include/linux/migrate.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 #ifndef _LINUX_MIGRATE_H
   3 #define _LINUX_MIGRATE_H
   4
   5 #include <linux/mm.h>
   6 #include <linux/mempolicy.h>
   7 #include <linux/migrate_mode.h>
   8 #include <linux/hugetlb.h>
   9
  10 typedef struct page *new_page_t(struct page *page, unsigned long private,
  11                                 int **reason);
  12 typedef void free_page_t(struct page *page, unsigned long private);
  13
  14 /*
  15  * Return values from addresss_space_operations.migratepage():
  16  * - negative errno on page migration failure;
  17  * - zero on page migration success;
  18  */
  19 #define MIGRATEPAGE_SUCCESS             0
  20
  21 enum migrate_reason {
  22         MR_COMPACTION,
  23         MR_MEMORY_FAILURE,
  24         MR_MEMORY_HOTPLUG,
  25         MR_SYSCALL,             /* also applies to cpusets */
  26         MR_MEMPOLICY_MBIND,
  27         MR_NUMA_MISPLACED,
  28         MR_CMA,
  29         MR_TYPES
  30 };
  31
  32 /* In mm/debug.c; also keep sync with include/trace/events/migrate.h */
  33 extern char *migrate_reason_names[MR_TYPES];
  34
  35 static inline struct page *new_page_nodemask(struct page *page,
  36                                 int preferred_nid, nodemask_t *nodemask)
  37 {
  38         gfp_t gfp_mask = GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL;
  39         unsigned int order = 0;
  40         struct page *new_page = NULL;
  41
  42         if (PageHuge(page))
  43                 return alloc_huge_page_nodemask(page_hstate(compound_head(page)),
  44                                 preferred_nid, nodemask);
  45
  46         if (thp_migration_supported() && PageTransHuge(page)) {
  47                 order = HPAGE_PMD_ORDER;
  48                 gfp_mask |= GFP_TRANSHUGE;
  49         }
  50
  51         if (PageHighMem(page) || (zone_idx(page_zone(page)) == ZONE_MOVABLE))
  52                 gfp_mask |= __GFP_HIGHMEM;
  53
  54         new_page = __alloc_pages_nodemask(gfp_mask, order,
  55                                 preferred_nid, nodemask);
  56
  57         if (new_page && PageTransHuge(new_page))
  58                 prep_transhuge_page(new_page);
  59
  60         return new_page;
  61 }
  62
  63 #ifdef CONFIG_MIGRATION
  64
  65 extern void putback_movable_pages(struct list_head *l);
  66 extern int migrate_page(struct address_space *mapping,
  67                         struct page *newpage, struct page *page,
  68                         enum migrate_mode mode);
  69 extern int migrate_pages(struct list_head *l, new_page_t new, free_page_t free,
  70                 unsigned long private, enum migrate_mode mode, int reason);
  71 extern int isolate_movable_page(struct page *page, isolate_mode_t mode);
  72 extern void putback_movable_page(struct page *page);
  73
  74 extern int migrate_prep(void);
  75 extern int migrate_prep_local(void);
  76 extern void migrate_page_states(struct page *newpage, struct page *page);
  77 extern void migrate_page_copy(struct page *newpage, struct page *page);
  78 extern int migrate_huge_page_move_mapping(struct address_space *mapping,
  79                                   struct page *newpage, struct page *page);
  80 extern int migrate_page_move_mapping(struct address_space *mapping,
  81                 struct page *newpage, struct page *page,
  82                 struct buffer_head *head, enum migrate_mode mode,
  83                 int extra_count);
  84 #else
  85
  86 static inline void putback_movable_pages(struct list_head *l) {}
  87 static inline int migrate_pages(struct list_head *l, new_page_t new,
  88                 free_page_t free, unsigned long private, enum migrate_mode mode,
  89                 int reason)
  90         { return -ENOSYS; }
  91 static inline int isolate_movable_page(struct page *page, isolate_mode_t mode)
  92         { return -EBUSY; }
  93
  94 static inline int migrate_prep(void) { return -ENOSYS; }
  95 static inline int migrate_prep_local(void) { return -ENOSYS; }
