Commit | Line | Data |
---|---|---|
cddb8a5c AA |
1 | /* |
2 | * linux/mm/mmu_notifier.c | |
3 | * | |
4 | * Copyright (C) 2008 Qumranet, Inc. | |
5 | * Copyright (C) 2008 SGI | |
6 | * Christoph Lameter <clameter@sgi.com> | |
7 | * | |
8 | * This work is licensed under the terms of the GNU GPL, version 2. See | |
9 | * the COPYING file in the top-level directory. | |
10 | */ | |
11 | ||
12 | #include <linux/rculist.h> | |
13 | #include <linux/mmu_notifier.h> | |
b95f1b31 | 14 | #include <linux/export.h> |
cddb8a5c AA |
15 | #include <linux/mm.h> |
16 | #include <linux/err.h> | |
21a92735 | 17 | #include <linux/srcu.h> |
cddb8a5c AA |
18 | #include <linux/rcupdate.h> |
19 | #include <linux/sched.h> | |
5a0e3ad6 | 20 | #include <linux/slab.h> |
cddb8a5c | 21 | |
21a92735 | 22 | /* global SRCU for all MMs */ |
70400303 | 23 | static struct srcu_struct srcu; |
21a92735 | 24 | |
cddb8a5c AA |
25 | /* |
26 | * This function can't run concurrently against mmu_notifier_register | |
27 | * because mm->mm_users > 0 during mmu_notifier_register and exit_mmap | |
28 | * runs with mm_users == 0. Other tasks may still invoke mmu notifiers | |
29 | * in parallel despite there being no task using this mm any more, | |
30 | * through the vmas outside of the exit_mmap context, such as with | |
31 | * vmtruncate. This serializes against mmu_notifier_unregister with | |
21a92735 SG |
32 | * the mmu_notifier_mm->lock in addition to SRCU and it serializes |
33 | * against the other mmu notifiers with SRCU. struct mmu_notifier_mm | |
cddb8a5c AA |
34 | * can't go away from under us as exit_mmap holds an mm_count pin |
35 | * itself. | |
36 | */ | |
37 | void __mmu_notifier_release(struct mm_struct *mm) | |
38 | { | |
39 | struct mmu_notifier *mn; | |
21a92735 | 40 | int id; |
3ad3d901 XG |
41 | |
42 | /* | |
751efd86 RH |
43 | * srcu_read_lock() here will block synchronize_srcu() in |
44 | * mmu_notifier_unregister() until all registered | |
45 | * ->release() callouts this function makes have | |
46 | * returned. | |
3ad3d901 | 47 | */ |
21a92735 | 48 | id = srcu_read_lock(&srcu); |
cddb8a5c AA |
49 | spin_lock(&mm->mmu_notifier_mm->lock); |
50 | while (unlikely(!hlist_empty(&mm->mmu_notifier_mm->list))) { | |
51 | mn = hlist_entry(mm->mmu_notifier_mm->list.first, | |
52 | struct mmu_notifier, | |
53 | hlist); | |
751efd86 | 54 | |
cddb8a5c | 55 | /* |
751efd86 RH |
56 | * Unlink. This will prevent mmu_notifier_unregister() |
57 | * from also making the ->release() callout. | |
cddb8a5c AA |
58 | */ |
59 | hlist_del_init_rcu(&mn->hlist); | |
751efd86 RH |
60 | spin_unlock(&mm->mmu_notifier_mm->lock); |
61 | ||
62 | /* | |
63 | * Clear sptes. (see 'release' description in mmu_notifier.h) | |
64 | */ | |
65 | if (mn->ops->release) | |
66 | mn->ops->release(mn, mm); | |
67 | ||
68 | spin_lock(&mm->mmu_notifier_mm->lock); | |
cddb8a5c AA |
69 | } |
70 | spin_unlock(&mm->mmu_notifier_mm->lock); | |
71 | ||
72 | /* | |
751efd86 RH |
73 | * All callouts to ->release() which we have done are complete. |
74 | * Allow synchronize_srcu() in mmu_notifier_unregister() to complete | |
75 | */ | |
76 | srcu_read_unlock(&srcu, id); | |
77 | ||
78 | /* | |
79 | * mmu_notifier_unregister() may have unlinked a notifier and may | |
80 | * still be calling out to it. Additionally, other notifiers | |
81 | * may have been active via vmtruncate() et. al. Block here | |
