include/linux/mempolicy.h

   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
  10 #include <linux/mmzone.h>
  11 #include <linux/dax.h>
  12 #include <linux/slab.h>
  13 #include <linux/rbtree.h>
  14 #include <linux/spinlock.h>
  15 #include <linux/nodemask.h>
  16 #include <linux/pagemap.h>
  17 #include <uapi/linux/mempolicy.h>
  18
  19 struct mm_struct;
  20
  21 #ifdef CONFIG_NUMA
  22
  23 /*
  24  * Describe a memory policy.
  25  *
  26  * A mempolicy can be either associated with a process or with a VMA.
  27  * For VMA related allocations the VMA policy is preferred, otherwise
  28  * the process policy is used. Interrupts ignore the memory policy
  29  * of the current process.
  30  *
  31  * Locking policy for interlave:
  32  * In process context there is no locking because only the process accesses
  33  * its own state. All vma manipulation is somewhat protected by a down_read on
  34  * mmap_sem.
  35  *
  36  * Freeing policy:
  37  * Mempolicy objects are reference counted.  A mempolicy will be freed when
  38  * mpol_put() decrements the reference count to zero.
  39  *
  40  * Duplicating policy objects:
  41  * mpol_dup() allocates a new mempolicy and copies the specified mempolicy
  42  * to the new storage.  The reference count of the new object is initialized
  43  * to 1, representing the caller of mpol_dup().
  44  */
  45 struct mempolicy {
  46         atomic_t refcnt;
  47         unsigned short mode;    /* See MPOL_* above */
  48         unsigned short flags;   /* See set_mempolicy() MPOL_F_* above */
  49         union {
  50                 short            preferred_node; /* preferred */
  51                 nodemask_t       nodes;         /* interleave/bind */
  52                 /* undefined for default */
  53         } v;
  54         union {
  55                 nodemask_t cpuset_mems_allowed; /* relative to these nodes */
  56                 nodemask_t user_nodemask;       /* nodemask passed by user */
  57         } w;
  58 };
  59
  60 /*
  61  * Support for managing mempolicy data objects (clone, copy, destroy)
  62  * The default fast path of a NULL MPOL_DEFAULT policy is always inlined.
  63  */
  64
  65 extern void __mpol_put(struct mempolicy *pol);
  66 static inline void mpol_put(struct mempolicy *pol)
  67 {
  68         if (pol)
  69                 __mpol_put(pol);
  70 }
  71
  72 /*
  73  * Does mempolicy pol need explicit unref after use?
  74  * Currently only needed for shared policies.
  75  */
  76 static inline int mpol_needs_cond_ref(struct mempolicy *pol)
  77 {
  78         return (pol && (pol->flags & MPOL_F_SHARED));
  79 }
  80
  81 static inline void mpol_cond_put(struct mempolicy *pol)
  82 {
  83         if (mpol_needs_cond_ref(pol))
  84                 __mpol_put(pol);
  85 }
  86
  87 extern struct mempolicy *__mpol_dup(struct mempolicy *pol);
  88 static inline struct mempolicy *mpol_dup(struct mempolicy *pol)
  89 {
  90         if (pol)
  91                 pol = __mpol_dup(pol);
  92         return pol;
  93 }
  94
  95 #define vma_policy(vma) ((vma)->vm_policy)
  96
  97 static inline void mpol_get(struct mempolicy *pol)
  98 {
  99         if (pol)
 100                 atomic_inc(&pol->refcnt);
 101 }
 102
 103 extern bool __mpol_equal(struct mempolicy *a, struct mempolicy *b);
 104 static inline bool mpol_equal(struct mempolicy *a, struct mempolicy *b)
 105 {
 106         if (a == b)
 107                 return true;
 108         return __mpol_equal(a, b);
 109 }
 110
 111 /*
 112  * Tree of shared policies for a shared memory region.
 113  * Maintain the policies in a pseudo mm that contains vmas. The vmas
 114  * carry the policy. As a special twist the pseudo mm is indexed in pages, not
 115  * bytes, so that we can work with shared memory segments bigger than
 116  * unsigned long.
