mm/ksm.c

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