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 #include <linux/pagemap.h>
  63
  64 struct mm_struct;
  65
  66 #ifdef CONFIG_NUMA
  67
  68 /*
  69  * Describe a memory policy.
  70  *
  71  * A mempolicy can be either associated with a process or with a VMA.
  72  * For VMA related allocations the VMA policy is preferred, otherwise
  73  * the process policy is used. Interrupts ignore the memory policy
  74  * of the current process.
  75  *
  76  * Locking policy for interlave:
  77  * In process context there is no locking because only the process accesses
  78  * its own state. All vma manipulation is somewhat protected by a down_read on
  79  * mmap_sem.
  80  *
  81  * Freeing policy:
  82  * Mempolicy objects are reference counted.  A mempolicy will be freed when
  83  * mpol_put() decrements the reference count to zero.
  84  *
  85  * Duplicating policy objects:
  86  * mpol_dup() allocates a new mempolicy and copies the specified mempolicy
  87  * to the new storage.  The reference count of the new object is initialized
  88  * to 1, representing the caller of mpol_dup().
  89  */
  90 struct mempolicy {
  91         atomic_t refcnt;
  92         unsigned short mode;    /* See MPOL_* above */
  93         unsigned short flags;   /* See set_mempolicy() MPOL_F_* above */
  94         union {
  95                 short            preferred_node; /* preferred */
  96                 nodemask_t       nodes;         /* interleave/bind */
  97                 /* undefined for default */
  98         } v;
  99         union {
 100                 nodemask_t cpuset_mems_allowed; /* relative to these nodes */
 101                 nodemask_t user_nodemask;       /* nodemask passed by user */
 102         } w;
 103 };
 104
 105 /*
 106  * Support for managing mempolicy data objects (clone, copy, destroy)
 107  * The default fast path of a NULL MPOL_DEFAULT policy is always inlined.
 108  */
 109
 110 extern void __mpol_put(struct mempolicy *pol);
 111 static inline void mpol_put(struct mempolicy *pol)
 112 {
 113         if (pol)
 114                 __mpol_put(pol);
 115 }
 116
 117 /*
 118  * Does mempolicy pol need explicit unref after use?
 119  * Currently only needed for shared policies.
 120  */
 121 static inline int mpol_needs_cond_ref(struct mempolicy *pol)
 122 {
 123         return (pol && (pol->flags & MPOL_F_SHARED));
 124 }
 125
 126 static inline void mpol_cond_put(struct mempolicy *pol)
 127 {
 128         if (mpol_needs_cond_ref(pol))
 129                 __mpol_put(pol);
 130 }
 131
 132 extern struct mempolicy *__mpol_cond_copy(struct mempolicy *tompol,
 133                                           struct mempolicy *frompol);
 134 static inline struct mempolicy *mpol_cond_copy(struct mempolicy *tompol,
 135                                                 struct mempolicy *frompol)
 136 {
 137         if (!frompol)
 138                 return frompol;
 139         return __mpol_cond_copy(tompol, frompol);
 140 }
 141
 142 extern struct mempolicy *__mpol_dup(struct mempolicy *pol);
 143 static inline struct mempolicy *mpol_dup(struct mempolicy *pol)
 144 {
 145         if (pol)
 146                 pol = __mpol_dup(pol);
 147         return pol;
 148 }
 149
 150 #define vma_policy(vma) ((vma)->vm_policy)
 151 #define vma_set_policy(vma, pol) ((vma)->vm_policy = (pol))
 152
 153 static inline void mpol_get(struct mempolicy *pol)
 154 {
 155         if (pol)
 156                 atomic_inc(&pol->refcnt);
 157 }
 158
 159 extern int __mpol_equal(struct mempolicy *a, struct mempolicy *b);
 160 static inline int mpol_equal(struct mempolicy *a, struct mempolicy *b)
 161 {
 162         if (a == b)
 163                 return 1;
 164         return __mpol_equal(a, b);
 165 }
 166
 167 /*
 168  * Tree of shared policies for a shared memory region.
 169  * Maintain the policies in a pseudo mm that contains vmas. The vmas
 170  * carry the policy. As a special twist the pseudo mm is indexed in pages, not
 171  * bytes, so that we can work with shared memory segments bigger than
 172  * unsigned long.
