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 */ 15extern 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/mm/active_mm.rst> for an in-depth explanation 32 * of &mm_struct.mm_count vs &mm_struct.mm_users. 33 */ 34static inline void mmgrab(struct mm_struct *mm) 35{ 36 atomic_inc(&mm->mm_count); 37} 38 39static inline void smp_mb__after_mmgrab(void) 40{ 41 smp_mb__after_atomic(); 42} 43 44extern void __mmdrop(struct mm_struct *mm); 45 46static inline void mmdrop(struct mm_struct *mm) 47{ 48 /* 49 * The implicit full barrier implied by atomic_dec_and_test() is 50 * required by the membarrier system call before returning to 51 * user-space, after storing to rq->curr. 52 */ 53 if (unlikely(atomic_dec_and_test(&mm->mm_count))) 54 __mmdrop(mm); 55} 56 57#ifdef CONFIG_PREEMPT_RT 58/* 59 * RCU callback for delayed mm drop. Not strictly RCU, but call_rcu() is 60 * by far the least expensive way to do that. 61 */ 62static inline void __mmdrop_delayed(struct rcu_head *rhp) 63{ 64 struct mm_struct *mm = container_of(rhp, struct mm_struct, delayed_drop); 65 66 __mmdrop(mm); 67} 68 69/* 70 * Invoked from finish_task_switch(). Delegates the heavy lifting on RT 71 * kernels via RCU. 72 */ 73static inline void mmdrop_sched(struct mm_struct *mm) 74{ 75 /* Provides a full memory barrier. See mmdrop() */ 76 if (atomic_dec_and_test(&mm->mm_count)) 77 call_rcu(&mm->delayed_drop, __mmdrop_delayed); 78} 79#else 80static inline void mmdrop_sched(struct mm_struct *mm) 81{ 82 mmdrop(mm); 83} 84#endif 85 86/* Helpers for lazy TLB mm refcounting */ 87static inline void mmgrab_lazy_tlb(struct mm_struct *mm) 88{ 89 if (IS_ENABLED(CONFIG_MMU_LAZY_TLB_REFCOUNT)) 90 mmgrab(mm); 91} 92 93static inline void mmdrop_lazy_tlb(struct mm_struct *mm) 94{ 95 if (IS_ENABLED(CONFIG_MMU_LAZY_TLB_REFCOUNT)) { 96 mmdrop(mm); 97 } else { 98 /* 99 * mmdrop_lazy_tlb must provide a full memory barrier, see the 100 * membarrier comment finish_task_switch which relies on this. 101 */ 102 smp_mb(); 103 } 104} 105 106static inline void mmdrop_lazy_tlb_sched(struct mm_struct *mm) 107{ 108 if (IS_ENABLED(CONFIG_MMU_LAZY_TLB_REFCOUNT)) 109 mmdrop_sched(mm); 110 else 111 smp_mb(); /* see mmdrop_lazy_tlb() above */ 112} 113 114/** 115 * mmget() - Pin the address space associated with a &struct mm_struct. 116 * @mm: The address space to pin. 117 * 118 * Make sure that the address space of the given &struct mm_struct doesn't 119 * go away. This does not protect against parts of the address space being 120 * modified or freed, however. 121 * 122 * Never use this function to pin this address space for an 123 * unbounded/indefinite amount of time. 124 * 125 * Use mmput() to release the reference acquired by mmget(). 126 * 127 * See also <Documentation/mm/active_mm.rst> for an in-depth explanation 128 * of &mm_struct.mm_count vs &mm_struct.mm_users. 129 */ 130static inline void mmget(struct mm_struct *mm) 131{ 132 atomic_inc(&mm->mm_users); 133} 134 135static inline bool mmget_not_zero(struct mm_struct *mm) 136{ 137 return atomic_inc_not_zero(&mm->mm_users); 138} 139 140/* mmput gets rid of the mappings and all user-space */ 141extern void mmput(struct mm_struct *); 142#ifdef CONFIG_MMU 143/* same as above but performs the slow path from the async context. Can 144 * be called from the atomic context as well 145 */ 146void mmput_async(struct mm_struct *); 147#endif 148 149/* Grab a reference to a task's mm, if it is not already going away */ 150extern struct mm_struct *get_task_mm(struct task_struct *task); 151/* 152 * Grab a reference to a task's mm, if it is not already going away 153 * and ptrace_may_access with the mode parameter passed to it 154 * succeeds. 