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/swap.h>
34 #include <linux/ksm.h>
36 #include <asm/tlbflush.h>
40 * A few notes about the KSM scanning process,
41 * to make it easier to understand the data structures below:
43 * In order to reduce excessive scanning, KSM sorts the memory pages by their
44 * contents into a data structure that holds pointers to the pages' locations.
46 * Since the contents of the pages may change at any moment, KSM cannot just
47 * insert the pages into a normal sorted tree and expect it to find anything.
48 * Therefore KSM uses two data structures - the stable and the unstable tree.
50 * The stable tree holds pointers to all the merged pages (ksm pages), sorted
51 * by their contents. Because each such page is write-protected, searching on
52 * this tree is fully assured to be working (except when pages are unmapped),
53 * and therefore this tree is called the stable tree.
55 * In addition to the stable tree, KSM uses a second data structure called the
56 * unstable tree: this tree holds pointers to pages which have been found to
57 * be "unchanged for a period of time". The unstable tree sorts these pages
58 * by their contents, but since they are not write-protected, KSM cannot rely
59 * upon the unstable tree to work correctly - the unstable tree is liable to
60 * be corrupted as its contents are modified, and so it is called unstable.
62 * KSM solves this problem by several techniques:
64 * 1) The unstable tree is flushed every time KSM completes scanning all
65 * memory areas, and then the tree is rebuilt again from the beginning.
66 * 2) KSM will only insert into the unstable tree, pages whose hash value
67 * has not changed since the previous scan of all memory areas.
68 * 3) The unstable tree is a RedBlack Tree - so its balancing is based on the
69 * colors of the nodes and not on their contents, assuring that even when
70 * the tree gets "corrupted" it won't get out of balance, so scanning time
71 * remains the same (also, searching and inserting nodes in an rbtree uses
72 * the same algorithm, so we have no overhead when we flush and rebuild).
73 * 4) KSM never flushes the stable tree, which means that even if it were to
74 * take 10 attempts to find a page in the unstable tree, once it is found,
75 * it is secured in the stable tree. (When we scan a new page, we first
76 * compare it against the stable tree, and then against the unstable tree.)
80 * struct mm_slot - ksm information per mm that is being scanned
81 * @link: link to the mm_slots hash list
82 * @mm_list: link into the mm_slots list, rooted in ksm_mm_head
83 * @rmap_list: head for this mm_slot's singly-linked list of rmap_items
84 * @mm: the mm that this information is valid for
87 struct hlist_node link;
88 struct list_head mm_list;
89 struct rmap_item *rmap_list;
94 * struct ksm_scan - cursor for scanning
95 * @mm_slot: the current mm_slot we are scanning
96 * @address: the next address inside that to be scanned
97 * @rmap_list: link to the next rmap to be scanned in the rmap_list
98 * @seqnr: count of completed full scans (needed when removing unstable node)
100 * There is only the one ksm_scan instance of this cursor structure.
103 struct mm_slot *mm_slot;
104 unsigned long address;
105 struct rmap_item **rmap_list;
110 * struct stable_node - node of the stable rbtree
111 * @page: pointer to struct page of the ksm page
112 * @node: rb node of this ksm page in the stable tree
113 * @hlist: hlist head of rmap_items using this ksm page
118 struct hlist_head hlist;
122 * struct rmap_item - reverse mapping item for virtual addresses
123 * @rmap_list: next rmap_item in mm_slot's singly-linked rmap_list
124 * @filler: unused space we're making available in this patch
125 * @mm: the memory structure this rmap_item is pointing into
126 * @address: the virtual address this rmap_item tracks (+ flags in low bits)
127 * @oldchecksum: previous checksum of the page at that virtual address
128 * @node: rb node of this rmap_item in the unstable tree
129 * @head: pointer to stable_node heading this list in the stable tree
130 * @hlist: link into hlist of rmap_items hanging off that stable_node
133 struct rmap_item *rmap_list;
134 unsigned long filler;
135 struct mm_struct *mm;
136 unsigned long address; /* + low bits used for flags below */
137 unsigned int oldchecksum; /* when unstable */
139 struct rb_node node; /* when node of unstable tree */
140 struct { /* when listed from stable tree */
141 struct stable_node *head;
142 struct hlist_node hlist;
147 #define SEQNR_MASK 0x0ff /* low bits of unstable tree seqnr */
148 #define UNSTABLE_FLAG 0x100 /* is a node of the unstable tree */
149 #define STABLE_FLAG 0x200 /* is listed from the stable tree */
151 /* The stable and unstable tree heads */
152 static struct rb_root root_stable_tree = RB_ROOT;
153 static struct rb_root root_unstable_tree = RB_ROOT;
155 #define MM_SLOTS_HASH_HEADS 1024
156 static struct hlist_head *mm_slots_hash;
158 static struct mm_slot ksm_mm_head = {
159 .mm_list = LIST_HEAD_INIT(ksm_mm_head.mm_list),
161 static struct ksm_scan ksm_scan = {
162 .mm_slot = &ksm_mm_head,
165 static struct kmem_cache *rmap_item_cache;
166 static struct kmem_cache *stable_node_cache;
167 static struct kmem_cache *mm_slot_cache;
169 /* The number of nodes in the stable tree */
170 static unsigned long ksm_pages_shared;
172 /* The number of page slots additionally sharing those nodes */
173 static unsigned long ksm_pages_sharing;
175 /* The number of nodes in the unstable tree */
176 static unsigned long ksm_pages_unshared;
178 /* The number of rmap_items in use: to calculate pages_volatile */
179 static unsigned long ksm_rmap_items;
181 /* Limit on the number of unswappable pages used */
182 static unsigned long ksm_max_kernel_pages;
184 /* Number of pages ksmd should scan in one batch */
185 static unsigned int ksm_thread_pages_to_scan = 100;
187 /* Milliseconds ksmd should sleep between batches */
188 static unsigned int ksm_thread_sleep_millisecs = 20;
190 #define KSM_RUN_STOP 0
191 #define KSM_RUN_MERGE 1
192 #define KSM_RUN_UNMERGE 2
193 static unsigned int ksm_run = KSM_RUN_STOP;
195 static DECLARE_WAIT_QUEUE_HEAD(ksm_thread_wait);
196 static DEFINE_MUTEX(ksm_thread_mutex);
197 static DEFINE_SPINLOCK(ksm_mmlist_lock);
200 * Temporary hack for page_referenced_ksm() and try_to_unmap_ksm(),
201 * later we rework things a little to get the right vma to them.
