Commit | Line | Data |
---|---|---|
f8af4da3 | 1 | /* |
31dbd01f IE |
2 | * Memory merging support. |
3 | * | |
4 | * This code enables dynamic sharing of identical pages found in different | |
5 | * memory areas, even if they are not shared by fork() | |
6 | * | |
36b2528d | 7 | * Copyright (C) 2008-2009 Red Hat, Inc. |
31dbd01f IE |
8 | * Authors: |
9 | * Izik Eidus | |
10 | * Andrea Arcangeli | |
11 | * Chris Wright | |
36b2528d | 12 | * Hugh Dickins |
31dbd01f IE |
13 | * |
14 | * This work is licensed under the terms of the GNU GPL, version 2. | |
f8af4da3 HD |
15 | */ |
16 | ||
17 | #include <linux/errno.h> | |
31dbd01f IE |
18 | #include <linux/mm.h> |
19 | #include <linux/fs.h> | |
f8af4da3 | 20 | #include <linux/mman.h> |
31dbd01f IE |
21 | #include <linux/sched.h> |
22 | #include <linux/rwsem.h> | |
23 | #include <linux/pagemap.h> | |
24 | #include <linux/rmap.h> | |
25 | #include <linux/spinlock.h> | |
26 | #include <linux/jhash.h> | |
27 | #include <linux/delay.h> | |
28 | #include <linux/kthread.h> | |
29 | #include <linux/wait.h> | |
30 | #include <linux/slab.h> | |
31 | #include <linux/rbtree.h> | |
62b61f61 | 32 | #include <linux/memory.h> |
31dbd01f | 33 | #include <linux/mmu_notifier.h> |
2c6854fd | 34 | #include <linux/swap.h> |
f8af4da3 | 35 | #include <linux/ksm.h> |
d9f8984c | 36 | #include <linux/hash.h> |
878aee7d | 37 | #include <linux/freezer.h> |
f8af4da3 | 38 | |
31dbd01f | 39 | #include <asm/tlbflush.h> |
73848b46 | 40 | #include "internal.h" |
31dbd01f IE |
41 | |
42 | /* | |
43 | * A few notes about the KSM scanning process, | |
44 | * to make it easier to understand the data structures below: | |
45 | * | |
46 | * In order to reduce excessive scanning, KSM sorts the memory pages by their | |
47 | * contents into a data structure that holds pointers to the pages' locations. | |
48 | * | |
49 | * Since the contents of the pages may change at any moment, KSM cannot just | |
50 | * insert the pages into a normal sorted tree and expect it to find anything. | |
51 | * Therefore KSM uses two data structures - the stable and the unstable tree. | |
52 | * | |
53 | * The stable tree holds pointers to all the merged pages (ksm pages), sorted | |
54 | * by their contents. Because each such page is write-protected, searching on | |
55 | * this tree is fully assured to be working (except when pages are unmapped), | |
56 | * and therefore this tree is called the stable tree. | |
57 | * | |
58 | * In addition to the stable tree, KSM uses a second data structure called the | |
59 | * unstable tree: this tree holds pointers to pages which have been found to | |
60 | * be "unchanged for a period of time". The unstable tree sorts these pages | |
61 | * by their contents, but since they are not write-protected, KSM cannot rely | |
62 | * upon the unstable tree to work correctly - the unstable tree is liable to | |
63 | * be corrupted as its contents are modified, and so it is called unstable. | |
64 | * | |
65 | * KSM solves this problem by several techniques: | |
66 | * | |
67 | * 1) The unstable tree is flushed every time KSM completes scanning all | |
68 | * memory areas, and then the tree is rebuilt again from the beginning. | |
69 | * 2) KSM will only insert into the unstable tree, pages whose hash value | |
70 | * has not changed since the previous scan of all memory areas. | |
71 | * 3) The unstable tree is a RedBlack Tree - so its balancing is based on the | |
72 | * colors of the nodes and not on their contents, assuring that even when | |
73 | * the tree gets "corrupted" it won't get out of balance, so scanning time | |
74 | * remains the same (also, searching and inserting nodes in an rbtree uses | |
75 | * the same algorithm, so we have no overhead when we flush and rebuild). | |
76 | * 4) KSM never flushes the stable tree, which means that even if it were to | |
77 | * take 10 attempts to find a page in the unstable tree, once it is found, | |
78 | * it is secured in the stable tree. (When we scan a new page, we first | |
79 | * compare it against the stable tree, and then against the unstable tree.) | |
80 | */ | |
81 | ||
82 | /** | |
83 | * struct mm_slot - ksm information per mm that is being scanned | |
84 | * @link: link to the mm_slots hash list | |
85 | * @mm_list: link into the mm_slots list, rooted in ksm_mm_head | |
6514d511 | 86 | * @rmap_list: head for this mm_slot's singly-linked list of rmap_items |
31dbd01f IE |
87 | * @mm: the mm that this information is valid for |
88 | */ | |
89 | struct mm_slot { | |
90 | struct hlist_node link; | |
91 | struct list_head mm_list; | |
6514d511 | 92 | struct rmap_item *rmap_list; |
31dbd01f IE |
93 | struct mm_struct *mm; |
94 | }; | |
95 | ||
96 | /** | |
97 | * struct ksm_scan - cursor for scanning | |
98 | * @mm_slot: the current mm_slot we are scanning | |
99 | * @address: the next address inside that to be scanned | |
6514d511 | 100 | * @rmap_list: link to the next rmap to be scanned in the rmap_list |
31dbd01f IE |
101 | * @seqnr: count of completed full scans (needed when removing unstable node) |
102 | * | |
103 | * There is only the one ksm_scan instance of this cursor structure. | |
104 | */ | |
105 | struct ksm_scan { | |
106 | struct mm_slot *mm_slot; | |
107 | unsigned long address; | |
6514d511 | 108 | struct rmap_item **rmap_list; |
31dbd01f IE |
109 | unsigned long seqnr; |
110 | }; | |
111 | ||
7b6ba2c7 HD |
112 | /** |
113 | * struct stable_node - node of the stable rbtree | |
114 | * @node: rb node of this ksm page in the stable tree | |
115 | * @hlist: hlist head of rmap_items using this ksm page | |
62b61f61 | 116 | * @kpfn: page frame number of this ksm page |
7b6ba2c7 HD |
117 | */ |
118 | struct stable_node { | |
119 | struct rb_node node; | |
120 | struct hlist_head hlist; | |
62b61f61 | 121 | unsigned long kpfn; |
7b6ba2c7 HD |
122 | }; |
123 | ||
31dbd01f IE |
124 | /** |
125 | * struct rmap_item - reverse mapping item for virtual addresses | |
6514d511 | 126 | * @rmap_list: next rmap_item in mm_slot's singly-linked rmap_list |
db114b83 | 127 | * @anon_vma: pointer to anon_vma for this mm,address, when in stable tree |
31dbd01f IE |
128 | * @mm: the memory structure this rmap_item is pointing into |
129 | * @address: the virtual address this rmap_item tracks (+ flags in low bits) | |
130 | * @oldchecksum: previous checksum of the page at that virtual address | |
7b6ba2c7 HD |
131 | * @node: rb node of this rmap_item in the unstable tree |
132 | * @head: pointer to stable_node heading this list in the stable tree | |
133 | * @hlist: link into hlist of rmap_items hanging off that stable_node | |
31dbd01f IE |
134 | */ |
135 | struct rmap_item { | |
6514d511 | 136 | struct rmap_item *rmap_list; |
db114b83 | 137 | struct anon_vma *anon_vma; /* when stable */ |
31dbd01f IE |
138 | struct mm_struct *mm; |
139 | unsigned long address; /* + low bits used for flags below */ | |
7b6ba2c7 | 140 | unsigned int oldchecksum; /* when unstable */ |
31dbd01f | 141 | union { |
7b6ba2c7 HD |
142 | struct rb_node node; /* when node of unstable tree */ |
143 | struct { /* when listed from stable tree */ | |
144 | struct stable_node *head; | |
145 | struct hlist_node hlist; | |
146 | }; | |
31dbd01f IE |
147 | }; |
148 | }; | |
149 | ||
150 | #define SEQNR_MASK 0x0ff /* low bits of unstable tree seqnr */ | |
7b6ba2c7 HD |
151 | #define UNSTABLE_FLAG 0x100 /* is a node of the unstable tree */ |
152 | #define STABLE_FLAG 0x200 /* is listed from the stable tree */ | |
31dbd01f IE |
153 | |
154 | /* The stable and unstable tree heads */ | |
155 | static struct rb_root root_stable_tree = RB_ROOT; | |
156 | static struct rb_root root_unstable_tree = RB_ROOT; | |
157 | ||
d9f8984c LJ |
158 | #define MM_SLOTS_HASH_SHIFT 10 |
159 | #define MM_SLOTS_HASH_HEADS (1 << MM_SLOTS_HASH_SHIFT) | |
160 | static struct hlist_head mm_slots_hash[MM_SLOTS_HASH_HEADS]; | |
31dbd01f IE |
161 | |
162 | static struct mm_slot ksm_mm_head = { | |
163 | .mm_list = LIST_HEAD_INIT(ksm_mm_head.mm_list), | |
164 | }; | |
165 | static struct ksm_scan ksm_scan = { | |
166 | .mm_slot = &ksm_mm_head, | |
167 | }; | |
168 | ||
169 | static struct kmem_cache *rmap_item_cache; | |
7b6ba2c7 | 170 | static struct kmem_cache *stable_node_cache; |
31dbd01f IE |
171 | static struct kmem_cache *mm_slot_cache; |
172 | ||
173 | /* The number of nodes in the stable tree */ | |
b4028260 | 174 | static unsigned long ksm_pages_shared; |
31dbd01f | 175 | |
e178dfde | 176 | /* The number of page slots additionally sharing those nodes */ |
b4028260 | 177 | static unsigned long ksm_pages_sharing; |
31dbd01f | 178 | |
473b0ce4 HD |
179 | /* The number of nodes in the unstable tree */ |
180 | static unsigned long ksm_pages_unshared; | |
181 | ||
182 | /* The number of rmap_items in use: to calculate pages_volatile */ | |
183 | static unsigned long ksm_rmap_items; | |
184 | ||
31dbd01f | 185 | /* Number of pages ksmd should scan in one batch */ |
2c6854fd | 186 | static unsigned int ksm_thread_pages_to_scan = 100; |
31dbd01f IE |
187 | |
188 | /* Milliseconds ksmd should sleep between batches */ | |
2ffd8679 | 189 | static unsigned int ksm_thread_sleep_millisecs = 20; |
31dbd01f IE |
190 | |
191 | #define KSM_RUN_STOP 0 | |
192 | #define KSM_RUN_MERGE 1 | |
193 | #define KSM_RUN_UNMERGE 2 | |
2c6854fd | 194 | static unsigned int ksm_run = KSM_RUN_STOP; |
31dbd01f IE |
195 | |
196 | static DECLARE_WAIT_QUEUE_HEAD(ksm_thread_wait); | |
197 | static DEFINE_MUTEX(ksm_thread_mutex); | |
198 | static DEFINE_SPINLOCK(ksm_mmlist_lock); | |
199 | ||
200 | #define KSM_KMEM_CACHE(__struct, __flags) kmem_cache_create("ksm_"#__struct,\ | |
201 | sizeof(struct __struct), __alignof__(struct __struct),\ | |
202 | (__flags), NULL) | |
203 | ||
204 | static int __init ksm_slab_init(void) | |
205 | { | |
206 | rmap_item_cache = KSM_KMEM_CACHE(rmap_item, 0); | |
207 | if (!rmap_item_cache) | |
208 | goto out; | |
209 | ||
7b6ba2c7 HD |
210 | stable_node_cache = KSM_KMEM_CACHE(stable_node, 0); |
