Loading...
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
12struct mmu_notifier_subscriptions;
13struct mmu_notifier;
14struct mmu_notifier_range;
15struct 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 */
51enum 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
64struct 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/vm/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 */
237struct 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 */
251struct 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
257struct 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
268extern struct lockdep_map __mmu_notifier_invalidate_range_start_map;
269#endif
270
271struct mmu_notifier_range {
272 struct vm_area_struct *vma;
273 struct mm_struct *mm;
274 unsigned long start;
275 unsigned long end;
276 unsigned flags;
277 enum mmu_notifier_event event;
278 void *owner;
279};
280
281static inline int mm_has_notifiers(struct mm_struct *mm)
282{
283 return unlikely(mm->notifier_subscriptions);
284}
285
286struct mmu_notifier *mmu_notifier_get_locked(const struct mmu_notifier_ops *ops,
287 struct mm_struct *mm);
288static inline struct mmu_notifier *
289mmu_notifier_get(const struct mmu_notifier_ops *ops, struct mm_struct *mm)
290{
291 struct mmu_notifier *ret;
292
293 mmap_write_lock(mm);
294 ret = mmu_notifier_get_locked(ops, mm);
295 mmap_write_unlock(mm);
296 return ret;
297}
298void mmu_notifier_put(struct mmu_notifier *subscription);
299void mmu_notifier_synchronize(void);
300
301extern int mmu_notifier_register(struct mmu_notifier *subscription,
302 struct mm_struct *mm);
303extern int __mmu_notifier_register(struct mmu_notifier *subscription,
304 struct mm_struct *mm);
305extern void mmu_notifier_unregister(struct mmu_notifier *subscription,
306 struct mm_struct *mm);
307
308unsigned long
309mmu_interval_read_begin(struct mmu_interval_notifier *interval_sub);
310int mmu_interval_notifier_insert(struct mmu_interval_notifier *interval_sub,
311 struct mm_struct *mm, unsigned long start,
312 unsigned long length,
313 const struct mmu_interval_notifier_ops *ops);
314int mmu_interval_notifier_insert_locked(
315 struct mmu_interval_notifier *interval_sub, struct mm_struct *mm,
316 unsigned long start, unsigned long length,
317 const struct mmu_interval_notifier_ops *ops);
318void mmu_interval_notifier_remove(struct mmu_interval_notifier *interval_sub);
319
320/**
321 * mmu_interval_set_seq - Save the invalidation sequence
322 * @interval_sub - The subscription passed to invalidate
323 * @cur_seq - The cur_seq passed to the invalidate() callback
324 *
325 * This must be called unconditionally from the invalidate callback of a
326 * struct mmu_interval_notifier_ops under the same lock that is used to call
327 * mmu_interval_read_retry(). It updates the sequence number for later use by
328 * mmu_interval_read_retry(). The provided cur_seq will always be odd.
329 *
330 * If the caller does not call mmu_interval_read_begin() or
331 * mmu_interval_read_retry() then this call is not required.
332 */
333static inline void
334mmu_interval_set_seq(struct mmu_interval_notifier *interval_sub,
335 unsigned long cur_seq)
336{
337 WRITE_ONCE(interval_sub->invalidate_seq, cur_seq);
338}
339
340/**
341 * mmu_interval_read_retry - End a read side critical section against a VA range
342 * interval_sub: The subscription
343 * seq: The return of the paired mmu_interval_read_begin()
344 *
345 * This MUST be called under a user provided lock that is also held
346 * unconditionally by op->invalidate() when it calls mmu_interval_set_seq().
347 *
348 * Each call should be paired with a single mmu_interval_read_begin() and
349 * should be used to conclude the read side.
350 *
351 * Returns true if an invalidation collided with this critical section, and
352 * the caller should retry.
