1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef _LINUX_SCHED_MM_H 3 #define _LINUX_SCHED_MM_H 4 5 #include <linux/kernel.h> 6 #include <linux/atomic.h> 7 #include <linux/sched.h> 8 #include <linux/mm_types.h> 9 #include <linux/gfp.h> 10 #include <linux/sync_core.h> 11 12 /* 13 * Routines for handling mm_structs 14 */ 15 extern struct mm_struct *mm_alloc(void); 16 17 /** 18 * mmgrab() - Pin a &struct mm_struct. 19 * @mm: The &struct mm_struct to pin. 20 * 21 * Make sure that @mm will not get freed even after the owning task 22 * exits. This doesn't guarantee that the associated address space 23 * will still exist later on and mmget_not_zero() has to be used before 24 * accessing it. 25 * 26 * This is a preferred way to pin @mm for a longer/unbounded amount 27 * of time. 28 * 29 * Use mmdrop() to release the reference acquired by mmgrab(). 30 * 31 * See also <Documentation/vm/active_mm.rst> for an in-depth explanation 32 * of &mm_struct.mm_count vs &mm_struct.mm_users. 33 */ 34 static inline void mmgrab(struct mm_struct *mm) 35 { 36 atomic_inc(&mm->mm_count); 37 } 38 39 extern void __mmdrop(struct mm_struct *mm); 40 41 static inline void mmdrop(struct mm_struct *mm) 42 { 43 /* 44 * The implicit full barrier implied by atomic_dec_and_test() is 45 * required by the membarrier system call before returning to 46 * user-space, after storing to rq->curr. 47 */ 48 if (unlikely(atomic_dec_and_test(&mm->mm_count))) 49 __mmdrop(mm); 50 } 51 52 #ifdef CONFIG_PREEMPT_RT 53 /* 54 * RCU callback for delayed mm drop. Not strictly RCU, but call_rcu() is 55 * by far the least expensive way to do that. 56 */ 57 static inline void __mmdrop_delayed(struct rcu_head *rhp) 58 { 59 struct mm_struct *mm = container_of(rhp, struct mm_struct, delayed_drop); 60 61 __mmdrop(mm); 62 } 63 64 /* 65 * Invoked from finish_task_switch(). Delegates the heavy lifting on RT 66 * kernels via RCU. 67 */ 68 static inline void mmdrop_sched(struct mm_struct *mm) 69 { 70 /* Provides a full memory barrier. See mmdrop() */ 71 if (atomic_dec_and_test(&mm->mm_count)) 72 call_rcu(&mm->delayed_drop, __mmdrop_delayed); 73 } 74 #else 75 static inline void mmdrop_sched(struct mm_struct *mm) 76 { 77 mmdrop(mm); 78 } 79 #endif 80 81 /** 82 * mmget() - Pin the address space associated with a &struct mm_struct. 83 * @mm: The address space to pin. 84 * 85 * Make sure that the address space of the given &struct mm_struct doesn't 86 * go away. This does not protect against parts of the address space being 87 * modified or freed, however. 88 * 89 * Never use this function to pin this address space for an 90 * unbounded/indefinite amount of time. 91 * 92 * Use mmput() to release the reference acquired by mmget(). 93 * 94 * See also <Documentation/vm/active_mm.rst> for an in-depth explanation 95 * of &mm_struct.mm_count vs &mm_struct.mm_users. 96 */ 97 static inline void mmget(struct mm_struct *mm) 98 { 99 atomic_inc(&mm->mm_users); 100 } 101 102 static inline bool mmget_not_zero(struct mm_struct *mm) 103 { 104 return atomic_inc_not_zero(&mm->mm_users); 105 } 106 107 /* mmput gets rid of the mappings and all user-space */ 108 extern void mmput(struct mm_struct *); 109 #ifdef CONFIG_MMU 110 /* same as above but performs the slow path from the async context. Can 111 * be called from the atomic context as well 112 */ 113 void mmput_async(struct mm_struct *); 114 #endif 115 116 /* Grab a reference to a task's mm, if it is not already going away */ 117 extern struct mm_struct *get_task_mm(struct task_struct *task); 118 /* 119 * Grab a reference to a task's mm, if it is not already going away 120 * and ptrace_may_access with the mode parameter passed to it 121 * succeeds. 