203 static DEFINE_SPINLOCK(ksm_fallback_vma_lock);
204 static struct vm_area_struct ksm_fallback_vma;
206 #define KSM_KMEM_CACHE(__struct, __flags) kmem_cache_create("ksm_"#__struct,\
207 sizeof(struct __struct), __alignof__(struct __struct),\
210 static int __init ksm_slab_init(void)
212 rmap_item_cache = KSM_KMEM_CACHE(rmap_item, 0);
213 if (!rmap_item_cache)
216 stable_node_cache = KSM_KMEM_CACHE(stable_node, 0);
217 if (!stable_node_cache)
220 mm_slot_cache = KSM_KMEM_CACHE(mm_slot, 0);
227 kmem_cache_destroy(stable_node_cache);
229 kmem_cache_destroy(rmap_item_cache);
234 static void __init ksm_slab_free(void)
236 kmem_cache_destroy(mm_slot_cache);
237 kmem_cache_destroy(stable_node_cache);
238 kmem_cache_destroy(rmap_item_cache);
239 mm_slot_cache = NULL;
242 static inline struct rmap_item *alloc_rmap_item(void)
244 struct rmap_item *rmap_item;
246 rmap_item = kmem_cache_zalloc(rmap_item_cache, GFP_KERNEL);
252 static inline void free_rmap_item(struct rmap_item *rmap_item)
255 rmap_item->mm = NULL; /* debug safety */
256 kmem_cache_free(rmap_item_cache, rmap_item);
259 static inline struct stable_node *alloc_stable_node(void)
261 return kmem_cache_alloc(stable_node_cache, GFP_KERNEL);
264 static inline void free_stable_node(struct stable_node *stable_node)
266 kmem_cache_free(stable_node_cache, stable_node);
269 static inline struct mm_slot *alloc_mm_slot(void)
271 if (!mm_slot_cache) /* initialization failed */
273 return kmem_cache_zalloc(mm_slot_cache, GFP_KERNEL);
276 static inline void free_mm_slot(struct mm_slot *mm_slot)
278 kmem_cache_free(mm_slot_cache, mm_slot);
281 static int __init mm_slots_hash_init(void)
283 mm_slots_hash = kzalloc(MM_SLOTS_HASH_HEADS * sizeof(struct hlist_head),
290 static void __init mm_slots_hash_free(void)
292 kfree(mm_slots_hash);
295 static struct mm_slot *get_mm_slot(struct mm_struct *mm)
297 struct mm_slot *mm_slot;
298 struct hlist_head *bucket;
299 struct hlist_node *node;
301 bucket = &mm_slots_hash[((unsigned long)mm / sizeof(struct mm_struct))
302 % MM_SLOTS_HASH_HEADS];
303 hlist_for_each_entry(mm_slot, node, bucket, link) {
304 if (mm == mm_slot->mm)
310 static void insert_to_mm_slots_hash(struct mm_struct *mm,
311 struct mm_slot *mm_slot)
313 struct hlist_head *bucket;
315 bucket = &mm_slots_hash[((unsigned long)mm / sizeof(struct mm_struct))
316 % MM_SLOTS_HASH_HEADS];
318 hlist_add_head(&mm_slot->link, bucket);
321 static inline int in_stable_tree(struct rmap_item *rmap_item)
323 return rmap_item->address & STABLE_FLAG;
327 * ksmd, and unmerge_and_remove_all_rmap_items(), must not touch an mm's
328 * page tables after it has passed through ksm_exit() - which, if necessary,
329 * takes mmap_sem briefly to serialize against them. ksm_exit() does not set
330 * a special flag: they can just back out as soon as mm_users goes to zero.
331 * ksm_test_exit() is used throughout to make this test for exit: in some
332 * places for correctness, in some places just to avoid unnecessary work.
334 static inline bool ksm_test_exit(struct mm_struct *mm)
336 return atomic_read(&mm->mm_users) == 0;
340 * We use break_ksm to break COW on a ksm page: it's a stripped down
342 * if (get_user_pages(current, mm, addr, 1, 1, 1, &page, NULL) == 1)
345 * but taking great care only to touch a ksm page, in a VM_MERGEABLE vma,
346 * in case the application has unmapped and remapped mm,addr meanwhile.
347 * Could a ksm page appear anywhere else? Actually yes, in a VM_PFNMAP
348 * mmap of /dev/mem or /dev/kmem, where we would not want to touch it.
350 static int break_ksm(struct vm_area_struct *vma, unsigned long addr)
357 page = follow_page(vma, addr, FOLL_GET);
361 ret = handle_mm_fault(vma->vm_mm, vma, addr,
364 ret = VM_FAULT_WRITE;
366 } while (!(ret & (VM_FAULT_WRITE | VM_FAULT_SIGBUS | VM_FAULT_OOM)));
368 * We must loop because handle_mm_fault() may back out if there's
369 * any difficulty e.g. if pte accessed bit gets updated concurrently.
371 * VM_FAULT_WRITE is what we have been hoping for: it indicates that
372 * COW has been broken, even if the vma does not permit VM_WRITE;
373 * but note that a concurrent fault might break PageKsm for us.
375 * VM_FAULT_SIGBUS could occur if we race with truncation of the
376 * backing file, which also invalidates anonymous pages: that's
377 * okay, that truncation will have unmapped the PageKsm for us.
379 * VM_FAULT_OOM: at the time of writing (late July 2009), setting
380 * aside mem_cgroup limits, VM_FAULT_OOM would only be set if the
381 * current task has TIF_MEMDIE set, and will be OOM killed on return
382 * to user; and ksmd, having no mm, would never be chosen for that.
384 * But if the mm is in a limited mem_cgroup, then the fault may fail
385 * with VM_FAULT_OOM even if the current task is not TIF_MEMDIE; and
386 * even ksmd can fail in this way - though it's usually breaking ksm
387 * just to undo a merge it made a moment before, so unlikely to oom.
389 * That's a pity: we might therefore have more kernel pages allocated
390 * than we're counting as nodes in the stable tree; but ksm_do_scan
391 * will retry to break_cow on each pass, so should recover the page
392 * in due course. The important thing is to not let VM_MERGEABLE
393 * be cleared while any such pages might remain in the area.
