mm, oom: change type of oom_score_adj to short
[linux-3.10.git] / mm / ksm.c
index 8b76008..b4d5a9d 100644 (file)
--- a/mm/ksm.c
+++ b/mm/ksm.c
 /*
- * Initial dummy version just to illustrate KSM's interface to other files.
+ * Memory merging support.
+ *
+ * This code enables dynamic sharing of identical pages found in different
+ * memory areas, even if they are not shared by fork()
+ *
+ * Copyright (C) 2008-2009 Red Hat, Inc.
+ * Authors:
+ *     Izik Eidus
+ *     Andrea Arcangeli
+ *     Chris Wright
+ *     Hugh Dickins
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2.
  */
 
 #include <linux/errno.h>
+#include <linux/mm.h>
+#include <linux/fs.h>
 #include <linux/mman.h>
+#include <linux/sched.h>
+#include <linux/rwsem.h>
+#include <linux/pagemap.h>
+#include <linux/rmap.h>
+#include <linux/spinlock.h>
+#include <linux/jhash.h>
+#include <linux/delay.h>
+#include <linux/kthread.h>
+#include <linux/wait.h>
+#include <linux/slab.h>
+#include <linux/rbtree.h>
+#include <linux/memory.h>
+#include <linux/mmu_notifier.h>
+#include <linux/swap.h>
 #include <linux/ksm.h>
+#include <linux/hash.h>
+#include <linux/freezer.h>
+#include <linux/oom.h>
+
+#include <asm/tlbflush.h>
+#include "internal.h"
+
+/*
+ * A few notes about the KSM scanning process,
+ * to make it easier to understand the data structures below:
+ *
+ * In order to reduce excessive scanning, KSM sorts the memory pages by their
+ * contents into a data structure that holds pointers to the pages' locations.
+ *
+ * Since the contents of the pages may change at any moment, KSM cannot just
+ * insert the pages into a normal sorted tree and expect it to find anything.
+ * Therefore KSM uses two data structures - the stable and the unstable tree.
+ *
+ * The stable tree holds pointers to all the merged pages (ksm pages), sorted
+ * by their contents.  Because each such page is write-protected, searching on
+ * this tree is fully assured to be working (except when pages are unmapped),
+ * and therefore this tree is called the stable tree.
+ *
+ * In addition to the stable tree, KSM uses a second data structure called the
+ * unstable tree: this tree holds pointers to pages which have been found to
+ * be "unchanged for a period of time".  The unstable tree sorts these pages
+ * by their contents, but since they are not write-protected, KSM cannot rely
+ * upon the unstable tree to work correctly - the unstable tree is liable to
+ * be corrupted as its contents are modified, and so it is called unstable.
+ *
+ * KSM solves this problem by several techniques:
+ *
+ * 1) The unstable tree is flushed every time KSM completes scanning all
+ *    memory areas, and then the tree is rebuilt again from the beginning.
+ * 2) KSM will only insert into the unstable tree, pages whose hash value
+ *    has not changed since the previous scan of all memory areas.
+ * 3) The unstable tree is a RedBlack Tree - so its balancing is based on the
+ *    colors of the nodes and not on their contents, assuring that even when
+ *    the tree gets "corrupted" it won't get out of balance, so scanning time
+ *    remains the same (also, searching and inserting nodes in an rbtree uses
+ *    the same algorithm, so we have no overhead when we flush and rebuild).
+ * 4) KSM never flushes the stable tree, which means that even if it were to
+ *    take 10 attempts to find a page in the unstable tree, once it is found,
+ *    it is secured in the stable tree.  (When we scan a new page, we first
+ *    compare it against the stable tree, and then against the unstable tree.)
+ */
+
+/**
+ * struct mm_slot - ksm information per mm that is being scanned
+ * @link: link to the mm_slots hash list
+ * @mm_list: link into the mm_slots list, rooted in ksm_mm_head
+ * @rmap_list: head for this mm_slot's singly-linked list of rmap_items
+ * @mm: the mm that this information is valid for
+ */
+struct mm_slot {
+       struct hlist_node link;
+       struct list_head mm_list;
+       struct rmap_item *rmap_list;
+       struct mm_struct *mm;
+};
+
+/**
+ * struct ksm_scan - cursor for scanning
+ * @mm_slot: the current mm_slot we are scanning
+ * @address: the next address inside that to be scanned
+ * @rmap_list: link to the next rmap to be scanned in the rmap_list
+ * @seqnr: count of completed full scans (needed when removing unstable node)
+ *
+ * There is only the one ksm_scan instance of this cursor structure.
+ */
+struct ksm_scan {
+       struct mm_slot *mm_slot;
+       unsigned long address;
+       struct rmap_item **rmap_list;
+       unsigned long seqnr;
+};
+
+/**
+ * struct stable_node - node of the stable rbtree
+ * @node: rb node of this ksm page in the stable tree
+ * @hlist: hlist head of rmap_items using this ksm page
+ * @kpfn: page frame number of this ksm page
+ */
+struct stable_node {
+       struct rb_node node;
+       struct hlist_head hlist;
+       unsigned long kpfn;
+};
+
+/**
+ * struct rmap_item - reverse mapping item for virtual addresses
+ * @rmap_list: next rmap_item in mm_slot's singly-linked rmap_list
+ * @anon_vma: pointer to anon_vma for this mm,address, when in stable tree
+ * @mm: the memory structure this rmap_item is pointing into
+ * @address: the virtual address this rmap_item tracks (+ flags in low bits)
+ * @oldchecksum: previous checksum of the page at that virtual address
+ * @node: rb node of this rmap_item in the unstable tree
+ * @head: pointer to stable_node heading this list in the stable tree
+ * @hlist: link into hlist of rmap_items hanging off that stable_node
+ */
+struct rmap_item {
+       struct rmap_item *rmap_list;
+       struct anon_vma *anon_vma;      /* when stable */
+       struct mm_struct *mm;
+       unsigned long address;          /* + low bits used for flags below */
+       unsigned int oldchecksum;       /* when unstable */
+       union {
+               struct rb_node node;    /* when node of unstable tree */
+               struct {                /* when listed from stable tree */
+                       struct stable_node *head;
+                       struct hlist_node hlist;
+               };
+       };
+};
+
+#define SEQNR_MASK     0x0ff   /* low bits of unstable tree seqnr */
+#define UNSTABLE_FLAG  0x100   /* is a node of the unstable tree */
+#define STABLE_FLAG    0x200   /* is listed from the stable tree */
+
+/* The stable and unstable tree heads */
+static struct rb_root root_stable_tree = RB_ROOT;
+static struct rb_root root_unstable_tree = RB_ROOT;
+
+#define MM_SLOTS_HASH_SHIFT 10
+#define MM_SLOTS_HASH_HEADS (1 << MM_SLOTS_HASH_SHIFT)
+static struct hlist_head mm_slots_hash[MM_SLOTS_HASH_HEADS];
+
+static struct mm_slot ksm_mm_head = {
+       .mm_list = LIST_HEAD_INIT(ksm_mm_head.mm_list),
+};
+static struct ksm_scan ksm_scan = {
+       .mm_slot = &ksm_mm_head,
+};
+
+static struct kmem_cache *rmap_item_cache;
+static struct kmem_cache *stable_node_cache;
+static struct kmem_cache *mm_slot_cache;
+
+/* The number of nodes in the stable tree */
+static unsigned long ksm_pages_shared;
+
+/* The number of page slots additionally sharing those nodes */
+static unsigned long ksm_pages_sharing;
+
+/* The number of nodes in the unstable tree */
+static unsigned long ksm_pages_unshared;
+
+/* The number of rmap_items in use: to calculate pages_volatile */
+static unsigned long ksm_rmap_items;
+
+/* Number of pages ksmd should scan in one batch */
+static unsigned int ksm_thread_pages_to_scan = 100;
+
+/* Milliseconds ksmd should sleep between batches */
+static unsigned int ksm_thread_sleep_millisecs = 20;
+
+#define KSM_RUN_STOP   0
+#define KSM_RUN_MERGE  1
+#define KSM_RUN_UNMERGE        2
+static unsigned int ksm_run = KSM_RUN_STOP;
+
+static DECLARE_WAIT_QUEUE_HEAD(ksm_thread_wait);
+static DEFINE_MUTEX(ksm_thread_mutex);
+static DEFINE_SPINLOCK(ksm_mmlist_lock);
+
+#define KSM_KMEM_CACHE(__struct, __flags) kmem_cache_create("ksm_"#__struct,\
+               sizeof(struct __struct), __alignof__(struct __struct),\
+               (__flags), NULL)
+
+static int __init ksm_slab_init(void)
+{
+       rmap_item_cache = KSM_KMEM_CACHE(rmap_item, 0);
+       if (!rmap_item_cache)
+               goto out;
+
+       stable_node_cache = KSM_KMEM_CACHE(stable_node, 0);
+       if (!stable_node_cache)
+               goto out_free1;
+
+       mm_slot_cache = KSM_KMEM_CACHE(mm_slot, 0);
+       if (!mm_slot_cache)
+               goto out_free2;
+
+       return 0;
+
+out_free2:
+       kmem_cache_destroy(stable_node_cache);
+out_free1:
+       kmem_cache_destroy(rmap_item_cache);
+out:
+       return -ENOMEM;
+}
+
+static void __init ksm_slab_free(void)
