1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_MMU_NOTIFIER_H
3 #define _LINUX_MMU_NOTIFIER_H
4 
5 #include <linux/list.h>
6 #include <linux/spinlock.h>
7 #include <linux/mm_types.h>
8 #include <linux/mmap_lock.h>
9 #include <linux/srcu.h>
10 #include <linux/interval_tree.h>
11 
12 struct mmu_notifier_subscriptions;
13 struct mmu_notifier;
14 struct mmu_notifier_range;
15 struct mmu_interval_notifier;
16 
17 /**
18  * enum mmu_notifier_event - reason for the mmu notifier callback
19  * @MMU_NOTIFY_UNMAP: either munmap() that unmap the range or a mremap() that
20  * move the range
21  *
22  * @MMU_NOTIFY_CLEAR: clear page table entry (many reasons for this like
23  * madvise() or replacing a page by another one, ...).
24  *
25  * @MMU_NOTIFY_PROTECTION_VMA: update is due to protection change for the range
26  * ie using the vma access permission (vm_page_prot) to update the whole range
27  * is enough no need to inspect changes to the CPU page table (mprotect()
28  * syscall)
29  *
30  * @MMU_NOTIFY_PROTECTION_PAGE: update is due to change in read/write flag for
31  * pages in the range so to mirror those changes the user must inspect the CPU
32  * page table (from the end callback).
33  *
34  * @MMU_NOTIFY_SOFT_DIRTY: soft dirty accounting (still same page and same
35  * access flags). User should soft dirty the page in the end callback to make
36  * sure that anyone relying on soft dirtiness catch pages that might be written
37  * through non CPU mappings.
38  *
39  * @MMU_NOTIFY_RELEASE: used during mmu_interval_notifier invalidate to signal
40  * that the mm refcount is zero and the range is no longer accessible.
41  *
42  * @MMU_NOTIFY_MIGRATE: used during migrate_vma_collect() invalidate to signal
43  * a device driver to possibly ignore the invalidation if the
44  * owner field matches the driver's device private pgmap owner.
45  *
46  * @MMU_NOTIFY_EXCLUSIVE: to signal a device driver that the device will no
47  * longer have exclusive access to the page. When sent during creation of an
48  * exclusive range the owner will be initialised to the value provided by the
49  * caller of make_device_exclusive_range(), otherwise the owner will be NULL.
50  */
51 enum mmu_notifier_event {
52 	MMU_NOTIFY_UNMAP = 0,
53 	MMU_NOTIFY_CLEAR,
54 	MMU_NOTIFY_PROTECTION_VMA,
55 	MMU_NOTIFY_PROTECTION_PAGE,
56 	MMU_NOTIFY_SOFT_DIRTY,
57 	MMU_NOTIFY_RELEASE,
58 	MMU_NOTIFY_MIGRATE,
59 	MMU_NOTIFY_EXCLUSIVE,
60 };
61 
62 #define MMU_NOTIFIER_RANGE_BLOCKABLE (1 << 0)
63 
64 struct mmu_notifier_ops {
65 	/*
66 	 * Called either by mmu_notifier_unregister or when the mm is
67 	 * being destroyed by exit_mmap, always before all pages are
68 	 * freed. This can run concurrently with other mmu notifier
69 	 * methods (the ones invoked outside the mm context) and it
70 	 * should tear down all secondary mmu mappings and freeze the
71 	 * secondary mmu. If this method isn't implemented you've to
72 	 * be sure that nothing could possibly write to the pages
73 	 * through the secondary mmu by the time the last thread with
74 	 * tsk->mm == mm exits.
75 	 *
76 	 * As side note: the pages freed after ->release returns could
77 	 * be immediately reallocated by the gart at an alias physical
78 	 * address with a different cache model, so if ->release isn't
79 	 * implemented because all _software_ driven memory accesses
80 	 * through the secondary mmu are terminated by the time the
81 	 * last thread of this mm quits, you've also to be sure that
82 	 * speculative _hardware_ operations can't allocate dirty
83 	 * cachelines in the cpu that could not be snooped and made
84 	 * coherent with the other read and write operations happening
85 	 * through the gart alias address, so leading to memory
86 	 * corruption.