353 */
354static inline bool
355mmu_interval_read_retry(struct mmu_interval_notifier *interval_sub,
356 unsigned long seq)
357{
358 return interval_sub->invalidate_seq != seq;
359}
360
361/**
362 * mmu_interval_check_retry - Test if a collision has occurred
363 * interval_sub: The subscription
364 * seq: The return of the matching mmu_interval_read_begin()
365 *
366 * This can be used in the critical section between mmu_interval_read_begin()
367 * and mmu_interval_read_retry(). A return of true indicates an invalidation
368 * has collided with this critical region and a future
369 * mmu_interval_read_retry() will return true.
370 *
371 * False is not reliable and only suggests a collision may not have
372 * occurred. It can be called many times and does not have to hold the user
373 * provided lock.
374 *
375 * This call can be used as part of loops and other expensive operations to
376 * expedite a retry.
377 */
378static inline bool
379mmu_interval_check_retry(struct mmu_interval_notifier *interval_sub,
380 unsigned long seq)
381{
382 /* Pairs with the WRITE_ONCE in mmu_interval_set_seq() */
383 return READ_ONCE(interval_sub->invalidate_seq) != seq;
384}
385
386extern void __mmu_notifier_subscriptions_destroy(struct mm_struct *mm);
387extern void __mmu_notifier_release(struct mm_struct *mm);
388extern int __mmu_notifier_clear_flush_young(struct mm_struct *mm,
389 unsigned long start,
390 unsigned long end);
391extern int __mmu_notifier_clear_young(struct mm_struct *mm,
392 unsigned long start,
393 unsigned long end);
394extern int __mmu_notifier_test_young(struct mm_struct *mm,
395 unsigned long address);
396extern void __mmu_notifier_change_pte(struct mm_struct *mm,
397 unsigned long address, pte_t pte);
398extern int __mmu_notifier_invalidate_range_start(struct mmu_notifier_range *r);
399extern void __mmu_notifier_invalidate_range_end(struct mmu_notifier_range *r,
400 bool only_end);
401extern void __mmu_notifier_invalidate_range(struct mm_struct *mm,
402 unsigned long start, unsigned long end);
403extern bool
404mmu_notifier_range_update_to_read_only(const struct mmu_notifier_range *range);
405
406static inline bool
407mmu_notifier_range_blockable(const struct mmu_notifier_range *range)
408{
409 return (range->flags & MMU_NOTIFIER_RANGE_BLOCKABLE);
410}
411
412static inline void mmu_notifier_release(struct mm_struct *mm)
413{
414 if (mm_has_notifiers(mm))
415 __mmu_notifier_release(mm);
416}
417
418static inline int mmu_notifier_clear_flush_young(struct mm_struct *mm,
419 unsigned long start,
420 unsigned long end)
421{
422 if (mm_has_notifiers(mm))
423 return __mmu_notifier_clear_flush_young(mm, start, end);
424 return 0;
425}
426
427static inline int mmu_notifier_clear_young(struct mm_struct *mm,
428 unsigned long start,
429 unsigned long end)
430{
431 if (mm_has_notifiers(mm))
432 return __mmu_notifier_clear_young(mm, start, end);
433 return 0;
434}
435
436static inline int mmu_notifier_test_young(struct mm_struct *mm,
437 unsigned long address)
438{
439 if (mm_has_notifiers(mm))
440 return __mmu_notifier_test_young(mm, address);
441 return 0;
442}
443
444static inline void mmu_notifier_change_pte(struct mm_struct *mm,
445 unsigned long address, pte_t pte)
446{
447 if (mm_has_notifiers(mm))
448 __mmu_notifier_change_pte(mm, address, pte);
449}
450
451static inline void
