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/memory.h>
33 #include <linux/mmu_notifier.h>
34 #include <linux/swap.h>
35 #include <linux/ksm.h>
36 #include <linux/hashtable.h>
37 #include <linux/freezer.h>
38 #include <linux/oom.h>
39 #include <linux/numa.h>
41 #include <asm/tlbflush.h>
46 #define DO_NUMA(x) do { (x); } while (0)
49 #define DO_NUMA(x) do { } while (0)
53 * A few notes about the KSM scanning process,
54 * to make it easier to understand the data structures below:
56 * In order to reduce excessive scanning, KSM sorts the memory pages by their
57 * contents into a data structure that holds pointers to the pages' locations.
59 * Since the contents of the pages may change at any moment, KSM cannot just
60 * insert the pages into a normal sorted tree and expect it to find anything.
61 * Therefore KSM uses two data structures - the stable and the unstable tree.
63 * The stable tree holds pointers to all the merged pages (ksm pages), sorted
64 * by their contents. Because each such page is write-protected, searching on
65 * this tree is fully assured to be working (except when pages are unmapped),
66 * and therefore this tree is called the stable tree.
68 * In addition to the stable tree, KSM uses a second data structure called the
69 * unstable tree: this tree holds pointers to pages which have been found to
70 * be "unchanged for a period of time". The unstable tree sorts these pages
71 * by their contents, but since they are not write-protected, KSM cannot rely
72 * upon the unstable tree to work correctly - the unstable tree is liable to
73 * be corrupted as its contents are modified, and so it is called unstable.
75 * KSM solves this problem by several techniques:
77 * 1) The unstable tree is flushed every time KSM completes scanning all
78 * memory areas, and then the tree is rebuilt again from the beginning.
79 * 2) KSM will only insert into the unstable tree, pages whose hash value
80 * has not changed since the previous scan of all memory areas.
81 * 3) The unstable tree is a RedBlack Tree - so its balancing is based on the
82 * colors of the nodes and not on their contents, assuring that even when
83 * the tree gets "corrupted" it won't get out of balance, so scanning time
84 * remains the same (also, searching and inserting nodes in an rbtree uses
85 * the same algorithm, so we have no overhead when we flush and rebuild).
86 * 4) KSM never flushes the stable tree, which means that even if it were to
87 * take 10 attempts to find a page in the unstable tree, once it is found,
88 * it is secured in the stable tree. (When we scan a new page, we first
89 * compare it against the stable tree, and then against the unstable tree.)
91 * If the merge_across_nodes tunable is unset, then KSM maintains multiple
92 * stable trees and multiple unstable trees: one of each for each NUMA node.
96 * struct mm_slot - ksm information per mm that is being scanned
97 * @link: link to the mm_slots hash list
98 * @mm_list: link into the mm_slots list, rooted in ksm_mm_head
99 * @rmap_list: head for this mm_slot's singly-linked list of rmap_items
100 * @mm: the mm that this information is valid for
103 struct hlist_node link
;
104 struct list_head mm_list
;
105 struct rmap_item
*rmap_list
;
106 struct mm_struct
*mm
;
110 * struct ksm_scan - cursor for scanning
111 * @mm_slot: the current mm_slot we are scanning
112 * @address: the next address inside that to be scanned
113 * @rmap_list: link to the next rmap to be scanned in the rmap_list
114 * @seqnr: count of completed full scans (needed when removing unstable node)
116 * There is only the one ksm_scan instance of this cursor structure.
119 struct mm_slot
*mm_slot
;
120 unsigned long address
;
121 struct rmap_item
**rmap_list
;
126 * struct stable_node - node of the stable rbtree
127 * @node: rb node of this ksm page in the stable tree
128 * @head: (overlaying parent) &migrate_nodes indicates temporarily on that list
129 * @list: linked into migrate_nodes, pending placement in the proper node tree
130 * @hlist: hlist head of rmap_items using this ksm page
131 * @kpfn: page frame number of this ksm page (perhaps temporarily on wrong nid)
132 * @nid: NUMA node id of stable tree in which linked (may not match kpfn)
136 struct rb_node node
; /* when node of stable tree */
137 struct { /* when listed for migration */
138 struct list_head
*head
;
139 struct list_head list
;
142 struct hlist_head hlist
;
150 * struct rmap_item - reverse mapping item for virtual addresses
151 * @rmap_list: next rmap_item in mm_slot's singly-linked rmap_list
152 * @anon_vma: pointer to anon_vma for this mm,address, when in stable tree
153 * @nid: NUMA node id of unstable tree in which linked (may not match page)
154 * @mm: the memory structure this rmap_item is pointing into
155 * @address: the virtual address this rmap_item tracks (+ flags in low bits)
156 * @oldchecksum: previous checksum of the page at that virtual address
157 * @node: rb node of this rmap_item in the unstable tree
158 * @head: pointer to stable_node heading this list in the stable tree
159 * @hlist: link into hlist of rmap_items hanging off that stable_node
162 struct rmap_item
*rmap_list
;
164 struct anon_vma
*anon_vma
; /* when stable */
166 int nid
; /* when node of unstable tree */
169 struct mm_struct
*mm
;
170 unsigned long address
; /* + low bits used for flags below */
171 unsigned int oldchecksum
; /* when unstable */
173 struct rb_node node
; /* when node of unstable tree */
174 struct { /* when listed from stable tree */
175 struct stable_node
*head
;
176 struct hlist_node hlist
;
181 #define SEQNR_MASK 0x0ff /* low bits of unstable tree seqnr */
182 #define UNSTABLE_FLAG 0x100 /* is a node of the unstable tree */
183 #define STABLE_FLAG 0x200 /* is listed from the stable tree */
185 /* The stable and unstable tree heads */
186 static struct rb_root root_unstable_tree
[MAX_NUMNODES
];
187 static struct rb_root root_stable_tree
[MAX_NUMNODES
];
189 /* Recently migrated nodes of stable tree, pending proper placement */
190 static LIST_HEAD(migrate_nodes
);
192 #define MM_SLOTS_HASH_BITS 10
193 static DEFINE_HASHTABLE(mm_slots_hash
, MM_SLOTS_HASH_BITS
);
195 static struct mm_slot ksm_mm_head
= {
196 .mm_list
= LIST_HEAD_INIT(ksm_mm_head
.mm_list
),
198 static struct ksm_scan ksm_scan
= {
199 .mm_slot
= &ksm_mm_head
,
202 static struct kmem_cache
*rmap_item_cache
;
203 static struct kmem_cache
*stable_node_cache
;
204 static struct kmem_cache
*mm_slot_cache
;
206 /* The number of nodes in the stable tree */
207 static unsigned long ksm_pages_shared
;
209 /* The number of page slots additionally sharing those nodes */
210 static unsigned long ksm_pages_sharing
;
212 /* The number of nodes in the unstable tree */
213 static unsigned long ksm_pages_unshared
;
215 /* The number of rmap_items in use: to calculate pages_volatile */
216 static unsigned long ksm_rmap_items
;
218 /* Number of pages ksmd should scan in one batch */
219 static unsigned int ksm_thread_pages_to_scan
= 100;
221 /* Milliseconds ksmd should sleep between batches */
222 static unsigned int ksm_thread_sleep_millisecs
= 20;
225 /* Zeroed when merging across nodes is not allowed */
226 static unsigned int ksm_merge_across_nodes
= 1;
228 #define ksm_merge_across_nodes 1U
231 #define KSM_RUN_STOP 0
232 #define KSM_RUN_MERGE 1
233 #define KSM_RUN_UNMERGE 2
234 #define KSM_RUN_OFFLINE 4
235 static unsigned long ksm_run
= KSM_RUN_STOP
;
236 static void wait_while_offlining(void);
238 static DECLARE_WAIT_QUEUE_HEAD(ksm_thread_wait
);
239 static DEFINE_MUTEX(ksm_thread_mutex
);
240 static DEFINE_SPINLOCK(ksm_mmlist_lock
);
242 #define KSM_KMEM_CACHE(__struct, __flags) kmem_cache_create("ksm_"#__struct,\
243 sizeof(struct __struct), __alignof__(struct __struct),\
246 static int __init
ksm_slab_init(void)
248 rmap_item_cache
= KSM_KMEM_CACHE(rmap_item
, 0);
249 if (!rmap_item_cache
)
252 stable_node_cache
= KSM_KMEM_CACHE(stable_node
, 0);
253 if (!stable_node_cache
)
256 mm_slot_cache
= KSM_KMEM_CACHE(mm_slot
, 0);
263 kmem_cache_destroy(stable_node_cache
);
265 kmem_cache_destroy(rmap_item_cache
);
270 static void __init
ksm_slab_free(void)
272 kmem_cache_destroy(mm_slot_cache
);
273 kmem_cache_destroy(stable_node_cache
);
274 kmem_cache_destroy(rmap_item_cache
);
275 mm_slot_cache
= NULL
;
278 static inline struct rmap_item
*alloc_rmap_item(void)
280 struct rmap_item
*rmap_item
;
282 rmap_item
= kmem_cache_zalloc(rmap_item_cache
, GFP_KERNEL
);
288 static inline void free_rmap_item(struct rmap_item
*rmap_item
)
291 rmap_item
->mm
= NULL
; /* debug safety */
292 kmem_cache_free(rmap_item_cache
, rmap_item
);
295 static inline struct stable_node
*alloc_stable_node(void)
297 return kmem_cache_alloc(stable_node_cache
, GFP_KERNEL
);
300 static inline void free_stable_node(struct stable_node
*stable_node
)
302 kmem_cache_free(stable_node_cache
, stable_node
);
305 static inline struct mm_slot
*alloc_mm_slot(void)
307 if (!mm_slot_cache
) /* initialization failed */
309 return kmem_cache_zalloc(mm_slot_cache
, GFP_KERNEL
);
312 static inline void free_mm_slot(struct mm_slot
*mm_slot
)
314 kmem_cache_free(mm_slot_cache
, mm_slot
);
317 static struct mm_slot
*get_mm_slot(struct mm_struct
*mm
)
319 struct hlist_node
*node
;
320 struct mm_slot
*slot
;
322 hash_for_each_possible(mm_slots_hash
, slot
, node
, link
, (unsigned long)mm
)
329 static void insert_to_mm_slots_hash(struct mm_struct
*mm
,
330 struct mm_slot
*mm_slot
)
333 hash_add(mm_slots_hash
, &mm_slot
->link
, (unsigned long)mm
);
337 * ksmd, and unmerge_and_remove_all_rmap_items(), must not touch an mm's
338 * page tables after it has passed through ksm_exit() - which, if necessary,
339 * takes mmap_sem briefly to serialize against them. ksm_exit() does not set
340 * a special flag: they can just back out as soon as mm_users goes to zero.