395 return (ret & VM_FAULT_OOM) ? -ENOMEM : 0;
398 static void break_cow(struct rmap_item *rmap_item)
400 struct mm_struct *mm = rmap_item->mm;
401 unsigned long addr = rmap_item->address;
402 struct vm_area_struct *vma;
404 down_read(&mm->mmap_sem);
405 if (ksm_test_exit(mm))
407 vma = find_vma(mm, addr);
408 if (!vma || vma->vm_start > addr)
410 if (!(vma->vm_flags & VM_MERGEABLE) || !vma->anon_vma)
412 break_ksm(vma, addr);
414 up_read(&mm->mmap_sem);
417 static struct page *get_mergeable_page(struct rmap_item *rmap_item)
419 struct mm_struct *mm = rmap_item->mm;
420 unsigned long addr = rmap_item->address;
421 struct vm_area_struct *vma;
424 down_read(&mm->mmap_sem);
425 if (ksm_test_exit(mm))
427 vma = find_vma(mm, addr);
428 if (!vma || vma->vm_start > addr)
430 if (!(vma->vm_flags & VM_MERGEABLE) || !vma->anon_vma)
433 page = follow_page(vma, addr, FOLL_GET);
436 if (PageAnon(page)) {
437 flush_anon_page(vma, page, addr);
438 flush_dcache_page(page);
443 up_read(&mm->mmap_sem);
448 * Removing rmap_item from stable or unstable tree.
449 * This function will clean the information from the stable/unstable tree.
451 static void remove_rmap_item_from_tree(struct rmap_item *rmap_item)
453 if (rmap_item->address & STABLE_FLAG) {
454 struct stable_node *stable_node;
457 stable_node = rmap_item->head;
458 page = stable_node->page;
461 hlist_del(&rmap_item->hlist);
462 if (stable_node->hlist.first) {
466 set_page_stable_node(page, NULL);
470 rb_erase(&stable_node->node, &root_stable_tree);
471 free_stable_node(stable_node);
475 rmap_item->address &= PAGE_MASK;
477 } else if (rmap_item->address & UNSTABLE_FLAG) {
480 * Usually ksmd can and must skip the rb_erase, because
481 * root_unstable_tree was already reset to RB_ROOT.
482 * But be careful when an mm is exiting: do the rb_erase
483 * if this rmap_item was inserted by this scan, rather
484 * than left over from before.
486 age = (unsigned char)(ksm_scan.seqnr - rmap_item->address);
489 rb_erase(&rmap_item->node, &root_unstable_tree);
491 ksm_pages_unshared--;
492 rmap_item->address &= PAGE_MASK;
495 cond_resched(); /* we're called from many long loops */
498 static void remove_trailing_rmap_items(struct mm_slot *mm_slot,
499 struct rmap_item **rmap_list)
502 struct rmap_item *rmap_item = *rmap_list;
503 *rmap_list = rmap_item->rmap_list;
504 remove_rmap_item_from_tree(rmap_item);
505 free_rmap_item(rmap_item);
510 * Though it's very tempting to unmerge in_stable_tree(rmap_item)s rather
511 * than check every pte of a given vma, the locking doesn't quite work for
512 * that - an rmap_item is assigned to the stable tree after inserting ksm
513 * page and upping mmap_sem. Nor does it fit with the way we skip dup'ing
514 * rmap_items from parent to child at fork time (so as not to waste time
515 * if exit comes before the next scan reaches it).
517 * Similarly, although we'd like to remove rmap_items (so updating counts
518 * and freeing memory) when unmerging an area, it's easier to leave that
519 * to the next pass of ksmd - consider, for example, how ksmd might be
520 * in cmp_and_merge_page on one of the rmap_items we would be removing.
522 static int unmerge_ksm_pages(struct vm_area_struct *vma,
523 unsigned long start, unsigned long end)
528 for (addr = start; addr < end && !err; addr += PAGE_SIZE) {
529 if (ksm_test_exit(vma->vm_mm))
531 if (signal_pending(current))
534 err = break_ksm(vma, addr);
541 * Only called through the sysfs control interface:
543 static int unmerge_and_remove_all_rmap_items(void)
545 struct mm_slot *mm_slot;
546 struct mm_struct *mm;
547 struct vm_area_struct *vma;
550 spin_lock(&ksm_mmlist_lock);
551 ksm_scan.mm_slot = list_entry(ksm_mm_head.mm_list.next,
552 struct mm_slot, mm_list);
553 spin_unlock(&ksm_mmlist_lock);
555 for (mm_slot = ksm_scan.mm_slot;
556 mm_slot != &ksm_mm_head; mm_slot = ksm_scan.mm_slot) {
558 down_read(&mm->mmap_sem);
559 for (vma = mm->mmap; vma; vma = vma->vm_next) {
560 if (ksm_test_exit(mm))
562 if (!(vma->vm_flags & VM_MERGEABLE) || !vma->anon_vma)
564 err = unmerge_ksm_pages(vma,
565 vma->vm_start, vma->vm_end);
570 remove_trailing_rmap_items(mm_slot, &mm_slot->rmap_list);
572 spin_lock(&ksm_mmlist_lock);
573 ksm_scan.mm_slot = list_entry(mm_slot->mm_list.next,
574 struct mm_slot, mm_list);
575 if (ksm_test_exit(mm)) {
576 hlist_del(&mm_slot->link);
577 list_del(&mm_slot->mm_list);
578 spin_unlock(&ksm_mmlist_lock);
580 free_mm_slot(mm_slot);
581 clear_bit(MMF_VM_MERGEABLE, &mm->flags);
582 up_read(&mm->mmap_sem);
585 spin_unlock(&ksm_mmlist_lock);
586 up_read(&mm->mmap_sem);
594 up_read(&mm->mmap_sem);
595 spin_lock(&ksm_mmlist_lock);
596 ksm_scan.mm_slot = &ksm_mm_head;
597 spin_unlock(&ksm_mmlist_lock);
600 #endif /* CONFIG_SYSFS */
602 static u32 calc_checksum(struct page *page)
605 void *addr = kmap_atomic(page, KM_USER0);
606 checksum = jhash2(addr, PAGE_SIZE / 4, 17);
607 kunmap_atomic(addr, KM_USER0);
611 static int memcmp_pages(struct page *page1, struct page *page2)
616 addr1 = kmap_atomic(page1, KM_USER0);
617 addr2 = kmap_atomic(page2, KM_USER1);
618 ret = memcmp(addr1, addr2, PAGE_SIZE);
619 kunmap_atomic(addr2, KM_USER1);
620 kunmap_atomic(addr1, KM_USER0);
624 static inline int pages_identical(struct page *page1, struct page *page2)
626 return !memcmp_pages(page1, page2);
629 static int write_protect_page(struct vm_area_struct *vma, struct page *page,
632 struct mm_struct *mm = vma->vm_mm;
639 addr = page_address_in_vma(page, vma);
643 ptep = page_check_address(page, mm, addr, &ptl, 0);
647 if (pte_write(*ptep)) {
650 swapped = PageSwapCache(page);
651 flush_cache_page(vma, addr, page_to_pfn(page));
653 * Ok this is tricky, when get_user_pages_fast() run it doesnt
654 * take any lock, therefore the check that we are going to make
655 * with the pagecount against the mapcount is racey and
656 * O_DIRECT can happen right after the check.
