include/linux/mempolicy.h

   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
  11 /* Policies */
  12 #define MPOL_DEFAULT    0
  13 #define MPOL_PREFERRED  1
  14 #define MPOL_BIND       2
  15 #define MPOL_INTERLEAVE 3
  16
  17 #define MPOL_MAX MPOL_INTERLEAVE
  18
  19 /* Flags for get_mem_policy */
  20 #define MPOL_F_NODE     (1<<0)  /* return next IL mode instead of node mask */
  21 #define MPOL_F_ADDR     (1<<1)  /* look up vma using address */
  22
  23 /* Flags for mbind */
  24 #define MPOL_MF_STRICT  (1<<0)  /* Verify existing pages in the mapping */
  25 #define MPOL_MF_MOVE    (1<<1)  /* Move pages owned by this process to conform to mapping */
  26 #define MPOL_MF_MOVE_ALL (1<<2) /* Move every page to conform to mapping */
  27 #define MPOL_MF_INTERNAL (1<<3) /* Internal flags start here */
  28
  29 #ifdef __KERNEL__
  30
  31 #include <linux/config.h>
  32 #include <linux/mmzone.h>
  33 #include <linux/slab.h>
  34 #include <linux/rbtree.h>
  35 #include <linux/spinlock.h>
  36 #include <linux/nodemask.h>
  37
  38 struct vm_area_struct;
  39
  40 #ifdef CONFIG_NUMA
  41
  42 /*
  43  * Describe a memory policy.
  44  *
  45  * A mempolicy can be either associated with a process or with a VMA.
  46  * For VMA related allocations the VMA policy is preferred, otherwise
  47  * the process policy is used. Interrupts ignore the memory policy
  48  * of the current process.
  49  *
  50  * Locking policy for interlave:
  51  * In process context there is no locking because only the process accesses
  52  * its own state. All vma manipulation is somewhat protected by a down_read on
  53  * mmap_sem.
  54  *
  55  * Freeing policy:
  56  * When policy is MPOL_BIND v.zonelist is kmalloc'ed and must be kfree'd.
  57  * All other policies don't have any external state. mpol_free() handles this.
  58  *
  59  * Copying policy objects:
  60  * For MPOL_BIND the zonelist must be always duplicated. mpol_clone() does this.
  61  */
  62 struct mempolicy {
  63         atomic_t refcnt;
  64         short policy;   /* See MPOL_* above */
  65         union {
  66                 struct zonelist  *zonelist;     /* bind */
  67                 short            preferred_node; /* preferred */
  68                 nodemask_t       nodes;         /* interleave */
  69                 /* undefined for default */
  70         } v;
  71         nodemask_t cpuset_mems_allowed; /* mempolicy relative to these nodes */
  72 };
  73
  74 /*
  75  * Support for managing mempolicy data objects (clone, copy, destroy)
  76  * The default fast path of a NULL MPOL_DEFAULT policy is always inlined.
  77  */
  78
  79 extern void __mpol_free(struct mempolicy *pol);
  80 static inline void mpol_free(struct mempolicy *pol)
  81 {
  82         if (pol)
  83                 __mpol_free(pol);
  84 }
  85
  86 extern struct mempolicy *__mpol_copy(struct mempolicy *pol);
  87 static inline struct mempolicy *mpol_copy(struct mempolicy *pol)
  88 {
  89         if (pol)
  90                 pol = __mpol_copy(pol);
  91         return pol;
  92 }
  93
  94 #define vma_policy(vma) ((vma)->vm_policy)
  95 #define vma_set_policy(vma, pol) ((vma)->vm_policy = (pol))
  96
  97 static inline void mpol_get(struct mempolicy *pol)
  98 {
  99         if (pol)
 100                 atomic_inc(&pol->refcnt);
 101 }
 102
 103 extern int __mpol_equal(struct mempolicy *a, struct mempolicy *b);
 104 static inline int mpol_equal(struct mempolicy *a, struct mempolicy *b)
 105 {
 106         if (a == b)
 107                 return 1;
 108         return __mpol_equal(a, b);
 109 }
 110 #define vma_mpol_equal(a,b) mpol_equal(vma_policy(a), vma_policy(b))
 111
 112 /* Could later add inheritance of the process policy here. */
 113
 114 #define mpol_set_vma_default(vma) ((vma)->vm_policy = NULL)
 115
 116 /*
 117  * Tree of shared policies for a shared memory region.
 118  * Maintain the policies in a pseudo mm that contains vmas. The vmas
 119  * carry the policy. As a special twist the pseudo mm is indexed in pages, not
 120  * bytes, so that we can work with shared memory segments bigger than
 121  * unsigned long.
