blob: 9e6caa8ecd1938621a0607cf8542e78cea73f91a [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001/* SPDX-License-Identifier: GPL-2.0 */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002#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>
Sagi Grimberg21a92732012-10-08 16:29:24 -07008#include <linux/srcu.h>
Jason Gunthorpe99cb2522019-11-12 16:22:19 -04009#include <linux/interval_tree.h>
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070010
Jason Gunthorpe56f434f2019-11-12 16:22:18 -040011struct mmu_notifier_mm;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070012struct mmu_notifier;
Jason Gunthorpe56f434f2019-11-12 16:22:18 -040013struct mmu_notifier_range;
Jason Gunthorpe99cb2522019-11-12 16:22:19 -040014struct mmu_interval_notifier;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070015
Jérôme Glissed87f0552019-05-13 17:20:45 -070016/**
17 * enum mmu_notifier_event - reason for the mmu notifier callback
18 * @MMU_NOTIFY_UNMAP: either munmap() that unmap the range or a mremap() that
19 * move the range
20 *
21 * @MMU_NOTIFY_CLEAR: clear page table entry (many reasons for this like
22 * madvise() or replacing a page by another one, ...).
23 *
24 * @MMU_NOTIFY_PROTECTION_VMA: update is due to protection change for the range
25 * ie using the vma access permission (vm_page_prot) to update the whole range
26 * is enough no need to inspect changes to the CPU page table (mprotect()
27 * syscall)
28 *
29 * @MMU_NOTIFY_PROTECTION_PAGE: update is due to change in read/write flag for
30 * pages in the range so to mirror those changes the user must inspect the CPU
31 * page table (from the end callback).
32 *
33 * @MMU_NOTIFY_SOFT_DIRTY: soft dirty accounting (still same page and same
34 * access flags). User should soft dirty the page in the end callback to make
35 * sure that anyone relying on soft dirtyness catch pages that might be written
36 * through non CPU mappings.
Jason Gunthorpe99cb2522019-11-12 16:22:19 -040037 *
38 * @MMU_NOTIFY_RELEASE: used during mmu_interval_notifier invalidate to signal
39 * that the mm refcount is zero and the range is no longer accessible.
Jérôme Glissed87f0552019-05-13 17:20:45 -070040 */
41enum mmu_notifier_event {
42 MMU_NOTIFY_UNMAP = 0,
43 MMU_NOTIFY_CLEAR,
44 MMU_NOTIFY_PROTECTION_VMA,
45 MMU_NOTIFY_PROTECTION_PAGE,
46 MMU_NOTIFY_SOFT_DIRTY,
Jason Gunthorpe99cb2522019-11-12 16:22:19 -040047 MMU_NOTIFY_RELEASE,
Jérôme Glissed87f0552019-05-13 17:20:45 -070048};
49
Jérôme Glisse27560ee2019-05-13 17:20:42 -070050#define MMU_NOTIFIER_RANGE_BLOCKABLE (1 << 0)
51
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070052struct mmu_notifier_ops {
53 /*
54 * Called either by mmu_notifier_unregister or when the mm is
55 * being destroyed by exit_mmap, always before all pages are
56 * freed. This can run concurrently with other mmu notifier
57 * methods (the ones invoked outside the mm context) and it
58 * should tear down all secondary mmu mappings and freeze the
59 * secondary mmu. If this method isn't implemented you've to
60 * be sure that nothing could possibly write to the pages
61 * through the secondary mmu by the time the last thread with
62 * tsk->mm == mm exits.
63 *
64 * As side note: the pages freed after ->release returns could
65 * be immediately reallocated by the gart at an alias physical
66 * address with a different cache model, so if ->release isn't
67 * implemented because all _software_ driven memory accesses
68 * through the secondary mmu are terminated by the time the
69 * last thread of this mm quits, you've also to be sure that
70 * speculative _hardware_ operations can't allocate dirty
71 * cachelines in the cpu that could not be snooped and made
72 * coherent with the other read and write operations happening
73 * through the gart alias address, so leading to memory
74 * corruption.
75 */
76 void (*release)(struct mmu_notifier *mn,
77 struct mm_struct *mm);
78
79 /*
80 * clear_flush_young is called after the VM is
81 * test-and-clearing the young/accessed bitflag in the
82 * pte. This way the VM will provide proper aging to the
83 * accesses to the page through the secondary MMUs and not
84 * only to the ones through the Linux pte.
Andres Lagar-Cavilla57128462014-09-22 14:54:42 -070085 * Start-end is necessary in case the secondary MMU is mapping the page
86 * at a smaller granularity than the primary MMU.
