
| 334 | |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 335 | if (!page_freeze_refs(page, expected_count)) { |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 336 | spin_unlock_irq(&mapping->tree_lock); |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 337 | return -EAGAIN; |
| 338 | } |
| 339 | |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 340 | /* |
Mel Gorman | b969c4a | 2012-01-12 17:19:34 -0800 | [diff] [blame] | 341 | * In the async migration case of moving a page with buffers, lock the |
| 342 | * buffers using trylock before the mapping is moved. If the mapping |
| 343 | * was moved, we later failed to lock the buffers and could not move |
| 344 | * the mapping back due to an elevated page count, we would have to |
| 345 | * block waiting on other references to be dropped. |
| 346 | */ |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 347 | if (mode == MIGRATE_ASYNC && head && |
| 348 | !buffer_migrate_lock_buffers(head, mode)) { |
Mel Gorman | b969c4a | 2012-01-12 17:19:34 -0800 | [diff] [blame] | 349 | page_unfreeze_refs(page, expected_count); |
| 350 | spin_unlock_irq(&mapping->tree_lock); |
| 351 | return -EAGAIN; |
| 352 | } |
| 353 | |
| 354 | /* |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 355 | * Now we know that no one else is looking at the page. |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 356 | */ |
Nick Piggin | 7cf9c2c | 2006-12-06 20:33:44 -0800 | [diff] [blame] | 357 | get_page(newpage); /* add cache reference */ |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 358 | if (PageSwapCache(page)) { |
| 359 | SetPageSwapCache(newpage); |
| 360 | set_page_private(newpage, page_private(page)); |
| 361 | } |
| 362 | |
Nick Piggin | 7cf9c2c | 2006-12-06 20:33:44 -0800 | [diff] [blame] | 363 | radix_tree_replace_slot(pslot, newpage); |
| 364 | |
| 365 | /* |
Jacobo Giralt | 937a94c | 2012-01-10 15:07:11 -0800 | [diff] [blame] | 366 | * Drop cache reference from old page by unfreezing |
| 367 | * to one less reference. |
Nick Piggin | 7cf9c2c | 2006-12-06 20:33:44 -0800 | [diff] [blame] | 368 | * We know this isn't the last reference. |
| 369 | */ |
Jacobo Giralt | 937a94c | 2012-01-10 15:07:11 -0800 | [diff] [blame] | 370 | page_unfreeze_refs(page, expected_count - 1); |
Nick Piggin | 7cf9c2c | 2006-12-06 20:33:44 -0800 | [diff] [blame] | 371 | |
Christoph Lameter | 0e8c7d0 | 2007-04-23 14:41:09 -0700 | [diff] [blame] | 372 | /* |
| 373 | * If moved to a different zone then also account |
| 374 | * the page for that zone. Other VM counters will be |
| 375 | * taken care of when we establish references to the |
| 376 | * new page and drop references to the old page. |
| 377 | * |
| 378 | * Note that anonymous pages are accounted for |
| 379 | * via NR_FILE_PAGES and NR_ANON_PAGES if they |
| 380 | * are mapped to swap space. |
| 381 | */ |
| 382 | __dec_zone_page_state(page, NR_FILE_PAGES); |
| 383 | __inc_zone_page_state(newpage, NR_FILE_PAGES); |
Andrea Arcangeli | 99a15e2 | 2011-06-16 12:56:19 -0700 | [diff] [blame] | 384 | if (!PageSwapCache(page) && PageSwapBacked(page)) { |
KOSAKI Motohiro | 4b02108 | 2009-09-21 17:01:33 -0700 | [diff] [blame] | 385 | __dec_zone_page_state(page, NR_SHMEM); |
| 386 | __inc_zone_page_state(newpage, NR_SHMEM); |
| 387 | } |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 388 | spin_unlock_irq(&mapping->tree_lock); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 389 | |
Rafael Aquini | 78bd520 | 2012-12-11 16:02:31 -0800 | [diff] [blame] | 390 | return MIGRATEPAGE_SUCCESS; |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 391 | } |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 392 | |
| 393 | /* |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 394 | * The expected number of remaining references is the same as that |
| 395 | * of migrate_page_move_mapping(). |
| 396 | */ |
| 397 | int migrate_huge_page_move_mapping(struct address_space *mapping, |
| 398 | struct page *newpage, struct page *page) |
| 399 | { |
| 400 | int expected_count; |
| 401 | void **pslot; |
| 402 | |
| 403 | if (!mapping) { |
| 404 | if (page_count(page) != 1) |
| 405 | return -EAGAIN; |
Rafael Aquini | 78bd520 | 2012-12-11 16:02:31 -0800 | [diff] [blame] | 406 | return MIGRATEPAGE_SUCCESS; |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 407 | } |
| 408 | |
| 409 | spin_lock_irq(&mapping->tree_lock); |
| 410 | |
| 411 | pslot = radix_tree_lookup_slot(&mapping->page_tree, |
| 412 | page_index(page)); |
| 413 | |
| 414 | expected_count = 2 + page_has_private(page); |
| 415 | if (page_count(page) != expected_count || |
Mel Gorman | 29c1f67 | 2011-01-13 15:47:21 -0800 | [diff] [blame] | 416 | radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) { |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 417 | spin_unlock_irq(&mapping->tree_lock); |
| 418 | return -EAGAIN; |
| 419 | } |
| 420 | |
| 421 | if (!page_freeze_refs(page, expected_count)) { |
| 422 | spin_unlock_irq(&mapping->tree_lock); |
| 423 | return -EAGAIN; |
| 424 | } |
| 425 | |
| 426 | get_page(newpage); |
| 427 | |
| 428 | radix_tree_replace_slot(pslot, newpage); |
| 429 | |
Jacobo Giralt | 937a94c | 2012-01-10 15:07:11 -0800 | [diff] [blame] | 430 | page_unfreeze_refs(page, expected_count - 1); |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 431 | |
| 432 | spin_unlock_irq(&mapping->tree_lock); |
Rafael Aquini | 78bd520 | 2012-12-11 16:02:31 -0800 | [diff] [blame] | 433 | return MIGRATEPAGE_SUCCESS; |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 434 | } |
| 435 | |
| 436 | /* |
Dave Hansen | 30b0a10 | 2013-11-21 14:31:58 -0800 | [diff] [blame] | 437 | * Gigantic pages are so large that we do not guarantee that page++ pointer |
| 438 | * arithmetic will work across the entire page. We need something more |
| 439 | * specialized. |
| 440 | */ |
| 441 | static void __copy_gigantic_page(struct page *dst, struct page *src, |
| 442 | int nr_pages) |
| 443 | { |
| 444 | int i; |
| 445 | struct page *dst_base = dst; |
| 446 | struct page *src_base = src; |
| 447 | |
| 448 | for (i = 0; i < nr_pages; ) { |
| 449 | cond_resched(); |
| 450 | copy_highpage(dst, src); |
| 451 | |
| 452 | i++; |
| 453 | dst = mem_map_next(dst, dst_base, i); |
| 454 | src = mem_map_next(src, src_base, i); |
| 455 | } |
| 456 | } |
| 457 | |
| 458 | static void copy_huge_page(struct page *dst, struct page *src) |
| 459 | { |
| 460 | int i; |
| 461 | int nr_pages; |
| 462 | |
| 463 | if (PageHuge(src)) { |
| 464 | /* hugetlbfs page */ |
| 465 | struct hstate *h = page_hstate(src); |
| 466 | nr_pages = pages_per_huge_page(h); |
| 467 | |
| 468 | if (unlikely(nr_pages > MAX_ORDER_NR_PAGES)) { |
| 469 | __copy_gigantic_page(dst, src, nr_pages); |
| 470 | return; |
| 471 | } |
| 472 | } else { |
| 473 | /* thp page */ |
| 474 | BUG_ON(!PageTransHuge(src)); |
| 475 | nr_pages = hpage_nr_pages(src); |
| 476 | } |
| 477 | |
| 478 | for (i = 0; i < nr_pages; i++) { |
| 479 | cond_resched(); |
| 480 | copy_highpage(dst + i, src + i); |
| 481 | } |
| 482 | } |
| 483 | |
| 484 | /* |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 485 | * Copy the page to its new location |
| 486 | */ |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 487 | void migrate_page_copy(struct page *newpage, struct page *page) |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 488 | { |
Rik van Riel | 7851a45 | 2013-10-07 11:29:23 +0100 | [diff] [blame] | 489 | int cpupid; |
| 490 | |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 491 | if (PageHuge(page) || PageTransHuge(page)) |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 492 | copy_huge_page(newpage, page); |
| 493 | else |
| 494 | copy_highpage(newpage, page); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 495 | |
| 496 | if (PageError(page)) |
| 497 | SetPageError(newpage); |
| 498 | if (PageReferenced(page)) |
| 499 | SetPageReferenced(newpage); |
| 500 | if (PageUptodate(page)) |
| 501 | SetPageUptodate(newpage); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 502 | if (TestClearPageActive(page)) { |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 503 | VM_BUG_ON_PAGE(PageUnevictable(page), page); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 504 | SetPageActive(newpage); |
Lee Schermerhorn | 418b27e | 2009-12-14 17:59:54 -0800 | [diff] [blame] | 505 | } else if (TestClearPageUnevictable(page)) |
| 506 | SetPageUnevictable(newpage); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 507 | if (PageChecked(page)) |
| 508 | SetPageChecked(newpage); |
| 509 | if (PageMappedToDisk(page)) |
| 510 | SetPageMappedToDisk(newpage); |
| 511 | |
| 512 | if (PageDirty(page)) { |
| 513 | clear_page_dirty_for_io(page); |
Nick Piggin | 3a902c5 | 2008-04-30 00:55:16 -0700 | [diff] [blame] | 514 | /* |
| 515 | * Want to mark the page and the radix tree as dirty, and |
| 516 | * redo the accounting that clear_page_dirty_for_io undid, |
| 517 | * but we can't use set_page_dirty because that function |
| 518 | * is actually a signal that all of the page has become dirty. |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 519 | * Whereas only part of our page may be dirty. |
Nick Piggin | 3a902c5 | 2008-04-30 00:55:16 -0700 | [diff] [blame] | 520 | */ |
Hugh Dickins | 752dc18 | 2012-06-02 00:27:47 -0700 | [diff] [blame] | 521 | if (PageSwapBacked(page)) |
| 522 | SetPageDirty(newpage); |
| 523 | else |
| 524 | __set_page_dirty_nobuffers(newpage); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 525 | } |
| 526 | |
Rik van Riel | 7851a45 | 2013-10-07 11:29:23 +0100 | [diff] [blame] | 527 | /* |
| 528 | * Copy NUMA information to the new page, to prevent over-eager |
| 529 | * future migrations of this same page. |
| 530 | */ |
| 531 | cpupid = page_cpupid_xchg_last(page, -1); |
| 532 | page_cpupid_xchg_last(newpage, cpupid); |
| 533 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 534 | mlock_migrate_page(newpage, page); |
Hugh Dickins | e9995ef | 2009-12-14 17:59:31 -0800 | [diff] [blame] | 535 | ksm_migrate_page(newpage, page); |
Hugh Dickins | c8d6553 | 2013-02-22 16:35:10 -0800 | [diff] [blame] | 536 | /* |
| 537 | * Please do not reorder this without considering how mm/ksm.c's |
| 538 | * get_ksm_page() depends upon ksm_migrate_page() and PageSwapCache(). |
| 539 | */ |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 540 | ClearPageSwapCache(page); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 541 | ClearPagePrivate(page); |
| 542 | set_page_private(page, 0); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 543 | |
| 544 | /* |
| 545 | * If any waiters have accumulated on the new page then |
| 546 | * wake them up. |
| 547 | */ |
| 548 | if (PageWriteback(newpage)) |
| 549 | end_page_writeback(newpage); |
| 550 | } |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 551 | |
Christoph Lameter | 1d8b85c | 2006-06-23 02:03:28 -0700 | [diff] [blame] | 552 | /************************************************************ |
| 553 | * Migration functions |
| 554 | ***********************************************************/ |
| 555 | |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 556 | /* |
| 557 | * Common logic to directly migrate a single page suitable for |
David Howells | 266cf65 | 2009-04-03 16:42:36 +0100 | [diff] [blame] | 558 | * pages that do not use PagePrivate/PagePrivate2. |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 559 | * |