341 * ksm_test_exit() is used throughout to make this test for exit: in some
342 * places for correctness, in some places just to avoid unnecessary work.
344 static inline bool ksm_test_exit(struct mm_struct
*mm
)
346 return atomic_read(&mm
->mm_users
) == 0;
350 * We use break_ksm to break COW on a ksm page: it's a stripped down
352 * if (get_user_pages(current, mm, addr, 1, 1, 1, &page, NULL) == 1)
355 * but taking great care only to touch a ksm page, in a VM_MERGEABLE vma,
356 * in case the application has unmapped and remapped mm,addr meanwhile.
357 * Could a ksm page appear anywhere else? Actually yes, in a VM_PFNMAP
358 * mmap of /dev/mem or /dev/kmem, where we would not want to touch it.
360 static int break_ksm(struct vm_area_struct
*vma
, unsigned long addr
)
367 page
= follow_page(vma
, addr
, FOLL_GET
);
368 if (IS_ERR_OR_NULL(page
))
371 ret
= handle_mm_fault(vma
->vm_mm
, vma
, addr
,
374 ret
= VM_FAULT_WRITE
;
376 } while (!(ret
& (VM_FAULT_WRITE
| VM_FAULT_SIGBUS
| VM_FAULT_OOM
)));
378 * We must loop because handle_mm_fault() may back out if there's
379 * any difficulty e.g. if pte accessed bit gets updated concurrently.
381 * VM_FAULT_WRITE is what we have been hoping for: it indicates that
382 * COW has been broken, even if the vma does not permit VM_WRITE;
383 * but note that a concurrent fault might break PageKsm for us.
385 * VM_FAULT_SIGBUS could occur if we race with truncation of the
386 * backing file, which also invalidates anonymous pages: that's
387 * okay, that truncation will have unmapped the PageKsm for us.
389 * VM_FAULT_OOM: at the time of writing (late July 2009), setting
390 * aside mem_cgroup limits, VM_FAULT_OOM would only be set if the
391 * current task has TIF_MEMDIE set, and will be OOM killed on return
392 * to user; and ksmd, having no mm, would never be chosen for that.
394 * But if the mm is in a limited mem_cgroup, then the fault may fail
395 * with VM_FAULT_OOM even if the current task is not TIF_MEMDIE; and
396 * even ksmd can fail in this way - though it's usually breaking ksm
397 * just to undo a merge it made a moment before, so unlikely to oom.
399 * That's a pity: we might therefore have more kernel pages allocated
400 * than we're counting as nodes in the stable tree; but ksm_do_scan
401 * will retry to break_cow on each pass, so should recover the page
402 * in due course. The important thing is to not let VM_MERGEABLE
403 * be cleared while any such pages might remain in the area.
405 return (ret
& VM_FAULT_OOM
) ? -ENOMEM
: 0;
408 static struct vm_area_struct
*find_mergeable_vma(struct mm_struct
*mm
,
411 struct vm_area_struct
*vma
;
412 if (ksm_test_exit(mm
))
414 vma
= find_vma(mm
, addr
);
415 if (!vma
|| vma
->vm_start
> addr
)
417 if (!(vma
->vm_flags
& VM_MERGEABLE
) || !vma
->anon_vma
)
422 static void break_cow(struct rmap_item
*rmap_item
)
424 struct mm_struct
*mm
= rmap_item
->mm
;
425 unsigned long addr
= rmap_item
->address
;
426 struct vm_area_struct
*vma
;
429 * It is not an accident that whenever we want to break COW
430 * to undo, we also need to drop a reference to the anon_vma.
432 put_anon_vma(rmap_item
->anon_vma
);
434 down_read(&mm
->mmap_sem
);
435 vma
= find_mergeable_vma(mm
, addr
);
437 break_ksm(vma
, addr
);
438 up_read(&mm
->mmap_sem
);
441 static struct page
*page_trans_compound_anon(struct page
*page
)
443 if (PageTransCompound(page
)) {
444 struct page
*head
= compound_trans_head(page
);
446 * head may actually be splitted and freed from under
447 * us but it's ok here.
455 static struct page
*get_mergeable_page(struct rmap_item
*rmap_item
)
457 struct mm_struct
*mm
= rmap_item
->mm
;
458 unsigned long addr
= rmap_item
->address
;
459 struct vm_area_struct
*vma
;
462 down_read(&mm
->mmap_sem
);
463 vma
= find_mergeable_vma(mm
, addr
);
467 page
= follow_page(vma
, addr
, FOLL_GET
);
468 if (IS_ERR_OR_NULL(page
))
470 if (PageAnon(page
) || page_trans_compound_anon(page
)) {
471 flush_anon_page(vma
, page
, addr
);
472 flush_dcache_page(page
);
477 up_read(&mm
->mmap_sem
);
482 * This helper is used for getting right index into array of tree roots.
483 * When merge_across_nodes knob is set to 1, there are only two rb-trees for
484 * stable and unstable pages from all nodes with roots in index 0. Otherwise,
485 * every node has its own stable and unstable tree.
487 static inline int get_kpfn_nid(unsigned long kpfn
)
489 return ksm_merge_across_nodes
? 0 : pfn_to_nid(kpfn
);
492 static void remove_node_from_stable_tree(struct stable_node
*stable_node
)
494 struct rmap_item
*rmap_item
;
495 struct hlist_node
*hlist
;
497 hlist_for_each_entry(rmap_item
, hlist
, &stable_node
->hlist
, hlist
) {
498 if (rmap_item
->hlist
.next
)
502 put_anon_vma(rmap_item
->anon_vma
);
503 rmap_item
->address
&= PAGE_MASK
;
507 if (stable_node
->head
== &migrate_nodes
)
508 list_del(&stable_node
->list
);
510 rb_erase(&stable_node
->node
,
511 &root_stable_tree
[NUMA(stable_node
->nid
)]);
512 free_stable_node(stable_node
);
516 * get_ksm_page: checks if the page indicated by the stable node
517 * is still its ksm page, despite having held no reference to it.
518 * In which case we can trust the content of the page, and it
519 * returns the gotten page; but if the page has now been zapped,
520 * remove the stale node from the stable tree and return NULL.
521 * But beware, the stable node's page might be being migrated.
523 * You would expect the stable_node to hold a reference to the ksm page.
524 * But if it increments the page's count, swapping out has to wait for
525 * ksmd to come around again before it can free the page, which may take
526 * seconds or even minutes: much too unresponsive. So instead we use a
527 * "keyhole reference": access to the ksm page from the stable node peeps
528 * out through its keyhole to see if that page still holds the right key,
529 * pointing back to this stable node. This relies on freeing a PageAnon
530 * page to reset its page->mapping to NULL, and relies on no other use of
531 * a page to put something that might look like our key in page->mapping.
532 * is on its way to being freed; but it is an anomaly to bear in mind.
534 static struct page
*get_ksm_page(struct stable_node
*stable_node
, bool lock_it
)
537 void *expected_mapping
;
540 expected_mapping
= (void *)stable_node
+
541 (PAGE_MAPPING_ANON
| PAGE_MAPPING_KSM
);
543 kpfn
= ACCESS_ONCE(stable_node
->kpfn
);
544 page
= pfn_to_page(kpfn
);
547 * page is computed from kpfn, so on most architectures reading
548 * page->mapping is naturally ordered after reading node->kpfn,
549 * but on Alpha we need to be more careful.
551 smp_read_barrier_depends();
552 if (ACCESS_ONCE(page
->mapping
) != expected_mapping
)
556 * We cannot do anything with the page while its refcount is 0.
557 * Usually 0 means free, or tail of a higher-order page: in which
558 * case this node is no longer referenced, and should be freed;
559 * however, it might mean that the page is under page_freeze_refs().
560 * The __remove_mapping() case is easy, again the node is now stale;
561 * but if page is swapcache in migrate_page_move_mapping(), it might
562 * still be our page, in which case it's essential to keep the node.
