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/vm/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 extern void __mmdrop(struct mm_struct *mm);
41 static inline void mmdrop(struct mm_struct *mm)
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.
48 if (unlikely(atomic_dec_and_test(&mm->mm_count)))
52 #ifdef CONFIG_PREEMPT_RT
54 * RCU callback for delayed mm drop. Not strictly RCU, but call_rcu() is
55 * by far the least expensive way to do that.
57 static inline void __mmdrop_delayed(struct rcu_head *rhp)
59 struct mm_struct *mm = container_of(rhp, struct mm_struct, delayed_drop);
65 * Invoked from finish_task_switch(). Delegates the heavy lifting on RT
68 static inline void mmdrop_sched(struct mm_struct *mm)
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);
75 static inline void mmdrop_sched(struct mm_struct *mm)
82 * mmget() - Pin the address space associated with a &struct mm_struct.
83 * @mm: The address space to pin.
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.
89 * Never use this function to pin this address space for an
90 * unbounded/indefinite amount of time.
92 * Use mmput() to release the reference acquired by mmget().
94 * See also <Documentation/vm/active_mm.rst> for an in-depth explanation
95 * of &mm_struct.mm_count vs &mm_struct.mm_users.
97 static inline void mmget(struct mm_struct *mm)
99 atomic_inc(&mm->mm_users);
102 static inline bool mmget_not_zero(struct mm_struct *mm)
104 return atomic_inc_not_zero(&mm->mm_users);
107 /* mmput gets rid of the mappings and all user-space */
108 extern void mmput(struct mm_struct *);
110 /* same as above but performs the slow path from the async context. Can
111 * be called from the atomic context as well
113 void mmput_async(struct mm_struct *);
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);
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
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 *);
130 extern void mm_update_next_owner(struct mm_struct *mm);
132 static inline void mm_update_next_owner(struct mm_struct *mm)
135 #endif /* CONFIG_MEMCG */
138 extern void arch_pick_mmap_layout(struct mm_struct *mm,
139 struct rlimit *rlim_stack);
141 arch_get_unmapped_area(struct file *, unsigned long, unsigned long,
142 unsigned long, 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);
148 static inline void arch_pick_mmap_layout(struct mm_struct *mm,
149 struct rlimit *rlim_stack) {}
152 static inline bool in_vfork(struct task_struct *tsk)
157 * need RCU to access ->real_parent if CLONE_VM was used along with
160 * We check real_parent->mm == tsk->mm because CLONE_VFORK does not
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.
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.
172 ret = tsk->vfork_done &&
173 rcu_dereference(tsk->real_parent)->mm == tsk->mm;
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
185 static inline gfp_t current_gfp_context(gfp_t flags)
187 unsigned int pflags = READ_ONCE(current->flags);
189 if (unlikely(pflags & (PF_MEMALLOC_NOIO | PF_MEMALLOC_NOFS | PF_MEMALLOC_PIN))) {
191 * NOIO implies both NOIO and NOFS and it is a weaker context
192 * so always make sure it makes precedence
194 if (pflags & PF_MEMALLOC_NOIO)
195 flags &= ~(__GFP_IO | __GFP_FS);
196 else if (pflags & PF_MEMALLOC_NOFS)
199 if (pflags & PF_MEMALLOC_PIN)
200 flags &= ~__GFP_MOVABLE;
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);
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) { }
217 /* Any memory-allocation retry loop should use
218 * memalloc_retry_wait(), and pass the flags for the most
219 * constrained allocation attempt that might have failed.
220 * This provides useful documentation of where loops are,
221 * and a central place to fine tune the waiting as the MM
222 * implementation changes.
224 static inline void memalloc_retry_wait(gfp_t gfp_flags)
226 /* We use io_schedule_timeout because waiting for memory
227 * typically included waiting for dirty pages to be
228 * written out, which requires IO.
230 __set_current_state(TASK_UNINTERRUPTIBLE);
231 gfp_flags = current_gfp_context(gfp_flags);
232 if (gfpflags_allow_blocking(gfp_flags) &&
233 !(gfp_flags & __GFP_NORETRY))
234 /* Probably waited already, no need for much more */
235 io_schedule_timeout(1);
237 /* Probably didn't wait, and has now released a lock,
238 * so now is a good time to wait
240 io_schedule_timeout(HZ/50);
244 * might_alloc - Mark possible allocation sites
245 * @gfp_mask: gfp_t flags that would be used to allocate
247 * Similar to might_sleep() and other annotations, this can be used in functions
248 * that might allocate, but often don't. Compiles to nothing without
249 * CONFIG_LOCKDEP. Includes a conditional might_sleep() if @gfp allows blocking.
251 static inline void might_alloc(gfp_t gfp_mask)
253 fs_reclaim_acquire(gfp_mask);
254 fs_reclaim_release(gfp_mask);
256 might_sleep_if(gfpflags_allow_blocking(gfp_mask));
260 * memalloc_noio_save - Marks implicit GFP_NOIO allocation scope.
262 * This functions marks the beginning of the GFP_NOIO allocation scope.
