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
  54 #ifdef __KERNEL__
  55
  56 #include <linux/mmzone.h>
  57 #include <linux/slab.h>
  58 #include <linux/rbtree.h>
  59 #include <linux/spinlock.h>
  60 #include <linux/nodemask.h>
  61
  62 struct mm_struct;
  63
  64 #ifdef CONFIG_NUMA
  65
  66 /*
  67  * Describe a memory policy.
  68  *
  69  * A mempolicy can be either associated with a process or with a VMA.
  70  * For VMA related allocations the VMA policy is preferred, otherwise
  71  * the process policy is used. Interrupts ignore the memory policy
  72  * of the current process.
  73  *
  74  * Locking policy for interlave:
  75  * In process context there is no locking because only the process accesses
  76  * its own state. All vma manipulation is somewhat protected by a down_read on
  77  * mmap_sem.
  78  *
  79  * Freeing policy:
  80  * Mempolicy objects are reference counted.  A mempolicy will be freed when
  81  * mpol_put() decrements the reference count to zero.
  82  *
  83  * Duplicating policy objects:
  84  * mpol_dup() allocates a new mempolicy and copies the specified mempolicy
  85  * to the new storage.  The reference count of the new object is initialized
  86  * to 1, representing the caller of mpol_dup().
  87  */
  88 struct mempolicy {
  89         atomic_t refcnt;
  90         unsigned short mode;    /* See MPOL_* above */
  91         unsigned short flags;   /* See set_mempolicy() MPOL_F_* above */
  92         union {
  93                 short            preferred_node; /* preferred */
  94                 nodemask_t       nodes;         /* interleave/bind */
  95                 /* undefined for default */
  96         } v;
  97         union {
  98                 nodemask_t cpuset_mems_allowed; /* relative to these nodes */
  99                 nodemask_t user_nodemask;       /* nodemask passed by user */
 100         } w;
 101 };
 102
 103 /*
 104  * Support for managing mempolicy data objects (clone, copy, destroy)
 105  * The default fast path of a NULL MPOL_DEFAULT policy is always inlined.
 106  */
 107
 108 extern void __mpol_put(struct mempolicy *pol);
 109 static inline void mpol_put(struct mempolicy *pol)
 110 {
 111         if (pol)
 112                 __mpol_put(pol);
 113 }
 114
 115 /*
 116  * Does mempolicy pol need explicit unref after use?
 117  * Currently only needed for shared policies.
 118  */
 119 static inline int mpol_needs_cond_ref(struct mempolicy *pol)
 120 {
 121         return (pol && (pol->flags & MPOL_F_SHARED));
 122 }
 123
 124 static inline void mpol_cond_put(struct mempolicy *pol)
 125 {
 126         if (mpol_needs_cond_ref(pol))
 127                 __mpol_put(pol);
 128 }
 129
 130 extern struct mempolicy *__mpol_cond_copy(struct mempolicy *tompol,
 131                                           struct mempolicy *frompol);
 132 static inline struct mempolicy *mpol_cond_copy(struct mempolicy *tompol,
 133                                                 struct mempolicy *frompol)
 134 {
 135         if (!frompol)
 136                 return frompol;
 137         return __mpol_cond_copy(tompol, frompol);
 138 }
 139
 140 extern struct mempolicy *__mpol_dup(struct mempolicy *pol);
 141 static inline struct mempolicy *mpol_dup(struct mempolicy *pol)
 142 {
 143         if (pol)
 144                 pol = __mpol_dup(pol);
 145         return pol;
 146 }
 147
 148 #define vma_policy(vma) ((vma)->vm_policy)
 149 #define vma_set_policy(vma, pol) ((vma)->vm_policy = (pol))
 150
 151 static inline void mpol_get(struct mempolicy *pol)
 152 {
 153         if (pol)
 154                 atomic_inc(&pol->refcnt);
 155 }
 156
 157 extern int __mpol_equal(struct mempolicy *a, struct mempolicy *b);
 158 static inline int mpol_equal(struct mempolicy *a, struct mempolicy *b)
 159 {
 160         if (a == b)
 161                 return 1;
 162         return __mpol_equal(a, b);
 163 }
 164
 165 /*
 166  * Tree of shared policies for a shared memory region.
 167  * Maintain the policies in a pseudo mm that contains vmas. The vmas
 168  * carry the policy. As a special twist the pseudo mm is indexed in pages, not
 169  * bytes, so that we can work with shared memory segments bigger than
 170  * unsigned long.