564 while (!get_page_unless_zero(page
)) {
566 * Another check for page->mapping != expected_mapping would
567 * work here too. We have chosen the !PageSwapCache test to
568 * optimize the common case, when the page is or is about to
569 * be freed: PageSwapCache is cleared (under spin_lock_irq)
570 * in the freeze_refs section of __remove_mapping(); but Anon
571 * page->mapping reset to NULL later, in free_pages_prepare().
573 if (!PageSwapCache(page
))
578 if (ACCESS_ONCE(page
->mapping
) != expected_mapping
) {
585 if (ACCESS_ONCE(page
->mapping
) != expected_mapping
) {
595 * We come here from above when page->mapping or !PageSwapCache
596 * suggests that the node is stale; but it might be under migration.
597 * We need smp_rmb(), matching the smp_wmb() in ksm_migrate_page(),
598 * before checking whether node->kpfn has been changed.
601 if (ACCESS_ONCE(stable_node
->kpfn
) != kpfn
)
603 remove_node_from_stable_tree(stable_node
);
608 * Removing rmap_item from stable or unstable tree.
609 * This function will clean the information from the stable/unstable tree.
611 static void remove_rmap_item_from_tree(struct rmap_item
*rmap_item
)
613 if (rmap_item
->address
& STABLE_FLAG
) {
614 struct stable_node
*stable_node
;
617 stable_node
= rmap_item
->head
;
618 page
= get_ksm_page(stable_node
, true);
622 hlist_del(&rmap_item
->hlist
);
626 if (stable_node
->hlist
.first
)
631 put_anon_vma(rmap_item
->anon_vma
);
632 rmap_item
->address
&= PAGE_MASK
;
634 } else if (rmap_item
->address
& UNSTABLE_FLAG
) {
637 * Usually ksmd can and must skip the rb_erase, because
638 * root_unstable_tree was already reset to RB_ROOT.
639 * But be careful when an mm is exiting: do the rb_erase
640 * if this rmap_item was inserted by this scan, rather
641 * than left over from before.
643 age
= (unsigned char)(ksm_scan
.seqnr
- rmap_item
->address
);
646 rb_erase(&rmap_item
->node
,
647 &root_unstable_tree
[NUMA(rmap_item
->nid
)]);
648 ksm_pages_unshared
--;
649 rmap_item
->address
&= PAGE_MASK
;
652 cond_resched(); /* we're called from many long loops */
655 static void remove_trailing_rmap_items(struct mm_slot
*mm_slot
,
656 struct rmap_item
**rmap_list
)
659 struct rmap_item
*rmap_item
= *rmap_list
;
660 *rmap_list
= rmap_item
->rmap_list
;
661 remove_rmap_item_from_tree(rmap_item
);
662 free_rmap_item(rmap_item
);
667 * Though it's very tempting to unmerge rmap_items from stable tree rather
668 * than check every pte of a given vma, the locking doesn't quite work for
669 * that - an rmap_item is assigned to the stable tree after inserting ksm
670 * page and upping mmap_sem. Nor does it fit with the way we skip dup'ing
671 * rmap_items from parent to child at fork time (so as not to waste time
672 * if exit comes before the next scan reaches it).
674 * Similarly, although we'd like to remove rmap_items (so updating counts
675 * and freeing memory) when unmerging an area, it's easier to leave that
676 * to the next pass of ksmd - consider, for example, how ksmd might be
677 * in cmp_and_merge_page on one of the rmap_items we would be removing.
679 static int unmerge_ksm_pages(struct vm_area_struct
*vma
,
680 unsigned long start
, unsigned long end
)
685 for (addr
= start
; addr
< end
&& !err
; addr
+= PAGE_SIZE
) {
686 if (ksm_test_exit(vma
->vm_mm
))
688 if (signal_pending(current
))
691 err
= break_ksm(vma
, addr
);
698 * Only called through the sysfs control interface:
700 static int remove_stable_node(struct stable_node
*stable_node
)
705 page
= get_ksm_page(stable_node
, true);
708 * get_ksm_page did remove_node_from_stable_tree itself.
713 if (WARN_ON_ONCE(page_mapped(page
))) {
715 * This should not happen: but if it does, just refuse to let
716 * merge_across_nodes be switched - there is no need to panic.
721 * The stable node did not yet appear stale to get_ksm_page(),
722 * since that allows for an unmapped ksm page to be recognized
723 * right up until it is freed; but the node is safe to remove.
724 * This page might be in a pagevec waiting to be freed,
725 * or it might be PageSwapCache (perhaps under writeback),
726 * or it might have been removed from swapcache a moment ago.
728 set_page_stable_node(page
, NULL
);
729 remove_node_from_stable_tree(stable_node
);
738 static int remove_all_stable_nodes(void)
740 struct stable_node
*stable_node
;
741 struct list_head
*this, *next
;
745 for (nid
= 0; nid
< nr_node_ids
; nid
++) {
746 while (root_stable_tree
[nid
].rb_node
) {
747 stable_node
= rb_entry(root_stable_tree
[nid
].rb_node
,
748 struct stable_node
, node
);
749 if (remove_stable_node(stable_node
)) {
751 break; /* proceed to next nid */
756 list_for_each_safe(this, next
, &migrate_nodes
) {
757 stable_node
= list_entry(this, struct stable_node
, list
);
758 if (remove_stable_node(stable_node
))
765 static int unmerge_and_remove_all_rmap_items(void)
767 struct mm_slot
*mm_slot
;
768 struct mm_struct
*mm
;
769 struct vm_area_struct
*vma
;
772 spin_lock(&ksm_mmlist_lock
);
773 ksm_scan
.mm_slot
= list_entry(ksm_mm_head
.mm_list
.next
,
774 struct mm_slot
, mm_list
);
775 spin_unlock(&ksm_mmlist_lock
);
777 for (mm_slot
= ksm_scan
.mm_slot
;
778 mm_slot
!= &ksm_mm_head
; mm_slot
= ksm_scan
.mm_slot
) {
780 down_read(&mm
->mmap_sem
);
781 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
782 if (ksm_test_exit(mm
))
784 if (!(vma
->vm_flags
& VM_MERGEABLE
) || !vma
->anon_vma
)
786 err
= unmerge_ksm_pages(vma
,
787 vma
->vm_start
, vma
->vm_end
);
792 remove_trailing_rmap_items(mm_slot
, &mm_slot
->rmap_list
);
794 spin_lock(&ksm_mmlist_lock
);
795 ksm_scan
.mm_slot
= list_entry(mm_slot
->mm_list
.next
,
796 struct mm_slot
, mm_list
);
797 if (ksm_test_exit(mm
)) {
798 hash_del(&mm_slot
->link
);
799 list_del(&mm_slot
->mm_list
);
800 spin_unlock(&ksm_mmlist_lock
);
802 free_mm_slot(mm_slot
);
803 clear_bit(MMF_VM_MERGEABLE
, &mm
->flags
);
804 up_read(&mm
->mmap_sem
);
807 spin_unlock(&ksm_mmlist_lock
);
808 up_read(&mm
->mmap_sem
);
812 /* Clean up stable nodes, but don't worry if some are still busy */
813 remove_all_stable_nodes();
818 up_read(&mm
->mmap_sem
);
819 spin_lock(&ksm_mmlist_lock
);
820 ksm_scan
.mm_slot
= &ksm_mm_head
;
821 spin_unlock(&ksm_mmlist_lock
);
824 #endif /* CONFIG_SYSFS */
826 static u32
calc_checksum(struct page
*page
)
829 void *addr
= kmap_atomic(page
);
830 checksum
= jhash2(addr
, PAGE_SIZE
/ 4, 17);
835 static int memcmp_pages(struct page
*page1
, struct page
*page2
)
840 addr1
= kmap_atomic(page1
);
841 addr2
= kmap_atomic(page2
);
842 ret
= memcmp(addr1
, addr2
, PAGE_SIZE
);
843 kunmap_atomic(addr2
);
844 kunmap_atomic(addr1
);
848 static inline int pages_identical(struct page
*page1
, struct page
*page2
)
850 return !memcmp_pages(page1
, page2
);
853 static int write_protect_page(struct vm_area_struct
*vma
, struct page
*page
,
856 struct mm_struct
*mm
= vma
->vm_mm
;
862 unsigned long mmun_start
; /* For mmu_notifiers */
863 unsigned long mmun_end
; /* For mmu_notifiers */
865 addr
= page_address_in_vma(page
, vma
);
869 BUG_ON(PageTransCompound(page
));
872 mmun_end
= addr
+ PAGE_SIZE
;
873 mmu_notifier_invalidate_range_start(mm
, mmun_start
, mmun_end
);
875 ptep
= page_check_address(page
, mm
, addr
, &ptl
, 0);
879 if (pte_write(*ptep
) || pte_dirty(*ptep
)) {
882 swapped
= PageSwapCache(page
);
883 flush_cache_page(vma
, addr
, page_to_pfn(page
));
885 * Ok this is tricky, when get_user_pages_fast() run it doesn't
886 * take any lock, therefore the check that we are going to make
887 * with the pagecount against the mapcount is racey and
888 * O_DIRECT can happen right after the check.
889 * So we clear the pte and flush the tlb before the check
890 * this assure us that no O_DIRECT can happen after the check
891 * or in the middle of the check.
