xref: /linux-6.15/include/linux/migrate.h (revision bb4e9af0)
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 const 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 }
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, int extra_count);
81 #else
82 
83 static inline void putback_movable_pages(struct list_head *l) {}
84 static inline int migrate_pages(struct list_head *l, new_page_t new,
85 		free_page_t free, unsigned long private, enum migrate_mode mode,
86 		int reason)
87 	{ return -ENOSYS; }
88 static inline int isolate_movable_page(struct page *page, isolate_mode_t mode)
89 	{ return -EBUSY; }
90 
91 static inline int migrate_prep(void) { return -ENOSYS; }
92 static inline int migrate_prep_local(void) { return -ENOSYS; }
93 
94 static inline void migrate_page_states(struct page *newpage, struct page *page)
95 {
96 }
97 
98 static inline void migrate_page_copy(struct page *newpage,
99 				     struct page *page) {}
100 
101 static inline int migrate_huge_page_move_mapping(struct address_space *mapping,
102 				  struct page *newpage, struct page *page)
103 {
104 	return -ENOSYS;
105 }
106 
107 #endif /* CONFIG_MIGRATION */
108 
109 #ifdef CONFIG_COMPACTION
110 extern int PageMovable(struct page *page);
111 extern void __SetPageMovable(struct page *page, struct address_space *mapping);
112 extern void __ClearPageMovable(struct page *page);
113 #else
114 static inline int PageMovable(struct page *page) { return 0; };
115 static inline void __SetPageMovable(struct page *page,
116 				struct address_space *mapping)
117 {
118 }
119 static inline void __ClearPageMovable(struct page *page)
120 {
121 }
122 #endif
123 
124 #ifdef CONFIG_NUMA_BALANCING
125 extern bool pmd_trans_migrating(pmd_t pmd);
126 extern int migrate_misplaced_page(struct page *page,
127 				  struct vm_area_struct *vma, int node);
128 #else
129 static inline bool pmd_trans_migrating(pmd_t pmd)
130 {
131 	return false;
132 }
133 static inline int migrate_misplaced_page(struct page *page,
134 					 struct vm_area_struct *vma, int node)
135 {
136 	return -EAGAIN; /* can't migrate now */
137 }
138 #endif /* CONFIG_NUMA_BALANCING */
139 
140 #if defined(CONFIG_NUMA_BALANCING) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
141 extern int migrate_misplaced_transhuge_page(struct mm_struct *mm,
142 			struct vm_area_struct *vma,
143 			pmd_t *pmd, pmd_t entry,
144 			unsigned long address,
145 			struct page *page, int node);
146 #else
147 static inline int migrate_misplaced_transhuge_page(struct mm_struct *mm,
148 			struct vm_area_struct *vma,
149 			pmd_t *pmd, pmd_t entry,
150 			unsigned long address,
151 			struct page *page, int node)
152 {
153 	return -EAGAIN;
154 }
155 #endif /* CONFIG_NUMA_BALANCING && CONFIG_TRANSPARENT_HUGEPAGE*/
156 
157 
158 #ifdef CONFIG_MIGRATION
159 
160 /*
161  * Watch out for PAE architecture, which has an unsigned long, and might not
162  * have enough bits to store all physical address and flags. So far we have
163  * enough room for all our flags.
164  */
165 #define MIGRATE_PFN_VALID	(1UL << 0)
166 #define MIGRATE_PFN_MIGRATE	(1UL << 1)
167 #define MIGRATE_PFN_LOCKED	(1UL << 2)
168 #define MIGRATE_PFN_WRITE	(1UL << 3)
169 #define MIGRATE_PFN_DEVICE	(1UL << 4)
170 #define MIGRATE_PFN_ERROR	(1UL << 5)
171 #define MIGRATE_PFN_SHIFT	6
172 
173 static inline struct page *migrate_pfn_to_page(unsigned long mpfn)
174 {
175 	if (!(mpfn & MIGRATE_PFN_VALID))
176 		return NULL;
177 	return pfn_to_page(mpfn >> MIGRATE_PFN_SHIFT);
178 }
179 
180 static inline unsigned long migrate_pfn(unsigned long pfn)
181 {
182 	return (pfn << MIGRATE_PFN_SHIFT) | MIGRATE_PFN_VALID;
183 }
184 
185 /*
186  * struct migrate_vma_ops - migrate operation callback
187  *
188  * @alloc_and_copy: alloc destination memory and copy source memory to it
189  * @finalize_and_map: allow caller to map the successfully migrated pages
190  *
191  *
192  * The alloc_and_copy() callback happens once all source pages have been locked,
193  * unmapped and checked (checked whether pinned or not). All pages that can be
194  * migrated will have an entry in the src array set with the pfn value of the
195  * page and with the MIGRATE_PFN_VALID and MIGRATE_PFN_MIGRATE flag set (other
196  * flags might be set but should be ignored by the callback).
