blob: cc49b232b6e424d80b865fffd88ad61681260887 [file] [log] [blame]
lh9ed821d2023-04-07 01:36:19 -07001#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
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
17/* Policies */
18enum {
19 MPOL_DEFAULT,
20 MPOL_PREFERRED,
21 MPOL_BIND,
22 MPOL_INTERLEAVE,
23 MPOL_MAX, /* always last member of enum */
24};
25
26enum mpol_rebind_step {
27 MPOL_REBIND_ONCE, /* do rebind work at once(not by two step) */
28 MPOL_REBIND_STEP1, /* first step(set all the newly nodes) */
29 MPOL_REBIND_STEP2, /* second step(clean all the disallowed nodes)*/
30 MPOL_REBIND_NSTEP,
31};
32
33/* Flags for set_mempolicy */
34#define MPOL_F_STATIC_NODES (1 << 15)
35#define MPOL_F_RELATIVE_NODES (1 << 14)
36
37/*
38 * MPOL_MODE_FLAGS is the union of all possible optional mode flags passed to
39 * either set_mempolicy() or mbind().
40 */
41#define MPOL_MODE_FLAGS (MPOL_F_STATIC_NODES | MPOL_F_RELATIVE_NODES)
42
43/* Flags for get_mempolicy */
44#define MPOL_F_NODE (1<<0) /* return next IL mode instead of node mask */
45#define MPOL_F_ADDR (1<<1) /* look up vma using address */
46#define MPOL_F_MEMS_ALLOWED (1<<2) /* return allowed memories */
47
48/* Flags for mbind */
49#define MPOL_MF_STRICT (1<<0) /* Verify existing pages in the mapping */
50#define MPOL_MF_MOVE (1<<1) /* Move pages owned by this process to conform to mapping */
51#define MPOL_MF_MOVE_ALL (1<<2) /* Move every page to conform to mapping */
52#define MPOL_MF_INTERNAL (1<<3) /* Internal flags start here */
53
54/*
55 * Internal flags that share the struct mempolicy flags word with
56 * "mode flags". These flags are allocated from bit 0 up, as they
57 * are never OR'ed into the mode in mempolicy API arguments.
58 */
59#define MPOL_F_SHARED (1 << 0) /* identify shared policies */
60#define MPOL_F_LOCAL (1 << 1) /* preferred local allocation */
61#define MPOL_F_REBINDING (1 << 2) /* identify policies in rebinding */
62
63#ifdef __KERNEL__
64
65#include <linux/mmzone.h>
66#include <linux/slab.h>
67#include <linux/rbtree.h>
68#include <linux/spinlock.h>
69#include <linux/nodemask.h>
70#include <linux/pagemap.h>
71
72struct mm_struct;
73
74#ifdef CONFIG_NUMA
75
76/*
77 * Describe a memory policy.
78 *
79 * A mempolicy can be either associated with a process or with a VMA.
80 * For VMA related allocations the VMA policy is preferred, otherwise
81 * the process policy is used. Interrupts ignore the memory policy
82 * of the current process.
83 *
84 * Locking policy for interlave:
85 * In process context there is no locking because only the process accesses
86 * its own state. All vma manipulation is somewhat protected by a down_read on
87 * mmap_sem.
88 *
89 * Freeing policy:
90 * Mempolicy objects are reference counted. A mempolicy will be freed when
91 * mpol_put() decrements the reference count to zero.
92 *
93 * Duplicating policy objects:
94 * mpol_dup() allocates a new mempolicy and copies the specified mempolicy
95 * to the new storage. The reference count of the new object is initialized
96 * to 1, representing the caller of mpol_dup().
97 */
98struct mempolicy {
99 atomic_t refcnt;
100 unsigned short mode; /* See MPOL_* above */
101 unsigned short flags; /* See set_mempolicy() MPOL_F_* above */
102 union {
103 short preferred_node; /* preferred */
104 nodemask_t nodes; /* interleave/bind */
105 /* undefined for default */
106 } v;
107 union {
108 nodemask_t cpuset_mems_allowed; /* relative to these nodes */
109 nodemask_t user_nodemask; /* nodemask passed by user */
110 } w;
111};
112
113/*
114 * Support for managing mempolicy data objects (clone, copy, destroy)
115 * The default fast path of a NULL MPOL_DEFAULT policy is always inlined.
116 */
117
118extern void __mpol_put(struct mempolicy *pol);
119static inline void mpol_put(struct mempolicy *pol)
120{
121 if (pol)
122 __mpol_put(pol);
123}
124
125/*
126 * Does mempolicy pol need explicit unref after use?
127 * Currently only needed for shared policies.
128 */
129static inline int mpol_needs_cond_ref(struct mempolicy *pol)
130{
131 return (pol && (pol->flags & MPOL_F_SHARED));
132}
133
134static inline void mpol_cond_put(struct mempolicy *pol)
135{
136 if (mpol_needs_cond_ref(pol))
137 __mpol_put(pol);
138}
139
140extern struct mempolicy *__mpol_dup(struct mempolicy *pol);
141static inline struct mempolicy *mpol_dup(struct mempolicy *pol)
142{
143 if (pol)
144 pol = __mpol_dup(pol);
145 return pol;
146}
147
148#define vma_policy(vma) ((vma)->vm_policy)
149#define vma_set_policy(vma, pol) ((vma)->vm_policy = (pol))
150
151static inline void mpol_get(struct mempolicy *pol)
152{
153 if (pol)
154 atomic_inc(&pol->refcnt);
155}
156
157extern bool __mpol_equal(struct mempolicy *a, struct mempolicy *b);
158static inline bool mpol_equal(struct mempolicy *a, struct mempolicy *b)
159{
160 if (a == b)
161 return true;
162 return __mpol_equal(a, b);
163}
164
165/*
166 * Tree of shared policies for a shared memory region.