+{
+       kmem_cache_destroy(mm_slot_cache);
+       kmem_cache_destroy(stable_node_cache);
+       kmem_cache_destroy(rmap_item_cache);
+       mm_slot_cache = NULL;
+}
+
+static inline struct rmap_item *alloc_rmap_item(void)
+{
+       struct rmap_item *rmap_item;
+
+       rmap_item = kmem_cache_zalloc(rmap_item_cache, GFP_KERNEL);
+       if (rmap_item)
+               ksm_rmap_items++;
+       return rmap_item;
+}
+
+static inline void free_rmap_item(struct rmap_item *rmap_item)
+{
+       ksm_rmap_items--;
+       rmap_item->mm = NULL;   /* debug safety */
+       kmem_cache_free(rmap_item_cache, rmap_item);
+}
+
+static inline struct stable_node *alloc_stable_node(void)
+{
+       return kmem_cache_alloc(stable_node_cache, GFP_KERNEL);
+}
+
+static inline void free_stable_node(struct stable_node *stable_node)
+{
+       kmem_cache_free(stable_node_cache, stable_node);
+}
+
+static inline struct mm_slot *alloc_mm_slot(void)
+{
+       if (!mm_slot_cache)     /* initialization failed */
+               return NULL;
+       return kmem_cache_zalloc(mm_slot_cache, GFP_KERNEL);
+}
+
+static inline void free_mm_slot(struct mm_slot *mm_slot)
+{
+       kmem_cache_free(mm_slot_cache, mm_slot);
+}
+
+static struct mm_slot *get_mm_slot(struct mm_struct *mm)
+{
+       struct mm_slot *mm_slot;
+       struct hlist_head *bucket;
+       struct hlist_node *node;
+
+       bucket = &mm_slots_hash[hash_ptr(mm, MM_SLOTS_HASH_SHIFT)];
+       hlist_for_each_entry(mm_slot, node, bucket, link) {
+               if (mm == mm_slot->mm)
+                       return mm_slot;
+       }
+       return NULL;
+}
+
+static void insert_to_mm_slots_hash(struct mm_struct *mm,
+                                   struct mm_slot *mm_slot)
+{
+       struct hlist_head *bucket;
+
+       bucket = &mm_slots_hash[hash_ptr(mm, MM_SLOTS_HASH_SHIFT)];
+       mm_slot->mm = mm;
+       hlist_add_head(&mm_slot->link, bucket);
+}
+
+static inline int in_stable_tree(struct rmap_item *rmap_item)
+{
+       return rmap_item->address & STABLE_FLAG;
+}
+
+/*
+ * ksmd, and unmerge_and_remove_all_rmap_items(), must not touch an mm's
+ * page tables after it has passed through ksm_exit() - which, if necessary,
+ * takes mmap_sem briefly to serialize against them.  ksm_exit() does not set
+ * a special flag: they can just back out as soon as mm_users goes to zero.
+ * ksm_test_exit() is used throughout to make this test for exit: in some
+ * places for correctness, in some places just to avoid unnecessary work.
+ */
+static inline bool ksm_test_exit(struct mm_struct *mm)
+{
+       return atomic_read(&mm->mm_users) == 0;
+}
+
+/*
+ * We use break_ksm to break COW on a ksm page: it's a stripped down
+ *
+ *     if (get_user_pages(current, mm, addr, 1, 1, 1, &page, NULL) == 1)
+ *             put_page(page);
+ *
+ * but taking great care only to touch a ksm page, in a VM_MERGEABLE vma,
+ * in case the application has unmapped and remapped mm,addr meanwhile.
+ * Could a ksm page appear anywhere else?  Actually yes, in a VM_PFNMAP
+ * mmap of /dev/mem or /dev/kmem, where we would not want to touch it.
+ */
+static int break_ksm(struct vm_area_struct *vma, unsigned long addr)
+{
+       struct page *page;
+       int ret = 0;
+
+       do {
+               cond_resched();
+               page = follow_page(vma, addr, FOLL_GET);
+               if (IS_ERR_OR_NULL(page))
+                       break;
+               if (PageKsm(page))
+                       ret = handle_mm_fault(vma->vm_mm, vma, addr,
+                                                       FAULT_FLAG_WRITE);
+               else
+                       ret = VM_FAULT_WRITE;
+               put_page(page);
+       } while (!(ret & (VM_FAULT_WRITE | VM_FAULT_SIGBUS | VM_FAULT_OOM)));
+       /*
+        * We must loop because handle_mm_fault() may back out if there's
+        * any difficulty e.g. if pte accessed bit gets updated concurrently.
+        *
+        * VM_FAULT_WRITE is what we have been hoping for: it indicates that
+        * COW has been broken, even if the vma does not permit VM_WRITE;
+        * but note that a concurrent fault might break PageKsm for us.
+        *
+        * VM_FAULT_SIGBUS could occur if we race with truncation of the
+        * backing file, which also invalidates anonymous pages: that's
+        * okay, that truncation will have unmapped the PageKsm for us.
+        *
+        * VM_FAULT_OOM: at the time of writing (late July 2009), setting
+        * aside mem_cgroup limits, VM_FAULT_OOM would only be set if the
+        * current task has TIF_MEMDIE set, and will be OOM killed on return
+        * to user; and ksmd, having no mm, would never be chosen for that.
+        *
+        * But if the mm is in a limited mem_cgroup, then the fault may fail
+        * with VM_FAULT_OOM even if the current task is not TIF_MEMDIE; and
+        * even ksmd can fail in this way - though it's usually breaking ksm
+        * just to undo a merge it made a moment before, so unlikely to oom.
+        *
+        * That's a pity: we might therefore have more kernel pages allocated
+        * than we're counting as nodes in the stable tree; but ksm_do_scan
+        * will retry to break_cow on each pass, so should recover the page
+        * in due course.  The important thing is to not let VM_MERGEABLE
+        * be cleared while any such pages might remain in the area.
+        */
+       return (ret & VM_FAULT_OOM) ? -ENOMEM : 0;
+}
+
+static struct vm_area_struct *find_mergeable_vma(struct mm_struct *mm,
+               unsigned long addr)
+{
+       struct vm_area_struct *vma;
+       if (ksm_test_exit(mm))
+               return NULL;
+       vma = find_vma(mm, addr);
+       if (!vma || vma->vm_start > addr)
+               return NULL;
+       if (!(vma->vm_flags & VM_MERGEABLE) || !vma->anon_vma)
+               return NULL;
+       return vma;
+}
+
+static void break_cow(struct rmap_item *rmap_item)
+{
+       struct mm_struct *mm = rmap_item->mm;
+       unsigned long addr = rmap_item->address;
+       struct vm_area_struct *vma;
+
+       /*
+        * It is not an accident that whenever we want to break COW
+        * to undo, we also need to drop a reference to the anon_vma.
+        */
+       put_anon_vma(rmap_item->anon_vma);
+
+       down_read(&mm->mmap_sem);
+       vma = find_mergeable_vma(mm, addr);
+       if (vma)
+               break_ksm(vma, addr);
+       up_read(&mm->mmap_sem);
+}
+
+static struct page *page_trans_compound_anon(struct page *page)
+{
+       if (PageTransCompound(page)) {
+               struct page *head = compound_trans_head(page);
+               /*
+                * head may actually be splitted and freed from under
+                * us but it's ok here.
+                */
+               if (PageAnon(head))
+                       return head;
+       }
+       return NULL;
+}
+
+static struct page *get_mergeable_page(struct rmap_item *rmap_item)
+{
+       struct mm_struct *mm = rmap_item->mm;
+       unsigned long addr = rmap_item->address;
+       struct vm_area_struct *vma;
+       struct page *page;
+
+       down_read(&mm->mmap_sem);
+       vma = find_mergeable_vma(mm, addr);
+       if (!vma)
+               goto out;
+
+       page = follow_page(vma, addr, FOLL_GET);
+       if (IS_ERR_OR_NULL(page))
+               goto out;
+       if (PageAnon(page) || page_trans_compound_anon(page)) {
+               flush_anon_page(vma, page, addr);
+               flush_dcache_page(page);
+       } else {
+               put_page(page);
+out:           page = NULL;
+       }
+       up_read(&mm->mmap_sem);
+       return page;
+}
+
+static void remove_node_from_stable_tree(struct stable_node *stable_node)
+{
+       struct rmap_item *rmap_item;
+       struct hlist_node *hlist;
+
+       hlist_for_each_entry(rmap_item, hlist, &stable_node->hlist, hlist) {
+               if (rmap_item->hlist.next)
+                       ksm_pages_sharing--;
+               else
+                       ksm_pages_shared--;
+               put_anon_vma(rmap_item->anon_vma);
+               rmap_item->address &= PAGE_MASK;
+               cond_resched();
+       }
+
+       rb_erase(&stable_node->node, &root_stable_tree);
+       free_stable_node(stable_node);
+}
+
+/*
+ * get_ksm_page: checks if the page indicated by the stable node
+ * is still its ksm page, despite having held no reference to it.
+ * In which case we can trust the content of the page, and it
+ * returns the gotten page; but if the page has now been zapped,
+ * remove the stale node from the stable tree and return NULL.