82 | * to ensure that all notifier callouts for this mm have been | |
83 | * completed and the sptes are really cleaned up before returning | |
84 | * to exit_mmap(). | |
cddb8a5c | 85 | */ |
21a92735 | 86 | synchronize_srcu(&srcu); |
cddb8a5c AA |
87 | } |
88 | ||
89 | /* | |
90 | * If no young bitflag is supported by the hardware, ->clear_flush_young can | |
91 | * unmap the address and return 1 or 0 depending if the mapping previously | |
92 | * existed or not. | |
93 | */ | |
94 | int __mmu_notifier_clear_flush_young(struct mm_struct *mm, | |
95 | unsigned long address) | |
96 | { | |
97 | struct mmu_notifier *mn; | |
21a92735 | 98 | int young = 0, id; |
cddb8a5c | 99 | |
21a92735 | 100 | id = srcu_read_lock(&srcu); |
b67bfe0d | 101 | hlist_for_each_entry_rcu(mn, &mm->mmu_notifier_mm->list, hlist) { |
cddb8a5c AA |
102 | if (mn->ops->clear_flush_young) |
103 | young |= mn->ops->clear_flush_young(mn, mm, address); | |
104 | } | |
21a92735 | 105 | srcu_read_unlock(&srcu, id); |
cddb8a5c AA |
106 | |
107 | return young; | |
108 | } | |
109 | ||
8ee53820 AA |
110 | int __mmu_notifier_test_young(struct mm_struct *mm, |
111 | unsigned long address) | |
112 | { | |
113 | struct mmu_notifier *mn; | |
21a92735 | 114 | int young = 0, id; |
8ee53820 | 115 | |
21a92735 | 116 | id = srcu_read_lock(&srcu); |
b67bfe0d | 117 | hlist_for_each_entry_rcu(mn, &mm->mmu_notifier_mm->list, hlist) { |
8ee53820 AA |
118 | if (mn->ops->test_young) { |
119 | young = mn->ops->test_young(mn, mm, address); | |
120 | if (young) | |
121 | break; | |
122 | } | |
123 | } | |
21a92735 | 124 | srcu_read_unlock(&srcu, id); |
8ee53820 AA |
125 | |
126 | return young; | |
127 | } | |
128 | ||
828502d3 IE |
129 | void __mmu_notifier_change_pte(struct mm_struct *mm, unsigned long address, |
130 | pte_t pte) | |
131 | { | |
132 | struct mmu_notifier *mn; | |
21a92735 | 133 | int id; |
828502d3 | 134 | |
21a92735 | 135 | id = srcu_read_lock(&srcu); |
b67bfe0d | 136 | hlist_for_each_entry_rcu(mn, &mm->mmu_notifier_mm->list, hlist) { |
828502d3 IE |
137 | if (mn->ops->change_pte) |
138 | mn->ops->change_pte(mn, mm, address, pte); | |
828502d3 | 139 | } |
21a92735 | 140 | srcu_read_unlock(&srcu, id); |
828502d3 IE |
141 | } |
142 | ||
cddb8a5c AA |
143 | void __mmu_notifier_invalidate_page(struct mm_struct *mm, |
144 | unsigned long address) | |
145 | { | |
146 | struct mmu_notifier *mn; | |
21a92735 | 147 | int id; |
cddb8a5c | 148 | |
21a92735 | 149 | id = srcu_read_lock(&srcu); |
b67bfe0d | 150 | hlist_for_each_entry_rcu(mn, &mm->mmu_notifier_mm->list, hlist) { |
cddb8a5c AA |
151 | if (mn->ops->invalidate_page) |
152 | mn->ops->invalidate_page(mn, mm, address); | |
153 | } | |
21a92735 | 154 | srcu_read_unlock(&srcu, id); |
cddb8a5c AA |
155 | } |
156 | ||
157 | void __mmu_notifier_invalidate_range_start(struct mm_struct *mm, | |
158 | unsigned long start, unsigned long end) | |
159 | { | |
160 | struct mmu_notifier *mn; | |
21a92735 | 161 | int id; |
cddb8a5c | 162 | |
21a92735 | 163 | id = srcu_read_lock(&srcu); |
b67bfe0d | 164 | hlist_for_each_entry_rcu(mn, &mm->mmu_notifier_mm->list, hlist) { |
cddb8a5c AA |
165 | if (mn->ops->invalidate_range_start) |
166 | mn->ops->invalidate_range_start(mn, mm, start, end); | |
167 | } | |
21a92735 | 168 | srcu_read_unlock(&srcu, id); |
cddb8a5c | 169 | } |
