Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | #ifndef _LINUX_MEMPOLICY_H |
2 | #define _LINUX_MEMPOLICY_H 1 | |
3 | ||
4 | #include <linux/errno.h> | |
5 | ||
6 | /* | |
7 | * NUMA memory policies for Linux. | |
8 | * Copyright 2003,2004 Andi Kleen SuSE Labs | |
9 | */ | |
10 | ||
028fec41 DR |
11 | /* |
12 | * Both the MPOL_* mempolicy mode and the MPOL_F_* optional mode flags are | |
13 | * passed by the user to either set_mempolicy() or mbind() in an 'int' actual. | |
14 | * The MPOL_MODE_FLAGS macro determines the legal set of optional mode flags. | |
15 | */ | |
16 | ||
1da177e4 | 17 | /* Policies */ |
a3b51e01 DR |
18 | enum { |
19 | MPOL_DEFAULT, | |
20 | MPOL_PREFERRED, | |
21 | MPOL_BIND, | |
22 | MPOL_INTERLEAVE, | |
23 | MPOL_MAX, /* always last member of enum */ | |
24 | }; | |
1da177e4 | 25 | |
028fec41 | 26 | /* Flags for set_mempolicy */ |
f5b087b5 | 27 | #define MPOL_F_STATIC_NODES (1 << 15) |
4c50bc01 | 28 | #define MPOL_F_RELATIVE_NODES (1 << 14) |
f5b087b5 | 29 | |
028fec41 DR |
30 | /* |
31 | * MPOL_MODE_FLAGS is the union of all possible optional mode flags passed to | |
32 | * either set_mempolicy() or mbind(). | |
33 | */ | |
4c50bc01 | 34 | #define MPOL_MODE_FLAGS (MPOL_F_STATIC_NODES | MPOL_F_RELATIVE_NODES) |
028fec41 DR |
35 | |
36 | /* Flags for get_mempolicy */ | |
1da177e4 LT |
37 | #define MPOL_F_NODE (1<<0) /* return next IL mode instead of node mask */ |
38 | #define MPOL_F_ADDR (1<<1) /* look up vma using address */ | |
754af6f5 | 39 | #define MPOL_F_MEMS_ALLOWED (1<<2) /* return allowed memories */ |
1da177e4 LT |
40 | |
41 | /* Flags for mbind */ | |
42 | #define MPOL_MF_STRICT (1<<0) /* Verify existing pages in the mapping */ | |
dc9aa5b9 CL |
43 | #define MPOL_MF_MOVE (1<<1) /* Move pages owned by this process to conform to mapping */ |
44 | #define MPOL_MF_MOVE_ALL (1<<2) /* Move every page to conform to mapping */ | |
45 | #define MPOL_MF_INTERNAL (1<<3) /* Internal flags start here */ | |
1da177e4 | 46 | |
aab0b102 LS |
47 | /* |
48 | * Internal flags that share the struct mempolicy flags word with | |
49 | * "mode flags". These flags are allocated from bit 0 up, as they | |
50 | * are never OR'ed into the mode in mempolicy API arguments. | |
51 | */ | |
52 | #define MPOL_F_SHARED (1 << 0) /* identify shared policies */ | |
fc36b8d3 | 53 | #define MPOL_F_LOCAL (1 << 1) /* preferred local allocation */ |
aab0b102 | 54 | |
1da177e4 LT |
55 | #ifdef __KERNEL__ |
56 | ||
1da177e4 | 57 | #include <linux/mmzone.h> |
1da177e4 LT |
58 | #include <linux/slab.h> |
59 | #include <linux/rbtree.h> | |
60 | #include <linux/spinlock.h> | |
dfcd3c0d | 61 | #include <linux/nodemask.h> |
83d1674a | 62 | #include <linux/pagemap.h> |
1da177e4 | 63 | |
45b35a5c | 64 | struct mm_struct; |
1da177e4 LT |
65 | |
66 | #ifdef CONFIG_NUMA | |
67 | ||
68 | /* | |
69 | * Describe a memory policy. | |
70 | * | |
71 | * A mempolicy can be either associated with a process or with a VMA. | |
72 | * For VMA related allocations the VMA policy is preferred, otherwise | |
73 | * the process policy is used. Interrupts ignore the memory policy | |
74 | * of the current process. | |
75 | * | |
76 | * Locking policy for interlave: | |
77 | * In process context there is no locking because only the process accesses | |
78 | * its own state. All vma manipulation is somewhat protected by a down_read on | |
