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 enum mpol_rebind_step {
  27         MPOL_REBIND_ONCE,       /* do rebind work at once(not by two step) */
  28         MPOL_REBIND_STEP1,      /* first step(set all the newly nodes) */
  29         MPOL_REBIND_STEP2,      /* second step(clean all the disallowed nodes)*/
  30         MPOL_REBIND_NSTEP,
  31 };
  32
  33 /* Flags for set_mempolicy */
  34 #define MPOL_F_STATIC_NODES     (1 << 15)
  35 #define MPOL_F_RELATIVE_NODES   (1 << 14)
  36
  37 /*
  38  * MPOL_MODE_FLAGS is the union of all possible optional mode flags passed to
  39  * either set_mempolicy() or mbind().
  40  */
  41 #define MPOL_MODE_FLAGS (MPOL_F_STATIC_NODES | MPOL_F_RELATIVE_NODES)
  42
  43 /* Flags for get_mempolicy */
  44 #define MPOL_F_NODE     (1<<0)  /* return next IL mode instead of node mask */
  45 #define MPOL_F_ADDR     (1<<1)  /* look up vma using address */
  46 #define MPOL_F_MEMS_ALLOWED (1<<2) /* return allowed memories */
  47
  48 /* Flags for mbind */
  49 #define MPOL_MF_STRICT  (1<<0)  /* Verify existing pages in the mapping */
  50 #define MPOL_MF_MOVE    (1<<1)  /* Move pages owned by this process to conform to mapping */
  51 #define MPOL_MF_MOVE_ALL (1<<2) /* Move every page to conform to mapping */
  52 #define MPOL_MF_INTERNAL (1<<3) /* Internal flags start here */
  53
  54 /*
  55  * Internal flags that share the struct mempolicy flags word with
  56  * "mode flags".  These flags are allocated from bit 0 up, as they
  57  * are never OR'ed into the mode in mempolicy API arguments.
  58  */
  59 #define MPOL_F_SHARED  (1 << 0) /* identify shared policies */
  60 #define MPOL_F_LOCAL   (1 << 1) /* preferred local allocation */
  61 #define MPOL_F_REBINDING (1 << 2)       /* identify policies in rebinding */
  62
  63 #ifdef __KERNEL__
  64
  65 #include <linux/mmzone.h>
  66 #include <linux/slab.h>
  67 #include <linux/rbtree.h>
  68 #include <linux/spinlock.h>
  69 #include <linux/nodemask.h>
  70 #include <linux/pagemap.h>
  71
  72 struct mm_struct;
  73
  74 #ifdef CONFIG_NUMA
  75
  76 /*
  77  * Describe a memory policy.
  78  *
  79  * A mempolicy can be either associated with a process or with a VMA.
  80  * For VMA related allocations the VMA policy is preferred, otherwise
  81  * the process policy is used. Interrupts ignore the memory policy
  82  * of the current process.
  83  *
  84  * Locking policy for interlave:
  85  * In process context there is no locking because only the process accesses
  86  * its own state. All vma manipulation is somewhat protected by a down_read on
  87  * mmap_sem.
  88  *
  89  * Freeing policy:
  90  * Mempolicy objects are reference counted.  A mempolicy will be freed when
  91  * mpol_put() decrements the reference count to zero.
  92  *
  93  * Duplicating policy objects:
  94  * mpol_dup() allocates a new mempolicy and copies the specified mempolicy
  95  * to the new storage.  The reference count of the new object is initialized
  96  * to 1, representing the caller of mpol_dup().
  97  */
  98 struct mempolicy {
  99         atomic_t refcnt;
 100         unsigned short mode;    /* See MPOL_* above */
 101         unsigned short flags;   /* See set_mempolicy() MPOL_F_* above */
 102         union {
 103                 short            preferred_node; /* preferred */
 104                 nodemask_t       nodes;         /* interleave/bind */
 105                 /* undefined for default */
 106         } v;
 107         union {
 108                 nodemask_t cpuset_mems_allowed; /* relative to these nodes */
 109                 nodemask_t user_nodemask;       /* nodemask passed by user */
 110         } w;
 111 };
 112
 113 /*
 114  * Support for managing mempolicy data objects (clone, copy, destroy)
 115  * The default fast path of a NULL MPOL_DEFAULT policy is always inlined.