155 */ 156extern struct mm_struct *mm_access(struct task_struct *task, unsigned int mode); 157/* Remove the current tasks stale references to the old mm_struct on exit() */ 158extern void exit_mm_release(struct task_struct *, struct mm_struct *); 159/* Remove the current tasks stale references to the old mm_struct on exec() */ 160extern void exec_mm_release(struct task_struct *, struct mm_struct *); 161 162#ifdef CONFIG_MEMCG 163extern void mm_update_next_owner(struct mm_struct *mm); 164#else 165static inline void mm_update_next_owner(struct mm_struct *mm) 166{ 167} 168#endif /* CONFIG_MEMCG */ 169 170#ifdef CONFIG_MMU 171#ifndef arch_get_mmap_end 172#define arch_get_mmap_end(addr, len, flags) (TASK_SIZE) 173#endif 174 175#ifndef arch_get_mmap_base 176#define arch_get_mmap_base(addr, base) (base) 177#endif 178 179extern void arch_pick_mmap_layout(struct mm_struct *mm, 180 struct rlimit *rlim_stack); 181extern unsigned long 182arch_get_unmapped_area(struct file *, unsigned long, unsigned long, 183 unsigned long, unsigned long); 184extern unsigned long 185arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr, 186 unsigned long len, unsigned long pgoff, 187 unsigned long flags); 188 189unsigned long 190generic_get_unmapped_area(struct file *filp, unsigned long addr, 191 unsigned long len, unsigned long pgoff, 192 unsigned long flags); 193unsigned long 194generic_get_unmapped_area_topdown(struct file *filp, unsigned long addr, 195 unsigned long len, unsigned long pgoff, 196 unsigned long flags); 197#else 198static inline void arch_pick_mmap_layout(struct mm_struct *mm, 199 struct rlimit *rlim_stack) {} 200#endif 201 202static inline bool in_vfork(struct task_struct *tsk) 203{ 204 bool ret; 205 206 /* 207 * need RCU to access ->real_parent if CLONE_VM was used along with 208 * CLONE_PARENT. 209 * 210 * We check real_parent->mm == tsk->mm because CLONE_VFORK does not 211 * imply CLONE_VM 212 * 213 * CLONE_VFORK can be used with CLONE_PARENT/CLONE_THREAD and thus 214 * ->real_parent is not necessarily the task doing vfork(), so in 215 * theory we can't rely on task_lock() if we want to dereference it. 216 * 217 * And in this case we can't trust the real_parent->mm == tsk->mm 218 * check, it can be false negative. But we do not care, if init or 219 * another oom-unkillable task does this it should blame itself. 220 */ 221 rcu_read_lock(); 222 ret = tsk->vfork_done && 223 rcu_dereference(tsk->real_parent)->mm == tsk->mm; 224 rcu_read_unlock(); 225 226 return ret; 227} 228 229/* 230 * Applies per-task gfp context to the given allocation flags. 231 * PF_MEMALLOC_NOIO implies GFP_NOIO 232 * PF_MEMALLOC_NOFS implies GFP_NOFS 233 * PF_MEMALLOC_PIN implies !GFP_MOVABLE 234 */ 235static inline gfp_t current_gfp_context(gfp_t flags) 236{ 237 unsigned int pflags = READ_ONCE(current->flags); 238 239 if (unlikely(pflags & (PF_MEMALLOC_NOIO | 240 PF_MEMALLOC_NOFS | 241 PF_MEMALLOC_NORECLAIM | 242 PF_MEMALLOC_NOWARN | 243 PF_MEMALLOC_PIN))) { 244 /* 245 * Stronger flags before weaker flags: 246 * NORECLAIM implies NOIO, which in turn implies NOFS 247 */ 248 if (pflags & PF_MEMALLOC_NORECLAIM) 249 flags &= ~__GFP_DIRECT_RECLAIM; 250 else if (pflags & PF_MEMALLOC_NOIO) 251 flags &= ~(__GFP_IO | __GFP_FS); 252 else if (pflags & PF_MEMALLOC_NOFS) 253 flags &= ~__GFP_FS; 254 255 if (pflags & PF_MEMALLOC_NOWARN) 256 flags |= __GFP_NOWARN; 257 258 if (pflags & PF_MEMALLOC_PIN) 259 flags &= ~__GFP_MOVABLE; 260 } 261 return flags; 262} 263 264#ifdef CONFIG_LOCKDEP 265extern void __fs_reclaim_acquire(unsigned long ip); 266extern void __fs_reclaim_release(unsigned long ip); 267extern void fs_reclaim_acquire(gfp_t gfp_mask); 268extern void fs_reclaim_release(gfp_t gfp_mask); 269#else 270static inline void __fs_reclaim_acquire(unsigned long ip) { } 271static inline void __fs_reclaim_release(unsigned long ip) { } 272static inline void fs_reclaim_acquire(gfp_t gfp_mask) { } 273static inline void fs_reclaim_release(gfp_t gfp_mask) { } 274#endif 275 276/* Any memory-allocation retry loop should use 277 * memalloc_retry_wait(), and pass the flags for the most 278 * constrained allocation attempt that might have failed. 279 * This provides useful documentation of where loops are, 280 * and a central place to fine tune the waiting as the MM 281 * implementation changes. 282 */ 283static inline void memalloc_retry_wait(gfp_t gfp_flags) 284{ 285 /* We use io_schedule_timeout because waiting for memory 286 * typically included waiting for dirty pages to be 287 * written out, which requires IO. 288 */ 289 __set_current_state(TASK_UNINTERRUPTIBLE); 290 gfp_flags = current_gfp_context(gfp_flags); 291 if (gfpflags_allow_blocking(gfp_flags) && 292 !(gfp_flags & __GFP_NORETRY)) 293 /* Probably waited already, no need for much more */ 294 io_schedule_timeout(1); 295 else 296 /* Probably didn't wait, and has now released a lock, 297 * so now is a good time to wait 298 */ 299 io_schedule_timeout(HZ/50); 300} 301 302/** 303 * might_alloc - Mark possible allocation sites 304 * @gfp_mask: gfp_t flags that would be used to allocate 305 * 306 * Similar to might_sleep() and other annotations, this can be used in functions 307 * that might allocate, but often don't. Compiles to nothing without 308 * CONFIG_LOCKDEP. Includes a conditional might_sleep() if @gfp allows blocking. 309 */ 310static inline void might_alloc(gfp_t gfp_mask) 311{ 312 fs_reclaim_acquire(gfp_mask); 313 fs_reclaim_release(gfp_mask); 314 315 might_sleep_if(gfpflags_allow_blocking(gfp_mask)); 316} 317 318/** 319 * memalloc_flags_save - Add a PF_* flag to current->flags, save old value 320 * 321 * This allows PF_* flags to be conveniently added, irrespective of current 322 * value, and then the old version restored with memalloc_flags_restore(). 323 */ 324static inline unsigned memalloc_flags_save(unsigned flags) 325{ 326 unsigned oldflags = ~current->flags & flags; 327 current->flags |= flags; 328 return oldflags; 329} 330 331static inline void memalloc_flags_restore(unsigned flags) 332{ 333 current->flags &= ~flags; 334} 335 336/** 337 * memalloc_noio_save - Marks implicit GFP_NOIO allocation scope. 338 * 339 * This functions marks the beginning of the GFP_NOIO allocation scope. 340 * All further allocations will implicitly drop __GFP_IO flag and so 341 * they are safe for the IO critical section from the allocation recursion 342 * point of view. Use memalloc_noio_restore to end the scope with flags 343 * returned by this function. 344 * 345 * Context: This function is safe to be used from any context. 346 * Return: The saved flags to be passed to memalloc_noio_restore. 347 */ 348static inline unsigned int memalloc_noio_save(void) 349{ 350 return memalloc_flags_save(PF_MEMALLOC_NOIO); 351} 352 353/** 354 * memalloc_noio_restore - Ends the implicit GFP_NOIO scope. 355 * @flags: Flags to restore. 356 * 357 * Ends the implicit GFP_NOIO scope started by memalloc_noio_save function. 