include/linux/mempolicy.h

   1 /*
   2  * NUMA memory policies for Linux.
   3  * Copyright 2003,2004 Andi Kleen SuSE Labs
   4  */
   5 #ifndef _LINUX_MEMPOLICY_H
   6 #define _LINUX_MEMPOLICY_H 1
   7
   8
   9 #include <linux/mmzone.h>
  10 #include <linux/slab.h>
  11 #include <linux/rbtree.h>
  12 #include <linux/spinlock.h>
  13 #include <linux/nodemask.h>
  14 #include <linux/pagemap.h>
  15 #include <uapi/linux/mempolicy.h>
  16
  17 struct mm_struct;
  18
  19 #ifdef CONFIG_NUMA
  20
  21 /*
  22  * Describe a memory policy.
  23  *
  24  * A mempolicy can be either associated with a process or with a VMA.
  25  * For VMA related allocations the VMA policy is preferred, otherwise
  26  * the process policy is used. Interrupts ignore the memory policy
  27  * of the current process.
  28  *
  29  * Locking policy for interlave:
  30  * In process context there is no locking because only the process accesses
  31  * its own state. All vma manipulation is somewhat protected by a down_read on
  32  * mmap_sem.
  33  *
  34  * Freeing policy:
  35  * Mempolicy objects are reference counted.  A mempolicy will be freed when
  36  * mpol_put() decrements the reference count to zero.
  37  *
  38  * Duplicating policy objects:
  39  * mpol_dup() allocates a new mempolicy and copies the specified mempolicy
  40  * to the new storage.  The reference count of the new object is initialized
  41  * to 1, representing the caller of mpol_dup().
  42  */
  43 struct mempolicy {
  44         atomic_t refcnt;
  45         unsigned short mode;    /* See MPOL_* above */
  46         unsigned short flags;   /* See set_mempolicy() MPOL_F_* above */
  47         union {
  48                 short            preferred_node; /* preferred */
  49                 nodemask_t       nodes;         /* interleave/bind */
  50                 /* undefined for default */
  51         } v;
  52         union {
  53                 nodemask_t cpuset_mems_allowed; /* relative to these nodes */
  54                 nodemask_t user_nodemask;       /* nodemask passed by user */
  55         } w;
  56 };
  57
  58 /*
  59  * Support for managing mempolicy data objects (clone, copy, destroy)
  60  * The default fast path of a NULL MPOL_DEFAULT policy is always inlined.
  61  */
  62
  63 extern void __mpol_put(struct mempolicy *pol);
  64 static inline void mpol_put(struct mempolicy *pol)
  65 {
  66         if (pol)
  67                 __mpol_put(pol);
  68 }
  69
  70 /*
  71  * Does mempolicy pol need explicit unref after use?
  72  * Currently only needed for shared policies.
  73  */
  74 static inline int mpol_needs_cond_ref(struct mempolicy *pol)
  75 {
  76         return (pol && (pol->flags & MPOL_F_SHARED));
  77 }
  78
  79 static inline void mpol_cond_put(struct mempolicy *pol)
  80 {
  81         if (mpol_needs_cond_ref(pol))
  82                 __mpol_put(pol);
  83 }
  84
  85 extern struct mempolicy *__mpol_dup(struct mempolicy *pol);
  86 static inline struct mempolicy *mpol_dup(struct mempolicy *pol)
  87 {
  88         if (pol)
  89                 pol = __mpol_dup(pol);
  90         return pol;
  91 }
  92
  93 #define vma_policy(vma) ((vma)->vm_policy)
  94 #define vma_set_policy(vma, pol) ((vma)->vm_policy = (pol))
  95
  96 static inline void mpol_get(struct mempolicy *pol)
  97 {
  98         if (pol)
  99                 atomic_inc(&pol->refcnt);
 100 }
 101
 102 extern bool __mpol_equal(struct mempolicy *a, struct mempolicy *b);
 103 static inline bool mpol_equal(struct mempolicy *a, struct mempolicy *b)
 104 {
 105         if (a == b)
 106                 return true;
 107         return __mpol_equal(a, b);
 108 }
 109
 110 /*
 111  * Tree of shared policies for a shared memory region.
 112  * Maintain the policies in a pseudo mm that contains vmas. The vmas
 113  * carry the policy. As a special twist the pseudo mm is indexed in pages, not
 114  * bytes, so that we can work with shared memory segments bigger than
 115  * unsigned long.