 116  */
 117
 118 extern void __mpol_put(struct mempolicy *pol);
 119 static inline void mpol_put(struct mempolicy *pol)
 120 {
 121         if (pol)
 122                 __mpol_put(pol);
 123 }
 124
 125 /*
 126  * Does mempolicy pol need explicit unref after use?
 127  * Currently only needed for shared policies.
 128  */
 129 static inline int mpol_needs_cond_ref(struct mempolicy *pol)
 130 {
 131         return (pol && (pol->flags & MPOL_F_SHARED));
 132 }
 133
 134 static inline void mpol_cond_put(struct mempolicy *pol)
 135 {
 136         if (mpol_needs_cond_ref(pol))
 137                 __mpol_put(pol);
 138 }
 139
 140 extern struct mempolicy *__mpol_cond_copy(struct mempolicy *tompol,
 141                                           struct mempolicy *frompol);
 142 static inline struct mempolicy *mpol_cond_copy(struct mempolicy *tompol,
 143                                                 struct mempolicy *frompol)
 144 {
 145         if (!frompol)
 146                 return frompol;
 147         return __mpol_cond_copy(tompol, frompol);
 148 }
 149
 150 extern struct mempolicy *__mpol_dup(struct mempolicy *pol);
 151 static inline struct mempolicy *mpol_dup(struct mempolicy *pol)
 152 {
 153         if (pol)
 154                 pol = __mpol_dup(pol);
 155         return pol;
 156 }
 157
 158 #define vma_policy(vma) ((vma)->vm_policy)
 159 #define vma_set_policy(vma, pol) ((vma)->vm_policy = (pol))
 160
 161 static inline void mpol_get(struct mempolicy *pol)
 162 {
 163         if (pol)
 164                 atomic_inc(&pol->refcnt);
 165 }
 166
 167 extern bool __mpol_equal(struct mempolicy *a, struct mempolicy *b);
 168 static inline bool mpol_equal(struct mempolicy *a, struct mempolicy *b)
 169 {
 170         if (a == b)
 171                 return true;
 172         return __mpol_equal(a, b);
 173 }
 174
 175 /*
 176  * Tree of shared policies for a shared memory region.
 177  * Maintain the policies in a pseudo mm that contains vmas. The vmas
 178  * carry the policy. As a special twist the pseudo mm is indexed in pages, not
 179  * bytes, so that we can work with shared memory segments bigger than
 180  * unsigned long.
 181  */
 182
 183 struct sp_node {
 184         struct rb_node nd;
 185         unsigned long start, end;
 186         struct mempolicy *policy;
 187 };
 188
 189 struct shared_policy {
 190         struct rb_root root;
 191         spinlock_t lock;
 192 };
 193
 194 void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol);
 195 int mpol_set_shared_policy(struct shared_policy *info,
 196                                 struct vm_area_struct *vma,
 197                                 struct mempolicy *new);
 198 void mpol_free_shared_policy(struct shared_policy *p);
 199 struct mempolicy *mpol_shared_policy_lookup(struct shared_policy *sp,
 200                                             unsigned long idx);
 201
 202 struct mempolicy *get_vma_policy(struct task_struct *tsk,
 203                 struct vm_area_struct *vma, unsigned long addr);
 204
 205 extern void numa_default_policy(void);
 206 extern void numa_policy_init(void);
 207 extern void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new,
 208                                 enum mpol_rebind_step step);
 209 extern void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new);
 210 extern void mpol_fix_fork_child_flag(struct task_struct *p);
 211
 212 extern struct zonelist *huge_zonelist(struct vm_area_struct *vma,
 213                                 unsigned long addr, gfp_t gfp_flags,
 214                                 struct mempolicy **mpol, nodemask_t **nodemask);
 215 extern bool init_nodemask_of_mempolicy(nodemask_t *mask);
 216 extern bool mempolicy_nodemask_intersects(struct task_struct *tsk,
 217                                 const nodemask_t *mask);
 218 extern unsigned slab_node(struct mempolicy *policy);
 219
 220 extern enum zone_type policy_zone;
 221
 222 static inline void check_highest_zone(enum zone_type k)
 223 {
 224         if (k > policy_zone && k != ZONE_MOVABLE)
 225                 policy_zone = k;
 226 }
 227
 228 int do_migrate_pages(struct mm_struct *mm,
 229         const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags);
 230
 231
 232 #ifdef CONFIG_TMPFS
 233 extern int mpol_parse_str(char *str, struct mempolicy **mpol, int no_context);
 234 #endif
 235
 236 extern int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol,
 237                         int no_context);
 238
 239 /* Check if a vma is migratable */
 240 static inline int vma_migratable(struct vm_area_struct *vma)
 241 {
 242         if (vma->vm_flags & (VM_IO|VM_HUGETLB|VM_PFNMAP|VM_RESERVED))
 243                 return 0;
 244         /*
 245          * Migration allocates pages in the highest zone. If we cannot
 246          * do so then migration (at least from node to node) is not
 247          * possible.