358 * Always make sure that the given flags is the return value from the 359 * pairing memalloc_noio_save call. 360 */ 361static inline void memalloc_noio_restore(unsigned int flags) 362{ 363 memalloc_flags_restore(flags); 364} 365 366/** 367 * memalloc_nofs_save - Marks implicit GFP_NOFS allocation scope. 368 * 369 * This functions marks the beginning of the GFP_NOFS allocation scope. 370 * All further allocations will implicitly drop __GFP_FS flag and so 371 * they are safe for the FS critical section from the allocation recursion 372 * point of view. Use memalloc_nofs_restore to end the scope with flags 373 * returned by this function. 374 * 375 * Context: This function is safe to be used from any context. 376 * Return: The saved flags to be passed to memalloc_nofs_restore. 377 */ 378static inline unsigned int memalloc_nofs_save(void) 379{ 380 return memalloc_flags_save(PF_MEMALLOC_NOFS); 381} 382 383/** 384 * memalloc_nofs_restore - Ends the implicit GFP_NOFS scope. 385 * @flags: Flags to restore. 386 * 387 * Ends the implicit GFP_NOFS scope started by memalloc_nofs_save function. 388 * Always make sure that the given flags is the return value from the 389 * pairing memalloc_nofs_save call. 390 */ 391static inline void memalloc_nofs_restore(unsigned int flags) 392{ 393 memalloc_flags_restore(flags); 394} 395 396/** 397 * memalloc_noreclaim_save - Marks implicit __GFP_MEMALLOC scope. 398 * 399 * This function marks the beginning of the __GFP_MEMALLOC allocation scope. 400 * All further allocations will implicitly add the __GFP_MEMALLOC flag, which 401 * prevents entering reclaim and allows access to all memory reserves. This 402 * should only be used when the caller guarantees the allocation will allow more 403 * memory to be freed very shortly, i.e. it needs to allocate some memory in 404 * the process of freeing memory, and cannot reclaim due to potential recursion. 405 * 406 * Users of this scope have to be extremely careful to not deplete the reserves 407 * completely and implement a throttling mechanism which controls the 408 * consumption of the reserve based on the amount of freed memory. Usage of a 409 * pre-allocated pool (e.g. mempool) should be always considered before using 410 * this scope. 411 * 412 * Individual allocations under the scope can opt out using __GFP_NOMEMALLOC 413 * 414 * Context: This function should not be used in an interrupt context as that one 415 * does not give PF_MEMALLOC access to reserves. 416 * See __gfp_pfmemalloc_flags(). 417 * Return: The saved flags to be passed to memalloc_noreclaim_restore. 418 */ 419static inline unsigned int memalloc_noreclaim_save(void) 420{ 421 return memalloc_flags_save(PF_MEMALLOC); 422} 423 424/** 425 * memalloc_noreclaim_restore - Ends the implicit __GFP_MEMALLOC scope. 426 * @flags: Flags to restore. 427 * 428 * Ends the implicit __GFP_MEMALLOC scope started by memalloc_noreclaim_save 429 * function. Always make sure that the given flags is the return value from the 430 * pairing memalloc_noreclaim_save call. 431 */ 432static inline void memalloc_noreclaim_restore(unsigned int flags) 433{ 434 memalloc_flags_restore(flags); 435} 436 437/** 438 * memalloc_pin_save - Marks implicit ~__GFP_MOVABLE scope. 439 * 440 * This function marks the beginning of the ~__GFP_MOVABLE allocation scope. 441 * All further allocations will implicitly remove the __GFP_MOVABLE flag, which 442 * will constraint the allocations to zones that allow long term pinning, i.e. 443 * not ZONE_MOVABLE zones. 