+ *
+ * You would expect the stable_node to hold a reference to the ksm page.
+ * But if it increments the page's count, swapping out has to wait for
+ * ksmd to come around again before it can free the page, which may take
+ * seconds or even minutes: much too unresponsive.  So instead we use a
+ * "keyhole reference": access to the ksm page from the stable node peeps
+ * out through its keyhole to see if that page still holds the right key,
+ * pointing back to this stable node.  This relies on freeing a PageAnon
+ * page to reset its page->mapping to NULL, and relies on no other use of
+ * a page to put something that might look like our key in page->mapping.
+ *
+ * include/linux/pagemap.h page_cache_get_speculative() is a good reference,
+ * but this is different - made simpler by ksm_thread_mutex being held, but
+ * interesting for assuming that no other use of the struct page could ever
+ * put our expected_mapping into page->mapping (or a field of the union which
+ * coincides with page->mapping).  The RCU calls are not for KSM at all, but
+ * to keep the page_count protocol described with page_cache_get_speculative.
+ *
+ * Note: it is possible that get_ksm_page() will return NULL one moment,
+ * then page the next, if the page is in between page_freeze_refs() and
+ * page_unfreeze_refs(): this shouldn't be a problem anywhere, the page
+ * is on its way to being freed; but it is an anomaly to bear in mind.
+ */
+static struct page *get_ksm_page(struct stable_node *stable_node)
+{
+       struct page *page;
+       void *expected_mapping;
+
+       page = pfn_to_page(stable_node->kpfn);
+       expected_mapping = (void *)stable_node +
+                               (PAGE_MAPPING_ANON | PAGE_MAPPING_KSM);
+       rcu_read_lock();
+       if (page->mapping != expected_mapping)
+               goto stale;
+       if (!get_page_unless_zero(page))
+               goto stale;
+       if (page->mapping != expected_mapping) {
+               put_page(page);
+               goto stale;
+       }
+       rcu_read_unlock();
+       return page;
+stale:
+       rcu_read_unlock();
+       remove_node_from_stable_tree(stable_node);
+       return NULL;
+}
+
+/*
+ * Removing rmap_item from stable or unstable tree.
+ * This function will clean the information from the stable/unstable tree.
+ */
+static void remove_rmap_item_from_tree(struct rmap_item *rmap_item)
+{
+       if (rmap_item->address & STABLE_FLAG) {
+               struct stable_node *stable_node;
+               struct page *page;
+
+               stable_node = rmap_item->head;
+               page = get_ksm_page(stable_node);
+               if (!page)
+                       goto out;
+
+               lock_page(page);
+               hlist_del(&rmap_item->hlist);
+               unlock_page(page);
+               put_page(page);
+
+               if (stable_node->hlist.first)
+                       ksm_pages_sharing--;
+               else
+                       ksm_pages_shared--;
+
+               put_anon_vma(rmap_item->anon_vma);
+               rmap_item->address &= PAGE_MASK;
+
+       } else if (rmap_item->address & UNSTABLE_FLAG) {
+               unsigned char age;
+               /*
+                * Usually ksmd can and must skip the rb_erase, because
+                * root_unstable_tree was already reset to RB_ROOT.
+                * But be careful when an mm is exiting: do the rb_erase
+                * if this rmap_item was inserted by this scan, rather
+                * than left over from before.
+                */
+               age = (unsigned char)(ksm_scan.seqnr - rmap_item->address);
+               BUG_ON(age > 1);
+               if (!age)
+                       rb_erase(&rmap_item->node, &root_unstable_tree);
+
+               ksm_pages_unshared--;
+               rmap_item->address &= PAGE_MASK;
+       }
+out:
+       cond_resched();         /* we're called from many long loops */
+}
+
+static void remove_trailing_rmap_items(struct mm_slot *mm_slot,
+                                      struct rmap_item **rmap_list)
+{
+       while (*rmap_list) {
+               struct rmap_item *rmap_item = *rmap_list;
+               *rmap_list = rmap_item->rmap_list;
+               remove_rmap_item_from_tree(rmap_item);
+               free_rmap_item(rmap_item);
+       }
+}
+
+/*
+ * Though it's very tempting to unmerge in_stable_tree(rmap_item)s rather
+ * than check every pte of a given vma, the locking doesn't quite work for
+ * that - an rmap_item is assigned to the stable tree after inserting ksm
+ * page and upping mmap_sem.  Nor does it fit with the way we skip dup'ing
+ * rmap_items from parent to child at fork time (so as not to waste time
+ * if exit comes before the next scan reaches it).
+ *
+ * Similarly, although we'd like to remove rmap_items (so updating counts
+ * and freeing memory) when unmerging an area, it's easier to leave that
+ * to the next pass of ksmd - consider, for example, how ksmd might be
+ * in cmp_and_merge_page on one of the rmap_items we would be removing.
+ */
+static int unmerge_ksm_pages(struct vm_area_struct *vma,
+                            unsigned long start, unsigned long end)
+{
+       unsigned long addr;
+       int err = 0;
+
+       for (addr = start; addr < end && !err; addr += PAGE_SIZE) {
+               if (ksm_test_exit(vma->vm_mm))
+                       break;
+               if (signal_pending(current))
+                       err = -ERESTARTSYS;
+               else
+                       err = break_ksm(vma, addr);
+       }
+       return err;
+}
+
+#ifdef CONFIG_SYSFS
+/*
+ * Only called through the sysfs control interface:
+ */
+static int unmerge_and_remove_all_rmap_items(void)
+{
+       struct mm_slot *mm_slot;
+       struct mm_struct *mm;
+       struct vm_area_struct *vma;
+       int err = 0;
+
+       spin_lock(&ksm_mmlist_lock);
+       ksm_scan.mm_slot = list_entry(ksm_mm_head.mm_list.next,
+                                               struct mm_slot, mm_list);
+       spin_unlock(&ksm_mmlist_lock);
+
+       for (mm_slot = ksm_scan.mm_slot;
+                       mm_slot != &ksm_mm_head; mm_slot = ksm_scan.mm_slot) {
+               mm = mm_slot->mm;
+               down_read(&mm->mmap_sem);
+               for (vma = mm->mmap; vma; vma = vma->vm_next) {
+                       if (ksm_test_exit(mm))
+                               break;
+                       if (!(vma->vm_flags & VM_MERGEABLE) || !vma->anon_vma)
+                               continue;
+                       err = unmerge_ksm_pages(vma,
+                                               vma->vm_start, vma->vm_end);
+                       if (err)
+                               goto error;
+               }
+
+               remove_trailing_rmap_items(mm_slot, &mm_slot->rmap_list);
+
+               spin_lock(&ksm_mmlist_lock);
+               ksm_scan.mm_slot = list_entry(mm_slot->mm_list.next,
+                                               struct mm_slot, mm_list);
+               if (ksm_test_exit(mm)) {
+                       hlist_del(&mm_slot->link);
+                       list_del(&mm_slot->mm_list);
+                       spin_unlock(&ksm_mmlist_lock);
+
+                       free_mm_slot(mm_slot);
+                       clear_bit(MMF_VM_MERGEABLE, &mm->flags);
+                       up_read(&mm->mmap_sem);
+                       mmdrop(mm);
+               } else {
+                       spin_unlock(&ksm_mmlist_lock);
+                       up_read(&mm->mmap_sem);
+               }
+       }
+
+       ksm_scan.seqnr = 0;
+       return 0;
+
+error:
+       up_read(&mm->mmap_sem);
+       spin_lock(&ksm_mmlist_lock);
+       ksm_scan.mm_slot = &ksm_mm_head;
+       spin_unlock(&ksm_mmlist_lock);
+       return err;
+}
+#endif /* CONFIG_SYSFS */
+
+static u32 calc_checksum(struct page *page)
+{
+       u32 checksum;
+       void *addr = kmap_atomic(page);
+       checksum = jhash2(addr, PAGE_SIZE / 4, 17);
+       kunmap_atomic(addr);
+       return checksum;
+}
+
+static int memcmp_pages(struct page *page1, struct page *page2)
+{
+       char *addr1, *addr2;
+       int ret;
+
+       addr1 = kmap_atomic(page1);
+       addr2 = kmap_atomic(page2);
+       ret = memcmp(addr1, addr2, PAGE_SIZE);
+       kunmap_atomic(addr2);
+       kunmap_atomic(addr1);
+       return ret;
+}
+
+static inline int pages_identical(struct page *page1, struct page *page2)
+{
+       return !memcmp_pages(page1, page2);
+}
+
+static int write_protect_page(struct vm_area_struct *vma, struct page *page,
+                             pte_t *orig_pte)
+{
+       struct mm_struct *mm = vma->vm_mm;
+       unsigned long addr;
+       pte_t *ptep;
+       spinlock_t *ptl;
+       int swapped;
+       int err = -EFAULT;
+       unsigned long mmun_start;       /* For mmu_notifiers */
+       unsigned long mmun_end;         /* For mmu_notifiers */
+
+       addr = page_address_in_vma(page, vma);
+       if (addr == -EFAULT)
+               goto out;
+
+       BUG_ON(PageTransCompound(page));
+
+       mmun_start = addr;
+       mmun_end   = addr + PAGE_SIZE;
+       mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
+
+       ptep = page_check_address(page, mm, addr, &ptl, 0);
+       if (!ptep)
+               goto out_mn;
+
+       if (pte_write(*ptep) || pte_dirty(*ptep)) {
+               pte_t entry;
+
+               swapped = PageSwapCache(page);
+               flush_cache_page(vma, addr, page_to_pfn(page));
+               /*
+                * Ok this is tricky, when get_user_pages_fast() run it doesn't
+                * take any lock, therefore the check that we are going to make
+                * with the pagecount against the mapcount is racey and
+                * O_DIRECT can happen right after the check.