452mmu_notifier_invalidate_range_start(struct mmu_notifier_range *range)
453{
454 might_sleep();
455
456 lock_map_acquire(&__mmu_notifier_invalidate_range_start_map);
457 if (mm_has_notifiers(range->mm)) {
458 range->flags |= MMU_NOTIFIER_RANGE_BLOCKABLE;
459 __mmu_notifier_invalidate_range_start(range);
460 }
461 lock_map_release(&__mmu_notifier_invalidate_range_start_map);
462}
463
464static inline int
465mmu_notifier_invalidate_range_start_nonblock(struct mmu_notifier_range *range)
466{
467 int ret = 0;
468
469 lock_map_acquire(&__mmu_notifier_invalidate_range_start_map);
470 if (mm_has_notifiers(range->mm)) {
471 range->flags &= ~MMU_NOTIFIER_RANGE_BLOCKABLE;
472 ret = __mmu_notifier_invalidate_range_start(range);
473 }
474 lock_map_release(&__mmu_notifier_invalidate_range_start_map);
475 return ret;
476}
477
478static inline void
479mmu_notifier_invalidate_range_end(struct mmu_notifier_range *range)
480{
481 if (mmu_notifier_range_blockable(range))
482 might_sleep();
483
484 if (mm_has_notifiers(range->mm))
485 __mmu_notifier_invalidate_range_end(range, false);
486}
487
488static inline void
489mmu_notifier_invalidate_range_only_end(struct mmu_notifier_range *range)
490{
491 if (mm_has_notifiers(range->mm))
492 __mmu_notifier_invalidate_range_end(range, true);
493}
494
495static inline void mmu_notifier_invalidate_range(struct mm_struct *mm,
496 unsigned long start, unsigned long end)
497{
498 if (mm_has_notifiers(mm))
499 __mmu_notifier_invalidate_range(mm, start, end);
500}
501
502static inline void mmu_notifier_subscriptions_init(struct mm_struct *mm)
503{
504 mm->notifier_subscriptions = NULL;
505}
506
507static inline void mmu_notifier_subscriptions_destroy(struct mm_struct *mm)
508{
509 if (mm_has_notifiers(mm))
510 __mmu_notifier_subscriptions_destroy(mm);
511}
512
513
514static inline void mmu_notifier_range_init(struct mmu_notifier_range *range,
515 enum mmu_notifier_event event,
516 unsigned flags,
517 struct vm_area_struct *vma,
518 struct mm_struct *mm,
519 unsigned long start,
520 unsigned long end)
521{
522 range->vma = vma;
523 range->event = event;
524 range->mm = mm;
525 range->start = start;
526 range->end = end;
527 range->flags = flags;
528}
529
530static inline void mmu_notifier_range_init_owner(
531 struct mmu_notifier_range *range,
532 enum mmu_notifier_event event, unsigned int flags,
533 struct vm_area_struct *vma, struct mm_struct *mm,
534 unsigned long start, unsigned long end, void *owner)
535{
536 mmu_notifier_range_init(range, event, flags, vma, mm, start, end);
537 range->owner = owner;
538}
539
540#define ptep_clear_flush_young_notify(__vma, __address, __ptep) \
541({ \
542 int __young; \
543 struct vm_area_struct *___vma = __vma; \
544 unsigned long ___address = __address; \
545 __young = ptep_clear_flush_young(___vma, ___address, __ptep); \
546 __young |= mmu_notifier_clear_flush_young(___vma->vm_mm, \
547 ___address, \
548 ___address + \
549 PAGE_SIZE); \
550 __young; \
551})
552
553#define pmdp_clear_flush_young_notify(__vma, __address, __pmdp) \
554({ \
555 int __young; \
556 struct vm_area_struct *___vma = __vma; \
557 unsigned long ___address = __address; \
558 __young = pmdp_clear_flush_young(___vma, ___address, __pmdp); \
559 __young |= mmu_notifier_clear_flush_young(___vma->vm_mm, \
560 ___address, \
561 ___address + \
562 PMD_SIZE); \
563 __young; \