 122  */
 123
 124 struct sp_node {
 125         struct rb_node nd;
 126         unsigned long start, end;
 127         struct mempolicy *policy;
 128 };
 129
 130 struct shared_policy {
 131         struct rb_root root;
 132         spinlock_t lock;
 133 };
 134
 135 static inline void mpol_shared_policy_init(struct shared_policy *info)
 136 {
 137         info->root = RB_ROOT;
 138         spin_lock_init(&info->lock);
 139 }
 140
 141 int mpol_set_shared_policy(struct shared_policy *info,
 142                                 struct vm_area_struct *vma,
 143                                 struct mempolicy *new);
 144 void mpol_free_shared_policy(struct shared_policy *p);
 145 struct mempolicy *mpol_shared_policy_lookup(struct shared_policy *sp,
 146                                             unsigned long idx);
 147
 148 extern void numa_default_policy(void);
 149 extern void numa_policy_init(void);
 150 extern void mpol_rebind_policy(struct mempolicy *pol, const nodemask_t *new);
 151 extern void mpol_rebind_task(struct task_struct *tsk,
 152                                         const nodemask_t *new);
 153 extern void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new);
 154 #define set_cpuset_being_rebound(x) (cpuset_being_rebound = (x))
 155
 156 #ifdef CONFIG_CPUSET
 157 #define current_cpuset_is_being_rebound() \
 158                                 (cpuset_being_rebound == current->cpuset)
 159 #else
 160 #define current_cpuset_is_being_rebound() 0
 161 #endif
 162
 163 extern struct mempolicy default_policy;
 164 extern struct zonelist *huge_zonelist(struct vm_area_struct *vma,
 165                 unsigned long addr);
 166
 167 extern int policy_zone;
 168
 169 static inline void check_highest_zone(int k)
 170 {
 171         if (k > policy_zone)
 172                 policy_zone = k;
 173 }
 174
 175 int do_migrate_pages(struct mm_struct *mm,
 176         const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags);
 177
 178 extern void *cpuset_being_rebound;      /* Trigger mpol_copy vma rebind */
 179
 180 #else
 181
 182 struct mempolicy {};
 183
 184 static inline int mpol_equal(struct mempolicy *a, struct mempolicy *b)
 185 {
 186         return 1;
 187 }
 188 #define vma_mpol_equal(a,b) 1
 189
 190 #define mpol_set_vma_default(vma) do {} while(0)
 191
 192 static inline void mpol_free(struct mempolicy *p)
 193 {
 194 }
 195
 196 static inline void mpol_get(struct mempolicy *pol)
 197 {
 198 }
 199
 200 static inline struct mempolicy *mpol_copy(struct mempolicy *old)
 201 {
 202         return NULL;
 203 }
 204
 205 struct shared_policy {};
 206
 207 static inline int mpol_set_shared_policy(struct shared_policy *info,
 208                                         struct vm_area_struct *vma,
 209                                         struct mempolicy *new)
 210 {
 211         return -EINVAL;
 212 }
 213
 214 static inline void mpol_shared_policy_init(struct shared_policy *info)
 215 {
 216 }
 217
 218 static inline void mpol_free_shared_policy(struct shared_policy *p)
 219 {
 220 }
 221
 222 static inline struct mempolicy *
 223 mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx)
 224 {
 225         return NULL;
 226 }
 227
 228 #define vma_policy(vma) NULL
 229 #define vma_set_policy(vma, pol) do {} while(0)
 230
 231 static inline void numa_policy_init(void)
 232 {
 233 }
 234
 235 static inline void numa_default_policy(void)
 236 {
 237 }
 238
 239 static inline void mpol_rebind_policy(struct mempolicy *pol,
 240                                         const nodemask_t *new)
 241 {
 242 }
 243
 244 static inline void mpol_rebind_task(struct task_struct *tsk,
 245                                         const nodemask_t *new)
 246 {
 247 }
 248
 249 static inline void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new)
 250 {
 251 }
 252
 253 #define set_cpuset_being_rebound(x) do {} while (0)
 254
 255 static inline struct zonelist *huge_zonelist(struct vm_area_struct *vma,
 256                 unsigned long addr)
 257 {
 258         return NODE_DATA(0)->node_zonelists + gfp_zone(GFP_HIGHUSER);
 259 }
 260
 261 static inline int do_migrate_pages(struct mm_struct *mm,
 262                         const nodemask_t *from_nodes,
 263                         const nodemask_t *to_nodes, int flags)
 264 {
 265         return 0;
 266 }
 267
 268 static inline void check_highest_zone(int k)
 269 {
 270 }
 271 #endif /* CONFIG_NUMA */
 272 #endif /* __KERNEL__ */
 273
 274 #endif