2 * Memory merging support.
4 * This code enables dynamic sharing of identical pages found in different
5 * memory areas, even if they are not shared by fork()
7 * Copyright (C) 2008-2009 Red Hat, Inc.
14 * This work is licensed under the terms of the GNU GPL, version 2.
17 #include <linux/errno.h>
20 #include <linux/mman.h>
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>
32 #include <linux/mmu_notifier.h>
33 #include <linux/ksm.h>
35 #include <asm/tlbflush.h>
38 * A few notes about the KSM scanning process,
39 * to make it easier to understand the data structures below:
41 * In order to reduce excessive scanning, KSM sorts the memory pages by their
42 * contents into a data structure that holds pointers to the pages' locations.
44 * Since the contents of the pages may change at any moment, KSM cannot just
45 * insert the pages into a normal sorted tree and expect it to find anything.
46 * Therefore KSM uses two data structures - the stable and the unstable tree.
48 * The stable tree holds pointers to all the merged pages (ksm pages), sorted
49 * by their contents. Because each such page is write-protected, searching on
50 * this tree is fully assured to be working (except when pages are unmapped),
51 * and therefore this tree is called the stable tree.
53 * In addition to the stable tree, KSM uses a second data structure called the
54 * unstable tree: this tree holds pointers to pages which have been found to
55 * be "unchanged for a period of time". The unstable tree sorts these pages
56 * by their contents, but since they are not write-protected, KSM cannot rely
57 * upon the unstable tree to work correctly - the unstable tree is liable to
58 * be corrupted as its contents are modified, and so it is called unstable.
60 * KSM solves this problem by several techniques:
62 * 1) The unstable tree is flushed every time KSM completes scanning all
63 * memory areas, and then the tree is rebuilt again from the beginning.
64 * 2) KSM will only insert into the unstable tree, pages whose hash value
65 * has not changed since the previous scan of all memory areas.
66 * 3) The unstable tree is a RedBlack Tree - so its balancing is based on the
67 * colors of the nodes and not on their contents, assuring that even when
68 * the tree gets "corrupted" it won't get out of balance, so scanning time
69 * remains the same (also, searching and inserting nodes in an rbtree uses
70 * the same algorithm, so we have no overhead when we flush and rebuild).
71 * 4) KSM never flushes the stable tree, which means that even if it were to
72 * take 10 attempts to find a page in the unstable tree, once it is found,
73 * it is secured in the stable tree. (When we scan a new page, we first
74 * compare it against the stable tree, and then against the unstable tree.)
78 * struct mm_slot - ksm information per mm that is being scanned
79 * @link: link to the mm_slots hash list
80 * @mm_list: link into the mm_slots list, rooted in ksm_mm_head
81 * @rmap_list: head for this mm_slot's list of rmap_items
82 * @mm: the mm that this information is valid for
85 struct hlist_node link
;
86 struct list_head mm_list
;
87 struct list_head rmap_list
;
92 * struct ksm_scan - cursor for scanning
93 * @mm_slot: the current mm_slot we are scanning
94 * @address: the next address inside that to be scanned
95 * @rmap_item: the current rmap that we are scanning inside the rmap_list
96 * @seqnr: count of completed full scans (needed when removing unstable node)
98 * There is only the one ksm_scan instance of this cursor structure.
101 struct mm_slot
*mm_slot
;
102 unsigned long address
;
103 struct rmap_item
*rmap_item
;
108 * struct rmap_item - reverse mapping item for virtual addresses
109 * @link: link into mm_slot's rmap_list (rmap_list is per mm)
110 * @mm: the memory structure this rmap_item is pointing into
111 * @address: the virtual address this rmap_item tracks (+ flags in low bits)
112 * @oldchecksum: previous checksum of the page at that virtual address
113 * @node: rb_node of this rmap_item in either unstable or stable tree
114 * @next: next rmap_item hanging off the same node of the stable tree
115 * @prev: previous rmap_item hanging off the same node of the stable tree
118 struct list_head link
;
119 struct mm_struct
*mm
;
120 unsigned long address
; /* + low bits used for flags below */
122 unsigned int oldchecksum
; /* when unstable */
123 struct rmap_item
*next
; /* when stable */
126 struct rb_node node
; /* when tree node */
127 struct rmap_item
*prev
; /* in stable list */
131 #define SEQNR_MASK 0x0ff /* low bits of unstable tree seqnr */
132 #define NODE_FLAG 0x100 /* is a node of unstable or stable tree */
133 #define STABLE_FLAG 0x200 /* is a node or list item of stable tree */
135 /* The stable and unstable tree heads */
136 static struct rb_root root_stable_tree
= RB_ROOT
;
137 static struct rb_root root_unstable_tree
= RB_ROOT
;
139 #define MM_SLOTS_HASH_HEADS 1024
140 static struct hlist_head
*mm_slots_hash
;
142 static struct mm_slot ksm_mm_head
= {
143 .mm_list
= LIST_HEAD_INIT(ksm_mm_head
.mm_list
),
145 static struct ksm_scan ksm_scan
= {
146 .mm_slot
= &ksm_mm_head
,
149 static struct kmem_cache
*rmap_item_cache
;
150 static struct kmem_cache
*mm_slot_cache
;
152 /* The number of nodes in the stable tree */
153 static unsigned long ksm_pages_shared
;
155 /* The number of page slots additionally sharing those nodes */
156 static unsigned long ksm_pages_sharing
;
158 /* The number of nodes in the unstable tree */
159 static unsigned long ksm_pages_unshared
;
161 /* The number of rmap_items in use: to calculate pages_volatile */
162 static unsigned long ksm_rmap_items
;
164 /* Limit on the number of unswappable pages used */
165 static unsigned long ksm_max_kernel_pages
;
167 /* Number of pages ksmd should scan in one batch */
