1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_SCHED_MM_H
3 #define _LINUX_SCHED_MM_H
4 
5 #include <linux/kernel.h>
6 #include <linux/atomic.h>
7 #include <linux/sched.h>
8 #include <linux/mm_types.h>
9 #include <linux/gfp.h>
10 #include <linux/sync_core.h>
11 
12 /*
13  * Routines for handling mm_structs
14  */
15 extern struct mm_struct *mm_alloc(void);
16 
17 /**
18  * mmgrab() - Pin a &struct mm_struct.
19  * @mm: The &struct mm_struct to pin.
20  *
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
24  * accessing it.
25  *
26  * This is a preferred way to pin @mm for a longer/unbounded amount
27  * of time.
28  *
29  * Use mmdrop() to release the reference acquired by mmgrab().
30  *
31  * See also <Documentation/mm/active_mm.rst> for an in-depth explanation
32  * of &mm_struct.mm_count vs &mm_struct.mm_users.
33  */
mmgrab(struct mm_struct * mm)34 static inline void mmgrab(struct mm_struct *mm)
35 {
36 	atomic_inc(&mm->mm_count);
37 }
38 
smp_mb__after_mmgrab(void)39 static inline void smp_mb__after_mmgrab(void)
40 {
41 	smp_mb__after_atomic();
42 }
43 
44 extern void __mmdrop(struct mm_struct *mm);
45 
mmdrop(struct mm_struct * mm)46 static inline void mmdrop(struct mm_struct *mm)
47 {
48 	/*
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.
52 	 */
53 	if (unlikely(atomic_dec_and_test(&mm->mm_count)))
54 		__mmdrop(mm);
55 }
56 
57 #ifdef CONFIG_PREEMPT_RT
58 /*
59  * RCU callback for delayed mm drop. Not strictly RCU, but call_rcu() is
60  * by far the least expensive way to do that.
61  */
__mmdrop_delayed(struct rcu_head * rhp)62 static inline void __mmdrop_delayed(struct rcu_head *rhp)
63 {
64 	struct mm_struct *mm = container_of(rhp, struct mm_struct, delayed_drop);
65 
66 	__mmdrop(mm);
67 }
68 
69 /*
70  * Invoked from finish_task_switch(). Delegates the heavy lifting on RT
71  * kernels via RCU.
72  */
mmdrop_sched(struct mm_struct * mm)73 static inline void mmdrop_sched(struct mm_struct *mm)
74 {
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);
78 }
79 #else
mmdrop_sched(struct mm_struct * mm)80 static inline void mmdrop_sched(struct mm_struct *mm)
81 {
82 	mmdrop(mm);
83 }
84 #endif
85 
86 /* Helpers for lazy TLB mm refcounting */
mmgrab_lazy_tlb(struct mm_struct * mm)87 static inline void mmgrab_lazy_tlb(struct mm_struct *mm)
88 {
89 	if (IS_ENABLED(CONFIG_MMU_LAZY_TLB_REFCOUNT))
90 		mmgrab(mm);
91 }
92 
mmdrop_lazy_tlb(struct mm_struct * mm)93 static inline void mmdrop_lazy_tlb(struct mm_struct *mm)
94 {
95 	if (IS_ENABLED(CONFIG_MMU_LAZY_TLB_REFCOUNT)) {
96 		mmdrop(mm);
97 	} else {
98 		/*
99 		 * mmdrop_lazy_tlb must provide a full memory barrier, see the
100 		 * membarrier comment finish_task_switch which relies on this.
101 		 */
102 		smp_mb();
103 	}
104 }
105 
mmdrop_lazy_tlb_sched(struct mm_struct * mm)106 static inline void mmdrop_lazy_tlb_sched(struct mm_struct *mm)
107 {
108 	if (IS_ENABLED(CONFIG_MMU_LAZY_TLB_REFCOUNT))
109 		mmdrop_sched(mm);
110 	else
111 		smp_mb(); /* see mmdrop_lazy_tlb() above */
112 }
113 
114 /**
115  * mmget() - Pin the address space associated with a &struct mm_struct.
116  * @mm: The address space to pin.
117  *
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.
121  *
122  * Never use this function to pin this address space for an
123  * unbounded/indefinite amount of time.
124  *
125  * Use mmput() to release the reference acquired by mmget().
126  *
127  * See also <Documentation/mm/active_mm.rst> for an in-depth explanation
128  * of &mm_struct.mm_count vs &mm_struct.mm_users.