  96
  97 static inline void migrate_page_states(struct page *newpage, struct page *page)
  98 {
  99 }
 100
 101 static inline void migrate_page_copy(struct page *newpage,
 102                                      struct page *page) {}
 103
 104 static inline int migrate_huge_page_move_mapping(struct address_space *mapping,
 105                                   struct page *newpage, struct page *page)
 106 {
 107         return -ENOSYS;
 108 }
 109
 110 #endif /* CONFIG_MIGRATION */
 111
 112 #ifdef CONFIG_COMPACTION
 113 extern int PageMovable(struct page *page);
 114 extern void __SetPageMovable(struct page *page, struct address_space *mapping);
 115 extern void __ClearPageMovable(struct page *page);
 116 #else
 117 static inline int PageMovable(struct page *page) { return 0; };
 118 static inline void __SetPageMovable(struct page *page,
 119                                 struct address_space *mapping)
 120 {
 121 }
 122 static inline void __ClearPageMovable(struct page *page)
 123 {
 124 }
 125 #endif
 126
 127 #ifdef CONFIG_NUMA_BALANCING
 128 extern bool pmd_trans_migrating(pmd_t pmd);
 129 extern int migrate_misplaced_page(struct page *page,
 130                                   struct vm_area_struct *vma, int node);
 131 #else
 132 static inline bool pmd_trans_migrating(pmd_t pmd)
 133 {
 134         return false;
 135 }
 136 static inline int migrate_misplaced_page(struct page *page,
 137                                          struct vm_area_struct *vma, int node)
 138 {
 139         return -EAGAIN; /* can't migrate now */
 140 }
 141 #endif /* CONFIG_NUMA_BALANCING */
 142
 143 #if defined(CONFIG_NUMA_BALANCING) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
 144 extern int migrate_misplaced_transhuge_page(struct mm_struct *mm,
 145                         struct vm_area_struct *vma,
 146                         pmd_t *pmd, pmd_t entry,
 147                         unsigned long address,
 148                         struct page *page, int node);
 149 #else
 150 static inline int migrate_misplaced_transhuge_page(struct mm_struct *mm,
 151                         struct vm_area_struct *vma,
 152                         pmd_t *pmd, pmd_t entry,
 153                         unsigned long address,
 154                         struct page *page, int node)
 155 {
 156         return -EAGAIN;
 157 }
 158 #endif /* CONFIG_NUMA_BALANCING && CONFIG_TRANSPARENT_HUGEPAGE*/
 159
 160
 161 #ifdef CONFIG_MIGRATION
 162
 163 /*
 164  * Watch out for PAE architecture, which has an unsigned long, and might not
 165  * have enough bits to store all physical address and flags. So far we have
 166  * enough room for all our flags.
 167  */
 168 #define MIGRATE_PFN_VALID       (1UL << 0)
 169 #define MIGRATE_PFN_MIGRATE     (1UL << 1)
 170 #define MIGRATE_PFN_LOCKED      (1UL << 2)
 171 #define MIGRATE_PFN_WRITE       (1UL << 3)
 172 #define MIGRATE_PFN_DEVICE      (1UL << 4)
 173 #define MIGRATE_PFN_ERROR       (1UL << 5)
 174 #define MIGRATE_PFN_SHIFT       6
 175
 176 static inline struct page *migrate_pfn_to_page(unsigned long mpfn)
 177 {
 178         if (!(mpfn & MIGRATE_PFN_VALID))
 179                 return NULL;
 180         return pfn_to_page(mpfn >> MIGRATE_PFN_SHIFT);
 181 }
 182
 183 static inline unsigned long migrate_pfn(unsigned long pfn)
 184 {
 185         return (pfn << MIGRATE_PFN_SHIFT) | MIGRATE_PFN_VALID;
 186 }
 187
 188 /*
 189  * struct migrate_vma_ops - migrate operation callback
 190  *
 191  * @alloc_and_copy: alloc destination memory and copy source memory to it
 192  * @finalize_and_map: allow caller to map the successfully migrated pages
 193  *
 194  *
 195  * The alloc_and_copy() callback happens once all source pages have been locked,
 196  * unmapped and checked (checked whether pinned or not). All pages that can be
 197  * migrated will have an entry in the src array set with the pfn value of the
 198  * page and with the MIGRATE_PFN_VALID and MIGRATE_PFN_MIGRATE flag set (other
 199  * flags might be set but should be ignored by the callback).