893 entry
= ptep_clear_flush(vma
, addr
, ptep
);
895 * Check that no O_DIRECT or similar I/O is in progress on the
898 if (page_mapcount(page
) + 1 + swapped
!= page_count(page
)) {
899 set_pte_at(mm
, addr
, ptep
, entry
);
902 if (pte_dirty(entry
))
903 set_page_dirty(page
);
904 entry
= pte_mkclean(pte_wrprotect(entry
));
905 set_pte_at_notify(mm
, addr
, ptep
, entry
);
911 pte_unmap_unlock(ptep
, ptl
);
913 mmu_notifier_invalidate_range_end(mm
, mmun_start
, mmun_end
);
919 * replace_page - replace page in vma by new ksm page
920 * @vma: vma that holds the pte pointing to page
921 * @page: the page we are replacing by kpage
922 * @kpage: the ksm page we replace page by
923 * @orig_pte: the original value of the pte
925 * Returns 0 on success, -EFAULT on failure.
927 static int replace_page(struct vm_area_struct
*vma
, struct page
*page
,
928 struct page
*kpage
, pte_t orig_pte
)
930 struct mm_struct
*mm
= vma
->vm_mm
;
936 unsigned long mmun_start
; /* For mmu_notifiers */
937 unsigned long mmun_end
; /* For mmu_notifiers */
939 addr
= page_address_in_vma(page
, vma
);
943 pmd
= mm_find_pmd(mm
, addr
);
946 BUG_ON(pmd_trans_huge(*pmd
));
949 mmun_end
= addr
+ PAGE_SIZE
;
950 mmu_notifier_invalidate_range_start(mm
, mmun_start
, mmun_end
);
952 ptep
= pte_offset_map_lock(mm
, pmd
, addr
, &ptl
);
953 if (!pte_same(*ptep
, orig_pte
)) {
954 pte_unmap_unlock(ptep
, ptl
);
959 page_add_anon_rmap(kpage
, vma
, addr
);
961 flush_cache_page(vma
, addr
, pte_pfn(*ptep
));
962 ptep_clear_flush(vma
, addr
, ptep
);
963 set_pte_at_notify(mm
, addr
, ptep
, mk_pte(kpage
, vma
->vm_page_prot
));
965 page_remove_rmap(page
);
966 if (!page_mapped(page
))
967 try_to_free_swap(page
);
970 pte_unmap_unlock(ptep
, ptl
);
973 mmu_notifier_invalidate_range_end(mm
, mmun_start
, mmun_end
);
978 static int page_trans_compound_anon_split(struct page
*page
)
981 struct page
*transhuge_head
= page_trans_compound_anon(page
);
982 if (transhuge_head
) {
983 /* Get the reference on the head to split it. */
984 if (get_page_unless_zero(transhuge_head
)) {
986 * Recheck we got the reference while the head
987 * was still anonymous.
989 if (PageAnon(transhuge_head
))
990 ret
= split_huge_page(transhuge_head
);
993 * Retry later if split_huge_page run
997 put_page(transhuge_head
);
999 /* Retry later if split_huge_page run from under us. */
1006 * try_to_merge_one_page - take two pages and merge them into one
1007 * @vma: the vma that holds the pte pointing to page
1008 * @page: the PageAnon page that we want to replace with kpage
1009 * @kpage: the PageKsm page that we want to map instead of page,
1010 * or NULL the first time when we want to use page as kpage.
1012 * This function returns 0 if the pages were merged, -EFAULT otherwise.
1014 static int try_to_merge_one_page(struct vm_area_struct
*vma
,
1015 struct page
*page
, struct page
*kpage
)
1017 pte_t orig_pte
= __pte(0);
1020 if (page
== kpage
) /* ksm page forked */
1023 if (!(vma
->vm_flags
& VM_MERGEABLE
))
1025 if (PageTransCompound(page
) && page_trans_compound_anon_split(page
))
1027 BUG_ON(PageTransCompound(page
));
1028 if (!PageAnon(page
))
1032 * We need the page lock to read a stable PageSwapCache in
1033 * write_protect_page(). We use trylock_page() instead of
1034 * lock_page() because we don't want to wait here - we
1035 * prefer to continue scanning and merging different pages,
1036 * then come back to this page when it is unlocked.
1038 if (!trylock_page(page
))
1041 * If this anonymous page is mapped only here, its pte may need
1042 * to be write-protected. If it's mapped elsewhere, all of its
1043 * ptes are necessarily already write-protected. But in either
1044 * case, we need to lock and check page_count is not raised.
1046 if (write_protect_page(vma
, page
, &orig_pte
) == 0) {
1049 * While we hold page lock, upgrade page from
1050 * PageAnon+anon_vma to PageKsm+NULL stable_node:
1051 * stable_tree_insert() will update stable_node.
1053 set_page_stable_node(page
, NULL
);
1054 mark_page_accessed(page
);
1056 } else if (pages_identical(page
, kpage
))
1057 err
= replace_page(vma
, page
, kpage
, orig_pte
);
1060 if ((vma
->vm_flags
& VM_LOCKED
) && kpage
&& !err
) {
1061 munlock_vma_page(page
);
1062 if (!PageMlocked(kpage
)) {
1065 mlock_vma_page(kpage
);
1066 page
= kpage
; /* for final unlock */
1076 * try_to_merge_with_ksm_page - like try_to_merge_two_pages,
1077 * but no new kernel page is allocated: kpage must already be a ksm page.
1079 * This function returns 0 if the pages were merged, -EFAULT otherwise.
1081 static int try_to_merge_with_ksm_page(struct rmap_item
*rmap_item
,
1082 struct page
*page
, struct page
*kpage
)
1084 struct mm_struct
*mm
= rmap_item
->mm
;
1085 struct vm_area_struct
*vma
;
1088 down_read(&mm
->mmap_sem
);
1089 if (ksm_test_exit(mm
))
1091 vma
= find_vma(mm
, rmap_item
->address
);
1092 if (!vma
|| vma
->vm_start
> rmap_item
->address
)
1095 err
= try_to_merge_one_page(vma
, page
, kpage
);
1099 /* Unstable nid is in union with stable anon_vma: remove first */
1100 remove_rmap_item_from_tree(rmap_item
);
1102 /* Must get reference to anon_vma while still holding mmap_sem */
1103 rmap_item
->anon_vma
= vma
->anon_vma
;
1104 get_anon_vma(vma
->anon_vma
);
1106 up_read(&mm
->mmap_sem
);
1111 * try_to_merge_two_pages - take two identical pages and prepare them
1112 * to be merged into one page.
1114 * This function returns the kpage if we successfully merged two identical
1115 * pages into one ksm page, NULL otherwise.
1117 * Note that this function upgrades page to ksm page: if one of the pages
1118 * is already a ksm page, try_to_merge_with_ksm_page should be used.
1120 static struct page
*try_to_merge_two_pages(struct rmap_item
*rmap_item
,
1122 struct rmap_item
*tree_rmap_item
,
1123 struct page
*tree_page
)
1127 err
= try_to_merge_with_ksm_page(rmap_item
, page
, NULL
);
1129 err
= try_to_merge_with_ksm_page(tree_rmap_item
,
1132 * If that fails, we have a ksm page with only one pte
1133 * pointing to it: so break it.
1136 break_cow(rmap_item
);
1138 return err
? NULL
: page
;
1142 * stable_tree_search - search for page inside the stable tree
1144 * This function checks if there is a page inside the stable tree
1145 * with identical content to the page that we are scanning right now.
1147 * This function returns the stable tree node of identical content if found,
1150 static struct page
*stable_tree_search(struct page
*page
)
1153 struct rb_node
**new;
1154 struct rb_node
*parent
;
1155 struct stable_node
*stable_node
;
1156 struct stable_node
*page_node
;
1158 page_node
= page_stable_node(page
);
1159 if (page_node
&& page_node
->head
!= &migrate_nodes
) {
1160 /* ksm page forked */
1165 nid
= get_kpfn_nid(page_to_pfn(page
));
1167 new = &root_stable_tree
[nid
].rb_node
;
1171 struct page
*tree_page
;
1175 stable_node
= rb_entry(*new, struct stable_node
, node
);
1176 tree_page
= get_ksm_page(stable_node
, false);
1180 ret
= memcmp_pages(page
, tree_page
);
1181 put_page(tree_page
);
1185 new = &parent
->rb_left
;
1187 new = &parent
->rb_right
;
1190 * Lock and unlock the stable_node's page (which
1191 * might already have been migrated) so that page
1192 * migration is sure to notice its raised count.
1193 * It would be more elegant to return stable_node
1194 * than kpage, but that involves more changes.
1196 tree_page
= get_ksm_page(stable_node
, true);
1198 unlock_page(tree_page
);
1199 if (get_kpfn_nid(stable_node
->kpfn
) !=
1200 NUMA(stable_node
->nid
)) {
1201 put_page(tree_page
);
1207 * There is now a place for page_node, but the tree may
1208 * have been rebalanced, so re-evaluate parent and new.
1219 list_del(&page_node
->list
);
1220 DO_NUMA(page_node
->nid
= nid
);
1221 rb_link_node(&page_node
->node
, parent
, new);
1222 rb_insert_color(&page_node
->node
, &root_stable_tree
[nid
]);
1228 list_del(&page_node
->list
);
1229 DO_NUMA(page_node
->nid
= nid
);
1230 rb_replace_node(&stable_node
->node
,
1231 &page_node
->node
, &root_stable_tree
[nid
]);
1234 rb_erase(&stable_node
->node
, &root_stable_tree
[nid
]);
1237 stable_node
->head
= &migrate_nodes
;
1238 list_add(&stable_node
->list
, stable_node
->head
);
1243 * stable_tree_insert - insert stable tree node pointing to new ksm page
1244 * into the stable tree.