211 | if (!stable_node_cache) | |
212 | goto out_free1; | |
213 | ||
31dbd01f IE |
214 | mm_slot_cache = KSM_KMEM_CACHE(mm_slot, 0); |
215 | if (!mm_slot_cache) | |
7b6ba2c7 | 216 | goto out_free2; |
31dbd01f IE |
217 | |
218 | return 0; | |
219 | ||
7b6ba2c7 HD |
220 | out_free2: |
221 | kmem_cache_destroy(stable_node_cache); | |
222 | out_free1: | |
31dbd01f IE |
223 | kmem_cache_destroy(rmap_item_cache); |
224 | out: | |
225 | return -ENOMEM; | |
226 | } | |
227 | ||
228 | static void __init ksm_slab_free(void) | |
229 | { | |
230 | kmem_cache_destroy(mm_slot_cache); | |
7b6ba2c7 | 231 | kmem_cache_destroy(stable_node_cache); |
31dbd01f IE |
232 | kmem_cache_destroy(rmap_item_cache); |
233 | mm_slot_cache = NULL; | |
234 | } | |
235 | ||
236 | static inline struct rmap_item *alloc_rmap_item(void) | |
237 | { | |
473b0ce4 HD |
238 | struct rmap_item *rmap_item; |
239 | ||
240 | rmap_item = kmem_cache_zalloc(rmap_item_cache, GFP_KERNEL); | |
241 | if (rmap_item) | |
242 | ksm_rmap_items++; | |
243 | return rmap_item; | |
31dbd01f IE |
244 | } |
245 | ||
246 | static inline void free_rmap_item(struct rmap_item *rmap_item) | |
247 | { | |
473b0ce4 | 248 | ksm_rmap_items--; |
31dbd01f IE |
249 | rmap_item->mm = NULL; /* debug safety */ |
250 | kmem_cache_free(rmap_item_cache, rmap_item); | |
251 | } | |
252 | ||
7b6ba2c7 HD |
253 | static inline struct stable_node *alloc_stable_node(void) |
254 | { | |
255 | return kmem_cache_alloc(stable_node_cache, GFP_KERNEL); | |
256 | } | |
257 | ||
258 | static inline void free_stable_node(struct stable_node *stable_node) | |
259 | { | |
260 | kmem_cache_free(stable_node_cache, stable_node); | |
261 | } | |
262 | ||
31dbd01f IE |
263 | static inline struct mm_slot *alloc_mm_slot(void) |
264 | { | |
265 | if (!mm_slot_cache) /* initialization failed */ | |
266 | return NULL; | |
267 | return kmem_cache_zalloc(mm_slot_cache, GFP_KERNEL); | |
268 | } | |
269 | ||
270 | static inline void free_mm_slot(struct mm_slot *mm_slot) | |
271 | { | |
272 | kmem_cache_free(mm_slot_cache, mm_slot); | |
273 | } | |
274 | ||
31dbd01f IE |
275 | static struct mm_slot *get_mm_slot(struct mm_struct *mm) |
276 | { | |
277 | struct mm_slot *mm_slot; | |
278 | struct hlist_head *bucket; | |
279 | struct hlist_node *node; | |
280 | ||
d9f8984c | 281 | bucket = &mm_slots_hash[hash_ptr(mm, MM_SLOTS_HASH_SHIFT)]; |
31dbd01f IE |
282 | hlist_for_each_entry(mm_slot, node, bucket, link) { |
283 | if (mm == mm_slot->mm) | |
284 | return mm_slot; | |
285 | } | |
286 | return NULL; | |
287 | } | |
288 | ||
289 | static void insert_to_mm_slots_hash(struct mm_struct *mm, | |
290 | struct mm_slot *mm_slot) | |
291 | { | |
292 | struct hlist_head *bucket; | |
293 | ||
d9f8984c | 294 | bucket = &mm_slots_hash[hash_ptr(mm, MM_SLOTS_HASH_SHIFT)]; |
31dbd01f | 295 | mm_slot->mm = mm; |
31dbd01f IE |
296 | hlist_add_head(&mm_slot->link, bucket); |
297 | } | |
298 | ||
299 | static inline int in_stable_tree(struct rmap_item *rmap_item) | |
300 | { | |
301 | return rmap_item->address & STABLE_FLAG; | |
302 | } | |
303 | ||
db114b83 HD |
304 | static void hold_anon_vma(struct rmap_item *rmap_item, |
305 | struct anon_vma *anon_vma) | |
306 | { | |
307 | rmap_item->anon_vma = anon_vma; | |
76545066 | 308 | get_anon_vma(anon_vma); |
db114b83 HD |
309 | } |
310 | ||
76545066 | 311 | static void ksm_drop_anon_vma(struct rmap_item *rmap_item) |
db114b83 HD |
312 | { |
313 | struct anon_vma *anon_vma = rmap_item->anon_vma; | |
314 | ||
76545066 | 315 | drop_anon_vma(anon_vma); |
db114b83 HD |
316 | } |
317 | ||
a913e182 HD |
318 | /* |
319 | * ksmd, and unmerge_and_remove_all_rmap_items(), must not touch an mm's | |
320 | * page tables after it has passed through ksm_exit() - which, if necessary, | |
321 | * takes mmap_sem briefly to serialize against them. ksm_exit() does not set | |
322 | * a special flag: they can just back out as soon as mm_users goes to zero. | |
323 | * ksm_test_exit() is used throughout to make this test for exit: in some | |
324 | * places for correctness, in some places just to avoid unnecessary work. | |
325 | */ | |
326 | static inline bool ksm_test_exit(struct mm_struct *mm) | |
327 | { | |
328 | return atomic_read(&mm->mm_users) == 0; | |
329 | } | |
330 | ||
31dbd01f IE |
331 | /* |
332 | * We use break_ksm to break COW on a ksm page: it's a stripped down | |
333 | * | |
334 | * if (get_user_pages(current, mm, addr, 1, 1, 1, &page, NULL) == 1) | |
335 | * put_page(page); | |
336 | * | |
337 | * but taking great care only to touch a ksm page, in a VM_MERGEABLE vma, | |
338 | * in case the application has unmapped and remapped mm,addr meanwhile. | |
339 | * Could a ksm page appear anywhere else? Actually yes, in a VM_PFNMAP | |
340 | * mmap of /dev/mem or /dev/kmem, where we would not want to touch it. | |
341 | */ | |
d952b791 | 342 | static int break_ksm(struct vm_area_struct *vma, unsigned long addr) |
31dbd01f IE |
343 | { |
344 | struct page *page; | |
d952b791 | 345 | int ret = 0; |
31dbd01f IE |
346 | |
347 | do { | |
348 | cond_resched(); | |
349 | page = follow_page(vma, addr, FOLL_GET); | |
22eccdd7 | 350 | if (IS_ERR_OR_NULL(page)) |
31dbd01f IE |
351 | break; |
352 | if (PageKsm(page)) | |
353 | ret = handle_mm_fault(vma->vm_mm, vma, addr, | |
354 | FAULT_FLAG_WRITE); | |
355 | else | |
356 | ret = VM_FAULT_WRITE; | |
357 | put_page(page); | |
d952b791 HD |
358 | } while (!(ret & (VM_FAULT_WRITE | VM_FAULT_SIGBUS | VM_FAULT_OOM))); |
359 | /* | |
360 | * We must loop because handle_mm_fault() may back out if there's | |
361 | * any difficulty e.g. if pte accessed bit gets updated concurrently. | |
362 | * | |
363 | * VM_FAULT_WRITE is what we have been hoping for: it indicates that | |
364 | * COW has been broken, even if the vma does not permit VM_WRITE; | |
365 | * but note that a concurrent fault might break PageKsm for us. | |
366 | * | |
367 | * VM_FAULT_SIGBUS could occur if we race with truncation of the | |
368 | * backing file, which also invalidates anonymous pages: that's | |
369 | * okay, that truncation will have unmapped the PageKsm for us. | |
370 | * | |
371 | * VM_FAULT_OOM: at the time of writing (late July 2009), setting | |
372 | * aside mem_cgroup limits, VM_FAULT_OOM would only be set if the | |
373 | * current task has TIF_MEMDIE set, and will be OOM killed on return | |
374 | * to user; and ksmd, having no mm, would never be chosen for that. | |
375 | * | |
376 | * But if the mm is in a limited mem_cgroup, then the fault may fail | |
377 | * with VM_FAULT_OOM even if the current task is not TIF_MEMDIE; and | |
378 | * even ksmd can fail in this way - though it's usually breaking ksm | |
379 | * just to undo a merge it made a moment before, so unlikely to oom. | |
380 | * | |
381 | * That's a pity: we might therefore have more kernel pages allocated | |
382 | * than we're counting as nodes in the stable tree; but ksm_do_scan | |
383 | * will retry to break_cow on each pass, so should recover the page | |
384 | * in due course. The important thing is to not let VM_MERGEABLE | |
385 | * be cleared while any such pages might remain in the area. | |
386 | */ | |
387 | return (ret & VM_FAULT_OOM) ? -ENOMEM : 0; | |
31dbd01f IE |
388 | } |
389 | ||
8dd3557a | 390 | static void break_cow(struct rmap_item *rmap_item) |
31dbd01f | 391 | { |
8dd3557a HD |
392 | struct mm_struct *mm = rmap_item->mm; |
393 | unsigned long addr = rmap_item->address; | |
31dbd01f IE |
394 | struct vm_area_struct *vma; |
395 | ||
4035c07a HD |
396 | /* |
397 | * It is not an accident that whenever we want to break COW | |
398 | * to undo, we also need to drop a reference to the anon_vma. | |
399 | */ | |
76545066 | 400 | ksm_drop_anon_vma(rmap_item); |
4035c07a | 401 | |
81464e30 | 402 | down_read(&mm->mmap_sem); |
9ba69294 HD |
403 | if (ksm_test_exit(mm)) |
404 | goto out; | |
31dbd01f IE |
405 | vma = find_vma(mm, addr); |
406 | if (!vma || vma->vm_start > addr) | |
81464e30 | 407 | goto out; |
31dbd01f | 408 | if (!(vma->vm_flags & VM_MERGEABLE) || !vma->anon_vma) |
81464e30 | 409 | goto out; |
31dbd01f | 410 | break_ksm(vma, addr); |
81464e30 | 411 | out: |
31dbd01f IE |
412 | up_read(&mm->mmap_sem); |
413 | } | |
414 | ||
29ad768c AA |
415 | static struct page *page_trans_compound_anon(struct page *page) |
416 | { | |
417 | if (PageTransCompound(page)) { | |
418 | struct page *head; | |
419 | head = compound_head(page); | |
420 | /* | |
421 | * head may be a dangling pointer. | |
422 | * __split_huge_page_refcount clears PageTail | |
423 | * before overwriting first_page, so if | |
424 | * PageTail is still there it means the head | |
425 | * pointer isn't dangling. | |
426 | */ | |
427 | if (head != page) { | |
428 | smp_rmb(); | |
429 | if (!PageTransCompound(page)) | |
430 | return NULL; | |
431 | } | |
432 | if (PageAnon(head)) | |
433 | return head; | |
434 | } | |
435 | return NULL; | |
436 | } | |
437 | ||
31dbd01f IE |
438 | static struct page *get_mergeable_page(struct rmap_item *rmap_item) |
439 | { | |
440 | struct mm_struct *mm = rmap_item->mm; | |
441 | unsigned long addr = rmap_item->address; | |
442 | struct vm_area_struct *vma; | |
443 | struct page *page; | |
444 | ||
445 | down_read(&mm->mmap_sem); | |
9ba69294 HD |
446 | if (ksm_test_exit(mm)) |
447 | goto out; | |
31dbd01f IE |
448 | vma = find_vma(mm, addr); |
449 | if (!vma || vma->vm_start > addr) | |
450 | goto out; | |
451 | if (!(vma->vm_flags & VM_MERGEABLE) || !vma->anon_vma) | |
452 | goto out; | |
453 | ||
454 | page = follow_page(vma, addr, FOLL_GET); | |
22eccdd7 | 455 | if (IS_ERR_OR_NULL(page)) |
31dbd01f | 456 | goto out; |
29ad768c | 457 | if (PageAnon(page) || page_trans_compound_anon(page)) { |
31dbd01f IE |
458 | flush_anon_page(vma, page, addr); |
459 | flush_dcache_page(page); | |
460 | } else { | |
461 | put_page(page); | |
462 | out: page = NULL; | |
463 | } | |
464 | up_read(&mm->mmap_sem); | |
465 | return page; | |
466 | } | |
467 | ||
4035c07a HD |
468 | static void remove_node_from_stable_tree(struct stable_node *stable_node) |
