| 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
| 2 | /* |
| 3 | * NUMA memory policies for Linux. |
| 4 | * Copyright 2003,2004 Andi Kleen SuSE Labs |
| 5 | */ |
| 6 | #ifndef _LINUX_MEMPOLICY_H |
| 7 | #define _LINUX_MEMPOLICY_H 1 |
| 8 | |
| 9 | |
| 10 | #include <linux/mmzone.h> |
| 11 | #include <linux/dax.h> |
| 12 | #include <linux/slab.h> |
| 13 | #include <linux/rbtree.h> |
| 14 | #include <linux/spinlock.h> |
| 15 | #include <linux/nodemask.h> |
| 16 | #include <linux/pagemap.h> |
| 17 | #include <uapi/linux/mempolicy.h> |
| 18 | |
| 19 | struct mm_struct; |
| 20 | |
| 21 | #ifdef CONFIG_NUMA |
| 22 | |
| 23 | /* |
| 24 | * Describe a memory policy. |
| 25 | * |
| 26 | * A mempolicy can be either associated with a process or with a VMA. |
| 27 | * For VMA related allocations the VMA policy is preferred, otherwise |
| 28 | * the process policy is used. Interrupts ignore the memory policy |
| 29 | * of the current process. |
| 30 | * |
| 31 | * Locking policy for interlave: |
| 32 | * In process context there is no locking because only the process accesses |
| 33 | * its own state. All vma manipulation is somewhat protected by a down_read on |
| 34 | * mmap_sem. |
| 35 | * |
| 36 | * Freeing policy: |
| 37 | * Mempolicy objects are reference counted. A mempolicy will be freed when |
| 38 | * mpol_put() decrements the reference count to zero. |
| 39 | * |
| 40 | * Duplicating policy objects: |
| 41 | * mpol_dup() allocates a new mempolicy and copies the specified mempolicy |
| 42 | * to the new storage. The reference count of the new object is initialized |
| 43 | * to 1, representing the caller of mpol_dup(). |
| 44 | */ |
| 45 | struct mempolicy { |
| 46 | atomic_t refcnt; |
| 47 | unsigned short mode; /* See MPOL_* above */ |
| 48 | unsigned short flags; /* See set_mempolicy() MPOL_F_* above */ |
| 49 | union { |
| 50 | short preferred_node; /* preferred */ |
| 51 | nodemask_t nodes; /* interleave/bind */ |
| 52 | /* undefined for default */ |
| 53 | } v; |
| 54 | union { |
| 55 | nodemask_t cpuset_mems_allowed; /* relative to these nodes */ |
| 56 | nodemask_t user_nodemask; /* nodemask passed by user */ |
| 57 | } w; |
| 58 | }; |
| 59 | |
| 60 | /* |
| 61 | * Support for managing mempolicy data objects (clone, copy, destroy) |
| 62 | * The default fast path of a NULL MPOL_DEFAULT policy is always inlined. |
| 63 | */ |
| 64 | |
| 65 | extern void __mpol_put(struct mempolicy *pol); |
| 66 | static inline void mpol_put(struct mempolicy *pol) |
| 67 | { |
| 68 | if (pol) |
| 69 | __mpol_put(pol); |
| 70 | } |
| 71 | |
| 72 | /* |
| 73 | * Does mempolicy pol need explicit unref after use? |
| 74 | * Currently only needed for shared policies. |
| 75 | */ |
| 76 | static inline int mpol_needs_cond_ref(struct mempolicy *pol) |
| 77 | { |
| 78 | return (pol && (pol->flags & MPOL_F_SHARED)); |
| 79 | } |
| 80 | |
| 81 | static inline void mpol_cond_put(struct mempolicy *pol) |
| 82 | { |
| 83 | if (mpol_needs_cond_ref(pol)) |
| 84 | __mpol_put(pol); |
| 85 | } |
| 86 | |
| 87 | extern struct mempolicy *__mpol_dup(struct mempolicy *pol); |
| 88 | static inline struct mempolicy *mpol_dup(struct mempolicy *pol) |
| 89 | { |
| 90 | if (pol) |
| 91 | pol = __mpol_dup(pol); |
| 92 | return pol; |
| 93 | } |
| 94 | |
| 95 | #define vma_policy(vma) ((vma)->vm_policy) |
| 96 | |
| 97 | static inline void mpol_get(struct mempolicy *pol) |
| 98 | { |
| 99 | if (pol) |
