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 extern void arch_pick_mmap_layout(struct mm_struct *mm,
110 struct rlimit *rlim_stack);
112 arch_get_unmapped_area(struct file *, unsigned long, unsigned long,
113 unsigned long, unsigned long);
115 arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr,
116 unsigned long len, unsigned long pgoff,
117 unsigned long flags);
119 static inline void arch_pick_mmap_layout(struct mm_struct *mm,
120 struct rlimit *rlim_stack) {}
123 static inline bool in_vfork(struct task_struct *tsk)
128 * need RCU to access ->real_parent if CLONE_VM was used along with
131 * We check real_parent->mm == tsk->mm because CLONE_VFORK does not
134 * CLONE_VFORK can be used with CLONE_PARENT/CLONE_THREAD and thus
135 * ->real_parent is not necessarily the task doing vfork(), so in
136 * theory we can't rely on task_lock() if we want to dereference it.
138 * And in this case we can't trust the real_parent->mm == tsk->mm
139 * check, it can be false negative. But we do not care, if init or
140 * another oom-unkillable task does this it should blame itself.
143 ret = tsk->vfork_done && tsk->real_parent->mm == tsk->mm;
150 * Applies per-task gfp context to the given allocation flags.
151 * PF_MEMALLOC_NOIO implies GFP_NOIO
152 * PF_MEMALLOC_NOFS implies GFP_NOFS
154 static inline gfp_t current_gfp_context(gfp_t flags)
156 unsigned int pflags = READ_ONCE(current->flags);
158 if (unlikely(pflags & (PF_MEMALLOC_NOIO | PF_MEMALLOC_NOFS))) {
160 * NOIO implies both NOIO and NOFS and it is a weaker context
161 * so always make sure it makes precedence
163 if (pflags & PF_MEMALLOC_NOIO)
164 flags &= ~(__GFP_IO | __GFP_FS);
165 else if (pflags & PF_MEMALLOC_NOFS)
171 #ifdef CONFIG_LOCKDEP
172 extern void __fs_reclaim_acquire(void);
173 extern void __fs_reclaim_release(void);
174 extern void fs_reclaim_acquire(gfp_t gfp_mask);
175 extern void fs_reclaim_release(gfp_t gfp_mask);
177 static inline void __fs_reclaim_acquire(void) { }
178 static inline void __fs_reclaim_release(void) { }
179 static inline void fs_reclaim_acquire(gfp_t gfp_mask) { }
180 static inline void fs_reclaim_release(gfp_t gfp_mask) { }
184 * might_alloc - Mark possible allocation sites
185 * @gfp_mask: gfp_t flags that would be used to allocate
187 * Similar to might_sleep() and other annotations, this can be used in functions
188 * that might allocate, but often don't. Compiles to nothing without
189 * CONFIG_LOCKDEP. Includes a conditional might_sleep() if @gfp allows blocking.
191 static inline void might_alloc(gfp_t gfp_mask)
193 fs_reclaim_acquire(gfp_mask);
194 fs_reclaim_release(gfp_mask);
196 might_sleep_if(gfpflags_allow_blocking(gfp_mask));
200 * memalloc_noio_save - Marks implicit GFP_NOIO allocation scope.
202 * This functions marks the beginning of the GFP_NOIO allocation scope.
203 * All further allocations will implicitly drop __GFP_IO flag and so
204 * they are safe for the IO critical section from the allocation recursion
205 * point of view. Use memalloc_noio_restore to end the scope with flags
206 * returned by this function.
208 * This function is safe to be used from any context.
210 static inline unsigned int memalloc_noio_save(void)
212 unsigned int flags = current->flags & PF_MEMALLOC_NOIO;
213 current->flags |= PF_MEMALLOC_NOIO;
218 * memalloc_noio_restore - Ends the implicit GFP_NOIO scope.
219 * @flags: Flags to restore.
221 * Ends the implicit GFP_NOIO scope started by memalloc_noio_save function.
222 * Always make sure that the given flags is the return value from the
223 * pairing memalloc_noio_save call.
225 static inline void memalloc_noio_restore(unsigned int flags)
227 current->flags = (current->flags & ~PF_MEMALLOC_NOIO) | flags;
231 * memalloc_nofs_save - Marks implicit GFP_NOFS allocation scope.
233 * This functions marks the beginning of the GFP_NOFS allocation scope.