564})
565
566#define ptep_clear_young_notify(__vma, __address, __ptep) \
567({ \
568 int __young; \
569 struct vm_area_struct *___vma = __vma; \
570 unsigned long ___address = __address; \
571 __young = ptep_test_and_clear_young(___vma, ___address, __ptep);\
572 __young |= mmu_notifier_clear_young(___vma->vm_mm, ___address, \
573 ___address + PAGE_SIZE); \
574 __young; \
575})
576
577#define pmdp_clear_young_notify(__vma, __address, __pmdp) \
578({ \
579 int __young; \
580 struct vm_area_struct *___vma = __vma; \
581 unsigned long ___address = __address; \
582 __young = pmdp_test_and_clear_young(___vma, ___address, __pmdp);\
583 __young |= mmu_notifier_clear_young(___vma->vm_mm, ___address, \
584 ___address + PMD_SIZE); \
585 __young; \
586})
587
588#define ptep_clear_flush_notify(__vma, __address, __ptep) \
589({ \
590 unsigned long ___addr = __address & PAGE_MASK; \
591 struct mm_struct *___mm = (__vma)->vm_mm; \
592 pte_t ___pte; \
593 \
594 ___pte = ptep_clear_flush(__vma, __address, __ptep); \
595 mmu_notifier_invalidate_range(___mm, ___addr, \
596 ___addr + PAGE_SIZE); \
597 \
598 ___pte; \
599})
600
601#define pmdp_huge_clear_flush_notify(__vma, __haddr, __pmd) \
602({ \
603 unsigned long ___haddr = __haddr & HPAGE_PMD_MASK; \
604 struct mm_struct *___mm = (__vma)->vm_mm; \
605 pmd_t ___pmd; \
606 \
607 ___pmd = pmdp_huge_clear_flush(__vma, __haddr, __pmd); \
608 mmu_notifier_invalidate_range(___mm, ___haddr, \
609 ___haddr + HPAGE_PMD_SIZE); \
610 \
611 ___pmd; \
612})
613
614#define pudp_huge_clear_flush_notify(__vma, __haddr, __pud) \
615({ \
616 unsigned long ___haddr = __haddr & HPAGE_PUD_MASK; \
617 struct mm_struct *___mm = (__vma)->vm_mm; \
618 pud_t ___pud; \
619 \
620 ___pud = pudp_huge_clear_flush(__vma, __haddr, __pud); \
621 mmu_notifier_invalidate_range(___mm, ___haddr, \
622 ___haddr + HPAGE_PUD_SIZE); \
623 \
624 ___pud; \
625})
626
627/*
628 * set_pte_at_notify() sets the pte _after_ running the notifier.
629 * This is safe to start by updating the secondary MMUs, because the primary MMU
630 * pte invalidate must have already happened with a ptep_clear_flush() before
631 * set_pte_at_notify() has been invoked. Updating the secondary MMUs first is
632 * required when we change both the protection of the mapping from read-only to
633 * read-write and the pfn (like during copy on write page faults). Otherwise the
634 * old page would remain mapped readonly in the secondary MMUs after the new
635 * page is already writable by some CPU through the primary MMU.
636 */
637#define set_pte_at_notify(__mm, __address, __ptep, __pte) \
638({ \
639 struct mm_struct *___mm = __mm; \
640 unsigned long ___address = __address; \
641 pte_t ___pte = __pte; \
642 \
643 mmu_notifier_change_pte(___mm, ___address, ___pte); \
644 set_pte_at(___mm, ___address, __ptep, ___pte); \
645})
646
647#else /* CONFIG_MMU_NOTIFIER */
648
649struct mmu_notifier_range {
650 unsigned long start;
651 unsigned long end;
652};
653
654static inline void _mmu_notifier_range_init(struct mmu_notifier_range *range,
655 unsigned long start,
656 unsigned long end)
657{
658 range->start = start;
659 range->end = end;
660}
661
662#define mmu_notifier_range_init(range,event,flags,vma,mm,start,end) \
663 _mmu_notifier_range_init(range, start, end)
664#define mmu_notifier_range_init_owner(range, event, flags, vma, mm, start, \