 117  */
 118
 119 struct sp_node {
 120         struct rb_node nd;
 121         unsigned long start, end;
 122         struct mempolicy *policy;
 123 };
 124
 125 struct shared_policy {
 126         struct rb_root root;
 127         rwlock_t lock;
 128 };
 129
 130 int vma_dup_policy(struct vm_area_struct *src, struct vm_area_struct *dst);
 131 void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol);
 132 int mpol_set_shared_policy(struct shared_policy *info,
 133                                 struct vm_area_struct *vma,
 134                                 struct mempolicy *new);
 135 void mpol_free_shared_policy(struct shared_policy *p);
 136 struct mempolicy *mpol_shared_policy_lookup(struct shared_policy *sp,
 137                                             unsigned long idx);
 138
 139 struct mempolicy *get_task_policy(struct task_struct *p);
 140 struct mempolicy *__get_vma_policy(struct vm_area_struct *vma,
 141                 unsigned long addr);
 142 struct mempolicy *get_vma_policy(struct vm_area_struct *vma,
 143                                                 unsigned long addr);
 144 bool vma_policy_mof(struct vm_area_struct *vma);
 145
 146 extern void numa_default_policy(void);
 147 extern void numa_policy_init(void);
 148 extern void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new);
 149 extern void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new);
 150
 151 extern int huge_node(struct vm_area_struct *vma,
 152                                 unsigned long addr, gfp_t gfp_flags,
 153                                 struct mempolicy **mpol, nodemask_t **nodemask);
 154 extern bool init_nodemask_of_mempolicy(nodemask_t *mask);
 155 extern bool mempolicy_nodemask_intersects(struct task_struct *tsk,
 156                                 const nodemask_t *mask);
 157 extern unsigned int mempolicy_slab_node(void);
 158
 159 extern enum zone_type policy_zone;
 160
 161 static inline void check_highest_zone(enum zone_type k)
 162 {
 163         if (k > policy_zone && k != ZONE_MOVABLE)
 164                 policy_zone = k;
 165 }
 166
 167 int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
 168                      const nodemask_t *to, int flags);
 169
 170
 171 #ifdef CONFIG_TMPFS
 172 extern int mpol_parse_str(char *str, struct mempolicy **mpol);
 173 #endif
 174
 175 extern void mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol);
 176
 177 /* Check if a vma is migratable */
 178 static inline bool vma_migratable(struct vm_area_struct *vma)
 179 {
 180         if (vma->vm_flags & (VM_IO | VM_PFNMAP))
 181                 return false;
 182
 183         /*
 184          * DAX device mappings require predictable access latency, so avoid
 185          * incurring periodic faults.
 186          */
 187         if (vma_is_dax(vma))
 188                 return false;
 189
 190 #ifndef CONFIG_ARCH_ENABLE_HUGEPAGE_MIGRATION
 191         if (vma->vm_flags & VM_HUGETLB)
 192                 return false;
 193 #endif
 194
 195         /*
 196          * Migration allocates pages in the highest zone. If we cannot
 197          * do so then migration (at least from node to node) is not
 198          * possible.
 199          */
 200         if (vma->vm_file &&
 201                 gfp_zone(mapping_gfp_mask(vma->vm_file->f_mapping))
 202                                                                 < policy_zone)
 203                         return false;
 204         return true;
 205 }
 206
 207 extern int mpol_misplaced(struct page *, struct vm_area_struct *, unsigned long);
 208 extern void mpol_put_task_policy(struct task_struct *);
 209
 210 #else
 211
 212 struct mempolicy {};
 213
 214 static inline bool mpol_equal(struct mempolicy *a, struct mempolicy *b)
 215 {
 216         return true;
 217 }
 218
 219 static inline void mpol_put(struct mempolicy *p)
 220 {
 221 }
 222
 223 static inline void mpol_cond_put(struct mempolicy *pol)
 224 {
 225 }
 226
 227 static inline void mpol_get(struct mempolicy *pol)
 228 {
 229 }
 230
 231 struct shared_policy {};
 232
 233 static inline void mpol_shared_policy_init(struct shared_policy *sp,
 234                                                 struct mempolicy *mpol)
 235 {
 236 }
 237
 238 static inline void mpol_free_shared_policy(struct shared_policy *p)
 239 {
 240 }
 241
 242 static inline struct mempolicy *
 243 mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx)
 244 {
 245         return NULL;
 246 }
 247
 248 #define vma_policy(vma) NULL
 249
 250 static inline int
 251 vma_dup_policy(struct vm_area_struct *src, struct vm_area_struct *dst)
 252 {
 253         return 0;
 254 }
 255
 256 static inline void numa_policy_init(void)
 257 {
 258 }
 259
 260 static inline void numa_default_policy(void)
 261 {
 262 }
 263
 264 static inline void mpol_rebind_task(struct task_struct *tsk,
 265                                 const nodemask_t *new)
 266 {
 267 }
 268
 269 static inline void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new)
 270 {
 271 }
 272
 273 static inline int huge_node(struct vm_area_struct *vma,
 274                                 unsigned long addr, gfp_t gfp_flags,
 275                                 struct mempolicy **mpol, nodemask_t **nodemask)
 276 {
 277         *mpol = NULL;
 278         *nodemask = NULL;
 279         return 0;
 280 }
 281
 282 static inline bool init_nodemask_of_mempolicy(nodemask_t *m)
 283 {
 284         return false;
 285 }
 286
 287 static inline int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
 288                                    const nodemask_t *to, int flags)
 289 {
 290         return 0;
 291 }
 292
 293 static inline void check_highest_zone(int k)
 294 {
 295 }
 296
 297 #ifdef CONFIG_TMPFS
 298 static inline int mpol_parse_str(char *str, struct mempolicy **mpol)
 299 {
 300         return 1;       /* error */
 301 }
 302 #endif
 303
 304 static inline int mpol_misplaced(struct page *page, struct vm_area_struct *vma,
 305                                  unsigned long address)
 306 {
 307         return -1; /* no node preference */
 308 }
 309
 310 static inline void mpol_put_task_policy(struct task_struct *task)
 311 {
 312 }
 313 #endif /* CONFIG_NUMA */
 314 #endif