1246 * This function returns the stable tree node just allocated on success,
1249 static struct stable_node
*stable_tree_insert(struct page
*kpage
)
1253 struct rb_node
**new;
1254 struct rb_node
*parent
= NULL
;
1255 struct stable_node
*stable_node
;
1257 kpfn
= page_to_pfn(kpage
);
1258 nid
= get_kpfn_nid(kpfn
);
1259 new = &root_stable_tree
[nid
].rb_node
;
1262 struct page
*tree_page
;
1266 stable_node
= rb_entry(*new, struct stable_node
, node
);
1267 tree_page
= get_ksm_page(stable_node
, false);
1271 ret
= memcmp_pages(kpage
, tree_page
);
1272 put_page(tree_page
);
1276 new = &parent
->rb_left
;
1278 new = &parent
->rb_right
;
1281 * It is not a bug that stable_tree_search() didn't
1282 * find this node: because at that time our page was
1283 * not yet write-protected, so may have changed since.
1289 stable_node
= alloc_stable_node();
1293 INIT_HLIST_HEAD(&stable_node
->hlist
);
1294 stable_node
->kpfn
= kpfn
;
1295 set_page_stable_node(kpage
, stable_node
);
1296 DO_NUMA(stable_node
->nid
= nid
);
1297 rb_link_node(&stable_node
->node
, parent
, new);
1298 rb_insert_color(&stable_node
->node
, &root_stable_tree
[nid
]);
1304 * unstable_tree_search_insert - search for identical page,
1305 * else insert rmap_item into the unstable tree.
1307 * This function searches for a page in the unstable tree identical to the
1308 * page currently being scanned; and if no identical page is found in the
1309 * tree, we insert rmap_item as a new object into the unstable tree.
1311 * This function returns pointer to rmap_item found to be identical
1312 * to the currently scanned page, NULL otherwise.
1314 * This function does both searching and inserting, because they share
1315 * the same walking algorithm in an rbtree.
1318 struct rmap_item
*unstable_tree_search_insert(struct rmap_item
*rmap_item
,
1320 struct page
**tree_pagep
)
1322 struct rb_node
**new;
1323 struct rb_root
*root
;
1324 struct rb_node
*parent
= NULL
;
1327 nid
= get_kpfn_nid(page_to_pfn(page
));
1328 root
= &root_unstable_tree
[nid
];
1329 new = &root
->rb_node
;
1332 struct rmap_item
*tree_rmap_item
;
1333 struct page
*tree_page
;
1337 tree_rmap_item
= rb_entry(*new, struct rmap_item
, node
);
1338 tree_page
= get_mergeable_page(tree_rmap_item
);
1339 if (IS_ERR_OR_NULL(tree_page
))
1343 * Don't substitute a ksm page for a forked page.
1345 if (page
== tree_page
) {
1346 put_page(tree_page
);
1350 ret
= memcmp_pages(page
, tree_page
);
1354 put_page(tree_page
);
1355 new = &parent
->rb_left
;
1356 } else if (ret
> 0) {
1357 put_page(tree_page
);
1358 new = &parent
->rb_right
;
1359 } else if (!ksm_merge_across_nodes
&&
1360 page_to_nid(tree_page
) != nid
) {
1362 * If tree_page has been migrated to another NUMA node,
1363 * it will be flushed out and put in the right unstable
1364 * tree next time: only merge with it when across_nodes.
1366 put_page(tree_page
);
1369 *tree_pagep
= tree_page
;
1370 return tree_rmap_item
;
1374 rmap_item
->address
|= UNSTABLE_FLAG
;
1375 rmap_item
->address
|= (ksm_scan
.seqnr
& SEQNR_MASK
);
1376 DO_NUMA(rmap_item
->nid
= nid
);
1377 rb_link_node(&rmap_item
->node
, parent
, new);
1378 rb_insert_color(&rmap_item
->node
, root
);
1380 ksm_pages_unshared
++;
1385 * stable_tree_append - add another rmap_item to the linked list of
1386 * rmap_items hanging off a given node of the stable tree, all sharing
1387 * the same ksm page.
1389 static void stable_tree_append(struct rmap_item
*rmap_item
,
1390 struct stable_node
*stable_node
)
1392 rmap_item
->head
= stable_node
;
1393 rmap_item
->address
|= STABLE_FLAG
;
1394 hlist_add_head(&rmap_item
->hlist
, &stable_node
->hlist
);
1396 if (rmap_item
->hlist
.next
)
1397 ksm_pages_sharing
++;
1403 * cmp_and_merge_page - first see if page can be merged into the stable tree;
1404 * if not, compare checksum to previous and if it's the same, see if page can
1405 * be inserted into the unstable tree, or merged with a page already there and
1406 * both transferred to the stable tree.
1408 * @page: the page that we are searching identical page to.
1409 * @rmap_item: the reverse mapping into the virtual address of this page
1411 static void cmp_and_merge_page(struct page
*page
, struct rmap_item
*rmap_item
)
1413 struct rmap_item
*tree_rmap_item
;
1414 struct page
*tree_page
= NULL
;
1415 struct stable_node
*stable_node
;
1417 unsigned int checksum
;
1420 stable_node
= page_stable_node(page
);
1422 if (stable_node
->head
!= &migrate_nodes
&&
1423 get_kpfn_nid(stable_node
->kpfn
) != NUMA(stable_node
->nid
)) {
1424 rb_erase(&stable_node
->node
,
1425 &root_stable_tree
[NUMA(stable_node
->nid
)]);
1426 stable_node
->head
= &migrate_nodes
;
1427 list_add(&stable_node
->list
, stable_node
->head
);
1429 if (stable_node
->head
!= &migrate_nodes
&&
1430 rmap_item
->head
== stable_node
)
1434 /* We first start with searching the page inside the stable tree */
1435 kpage
= stable_tree_search(page
);
1436 if (kpage
== page
&& rmap_item
->head
== stable_node
) {
1441 remove_rmap_item_from_tree(rmap_item
);
1444 err
= try_to_merge_with_ksm_page(rmap_item
, page
, kpage
);
1447 * The page was successfully merged:
1448 * add its rmap_item to the stable tree.
1451 stable_tree_append(rmap_item
, page_stable_node(kpage
));
1459 * If the hash value of the page has changed from the last time
1460 * we calculated it, this page is changing frequently: therefore we
1461 * don't want to insert it in the unstable tree, and we don't want
1462 * to waste our time searching for something identical to it there.
1464 checksum
= calc_checksum(page
);
1465 if (rmap_item
->oldchecksum
!= checksum
) {
1466 rmap_item
->oldchecksum
= checksum
;
1471 unstable_tree_search_insert(rmap_item
, page
, &tree_page
);
1472 if (tree_rmap_item
) {
1473 kpage
= try_to_merge_two_pages(rmap_item
, page
,
1474 tree_rmap_item
, tree_page
);
1475 put_page(tree_page
);
1478 * The pages were successfully merged: insert new
1479 * node in the stable tree and add both rmap_items.
1482 stable_node
= stable_tree_insert(kpage
);
1484 stable_tree_append(tree_rmap_item
, stable_node
);
1485 stable_tree_append(rmap_item
, stable_node
);
1490 * If we fail to insert the page into the stable tree,
1491 * we will have 2 virtual addresses that are pointing
1492 * to a ksm page left outside the stable tree,
1493 * in which case we need to break_cow on both.
1496 break_cow(tree_rmap_item
);
1497 break_cow(rmap_item
);
1503 static struct rmap_item
*get_next_rmap_item(struct mm_slot
*mm_slot
,
1504 struct rmap_item
**rmap_list
,
1507 struct rmap_item
*rmap_item
;
1509 while (*rmap_list
) {
1510 rmap_item
= *rmap_list
;
1511 if ((rmap_item
->address
& PAGE_MASK
) == addr
)
1513 if (rmap_item
->address
> addr
)
1515 *rmap_list
= rmap_item
->rmap_list
;
1516 remove_rmap_item_from_tree(rmap_item
);
1517 free_rmap_item(rmap_item
);
1520 rmap_item
= alloc_rmap_item();
1522 /* It has already been zeroed */
1523 rmap_item
->mm
= mm_slot
->mm
;
1524 rmap_item
->address
= addr
;
1525 rmap_item
->rmap_list
= *rmap_list
;
1526 *rmap_list
= rmap_item
;
1531 static struct rmap_item
*scan_get_next_rmap_item(struct page
**page
)
1533 struct mm_struct
*mm
;
1534 struct mm_slot
*slot
;
1535 struct vm_area_struct
*vma
;
1536 struct rmap_item
*rmap_item
;
1539 if (list_empty(&ksm_mm_head
.mm_list
))
1542 slot
= ksm_scan
.mm_slot
;
1543 if (slot
== &ksm_mm_head
) {
1545 * A number of pages can hang around indefinitely on per-cpu
1546 * pagevecs, raised page count preventing write_protect_page
1547 * from merging them. Though it doesn't really matter much,
1548 * it is puzzling to see some stuck in pages_volatile until
1549 * other activity jostles them out, and they also prevented
1550 * LTP's KSM test from succeeding deterministically; so drain
1551 * them here (here rather than on entry to ksm_do_scan(),
1552 * so we don't IPI too often when pages_to_scan is set low).