168 static unsigned int ksm_thread_pages_to_scan
;
170 /* Milliseconds ksmd should sleep between batches */
171 static unsigned int ksm_thread_sleep_millisecs
;
173 #define KSM_RUN_STOP 0
174 #define KSM_RUN_MERGE 1
175 #define KSM_RUN_UNMERGE 2
176 static unsigned int ksm_run
;
178 static DECLARE_WAIT_QUEUE_HEAD(ksm_thread_wait
);
179 static DEFINE_MUTEX(ksm_thread_mutex
);
180 static DEFINE_SPINLOCK(ksm_mmlist_lock
);
182 #define KSM_KMEM_CACHE(__struct, __flags) kmem_cache_create("ksm_"#__struct,\
183 sizeof(struct __struct), __alignof__(struct __struct),\
186 static int __init
ksm_slab_init(void)
188 rmap_item_cache
= KSM_KMEM_CACHE(rmap_item
, 0);
189 if (!rmap_item_cache
)
192 mm_slot_cache
= KSM_KMEM_CACHE(mm_slot
, 0);
199 kmem_cache_destroy(rmap_item_cache
);
204 static void __init
ksm_slab_free(void)
206 kmem_cache_destroy(mm_slot_cache
);
207 kmem_cache_destroy(rmap_item_cache
);
208 mm_slot_cache
= NULL
;
211 static inline struct rmap_item
*alloc_rmap_item(void)
213 struct rmap_item
*rmap_item
;
215 rmap_item
= kmem_cache_zalloc(rmap_item_cache
, GFP_KERNEL
);
221 static inline void free_rmap_item(struct rmap_item
*rmap_item
)
224 rmap_item
->mm
= NULL
; /* debug safety */
225 kmem_cache_free(rmap_item_cache
, rmap_item
);
228 static inline struct mm_slot
*alloc_mm_slot(void)
230 if (!mm_slot_cache
) /* initialization failed */
232 return kmem_cache_zalloc(mm_slot_cache
, GFP_KERNEL
);
235 static inline void free_mm_slot(struct mm_slot
*mm_slot
)
237 kmem_cache_free(mm_slot_cache
, mm_slot
);
240 static int __init
mm_slots_hash_init(void)
242 mm_slots_hash
= kzalloc(MM_SLOTS_HASH_HEADS
* sizeof(struct hlist_head
),
249 static void __init
mm_slots_hash_free(void)
251 kfree(mm_slots_hash
);
254 static struct mm_slot
*get_mm_slot(struct mm_struct
*mm
)
256 struct mm_slot
*mm_slot
;
257 struct hlist_head
*bucket
;
258 struct hlist_node
*node
;
260 bucket
= &mm_slots_hash
[((unsigned long)mm
/ sizeof(struct mm_struct
))
261 % MM_SLOTS_HASH_HEADS
];
262 hlist_for_each_entry(mm_slot
, node
, bucket
, link
) {
263 if (mm
== mm_slot
->mm
)
269 static void insert_to_mm_slots_hash(struct mm_struct
*mm
,
270 struct mm_slot
*mm_slot
)
272 struct hlist_head
*bucket
;
274 bucket
= &mm_slots_hash
[((unsigned long)mm
/ sizeof(struct mm_struct
))
275 % MM_SLOTS_HASH_HEADS
];
277 INIT_LIST_HEAD(&mm_slot
->rmap_list
);
278 hlist_add_head(&mm_slot
->link
, bucket
);
281 static inline int in_stable_tree(struct rmap_item
*rmap_item
)
283 return rmap_item
->address
& STABLE_FLAG
;
287 * We use break_ksm to break COW on a ksm page: it's a stripped down
289 * if (get_user_pages(current, mm, addr, 1, 1, 1, &page, NULL) == 1)
292 * but taking great care only to touch a ksm page, in a VM_MERGEABLE vma,
293 * in case the application has unmapped and remapped mm,addr meanwhile.
294 * Could a ksm page appear anywhere else? Actually yes, in a VM_PFNMAP
295 * mmap of /dev/mem or /dev/kmem, where we would not want to touch it.
297 static int break_ksm(struct vm_area_struct
*vma
, unsigned long addr
)
304 page
= follow_page(vma
, addr
, FOLL_GET
);
308 ret
= handle_mm_fault(vma
->vm_mm
, vma
, addr
,
311 ret
= VM_FAULT_WRITE
;
313 } while (!(ret
& (VM_FAULT_WRITE
| VM_FAULT_SIGBUS
| VM_FAULT_OOM
)));
315 * We must loop because handle_mm_fault() may back out if there's
316 * any difficulty e.g. if pte accessed bit gets updated concurrently.
318 * VM_FAULT_WRITE is what we have been hoping for: it indicates that
319 * COW has been broken, even if the vma does not permit VM_WRITE;
320 * but note that a concurrent fault might break PageKsm for us.
322 * VM_FAULT_SIGBUS could occur if we race with truncation of the
323 * backing file, which also invalidates anonymous pages: that's
324 * okay, that truncation will have unmapped the PageKsm for us.
326 * VM_FAULT_OOM: at the time of writing (late July 2009), setting
327 * aside mem_cgroup limits, VM_FAULT_OOM would only be set if the
328 * current task has TIF_MEMDIE set, and will be OOM killed on return
329 * to user; and ksmd, having no mm, would never be chosen for that.
331 * But if the mm is in a limited mem_cgroup, then the fault may fail
332 * with VM_FAULT_OOM even if the current task is not TIF_MEMDIE; and
333 * even ksmd can fail in this way - though it's usually breaking ksm
334 * just to undo a merge it made a moment before, so unlikely to oom.
336 * That's a pity: we might therefore have more kernel pages allocated
337 * than we're counting as nodes in the stable tree; but ksm_do_scan
338 * will retry to break_cow on each pass, so should recover the page
339 * in due course. The important thing is to not let VM_MERGEABLE
340 * be cleared while any such pages might remain in the area.
342 return (ret
& VM_FAULT_OOM
) ? -ENOMEM
: 0;
345 static void break_cow(struct mm_struct
*mm
, unsigned long addr
)
347 struct vm_area_struct
*vma
;
349 down_read(&mm
->mmap_sem
);
350 vma
= find_vma(mm
, addr
);
351 if (!vma
|| vma
->vm_start
> addr
)
353 if (!(vma
->vm_flags
& VM_MERGEABLE
) || !vma
->anon_vma
)
355 break_ksm(vma
, addr
);
357 up_read(&mm
->mmap_sem
);
360 static struct page
*get_mergeable_page(struct rmap_item
*rmap_item
)
362 struct mm_struct
*mm
= rmap_item
->mm
;
363 unsigned long addr
= rmap_item
->address
;
364 struct vm_area_struct
*vma
;
367 down_read(&mm
->mmap_sem
);
368 vma
= find_vma(mm
, addr
);
369 if (!vma
|| vma
->vm_start
> addr
)
371 if (!(vma
->vm_flags
& VM_MERGEABLE
) || !vma
->anon_vma
)
374 page
= follow_page(vma
, addr
, FOLL_GET
);
377 if (PageAnon(page
)) {
378 flush_anon_page(vma
, page
, addr
);
379 flush_dcache_page(page
);
384 up_read(&mm
->mmap_sem
);
389 * get_ksm_page: checks if the page at the virtual address in rmap_item
390 * is still PageKsm, in which case we can trust the content of the page,
391 * and it returns the gotten page; but NULL if the page has been zapped.