87 	 */
88 	void (*release)(struct mmu_notifier *subscription,
89 			struct mm_struct *mm);
90 
91 	/*
92 	 * clear_flush_young is called after the VM is
93 	 * test-and-clearing the young/accessed bitflag in the
94 	 * pte. This way the VM will provide proper aging to the
95 	 * accesses to the page through the secondary MMUs and not
96 	 * only to the ones through the Linux pte.
97 	 * Start-end is necessary in case the secondary MMU is mapping the page
98 	 * at a smaller granularity than the primary MMU.
99 	 */
100 	int (*clear_flush_young)(struct mmu_notifier *subscription,
101 				 struct mm_struct *mm,
102 				 unsigned long start,
103 				 unsigned long end);
104 
105 	/*
106 	 * clear_young is a lightweight version of clear_flush_young. Like the
107 	 * latter, it is supposed to test-and-clear the young/accessed bitflag
108 	 * in the secondary pte, but it may omit flushing the secondary tlb.
109 	 */
110 	int (*clear_young)(struct mmu_notifier *subscription,
111 			   struct mm_struct *mm,
112 			   unsigned long start,
113 			   unsigned long end);
114 
115 	/*
116 	 * test_young is called to check the young/accessed bitflag in
117 	 * the secondary pte. This is used to know if the page is
118 	 * frequently used without actually clearing the flag or tearing
119 	 * down the secondary mapping on the page.
120 	 */
121 	int (*test_young)(struct mmu_notifier *subscription,
122 			  struct mm_struct *mm,
123 			  unsigned long address);
124 
125 	/*
126 	 * change_pte is called in cases that pte mapping to page is changed:
127 	 * for example, when ksm remaps pte to point to a new shared page.
128 	 */
129 	void (*change_pte)(struct mmu_notifier *subscription,
130 			   struct mm_struct *mm,
131 			   unsigned long address,
132 			   pte_t pte);
133 
134 	/*
135 	 * invalidate_range_start() and invalidate_range_end() must be
136 	 * paired and are called only when the mmap_lock and/or the
137 	 * locks protecting the reverse maps are held. If the subsystem
138 	 * can't guarantee that no additional references are taken to
139 	 * the pages in the range, it has to implement the
140 	 * invalidate_range() notifier to remove any references taken
141 	 * after invalidate_range_start().
142 	 *
143 	 * Invalidation of multiple concurrent ranges may be
144 	 * optionally permitted by the driver. Either way the
145 	 * establishment of sptes is forbidden in the range passed to
146 	 * invalidate_range_begin/end for the whole duration of the
147 	 * invalidate_range_begin/end critical section.
148 	 *
149 	 * invalidate_range_start() is called when all pages in the
150 	 * range are still mapped and have at least a refcount of one.
151 	 *
152 	 * invalidate_range_end() is called when all pages in the
153 	 * range have been unmapped and the pages have been freed by
154 	 * the VM.
155 	 *
156 	 * The VM will remove the page table entries and potentially
157 	 * the page between invalidate_range_start() and
158 	 * invalidate_range_end(). If the page must not be freed
159 	 * because of pending I/O or other circumstances then the
160 	 * invalidate_range_start() callback (or the initial mapping
161 	 * by the driver) must make sure that the refcount is kept
162 	 * elevated.
163 	 *
164 	 * If the driver increases the refcount when the pages are
165 	 * initially mapped into an address space then either
166 	 * invalidate_range_start() or invalidate_range_end() may
167 	 * decrease the refcount. If the refcount is decreased on
168 	 * invalidate_range_start() then the VM can free pages as page
169 	 * table entries are removed.  If the refcount is only
170 	 * dropped on invalidate_range_end() then the driver itself
171 	 * will drop the last refcount but it must take care to flush
172 	 * any secondary tlb before doing the final free on the
173 	 * page. Pages will no longer be referenced by the linux
174 	 * address space but may still be referenced by sptes until
175 	 * the last refcount is dropped.
176 	 *
177 	 * If blockable argument is set to false then the callback cannot
178 	 * sleep and has to return with -EAGAIN if sleeping would be required.
179 	 * 0 should be returned otherwise. Please note that notifiers that can
180 	 * fail invalidate_range_start are not allowed to implement
181 	 * invalidate_range_end, as there is no mechanism for informing the
182 	 * notifier that its start failed.