469 | { | |
470 | struct rmap_item *rmap_item; | |
471 | struct hlist_node *hlist; | |
472 | ||
473 | hlist_for_each_entry(rmap_item, hlist, &stable_node->hlist, hlist) { | |
474 | if (rmap_item->hlist.next) | |
475 | ksm_pages_sharing--; | |
476 | else | |
477 | ksm_pages_shared--; | |
76545066 | 478 | ksm_drop_anon_vma(rmap_item); |
4035c07a HD |
479 | rmap_item->address &= PAGE_MASK; |
480 | cond_resched(); | |
481 | } | |
482 | ||
483 | rb_erase(&stable_node->node, &root_stable_tree); | |
484 | free_stable_node(stable_node); | |
485 | } | |
486 | ||
487 | /* | |
488 | * get_ksm_page: checks if the page indicated by the stable node | |
489 | * is still its ksm page, despite having held no reference to it. | |
490 | * In which case we can trust the content of the page, and it | |
491 | * returns the gotten page; but if the page has now been zapped, | |
492 | * remove the stale node from the stable tree and return NULL. | |
493 | * | |
494 | * You would expect the stable_node to hold a reference to the ksm page. | |
495 | * But if it increments the page's count, swapping out has to wait for | |
496 | * ksmd to come around again before it can free the page, which may take | |
497 | * seconds or even minutes: much too unresponsive. So instead we use a | |
498 | * "keyhole reference": access to the ksm page from the stable node peeps | |
499 | * out through its keyhole to see if that page still holds the right key, | |
500 | * pointing back to this stable node. This relies on freeing a PageAnon | |
501 | * page to reset its page->mapping to NULL, and relies on no other use of | |
502 | * a page to put something that might look like our key in page->mapping. | |
503 | * | |
504 | * include/linux/pagemap.h page_cache_get_speculative() is a good reference, | |
505 | * but this is different - made simpler by ksm_thread_mutex being held, but | |
506 | * interesting for assuming that no other use of the struct page could ever | |
507 | * put our expected_mapping into page->mapping (or a field of the union which | |
508 | * coincides with page->mapping). The RCU calls are not for KSM at all, but | |
509 | * to keep the page_count protocol described with page_cache_get_speculative. | |
510 | * | |
511 | * Note: it is possible that get_ksm_page() will return NULL one moment, | |
512 | * then page the next, if the page is in between page_freeze_refs() and | |
513 | * page_unfreeze_refs(): this shouldn't be a problem anywhere, the page | |
514 | * is on its way to being freed; but it is an anomaly to bear in mind. | |
515 | */ | |
516 | static struct page *get_ksm_page(struct stable_node *stable_node) | |
517 | { | |
518 | struct page *page; | |
519 | void *expected_mapping; | |
520 | ||
62b61f61 | 521 | page = pfn_to_page(stable_node->kpfn); |
4035c07a HD |
522 | expected_mapping = (void *)stable_node + |
523 | (PAGE_MAPPING_ANON | PAGE_MAPPING_KSM); | |
524 | rcu_read_lock(); | |
525 | if (page->mapping != expected_mapping) | |
526 | goto stale; | |
527 | if (!get_page_unless_zero(page)) | |
528 | goto stale; | |
529 | if (page->mapping != expected_mapping) { | |
530 | put_page(page); | |
531 | goto stale; | |
532 | } | |
533 | rcu_read_unlock(); | |
534 | return page; | |
535 | stale: | |
536 | rcu_read_unlock(); | |
537 | remove_node_from_stable_tree(stable_node); | |
538 | return NULL; | |
539 | } | |
540 | ||
31dbd01f IE |
541 | /* |
542 | * Removing rmap_item from stable or unstable tree. | |
543 | * This function will clean the information from the stable/unstable tree. | |
544 | */ | |
545 | static void remove_rmap_item_from_tree(struct rmap_item *rmap_item) | |
546 | { | |
7b6ba2c7 HD |
547 | if (rmap_item->address & STABLE_FLAG) { |
548 | struct stable_node *stable_node; | |
5ad64688 | 549 | struct page *page; |
31dbd01f | 550 | |
7b6ba2c7 | 551 | stable_node = rmap_item->head; |
4035c07a HD |
552 | page = get_ksm_page(stable_node); |
553 | if (!page) | |
554 | goto out; | |
5ad64688 | 555 | |
4035c07a | 556 | lock_page(page); |
7b6ba2c7 | 557 | hlist_del(&rmap_item->hlist); |
4035c07a HD |
558 | unlock_page(page); |
559 | put_page(page); | |
08beca44 | 560 | |
4035c07a HD |
561 | if (stable_node->hlist.first) |
562 | ksm_pages_sharing--; | |
563 | else | |
7b6ba2c7 | 564 | ksm_pages_shared--; |
31dbd01f | 565 | |
76545066 | 566 | ksm_drop_anon_vma(rmap_item); |
93d17715 | 567 | rmap_item->address &= PAGE_MASK; |
31dbd01f | 568 | |
7b6ba2c7 | 569 | } else if (rmap_item->address & UNSTABLE_FLAG) { |
31dbd01f IE |
570 | unsigned char age; |
571 | /* | |
9ba69294 | 572 | * Usually ksmd can and must skip the rb_erase, because |
31dbd01f | 573 | * root_unstable_tree was already reset to RB_ROOT. |
9ba69294 HD |
574 | * But be careful when an mm is exiting: do the rb_erase |
575 | * if this rmap_item was inserted by this scan, rather | |
576 | * than left over from before. | |
31dbd01f IE |
577 | */ |
578 | age = (unsigned char)(ksm_scan.seqnr - rmap_item->address); | |
cd551f97 | 579 | BUG_ON(age > 1); |
31dbd01f IE |
580 | if (!age) |
581 | rb_erase(&rmap_item->node, &root_unstable_tree); | |
93d17715 | 582 | |
473b0ce4 | 583 | ksm_pages_unshared--; |
93d17715 | 584 | rmap_item->address &= PAGE_MASK; |
31dbd01f | 585 | } |
4035c07a | 586 | out: |
31dbd01f IE |
587 | cond_resched(); /* we're called from many long loops */ |
588 | } | |
589 | ||
31dbd01f | 590 | static void remove_trailing_rmap_items(struct mm_slot *mm_slot, |
6514d511 | 591 | struct rmap_item **rmap_list) |
31dbd01f | 592 | { |
6514d511 HD |
593 | while (*rmap_list) { |
594 | struct rmap_item *rmap_item = *rmap_list; | |
595 | *rmap_list = rmap_item->rmap_list; | |
31dbd01f | 596 | remove_rmap_item_from_tree(rmap_item); |
31dbd01f IE |
597 | free_rmap_item(rmap_item); |
598 | } | |
599 | } | |
600 | ||
601 | /* | |
602 | * Though it's very tempting to unmerge in_stable_tree(rmap_item)s rather | |
603 | * than check every pte of a given vma, the locking doesn't quite work for | |
604 | * that - an rmap_item is assigned to the stable tree after inserting ksm | |
605 | * page and upping mmap_sem. Nor does it fit with the way we skip dup'ing | |
606 | * rmap_items from parent to child at fork time (so as not to waste time | |
607 | * if exit comes before the next scan reaches it). | |
81464e30 HD |
608 | * |
609 | * Similarly, although we'd like to remove rmap_items (so updating counts | |
610 | * and freeing memory) when unmerging an area, it's easier to leave that | |
611 | * to the next pass of ksmd - consider, for example, how ksmd might be | |
612 | * in cmp_and_merge_page on one of the rmap_items we would be removing. | |
31dbd01f | 613 | */ |
d952b791 HD |
614 | static int unmerge_ksm_pages(struct vm_area_struct *vma, |
615 | unsigned long start, unsigned long end) | |
31dbd01f IE |
616 | { |
617 | unsigned long addr; | |
d952b791 | 618 | int err = 0; |
31dbd01f | 619 | |
d952b791 | 620 | for (addr = start; addr < end && !err; addr += PAGE_SIZE) { |
9ba69294 HD |
621 | if (ksm_test_exit(vma->vm_mm)) |
622 | break; | |
d952b791 HD |
623 | if (signal_pending(current)) |
624 | err = -ERESTARTSYS; | |
625 | else | |
626 | err = break_ksm(vma, addr); | |
627 | } | |
628 | return err; | |
31dbd01f IE |
629 | } |
630 | ||
2ffd8679 HD |
631 | #ifdef CONFIG_SYSFS |
632 | /* | |
633 | * Only called through the sysfs control interface: | |
634 | */ | |
d952b791 | 635 | static int unmerge_and_remove_all_rmap_items(void) |
31dbd01f IE |
636 | { |
637 | struct mm_slot *mm_slot; | |
638 | struct mm_struct *mm; | |
639 | struct vm_area_struct *vma; | |
d952b791 HD |
640 | int err = 0; |
641 | ||
642 | spin_lock(&ksm_mmlist_lock); | |
9ba69294 | 643 | ksm_scan.mm_slot = list_entry(ksm_mm_head.mm_list.next, |
d952b791 HD |
644 | struct mm_slot, mm_list); |
645 | spin_unlock(&ksm_mmlist_lock); | |
31dbd01f | 646 | |
9ba69294 HD |
647 | for (mm_slot = ksm_scan.mm_slot; |
648 | mm_slot != &ksm_mm_head; mm_slot = ksm_scan.mm_slot) { | |
31dbd01f IE |
649 | mm = mm_slot->mm; |
650 | down_read(&mm->mmap_sem); | |
651 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
9ba69294 HD |
652 | if (ksm_test_exit(mm)) |
653 | break; | |
31dbd01f IE |
654 | if (!(vma->vm_flags & VM_MERGEABLE) || !vma->anon_vma) |
655 | continue; | |
d952b791 HD |
656 | err = unmerge_ksm_pages(vma, |
657 | vma->vm_start, vma->vm_end); | |
9ba69294 HD |
658 | if (err) |
659 | goto error; | |
31dbd01f | 660 | } |
9ba69294 | 661 | |
6514d511 | 662 | remove_trailing_rmap_items(mm_slot, &mm_slot->rmap_list); |
d952b791 HD |
663 | |
664 | spin_lock(&ksm_mmlist_lock); | |
9ba69294 | 665 | ksm_scan.mm_slot = list_entry(mm_slot->mm_list.next, |
d952b791 | 666 | struct mm_slot, mm_list); |
9ba69294 HD |
667 | if (ksm_test_exit(mm)) { |
668 | hlist_del(&mm_slot->link); | |
669 | list_del(&mm_slot->mm_list); | |
670 | spin_unlock(&ksm_mmlist_lock); | |
671 | ||
672 | free_mm_slot(mm_slot); | |
673 | clear_bit(MMF_VM_MERGEABLE, &mm->flags); | |
674 | up_read(&mm->mmap_sem); | |
675 | mmdrop(mm); | |
676 | } else { | |
677 | spin_unlock(&ksm_mmlist_lock); | |
678 | up_read(&mm->mmap_sem); | |
679 | } | |
31dbd01f IE |
680 | } |
681 | ||
d952b791 | 682 | ksm_scan.seqnr = 0; |
9ba69294 HD |
683 | return 0; |
684 | ||
685 | error: | |
686 | up_read(&mm->mmap_sem); | |
31dbd01f | 687 | spin_lock(&ksm_mmlist_lock); |
d952b791 | 688 | ksm_scan.mm_slot = &ksm_mm_head; |
31dbd01f | 689 | spin_unlock(&ksm_mmlist_lock); |
d952b791 | 690 | return err; |
31dbd01f | 691 | } |
2ffd8679 | 692 | #endif /* CONFIG_SYSFS */ |
31dbd01f | 693 | |
31dbd01f IE |
694 | static u32 calc_checksum(struct page *page) |
695 | { | |
696 | u32 checksum; | |
697 | void *addr = kmap_atomic(page, KM_USER0); | |
698 | checksum = jhash2(addr, PAGE_SIZE / 4, 17); | |
699 | kunmap_atomic(addr, KM_USER0); | |
700 | return checksum; | |
701 | } | |
702 | ||
703 | static int memcmp_pages(struct page *page1, struct page *page2) | |
704 | { | |
705 | char *addr1, *addr2; | |
706 | int ret; | |
707 | ||
