1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 * NUMA memory policies for Linux.
4 * Copyright 2003,2004 Andi Kleen SuSE Labs
5 */
6#ifndef _LINUX_MEMPOLICY_H
7#define _LINUX_MEMPOLICY_H 1
8
9#include <linux/sched.h>
10#include <linux/mmzone.h>
11#include <linux/slab.h>
12#include <linux/rbtree.h>
13#include <linux/spinlock.h>
14#include <linux/nodemask.h>
15#include <linux/pagemap.h>
16#include <uapi/linux/mempolicy.h>
17
18struct mm_struct;
19
20#define NO_INTERLEAVE_INDEX (-1UL)	/* use task il_prev for interleaving */
21
22#ifdef CONFIG_NUMA
23
24/*
25 * Describe a memory policy.
26 *
27 * A mempolicy can be either associated with a process or with a VMA.
28 * For VMA related allocations the VMA policy is preferred, otherwise
29 * the process policy is used. Interrupts ignore the memory policy
30 * of the current process.
31 *
32 * Locking policy for interleave:
33 * In process context there is no locking because only the process accesses
34 * its own state. All vma manipulation is somewhat protected by a down_read on
35 * mmap_lock.
36 *
37 * Freeing policy:
38 * Mempolicy objects are reference counted.  A mempolicy will be freed when
39 * mpol_put() decrements the reference count to zero.
40 *
41 * Duplicating policy objects:
42 * mpol_dup() allocates a new mempolicy and copies the specified mempolicy
43 * to the new storage.  The reference count of the new object is initialized
44 * to 1, representing the caller of mpol_dup().
45 */
46struct mempolicy {
47	atomic_t refcnt;
48	unsigned short mode; 	/* See MPOL_* above */
49	unsigned short flags;	/* See set_mempolicy() MPOL_F_* above */
50	nodemask_t nodes;	/* interleave/bind/perfer */
51	int home_node;		/* Home node to use for MPOL_BIND and MPOL_PREFERRED_MANY */
52
53	union {
54		nodemask_t cpuset_mems_allowed;	/* relative to these nodes */
55		nodemask_t user_nodemask;	/* nodemask passed by user */
56	} w;
57};
58
59/*
60 * Support for managing mempolicy data objects (clone, copy, destroy)
61 * The default fast path of a NULL MPOL_DEFAULT policy is always inlined.
62 */
63
64extern void __mpol_put(struct mempolicy *pol);
65static inline void mpol_put(struct mempolicy *pol)
66{
67	if (pol)
68		__mpol_put(pol);
69}
70
71/*
72 * Does mempolicy pol need explicit unref after use?
73 * Currently only needed for shared policies.
74 */
75static inline int mpol_needs_cond_ref(struct mempolicy *pol)
76{
77	return (pol && (pol->flags & MPOL_F_SHARED));
78}
79
80static inline void mpol_cond_put(struct mempolicy *pol)
81{
82	if (mpol_needs_cond_ref(pol))
83		__mpol_put(pol);
84}
85
86extern struct mempolicy *__mpol_dup(struct mempolicy *pol);
87static inline struct mempolicy *mpol_dup(struct mempolicy *pol)
88{
89	if (pol)
90		pol = __mpol_dup(pol);
91	return pol;
92}
93
94static inline void mpol_get(struct mempolicy *pol)
95{
96	if (pol)
97		atomic_inc(&pol->refcnt);
98}
99
100extern bool __mpol_equal(struct mempolicy *a, struct mempolicy *b);
101static inline bool mpol_equal(struct mempolicy *a, struct mempolicy *b)
102{
103	if (a == b)
104		return true;
105	return __mpol_equal(a, b);
106}
107
108/*
109 * Tree of shared policies for a shared memory region.
