xref: /linux-6.15/include/linux/mempolicy.h (revision bb66fc67)
1 /*
2  * NUMA memory policies for Linux.
3  * Copyright 2003,2004 Andi Kleen SuSE Labs
4  */
5 #ifndef _LINUX_MEMPOLICY_H
6 #define _LINUX_MEMPOLICY_H 1
7 
8 
9 #include <linux/mmzone.h>
10 #include <linux/slab.h>
11 #include <linux/rbtree.h>
12 #include <linux/spinlock.h>
13 #include <linux/nodemask.h>
14 #include <linux/pagemap.h>
15 #include <uapi/linux/mempolicy.h>
16 
17 struct mm_struct;
18 
19 #ifdef CONFIG_NUMA
20 
21 /*
22  * Describe a memory policy.
23  *
24  * A mempolicy can be either associated with a process or with a VMA.
25  * For VMA related allocations the VMA policy is preferred, otherwise
26  * the process policy is used. Interrupts ignore the memory policy
27  * of the current process.
28  *
29  * Locking policy for interlave:
30  * In process context there is no locking because only the process accesses
31  * its own state. All vma manipulation is somewhat protected by a down_read on
32  * mmap_sem.
33  *
34  * Freeing policy:
35  * Mempolicy objects are reference counted.  A mempolicy will be freed when
36  * mpol_put() decrements the reference count to zero.
37  *
38  * Duplicating policy objects:
39  * mpol_dup() allocates a new mempolicy and copies the specified mempolicy
40  * to the new storage.  The reference count of the new object is initialized
41  * to 1, representing the caller of mpol_dup().
42  */
43 struct mempolicy {
44 	atomic_t refcnt;
45 	unsigned short mode; 	/* See MPOL_* above */
46 	unsigned short flags;	/* See set_mempolicy() MPOL_F_* above */
47 	union {
48 		short 		 preferred_node; /* preferred */
49 		nodemask_t	 nodes;		/* interleave/bind */
50 		/* undefined for default */
51 	} v;
52 	union {
53 		nodemask_t cpuset_mems_allowed;	/* relative to these nodes */
54 		nodemask_t user_nodemask;	/* nodemask passed by user */
55 	} w;
56 };
57 
58 /*
59  * Support for managing mempolicy data objects (clone, copy, destroy)
60  * The default fast path of a NULL MPOL_DEFAULT policy is always inlined.
61  */
62 
63 extern void __mpol_put(struct mempolicy *pol);
64 static inline void mpol_put(struct mempolicy *pol)
65 {
66 	if (pol)
67 		__mpol_put(pol);
68 }
69 
70 /*
71  * Does mempolicy pol need explicit unref after use?
72  * Currently only needed for shared policies.
73  */
74 static inline int mpol_needs_cond_ref(struct mempolicy *pol)
75 {
76 	return (pol && (pol->flags & MPOL_F_SHARED));
77 }
78 
79 static inline void mpol_cond_put(struct mempolicy *pol)
80 {
81 	if (mpol_needs_cond_ref(pol))
82 		__mpol_put(pol);
83 }
84 
85 extern struct mempolicy *__mpol_dup(struct mempolicy *pol);
86 static inline struct mempolicy *mpol_dup(struct mempolicy *pol)
87 {
88 	if (pol)
89 		pol = __mpol_dup(pol);
90 	return pol;
91 }
92 
93 #define vma_policy(vma) ((vma)->vm_policy)
94 
95 static inline void mpol_get(struct mempolicy *pol)
96 {
97 	if (pol)
98 		atomic_inc(&pol->refcnt);
99 }
100 
101 extern bool __mpol_equal(struct mempolicy *a, struct mempolicy *b);
102 static inline bool mpol_equal(struct mempolicy *a, struct mempolicy *b)
103 {
104 	if (a == b)
105 		return true;
106 	return __mpol_equal(a, b);
107 }
108 
109 /*
110  * Tree of shared policies for a shared memory region.
111  * Maintain the policies in a pseudo mm that contains vmas. The vmas
112  * carry the policy. As a special twist the pseudo mm is indexed in pages, not
113  * bytes, so that we can work with shared memory segments bigger than
114  * unsigned long.