183 	 */
184 	int (*invalidate_range_start)(struct mmu_notifier *subscription,
185 				      const struct mmu_notifier_range *range);
186 	void (*invalidate_range_end)(struct mmu_notifier *subscription,
187 				     const struct mmu_notifier_range *range);
188 
189 	/*
190 	 * invalidate_range() is either called between
191 	 * invalidate_range_start() and invalidate_range_end() when the
192 	 * VM has to free pages that where unmapped, but before the
193 	 * pages are actually freed, or outside of _start()/_end() when
194 	 * a (remote) TLB is necessary.
195 	 *
196 	 * If invalidate_range() is used to manage a non-CPU TLB with
197 	 * shared page-tables, it not necessary to implement the
198 	 * invalidate_range_start()/end() notifiers, as
199 	 * invalidate_range() already catches the points in time when an
200 	 * external TLB range needs to be flushed. For more in depth
201 	 * discussion on this see Documentation/mm/mmu_notifier.rst
202 	 *
203 	 * Note that this function might be called with just a sub-range
204 	 * of what was passed to invalidate_range_start()/end(), if
205 	 * called between those functions.
206 	 */
207 	void (*invalidate_range)(struct mmu_notifier *subscription,
208 				 struct mm_struct *mm,
209 				 unsigned long start,
210 				 unsigned long end);
211 
212 	/*
213 	 * These callbacks are used with the get/put interface to manage the
214 	 * lifetime of the mmu_notifier memory. alloc_notifier() returns a new
215 	 * notifier for use with the mm.
216 	 *
217 	 * free_notifier() is only called after the mmu_notifier has been
218 	 * fully put, calls to any ops callback are prevented and no ops
219 	 * callbacks are currently running. It is called from a SRCU callback
220 	 * and cannot sleep.
221 	 */
222 	struct mmu_notifier *(*alloc_notifier)(struct mm_struct *mm);
223 	void (*free_notifier)(struct mmu_notifier *subscription);
224 };
225 
226 /*
227  * The notifier chains are protected by mmap_lock and/or the reverse map
228  * semaphores. Notifier chains are only changed when all reverse maps and
229  * the mmap_lock locks are taken.
230  *
231  * Therefore notifier chains can only be traversed when either
232  *
233  * 1. mmap_lock is held.
234  * 2. One of the reverse map locks is held (i_mmap_rwsem or anon_vma->rwsem).
235  * 3. No other concurrent thread can access the list (release)
236  */
237 struct mmu_notifier {
238 	struct hlist_node hlist;
239 	const struct mmu_notifier_ops *ops;
240 	struct mm_struct *mm;
241 	struct rcu_head rcu;
242 	unsigned int users;
243 };
244 
245 /**
246  * struct mmu_interval_notifier_ops
247  * @invalidate: Upon return the caller must stop using any SPTEs within this
248  *              range. This function can sleep. Return false only if sleeping
249  *              was required but mmu_notifier_range_blockable(range) is false.
250  */
251 struct mmu_interval_notifier_ops {
252 	bool (*invalidate)(struct mmu_interval_notifier *interval_sub,
253 			   const struct mmu_notifier_range *range,
254 			   unsigned long cur_seq);
255 };
256 
257 struct mmu_interval_notifier {
258 	struct interval_tree_node interval_tree;
259 	const struct mmu_interval_notifier_ops *ops;
260 	struct mm_struct *mm;
261 	struct hlist_node deferred_item;
262 	unsigned long invalidate_seq;
263 };
264 
265 #ifdef CONFIG_MMU_NOTIFIER
266 
267 #ifdef CONFIG_LOCKDEP
268 extern struct lockdep_map __mmu_notifier_invalidate_range_start_map;
269 #endif
270 
271 struct mmu_notifier_range {
272 	struct mm_struct *mm;
273 	unsigned long start;
274 	unsigned long end;
275 	unsigned flags;
276 	enum mmu_notifier_event event;
277 	void *owner;
278 };
279 
mm_has_notifiers(struct mm_struct * mm)280 static inline int mm_has_notifiers(struct mm_struct *mm)
281 {
282 	return unlikely(mm->notifier_subscriptions);
283 }
284 
285 struct mmu_notifier *mmu_notifier_get_locked(const struct mmu_notifier_ops *ops,
286 					     struct mm_struct *mm);
287 static inline struct mmu_notifier *