122 */ 123 extern struct mm_struct *mm_access(struct task_struct *task, unsigned int mode); 124 /* Remove the current tasks stale references to the old mm_struct on exit() */ 125 extern void exit_mm_release(struct task_struct *, struct mm_struct *); 126 /* Remove the current tasks stale references to the old mm_struct on exec() */ 127 extern void exec_mm_release(struct task_struct *, struct mm_struct *); 128 129 #ifdef CONFIG_MEMCG 130 extern void mm_update_next_owner(struct mm_struct *mm); 131 #else 132 static inline void mm_update_next_owner(struct mm_struct *mm) 133 { 134 } 135 #endif /* CONFIG_MEMCG */ 136 137 #ifdef CONFIG_MMU 138 extern void arch_pick_mmap_layout(struct mm_struct *mm, 139 struct rlimit *rlim_stack); 140 extern unsigned long 141 arch_get_unmapped_area(struct file *, unsigned long, unsigned long, 142 unsigned long, unsigned long); 143 extern unsigned long 144 arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr, 145 unsigned long len, unsigned long pgoff, 146 unsigned long flags); 147 #else 148 static inline void arch_pick_mmap_layout(struct mm_struct *mm, 149 struct rlimit *rlim_stack) {} 150 #endif 151 152 static inline bool in_vfork(struct task_struct *tsk) 153 { 154 bool ret; 155 156 /* 157 * need RCU to access ->real_parent if CLONE_VM was used along with 158 * CLONE_PARENT. 159 * 160 * We check real_parent->mm == tsk->mm because CLONE_VFORK does not 161 * imply CLONE_VM 162 * 163 * CLONE_VFORK can be used with CLONE_PARENT/CLONE_THREAD and thus 164 * ->real_parent is not necessarily the task doing vfork(), so in 165 * theory we can't rely on task_lock() if we want to dereference it. 166 * 167 * And in this case we can't trust the real_parent->mm == tsk->mm 168 * check, it can be false negative. But we do not care, if init or 169 * another oom-unkillable task does this it should blame itself. 170 */ 171 rcu_read_lock(); 172 ret = tsk->vfork_done && 173 rcu_dereference(tsk->real_parent)->mm == tsk->mm; 174 rcu_read_unlock(); 175 176 return ret; 177 } 178 179 /* 180 * Applies per-task gfp context to the given allocation flags. 181 * PF_MEMALLOC_NOIO implies GFP_NOIO 182 * PF_MEMALLOC_NOFS implies GFP_NOFS 183 * PF_MEMALLOC_PIN implies !GFP_MOVABLE 184 */ 185 static inline gfp_t current_gfp_context(gfp_t flags) 186 { 187 unsigned int pflags = READ_ONCE(current->flags); 188 189 if (unlikely(pflags & (PF_MEMALLOC_NOIO | PF_MEMALLOC_NOFS | PF_MEMALLOC_PIN))) { 190 /* 191 * NOIO implies both NOIO and NOFS and it is a weaker context 192 * so always make sure it makes precedence 193 */ 194 if (pflags & PF_MEMALLOC_NOIO) 195 flags &= ~(__GFP_IO | __GFP_FS); 196 else if (pflags & PF_MEMALLOC_NOFS) 197 flags &= ~__GFP_FS; 198 199 if (pflags & PF_MEMALLOC_PIN) 200 flags &= ~__GFP_MOVABLE; 201 } 202 return flags; 203 } 204 205 #ifdef CONFIG_LOCKDEP 206 extern void __fs_reclaim_acquire(unsigned long ip); 207 extern void __fs_reclaim_release(unsigned long ip); 