657 * So we clear the pte and flush the tlb before the check
658 * this assure us that no O_DIRECT can happen after the check
659 * or in the middle of the check.
661 entry = ptep_clear_flush(vma, addr, ptep);
663 * Check that no O_DIRECT or similar I/O is in progress on the
666 if (page_mapcount(page) + 1 + swapped != page_count(page)) {
667 set_pte_at_notify(mm, addr, ptep, entry);
670 entry = pte_wrprotect(entry);
671 set_pte_at_notify(mm, addr, ptep, entry);
677 pte_unmap_unlock(ptep, ptl);
683 * replace_page - replace page in vma by new ksm page
684 * @vma: vma that holds the pte pointing to page
685 * @page: the page we are replacing by kpage
686 * @kpage: the ksm page we replace page by
687 * @orig_pte: the original value of the pte
689 * Returns 0 on success, -EFAULT on failure.
691 static int replace_page(struct vm_area_struct *vma, struct page *page,
692 struct page *kpage, pte_t orig_pte)
694 struct mm_struct *mm = vma->vm_mm;
703 addr = page_address_in_vma(page, vma);
707 pgd = pgd_offset(mm, addr);
708 if (!pgd_present(*pgd))
711 pud = pud_offset(pgd, addr);
712 if (!pud_present(*pud))
715 pmd = pmd_offset(pud, addr);
716 if (!pmd_present(*pmd))
719 ptep = pte_offset_map_lock(mm, pmd, addr, &ptl);
720 if (!pte_same(*ptep, orig_pte)) {
721 pte_unmap_unlock(ptep, ptl);
726 page_add_anon_rmap(kpage, vma, addr);
728 flush_cache_page(vma, addr, pte_pfn(*ptep));
729 ptep_clear_flush(vma, addr, ptep);
730 set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot));
732 page_remove_rmap(page);
735 pte_unmap_unlock(ptep, ptl);
742 * try_to_merge_one_page - take two pages and merge them into one
743 * @vma: the vma that holds the pte pointing to page
744 * @page: the PageAnon page that we want to replace with kpage
745 * @kpage: the PageKsm page that we want to map instead of page
747 * This function returns 0 if the pages were merged, -EFAULT otherwise.
749 static int try_to_merge_one_page(struct vm_area_struct *vma,
750 struct page *page, struct page *kpage)
752 pte_t orig_pte = __pte(0);
755 if (!(vma->vm_flags & VM_MERGEABLE))
761 * We need the page lock to read a stable PageSwapCache in
762 * write_protect_page(). We use trylock_page() instead of
763 * lock_page() because we don't want to wait here - we
764 * prefer to continue scanning and merging different pages,
765 * then come back to this page when it is unlocked.
767 if (!trylock_page(page))
770 * If this anonymous page is mapped only here, its pte may need
771 * to be write-protected. If it's mapped elsewhere, all of its
772 * ptes are necessarily already write-protected. But in either
773 * case, we need to lock and check page_count is not raised.
775 if (write_protect_page(vma, page, &orig_pte) == 0 &&
776 pages_identical(page, kpage))
777 err = replace_page(vma, page, kpage, orig_pte);
779 if ((vma->vm_flags & VM_LOCKED) && !err) {
780 munlock_vma_page(page);
781 if (!PageMlocked(kpage)) {
785 mlock_vma_page(kpage);
786 page = kpage; /* for final unlock */
796 * try_to_merge_with_ksm_page - like try_to_merge_two_pages,
797 * but no new kernel page is allocated: kpage must already be a ksm page.
799 * This function returns 0 if the pages were merged, -EFAULT otherwise.
801 static int try_to_merge_with_ksm_page(struct rmap_item *rmap_item,
802 struct page *page, struct page *kpage)
804 struct mm_struct *mm = rmap_item->mm;
805 struct vm_area_struct *vma;
808 if (page == kpage) /* ksm page forked */
811 down_read(&mm->mmap_sem);
812 if (ksm_test_exit(mm))
814 vma = find_vma(mm, rmap_item->address);
815 if (!vma || vma->vm_start > rmap_item->address)
818 err = try_to_merge_one_page(vma, page, kpage);
820 up_read(&mm->mmap_sem);
825 * try_to_merge_two_pages - take two identical pages and prepare them
826 * to be merged into one page.
828 * This function returns the kpage if we successfully merged two identical
829 * pages into one ksm page, NULL otherwise.
831 * Note that this function allocates a new kernel page: if one of the pages
832 * is already a ksm page, try_to_merge_with_ksm_page should be used.
834 static struct page *try_to_merge_two_pages(struct rmap_item *rmap_item,
836 struct rmap_item *tree_rmap_item,
837 struct page *tree_page)
839 struct mm_struct *mm = rmap_item->mm;
840 struct vm_area_struct *vma;
845 * The number of nodes in the stable tree
846 * is the number of kernel pages that we hold.
848 if (ksm_max_kernel_pages &&
849 ksm_max_kernel_pages <= ksm_pages_shared)
852 kpage = alloc_page(GFP_HIGHUSER);
856 down_read(&mm->mmap_sem);
857 if (ksm_test_exit(mm))
859 vma = find_vma(mm, rmap_item->address);
860 if (!vma || vma->vm_start > rmap_item->address)
863 copy_user_highpage(kpage, page, rmap_item->address, vma);
866 __SetPageUptodate(kpage);
867 SetPageSwapBacked(kpage);
868 set_page_stable_node(kpage, NULL); /* mark it PageKsm */
869 lru_cache_add_lru(kpage, LRU_ACTIVE_ANON);
871 err = try_to_merge_one_page(vma, page, kpage);
873 up_read(&mm->mmap_sem);
876 err = try_to_merge_with_ksm_page(tree_rmap_item,
879 * If that fails, we have a ksm page with only one pte
880 * pointing to it: so break it.
883 break_cow(rmap_item);
893 * stable_tree_search - search for page inside the stable tree
895 * This function checks if there is a page inside the stable tree
896 * with identical content to the page that we are scanning right now.
