include/linux/mempolicy.h

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