 171  */
 172
 173 struct sp_node {
 174         struct rb_node nd;
 175         unsigned long start, end;
 176         struct mempolicy *policy;
 177 };
 178
 179 struct shared_policy {
 180         struct rb_root root;
 181         spinlock_t lock;
 182 };
 183
 184 void mpol_shared_policy_init(struct shared_policy *info, unsigned short mode,
 185                                 unsigned short flags, nodemask_t *nodes);
 186 int mpol_set_shared_policy(struct shared_policy *info,
 187                                 struct vm_area_struct *vma,
 188                                 struct mempolicy *new);
 189 void mpol_free_shared_policy(struct shared_policy *p);
 190 struct mempolicy *mpol_shared_policy_lookup(struct shared_policy *sp,
 191                                             unsigned long idx);
 192
 193 extern void numa_default_policy(void);
 194 extern void numa_policy_init(void);
 195 extern void mpol_rebind_task(struct task_struct *tsk,
 196                                         const nodemask_t *new);
 197 extern void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new);
 198 extern void mpol_fix_fork_child_flag(struct task_struct *p);
 199
 200 extern struct zonelist *huge_zonelist(struct vm_area_struct *vma,
 201                                 unsigned long addr, gfp_t gfp_flags,
 202                                 struct mempolicy **mpol, nodemask_t **nodemask);
 203 extern unsigned slab_node(struct mempolicy *policy);
 204
 205 extern enum zone_type policy_zone;
 206
 207 static inline void check_highest_zone(enum zone_type k)
 208 {
 209         if (k > policy_zone && k != ZONE_MOVABLE)
 210                 policy_zone = k;
 211 }
 212
 213 int do_migrate_pages(struct mm_struct *mm,
 214         const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags);
 215
 216 #else
 217
 218 struct mempolicy {};
 219
 220 static inline int mpol_equal(struct mempolicy *a, struct mempolicy *b)
 221 {
 222         return 1;
 223 }
 224
 225 static inline void mpol_put(struct mempolicy *p)
 226 {
 227 }
 228
 229 static inline void mpol_cond_put(struct mempolicy *pol)
 230 {
 231 }
 232
 233 static inline struct mempolicy *mpol_cond_copy(struct mempolicy *to,
 234                                                 struct mempolicy *from)
 235 {
 236         return from;
 237 }
 238
 239 static inline void mpol_get(struct mempolicy *pol)
 240 {
 241 }
 242
 243 static inline struct mempolicy *mpol_dup(struct mempolicy *old)
 244 {
 245         return NULL;
 246 }
 247
 248 struct shared_policy {};
 249
 250 static inline int mpol_set_shared_policy(struct shared_policy *info,
 251                                         struct vm_area_struct *vma,
 252                                         struct mempolicy *new)
 253 {
 254         return -EINVAL;
 255 }
 256
 257 static inline void mpol_shared_policy_init(struct shared_policy *info,
 258                 unsigned short mode, unsigned short flags, nodemask_t *nodes)
 259 {
 260 }
 261
 262 static inline void mpol_free_shared_policy(struct shared_policy *p)
 263 {
 264 }
 265
 266 static inline struct mempolicy *
 267 mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx)
 268 {
 269         return NULL;
 270 }
 271
 272 #define vma_policy(vma) NULL
 273 #define vma_set_policy(vma, pol) do {} while(0)
 274
 275 static inline void numa_policy_init(void)
 276 {
 277 }
 278
 279 static inline void numa_default_policy(void)
 280 {
 281 }
 282
 283 static inline void mpol_rebind_task(struct task_struct *tsk,
 284                                         const nodemask_t *new)
 285 {
 286 }
 287
 288 static inline void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new)
 289 {
 290 }
 291
 292 static inline void mpol_fix_fork_child_flag(struct task_struct *p)
 293 {
 294 }
 295
 296 static inline struct zonelist *huge_zonelist(struct vm_area_struct *vma,
 297                                 unsigned long addr, gfp_t gfp_flags,
 298                                 struct mempolicy **mpol, nodemask_t **nodemask)
 299 {
 300         *mpol = NULL;
 301         *nodemask = NULL;
 302         return node_zonelist(0, gfp_flags);
 303 }
 304
 305 static inline int do_migrate_pages(struct mm_struct *mm,
 306                         const nodemask_t *from_nodes,
 307                         const nodemask_t *to_nodes, int flags)
 308 {
 309         return 0;
 310 }
 311
 312 static inline void check_highest_zone(int k)
 313 {
 314 }
 315 #endif /* CONFIG_NUMA */
 316 #endif /* __KERNEL__ */
 317
 318 #endif