898 * This function returns the stable tree node of identical content if found,
901 static struct stable_node *stable_tree_search(struct page *page)
903 struct rb_node *node = root_stable_tree.rb_node;
904 struct stable_node *stable_node;
906 stable_node = page_stable_node(page);
907 if (stable_node) { /* ksm page forked */
916 stable_node = rb_entry(node, struct stable_node, node);
918 ret = memcmp_pages(page, stable_node->page);
921 node = node->rb_left;
923 node = node->rb_right;
925 get_page(stable_node->page);
934 * stable_tree_insert - insert rmap_item pointing to new ksm page
935 * into the stable tree.
937 * This function returns the stable tree node just allocated on success,
940 static struct stable_node *stable_tree_insert(struct page *kpage)
942 struct rb_node **new = &root_stable_tree.rb_node;
943 struct rb_node *parent = NULL;
944 struct stable_node *stable_node;
950 stable_node = rb_entry(*new, struct stable_node, node);
952 ret = memcmp_pages(kpage, stable_node->page);
956 new = &parent->rb_left;
958 new = &parent->rb_right;
961 * It is not a bug that stable_tree_search() didn't
962 * find this node: because at that time our page was
963 * not yet write-protected, so may have changed since.
969 stable_node = alloc_stable_node();
973 rb_link_node(&stable_node->node, parent, new);
974 rb_insert_color(&stable_node->node, &root_stable_tree);
976 INIT_HLIST_HEAD(&stable_node->hlist);
979 stable_node->page = kpage;
980 set_page_stable_node(kpage, stable_node);
986 * unstable_tree_search_insert - search for identical page,
987 * else insert rmap_item into the unstable tree.
989 * This function searches for a page in the unstable tree identical to the
990 * page currently being scanned; and if no identical page is found in the
991 * tree, we insert rmap_item as a new object into the unstable tree.
993 * This function returns pointer to rmap_item found to be identical
994 * to the currently scanned page, NULL otherwise.
996 * This function does both searching and inserting, because they share
997 * the same walking algorithm in an rbtree.
1000 struct rmap_item *unstable_tree_search_insert(struct rmap_item *rmap_item,
1002 struct page **tree_pagep)
1005 struct rb_node **new = &root_unstable_tree.rb_node;
1006 struct rb_node *parent = NULL;
1009 struct rmap_item *tree_rmap_item;
1010 struct page *tree_page;
1014 tree_rmap_item = rb_entry(*new, struct rmap_item, node);
1015 tree_page = get_mergeable_page(tree_rmap_item);
1020 * Don't substitute a ksm page for a forked page.
1022 if (page == tree_page) {
1023 put_page(tree_page);
1027 ret = memcmp_pages(page, tree_page);
1031 put_page(tree_page);
1032 new = &parent->rb_left;
1033 } else if (ret > 0) {
1034 put_page(tree_page);
1035 new = &parent->rb_right;
1037 *tree_pagep = tree_page;
1038 return tree_rmap_item;
1042 rmap_item->address |= UNSTABLE_FLAG;
1043 rmap_item->address |= (ksm_scan.seqnr & SEQNR_MASK);
1044 rb_link_node(&rmap_item->node, parent, new);
1045 rb_insert_color(&rmap_item->node, &root_unstable_tree);
1047 ksm_pages_unshared++;
1052 * stable_tree_append - add another rmap_item to the linked list of
1053 * rmap_items hanging off a given node of the stable tree, all sharing
1054 * the same ksm page.
1056 static void stable_tree_append(struct rmap_item *rmap_item,
1057 struct stable_node *stable_node)
1059 rmap_item->head = stable_node;
1060 rmap_item->address |= STABLE_FLAG;
1061 hlist_add_head(&rmap_item->hlist, &stable_node->hlist);
1063 if (rmap_item->hlist.next)
1064 ksm_pages_sharing++;
1070 * cmp_and_merge_page - first see if page can be merged into the stable tree;
1071 * if not, compare checksum to previous and if it's the same, see if page can
1072 * be inserted into the unstable tree, or merged with a page already there and
1073 * both transferred to the stable tree.
1075 * @page: the page that we are searching identical page to.
1076 * @rmap_item: the reverse mapping into the virtual address of this page
1078 static void cmp_and_merge_page(struct page *page, struct rmap_item *rmap_item)
1080 struct rmap_item *tree_rmap_item;
1081 struct page *tree_page = NULL;
1082 struct stable_node *stable_node;
1084 unsigned int checksum;
1087 remove_rmap_item_from_tree(rmap_item);
1089 /* We first start with searching the page inside the stable tree */
1090 stable_node = stable_tree_search(page);
1092 kpage = stable_node->page;
1093 err = try_to_merge_with_ksm_page(rmap_item, page, kpage);
1096 * The page was successfully merged:
1097 * add its rmap_item to the stable tree.
1100 stable_tree_append(rmap_item, stable_node);
1108 * A ksm page might have got here by fork, but its other
1109 * references have already been removed from the stable tree.
1110 * Or it might be left over from a break_ksm which failed
1111 * when the mem_cgroup had reached its limit: try again now.
1114 break_cow(rmap_item);
1117 * In case the hash value of the page was changed from the last time we
1118 * have calculated it, this page to be changed frequely, therefore we
1119 * don't want to insert it to the unstable tree, and we don't want to
1120 * waste our time to search if there is something identical to it there.
1122 checksum = calc_checksum(page);
1123 if (rmap_item->oldchecksum != checksum) {
1124 rmap_item->oldchecksum = checksum;
1129 unstable_tree_search_insert(rmap_item, page, &tree_page);
1130 if (tree_rmap_item) {
1131 kpage = try_to_merge_two_pages(rmap_item, page,
1132 tree_rmap_item, tree_page);
1133 put_page(tree_page);
1135 * As soon as we merge this page, we want to remove the
1136 * rmap_item of the page we have merged with from the unstable
1137 * tree, and insert it instead as new node in the stable tree.
1140 remove_rmap_item_from_tree(tree_rmap_item);
1143 stable_node = stable_tree_insert(kpage);
1145 stable_tree_append(tree_rmap_item, stable_node);
1146 stable_tree_append(rmap_item, stable_node);
1152 * If we fail to insert the page into the stable tree,
1153 * we will have 2 virtual addresses that are pointing
1154 * to a ksm page left outside the stable tree,
1155 * in which case we need to break_cow on both.