| 560 | * Pages are locked upon entry and exit. |
| 561 | */ |
Christoph Lameter | 2d1db3b | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 562 | int migrate_page(struct address_space *mapping, |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 563 | struct page *newpage, struct page *page, |
| 564 | enum migrate_mode mode) |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 565 | { |
| 566 | int rc; |
| 567 | |
| 568 | BUG_ON(PageWriteback(page)); /* Writeback must be complete */ |
| 569 | |
Benjamin LaHaise | 8e321fe | 2013-12-21 17:56:08 -0500 | [diff] [blame] | 570 | rc = migrate_page_move_mapping(mapping, newpage, page, NULL, mode, 0); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 571 | |
Rafael Aquini | 78bd520 | 2012-12-11 16:02:31 -0800 | [diff] [blame] | 572 | if (rc != MIGRATEPAGE_SUCCESS) |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 573 | return rc; |
| 574 | |
| 575 | migrate_page_copy(newpage, page); |
Rafael Aquini | 78bd520 | 2012-12-11 16:02:31 -0800 | [diff] [blame] | 576 | return MIGRATEPAGE_SUCCESS; |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 577 | } |
| 578 | EXPORT_SYMBOL(migrate_page); |
| 579 | |
David Howells | 9361401 | 2006-09-30 20:45:40 +0200 | [diff] [blame] | 580 | #ifdef CONFIG_BLOCK |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 581 | /* |
Christoph Lameter | 1d8b85c | 2006-06-23 02:03:28 -0700 | [diff] [blame] | 582 | * Migration function for pages with buffers. This function can only be used |
| 583 | * if the underlying filesystem guarantees that no other references to "page" |
| 584 | * exist. |
| 585 | */ |
Christoph Lameter | 2d1db3b | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 586 | int buffer_migrate_page(struct address_space *mapping, |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 587 | struct page *newpage, struct page *page, enum migrate_mode mode) |
Christoph Lameter | 1d8b85c | 2006-06-23 02:03:28 -0700 | [diff] [blame] | 588 | { |
Christoph Lameter | 1d8b85c | 2006-06-23 02:03:28 -0700 | [diff] [blame] | 589 | struct buffer_head *bh, *head; |
| 590 | int rc; |
| 591 | |
Christoph Lameter | 1d8b85c | 2006-06-23 02:03:28 -0700 | [diff] [blame] | 592 | if (!page_has_buffers(page)) |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 593 | return migrate_page(mapping, newpage, page, mode); |
Christoph Lameter | 1d8b85c | 2006-06-23 02:03:28 -0700 | [diff] [blame] | 594 | |
| 595 | head = page_buffers(page); |
| 596 | |
Benjamin LaHaise | 8e321fe | 2013-12-21 17:56:08 -0500 | [diff] [blame] | 597 | rc = migrate_page_move_mapping(mapping, newpage, page, head, mode, 0); |
Christoph Lameter | 1d8b85c | 2006-06-23 02:03:28 -0700 | [diff] [blame] | 598 | |
Rafael Aquini | 78bd520 | 2012-12-11 16:02:31 -0800 | [diff] [blame] | 599 | if (rc != MIGRATEPAGE_SUCCESS) |
Christoph Lameter | 1d8b85c | 2006-06-23 02:03:28 -0700 | [diff] [blame] | 600 | return rc; |
| 601 | |
Mel Gorman | b969c4a | 2012-01-12 17:19:34 -0800 | [diff] [blame] | 602 | /* |
| 603 | * In the async case, migrate_page_move_mapping locked the buffers |
| 604 | * with an IRQ-safe spinlock held. In the sync case, the buffers |
| 605 | * need to be locked now |
| 606 | */ |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 607 | if (mode != MIGRATE_ASYNC) |
| 608 | BUG_ON(!buffer_migrate_lock_buffers(head, mode)); |
Christoph Lameter | 1d8b85c | 2006-06-23 02:03:28 -0700 | [diff] [blame] | 609 | |
| 610 | ClearPagePrivate(page); |
| 611 | set_page_private(newpage, page_private(page)); |
| 612 | set_page_private(page, 0); |
| 613 | put_page(page); |
| 614 | get_page(newpage); |
| 615 | |
| 616 | bh = head; |
| 617 | do { |
| 618 | set_bh_page(bh, newpage, bh_offset(bh)); |
| 619 | bh = bh->b_this_page; |
| 620 | |
| 621 | } while (bh != head); |
| 622 | |
| 623 | SetPagePrivate(newpage); |
| 624 | |
| 625 | migrate_page_copy(newpage, page); |
| 626 | |
| 627 | bh = head; |
| 628 | do { |
| 629 | unlock_buffer(bh); |
| 630 | put_bh(bh); |
| 631 | bh = bh->b_this_page; |
| 632 | |
| 633 | } while (bh != head); |
| 634 | |
Rafael Aquini | 78bd520 | 2012-12-11 16:02:31 -0800 | [diff] [blame] | 635 | return MIGRATEPAGE_SUCCESS; |
Christoph Lameter | 1d8b85c | 2006-06-23 02:03:28 -0700 | [diff] [blame] | 636 | } |
| 637 | EXPORT_SYMBOL(buffer_migrate_page); |
David Howells | 9361401 | 2006-09-30 20:45:40 +0200 | [diff] [blame] | 638 | #endif |
Christoph Lameter | 1d8b85c | 2006-06-23 02:03:28 -0700 | [diff] [blame] | 639 | |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 640 | /* |
| 641 | * Writeback a page to clean the dirty state |
| 642 | */ |
| 643 | static int writeout(struct address_space *mapping, struct page *page) |
| 644 | { |
| 645 | struct writeback_control wbc = { |
| 646 | .sync_mode = WB_SYNC_NONE, |
| 647 | .nr_to_write = 1, |
| 648 | .range_start = 0, |
| 649 | .range_end = LLONG_MAX, |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 650 | .for_reclaim = 1 |
| 651 | }; |
| 652 | int rc; |
| 653 | |
| 654 | if (!mapping->a_ops->writepage) |
| 655 | /* No write method for the address space */ |
| 656 | return -EINVAL; |
| 657 | |
| 658 | if (!clear_page_dirty_for_io(page)) |
| 659 | /* Someone else already triggered a write */ |
| 660 | return -EAGAIN; |
| 661 | |
| 662 | /* |
| 663 | * A dirty page may imply that the underlying filesystem has |
| 664 | * the page on some queue. So the page must be clean for |
| 665 | * migration. Writeout may mean we loose the lock and the |
| 666 | * page state is no longer what we checked for earlier. |
| 667 | * At this point we know that the migration attempt cannot |
| 668 | * be successful. |
| 669 | */ |
| 670 | remove_migration_ptes(page, page); |
| 671 | |
| 672 | rc = mapping->a_ops->writepage(page, &wbc); |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 673 | |
| 674 | if (rc != AOP_WRITEPAGE_ACTIVATE) |
| 675 | /* unlocked. Relock */ |
| 676 | lock_page(page); |
| 677 | |
Hugh Dickins | bda8550 | 2008-11-19 15:36:36 -0800 | [diff] [blame] | 678 | return (rc < 0) ? -EIO : -EAGAIN; |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 679 | } |
| 680 | |
| 681 | /* |
| 682 | * Default handling if a filesystem does not provide a migration function. |
| 683 | */ |
Christoph Lameter | 8351a6e | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 684 | static int fallback_migrate_page(struct address_space *mapping, |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 685 | struct page *newpage, struct page *page, enum migrate_mode mode) |
Christoph Lameter | 8351a6e | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 686 | { |
Mel Gorman | b969c4a | 2012-01-12 17:19:34 -0800 | [diff] [blame] | 687 | if (PageDirty(page)) { |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 688 | /* Only writeback pages in full synchronous migration */ |
| 689 | if (mode != MIGRATE_SYNC) |
Mel Gorman | b969c4a | 2012-01-12 17:19:34 -0800 | [diff] [blame] | 690 | return -EBUSY; |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 691 | return writeout(mapping, page); |
Mel Gorman | b969c4a | 2012-01-12 17:19:34 -0800 | [diff] [blame] | 692 | } |
Christoph Lameter | 8351a6e | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 693 | |
| 694 | /* |
| 695 | * Buffers may be managed in a filesystem specific way. |
| 696 | * We must have no buffers or drop them. |
| 697 | */ |
David Howells | 266cf65 | 2009-04-03 16:42:36 +0100 | [diff] [blame] | 698 | if (page_has_private(page) && |
Christoph Lameter | 8351a6e | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 699 | !try_to_release_page(page, GFP_KERNEL)) |
| 700 | return -EAGAIN; |
| 701 | |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 702 | return migrate_page(mapping, newpage, page, mode); |
Christoph Lameter | 8351a6e | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 703 | } |
| 704 | |
Christoph Lameter | 1d8b85c | 2006-06-23 02:03:28 -0700 | [diff] [blame] | 705 | /* |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 706 | * Move a page to a newly allocated page |
| 707 | * The page is locked and all ptes have been successfully removed. |
| 708 | * |
| 709 | * The new page will have replaced the old page if this function |
| 710 | * is successful. |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 711 | * |
| 712 | * Return value: |
| 713 | * < 0 - error code |
Rafael Aquini | 78bd520 | 2012-12-11 16:02:31 -0800 | [diff] [blame] | 714 | * MIGRATEPAGE_SUCCESS - success |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 715 | */ |
Mel Gorman | 3fe2011 | 2010-05-24 14:32:20 -0700 | [diff] [blame] | 716 | static int move_to_new_page(struct page *newpage, struct page *page, |
Hugh Dickins | 2ebba6b | 2014-12-12 16:56:19 -0800 | [diff] [blame] | 717 | int page_was_mapped, enum migrate_mode mode) |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 718 | { |
| 719 | struct address_space *mapping; |
| 720 | int rc; |
| 721 | |
| 722 | /* |
| 723 | * Block others from accessing the page when we get around to |
| 724 | * establishing additional references. We are the only one |
| 725 | * holding a reference to the new page at this point. |
| 726 | */ |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 727 | if (!trylock_page(newpage)) |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 728 | BUG(); |
| 729 | |
| 730 | /* Prepare mapping for the new page.*/ |
| 731 | newpage->index = page->index; |
| 732 | newpage->mapping = page->mapping; |
Rik van Riel | b2e1853 | 2008-10-18 20:26:30 -0700 | [diff] [blame] | 733 | if (PageSwapBacked(page)) |
| 734 | SetPageSwapBacked(newpage); |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 735 | |
| 736 | mapping = page_mapping(page); |
| 737 | if (!mapping) |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 738 | rc = migrate_page(mapping, newpage, page, mode); |
Mel Gorman | b969c4a | 2012-01-12 17:19:34 -0800 | [diff] [blame] | 739 | else if (mapping->a_ops->migratepage) |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 740 | /* |
Mel Gorman | b969c4a | 2012-01-12 17:19:34 -0800 | [diff] [blame] | 741 | * Most pages have a mapping and most filesystems provide a |
| 742 | * migratepage callback. Anonymous pages are part of swap |
| 743 | * space which also has its own migratepage callback. This |
| 744 | * is the most common path for page migration. |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 745 | */ |
Mel Gorman | b969c4a | 2012-01-12 17:19:34 -0800 | [diff] [blame] | 746 | rc = mapping->a_ops->migratepage(mapping, |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 747 | newpage, page, mode); |
Mel Gorman | b969c4a | 2012-01-12 17:19:34 -0800 | [diff] [blame] | 748 | else |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 749 | rc = fallback_migrate_page(mapping, newpage, page, mode); |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 750 | |
Rafael Aquini | 78bd520 | 2012-12-11 16:02:31 -0800 | [diff] [blame] | 751 | if (rc != MIGRATEPAGE_SUCCESS) { |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 752 | newpage->mapping = NULL; |
Mel Gorman | 3fe2011 | 2010-05-24 14:32:20 -0700 | [diff] [blame] | 753 | } else { |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 754 | mem_cgroup_migrate(page, newpage, false); |