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070087 */
88 int (*clear_flush_young)(struct mmu_notifier *mn,
89 struct mm_struct *mm,
Andres Lagar-Cavilla57128462014-09-22 14:54:42 -070090 unsigned long start,
91 unsigned long end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070092
93 /*
Vladimir Davydov1d7715c2015-09-09 15:35:41 -070094 * clear_young is a lightweight version of clear_flush_young. Like the
95 * latter, it is supposed to test-and-clear the young/accessed bitflag
96 * in the secondary pte, but it may omit flushing the secondary tlb.
97 */
98 int (*clear_young)(struct mmu_notifier *mn,
99 struct mm_struct *mm,
100 unsigned long start,
101 unsigned long end);
102
103 /*
Andrea Arcangeli8ee53822011-01-13 15:47:10 -0800104 * test_young is called to check the young/accessed bitflag in
105 * the secondary pte. This is used to know if the page is
106 * frequently used without actually clearing the flag or tearing
107 * down the secondary mapping on the page.
108 */
109 int (*test_young)(struct mmu_notifier *mn,
110 struct mm_struct *mm,
111 unsigned long address);
112
113 /*
Izik Eidus828502d2009-09-21 17:01:51 -0700114 * change_pte is called in cases that pte mapping to page is changed:
115 * for example, when ksm remaps pte to point to a new shared page.
116 */
117 void (*change_pte)(struct mmu_notifier *mn,
118 struct mm_struct *mm,
119 unsigned long address,
120 pte_t pte);
121
122 /*
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700123 * invalidate_range_start() and invalidate_range_end() must be
124 * paired and are called only when the mmap_sem and/or the
Joerg Roedel0f0a3272014-11-13 13:46:09 +1100125 * locks protecting the reverse maps are held. If the subsystem
126 * can't guarantee that no additional references are taken to
127 * the pages in the range, it has to implement the
128 * invalidate_range() notifier to remove any references taken
129 * after invalidate_range_start().
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700130 *
131 * Invalidation of multiple concurrent ranges may be
132 * optionally permitted by the driver. Either way the
133 * establishment of sptes is forbidden in the range passed to
134 * invalidate_range_begin/end for the whole duration of the
135 * invalidate_range_begin/end critical section.
136 *
137 * invalidate_range_start() is called when all pages in the
138 * range are still mapped and have at least a refcount of one.
139 *
140 * invalidate_range_end() is called when all pages in the
141 * range have been unmapped and the pages have been freed by
142 * the VM.
143 *
144 * The VM will remove the page table entries and potentially
145 * the page between invalidate_range_start() and
146 * invalidate_range_end(). If the page must not be freed
147 * because of pending I/O or other circumstances then the
148 * invalidate_range_start() callback (or the initial mapping
149 * by the driver) must make sure that the refcount is kept
150 * elevated.
151 *
152 * If the driver increases the refcount when the pages are
153 * initially mapped into an address space then either
154 * invalidate_range_start() or invalidate_range_end() may
155 * decrease the refcount. If the refcount is decreased on
156 * invalidate_range_start() then the VM can free pages as page
157 * table entries are removed. If the refcount is only
158 * droppped on invalidate_range_end() then the driver itself
159 * will drop the last refcount but it must take care to flush
160 * any secondary tlb before doing the final free on the
161 * page. Pages will no longer be referenced by the linux
162 * address space but may still be referenced by sptes until
163 * the last refcount is dropped.
David Rientjes5ff70912018-01-31 16:18:32 -0800164 *
Michal Hocko93065ac2018-08-21 21:52:33 -0700165 * If blockable argument is set to false then the callback cannot
166 * sleep and has to return with -EAGAIN. 0 should be returned
Michal Hocko33490af2018-10-26 15:03:35 -0700167 * otherwise. Please note that if invalidate_range_start approves
168 * a non-blocking behavior then the same applies to
169 * invalidate_range_end.
Michal Hocko93065ac2018-08-21 21:52:33 -0700170 *
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700171 */
Michal Hocko93065ac2018-08-21 21:52:33 -0700172 int (*invalidate_range_start)(struct mmu_notifier *mn,
Jérôme Glisse5d6527a2018-12-28 00:38:05 -0800173 const struct mmu_notifier_range *range);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700174 void (*invalidate_range_end)(struct mmu_notifier *mn,
Jérôme Glisse5d6527a2018-12-28 00:38:05 -0800175 const struct mmu_notifier_range *range);
Joerg Roedel0f0a3272014-11-13 13:46:09 +1100176
177 /*
178 * invalidate_range() is either called between
179 * invalidate_range_start() and invalidate_range_end() when the
180 * VM has to free pages that where unmapped, but before the
181 * pages are actually freed, or outside of _start()/_end() when
182 * a (remote) TLB is necessary.