665 end, owner) \
666 _mmu_notifier_range_init(range, start, end)
667
668static inline bool
669mmu_notifier_range_blockable(const struct mmu_notifier_range *range)
670{
671 return true;
672}
673
674static inline int mm_has_notifiers(struct mm_struct *mm)
675{
676 return 0;
677}
678
679static inline void mmu_notifier_release(struct mm_struct *mm)
680{
681}
682
683static inline int mmu_notifier_clear_flush_young(struct mm_struct *mm,
684 unsigned long start,
685 unsigned long end)
686{
687 return 0;
688}
689
690static inline int mmu_notifier_test_young(struct mm_struct *mm,
691 unsigned long address)
692{
693 return 0;
694}
695
696static inline void mmu_notifier_change_pte(struct mm_struct *mm,
697 unsigned long address, pte_t pte)
698{
699}
700
701static inline void
702mmu_notifier_invalidate_range_start(struct mmu_notifier_range *range)
703{
704}
705
706static inline int
707mmu_notifier_invalidate_range_start_nonblock(struct mmu_notifier_range *range)
708{
709 return 0;
710}
711
712static inline
713void mmu_notifier_invalidate_range_end(struct mmu_notifier_range *range)
714{
715}
716
717static inline void
718mmu_notifier_invalidate_range_only_end(struct mmu_notifier_range *range)
719{
720}
721
722static inline void mmu_notifier_invalidate_range(struct mm_struct *mm,
723 unsigned long start, unsigned long end)
724{
725}
726
727static inline void mmu_notifier_subscriptions_init(struct mm_struct *mm)
728{
729}
730
731static inline void mmu_notifier_subscriptions_destroy(struct mm_struct *mm)
732{
733}
734
735#define mmu_notifier_range_update_to_read_only(r) false
736
737#define ptep_clear_flush_young_notify ptep_clear_flush_young
738#define pmdp_clear_flush_young_notify pmdp_clear_flush_young
739#define ptep_clear_young_notify ptep_test_and_clear_young
740#define pmdp_clear_young_notify pmdp_test_and_clear_young
741#define ptep_clear_flush_notify ptep_clear_flush
742#define pmdp_huge_clear_flush_notify pmdp_huge_clear_flush
743#define pudp_huge_clear_flush_notify pudp_huge_clear_flush
744#define set_pte_at_notify set_pte_at
745
746static inline void mmu_notifier_synchronize(void)
747{
748}
749
750#endif /* CONFIG_MMU_NOTIFIER */
751
752#endif /* _LINUX_MMU_NOTIFIER_H */
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
12struct mmu_notifier_subscriptions;
13struct mmu_notifier;
14struct mmu_notifier_range;
15struct 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 */
51enum 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
64struct 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 * arch_invalidate_secondary_tlbs() is used to manage a non-CPU TLB
191 * which shares page-tables with the CPU. The
192 * invalidate_range_start()/end() callbacks should not be implemented as
193 * invalidate_secondary_tlbs() already catches the points in time when
194 * an external TLB needs to be flushed.
195 *
196 * This requires arch_invalidate_secondary_tlbs() to be called while
197 * holding the ptl spin-lock and therefore this callback is not allowed
198 * to sleep.
199 *
200 * This is called by architecture code whenever invalidating a TLB
201 * entry. It is assumed that any secondary TLB has the same rules for
202 * when invalidations are required. If this is not the case architecture
203 * code will need to call this explicitly when required for secondary
204 * TLB invalidation.