b8072f09 | 79 | * mmap_sem. |
1da177e4 LT |
80 | * |
81 | * Freeing policy: | |
19770b32 | 82 | * Mempolicy objects are reference counted. A mempolicy will be freed when |
f0be3d32 | 83 | * mpol_put() decrements the reference count to zero. |
1da177e4 | 84 | * |
846a16bf LS |
85 | * Duplicating policy objects: |
86 | * mpol_dup() allocates a new mempolicy and copies the specified mempolicy | |
19770b32 | 87 | * to the new storage. The reference count of the new object is initialized |
846a16bf | 88 | * to 1, representing the caller of mpol_dup(). |
1da177e4 LT |
89 | */ |
90 | struct mempolicy { | |
91 | atomic_t refcnt; | |
45c4745a | 92 | unsigned short mode; /* See MPOL_* above */ |
028fec41 | 93 | unsigned short flags; /* See set_mempolicy() MPOL_F_* above */ |
1da177e4 | 94 | union { |
1da177e4 | 95 | short preferred_node; /* preferred */ |
19770b32 | 96 | nodemask_t nodes; /* interleave/bind */ |
1da177e4 LT |
97 | /* undefined for default */ |
98 | } v; | |
f5b087b5 DR |
99 | union { |
100 | nodemask_t cpuset_mems_allowed; /* relative to these nodes */ | |
101 | nodemask_t user_nodemask; /* nodemask passed by user */ | |
102 | } w; | |
1da177e4 LT |
103 | }; |
104 | ||
105 | /* | |
106 | * Support for managing mempolicy data objects (clone, copy, destroy) | |
107 | * The default fast path of a NULL MPOL_DEFAULT policy is always inlined. | |
108 | */ | |
109 | ||
f0be3d32 LS |
110 | extern void __mpol_put(struct mempolicy *pol); |
111 | static inline void mpol_put(struct mempolicy *pol) | |
1da177e4 LT |
112 | { |
113 | if (pol) | |
f0be3d32 | 114 | __mpol_put(pol); |
1da177e4 LT |
115 | } |
116 | ||
52cd3b07 LS |
117 | /* |
118 | * Does mempolicy pol need explicit unref after use? | |
119 | * Currently only needed for shared policies. | |
120 | */ | |
121 | static inline int mpol_needs_cond_ref(struct mempolicy *pol) | |
122 | { | |
123 | return (pol && (pol->flags & MPOL_F_SHARED)); | |
124 | } | |
125 | ||
126 | static inline void mpol_cond_put(struct mempolicy *pol) | |
127 | { | |
128 | if (mpol_needs_cond_ref(pol)) | |
129 | __mpol_put(pol); | |
130 | } | |
131 | ||
132 | extern struct mempolicy *__mpol_cond_copy(struct mempolicy *tompol, | |
133 | struct mempolicy *frompol); | |
134 | static inline struct mempolicy *mpol_cond_copy(struct mempolicy *tompol, | |
135 | struct mempolicy *frompol) | |
136 | { | |
137 | if (!frompol) | |
138 | return frompol; | |
139 | return __mpol_cond_copy(tompol, frompol); | |
140 | } | |
141 | ||
846a16bf LS |
142 | extern struct mempolicy *__mpol_dup(struct mempolicy *pol); |
143 | static inline struct mempolicy *mpol_dup(struct mempolicy *pol) | |
1da177e4 LT |
144 | { |
145 | if (pol) | |
846a16bf | 146 | pol = __mpol_dup(pol); |
1da177e4 LT |
147 | return pol; |
148 | } | |
149 | ||
150 | #define vma_policy(vma) ((vma)->vm_policy) | |
151 | #define vma_set_policy(vma, pol) ((vma)->vm_policy = (pol)) | |
152 | ||
153 | static inline void mpol_get(struct mempolicy *pol) | |
154 | { | |
155 | if (pol) | |
156 | atomic_inc(&pol->refcnt); | |
157 | } | |
158 | ||
159 | extern int __mpol_equal(struct mempolicy *a, struct mempolicy *b); | |
160 | static inline int mpol_equal(struct mempolicy *a, struct mempolicy *b) | |
161 | { | |
162 | if (a == b) | |
163 | return 1; | |
164 | return __mpol_equal(a, b); | |
165 | } | |
1da177e4 | 166 | |
1da177e4 LT |
167 | /* |
168 | * Tree of shared policies for a shared memory region. | |