1554 lru_add_drain_all();
1557 * Whereas stale stable_nodes on the stable_tree itself
1558 * get pruned in the regular course of stable_tree_search(),
1559 * those moved out to the migrate_nodes list can accumulate:
1560 * so prune them once before each full scan.
1562 if (!ksm_merge_across_nodes
) {
1563 struct stable_node
*stable_node
;
1564 struct list_head
*this, *next
;
1567 list_for_each_safe(this, next
, &migrate_nodes
) {
1568 stable_node
= list_entry(this,
1569 struct stable_node
, list
);
1570 page
= get_ksm_page(stable_node
, false);
1577 for (nid
= 0; nid
< nr_node_ids
; nid
++)
1578 root_unstable_tree
[nid
] = RB_ROOT
;
1580 spin_lock(&ksm_mmlist_lock
);
1581 slot
= list_entry(slot
->mm_list
.next
, struct mm_slot
, mm_list
);
1582 ksm_scan
.mm_slot
= slot
;
1583 spin_unlock(&ksm_mmlist_lock
);
1585 * Although we tested list_empty() above, a racing __ksm_exit
1586 * of the last mm on the list may have removed it since then.
1588 if (slot
== &ksm_mm_head
)
1591 ksm_scan
.address
= 0;
1592 ksm_scan
.rmap_list
= &slot
->rmap_list
;
1596 down_read(&mm
->mmap_sem
);
1597 if (ksm_test_exit(mm
))
1600 vma
= find_vma(mm
, ksm_scan
.address
);
1602 for (; vma
; vma
= vma
->vm_next
) {
1603 if (!(vma
->vm_flags
& VM_MERGEABLE
))
1605 if (ksm_scan
.address
< vma
->vm_start
)
1606 ksm_scan
.address
= vma
->vm_start
;
1608 ksm_scan
.address
= vma
->vm_end
;
1610 while (ksm_scan
.address
< vma
->vm_end
) {
1611 if (ksm_test_exit(mm
))
1613 *page
= follow_page(vma
, ksm_scan
.address
, FOLL_GET
);
1614 if (IS_ERR_OR_NULL(*page
)) {
1615 ksm_scan
.address
+= PAGE_SIZE
;
1619 if (PageAnon(*page
) ||
1620 page_trans_compound_anon(*page
)) {
1621 flush_anon_page(vma
, *page
, ksm_scan
.address
);
1622 flush_dcache_page(*page
);
1623 rmap_item
= get_next_rmap_item(slot
,
1624 ksm_scan
.rmap_list
, ksm_scan
.address
);
1626 ksm_scan
.rmap_list
=
1627 &rmap_item
->rmap_list
;
1628 ksm_scan
.address
+= PAGE_SIZE
;
1631 up_read(&mm
->mmap_sem
);
1635 ksm_scan
.address
+= PAGE_SIZE
;
1640 if (ksm_test_exit(mm
)) {
1641 ksm_scan
.address
= 0;
1642 ksm_scan
.rmap_list
= &slot
->rmap_list
;
1645 * Nuke all the rmap_items that are above this current rmap:
1646 * because there were no VM_MERGEABLE vmas with such addresses.
1648 remove_trailing_rmap_items(slot
, ksm_scan
.rmap_list
);
1650 spin_lock(&ksm_mmlist_lock
);
1651 ksm_scan
.mm_slot
= list_entry(slot
->mm_list
.next
,
1652 struct mm_slot
, mm_list
);
1653 if (ksm_scan
.address
== 0) {
1655 * We've completed a full scan of all vmas, holding mmap_sem
1656 * throughout, and found no VM_MERGEABLE: so do the same as
1657 * __ksm_exit does to remove this mm from all our lists now.
1658 * This applies either when cleaning up after __ksm_exit
1659 * (but beware: we can reach here even before __ksm_exit),
1660 * or when all VM_MERGEABLE areas have been unmapped (and
1661 * mmap_sem then protects against race with MADV_MERGEABLE).
1663 hash_del(&slot
->link
);
1664 list_del(&slot
->mm_list
);
1665 spin_unlock(&ksm_mmlist_lock
);
1668 clear_bit(MMF_VM_MERGEABLE
, &mm
->flags
);
1669 up_read(&mm
->mmap_sem
);
1672 spin_unlock(&ksm_mmlist_lock
);
1673 up_read(&mm
->mmap_sem
);
1676 /* Repeat until we've completed scanning the whole list */
1677 slot
= ksm_scan
.mm_slot
;
1678 if (slot
!= &ksm_mm_head
)
1686 * ksm_do_scan - the ksm scanner main worker function.
1687 * @scan_npages - number of pages we want to scan before we return.
1689 static void ksm_do_scan(unsigned int scan_npages
)
1691 struct rmap_item
*rmap_item
;
1692 struct page
*uninitialized_var(page
);
1694 while (scan_npages
-- && likely(!freezing(current
))) {
1696 rmap_item
= scan_get_next_rmap_item(&page
);
1699 cmp_and_merge_page(page
, rmap_item
);
1704 static int ksmd_should_run(void)
1706 return (ksm_run
& KSM_RUN_MERGE
) && !list_empty(&ksm_mm_head
.mm_list
);
1709 static int ksm_scan_thread(void *nothing
)
1712 set_user_nice(current
, 5);
1714 while (!kthread_should_stop()) {
1715 mutex_lock(&ksm_thread_mutex
);
1716 wait_while_offlining();
1717 if (ksmd_should_run())
1718 ksm_do_scan(ksm_thread_pages_to_scan
);
1719 mutex_unlock(&ksm_thread_mutex
);
1723 if (ksmd_should_run()) {
1724 schedule_timeout_interruptible(
1725 msecs_to_jiffies(ksm_thread_sleep_millisecs
));
1727 wait_event_freezable(ksm_thread_wait
,
1728 ksmd_should_run() || kthread_should_stop());
1734 int ksm_madvise(struct vm_area_struct
*vma
, unsigned long start
,
1735 unsigned long end
, int advice
, unsigned long *vm_flags
)
1737 struct mm_struct
*mm
= vma
->vm_mm
;
1741 case MADV_MERGEABLE
:
1743 * Be somewhat over-protective for now!
1745 if (*vm_flags
& (VM_MERGEABLE
| VM_SHARED
| VM_MAYSHARE
|
1746 VM_PFNMAP
| VM_IO
| VM_DONTEXPAND
|
1747 VM_HUGETLB
| VM_NONLINEAR
| VM_MIXEDMAP
))
1748 return 0; /* just ignore the advice */
1751 if (*vm_flags
& VM_SAO
)
1755 if (!test_bit(MMF_VM_MERGEABLE
, &mm
->flags
)) {
1756 err
= __ksm_enter(mm
);
1761 *vm_flags
|= VM_MERGEABLE
;
1764 case MADV_UNMERGEABLE
:
1765 if (!(*vm_flags
& VM_MERGEABLE
))
1766 return 0; /* just ignore the advice */
1768 if (vma
->anon_vma
) {
1769 err
= unmerge_ksm_pages(vma
, start
, end
);
1774 *vm_flags
&= ~VM_MERGEABLE
;
1781 int __ksm_enter(struct mm_struct
*mm
)
1783 struct mm_slot
*mm_slot
;
1786 mm_slot
= alloc_mm_slot();
1790 /* Check ksm_run too? Would need tighter locking */
1791 needs_wakeup
= list_empty(&ksm_mm_head
.mm_list
);
1793 spin_lock(&ksm_mmlist_lock
);
1794 insert_to_mm_slots_hash(mm
, mm_slot
);
1796 * When KSM_RUN_MERGE (or KSM_RUN_STOP),
1797 * insert just behind the scanning cursor, to let the area settle
1798 * down a little; when fork is followed by immediate exec, we don't
1799 * want ksmd to waste time setting up and tearing down an rmap_list.
1801 * But when KSM_RUN_UNMERGE, it's important to insert ahead of its
1802 * scanning cursor, otherwise KSM pages in newly forked mms will be
1803 * missed: then we might as well insert at the end of the list.
1805 if (ksm_run
& KSM_RUN_UNMERGE
)
1806 list_add_tail(&mm_slot
->mm_list
, &ksm_mm_head
.mm_list
);
1808 list_add_tail(&mm_slot
->mm_list
, &ksm_scan
.mm_slot
->mm_list
);
1809 spin_unlock(&ksm_mmlist_lock
);
1811 set_bit(MMF_VM_MERGEABLE
, &mm
->flags
);
1812 atomic_inc(&mm
->mm_count
);
1815 wake_up_interruptible(&ksm_thread_wait
);
1820 void __ksm_exit(struct mm_struct
*mm
)
1822 struct mm_slot
*mm_slot
;
1823 int easy_to_free
= 0;
1826 * This process is exiting: if it's straightforward (as is the
1827 * case when ksmd was never running), free mm_slot immediately.
1828 * But if it's at the cursor or has rmap_items linked to it, use
1829 * mmap_sem to synchronize with any break_cows before pagetables
1830 * are freed, and leave the mm_slot on the list for ksmd to free.
1831 * Beware: ksm may already have noticed it exiting and freed the slot.
