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)))
53 * mmget() - Pin the address space associated with a &struct mm_struct.
54 * @mm: The address space to pin.
56 * Make sure that the address space of the given &struct mm_struct doesn't
57 * go away. This does not protect against parts of the address space being
58 * modified or freed, however.
60 * Never use this function to pin this address space for an
61 * unbounded/indefinite amount of time.
63 * Use mmput() to release the reference acquired by mmget().
65 * See also <Documentation/vm/active_mm.rst> for an in-depth explanation
66 * of &mm_struct.mm_count vs &mm_struct.mm_users.
68 static inline void mmget(struct mm_struct *mm)
70 atomic_inc(&mm->mm_users);
73 static inline bool mmget_not_zero(struct mm_struct *mm)
75 return atomic_inc_not_zero(&mm->mm_users);
78 /* mmput gets rid of the mappings and all user-space */
79 extern void mmput(struct mm_struct *);
81 /* same as above but performs the slow path from the async context. Can
82 * be called from the atomic context as well
84 void mmput_async(struct mm_struct *);
87 /* Grab a reference to a task's mm, if it is not already going away */
88 extern struct mm_struct *get_task_mm(struct task_struct *task);
90 * Grab a reference to a task's mm, if it is not already going away
91 * and ptrace_may_access with the mode parameter passed to it
94 extern struct mm_struct *mm_access(struct task_struct *task, unsigned int mode);
95 /* Remove the current tasks stale references to the old mm_struct on exit() */
96 extern void exit_mm_release(struct task_struct *, struct mm_struct *);
97 /* Remove the current tasks stale references to the old mm_struct on exec() */
98 extern void exec_mm_release(struct task_struct *, struct mm_struct *);
101 extern void mm_update_next_owner(struct mm_struct *mm);
103 static inline void mm_update_next_owner(struct mm_struct *mm)
106 #endif /* CONFIG_MEMCG */
109 #ifndef arch_get_mmap_end
110 #define arch_get_mmap_end(addr) (TASK_SIZE)
113 #ifndef arch_get_mmap_base
114 #define arch_get_mmap_base(addr, base) (base)
117 extern void arch_pick_mmap_layout(struct mm_struct *mm,
118 struct rlimit *rlim_stack);
120 arch_get_unmapped_area(struct file *, unsigned long, unsigned long,
121 unsigned long, unsigned long);
123 arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr,
124 unsigned long len, unsigned long pgoff,
125 unsigned long flags);
127 static inline void arch_pick_mmap_layout(struct mm_struct *mm,
128 struct rlimit *rlim_stack) {}
131 static inline bool in_vfork(struct task_struct *tsk)
136 * need RCU to access ->real_parent if CLONE_VM was used along with
139 * We check real_parent->mm == tsk->mm because CLONE_VFORK does not
142 * CLONE_VFORK can be used with CLONE_PARENT/CLONE_THREAD and thus
143 * ->real_parent is not necessarily the task doing vfork(), so in
144 * theory we can't rely on task_lock() if we want to dereference it.
146 * And in this case we can't trust the real_parent->mm == tsk->mm
147 * check, it can be false negative. But we do not care, if init or
148 * another oom-unkillable task does this it should blame itself.
151 ret = tsk->vfork_done &&
152 rcu_dereference(tsk->real_parent)->mm == tsk->mm;
159 * Applies per-task gfp context to the given allocation flags.
160 * PF_MEMALLOC_NOIO implies GFP_NOIO
161 * PF_MEMALLOC_NOFS implies GFP_NOFS
162 * PF_MEMALLOC_PIN implies !GFP_MOVABLE
164 static inline gfp_t current_gfp_context(gfp_t flags)
166 unsigned int pflags = READ_ONCE(current->flags);
168 if (unlikely(pflags & (PF_MEMALLOC_NOIO | PF_MEMALLOC_NOFS | PF_MEMALLOC_PIN))) {
170 * NOIO implies both NOIO and NOFS and it is a weaker context
171 * so always make sure it makes precedence
173 if (pflags & PF_MEMALLOC_NOIO)
174 flags &= ~(__GFP_IO | __GFP_FS);
175 else if (pflags & PF_MEMALLOC_NOFS)
178 if (pflags & PF_MEMALLOC_PIN)
179 flags &= ~__GFP_MOVABLE;
184 #ifdef CONFIG_LOCKDEP
185 extern void __fs_reclaim_acquire(unsigned long ip);
186 extern void __fs_reclaim_release(unsigned long ip);
187 extern void fs_reclaim_acquire(gfp_t gfp_mask);
188 extern void fs_reclaim_release(gfp_t gfp_mask);
190 static inline void __fs_reclaim_acquire(unsigned long ip) { }
191 static inline void __fs_reclaim_release(unsigned long ip) { }
192 static inline void fs_reclaim_acquire(gfp_t gfp_mask) { }
193 static inline void fs_reclaim_release(gfp_t gfp_mask) { }
197 * might_alloc - Mark possible allocation sites
198 * @gfp_mask: gfp_t flags that would be used to allocate
200 * Similar to might_sleep() and other annotations, this can be used in functions
201 * that might allocate, but often don't. Compiles to nothing without
202 * CONFIG_LOCKDEP. Includes a conditional might_sleep() if @gfp allows blocking.
204 static inline void might_alloc(gfp_t gfp_mask)
206 fs_reclaim_acquire(gfp_mask);
207 fs_reclaim_release(gfp_mask);
209 might_sleep_if(gfpflags_allow_blocking(gfp_mask));
213 * memalloc_noio_save - Marks implicit GFP_NOIO allocation scope.