708 | addr1 = kmap_atomic(page1, KM_USER0); | |
709 | addr2 = kmap_atomic(page2, KM_USER1); | |
710 | ret = memcmp(addr1, addr2, PAGE_SIZE); | |
711 | kunmap_atomic(addr2, KM_USER1); | |
712 | kunmap_atomic(addr1, KM_USER0); | |
713 | return ret; | |
714 | } | |
715 | ||
716 | static inline int pages_identical(struct page *page1, struct page *page2) | |
717 | { | |
718 | return !memcmp_pages(page1, page2); | |
719 | } | |
720 | ||
721 | static int write_protect_page(struct vm_area_struct *vma, struct page *page, | |
722 | pte_t *orig_pte) | |
723 | { | |
724 | struct mm_struct *mm = vma->vm_mm; | |
725 | unsigned long addr; | |
726 | pte_t *ptep; | |
727 | spinlock_t *ptl; | |
728 | int swapped; | |
729 | int err = -EFAULT; | |
730 | ||
731 | addr = page_address_in_vma(page, vma); | |
732 | if (addr == -EFAULT) | |
733 | goto out; | |
734 | ||
29ad768c | 735 | BUG_ON(PageTransCompound(page)); |
31dbd01f IE |
736 | ptep = page_check_address(page, mm, addr, &ptl, 0); |
737 | if (!ptep) | |
738 | goto out; | |
739 | ||
4e31635c | 740 | if (pte_write(*ptep) || pte_dirty(*ptep)) { |
31dbd01f IE |
741 | pte_t entry; |
742 | ||
743 | swapped = PageSwapCache(page); | |
744 | flush_cache_page(vma, addr, page_to_pfn(page)); | |
745 | /* | |
746 | * Ok this is tricky, when get_user_pages_fast() run it doesnt | |
747 | * take any lock, therefore the check that we are going to make | |
748 | * with the pagecount against the mapcount is racey and | |
749 | * O_DIRECT can happen right after the check. | |
750 | * So we clear the pte and flush the tlb before the check | |
751 | * this assure us that no O_DIRECT can happen after the check | |
752 | * or in the middle of the check. | |
753 | */ | |
754 | entry = ptep_clear_flush(vma, addr, ptep); | |
755 | /* | |
756 | * Check that no O_DIRECT or similar I/O is in progress on the | |
757 | * page | |
758 | */ | |
31e855ea | 759 | if (page_mapcount(page) + 1 + swapped != page_count(page)) { |
cb532375 | 760 | set_pte_at(mm, addr, ptep, entry); |
31dbd01f IE |
761 | goto out_unlock; |
762 | } | |
4e31635c HD |
763 | if (pte_dirty(entry)) |
764 | set_page_dirty(page); | |
765 | entry = pte_mkclean(pte_wrprotect(entry)); | |
31dbd01f IE |
766 | set_pte_at_notify(mm, addr, ptep, entry); |
767 | } | |
768 | *orig_pte = *ptep; | |
769 | err = 0; | |
770 | ||
771 | out_unlock: | |
772 | pte_unmap_unlock(ptep, ptl); | |
773 | out: | |
774 | return err; | |
775 | } | |
776 | ||
777 | /** | |
778 | * replace_page - replace page in vma by new ksm page | |
8dd3557a HD |
779 | * @vma: vma that holds the pte pointing to page |
780 | * @page: the page we are replacing by kpage | |
781 | * @kpage: the ksm page we replace page by | |
31dbd01f IE |
782 | * @orig_pte: the original value of the pte |
783 | * | |
784 | * Returns 0 on success, -EFAULT on failure. | |
785 | */ | |
8dd3557a HD |
786 | static int replace_page(struct vm_area_struct *vma, struct page *page, |
787 | struct page *kpage, pte_t orig_pte) | |
31dbd01f IE |
788 | { |
789 | struct mm_struct *mm = vma->vm_mm; | |
790 | pgd_t *pgd; | |
791 | pud_t *pud; | |
792 | pmd_t *pmd; | |
793 | pte_t *ptep; | |
794 | spinlock_t *ptl; | |
795 | unsigned long addr; | |
31dbd01f IE |
796 | int err = -EFAULT; |
797 | ||
8dd3557a | 798 | addr = page_address_in_vma(page, vma); |
31dbd01f IE |
799 | if (addr == -EFAULT) |
800 | goto out; | |
801 | ||
802 | pgd = pgd_offset(mm, addr); | |
803 | if (!pgd_present(*pgd)) | |
804 | goto out; | |
805 | ||
806 | pud = pud_offset(pgd, addr); | |
807 | if (!pud_present(*pud)) | |
808 | goto out; | |
809 | ||
810 | pmd = pmd_offset(pud, addr); | |
29ad768c | 811 | BUG_ON(pmd_trans_huge(*pmd)); |
31dbd01f IE |
812 | if (!pmd_present(*pmd)) |
813 | goto out; | |
814 | ||
815 | ptep = pte_offset_map_lock(mm, pmd, addr, &ptl); | |
816 | if (!pte_same(*ptep, orig_pte)) { | |
817 | pte_unmap_unlock(ptep, ptl); | |
818 | goto out; | |
819 | } | |
820 | ||
8dd3557a | 821 | get_page(kpage); |
5ad64688 | 822 | page_add_anon_rmap(kpage, vma, addr); |
31dbd01f IE |
823 | |
824 | flush_cache_page(vma, addr, pte_pfn(*ptep)); | |
825 | ptep_clear_flush(vma, addr, ptep); | |
8dd3557a | 826 | set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot)); |
31dbd01f | 827 | |
8dd3557a | 828 | page_remove_rmap(page); |
ae52a2ad HD |
829 | if (!page_mapped(page)) |
830 | try_to_free_swap(page); | |
8dd3557a | 831 | put_page(page); |
31dbd01f IE |
832 | |
833 | pte_unmap_unlock(ptep, ptl); | |
834 | err = 0; | |
835 | out: | |
836 | return err; | |
837 | } | |
838 | ||
29ad768c AA |
839 | static int page_trans_compound_anon_split(struct page *page) |
840 | { | |
841 | int ret = 0; | |
842 | struct page *transhuge_head = page_trans_compound_anon(page); | |
843 | if (transhuge_head) { | |
844 | /* Get the reference on the head to split it. */ | |
845 | if (get_page_unless_zero(transhuge_head)) { | |
846 | /* | |
847 | * Recheck we got the reference while the head | |
848 | * was still anonymous. | |
849 | */ | |
850 | if (PageAnon(transhuge_head)) | |
851 | ret = split_huge_page(transhuge_head); | |
852 | else | |
853 | /* | |
854 | * Retry later if split_huge_page run | |
855 | * from under us. | |
856 | */ | |
857 | ret = 1; | |
858 | put_page(transhuge_head); | |
859 | } else | |
860 | /* Retry later if split_huge_page run from under us. */ | |
861 | ret = 1; | |
862 | } | |
863 | return ret; | |
864 | } | |
865 | ||
31dbd01f IE |
866 | /* |
867 | * try_to_merge_one_page - take two pages and merge them into one | |
8dd3557a HD |
868 | * @vma: the vma that holds the pte pointing to page |
869 | * @page: the PageAnon page that we want to replace with kpage | |
80e14822 HD |
870 | * @kpage: the PageKsm page that we want to map instead of page, |
871 | * or NULL the first time when we want to use page as kpage. | |
31dbd01f IE |
872 | * |
873 | * This function returns 0 if the pages were merged, -EFAULT otherwise. | |
874 | */ | |
875 | static int try_to_merge_one_page(struct vm_area_struct *vma, | |
8dd3557a | 876 | struct page *page, struct page *kpage) |
31dbd01f IE |
877 | { |
878 | pte_t orig_pte = __pte(0); | |
879 | int err = -EFAULT; | |
880 | ||
db114b83 HD |
881 | if (page == kpage) /* ksm page forked */ |
882 | return 0; | |
883 | ||
31dbd01f IE |
884 | if (!(vma->vm_flags & VM_MERGEABLE)) |
885 | goto out; | |
29ad768c AA |
886 | if (PageTransCompound(page) && page_trans_compound_anon_split(page)) |
887 | goto out; | |
888 | BUG_ON(PageTransCompound(page)); | |
8dd3557a | 889 | if (!PageAnon(page)) |
31dbd01f IE |
890 | goto out; |
891 | ||
31dbd01f IE |
892 | /* |
893 | * We need the page lock to read a stable PageSwapCache in | |
894 | * write_protect_page(). We use trylock_page() instead of | |
895 | * lock_page() because we don't want to wait here - we | |
896 | * prefer to continue scanning and merging different pages, | |
897 | * then come back to this page when it is unlocked. | |
898 | */ | |
8dd3557a | 899 | if (!trylock_page(page)) |
31e855ea | 900 | goto out; |
31dbd01f IE |
901 | /* |
902 | * If this anonymous page is mapped only here, its pte may need | |
903 | * to be write-protected. If it's mapped elsewhere, all of its | |
904 | * ptes are necessarily already write-protected. But in either | |
905 | * case, we need to lock and check page_count is not raised. | |
906 | */ | |
80e14822 HD |
907 | if (write_protect_page(vma, page, &orig_pte) == 0) { |
908 | if (!kpage) { | |
909 | /* | |
910 | * While we hold page lock, upgrade page from | |
911 | * PageAnon+anon_vma to PageKsm+NULL stable_node: | |
912 | * stable_tree_insert() will update stable_node. | |
913 | */ | |
914 | set_page_stable_node(page, NULL); | |
915 | mark_page_accessed(page); | |
916 | err = 0; | |
917 | } else if (pages_identical(page, kpage)) | |
918 | err = replace_page(vma, page, kpage, orig_pte); | |
919 | } | |
31dbd01f | 920 | |
80e14822 | 921 | if ((vma->vm_flags & VM_LOCKED) && kpage && !err) { |
73848b46 | 922 | munlock_vma_page(page); |
5ad64688 HD |
923 | if (!PageMlocked(kpage)) { |
924 | unlock_page(page); | |
5ad64688 HD |
925 | lock_page(kpage); |
926 | mlock_vma_page(kpage); | |
927 | page = kpage; /* for final unlock */ | |
928 | } | |
929 | } | |
73848b46 | 930 | |
8dd3557a | 931 | unlock_page(page); |
31dbd01f IE |
932 | out: |
933 | return err; | |
934 | } | |
935 | ||
81464e30 HD |
936 | /* |
937 | * try_to_merge_with_ksm_page - like try_to_merge_two_pages, | |
938 | * but no new kernel page is allocated: kpage must already be a ksm page. | |
8dd3557a HD |
939 | * |
940 | * This function returns 0 if the pages were merged, -EFAULT otherwise. | |
81464e30 | 941 | */ |
8dd3557a HD |
942 | static int try_to_merge_with_ksm_page(struct rmap_item *rmap_item, |
943 | struct page *page, struct page *kpage) | |
81464e30 | 944 | { |
8dd3557a | 945 | struct mm_struct *mm = rmap_item->mm; |
81464e30 HD |
946 | struct vm_area_struct *vma; |
947 | int err = -EFAULT; | |
948 | ||
8dd3557a HD |
949 | down_read(&mm->mmap_sem); |
950 | if (ksm_test_exit(mm)) | |
9ba69294 | 951 | goto out; |
8dd3557a HD |
952 | vma = find_vma(mm, rmap_item->address); |
953 | if (!vma || vma->vm_start > rmap_item->address) | |
81464e30 HD |
954 | goto out; |
955 | ||
8dd3557a | 956 | err = try_to_merge_one_page(vma, page, kpage); |
db114b83 HD |
957 | if (err) |
958 | goto out; | |
959 | ||
960 | /* Must get reference to anon_vma while still holding mmap_sem */ | |
961 | hold_anon_vma(rmap_item, vma->anon_vma); | |
81464e30 | 962 | out: |
8dd3557a | 963 | up_read(&mm->mmap_sem); |
81464e30 HD |
964 | return err; |
965 | } | |
966 | ||
31dbd01f IE |
967 | /* |
968 | * try_to_merge_two_pages - take two identical pages and prepare them | |
969 | * to be merged into one page. | |
970 | * | |
8dd3557a HD |
971 | * This function returns the kpage if we successfully merged two identical |
972 | * pages into one ksm page, NULL otherwise. | |
31dbd01f | 973 | * |
80e14822 | 974 | * Note that this function upgrades page to ksm page: if one of the pages |
31dbd01f IE |