129  */
mmget(struct mm_struct * mm)130 static inline void mmget(struct mm_struct *mm)
131 {
132 	atomic_inc(&mm->mm_users);
133 }
134 
mmget_not_zero(struct mm_struct * mm)135 static inline bool mmget_not_zero(struct mm_struct *mm)
136 {
137 	return atomic_inc_not_zero(&mm->mm_users);
138 }
139 
140 /* mmput gets rid of the mappings and all user-space */
141 extern void mmput(struct mm_struct *);
142 #ifdef CONFIG_MMU
143 /* same as above but performs the slow path from the async context. Can
144  * be called from the atomic context as well
145  */
146 void mmput_async(struct mm_struct *);
147 #endif
148 
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);
151 /*
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
154  * succeeds.
155  */
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 *);
161 
162 #ifdef CONFIG_MEMCG
163 extern void mm_update_next_owner(struct mm_struct *mm);
164 #else
mm_update_next_owner(struct mm_struct * mm)165 static inline void mm_update_next_owner(struct mm_struct *mm)
166 {
167 }
168 #endif /* CONFIG_MEMCG */
169 
170 #ifdef CONFIG_MMU
171 #ifndef arch_get_mmap_end
172 #define arch_get_mmap_end(addr, len, flags)	(TASK_SIZE)
173 #endif
174 
175 #ifndef arch_get_mmap_base
176 #define arch_get_mmap_base(addr, base) (base)
177 #endif
178 
179 extern void arch_pick_mmap_layout(struct mm_struct *mm,
180 				  struct rlimit *rlim_stack);
181 extern unsigned long
182 arch_get_unmapped_area(struct file *, unsigned long, unsigned long,
183 		       unsigned long, unsigned long);
184 extern 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);
188 
189 unsigned long
190 generic_get_unmapped_area(struct file *filp, unsigned long addr,
191 			  unsigned long len, unsigned long pgoff,
192 			  unsigned long flags);
193 unsigned long
194 generic_get_unmapped_area_topdown(struct file *filp, unsigned long addr,
195 				  unsigned long len, unsigned long pgoff,
196 				  unsigned long flags);
197 #else
arch_pick_mmap_layout(struct mm_struct * mm,struct rlimit * rlim_stack)198 static inline void arch_pick_mmap_layout(struct mm_struct *mm,
199 					 struct rlimit *rlim_stack) {}
200 #endif
201 
in_vfork(struct task_struct * tsk)202 static inline bool in_vfork(struct task_struct *tsk)
203 {
204 	bool ret;
205 
206 	/*
207 	 * need RCU to access ->real_parent if CLONE_VM was used along with
208 	 * CLONE_PARENT.
209 	 *
210 	 * We check real_parent->mm == tsk->mm because CLONE_VFORK does not
211 	 * imply CLONE_VM
212 	 *
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.
216 	 *
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.
220 	 */
221 	rcu_read_lock();
222 	ret = tsk->vfork_done &&
223 			rcu_dereference(tsk->real_parent)->mm == tsk->mm;
224 	rcu_read_unlock();
225 
226 	return ret;
227 }
228 
229 /*
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
234  */
current_gfp_context(gfp_t flags)235 static inline gfp_t current_gfp_context(gfp_t flags)
236 {
237 	unsigned int pflags = READ_ONCE(current->flags);
238 
239 	if (unlikely(pflags & (PF_MEMALLOC_NOIO | PF_MEMALLOC_NOFS | PF_MEMALLOC_PIN))) {
240 		/*
241 		 * NOIO implies both NOIO and NOFS and it is a weaker context
242 		 * so always make sure it makes precedence
243 		 */
244 		if (pflags & PF_MEMALLOC_NOIO)
245 			flags &= ~(__GFP_IO | __GFP_FS);
246 		else if (pflags & PF_MEMALLOC_NOFS)
247 			flags &= ~__GFP_FS;
248 
249 		if (pflags & PF_MEMALLOC_PIN)
250 			flags &= ~__GFP_MOVABLE;
251 	}
252 	return flags;
253 }
254 
255 #ifdef CONFIG_LOCKDEP
256 extern void __fs_reclaim_acquire(unsigned long ip);
257 extern void __fs_reclaim_release(unsigned long ip);
258 extern void fs_reclaim_acquire(gfp_t gfp_mask);
259 extern void fs_reclaim_release(gfp_t gfp_mask);
260 #else
__fs_reclaim_acquire(unsigned long ip)261 static inline void __fs_reclaim_acquire(unsigned long ip) { }
__fs_reclaim_release(unsigned long ip)262 static inline void __fs_reclaim_release(unsigned long ip) { }
fs_reclaim_acquire(gfp_t gfp_mask)263 static inline void fs_reclaim_acquire(gfp_t gfp_mask) { }
fs_reclaim_release(gfp_t gfp_mask)264 static inline void fs_reclaim_release(gfp_t gfp_mask) { }
265 #endif
266 
267 /* Any memory-allocation retry loop should use
268  * memalloc_retry_wait(), and pass the flags for the most
269  * constrained allocation attempt that might have failed.