183 *
184 * If invalidate_range() is used to manage a non-CPU TLB with
185 * shared page-tables, it not necessary to implement the
186 * invalidate_range_start()/end() notifiers, as
187 * invalidate_range() alread catches the points in time when an
Jérôme Glisse0f108512017-11-15 17:34:07 -0800188 * external TLB range needs to be flushed. For more in depth
Mike Rapoportad56b732018-03-21 21:22:47 +0200189 * discussion on this see Documentation/vm/mmu_notifier.rst
Joerg Roedel0f0a3272014-11-13 13:46:09 +1100190 *
Joerg Roedel0f0a3272014-11-13 13:46:09 +1100191 * Note that this function might be called with just a sub-range
192 * of what was passed to invalidate_range_start()/end(), if
193 * called between those functions.
194 */
195 void (*invalidate_range)(struct mmu_notifier *mn, struct mm_struct *mm,
196 unsigned long start, unsigned long end);
Jason Gunthorpe2c7933f2019-08-06 20:15:40 -0300197
198 /*
199 * These callbacks are used with the get/put interface to manage the
200 * lifetime of the mmu_notifier memory. alloc_notifier() returns a new
201 * notifier for use with the mm.
202 *
203 * free_notifier() is only called after the mmu_notifier has been
204 * fully put, calls to any ops callback are prevented and no ops
205 * callbacks are currently running. It is called from a SRCU callback
206 * and cannot sleep.
207 */
208 struct mmu_notifier *(*alloc_notifier)(struct mm_struct *mm);
209 void (*free_notifier)(struct mmu_notifier *mn);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700210};
211
212/*
213 * The notifier chains are protected by mmap_sem and/or the reverse map
214 * semaphores. Notifier chains are only changed when all reverse maps and
215 * the mmap_sem locks are taken.
216 *
217 * Therefore notifier chains can only be traversed when either
218 *
219 * 1. mmap_sem is held.
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800220 * 2. One of the reverse map locks is held (i_mmap_rwsem or anon_vma->rwsem).
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700221 * 3. No other concurrent thread can access the list (release)
222 */
223struct mmu_notifier {
224 struct hlist_node hlist;
225 const struct mmu_notifier_ops *ops;
Jason Gunthorpe2c7933f2019-08-06 20:15:40 -0300226 struct mm_struct *mm;
227 struct rcu_head rcu;
228 unsigned int users;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700229};
230
Jason Gunthorpe99cb2522019-11-12 16:22:19 -0400231/**
232 * struct mmu_interval_notifier_ops
233 * @invalidate: Upon return the caller must stop using any SPTEs within this
234 * range. This function can sleep. Return false only if sleeping
235 * was required but mmu_notifier_range_blockable(range) is false.
236 */
237struct mmu_interval_notifier_ops {
238 bool (*invalidate)(struct mmu_interval_notifier *mni,
239 const struct mmu_notifier_range *range,
240 unsigned long cur_seq);
241};
242
243struct mmu_interval_notifier {
244 struct interval_tree_node interval_tree;
245 const struct mmu_interval_notifier_ops *ops;
246 struct mm_struct *mm;
247 struct hlist_node deferred_item;
248 unsigned long invalidate_seq;
249};
250
Jason Gunthorpe56f434f2019-11-12 16:22:18 -0400251#ifdef CONFIG_MMU_NOTIFIER
252
253#ifdef CONFIG_LOCKDEP
254extern struct lockdep_map __mmu_notifier_invalidate_range_start_map;
255#endif
256
257struct mmu_notifier_range {
258 struct vm_area_struct *vma;
259 struct mm_struct *mm;
260 unsigned long start;
261 unsigned long end;
262 unsigned flags;
263 enum mmu_notifier_event event;
264};
265
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700266static inline int mm_has_notifiers(struct mm_struct *mm)
267{
268 return unlikely(mm->mmu_notifier_mm);
269}
270