+                * So we clear the pte and flush the tlb before the check
+                * this assure us that no O_DIRECT can happen after the check
+                * or in the middle of the check.
+                */
+               entry = ptep_clear_flush(vma, addr, ptep);
+               /*
+                * Check that no O_DIRECT or similar I/O is in progress on the
+                * page
+                */
+               if (page_mapcount(page) + 1 + swapped != page_count(page)) {
+                       set_pte_at(mm, addr, ptep, entry);
+                       goto out_unlock;
+               }
+               if (pte_dirty(entry))
+                       set_page_dirty(page);
+               entry = pte_mkclean(pte_wrprotect(entry));
+               set_pte_at_notify(mm, addr, ptep, entry);
+       }
+       *orig_pte = *ptep;
+       err = 0;
+
+out_unlock:
+       pte_unmap_unlock(ptep, ptl);
+out_mn:
+       mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
+out:
+       return err;
+}
+
+/**
+ * replace_page - replace page in vma by new ksm page
+ * @vma:      vma that holds the pte pointing to page
+ * @page:     the page we are replacing by kpage
+ * @kpage:    the ksm page we replace page by
+ * @orig_pte: the original value of the pte
+ *
+ * Returns 0 on success, -EFAULT on failure.
+ */
+static int replace_page(struct vm_area_struct *vma, struct page *page,
+                       struct page *kpage, pte_t orig_pte)
+{
+       struct mm_struct *mm = vma->vm_mm;
+       pmd_t *pmd;
+       pte_t *ptep;
+       spinlock_t *ptl;
+       unsigned long addr;
+       int err = -EFAULT;
+       unsigned long mmun_start;       /* For mmu_notifiers */
+       unsigned long mmun_end;         /* For mmu_notifiers */
+
+       addr = page_address_in_vma(page, vma);
+       if (addr == -EFAULT)
+               goto out;
+
+       pmd = mm_find_pmd(mm, addr);
+       if (!pmd)
+               goto out;
+       BUG_ON(pmd_trans_huge(*pmd));
+
+       mmun_start = addr;
+       mmun_end   = addr + PAGE_SIZE;
+       mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
+
+       ptep = pte_offset_map_lock(mm, pmd, addr, &ptl);
+       if (!pte_same(*ptep, orig_pte)) {
+               pte_unmap_unlock(ptep, ptl);
+               goto out_mn;
+       }
+
+       get_page(kpage);
+       page_add_anon_rmap(kpage, vma, addr);
+
+       flush_cache_page(vma, addr, pte_pfn(*ptep));
+       ptep_clear_flush(vma, addr, ptep);
+       set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot));
+
+       page_remove_rmap(page);
+       if (!page_mapped(page))
+               try_to_free_swap(page);
+       put_page(page);
+
+       pte_unmap_unlock(ptep, ptl);
+       err = 0;
+out_mn:
+       mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
+out:
+       return err;
+}
+
+static int page_trans_compound_anon_split(struct page *page)
+{
+       int ret = 0;
+       struct page *transhuge_head = page_trans_compound_anon(page);
+       if (transhuge_head) {
+               /* Get the reference on the head to split it. */
+               if (get_page_unless_zero(transhuge_head)) {
+                       /*
+                        * Recheck we got the reference while the head
+                        * was still anonymous.
+                        */
+                       if (PageAnon(transhuge_head))
+                               ret = split_huge_page(transhuge_head);
+                       else
+                               /*
+                                * Retry later if split_huge_page run
+                                * from under us.
+                                */
+                               ret = 1;
+                       put_page(transhuge_head);
+               } else
+                       /* Retry later if split_huge_page run from under us. */
+                       ret = 1;
+       }
+       return ret;
+}
+
+/*
+ * try_to_merge_one_page - take two pages and merge them into one
+ * @vma: the vma that holds the pte pointing to page
+ * @page: the PageAnon page that we want to replace with kpage
+ * @kpage: the PageKsm page that we want to map instead of page,
+ *         or NULL the first time when we want to use page as kpage.
+ *
+ * This function returns 0 if the pages were merged, -EFAULT otherwise.
+ */
+static int try_to_merge_one_page(struct vm_area_struct *vma,
+                                struct page *page, struct page *kpage)
+{
+       pte_t orig_pte = __pte(0);
+       int err = -EFAULT;
+
+       if (page == kpage)                      /* ksm page forked */
+               return 0;
+
+       if (!(vma->vm_flags & VM_MERGEABLE))
+               goto out;
+       if (PageTransCompound(page) && page_trans_compound_anon_split(page))
+               goto out;
+       BUG_ON(PageTransCompound(page));
+       if (!PageAnon(page))
+               goto out;
+
+       /*
+        * We need the page lock to read a stable PageSwapCache in
+        * write_protect_page().  We use trylock_page() instead of
+        * lock_page() because we don't want to wait here - we
+        * prefer to continue scanning and merging different pages,
+        * then come back to this page when it is unlocked.
+        */
+       if (!trylock_page(page))
+               goto out;
+       /*
+        * If this anonymous page is mapped only here, its pte may need
+        * to be write-protected.  If it's mapped elsewhere, all of its
+        * ptes are necessarily already write-protected.  But in either
+        * case, we need to lock and check page_count is not raised.
+        */
+       if (write_protect_page(vma, page, &orig_pte) == 0) {
+               if (!kpage) {
+                       /*
+                        * While we hold page lock, upgrade page from
+                        * PageAnon+anon_vma to PageKsm+NULL stable_node:
+                        * stable_tree_insert() will update stable_node.
+                        */
+                       set_page_stable_node(page, NULL);
+                       mark_page_accessed(page);
+                       err = 0;
+               } else if (pages_identical(page, kpage))
+                       err = replace_page(vma, page, kpage, orig_pte);
+       }
+
+       if ((vma->vm_flags & VM_LOCKED) && kpage && !err) {
+               munlock_vma_page(page);
+               if (!PageMlocked(kpage)) {
+                       unlock_page(page);
+                       lock_page(kpage);
+                       mlock_vma_page(kpage);
+                       page = kpage;           /* for final unlock */
+               }
+       }
+
+       unlock_page(page);
+out:
+       return err;
+}
+
+/*
+ * try_to_merge_with_ksm_page - like try_to_merge_two_pages,
+ * but no new kernel page is allocated: kpage must already be a ksm page.
+ *
+ * This function returns 0 if the pages were merged, -EFAULT otherwise.
+ */
+static int try_to_merge_with_ksm_page(struct rmap_item *rmap_item,
+                                     struct page *page, struct page *kpage)
+{
+       struct mm_struct *mm = rmap_item->mm;
+       struct vm_area_struct *vma;
+       int err = -EFAULT;
+
+       down_read(&mm->mmap_sem);
+       if (ksm_test_exit(mm))
+               goto out;
+       vma = find_vma(mm, rmap_item->address);
+       if (!vma || vma->vm_start > rmap_item->address)
+               goto out;
+
+       err = try_to_merge_one_page(vma, page, kpage);
+       if (err)
+               goto out;
+
+       /* Must get reference to anon_vma while still holding mmap_sem */
+       rmap_item->anon_vma = vma->anon_vma;
+       get_anon_vma(vma->anon_vma);
+out:
+       up_read(&mm->mmap_sem);
+       return err;
+}
+
+/*
+ * try_to_merge_two_pages - take two identical pages and prepare them
+ * to be merged into one page.
+ *
+ * This function returns the kpage if we successfully merged two identical
+ * pages into one ksm page, NULL otherwise.
+ *
+ * Note that this function upgrades page to ksm page: if one of the pages
+ * is already a ksm page, try_to_merge_with_ksm_page should be used.
+ */
+static struct page *try_to_merge_two_pages(struct rmap_item *rmap_item,
+                                          struct page *page,
+                                          struct rmap_item *tree_rmap_item,
+                                          struct page *tree_page)
+{
+       int err;
+
+       err = try_to_merge_with_ksm_page(rmap_item, page, NULL);
+       if (!err) {
+               err = try_to_merge_with_ksm_page(tree_rmap_item,
+                                                       tree_page, page);
+               /*
+                * If that fails, we have a ksm page with only one pte
+                * pointing to it: so break it.
+                */
+               if (err)
+                       break_cow(rmap_item);
+       }
+       return err ? NULL : page;
+}
+
+/*
+ * stable_tree_search - search for page inside the stable tree
+ *
+ * This function checks if there is a page inside the stable tree
+ * with identical content to the page that we are scanning right now.
+ *
+ * This function returns the stable tree node of identical content if found,
+ * NULL otherwise.