270  * This provides useful documentation of where loops are,
271  * and a central place to fine tune the waiting as the MM
272  * implementation changes.
273  */
memalloc_retry_wait(gfp_t gfp_flags)274 static inline void memalloc_retry_wait(gfp_t gfp_flags)
275 {
276 	/* We use io_schedule_timeout because waiting for memory
277 	 * typically included waiting for dirty pages to be
278 	 * written out, which requires IO.
279 	 */
280 	__set_current_state(TASK_UNINTERRUPTIBLE);
281 	gfp_flags = current_gfp_context(gfp_flags);
282 	if (gfpflags_allow_blocking(gfp_flags) &&
283 	    !(gfp_flags & __GFP_NORETRY))
284 		/* Probably waited already, no need for much more */
285 		io_schedule_timeout(1);
286 	else
287 		/* Probably didn't wait, and has now released a lock,
288 		 * so now is a good time to wait
289 		 */
290 		io_schedule_timeout(HZ/50);
291 }
292 
293 /**
294  * might_alloc - Mark possible allocation sites
295  * @gfp_mask: gfp_t flags that would be used to allocate
296  *
297  * Similar to might_sleep() and other annotations, this can be used in functions
298  * that might allocate, but often don't. Compiles to nothing without
299  * CONFIG_LOCKDEP. Includes a conditional might_sleep() if @gfp allows blocking.
300  */
might_alloc(gfp_t gfp_mask)301 static inline void might_alloc(gfp_t gfp_mask)
302 {
303 	fs_reclaim_acquire(gfp_mask);
304 	fs_reclaim_release(gfp_mask);
305 
306 	might_sleep_if(gfpflags_allow_blocking(gfp_mask));
307 }
308 
309 /**
310  * memalloc_noio_save - Marks implicit GFP_NOIO allocation scope.
311  *
312  * This functions marks the beginning of the GFP_NOIO allocation scope.
313  * All further allocations will implicitly drop __GFP_IO flag and so
314  * they are safe for the IO critical section from the allocation recursion
315  * point of view. Use memalloc_noio_restore to end the scope with flags
316  * returned by this function.
317  *
318  * This function is safe to be used from any context.
319  */
memalloc_noio_save(void)320 static inline unsigned int memalloc_noio_save(void)
321 {
322 	unsigned int flags = current->flags & PF_MEMALLOC_NOIO;
323 	current->flags |= PF_MEMALLOC_NOIO;
324 	return flags;
325 }
326 
327 /**
328  * memalloc_noio_restore - Ends the implicit GFP_NOIO scope.
329  * @flags: Flags to restore.
330  *
331  * Ends the implicit GFP_NOIO scope started by memalloc_noio_save function.
332  * Always make sure that the given flags is the return value from the
333  * pairing memalloc_noio_save call.
334  */
memalloc_noio_restore(unsigned int flags)335 static inline void memalloc_noio_restore(unsigned int flags)
336 {
337 	current->flags = (current->flags & ~PF_MEMALLOC_NOIO) | flags;
338 }
339 
340 /**
341  * memalloc_nofs_save - Marks implicit GFP_NOFS allocation scope.
342  *
343  * This functions marks the beginning of the GFP_NOFS allocation scope.
344  * All further allocations will implicitly drop __GFP_FS flag and so
345  * they are safe for the FS critical section from the allocation recursion
346  * point of view. Use memalloc_nofs_restore to end the scope with flags
347  * returned by this function.
348  *
349  * This function is safe to be used from any context.
350  */
memalloc_nofs_save(void)351 static inline unsigned int memalloc_nofs_save(void)
352 {
353 	unsigned int flags = current->flags & PF_MEMALLOC_NOFS;
354 	current->flags |= PF_MEMALLOC_NOFS;
355 	return flags;
356 }
357 
358 /**
359  * memalloc_nofs_restore - Ends the implicit GFP_NOFS scope.
360  * @flags: Flags to restore.
361  *
362  * Ends the implicit GFP_NOFS scope started by memalloc_nofs_save function.
363  * Always make sure that the given flags is the return value from the
364  * pairing memalloc_nofs_save call.
