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