Hugh Dickins | 2ebba6b | 2014-12-12 16:56:19 -0800 | [diff] [blame] | 755 | if (page_was_mapped) |
Mel Gorman | 3fe2011 | 2010-05-24 14:32:20 -0700 | [diff] [blame] | 756 | remove_migration_ptes(page, newpage); |
Konstantin Khlebnikov | 35512ec | 2012-02-03 15:37:13 -0800 | [diff] [blame] | 757 | page->mapping = NULL; |
Mel Gorman | 3fe2011 | 2010-05-24 14:32:20 -0700 | [diff] [blame] | 758 | } |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 759 | |
| 760 | unlock_page(newpage); |
| 761 | |
| 762 | return rc; |
| 763 | } |
| 764 | |
Minchan Kim | 0dabec9 | 2011-10-31 17:06:57 -0700 | [diff] [blame] | 765 | static int __unmap_and_move(struct page *page, struct page *newpage, |
Hugh Dickins | 9c620e2 | 2013-02-22 16:35:14 -0800 | [diff] [blame] | 766 | int force, enum migrate_mode mode) |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 767 | { |
Minchan Kim | 0dabec9 | 2011-10-31 17:06:57 -0700 | [diff] [blame] | 768 | int rc = -EAGAIN; |
Hugh Dickins | 2ebba6b | 2014-12-12 16:56:19 -0800 | [diff] [blame] | 769 | int page_was_mapped = 0; |
Mel Gorman | 3f6c827 | 2010-05-24 14:32:17 -0700 | [diff] [blame] | 770 | struct anon_vma *anon_vma = NULL; |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 771 | |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 772 | if (!trylock_page(page)) { |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 773 | if (!force || mode == MIGRATE_ASYNC) |
Minchan Kim | 0dabec9 | 2011-10-31 17:06:57 -0700 | [diff] [blame] | 774 | goto out; |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 775 | |
| 776 | /* |
| 777 | * It's not safe for direct compaction to call lock_page. |
| 778 | * For example, during page readahead pages are added locked |
| 779 | * to the LRU. Later, when the IO completes the pages are |
| 780 | * marked uptodate and unlocked. However, the queueing |
| 781 | * could be merging multiple pages for one bio (e.g. |
| 782 | * mpage_readpages). If an allocation happens for the |
| 783 | * second or third page, the process can end up locking |
| 784 | * the same page twice and deadlocking. Rather than |
| 785 | * trying to be clever about what pages can be locked, |
| 786 | * avoid the use of lock_page for direct compaction |
| 787 | * altogether. |
| 788 | */ |
| 789 | if (current->flags & PF_MEMALLOC) |
Minchan Kim | 0dabec9 | 2011-10-31 17:06:57 -0700 | [diff] [blame] | 790 | goto out; |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 791 | |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 792 | lock_page(page); |
| 793 | } |
| 794 | |
| 795 | if (PageWriteback(page)) { |
Andrea Arcangeli | 11bc82d | 2011-03-22 16:33:11 -0700 | [diff] [blame] | 796 | /* |
Jianguo Wu | fed5b64 | 2013-04-29 15:07:58 -0700 | [diff] [blame] | 797 | * Only in the case of a full synchronous migration is it |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 798 | * necessary to wait for PageWriteback. In the async case, |
| 799 | * the retry loop is too short and in the sync-light case, |
| 800 | * the overhead of stalling is too much |
Andrea Arcangeli | 11bc82d | 2011-03-22 16:33:11 -0700 | [diff] [blame] | 801 | */ |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 802 | if (mode != MIGRATE_SYNC) { |
Andrea Arcangeli | 11bc82d | 2011-03-22 16:33:11 -0700 | [diff] [blame] | 803 | rc = -EBUSY; |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 804 | goto out_unlock; |
Andrea Arcangeli | 11bc82d | 2011-03-22 16:33:11 -0700 | [diff] [blame] | 805 | } |
| 806 | if (!force) |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 807 | goto out_unlock; |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 808 | wait_on_page_writeback(page); |
| 809 | } |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 810 | /* |
KAMEZAWA Hiroyuki | dc386d4 | 2007-07-26 10:41:07 -0700 | [diff] [blame] | 811 | * By try_to_unmap(), page->mapcount goes down to 0 here. In this case, |
| 812 | * we cannot notice that anon_vma is freed while we migrates a page. |
Hugh Dickins | 1ce82b6 | 2011-01-13 15:47:30 -0800 | [diff] [blame] | 813 | * This get_anon_vma() delays freeing anon_vma pointer until the end |
KAMEZAWA Hiroyuki | dc386d4 | 2007-07-26 10:41:07 -0700 | [diff] [blame] | 814 | * of migration. File cache pages are no problem because of page_lock() |
KAMEZAWA Hiroyuki | 989f89c | 2007-08-30 23:56:21 -0700 | [diff] [blame] | 815 | * File Caches may use write_page() or lock_page() in migration, then, |
| 816 | * just care Anon page here. |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 817 | */ |
Hugh Dickins | b79bc0a | 2013-02-22 16:35:13 -0800 | [diff] [blame] | 818 | if (PageAnon(page) && !PageKsm(page)) { |
Hugh Dickins | 1ce82b6 | 2011-01-13 15:47:30 -0800 | [diff] [blame] | 819 | /* |
Ingo Molnar | 4fc3f1d | 2012-12-02 19:56:50 +0000 | [diff] [blame] | 820 | * Only page_lock_anon_vma_read() understands the subtleties of |
Hugh Dickins | 1ce82b6 | 2011-01-13 15:47:30 -0800 | [diff] [blame] | 821 | * getting a hold on an anon_vma from outside one of its mms. |
| 822 | */ |
Peter Zijlstra | 746b18d | 2011-05-24 17:12:10 -0700 | [diff] [blame] | 823 | anon_vma = page_get_anon_vma(page); |
Hugh Dickins | 1ce82b6 | 2011-01-13 15:47:30 -0800 | [diff] [blame] | 824 | if (anon_vma) { |
| 825 | /* |
Peter Zijlstra | 746b18d | 2011-05-24 17:12:10 -0700 | [diff] [blame] | 826 | * Anon page |
Hugh Dickins | 1ce82b6 | 2011-01-13 15:47:30 -0800 | [diff] [blame] | 827 | */ |
Hugh Dickins | 1ce82b6 | 2011-01-13 15:47:30 -0800 | [diff] [blame] | 828 | } else if (PageSwapCache(page)) { |
Mel Gorman | 3fe2011 | 2010-05-24 14:32:20 -0700 | [diff] [blame] | 829 | /* |
| 830 | * We cannot be sure that the anon_vma of an unmapped |
| 831 | * swapcache page is safe to use because we don't |
| 832 | * know in advance if the VMA that this page belonged |
| 833 | * to still exists. If the VMA and others sharing the |
| 834 | * data have been freed, then the anon_vma could |
| 835 | * already be invalid. |
| 836 | * |
| 837 | * To avoid this possibility, swapcache pages get |
| 838 | * migrated but are not remapped when migration |
| 839 | * completes |
| 840 | */ |
Mel Gorman | 3fe2011 | 2010-05-24 14:32:20 -0700 | [diff] [blame] | 841 | } else { |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 842 | goto out_unlock; |
Mel Gorman | 3fe2011 | 2010-05-24 14:32:20 -0700 | [diff] [blame] | 843 | } |
KAMEZAWA Hiroyuki | 989f89c | 2007-08-30 23:56:21 -0700 | [diff] [blame] | 844 | } |
Shaohua Li | 62e1c55 | 2008-02-04 22:29:33 -0800 | [diff] [blame] | 845 | |
Konstantin Khlebnikov | d6d86c0 | 2014-10-09 15:29:27 -0700 | [diff] [blame] | 846 | if (unlikely(isolated_balloon_page(page))) { |
Rafael Aquini | bf6bddf | 2012-12-11 16:02:42 -0800 | [diff] [blame] | 847 | /* |
| 848 | * A ballooned page does not need any special attention from |
| 849 | * physical to virtual reverse mapping procedures. |
| 850 | * Skip any attempt to unmap PTEs or to remap swap cache, |
| 851 | * in order to avoid burning cycles at rmap level, and perform |
| 852 | * the page migration right away (proteced by page lock). |
| 853 | */ |
| 854 | rc = balloon_page_migrate(newpage, page, mode); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 855 | goto out_unlock; |
Rafael Aquini | bf6bddf | 2012-12-11 16:02:42 -0800 | [diff] [blame] | 856 | } |
| 857 | |
KAMEZAWA Hiroyuki | dc386d4 | 2007-07-26 10:41:07 -0700 | [diff] [blame] | 858 | /* |
Shaohua Li | 62e1c55 | 2008-02-04 22:29:33 -0800 | [diff] [blame] | 859 | * Corner case handling: |
| 860 | * 1. When a new swap-cache page is read into, it is added to the LRU |
| 861 | * and treated as swapcache but it has no rmap yet. |
| 862 | * Calling try_to_unmap() against a page->mapping==NULL page will |
| 863 | * trigger a BUG. So handle it here. |
| 864 | * 2. An orphaned page (see truncate_complete_page) might have |
| 865 | * fs-private metadata. The page can be picked up due to memory |
| 866 | * offlining. Everywhere else except page reclaim, the page is |
| 867 | * invisible to the vm, so the page can not be migrated. So try to |
| 868 | * free the metadata, so the page can be freed. |
KAMEZAWA Hiroyuki | dc386d4 | 2007-07-26 10:41:07 -0700 | [diff] [blame] | 869 | */ |
Shaohua Li | 62e1c55 | 2008-02-04 22:29:33 -0800 | [diff] [blame] | 870 | if (!page->mapping) { |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 871 | VM_BUG_ON_PAGE(PageAnon(page), page); |
Hugh Dickins | 1ce82b6 | 2011-01-13 15:47:30 -0800 | [diff] [blame] | 872 | if (page_has_private(page)) { |
Shaohua Li | 62e1c55 | 2008-02-04 22:29:33 -0800 | [diff] [blame] | 873 | try_to_free_buffers(page); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 874 | goto out_unlock; |
Shaohua Li | 62e1c55 | 2008-02-04 22:29:33 -0800 | [diff] [blame] | 875 | } |
Shaohua Li | abfc348 | 2009-09-21 17:01:19 -0700 | [diff] [blame] | 876 | goto skip_unmap; |
Shaohua Li | 62e1c55 | 2008-02-04 22:29:33 -0800 | [diff] [blame] | 877 | } |
| 878 | |
KAMEZAWA Hiroyuki | dc386d4 | 2007-07-26 10:41:07 -0700 | [diff] [blame] | 879 | /* Establish migration ptes or remove ptes */ |
Hugh Dickins | 2ebba6b | 2014-12-12 16:56:19 -0800 | [diff] [blame] | 880 | if (page_mapped(page)) { |
| 881 | try_to_unmap(page, |
| 882 | TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS); |
| 883 | page_was_mapped = 1; |
| 884 | } |
KAMEZAWA Hiroyuki | dc386d4 | 2007-07-26 10:41:07 -0700 | [diff] [blame] | 885 | |
Shaohua Li | abfc348 | 2009-09-21 17:01:19 -0700 | [diff] [blame] | 886 | skip_unmap: |
Christoph Lameter | e6a1530 | 2006-06-25 05:46:49 -0700 | [diff] [blame] | 887 | if (!page_mapped(page)) |
Hugh Dickins | 2ebba6b | 2014-12-12 16:56:19 -0800 | [diff] [blame] | 888 | rc = move_to_new_page(newpage, page, page_was_mapped, mode); |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 889 | |
Hugh Dickins | 2ebba6b | 2014-12-12 16:56:19 -0800 | [diff] [blame] | 890 | if (rc && page_was_mapped) |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 891 | remove_migration_ptes(page, page); |
Mel Gorman | 3f6c827 | 2010-05-24 14:32:17 -0700 | [diff] [blame] | 892 | |
| 893 | /* Drop an anon_vma reference if we took one */ |
Rik van Riel | 7654506 | 2010-08-09 17:18:41 -0700 | [diff] [blame] | 894 | if (anon_vma) |
Peter Zijlstra | 9e60109 | 2011-03-22 16:32:46 -0700 | [diff] [blame] | 895 | put_anon_vma(anon_vma); |
Mel Gorman | 3f6c827 | 2010-05-24 14:32:17 -0700 | [diff] [blame] | 896 | |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 897 | out_unlock: |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 898 | unlock_page(page); |
Minchan Kim | 0dabec9 | 2011-10-31 17:06:57 -0700 | [diff] [blame] | 899 | out: |
| 900 | return rc; |
| 901 | } |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 902 | |
Minchan Kim | 0dabec9 | 2011-10-31 17:06:57 -0700 | [diff] [blame] | 903 | /* |
| 904 | * Obtain the lock on page, remove all ptes and migrate the page |
| 905 | * to the newly allocated page in newpage. |
| 906 | */ |
David Rientjes | 68711a7 | 2014-06-04 16:08:25 -0700 | [diff] [blame] | 907 | static int unmap_and_move(new_page_t get_new_page, free_page_t put_new_page, |
| 908 | unsigned long private, struct page *page, int force, |
| 909 | enum migrate_mode mode) |
Minchan Kim | 0dabec9 | 2011-10-31 17:06:57 -0700 | [diff] [blame] | 910 | { |
| 911 | int rc = 0; |
| 912 | int *result = NULL; |