Jason Gunthorpe2c7933f2019-08-06 20:15:40 -0300271struct mmu_notifier *mmu_notifier_get_locked(const struct mmu_notifier_ops *ops,
272 struct mm_struct *mm);
273static inline struct mmu_notifier *
274mmu_notifier_get(const struct mmu_notifier_ops *ops, struct mm_struct *mm)
275{
276 struct mmu_notifier *ret;
277
278 down_write(&mm->mmap_sem);
279 ret = mmu_notifier_get_locked(ops, mm);
280 up_write(&mm->mmap_sem);
281 return ret;
282}
283void mmu_notifier_put(struct mmu_notifier *mn);
284void mmu_notifier_synchronize(void);
285
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700286extern int mmu_notifier_register(struct mmu_notifier *mn,
287 struct mm_struct *mm);
288extern int __mmu_notifier_register(struct mmu_notifier *mn,
289 struct mm_struct *mm);
290extern void mmu_notifier_unregister(struct mmu_notifier *mn,
291 struct mm_struct *mm);
Jason Gunthorpe99cb2522019-11-12 16:22:19 -0400292
293unsigned long mmu_interval_read_begin(struct mmu_interval_notifier *mni);
294int mmu_interval_notifier_insert(struct mmu_interval_notifier *mni,
295 struct mm_struct *mm, unsigned long start,
296 unsigned long length,
297 const struct mmu_interval_notifier_ops *ops);
298int mmu_interval_notifier_insert_locked(
299 struct mmu_interval_notifier *mni, struct mm_struct *mm,
300 unsigned long start, unsigned long length,
301 const struct mmu_interval_notifier_ops *ops);
302void mmu_interval_notifier_remove(struct mmu_interval_notifier *mni);
303
304/**
305 * mmu_interval_set_seq - Save the invalidation sequence
306 * @mni - The mni passed to invalidate
307 * @cur_seq - The cur_seq passed to the invalidate() callback
308 *
309 * This must be called unconditionally from the invalidate callback of a
310 * struct mmu_interval_notifier_ops under the same lock that is used to call
311 * mmu_interval_read_retry(). It updates the sequence number for later use by
312 * mmu_interval_read_retry(). The provided cur_seq will always be odd.
313 *
314 * If the caller does not call mmu_interval_read_begin() or
315 * mmu_interval_read_retry() then this call is not required.
316 */
317static inline void mmu_interval_set_seq(struct mmu_interval_notifier *mni,
318 unsigned long cur_seq)
319{
320 WRITE_ONCE(mni->invalidate_seq, cur_seq);
321}
322
323/**
324 * mmu_interval_read_retry - End a read side critical section against a VA range
325 * mni: The range
326 * seq: The return of the paired mmu_interval_read_begin()
327 *
328 * This MUST be called under a user provided lock that is also held
329 * unconditionally by op->invalidate() when it calls mmu_interval_set_seq().
330 *
331 * Each call should be paired with a single mmu_interval_read_begin() and
332 * should be used to conclude the read side.
333 *
334 * Returns true if an invalidation collided with this critical section, and
335 * the caller should retry.
336 */
337static inline bool mmu_interval_read_retry(struct mmu_interval_notifier *mni,
338 unsigned long seq)
339{
340 return mni->invalidate_seq != seq;
341}
342
343/**
344 * mmu_interval_check_retry - Test if a collision has occurred
345 * mni: The range
346 * seq: The return of the matching mmu_interval_read_begin()
347 *
348 * This can be used in the critical section between mmu_interval_read_begin()
349 * and mmu_interval_read_retry(). A return of true indicates an invalidation
350 * has collided with this critical region and a future
351 * mmu_interval_read_retry() will return true.
352 *
353 * False is not reliable and only suggests a collision may not have
354 * occured. It can be called many times and does not have to hold the user
355 * provided lock.
356 *
357 * This call can be used as part of loops and other expensive operations to
358 * expedite a retry.