169 | * Maintain the policies in a pseudo mm that contains vmas. The vmas | |
170 | * carry the policy. As a special twist the pseudo mm is indexed in pages, not | |
171 | * bytes, so that we can work with shared memory segments bigger than | |
172 | * unsigned long. | |
173 | */ | |
174 | ||
175 | struct sp_node { | |
176 | struct rb_node nd; | |
177 | unsigned long start, end; | |
178 | struct mempolicy *policy; | |
179 | }; | |
180 | ||
181 | struct shared_policy { | |
182 | struct rb_root root; | |
183 | spinlock_t lock; | |
184 | }; | |
185 | ||
71fe804b | 186 | void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol); |
1da177e4 LT |
187 | int mpol_set_shared_policy(struct shared_policy *info, |
188 | struct vm_area_struct *vma, | |
189 | struct mempolicy *new); | |
190 | void mpol_free_shared_policy(struct shared_policy *p); | |
191 | struct mempolicy *mpol_shared_policy_lookup(struct shared_policy *sp, | |
192 | unsigned long idx); | |
193 | ||
194 | extern void numa_default_policy(void); | |
195 | extern void numa_policy_init(void); | |
74cb2155 PJ |
196 | extern void mpol_rebind_task(struct task_struct *tsk, |
197 | const nodemask_t *new); | |
4225399a | 198 | extern void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new); |
c61afb18 | 199 | extern void mpol_fix_fork_child_flag(struct task_struct *p); |
4225399a | 200 | |
5da7ca86 | 201 | extern struct zonelist *huge_zonelist(struct vm_area_struct *vma, |
19770b32 MG |
202 | unsigned long addr, gfp_t gfp_flags, |
203 | struct mempolicy **mpol, nodemask_t **nodemask); | |
06808b08 | 204 | extern bool init_nodemask_of_mempolicy(nodemask_t *mask); |
dc85da15 | 205 | extern unsigned slab_node(struct mempolicy *policy); |
1da177e4 | 206 | |
2f6726e5 | 207 | extern enum zone_type policy_zone; |
4be38e35 | 208 | |
2f6726e5 | 209 | static inline void check_highest_zone(enum zone_type k) |
4be38e35 | 210 | { |
b377fd39 | 211 | if (k > policy_zone && k != ZONE_MOVABLE) |
4be38e35 CL |
212 | policy_zone = k; |
213 | } | |
214 | ||
39743889 CL |
215 | int do_migrate_pages(struct mm_struct *mm, |
216 | const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags); | |
217 | ||
095f1fc4 LS |
218 | |
219 | #ifdef CONFIG_TMPFS | |
71fe804b | 220 | extern int mpol_parse_str(char *str, struct mempolicy **mpol, int no_context); |
095f1fc4 | 221 | |
71fe804b LS |
222 | extern int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol, |
223 | int no_context); | |
095f1fc4 | 224 | #endif |
83d1674a GS |
225 | |
226 | /* Check if a vma is migratable */ | |
227 | static inline int vma_migratable(struct vm_area_struct *vma) | |
228 | { | |
229 | if (vma->vm_flags & (VM_IO|VM_HUGETLB|VM_PFNMAP|VM_RESERVED)) | |
230 | return 0; | |
231 | /* | |
232 | * Migration allocates pages in the highest zone. If we cannot | |
233 | * do so then migration (at least from node to node) is not | |
234 | * possible. | |
235 | */ | |
236 | if (vma->vm_file && | |
237 | gfp_zone(mapping_gfp_mask(vma->vm_file->f_mapping)) | |
238 | < policy_zone) | |
239 | return 0; | |
240 | return 1; | |
241 | } | |
242 | ||
1da177e4 LT |
243 | #else |
244 | ||
245 | struct mempolicy {}; | |
246 | ||
247 | static inline int mpol_equal(struct mempolicy *a, struct mempolicy *b) | |
248 | { | |
249 | return 1; | |
250 | } | |
1da177e4 | 251 | |
f0be3d32 | 252 | static inline void mpol_put(struct mempolicy *p) |