mmu_notifier_get(const struct mmu_notifier_ops * ops,struct mm_struct * mm)288 mmu_notifier_get(const struct mmu_notifier_ops *ops, struct mm_struct *mm)
289 {
290 	struct mmu_notifier *ret;
291 
292 	mmap_write_lock(mm);
293 	ret = mmu_notifier_get_locked(ops, mm);
294 	mmap_write_unlock(mm);
295 	return ret;
296 }
297 void mmu_notifier_put(struct mmu_notifier *subscription);
298 void mmu_notifier_synchronize(void);
299 
300 extern int mmu_notifier_register(struct mmu_notifier *subscription,
301 				 struct mm_struct *mm);
302 extern int __mmu_notifier_register(struct mmu_notifier *subscription,
303 				   struct mm_struct *mm);
304 extern void mmu_notifier_unregister(struct mmu_notifier *subscription,
305 				    struct mm_struct *mm);
306 
307 unsigned long
308 mmu_interval_read_begin(struct mmu_interval_notifier *interval_sub);
309 int mmu_interval_notifier_insert(struct mmu_interval_notifier *interval_sub,
310 				 struct mm_struct *mm, unsigned long start,
311 				 unsigned long length,
312 				 const struct mmu_interval_notifier_ops *ops);
313 int mmu_interval_notifier_insert_locked(
314 	struct mmu_interval_notifier *interval_sub, struct mm_struct *mm,
315 	unsigned long start, unsigned long length,
316 	const struct mmu_interval_notifier_ops *ops);
317 void mmu_interval_notifier_remove(struct mmu_interval_notifier *interval_sub);
318 
319 /**
320  * mmu_interval_set_seq - Save the invalidation sequence
321  * @interval_sub - The subscription passed to invalidate
322  * @cur_seq - The cur_seq passed to the invalidate() callback
323  *
324  * This must be called unconditionally from the invalidate callback of a
325  * struct mmu_interval_notifier_ops under the same lock that is used to call
326  * mmu_interval_read_retry(). It updates the sequence number for later use by
327  * mmu_interval_read_retry(). The provided cur_seq will always be odd.
328  *
329  * If the caller does not call mmu_interval_read_begin() or
330  * mmu_interval_read_retry() then this call is not required.
331  */
332 static inline void
mmu_interval_set_seq(struct mmu_interval_notifier * interval_sub,unsigned long cur_seq)333 mmu_interval_set_seq(struct mmu_interval_notifier *interval_sub,
334 		     unsigned long cur_seq)
335 {
336 	WRITE_ONCE(interval_sub->invalidate_seq, cur_seq);
337 }
338 
339 /**
340  * mmu_interval_read_retry - End a read side critical section against a VA range
341  * interval_sub: The subscription
342  * seq: The return of the paired mmu_interval_read_begin()
343  *
344  * This MUST be called under a user provided lock that is also held
345  * unconditionally by op->invalidate() when it calls mmu_interval_set_seq().
346  *
347  * Each call should be paired with a single mmu_interval_read_begin() and
348  * should be used to conclude the read side.
349  *
350  * Returns true if an invalidation collided with this critical section, and
351  * the caller should retry.
352  */
353 static inline bool
mmu_interval_read_retry(struct mmu_interval_notifier * interval_sub,unsigned long seq)354 mmu_interval_read_retry(struct mmu_interval_notifier *interval_sub,
355 			unsigned long seq)
356 {
357 	return interval_sub->invalidate_seq != seq;
358 }
359 
360 /**
361  * mmu_interval_check_retry - Test if a collision has occurred
362  * interval_sub: The subscription
363  * seq: The return of the matching mmu_interval_read_begin()
364  *
365  * This can be used in the critical section between mmu_interval_read_begin()
366  * and mmu_interval_read_retry().  A return of true indicates an invalidation
367  * has collided with this critical region and a future
368  * mmu_interval_read_retry() will return true.
369  *
370  * False is not reliable and only suggests a collision may not have
371  * occurred. It can be called many times and does not have to hold the user
372  * provided lock.
373  *
374  * This call can be used as part of loops and other expensive operations to
375  * expedite a retry.