975 | * is already a ksm page, try_to_merge_with_ksm_page should be used. |
976 | */ | |
8dd3557a HD |
977 | static struct page *try_to_merge_two_pages(struct rmap_item *rmap_item, |
978 | struct page *page, | |
979 | struct rmap_item *tree_rmap_item, | |
980 | struct page *tree_page) | |
31dbd01f | 981 | { |
80e14822 | 982 | int err; |
31dbd01f | 983 | |
80e14822 | 984 | err = try_to_merge_with_ksm_page(rmap_item, page, NULL); |
31dbd01f | 985 | if (!err) { |
8dd3557a | 986 | err = try_to_merge_with_ksm_page(tree_rmap_item, |
80e14822 | 987 | tree_page, page); |
31dbd01f | 988 | /* |
81464e30 HD |
989 | * If that fails, we have a ksm page with only one pte |
990 | * pointing to it: so break it. | |
31dbd01f | 991 | */ |
4035c07a | 992 | if (err) |
8dd3557a | 993 | break_cow(rmap_item); |
31dbd01f | 994 | } |
80e14822 | 995 | return err ? NULL : page; |
31dbd01f IE |
996 | } |
997 | ||
31dbd01f | 998 | /* |
8dd3557a | 999 | * stable_tree_search - search for page inside the stable tree |
31dbd01f IE |
1000 | * |
1001 | * This function checks if there is a page inside the stable tree | |
1002 | * with identical content to the page that we are scanning right now. | |
1003 | * | |
7b6ba2c7 | 1004 | * This function returns the stable tree node of identical content if found, |
31dbd01f IE |
1005 | * NULL otherwise. |
1006 | */ | |
62b61f61 | 1007 | static struct page *stable_tree_search(struct page *page) |
31dbd01f IE |
1008 | { |
1009 | struct rb_node *node = root_stable_tree.rb_node; | |
7b6ba2c7 | 1010 | struct stable_node *stable_node; |
31dbd01f | 1011 | |
08beca44 HD |
1012 | stable_node = page_stable_node(page); |
1013 | if (stable_node) { /* ksm page forked */ | |
1014 | get_page(page); | |
62b61f61 | 1015 | return page; |
08beca44 HD |
1016 | } |
1017 | ||
31dbd01f | 1018 | while (node) { |
4035c07a | 1019 | struct page *tree_page; |
31dbd01f IE |
1020 | int ret; |
1021 | ||
08beca44 | 1022 | cond_resched(); |
7b6ba2c7 | 1023 | stable_node = rb_entry(node, struct stable_node, node); |
4035c07a HD |
1024 | tree_page = get_ksm_page(stable_node); |
1025 | if (!tree_page) | |
1026 | return NULL; | |
31dbd01f | 1027 | |
4035c07a | 1028 | ret = memcmp_pages(page, tree_page); |
31dbd01f | 1029 | |
4035c07a HD |
1030 | if (ret < 0) { |
1031 | put_page(tree_page); | |
31dbd01f | 1032 | node = node->rb_left; |
4035c07a HD |
1033 | } else if (ret > 0) { |
1034 | put_page(tree_page); | |
31dbd01f | 1035 | node = node->rb_right; |
4035c07a | 1036 | } else |
62b61f61 | 1037 | return tree_page; |
31dbd01f IE |
1038 | } |
1039 | ||
1040 | return NULL; | |
1041 | } | |
1042 | ||
1043 | /* | |
1044 | * stable_tree_insert - insert rmap_item pointing to new ksm page | |
1045 | * into the stable tree. | |
1046 | * | |
7b6ba2c7 HD |
1047 | * This function returns the stable tree node just allocated on success, |
1048 | * NULL otherwise. | |
31dbd01f | 1049 | */ |
7b6ba2c7 | 1050 | static struct stable_node *stable_tree_insert(struct page *kpage) |
31dbd01f IE |
1051 | { |
1052 | struct rb_node **new = &root_stable_tree.rb_node; | |
1053 | struct rb_node *parent = NULL; | |
7b6ba2c7 | 1054 | struct stable_node *stable_node; |
31dbd01f IE |
1055 | |
1056 | while (*new) { | |
4035c07a | 1057 | struct page *tree_page; |
31dbd01f IE |
1058 | int ret; |
1059 | ||
08beca44 | 1060 | cond_resched(); |
7b6ba2c7 | 1061 | stable_node = rb_entry(*new, struct stable_node, node); |
4035c07a HD |
1062 | tree_page = get_ksm_page(stable_node); |
1063 | if (!tree_page) | |
1064 | return NULL; | |
31dbd01f | 1065 | |
4035c07a HD |
1066 | ret = memcmp_pages(kpage, tree_page); |
1067 | put_page(tree_page); | |
31dbd01f IE |
1068 | |
1069 | parent = *new; | |
1070 | if (ret < 0) | |
1071 | new = &parent->rb_left; | |
1072 | else if (ret > 0) | |
1073 | new = &parent->rb_right; | |
1074 | else { | |
1075 | /* | |
1076 | * It is not a bug that stable_tree_search() didn't | |
1077 | * find this node: because at that time our page was | |
1078 | * not yet write-protected, so may have changed since. | |
1079 | */ | |
1080 | return NULL; | |
1081 | } | |
1082 | } | |
1083 | ||
7b6ba2c7 HD |
1084 | stable_node = alloc_stable_node(); |
1085 | if (!stable_node) | |
1086 | return NULL; | |
31dbd01f | 1087 | |
7b6ba2c7 HD |
1088 | rb_link_node(&stable_node->node, parent, new); |
1089 | rb_insert_color(&stable_node->node, &root_stable_tree); | |
1090 | ||
1091 | INIT_HLIST_HEAD(&stable_node->hlist); | |
1092 | ||
62b61f61 | 1093 | stable_node->kpfn = page_to_pfn(kpage); |
08beca44 HD |
1094 | set_page_stable_node(kpage, stable_node); |
1095 | ||
7b6ba2c7 | 1096 | return stable_node; |
31dbd01f IE |
1097 | } |
1098 | ||
1099 | /* | |
8dd3557a HD |
1100 | * unstable_tree_search_insert - search for identical page, |
1101 | * else insert rmap_item into the unstable tree. | |
31dbd01f IE |
1102 | * |
1103 | * This function searches for a page in the unstable tree identical to the | |
1104 | * page currently being scanned; and if no identical page is found in the | |
1105 | * tree, we insert rmap_item as a new object into the unstable tree. | |
1106 | * | |
1107 | * This function returns pointer to rmap_item found to be identical | |
1108 | * to the currently scanned page, NULL otherwise. | |
1109 | * | |
1110 | * This function does both searching and inserting, because they share | |
1111 | * the same walking algorithm in an rbtree. | |
1112 | */ | |
8dd3557a HD |
1113 | static |
1114 | struct rmap_item *unstable_tree_search_insert(struct rmap_item *rmap_item, | |
1115 | struct page *page, | |
1116 | struct page **tree_pagep) | |
1117 | ||
31dbd01f IE |
1118 | { |
1119 | struct rb_node **new = &root_unstable_tree.rb_node; | |
1120 | struct rb_node *parent = NULL; | |
1121 | ||
1122 | while (*new) { | |
1123 | struct rmap_item *tree_rmap_item; | |
8dd3557a | 1124 | struct page *tree_page; |
31dbd01f IE |
1125 | int ret; |
1126 | ||
d178f27f | 1127 | cond_resched(); |
31dbd01f | 1128 | tree_rmap_item = rb_entry(*new, struct rmap_item, node); |
8dd3557a | 1129 | tree_page = get_mergeable_page(tree_rmap_item); |
22eccdd7 | 1130 | if (IS_ERR_OR_NULL(tree_page)) |
31dbd01f IE |
1131 | return NULL; |
1132 | ||
1133 | /* | |
8dd3557a | 1134 | * Don't substitute a ksm page for a forked page. |
31dbd01f | 1135 | */ |
8dd3557a HD |
1136 | if (page == tree_page) { |
1137 | put_page(tree_page); | |
31dbd01f IE |
1138 | return NULL; |
1139 | } | |
1140 | ||
8dd3557a | 1141 | ret = memcmp_pages(page, tree_page); |
31dbd01f IE |
1142 | |
1143 | parent = *new; | |
1144 | if (ret < 0) { | |
8dd3557a | 1145 | put_page(tree_page); |
31dbd01f IE |
1146 | new = &parent->rb_left; |
1147 | } else if (ret > 0) { | |
8dd3557a | 1148 | put_page(tree_page); |
31dbd01f IE |
1149 | new = &parent->rb_right; |
1150 | } else { | |
8dd3557a | 1151 | *tree_pagep = tree_page; |
31dbd01f IE |
1152 | return tree_rmap_item; |
1153 | } | |
1154 | } | |
1155 | ||
7b6ba2c7 | 1156 | rmap_item->address |= UNSTABLE_FLAG; |
31dbd01f IE |
1157 | rmap_item->address |= (ksm_scan.seqnr & SEQNR_MASK); |
1158 | rb_link_node(&rmap_item->node, parent, new); | |
1159 | rb_insert_color(&rmap_item->node, &root_unstable_tree); | |
1160 | ||
473b0ce4 | 1161 | ksm_pages_unshared++; |
31dbd01f IE |
1162 | return NULL; |
1163 | } | |
1164 | ||
1165 | /* | |
1166 | * stable_tree_append - add another rmap_item to the linked list of | |
1167 | * rmap_items hanging off a given node of the stable tree, all sharing | |
1168 | * the same ksm page. | |
1169 | */ | |
1170 | static void stable_tree_append(struct rmap_item *rmap_item, | |
7b6ba2c7 | 1171 | struct stable_node *stable_node) |
31dbd01f | 1172 | { |
7b6ba2c7 | 1173 | rmap_item->head = stable_node; |
31dbd01f | 1174 | rmap_item->address |= STABLE_FLAG; |
7b6ba2c7 | 1175 | hlist_add_head(&rmap_item->hlist, &stable_node->hlist); |
e178dfde | 1176 | |
7b6ba2c7 HD |
1177 | if (rmap_item->hlist.next) |
1178 | ksm_pages_sharing++; | |
1179 | else | |
1180 | ksm_pages_shared++; | |
31dbd01f IE |
1181 | } |
1182 | ||
1183 | /* | |
81464e30 HD |
1184 | * cmp_and_merge_page - first see if page can be merged into the stable tree; |
1185 | * if not, compare checksum to previous and if it's the same, see if page can | |
1186 | * be inserted into the unstable tree, or merged with a page already there and | |
1187 | * both transferred to the stable tree. | |
31dbd01f IE |
1188 | * |
1189 | * @page: the page that we are searching identical page to. | |
1190 | * @rmap_item: the reverse mapping into the virtual address of this page | |
1191 | */ | |
1192 | static void cmp_and_merge_page(struct page *page, struct rmap_item *rmap_item) | |
1193 | { | |
31dbd01f | 1194 | struct rmap_item *tree_rmap_item; |
8dd3557a | 1195 | struct page *tree_page = NULL; |
7b6ba2c7 | 1196 | struct stable_node *stable_node; |
8dd3557a | 1197 | struct page *kpage; |
31dbd01f IE |
1198 | unsigned int checksum; |
1199 | int err; | |
1200 | ||
93d17715 | 1201 | remove_rmap_item_from_tree(rmap_item); |
31dbd01f IE |
1202 | |
1203 | /* We first start with searching the page inside the stable tree */ | |
62b61f61 HD |
1204 | kpage = stable_tree_search(page); |
1205 | if (kpage) { | |
08beca44 | 1206 | err = try_to_merge_with_ksm_page(rmap_item, page, kpage); |
31dbd01f IE |
1207 | if (!err) { |
1208 | /* | |
1209 | * The page was successfully merged: | |
1210 | * add its rmap_item to the stable tree. | |
1211 | */ | |
5ad64688 | 1212 | lock_page(kpage); |
62b61f61 | 1213 | stable_tree_append(rmap_item, page_stable_node(kpage)); |
5ad64688 | 1214 | unlock_page(kpage); |
31dbd01f | 1215 | } |
8dd3557a | 1216 | put_page(kpage); |
31dbd01f IE |
1217 | return; |
1218 | } | |
1219 | ||
1220 | /* | |
4035c07a HD |
1221 | * If the hash value of the page has changed from the last time |
1222 | * we calculated it, this page is changing frequently: therefore we | |
1223 | * don't want to insert it in the unstable tree, and we don't want | |
1224 | * to waste our time searching for something identical to it there. | |
31dbd01f IE |