| 100 | atomic_inc(&pol->refcnt); |
| 101 | } |
| 102 | |
| 103 | extern bool __mpol_equal(struct mempolicy *a, struct mempolicy *b); |
| 104 | static inline bool mpol_equal(struct mempolicy *a, struct mempolicy *b) |
| 105 | { |
| 106 | if (a == b) |
| 107 | return true; |
| 108 | return __mpol_equal(a, b); |
| 109 | } |
| 110 | |
| 111 | /* |
| 112 | * Tree of shared policies for a shared memory region. |
| 113 | * Maintain the policies in a pseudo mm that contains vmas. The vmas |
| 114 | * carry the policy. As a special twist the pseudo mm is indexed in pages, not |
| 115 | * bytes, so that we can work with shared memory segments bigger than |
| 116 | * unsigned long. |
| 117 | */ |
| 118 | |
| 119 | struct sp_node { |
| 120 | struct rb_node nd; |
| 121 | unsigned long start, end; |
| 122 | struct mempolicy *policy; |
| 123 | }; |
| 124 | |
| 125 | struct shared_policy { |
| 126 | struct rb_root root; |
| 127 | rwlock_t lock; |
| 128 | }; |
| 129 | |
| 130 | int vma_dup_policy(struct vm_area_struct *src, struct vm_area_struct *dst); |
| 131 | void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol); |
| 132 | int mpol_set_shared_policy(struct shared_policy *info, |
| 133 | struct vm_area_struct *vma, |
| 134 | struct mempolicy *new); |
| 135 | void mpol_free_shared_policy(struct shared_policy *p); |
| 136 | struct mempolicy *mpol_shared_policy_lookup(struct shared_policy *sp, |
| 137 | unsigned long idx); |
| 138 | |
| 139 | struct mempolicy *get_task_policy(struct task_struct *p); |
| 140 | struct mempolicy *__get_vma_policy(struct vm_area_struct *vma, |
| 141 | unsigned long addr); |
| 142 | bool vma_policy_mof(struct vm_area_struct *vma); |
| 143 | |
| 144 | extern void numa_default_policy(void); |
| 145 | extern void numa_policy_init(void); |
| 146 | extern void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new); |
| 147 | extern void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new); |
| 148 | |
| 149 | extern int huge_node(struct vm_area_struct *vma, |
| 150 | unsigned long addr, gfp_t gfp_flags, |
| 151 | struct mempolicy **mpol, nodemask_t **nodemask); |
| 152 | extern bool init_nodemask_of_mempolicy(nodemask_t *mask); |
| 153 | extern bool mempolicy_nodemask_intersects(struct task_struct *tsk, |
| 154 | const nodemask_t *mask); |
| 155 | extern unsigned int mempolicy_slab_node(void); |
| 156 | |
| 157 | extern enum zone_type policy_zone; |
| 158 | |
| 159 | static inline void check_highest_zone(enum zone_type k) |
| 160 | { |
| 161 | if (k > policy_zone && k != ZONE_MOVABLE) |
| 162 | policy_zone = k; |
| 163 | } |
| 164 | |
| 165 | int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from, |
| 166 | const nodemask_t *to, int flags); |
| 167 | |
| 168 | |
| 169 | #ifdef CONFIG_TMPFS |
| 170 | extern int mpol_parse_str(char *str, struct mempolicy **mpol); |
| 171 | #endif |
| 172 | |
| 173 | extern void mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol); |
| 174 | |
| 175 | /* Check if a vma is migratable */ |
| 176 | static inline bool vma_migratable(struct vm_area_struct *vma) |
| 177 | { |
| 178 | if (vma->vm_flags & (VM_IO | VM_PFNMAP)) |
| 179 | return false; |
| 180 | |
| 181 | /* |
| 182 | * DAX device mappings require predictable access latency, so avoid |
| 183 | * incurring periodic faults. |
| 184 | */ |
| 185 | if (vma_is_dax(vma)) |
| 186 | return false; |
| 187 | |
| 188 | #ifndef CONFIG_ARCH_ENABLE_HUGEPAGE_MIGRATION |
| 189 | if (vma->vm_flags & VM_HUGETLB) |