205 */
206 void (*arch_invalidate_secondary_tlbs)(
207 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 */
237struct 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 */
251struct 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
257struct 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
268extern struct lockdep_map __mmu_notifier_invalidate_range_start_map;
269#endif
270
271struct 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
280static inline int mm_has_notifiers(struct mm_struct *mm)
281{
282 return unlikely(mm->notifier_subscriptions);
283}
284
285struct mmu_notifier *mmu_notifier_get_locked(const struct mmu_notifier_ops *ops,
286 struct mm_struct *mm);
287static inline struct mmu_notifier *
288mmu_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}
297void mmu_notifier_put(struct mmu_notifier *subscription);
298void mmu_notifier_synchronize(void);
299
300extern int mmu_notifier_register(struct mmu_notifier *subscription,
301 struct mm_struct *mm);
302extern int __mmu_notifier_register(struct mmu_notifier *subscription,
303 struct mm_struct *mm);
304extern void mmu_notifier_unregister(struct mmu_notifier *subscription,
305 struct mm_struct *mm);
306
307unsigned long
308mmu_interval_read_begin(struct mmu_interval_notifier *interval_sub);
309int 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);
313int 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);
317void 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 */
332static inline void
333mmu_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 */
353static inline bool
354mmu_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 */
377static inline bool
378mmu_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
385extern void __mmu_notifier_subscriptions_destroy(struct mm_struct *mm);
386extern void __mmu_notifier_release(struct mm_struct *mm);
387extern int __mmu_notifier_clear_flush_young(struct mm_struct *mm,
388 unsigned long start,
389 unsigned long end);
390extern int __mmu_notifier_clear_young(struct mm_struct *mm,
391 unsigned long start,
392 unsigned long end);
393extern int __mmu_notifier_test_young(struct mm_struct *mm,
394 unsigned long address);
395extern void __mmu_notifier_change_pte(struct mm_struct *mm,
396 unsigned long address, pte_t pte);
397extern int __mmu_notifier_invalidate_range_start(struct mmu_notifier_range *r);
398extern void __mmu_notifier_invalidate_range_end(struct mmu_notifier_range *r);
399extern void __mmu_notifier_arch_invalidate_secondary_tlbs(struct mm_struct *mm,
400 unsigned long start, unsigned long end);
401extern bool
402mmu_notifier_range_update_to_read_only(const struct mmu_notifier_range *range);
403
404static inline bool
405mmu_notifier_range_blockable(const struct mmu_notifier_range *range)
406{
407 return (range->flags & MMU_NOTIFIER_RANGE_BLOCKABLE);
408}
409
410static inline void mmu_notifier_release(struct mm_struct *mm)
411{
412 if (mm_has_notifiers(mm))
413 __mmu_notifier_release(mm);
414}
415
416static inline int mmu_notifier_clear_flush_young(struct mm_struct *mm,
417 unsigned long start,
418 unsigned long end)
419{
420 if (mm_has_notifiers(mm))
421 return __mmu_notifier_clear_flush_young(mm, start, end);
422 return 0;
423}
424
425static inline int mmu_notifier_clear_young(struct mm_struct *mm,
426 unsigned long start,
427 unsigned long end)
428{
429 if (mm_has_notifiers(mm))
430 return __mmu_notifier_clear_young(mm, start, end);
431 return 0;
432}
433
434static inline int mmu_notifier_test_young(struct mm_struct *mm,
435 unsigned long address)
436{
437 if (mm_has_notifiers(mm))
438 return __mmu_notifier_test_young(mm, address);
439 return 0;
440}
441
442static inline void mmu_notifier_change_pte(struct mm_struct *mm,
443 unsigned long address, pte_t pte)
444{
445 if (mm_has_notifiers(mm))
446 __mmu_notifier_change_pte(mm, address, pte);
447}
448
449static inline void
450mmu_notifier_invalidate_range_start(struct mmu_notifier_range *range)
451{
452 might_sleep();
453
454 lock_map_acquire(&__mmu_notifier_invalidate_range_start_map);
455 if (mm_has_notifiers(range->mm)) {
456 range->flags |= MMU_NOTIFIER_RANGE_BLOCKABLE;
457 __mmu_notifier_invalidate_range_start(range);
458 }
459 lock_map_release(&__mmu_notifier_invalidate_range_start_map);
460}
461
462/*
463 * This version of mmu_notifier_invalidate_range_start() avoids blocking, but it
464 * can return an error if a notifier can't proceed without blocking, in which
465 * case you're not allowed to modify PTEs in the specified range.
466 *
467 * This is mainly intended for OOM handling.