1834 spin_lock(&ksm_mmlist_lock
);
1835 mm_slot
= get_mm_slot(mm
);
1836 if (mm_slot
&& ksm_scan
.mm_slot
!= mm_slot
) {
1837 if (!mm_slot
->rmap_list
) {
1838 hash_del(&mm_slot
->link
);
1839 list_del(&mm_slot
->mm_list
);
1842 list_move(&mm_slot
->mm_list
,
1843 &ksm_scan
.mm_slot
->mm_list
);
1846 spin_unlock(&ksm_mmlist_lock
);
1849 free_mm_slot(mm_slot
);
1850 clear_bit(MMF_VM_MERGEABLE
, &mm
->flags
);
1852 } else if (mm_slot
) {
1853 down_write(&mm
->mmap_sem
);
1854 up_write(&mm
->mmap_sem
);
1858 struct page
*ksm_might_need_to_copy(struct page
*page
,
1859 struct vm_area_struct
*vma
, unsigned long address
)
1861 struct anon_vma
*anon_vma
= page_anon_vma(page
);
1862 struct page
*new_page
;
1864 if (PageKsm(page
)) {
1865 if (page_stable_node(page
) &&
1866 !(ksm_run
& KSM_RUN_UNMERGE
))
1867 return page
; /* no need to copy it */
1868 } else if (!anon_vma
) {
1869 return page
; /* no need to copy it */
1870 } else if (anon_vma
->root
== vma
->anon_vma
->root
&&
1871 page
->index
== linear_page_index(vma
, address
)) {
1872 return page
; /* still no need to copy it */
1874 if (!PageUptodate(page
))
1875 return page
; /* let do_swap_page report the error */
1877 new_page
= alloc_page_vma(GFP_HIGHUSER_MOVABLE
, vma
, address
);
1879 copy_user_highpage(new_page
, page
, address
, vma
);
1881 SetPageDirty(new_page
);
1882 __SetPageUptodate(new_page
);
1883 __set_page_locked(new_page
);
1889 int page_referenced_ksm(struct page
*page
, struct mem_cgroup
*memcg
,
1890 unsigned long *vm_flags
)
1892 struct stable_node
*stable_node
;
1893 struct rmap_item
*rmap_item
;
1894 struct hlist_node
*hlist
;
1895 unsigned int mapcount
= page_mapcount(page
);
1897 int search_new_forks
= 0;
1899 VM_BUG_ON(!PageKsm(page
));
1900 VM_BUG_ON(!PageLocked(page
));
1902 stable_node
= page_stable_node(page
);
1906 hlist_for_each_entry(rmap_item
, hlist
, &stable_node
->hlist
, hlist
) {
1907 struct anon_vma
*anon_vma
= rmap_item
->anon_vma
;
1908 struct anon_vma_chain
*vmac
;
1909 struct vm_area_struct
*vma
;
1911 anon_vma_lock_read(anon_vma
);
1912 anon_vma_interval_tree_foreach(vmac
, &anon_vma
->rb_root
,
1915 if (rmap_item
->address
< vma
->vm_start
||
1916 rmap_item
->address
>= vma
->vm_end
)
1919 * Initially we examine only the vma which covers this
1920 * rmap_item; but later, if there is still work to do,
1921 * we examine covering vmas in other mms: in case they
1922 * were forked from the original since ksmd passed.
1924 if ((rmap_item
->mm
== vma
->vm_mm
) == search_new_forks
)
1927 if (memcg
&& !mm_match_cgroup(vma
->vm_mm
, memcg
))
1930 referenced
+= page_referenced_one(page
, vma
,
1931 rmap_item
->address
, &mapcount
, vm_flags
);
1932 if (!search_new_forks
|| !mapcount
)
1935 anon_vma_unlock_read(anon_vma
);
1939 if (!search_new_forks
++)
1945 int try_to_unmap_ksm(struct page
*page
, enum ttu_flags flags
)
1947 struct stable_node
*stable_node
;
1948 struct hlist_node
*hlist
;
1949 struct rmap_item
*rmap_item
;
1950 int ret
= SWAP_AGAIN
;
1951 int search_new_forks
= 0;
1953 VM_BUG_ON(!PageKsm(page
));
1954 VM_BUG_ON(!PageLocked(page
));
1956 stable_node
= page_stable_node(page
);
1960 hlist_for_each_entry(rmap_item
, hlist
, &stable_node
->hlist
, hlist
) {
1961 struct anon_vma
*anon_vma
= rmap_item
->anon_vma
;
1962 struct anon_vma_chain
*vmac
;
1963 struct vm_area_struct
*vma
;
1965 anon_vma_lock_read(anon_vma
);
1966 anon_vma_interval_tree_foreach(vmac
, &anon_vma
->rb_root
,
1969 if (rmap_item
->address
< vma
->vm_start
||
1970 rmap_item
->address
>= vma
->vm_end
)
1973 * Initially we examine only the vma which covers this
1974 * rmap_item; but later, if there is still work to do,
1975 * we examine covering vmas in other mms: in case they
1976 * were forked from the original since ksmd passed.
1978 if ((rmap_item
->mm
== vma
->vm_mm
) == search_new_forks
)
1981 ret
= try_to_unmap_one(page
, vma
,
1982 rmap_item
->address
, flags
);
1983 if (ret
!= SWAP_AGAIN
|| !page_mapped(page
)) {
1984 anon_vma_unlock_read(anon_vma
);
1988 anon_vma_unlock_read(anon_vma
);
1990 if (!search_new_forks
++)
1996 #ifdef CONFIG_MIGRATION
1997 int rmap_walk_ksm(struct page
*page
, int (*rmap_one
)(struct page
*,
1998 struct vm_area_struct
*, unsigned long, void *), void *arg
)
2000 struct stable_node
*stable_node
;
2001 struct hlist_node
*hlist
;
2002 struct rmap_item
*rmap_item
;
2003 int ret
= SWAP_AGAIN
;
2004 int search_new_forks
= 0;
2006 VM_BUG_ON(!PageKsm(page
));
2007 VM_BUG_ON(!PageLocked(page
));
2009 stable_node
= page_stable_node(page
);
2013 hlist_for_each_entry(rmap_item
, hlist
, &stable_node
->hlist
, hlist
) {
2014 struct anon_vma
*anon_vma
= rmap_item
->anon_vma
;
2015 struct anon_vma_chain
*vmac
;
2016 struct vm_area_struct
*vma
;
2018 anon_vma_lock_read(anon_vma
);
2019 anon_vma_interval_tree_foreach(vmac
, &anon_vma
->rb_root
,
2022 if (rmap_item
->address
< vma
->vm_start
||
2023 rmap_item
->address
>= vma
->vm_end
)
2026 * Initially we examine only the vma which covers this
2027 * rmap_item; but later, if there is still work to do,
2028 * we examine covering vmas in other mms: in case they
2029 * were forked from the original since ksmd passed.
2031 if ((rmap_item
->mm
== vma
->vm_mm
) == search_new_forks
)
2034 ret
= rmap_one(page
, vma
, rmap_item
->address
, arg
);
2035 if (ret
!= SWAP_AGAIN
) {
2036 anon_vma_unlock_read(anon_vma
);
2040 anon_vma_unlock_read(anon_vma
);
2042 if (!search_new_forks
++)
2048 void ksm_migrate_page(struct page
*newpage
, struct page
*oldpage
)
2050 struct stable_node
*stable_node
;
2052 VM_BUG_ON(!PageLocked(oldpage
));
2053 VM_BUG_ON(!PageLocked(newpage
));
2054 VM_BUG_ON(newpage
->mapping
!= oldpage
->mapping
);
2056 stable_node
= page_stable_node(newpage
);
2058 VM_BUG_ON(stable_node
->kpfn
!= page_to_pfn(oldpage
));
2059 stable_node
->kpfn
= page_to_pfn(newpage
);
2061 * newpage->mapping was set in advance; now we need smp_wmb()
2062 * to make sure that the new stable_node->kpfn is visible
2063 * to get_ksm_page() before it can see that oldpage->mapping
2064 * has gone stale (or that PageSwapCache has been cleared).
2067 set_page_stable_node(oldpage
, NULL
);
2070 #endif /* CONFIG_MIGRATION */
2072 #ifdef CONFIG_MEMORY_HOTREMOVE
2073 static int just_wait(void *word
)
2079 static void wait_while_offlining(void)
2081 while (ksm_run
& KSM_RUN_OFFLINE
) {
2082 mutex_unlock(&ksm_thread_mutex
);
2083 wait_on_bit(&ksm_run
, ilog2(KSM_RUN_OFFLINE
),
2084 just_wait
, TASK_UNINTERRUPTIBLE
);
2085 mutex_lock(&ksm_thread_mutex
);
2089 static void ksm_check_stable_tree(unsigned long start_pfn
,
2090 unsigned long end_pfn
)
2092 struct stable_node
*stable_node
;
2093 struct list_head
*this, *next
;
2094 struct rb_node
*node
;
2097 for (nid
= 0; nid
< nr_node_ids
; nid
++) {
2098 node
= rb_first(&root_stable_tree
[nid
]);
2100 stable_node
= rb_entry(node
, struct stable_node
, node
);
2101 if (stable_node
->kpfn
>= start_pfn
&&
2102 stable_node
->kpfn
< end_pfn
) {
2104 * Don't get_ksm_page, page has already gone:
2105 * which is why we keep kpfn instead of page*
2107 remove_node_from_stable_tree(stable_node
);
2108 node
= rb_first(&root_stable_tree
[nid
]);
2110 node
= rb_next(node
);
2114 list_for_each_safe(this, next
, &migrate_nodes
) {
2115 stable_node
= list_entry(this, struct stable_node
, list
);
2116 if (stable_node
->kpfn
>= start_pfn
&&
2117 stable_node
->kpfn
< end_pfn
)
2118 remove_node_from_stable_tree(stable_node
);
2123 static int ksm_memory_callback(struct notifier_block
*self
,
2124 unsigned long action
, void *arg
)
2126 struct memory_notify
*mn
= arg
;
2129 case MEM_GOING_OFFLINE
:
2131 * Prevent ksm_do_scan(), unmerge_and_remove_all_rmap_items()
2132 * and remove_all_stable_nodes() while memory is going offline:
2133 * it is unsafe for them to touch the stable tree at this time.
