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 /*
  12  * Both the MPOL_* mempolicy mode and the MPOL_F_* optional mode flags are
  13  * passed by the user to either set_mempolicy() or mbind() in an 'int' actual.
  14  * The MPOL_MODE_FLAGS macro determines the legal set of optional mode flags.
  15  */
  16
  17 /* Policies */
  18 enum {
  19         MPOL_DEFAULT,
  20         MPOL_PREFERRED,
  21         MPOL_BIND,
  22         MPOL_INTERLEAVE,
  23         MPOL_MAX,       /* always last member of enum */
  24 };
  25
  26 /* Flags for set_mempolicy */
  27 #define MPOL_F_STATIC_NODES     (1 << 15)
  28 #define MPOL_F_RELATIVE_NODES   (1 << 14)
  29
  30 /*
  31  * MPOL_MODE_FLAGS is the union of all possible optional mode flags passed to
  32  * either set_mempolicy() or mbind().
  33  */
  34 #define MPOL_MODE_FLAGS (MPOL_F_STATIC_NODES | MPOL_F_RELATIVE_NODES)
  35
  36 /* Flags for get_mempolicy */
  37 #define MPOL_F_NODE     (1<<0)  /* return next IL mode instead of node mask */
  38 #define MPOL_F_ADDR     (1<<1)  /* look up vma using address */
  39 #define MPOL_F_MEMS_ALLOWED (1<<2) /* return allowed memories */
  40
  41 /* Flags for mbind */
  42 #define MPOL_MF_STRICT  (1<<0)  /* Verify existing pages in the mapping */
  43 #define MPOL_MF_MOVE    (1<<1)  /* Move pages owned by this process to conform to mapping */
  44 #define MPOL_MF_MOVE_ALL (1<<2) /* Move every page to conform to mapping */
  45 #define MPOL_MF_INTERNAL (1<<3) /* Internal flags start here */
  46
  47 /*
  48  * Internal flags that share the struct mempolicy flags word with
  49  * "mode flags".  These flags are allocated from bit 0 up, as they
  50  * are never OR'ed into the mode in mempolicy API arguments.
  51  */
  52 #define MPOL_F_SHARED  (1 << 0) /* identify shared policies */
  53 #define MPOL_F_LOCAL   (1 << 1) /* preferred local allocation */
  54
  55 #ifdef __KERNEL__
  56
  57 #include <linux/mmzone.h>
  58 #include <linux/slab.h>
  59 #include <linux/rbtree.h>
  60 #include <linux/spinlock.h>
  61 #include <linux/nodemask.h>
  62
  63 struct mm_struct;
  64
  65 #ifdef CONFIG_NUMA
  66
  67 /*
  68  * Describe a memory policy.
  69  *
  70  * A mempolicy can be either associated with a process or with a VMA.
  71  * For VMA related allocations the VMA policy is preferred, otherwise
  72  * the process policy is used. Interrupts ignore the memory policy
  73  * of the current process.
  74  *
  75  * Locking policy for interlave:
  76  * In process context there is no locking because only the process accesses
  77  * its own state. All vma manipulation is somewhat protected by a down_read on
  78  * mmap_sem.
  79  *
  80  * Freeing policy:
  81  * Mempolicy objects are reference counted.  A mempolicy will be freed when
  82  * mpol_put() decrements the reference count to zero.
  83  *
  84  * Duplicating policy objects:
  85  * mpol_dup() allocates a new mempolicy and copies the specified mempolicy
  86  * to the new storage.  The reference count of the new object is initialized
  87  * to 1, representing the caller of mpol_dup().
  88  */
  89 struct mempolicy {
  90         atomic_t refcnt;
  91         unsigned short mode;    /* See MPOL_* above */
  92         unsigned short flags;   /* See set_mempolicy() MPOL_F_* above */
  93         union {
  94                 short            preferred_node; /* preferred */
  95                 nodemask_t       nodes;         /* interleave/bind */
  96                 /* undefined for default */
  97         } v;
  98         union {
  99                 nodemask_t cpuset_mems_allowed; /* relative to these nodes */
 100                 nodemask_t user_nodemask;       /* nodemask passed by user */
 101         } w;
 102 };
 103
 104 /*
 105  * Support for managing mempolicy data objects (clone, copy, destroy)
 106  * The default fast path of a NULL MPOL_DEFAULT policy is always inlined.