1158 break_cow(tree_rmap_item);
1159 break_cow(rmap_item);
1165 static struct rmap_item *get_next_rmap_item(struct mm_slot *mm_slot,
1166 struct rmap_item **rmap_list,
1169 struct rmap_item *rmap_item;
1171 while (*rmap_list) {
1172 rmap_item = *rmap_list;
1173 if ((rmap_item->address & PAGE_MASK) == addr)
1175 if (rmap_item->address > addr)
1177 *rmap_list = rmap_item->rmap_list;
1178 remove_rmap_item_from_tree(rmap_item);
1179 free_rmap_item(rmap_item);
1182 rmap_item = alloc_rmap_item();
1184 /* It has already been zeroed */
1185 rmap_item->mm = mm_slot->mm;
1186 rmap_item->address = addr;
1187 rmap_item->rmap_list = *rmap_list;
1188 *rmap_list = rmap_item;
1193 static struct rmap_item *scan_get_next_rmap_item(struct page **page)
1195 struct mm_struct *mm;
1196 struct mm_slot *slot;
1197 struct vm_area_struct *vma;
1198 struct rmap_item *rmap_item;
1200 if (list_empty(&ksm_mm_head.mm_list))
1203 slot = ksm_scan.mm_slot;
1204 if (slot == &ksm_mm_head) {
1205 root_unstable_tree = RB_ROOT;
1207 spin_lock(&ksm_mmlist_lock);
1208 slot = list_entry(slot->mm_list.next, struct mm_slot, mm_list);
1209 ksm_scan.mm_slot = slot;
1210 spin_unlock(&ksm_mmlist_lock);
1212 ksm_scan.address = 0;
1213 ksm_scan.rmap_list = &slot->rmap_list;
1217 down_read(&mm->mmap_sem);
1218 if (ksm_test_exit(mm))
1221 vma = find_vma(mm, ksm_scan.address);
1223 for (; vma; vma = vma->vm_next) {
1224 if (!(vma->vm_flags & VM_MERGEABLE))
1226 if (ksm_scan.address < vma->vm_start)
1227 ksm_scan.address = vma->vm_start;
1229 ksm_scan.address = vma->vm_end;
1231 while (ksm_scan.address < vma->vm_end) {
1232 if (ksm_test_exit(mm))
1234 *page = follow_page(vma, ksm_scan.address, FOLL_GET);
1235 if (*page && PageAnon(*page)) {
1236 flush_anon_page(vma, *page, ksm_scan.address);
1237 flush_dcache_page(*page);
1238 rmap_item = get_next_rmap_item(slot,
1239 ksm_scan.rmap_list, ksm_scan.address);
1241 ksm_scan.rmap_list =
1242 &rmap_item->rmap_list;
1243 ksm_scan.address += PAGE_SIZE;
1246 up_read(&mm->mmap_sem);
1251 ksm_scan.address += PAGE_SIZE;
1256 if (ksm_test_exit(mm)) {
1257 ksm_scan.address = 0;
1258 ksm_scan.rmap_list = &slot->rmap_list;
1261 * Nuke all the rmap_items that are above this current rmap:
1262 * because there were no VM_MERGEABLE vmas with such addresses.
1264 remove_trailing_rmap_items(slot, ksm_scan.rmap_list);
1266 spin_lock(&ksm_mmlist_lock);
1267 ksm_scan.mm_slot = list_entry(slot->mm_list.next,
1268 struct mm_slot, mm_list);
1269 if (ksm_scan.address == 0) {
1271 * We've completed a full scan of all vmas, holding mmap_sem
1272 * throughout, and found no VM_MERGEABLE: so do the same as
1273 * __ksm_exit does to remove this mm from all our lists now.
1274 * This applies either when cleaning up after __ksm_exit
1275 * (but beware: we can reach here even before __ksm_exit),
1276 * or when all VM_MERGEABLE areas have been unmapped (and
1277 * mmap_sem then protects against race with MADV_MERGEABLE).
1279 hlist_del(&slot->link);
1280 list_del(&slot->mm_list);
1281 spin_unlock(&ksm_mmlist_lock);
1284 clear_bit(MMF_VM_MERGEABLE, &mm->flags);
1285 up_read(&mm->mmap_sem);
1288 spin_unlock(&ksm_mmlist_lock);
1289 up_read(&mm->mmap_sem);
1292 /* Repeat until we've completed scanning the whole list */
1293 slot = ksm_scan.mm_slot;
1294 if (slot != &ksm_mm_head)
1302 * ksm_do_scan - the ksm scanner main worker function.
1303 * @scan_npages - number of pages we want to scan before we return.
1305 static void ksm_do_scan(unsigned int scan_npages)
1307 struct rmap_item *rmap_item;
1310 while (scan_npages--) {
1312 rmap_item = scan_get_next_rmap_item(&page);
1315 if (!PageKsm(page) || !in_stable_tree(rmap_item))
1316 cmp_and_merge_page(page, rmap_item);
1321 static int ksmd_should_run(void)
1323 return (ksm_run & KSM_RUN_MERGE) && !list_empty(&ksm_mm_head.mm_list);
1326 static int ksm_scan_thread(void *nothing)
1328 set_user_nice(current, 5);
1330 while (!kthread_should_stop()) {
1331 mutex_lock(&ksm_thread_mutex);
1332 if (ksmd_should_run())
1333 ksm_do_scan(ksm_thread_pages_to_scan);
1334 mutex_unlock(&ksm_thread_mutex);
1336 if (ksmd_should_run()) {
1337 schedule_timeout_interruptible(
1338 msecs_to_jiffies(ksm_thread_sleep_millisecs));
1340 wait_event_interruptible(ksm_thread_wait,
1341 ksmd_should_run() || kthread_should_stop());
1347 int ksm_madvise(struct vm_area_struct *vma, unsigned long start,
1348 unsigned long end, int advice, unsigned long *vm_flags)
1350 struct mm_struct *mm = vma->vm_mm;
1354 case MADV_MERGEABLE:
1356 * Be somewhat over-protective for now!