2134 * But unmerge_ksm_pages(), rmap lookups and other entry points
2135 * which do not need the ksm_thread_mutex are all safe.
2137 mutex_lock(&ksm_thread_mutex
);
2138 ksm_run
|= KSM_RUN_OFFLINE
;
2139 mutex_unlock(&ksm_thread_mutex
);
2144 * Most of the work is done by page migration; but there might
2145 * be a few stable_nodes left over, still pointing to struct
2146 * pages which have been offlined: prune those from the tree,
2147 * otherwise get_ksm_page() might later try to access a
2148 * non-existent struct page.
2150 ksm_check_stable_tree(mn
->start_pfn
,
2151 mn
->start_pfn
+ mn
->nr_pages
);
2154 case MEM_CANCEL_OFFLINE
:
2155 mutex_lock(&ksm_thread_mutex
);
2156 ksm_run
&= ~KSM_RUN_OFFLINE
;
2157 mutex_unlock(&ksm_thread_mutex
);
2159 smp_mb(); /* wake_up_bit advises this */
2160 wake_up_bit(&ksm_run
, ilog2(KSM_RUN_OFFLINE
));
2166 static void wait_while_offlining(void)
2169 #endif /* CONFIG_MEMORY_HOTREMOVE */
2173 * This all compiles without CONFIG_SYSFS, but is a waste of space.
2176 #define KSM_ATTR_RO(_name) \
2177 static struct kobj_attribute _name##_attr = __ATTR_RO(_name)
2178 #define KSM_ATTR(_name) \
2179 static struct kobj_attribute _name##_attr = \
2180 __ATTR(_name, 0644, _name##_show, _name##_store)
2182 static ssize_t
sleep_millisecs_show(struct kobject
*kobj
,
2183 struct kobj_attribute
*attr
, char *buf
)
2185 return sprintf(buf
, "%u\n", ksm_thread_sleep_millisecs
);
2188 static ssize_t
sleep_millisecs_store(struct kobject
*kobj
,
2189 struct kobj_attribute
*attr
,
2190 const char *buf
, size_t count
)
2192 unsigned long msecs
;
2195 err
= strict_strtoul(buf
, 10, &msecs
);
2196 if (err
|| msecs
> UINT_MAX
)
2199 ksm_thread_sleep_millisecs
= msecs
;
2203 KSM_ATTR(sleep_millisecs
);
2205 static ssize_t
pages_to_scan_show(struct kobject
*kobj
,
2206 struct kobj_attribute
*attr
, char *buf
)
2208 return sprintf(buf
, "%u\n", ksm_thread_pages_to_scan
);
2211 static ssize_t
pages_to_scan_store(struct kobject
*kobj
,
2212 struct kobj_attribute
*attr
,
2213 const char *buf
, size_t count
)
2216 unsigned long nr_pages
;
2218 err
= strict_strtoul(buf
, 10, &nr_pages
);
2219 if (err
|| nr_pages
> UINT_MAX
)
2222 ksm_thread_pages_to_scan
= nr_pages
;
2226 KSM_ATTR(pages_to_scan
);
2228 static ssize_t
run_show(struct kobject
*kobj
, struct kobj_attribute
*attr
,
2231 return sprintf(buf
, "%lu\n", ksm_run
);
2234 static ssize_t
run_store(struct kobject
*kobj
, struct kobj_attribute
*attr
,
2235 const char *buf
, size_t count
)
2238 unsigned long flags
;
2240 err
= strict_strtoul(buf
, 10, &flags
);
2241 if (err
|| flags
> UINT_MAX
)
2243 if (flags
> KSM_RUN_UNMERGE
)
2247 * KSM_RUN_MERGE sets ksmd running, and 0 stops it running.
2248 * KSM_RUN_UNMERGE stops it running and unmerges all rmap_items,
2249 * breaking COW to free the pages_shared (but leaves mm_slots
2250 * on the list for when ksmd may be set running again).
2253 mutex_lock(&ksm_thread_mutex
);
2254 wait_while_offlining();
2255 if (ksm_run
!= flags
) {
2257 if (flags
& KSM_RUN_UNMERGE
) {
2258 set_current_oom_origin();
2259 err
= unmerge_and_remove_all_rmap_items();
2260 clear_current_oom_origin();
2262 ksm_run
= KSM_RUN_STOP
;
2267 mutex_unlock(&ksm_thread_mutex
);
2269 if (flags
& KSM_RUN_MERGE
)
2270 wake_up_interruptible(&ksm_thread_wait
);
2277 static ssize_t
merge_across_nodes_show(struct kobject
*kobj
,
2278 struct kobj_attribute
*attr
, char *buf
)
2280 return sprintf(buf
, "%u\n", ksm_merge_across_nodes
);
2283 static ssize_t
merge_across_nodes_store(struct kobject
*kobj
,
2284 struct kobj_attribute
*attr
,
2285 const char *buf
, size_t count
)
2290 err
= kstrtoul(buf
, 10, &knob
);
2296 mutex_lock(&ksm_thread_mutex
);
2297 wait_while_offlining();
2298 if (ksm_merge_across_nodes
!= knob
) {
2299 if (ksm_pages_shared
|| remove_all_stable_nodes())
2302 ksm_merge_across_nodes
= knob
;
2304 mutex_unlock(&ksm_thread_mutex
);
2306 return err
? err
: count
;
2308 KSM_ATTR(merge_across_nodes
);
2311 static ssize_t
pages_shared_show(struct kobject
*kobj
,
2312 struct kobj_attribute
*attr
, char *buf
)
2314 return sprintf(buf
, "%lu\n", ksm_pages_shared
);
2316 KSM_ATTR_RO(pages_shared
);
2318 static ssize_t
pages_sharing_show(struct kobject
*kobj
,
2319 struct kobj_attribute
*attr
, char *buf
)
2321 return sprintf(buf
, "%lu\n", ksm_pages_sharing
);
2323 KSM_ATTR_RO(pages_sharing
);
2325 static ssize_t
pages_unshared_show(struct kobject
*kobj
,
2326 struct kobj_attribute
*attr
, char *buf
)
2328 return sprintf(buf
, "%lu\n", ksm_pages_unshared
);
2330 KSM_ATTR_RO(pages_unshared
);
2332 static ssize_t
pages_volatile_show(struct kobject
*kobj
,
2333 struct kobj_attribute
*attr
, char *buf
)
2335 long ksm_pages_volatile
;
2337 ksm_pages_volatile
= ksm_rmap_items
- ksm_pages_shared
2338 - ksm_pages_sharing
- ksm_pages_unshared
;
2340 * It was not worth any locking to calculate that statistic,
2341 * but it might therefore sometimes be negative: conceal that.
2343 if (ksm_pages_volatile
< 0)
2344 ksm_pages_volatile
= 0;
2345 return sprintf(buf
, "%ld\n", ksm_pages_volatile
);
2347 KSM_ATTR_RO(pages_volatile
);
2349 static ssize_t
full_scans_show(struct kobject
*kobj
,
2350 struct kobj_attribute
*attr
, char *buf
)
2352 return sprintf(buf
, "%lu\n", ksm_scan
.seqnr
);
2354 KSM_ATTR_RO(full_scans
);
2356 static struct attribute
*ksm_attrs
[] = {
2357 &sleep_millisecs_attr
.attr
,
2358 &pages_to_scan_attr
.attr
,
2360 &pages_shared_attr
.attr
,
2361 &pages_sharing_attr
.attr
,
2362 &pages_unshared_attr
.attr
,
2363 &pages_volatile_attr
.attr
,
2364 &full_scans_attr
.attr
,
2366 &merge_across_nodes_attr
.attr
,
2371 static struct attribute_group ksm_attr_group
= {
2375 #endif /* CONFIG_SYSFS */
2377 static int __init
ksm_init(void)
2379 struct task_struct
*ksm_thread
;
2383 err
= ksm_slab_init();
2387 for (nid
= 0; nid
< nr_node_ids
; nid
++)
2388 root_stable_tree
[nid
] = RB_ROOT
;
2390 ksm_thread
= kthread_run(ksm_scan_thread
, NULL
, "ksmd");
2391 if (IS_ERR(ksm_thread
)) {
2392 printk(KERN_ERR
"ksm: creating kthread failed\n");
2393 err
= PTR_ERR(ksm_thread
);
2398 err
= sysfs_create_group(mm_kobj
, &ksm_attr_group
);
2400 printk(KERN_ERR
"ksm: register sysfs failed\n");
2401 kthread_stop(ksm_thread
);
2405 ksm_run
= KSM_RUN_MERGE
; /* no way for user to start it */
2407 #endif /* CONFIG_SYSFS */
2409 #ifdef CONFIG_MEMORY_HOTREMOVE
2410 /* There is no significance to this priority 100 */
2411 hotplug_memory_notifier(ksm_memory_callback
, 100);
2420 module_init(ksm_init
)