 116  */
 117
 118 struct sp_node {
 119         struct rb_node nd;
 120         unsigned long start, end;
 121         struct mempolicy *policy;
 122 };
 123
 124 struct shared_policy {
 125         struct rb_root root;
 126         spinlock_t lock;
 127 };
 128
 129 void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol);
 130 int mpol_set_shared_policy(struct shared_policy *info,
 131                                 struct vm_area_struct *vma,
 132                                 struct mempolicy *new);
 133 void mpol_free_shared_policy(struct shared_policy *p);
 134 struct mempolicy *mpol_shared_policy_lookup(struct shared_policy *sp,
 135                                             unsigned long idx);
 136
 137 struct mempolicy *get_vma_policy(struct task_struct *tsk,
 138                 struct vm_area_struct *vma, unsigned long addr);
 139
 140 extern void numa_default_policy(void);
 141 extern void numa_policy_init(void);
 142 extern void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new,
 143                                 enum mpol_rebind_step step);
 144 extern void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new);
 145 extern void mpol_fix_fork_child_flag(struct task_struct *p);
 146
 147 extern struct zonelist *huge_zonelist(struct vm_area_struct *vma,
 148                                 unsigned long addr, gfp_t gfp_flags,
 149                                 struct mempolicy **mpol, nodemask_t **nodemask);
 150 extern bool init_nodemask_of_mempolicy(nodemask_t *mask);
 151 extern bool mempolicy_nodemask_intersects(struct task_struct *tsk,
 152                                 const nodemask_t *mask);
 153 extern unsigned slab_node(void);
 154
 155 extern enum zone_type policy_zone;
 156
 157 static inline void check_highest_zone(enum zone_type k)
 158 {
 159         if (k > policy_zone && k != ZONE_MOVABLE)
 160                 policy_zone = k;
 161 }
 162
 163 int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
 164                      const nodemask_t *to, int flags);
 165
 166
 167 #ifdef CONFIG_TMPFS
 168 extern int mpol_parse_str(char *str, struct mempolicy **mpol);
 169 #endif
 170
 171 extern int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol);
 172
 173 /* Check if a vma is migratable */
 174 static inline int vma_migratable(struct vm_area_struct *vma)
 175 {
 176         if (vma->vm_flags & (VM_IO | VM_HUGETLB | VM_PFNMAP))
 177                 return 0;
 178         /*
 179          * Migration allocates pages in the highest zone. If we cannot
 180          * do so then migration (at least from node to node) is not
 181          * possible.
 182          */
 183         if (vma->vm_file &&
 184                 gfp_zone(mapping_gfp_mask(vma->vm_file->f_mapping))
 185                                                                 < policy_zone)
 186                         return 0;
 187         return 1;
 188 }
 189
 190 extern int mpol_misplaced(struct page *, struct vm_area_struct *, unsigned long);
 191
 192 #else
 193
 194 struct mempolicy {};
 195
 196 static inline bool mpol_equal(struct mempolicy *a, struct mempolicy *b)
 197 {
 198         return true;
 199 }
 200
 201 static inline void mpol_put(struct mempolicy *p)
 202 {
 203 }
 204
 205 static inline void mpol_cond_put(struct mempolicy *pol)
 206 {
 207 }
 208
 209 static inline void mpol_get(struct mempolicy *pol)
 210 {
 211 }
 212
 213 static inline struct mempolicy *mpol_dup(struct mempolicy *old)
 214 {
 215         return NULL;
 216 }
 217
 218 struct shared_policy {};
 219
 220 static inline int mpol_set_shared_policy(struct shared_policy *info,
 221                                         struct vm_area_struct *vma,
 222                                         struct mempolicy *new)
 223 {
 224         return -EINVAL;
 225 }
 226
 227 static inline void mpol_shared_policy_init(struct shared_policy *sp,
 228                                                 struct mempolicy *mpol)
 229 {
 230 }
 231
 232 static inline void mpol_free_shared_policy(struct shared_policy *p)
 233 {
 234 }
 235
 236 static inline struct mempolicy *
 237 mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx)
 238 {
 239         return NULL;
 240 }
 241
 242 #define vma_policy(vma) NULL
 243 #define vma_set_policy(vma, pol) do {} while(0)
 244
 245 static inline void numa_policy_init(void)
 246 {
 247 }
 248
 249 static inline void numa_default_policy(void)
 250 {
 251 }
 252
 253 static inline void mpol_rebind_task(struct task_struct *tsk,
 254                                 const nodemask_t *new,
 255                                 enum mpol_rebind_step step)
 256 {
 257 }
 258
 259 static inline void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new)
 260 {
 261 }
 262
 263 static inline void mpol_fix_fork_child_flag(struct task_struct *p)
 264 {
 265 }
 266
 267 static inline struct zonelist *huge_zonelist(struct vm_area_struct *vma,
 268                                 unsigned long addr, gfp_t gfp_flags,
 269                                 struct mempolicy **mpol, nodemask_t **nodemask)
 270 {
 271         *mpol = NULL;
 272         *nodemask = NULL;
 273         return node_zonelist(0, gfp_flags);
 274 }
 275
 276 static inline bool init_nodemask_of_mempolicy(nodemask_t *m)
 277 {
 278         return false;
 279 }
 280
 281 static inline bool mempolicy_nodemask_intersects(struct task_struct *tsk,
 282                         const nodemask_t *mask)
 283 {
 284         return false;
 285 }
 286
 287 static inline int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
 288                                    const nodemask_t *to, int flags)
 289 {
 290         return 0;
 291 }
 292
 293 static inline void check_highest_zone(int k)
 294 {
 295 }
 296
 297 #ifdef CONFIG_TMPFS
 298 static inline int mpol_parse_str(char *str, struct mempolicy **mpol)
 299 {
 300         return 1;       /* error */
 301 }
 302 #endif
 303
 304 static inline int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol)
 305 {
 306         return 0;
 307 }
 308
 309 static inline int mpol_misplaced(struct page *page, struct vm_area_struct *vma,
 310                                  unsigned long address)
 311 {
 312         return -1; /* no node preference */
 313 }
 314
 315 #endif /* CONFIG_NUMA */
 316 #endif