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 typedef void free_page_t(struct page *page, unsigned long private);
12
13 /*
14 * Return values from addresss_space_operations.migratepage():
15 * - negative errno on page migration failure;
16 * - zero on page migration success;
17 */
18 #define MIGRATEPAGE_SUCCESS 0
19
20 enum migrate_reason {
21 MR_COMPACTION,
22 MR_MEMORY_FAILURE,
23 MR_MEMORY_HOTPLUG,
24 MR_SYSCALL, /* also applies to cpusets */
25 MR_MEMPOLICY_MBIND,
26 MR_NUMA_MISPLACED,
27 MR_CONTIG_RANGE,
28 MR_TYPES
29 };
30
31 /* In mm/debug.c; also keep sync with include/trace/events/migrate.h */
32 extern char *migrate_reason_names[MR_TYPES];
33
new_page_nodemask(struct page * page,int preferred_nid,nodemask_t * nodemask)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 }
61
62 #ifdef CONFIG_MIGRATION
63
64 extern void putback_movable_pages(struct list_head *l);
65 extern int migrate_page(struct address_space *mapping,
66 struct page *newpage, struct page *page,
67 enum migrate_mode mode);
68 extern int migrate_pages(struct list_head *l, new_page_t new, free_page_t free,
69 unsigned long private, enum migrate_mode mode, int reason);
70 extern int isolate_movable_page(struct page *page, isolate_mode_t mode);
71 extern void putback_movable_page(struct page *page);
72
73 extern int migrate_prep(void);
74 extern int migrate_prep_local(void);
75 extern void migrate_page_states(struct page *newpage, struct page *page);
76 extern void migrate_page_copy(struct page *newpage, struct page *page);
77 extern int migrate_huge_page_move_mapping(struct address_space *mapping,
78 struct page *newpage, struct page *page);
79 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);
83 #else
84
putback_movable_pages(struct list_head * l)85 static inline void putback_movable_pages(struct list_head *l) {}
migrate_pages(struct list_head * l,new_page_t new,free_page_t free,unsigned long private,enum migrate_mode mode,int reason)86 static inline int migrate_pages(struct list_head *l, new_page_t new,
87 free_page_t free, unsigned long private, enum migrate_mode mode,
88 int reason)
89 { return -ENOSYS; }
isolate_movable_page(struct page * page,isolate_mode_t mode)90 static inline int isolate_movable_page(struct page *page, isolate_mode_t mode)
91 { return -EBUSY; }
92
migrate_prep(void)93 static inline int migrate_prep(void) { return -ENOSYS; }
migrate_prep_local(void)94 static inline int migrate_prep_local(void) { return -ENOSYS; }
95
migrate_page_states(struct page * newpage,struct page * page)96 static inline void migrate_page_states(struct page *newpage, struct page *page)
97 {
98 }
99
migrate_page_copy(struct page * newpage,struct page * page)100 static inline void migrate_page_copy(struct page *newpage,
101 struct page *page) {}
102
migrate_huge_page_move_mapping(struct address_space * mapping,struct page * newpage,struct page * page)103 static inline int migrate_huge_page_move_mapping(struct address_space *mapping,
104 struct page *newpage, struct page *page)
105 {
106 return -ENOSYS;
107 }
108
109 #endif /* CONFIG_MIGRATION */
110
111 #ifdef CONFIG_COMPACTION
112 extern int PageMovable(struct page *page);
113 extern void __SetPageMovable(struct page *page, struct address_space *mapping);
114 extern void __ClearPageMovable(struct page *page);
115 #else
PageMovable(struct page * page)116 static inline int PageMovable(struct page *page) { return 0; };
__SetPageMovable(struct page * page,struct address_space * mapping)117 static inline void __SetPageMovable(struct page *page,
118 struct address_space *mapping)
119 {
120 }
__ClearPageMovable(struct page * page)121 static inline void __ClearPageMovable(struct page *page)
122 {
123 }
124 #endif
125
126 #ifdef CONFIG_NUMA_BALANCING
127 extern bool pmd_trans_migrating(pmd_t pmd);
128 extern int migrate_misplaced_page(struct page *page,
129 struct vm_area_struct *vma, int node);
130 #else
pmd_trans_migrating(pmd_t pmd)131 static inline bool pmd_trans_migrating(pmd_t pmd)
132 {
133 return false;
134 }
migrate_misplaced_page(struct page * page,struct vm_area_struct * vma,int node)135 static inline int migrate_misplaced_page(struct page *page,
136 struct vm_area_struct *vma, int node)
137 {
138 return -EAGAIN; /* can't migrate now */
139 }
140 #endif /* CONFIG_NUMA_BALANCING */
141
142 #if defined(CONFIG_NUMA_BALANCING) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
143 extern int migrate_misplaced_transhuge_page(struct mm_struct *mm,
144 struct vm_area_struct *vma,
145 pmd_t *pmd, pmd_t entry,
146 unsigned long address,
147 struct page *page, int node);
148 #else
migrate_misplaced_transhuge_page(struct mm_struct * mm,struct vm_area_struct * vma,pmd_t * pmd,pmd_t entry,unsigned long address,struct page * page,int node)149 static inline int migrate_misplaced_transhuge_page(struct mm_struct *mm,
150 struct vm_area_struct *vma,
151 pmd_t *pmd, pmd_t entry,
152 unsigned long address,
153 struct page *page, int node)
154 {
155 return -EAGAIN;
156 }
157 #endif /* CONFIG_NUMA_BALANCING && CONFIG_TRANSPARENT_HUGEPAGE*/
158
159
160 #ifdef CONFIG_MIGRATION
161
162 /*
163 * Watch out for PAE architecture, which has an unsigned long, and might not
164 * have enough bits to store all physical address and flags. So far we have
165 * enough room for all our flags.
166 */
167 #define MIGRATE_PFN_VALID (1UL << 0)
168 #define MIGRATE_PFN_MIGRATE (1UL << 1)
169 #define MIGRATE_PFN_LOCKED (1UL << 2)
170 #define MIGRATE_PFN_WRITE (1UL << 3)
171 #define MIGRATE_PFN_DEVICE (1UL << 4)
172 #define MIGRATE_PFN_ERROR (1UL << 5)
173 #define MIGRATE_PFN_SHIFT 6
174
migrate_pfn_to_page(unsigned long mpfn)175 static inline struct page *migrate_pfn_to_page(unsigned long mpfn)
176 {
177 if (!(mpfn & MIGRATE_PFN_VALID))
178 return NULL;
179 return pfn_to_page(mpfn >> MIGRATE_PFN_SHIFT);
180 }
181
migrate_pfn(unsigned long pfn)182 static inline unsigned long migrate_pfn(unsigned long pfn)
183 {
184 return (pfn << MIGRATE_PFN_SHIFT) | MIGRATE_PFN_VALID;
185 }
186
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);
266 };
267
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
migrate_vma(const struct migrate_vma_ops * ops,struct vm_area_struct * vma,unsigned long start,unsigned long end,unsigned long * src,unsigned long * dst,void * private)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) */
288
289 #endif /* CONFIG_MIGRATION */
290
291 #endif /* _LINUX_MIGRATE_H */
292