359 */
360static inline bool mmu_interval_check_retry(struct mmu_interval_notifier *mni,
361 unsigned long seq)
362{
363 /* Pairs with the WRITE_ONCE in mmu_interval_set_seq() */
364 return READ_ONCE(mni->invalidate_seq) != seq;
365}
366
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700367extern void __mmu_notifier_mm_destroy(struct mm_struct *mm);
368extern void __mmu_notifier_release(struct mm_struct *mm);
369extern int __mmu_notifier_clear_flush_young(struct mm_struct *mm,
Andres Lagar-Cavilla57128462014-09-22 14:54:42 -0700370 unsigned long start,
371 unsigned long end);
Vladimir Davydov1d7715c2015-09-09 15:35:41 -0700372extern int __mmu_notifier_clear_young(struct mm_struct *mm,
373 unsigned long start,
374 unsigned long end);
Andrea Arcangeli8ee53822011-01-13 15:47:10 -0800375extern int __mmu_notifier_test_young(struct mm_struct *mm,
376 unsigned long address);
Izik Eidus828502d2009-09-21 17:01:51 -0700377extern void __mmu_notifier_change_pte(struct mm_struct *mm,
378 unsigned long address, pte_t pte);
Jérôme Glisseac46d4f2018-12-28 00:38:09 -0800379extern int __mmu_notifier_invalidate_range_start(struct mmu_notifier_range *r);
380extern void __mmu_notifier_invalidate_range_end(struct mmu_notifier_range *r,
Jérôme Glisse4645b9f2017-11-15 17:34:11 -0800381 bool only_end);
Joerg Roedel0f0a3272014-11-13 13:46:09 +1100382extern void __mmu_notifier_invalidate_range(struct mm_struct *mm,
383 unsigned long start, unsigned long end);
Jérôme Glissec6d23412019-05-13 17:21:00 -0700384extern bool
385mmu_notifier_range_update_to_read_only(const struct mmu_notifier_range *range);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700386
Jérôme Glisse4a83bfe2019-05-13 17:20:34 -0700387static inline bool
388mmu_notifier_range_blockable(const struct mmu_notifier_range *range)
389{
Jérôme Glisse27560ee2019-05-13 17:20:42 -0700390 return (range->flags & MMU_NOTIFIER_RANGE_BLOCKABLE);
Jérôme Glisse4a83bfe2019-05-13 17:20:34 -0700391}
392
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700393static inline void mmu_notifier_release(struct mm_struct *mm)
394{
395 if (mm_has_notifiers(mm))
396 __mmu_notifier_release(mm);
397}
398
399static inline int mmu_notifier_clear_flush_young(struct mm_struct *mm,
Andres Lagar-Cavilla57128462014-09-22 14:54:42 -0700400 unsigned long start,
401 unsigned long end)
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700402{
403 if (mm_has_notifiers(mm))
Andres Lagar-Cavilla57128462014-09-22 14:54:42 -0700404 return __mmu_notifier_clear_flush_young(mm, start, end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700405 return 0;
406}
407
Vladimir Davydov1d7715c2015-09-09 15:35:41 -0700408static inline int mmu_notifier_clear_young(struct mm_struct *mm,
409 unsigned long start,
410 unsigned long end)
411{
412 if (mm_has_notifiers(mm))
413 return __mmu_notifier_clear_young(mm, start, end);
414 return 0;
415}
416
Andrea Arcangeli8ee53822011-01-13 15:47:10 -0800417static inline int mmu_notifier_test_young(struct mm_struct *mm,
418 unsigned long address)
419{
420 if (mm_has_notifiers(mm))
421 return __mmu_notifier_test_young(mm, address);
422 return 0;
423}
424
Izik Eidus828502d2009-09-21 17:01:51 -0700425static inline void mmu_notifier_change_pte(struct mm_struct *mm,
426 unsigned long address, pte_t pte)
427{
428 if (mm_has_notifiers(mm))
429 __mmu_notifier_change_pte(mm, address, pte);
430}
431
Jérôme Glisseac46d4f2018-12-28 00:38:09 -0800432static inline void
433mmu_notifier_invalidate_range_start(struct mmu_notifier_range *range)
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700434{
Daniel Vetter810e24e2019-08-26 22:14:25 +0200435 might_sleep();
436
Daniel Vetter23b68392019-08-26 22:14:21 +0200437 lock_map_acquire(&__mmu_notifier_invalidate_range_start_map);