234 * All further allocations will implicitly drop __GFP_FS flag and so
235 * they are safe for the FS critical section from the allocation recursion
236 * point of view. Use memalloc_nofs_restore to end the scope with flags
237 * returned by this function.
239 * This function is safe to be used from any context.
241 static inline unsigned int memalloc_nofs_save(void)
243 unsigned int flags = current->flags & PF_MEMALLOC_NOFS;
244 current->flags |= PF_MEMALLOC_NOFS;
249 * memalloc_nofs_restore - Ends the implicit GFP_NOFS scope.
250 * @flags: Flags to restore.
252 * Ends the implicit GFP_NOFS scope started by memalloc_nofs_save function.
253 * Always make sure that the given flags is the return value from the
254 * pairing memalloc_nofs_save call.
256 static inline void memalloc_nofs_restore(unsigned int flags)
258 current->flags = (current->flags & ~PF_MEMALLOC_NOFS) | flags;
261 static inline unsigned int memalloc_noreclaim_save(void)
263 unsigned int flags = current->flags & PF_MEMALLOC;
264 current->flags |= PF_MEMALLOC;
268 static inline void memalloc_noreclaim_restore(unsigned int flags)
270 current->flags = (current->flags & ~PF_MEMALLOC) | flags;
274 static inline unsigned int memalloc_nocma_save(void)
276 unsigned int flags = current->flags & PF_MEMALLOC_NOCMA;
278 current->flags |= PF_MEMALLOC_NOCMA;
282 static inline void memalloc_nocma_restore(unsigned int flags)
284 current->flags = (current->flags & ~PF_MEMALLOC_NOCMA) | flags;
287 static inline unsigned int memalloc_nocma_save(void)
292 static inline void memalloc_nocma_restore(unsigned int flags)
298 DECLARE_PER_CPU(struct mem_cgroup *, int_active_memcg);
300 * set_active_memcg - Starts the remote memcg charging scope.
301 * @memcg: memcg to charge.
303 * This function marks the beginning of the remote memcg charging scope. All the
304 * __GFP_ACCOUNT allocations till the end of the scope will be charged to the
307 * NOTE: This function can nest. Users must save the return value and
308 * reset the previous value after their own charging scope is over.
310 static inline struct mem_cgroup *
311 set_active_memcg(struct mem_cgroup *memcg)
313 struct mem_cgroup *old;
315 if (in_interrupt()) {
316 old = this_cpu_read(int_active_memcg);
317 this_cpu_write(int_active_memcg, memcg);
319 old = current->active_memcg;
320 current->active_memcg = memcg;
326 static inline struct mem_cgroup *
327 set_active_memcg(struct mem_cgroup *memcg)
333 #ifdef CONFIG_MEMBARRIER
335 MEMBARRIER_STATE_PRIVATE_EXPEDITED_READY = (1U << 0),
336 MEMBARRIER_STATE_PRIVATE_EXPEDITED = (1U << 1),
337 MEMBARRIER_STATE_GLOBAL_EXPEDITED_READY = (1U << 2),
338 MEMBARRIER_STATE_GLOBAL_EXPEDITED = (1U << 3),
339 MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE_READY = (1U << 4),
340 MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE = (1U << 5),
341 MEMBARRIER_STATE_PRIVATE_EXPEDITED_RSEQ_READY = (1U << 6),
342 MEMBARRIER_STATE_PRIVATE_EXPEDITED_RSEQ = (1U << 7),
346 MEMBARRIER_FLAG_SYNC_CORE = (1U << 0),
347 MEMBARRIER_FLAG_RSEQ = (1U << 1),
350 #ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
351 #include <asm/membarrier.h>
354 static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
356 if (current->mm != mm)
358 if (likely(!(atomic_read(&mm->membarrier_state) &
359 MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE)))
361 sync_core_before_usermode();
364 extern void membarrier_exec_mmap(struct mm_struct *mm);
366 extern void membarrier_update_current_mm(struct mm_struct *next_mm);
369 #ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
370 static inline void membarrier_arch_switch_mm(struct mm_struct *prev,
371 struct mm_struct *next,
372 struct task_struct *tsk)
376 static inline void membarrier_exec_mmap(struct mm_struct *mm)
379 static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
382 static inline void membarrier_update_current_mm(struct mm_struct *next_mm)
387 #endif /* _LINUX_SCHED_MM_H */