393 static struct page
*get_ksm_page(struct rmap_item
*rmap_item
)
397 page
= get_mergeable_page(rmap_item
);
398 if (page
&& !PageKsm(page
)) {
406 * Removing rmap_item from stable or unstable tree.
407 * This function will clean the information from the stable/unstable tree.
409 static void remove_rmap_item_from_tree(struct rmap_item
*rmap_item
)
411 if (in_stable_tree(rmap_item
)) {
412 struct rmap_item
*next_item
= rmap_item
->next
;
414 if (rmap_item
->address
& NODE_FLAG
) {
416 rb_replace_node(&rmap_item
->node
,
419 next_item
->address
|= NODE_FLAG
;
422 rb_erase(&rmap_item
->node
, &root_stable_tree
);
426 struct rmap_item
*prev_item
= rmap_item
->prev
;
428 BUG_ON(prev_item
->next
!= rmap_item
);
429 prev_item
->next
= next_item
;
431 BUG_ON(next_item
->prev
!= rmap_item
);
432 next_item
->prev
= rmap_item
->prev
;
437 rmap_item
->next
= NULL
;
439 } else if (rmap_item
->address
& NODE_FLAG
) {
442 * ksm_thread can and must skip the rb_erase, because
443 * root_unstable_tree was already reset to RB_ROOT.
444 * But __ksm_exit has to be careful: do the rb_erase
445 * if it's interrupting a scan, and this rmap_item was
446 * inserted by this scan rather than left from before.
448 age
= (unsigned char)(ksm_scan
.seqnr
- rmap_item
->address
);
451 rb_erase(&rmap_item
->node
, &root_unstable_tree
);
452 ksm_pages_unshared
--;
455 rmap_item
->address
&= PAGE_MASK
;
457 cond_resched(); /* we're called from many long loops */
460 static void remove_trailing_rmap_items(struct mm_slot
*mm_slot
,
461 struct list_head
*cur
)
463 struct rmap_item
*rmap_item
;
465 while (cur
!= &mm_slot
->rmap_list
) {
466 rmap_item
= list_entry(cur
, struct rmap_item
, link
);
468 remove_rmap_item_from_tree(rmap_item
);
469 list_del(&rmap_item
->link
);
470 free_rmap_item(rmap_item
);
475 * Though it's very tempting to unmerge in_stable_tree(rmap_item)s rather
476 * than check every pte of a given vma, the locking doesn't quite work for
477 * that - an rmap_item is assigned to the stable tree after inserting ksm
478 * page and upping mmap_sem. Nor does it fit with the way we skip dup'ing
479 * rmap_items from parent to child at fork time (so as not to waste time
480 * if exit comes before the next scan reaches it).
482 * Similarly, although we'd like to remove rmap_items (so updating counts
483 * and freeing memory) when unmerging an area, it's easier to leave that
484 * to the next pass of ksmd - consider, for example, how ksmd might be
485 * in cmp_and_merge_page on one of the rmap_items we would be removing.
487 static int unmerge_ksm_pages(struct vm_area_struct
*vma
,
488 unsigned long start
, unsigned long end
)
493 for (addr
= start
; addr
< end
&& !err
; addr
+= PAGE_SIZE
) {
494 if (signal_pending(current
))
497 err
= break_ksm(vma
, addr
);
502 static int unmerge_and_remove_all_rmap_items(void)
504 struct mm_slot
*mm_slot
;
505 struct mm_struct
*mm
;
506 struct vm_area_struct
*vma
;
509 spin_lock(&ksm_mmlist_lock
);
510 mm_slot
= list_entry(ksm_mm_head
.mm_list
.next
,
511 struct mm_slot
, mm_list
);
512 spin_unlock(&ksm_mmlist_lock
);
514 while (mm_slot
!= &ksm_mm_head
) {
516 down_read(&mm
->mmap_sem
);
517 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
518 if (!(vma
->vm_flags
& VM_MERGEABLE
) || !vma
->anon_vma
)
520 err
= unmerge_ksm_pages(vma
,
521 vma
->vm_start
, vma
->vm_end
);
523 up_read(&mm
->mmap_sem
);
527 remove_trailing_rmap_items(mm_slot
, mm_slot
->rmap_list
.next
);
528 up_read(&mm
->mmap_sem
);
530 spin_lock(&ksm_mmlist_lock
);
531 mm_slot
= list_entry(mm_slot
->mm_list
.next
,
532 struct mm_slot
, mm_list
);
533 spin_unlock(&ksm_mmlist_lock
);
538 spin_lock(&ksm_mmlist_lock
);
539 ksm_scan
.mm_slot
= &ksm_mm_head
;
540 spin_unlock(&ksm_mmlist_lock
);
544 static u32
calc_checksum(struct page
*page
)
547 void *addr
= kmap_atomic(page
, KM_USER0
);
548 checksum
= jhash2(addr
, PAGE_SIZE
/ 4, 17);
549 kunmap_atomic(addr
, KM_USER0
);
553 static int memcmp_pages(struct page
*page1
, struct page
*page2
)
558 addr1
= kmap_atomic(page1
, KM_USER0
);
559 addr2
= kmap_atomic(page2
, KM_USER1
);
560 ret
= memcmp(addr1
, addr2
, PAGE_SIZE
);
561 kunmap_atomic(addr2
, KM_USER1
);
562 kunmap_atomic(addr1
, KM_USER0
);
566 static inline int pages_identical(struct page
*page1
, struct page
*page2
)
568 return !memcmp_pages(page1
, page2
);
571 static int write_protect_page(struct vm_area_struct
*vma
, struct page
*page
,
574 struct mm_struct
*mm
= vma
->vm_mm
;
581 addr
= page_address_in_vma(page
, vma
);
585 ptep
= page_check_address(page
, mm
, addr
, &ptl
, 0);
589 if (pte_write(*ptep
)) {
592 swapped
= PageSwapCache(page
);
593 flush_cache_page(vma
, addr
, page_to_pfn(page
));
595 * Ok this is tricky, when get_user_pages_fast() run it doesnt
596 * take any lock, therefore the check that we are going to make
597 * with the pagecount against the mapcount is racey and
598 * O_DIRECT can happen right after the check.