376  */
377 static inline bool
mmu_interval_check_retry(struct mmu_interval_notifier * interval_sub,unsigned long seq)378 mmu_interval_check_retry(struct mmu_interval_notifier *interval_sub,
379 			 unsigned long seq)
380 {
381 	/* Pairs with the WRITE_ONCE in mmu_interval_set_seq() */
382 	return READ_ONCE(interval_sub->invalidate_seq) != seq;
383 }
384 
385 extern void __mmu_notifier_subscriptions_destroy(struct mm_struct *mm);
386 extern void __mmu_notifier_release(struct mm_struct *mm);
387 extern int __mmu_notifier_clear_flush_young(struct mm_struct *mm,
388 					  unsigned long start,
389 					  unsigned long end);
390 extern int __mmu_notifier_clear_young(struct mm_struct *mm,
391 				      unsigned long start,
392 				      unsigned long end);
393 extern int __mmu_notifier_test_young(struct mm_struct *mm,
394 				     unsigned long address);
395 extern void __mmu_notifier_change_pte(struct mm_struct *mm,
396 				      unsigned long address, pte_t pte);
397 extern int __mmu_notifier_invalidate_range_start(struct mmu_notifier_range *r);
398 extern void __mmu_notifier_invalidate_range_end(struct mmu_notifier_range *r,
399 				  bool only_end);
400 extern void __mmu_notifier_invalidate_range(struct mm_struct *mm,
401 				  unsigned long start, unsigned long end);
402 extern bool
403 mmu_notifier_range_update_to_read_only(const struct mmu_notifier_range *range);
404 
405 static inline bool
mmu_notifier_range_blockable(const struct mmu_notifier_range * range)406 mmu_notifier_range_blockable(const struct mmu_notifier_range *range)
407 {
408 	return (range->flags & MMU_NOTIFIER_RANGE_BLOCKABLE);
409 }
410 
mmu_notifier_release(struct mm_struct * mm)411 static inline void mmu_notifier_release(struct mm_struct *mm)
412 {
413 	if (mm_has_notifiers(mm))
414 		__mmu_notifier_release(mm);
415 }
416 
mmu_notifier_clear_flush_young(struct mm_struct * mm,unsigned long start,unsigned long end)417 static inline int mmu_notifier_clear_flush_young(struct mm_struct *mm,
418 					  unsigned long start,
419 					  unsigned long end)
420 {
421 	if (mm_has_notifiers(mm))
422 		return __mmu_notifier_clear_flush_young(mm, start, end);
423 	return 0;
424 }
425 
mmu_notifier_clear_young(struct mm_struct * mm,unsigned long start,unsigned long end)426 static inline int mmu_notifier_clear_young(struct mm_struct *mm,
427 					   unsigned long start,
428 					   unsigned long end)
429 {
430 	if (mm_has_notifiers(mm))
431 		return __mmu_notifier_clear_young(mm, start, end);
432 	return 0;
433 }
434 
mmu_notifier_test_young(struct mm_struct * mm,unsigned long address)435 static inline int mmu_notifier_test_young(struct mm_struct *mm,
436 					  unsigned long address)
437 {
438 	if (mm_has_notifiers(mm))
439 		return __mmu_notifier_test_young(mm, address);
440 	return 0;
441 }
442 
mmu_notifier_change_pte(struct mm_struct * mm,unsigned long address,pte_t pte)443 static inline void mmu_notifier_change_pte(struct mm_struct *mm,
444 					   unsigned long address, pte_t pte)
445 {
446 	if (mm_has_notifiers(mm))
447 		__mmu_notifier_change_pte(mm, address, pte);
448 }
449 
450 static inline void
mmu_notifier_invalidate_range_start(struct mmu_notifier_range * range)451 mmu_notifier_invalidate_range_start(struct mmu_notifier_range *range)
452 {
453 	might_sleep();
454 
455 	lock_map_acquire(&__mmu_notifier_invalidate_range_start_map);
456 	if (mm_has_notifiers(range->mm)) {
457 		range->flags |= MMU_NOTIFIER_RANGE_BLOCKABLE;
