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>
11 #include <linux/ioasid.h>
14 * Routines for handling mm_structs
16 extern struct mm_struct *mm_alloc(void);
19 * mmgrab() - Pin a &struct mm_struct.
20 * @mm: The &struct mm_struct to pin.
22 * Make sure that @mm will not get freed even after the owning task
23 * exits. This doesn't guarantee that the associated address space
24 * will still exist later on and mmget_not_zero() has to be used before
27 * This is a preferred way to pin @mm for a longer/unbounded amount
30 * Use mmdrop() to release the reference acquired by mmgrab().
32 * See also <Documentation/vm/active_mm.rst> for an in-depth explanation
33 * of &mm_struct.mm_count vs &mm_struct.mm_users.
35 static inline void mmgrab(struct mm_struct *mm)
37 atomic_inc(&mm->mm_count);
40 extern void __mmdrop(struct mm_struct *mm);
42 static inline void mmdrop(struct mm_struct *mm)
45 * The implicit full barrier implied by atomic_dec_and_test() is
46 * required by the membarrier system call before returning to
47 * user-space, after storing to rq->curr.
49 if (unlikely(atomic_dec_and_test(&mm->mm_count)))
53 #ifdef CONFIG_PREEMPT_RT
55 * RCU callback for delayed mm drop. Not strictly RCU, but call_rcu() is
56 * by far the least expensive way to do that.
58 static inline void __mmdrop_delayed(struct rcu_head *rhp)
60 struct mm_struct *mm = container_of(rhp, struct mm_struct, delayed_drop);
66 * Invoked from finish_task_switch(). Delegates the heavy lifting on RT
69 static inline void mmdrop_sched(struct mm_struct *mm)
71 /* Provides a full memory barrier. See mmdrop() */
72 if (atomic_dec_and_test(&mm->mm_count))
73 call_rcu(&mm->delayed_drop, __mmdrop_delayed);
76 static inline void mmdrop_sched(struct mm_struct *mm)
83 * mmget() - Pin the address space associated with a &struct mm_struct.
84 * @mm: The address space to pin.
86 * Make sure that the address space of the given &struct mm_struct doesn't
87 * go away. This does not protect against parts of the address space being
88 * modified or freed, however.
90 * Never use this function to pin this address space for an
91 * unbounded/indefinite amount of time.
93 * Use mmput() to release the reference acquired by mmget().
95 * See also <Documentation/vm/active_mm.rst> for an in-depth explanation
96 * of &mm_struct.mm_count vs &mm_struct.mm_users.
98 static inline void mmget(struct mm_struct *mm)
100 atomic_inc(&mm->mm_users);
103 static inline bool mmget_not_zero(struct mm_struct *mm)
105 return atomic_inc_not_zero(&mm->mm_users);
108 /* mmput gets rid of the mappings and all user-space */
109 extern void mmput(struct mm_struct *);
111 /* same as above but performs the slow path from the async context. Can
112 * be called from the atomic context as well
114 void mmput_async(struct mm_struct *);
117 /* Grab a reference to a task's mm, if it is not already going away */
118 extern struct mm_struct *get_task_mm(struct task_struct *task);
120 * Grab a reference to a task's mm, if it is not already going away
121 * and ptrace_may_access with the mode parameter passed to it
124 extern struct mm_struct *mm_access(struct task_struct *task, unsigned int mode);
125 /* Remove the current tasks stale references to the old mm_struct on exit() */
126 extern void exit_mm_release(struct task_struct *, struct mm_struct *);
127 /* Remove the current tasks stale references to the old mm_struct on exec() */
128 extern void exec_mm_release(struct task_struct *, struct mm_struct *);
131 extern void mm_update_next_owner(struct mm_struct *mm);
133 static inline void mm_update_next_owner(struct mm_struct *mm)
136 #endif /* CONFIG_MEMCG */
139 extern void arch_pick_mmap_layout(struct mm_struct *mm,
140 struct rlimit *rlim_stack);
142 arch_get_unmapped_area(struct file *, unsigned long, unsigned long,
143 unsigned long, unsigned long);
145 arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr,
146 unsigned long len, unsigned long pgoff,
147 unsigned long flags);
149 static inline void arch_pick_mmap_layout(struct mm_struct *mm,
150 struct rlimit *rlim_stack) {}
153 static inline bool in_vfork(struct task_struct *tsk)
158 * need RCU to access ->real_parent if CLONE_VM was used along with
161 * We check real_parent->mm == tsk->mm because CLONE_VFORK does not
164 * CLONE_VFORK can be used with CLONE_PARENT/CLONE_THREAD and thus
165 * ->real_parent is not necessarily the task doing vfork(), so in
166 * theory we can't rely on task_lock() if we want to dereference it.