599 * So we clear the pte and flush the tlb before the check
600 * this assure us that no O_DIRECT can happen after the check
601 * or in the middle of the check.
603 entry
= ptep_clear_flush(vma
, addr
, ptep
);
605 * Check that no O_DIRECT or similar I/O is in progress on the
608 if ((page_mapcount(page
) + 2 + swapped
) != page_count(page
)) {
609 set_pte_at_notify(mm
, addr
, ptep
, entry
);
612 entry
= pte_wrprotect(entry
);
613 set_pte_at_notify(mm
, addr
, ptep
, entry
);
619 pte_unmap_unlock(ptep
, ptl
);
625 * replace_page - replace page in vma by new ksm page
626 * @vma: vma that holds the pte pointing to oldpage
627 * @oldpage: the page we are replacing by newpage
628 * @newpage: the ksm page we replace oldpage by
629 * @orig_pte: the original value of the pte
631 * Returns 0 on success, -EFAULT on failure.
633 static int replace_page(struct vm_area_struct
*vma
, struct page
*oldpage
,
634 struct page
*newpage
, pte_t orig_pte
)
636 struct mm_struct
*mm
= vma
->vm_mm
;
646 prot
= vm_get_page_prot(vma
->vm_flags
& ~VM_WRITE
);
648 addr
= page_address_in_vma(oldpage
, vma
);
652 pgd
= pgd_offset(mm
, addr
);
653 if (!pgd_present(*pgd
))
656 pud
= pud_offset(pgd
, addr
);
657 if (!pud_present(*pud
))
660 pmd
= pmd_offset(pud
, addr
);
661 if (!pmd_present(*pmd
))
664 ptep
= pte_offset_map_lock(mm
, pmd
, addr
, &ptl
);
665 if (!pte_same(*ptep
, orig_pte
)) {
666 pte_unmap_unlock(ptep
, ptl
);
671 page_add_ksm_rmap(newpage
);
673 flush_cache_page(vma
, addr
, pte_pfn(*ptep
));
674 ptep_clear_flush(vma
, addr
, ptep
);
675 set_pte_at_notify(mm
, addr
, ptep
, mk_pte(newpage
, prot
));
677 page_remove_rmap(oldpage
);
680 pte_unmap_unlock(ptep
, ptl
);
687 * try_to_merge_one_page - take two pages and merge them into one
688 * @vma: the vma that hold the pte pointing into oldpage
689 * @oldpage: the page that we want to replace with newpage
690 * @newpage: the page that we want to map instead of oldpage
693 * oldpage should be a PageAnon page, while newpage should be a PageKsm page,
694 * or a newly allocated kernel page which page_add_ksm_rmap will make PageKsm.
696 * This function returns 0 if the pages were merged, -EFAULT otherwise.
698 static int try_to_merge_one_page(struct vm_area_struct
*vma
,
699 struct page
*oldpage
,
700 struct page
*newpage
)
702 pte_t orig_pte
= __pte(0);
705 if (!(vma
->vm_flags
& VM_MERGEABLE
))
708 if (!PageAnon(oldpage
))
715 * We need the page lock to read a stable PageSwapCache in
716 * write_protect_page(). We use trylock_page() instead of
717 * lock_page() because we don't want to wait here - we
718 * prefer to continue scanning and merging different pages,
719 * then come back to this page when it is unlocked.
721 if (!trylock_page(oldpage
))
724 * If this anonymous page is mapped only here, its pte may need
725 * to be write-protected. If it's mapped elsewhere, all of its
726 * ptes are necessarily already write-protected. But in either
727 * case, we need to lock and check page_count is not raised.
729 if (write_protect_page(vma
, oldpage
, &orig_pte
)) {
730 unlock_page(oldpage
);
733 unlock_page(oldpage
);
735 if (pages_identical(oldpage
, newpage
))
736 err
= replace_page(vma
, oldpage
, newpage
, orig_pte
);
746 * try_to_merge_with_ksm_page - like try_to_merge_two_pages,
747 * but no new kernel page is allocated: kpage must already be a ksm page.
749 static int try_to_merge_with_ksm_page(struct mm_struct
*mm1
,
754 struct vm_area_struct
*vma
;
757 down_read(&mm1
->mmap_sem
);
758 vma
= find_vma(mm1
, addr1
);
759 if (!vma
|| vma
->vm_start
> addr1
)
762 err
= try_to_merge_one_page(vma
, page1
, kpage
);
764 up_read(&mm1
->mmap_sem
);
769 * try_to_merge_two_pages - take two identical pages and prepare them
770 * to be merged into one page.
772 * This function returns 0 if we successfully mapped two identical pages
773 * into one page, -EFAULT otherwise.
775 * Note that this function allocates a new kernel page: if one of the pages
776 * is already a ksm page, try_to_merge_with_ksm_page should be used.
778 static int try_to_merge_two_pages(struct mm_struct
*mm1
, unsigned long addr1
,
779 struct page
*page1
, struct mm_struct
*mm2
,
780 unsigned long addr2
, struct page
*page2
)
782 struct vm_area_struct
*vma
;
787 * The number of nodes in the stable tree
788 * is the number of kernel pages that we hold.
790 if (ksm_max_kernel_pages
&&
791 ksm_max_kernel_pages
<= ksm_pages_shared
)
794 kpage
= alloc_page(GFP_HIGHUSER
);
798 down_read(&mm1
->mmap_sem
);
799 vma
= find_vma(mm1
, addr1
);
800 if (!vma
|| vma
->vm_start
> addr1
) {
801 up_read(&mm1
->mmap_sem
);
805 copy_user_highpage(kpage
, page1
, addr1
, vma
);
806 err
= try_to_merge_one_page(vma
, page1
, kpage
);
807 up_read(&mm1
->mmap_sem
);
810 err
= try_to_merge_with_ksm_page(mm2
, addr2
, page2
, kpage
);
812 * If that fails, we have a ksm page with only one pte
813 * pointing to it: so break it.