+ */
+static struct page *stable_tree_search(struct page *page)
+{
+       struct rb_node *node = root_stable_tree.rb_node;
+       struct stable_node *stable_node;
+
+       stable_node = page_stable_node(page);
+       if (stable_node) {                      /* ksm page forked */
+               get_page(page);
+               return page;
+       }
+
+       while (node) {
+               struct page *tree_page;
+               int ret;
+
+               cond_resched();
+               stable_node = rb_entry(node, struct stable_node, node);
+               tree_page = get_ksm_page(stable_node);
+               if (!tree_page)
+                       return NULL;
+
+               ret = memcmp_pages(page, tree_page);
+
+               if (ret < 0) {
+                       put_page(tree_page);
+                       node = node->rb_left;
+               } else if (ret > 0) {
+                       put_page(tree_page);
+                       node = node->rb_right;
+               } else
+                       return tree_page;
+       }
+
+       return NULL;
+}
+
+/*
+ * stable_tree_insert - insert rmap_item pointing to new ksm page
+ * into the stable tree.
+ *
+ * This function returns the stable tree node just allocated on success,
+ * NULL otherwise.
+ */
+static struct stable_node *stable_tree_insert(struct page *kpage)
+{
+       struct rb_node **new = &root_stable_tree.rb_node;
+       struct rb_node *parent = NULL;
+       struct stable_node *stable_node;
+
+       while (*new) {
+               struct page *tree_page;
+               int ret;
+
+               cond_resched();
+               stable_node = rb_entry(*new, struct stable_node, node);
+               tree_page = get_ksm_page(stable_node);
+               if (!tree_page)
+                       return NULL;
+
+               ret = memcmp_pages(kpage, tree_page);
+               put_page(tree_page);
+
+               parent = *new;
+               if (ret < 0)
+                       new = &parent->rb_left;
+               else if (ret > 0)
+                       new = &parent->rb_right;
+               else {
+                       /*
+                        * It is not a bug that stable_tree_search() didn't
+                        * find this node: because at that time our page was
+                        * not yet write-protected, so may have changed since.
+                        */
+                       return NULL;
+               }
+       }
+
+       stable_node = alloc_stable_node();
+       if (!stable_node)
+               return NULL;
+
+       rb_link_node(&stable_node->node, parent, new);
+       rb_insert_color(&stable_node->node, &root_stable_tree);
+
+       INIT_HLIST_HEAD(&stable_node->hlist);
+
+       stable_node->kpfn = page_to_pfn(kpage);
+       set_page_stable_node(kpage, stable_node);
+
+       return stable_node;
+}
+
+/*
+ * unstable_tree_search_insert - search for identical page,
+ * else insert rmap_item into the unstable tree.
+ *
+ * This function searches for a page in the unstable tree identical to the
+ * page currently being scanned; and if no identical page is found in the
+ * tree, we insert rmap_item as a new object into the unstable tree.
+ *
+ * This function returns pointer to rmap_item found to be identical
+ * to the currently scanned page, NULL otherwise.
+ *
+ * This function does both searching and inserting, because they share
+ * the same walking algorithm in an rbtree.
+ */
+static
+struct rmap_item *unstable_tree_search_insert(struct rmap_item *rmap_item,
+                                             struct page *page,
+                                             struct page **tree_pagep)
+
+{
+       struct rb_node **new = &root_unstable_tree.rb_node;
+       struct rb_node *parent = NULL;
+
+       while (*new) {
+               struct rmap_item *tree_rmap_item;
+               struct page *tree_page;
+               int ret;
+
+               cond_resched();
+               tree_rmap_item = rb_entry(*new, struct rmap_item, node);
+               tree_page = get_mergeable_page(tree_rmap_item);
+               if (IS_ERR_OR_NULL(tree_page))
+                       return NULL;
+
+               /*
+                * Don't substitute a ksm page for a forked page.
+                */
+               if (page == tree_page) {
+                       put_page(tree_page);
+                       return NULL;
+               }
+
+               ret = memcmp_pages(page, tree_page);
+
+               parent = *new;
+               if (ret < 0) {
+                       put_page(tree_page);
+                       new = &parent->rb_left;
+               } else if (ret > 0) {
+                       put_page(tree_page);
+                       new = &parent->rb_right;
+               } else {
+                       *tree_pagep = tree_page;
+                       return tree_rmap_item;
+               }
+       }
+
+       rmap_item->address |= UNSTABLE_FLAG;
+       rmap_item->address |= (ksm_scan.seqnr & SEQNR_MASK);
+       rb_link_node(&rmap_item->node, parent, new);
+       rb_insert_color(&rmap_item->node, &root_unstable_tree);
+
+       ksm_pages_unshared++;
+       return NULL;
+}
+
+/*
+ * stable_tree_append - add another rmap_item to the linked list of
+ * rmap_items hanging off a given node of the stable tree, all sharing
+ * the same ksm page.
+ */
+static void stable_tree_append(struct rmap_item *rmap_item,
+                              struct stable_node *stable_node)
+{
+       rmap_item->head = stable_node;
+       rmap_item->address |= STABLE_FLAG;
+       hlist_add_head(&rmap_item->hlist, &stable_node->hlist);
+
+       if (rmap_item->hlist.next)
+               ksm_pages_sharing++;
+       else
+               ksm_pages_shared++;
+}
+
+/*
+ * cmp_and_merge_page - first see if page can be merged into the stable tree;
+ * if not, compare checksum to previous and if it's the same, see if page can
+ * be inserted into the unstable tree, or merged with a page already there and
+ * both transferred to the stable tree.
+ *
+ * @page: the page that we are searching identical page to.
+ * @rmap_item: the reverse mapping into the virtual address of this page
+ */
+static void cmp_and_merge_page(struct page *page, struct rmap_item *rmap_item)
+{
+       struct rmap_item *tree_rmap_item;
+       struct page *tree_page = NULL;
+       struct stable_node *stable_node;
+       struct page *kpage;
+       unsigned int checksum;
+       int err;
+
+       remove_rmap_item_from_tree(rmap_item);
+
+       /* We first start with searching the page inside the stable tree */
+       kpage = stable_tree_search(page);
+       if (kpage) {
+               err = try_to_merge_with_ksm_page(rmap_item, page, kpage);
+               if (!err) {
+                       /*
+                        * The page was successfully merged:
+                        * add its rmap_item to the stable tree.
+                        */
+                       lock_page(kpage);
+                       stable_tree_append(rmap_item, page_stable_node(kpage));
+                       unlock_page(kpage);
+               }
+               put_page(kpage);
+               return;
+       }
+
+       /*
+        * If the hash value of the page has changed from the last time
+        * we calculated it, this page is changing frequently: therefore we
+        * don't want to insert it in the unstable tree, and we don't want
+        * to waste our time searching for something identical to it there.
+        */
+       checksum = calc_checksum(page);
+       if (rmap_item->oldchecksum != checksum) {
+               rmap_item->oldchecksum = checksum;
+               return;
+       }
+
+       tree_rmap_item =
+               unstable_tree_search_insert(rmap_item, page, &tree_page);
+       if (tree_rmap_item) {
+               kpage = try_to_merge_two_pages(rmap_item, page,
+                                               tree_rmap_item, tree_page);
+               put_page(tree_page);
+               /*
+                * As soon as we merge this page, we want to remove the
+                * rmap_item of the page we have merged with from the unstable
+                * tree, and insert it instead as new node in the stable tree.
+                */
+               if (kpage) {
+                       remove_rmap_item_from_tree(tree_rmap_item);
+
+                       lock_page(kpage);
+                       stable_node = stable_tree_insert(kpage);
+                       if (stable_node) {
+                               stable_tree_append(tree_rmap_item, stable_node);
+                               stable_tree_append(rmap_item, stable_node);
+                       }
+                       unlock_page(kpage);
+
+                       /*
+                        * If we fail to insert the page into the stable tree,
+                        * we will have 2 virtual addresses that are pointing
+                        * to a ksm page left outside the stable tree,
+                        * in which case we need to break_cow on both.
+                        */
+                       if (!stable_node) {
+                               break_cow(tree_rmap_item);
+                               break_cow(rmap_item);
+                       }
+               }
+       }
+}
+
+static struct rmap_item *get_next_rmap_item(struct mm_slot *mm_slot,
+                                           struct rmap_item **rmap_list,
+                                           unsigned long addr)
+{
+       struct rmap_item *rmap_item;
+
+       while (*rmap_list) {
+               rmap_item = *rmap_list;
+               if ((rmap_item->address & PAGE_MASK) == addr)
+                       return rmap_item;
+               if (rmap_item->address > addr)
+                       break;
+               *rmap_list = rmap_item->rmap_list;
+               remove_rmap_item_from_tree(rmap_item);
+               free_rmap_item(rmap_item);
+       }
+
+       rmap_item = alloc_rmap_item();
+       if (rmap_item) {
+               /* It has already been zeroed */
+               rmap_item->mm = mm_slot->mm;
+               rmap_item->address = addr;
+               rmap_item->rmap_list = *rmap_list;
+               *rmap_list = rmap_item;
+       }
+       return rmap_item;
+}
+
+static struct rmap_item *scan_get_next_rmap_item(struct page **page)
+{
+       struct mm_struct *mm;
+       struct mm_slot *slot;
+       struct vm_area_struct *vma;
+       struct rmap_item *rmap_item;
+
+       if (list_empty(&ksm_mm_head.mm_list))
+               return NULL;
+
+       slot = ksm_scan.mm_slot;
+       if (slot == &ksm_mm_head) {
+               /*
+                * A number of pages can hang around indefinitely on per-cpu
+                * pagevecs, raised page count preventing write_protect_page
+                * from merging them.  Though it doesn't really matter much,
+                * it is puzzling to see some stuck in pages_volatile until
+                * other activity jostles them out, and they also prevented
+                * LTP's KSM test from succeeding deterministically; so drain
+                * them here (here rather than on entry to ksm_do_scan(),
+                * so we don't IPI too often when pages_to_scan is set low).