168 * And in this case we can't trust the real_parent->mm == tsk->mm
169 * check, it can be false negative. But we do not care, if init or
170 * another oom-unkillable task does this it should blame itself.
173 ret = tsk->vfork_done &&
174 rcu_dereference(tsk->real_parent)->mm == tsk->mm;
181 * Applies per-task gfp context to the given allocation flags.
182 * PF_MEMALLOC_NOIO implies GFP_NOIO
183 * PF_MEMALLOC_NOFS implies GFP_NOFS
184 * PF_MEMALLOC_PIN implies !GFP_MOVABLE
186 static inline gfp_t current_gfp_context(gfp_t flags)
188 unsigned int pflags = READ_ONCE(current->flags);
190 if (unlikely(pflags & (PF_MEMALLOC_NOIO | PF_MEMALLOC_NOFS | PF_MEMALLOC_PIN))) {
192 * NOIO implies both NOIO and NOFS and it is a weaker context
193 * so always make sure it makes precedence
195 if (pflags & PF_MEMALLOC_NOIO)
196 flags &= ~(__GFP_IO | __GFP_FS);
197 else if (pflags & PF_MEMALLOC_NOFS)
200 if (pflags & PF_MEMALLOC_PIN)
201 flags &= ~__GFP_MOVABLE;
206 #ifdef CONFIG_LOCKDEP
207 extern void __fs_reclaim_acquire(unsigned long ip);
208 extern void __fs_reclaim_release(unsigned long ip);
209 extern void fs_reclaim_acquire(gfp_t gfp_mask);
210 extern void fs_reclaim_release(gfp_t gfp_mask);
212 static inline void __fs_reclaim_acquire(unsigned long ip) { }
213 static inline void __fs_reclaim_release(unsigned long ip) { }
214 static inline void fs_reclaim_acquire(gfp_t gfp_mask) { }
215 static inline void fs_reclaim_release(gfp_t gfp_mask) { }
218 /* Any memory-allocation retry loop should use
219 * memalloc_retry_wait(), and pass the flags for the most
220 * constrained allocation attempt that might have failed.
221 * This provides useful documentation of where loops are,
222 * and a central place to fine tune the waiting as the MM
223 * implementation changes.
225 static inline void memalloc_retry_wait(gfp_t gfp_flags)
227 /* We use io_schedule_timeout because waiting for memory
228 * typically included waiting for dirty pages to be
229 * written out, which requires IO.
231 __set_current_state(TASK_UNINTERRUPTIBLE);
232 gfp_flags = current_gfp_context(gfp_flags);
233 if (gfpflags_allow_blocking(gfp_flags) &&
234 !(gfp_flags & __GFP_NORETRY))
235 /* Probably waited already, no need for much more */
236 io_schedule_timeout(1);
238 /* Probably didn't wait, and has now released a lock,
239 * so now is a good time to wait
241 io_schedule_timeout(HZ/50);
245 * might_alloc - Mark possible allocation sites
246 * @gfp_mask: gfp_t flags that would be used to allocate
248 * Similar to might_sleep() and other annotations, this can be used in functions
249 * that might allocate, but often don't. Compiles to nothing without
250 * CONFIG_LOCKDEP. Includes a conditional might_sleep() if @gfp allows blocking.
252 static inline void might_alloc(gfp_t gfp_mask)
254 fs_reclaim_acquire(gfp_mask);
255 fs_reclaim_release(gfp_mask);
257 might_sleep_if(gfpflags_allow_blocking(gfp_mask));
261 * memalloc_noio_save - Marks implicit GFP_NOIO allocation scope.
263 * This functions marks the beginning of the GFP_NOIO allocation scope.
264 * All further allocations will implicitly drop __GFP_IO flag and so
265 * they are safe for the IO critical section from the allocation recursion
266 * point of view. Use memalloc_noio_restore to end the scope with flags
267 * returned by this function.
269 * This function is safe to be used from any context.
271 static inline unsigned int memalloc_noio_save(void)
273 unsigned int flags = current->flags & PF_MEMALLOC_NOIO;
274 current->flags |= PF_MEMALLOC_NOIO;
279 * memalloc_noio_restore - Ends the implicit GFP_NOIO scope.
280 * @flags: Flags to restore.
282 * Ends the implicit GFP_NOIO scope started by memalloc_noio_save function.
283 * Always make sure that the given flags is the return value from the
284 * pairing memalloc_noio_save call.
286 static inline void memalloc_noio_restore(unsigned int flags)
288 current->flags = (current->flags & ~PF_MEMALLOC_NOIO) | flags;
292 * memalloc_nofs_save - Marks implicit GFP_NOFS allocation scope.
294 * This functions marks the beginning of the GFP_NOFS allocation scope.
295 * All further allocations will implicitly drop __GFP_FS flag and so
296 * they are safe for the FS critical section from the allocation recursion
297 * point of view. Use memalloc_nofs_restore to end the scope with flags
298 * returned by this function.