458 		__mmu_notifier_invalidate_range_start(range);
459 	}
460 	lock_map_release(&__mmu_notifier_invalidate_range_start_map);
461 }
462 
463 static inline int
mmu_notifier_invalidate_range_start_nonblock(struct mmu_notifier_range * range)464 mmu_notifier_invalidate_range_start_nonblock(struct mmu_notifier_range *range)
465 {
466 	int ret = 0;
467 
468 	lock_map_acquire(&__mmu_notifier_invalidate_range_start_map);
469 	if (mm_has_notifiers(range->mm)) {
470 		range->flags &= ~MMU_NOTIFIER_RANGE_BLOCKABLE;
471 		ret = __mmu_notifier_invalidate_range_start(range);
472 	}
473 	lock_map_release(&__mmu_notifier_invalidate_range_start_map);
474 	return ret;
475 }
476 
477 static inline void
mmu_notifier_invalidate_range_end(struct mmu_notifier_range * range)478 mmu_notifier_invalidate_range_end(struct mmu_notifier_range *range)
479 {
480 	if (mmu_notifier_range_blockable(range))
481 		might_sleep();
482 
483 	if (mm_has_notifiers(range->mm))
484 		__mmu_notifier_invalidate_range_end(range, false);
485 }
486 
487 static inline void
mmu_notifier_invalidate_range_only_end(struct mmu_notifier_range * range)488 mmu_notifier_invalidate_range_only_end(struct mmu_notifier_range *range)
489 {
490 	if (mm_has_notifiers(range->mm))
491 		__mmu_notifier_invalidate_range_end(range, true);
492 }
493 
mmu_notifier_invalidate_range(struct mm_struct * mm,unsigned long start,unsigned long end)494 static inline void mmu_notifier_invalidate_range(struct mm_struct *mm,
495 				  unsigned long start, unsigned long end)
496 {
497 	if (mm_has_notifiers(mm))
498 		__mmu_notifier_invalidate_range(mm, start, end);
499 }
500 
mmu_notifier_subscriptions_init(struct mm_struct * mm)501 static inline void mmu_notifier_subscriptions_init(struct mm_struct *mm)
502 {
503 	mm->notifier_subscriptions = NULL;
504 }
505 
mmu_notifier_subscriptions_destroy(struct mm_struct * mm)506 static inline void mmu_notifier_subscriptions_destroy(struct mm_struct *mm)
507 {
508 	if (mm_has_notifiers(mm))
509 		__mmu_notifier_subscriptions_destroy(mm);
510 }
511 
512 
mmu_notifier_range_init(struct mmu_notifier_range * range,enum mmu_notifier_event event,unsigned flags,struct mm_struct * mm,unsigned long start,unsigned long end)513 static inline void mmu_notifier_range_init(struct mmu_notifier_range *range,
514 					   enum mmu_notifier_event event,
515 					   unsigned flags,
516 					   struct mm_struct *mm,
517 					   unsigned long start,
518 					   unsigned long end)
519 {
520 	range->event = event;
521 	range->mm = mm;
522 	range->start = start;
523 	range->end = end;
524 	range->flags = flags;
525 }
526 
mmu_notifier_range_init_owner(struct mmu_notifier_range * range,enum mmu_notifier_event event,unsigned int flags,struct mm_struct * mm,unsigned long start,unsigned long end,void * owner)527 static inline void mmu_notifier_range_init_owner(
528 			struct mmu_notifier_range *range,
529 			enum mmu_notifier_event event, unsigned int flags,
530 			struct mm_struct *mm, unsigned long start,
531 			unsigned long end, void *owner)
532 {
533 	mmu_notifier_range_init(range, event, flags, mm, start, end);
534 	range->owner = owner;
535 }
536 
537 #define ptep_clear_flush_young_notify(__vma, __address, __ptep)		\
538 ({									\
539 	int __young;							\
540 	struct vm_area_struct *___vma = __vma;				\
541 	unsigned long ___address = __address;				\
542 	__young = ptep_clear_flush_young(___vma, ___address, __ptep);	\