468 */
469static inline int __must_check
470mmu_notifier_invalidate_range_start_nonblock(struct mmu_notifier_range *range)
471{
472 int ret = 0;
473
474 lock_map_acquire(&__mmu_notifier_invalidate_range_start_map);
475 if (mm_has_notifiers(range->mm)) {
476 range->flags &= ~MMU_NOTIFIER_RANGE_BLOCKABLE;
477 ret = __mmu_notifier_invalidate_range_start(range);
478 }
479 lock_map_release(&__mmu_notifier_invalidate_range_start_map);
480 return ret;
481}
482
483static inline void
484mmu_notifier_invalidate_range_end(struct mmu_notifier_range *range)
485{
486 if (mmu_notifier_range_blockable(range))
487 might_sleep();
488
489 if (mm_has_notifiers(range->mm))
490 __mmu_notifier_invalidate_range_end(range);
491}
492
493static inline void mmu_notifier_arch_invalidate_secondary_tlbs(struct mm_struct *mm,
494 unsigned long start, unsigned long end)
495{
496 if (mm_has_notifiers(mm))
497 __mmu_notifier_arch_invalidate_secondary_tlbs(mm, start, end);
498}
499
500static inline void mmu_notifier_subscriptions_init(struct mm_struct *mm)
501{
502 mm->notifier_subscriptions = NULL;
503}
504
505static inline void mmu_notifier_subscriptions_destroy(struct mm_struct *mm)
506{
507 if (mm_has_notifiers(mm))
508 __mmu_notifier_subscriptions_destroy(mm);
509}
510
511
512static inline void mmu_notifier_range_init(struct mmu_notifier_range *range,
513 enum mmu_notifier_event event,
514 unsigned flags,
515 struct mm_struct *mm,
516 unsigned long start,
517 unsigned long end)
518{
519 range->event = event;
520 range->mm = mm;
521 range->start = start;
522 range->end = end;
523 range->flags = flags;
524}
525
526static inline void mmu_notifier_range_init_owner(
527 struct mmu_notifier_range *range,
528 enum mmu_notifier_event event, unsigned int flags,
529 struct mm_struct *mm, unsigned long start,
530 unsigned long end, void *owner)
531{
532 mmu_notifier_range_init(range, event, flags, mm, start, end);
533 range->owner = owner;
534}
535
536#define ptep_clear_flush_young_notify(__vma, __address, __ptep) \
537({ \
538 int __young; \
539 struct vm_area_struct *___vma = __vma; \
540 unsigned long ___address = __address; \
541 __young = ptep_clear_flush_young(___vma, ___address, __ptep); \
542 __young |= mmu_notifier_clear_flush_young(___vma->vm_mm, \
543 ___address, \
544 ___address + \
545 PAGE_SIZE); \
546 __young; \
547})
548
549#define pmdp_clear_flush_young_notify(__vma, __address, __pmdp) \
550({ \
551 int __young; \
552 struct vm_area_struct *___vma = __vma; \
553 unsigned long ___address = __address; \
554 __young = pmdp_clear_flush_young(___vma, ___address, __pmdp); \
555 __young |= mmu_notifier_clear_flush_young(___vma->vm_mm, \
556 ___address, \
557 ___address + \
558 PMD_SIZE); \
559 __young; \
560})
561
562#define ptep_clear_young_notify(__vma, __address, __ptep) \
563({ \
564 int __young; \
565 struct vm_area_struct *___vma = __vma; \
566 unsigned long ___address = __address; \
567 __young = ptep_test_and_clear_young(___vma, ___address, __ptep);\
568 __young |= mmu_notifier_clear_young(___vma->vm_mm, ___address, \
569 ___address + PAGE_SIZE); \
570 __young; \
571})
572
573#define pmdp_clear_young_notify(__vma, __address, __pmdp) \
574({ \
575 int __young; \
576 struct vm_area_struct *___vma = __vma; \
577 unsigned long ___address = __address; \
578 __young = pmdp_test_and_clear_young(___vma, ___address, __pmdp);\
579 __young |= mmu_notifier_clear_young(___vma->vm_mm, ___address, \
580 ___address + PMD_SIZE); \
581 __young; \
582})
583
584/*
585 * set_pte_at_notify() sets the pte _after_ running the notifier.