300 * This function is safe to be used from any context.
302 static inline unsigned int memalloc_nofs_save(void)
304 unsigned int flags = current->flags & PF_MEMALLOC_NOFS;
305 current->flags |= PF_MEMALLOC_NOFS;
310 * memalloc_nofs_restore - Ends the implicit GFP_NOFS scope.
311 * @flags: Flags to restore.
313 * Ends the implicit GFP_NOFS scope started by memalloc_nofs_save function.
314 * Always make sure that the given flags is the return value from the
315 * pairing memalloc_nofs_save call.
317 static inline void memalloc_nofs_restore(unsigned int flags)
319 current->flags = (current->flags & ~PF_MEMALLOC_NOFS) | flags;
322 static inline unsigned int memalloc_noreclaim_save(void)
324 unsigned int flags = current->flags & PF_MEMALLOC;
325 current->flags |= PF_MEMALLOC;
329 static inline void memalloc_noreclaim_restore(unsigned int flags)
331 current->flags = (current->flags & ~PF_MEMALLOC) | flags;
334 static inline unsigned int memalloc_pin_save(void)
336 unsigned int flags = current->flags & PF_MEMALLOC_PIN;
338 current->flags |= PF_MEMALLOC_PIN;
342 static inline void memalloc_pin_restore(unsigned int flags)
344 current->flags = (current->flags & ~PF_MEMALLOC_PIN) | flags;
348 DECLARE_PER_CPU(struct mem_cgroup *, int_active_memcg);
350 * set_active_memcg - Starts the remote memcg charging scope.
351 * @memcg: memcg to charge.
353 * This function marks the beginning of the remote memcg charging scope. All the
354 * __GFP_ACCOUNT allocations till the end of the scope will be charged to the
357 * NOTE: This function can nest. Users must save the return value and
358 * reset the previous value after their own charging scope is over.
360 static inline struct mem_cgroup *
361 set_active_memcg(struct mem_cgroup *memcg)
363 struct mem_cgroup *old;
366 old = this_cpu_read(int_active_memcg);
367 this_cpu_write(int_active_memcg, memcg);
369 old = current->active_memcg;
370 current->active_memcg = memcg;
376 static inline struct mem_cgroup *
377 set_active_memcg(struct mem_cgroup *memcg)
383 #ifdef CONFIG_MEMBARRIER
385 MEMBARRIER_STATE_PRIVATE_EXPEDITED_READY = (1U << 0),
386 MEMBARRIER_STATE_PRIVATE_EXPEDITED = (1U << 1),
387 MEMBARRIER_STATE_GLOBAL_EXPEDITED_READY = (1U << 2),
388 MEMBARRIER_STATE_GLOBAL_EXPEDITED = (1U << 3),
389 MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE_READY = (1U << 4),
390 MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE = (1U << 5),
391 MEMBARRIER_STATE_PRIVATE_EXPEDITED_RSEQ_READY = (1U << 6),
392 MEMBARRIER_STATE_PRIVATE_EXPEDITED_RSEQ = (1U << 7),
396 MEMBARRIER_FLAG_SYNC_CORE = (1U << 0),
397 MEMBARRIER_FLAG_RSEQ = (1U << 1),
400 #ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
401 #include <asm/membarrier.h>
404 static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
406 if (current->mm != mm)
408 if (likely(!(atomic_read(&mm->membarrier_state) &
409 MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE)))
411 sync_core_before_usermode();
414 extern void membarrier_exec_mmap(struct mm_struct *mm);
416 extern void membarrier_update_current_mm(struct mm_struct *next_mm);
419 #ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
420 static inline void membarrier_arch_switch_mm(struct mm_struct *prev,
421 struct mm_struct *next,
422 struct task_struct *tsk)
426 static inline void membarrier_exec_mmap(struct mm_struct *mm)
429 static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
432 static inline void membarrier_update_current_mm(struct mm_struct *next_mm)
437 #ifdef CONFIG_IOMMU_SVA
438 static inline void mm_pasid_init(struct mm_struct *mm)
440 mm->pasid = INVALID_IOASID;
443 /* Associate a PASID with an mm_struct: */
444 static inline void mm_pasid_set(struct mm_struct *mm, u32 pasid)
449 static inline void mm_pasid_drop(struct mm_struct *mm)
451 if (pasid_valid(mm->pasid)) {
452 ioasid_free(mm->pasid);
453 mm->pasid = INVALID_IOASID;
457 static inline void mm_pasid_init(struct mm_struct *mm) {}
458 static inline void mm_pasid_set(struct mm_struct *mm, u32 pasid) {}
459 static inline void mm_pasid_drop(struct mm_struct *mm) {}
462 #endif /* _LINUX_SCHED_MM_H */