fa794199 | 170 | EXPORT_SYMBOL_GPL(__mmu_notifier_invalidate_range_start); |
cddb8a5c AA |
171 | |
172 | void __mmu_notifier_invalidate_range_end(struct mm_struct *mm, | |
173 | unsigned long start, unsigned long end) | |
174 | { | |
175 | struct mmu_notifier *mn; | |
21a92735 | 176 | int id; |
cddb8a5c | 177 | |
21a92735 | 178 | id = srcu_read_lock(&srcu); |
b67bfe0d | 179 | hlist_for_each_entry_rcu(mn, &mm->mmu_notifier_mm->list, hlist) { |
cddb8a5c AA |
180 | if (mn->ops->invalidate_range_end) |
181 | mn->ops->invalidate_range_end(mn, mm, start, end); | |
182 | } | |
21a92735 | 183 | srcu_read_unlock(&srcu, id); |
cddb8a5c | 184 | } |
fa794199 | 185 | EXPORT_SYMBOL_GPL(__mmu_notifier_invalidate_range_end); |
cddb8a5c AA |
186 | |
187 | static int do_mmu_notifier_register(struct mmu_notifier *mn, | |
188 | struct mm_struct *mm, | |
189 | int take_mmap_sem) | |
190 | { | |
191 | struct mmu_notifier_mm *mmu_notifier_mm; | |
192 | int ret; | |
193 | ||
194 | BUG_ON(atomic_read(&mm->mm_users) <= 0); | |
195 | ||
21a92735 | 196 | /* |
35cfa2b0 GS |
197 | * Verify that mmu_notifier_init() already run and the global srcu is |
198 | * initialized. | |
199 | */ | |
21a92735 SG |
200 | BUG_ON(!srcu.per_cpu_ref); |
201 | ||
35cfa2b0 GS |
202 | ret = -ENOMEM; |
203 | mmu_notifier_mm = kmalloc(sizeof(struct mmu_notifier_mm), GFP_KERNEL); | |
204 | if (unlikely(!mmu_notifier_mm)) | |
205 | goto out; | |
206 | ||
cddb8a5c AA |
207 | if (take_mmap_sem) |
208 | down_write(&mm->mmap_sem); | |
209 | ret = mm_take_all_locks(mm); | |
210 | if (unlikely(ret)) | |
35cfa2b0 | 211 | goto out_clean; |
cddb8a5c AA |
212 | |
213 | if (!mm_has_notifiers(mm)) { | |
214 | INIT_HLIST_HEAD(&mmu_notifier_mm->list); | |
215 | spin_lock_init(&mmu_notifier_mm->lock); | |
e0f3c3f7 | 216 | |
cddb8a5c | 217 | mm->mmu_notifier_mm = mmu_notifier_mm; |
35cfa2b0 | 218 | mmu_notifier_mm = NULL; |
cddb8a5c AA |
219 | } |
220 | atomic_inc(&mm->mm_count); | |
221 | ||
222 | /* | |
223 | * Serialize the update against mmu_notifier_unregister. A | |
224 | * side note: mmu_notifier_release can't run concurrently with | |
225 | * us because we hold the mm_users pin (either implicitly as | |
226 | * current->mm or explicitly with get_task_mm() or similar). | |
227 | * We can't race against any other mmu notifier method either | |
228 | * thanks to mm_take_all_locks(). | |
229 | */ | |
230 | spin_lock(&mm->mmu_notifier_mm->lock); | |
231 | hlist_add_head(&mn->hlist, &mm->mmu_notifier_mm->list); | |
232 | spin_unlock(&mm->mmu_notifier_mm->lock); | |
233 | ||
234 | mm_drop_all_locks(mm); | |
35cfa2b0 | 235 | out_clean: |
cddb8a5c AA |
236 | if (take_mmap_sem) |
237 | up_write(&mm->mmap_sem); | |
35cfa2b0 GS |
238 | kfree(mmu_notifier_mm); |
239 | out: | |
cddb8a5c AA |
240 | BUG_ON(atomic_read(&mm->mm_users) <= 0); |
241 | return ret; | |
242 | } | |
243 | ||
244 | /* | |
245 | * Must not hold mmap_sem nor any other VM related lock when calling | |
246 | * this registration function. Must also ensure mm_users can't go down | |
247 | * to zero while this runs to avoid races with mmu_notifier_release, | |
248 | * so mm has to be current->mm or the mm should be pinned safely such | |
249 | * as with get_task_mm(). If the mm is not current->mm, the mm_users | |
250 | * pin should be released by calling mmput after mmu_notifier_register | |
251 | * returns. mmu_notifier_unregister must be always called to | |