| 190 | return false; |
| 191 | #endif |
| 192 | |
| 193 | /* |
| 194 | * Migration allocates pages in the highest zone. If we cannot |
| 195 | * do so then migration (at least from node to node) is not |
| 196 | * possible. |
| 197 | */ |
| 198 | if (vma->vm_file && |
| 199 | gfp_zone(mapping_gfp_mask(vma->vm_file->f_mapping)) |
| 200 | < policy_zone) |
| 201 | return false; |
| 202 | return true; |
| 203 | } |
| 204 | |
| 205 | extern int mpol_misplaced(struct page *, struct vm_area_struct *, unsigned long); |
| 206 | extern void mpol_put_task_policy(struct task_struct *); |
| 207 | |
| 208 | #else |
| 209 | |
| 210 | struct mempolicy {}; |
| 211 | |
| 212 | static inline bool mpol_equal(struct mempolicy *a, struct mempolicy *b) |
| 213 | { |
| 214 | return true; |
| 215 | } |
| 216 | |
| 217 | static inline void mpol_put(struct mempolicy *p) |
| 218 | { |
| 219 | } |
| 220 | |
| 221 | static inline void mpol_cond_put(struct mempolicy *pol) |
| 222 | { |
| 223 | } |
| 224 | |
| 225 | static inline void mpol_get(struct mempolicy *pol) |
| 226 | { |
| 227 | } |
| 228 | |
| 229 | struct shared_policy {}; |
| 230 | |
| 231 | static inline void mpol_shared_policy_init(struct shared_policy *sp, |
| 232 | struct mempolicy *mpol) |
| 233 | { |
| 234 | } |
| 235 | |
| 236 | static inline void mpol_free_shared_policy(struct shared_policy *p) |
| 237 | { |
| 238 | } |
| 239 | |
| 240 | static inline struct mempolicy * |
| 241 | mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx) |
| 242 | { |
| 243 | return NULL; |
| 244 | } |
| 245 | |
| 246 | #define vma_policy(vma) NULL |
| 247 | |
| 248 | static inline int |
| 249 | vma_dup_policy(struct vm_area_struct *src, struct vm_area_struct *dst) |
| 250 | { |
| 251 | return 0; |
| 252 | } |
| 253 | |
| 254 | static inline void numa_policy_init(void) |
| 255 | { |
| 256 | } |
| 257 | |
| 258 | static inline void numa_default_policy(void) |
| 259 | { |
| 260 | } |
| 261 | |
| 262 | static inline void mpol_rebind_task(struct task_struct *tsk, |
| 263 | const nodemask_t *new) |
| 264 | { |
| 265 | } |
| 266 | |
| 267 | static inline void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new) |
| 268 | { |
| 269 | } |
| 270 | |
| 271 | static inline int huge_node(struct vm_area_struct *vma, |
| 272 | unsigned long addr, gfp_t gfp_flags, |
| 273 | struct mempolicy **mpol, nodemask_t **nodemask) |
| 274 | { |
| 275 | *mpol = NULL; |
| 276 | *nodemask = NULL; |
| 277 | return 0; |
| 278 | } |
| 279 | |
| 280 | static inline bool init_nodemask_of_mempolicy(nodemask_t *m) |
| 281 | { |
| 282 | return false; |
| 283 | } |
| 284 | |
| 285 | static inline int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from, |
| 286 | const nodemask_t *to, int flags) |
| 287 | { |
| 288 | return 0; |
| 289 | } |
| 290 | |
| 291 | static inline void check_highest_zone(int k) |
| 292 | { |
| 293 | } |
| 294 | |
| 295 | #ifdef CONFIG_TMPFS |
| 296 | static inline int mpol_parse_str(char *str, struct mempolicy **mpol) |
| 297 | { |
| 298 | return 1; /* error */ |
| 299 | } |
| 300 | #endif |
| 301 | |
| 302 | static inline int mpol_misplaced(struct page *page, struct vm_area_struct *vma, |
| 303 | unsigned long address) |
| 304 | { |
| 305 | return -1; /* no node preference */ |
| 306 | } |
| 307 | |
| 308 | static inline void mpol_put_task_policy(struct task_struct *task) |
| 309 | { |
| 310 | } |
| 311 | #endif /* CONFIG_NUMA */ |
| 312 | #endif |