444 * 445 * Return: The saved flags to be passed to memalloc_pin_restore. 446 */ 447static inline unsigned int memalloc_pin_save(void) 448{ 449 return memalloc_flags_save(PF_MEMALLOC_PIN); 450} 451 452/** 453 * memalloc_pin_restore - Ends the implicit ~__GFP_MOVABLE scope. 454 * @flags: Flags to restore. 455 * 456 * Ends the implicit ~__GFP_MOVABLE scope started by memalloc_pin_save function. 457 * Always make sure that the given flags is the return value from the pairing 458 * memalloc_pin_save call. 459 */ 460static inline void memalloc_pin_restore(unsigned int flags) 461{ 462 memalloc_flags_restore(flags); 463} 464 465#ifdef CONFIG_MEMCG 466DECLARE_PER_CPU(struct mem_cgroup *, int_active_memcg); 467/** 468 * set_active_memcg - Starts the remote memcg charging scope. 469 * @memcg: memcg to charge. 470 * 471 * This function marks the beginning of the remote memcg charging scope. All the 472 * __GFP_ACCOUNT allocations till the end of the scope will be charged to the 473 * given memcg. 474 * 475 * Please, make sure that caller has a reference to the passed memcg structure, 476 * so its lifetime is guaranteed to exceed the scope between two 477 * set_active_memcg() calls. 478 * 479 * NOTE: This function can nest. Users must save the return value and 480 * reset the previous value after their own charging scope is over. 481 */ 482static inline struct mem_cgroup * 483set_active_memcg(struct mem_cgroup *memcg) 484{ 485 struct mem_cgroup *old; 486 487 if (!in_task()) { 488 old = this_cpu_read(int_active_memcg); 489 this_cpu_write(int_active_memcg, memcg); 490 } else { 491 old = current->active_memcg; 492 current->active_memcg = memcg; 493 } 494 495 return old; 496} 497#else 498static inline struct mem_cgroup * 499set_active_memcg(struct mem_cgroup *memcg) 500{ 501 return NULL; 502} 503#endif 504 505#ifdef CONFIG_MEMBARRIER 506enum { 507 MEMBARRIER_STATE_PRIVATE_EXPEDITED_READY = (1U << 0), 508 MEMBARRIER_STATE_PRIVATE_EXPEDITED = (1U << 1), 509 MEMBARRIER_STATE_GLOBAL_EXPEDITED_READY = (1U << 2), 510 MEMBARRIER_STATE_GLOBAL_EXPEDITED = (1U << 3), 511 MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE_READY = (1U << 4), 512 MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE = (1U << 5), 513 MEMBARRIER_STATE_PRIVATE_EXPEDITED_RSEQ_READY = (1U << 6), 514 MEMBARRIER_STATE_PRIVATE_EXPEDITED_RSEQ = (1U << 7), 515}; 516 517enum { 518 MEMBARRIER_FLAG_SYNC_CORE = (1U << 0), 519 MEMBARRIER_FLAG_RSEQ = (1U << 1), 520}; 521 522#ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS 523#include <asm/membarrier.h> 524#endif 525 526static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm) 527{ 528 if (current->mm != mm) 529 return; 530 if (likely(!(atomic_read(&mm->membarrier_state) & 531 MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE))) 532 return; 533 sync_core_before_usermode(); 534} 535 536extern void membarrier_exec_mmap(struct mm_struct *mm); 537 538extern void membarrier_update_current_mm(struct mm_struct *next_mm); 539 540#else 541#ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS 542static inline void membarrier_arch_switch_mm(struct mm_struct *prev, 543 struct mm_struct *next, 544 struct task_struct *tsk) 545{ 546} 547#endif 548static inline void membarrier_exec_mmap(struct mm_struct *mm) 549{ 550} 551static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm) 552{ 553} 554static inline void membarrier_update_current_mm(struct mm_struct *next_mm) 555{ 556} 557#endif 558 559#endif /* _LINUX_SCHED_MM_H */ 560