+                */
+               lru_add_drain_all();
+
+               root_unstable_tree = RB_ROOT;
+
+               spin_lock(&ksm_mmlist_lock);
+               slot = list_entry(slot->mm_list.next, struct mm_slot, mm_list);
+               ksm_scan.mm_slot = slot;
+               spin_unlock(&ksm_mmlist_lock);
+               /*
+                * Although we tested list_empty() above, a racing __ksm_exit
+                * of the last mm on the list may have removed it since then.
+                */
+               if (slot == &ksm_mm_head)
+                       return NULL;
+next_mm:
+               ksm_scan.address = 0;
+               ksm_scan.rmap_list = &slot->rmap_list;
+       }
+
+       mm = slot->mm;
+       down_read(&mm->mmap_sem);
+       if (ksm_test_exit(mm))
+               vma = NULL;
+       else
+               vma = find_vma(mm, ksm_scan.address);
+
+       for (; vma; vma = vma->vm_next) {
+               if (!(vma->vm_flags & VM_MERGEABLE))
+                       continue;
+               if (ksm_scan.address < vma->vm_start)
+                       ksm_scan.address = vma->vm_start;
+               if (!vma->anon_vma)
+                       ksm_scan.address = vma->vm_end;
+
+               while (ksm_scan.address < vma->vm_end) {
+                       if (ksm_test_exit(mm))
+                               break;
+                       *page = follow_page(vma, ksm_scan.address, FOLL_GET);
+                       if (IS_ERR_OR_NULL(*page)) {
+                               ksm_scan.address += PAGE_SIZE;
+                               cond_resched();
+                               continue;
+                       }
+                       if (PageAnon(*page) ||
+                           page_trans_compound_anon(*page)) {
+                               flush_anon_page(vma, *page, ksm_scan.address);
+                               flush_dcache_page(*page);
+                               rmap_item = get_next_rmap_item(slot,
+                                       ksm_scan.rmap_list, ksm_scan.address);
+                               if (rmap_item) {
+                                       ksm_scan.rmap_list =
+                                                       &rmap_item->rmap_list;
+                                       ksm_scan.address += PAGE_SIZE;
+                               } else
+                                       put_page(*page);
+                               up_read(&mm->mmap_sem);
+                               return rmap_item;
+                       }
+                       put_page(*page);
+                       ksm_scan.address += PAGE_SIZE;
+                       cond_resched();
+               }
+       }
+
+       if (ksm_test_exit(mm)) {
+               ksm_scan.address = 0;
+               ksm_scan.rmap_list = &slot->rmap_list;
+       }
+       /*
+        * Nuke all the rmap_items that are above this current rmap:
+        * because there were no VM_MERGEABLE vmas with such addresses.
+        */
+       remove_trailing_rmap_items(slot, ksm_scan.rmap_list);
+
+       spin_lock(&ksm_mmlist_lock);
+       ksm_scan.mm_slot = list_entry(slot->mm_list.next,
+                                               struct mm_slot, mm_list);
+       if (ksm_scan.address == 0) {
+               /*
+                * We've completed a full scan of all vmas, holding mmap_sem
+                * throughout, and found no VM_MERGEABLE: so do the same as
+                * __ksm_exit does to remove this mm from all our lists now.
+                * This applies either when cleaning up after __ksm_exit
+                * (but beware: we can reach here even before __ksm_exit),
+                * or when all VM_MERGEABLE areas have been unmapped (and
+                * mmap_sem then protects against race with MADV_MERGEABLE).
+                */
+               hlist_del(&slot->link);
+               list_del(&slot->mm_list);
+               spin_unlock(&ksm_mmlist_lock);
+
+               free_mm_slot(slot);
+               clear_bit(MMF_VM_MERGEABLE, &mm->flags);
+               up_read(&mm->mmap_sem);
+               mmdrop(mm);
+       } else {
+               spin_unlock(&ksm_mmlist_lock);
+               up_read(&mm->mmap_sem);
+       }
+
+       /* Repeat until we've completed scanning the whole list */
+       slot = ksm_scan.mm_slot;
+       if (slot != &ksm_mm_head)
+               goto next_mm;
+
+       ksm_scan.seqnr++;
+       return NULL;
+}
+
+/**
+ * ksm_do_scan  - the ksm scanner main worker function.
+ * @scan_npages - number of pages we want to scan before we return.
+ */
+static void ksm_do_scan(unsigned int scan_npages)
+{
+       struct rmap_item *rmap_item;
+       struct page *uninitialized_var(page);
+
+       while (scan_npages-- && likely(!freezing(current))) {
+               cond_resched();
+               rmap_item = scan_get_next_rmap_item(&page);
+               if (!rmap_item)
+                       return;
+               if (!PageKsm(page) || !in_stable_tree(rmap_item))
+                       cmp_and_merge_page(page, rmap_item);
+               put_page(page);
+       }
+}
+
+static int ksmd_should_run(void)
+{
+       return (ksm_run & KSM_RUN_MERGE) && !list_empty(&ksm_mm_head.mm_list);
+}
+
+static int ksm_scan_thread(void *nothing)
+{
+       set_freezable();
+       set_user_nice(current, 5);
+
+       while (!kthread_should_stop()) {
+               mutex_lock(&ksm_thread_mutex);
+               if (ksmd_should_run())
+                       ksm_do_scan(ksm_thread_pages_to_scan);
+               mutex_unlock(&ksm_thread_mutex);
+
+               try_to_freeze();
+
+               if (ksmd_should_run()) {
+                       schedule_timeout_interruptible(
+                               msecs_to_jiffies(ksm_thread_sleep_millisecs));
+               } else {
+                       wait_event_freezable(ksm_thread_wait,
+                               ksmd_should_run() || kthread_should_stop());
+               }
+       }
+       return 0;
+}
 
 int ksm_madvise(struct vm_area_struct *vma, unsigned long start,
                unsigned long end, int advice, unsigned long *vm_flags)
 {
        struct mm_struct *mm = vma->vm_mm;
+       int err;
 
        switch (advice) {
        case MADV_MERGEABLE:
@@ -18,13 +1476,19 @@ int ksm_madvise(struct vm_area_struct *vma, unsigned long start,
                 */
                if (*vm_flags & (VM_MERGEABLE | VM_SHARED  | VM_MAYSHARE   |
                                 VM_PFNMAP    | VM_IO      | VM_DONTEXPAND |
-                                VM_RESERVED  | VM_HUGETLB | VM_INSERTPAGE |
-                                VM_MIXEDMAP  | VM_SAO))
+                                VM_HUGETLB | VM_NONLINEAR | VM_MIXEDMAP))
                        return 0;               /* just ignore the advice */
 
-               if (!test_bit(MMF_VM_MERGEABLE, &mm->flags))
-                       if (__ksm_enter(mm) < 0)
-                               return -EAGAIN;
+#ifdef VM_SAO
+               if (*vm_flags & VM_SAO)
+                       return 0;
+#endif
+
+               if (!test_bit(MMF_VM_MERGEABLE, &mm->flags)) {
+                       err = __ksm_enter(mm);
+                       if (err)
+                               return err;
+               }
 
                *vm_flags |= VM_MERGEABLE;
                break;
@@ -33,7 +1497,11 @@ int ksm_madvise(struct vm_area_struct *vma, unsigned long start,
                if (!(*vm_flags & VM_MERGEABLE))
                        return 0;               /* just ignore the advice */
 
-               /* Unmerge any merged pages here */
+               if (vma->anon_vma) {
+                       err = unmerge_ksm_pages(vma, start, end);
+                       if (err)
+                               return err;
+               }
 
                *vm_flags &= ~VM_MERGEABLE;
                break;
@@ -44,13 +1512,537 @@ int ksm_madvise(struct vm_area_struct *vma, unsigned long start,
 
 int __ksm_enter(struct mm_struct *mm)
 {
-       /* Allocate a structure to track mm and link it into KSM's list */
+       struct mm_slot *mm_slot;
+       int needs_wakeup;
+
+       mm_slot = alloc_mm_slot();
+       if (!mm_slot)
+               return -ENOMEM;
+
+       /* Check ksm_run too?  Would need tighter locking */
+       needs_wakeup = list_empty(&ksm_mm_head.mm_list);
+
+       spin_lock(&ksm_mmlist_lock);
+       insert_to_mm_slots_hash(mm, mm_slot);
+       /*
+        * Insert just behind the scanning cursor, to let the area settle
+        * down a little; when fork is followed by immediate exec, we don't
+        * want ksmd to waste time setting up and tearing down an rmap_list.