 107  */
 108
 109 extern void __mpol_put(struct mempolicy *pol);
 110 static inline void mpol_put(struct mempolicy *pol)
 111 {
 112         if (pol)
 113                 __mpol_put(pol);
 114 }
 115
 116 /*
 117  * Does mempolicy pol need explicit unref after use?
 118  * Currently only needed for shared policies.
 119  */
 120 static inline int mpol_needs_cond_ref(struct mempolicy *pol)
 121 {
 122         return (pol && (pol->flags & MPOL_F_SHARED));
 123 }
 124
 125 static inline void mpol_cond_put(struct mempolicy *pol)
 126 {
 127         if (mpol_needs_cond_ref(pol))
 128                 __mpol_put(pol);
 129 }
 130
 131 extern struct mempolicy *__mpol_cond_copy(struct mempolicy *tompol,
 132                                           struct mempolicy *frompol);
 133 static inline struct mempolicy *mpol_cond_copy(struct mempolicy *tompol,
 134                                                 struct mempolicy *frompol)
 135 {
 136         if (!frompol)
 137                 return frompol;
 138         return __mpol_cond_copy(tompol, frompol);
 139 }
 140
 141 extern struct mempolicy *__mpol_dup(struct mempolicy *pol);
 142 static inline struct mempolicy *mpol_dup(struct mempolicy *pol)
 143 {
 144         if (pol)
 145                 pol = __mpol_dup(pol);
 146         return pol;
 147 }
 148
 149 #define vma_policy(vma) ((vma)->vm_policy)
 150 #define vma_set_policy(vma, pol) ((vma)->vm_policy = (pol))
 151
 152 static inline void mpol_get(struct mempolicy *pol)
 153 {
 154         if (pol)
 155                 atomic_inc(&pol->refcnt);
 156 }
 157
 158 extern int __mpol_equal(struct mempolicy *a, struct mempolicy *b);
 159 static inline int mpol_equal(struct mempolicy *a, struct mempolicy *b)
 160 {
 161         if (a == b)
 162                 return 1;
 163         return __mpol_equal(a, b);
 164 }
 165
 166 /*
 167  * Tree of shared policies for a shared memory region.
 168  * Maintain the policies in a pseudo mm that contains vmas. The vmas
 169  * carry the policy. As a special twist the pseudo mm is indexed in pages, not
 170  * bytes, so that we can work with shared memory segments bigger than
 171  * unsigned long.
 172  */
 173
 174 struct sp_node {
 175         struct rb_node nd;
 176         unsigned long start, end;
 177         struct mempolicy *policy;
 178 };
 179
 180 struct shared_policy {
 181         struct rb_root root;
 182         spinlock_t lock;
 183 };
 184
 185 void mpol_shared_policy_init(struct shared_policy *info, unsigned short mode,
 186                                 unsigned short flags, nodemask_t *nodes);
 187 int mpol_set_shared_policy(struct shared_policy *info,
 188                                 struct vm_area_struct *vma,
 189                                 struct mempolicy *new);
 190 void mpol_free_shared_policy(struct shared_policy *p);
 191 struct mempolicy *mpol_shared_policy_lookup(struct shared_policy *sp,
 192                                             unsigned long idx);
 193
 194 extern void numa_default_policy(void);
 195 extern void numa_policy_init(void);
 196 extern void mpol_rebind_task(struct task_struct *tsk,
 197                                         const nodemask_t *new);
 198 extern void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new);
 199 extern void mpol_fix_fork_child_flag(struct task_struct *p);
 200
 201 extern struct zonelist *huge_zonelist(struct vm_area_struct *vma,
 202                                 unsigned long addr, gfp_t gfp_flags,
 203                                 struct mempolicy **mpol, nodemask_t **nodemask);
 204 extern unsigned slab_node(struct mempolicy *policy);
 205
 206 extern enum zone_type policy_zone;