115  */
116 
117 struct sp_node {
118 	struct rb_node nd;
119 	unsigned long start, end;
120 	struct mempolicy *policy;
121 };
122 
123 struct shared_policy {
124 	struct rb_root root;
125 	spinlock_t lock;
126 };
127 
128 int vma_dup_policy(struct vm_area_struct *src, struct vm_area_struct *dst);
129 void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol);
130 int mpol_set_shared_policy(struct shared_policy *info,
131 				struct vm_area_struct *vma,
132 				struct mempolicy *new);
133 void mpol_free_shared_policy(struct shared_policy *p);
134 struct mempolicy *mpol_shared_policy_lookup(struct shared_policy *sp,
135 					    unsigned long idx);
136 
137 struct mempolicy *get_vma_policy(struct task_struct *tsk,
138 		struct vm_area_struct *vma, unsigned long addr);
139 bool vma_policy_mof(struct task_struct *task, struct vm_area_struct *vma);
140 
141 extern void numa_default_policy(void);
142 extern void numa_policy_init(void);
143 extern void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new,
144 				enum mpol_rebind_step step);
145 extern void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new);
146 
147 extern struct zonelist *huge_zonelist(struct vm_area_struct *vma,
148 				unsigned long addr, gfp_t gfp_flags,
149 				struct mempolicy **mpol, nodemask_t **nodemask);
150 extern bool init_nodemask_of_mempolicy(nodemask_t *mask);
151 extern bool mempolicy_nodemask_intersects(struct task_struct *tsk,
152 				const nodemask_t *mask);
153 extern unsigned int mempolicy_slab_node(void);
154 
155 extern enum zone_type policy_zone;
156 
157 static inline void check_highest_zone(enum zone_type k)
158 {
159 	if (k > policy_zone && k != ZONE_MOVABLE)
160 		policy_zone = k;
161 }
162 
163 int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
164 		     const nodemask_t *to, int flags);
165 
166 
167 #ifdef CONFIG_TMPFS
168 extern int mpol_parse_str(char *str, struct mempolicy **mpol);
169 #endif
170 
171 extern void mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol);
172 
173 /* Check if a vma is migratable */
174 static inline int vma_migratable(struct vm_area_struct *vma)
175 {
176 	if (vma->vm_flags & (VM_IO | VM_PFNMAP))
177 		return 0;
178 	/*
179 	 * Migration allocates pages in the highest zone. If we cannot
180 	 * do so then migration (at least from node to node) is not
181 	 * possible.
182 	 */
183 	if (vma->vm_file &&
184 		gfp_zone(mapping_gfp_mask(vma->vm_file->f_mapping))
185 								< policy_zone)
186 			return 0;
187 	return 1;
188 }
189 
190 extern int mpol_misplaced(struct page *, struct vm_area_struct *, unsigned long);
191 
192 #else
193 
194 struct mempolicy {};
195 
196 static inline bool mpol_equal(struct mempolicy *a, struct mempolicy *b)
197 {
198 	return true;
199 }
200 
201 static inline void mpol_put(struct mempolicy *p)
202 {
203 }
204 
205 static inline void mpol_cond_put(struct mempolicy *pol)
206 {
207 }
208 
209 static inline void mpol_get(struct mempolicy *pol)
210 {
211 }
212 
213 struct shared_policy {};
214 
215 static inline void mpol_shared_policy_init(struct shared_policy *sp,
216 						struct mempolicy *mpol)
217 {
218 }
219 
220 static inline void mpol_free_shared_policy(struct shared_policy *p)
221 {
222 }
223 
224 #define vma_policy(vma) NULL
225 
226 static inline int
227 vma_dup_policy(struct vm_area_struct *src, struct vm_area_struct *dst)
228 {
229 	return 0;
230 }
231 
232 static inline void numa_policy_init(void)
233 {
234 }
235 
236 static inline void numa_default_policy(void)
237 {
238 }
239 
240 static inline void mpol_rebind_task(struct task_struct *tsk,
241 				const nodemask_t *new,
242 				enum mpol_rebind_step step)
243 {
244 }
245 
246 static inline void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new)
247 {
248 }
249 
250 static inline struct zonelist *huge_zonelist(struct vm_area_struct *vma,
251 				unsigned long addr, gfp_t gfp_flags,
252 				struct mempolicy **mpol, nodemask_t **nodemask)
253 {
254 	*mpol = NULL;
255 	*nodemask = NULL;
256 	return node_zonelist(0, gfp_flags);
257 }
258 
259 static inline bool init_nodemask_of_mempolicy(nodemask_t *m)
260 {
261 	return false;
262 }
263 
264 static inline int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
265 				   const nodemask_t *to, int flags)
266 {
267 	return 0;
268 }
269 
270 static inline void check_highest_zone(int k)
271 {
272 }
273 
274 #ifdef CONFIG_TMPFS
275 static inline int mpol_parse_str(char *str, struct mempolicy **mpol)
276 {
277 	return 1;	/* error */
278 }
279 #endif
280 
281 static inline int mpol_misplaced(struct page *page, struct vm_area_struct *vma,
282 				 unsigned long address)
283 {
284 	return -1; /* no node preference */
285 }
286 
287 #endif /* CONFIG_NUMA */
288 #endif
289