816 break_cow(mm1
, addr1
);
824 * stable_tree_search - search page inside the stable tree
825 * @page: the page that we are searching identical pages to.
826 * @page2: pointer into identical page that we are holding inside the stable
827 * tree that we have found.
828 * @rmap_item: the reverse mapping item
830 * This function checks if there is a page inside the stable tree
831 * with identical content to the page that we are scanning right now.
833 * This function return rmap_item pointer to the identical item if found,
836 static struct rmap_item
*stable_tree_search(struct page
*page
,
838 struct rmap_item
*rmap_item
)
840 struct rb_node
*node
= root_stable_tree
.rb_node
;
843 struct rmap_item
*tree_rmap_item
, *next_rmap_item
;
846 tree_rmap_item
= rb_entry(node
, struct rmap_item
, node
);
847 while (tree_rmap_item
) {
848 BUG_ON(!in_stable_tree(tree_rmap_item
));
850 page2
[0] = get_ksm_page(tree_rmap_item
);
853 next_rmap_item
= tree_rmap_item
->next
;
854 remove_rmap_item_from_tree(tree_rmap_item
);
855 tree_rmap_item
= next_rmap_item
;
860 ret
= memcmp_pages(page
, page2
[0]);
864 node
= node
->rb_left
;
865 } else if (ret
> 0) {
867 node
= node
->rb_right
;
869 return tree_rmap_item
;
877 * stable_tree_insert - insert rmap_item pointing to new ksm page
878 * into the stable tree.
880 * @page: the page that we are searching identical page to inside the stable
882 * @rmap_item: pointer to the reverse mapping item.
884 * This function returns rmap_item if success, NULL otherwise.
886 static struct rmap_item
*stable_tree_insert(struct page
*page
,
887 struct rmap_item
*rmap_item
)
889 struct rb_node
**new = &root_stable_tree
.rb_node
;
890 struct rb_node
*parent
= NULL
;
893 struct rmap_item
*tree_rmap_item
, *next_rmap_item
;
894 struct page
*tree_page
;
897 tree_rmap_item
= rb_entry(*new, struct rmap_item
, node
);
898 while (tree_rmap_item
) {
899 BUG_ON(!in_stable_tree(tree_rmap_item
));
901 tree_page
= get_ksm_page(tree_rmap_item
);
904 next_rmap_item
= tree_rmap_item
->next
;
905 remove_rmap_item_from_tree(tree_rmap_item
);
906 tree_rmap_item
= next_rmap_item
;
911 ret
= memcmp_pages(page
, tree_page
);
916 new = &parent
->rb_left
;
918 new = &parent
->rb_right
;
921 * It is not a bug that stable_tree_search() didn't
922 * find this node: because at that time our page was
923 * not yet write-protected, so may have changed since.
929 rmap_item
->address
|= NODE_FLAG
| STABLE_FLAG
;
930 rmap_item
->next
= NULL
;
931 rb_link_node(&rmap_item
->node
, parent
, new);
932 rb_insert_color(&rmap_item
->node
, &root_stable_tree
);
939 * unstable_tree_search_insert - search and insert items into the unstable tree.
941 * @page: the page that we are going to search for identical page or to insert
942 * into the unstable tree
943 * @page2: pointer into identical page that was found inside the unstable tree
944 * @rmap_item: the reverse mapping item of page
946 * This function searches for a page in the unstable tree identical to the
947 * page currently being scanned; and if no identical page is found in the
948 * tree, we insert rmap_item as a new object into the unstable tree.
950 * This function returns pointer to rmap_item found to be identical
951 * to the currently scanned page, NULL otherwise.
953 * This function does both searching and inserting, because they share
954 * the same walking algorithm in an rbtree.
956 static struct rmap_item
*unstable_tree_search_insert(struct page
*page
,
958 struct rmap_item
*rmap_item
)
960 struct rb_node
**new = &root_unstable_tree
.rb_node
;
961 struct rb_node
*parent
= NULL
;
964 struct rmap_item
*tree_rmap_item
;
967 tree_rmap_item
= rb_entry(*new, struct rmap_item
, node
);
968 page2
[0] = get_mergeable_page(tree_rmap_item
);
973 * Don't substitute an unswappable ksm page
974 * just for one good swappable forked page.
976 if (page
== page2
[0]) {
981 ret
= memcmp_pages(page
, page2
[0]);
986 new = &parent
->rb_left
;
987 } else if (ret
> 0) {
989 new = &parent
->rb_right
;
991 return tree_rmap_item
;
995 rmap_item
->address
|= NODE_FLAG
;
996 rmap_item
->address
|= (ksm_scan
.seqnr
& SEQNR_MASK
);
997 rb_link_node(&rmap_item
->node
, parent
, new);
998 rb_insert_color(&rmap_item
->node
, &root_unstable_tree
);
1000 ksm_pages_unshared
++;
1005 * stable_tree_append - add another rmap_item to the linked list of
1006 * rmap_items hanging off a given node of the stable tree, all sharing
1007 * the same ksm page.
1009 static void stable_tree_append(struct rmap_item
*rmap_item
,
1010 struct rmap_item
*tree_rmap_item
)
1012 rmap_item
->next
= tree_rmap_item
->next
;
1013 rmap_item
->prev
= tree_rmap_item
;
1015 if (tree_rmap_item
->next
)
1016 tree_rmap_item
->next
->prev
= rmap_item
;
1018 tree_rmap_item
->next
= rmap_item
;
1019 rmap_item
->address
|= STABLE_FLAG
;
1021 ksm_pages_sharing
++;
1025 * cmp_and_merge_page - first see if page can be merged into the stable tree;
1026 * if not, compare checksum to previous and if it's the same, see if page can
1027 * be inserted into the unstable tree, or merged with a page already there and
1028 * both transferred to the stable tree.
1030 * @page: the page that we are searching identical page to.
1031 * @rmap_item: the reverse mapping into the virtual address of this page
1033 static void cmp_and_merge_page(struct page
*page
, struct rmap_item
*rmap_item
)
1035 struct page
*page2
[1];
1036 struct rmap_item
*tree_rmap_item
;
1037 unsigned int checksum
;
1040 if (in_stable_tree(rmap_item
))
1041 remove_rmap_item_from_tree(rmap_item
);
1043 /* We first start with searching the page inside the stable tree */
1044 tree_rmap_item
= stable_tree_search(page
, page2
, rmap_item
);
1045 if (tree_rmap_item
) {
1046 if (page
== page2
[0]) /* forked */
1049 err
= try_to_merge_with_ksm_page(rmap_item
->mm
,
1056 * The page was successfully merged:
1057 * add its rmap_item to the stable tree.