543 	__young |= mmu_notifier_clear_flush_young(___vma->vm_mm,	\
544 						  ___address,		\
545 						  ___address +		\
546 							PAGE_SIZE);	\
547 	__young;							\
548 })
549 
550 #define pmdp_clear_flush_young_notify(__vma, __address, __pmdp)		\
551 ({									\
552 	int __young;							\
553 	struct vm_area_struct *___vma = __vma;				\
554 	unsigned long ___address = __address;				\
555 	__young = pmdp_clear_flush_young(___vma, ___address, __pmdp);	\
556 	__young |= mmu_notifier_clear_flush_young(___vma->vm_mm,	\
557 						  ___address,		\
558 						  ___address +		\
559 							PMD_SIZE);	\
560 	__young;							\
561 })
562 
563 #define ptep_clear_young_notify(__vma, __address, __ptep)		\
564 ({									\
565 	int __young;							\
566 	struct vm_area_struct *___vma = __vma;				\
567 	unsigned long ___address = __address;				\
568 	__young = ptep_test_and_clear_young(___vma, ___address, __ptep);\
569 	__young |= mmu_notifier_clear_young(___vma->vm_mm, ___address,	\
570 					    ___address + PAGE_SIZE);	\
571 	__young;							\
572 })
573 
574 #define pmdp_clear_young_notify(__vma, __address, __pmdp)		\
575 ({									\
576 	int __young;							\
577 	struct vm_area_struct *___vma = __vma;				\
578 	unsigned long ___address = __address;				\
579 	__young = pmdp_test_and_clear_young(___vma, ___address, __pmdp);\
580 	__young |= mmu_notifier_clear_young(___vma->vm_mm, ___address,	\
581 					    ___address + PMD_SIZE);	\
582 	__young;							\
583 })
584 
585 #define	ptep_clear_flush_notify(__vma, __address, __ptep)		\
586 ({									\
587 	unsigned long ___addr = __address & PAGE_MASK;			\
588 	struct mm_struct *___mm = (__vma)->vm_mm;			\
589 	pte_t ___pte;							\
590 									\
591 	___pte = ptep_clear_flush(__vma, __address, __ptep);		\
592 	mmu_notifier_invalidate_range(___mm, ___addr,			\
593 					___addr + PAGE_SIZE);		\
594 									\
595 	___pte;								\
596 })
597 
598 #define pmdp_huge_clear_flush_notify(__vma, __haddr, __pmd)		\
599 ({									\
600 	unsigned long ___haddr = __haddr & HPAGE_PMD_MASK;		\
601 	struct mm_struct *___mm = (__vma)->vm_mm;			\
602 	pmd_t ___pmd;							\
603 									\
604 	___pmd = pmdp_huge_clear_flush(__vma, __haddr, __pmd);		\
605 	mmu_notifier_invalidate_range(___mm, ___haddr,			\
606 				      ___haddr + HPAGE_PMD_SIZE);	\
607 									\
608 	___pmd;								\
609 })
610 
611 #define pudp_huge_clear_flush_notify(__vma, __haddr, __pud)		\
612 ({									\
613 	unsigned long ___haddr = __haddr & HPAGE_PUD_MASK;		\
614 	struct mm_struct *___mm = (__vma)->vm_mm;			\
615 	pud_t ___pud;							\
616 									\
617 	___pud = pudp_huge_clear_flush(__vma, __haddr, __pud);		\
618 	mmu_notifier_invalidate_range(___mm, ___haddr,			\
619 				      ___haddr + HPAGE_PUD_SIZE);	\
620 									\
621 	___pud;								\
622 })
623 
624 /*
625  * set_pte_at_notify() sets the pte _after_ running the notifier.
626  * This is safe to start by updating the secondary MMUs, because the primary MMU
627  * pte invalidate must have already happened with a ptep_clear_flush() before
628  * set_pte_at_notify() has been invoked.  Updating the secondary MMUs first is
629  * required when we change both the protection of the mapping from read-only to
630  * read-write and the pfn (like during copy on write page faults). Otherwise the
631  * old page would remain mapped readonly in the secondary MMUs after the new
632  * page is already writable by some CPU through the primary MMU.