1225 | */ |
1226 | checksum = calc_checksum(page); | |
1227 | if (rmap_item->oldchecksum != checksum) { | |
1228 | rmap_item->oldchecksum = checksum; | |
1229 | return; | |
1230 | } | |
1231 | ||
8dd3557a HD |
1232 | tree_rmap_item = |
1233 | unstable_tree_search_insert(rmap_item, page, &tree_page); | |
31dbd01f | 1234 | if (tree_rmap_item) { |
8dd3557a HD |
1235 | kpage = try_to_merge_two_pages(rmap_item, page, |
1236 | tree_rmap_item, tree_page); | |
1237 | put_page(tree_page); | |
31dbd01f IE |
1238 | /* |
1239 | * As soon as we merge this page, we want to remove the | |
1240 | * rmap_item of the page we have merged with from the unstable | |
1241 | * tree, and insert it instead as new node in the stable tree. | |
1242 | */ | |
8dd3557a | 1243 | if (kpage) { |
93d17715 | 1244 | remove_rmap_item_from_tree(tree_rmap_item); |
473b0ce4 | 1245 | |
5ad64688 | 1246 | lock_page(kpage); |
7b6ba2c7 HD |
1247 | stable_node = stable_tree_insert(kpage); |
1248 | if (stable_node) { | |
1249 | stable_tree_append(tree_rmap_item, stable_node); | |
1250 | stable_tree_append(rmap_item, stable_node); | |
1251 | } | |
5ad64688 | 1252 | unlock_page(kpage); |
7b6ba2c7 | 1253 | |
31dbd01f IE |
1254 | /* |
1255 | * If we fail to insert the page into the stable tree, | |
1256 | * we will have 2 virtual addresses that are pointing | |
1257 | * to a ksm page left outside the stable tree, | |
1258 | * in which case we need to break_cow on both. | |
1259 | */ | |
7b6ba2c7 | 1260 | if (!stable_node) { |
8dd3557a HD |
1261 | break_cow(tree_rmap_item); |
1262 | break_cow(rmap_item); | |
31dbd01f IE |
1263 | } |
1264 | } | |
31dbd01f IE |
1265 | } |
1266 | } | |
1267 | ||
1268 | static struct rmap_item *get_next_rmap_item(struct mm_slot *mm_slot, | |
6514d511 | 1269 | struct rmap_item **rmap_list, |
31dbd01f IE |
1270 | unsigned long addr) |
1271 | { | |
1272 | struct rmap_item *rmap_item; | |
1273 | ||
6514d511 HD |
1274 | while (*rmap_list) { |
1275 | rmap_item = *rmap_list; | |
93d17715 | 1276 | if ((rmap_item->address & PAGE_MASK) == addr) |
31dbd01f | 1277 | return rmap_item; |
31dbd01f IE |
1278 | if (rmap_item->address > addr) |
1279 | break; | |
6514d511 | 1280 | *rmap_list = rmap_item->rmap_list; |
31dbd01f | 1281 | remove_rmap_item_from_tree(rmap_item); |
31dbd01f IE |
1282 | free_rmap_item(rmap_item); |
1283 | } | |
1284 | ||
1285 | rmap_item = alloc_rmap_item(); | |
1286 | if (rmap_item) { | |
1287 | /* It has already been zeroed */ | |
1288 | rmap_item->mm = mm_slot->mm; | |
1289 | rmap_item->address = addr; | |
6514d511 HD |
1290 | rmap_item->rmap_list = *rmap_list; |
1291 | *rmap_list = rmap_item; | |
31dbd01f IE |
1292 | } |
1293 | return rmap_item; | |
1294 | } | |
1295 | ||
1296 | static struct rmap_item *scan_get_next_rmap_item(struct page **page) | |
1297 | { | |
1298 | struct mm_struct *mm; | |
1299 | struct mm_slot *slot; | |
1300 | struct vm_area_struct *vma; | |
1301 | struct rmap_item *rmap_item; | |
1302 | ||
1303 | if (list_empty(&ksm_mm_head.mm_list)) | |
1304 | return NULL; | |
1305 | ||
1306 | slot = ksm_scan.mm_slot; | |
1307 | if (slot == &ksm_mm_head) { | |
1308 | root_unstable_tree = RB_ROOT; | |
1309 | ||
1310 | spin_lock(&ksm_mmlist_lock); | |
1311 | slot = list_entry(slot->mm_list.next, struct mm_slot, mm_list); | |
1312 | ksm_scan.mm_slot = slot; | |
1313 | spin_unlock(&ksm_mmlist_lock); | |
1314 | next_mm: | |
1315 | ksm_scan.address = 0; | |
6514d511 | 1316 | ksm_scan.rmap_list = &slot->rmap_list; |
31dbd01f IE |
1317 | } |
1318 | ||
1319 | mm = slot->mm; | |
1320 | down_read(&mm->mmap_sem); | |
9ba69294 HD |
1321 | if (ksm_test_exit(mm)) |
1322 | vma = NULL; | |
1323 | else | |
1324 | vma = find_vma(mm, ksm_scan.address); | |
1325 | ||
1326 | for (; vma; vma = vma->vm_next) { | |
31dbd01f IE |
1327 | if (!(vma->vm_flags & VM_MERGEABLE)) |
1328 | continue; | |
1329 | if (ksm_scan.address < vma->vm_start) | |
1330 | ksm_scan.address = vma->vm_start; | |
1331 | if (!vma->anon_vma) | |
1332 | ksm_scan.address = vma->vm_end; | |
1333 | ||
1334 | while (ksm_scan.address < vma->vm_end) { | |
9ba69294 HD |
1335 | if (ksm_test_exit(mm)) |
1336 | break; | |
31dbd01f | 1337 | *page = follow_page(vma, ksm_scan.address, FOLL_GET); |
21ae5b01 AA |
1338 | if (IS_ERR_OR_NULL(*page)) { |
1339 | ksm_scan.address += PAGE_SIZE; | |
1340 | cond_resched(); | |
1341 | continue; | |
1342 | } | |
29ad768c AA |
1343 | if (PageAnon(*page) || |
1344 | page_trans_compound_anon(*page)) { | |
31dbd01f IE |
1345 | flush_anon_page(vma, *page, ksm_scan.address); |
1346 | flush_dcache_page(*page); | |
1347 | rmap_item = get_next_rmap_item(slot, | |
6514d511 | 1348 | ksm_scan.rmap_list, ksm_scan.address); |
31dbd01f | 1349 | if (rmap_item) { |
6514d511 HD |
1350 | ksm_scan.rmap_list = |
1351 | &rmap_item->rmap_list; | |
31dbd01f IE |
1352 | ksm_scan.address += PAGE_SIZE; |
1353 | } else | |
1354 | put_page(*page); | |
1355 | up_read(&mm->mmap_sem); | |
1356 | return rmap_item; | |
1357 | } | |
21ae5b01 | 1358 | put_page(*page); |
31dbd01f IE |
1359 | ksm_scan.address += PAGE_SIZE; |
1360 | cond_resched(); | |
1361 | } | |
1362 | } | |
1363 | ||
9ba69294 HD |
1364 | if (ksm_test_exit(mm)) { |
1365 | ksm_scan.address = 0; | |
6514d511 | 1366 | ksm_scan.rmap_list = &slot->rmap_list; |
9ba69294 | 1367 | } |
31dbd01f IE |
1368 | /* |
1369 | * Nuke all the rmap_items that are above this current rmap: | |
1370 | * because there were no VM_MERGEABLE vmas with such addresses. | |
1371 | */ | |
6514d511 | 1372 | remove_trailing_rmap_items(slot, ksm_scan.rmap_list); |
31dbd01f IE |
1373 | |
1374 | spin_lock(&ksm_mmlist_lock); | |
cd551f97 HD |
1375 | ksm_scan.mm_slot = list_entry(slot->mm_list.next, |
1376 | struct mm_slot, mm_list); | |
1377 | if (ksm_scan.address == 0) { | |
1378 | /* | |
1379 | * We've completed a full scan of all vmas, holding mmap_sem | |
1380 | * throughout, and found no VM_MERGEABLE: so do the same as | |
1381 | * __ksm_exit does to remove this mm from all our lists now. | |
9ba69294 HD |
1382 | * This applies either when cleaning up after __ksm_exit |
1383 | * (but beware: we can reach here even before __ksm_exit), | |
1384 | * or when all VM_MERGEABLE areas have been unmapped (and | |
1385 | * mmap_sem then protects against race with MADV_MERGEABLE). | |
cd551f97 HD |
1386 | */ |
1387 | hlist_del(&slot->link); | |
1388 | list_del(&slot->mm_list); | |
9ba69294 HD |
1389 | spin_unlock(&ksm_mmlist_lock); |
1390 | ||
cd551f97 HD |
1391 | free_mm_slot(slot); |
1392 | clear_bit(MMF_VM_MERGEABLE, &mm->flags); | |
9ba69294 HD |
1393 | up_read(&mm->mmap_sem); |
1394 | mmdrop(mm); | |
1395 | } else { | |
1396 | spin_unlock(&ksm_mmlist_lock); | |
1397 | up_read(&mm->mmap_sem); | |
cd551f97 | 1398 | } |
31dbd01f IE |
1399 | |
1400 | /* Repeat until we've completed scanning the whole list */ | |
cd551f97 | 1401 | slot = ksm_scan.mm_slot; |
31dbd01f IE |
1402 | if (slot != &ksm_mm_head) |
1403 | goto next_mm; | |
1404 | ||
31dbd01f IE |
1405 | ksm_scan.seqnr++; |
1406 | return NULL; | |
1407 | } | |
1408 | ||
1409 | /** | |
1410 | * ksm_do_scan - the ksm scanner main worker function. | |
1411 | * @scan_npages - number of pages we want to scan before we return. | |
1412 | */ | |
1413 | static void ksm_do_scan(unsigned int scan_npages) | |
1414 | { | |
1415 | struct rmap_item *rmap_item; | |
22eccdd7 | 1416 | struct page *uninitialized_var(page); |
31dbd01f | 1417 | |
878aee7d | 1418 | while (scan_npages-- && likely(!freezing(current))) { |
31dbd01f IE |
1419 | cond_resched(); |
1420 | rmap_item = scan_get_next_rmap_item(&page); | |
1421 | if (!rmap_item) | |
1422 | return; | |
1423 | if (!PageKsm(page) || !in_stable_tree(rmap_item)) | |
1424 | cmp_and_merge_page(page, rmap_item); | |
1425 | put_page(page); | |
1426 | } | |
1427 | } | |
1428 | ||
6e158384 HD |
1429 | static int ksmd_should_run(void) |
1430 | { | |
1431 | return (ksm_run & KSM_RUN_MERGE) && !list_empty(&ksm_mm_head.mm_list); | |
1432 | } | |
1433 | ||
31dbd01f IE |
1434 | static int ksm_scan_thread(void *nothing) |
1435 | { | |
878aee7d | 1436 | set_freezable(); |
339aa624 | 1437 | set_user_nice(current, 5); |
31dbd01f IE |
1438 | |
1439 | while (!kthread_should_stop()) { | |
6e158384 HD |
1440 | mutex_lock(&ksm_thread_mutex); |
1441 | if (ksmd_should_run()) | |
31dbd01f | 1442 | ksm_do_scan(ksm_thread_pages_to_scan); |
6e158384 HD |
1443 | mutex_unlock(&ksm_thread_mutex); |
1444 | ||
878aee7d AA |
1445 | try_to_freeze(); |
1446 | ||
6e158384 | 1447 | if (ksmd_should_run()) { |
31dbd01f IE |
1448 | schedule_timeout_interruptible( |
1449 | msecs_to_jiffies(ksm_thread_sleep_millisecs)); | |
1450 | } else { | |
878aee7d | 1451 | wait_event_freezable(ksm_thread_wait, |
6e158384 | 1452 | ksmd_should_run() || kthread_should_stop()); |
31dbd01f IE |
1453 | } |
1454 | } | |
1455 | return 0; | |
1456 | } | |
1457 | ||
f8af4da3 HD |
1458 | int ksm_madvise(struct vm_area_struct *vma, unsigned long start, |
1459 | unsigned long end, int advice, unsigned long *vm_flags) | |
1460 | { | |
1461 | struct mm_struct *mm = vma->vm_mm; | |
d952b791 | 1462 | int err; |
f8af4da3 HD |
1463 | |
1464 | switch (advice) { | |
1465 | case MADV_MERGEABLE: | |
1466 | /* | |
1467 | * Be somewhat over-protective for now! | |
1468 | */ | |
1469 | if (*vm_flags & (VM_MERGEABLE | VM_SHARED | VM_MAYSHARE | | |
1470 | VM_PFNMAP | VM_IO | VM_DONTEXPAND | | |
1471 | VM_RESERVED | VM_HUGETLB | VM_INSERTPAGE | | |
5ad64688 | 1472 | VM_NONLINEAR | VM_MIXEDMAP | VM_SAO)) |
f8af4da3 HD |
1473 | return 0; /* just ignore the advice */ |
1474 | ||
d952b791 HD |
1475 | if (!test_bit(MMF_VM_MERGEABLE, &mm->flags)) { |
1476 | err = __ksm_enter(mm); | |
1477 | if (err) | |
1478 | return err; | |
1479 | } | |
f8af4da3 HD |
1480 | |
1481 | *vm_flags |= VM_MERGEABLE; | |
1482 | break; | |
1483 | ||
1484 | case MADV_UNMERGEABLE: | |
1485 | if (!(*vm_flags & VM_MERGEABLE)) | |
1486 | return 0; /* just ignore the advice */ | |
1487 | ||
d952b791 HD |
1488 | if (vma->anon_vma) { |
1489 | err = unmerge_ksm_pages(vma, start, end); | |