| 913 | struct page *newpage = get_new_page(page, private, &result); |
| 914 | |
| 915 | if (!newpage) |
| 916 | return -ENOMEM; |
| 917 | |
| 918 | if (page_count(page) == 1) { |
| 919 | /* page was freed from under us. So we are done. */ |
| 920 | goto out; |
| 921 | } |
| 922 | |
| 923 | if (unlikely(PageTransHuge(page))) |
| 924 | if (unlikely(split_huge_page(page))) |
| 925 | goto out; |
| 926 | |
Hugh Dickins | 9c620e2 | 2013-02-22 16:35:14 -0800 | [diff] [blame] | 927 | rc = __unmap_and_move(page, newpage, force, mode); |
Rafael Aquini | bf6bddf | 2012-12-11 16:02:42 -0800 | [diff] [blame] | 928 | |
Minchan Kim | 0dabec9 | 2011-10-31 17:06:57 -0700 | [diff] [blame] | 929 | out: |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 930 | if (rc != -EAGAIN) { |
Minchan Kim | 0dabec9 | 2011-10-31 17:06:57 -0700 | [diff] [blame] | 931 | /* |
| 932 | * A page that has been migrated has all references |
| 933 | * removed and will be freed. A page that has not been |
| 934 | * migrated will have kepts its references and be |
| 935 | * restored. |
| 936 | */ |
| 937 | list_del(&page->lru); |
KOSAKI Motohiro | a731286 | 2009-09-21 17:01:37 -0700 | [diff] [blame] | 938 | dec_zone_page_state(page, NR_ISOLATED_ANON + |
Johannes Weiner | 6c0b135 | 2009-09-21 17:02:59 -0700 | [diff] [blame] | 939 | page_is_file_cache(page)); |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 940 | putback_lru_page(page); |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 941 | } |
David Rientjes | 68711a7 | 2014-06-04 16:08:25 -0700 | [diff] [blame] | 942 | |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 943 | /* |
David Rientjes | 68711a7 | 2014-06-04 16:08:25 -0700 | [diff] [blame] | 944 | * If migration was not successful and there's a freeing callback, use |
| 945 | * it. Otherwise, putback_lru_page() will drop the reference grabbed |
| 946 | * during isolation. |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 947 | */ |
Hugh Dickins | 8bdd638 | 2014-07-26 12:58:23 -0700 | [diff] [blame] | 948 | if (rc != MIGRATEPAGE_SUCCESS && put_new_page) { |
| 949 | ClearPageSwapBacked(newpage); |
David Rientjes | 68711a7 | 2014-06-04 16:08:25 -0700 | [diff] [blame] | 950 | put_new_page(newpage, private); |
Konstantin Khlebnikov | d6d86c0 | 2014-10-09 15:29:27 -0700 | [diff] [blame] | 951 | } else if (unlikely(__is_movable_balloon_page(newpage))) { |
| 952 | /* drop our reference, page already in the balloon */ |
| 953 | put_page(newpage); |
Hugh Dickins | 8bdd638 | 2014-07-26 12:58:23 -0700 | [diff] [blame] | 954 | } else |
David Rientjes | 68711a7 | 2014-06-04 16:08:25 -0700 | [diff] [blame] | 955 | putback_lru_page(newpage); |
| 956 | |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 957 | if (result) { |
| 958 | if (rc) |
| 959 | *result = rc; |
| 960 | else |
| 961 | *result = page_to_nid(newpage); |
| 962 | } |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 963 | return rc; |
| 964 | } |
| 965 | |
| 966 | /* |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 967 | * Counterpart of unmap_and_move_page() for hugepage migration. |
| 968 | * |
| 969 | * This function doesn't wait the completion of hugepage I/O |
| 970 | * because there is no race between I/O and migration for hugepage. |
| 971 | * Note that currently hugepage I/O occurs only in direct I/O |
| 972 | * where no lock is held and PG_writeback is irrelevant, |
| 973 | * and writeback status of all subpages are counted in the reference |
| 974 | * count of the head page (i.e. if all subpages of a 2MB hugepage are |
| 975 | * under direct I/O, the reference of the head page is 512 and a bit more.) |
| 976 | * This means that when we try to migrate hugepage whose subpages are |
| 977 | * doing direct I/O, some references remain after try_to_unmap() and |
| 978 | * hugepage migration fails without data corruption. |
| 979 | * |
| 980 | * There is also no race when direct I/O is issued on the page under migration, |
| 981 | * because then pte is replaced with migration swap entry and direct I/O code |
| 982 | * will wait in the page fault for migration to complete. |
| 983 | */ |
| 984 | static int unmap_and_move_huge_page(new_page_t get_new_page, |
David Rientjes | 68711a7 | 2014-06-04 16:08:25 -0700 | [diff] [blame] | 985 | free_page_t put_new_page, unsigned long private, |
| 986 | struct page *hpage, int force, |
| 987 | enum migrate_mode mode) |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 988 | { |
| 989 | int rc = 0; |
| 990 | int *result = NULL; |
Hugh Dickins | 2ebba6b | 2014-12-12 16:56:19 -0800 | [diff] [blame] | 991 | int page_was_mapped = 0; |
Joonsoo Kim | 32665f2 | 2014-01-21 15:51:15 -0800 | [diff] [blame] | 992 | struct page *new_hpage; |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 993 | struct anon_vma *anon_vma = NULL; |
| 994 | |
Naoya Horiguchi | 83467ef | 2013-09-11 14:22:11 -0700 | [diff] [blame] | 995 | /* |
| 996 | * Movability of hugepages depends on architectures and hugepage size. |
| 997 | * This check is necessary because some callers of hugepage migration |
| 998 | * like soft offline and memory hotremove don't walk through page |
| 999 | * tables or check whether the hugepage is pmd-based or not before |
| 1000 | * kicking migration. |
| 1001 | */ |
Naoya Horiguchi | 100873d | 2014-06-04 16:10:56 -0700 | [diff] [blame] | 1002 | if (!hugepage_migration_supported(page_hstate(hpage))) { |
Joonsoo Kim | 32665f2 | 2014-01-21 15:51:15 -0800 | [diff] [blame] | 1003 | putback_active_hugepage(hpage); |
Naoya Horiguchi | 83467ef | 2013-09-11 14:22:11 -0700 | [diff] [blame] | 1004 | return -ENOSYS; |
Joonsoo Kim | 32665f2 | 2014-01-21 15:51:15 -0800 | [diff] [blame] | 1005 | } |
Naoya Horiguchi | 83467ef | 2013-09-11 14:22:11 -0700 | [diff] [blame] | 1006 | |
Joonsoo Kim | 32665f2 | 2014-01-21 15:51:15 -0800 | [diff] [blame] | 1007 | new_hpage = get_new_page(hpage, private, &result); |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 1008 | if (!new_hpage) |
| 1009 | return -ENOMEM; |
| 1010 | |
| 1011 | rc = -EAGAIN; |
| 1012 | |
| 1013 | if (!trylock_page(hpage)) { |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 1014 | if (!force || mode != MIGRATE_SYNC) |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 1015 | goto out; |
| 1016 | lock_page(hpage); |
| 1017 | } |
| 1018 | |
Peter Zijlstra | 746b18d | 2011-05-24 17:12:10 -0700 | [diff] [blame] | 1019 | if (PageAnon(hpage)) |
| 1020 | anon_vma = page_get_anon_vma(hpage); |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 1021 | |
Hugh Dickins | 2ebba6b | 2014-12-12 16:56:19 -0800 | [diff] [blame] | 1022 | if (page_mapped(hpage)) { |
| 1023 | try_to_unmap(hpage, |
| 1024 | TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS); |
| 1025 | page_was_mapped = 1; |
| 1026 | } |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 1027 | |
| 1028 | if (!page_mapped(hpage)) |
Hugh Dickins | 2ebba6b | 2014-12-12 16:56:19 -0800 | [diff] [blame] | 1029 | rc = move_to_new_page(new_hpage, hpage, page_was_mapped, mode); |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 1030 | |
Hugh Dickins | 2ebba6b | 2014-12-12 16:56:19 -0800 | [diff] [blame] | 1031 | if (rc != MIGRATEPAGE_SUCCESS && page_was_mapped) |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 1032 | remove_migration_ptes(hpage, hpage); |
| 1033 | |
Hugh Dickins | fd4a466 | 2011-01-13 15:47:31 -0800 | [diff] [blame] | 1034 | if (anon_vma) |
Peter Zijlstra | 9e60109 | 2011-03-22 16:32:46 -0700 | [diff] [blame] | 1035 | put_anon_vma(anon_vma); |
Aneesh Kumar K.V | 8e6ac7f | 2012-07-31 16:42:27 -0700 | [diff] [blame] | 1036 | |
David Rientjes | 68711a7 | 2014-06-04 16:08:25 -0700 | [diff] [blame] | 1037 | if (rc == MIGRATEPAGE_SUCCESS) |
Aneesh Kumar K.V | 8e6ac7f | 2012-07-31 16:42:27 -0700 | [diff] [blame] | 1038 | hugetlb_cgroup_migrate(hpage, new_hpage); |
| 1039 | |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 1040 | unlock_page(hpage); |
Hillf Danton | 0976133 | 2011-12-08 14:34:20 -0800 | [diff] [blame] | 1041 | out: |
Naoya Horiguchi | b8ec1ce | 2013-09-11 14:22:01 -0700 | [diff] [blame] | 1042 | if (rc != -EAGAIN) |
| 1043 | putback_active_hugepage(hpage); |
David Rientjes | 68711a7 | 2014-06-04 16:08:25 -0700 | [diff] [blame] | 1044 | |
| 1045 | /* |
| 1046 | * If migration was not successful and there's a freeing callback, use |
| 1047 | * it. Otherwise, put_page() will drop the reference grabbed during |
| 1048 | * isolation. |
| 1049 | */ |
| 1050 | if (rc != MIGRATEPAGE_SUCCESS && put_new_page) |
| 1051 | put_new_page(new_hpage, private); |
| 1052 | else |
| 1053 | put_page(new_hpage); |
| 1054 | |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 1055 | if (result) { |
| 1056 | if (rc) |
| 1057 | *result = rc; |
| 1058 | else |
| 1059 | *result = page_to_nid(new_hpage); |
| 1060 | } |
| 1061 | return rc; |
| 1062 | } |
| 1063 | |
| 1064 | /* |
Srivatsa S. Bhat | c73e5c9 | 2013-04-29 15:08:16 -0700 | [diff] [blame] | 1065 | * migrate_pages - migrate the pages specified in a list, to the free pages |
| 1066 | * supplied as the target for the page migration |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 1067 | * |
Srivatsa S. Bhat | c73e5c9 | 2013-04-29 15:08:16 -0700 | [diff] [blame] | 1068 | * @from: The list of pages to be migrated. |
| 1069 | * @get_new_page: The function used to allocate free pages to be used |
| 1070 | * as the target of the page migration. |
David Rientjes | 68711a7 | 2014-06-04 16:08:25 -0700 | [diff] [blame] | 1071 | * @put_new_page: The function used to free target pages if migration |
| 1072 | * fails, or NULL if no special handling is necessary. |
Srivatsa S. Bhat | c73e5c9 | 2013-04-29 15:08:16 -0700 | [diff] [blame] | 1073 | * @private: Private data to be passed on to get_new_page() |
| 1074 | * @mode: The migration mode that specifies the constraints for |
| 1075 | * page migration, if any. |
| 1076 | * @reason: The reason for page migration. |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 1077 | * |
Srivatsa S. Bhat | c73e5c9 | 2013-04-29 15:08:16 -0700 | [diff] [blame] | 1078 | * The function returns after 10 attempts or if no pages are movable any more |
| 1079 | * because the list has become empty or no retryable pages exist any more. |
| 1080 | * The caller should call putback_lru_pages() to return pages to the LRU |
Minchan Kim | 28bd657 | 2011-01-25 15:07:26 -0800 | [diff] [blame] | 1081 | * or free list only if ret != 0. |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 1082 | * |
Srivatsa S. Bhat | c73e5c9 | 2013-04-29 15:08:16 -0700 | [diff] [blame] | 1083 | * Returns the number of pages that were not migrated, or an error code. |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 1084 | */ |
Hugh Dickins | 9c620e2 | 2013-02-22 16:35:14 -0800 | [diff] [blame] | 1085 | int migrate_pages(struct list_head *from, new_page_t get_new_page, |
David Rientjes | 68711a7 | 2014-06-04 16:08:25 -0700 | [diff] [blame] | 1086 | free_page_t put_new_page, unsigned long private, |
| 1087 | enum migrate_mode mode, int reason) |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 1088 | { |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 1089 | int retry = 1; |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 1090 | int nr_failed = 0; |
Mel Gorman | 5647bc2 | 2012-10-19 10:46:20 +0100 | [diff] [blame] | 1091 | int nr_succeeded = 0; |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 1092 | int pass = 0; |