 248          */
 249         if (vma->vm_file &&
 250                 gfp_zone(mapping_gfp_mask(vma->vm_file->f_mapping))
 251                                                                 < policy_zone)
 252                         return 0;
 253         return 1;
 254 }
 255
 256 #else
 257
 258 struct mempolicy {};
 259
 260 static inline bool mpol_equal(struct mempolicy *a, struct mempolicy *b)
 261 {
 262         return true;
 263 }
 264
 265 static inline void mpol_put(struct mempolicy *p)
 266 {
 267 }
 268
 269 static inline void mpol_cond_put(struct mempolicy *pol)
 270 {
 271 }
 272
 273 static inline struct mempolicy *mpol_cond_copy(struct mempolicy *to,
 274                                                 struct mempolicy *from)
 275 {
 276         return from;
 277 }
 278
 279 static inline void mpol_get(struct mempolicy *pol)
 280 {
 281 }
 282
 283 static inline struct mempolicy *mpol_dup(struct mempolicy *old)
 284 {
 285         return NULL;
 286 }
 287
 288 struct shared_policy {};
 289
 290 static inline int mpol_set_shared_policy(struct shared_policy *info,
 291                                         struct vm_area_struct *vma,
 292                                         struct mempolicy *new)
 293 {
 294         return -EINVAL;
 295 }
 296
 297 static inline void mpol_shared_policy_init(struct shared_policy *sp,
 298                                                 struct mempolicy *mpol)
 299 {
 300 }
 301
 302 static inline void mpol_free_shared_policy(struct shared_policy *p)
 303 {
 304 }
 305
 306 static inline struct mempolicy *
 307 mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx)
 308 {
 309         return NULL;
 310 }
 311
 312 #define vma_policy(vma) NULL
 313 #define vma_set_policy(vma, pol) do {} while(0)
 314
 315 static inline void numa_policy_init(void)
 316 {
 317 }
 318
 319 static inline void numa_default_policy(void)
 320 {
 321 }
 322
 323 static inline void mpol_rebind_task(struct task_struct *tsk,
 324                                 const nodemask_t *new,
 325                                 enum mpol_rebind_step step)
 326 {
 327 }
 328
 329 static inline void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new)
 330 {
 331 }
 332
 333 static inline void mpol_fix_fork_child_flag(struct task_struct *p)
 334 {
 335 }
 336
 337 static inline struct zonelist *huge_zonelist(struct vm_area_struct *vma,
 338                                 unsigned long addr, gfp_t gfp_flags,
 339                                 struct mempolicy **mpol, nodemask_t **nodemask)
 340 {
 341         *mpol = NULL;
 342         *nodemask = NULL;
 343         return node_zonelist(0, gfp_flags);
 344 }
 345
 346 static inline bool init_nodemask_of_mempolicy(nodemask_t *m)
 347 {
 348         return false;
 349 }
 350
 351 static inline bool mempolicy_nodemask_intersects(struct task_struct *tsk,
 352                         const nodemask_t *mask)
 353 {
 354         return false;
 355 }
 356
 357 static inline int do_migrate_pages(struct mm_struct *mm,
 358                         const nodemask_t *from_nodes,
 359                         const nodemask_t *to_nodes, int flags)
 360 {
 361         return 0;
 362 }
 363
 364 static inline void check_highest_zone(int k)
 365 {
 366 }
 367
 368 #ifdef CONFIG_TMPFS
 369 static inline int mpol_parse_str(char *str, struct mempolicy **mpol,
 370                                 int no_context)
 371 {
 372         return 1;       /* error */
 373 }
 374 #endif
 375
 376 static inline int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol,
 377                                 int no_context)
 378 {
 379         return 0;
 380 }
 381
 382 #endif /* CONFIG_NUMA */
 383 #endif /* __KERNEL__ */
 384
 385 #endif