215 * This functions marks the beginning of the GFP_NOIO allocation scope.
216 * All further allocations will implicitly drop __GFP_IO flag and so
217 * they are safe for the IO critical section from the allocation recursion
218 * point of view. Use memalloc_noio_restore to end the scope with flags
219 * returned by this function.
221 * This function is safe to be used from any context.
223 static inline unsigned int memalloc_noio_save(void)
225 unsigned int flags = current->flags & PF_MEMALLOC_NOIO;
226 current->flags |= PF_MEMALLOC_NOIO;
231 * memalloc_noio_restore - Ends the implicit GFP_NOIO scope.
232 * @flags: Flags to restore.
234 * Ends the implicit GFP_NOIO scope started by memalloc_noio_save function.
235 * Always make sure that the given flags is the return value from the
236 * pairing memalloc_noio_save call.
238 static inline void memalloc_noio_restore(unsigned int flags)
240 current->flags = (current->flags & ~PF_MEMALLOC_NOIO) | flags;
244 * memalloc_nofs_save - Marks implicit GFP_NOFS allocation scope.
246 * This functions marks the beginning of the GFP_NOFS allocation scope.
247 * All further allocations will implicitly drop __GFP_FS flag and so
248 * they are safe for the FS critical section from the allocation recursion
249 * point of view. Use memalloc_nofs_restore to end the scope with flags
250 * returned by this function.
252 * This function is safe to be used from any context.
254 static inline unsigned int memalloc_nofs_save(void)
256 unsigned int flags = current->flags & PF_MEMALLOC_NOFS;
257 current->flags |= PF_MEMALLOC_NOFS;
262 * memalloc_nofs_restore - Ends the implicit GFP_NOFS scope.
263 * @flags: Flags to restore.
265 * Ends the implicit GFP_NOFS scope started by memalloc_nofs_save function.
266 * Always make sure that the given flags is the return value from the
267 * pairing memalloc_nofs_save call.
269 static inline void memalloc_nofs_restore(unsigned int flags)
271 current->flags = (current->flags & ~PF_MEMALLOC_NOFS) | flags;
274 static inline unsigned int memalloc_noreclaim_save(void)
276 unsigned int flags = current->flags & PF_MEMALLOC;
277 current->flags |= PF_MEMALLOC;
281 static inline void memalloc_noreclaim_restore(unsigned int flags)
283 current->flags = (current->flags & ~PF_MEMALLOC) | flags;
286 static inline unsigned int memalloc_pin_save(void)
288 unsigned int flags = current->flags & PF_MEMALLOC_PIN;
290 current->flags |= PF_MEMALLOC_PIN;
294 static inline void memalloc_pin_restore(unsigned int flags)
296 current->flags = (current->flags & ~PF_MEMALLOC_PIN) | flags;
300 DECLARE_PER_CPU(struct mem_cgroup *, int_active_memcg);
302 * set_active_memcg - Starts the remote memcg charging scope.
303 * @memcg: memcg to charge.
305 * This function marks the beginning of the remote memcg charging scope. All the
306 * __GFP_ACCOUNT allocations till the end of the scope will be charged to the
309 * NOTE: This function can nest. Users must save the return value and
310 * reset the previous value after their own charging scope is over.
312 static inline struct mem_cgroup *
313 set_active_memcg(struct mem_cgroup *memcg)
315 struct mem_cgroup *old;
318 old = this_cpu_read(int_active_memcg);
319 this_cpu_write(int_active_memcg, memcg);
321 old = current->active_memcg;
322 current->active_memcg = memcg;
328 static inline struct mem_cgroup *
329 set_active_memcg(struct mem_cgroup *memcg)
335 #ifdef CONFIG_MEMBARRIER
337 MEMBARRIER_STATE_PRIVATE_EXPEDITED_READY = (1U << 0),
338 MEMBARRIER_STATE_PRIVATE_EXPEDITED = (1U << 1),
339 MEMBARRIER_STATE_GLOBAL_EXPEDITED_READY = (1U << 2),
340 MEMBARRIER_STATE_GLOBAL_EXPEDITED = (1U << 3),
341 MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE_READY = (1U << 4),
342 MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE = (1U << 5),
343 MEMBARRIER_STATE_PRIVATE_EXPEDITED_RSEQ_READY = (1U << 6),
344 MEMBARRIER_STATE_PRIVATE_EXPEDITED_RSEQ = (1U << 7),
348 MEMBARRIER_FLAG_SYNC_CORE = (1U << 0),
349 MEMBARRIER_FLAG_RSEQ = (1U << 1),
352 #ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
353 #include <asm/membarrier.h>
356 static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
358 if (current->mm != mm)
360 if (likely(!(atomic_read(&mm->membarrier_state) &
361 MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE)))
363 sync_core_before_usermode();
366 extern void membarrier_exec_mmap(struct mm_struct *mm);
368 extern void membarrier_update_current_mm(struct mm_struct *next_mm);
371 #ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
372 static inline void membarrier_arch_switch_mm(struct mm_struct *prev,
373 struct mm_struct *next,
374 struct task_struct *tsk)
378 static inline void membarrier_exec_mmap(struct mm_struct *mm)
381 static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
384 static inline void membarrier_update_current_mm(struct mm_struct *next_mm)
389 #endif /* _LINUX_SCHED_MM_H */