586 * This is safe to start by updating the secondary MMUs, because the primary MMU
587 * pte invalidate must have already happened with a ptep_clear_flush() before
588 * set_pte_at_notify() has been invoked. Updating the secondary MMUs first is
589 * required when we change both the protection of the mapping from read-only to
590 * read-write and the pfn (like during copy on write page faults). Otherwise the
591 * old page would remain mapped readonly in the secondary MMUs after the new
592 * page is already writable by some CPU through the primary MMU.
593 */
594#define set_pte_at_notify(__mm, __address, __ptep, __pte) \
595({ \
596 struct mm_struct *___mm = __mm; \
597 unsigned long ___address = __address; \
598 pte_t ___pte = __pte; \
599 \
600 mmu_notifier_change_pte(___mm, ___address, ___pte); \
601 set_pte_at(___mm, ___address, __ptep, ___pte); \
602})
603
604#else /* CONFIG_MMU_NOTIFIER */
605
606struct mmu_notifier_range {
607 unsigned long start;
608 unsigned long end;
609};
610
611static inline void _mmu_notifier_range_init(struct mmu_notifier_range *range,
612 unsigned long start,
613 unsigned long end)
614{
615 range->start = start;
616 range->end = end;
617}
618
619#define mmu_notifier_range_init(range,event,flags,mm,start,end) \
620 _mmu_notifier_range_init(range, start, end)
621#define mmu_notifier_range_init_owner(range, event, flags, mm, start, \
622 end, owner) \
623 _mmu_notifier_range_init(range, start, end)
624
625static inline bool
626mmu_notifier_range_blockable(const struct mmu_notifier_range *range)
627{
628 return true;
629}
630
631static inline int mm_has_notifiers(struct mm_struct *mm)
632{
633 return 0;
634}
635
636static inline void mmu_notifier_release(struct mm_struct *mm)
637{
638}
639
640static inline int mmu_notifier_clear_flush_young(struct mm_struct *mm,
641 unsigned long start,
642 unsigned long end)
643{
644 return 0;
645}
646
647static inline int mmu_notifier_test_young(struct mm_struct *mm,
648 unsigned long address)
649{
650 return 0;
651}
652
653static inline void mmu_notifier_change_pte(struct mm_struct *mm,
654 unsigned long address, pte_t pte)
655{
656}
657
658static inline void
659mmu_notifier_invalidate_range_start(struct mmu_notifier_range *range)
660{
661}
662
663static inline int
664mmu_notifier_invalidate_range_start_nonblock(struct mmu_notifier_range *range)
665{
666 return 0;
667}
668
669static inline
670void mmu_notifier_invalidate_range_end(struct mmu_notifier_range *range)
671{
672}
673
674static inline void mmu_notifier_arch_invalidate_secondary_tlbs(struct mm_struct *mm,
675 unsigned long start, unsigned long end)
676{
677}
678
679static inline void mmu_notifier_subscriptions_init(struct mm_struct *mm)
680{
681}
682
683static inline void mmu_notifier_subscriptions_destroy(struct mm_struct *mm)
684{
685}
686
687#define mmu_notifier_range_update_to_read_only(r) false
688
689#define ptep_clear_flush_young_notify ptep_clear_flush_young
690#define pmdp_clear_flush_young_notify pmdp_clear_flush_young
691#define ptep_clear_young_notify ptep_test_and_clear_young
692#define pmdp_clear_young_notify pmdp_test_and_clear_young
693#define ptep_clear_flush_notify ptep_clear_flush
694#define pmdp_huge_clear_flush_notify pmdp_huge_clear_flush
695#define pudp_huge_clear_flush_notify pudp_huge_clear_flush
696#define set_pte_at_notify set_pte_at
697
698static inline void mmu_notifier_synchronize(void)
699{
700}
701
702#endif /* CONFIG_MMU_NOTIFIER */
703
704#endif /* _LINUX_MMU_NOTIFIER_H */