1059 stable_tree_append(rmap_item
, tree_rmap_item
);
1065 * A ksm page might have got here by fork, but its other
1066 * references have already been removed from the stable tree.
1067 * Or it might be left over from a break_ksm which failed
1068 * when the mem_cgroup had reached its limit: try again now.
1071 break_cow(rmap_item
->mm
, rmap_item
->address
);
1074 * In case the hash value of the page was changed from the last time we
1075 * have calculated it, this page to be changed frequely, therefore we
1076 * don't want to insert it to the unstable tree, and we don't want to
1077 * waste our time to search if there is something identical to it there.
1079 checksum
= calc_checksum(page
);
1080 if (rmap_item
->oldchecksum
!= checksum
) {
1081 rmap_item
->oldchecksum
= checksum
;
1085 tree_rmap_item
= unstable_tree_search_insert(page
, page2
, rmap_item
);
1086 if (tree_rmap_item
) {
1087 err
= try_to_merge_two_pages(rmap_item
->mm
,
1088 rmap_item
->address
, page
,
1090 tree_rmap_item
->address
, page2
[0]);
1092 * As soon as we merge this page, we want to remove the
1093 * rmap_item of the page we have merged with from the unstable
1094 * tree, and insert it instead as new node in the stable tree.
1097 rb_erase(&tree_rmap_item
->node
, &root_unstable_tree
);
1098 tree_rmap_item
->address
&= ~NODE_FLAG
;
1099 ksm_pages_unshared
--;
1102 * If we fail to insert the page into the stable tree,
1103 * we will have 2 virtual addresses that are pointing
1104 * to a ksm page left outside the stable tree,
1105 * in which case we need to break_cow on both.
1107 if (stable_tree_insert(page2
[0], tree_rmap_item
))
1108 stable_tree_append(rmap_item
, tree_rmap_item
);
1110 break_cow(tree_rmap_item
->mm
,
1111 tree_rmap_item
->address
);
1112 break_cow(rmap_item
->mm
, rmap_item
->address
);
1120 static struct rmap_item
*get_next_rmap_item(struct mm_slot
*mm_slot
,
1121 struct list_head
*cur
,
1124 struct rmap_item
*rmap_item
;
1126 while (cur
!= &mm_slot
->rmap_list
) {
1127 rmap_item
= list_entry(cur
, struct rmap_item
, link
);
1128 if ((rmap_item
->address
& PAGE_MASK
) == addr
) {
1129 if (!in_stable_tree(rmap_item
))
1130 remove_rmap_item_from_tree(rmap_item
);
1133 if (rmap_item
->address
> addr
)
1136 remove_rmap_item_from_tree(rmap_item
);
1137 list_del(&rmap_item
->link
);
1138 free_rmap_item(rmap_item
);
1141 rmap_item
= alloc_rmap_item();
1143 /* It has already been zeroed */
1144 rmap_item
->mm
= mm_slot
->mm
;
1145 rmap_item
->address
= addr
;
1146 list_add_tail(&rmap_item
->link
, cur
);
1151 static struct rmap_item
*scan_get_next_rmap_item(struct page
**page
)
1153 struct mm_struct
*mm
;
1154 struct mm_slot
*slot
;
1155 struct vm_area_struct
*vma
;
1156 struct rmap_item
*rmap_item
;
1158 if (list_empty(&ksm_mm_head
.mm_list
))
1161 slot
= ksm_scan
.mm_slot
;
1162 if (slot
== &ksm_mm_head
) {
1163 root_unstable_tree
= RB_ROOT
;
1165 spin_lock(&ksm_mmlist_lock
);
1166 slot
= list_entry(slot
->mm_list
.next
, struct mm_slot
, mm_list
);
1167 ksm_scan
.mm_slot
= slot
;
1168 spin_unlock(&ksm_mmlist_lock
);
1170 ksm_scan
.address
= 0;
1171 ksm_scan
.rmap_item
= list_entry(&slot
->rmap_list
,
1172 struct rmap_item
, link
);
1176 down_read(&mm
->mmap_sem
);
1177 for (vma
= find_vma(mm
, ksm_scan
.address
); vma
; vma
= vma
->vm_next
) {
1178 if (!(vma
->vm_flags
& VM_MERGEABLE
))
1180 if (ksm_scan
.address
< vma
->vm_start
)
1181 ksm_scan
.address
= vma
->vm_start
;
1183 ksm_scan
.address
= vma
->vm_end
;
1185 while (ksm_scan
.address
< vma
->vm_end
) {
1186 *page
= follow_page(vma
, ksm_scan
.address
, FOLL_GET
);
1187 if (*page
&& PageAnon(*page
)) {
1188 flush_anon_page(vma
, *page
, ksm_scan
.address
);
1189 flush_dcache_page(*page
);
1190 rmap_item
= get_next_rmap_item(slot
,
1191 ksm_scan
.rmap_item
->link
.next
,
1194 ksm_scan
.rmap_item
= rmap_item
;
1195 ksm_scan
.address
+= PAGE_SIZE
;
1198 up_read(&mm
->mmap_sem
);
1203 ksm_scan
.address
+= PAGE_SIZE
;
1209 * Nuke all the rmap_items that are above this current rmap:
1210 * because there were no VM_MERGEABLE vmas with such addresses.
1212 remove_trailing_rmap_items(slot
, ksm_scan
.rmap_item
->link
.next
);
1214 spin_lock(&ksm_mmlist_lock
);
1215 ksm_scan
.mm_slot
= list_entry(slot
->mm_list
.next
,
1216 struct mm_slot
, mm_list
);
1217 if (ksm_scan
.address
== 0) {
1219 * We've completed a full scan of all vmas, holding mmap_sem
1220 * throughout, and found no VM_MERGEABLE: so do the same as
1221 * __ksm_exit does to remove this mm from all our lists now.