167 * Maintain the policies in a pseudo mm that contains vmas. The vmas
168 * carry the policy. As a special twist the pseudo mm is indexed in pages, not
169 * bytes, so that we can work with shared memory segments bigger than
170 * unsigned long.
171 */
172
173struct sp_node {
174 struct rb_node nd;
175 unsigned long start, end;
176 struct mempolicy *policy;
177};
178
179struct shared_policy {
180 struct rb_root root;
181 struct mutex mutex;
182};
183
184void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol);
185int mpol_set_shared_policy(struct shared_policy *info,
186 struct vm_area_struct *vma,
187 struct mempolicy *new);
188void mpol_free_shared_policy(struct shared_policy *p);
189struct mempolicy *mpol_shared_policy_lookup(struct shared_policy *sp,
190 unsigned long idx);
191
192struct mempolicy *get_vma_policy(struct task_struct *tsk,
193 struct vm_area_struct *vma, unsigned long addr);
194
195extern void numa_default_policy(void);
196extern void numa_policy_init(void);
197extern void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new,
198 enum mpol_rebind_step step);
199extern void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new);
200extern void mpol_fix_fork_child_flag(struct task_struct *p);
201
202extern struct zonelist *huge_zonelist(struct vm_area_struct *vma,
203 unsigned long addr, gfp_t gfp_flags,
204 struct mempolicy **mpol, nodemask_t **nodemask);
205extern bool init_nodemask_of_mempolicy(nodemask_t *mask);
206extern bool mempolicy_nodemask_intersects(struct task_struct *tsk,
207 const nodemask_t *mask);
208extern unsigned slab_node(void);
209
210extern enum zone_type policy_zone;
211
212static inline void check_highest_zone(enum zone_type k)
213{
214 if (k > policy_zone && k != ZONE_MOVABLE)
215 policy_zone = k;
216}
217
218int do_migrate_pages(struct mm_struct *mm,
219 const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags);
220
221
222#ifdef CONFIG_TMPFS
223extern int mpol_parse_str(char *str, struct mempolicy **mpol, int no_context);
224#endif
225
226extern int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol,
227 int no_context);
228
229/* Check if a vma is migratable */
230static inline int vma_migratable(struct vm_area_struct *vma)
231{
232 if (vma->vm_flags & (VM_IO|VM_HUGETLB|VM_PFNMAP|VM_RESERVED))
233 return 0;
234 /*
235 * Migration allocates pages in the highest zone. If we cannot
236 * do so then migration (at least from node to node) is not
237 * possible.
238 */
239 if (vma->vm_file &&
240 gfp_zone(mapping_gfp_mask(vma->vm_file->f_mapping))
241 < policy_zone)
242 return 0;
243 return 1;
244}
245
246#else
247
248struct mempolicy {};
249
250static inline bool mpol_equal(struct mempolicy *a, struct mempolicy *b)
251{
252 return true;
253}
254
255static inline void mpol_put(struct mempolicy *p)
256{
257}
258
259static inline void mpol_cond_put(struct mempolicy *pol)
260{
261}
262
263static inline void mpol_get(struct mempolicy *pol)
264{
265}
266
267static inline struct mempolicy *mpol_dup(struct mempolicy *old)
268{
269 return NULL;
270}
271
272struct shared_policy {};
273
274static inline int mpol_set_shared_policy(struct shared_policy *info,
275 struct vm_area_struct *vma,
276 struct mempolicy *new)
277{
278 return -EINVAL;
279}
280
281static inline void mpol_shared_policy_init(struct shared_policy *sp,
282 struct mempolicy *mpol)
283{
284}
285
286static inline void mpol_free_shared_policy(struct shared_policy *p)
287{
288}
289
290static inline struct mempolicy *
291mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx)
292{
293 return NULL;
294}
295
296#define vma_policy(vma) NULL
297#define vma_set_policy(vma, pol) do {} while(0)
298
299static inline void numa_policy_init(void)
300{
301}
302
303static inline void numa_default_policy(void)
304{
305}
306
307static inline void mpol_rebind_task(struct task_struct *tsk,
308 const nodemask_t *new,
309 enum mpol_rebind_step step)
310{
311}
312
313static inline void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new)
314{
315}
316
317static inline void mpol_fix_fork_child_flag(struct task_struct *p)
318{
319}
320
321static inline struct zonelist *huge_zonelist(struct vm_area_struct *vma,
322 unsigned long addr, gfp_t gfp_flags,
323 struct mempolicy **mpol, nodemask_t **nodemask)
324{
325 *mpol = NULL;
326 *nodemask = NULL;
327 return node_zonelist(0, gfp_flags);
328}
329
330static inline bool init_nodemask_of_mempolicy(nodemask_t *m)
331{
332 return false;
333}
334
335static inline bool mempolicy_nodemask_intersects(struct task_struct *tsk,
336 const nodemask_t *mask)
337{
338 return false;
339}
340
341static inline int do_migrate_pages(struct mm_struct *mm,
342 const nodemask_t *from_nodes,
343 const nodemask_t *to_nodes, int flags)
344{
345 return 0;
346}
347
348static inline void check_highest_zone(int k)
349{
350}
351
352#ifdef CONFIG_TMPFS
353static inline int mpol_parse_str(char *str, struct mempolicy **mpol,
354 int no_context)
355{
356 return 1; /* error */
357}
358#endif
359
360static inline int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol,
361 int no_context)
362{
363 return 0;
364}
365
366#endif /* CONFIG_NUMA */
367#endif /* __KERNEL__ */
368
369#endif