| 1093 | struct page *page; |
| 1094 | struct page *page2; |
| 1095 | int swapwrite = current->flags & PF_SWAPWRITE; |
| 1096 | int rc; |
| 1097 | |
| 1098 | if (!swapwrite) |
| 1099 | current->flags |= PF_SWAPWRITE; |
| 1100 | |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 1101 | for(pass = 0; pass < 10 && retry; pass++) { |
| 1102 | retry = 0; |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 1103 | |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 1104 | list_for_each_entry_safe(page, page2, from, lru) { |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 1105 | cond_resched(); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 1106 | |
Naoya Horiguchi | 31caf66 | 2013-09-11 14:21:59 -0700 | [diff] [blame] | 1107 | if (PageHuge(page)) |
| 1108 | rc = unmap_and_move_huge_page(get_new_page, |
David Rientjes | 68711a7 | 2014-06-04 16:08:25 -0700 | [diff] [blame] | 1109 | put_new_page, private, page, |
| 1110 | pass > 2, mode); |
Naoya Horiguchi | 31caf66 | 2013-09-11 14:21:59 -0700 | [diff] [blame] | 1111 | else |
David Rientjes | 68711a7 | 2014-06-04 16:08:25 -0700 | [diff] [blame] | 1112 | rc = unmap_and_move(get_new_page, put_new_page, |
| 1113 | private, page, pass > 2, mode); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 1114 | |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 1115 | switch(rc) { |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 1116 | case -ENOMEM: |
| 1117 | goto out; |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 1118 | case -EAGAIN: |
Christoph Lameter | 2d1db3b | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 1119 | retry++; |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 1120 | break; |
Rafael Aquini | 78bd520 | 2012-12-11 16:02:31 -0800 | [diff] [blame] | 1121 | case MIGRATEPAGE_SUCCESS: |
Mel Gorman | 5647bc2 | 2012-10-19 10:46:20 +0100 | [diff] [blame] | 1122 | nr_succeeded++; |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 1123 | break; |
| 1124 | default: |
Naoya Horiguchi | 354a336 | 2014-01-21 15:51:14 -0800 | [diff] [blame] | 1125 | /* |
| 1126 | * Permanent failure (-EBUSY, -ENOSYS, etc.): |
| 1127 | * unlike -EAGAIN case, the failed page is |
| 1128 | * removed from migration page list and not |
| 1129 | * retried in the next outer loop. |
| 1130 | */ |
Christoph Lameter | 2d1db3b | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 1131 | nr_failed++; |
Christoph Lameter | e24f0b8 | 2006-06-23 02:03:51 -0700 | [diff] [blame] | 1132 | break; |
Christoph Lameter | 2d1db3b | 2006-06-23 02:03:33 -0700 | [diff] [blame] | 1133 | } |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 1134 | } |
| 1135 | } |
Rafael Aquini | 78bd520 | 2012-12-11 16:02:31 -0800 | [diff] [blame] | 1136 | rc = nr_failed + retry; |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 1137 | out: |
Mel Gorman | 5647bc2 | 2012-10-19 10:46:20 +0100 | [diff] [blame] | 1138 | if (nr_succeeded) |
| 1139 | count_vm_events(PGMIGRATE_SUCCESS, nr_succeeded); |
| 1140 | if (nr_failed) |
| 1141 | count_vm_events(PGMIGRATE_FAIL, nr_failed); |
Mel Gorman | 7b2a2d4 | 2012-10-19 14:07:31 +0100 | [diff] [blame] | 1142 | trace_mm_migrate_pages(nr_succeeded, nr_failed, mode, reason); |
| 1143 | |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 1144 | if (!swapwrite) |
| 1145 | current->flags &= ~PF_SWAPWRITE; |
| 1146 | |
Rafael Aquini | 78bd520 | 2012-12-11 16:02:31 -0800 | [diff] [blame] | 1147 | return rc; |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 1148 | } |
| 1149 | |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1150 | #ifdef CONFIG_NUMA |
| 1151 | /* |
| 1152 | * Move a list of individual pages |
| 1153 | */ |
| 1154 | struct page_to_node { |
| 1155 | unsigned long addr; |
| 1156 | struct page *page; |
| 1157 | int node; |
| 1158 | int status; |
| 1159 | }; |
| 1160 | |
| 1161 | static struct page *new_page_node(struct page *p, unsigned long private, |
| 1162 | int **result) |
| 1163 | { |
| 1164 | struct page_to_node *pm = (struct page_to_node *)private; |
| 1165 | |
| 1166 | while (pm->node != MAX_NUMNODES && pm->page != p) |
| 1167 | pm++; |
| 1168 | |
| 1169 | if (pm->node == MAX_NUMNODES) |
| 1170 | return NULL; |
| 1171 | |
| 1172 | *result = &pm->status; |
| 1173 | |
Naoya Horiguchi | e632a93 | 2013-09-11 14:22:04 -0700 | [diff] [blame] | 1174 | if (PageHuge(p)) |
| 1175 | return alloc_huge_page_node(page_hstate(compound_head(p)), |
| 1176 | pm->node); |
| 1177 | else |
| 1178 | return alloc_pages_exact_node(pm->node, |
Johannes Weiner | e97ca8e | 2014-03-10 15:49:43 -0700 | [diff] [blame] | 1179 | GFP_HIGHUSER_MOVABLE | __GFP_THISNODE, 0); |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1180 | } |
| 1181 | |
| 1182 | /* |
| 1183 | * Move a set of pages as indicated in the pm array. The addr |
| 1184 | * field must be set to the virtual address of the page to be moved |
| 1185 | * and the node number must contain a valid target node. |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1186 | * The pm array ends with node = MAX_NUMNODES. |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1187 | */ |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1188 | static int do_move_page_to_node_array(struct mm_struct *mm, |
| 1189 | struct page_to_node *pm, |
| 1190 | int migrate_all) |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1191 | { |
| 1192 | int err; |
| 1193 | struct page_to_node *pp; |
| 1194 | LIST_HEAD(pagelist); |
| 1195 | |
| 1196 | down_read(&mm->mmap_sem); |
| 1197 | |
| 1198 | /* |
| 1199 | * Build a list of pages to migrate |
| 1200 | */ |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1201 | for (pp = pm; pp->node != MAX_NUMNODES; pp++) { |
| 1202 | struct vm_area_struct *vma; |
| 1203 | struct page *page; |
| 1204 | |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1205 | err = -EFAULT; |
| 1206 | vma = find_vma(mm, pp->addr); |
Gleb Natapov | 70384dc | 2010-10-26 14:22:07 -0700 | [diff] [blame] | 1207 | if (!vma || pp->addr < vma->vm_start || !vma_migratable(vma)) |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1208 | goto set_status; |
| 1209 | |
Andrea Arcangeli | 500d65d | 2011-01-13 15:46:55 -0800 | [diff] [blame] | 1210 | page = follow_page(vma, pp->addr, FOLL_GET|FOLL_SPLIT); |
Linus Torvalds | 89f5b7d | 2008-06-20 11:18:25 -0700 | [diff] [blame] | 1211 | |
| 1212 | err = PTR_ERR(page); |
| 1213 | if (IS_ERR(page)) |
| 1214 | goto set_status; |
| 1215 | |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1216 | err = -ENOENT; |
| 1217 | if (!page) |
| 1218 | goto set_status; |
| 1219 | |
Hugh Dickins | 62b61f6 | 2009-12-14 17:59:33 -0800 | [diff] [blame] | 1220 | /* Use PageReserved to check for zero page */ |
Hugh Dickins | b79bc0a | 2013-02-22 16:35:13 -0800 | [diff] [blame] | 1221 | if (PageReserved(page)) |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1222 | goto put_and_set; |
| 1223 | |
| 1224 | pp->page = page; |
| 1225 | err = page_to_nid(page); |
| 1226 | |
| 1227 | if (err == pp->node) |
| 1228 | /* |
| 1229 | * Node already in the right place |
| 1230 | */ |
| 1231 | goto put_and_set; |
| 1232 | |
| 1233 | err = -EACCES; |
| 1234 | if (page_mapcount(page) > 1 && |
| 1235 | !migrate_all) |
| 1236 | goto put_and_set; |
| 1237 | |
Naoya Horiguchi | e632a93 | 2013-09-11 14:22:04 -0700 | [diff] [blame] | 1238 | if (PageHuge(page)) { |
| 1239 | isolate_huge_page(page, &pagelist); |
| 1240 | goto put_and_set; |
| 1241 | } |
| 1242 | |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 1243 | err = isolate_lru_page(page); |
KOSAKI Motohiro | 6d9c285 | 2009-12-14 17:58:11 -0800 | [diff] [blame] | 1244 | if (!err) { |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 1245 | list_add_tail(&page->lru, &pagelist); |
KOSAKI Motohiro | 6d9c285 | 2009-12-14 17:58:11 -0800 | [diff] [blame] | 1246 | inc_zone_page_state(page, NR_ISOLATED_ANON + |
| 1247 | page_is_file_cache(page)); |
| 1248 | } |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1249 | put_and_set: |
| 1250 | /* |
| 1251 | * Either remove the duplicate refcount from |
| 1252 | * isolate_lru_page() or drop the page ref if it was |
| 1253 | * not isolated. |
| 1254 | */ |
| 1255 | put_page(page); |
| 1256 | set_status: |
| 1257 | pp->status = err; |
| 1258 | } |
| 1259 | |
Brice Goglin | e78bbfa | 2008-10-18 20:27:15 -0700 | [diff] [blame] | 1260 | err = 0; |
Minchan Kim | cf608ac | 2010-10-26 14:21:29 -0700 | [diff] [blame] | 1261 | if (!list_empty(&pagelist)) { |
David Rientjes | 68711a7 | 2014-06-04 16:08:25 -0700 | [diff] [blame] | 1262 | err = migrate_pages(&pagelist, new_page_node, NULL, |
Hugh Dickins | 9c620e2 | 2013-02-22 16:35:14 -0800 | [diff] [blame] | 1263 | (unsigned long)pm, MIGRATE_SYNC, MR_SYSCALL); |
Minchan Kim | cf608ac | 2010-10-26 14:21:29 -0700 | [diff] [blame] | 1264 | if (err) |
Naoya Horiguchi | e632a93 | 2013-09-11 14:22:04 -0700 | [diff] [blame] | 1265 | putback_movable_pages(&pagelist); |
Minchan Kim | cf608ac | 2010-10-26 14:21:29 -0700 | [diff] [blame] | 1266 | } |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1267 | |
| 1268 | up_read(&mm->mmap_sem); |
| 1269 | return err; |
| 1270 | } |
| 1271 | |
| 1272 | /* |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1273 | * Migrate an array of page address onto an array of nodes and fill |
| 1274 | * the corresponding array of status. |
| 1275 | */ |
Christoph Lameter | 3268c63 | 2012-03-21 16:34:06 -0700 | [diff] [blame] | 1276 | static int do_pages_move(struct mm_struct *mm, nodemask_t task_nodes, |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1277 | unsigned long nr_pages, |
| 1278 | const void __user * __user *pages, |
| 1279 | const int __user *nodes, |
| 1280 | int __user *status, int flags) |
| 1281 | { |
Brice Goglin | 3140a22 | 2009-01-06 14:38:57 -0800 | [diff] [blame] | 1282 | struct page_to_node *pm; |
Brice Goglin | 3140a22 | 2009-01-06 14:38:57 -0800 | [diff] [blame] | 1283 | unsigned long chunk_nr_pages; |
| 1284 | unsigned long chunk_start; |
| 1285 | int err; |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1286 | |
Brice Goglin | 3140a22 | 2009-01-06 14:38:57 -0800 | [diff] [blame] | 1287 | err = -ENOMEM; |
| 1288 | pm = (struct page_to_node *)__get_free_page(GFP_KERNEL); |
| 1289 | if (!pm) |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1290 | goto out; |
Brice Goglin | 35282a2 | 2009-06-16 15:32:43 -0700 | [diff] [blame] | 1291 | |
| 1292 | migrate_prep(); |
| 1293 | |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1294 | /* |
Brice Goglin | 3140a22 | 2009-01-06 14:38:57 -0800 | [diff] [blame] | 1295 | * Store a chunk of page_to_node array in a page, |
| 1296 | * but keep the last one as a marker |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1297 | */ |
Brice Goglin | 3140a22 | 2009-01-06 14:38:57 -0800 | [diff] [blame] | 1298 | chunk_nr_pages = (PAGE_SIZE / sizeof(struct page_to_node)) - 1; |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1299 | |
Brice Goglin | 3140a22 | 2009-01-06 14:38:57 -0800 | [diff] [blame] | 1300 | for (chunk_start = 0; |
| 1301 | chunk_start < nr_pages; |
| 1302 | chunk_start += chunk_nr_pages) { |
| 1303 | int j; |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1304 | |
Brice Goglin | 3140a22 | 2009-01-06 14:38:57 -0800 | [diff] [blame] | 1305 | if (chunk_start + chunk_nr_pages > nr_pages) |