1223 hlist_del(&slot
->link
);
1224 list_del(&slot
->mm_list
);
1226 clear_bit(MMF_VM_MERGEABLE
, &mm
->flags
);
1228 spin_unlock(&ksm_mmlist_lock
);
1229 up_read(&mm
->mmap_sem
);
1231 /* Repeat until we've completed scanning the whole list */
1232 slot
= ksm_scan
.mm_slot
;
1233 if (slot
!= &ksm_mm_head
)
1237 * Bump seqnr here rather than at top, so that __ksm_exit
1238 * can skip rb_erase on unstable tree until we run again.
1245 * ksm_do_scan - the ksm scanner main worker function.
1246 * @scan_npages - number of pages we want to scan before we return.
1248 static void ksm_do_scan(unsigned int scan_npages
)
1250 struct rmap_item
*rmap_item
;
1253 while (scan_npages
--) {
1255 rmap_item
= scan_get_next_rmap_item(&page
);
1258 if (!PageKsm(page
) || !in_stable_tree(rmap_item
))
1259 cmp_and_merge_page(page
, rmap_item
);
1260 else if (page_mapcount(page
) == 1) {
1262 * Replace now-unshared ksm page by ordinary page.
1264 break_cow(rmap_item
->mm
, rmap_item
->address
);
1265 remove_rmap_item_from_tree(rmap_item
);
1266 rmap_item
->oldchecksum
= calc_checksum(page
);
1272 static int ksmd_should_run(void)
1274 return (ksm_run
& KSM_RUN_MERGE
) && !list_empty(&ksm_mm_head
.mm_list
);
1277 static int ksm_scan_thread(void *nothing
)
1279 set_user_nice(current
, 5);
1281 while (!kthread_should_stop()) {
1282 mutex_lock(&ksm_thread_mutex
);
1283 if (ksmd_should_run())
1284 ksm_do_scan(ksm_thread_pages_to_scan
);
1285 mutex_unlock(&ksm_thread_mutex
);
1287 if (ksmd_should_run()) {
1288 schedule_timeout_interruptible(
1289 msecs_to_jiffies(ksm_thread_sleep_millisecs
));
1291 wait_event_interruptible(ksm_thread_wait
,
1292 ksmd_should_run() || kthread_should_stop());
1298 int ksm_madvise(struct vm_area_struct
*vma
, unsigned long start
,
1299 unsigned long end
, int advice
, unsigned long *vm_flags
)
1301 struct mm_struct
*mm
= vma
->vm_mm
;
1305 case MADV_MERGEABLE
:
1307 * Be somewhat over-protective for now!
1309 if (*vm_flags
& (VM_MERGEABLE
| VM_SHARED
| VM_MAYSHARE
|
1310 VM_PFNMAP
| VM_IO
| VM_DONTEXPAND
|
1311 VM_RESERVED
| VM_HUGETLB
| VM_INSERTPAGE
|
1312 VM_MIXEDMAP
| VM_SAO
))
1313 return 0; /* just ignore the advice */
1315 if (!test_bit(MMF_VM_MERGEABLE
, &mm
->flags
)) {
1316 err
= __ksm_enter(mm
);
1321 *vm_flags
|= VM_MERGEABLE
;
1324 case MADV_UNMERGEABLE
:
1325 if (!(*vm_flags
& VM_MERGEABLE
))
1326 return 0; /* just ignore the advice */
1328 if (vma
->anon_vma
) {
1329 err
= unmerge_ksm_pages(vma
, start
, end
);
1334 *vm_flags
&= ~VM_MERGEABLE
;
1341 int __ksm_enter(struct mm_struct
*mm
)
1343 struct mm_slot
*mm_slot
;
1346 mm_slot
= alloc_mm_slot();
1350 /* Check ksm_run too? Would need tighter locking */
1351 needs_wakeup
= list_empty(&ksm_mm_head
.mm_list
);
1353 spin_lock(&ksm_mmlist_lock
);
1354 insert_to_mm_slots_hash(mm
, mm_slot
);
1356 * Insert just behind the scanning cursor, to let the area settle
1357 * down a little; when fork is followed by immediate exec, we don't
1358 * want ksmd to waste time setting up and tearing down an rmap_list.
1360 list_add_tail(&mm_slot
->mm_list
, &ksm_scan
.mm_slot
->mm_list
);
1361 spin_unlock(&ksm_mmlist_lock
);
1363 set_bit(MMF_VM_MERGEABLE
, &mm
->flags
);
1366 wake_up_interruptible(&ksm_thread_wait
);
1371 void __ksm_exit(struct mm_struct
*mm
)
1373 struct mm_slot
*mm_slot
;
1376 * This process is exiting: doesn't hold and doesn't need mmap_sem;
1377 * but we do need to exclude ksmd and other exiters while we modify
1378 * the various lists and trees.