1490 | if (err) | |
1491 | return err; | |
1492 | } | |
f8af4da3 HD |
1493 | |
1494 | *vm_flags &= ~VM_MERGEABLE; | |
1495 | break; | |
1496 | } | |
1497 | ||
1498 | return 0; | |
1499 | } | |
1500 | ||
1501 | int __ksm_enter(struct mm_struct *mm) | |
1502 | { | |
6e158384 HD |
1503 | struct mm_slot *mm_slot; |
1504 | int needs_wakeup; | |
1505 | ||
1506 | mm_slot = alloc_mm_slot(); | |
31dbd01f IE |
1507 | if (!mm_slot) |
1508 | return -ENOMEM; | |
1509 | ||
6e158384 HD |
1510 | /* Check ksm_run too? Would need tighter locking */ |
1511 | needs_wakeup = list_empty(&ksm_mm_head.mm_list); | |
1512 | ||
31dbd01f IE |
1513 | spin_lock(&ksm_mmlist_lock); |
1514 | insert_to_mm_slots_hash(mm, mm_slot); | |
1515 | /* | |
1516 | * Insert just behind the scanning cursor, to let the area settle | |
1517 | * down a little; when fork is followed by immediate exec, we don't | |
1518 | * want ksmd to waste time setting up and tearing down an rmap_list. | |
1519 | */ | |
1520 | list_add_tail(&mm_slot->mm_list, &ksm_scan.mm_slot->mm_list); | |
1521 | spin_unlock(&ksm_mmlist_lock); | |
1522 | ||
f8af4da3 | 1523 | set_bit(MMF_VM_MERGEABLE, &mm->flags); |
9ba69294 | 1524 | atomic_inc(&mm->mm_count); |
6e158384 HD |
1525 | |
1526 | if (needs_wakeup) | |
1527 | wake_up_interruptible(&ksm_thread_wait); | |
1528 | ||
f8af4da3 HD |
1529 | return 0; |
1530 | } | |
1531 | ||
1c2fb7a4 | 1532 | void __ksm_exit(struct mm_struct *mm) |
f8af4da3 | 1533 | { |
cd551f97 | 1534 | struct mm_slot *mm_slot; |
9ba69294 | 1535 | int easy_to_free = 0; |
cd551f97 | 1536 | |
31dbd01f | 1537 | /* |
9ba69294 HD |
1538 | * This process is exiting: if it's straightforward (as is the |
1539 | * case when ksmd was never running), free mm_slot immediately. | |
1540 | * But if it's at the cursor or has rmap_items linked to it, use | |
1541 | * mmap_sem to synchronize with any break_cows before pagetables | |
1542 | * are freed, and leave the mm_slot on the list for ksmd to free. | |
1543 | * Beware: ksm may already have noticed it exiting and freed the slot. | |
31dbd01f | 1544 | */ |
9ba69294 | 1545 | |
cd551f97 HD |
1546 | spin_lock(&ksm_mmlist_lock); |
1547 | mm_slot = get_mm_slot(mm); | |
9ba69294 | 1548 | if (mm_slot && ksm_scan.mm_slot != mm_slot) { |
6514d511 | 1549 | if (!mm_slot->rmap_list) { |
9ba69294 HD |
1550 | hlist_del(&mm_slot->link); |
1551 | list_del(&mm_slot->mm_list); | |
1552 | easy_to_free = 1; | |
1553 | } else { | |
1554 | list_move(&mm_slot->mm_list, | |
1555 | &ksm_scan.mm_slot->mm_list); | |
1556 | } | |
cd551f97 | 1557 | } |
cd551f97 HD |
1558 | spin_unlock(&ksm_mmlist_lock); |
1559 | ||
9ba69294 HD |
1560 | if (easy_to_free) { |
1561 | free_mm_slot(mm_slot); | |
1562 | clear_bit(MMF_VM_MERGEABLE, &mm->flags); | |
1563 | mmdrop(mm); | |
1564 | } else if (mm_slot) { | |
9ba69294 HD |
1565 | down_write(&mm->mmap_sem); |
1566 | up_write(&mm->mmap_sem); | |
9ba69294 | 1567 | } |
31dbd01f IE |
1568 | } |
1569 | ||
5ad64688 HD |
1570 | struct page *ksm_does_need_to_copy(struct page *page, |
1571 | struct vm_area_struct *vma, unsigned long address) | |
1572 | { | |
1573 | struct page *new_page; | |
1574 | ||
5ad64688 HD |
1575 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); |
1576 | if (new_page) { | |
1577 | copy_user_highpage(new_page, page, address, vma); | |
1578 | ||
1579 | SetPageDirty(new_page); | |
1580 | __SetPageUptodate(new_page); | |
1581 | SetPageSwapBacked(new_page); | |
1582 | __set_page_locked(new_page); | |
1583 | ||
1584 | if (page_evictable(new_page, vma)) | |
1585 | lru_cache_add_lru(new_page, LRU_ACTIVE_ANON); | |
1586 | else | |
1587 | add_page_to_unevictable_list(new_page); | |
1588 | } | |
1589 | ||
5ad64688 HD |
1590 | return new_page; |
1591 | } | |
1592 | ||
1593 | int page_referenced_ksm(struct page *page, struct mem_cgroup *memcg, | |
1594 | unsigned long *vm_flags) | |
1595 | { | |
1596 | struct stable_node *stable_node; | |
1597 | struct rmap_item *rmap_item; | |
1598 | struct hlist_node *hlist; | |
1599 | unsigned int mapcount = page_mapcount(page); | |
1600 | int referenced = 0; | |
db114b83 | 1601 | int search_new_forks = 0; |
5ad64688 HD |
1602 | |
1603 | VM_BUG_ON(!PageKsm(page)); | |
1604 | VM_BUG_ON(!PageLocked(page)); | |
1605 | ||
1606 | stable_node = page_stable_node(page); | |
1607 | if (!stable_node) | |
1608 | return 0; | |
db114b83 | 1609 | again: |
5ad64688 | 1610 | hlist_for_each_entry(rmap_item, hlist, &stable_node->hlist, hlist) { |
db114b83 | 1611 | struct anon_vma *anon_vma = rmap_item->anon_vma; |
5beb4930 | 1612 | struct anon_vma_chain *vmac; |
db114b83 | 1613 | struct vm_area_struct *vma; |
5ad64688 | 1614 | |
cba48b98 | 1615 | anon_vma_lock(anon_vma); |
5beb4930 RR |
1616 | list_for_each_entry(vmac, &anon_vma->head, same_anon_vma) { |
1617 | vma = vmac->vma; | |
db114b83 HD |
1618 | if (rmap_item->address < vma->vm_start || |
1619 | rmap_item->address >= vma->vm_end) | |
1620 | continue; | |
1621 | /* | |
1622 | * Initially we examine only the vma which covers this | |
1623 | * rmap_item; but later, if there is still work to do, | |
1624 | * we examine covering vmas in other mms: in case they | |
1625 | * were forked from the original since ksmd passed. | |
1626 | */ | |
1627 | if ((rmap_item->mm == vma->vm_mm) == search_new_forks) | |
1628 | continue; | |
1629 | ||
1630 | if (memcg && !mm_match_cgroup(vma->vm_mm, memcg)) | |
1631 | continue; | |
5ad64688 | 1632 | |
db114b83 | 1633 | referenced += page_referenced_one(page, vma, |
5ad64688 | 1634 | rmap_item->address, &mapcount, vm_flags); |
db114b83 HD |
1635 | if (!search_new_forks || !mapcount) |
1636 | break; | |
1637 | } | |
cba48b98 | 1638 | anon_vma_unlock(anon_vma); |
5ad64688 HD |
1639 | if (!mapcount) |
1640 | goto out; | |
1641 | } | |
db114b83 HD |
1642 | if (!search_new_forks++) |
1643 | goto again; | |
5ad64688 | 1644 | out: |
5ad64688 HD |
1645 | return referenced; |
1646 | } | |
1647 | ||
1648 | int try_to_unmap_ksm(struct page *page, enum ttu_flags flags) | |
1649 | { | |
1650 | struct stable_node *stable_node; | |
1651 | struct hlist_node *hlist; | |
1652 | struct rmap_item *rmap_item; | |
1653 | int ret = SWAP_AGAIN; | |
db114b83 | 1654 | int search_new_forks = 0; |
5ad64688 HD |
1655 | |
1656 | VM_BUG_ON(!PageKsm(page)); | |
1657 | VM_BUG_ON(!PageLocked(page)); | |
1658 | ||
1659 | stable_node = page_stable_node(page); | |
1660 | if (!stable_node) | |
1661 | return SWAP_FAIL; | |
db114b83 | 1662 | again: |
5ad64688 | 1663 | hlist_for_each_entry(rmap_item, hlist, &stable_node->hlist, hlist) { |
db114b83 | 1664 | struct anon_vma *anon_vma = rmap_item->anon_vma; |
5beb4930 | 1665 | struct anon_vma_chain *vmac; |
db114b83 | 1666 | struct vm_area_struct *vma; |
5ad64688 | 1667 | |
cba48b98 | 1668 | anon_vma_lock(anon_vma); |
5beb4930 RR |
1669 | list_for_each_entry(vmac, &anon_vma->head, same_anon_vma) { |
1670 | vma = vmac->vma; | |
db114b83 HD |
1671 | if (rmap_item->address < vma->vm_start || |
1672 | rmap_item->address >= vma->vm_end) | |
1673 | continue; | |
1674 | /* | |
1675 | * Initially we examine only the vma which covers this | |
1676 | * rmap_item; but later, if there is still work to do, | |
1677 | * we examine covering vmas in other mms: in case they | |
1678 | * were forked from the original since ksmd passed. | |
1679 | */ | |
1680 | if ((rmap_item->mm == vma->vm_mm) == search_new_forks) | |
1681 | continue; | |
1682 | ||
1683 | ret = try_to_unmap_one(page, vma, | |
1684 | rmap_item->address, flags); | |
1685 | if (ret != SWAP_AGAIN || !page_mapped(page)) { | |
cba48b98 | 1686 | anon_vma_unlock(anon_vma); |
db114b83 HD |
1687 | goto out; |
1688 | } | |
1689 | } | |
cba48b98 | 1690 | anon_vma_unlock(anon_vma); |
5ad64688 | 1691 | } |
db114b83 HD |
1692 | if (!search_new_forks++) |
1693 | goto again; | |
5ad64688 | 1694 | out: |
5ad64688 HD |
1695 | return ret; |
1696 | } | |
1697 | ||
e9995ef9 HD |
1698 | #ifdef CONFIG_MIGRATION |
1699 | int rmap_walk_ksm(struct page *page, int (*rmap_one)(struct page *, | |
1700 | struct vm_area_struct *, unsigned long, void *), void *arg) | |
1701 | { | |
1702 | struct stable_node *stable_node; | |
1703 | struct hlist_node *hlist; | |
1704 | struct rmap_item *rmap_item; | |
1705 | int ret = SWAP_AGAIN; | |
1706 | int search_new_forks = 0; | |
1707 | ||
1708 | VM_BUG_ON(!PageKsm(page)); | |
1709 | VM_BUG_ON(!PageLocked(page)); | |
1710 | ||
1711 | stable_node = page_stable_node(page); | |
1712 | if (!stable_node) | |
1713 | return ret; | |
1714 | again: | |
1715 | hlist_for_each_entry(rmap_item, hlist, &stable_node->hlist, hlist) { | |
1716 | struct anon_vma *anon_vma = rmap_item->anon_vma; | |
5beb4930 | 1717 | struct anon_vma_chain *vmac; |
e9995ef9 HD |
1718 | struct vm_area_struct *vma; |
1719 | ||
cba48b98 | 1720 | anon_vma_lock(anon_vma); |
5beb4930 RR |
1721 | list_for_each_entry(vmac, &anon_vma->head, same_anon_vma) { |
1722 | vma = vmac->vma; | |
e9995ef9 HD |
1723 | if (rmap_item->address < vma->vm_start || |
1724 | rmap_item->address >= vma->vm_end) | |
1725 | continue; | |
1726 | /* | |
1727 | * Initially we examine only the vma which covers this | |
1728 | * rmap_item; but later, if there is still work to do, | |
1729 | * we examine covering vmas in other mms: in case they | |
1730 | * were forked from the original since ksmd passed. | |
1731 | */ | |
1732 | if ((rmap_item->mm == vma->vm_mm) == search_new_forks) | |
1733 | continue; | |
1734 | ||
1735 | ret = rmap_one(page, vma, rmap_item->address, arg); | |
1736 | if (ret != SWAP_AGAIN) { | |
cba48b98 | 1737 | anon_vma_unlock(anon_vma); |
e9995ef9 HD |
1738 | goto out; |
1739 | } | |
1740 | } | |
cba48b98 | 1741 | anon_vma_unlock(anon_vma); |
e9995ef9 HD |
1742 | } |
1743 | if (!search_new_forks++) | |
1744 | goto again; | |
1745 | out: | |
1746 | return ret; | |
1747 | } | |
1748 | ||
1749 | void ksm_migrate_page(struct page *newpage, struct page *oldpage) | |
1750 | { | |
1751 | struct stable_node *stable_node; | |
1752 | ||
1753 | VM_BUG_ON(!PageLocked(oldpage)); | |