| 1306 | chunk_nr_pages = nr_pages - chunk_start; |
| 1307 | |
| 1308 | /* fill the chunk pm with addrs and nodes from user-space */ |
| 1309 | for (j = 0; j < chunk_nr_pages; j++) { |
| 1310 | const void __user *p; |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1311 | int node; |
| 1312 | |
Brice Goglin | 3140a22 | 2009-01-06 14:38:57 -0800 | [diff] [blame] | 1313 | err = -EFAULT; |
| 1314 | if (get_user(p, pages + j + chunk_start)) |
| 1315 | goto out_pm; |
| 1316 | pm[j].addr = (unsigned long) p; |
| 1317 | |
| 1318 | if (get_user(node, nodes + j + chunk_start)) |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1319 | goto out_pm; |
| 1320 | |
| 1321 | err = -ENODEV; |
Linus Torvalds | 6f5a55f | 2010-02-05 16:16:50 -0800 | [diff] [blame] | 1322 | if (node < 0 || node >= MAX_NUMNODES) |
| 1323 | goto out_pm; |
| 1324 | |
Lai Jiangshan | 389162c | 2012-12-12 13:51:30 -0800 | [diff] [blame] | 1325 | if (!node_state(node, N_MEMORY)) |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1326 | goto out_pm; |
| 1327 | |
| 1328 | err = -EACCES; |
| 1329 | if (!node_isset(node, task_nodes)) |
| 1330 | goto out_pm; |
| 1331 | |
Brice Goglin | 3140a22 | 2009-01-06 14:38:57 -0800 | [diff] [blame] | 1332 | pm[j].node = node; |
| 1333 | } |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1334 | |
Brice Goglin | 3140a22 | 2009-01-06 14:38:57 -0800 | [diff] [blame] | 1335 | /* End marker for this chunk */ |
| 1336 | pm[chunk_nr_pages].node = MAX_NUMNODES; |
| 1337 | |
| 1338 | /* Migrate this chunk */ |
| 1339 | err = do_move_page_to_node_array(mm, pm, |
| 1340 | flags & MPOL_MF_MOVE_ALL); |
| 1341 | if (err < 0) |
| 1342 | goto out_pm; |
| 1343 | |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1344 | /* Return status information */ |
Brice Goglin | 3140a22 | 2009-01-06 14:38:57 -0800 | [diff] [blame] | 1345 | for (j = 0; j < chunk_nr_pages; j++) |
| 1346 | if (put_user(pm[j].status, status + j + chunk_start)) { |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1347 | err = -EFAULT; |
Brice Goglin | 3140a22 | 2009-01-06 14:38:57 -0800 | [diff] [blame] | 1348 | goto out_pm; |
| 1349 | } |
| 1350 | } |
| 1351 | err = 0; |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1352 | |
| 1353 | out_pm: |
Brice Goglin | 3140a22 | 2009-01-06 14:38:57 -0800 | [diff] [blame] | 1354 | free_page((unsigned long)pm); |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1355 | out: |
| 1356 | return err; |
| 1357 | } |
| 1358 | |
| 1359 | /* |
Brice Goglin | 2f007e7 | 2008-10-18 20:27:16 -0700 | [diff] [blame] | 1360 | * Determine the nodes of an array of pages and store it in an array of status. |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1361 | */ |
Brice Goglin | 80bba12 | 2008-12-09 13:14:23 -0800 | [diff] [blame] | 1362 | static void do_pages_stat_array(struct mm_struct *mm, unsigned long nr_pages, |
| 1363 | const void __user **pages, int *status) |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1364 | { |
Brice Goglin | 2f007e7 | 2008-10-18 20:27:16 -0700 | [diff] [blame] | 1365 | unsigned long i; |
Brice Goglin | 2f007e7 | 2008-10-18 20:27:16 -0700 | [diff] [blame] | 1366 | |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1367 | down_read(&mm->mmap_sem); |
| 1368 | |
Brice Goglin | 2f007e7 | 2008-10-18 20:27:16 -0700 | [diff] [blame] | 1369 | for (i = 0; i < nr_pages; i++) { |
Brice Goglin | 80bba12 | 2008-12-09 13:14:23 -0800 | [diff] [blame] | 1370 | unsigned long addr = (unsigned long)(*pages); |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1371 | struct vm_area_struct *vma; |
| 1372 | struct page *page; |
KOSAKI Motohiro | c095adb | 2008-12-16 16:06:43 +0900 | [diff] [blame] | 1373 | int err = -EFAULT; |
Brice Goglin | 2f007e7 | 2008-10-18 20:27:16 -0700 | [diff] [blame] | 1374 | |
| 1375 | vma = find_vma(mm, addr); |
Gleb Natapov | 70384dc | 2010-10-26 14:22:07 -0700 | [diff] [blame] | 1376 | if (!vma || addr < vma->vm_start) |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1377 | goto set_status; |
| 1378 | |
Brice Goglin | 2f007e7 | 2008-10-18 20:27:16 -0700 | [diff] [blame] | 1379 | page = follow_page(vma, addr, 0); |
Linus Torvalds | 89f5b7d | 2008-06-20 11:18:25 -0700 | [diff] [blame] | 1380 | |
| 1381 | err = PTR_ERR(page); |
| 1382 | if (IS_ERR(page)) |
| 1383 | goto set_status; |
| 1384 | |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1385 | err = -ENOENT; |
| 1386 | /* Use PageReserved to check for zero page */ |
Hugh Dickins | b79bc0a | 2013-02-22 16:35:13 -0800 | [diff] [blame] | 1387 | if (!page || PageReserved(page)) |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1388 | goto set_status; |
| 1389 | |
| 1390 | err = page_to_nid(page); |
| 1391 | set_status: |
Brice Goglin | 80bba12 | 2008-12-09 13:14:23 -0800 | [diff] [blame] | 1392 | *status = err; |
| 1393 | |
| 1394 | pages++; |
| 1395 | status++; |
| 1396 | } |
| 1397 | |
| 1398 | up_read(&mm->mmap_sem); |
| 1399 | } |
| 1400 | |
| 1401 | /* |
| 1402 | * Determine the nodes of a user array of pages and store it in |
| 1403 | * a user array of status. |
| 1404 | */ |
| 1405 | static int do_pages_stat(struct mm_struct *mm, unsigned long nr_pages, |
| 1406 | const void __user * __user *pages, |
| 1407 | int __user *status) |
| 1408 | { |
| 1409 | #define DO_PAGES_STAT_CHUNK_NR 16 |
| 1410 | const void __user *chunk_pages[DO_PAGES_STAT_CHUNK_NR]; |
| 1411 | int chunk_status[DO_PAGES_STAT_CHUNK_NR]; |
Brice Goglin | 80bba12 | 2008-12-09 13:14:23 -0800 | [diff] [blame] | 1412 | |
H. Peter Anvin | 87b8d1a | 2010-02-18 16:13:40 -0800 | [diff] [blame] | 1413 | while (nr_pages) { |
| 1414 | unsigned long chunk_nr; |
Brice Goglin | 80bba12 | 2008-12-09 13:14:23 -0800 | [diff] [blame] | 1415 | |
H. Peter Anvin | 87b8d1a | 2010-02-18 16:13:40 -0800 | [diff] [blame] | 1416 | chunk_nr = nr_pages; |
| 1417 | if (chunk_nr > DO_PAGES_STAT_CHUNK_NR) |
| 1418 | chunk_nr = DO_PAGES_STAT_CHUNK_NR; |
| 1419 | |
| 1420 | if (copy_from_user(chunk_pages, pages, chunk_nr * sizeof(*chunk_pages))) |
| 1421 | break; |
Brice Goglin | 80bba12 | 2008-12-09 13:14:23 -0800 | [diff] [blame] | 1422 | |
| 1423 | do_pages_stat_array(mm, chunk_nr, chunk_pages, chunk_status); |
| 1424 | |
H. Peter Anvin | 87b8d1a | 2010-02-18 16:13:40 -0800 | [diff] [blame] | 1425 | if (copy_to_user(status, chunk_status, chunk_nr * sizeof(*status))) |
| 1426 | break; |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1427 | |
H. Peter Anvin | 87b8d1a | 2010-02-18 16:13:40 -0800 | [diff] [blame] | 1428 | pages += chunk_nr; |
| 1429 | status += chunk_nr; |
| 1430 | nr_pages -= chunk_nr; |
| 1431 | } |
| 1432 | return nr_pages ? -EFAULT : 0; |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1433 | } |
| 1434 | |
| 1435 | /* |
| 1436 | * Move a list of pages in the address space of the currently executing |
| 1437 | * process. |
| 1438 | */ |
Heiko Carstens | 938bb9f | 2009-01-14 14:14:30 +0100 | [diff] [blame] | 1439 | SYSCALL_DEFINE6(move_pages, pid_t, pid, unsigned long, nr_pages, |
| 1440 | const void __user * __user *, pages, |
| 1441 | const int __user *, nodes, |
| 1442 | int __user *, status, int, flags) |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1443 | { |
David Howells | c69e8d9 | 2008-11-14 10:39:19 +1100 | [diff] [blame] | 1444 | const struct cred *cred = current_cred(), *tcred; |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1445 | struct task_struct *task; |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1446 | struct mm_struct *mm; |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1447 | int err; |
Christoph Lameter | 3268c63 | 2012-03-21 16:34:06 -0700 | [diff] [blame] | 1448 | nodemask_t task_nodes; |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1449 | |
| 1450 | /* Check flags */ |
| 1451 | if (flags & ~(MPOL_MF_MOVE|MPOL_MF_MOVE_ALL)) |
| 1452 | return -EINVAL; |
| 1453 | |
| 1454 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) |
| 1455 | return -EPERM; |
| 1456 | |
| 1457 | /* Find the mm_struct */ |
Greg Thelen | a879bf5 | 2011-02-25 14:44:13 -0800 | [diff] [blame] | 1458 | rcu_read_lock(); |
Pavel Emelyanov | 228ebcb | 2007-10-18 23:40:16 -0700 | [diff] [blame] | 1459 | task = pid ? find_task_by_vpid(pid) : current; |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1460 | if (!task) { |
Greg Thelen | a879bf5 | 2011-02-25 14:44:13 -0800 | [diff] [blame] | 1461 | rcu_read_unlock(); |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1462 | return -ESRCH; |
| 1463 | } |
Christoph Lameter | 3268c63 | 2012-03-21 16:34:06 -0700 | [diff] [blame] | 1464 | get_task_struct(task); |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1465 | |
| 1466 | /* |
| 1467 | * Check if this process has the right to modify the specified |
| 1468 | * process. The right exists if the process has administrative |
| 1469 | * capabilities, superuser privileges or the same |
| 1470 | * userid as the target process. |
| 1471 | */ |
David Howells | c69e8d9 | 2008-11-14 10:39:19 +1100 | [diff] [blame] | 1472 | tcred = __task_cred(task); |
Eric W. Biederman | b38a86e | 2012-03-12 15:48:24 -0700 | [diff] [blame] | 1473 | if (!uid_eq(cred->euid, tcred->suid) && !uid_eq(cred->euid, tcred->uid) && |
| 1474 | !uid_eq(cred->uid, tcred->suid) && !uid_eq(cred->uid, tcred->uid) && |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1475 | !capable(CAP_SYS_NICE)) { |
David Howells | c69e8d9 | 2008-11-14 10:39:19 +1100 | [diff] [blame] | 1476 | rcu_read_unlock(); |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1477 | err = -EPERM; |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1478 | goto out; |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1479 | } |
David Howells | c69e8d9 | 2008-11-14 10:39:19 +1100 | [diff] [blame] | 1480 | rcu_read_unlock(); |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1481 | |
David Quigley | 86c3a76 | 2006-06-23 02:04:02 -0700 | [diff] [blame] | 1482 | err = security_task_movememory(task); |
| 1483 | if (err) |
Brice Goglin | 5e9a0f0 | 2008-10-18 20:27:17 -0700 | [diff] [blame] | 1484 | goto out; |
David Quigley | 86c3a76 | 2006-06-23 02:04:02 -0700 | [diff] [blame] | 1485 | |
Christoph Lameter | 3268c63 | 2012-03-21 16:34:06 -0700 | [diff] [blame] | 1486 | task_nodes = cpuset_mems_allowed(task); |
| 1487 | mm = get_task_mm(task); |
| 1488 | put_task_struct(task); |
| 1489 | |
Sasha Levin | 6e8b09e | 2012-04-25 16:01:53 -0700 | [diff] [blame] | 1490 | if (!mm) |
| 1491 | return -EINVAL; |
| 1492 | |
| 1493 | if (nodes) |
| 1494 | err = do_pages_move(mm, task_nodes, nr_pages, pages, |
| 1495 | nodes, status, flags); |
| 1496 | else |
| 1497 | err = do_pages_stat(mm, nr_pages, pages, status); |
Christoph Lameter | 3268c63 | 2012-03-21 16:34:06 -0700 | [diff] [blame] | 1498 | |
| 1499 | mmput(mm); |
| 1500 | return err; |
David Quigley | 86c3a76 | 2006-06-23 02:04:02 -0700 | [diff] [blame] | 1501 | |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1502 | out: |
Christoph Lameter | 3268c63 | 2012-03-21 16:34:06 -0700 | [diff] [blame] | 1503 | put_task_struct(task); |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1504 | return err; |
| 1505 | } |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1506 | |
Peter Zijlstra | 7039e1d | 2012-10-25 14:16:34 +0200 | [diff] [blame] | 1507 | #ifdef CONFIG_NUMA_BALANCING |
| 1508 | /* |