 173  */
 174
 175 struct sp_node {
 176         struct rb_node nd;
 177         unsigned long start, end;
 178         struct mempolicy *policy;
 179 };
 180
 181 struct shared_policy {
 182         struct rb_root root;
 183         spinlock_t lock;
 184 };
 185
 186 void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol);
 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, struct mempolicy **mpol, int no_context);
 220
 221 extern int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol,
 222                         int no_context);
 223 #endif
 224
 225 /* Check if a vma is migratable */
 226 static inline int vma_migratable(struct vm_area_struct *vma)
 227 {
 228         if (vma->vm_flags & (VM_IO|VM_HUGETLB|VM_PFNMAP|VM_RESERVED))
 229                 return 0;
 230         /*
 231          * Migration allocates pages in the highest zone. If we cannot
 232          * do so then migration (at least from node to node) is not
 233          * possible.
 234          */
 235         if (vma->vm_file &&
 236                 gfp_zone(mapping_gfp_mask(vma->vm_file->f_mapping))
 237                                                                 < policy_zone)
 238                         return 0;
 239         return 1;
 240 }
 241
 242 #else
 243
 244 struct mempolicy {};
 245
 246 static inline int mpol_equal(struct mempolicy *a, struct mempolicy *b)
 247 {
 248         return 1;
 249 }
 250
 251 static inline void mpol_put(struct mempolicy *p)
 252 {
 253 }
 254
 255 static inline void mpol_cond_put(struct mempolicy *pol)
 256 {
 257 }
 258
 259 static inline struct mempolicy *mpol_cond_copy(struct mempolicy *to,
 260                                                 struct mempolicy *from)
 261 {
 262         return from;
 263 }
 264
 265 static inline void mpol_get(struct mempolicy *pol)
 266 {
 267 }
 268
 269 static inline struct mempolicy *mpol_dup(struct mempolicy *old)
 270 {
 271         return NULL;
 272 }
 273
 274 struct shared_policy {};
 275
 276 static inline int mpol_set_shared_policy(struct shared_policy *info,
 277                                         struct vm_area_struct *vma,
 278                                         struct mempolicy *new)
 279 {
 280         return -EINVAL;
 281 }
 282
 283 static inline void mpol_shared_policy_init(struct shared_policy *sp,
 284                                                 struct mempolicy *mpol)
 285 {
 286 }
 287
 288 static inline void mpol_free_shared_policy(struct shared_policy *p)
 289 {
 290 }
 291
 292 static inline struct mempolicy *
 293 mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx)
 294 {
 295         return NULL;
 296 }
 297
 298 #define vma_policy(vma) NULL
 299 #define vma_set_policy(vma, pol) do {} while(0)
 300
 301 static inline void numa_policy_init(void)
 302 {
 303 }
 304
 305 static inline void numa_default_policy(void)
 306 {
 307 }
 308
 309 static inline void mpol_rebind_task(struct task_struct *tsk,
 310                                         const nodemask_t *new)
 311 {
 312 }
 313
 314 static inline void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new)
 315 {
 316 }
 317
 318 static inline void mpol_fix_fork_child_flag(struct task_struct *p)
 319 {
 320 }
 321
 322 static inline struct zonelist *huge_zonelist(struct vm_area_struct *vma,
 323                                 unsigned long addr, gfp_t gfp_flags,
 324                                 struct mempolicy **mpol, nodemask_t **nodemask)
 325 {
 326         *mpol = NULL;
 327         *nodemask = NULL;
 328         return node_zonelist(0, gfp_flags);
 329 }
 330
 331 static inline int do_migrate_pages(struct mm_struct *mm,
 332                         const nodemask_t *from_nodes,
 333                         const nodemask_t *to_nodes, int flags)
 334 {
 335         return 0;
 336 }
 337
 338 static inline void check_highest_zone(int k)
 339 {
 340 }
 341
 342 #ifdef CONFIG_TMPFS
 343 static inline int mpol_parse_str(char *str, struct mempolicy **mpol,
 344                                 int no_context)
 345 {
 346         return 1;       /* error */
 347 }
 348
 349 static inline int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol,
 350                                 int no_context)
 351 {
 352         return 0;
 353 }
 354 #endif
 355
 356 #endif /* CONFIG_NUMA */
 357 #endif /* __KERNEL__ */
 358
 359 #endif