Jérôme Glisseac46d4f2018-12-28 00:38:09 -0800438 if (mm_has_notifiers(range->mm)) {
Jérôme Glisse27560ee2019-05-13 17:20:42 -0700439 range->flags |= MMU_NOTIFIER_RANGE_BLOCKABLE;
Jérôme Glisseac46d4f2018-12-28 00:38:09 -0800440 __mmu_notifier_invalidate_range_start(range);
441 }
Daniel Vetter23b68392019-08-26 22:14:21 +0200442 lock_map_release(&__mmu_notifier_invalidate_range_start_map);
Michal Hocko93065ac2018-08-21 21:52:33 -0700443}
444
Jérôme Glisseac46d4f2018-12-28 00:38:09 -0800445static inline int
446mmu_notifier_invalidate_range_start_nonblock(struct mmu_notifier_range *range)
Michal Hocko93065ac2018-08-21 21:52:33 -0700447{
Daniel Vetter23b68392019-08-26 22:14:21 +0200448 int ret = 0;
449
450 lock_map_acquire(&__mmu_notifier_invalidate_range_start_map);
Jérôme Glisseac46d4f2018-12-28 00:38:09 -0800451 if (mm_has_notifiers(range->mm)) {
Jérôme Glisse27560ee2019-05-13 17:20:42 -0700452 range->flags &= ~MMU_NOTIFIER_RANGE_BLOCKABLE;
Daniel Vetter23b68392019-08-26 22:14:21 +0200453 ret = __mmu_notifier_invalidate_range_start(range);
Jérôme Glisseac46d4f2018-12-28 00:38:09 -0800454 }
Daniel Vetter23b68392019-08-26 22:14:21 +0200455 lock_map_release(&__mmu_notifier_invalidate_range_start_map);
456 return ret;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700457}
458
Jérôme Glisseac46d4f2018-12-28 00:38:09 -0800459static inline void
460mmu_notifier_invalidate_range_end(struct mmu_notifier_range *range)
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700461{
Daniel Vetter810e24e2019-08-26 22:14:25 +0200462 if (mmu_notifier_range_blockable(range))
463 might_sleep();
464
Jérôme Glisseac46d4f2018-12-28 00:38:09 -0800465 if (mm_has_notifiers(range->mm))
466 __mmu_notifier_invalidate_range_end(range, false);
Jérôme Glisse4645b9f2017-11-15 17:34:11 -0800467}
468
Jérôme Glisseac46d4f2018-12-28 00:38:09 -0800469static inline void
470mmu_notifier_invalidate_range_only_end(struct mmu_notifier_range *range)
Jérôme Glisse4645b9f2017-11-15 17:34:11 -0800471{
Jérôme Glisseac46d4f2018-12-28 00:38:09 -0800472 if (mm_has_notifiers(range->mm))
473 __mmu_notifier_invalidate_range_end(range, true);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700474}
475
Joerg Roedel1897bdc2014-11-13 13:46:09 +1100476static inline void mmu_notifier_invalidate_range(struct mm_struct *mm,
477 unsigned long start, unsigned long end)
478{
Joerg Roedel0f0a3272014-11-13 13:46:09 +1100479 if (mm_has_notifiers(mm))
480 __mmu_notifier_invalidate_range(mm, start, end);
Joerg Roedel1897bdc2014-11-13 13:46:09 +1100481}
482
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700483static inline void mmu_notifier_mm_init(struct mm_struct *mm)
484{
485 mm->mmu_notifier_mm = NULL;
486}
487
488static inline void mmu_notifier_mm_destroy(struct mm_struct *mm)
489{
490 if (mm_has_notifiers(mm))
491 __mmu_notifier_mm_destroy(mm);
492}
493
Jérôme Glisseac46d4f2018-12-28 00:38:09 -0800494
495static inline void mmu_notifier_range_init(struct mmu_notifier_range *range,
Jérôme Glisse6f4f13e2019-05-13 17:20:49 -0700496 enum mmu_notifier_event event,
497 unsigned flags,
498 struct vm_area_struct *vma,
Jérôme Glisseac46d4f2018-12-28 00:38:09 -0800499 struct mm_struct *mm,
500 unsigned long start,
501 unsigned long end)
502{
Jérôme Glissebf198b22019-05-13 17:20:57 -0700503 range->vma = vma;
504 range->event = event;
Jérôme Glisseac46d4f2018-12-28 00:38:09 -0800505 range->mm = mm;
506 range->start = start;
507 range->end = end;
Jérôme Glissebf198b22019-05-13 17:20:57 -0700508 range->flags = flags;
Jérôme Glisseac46d4f2018-12-28 00:38:09 -0800509}
510
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700511#define ptep_clear_flush_young_notify(__vma, __address, __ptep) \
512({ \
513 int __young; \
514 struct vm_area_struct *___vma = __vma; \
515 unsigned long ___address = __address; \