110 */
111struct shared_policy {
112	struct rb_root root;
113	rwlock_t lock;
114};
115struct sp_node {
116	struct rb_node nd;
117	pgoff_t start, end;
118	struct mempolicy *policy;
119};
120
121int vma_dup_policy(struct vm_area_struct *src, struct vm_area_struct *dst);
122void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol);
123int mpol_set_shared_policy(struct shared_policy *sp,
124			   struct vm_area_struct *vma, struct mempolicy *mpol);
125void mpol_free_shared_policy(struct shared_policy *sp);
126struct mempolicy *mpol_shared_policy_lookup(struct shared_policy *sp,
127					    pgoff_t idx);
128
129struct mempolicy *get_task_policy(struct task_struct *p);
130struct mempolicy *__get_vma_policy(struct vm_area_struct *vma,
131		unsigned long addr, pgoff_t *ilx);
132struct mempolicy *get_vma_policy(struct vm_area_struct *vma,
133		unsigned long addr, int order, pgoff_t *ilx);
134bool vma_policy_mof(struct vm_area_struct *vma);
135
136extern void numa_default_policy(void);
137extern void numa_policy_init(void);
138extern void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new);
139extern void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new);
140
141extern int huge_node(struct vm_area_struct *vma,
142				unsigned long addr, gfp_t gfp_flags,
143				struct mempolicy **mpol, nodemask_t **nodemask);
144extern bool init_nodemask_of_mempolicy(nodemask_t *mask);
145extern bool mempolicy_in_oom_domain(struct task_struct *tsk,
146				const nodemask_t *mask);
147extern unsigned int mempolicy_slab_node(void);
148
149extern enum zone_type policy_zone;
150
151static inline void check_highest_zone(enum zone_type k)
152{
153	if (k > policy_zone && k != ZONE_MOVABLE)
154		policy_zone = k;
155}
156
157int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
158		     const nodemask_t *to, int flags);
159
160
161#ifdef CONFIG_TMPFS
162extern int mpol_parse_str(char *str, struct mempolicy **mpol);
163#endif
164
165extern void mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol);
166
167/* Check if a vma is migratable */
168extern bool vma_migratable(struct vm_area_struct *vma);
169
170int mpol_misplaced(struct folio *folio, struct vm_fault *vmf,
171					unsigned long addr);
172extern void mpol_put_task_policy(struct task_struct *);
173
174static inline bool mpol_is_preferred_many(struct mempolicy *pol)
175{
176	return  (pol->mode == MPOL_PREFERRED_MANY);
177}
178
179extern bool apply_policy_zone(struct mempolicy *policy, enum zone_type zone);
180
181#else
182
183struct mempolicy {};
184
185static inline struct mempolicy *get_task_policy(struct task_struct *p)
186{
187	return NULL;
188}
189
190static inline bool mpol_equal(struct mempolicy *a, struct mempolicy *b)
191{
192	return true;
193}
194
195static inline void mpol_put(struct mempolicy *pol)
196{
197}
198
199static inline void mpol_cond_put(struct mempolicy *pol)
200{
201}
202
203static inline void mpol_get(struct mempolicy *pol)
204{
205}
206
207struct shared_policy {};
208
209static inline void mpol_shared_policy_init(struct shared_policy *sp,
210						struct mempolicy *mpol)
211{
212}
213
214static inline void mpol_free_shared_policy(struct shared_policy *sp)
215{
216}
217
218static inline struct mempolicy *
219mpol_shared_policy_lookup(struct shared_policy *sp, pgoff_t idx)
220{
221	return NULL;
222}
223
224static inline struct mempolicy *get_vma_policy(struct vm_area_struct *vma,
225				unsigned long addr, int order, pgoff_t *ilx)
226{
227	*ilx = 0;
228	return NULL;
229}
230
231static inline int
232vma_dup_policy(struct vm_area_struct *src, struct vm_area_struct *dst)
233{
234	return 0;
235}
236
237static inline void numa_policy_init(void)
238{
239}
240
241static inline void numa_default_policy(void)
242{
243}
244
245static inline void mpol_rebind_task(struct task_struct *tsk,
246				const nodemask_t *new)
247{
248}
249
250static inline void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new)
251{
252}
253
254static inline int huge_node(struct vm_area_struct *vma,
255				unsigned long addr, gfp_t gfp_flags,
256				struct mempolicy **mpol, nodemask_t **nodemask)
257{
258	*mpol = NULL;
259	*nodemask = NULL;
260	return 0;
261}
262
263static inline bool init_nodemask_of_mempolicy(nodemask_t *m)
264{
265	return false;
266}
267
268static inline int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
269				   const nodemask_t *to, int flags)
270{
271	return 0;
272}
273
274static inline void check_highest_zone(int k)
275{
276}
277
278#ifdef CONFIG_TMPFS
279static inline int mpol_parse_str(char *str, struct mempolicy **mpol)
280{
281	return 1;	/* error */
282}
283#endif
284
285static inline int mpol_misplaced(struct folio *folio,
286				 struct vm_fault *vmf,
287				 unsigned long address)
288{
289	return -1; /* no node preference */
290}
291
292static inline void mpol_put_task_policy(struct task_struct *task)
293{
294}
295
296static inline bool mpol_is_preferred_many(struct mempolicy *pol)
297{
298	return  false;
299}
300
301#endif /* CONFIG_NUMA */
302#endif
303