1754 | VM_BUG_ON(!PageLocked(newpage)); | |
1755 | VM_BUG_ON(newpage->mapping != oldpage->mapping); | |
1756 | ||
1757 | stable_node = page_stable_node(newpage); | |
1758 | if (stable_node) { | |
62b61f61 HD |
1759 | VM_BUG_ON(stable_node->kpfn != page_to_pfn(oldpage)); |
1760 | stable_node->kpfn = page_to_pfn(newpage); | |
e9995ef9 HD |
1761 | } |
1762 | } | |
1763 | #endif /* CONFIG_MIGRATION */ | |
1764 | ||
62b61f61 HD |
1765 | #ifdef CONFIG_MEMORY_HOTREMOVE |
1766 | static struct stable_node *ksm_check_stable_tree(unsigned long start_pfn, | |
1767 | unsigned long end_pfn) | |
1768 | { | |
1769 | struct rb_node *node; | |
1770 | ||
1771 | for (node = rb_first(&root_stable_tree); node; node = rb_next(node)) { | |
1772 | struct stable_node *stable_node; | |
1773 | ||
1774 | stable_node = rb_entry(node, struct stable_node, node); | |
1775 | if (stable_node->kpfn >= start_pfn && | |
1776 | stable_node->kpfn < end_pfn) | |
1777 | return stable_node; | |
1778 | } | |
1779 | return NULL; | |
1780 | } | |
1781 | ||
1782 | static int ksm_memory_callback(struct notifier_block *self, | |
1783 | unsigned long action, void *arg) | |
1784 | { | |
1785 | struct memory_notify *mn = arg; | |
1786 | struct stable_node *stable_node; | |
1787 | ||
1788 | switch (action) { | |
1789 | case MEM_GOING_OFFLINE: | |
1790 | /* | |
1791 | * Keep it very simple for now: just lock out ksmd and | |
1792 | * MADV_UNMERGEABLE while any memory is going offline. | |
a0b0f58c KM |
1793 | * mutex_lock_nested() is necessary because lockdep was alarmed |
1794 | * that here we take ksm_thread_mutex inside notifier chain | |
1795 | * mutex, and later take notifier chain mutex inside | |
1796 | * ksm_thread_mutex to unlock it. But that's safe because both | |
1797 | * are inside mem_hotplug_mutex. | |
62b61f61 | 1798 | */ |
a0b0f58c | 1799 | mutex_lock_nested(&ksm_thread_mutex, SINGLE_DEPTH_NESTING); |
62b61f61 HD |
1800 | break; |
1801 | ||
1802 | case MEM_OFFLINE: | |
1803 | /* | |
1804 | * Most of the work is done by page migration; but there might | |
1805 | * be a few stable_nodes left over, still pointing to struct | |
1806 | * pages which have been offlined: prune those from the tree. | |
1807 | */ | |
1808 | while ((stable_node = ksm_check_stable_tree(mn->start_pfn, | |
1809 | mn->start_pfn + mn->nr_pages)) != NULL) | |
1810 | remove_node_from_stable_tree(stable_node); | |
1811 | /* fallthrough */ | |
1812 | ||
1813 | case MEM_CANCEL_OFFLINE: | |
1814 | mutex_unlock(&ksm_thread_mutex); | |
1815 | break; | |
1816 | } | |
1817 | return NOTIFY_OK; | |
1818 | } | |
1819 | #endif /* CONFIG_MEMORY_HOTREMOVE */ | |
1820 | ||
2ffd8679 HD |
1821 | #ifdef CONFIG_SYSFS |
1822 | /* | |
1823 | * This all compiles without CONFIG_SYSFS, but is a waste of space. | |
1824 | */ | |
1825 | ||
31dbd01f IE |
1826 | #define KSM_ATTR_RO(_name) \ |
1827 | static struct kobj_attribute _name##_attr = __ATTR_RO(_name) | |
1828 | #define KSM_ATTR(_name) \ | |
1829 | static struct kobj_attribute _name##_attr = \ | |
1830 | __ATTR(_name, 0644, _name##_show, _name##_store) | |
1831 | ||
1832 | static ssize_t sleep_millisecs_show(struct kobject *kobj, | |
1833 | struct kobj_attribute *attr, char *buf) | |
1834 | { | |
1835 | return sprintf(buf, "%u\n", ksm_thread_sleep_millisecs); | |
1836 | } | |
1837 | ||
1838 | static ssize_t sleep_millisecs_store(struct kobject *kobj, | |
1839 | struct kobj_attribute *attr, | |
1840 | const char *buf, size_t count) | |
1841 | { | |
1842 | unsigned long msecs; | |
1843 | int err; | |
1844 | ||
1845 | err = strict_strtoul(buf, 10, &msecs); | |
1846 | if (err || msecs > UINT_MAX) | |
1847 | return -EINVAL; | |
1848 | ||
1849 | ksm_thread_sleep_millisecs = msecs; | |
1850 | ||
1851 | return count; | |
1852 | } | |
1853 | KSM_ATTR(sleep_millisecs); | |
1854 | ||
1855 | static ssize_t pages_to_scan_show(struct kobject *kobj, | |
1856 | struct kobj_attribute *attr, char *buf) | |
1857 | { | |
1858 | return sprintf(buf, "%u\n", ksm_thread_pages_to_scan); | |
1859 | } | |
1860 | ||
1861 | static ssize_t pages_to_scan_store(struct kobject *kobj, | |
1862 | struct kobj_attribute *attr, | |
1863 | const char *buf, size_t count) | |
1864 | { | |
1865 | int err; | |
1866 | unsigned long nr_pages; | |
1867 | ||
1868 | err = strict_strtoul(buf, 10, &nr_pages); | |
1869 | if (err || nr_pages > UINT_MAX) | |
1870 | return -EINVAL; | |
1871 | ||
1872 | ksm_thread_pages_to_scan = nr_pages; | |
1873 | ||
1874 | return count; | |
1875 | } | |
1876 | KSM_ATTR(pages_to_scan); | |
1877 | ||
1878 | static ssize_t run_show(struct kobject *kobj, struct kobj_attribute *attr, | |
1879 | char *buf) | |
1880 | { | |
1881 | return sprintf(buf, "%u\n", ksm_run); | |
1882 | } | |
1883 | ||
1884 | static ssize_t run_store(struct kobject *kobj, struct kobj_attribute *attr, | |
1885 | const char *buf, size_t count) | |
1886 | { | |
1887 | int err; | |
1888 | unsigned long flags; | |
1889 | ||
1890 | err = strict_strtoul(buf, 10, &flags); | |
1891 | if (err || flags > UINT_MAX) | |
1892 | return -EINVAL; | |
1893 | if (flags > KSM_RUN_UNMERGE) | |
1894 | return -EINVAL; | |
1895 | ||
1896 | /* | |
1897 | * KSM_RUN_MERGE sets ksmd running, and 0 stops it running. | |
1898 | * KSM_RUN_UNMERGE stops it running and unmerges all rmap_items, | |
d0f209f6 HD |
1899 | * breaking COW to free the pages_shared (but leaves mm_slots |
1900 | * on the list for when ksmd may be set running again). | |
31dbd01f IE |
1901 | */ |
1902 | ||
1903 | mutex_lock(&ksm_thread_mutex); | |
1904 | if (ksm_run != flags) { | |
1905 | ksm_run = flags; | |
d952b791 | 1906 | if (flags & KSM_RUN_UNMERGE) { |
35451bee | 1907 | current->flags |= PF_OOM_ORIGIN; |
d952b791 | 1908 | err = unmerge_and_remove_all_rmap_items(); |
35451bee | 1909 | current->flags &= ~PF_OOM_ORIGIN; |
d952b791 HD |
1910 | if (err) { |
1911 | ksm_run = KSM_RUN_STOP; | |
1912 | count = err; | |
1913 | } | |
1914 | } | |
31dbd01f IE |
1915 | } |
1916 | mutex_unlock(&ksm_thread_mutex); | |
1917 | ||
1918 | if (flags & KSM_RUN_MERGE) | |
1919 | wake_up_interruptible(&ksm_thread_wait); | |
1920 | ||
1921 | return count; | |
1922 | } | |
1923 | KSM_ATTR(run); | |
1924 | ||
b4028260 HD |
1925 | static ssize_t pages_shared_show(struct kobject *kobj, |
1926 | struct kobj_attribute *attr, char *buf) | |
1927 | { | |
1928 | return sprintf(buf, "%lu\n", ksm_pages_shared); | |
1929 | } | |
1930 | KSM_ATTR_RO(pages_shared); | |
1931 | ||
1932 | static ssize_t pages_sharing_show(struct kobject *kobj, | |
1933 | struct kobj_attribute *attr, char *buf) | |
1934 | { | |
e178dfde | 1935 | return sprintf(buf, "%lu\n", ksm_pages_sharing); |
b4028260 HD |
1936 | } |
1937 | KSM_ATTR_RO(pages_sharing); | |
1938 | ||
473b0ce4 HD |
1939 | static ssize_t pages_unshared_show(struct kobject *kobj, |
1940 | struct kobj_attribute *attr, char *buf) | |
1941 | { | |
1942 | return sprintf(buf, "%lu\n", ksm_pages_unshared); | |
1943 | } | |
1944 | KSM_ATTR_RO(pages_unshared); | |
1945 | ||
1946 | static ssize_t pages_volatile_show(struct kobject *kobj, | |
1947 | struct kobj_attribute *attr, char *buf) | |
1948 | { | |
1949 | long ksm_pages_volatile; | |
1950 | ||
1951 | ksm_pages_volatile = ksm_rmap_items - ksm_pages_shared | |
1952 | - ksm_pages_sharing - ksm_pages_unshared; | |
1953 | /* | |
1954 | * It was not worth any locking to calculate that statistic, | |
1955 | * but it might therefore sometimes be negative: conceal that. | |
1956 | */ | |
1957 | if (ksm_pages_volatile < 0) | |
1958 | ksm_pages_volatile = 0; | |
1959 | return sprintf(buf, "%ld\n", ksm_pages_volatile); | |
1960 | } | |
1961 | KSM_ATTR_RO(pages_volatile); | |
1962 | ||
1963 | static ssize_t full_scans_show(struct kobject *kobj, | |
1964 | struct kobj_attribute *attr, char *buf) | |
1965 | { | |
1966 | return sprintf(buf, "%lu\n", ksm_scan.seqnr); | |
1967 | } | |
1968 | KSM_ATTR_RO(full_scans); | |
1969 | ||
31dbd01f IE |
1970 | static struct attribute *ksm_attrs[] = { |
1971 | &sleep_millisecs_attr.attr, | |
1972 | &pages_to_scan_attr.attr, | |
1973 | &run_attr.attr, | |
b4028260 HD |
1974 | &pages_shared_attr.attr, |
1975 | &pages_sharing_attr.attr, | |
473b0ce4 HD |
1976 | &pages_unshared_attr.attr, |
1977 | &pages_volatile_attr.attr, | |
1978 | &full_scans_attr.attr, | |
31dbd01f IE |
1979 | NULL, |
1980 | }; | |
1981 | ||
1982 | static struct attribute_group ksm_attr_group = { | |
1983 | .attrs = ksm_attrs, | |
1984 | .name = "ksm", | |
1985 | }; | |
2ffd8679 | 1986 | #endif /* CONFIG_SYSFS */ |
31dbd01f IE |
1987 | |
1988 | static int __init ksm_init(void) | |
1989 | { | |
1990 | struct task_struct *ksm_thread; | |
1991 | int err; | |
1992 | ||
1993 | err = ksm_slab_init(); | |
1994 | if (err) | |
1995 | goto out; | |
1996 | ||
31dbd01f IE |
1997 | ksm_thread = kthread_run(ksm_scan_thread, NULL, "ksmd"); |
1998 | if (IS_ERR(ksm_thread)) { | |
1999 | printk(KERN_ERR "ksm: creating kthread failed\n"); | |
2000 | err = PTR_ERR(ksm_thread); | |
d9f8984c | 2001 | goto out_free; |
31dbd01f IE |
2002 | } |
2003 | ||
2ffd8679 | 2004 | #ifdef CONFIG_SYSFS |
31dbd01f IE |
2005 | err = sysfs_create_group(mm_kobj, &ksm_attr_group); |
2006 | if (err) { | |
2007 | printk(KERN_ERR "ksm: register sysfs failed\n"); | |
2ffd8679 | 2008 | kthread_stop(ksm_thread); |
d9f8984c | 2009 | goto out_free; |
31dbd01f | 2010 | } |
c73602ad HD |
2011 | #else |
2012 | ksm_run = KSM_RUN_MERGE; /* no way for user to start it */ | |
2013 | ||
2ffd8679 | 2014 | #endif /* CONFIG_SYSFS */ |
31dbd01f | 2015 | |
62b61f61 HD |
2016 | #ifdef CONFIG_MEMORY_HOTREMOVE |
2017 | /* | |
2018 | * Choose a high priority since the callback takes ksm_thread_mutex: | |
2019 | * later callbacks could only be taking locks which nest within that. | |
2020 | */ | |
2021 | hotplug_memory_notifier(ksm_memory_callback, 100); | |
2022 | #endif | |
31dbd01f IE |
2023 | return 0; |
2024 | ||
d9f8984c | 2025 | out_free: |
31dbd01f IE |
2026 | ksm_slab_free(); |
2027 | out: | |
2028 | return err; | |
f8af4da3 | 2029 | } |
31dbd01f | 2030 | module_init(ksm_init) |