 207
 208 static inline void check_highest_zone(enum zone_type k)
 209 {
 210         if (k > policy_zone && k != ZONE_MOVABLE)
 211                 policy_zone = k;
 212 }
 213
 214 int do_migrate_pages(struct mm_struct *mm,
 215         const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags);
 216
 217
 218 #ifdef CONFIG_TMPFS
 219 extern int mpol_parse_str(char *str, unsigned short *mode,
 220                         unsigned short *mode_flags, nodemask_t *policy_nodes);
 221
 222 extern int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol);
 223 #endif
 224 #else
 225
 226 struct mempolicy {};
 227
 228 static inline int mpol_equal(struct mempolicy *a, struct mempolicy *b)
 229 {
 230         return 1;
 231 }
 232
 233 static inline void mpol_put(struct mempolicy *p)
 234 {
 235 }
 236
 237 static inline void mpol_cond_put(struct mempolicy *pol)
 238 {
 239 }
 240
 241 static inline struct mempolicy *mpol_cond_copy(struct mempolicy *to,
 242                                                 struct mempolicy *from)
 243 {
 244         return from;
 245 }
 246
 247 static inline void mpol_get(struct mempolicy *pol)
 248 {
 249 }
 250
 251 static inline struct mempolicy *mpol_dup(struct mempolicy *old)
 252 {
 253         return NULL;
 254 }
 255
 256 struct shared_policy {};
 257
 258 static inline int mpol_set_shared_policy(struct shared_policy *info,
 259                                         struct vm_area_struct *vma,
 260                                         struct mempolicy *new)
 261 {
 262         return -EINVAL;
 263 }
 264
 265 static inline void mpol_shared_policy_init(struct shared_policy *info,
 266                 unsigned short mode, unsigned short flags, nodemask_t *nodes)
 267 {
 268 }
 269
 270 static inline void mpol_free_shared_policy(struct shared_policy *p)
 271 {
 272 }
 273
 274 static inline struct mempolicy *
 275 mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx)
 276 {
 277         return NULL;
 278 }
 279
 280 #define vma_policy(vma) NULL
 281 #define vma_set_policy(vma, pol) do {} while(0)
 282
 283 static inline void numa_policy_init(void)
 284 {
 285 }
 286
 287 static inline void numa_default_policy(void)
 288 {
 289 }
 290
 291 static inline void mpol_rebind_task(struct task_struct *tsk,
 292                                         const nodemask_t *new)
 293 {
 294 }
 295
 296 static inline void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new)
 297 {
 298 }
 299
 300 static inline void mpol_fix_fork_child_flag(struct task_struct *p)
 301 {
 302 }
 303
 304 static inline struct zonelist *huge_zonelist(struct vm_area_struct *vma,
 305                                 unsigned long addr, gfp_t gfp_flags,
 306                                 struct mempolicy **mpol, nodemask_t **nodemask)
 307 {
 308         *mpol = NULL;
 309         *nodemask = NULL;
 310         return node_zonelist(0, gfp_flags);
 311 }
 312
 313 static inline int do_migrate_pages(struct mm_struct *mm,
 314                         const nodemask_t *from_nodes,
 315                         const nodemask_t *to_nodes, int flags)
 316 {
 317         return 0;
 318 }
 319
 320 static inline void check_highest_zone(int k)
 321 {
 322 }
 323
 324 #ifdef CONFIG_TMPFS
 325 static inline int mpol_parse_str(char *value, unsigned short *policy,
 326                                 unsigned short flags, nodemask_t *policy_nodes)
 327 {
 328         return 1;
 329 }
 330
 331 static inline int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol)
 332 {
 333         return 0;
 334 }
 335 #endif
 336
 337 #endif /* CONFIG_NUMA */
 338 #endif /* __KERNEL__ */
 339
 340 #endif