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