365  */
memalloc_nofs_restore(unsigned int flags)366 static inline void memalloc_nofs_restore(unsigned int flags)
367 {
368 	current->flags = (current->flags & ~PF_MEMALLOC_NOFS) | flags;
369 }
370 
memalloc_noreclaim_save(void)371 static inline unsigned int memalloc_noreclaim_save(void)
372 {
373 	unsigned int flags = current->flags & PF_MEMALLOC;
374 	current->flags |= PF_MEMALLOC;
375 	return flags;
376 }
377 
memalloc_noreclaim_restore(unsigned int flags)378 static inline void memalloc_noreclaim_restore(unsigned int flags)
379 {
380 	current->flags = (current->flags & ~PF_MEMALLOC) | flags;
381 }
382 
memalloc_pin_save(void)383 static inline unsigned int memalloc_pin_save(void)
384 {
385 	unsigned int flags = current->flags & PF_MEMALLOC_PIN;
386 
387 	current->flags |= PF_MEMALLOC_PIN;
388 	return flags;
389 }
390 
memalloc_pin_restore(unsigned int flags)391 static inline void memalloc_pin_restore(unsigned int flags)
392 {
393 	current->flags = (current->flags & ~PF_MEMALLOC_PIN) | flags;
394 }
395 
396 #ifdef CONFIG_MEMCG
397 DECLARE_PER_CPU(struct mem_cgroup *, int_active_memcg);
398 /**
399  * set_active_memcg - Starts the remote memcg charging scope.
400  * @memcg: memcg to charge.
401  *
402  * This function marks the beginning of the remote memcg charging scope. All the
403  * __GFP_ACCOUNT allocations till the end of the scope will be charged to the
404  * given memcg.
405  *
406  * NOTE: This function can nest. Users must save the return value and
407  * reset the previous value after their own charging scope is over.
408  */
409 static inline struct mem_cgroup *
set_active_memcg(struct mem_cgroup * memcg)410 set_active_memcg(struct mem_cgroup *memcg)
411 {
412 	struct mem_cgroup *old;
413 
414 	if (!in_task()) {
415 		old = this_cpu_read(int_active_memcg);
416 		this_cpu_write(int_active_memcg, memcg);
417 	} else {
418 		old = current->active_memcg;
419 		current->active_memcg = memcg;
420 	}
421 
422 	return old;
423 }
424 #else
425 static inline struct mem_cgroup *
set_active_memcg(struct mem_cgroup * memcg)426 set_active_memcg(struct mem_cgroup *memcg)
427 {
428 	return NULL;
429 }
430 #endif
431 
432 #ifdef CONFIG_MEMBARRIER
433 enum {
434 	MEMBARRIER_STATE_PRIVATE_EXPEDITED_READY		= (1U << 0),
435 	MEMBARRIER_STATE_PRIVATE_EXPEDITED			= (1U << 1),
436 	MEMBARRIER_STATE_GLOBAL_EXPEDITED_READY			= (1U << 2),
437 	MEMBARRIER_STATE_GLOBAL_EXPEDITED			= (1U << 3),
438 	MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE_READY	= (1U << 4),
439 	MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE		= (1U << 5),
440 	MEMBARRIER_STATE_PRIVATE_EXPEDITED_RSEQ_READY		= (1U << 6),
441 	MEMBARRIER_STATE_PRIVATE_EXPEDITED_RSEQ			= (1U << 7),
442 };
443 
444 enum {
445 	MEMBARRIER_FLAG_SYNC_CORE	= (1U << 0),
446 	MEMBARRIER_FLAG_RSEQ		= (1U << 1),
447 };
448 
449 #ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
450 #include <asm/membarrier.h>
451 #endif
452 
membarrier_mm_sync_core_before_usermode(struct mm_struct * mm)453 static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
454 {
455 	if (current->mm != mm)
456 		return;
457 	if (likely(!(atomic_read(&mm->membarrier_state) &
458 		     MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE)))
459 		return;
460 	sync_core_before_usermode();
461 }
462 
463 extern void membarrier_exec_mmap(struct mm_struct *mm);
464 
465 extern void membarrier_update_current_mm(struct mm_struct *next_mm);
466 
467 #else
468 #ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
membarrier_arch_switch_mm(struct mm_struct * prev,struct mm_struct * next,struct task_struct * tsk)469 static inline void membarrier_arch_switch_mm(struct mm_struct *prev,
470 					     struct mm_struct *next,
471 					     struct task_struct *tsk)
472 {
473 }
474 #endif
membarrier_exec_mmap(struct mm_struct * mm)475 static inline void membarrier_exec_mmap(struct mm_struct *mm)
476 {
477 }
membarrier_mm_sync_core_before_usermode(struct mm_struct * mm)478 static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
479 {
480 }
membarrier_update_current_mm(struct mm_struct * next_mm)481 static inline void membarrier_update_current_mm(struct mm_struct *next_mm)
482 {
483 }
484 #endif
485 
486 #endif /* _LINUX_SCHED_MM_H */
487