1358 if (*vm_flags & (VM_MERGEABLE | VM_SHARED | VM_MAYSHARE |
1359 VM_PFNMAP | VM_IO | VM_DONTEXPAND |
1360 VM_RESERVED | VM_HUGETLB | VM_INSERTPAGE |
1361 VM_NONLINEAR | VM_MIXEDMAP | VM_SAO))
1362 return 0; /* just ignore the advice */
1364 if (!test_bit(MMF_VM_MERGEABLE, &mm->flags)) {
1365 err = __ksm_enter(mm);
1370 *vm_flags |= VM_MERGEABLE;
1373 case MADV_UNMERGEABLE:
1374 if (!(*vm_flags & VM_MERGEABLE))
1375 return 0; /* just ignore the advice */
1377 if (vma->anon_vma) {
1378 err = unmerge_ksm_pages(vma, start, end);
1383 *vm_flags &= ~VM_MERGEABLE;
1390 int __ksm_enter(struct mm_struct *mm)
1392 struct mm_slot *mm_slot;
1395 mm_slot = alloc_mm_slot();
1399 /* Check ksm_run too? Would need tighter locking */
1400 needs_wakeup = list_empty(&ksm_mm_head.mm_list);
1402 spin_lock(&ksm_mmlist_lock);
1403 insert_to_mm_slots_hash(mm, mm_slot);
1405 * Insert just behind the scanning cursor, to let the area settle
1406 * down a little; when fork is followed by immediate exec, we don't
1407 * want ksmd to waste time setting up and tearing down an rmap_list.
1409 list_add_tail(&mm_slot->mm_list, &ksm_scan.mm_slot->mm_list);
1410 spin_unlock(&ksm_mmlist_lock);
1412 set_bit(MMF_VM_MERGEABLE, &mm->flags);
1413 atomic_inc(&mm->mm_count);
1416 wake_up_interruptible(&ksm_thread_wait);
1421 void __ksm_exit(struct mm_struct *mm)
1423 struct mm_slot *mm_slot;
1424 int easy_to_free = 0;
1427 * This process is exiting: if it's straightforward (as is the
1428 * case when ksmd was never running), free mm_slot immediately.
1429 * But if it's at the cursor or has rmap_items linked to it, use
1430 * mmap_sem to synchronize with any break_cows before pagetables
1431 * are freed, and leave the mm_slot on the list for ksmd to free.
1432 * Beware: ksm may already have noticed it exiting and freed the slot.
1435 spin_lock(&ksm_mmlist_lock);
1436 mm_slot = get_mm_slot(mm);
1437 if (mm_slot && ksm_scan.mm_slot != mm_slot) {
1438 if (!mm_slot->rmap_list) {
1439 hlist_del(&mm_slot->link);
1440 list_del(&mm_slot->mm_list);
1443 list_move(&mm_slot->mm_list,
1444 &ksm_scan.mm_slot->mm_list);
1447 spin_unlock(&ksm_mmlist_lock);
1450 free_mm_slot(mm_slot);
1451 clear_bit(MMF_VM_MERGEABLE, &mm->flags);
1453 } else if (mm_slot) {
1454 down_write(&mm->mmap_sem);
1455 up_write(&mm->mmap_sem);
1459 struct page *ksm_does_need_to_copy(struct page *page,
1460 struct vm_area_struct *vma, unsigned long address)
1462 struct page *new_page;
1464 unlock_page(page); /* any racers will COW it, not modify it */
1466 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
1468 copy_user_highpage(new_page, page, address, vma);
1470 SetPageDirty(new_page);
1471 __SetPageUptodate(new_page);
1472 SetPageSwapBacked(new_page);
1473 __set_page_locked(new_page);
1475 if (page_evictable(new_page, vma))
1476 lru_cache_add_lru(new_page, LRU_ACTIVE_ANON);
1478 add_page_to_unevictable_list(new_page);
1481 page_cache_release(page);
1485 int page_referenced_ksm(struct page *page, struct mem_cgroup *memcg,
1486 unsigned long *vm_flags)
1488 struct stable_node *stable_node;
1489 struct rmap_item *rmap_item;
1490 struct hlist_node *hlist;
1491 unsigned int mapcount = page_mapcount(page);
1493 struct vm_area_struct *vma;
1495 VM_BUG_ON(!PageKsm(page));
1496 VM_BUG_ON(!PageLocked(page));
1498 stable_node = page_stable_node(page);
1503 * Temporary hack: really we need anon_vma in rmap_item, to
1504 * provide the correct vma, and to find recently forked instances.
1505 * Use zalloc to avoid weirdness if any other fields are involved.
1507 vma = kmem_cache_zalloc(vm_area_cachep, GFP_ATOMIC);
1509 spin_lock(&ksm_fallback_vma_lock);
1510 vma = &ksm_fallback_vma;
1513 hlist_for_each_entry(rmap_item, hlist, &stable_node->hlist, hlist) {
1514 if (memcg && !mm_match_cgroup(rmap_item->mm, memcg))
1517 vma->vm_mm = rmap_item->mm;
1518 vma->vm_start = rmap_item->address;
1519 vma->vm_end = vma->vm_start + PAGE_SIZE;
1521 referenced += page_referenced_one(page, vma,
1522 rmap_item->address, &mapcount, vm_flags);
1527 if (vma == &ksm_fallback_vma)
1528 spin_unlock(&ksm_fallback_vma_lock);
1530 kmem_cache_free(vm_area_cachep, vma);
1534 int try_to_unmap_ksm(struct page *page, enum ttu_flags flags)
1536 struct stable_node *stable_node;
1537 struct hlist_node *hlist;
1538 struct rmap_item *rmap_item;
1539 int ret = SWAP_AGAIN;
1540 struct vm_area_struct *vma;
1542 VM_BUG_ON(!PageKsm(page));
1543 VM_BUG_ON(!PageLocked(page));
1545 stable_node = page_stable_node(page);
1550 * Temporary hack: really we need anon_vma in rmap_item, to
1551 * provide the correct vma, and to find recently forked instances.
1552 * Use zalloc to avoid weirdness if any other fields are involved.
1554 if (TTU_ACTION(flags) != TTU_UNMAP)
1557 vma = kmem_cache_zalloc(vm_area_cachep, GFP_ATOMIC);
1559 spin_lock(&ksm_fallback_vma_lock);
1560 vma = &ksm_fallback_vma;
1563 hlist_for_each_entry(rmap_item, hlist, &stable_node->hlist, hlist) {
1564 vma->vm_mm = rmap_item->mm;
1565 vma->vm_start = rmap_item->address;
1566 vma->vm_end = vma->vm_start + PAGE_SIZE;
1568 ret = try_to_unmap_one(page, vma, rmap_item->address, flags);
1569 if (ret != SWAP_AGAIN || !page_mapped(page))
1573 if (vma == &ksm_fallback_vma)
1574 spin_unlock(&ksm_fallback_vma_lock);
1576 kmem_cache_free(vm_area_cachep, vma);
1582 * This all compiles without CONFIG_SYSFS, but is a waste of space.
