1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_SCHED_MM_H
3 #define _LINUX_SCHED_MM_H
5 #include <linux/kernel.h>
6 #include <linux/atomic.h>
7 #include <linux/sched.h>
8 #include <linux/mm_types.h>
10 #include <linux/sync_core.h>
13 * Routines for handling mm_structs
15 extern struct mm_struct *mm_alloc(void);
18 * mmgrab() - Pin a &struct mm_struct.
19 * @mm: The &struct mm_struct to pin.
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
26 * This is a preferred way to pin @mm for a longer/unbounded amount
29 * Use mmdrop() to release the reference acquired by mmgrab().
31 * See also <Documentation/mm/active_mm.rst> for an in-depth explanation
32 * of &mm_struct.mm_count vs &mm_struct.mm_users.
34 static inline void mmgrab(struct mm_struct *mm)
36 atomic_inc(&mm->mm_count);
39 static inline void smp_mb__after_mmgrab(void)
41 smp_mb__after_atomic();
44 extern void __mmdrop(struct mm_struct *mm);
46 static inline void mmdrop(struct mm_struct *mm)
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.
53 if (unlikely(atomic_dec_and_test(&mm->mm_count)))
57 #ifdef CONFIG_PREEMPT_RT
59 * RCU callback for delayed mm drop. Not strictly RCU, but call_rcu() is
60 * by far the least expensive way to do that.
62 static inline void __mmdrop_delayed(struct rcu_head *rhp)
64 struct mm_struct *mm = container_of(rhp, struct mm_struct, delayed_drop);
70 * Invoked from finish_task_switch(). Delegates the heavy lifting on RT
73 static inline void mmdrop_sched(struct mm_struct *mm)
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);
80 static inline void mmdrop_sched(struct mm_struct *mm)
86 /* Helpers for lazy TLB mm refcounting */
87 static inline void mmgrab_lazy_tlb(struct mm_struct *mm)
89 if (IS_ENABLED(CONFIG_MMU_LAZY_TLB_REFCOUNT))
93 static inline void mmdrop_lazy_tlb(struct mm_struct *mm)
95 if (IS_ENABLED(CONFIG_MMU_LAZY_TLB_REFCOUNT)) {
99 * mmdrop_lazy_tlb must provide a full memory barrier, see the
100 * membarrier comment finish_task_switch which relies on this.
106 static inline void mmdrop_lazy_tlb_sched(struct mm_struct *mm)
108 if (IS_ENABLED(CONFIG_MMU_LAZY_TLB_REFCOUNT))
111 smp_mb(); /* see mmdrop_lazy_tlb() above */
115 * mmget() - Pin the address space associated with a &struct mm_struct.
116 * @mm: The address space to pin.
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.
122 * Never use this function to pin this address space for an
123 * unbounded/indefinite amount of time.
125 * Use mmput() to release the reference acquired by mmget().
127 * See also <Documentation/mm/active_mm.rst> for an in-depth explanation
128 * of &mm_struct.mm_count vs &mm_struct.mm_users.
130 static inline void mmget(struct mm_struct *mm)
132 atomic_inc(&mm->mm_users);
135 static inline bool mmget_not_zero(struct mm_struct *mm)
137 return atomic_inc_not_zero(&mm->mm_users);
140 /* mmput gets rid of the mappings and all user-space */
141 extern void mmput(struct mm_struct *);
143 /* same as above but performs the slow path from the async context. Can
144 * be called from the atomic context as well
146 void mmput_async(struct mm_struct *);
149 /* Grab a reference to a task's mm, if it is not already going away */
150 extern struct mm_struct *get_task_mm(struct task_struct *task);
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
156 extern 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() */
158 extern void exit_mm_release(struct task_struct *, struct mm_struct *);
159 /* Remove the current tasks stale references to the old mm_struct on exec() */
160 extern void exec_mm_release(struct task_struct *, struct mm_struct *);
163 extern void mm_update_next_owner(struct mm_struct *mm);
165 static inline void mm_update_next_owner(struct mm_struct *mm)
168 #endif /* CONFIG_MEMCG */
171 #ifndef arch_get_mmap_end
172 #define arch_get_mmap_end(addr, len, flags) (TASK_SIZE)
175 #ifndef arch_get_mmap_base
176 #define arch_get_mmap_base(addr, base) (base)
179 extern void arch_pick_mmap_layout(struct mm_struct *mm,
180 struct rlimit *rlim_stack);
182 arch_get_unmapped_area(struct file *, unsigned long, unsigned long,
183 unsigned long, unsigned long);
185 arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr,
186 unsigned long len, unsigned long pgoff,
187 unsigned long flags);
190 generic_get_unmapped_area(struct file *filp, unsigned long addr,
191 unsigned long len, unsigned long pgoff,
192 unsigned long flags);
194 generic_get_unmapped_area_topdown(struct file *filp, unsigned long addr,
195 unsigned long len, unsigned long pgoff,
196 unsigned long flags);
198 static inline void arch_pick_mmap_layout(struct mm_struct *mm,
199 struct rlimit *rlim_stack) {}
202 static inline bool in_vfork(struct task_struct *tsk)
207 * need RCU to access ->real_parent if CLONE_VM was used along with
210 * We check real_parent->mm == tsk->mm because CLONE_VFORK does not
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.