208 extern void fs_reclaim_acquire(gfp_t gfp_mask); 209 extern void fs_reclaim_release(gfp_t gfp_mask); 210 #else 211 static inline void __fs_reclaim_acquire(unsigned long ip) { } 212 static inline void __fs_reclaim_release(unsigned long ip) { } 213 static inline void fs_reclaim_acquire(gfp_t gfp_mask) { } 214 static inline void fs_reclaim_release(gfp_t gfp_mask) { } 215 #endif 216 217 /** 218 * might_alloc - Mark possible allocation sites 219 * @gfp_mask: gfp_t flags that would be used to allocate 220 * 221 * Similar to might_sleep() and other annotations, this can be used in functions 222 * that might allocate, but often don't. Compiles to nothing without 223 * CONFIG_LOCKDEP. Includes a conditional might_sleep() if @gfp allows blocking. 224 */ 225 static inline void might_alloc(gfp_t gfp_mask) 226 { 227 fs_reclaim_acquire(gfp_mask); 228 fs_reclaim_release(gfp_mask); 229 230 might_sleep_if(gfpflags_allow_blocking(gfp_mask)); 231 } 232 233 /** 234 * memalloc_noio_save - Marks implicit GFP_NOIO allocation scope. 235 * 236 * This functions marks the beginning of the GFP_NOIO allocation scope. 237 * All further allocations will implicitly drop __GFP_IO flag and so 238 * they are safe for the IO critical section from the allocation recursion 239 * point of view. Use memalloc_noio_restore to end the scope with flags 240 * returned by this function. 241 * 242 * This function is safe to be used from any context. 243 */ 244 static inline unsigned int memalloc_noio_save(void) 245 { 246 unsigned int flags = current->flags & PF_MEMALLOC_NOIO; 247 current->flags |= PF_MEMALLOC_NOIO; 248 return flags; 249 } 250 251 /** 252 * memalloc_noio_restore - Ends the implicit GFP_NOIO scope. 253 * @flags: Flags to restore. 254 * 255 * Ends the implicit GFP_NOIO scope started by memalloc_noio_save function. 256 * Always make sure that the given flags is the return value from the 257 * pairing memalloc_noio_save call. 258 */ 259 static inline void memalloc_noio_restore(unsigned int flags) 260 { 261 current->flags = (current->flags & ~PF_MEMALLOC_NOIO) | flags; 262 } 263 264 /** 265 * memalloc_nofs_save - Marks implicit GFP_NOFS allocation scope. 266 * 267 * This functions marks the beginning of the GFP_NOFS allocation scope. 268 * All further allocations will implicitly drop __GFP_FS flag and so 269 * they are safe for the FS critical section from the allocation recursion 270 * point of view. Use memalloc_nofs_restore to end the scope with flags 271 * returned by this function. 272 * 273 * This function is safe to be used from any context. 274 */ 275 static inline unsigned int memalloc_nofs_save(void) 276 { 277 unsigned int flags = current->flags & PF_MEMALLOC_NOFS; 278 current->flags |= PF_MEMALLOC_NOFS; 279 return flags; 280 } 281 282 /** 283 * memalloc_nofs_restore - Ends the implicit GFP_NOFS scope. 284 * @flags: Flags to restore. 285 * 286 * Ends the implicit GFP_NOFS scope started by memalloc_nofs_save function. 