197  *
198  * The alloc_and_copy() callback can then allocate destination memory and copy
199  * source memory to it for all those entries (ie with MIGRATE_PFN_VALID and
200  * MIGRATE_PFN_MIGRATE flag set). Once these are allocated and copied, the
201  * callback must update each corresponding entry in the dst array with the pfn
202  * value of the destination page and with the MIGRATE_PFN_VALID and
203  * MIGRATE_PFN_LOCKED flags set (destination pages must have their struct pages
204  * locked, via lock_page()).
205  *
206  * At this point the alloc_and_copy() callback is done and returns.
207  *
208  * Note that the callback does not have to migrate all the pages that are
209  * marked with MIGRATE_PFN_MIGRATE flag in src array unless this is a migration
210  * from device memory to system memory (ie the MIGRATE_PFN_DEVICE flag is also
211  * set in the src array entry). If the device driver cannot migrate a device
212  * page back to system memory, then it must set the corresponding dst array
213  * entry to MIGRATE_PFN_ERROR. This will trigger a SIGBUS if CPU tries to
214  * access any of the virtual addresses originally backed by this page. Because
215  * a SIGBUS is such a severe result for the userspace process, the device
216  * driver should avoid setting MIGRATE_PFN_ERROR unless it is really in an
217  * unrecoverable state.
218  *
219  * For empty entry inside CPU page table (pte_none() or pmd_none() is true) we
220  * do set MIGRATE_PFN_MIGRATE flag inside the corresponding source array thus
221  * allowing device driver to allocate device memory for those unback virtual
222  * address. For this the device driver simply have to allocate device memory
223  * and properly set the destination entry like for regular migration. Note that
224  * this can still fails and thus inside the device driver must check if the
225  * migration was successful for those entry inside the finalize_and_map()
226  * callback just like for regular migration.
227  *
228  * THE alloc_and_copy() CALLBACK MUST NOT CHANGE ANY OF THE SRC ARRAY ENTRIES
229  * OR BAD THINGS WILL HAPPEN !
230  *
231  *
232  * The finalize_and_map() callback happens after struct page migration from
233  * source to destination (destination struct pages are the struct pages for the
234  * memory allocated by the alloc_and_copy() callback).  Migration can fail, and
235  * thus the finalize_and_map() allows the driver to inspect which pages were
236  * successfully migrated, and which were not. Successfully migrated pages will
237  * have the MIGRATE_PFN_MIGRATE flag set for their src array entry.
238  *
239  * It is safe to update device page table from within the finalize_and_map()
240  * callback because both destination and source page are still locked, and the
241  * mmap_sem is held in read mode (hence no one can unmap the range being
242  * migrated).
243  *
244  * Once callback is done cleaning up things and updating its page table (if it
245  * chose to do so, this is not an obligation) then it returns. At this point,
246  * the HMM core will finish up the final steps, and the migration is complete.
247  *
248  * THE finalize_and_map() CALLBACK MUST NOT CHANGE ANY OF THE SRC OR DST ARRAY
249  * ENTRIES OR BAD THINGS WILL HAPPEN !
250  */
251 struct migrate_vma_ops {
252 	void (*alloc_and_copy)(struct vm_area_struct *vma,
253 			       const unsigned long *src,
254 			       unsigned long *dst,
255 			       unsigned long start,
256 			       unsigned long end,
257 			       void *private);
258 	void (*finalize_and_map)(struct vm_area_struct *vma,
259 				 const unsigned long *src,
260 				 const unsigned long *dst,
261 				 unsigned long start,
262 				 unsigned long end,
263 				 void *private);
264 };
265 
266 #if defined(CONFIG_MIGRATE_VMA_HELPER)
267 int migrate_vma(const struct migrate_vma_ops *ops,
268 		struct vm_area_struct *vma,
269 		unsigned long start,
270 		unsigned long end,
271 		unsigned long *src,
272 		unsigned long *dst,
273 		void *private);
274 #else
275 static inline int migrate_vma(const struct migrate_vma_ops *ops,
276 			      struct vm_area_struct *vma,
277 			      unsigned long start,
278 			      unsigned long end,
279 			      unsigned long *src,
280 			      unsigned long *dst,
281 			      void *private)
282 {
283 	return -EINVAL;
284 }
285 #endif /* IS_ENABLED(CONFIG_MIGRATE_VMA_HELPER) */
286 
287 #endif /* CONFIG_MIGRATION */
288 
289 #endif /* _LINUX_MIGRATE_H */
290