+        */
+       list_add_tail(&mm_slot->mm_list, &ksm_scan.mm_slot->mm_list);
+       spin_unlock(&ksm_mmlist_lock);
+
        set_bit(MMF_VM_MERGEABLE, &mm->flags);
+       atomic_inc(&mm->mm_count);
+
+       if (needs_wakeup)
+               wake_up_interruptible(&ksm_thread_wait);
+
        return 0;
 }
 
 void __ksm_exit(struct mm_struct *mm)
 {
-       /* Unlink and free all KSM's structures which track this mm */
-       clear_bit(MMF_VM_MERGEABLE, &mm->flags);
+       struct mm_slot *mm_slot;
+       int easy_to_free = 0;
+
+       /*
+        * This process is exiting: if it's straightforward (as is the
+        * case when ksmd was never running), free mm_slot immediately.
+        * But if it's at the cursor or has rmap_items linked to it, use
+        * mmap_sem to synchronize with any break_cows before pagetables
+        * are freed, and leave the mm_slot on the list for ksmd to free.
+        * Beware: ksm may already have noticed it exiting and freed the slot.
+        */
+
+       spin_lock(&ksm_mmlist_lock);
+       mm_slot = get_mm_slot(mm);
+       if (mm_slot && ksm_scan.mm_slot != mm_slot) {
+               if (!mm_slot->rmap_list) {
+                       hlist_del(&mm_slot->link);
+                       list_del(&mm_slot->mm_list);
+                       easy_to_free = 1;
+               } else {
+                       list_move(&mm_slot->mm_list,
+                                 &ksm_scan.mm_slot->mm_list);
+               }
+       }
+       spin_unlock(&ksm_mmlist_lock);
+
+       if (easy_to_free) {
+               free_mm_slot(mm_slot);
+               clear_bit(MMF_VM_MERGEABLE, &mm->flags);
+               mmdrop(mm);
+       } else if (mm_slot) {
+               down_write(&mm->mmap_sem);
+               up_write(&mm->mmap_sem);
+       }
+}
+
+struct page *ksm_does_need_to_copy(struct page *page,
+                       struct vm_area_struct *vma, unsigned long address)
+{
+       struct page *new_page;
+
+       new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
+       if (new_page) {
+               copy_user_highpage(new_page, page, address, vma);
+
+               SetPageDirty(new_page);
+               __SetPageUptodate(new_page);
+               SetPageSwapBacked(new_page);
+               __set_page_locked(new_page);
+
+               if (!mlocked_vma_newpage(vma, new_page))
+                       lru_cache_add_lru(new_page, LRU_ACTIVE_ANON);
+               else
+                       add_page_to_unevictable_list(new_page);
+       }
+
+       return new_page;
+}
+
+int page_referenced_ksm(struct page *page, struct mem_cgroup *memcg,
+                       unsigned long *vm_flags)
+{
+       struct stable_node *stable_node;
+       struct rmap_item *rmap_item;
+       struct hlist_node *hlist;
+       unsigned int mapcount = page_mapcount(page);
+       int referenced = 0;
+       int search_new_forks = 0;
+
+       VM_BUG_ON(!PageKsm(page));
+       VM_BUG_ON(!PageLocked(page));
+
+       stable_node = page_stable_node(page);
+       if (!stable_node)
+               return 0;
+again:
+       hlist_for_each_entry(rmap_item, hlist, &stable_node->hlist, hlist) {
+               struct anon_vma *anon_vma = rmap_item->anon_vma;
+               struct anon_vma_chain *vmac;
+               struct vm_area_struct *vma;
+
+               anon_vma_lock(anon_vma);
+               anon_vma_interval_tree_foreach(vmac, &anon_vma->rb_root,
+                                              0, ULONG_MAX) {
+                       vma = vmac->vma;
+                       if (rmap_item->address < vma->vm_start ||
+                           rmap_item->address >= vma->vm_end)
+                               continue;
+                       /*
+                        * Initially we examine only the vma which covers this
+                        * rmap_item; but later, if there is still work to do,
+                        * we examine covering vmas in other mms: in case they
+                        * were forked from the original since ksmd passed.
+                        */
+                       if ((rmap_item->mm == vma->vm_mm) == search_new_forks)
+                               continue;
+
+                       if (memcg && !mm_match_cgroup(vma->vm_mm, memcg))
+                               continue;
+
+                       referenced += page_referenced_one(page, vma,
+                               rmap_item->address, &mapcount, vm_flags);
+                       if (!search_new_forks || !mapcount)
+                               break;
+               }
+               anon_vma_unlock(anon_vma);
+               if (!mapcount)
+                       goto out;
+       }
+       if (!search_new_forks++)
+               goto again;
+out:
+       return referenced;
+}
+
+int try_to_unmap_ksm(struct page *page, enum ttu_flags flags)
+{
+       struct stable_node *stable_node;
+       struct hlist_node *hlist;
+       struct rmap_item *rmap_item;
+       int ret = SWAP_AGAIN;
+       int search_new_forks = 0;
+
+       VM_BUG_ON(!PageKsm(page));
+       VM_BUG_ON(!PageLocked(page));
+
+       stable_node = page_stable_node(page);
+       if (!stable_node)
+               return SWAP_FAIL;
+again:
+       hlist_for_each_entry(rmap_item, hlist, &stable_node->hlist, hlist) {
+               struct anon_vma *anon_vma = rmap_item->anon_vma;
+               struct anon_vma_chain *vmac;
+               struct vm_area_struct *vma;
+
+               anon_vma_lock(anon_vma);
+               anon_vma_interval_tree_foreach(vmac, &anon_vma->rb_root,
+                                              0, ULONG_MAX) {
+                       vma = vmac->vma;
+                       if (rmap_item->address < vma->vm_start ||
+                           rmap_item->address >= vma->vm_end)
+                               continue;
+                       /*
+                        * Initially we examine only the vma which covers this
+                        * rmap_item; but later, if there is still work to do,
+                        * we examine covering vmas in other mms: in case they
+                        * were forked from the original since ksmd passed.
+                        */
+                       if ((rmap_item->mm == vma->vm_mm) == search_new_forks)
+                               continue;
+
+                       ret = try_to_unmap_one(page, vma,
+                                       rmap_item->address, flags);
+                       if (ret != SWAP_AGAIN || !page_mapped(page)) {
+                               anon_vma_unlock(anon_vma);
+                               goto out;
+                       }
+               }
+               anon_vma_unlock(anon_vma);
+       }
+       if (!search_new_forks++)
+               goto again;
+out:
+       return ret;
+}
+
+#ifdef CONFIG_MIGRATION
+int rmap_walk_ksm(struct page *page, int (*rmap_one)(struct page *,
+                 struct vm_area_struct *, unsigned long, void *), void *arg)
+{
+       struct stable_node *stable_node;
+       struct hlist_node *hlist;
+       struct rmap_item *rmap_item;
+       int ret = SWAP_AGAIN;
+       int search_new_forks = 0;
+
+       VM_BUG_ON(!PageKsm(page));
+       VM_BUG_ON(!PageLocked(page));
+
+       stable_node = page_stable_node(page);
+       if (!stable_node)
+               return ret;
+again:
+       hlist_for_each_entry(rmap_item, hlist, &stable_node->hlist, hlist) {
+               struct anon_vma *anon_vma = rmap_item->anon_vma;
+               struct anon_vma_chain *vmac;
+               struct vm_area_struct *vma;
+
+               anon_vma_lock(anon_vma);
+               anon_vma_interval_tree_foreach(vmac, &anon_vma->rb_root,
+                                              0, ULONG_MAX) {
+                       vma = vmac->vma;
+                       if (rmap_item->address < vma->vm_start ||
+                           rmap_item->address >= vma->vm_end)
+                               continue;
+                       /*
+                        * Initially we examine only the vma which covers this
+                        * rmap_item; but later, if there is still work to do,
+                        * we examine covering vmas in other mms: in case they
+                        * were forked from the original since ksmd passed.
+                        */
+                       if ((rmap_item->mm == vma->vm_mm) == search_new_forks)
+                               continue;
+
+                       ret = rmap_one(page, vma, rmap_item->address, arg);
+                       if (ret != SWAP_AGAIN) {
+                               anon_vma_unlock(anon_vma);
+                               goto out;
+                       }
+               }
+               anon_vma_unlock(anon_vma);
+       }
+       if (!search_new_forks++)
+               goto again;
+out:
+       return ret;
+}
+
+void ksm_migrate_page(struct page *newpage, struct page *oldpage)
+{
+       struct stable_node *stable_node;
+
+       VM_BUG_ON(!PageLocked(oldpage));
+       VM_BUG_ON(!PageLocked(newpage));
+       VM_BUG_ON(newpage->mapping != oldpage->mapping);
+
+       stable_node = page_stable_node(newpage);
+       if (stable_node) {
+               VM_BUG_ON(stable_node->kpfn != page_to_pfn(oldpage));
+               stable_node->kpfn = page_to_pfn(newpage);
+       }
+}
+#endif /* CONFIG_MIGRATION */
+
+#ifdef CONFIG_MEMORY_HOTREMOVE
+static struct stable_node *ksm_check_stable_tree(unsigned long start_pfn,
+                                                unsigned long end_pfn)
+{
+       struct rb_node *node;
+
+       for (node = rb_first(&root_stable_tree); node; node = rb_next(node)) {
+               struct stable_node *stable_node;
+
+               stable_node = rb_entry(node, struct stable_node, node);
+               if (stable_node->kpfn >= start_pfn &&
+                   stable_node->kpfn < end_pfn)
+                       return stable_node;
+       }
+       return NULL;
+}
+
+static int ksm_memory_callback(struct notifier_block *self,
+                              unsigned long action, void *arg)
+{
+       struct memory_notify *mn = arg;
+       struct stable_node *stable_node;
+
+       switch (action) {
+       case MEM_GOING_OFFLINE:
+               /*
+                * Keep it very simple for now: just lock out ksmd and
+                * MADV_UNMERGEABLE while any memory is going offline.