| 1509 | * Returns true if this is a safe migration target node for misplaced NUMA |
| 1510 | * pages. Currently it only checks the watermarks which crude |
| 1511 | */ |
| 1512 | static bool migrate_balanced_pgdat(struct pglist_data *pgdat, |
Mel Gorman | 3abef4e | 2013-02-22 16:34:27 -0800 | [diff] [blame] | 1513 | unsigned long nr_migrate_pages) |
Peter Zijlstra | 7039e1d | 2012-10-25 14:16:34 +0200 | [diff] [blame] | 1514 | { |
| 1515 | int z; |
| 1516 | for (z = pgdat->nr_zones - 1; z >= 0; z--) { |
| 1517 | struct zone *zone = pgdat->node_zones + z; |
| 1518 | |
| 1519 | if (!populated_zone(zone)) |
| 1520 | continue; |
| 1521 | |
Lisa Du | 6e543d5 | 2013-09-11 14:22:36 -0700 | [diff] [blame] | 1522 | if (!zone_reclaimable(zone)) |
Peter Zijlstra | 7039e1d | 2012-10-25 14:16:34 +0200 | [diff] [blame] | 1523 | continue; |
| 1524 | |
| 1525 | /* Avoid waking kswapd by allocating pages_to_migrate pages. */ |
| 1526 | if (!zone_watermark_ok(zone, 0, |
| 1527 | high_wmark_pages(zone) + |
| 1528 | nr_migrate_pages, |
| 1529 | 0, 0)) |
| 1530 | continue; |
| 1531 | return true; |
| 1532 | } |
| 1533 | return false; |
| 1534 | } |
| 1535 | |
| 1536 | static struct page *alloc_misplaced_dst_page(struct page *page, |
| 1537 | unsigned long data, |
| 1538 | int **result) |
| 1539 | { |
| 1540 | int nid = (int) data; |
| 1541 | struct page *newpage; |
| 1542 | |
| 1543 | newpage = alloc_pages_exact_node(nid, |
Johannes Weiner | e97ca8e | 2014-03-10 15:49:43 -0700 | [diff] [blame] | 1544 | (GFP_HIGHUSER_MOVABLE | |
| 1545 | __GFP_THISNODE | __GFP_NOMEMALLOC | |
| 1546 | __GFP_NORETRY | __GFP_NOWARN) & |
Peter Zijlstra | 7039e1d | 2012-10-25 14:16:34 +0200 | [diff] [blame] | 1547 | ~GFP_IOFS, 0); |
Hillf Danton | bac0382 | 2012-11-27 14:46:24 +0000 | [diff] [blame] | 1548 | |
Peter Zijlstra | 7039e1d | 2012-10-25 14:16:34 +0200 | [diff] [blame] | 1549 | return newpage; |
| 1550 | } |
| 1551 | |
| 1552 | /* |
Mel Gorman | a8f6077 | 2012-11-14 21:41:46 +0000 | [diff] [blame] | 1553 | * page migration rate limiting control. |
| 1554 | * Do not migrate more than @pages_to_migrate in a @migrate_interval_millisecs |
| 1555 | * window of time. Default here says do not migrate more than 1280M per second. |
Mel Gorman | e14808b | 2012-11-19 10:59:15 +0000 | [diff] [blame] | 1556 | * If a node is rate-limited then PTE NUMA updates are also rate-limited. However |
| 1557 | * as it is faults that reset the window, pte updates will happen unconditionally |
| 1558 | * if there has not been a fault since @pteupdate_interval_millisecs after the |
| 1559 | * throttle window closed. |
Mel Gorman | a8f6077 | 2012-11-14 21:41:46 +0000 | [diff] [blame] | 1560 | */ |
| 1561 | static unsigned int migrate_interval_millisecs __read_mostly = 100; |
Mel Gorman | e14808b | 2012-11-19 10:59:15 +0000 | [diff] [blame] | 1562 | static unsigned int pteupdate_interval_millisecs __read_mostly = 1000; |
Mel Gorman | a8f6077 | 2012-11-14 21:41:46 +0000 | [diff] [blame] | 1563 | static unsigned int ratelimit_pages __read_mostly = 128 << (20 - PAGE_SHIFT); |
| 1564 | |
Mel Gorman | e14808b | 2012-11-19 10:59:15 +0000 | [diff] [blame] | 1565 | /* Returns true if NUMA migration is currently rate limited */ |
| 1566 | bool migrate_ratelimited(int node) |
| 1567 | { |
| 1568 | pg_data_t *pgdat = NODE_DATA(node); |
| 1569 | |
| 1570 | if (time_after(jiffies, pgdat->numabalancing_migrate_next_window + |
| 1571 | msecs_to_jiffies(pteupdate_interval_millisecs))) |
| 1572 | return false; |
| 1573 | |
| 1574 | if (pgdat->numabalancing_migrate_nr_pages < ratelimit_pages) |
| 1575 | return false; |
| 1576 | |
| 1577 | return true; |
| 1578 | } |
| 1579 | |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1580 | /* Returns true if the node is migrate rate-limited after the update */ |
Mel Gorman | 1c30e01 | 2014-01-21 15:50:58 -0800 | [diff] [blame] | 1581 | static bool numamigrate_update_ratelimit(pg_data_t *pgdat, |
| 1582 | unsigned long nr_pages) |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1583 | { |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1584 | /* |
| 1585 | * Rate-limit the amount of data that is being migrated to a node. |
| 1586 | * Optimal placement is no good if the memory bus is saturated and |
| 1587 | * all the time is being spent migrating! |
| 1588 | */ |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1589 | if (time_after(jiffies, pgdat->numabalancing_migrate_next_window)) { |
Mel Gorman | 1c5e9c2 | 2014-01-21 15:50:59 -0800 | [diff] [blame] | 1590 | spin_lock(&pgdat->numabalancing_migrate_lock); |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1591 | pgdat->numabalancing_migrate_nr_pages = 0; |
| 1592 | pgdat->numabalancing_migrate_next_window = jiffies + |
| 1593 | msecs_to_jiffies(migrate_interval_millisecs); |
Mel Gorman | 1c5e9c2 | 2014-01-21 15:50:59 -0800 | [diff] [blame] | 1594 | spin_unlock(&pgdat->numabalancing_migrate_lock); |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1595 | } |
Mel Gorman | af1839d | 2014-01-21 15:51:01 -0800 | [diff] [blame] | 1596 | if (pgdat->numabalancing_migrate_nr_pages > ratelimit_pages) { |
| 1597 | trace_mm_numa_migrate_ratelimit(current, pgdat->node_id, |
| 1598 | nr_pages); |
Mel Gorman | 1c5e9c2 | 2014-01-21 15:50:59 -0800 | [diff] [blame] | 1599 | return true; |
Mel Gorman | af1839d | 2014-01-21 15:51:01 -0800 | [diff] [blame] | 1600 | } |
Mel Gorman | 1c5e9c2 | 2014-01-21 15:50:59 -0800 | [diff] [blame] | 1601 | |
| 1602 | /* |
| 1603 | * This is an unlocked non-atomic update so errors are possible. |
| 1604 | * The consequences are failing to migrate when we potentiall should |
| 1605 | * have which is not severe enough to warrant locking. If it is ever |
| 1606 | * a problem, it can be converted to a per-cpu counter. |
| 1607 | */ |
| 1608 | pgdat->numabalancing_migrate_nr_pages += nr_pages; |
| 1609 | return false; |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1610 | } |
| 1611 | |
Mel Gorman | 1c30e01 | 2014-01-21 15:50:58 -0800 | [diff] [blame] | 1612 | static int numamigrate_isolate_page(pg_data_t *pgdat, struct page *page) |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1613 | { |
Hugh Dickins | 340ef39 | 2013-02-22 16:34:33 -0800 | [diff] [blame] | 1614 | int page_lru; |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1615 | |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 1616 | VM_BUG_ON_PAGE(compound_order(page) && !PageTransHuge(page), page); |
Mel Gorman | 3abef4e | 2013-02-22 16:34:27 -0800 | [diff] [blame] | 1617 | |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1618 | /* Avoid migrating to a node that is nearly full */ |
Hugh Dickins | 340ef39 | 2013-02-22 16:34:33 -0800 | [diff] [blame] | 1619 | if (!migrate_balanced_pgdat(pgdat, 1UL << compound_order(page))) |
| 1620 | return 0; |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1621 | |
Hugh Dickins | 340ef39 | 2013-02-22 16:34:33 -0800 | [diff] [blame] | 1622 | if (isolate_lru_page(page)) |
| 1623 | return 0; |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1624 | |
| 1625 | /* |
Hugh Dickins | 340ef39 | 2013-02-22 16:34:33 -0800 | [diff] [blame] | 1626 | * migrate_misplaced_transhuge_page() skips page migration's usual |
| 1627 | * check on page_count(), so we must do it here, now that the page |
| 1628 | * has been isolated: a GUP pin, or any other pin, prevents migration. |
| 1629 | * The expected page count is 3: 1 for page's mapcount and 1 for the |
| 1630 | * caller's pin and 1 for the reference taken by isolate_lru_page(). |
| 1631 | */ |
| 1632 | if (PageTransHuge(page) && page_count(page) != 3) { |
| 1633 | putback_lru_page(page); |
| 1634 | return 0; |
| 1635 | } |
| 1636 | |
| 1637 | page_lru = page_is_file_cache(page); |
| 1638 | mod_zone_page_state(page_zone(page), NR_ISOLATED_ANON + page_lru, |
| 1639 | hpage_nr_pages(page)); |
| 1640 | |
| 1641 | /* |
| 1642 | * Isolating the page has taken another reference, so the |
| 1643 | * caller's reference can be safely dropped without the page |
| 1644 | * disappearing underneath us during migration. |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1645 | */ |
| 1646 | put_page(page); |
Hugh Dickins | 340ef39 | 2013-02-22 16:34:33 -0800 | [diff] [blame] | 1647 | return 1; |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1648 | } |
| 1649 | |
Mel Gorman | de466bd | 2013-12-18 17:08:42 -0800 | [diff] [blame] | 1650 | bool pmd_trans_migrating(pmd_t pmd) |
| 1651 | { |
| 1652 | struct page *page = pmd_page(pmd); |
| 1653 | return PageLocked(page); |
| 1654 | } |
| 1655 | |
| 1656 | void wait_migrate_huge_page(struct anon_vma *anon_vma, pmd_t *pmd) |
| 1657 | { |
| 1658 | struct page *page = pmd_page(*pmd); |
| 1659 | wait_on_page_locked(page); |
| 1660 | } |
| 1661 | |
Mel Gorman | a8f6077 | 2012-11-14 21:41:46 +0000 | [diff] [blame] | 1662 | /* |
Peter Zijlstra | 7039e1d | 2012-10-25 14:16:34 +0200 | [diff] [blame] | 1663 | * Attempt to migrate a misplaced page to the specified destination |
| 1664 | * node. Caller is expected to have an elevated reference count on |
| 1665 | * the page that will be dropped by this function before returning. |
| 1666 | */ |
Mel Gorman | 1bc115d | 2013-10-07 11:29:05 +0100 | [diff] [blame] | 1667 | int migrate_misplaced_page(struct page *page, struct vm_area_struct *vma, |
| 1668 | int node) |
Peter Zijlstra | 7039e1d | 2012-10-25 14:16:34 +0200 | [diff] [blame] | 1669 | { |
Mel Gorman | a8f6077 | 2012-11-14 21:41:46 +0000 | [diff] [blame] | 1670 | pg_data_t *pgdat = NODE_DATA(node); |
Hugh Dickins | 340ef39 | 2013-02-22 16:34:33 -0800 | [diff] [blame] | 1671 | int isolated; |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1672 | int nr_remaining; |
Peter Zijlstra | 7039e1d | 2012-10-25 14:16:34 +0200 | [diff] [blame] | 1673 | LIST_HEAD(migratepages); |
| 1674 | |
| 1675 | /* |
Mel Gorman | 1bc115d | 2013-10-07 11:29:05 +0100 | [diff] [blame] | 1676 | * Don't migrate file pages that are mapped in multiple processes |
| 1677 | * with execute permissions as they are probably shared libraries. |
Peter Zijlstra | 7039e1d | 2012-10-25 14:16:34 +0200 | [diff] [blame] | 1678 | */ |
Mel Gorman | 1bc115d | 2013-10-07 11:29:05 +0100 | [diff] [blame] | 1679 | if (page_mapcount(page) != 1 && page_is_file_cache(page) && |
| 1680 | (vma->vm_flags & VM_EXEC)) |
Peter Zijlstra | 7039e1d | 2012-10-25 14:16:34 +0200 | [diff] [blame] | 1681 | goto out; |
Peter Zijlstra | 7039e1d | 2012-10-25 14:16:34 +0200 | [diff] [blame] | 1682 | |
Mel Gorman | a8f6077 | 2012-11-14 21:41:46 +0000 | [diff] [blame] | 1683 | /* |
| 1684 | * Rate-limit the amount of data that is being migrated to a node. |
| 1685 | * Optimal placement is no good if the memory bus is saturated and |
| 1686 | * all the time is being spent migrating! |
| 1687 | */ |
Hugh Dickins | 340ef39 | 2013-02-22 16:34:33 -0800 | [diff] [blame] | 1688 | if (numamigrate_update_ratelimit(pgdat, 1)) |
Mel Gorman | a8f6077 | 2012-11-14 21:41:46 +0000 | [diff] [blame] | 1689 | goto out; |
Mel Gorman | a8f6077 | 2012-11-14 21:41:46 +0000 | [diff] [blame] | 1690 | |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1691 | isolated = numamigrate_isolate_page(pgdat, page); |
| 1692 | if (!isolated) |
| 1693 | goto out; |
Peter Zijlstra | 7039e1d | 2012-10-25 14:16:34 +0200 | [diff] [blame] | 1694 | |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1695 | list_add(&page->lru, &migratepages); |