516 __young = ptep_clear_flush_young(___vma, ___address, __ptep); \
517 __young |= mmu_notifier_clear_flush_young(___vma->vm_mm, \
Andres Lagar-Cavilla57128462014-09-22 14:54:42 -0700518 ___address, \
519 ___address + \
520 PAGE_SIZE); \
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700521 __young; \
522})
523
Andrea Arcangeli91a4ee22011-01-13 15:46:44 -0800524#define pmdp_clear_flush_young_notify(__vma, __address, __pmdp) \
525({ \
526 int __young; \
527 struct vm_area_struct *___vma = __vma; \
528 unsigned long ___address = __address; \
529 __young = pmdp_clear_flush_young(___vma, ___address, __pmdp); \
530 __young |= mmu_notifier_clear_flush_young(___vma->vm_mm, \
Andres Lagar-Cavilla57128462014-09-22 14:54:42 -0700531 ___address, \
532 ___address + \
533 PMD_SIZE); \
Andrea Arcangeli91a4ee22011-01-13 15:46:44 -0800534 __young; \
535})
536
Vladimir Davydov1d7715c2015-09-09 15:35:41 -0700537#define ptep_clear_young_notify(__vma, __address, __ptep) \
538({ \
539 int __young; \
540 struct vm_area_struct *___vma = __vma; \
541 unsigned long ___address = __address; \
542 __young = ptep_test_and_clear_young(___vma, ___address, __ptep);\
543 __young |= mmu_notifier_clear_young(___vma->vm_mm, ___address, \
544 ___address + PAGE_SIZE); \
545 __young; \
546})
547
548#define pmdp_clear_young_notify(__vma, __address, __pmdp) \
549({ \
550 int __young; \
551 struct vm_area_struct *___vma = __vma; \
552 unsigned long ___address = __address; \
553 __young = pmdp_test_and_clear_young(___vma, ___address, __pmdp);\
554 __young |= mmu_notifier_clear_young(___vma->vm_mm, ___address, \
555 ___address + PMD_SIZE); \
556 __young; \
557})
558
Joerg Roedel34ee6452014-11-13 13:46:09 +1100559#define ptep_clear_flush_notify(__vma, __address, __ptep) \
560({ \
561 unsigned long ___addr = __address & PAGE_MASK; \
562 struct mm_struct *___mm = (__vma)->vm_mm; \
563 pte_t ___pte; \
564 \
565 ___pte = ptep_clear_flush(__vma, __address, __ptep); \
566 mmu_notifier_invalidate_range(___mm, ___addr, \
567 ___addr + PAGE_SIZE); \
568 \
569 ___pte; \
570})
571
Aneesh Kumar K.V8809aa22015-06-24 16:57:44 -0700572#define pmdp_huge_clear_flush_notify(__vma, __haddr, __pmd) \
Joerg Roedel34ee6452014-11-13 13:46:09 +1100573({ \
574 unsigned long ___haddr = __haddr & HPAGE_PMD_MASK; \
575 struct mm_struct *___mm = (__vma)->vm_mm; \
576 pmd_t ___pmd; \
577 \
Aneesh Kumar K.V8809aa22015-06-24 16:57:44 -0700578 ___pmd = pmdp_huge_clear_flush(__vma, __haddr, __pmd); \
Joerg Roedel34ee6452014-11-13 13:46:09 +1100579 mmu_notifier_invalidate_range(___mm, ___haddr, \
580 ___haddr + HPAGE_PMD_SIZE); \
581 \
582 ___pmd; \
583})
584
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -0800585#define pudp_huge_clear_flush_notify(__vma, __haddr, __pud) \
586({ \
587 unsigned long ___haddr = __haddr & HPAGE_PUD_MASK; \
588 struct mm_struct *___mm = (__vma)->vm_mm; \
589 pud_t ___pud; \
590 \
591 ___pud = pudp_huge_clear_flush(__vma, __haddr, __pud); \
592 mmu_notifier_invalidate_range(___mm, ___haddr, \
593 ___haddr + HPAGE_PUD_SIZE); \
594 \
595 ___pud; \
596})
597
Xiao Guangrong48af0d72012-10-08 16:29:23 -0700598/*
599 * set_pte_at_notify() sets the pte _after_ running the notifier.
600 * This is safe to start by updating the secondary MMUs, because the primary MMU
601 * pte invalidate must have already happened with a ptep_clear_flush() before
602 * set_pte_at_notify() has been invoked. Updating the secondary MMUs first is
603 * required when we change both the protection of the mapping from read-only to
604 * read-write and the pfn (like during copy on write page faults). Otherwise the
605 * old page would remain mapped readonly in the secondary MMUs after the new
606 * page is already writable by some CPU through the primary MMU.