+                * mutex_lock_nested() is necessary because lockdep was alarmed
+                * that here we take ksm_thread_mutex inside notifier chain
+                * mutex, and later take notifier chain mutex inside
+                * ksm_thread_mutex to unlock it.   But that's safe because both
+                * are inside mem_hotplug_mutex.
+                */
+               mutex_lock_nested(&ksm_thread_mutex, SINGLE_DEPTH_NESTING);
+               break;
+
+       case MEM_OFFLINE:
+               /*
+                * Most of the work is done by page migration; but there might
+                * be a few stable_nodes left over, still pointing to struct
+                * pages which have been offlined: prune those from the tree.
+                */
+               while ((stable_node = ksm_check_stable_tree(mn->start_pfn,
+                                       mn->start_pfn + mn->nr_pages)) != NULL)
+                       remove_node_from_stable_tree(stable_node);
+               /* fallthrough */
+
+       case MEM_CANCEL_OFFLINE:
+               mutex_unlock(&ksm_thread_mutex);
+               break;
+       }
+       return NOTIFY_OK;
+}
+#endif /* CONFIG_MEMORY_HOTREMOVE */
+
+#ifdef CONFIG_SYSFS
+/*
+ * This all compiles without CONFIG_SYSFS, but is a waste of space.
+ */
+
+#define KSM_ATTR_RO(_name) \
+       static struct kobj_attribute _name##_attr = __ATTR_RO(_name)
+#define KSM_ATTR(_name) \
+       static struct kobj_attribute _name##_attr = \
+               __ATTR(_name, 0644, _name##_show, _name##_store)
+
+static ssize_t sleep_millisecs_show(struct kobject *kobj,
+                                   struct kobj_attribute *attr, char *buf)
+{
+       return sprintf(buf, "%u\n", ksm_thread_sleep_millisecs);
+}
+
+static ssize_t sleep_millisecs_store(struct kobject *kobj,
+                                    struct kobj_attribute *attr,
+                                    const char *buf, size_t count)
+{
+       unsigned long msecs;
+       int err;
+
+       err = strict_strtoul(buf, 10, &msecs);
+       if (err || msecs > UINT_MAX)
+               return -EINVAL;
+
+       ksm_thread_sleep_millisecs = msecs;
+
+       return count;
+}
+KSM_ATTR(sleep_millisecs);
+
+static ssize_t pages_to_scan_show(struct kobject *kobj,
+                                 struct kobj_attribute *attr, char *buf)
+{
+       return sprintf(buf, "%u\n", ksm_thread_pages_to_scan);
+}
+
+static ssize_t pages_to_scan_store(struct kobject *kobj,
+                                  struct kobj_attribute *attr,
+                                  const char *buf, size_t count)
+{
+       int err;
+       unsigned long nr_pages;
+
+       err = strict_strtoul(buf, 10, &nr_pages);
+       if (err || nr_pages > UINT_MAX)
+               return -EINVAL;
+
+       ksm_thread_pages_to_scan = nr_pages;
+
+       return count;
+}
+KSM_ATTR(pages_to_scan);
+
+static ssize_t run_show(struct kobject *kobj, struct kobj_attribute *attr,
+                       char *buf)
+{
+       return sprintf(buf, "%u\n", ksm_run);
+}
+
+static ssize_t run_store(struct kobject *kobj, struct kobj_attribute *attr,
+                        const char *buf, size_t count)
+{
+       int err;
+       unsigned long flags;
+
+       err = strict_strtoul(buf, 10, &flags);
+       if (err || flags > UINT_MAX)
+               return -EINVAL;
+       if (flags > KSM_RUN_UNMERGE)
+               return -EINVAL;
+
+       /*
+        * KSM_RUN_MERGE sets ksmd running, and 0 stops it running.
+        * KSM_RUN_UNMERGE stops it running and unmerges all rmap_items,
+        * breaking COW to free the pages_shared (but leaves mm_slots
+        * on the list for when ksmd may be set running again).
+        */
+
+       mutex_lock(&ksm_thread_mutex);
+       if (ksm_run != flags) {
+               ksm_run = flags;
+               if (flags & KSM_RUN_UNMERGE) {
+                       short oom_score_adj;
+
+                       oom_score_adj = test_set_oom_score_adj(OOM_SCORE_ADJ_MAX);
+                       err = unmerge_and_remove_all_rmap_items();
+                       compare_swap_oom_score_adj(OOM_SCORE_ADJ_MAX,
+                                                               oom_score_adj);
+                       if (err) {
+                               ksm_run = KSM_RUN_STOP;
+                               count = err;
+                       }
+               }
+       }
+       mutex_unlock(&ksm_thread_mutex);
+
+       if (flags & KSM_RUN_MERGE)
+               wake_up_interruptible(&ksm_thread_wait);
+
+       return count;
+}
+KSM_ATTR(run);
+
+static ssize_t pages_shared_show(struct kobject *kobj,
+                                struct kobj_attribute *attr, char *buf)
+{
+       return sprintf(buf, "%lu\n", ksm_pages_shared);
+}
+KSM_ATTR_RO(pages_shared);
+
+static ssize_t pages_sharing_show(struct kobject *kobj,
+                                 struct kobj_attribute *attr, char *buf)
+{
+       return sprintf(buf, "%lu\n", ksm_pages_sharing);
+}
+KSM_ATTR_RO(pages_sharing);
+
+static ssize_t pages_unshared_show(struct kobject *kobj,
+                                  struct kobj_attribute *attr, char *buf)
+{
+       return sprintf(buf, "%lu\n", ksm_pages_unshared);
+}
+KSM_ATTR_RO(pages_unshared);
+
+static ssize_t pages_volatile_show(struct kobject *kobj,
+                                  struct kobj_attribute *attr, char *buf)
+{
+       long ksm_pages_volatile;
+
+       ksm_pages_volatile = ksm_rmap_items - ksm_pages_shared
+                               - ksm_pages_sharing - ksm_pages_unshared;
+       /*
+        * It was not worth any locking to calculate that statistic,
+        * but it might therefore sometimes be negative: conceal that.
+        */
+       if (ksm_pages_volatile < 0)
+               ksm_pages_volatile = 0;
+       return sprintf(buf, "%ld\n", ksm_pages_volatile);
+}
+KSM_ATTR_RO(pages_volatile);
+
+static ssize_t full_scans_show(struct kobject *kobj,
+                              struct kobj_attribute *attr, char *buf)
+{
+       return sprintf(buf, "%lu\n", ksm_scan.seqnr);
+}
+KSM_ATTR_RO(full_scans);
+
+static struct attribute *ksm_attrs[] = {
+       &sleep_millisecs_attr.attr,
+       &pages_to_scan_attr.attr,
+       &run_attr.attr,
+       &pages_shared_attr.attr,
+       &pages_sharing_attr.attr,
+       &pages_unshared_attr.attr,
+       &pages_volatile_attr.attr,
+       &full_scans_attr.attr,
+       NULL,
+};
+
+static struct attribute_group ksm_attr_group = {
+       .attrs = ksm_attrs,
+       .name = "ksm",
+};
+#endif /* CONFIG_SYSFS */
+
+static int __init ksm_init(void)
+{
+       struct task_struct *ksm_thread;
+       int err;
+
+       err = ksm_slab_init();
+       if (err)
+               goto out;
+
+       ksm_thread = kthread_run(ksm_scan_thread, NULL, "ksmd");
+       if (IS_ERR(ksm_thread)) {
+               printk(KERN_ERR "ksm: creating kthread failed\n");
+               err = PTR_ERR(ksm_thread);
+               goto out_free;
+       }
+
+#ifdef CONFIG_SYSFS
+       err = sysfs_create_group(mm_kobj, &ksm_attr_group);
+       if (err) {
+               printk(KERN_ERR "ksm: register sysfs failed\n");
+               kthread_stop(ksm_thread);
+               goto out_free;
+       }
+#else
+       ksm_run = KSM_RUN_MERGE;        /* no way for user to start it */
+
+#endif /* CONFIG_SYSFS */
+
+#ifdef CONFIG_MEMORY_HOTREMOVE
+       /*
+        * Choose a high priority since the callback takes ksm_thread_mutex:
+        * later callbacks could only be taking locks which nest within that.
+        */
+       hotplug_memory_notifier(ksm_memory_callback, 100);
+#endif
+       return 0;
+
+out_free:
+       ksm_slab_free();
+out:
+       return err;
 }
+module_init(ksm_init)