287 * Always make sure that the given flags is the return value from the 288 * pairing memalloc_nofs_save call. 289 */ 290 static inline void memalloc_nofs_restore(unsigned int flags) 291 { 292 current->flags = (current->flags & ~PF_MEMALLOC_NOFS) | flags; 293 } 294 295 static inline unsigned int memalloc_noreclaim_save(void) 296 { 297 unsigned int flags = current->flags & PF_MEMALLOC; 298 current->flags |= PF_MEMALLOC; 299 return flags; 300 } 301 302 static inline void memalloc_noreclaim_restore(unsigned int flags) 303 { 304 current->flags = (current->flags & ~PF_MEMALLOC) | flags; 305 } 306 307 static inline unsigned int memalloc_pin_save(void) 308 { 309 unsigned int flags = current->flags & PF_MEMALLOC_PIN; 310 311 current->flags |= PF_MEMALLOC_PIN; 312 return flags; 313 } 314 315 static inline void memalloc_pin_restore(unsigned int flags) 316 { 317 current->flags = (current->flags & ~PF_MEMALLOC_PIN) | flags; 318 } 319 320 #ifdef CONFIG_MEMCG 321 DECLARE_PER_CPU(struct mem_cgroup *, int_active_memcg); 322 /** 323 * set_active_memcg - Starts the remote memcg charging scope. 324 * @memcg: memcg to charge. 325 * 326 * This function marks the beginning of the remote memcg charging scope. All the 327 * __GFP_ACCOUNT allocations till the end of the scope will be charged to the 328 * given memcg. 329 * 330 * NOTE: This function can nest. Users must save the return value and 331 * reset the previous value after their own charging scope is over. 332 */ 333 static inline struct mem_cgroup * 334 set_active_memcg(struct mem_cgroup *memcg) 335 { 336 struct mem_cgroup *old; 337 338 if (!in_task()) { 339 old = this_cpu_read(int_active_memcg); 340 this_cpu_write(int_active_memcg, memcg); 341 } else { 342 old = current->active_memcg; 343 current->active_memcg = memcg; 344 } 345 346 return old; 347 } 348 #else 349 static inline struct mem_cgroup * 350 set_active_memcg(struct mem_cgroup *memcg) 351 { 352 return NULL; 353 } 354 #endif 355 356 #ifdef CONFIG_MEMBARRIER 357 enum { 358 MEMBARRIER_STATE_PRIVATE_EXPEDITED_READY = (1U << 0), 359 MEMBARRIER_STATE_PRIVATE_EXPEDITED = (1U << 1), 360 MEMBARRIER_STATE_GLOBAL_EXPEDITED_READY = (1U << 2), 361 MEMBARRIER_STATE_GLOBAL_EXPEDITED = (1U << 3), 362 MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE_READY = (1U << 4), 363 MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE = (1U << 5), 364 MEMBARRIER_STATE_PRIVATE_EXPEDITED_RSEQ_READY = (1U << 6), 365 MEMBARRIER_STATE_PRIVATE_EXPEDITED_RSEQ = (1U << 7), 366 }; 367 368 enum { 369 MEMBARRIER_FLAG_SYNC_CORE = (1U << 0), 370 MEMBARRIER_FLAG_RSEQ = (1U << 1), 371 }; 372 373 #ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS 374 #include <asm/membarrier.h> 375 #endif 376 377 static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm) 378 { 379 if (current->mm != mm) 380 return; 381 if (likely(!(atomic_read(&mm->membarrier_state) & 382 MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE))) 383 return; 384 sync_core_before_usermode(); 385 } 386 387 extern void membarrier_exec_mmap(struct mm_struct *mm); 388 389 extern void membarrier_update_current_mm(struct mm_struct *next_mm); 390 391 #else 392 #ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS 393 static inline void membarrier_arch_switch_mm(struct mm_struct *prev, 394 struct mm_struct *next, 395 struct task_struct *tsk) 396 { 397 } 398 #endif 399 static inline void membarrier_exec_mmap(struct mm_struct *mm) 400 { 401 } 402 static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm) 403 { 404 } 405 static inline void membarrier_update_current_mm(struct mm_struct *next_mm) 406 { 407 } 408 #endif 409 410 #endif /* _LINUX_SCHED_MM_H */ 411