607 */
Izik Eidus828502d2009-09-21 17:01:51 -0700608#define set_pte_at_notify(__mm, __address, __ptep, __pte) \
609({ \
610 struct mm_struct *___mm = __mm; \
611 unsigned long ___address = __address; \
612 pte_t ___pte = __pte; \
613 \
Izik Eidus828502d2009-09-21 17:01:51 -0700614 mmu_notifier_change_pte(___mm, ___address, ___pte); \
Xiao Guangrong48af0d72012-10-08 16:29:23 -0700615 set_pte_at(___mm, ___address, __ptep, ___pte); \
Izik Eidus828502d2009-09-21 17:01:51 -0700616})
617
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700618#else /* CONFIG_MMU_NOTIFIER */
619
Jérôme Glisseac46d4f2018-12-28 00:38:09 -0800620struct mmu_notifier_range {
621 unsigned long start;
622 unsigned long end;
623};
624
625static inline void _mmu_notifier_range_init(struct mmu_notifier_range *range,
626 unsigned long start,
627 unsigned long end)
628{
629 range->start = start;
630 range->end = end;
631}
632
Jérôme Glisse6f4f13e2019-05-13 17:20:49 -0700633#define mmu_notifier_range_init(range,event,flags,vma,mm,start,end) \
Jérôme Glisseac46d4f2018-12-28 00:38:09 -0800634 _mmu_notifier_range_init(range, start, end)
635
Jérôme Glisse4a83bfe2019-05-13 17:20:34 -0700636static inline bool
637mmu_notifier_range_blockable(const struct mmu_notifier_range *range)
638{
639 return true;
640}
Jérôme Glisseac46d4f2018-12-28 00:38:09 -0800641
Michal Hocko4d4bbd82017-10-03 16:14:50 -0700642static inline int mm_has_notifiers(struct mm_struct *mm)
643{
644 return 0;
645}
646
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700647static inline void mmu_notifier_release(struct mm_struct *mm)
648{
649}
650
651static inline int mmu_notifier_clear_flush_young(struct mm_struct *mm,
Andres Lagar-Cavilla57128462014-09-22 14:54:42 -0700652 unsigned long start,
653 unsigned long end)
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700654{
655 return 0;
656}
657
Andrea Arcangeli8ee53822011-01-13 15:47:10 -0800658static inline int mmu_notifier_test_young(struct mm_struct *mm,
659 unsigned long address)
660{
661 return 0;
662}
663
Izik Eidus828502d2009-09-21 17:01:51 -0700664static inline void mmu_notifier_change_pte(struct mm_struct *mm,
665 unsigned long address, pte_t pte)
666{
667}
668
Jérôme Glisseac46d4f2018-12-28 00:38:09 -0800669static inline void
670mmu_notifier_invalidate_range_start(struct mmu_notifier_range *range)
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700671{
672}
673
Jérôme Glisseac46d4f2018-12-28 00:38:09 -0800674static inline int
675mmu_notifier_invalidate_range_start_nonblock(struct mmu_notifier_range *range)
Michal Hocko93065ac2018-08-21 21:52:33 -0700676{
677 return 0;
678}
679
Jérôme Glisseac46d4f2018-12-28 00:38:09 -0800680static inline
681void mmu_notifier_invalidate_range_end(struct mmu_notifier_range *range)
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700682{
683}
684
Jérôme Glisseac46d4f2018-12-28 00:38:09 -0800685static inline void
686mmu_notifier_invalidate_range_only_end(struct mmu_notifier_range *range)
Jérôme Glisse4645b9f2017-11-15 17:34:11 -0800687{
688}
689
Joerg Roedel1897bdc2014-11-13 13:46:09 +1100690static inline void mmu_notifier_invalidate_range(struct mm_struct *mm,
691 unsigned long start, unsigned long end)
692{
693}
694
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700695static inline void mmu_notifier_mm_init(struct mm_struct *mm)
696{
697}
698
699static inline void mmu_notifier_mm_destroy(struct mm_struct *mm)
700{
701}
702
Jérôme Glissec6d23412019-05-13 17:21:00 -0700703#define mmu_notifier_range_update_to_read_only(r) false
704
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700705#define ptep_clear_flush_young_notify ptep_clear_flush_young
Andrea Arcangeli91a4ee22011-01-13 15:46:44 -0800706#define pmdp_clear_flush_young_notify pmdp_clear_flush_young
Vladimir Davydov33c3fc72015-09-09 15:35:45 -0700707#define ptep_clear_young_notify ptep_test_and_clear_young
708#define pmdp_clear_young_notify pmdp_test_and_clear_young
Joerg Roedel34ee6452014-11-13 13:46:09 +1100709#define ptep_clear_flush_notify ptep_clear_flush
Aneesh Kumar K.V8809aa22015-06-24 16:57:44 -0700710#define pmdp_huge_clear_flush_notify pmdp_huge_clear_flush
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -0800711#define pudp_huge_clear_flush_notify pudp_huge_clear_flush
Izik Eidus828502d2009-09-21 17:01:51 -0700712#define set_pte_at_notify set_pte_at
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700713
Jason Gunthorpe2c7933f2019-08-06 20:15:40 -0300714static inline void mmu_notifier_synchronize(void)
715{
716}
717
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700718#endif /* CONFIG_MMU_NOTIFIER */
719
720#endif /* _LINUX_MMU_NOTIFIER_H */