252 | * unregister the notifier. mm_count is automatically pinned to allow | |
253 | * mmu_notifier_unregister to safely run at any time later, before or | |
254 | * after exit_mmap. ->release will always be called before exit_mmap | |
255 | * frees the pages. | |
256 | */ | |
257 | int mmu_notifier_register(struct mmu_notifier *mn, struct mm_struct *mm) | |
258 | { | |
259 | return do_mmu_notifier_register(mn, mm, 1); | |
260 | } | |
261 | EXPORT_SYMBOL_GPL(mmu_notifier_register); | |
262 | ||
263 | /* | |
264 | * Same as mmu_notifier_register but here the caller must hold the | |
265 | * mmap_sem in write mode. | |
266 | */ | |
267 | int __mmu_notifier_register(struct mmu_notifier *mn, struct mm_struct *mm) | |
268 | { | |
269 | return do_mmu_notifier_register(mn, mm, 0); | |
270 | } | |
271 | EXPORT_SYMBOL_GPL(__mmu_notifier_register); | |
272 | ||
273 | /* this is called after the last mmu_notifier_unregister() returned */ | |
274 | void __mmu_notifier_mm_destroy(struct mm_struct *mm) | |
275 | { | |
276 | BUG_ON(!hlist_empty(&mm->mmu_notifier_mm->list)); | |
277 | kfree(mm->mmu_notifier_mm); | |
278 | mm->mmu_notifier_mm = LIST_POISON1; /* debug */ | |
279 | } | |
280 | ||
281 | /* | |
282 | * This releases the mm_count pin automatically and frees the mm | |
283 | * structure if it was the last user of it. It serializes against | |
21a92735 SG |
284 | * running mmu notifiers with SRCU and against mmu_notifier_unregister |
285 | * with the unregister lock + SRCU. All sptes must be dropped before | |
cddb8a5c AA |
286 | * calling mmu_notifier_unregister. ->release or any other notifier |
287 | * method may be invoked concurrently with mmu_notifier_unregister, | |
288 | * and only after mmu_notifier_unregister returned we're guaranteed | |
289 | * that ->release or any other method can't run anymore. | |
290 | */ | |
291 | void mmu_notifier_unregister(struct mmu_notifier *mn, struct mm_struct *mm) | |
292 | { | |
293 | BUG_ON(atomic_read(&mm->mm_count) <= 0); | |
294 | ||
751efd86 | 295 | spin_lock(&mm->mmu_notifier_mm->lock); |
cddb8a5c | 296 | if (!hlist_unhashed(&mn->hlist)) { |
21a92735 | 297 | int id; |
3ad3d901 | 298 | |
cddb8a5c | 299 | /* |
751efd86 | 300 | * Ensure we synchronize up with __mmu_notifier_release(). |
cddb8a5c | 301 | */ |
751efd86 RH |
302 | id = srcu_read_lock(&srcu); |
303 | ||
304 | hlist_del_rcu(&mn->hlist); | |
305 | spin_unlock(&mm->mmu_notifier_mm->lock); | |
306 | ||
cddb8a5c AA |
307 | if (mn->ops->release) |
308 | mn->ops->release(mn, mm); | |
3ad3d901 | 309 | |
751efd86 RH |
310 | /* |
311 | * Allow __mmu_notifier_release() to complete. | |
312 | */ | |
313 | srcu_read_unlock(&srcu, id); | |
314 | } else | |
cddb8a5c AA |
315 | spin_unlock(&mm->mmu_notifier_mm->lock); |
316 | ||
317 | /* | |
751efd86 RH |
318 | * Wait for any running method to finish, including ->release() if it |
319 | * was run by __mmu_notifier_release() instead of us. | |
cddb8a5c | 320 | */ |
21a92735 | 321 | synchronize_srcu(&srcu); |
cddb8a5c AA |
322 | |
323 | BUG_ON(atomic_read(&mm->mm_count) <= 0); | |
324 | ||
325 | mmdrop(mm); | |
326 | } | |
327 | EXPORT_SYMBOL_GPL(mmu_notifier_unregister); | |
21a92735 SG |
328 | |
329 | static int __init mmu_notifier_init(void) | |
330 | { | |
331 | return init_srcu_struct(&srcu); | |
332 | } | |
333 | ||
334 | module_init(mmu_notifier_init); |