263 * All further allocations will implicitly drop __GFP_IO flag and so
264 * they are safe for the IO critical section from the allocation recursion
265 * point of view. Use memalloc_noio_restore to end the scope with flags
266 * returned by this function.
268 * This function is safe to be used from any context.
270 static inline unsigned int memalloc_noio_save(void)
272 unsigned int flags = current->flags & PF_MEMALLOC_NOIO;
273 current->flags |= PF_MEMALLOC_NOIO;
278 * memalloc_noio_restore - Ends the implicit GFP_NOIO scope.
279 * @flags: Flags to restore.
281 * Ends the implicit GFP_NOIO scope started by memalloc_noio_save function.
282 * Always make sure that the given flags is the return value from the
283 * pairing memalloc_noio_save call.
285 static inline void memalloc_noio_restore(unsigned int flags)
287 current->flags = (current->flags & ~PF_MEMALLOC_NOIO) | flags;
291 * memalloc_nofs_save - Marks implicit GFP_NOFS allocation scope.
293 * This functions marks the beginning of the GFP_NOFS allocation scope.
294 * All further allocations will implicitly drop __GFP_FS flag and so
295 * they are safe for the FS critical section from the allocation recursion
296 * point of view. Use memalloc_nofs_restore to end the scope with flags
297 * returned by this function.
299 * This function is safe to be used from any context.
301 static inline unsigned int memalloc_nofs_save(void)
303 unsigned int flags = current->flags & PF_MEMALLOC_NOFS;
304 current->flags |= PF_MEMALLOC_NOFS;
309 * memalloc_nofs_restore - Ends the implicit GFP_NOFS scope.
310 * @flags: Flags to restore.
312 * Ends the implicit GFP_NOFS scope started by memalloc_nofs_save function.
313 * Always make sure that the given flags is the return value from the
314 * pairing memalloc_nofs_save call.
316 static inline void memalloc_nofs_restore(unsigned int flags)
318 current->flags = (current->flags & ~PF_MEMALLOC_NOFS) | flags;
321 static inline unsigned int memalloc_noreclaim_save(void)
323 unsigned int flags = current->flags & PF_MEMALLOC;
324 current->flags |= PF_MEMALLOC;
328 static inline void memalloc_noreclaim_restore(unsigned int flags)
330 current->flags = (current->flags & ~PF_MEMALLOC) | flags;
333 static inline unsigned int memalloc_pin_save(void)
335 unsigned int flags = current->flags & PF_MEMALLOC_PIN;
337 current->flags |= PF_MEMALLOC_PIN;
341 static inline void memalloc_pin_restore(unsigned int flags)
343 current->flags = (current->flags & ~PF_MEMALLOC_PIN) | flags;
347 DECLARE_PER_CPU(struct mem_cgroup *, int_active_memcg);
349 * set_active_memcg - Starts the remote memcg charging scope.
350 * @memcg: memcg to charge.
352 * This function marks the beginning of the remote memcg charging scope. All the
353 * __GFP_ACCOUNT allocations till the end of the scope will be charged to the
356 * NOTE: This function can nest. Users must save the return value and
357 * reset the previous value after their own charging scope is over.
359 static inline struct mem_cgroup *
360 set_active_memcg(struct mem_cgroup *memcg)
362 struct mem_cgroup *old;
365 old = this_cpu_read(int_active_memcg);
366 this_cpu_write(int_active_memcg, memcg);
368 old = current->active_memcg;
369 current->active_memcg = memcg;
375 static inline struct mem_cgroup *
376 set_active_memcg(struct mem_cgroup *memcg)
382 #ifdef CONFIG_MEMBARRIER
384 MEMBARRIER_STATE_PRIVATE_EXPEDITED_READY = (1U << 0),
385 MEMBARRIER_STATE_PRIVATE_EXPEDITED = (1U << 1),
386 MEMBARRIER_STATE_GLOBAL_EXPEDITED_READY = (1U << 2),
387 MEMBARRIER_STATE_GLOBAL_EXPEDITED = (1U << 3),
388 MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE_READY = (1U << 4),
389 MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE = (1U << 5),
390 MEMBARRIER_STATE_PRIVATE_EXPEDITED_RSEQ_READY = (1U << 6),
391 MEMBARRIER_STATE_PRIVATE_EXPEDITED_RSEQ = (1U << 7),
395 MEMBARRIER_FLAG_SYNC_CORE = (1U << 0),
396 MEMBARRIER_FLAG_RSEQ = (1U << 1),
399 #ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
400 #include <asm/membarrier.h>
403 static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
405 if (current->mm != mm)
407 if (likely(!(atomic_read(&mm->membarrier_state) &
408 MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE)))
410 sync_core_before_usermode();
413 extern void membarrier_exec_mmap(struct mm_struct *mm);
415 extern void membarrier_update_current_mm(struct mm_struct *next_mm);
418 #ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
419 static inline void membarrier_arch_switch_mm(struct mm_struct *prev,
420 struct mm_struct *next,
421 struct task_struct *tsk)
425 static inline void membarrier_exec_mmap(struct mm_struct *mm)
428 static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
431 static inline void membarrier_update_current_mm(struct mm_struct *next_mm)
436 #endif /* _LINUX_SCHED_MM_H */