 200  *
 201  * The alloc_and_copy() callback can then allocate destination memory and copy
 202  * source memory to it for all those entries (ie with MIGRATE_PFN_VALID and
 203  * MIGRATE_PFN_MIGRATE flag set). Once these are allocated and copied, the
 204  * callback must update each corresponding entry in the dst array with the pfn
 205  * value of the destination page and with the MIGRATE_PFN_VALID and
 206  * MIGRATE_PFN_LOCKED flags set (destination pages must have their struct pages
 207  * locked, via lock_page()).
 208  *
 209  * At this point the alloc_and_copy() callback is done and returns.
 210  *
 211  * Note that the callback does not have to migrate all the pages that are
 212  * marked with MIGRATE_PFN_MIGRATE flag in src array unless this is a migration
 213  * from device memory to system memory (ie the MIGRATE_PFN_DEVICE flag is also
 214  * set in the src array entry). If the device driver cannot migrate a device
 215  * page back to system memory, then it must set the corresponding dst array
 216  * entry to MIGRATE_PFN_ERROR. This will trigger a SIGBUS if CPU tries to
 217  * access any of the virtual addresses originally backed by this page. Because
 218  * a SIGBUS is such a severe result for the userspace process, the device
 219  * driver should avoid setting MIGRATE_PFN_ERROR unless it is really in an
 220  * unrecoverable state.
 221  *
 222  * For empty entry inside CPU page table (pte_none() or pmd_none() is true) we
 223  * do set MIGRATE_PFN_MIGRATE flag inside the corresponding source array thus
 224  * allowing device driver to allocate device memory for those unback virtual
 225  * address. For this the device driver simply have to allocate device memory
 226  * and properly set the destination entry like for regular migration. Note that
 227  * this can still fails and thus inside the device driver must check if the
 228  * migration was successful for those entry inside the finalize_and_map()
 229  * callback just like for regular migration.
 230  *
 231  * THE alloc_and_copy() CALLBACK MUST NOT CHANGE ANY OF THE SRC ARRAY ENTRIES
 232  * OR BAD THINGS WILL HAPPEN !
 233  *
 234  *
 235  * The finalize_and_map() callback happens after struct page migration from
 236  * source to destination (destination struct pages are the struct pages for the
 237  * memory allocated by the alloc_and_copy() callback).  Migration can fail, and
 238  * thus the finalize_and_map() allows the driver to inspect which pages were
 239  * successfully migrated, and which were not. Successfully migrated pages will
 240  * have the MIGRATE_PFN_MIGRATE flag set for their src array entry.
 241  *
 242  * It is safe to update device page table from within the finalize_and_map()
 243  * callback because both destination and source page are still locked, and the
 244  * mmap_sem is held in read mode (hence no one can unmap the range being
 245  * migrated).
 246  *
 247  * Once callback is done cleaning up things and updating its page table (if it
 248  * chose to do so, this is not an obligation) then it returns. At this point,
 249  * the HMM core will finish up the final steps, and the migration is complete.
 250  *
 251  * THE finalize_and_map() CALLBACK MUST NOT CHANGE ANY OF THE SRC OR DST ARRAY
 252  * ENTRIES OR BAD THINGS WILL HAPPEN !
 253  */
 254 struct migrate_vma_ops {
 255         void (*alloc_and_copy)(struct vm_area_struct *vma,
 256                                const unsigned long *src,
 257                                unsigned long *dst,
 258                                unsigned long start,
 259                                unsigned long end,
 260                                void *private);
 261         void (*finalize_and_map)(struct vm_area_struct *vma,
 262                                  const unsigned long *src,
 263                                  const unsigned long *dst,
 264                                  unsigned long start,
 265                                  unsigned long end,
 266                                  void *private);
 267 };
 268
 269 #if defined(CONFIG_MIGRATE_VMA_HELPER)
 270 int migrate_vma(const struct migrate_vma_ops *ops,
 271                 struct vm_area_struct *vma,
 272                 unsigned long start,
 273                 unsigned long end,
 274                 unsigned long *src,
 275                 unsigned long *dst,
 276                 void *private);
 277 #else
 278 static inline int migrate_vma(const struct migrate_vma_ops *ops,
 279                               struct vm_area_struct *vma,
 280                               unsigned long start,
 281                               unsigned long end,
 282                               unsigned long *src,
 283                               unsigned long *dst,
 284                               void *private)
 285 {
 286         return -EINVAL;
 287 }
 288 #endif /* IS_ENABLED(CONFIG_MIGRATE_VMA_HELPER) */
 289
 290 #endif /* CONFIG_MIGRATION */
 291
 292 #endif /* _LINUX_MIGRATE_H */