1380 mutex_lock(&ksm_thread_mutex
);
1381 spin_lock(&ksm_mmlist_lock
);
1382 mm_slot
= get_mm_slot(mm
);
1383 if (!list_empty(&mm_slot
->rmap_list
)) {
1384 spin_unlock(&ksm_mmlist_lock
);
1385 remove_trailing_rmap_items(mm_slot
, mm_slot
->rmap_list
.next
);
1386 spin_lock(&ksm_mmlist_lock
);
1389 if (ksm_scan
.mm_slot
== mm_slot
) {
1390 ksm_scan
.mm_slot
= list_entry(
1391 mm_slot
->mm_list
.next
, struct mm_slot
, mm_list
);
1392 ksm_scan
.address
= 0;
1393 ksm_scan
.rmap_item
= list_entry(
1394 &ksm_scan
.mm_slot
->rmap_list
, struct rmap_item
, link
);
1395 if (ksm_scan
.mm_slot
== &ksm_mm_head
)
1399 hlist_del(&mm_slot
->link
);
1400 list_del(&mm_slot
->mm_list
);
1401 spin_unlock(&ksm_mmlist_lock
);
1403 free_mm_slot(mm_slot
);
1404 clear_bit(MMF_VM_MERGEABLE
, &mm
->flags
);
1405 mutex_unlock(&ksm_thread_mutex
);
1408 #define KSM_ATTR_RO(_name) \
1409 static struct kobj_attribute _name##_attr = __ATTR_RO(_name)
1410 #define KSM_ATTR(_name) \
1411 static struct kobj_attribute _name##_attr = \
1412 __ATTR(_name, 0644, _name##_show, _name##_store)
1414 static ssize_t
sleep_millisecs_show(struct kobject
*kobj
,
1415 struct kobj_attribute
*attr
, char *buf
)
1417 return sprintf(buf
, "%u\n", ksm_thread_sleep_millisecs
);
1420 static ssize_t
sleep_millisecs_store(struct kobject
*kobj
,
1421 struct kobj_attribute
*attr
,
1422 const char *buf
, size_t count
)
1424 unsigned long msecs
;
1427 err
= strict_strtoul(buf
, 10, &msecs
);
1428 if (err
|| msecs
> UINT_MAX
)
1431 ksm_thread_sleep_millisecs
= msecs
;
1435 KSM_ATTR(sleep_millisecs
);
1437 static ssize_t
pages_to_scan_show(struct kobject
*kobj
,
1438 struct kobj_attribute
*attr
, char *buf
)
1440 return sprintf(buf
, "%u\n", ksm_thread_pages_to_scan
);
1443 static ssize_t
pages_to_scan_store(struct kobject
*kobj
,
1444 struct kobj_attribute
*attr
,
1445 const char *buf
, size_t count
)
1448 unsigned long nr_pages
;
1450 err
= strict_strtoul(buf
, 10, &nr_pages
);
1451 if (err
|| nr_pages
> UINT_MAX
)
1454 ksm_thread_pages_to_scan
= nr_pages
;
1458 KSM_ATTR(pages_to_scan
);
1460 static ssize_t
run_show(struct kobject
*kobj
, struct kobj_attribute
*attr
,
1463 return sprintf(buf
, "%u\n", ksm_run
);
1466 static ssize_t
run_store(struct kobject
*kobj
, struct kobj_attribute
*attr
,
1467 const char *buf
, size_t count
)
1470 unsigned long flags
;
1472 err
= strict_strtoul(buf
, 10, &flags
);
1473 if (err
|| flags
> UINT_MAX
)
1475 if (flags
> KSM_RUN_UNMERGE
)
1479 * KSM_RUN_MERGE sets ksmd running, and 0 stops it running.
1480 * KSM_RUN_UNMERGE stops it running and unmerges all rmap_items,
1481 * breaking COW to free the unswappable pages_shared (but leaves
1482 * mm_slots on the list for when ksmd may be set running again).
1485 mutex_lock(&ksm_thread_mutex
);
1486 if (ksm_run
!= flags
) {
1488 if (flags
& KSM_RUN_UNMERGE
) {
1489 err
= unmerge_and_remove_all_rmap_items();
1491 ksm_run
= KSM_RUN_STOP
;
1496 mutex_unlock(&ksm_thread_mutex
);
1498 if (flags
& KSM_RUN_MERGE
)
1499 wake_up_interruptible(&ksm_thread_wait
);
1505 static ssize_t
max_kernel_pages_store(struct kobject
*kobj
,
1506 struct kobj_attribute
*attr
,
1507 const char *buf
, size_t count
)
1510 unsigned long nr_pages
;
1512 err
= strict_strtoul(buf
, 10, &nr_pages
);
1516 ksm_max_kernel_pages
= nr_pages
;
1521 static ssize_t
max_kernel_pages_show(struct kobject
*kobj
,
1522 struct kobj_attribute
*attr
, char *buf
)
1524 return sprintf(buf
, "%lu\n", ksm_max_kernel_pages
);
1526 KSM_ATTR(max_kernel_pages
);
1528 static ssize_t
pages_shared_show(struct kobject
*kobj
,
1529 struct kobj_attribute
*attr
, char *buf
)
1531 return sprintf(buf
, "%lu\n", ksm_pages_shared
);
1533 KSM_ATTR_RO(pages_shared
);
1535 static ssize_t
pages_sharing_show(struct kobject
*kobj
,
1536 struct kobj_attribute
*attr
, char *buf
)
1538 return sprintf(buf
, "%lu\n", ksm_pages_sharing
);
1540 KSM_ATTR_RO(pages_sharing
);
1542 static ssize_t
pages_unshared_show(struct kobject
*kobj
,
1543 struct kobj_attribute
*attr
, char *buf
)
1545 return sprintf(buf
, "%lu\n", ksm_pages_unshared
);
1547 KSM_ATTR_RO(pages_unshared
);
1549 static ssize_t
pages_volatile_show(struct kobject
*kobj
,
1550 struct kobj_attribute
*attr
, char *buf
)
1552 long ksm_pages_volatile
;
1554 ksm_pages_volatile
= ksm_rmap_items
- ksm_pages_shared
1555 - ksm_pages_sharing
- ksm_pages_unshared
;
1557 * It was not worth any locking to calculate that statistic,
1558 * but it might therefore sometimes be negative: conceal that.
1560 if (ksm_pages_volatile
< 0)
1561 ksm_pages_volatile
= 0;
1562 return sprintf(buf
, "%ld\n", ksm_pages_volatile
);
1564 KSM_ATTR_RO(pages_volatile
);
1566 static ssize_t
full_scans_show(struct kobject
*kobj
,
1567 struct kobj_attribute
*attr
, char *buf
)
1569 return sprintf(buf
, "%lu\n", ksm_scan
.seqnr
);
1571 KSM_ATTR_RO(full_scans
);
1573 static struct attribute
*ksm_attrs
[] = {
1574 &sleep_millisecs_attr
.attr
,
1575 &pages_to_scan_attr
.attr
,
1577 &max_kernel_pages_attr
.attr
,
1578 &pages_shared_attr
.attr
,
1579 &pages_sharing_attr
.attr
,
1580 &pages_unshared_attr
.attr
,
1581 &pages_volatile_attr
.attr
,
1582 &full_scans_attr
.attr
,
1586 static struct attribute_group ksm_attr_group
= {
1591 static int __init
ksm_init(void)
1593 struct task_struct
*ksm_thread
;
1596 err
= ksm_slab_init();
1600 err
= mm_slots_hash_init();
1604 ksm_thread
= kthread_run(ksm_scan_thread
, NULL
, "ksmd");
1605 if (IS_ERR(ksm_thread
)) {
1606 printk(KERN_ERR
"ksm: creating kthread failed\n");
1607 err
= PTR_ERR(ksm_thread
);
1611 err
= sysfs_create_group(mm_kobj
, &ksm_attr_group
);
1613 printk(KERN_ERR
"ksm: register sysfs failed\n");
1620 kthread_stop(ksm_thread
);
1622 mm_slots_hash_free();
1628 module_init(ksm_init
)