633  */
634 #define set_pte_at_notify(__mm, __address, __ptep, __pte)		\
635 ({									\
636 	struct mm_struct *___mm = __mm;					\
637 	unsigned long ___address = __address;				\
638 	pte_t ___pte = __pte;						\
639 									\
640 	mmu_notifier_change_pte(___mm, ___address, ___pte);		\
641 	set_pte_at(___mm, ___address, __ptep, ___pte);			\
642 })
643 
644 #else /* CONFIG_MMU_NOTIFIER */
645 
646 struct mmu_notifier_range {
647 	unsigned long start;
648 	unsigned long end;
649 };
650 
_mmu_notifier_range_init(struct mmu_notifier_range * range,unsigned long start,unsigned long end)651 static inline void _mmu_notifier_range_init(struct mmu_notifier_range *range,
652 					    unsigned long start,
653 					    unsigned long end)
654 {
655 	range->start = start;
656 	range->end = end;
657 }
658 
659 #define mmu_notifier_range_init(range,event,flags,mm,start,end)  \
660 	_mmu_notifier_range_init(range, start, end)
661 #define mmu_notifier_range_init_owner(range, event, flags, mm, start, \
662 					end, owner) \
663 	_mmu_notifier_range_init(range, start, end)
664 
665 static inline bool
mmu_notifier_range_blockable(const struct mmu_notifier_range * range)666 mmu_notifier_range_blockable(const struct mmu_notifier_range *range)
667 {
668 	return true;
669 }
670 
mm_has_notifiers(struct mm_struct * mm)671 static inline int mm_has_notifiers(struct mm_struct *mm)
672 {
673 	return 0;
674 }
675 
mmu_notifier_release(struct mm_struct * mm)676 static inline void mmu_notifier_release(struct mm_struct *mm)
677 {
678 }
679 
mmu_notifier_clear_flush_young(struct mm_struct * mm,unsigned long start,unsigned long end)680 static inline int mmu_notifier_clear_flush_young(struct mm_struct *mm,
681 					  unsigned long start,
682 					  unsigned long end)
683 {
684 	return 0;
685 }
686 
mmu_notifier_test_young(struct mm_struct * mm,unsigned long address)687 static inline int mmu_notifier_test_young(struct mm_struct *mm,
688 					  unsigned long address)
689 {
690 	return 0;
691 }
692 
mmu_notifier_change_pte(struct mm_struct * mm,unsigned long address,pte_t pte)693 static inline void mmu_notifier_change_pte(struct mm_struct *mm,
694 					   unsigned long address, pte_t pte)
695 {
696 }
697 
698 static inline void
mmu_notifier_invalidate_range_start(struct mmu_notifier_range * range)699 mmu_notifier_invalidate_range_start(struct mmu_notifier_range *range)
700 {
701 }
702 
703 static inline int
mmu_notifier_invalidate_range_start_nonblock(struct mmu_notifier_range * range)704 mmu_notifier_invalidate_range_start_nonblock(struct mmu_notifier_range *range)
705 {
706 	return 0;
707 }
708 
709 static inline
mmu_notifier_invalidate_range_end(struct mmu_notifier_range * range)710 void mmu_notifier_invalidate_range_end(struct mmu_notifier_range *range)
711 {
712 }
713 
714 static inline void
mmu_notifier_invalidate_range_only_end(struct mmu_notifier_range * range)715 mmu_notifier_invalidate_range_only_end(struct mmu_notifier_range *range)
716 {
717 }
718 
mmu_notifier_invalidate_range(struct mm_struct * mm,unsigned long start,unsigned long end)719 static inline void mmu_notifier_invalidate_range(struct mm_struct *mm,
720 				  unsigned long start, unsigned long end)
721 {
722 }
723 
mmu_notifier_subscriptions_init(struct mm_struct * mm)724 static inline void mmu_notifier_subscriptions_init(struct mm_struct *mm)
725 {
726 }
727 
mmu_notifier_subscriptions_destroy(struct mm_struct * mm)728 static inline void mmu_notifier_subscriptions_destroy(struct mm_struct *mm)
729 {
730 }
731 
732 #define mmu_notifier_range_update_to_read_only(r) false
733 
734 #define ptep_clear_flush_young_notify ptep_clear_flush_young
735 #define pmdp_clear_flush_young_notify pmdp_clear_flush_young
736 #define ptep_clear_young_notify ptep_test_and_clear_young
737 #define pmdp_clear_young_notify pmdp_test_and_clear_young
738 #define	ptep_clear_flush_notify ptep_clear_flush
739 #define pmdp_huge_clear_flush_notify pmdp_huge_clear_flush
740 #define pudp_huge_clear_flush_notify pudp_huge_clear_flush
741 #define set_pte_at_notify set_pte_at
742 
mmu_notifier_synchronize(void)743 static inline void mmu_notifier_synchronize(void)
744 {
745 }
746 
747 #endif /* CONFIG_MMU_NOTIFIER */
748 
749 #endif /* _LINUX_MMU_NOTIFIER_H */
750