1da177e4 LT |
253 | { |
254 | } | |
255 | ||
52cd3b07 LS |
256 | static inline void mpol_cond_put(struct mempolicy *pol) |
257 | { | |
258 | } | |
259 | ||
260 | static inline struct mempolicy *mpol_cond_copy(struct mempolicy *to, | |
261 | struct mempolicy *from) | |
262 | { | |
263 | return from; | |
264 | } | |
265 | ||
1da177e4 LT |
266 | static inline void mpol_get(struct mempolicy *pol) |
267 | { | |
268 | } | |
269 | ||
846a16bf | 270 | static inline struct mempolicy *mpol_dup(struct mempolicy *old) |
1da177e4 LT |
271 | { |
272 | return NULL; | |
273 | } | |
274 | ||
1da177e4 LT |
275 | struct shared_policy {}; |
276 | ||
277 | static inline int mpol_set_shared_policy(struct shared_policy *info, | |
278 | struct vm_area_struct *vma, | |
279 | struct mempolicy *new) | |
280 | { | |
281 | return -EINVAL; | |
282 | } | |
283 | ||
71fe804b LS |
284 | static inline void mpol_shared_policy_init(struct shared_policy *sp, |
285 | struct mempolicy *mpol) | |
1da177e4 LT |
286 | { |
287 | } | |
288 | ||
289 | static inline void mpol_free_shared_policy(struct shared_policy *p) | |
290 | { | |
291 | } | |
292 | ||
293 | static inline struct mempolicy * | |
294 | mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx) | |
295 | { | |
296 | return NULL; | |
297 | } | |
298 | ||
299 | #define vma_policy(vma) NULL | |
300 | #define vma_set_policy(vma, pol) do {} while(0) | |
301 | ||
302 | static inline void numa_policy_init(void) | |
303 | { | |
304 | } | |
305 | ||
306 | static inline void numa_default_policy(void) | |
307 | { | |
308 | } | |
309 | ||
74cb2155 | 310 | static inline void mpol_rebind_task(struct task_struct *tsk, |
68860ec1 PJ |
311 | const nodemask_t *new) |
312 | { | |
313 | } | |
314 | ||
4225399a PJ |
315 | static inline void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new) |
316 | { | |
317 | } | |
318 | ||
c61afb18 PJ |
319 | static inline void mpol_fix_fork_child_flag(struct task_struct *p) |
320 | { | |
321 | } | |
322 | ||
5da7ca86 | 323 | static inline struct zonelist *huge_zonelist(struct vm_area_struct *vma, |
19770b32 MG |
324 | unsigned long addr, gfp_t gfp_flags, |
325 | struct mempolicy **mpol, nodemask_t **nodemask) | |
5da7ca86 | 326 | { |
19770b32 MG |
327 | *mpol = NULL; |
328 | *nodemask = NULL; | |
0e88460d | 329 | return node_zonelist(0, gfp_flags); |
5da7ca86 CL |
330 | } |
331 | ||
06808b08 LS |
332 | static inline bool init_nodemask_of_mempolicy(nodemask_t *m) { return false; } |
333 | ||
45b07ef3 PJ |
334 | static inline int do_migrate_pages(struct mm_struct *mm, |
335 | const nodemask_t *from_nodes, | |
336 | const nodemask_t *to_nodes, int flags) | |
337 | { | |
338 | return 0; | |
339 | } | |
340 | ||
4be38e35 CL |
341 | static inline void check_highest_zone(int k) |
342 | { | |
343 | } | |
095f1fc4 LS |
344 | |
345 | #ifdef CONFIG_TMPFS | |
71fe804b LS |
346 | static inline int mpol_parse_str(char *str, struct mempolicy **mpol, |
347 | int no_context) | |
095f1fc4 | 348 | { |
71fe804b | 349 | return 1; /* error */ |
095f1fc4 LS |
350 | } |
351 | ||
71fe804b LS |
352 | static inline int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol, |
353 | int no_context) | |
095f1fc4 LS |
354 | { |
355 | return 0; | |
356 | } | |
357 | #endif | |
358 | ||
1da177e4 LT |
359 | #endif /* CONFIG_NUMA */ |
360 | #endif /* __KERNEL__ */ | |
361 | ||
362 | #endif |