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.
222 ret = tsk->vfork_done &&
223 rcu_dereference(tsk->real_parent)->mm == tsk->mm;
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
235 static inline gfp_t current_gfp_context(gfp_t flags)
237 unsigned int pflags = READ_ONCE(current->flags);
239 if (unlikely(pflags & (PF_MEMALLOC_NOIO |
241 PF_MEMALLOC_NORECLAIM |
245 * Stronger flags before weaker flags:
246 * NORECLAIM implies NOIO, which in turn implies NOFS
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)
255 if (pflags & PF_MEMALLOC_NOWARN)
256 flags |= __GFP_NOWARN;
258 if (pflags & PF_MEMALLOC_PIN)
259 flags &= ~__GFP_MOVABLE;
264 #ifdef CONFIG_LOCKDEP
265 extern void __fs_reclaim_acquire(unsigned long ip);
266 extern void __fs_reclaim_release(unsigned long ip);
267 extern void fs_reclaim_acquire(gfp_t gfp_mask);
268 extern void fs_reclaim_release(gfp_t gfp_mask);
270 static inline void __fs_reclaim_acquire(unsigned long ip) { }
271 static inline void __fs_reclaim_release(unsigned long ip) { }
272 static inline void fs_reclaim_acquire(gfp_t gfp_mask) { }
273 static inline void fs_reclaim_release(gfp_t gfp_mask) { }
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.
283 static inline void memalloc_retry_wait(gfp_t gfp_flags)
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.
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);
296 /* Probably didn't wait, and has now released a lock,
297 * so now is a good time to wait
299 io_schedule_timeout(HZ/50);
303 * might_alloc - Mark possible allocation sites
304 * @gfp_mask: gfp_t flags that would be used to allocate
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.
310 static inline void might_alloc(gfp_t gfp_mask)
312 fs_reclaim_acquire(gfp_mask);
313 fs_reclaim_release(gfp_mask);
315 might_sleep_if(gfpflags_allow_blocking(gfp_mask));
319 * memalloc_flags_save - Add a PF_* flag to current->flags, save old value
321 * This allows PF_* flags to be conveniently added, irrespective of current
322 * value, and then the old version restored with memalloc_flags_restore().
324 static inline unsigned memalloc_flags_save(unsigned flags)
326 unsigned oldflags = ~current->flags & flags;
327 current->flags |= flags;
331 static inline void memalloc_flags_restore(unsigned flags)
333 current->flags &= ~flags;
337 * memalloc_noio_save - Marks implicit GFP_NOIO allocation scope.
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.
345 * This function is safe to be used from any context.
347 static inline unsigned int memalloc_noio_save(void)
349 return memalloc_flags_save(PF_MEMALLOC_NOIO);
353 * memalloc_noio_restore - Ends the implicit GFP_NOIO scope.
354 * @flags: Flags to restore.
356 * Ends the implicit GFP_NOIO scope started by memalloc_noio_save function.
357 * Always make sure that the given flags is the return value from the
358 * pairing memalloc_noio_save call.