Hugh Dickins | 9c620e2 | 2013-02-22 16:35:14 -0800 | [diff] [blame] | 1696 | nr_remaining = migrate_pages(&migratepages, alloc_misplaced_dst_page, |
David Rientjes | 68711a7 | 2014-06-04 16:08:25 -0700 | [diff] [blame] | 1697 | NULL, node, MIGRATE_ASYNC, |
| 1698 | MR_NUMA_MISPLACED); |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1699 | if (nr_remaining) { |
Joonsoo Kim | 59c82b7 | 2014-01-21 15:51:17 -0800 | [diff] [blame] | 1700 | if (!list_empty(&migratepages)) { |
| 1701 | list_del(&page->lru); |
| 1702 | dec_zone_page_state(page, NR_ISOLATED_ANON + |
| 1703 | page_is_file_cache(page)); |
| 1704 | putback_lru_page(page); |
| 1705 | } |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1706 | isolated = 0; |
| 1707 | } else |
| 1708 | count_vm_numa_event(NUMA_PAGE_MIGRATE); |
Peter Zijlstra | 7039e1d | 2012-10-25 14:16:34 +0200 | [diff] [blame] | 1709 | BUG_ON(!list_empty(&migratepages)); |
Peter Zijlstra | 7039e1d | 2012-10-25 14:16:34 +0200 | [diff] [blame] | 1710 | return isolated; |
Hugh Dickins | 340ef39 | 2013-02-22 16:34:33 -0800 | [diff] [blame] | 1711 | |
| 1712 | out: |
| 1713 | put_page(page); |
| 1714 | return 0; |
Peter Zijlstra | 7039e1d | 2012-10-25 14:16:34 +0200 | [diff] [blame] | 1715 | } |
Mel Gorman | 220018d | 2012-12-05 09:32:56 +0000 | [diff] [blame] | 1716 | #endif /* CONFIG_NUMA_BALANCING */ |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1717 | |
Mel Gorman | 220018d | 2012-12-05 09:32:56 +0000 | [diff] [blame] | 1718 | #if defined(CONFIG_NUMA_BALANCING) && defined(CONFIG_TRANSPARENT_HUGEPAGE) |
Hugh Dickins | 340ef39 | 2013-02-22 16:34:33 -0800 | [diff] [blame] | 1719 | /* |
| 1720 | * Migrates a THP to a given target node. page must be locked and is unlocked |
| 1721 | * before returning. |
| 1722 | */ |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1723 | int migrate_misplaced_transhuge_page(struct mm_struct *mm, |
| 1724 | struct vm_area_struct *vma, |
| 1725 | pmd_t *pmd, pmd_t entry, |
| 1726 | unsigned long address, |
| 1727 | struct page *page, int node) |
| 1728 | { |
Kirill A. Shutemov | c4088eb | 2013-11-14 14:31:04 -0800 | [diff] [blame] | 1729 | spinlock_t *ptl; |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1730 | pg_data_t *pgdat = NODE_DATA(node); |
| 1731 | int isolated = 0; |
| 1732 | struct page *new_page = NULL; |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1733 | int page_lru = page_is_file_cache(page); |
Mel Gorman | f714f4f | 2013-12-18 17:08:33 -0800 | [diff] [blame] | 1734 | unsigned long mmun_start = address & HPAGE_PMD_MASK; |
| 1735 | unsigned long mmun_end = mmun_start + HPAGE_PMD_SIZE; |
Mel Gorman | 2b4847e | 2013-12-18 17:08:32 -0800 | [diff] [blame] | 1736 | pmd_t orig_entry; |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1737 | |
| 1738 | /* |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1739 | * Rate-limit the amount of data that is being migrated to a node. |
| 1740 | * Optimal placement is no good if the memory bus is saturated and |
| 1741 | * all the time is being spent migrating! |
| 1742 | */ |
Mel Gorman | d28d4335 | 2012-11-29 09:24:36 +0000 | [diff] [blame] | 1743 | if (numamigrate_update_ratelimit(pgdat, HPAGE_PMD_NR)) |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1744 | goto out_dropref; |
| 1745 | |
| 1746 | new_page = alloc_pages_node(node, |
Johannes Weiner | e97ca8e | 2014-03-10 15:49:43 -0700 | [diff] [blame] | 1747 | (GFP_TRANSHUGE | __GFP_THISNODE) & ~__GFP_WAIT, |
| 1748 | HPAGE_PMD_ORDER); |
Hugh Dickins | 340ef39 | 2013-02-22 16:34:33 -0800 | [diff] [blame] | 1749 | if (!new_page) |
| 1750 | goto out_fail; |
| 1751 | |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1752 | isolated = numamigrate_isolate_page(pgdat, page); |
Hugh Dickins | 340ef39 | 2013-02-22 16:34:33 -0800 | [diff] [blame] | 1753 | if (!isolated) { |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1754 | put_page(new_page); |
Hugh Dickins | 340ef39 | 2013-02-22 16:34:33 -0800 | [diff] [blame] | 1755 | goto out_fail; |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1756 | } |
| 1757 | |
Mel Gorman | b0943d6 | 2013-12-18 17:08:46 -0800 | [diff] [blame] | 1758 | if (mm_tlb_flush_pending(mm)) |
| 1759 | flush_tlb_range(vma, mmun_start, mmun_end); |
| 1760 | |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1761 | /* Prepare a page as a migration target */ |
| 1762 | __set_page_locked(new_page); |
| 1763 | SetPageSwapBacked(new_page); |
| 1764 | |
| 1765 | /* anon mapping, we can simply copy page->mapping to the new page: */ |
| 1766 | new_page->mapping = page->mapping; |
| 1767 | new_page->index = page->index; |
| 1768 | migrate_page_copy(new_page, page); |
| 1769 | WARN_ON(PageLRU(new_page)); |
| 1770 | |
| 1771 | /* Recheck the target PMD */ |
Mel Gorman | f714f4f | 2013-12-18 17:08:33 -0800 | [diff] [blame] | 1772 | mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); |
Kirill A. Shutemov | c4088eb | 2013-11-14 14:31:04 -0800 | [diff] [blame] | 1773 | ptl = pmd_lock(mm, pmd); |
Mel Gorman | 2b4847e | 2013-12-18 17:08:32 -0800 | [diff] [blame] | 1774 | if (unlikely(!pmd_same(*pmd, entry) || page_count(page) != 2)) { |
| 1775 | fail_putback: |
Kirill A. Shutemov | c4088eb | 2013-11-14 14:31:04 -0800 | [diff] [blame] | 1776 | spin_unlock(ptl); |
Mel Gorman | f714f4f | 2013-12-18 17:08:33 -0800 | [diff] [blame] | 1777 | mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1778 | |
| 1779 | /* Reverse changes made by migrate_page_copy() */ |
| 1780 | if (TestClearPageActive(new_page)) |
| 1781 | SetPageActive(page); |
| 1782 | if (TestClearPageUnevictable(new_page)) |
| 1783 | SetPageUnevictable(page); |
| 1784 | mlock_migrate_page(page, new_page); |
| 1785 | |
| 1786 | unlock_page(new_page); |
| 1787 | put_page(new_page); /* Free it */ |
| 1788 | |
Mel Gorman | a54a407 | 2013-10-07 11:28:46 +0100 | [diff] [blame] | 1789 | /* Retake the callers reference and putback on LRU */ |
| 1790 | get_page(page); |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1791 | putback_lru_page(page); |
Mel Gorman | a54a407 | 2013-10-07 11:28:46 +0100 | [diff] [blame] | 1792 | mod_zone_page_state(page_zone(page), |
| 1793 | NR_ISOLATED_ANON + page_lru, -HPAGE_PMD_NR); |
Mel Gorman | eb4489f6 | 2013-12-18 17:08:39 -0800 | [diff] [blame] | 1794 | |
| 1795 | goto out_unlock; |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1796 | } |
| 1797 | |
Mel Gorman | 2b4847e | 2013-12-18 17:08:32 -0800 | [diff] [blame] | 1798 | orig_entry = *pmd; |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1799 | entry = mk_pmd(new_page, vma->vm_page_prot); |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1800 | entry = pmd_mkhuge(entry); |
Mel Gorman | 2b4847e | 2013-12-18 17:08:32 -0800 | [diff] [blame] | 1801 | entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma); |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1802 | |
Mel Gorman | 2b4847e | 2013-12-18 17:08:32 -0800 | [diff] [blame] | 1803 | /* |
| 1804 | * Clear the old entry under pagetable lock and establish the new PTE. |
| 1805 | * Any parallel GUP will either observe the old page blocking on the |
| 1806 | * page lock, block on the page table lock or observe the new page. |
| 1807 | * The SetPageUptodate on the new page and page_add_new_anon_rmap |
| 1808 | * guarantee the copy is visible before the pagetable update. |
| 1809 | */ |
Mel Gorman | f714f4f | 2013-12-18 17:08:33 -0800 | [diff] [blame] | 1810 | flush_cache_range(vma, mmun_start, mmun_end); |
Mel Gorman | 11de992 | 2014-06-04 16:07:41 -0700 | [diff] [blame] | 1811 | page_add_anon_rmap(new_page, vma, mmun_start); |
Joerg Roedel | 34ee645 | 2014-11-13 13:46:09 +1100 | [diff] [blame] | 1812 | pmdp_clear_flush_notify(vma, mmun_start, pmd); |
Mel Gorman | f714f4f | 2013-12-18 17:08:33 -0800 | [diff] [blame] | 1813 | set_pmd_at(mm, mmun_start, pmd, entry); |
| 1814 | flush_tlb_range(vma, mmun_start, mmun_end); |
Stephen Rothwell | ce4a9cc | 2012-12-10 19:50:57 +1100 | [diff] [blame] | 1815 | update_mmu_cache_pmd(vma, address, &entry); |
Mel Gorman | 2b4847e | 2013-12-18 17:08:32 -0800 | [diff] [blame] | 1816 | |
| 1817 | if (page_count(page) != 2) { |
Mel Gorman | f714f4f | 2013-12-18 17:08:33 -0800 | [diff] [blame] | 1818 | set_pmd_at(mm, mmun_start, pmd, orig_entry); |
| 1819 | flush_tlb_range(vma, mmun_start, mmun_end); |
Joerg Roedel | 34ee645 | 2014-11-13 13:46:09 +1100 | [diff] [blame] | 1820 | mmu_notifier_invalidate_range(mm, mmun_start, mmun_end); |
Mel Gorman | 2b4847e | 2013-12-18 17:08:32 -0800 | [diff] [blame] | 1821 | update_mmu_cache_pmd(vma, address, &entry); |
| 1822 | page_remove_rmap(new_page); |
| 1823 | goto fail_putback; |
| 1824 | } |
| 1825 | |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 1826 | mem_cgroup_migrate(page, new_page, false); |
| 1827 | |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1828 | page_remove_rmap(page); |
Mel Gorman | 2b4847e | 2013-12-18 17:08:32 -0800 | [diff] [blame] | 1829 | |
Kirill A. Shutemov | c4088eb | 2013-11-14 14:31:04 -0800 | [diff] [blame] | 1830 | spin_unlock(ptl); |
Mel Gorman | f714f4f | 2013-12-18 17:08:33 -0800 | [diff] [blame] | 1831 | mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1832 | |
Mel Gorman | 11de992 | 2014-06-04 16:07:41 -0700 | [diff] [blame] | 1833 | /* Take an "isolate" reference and put new page on the LRU. */ |
| 1834 | get_page(new_page); |
| 1835 | putback_lru_page(new_page); |
| 1836 | |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1837 | unlock_page(new_page); |
| 1838 | unlock_page(page); |
| 1839 | put_page(page); /* Drop the rmap reference */ |
| 1840 | put_page(page); /* Drop the LRU isolation reference */ |
| 1841 | |
| 1842 | count_vm_events(PGMIGRATE_SUCCESS, HPAGE_PMD_NR); |
| 1843 | count_vm_numa_events(NUMA_PAGE_MIGRATE, HPAGE_PMD_NR); |
| 1844 | |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1845 | mod_zone_page_state(page_zone(page), |
| 1846 | NR_ISOLATED_ANON + page_lru, |
| 1847 | -HPAGE_PMD_NR); |
| 1848 | return isolated; |
| 1849 | |
Hugh Dickins | 340ef39 | 2013-02-22 16:34:33 -0800 | [diff] [blame] | 1850 | out_fail: |
| 1851 | count_vm_events(PGMIGRATE_FAIL, HPAGE_PMD_NR); |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1852 | out_dropref: |
Mel Gorman | 2b4847e | 2013-12-18 17:08:32 -0800 | [diff] [blame] | 1853 | ptl = pmd_lock(mm, pmd); |
| 1854 | if (pmd_same(*pmd, entry)) { |
| 1855 | entry = pmd_mknonnuma(entry); |
Mel Gorman | f714f4f | 2013-12-18 17:08:33 -0800 | [diff] [blame] | 1856 | set_pmd_at(mm, mmun_start, pmd, entry); |
Mel Gorman | 2b4847e | 2013-12-18 17:08:32 -0800 | [diff] [blame] | 1857 | update_mmu_cache_pmd(vma, address, &entry); |
| 1858 | } |
| 1859 | spin_unlock(ptl); |
Mel Gorman | a54a407 | 2013-10-07 11:28:46 +0100 | [diff] [blame] | 1860 | |
Mel Gorman | eb4489f6 | 2013-12-18 17:08:39 -0800 | [diff] [blame] | 1861 | out_unlock: |
Hugh Dickins | 340ef39 | 2013-02-22 16:34:33 -0800 | [diff] [blame] | 1862 | unlock_page(page); |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1863 | put_page(page); |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 1864 | return 0; |
| 1865 | } |
Peter Zijlstra | 7039e1d | 2012-10-25 14:16:34 +0200 | [diff] [blame] | 1866 | #endif /* CONFIG_NUMA_BALANCING */ |
| 1867 | |
| 1868 | #endif /* CONFIG_NUMA */ |