1585 #define KSM_ATTR_RO(_name) \
1586 static struct kobj_attribute _name##_attr = __ATTR_RO(_name)
1587 #define KSM_ATTR(_name) \
1588 static struct kobj_attribute _name##_attr = \
1589 __ATTR(_name, 0644, _name##_show, _name##_store)
1591 static ssize_t sleep_millisecs_show(struct kobject *kobj,
1592 struct kobj_attribute *attr, char *buf)
1594 return sprintf(buf, "%u\n", ksm_thread_sleep_millisecs);
1597 static ssize_t sleep_millisecs_store(struct kobject *kobj,
1598 struct kobj_attribute *attr,
1599 const char *buf, size_t count)
1601 unsigned long msecs;
1604 err = strict_strtoul(buf, 10, &msecs);
1605 if (err || msecs > UINT_MAX)
1608 ksm_thread_sleep_millisecs = msecs;
1612 KSM_ATTR(sleep_millisecs);
1614 static ssize_t pages_to_scan_show(struct kobject *kobj,
1615 struct kobj_attribute *attr, char *buf)
1617 return sprintf(buf, "%u\n", ksm_thread_pages_to_scan);
1620 static ssize_t pages_to_scan_store(struct kobject *kobj,
1621 struct kobj_attribute *attr,
1622 const char *buf, size_t count)
1625 unsigned long nr_pages;
1627 err = strict_strtoul(buf, 10, &nr_pages);
1628 if (err || nr_pages > UINT_MAX)
1631 ksm_thread_pages_to_scan = nr_pages;
1635 KSM_ATTR(pages_to_scan);
1637 static ssize_t run_show(struct kobject *kobj, struct kobj_attribute *attr,
1640 return sprintf(buf, "%u\n", ksm_run);
1643 static ssize_t run_store(struct kobject *kobj, struct kobj_attribute *attr,
1644 const char *buf, size_t count)
1647 unsigned long flags;
1649 err = strict_strtoul(buf, 10, &flags);
1650 if (err || flags > UINT_MAX)
1652 if (flags > KSM_RUN_UNMERGE)
1656 * KSM_RUN_MERGE sets ksmd running, and 0 stops it running.
1657 * KSM_RUN_UNMERGE stops it running and unmerges all rmap_items,
1658 * breaking COW to free the unswappable pages_shared (but leaves
1659 * mm_slots on the list for when ksmd may be set running again).
1662 mutex_lock(&ksm_thread_mutex);
1663 if (ksm_run != flags) {
1665 if (flags & KSM_RUN_UNMERGE) {
1666 current->flags |= PF_OOM_ORIGIN;
1667 err = unmerge_and_remove_all_rmap_items();
1668 current->flags &= ~PF_OOM_ORIGIN;
1670 ksm_run = KSM_RUN_STOP;
1675 mutex_unlock(&ksm_thread_mutex);
1677 if (flags & KSM_RUN_MERGE)
1678 wake_up_interruptible(&ksm_thread_wait);
1684 static ssize_t max_kernel_pages_store(struct kobject *kobj,
1685 struct kobj_attribute *attr,
1686 const char *buf, size_t count)
1689 unsigned long nr_pages;
1691 err = strict_strtoul(buf, 10, &nr_pages);
1695 ksm_max_kernel_pages = nr_pages;
1700 static ssize_t max_kernel_pages_show(struct kobject *kobj,
1701 struct kobj_attribute *attr, char *buf)
1703 return sprintf(buf, "%lu\n", ksm_max_kernel_pages);
1705 KSM_ATTR(max_kernel_pages);
1707 static ssize_t pages_shared_show(struct kobject *kobj,
1708 struct kobj_attribute *attr, char *buf)
1710 return sprintf(buf, "%lu\n", ksm_pages_shared);
1712 KSM_ATTR_RO(pages_shared);
1714 static ssize_t pages_sharing_show(struct kobject *kobj,
1715 struct kobj_attribute *attr, char *buf)
1717 return sprintf(buf, "%lu\n", ksm_pages_sharing);
1719 KSM_ATTR_RO(pages_sharing);
1721 static ssize_t pages_unshared_show(struct kobject *kobj,
1722 struct kobj_attribute *attr, char *buf)
1724 return sprintf(buf, "%lu\n", ksm_pages_unshared);
1726 KSM_ATTR_RO(pages_unshared);
1728 static ssize_t pages_volatile_show(struct kobject *kobj,
1729 struct kobj_attribute *attr, char *buf)
1731 long ksm_pages_volatile;
1733 ksm_pages_volatile = ksm_rmap_items - ksm_pages_shared
1734 - ksm_pages_sharing - ksm_pages_unshared;
1736 * It was not worth any locking to calculate that statistic,
1737 * but it might therefore sometimes be negative: conceal that.
1739 if (ksm_pages_volatile < 0)
1740 ksm_pages_volatile = 0;
1741 return sprintf(buf, "%ld\n", ksm_pages_volatile);
1743 KSM_ATTR_RO(pages_volatile);
1745 static ssize_t full_scans_show(struct kobject *kobj,
1746 struct kobj_attribute *attr, char *buf)
1748 return sprintf(buf, "%lu\n", ksm_scan.seqnr);
1750 KSM_ATTR_RO(full_scans);
1752 static struct attribute *ksm_attrs[] = {
1753 &sleep_millisecs_attr.attr,
1754 &pages_to_scan_attr.attr,
1756 &max_kernel_pages_attr.attr,
1757 &pages_shared_attr.attr,
1758 &pages_sharing_attr.attr,
1759 &pages_unshared_attr.attr,
1760 &pages_volatile_attr.attr,
1761 &full_scans_attr.attr,
1765 static struct attribute_group ksm_attr_group = {
1769 #endif /* CONFIG_SYSFS */
1771 static int __init ksm_init(void)
1773 struct task_struct *ksm_thread;
1776 ksm_max_kernel_pages = totalram_pages / 4;
1778 err = ksm_slab_init();
1782 err = mm_slots_hash_init();
1786 ksm_thread = kthread_run(ksm_scan_thread, NULL, "ksmd");
1787 if (IS_ERR(ksm_thread)) {
1788 printk(KERN_ERR "ksm: creating kthread failed\n");
1789 err = PTR_ERR(ksm_thread);
1794 err = sysfs_create_group(mm_kobj, &ksm_attr_group);
1796 printk(KERN_ERR "ksm: register sysfs failed\n");
1797 kthread_stop(ksm_thread);
1801 ksm_run = KSM_RUN_MERGE; /* no way for user to start it */
1803 #endif /* CONFIG_SYSFS */
1808 mm_slots_hash_free();
1814 module_init(ksm_init)