360 static inline void memalloc_noio_restore(unsigned int flags)
362 memalloc_flags_restore(flags);
366 * memalloc_nofs_save - Marks implicit GFP_NOFS allocation scope.
368 * This functions marks the beginning of the GFP_NOFS allocation scope.
369 * All further allocations will implicitly drop __GFP_FS flag and so
370 * they are safe for the FS critical section from the allocation recursion
371 * point of view. Use memalloc_nofs_restore to end the scope with flags
372 * returned by this function.
374 * This function is safe to be used from any context.
376 static inline unsigned int memalloc_nofs_save(void)
378 return memalloc_flags_save(PF_MEMALLOC_NOFS);
382 * memalloc_nofs_restore - Ends the implicit GFP_NOFS scope.
383 * @flags: Flags to restore.
385 * Ends the implicit GFP_NOFS scope started by memalloc_nofs_save function.
386 * Always make sure that the given flags is the return value from the
387 * pairing memalloc_nofs_save call.
389 static inline void memalloc_nofs_restore(unsigned int flags)
391 memalloc_flags_restore(flags);
394 static inline unsigned int memalloc_noreclaim_save(void)
396 return memalloc_flags_save(PF_MEMALLOC);
399 static inline void memalloc_noreclaim_restore(unsigned int flags)
401 memalloc_flags_restore(flags);
404 static inline unsigned int memalloc_pin_save(void)
406 return memalloc_flags_save(PF_MEMALLOC_PIN);
409 static inline void memalloc_pin_restore(unsigned int flags)
411 memalloc_flags_restore(flags);
415 DECLARE_PER_CPU(struct mem_cgroup *, int_active_memcg);
417 * set_active_memcg - Starts the remote memcg charging scope.
418 * @memcg: memcg to charge.
420 * This function marks the beginning of the remote memcg charging scope. All the
421 * __GFP_ACCOUNT allocations till the end of the scope will be charged to the
424 * Please, make sure that caller has a reference to the passed memcg structure,
425 * so its lifetime is guaranteed to exceed the scope between two
426 * set_active_memcg() calls.
428 * NOTE: This function can nest. Users must save the return value and
429 * reset the previous value after their own charging scope is over.
431 static inline struct mem_cgroup *
432 set_active_memcg(struct mem_cgroup *memcg)
434 struct mem_cgroup *old;
437 old = this_cpu_read(int_active_memcg);
438 this_cpu_write(int_active_memcg, memcg);
440 old = current->active_memcg;
441 current->active_memcg = memcg;
447 static inline struct mem_cgroup *
448 set_active_memcg(struct mem_cgroup *memcg)
454 #ifdef CONFIG_MEMBARRIER
456 MEMBARRIER_STATE_PRIVATE_EXPEDITED_READY = (1U << 0),
457 MEMBARRIER_STATE_PRIVATE_EXPEDITED = (1U << 1),
458 MEMBARRIER_STATE_GLOBAL_EXPEDITED_READY = (1U << 2),
459 MEMBARRIER_STATE_GLOBAL_EXPEDITED = (1U << 3),
460 MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE_READY = (1U << 4),
461 MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE = (1U << 5),
462 MEMBARRIER_STATE_PRIVATE_EXPEDITED_RSEQ_READY = (1U << 6),
463 MEMBARRIER_STATE_PRIVATE_EXPEDITED_RSEQ = (1U << 7),
467 MEMBARRIER_FLAG_SYNC_CORE = (1U << 0),
468 MEMBARRIER_FLAG_RSEQ = (1U << 1),
471 #ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
472 #include <asm/membarrier.h>
475 static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
477 if (current->mm != mm)
479 if (likely(!(atomic_read(&mm->membarrier_state) &
480 MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE)))
482 sync_core_before_usermode();
485 extern void membarrier_exec_mmap(struct mm_struct *mm);
487 extern void membarrier_update_current_mm(struct mm_struct *next_mm);
490 #ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
491 static inline void membarrier_arch_switch_mm(struct mm_struct *prev,
492 struct mm_struct *next,
493 struct task_struct *tsk)
497 static inline void membarrier_exec_mmap(struct mm_struct *mm)
500 static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
503 static inline void membarrier_update_current_mm(struct mm_struct *next_mm)
508 #endif /* _LINUX_SCHED_MM_H */