~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/mm/ksm.c

Version: ~ [ linux-5.0-rc6 ] ~ [ linux-4.20.10 ] ~ [ linux-4.19.23 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.101 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.158 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.174 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.134 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.63 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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/sched/mm.h>
 23 #include <linux/sched/coredump.h>
 24 #include <linux/rwsem.h>
 25 #include <linux/pagemap.h>
 26 #include <linux/rmap.h>
 27 #include <linux/spinlock.h>
 28 #include <linux/xxhash.h>
 29 #include <linux/delay.h>
 30 #include <linux/kthread.h>
 31 #include <linux/wait.h>
 32 #include <linux/slab.h>
 33 #include <linux/rbtree.h>
 34 #include <linux/memory.h>
 35 #include <linux/mmu_notifier.h>
 36 #include <linux/swap.h>
 37 #include <linux/ksm.h>
 38 #include <linux/hashtable.h>
 39 #include <linux/freezer.h>
 40 #include <linux/oom.h>
 41 #include <linux/numa.h>
 42 
 43 #include <asm/tlbflush.h>
 44 #include "internal.h"
 45 
 46 #ifdef CONFIG_NUMA
 47 #define NUMA(x)         (x)
 48 #define DO_NUMA(x)      do { (x); } while (0)
 49 #else
 50 #define NUMA(x)         (0)
 51 #define DO_NUMA(x)      do { } while (0)
 52 #endif
 53 
 54 /**
 55  * DOC: Overview
 56  *
 57  * A few notes about the KSM scanning process,
 58  * to make it easier to understand the data structures below:
 59  *
 60  * In order to reduce excessive scanning, KSM sorts the memory pages by their
 61  * contents into a data structure that holds pointers to the pages' locations.
 62  *
 63  * Since the contents of the pages may change at any moment, KSM cannot just
 64  * insert the pages into a normal sorted tree and expect it to find anything.
 65  * Therefore KSM uses two data structures - the stable and the unstable tree.
 66  *
 67  * The stable tree holds pointers to all the merged pages (ksm pages), sorted
 68  * by their contents.  Because each such page is write-protected, searching on
 69  * this tree is fully assured to be working (except when pages are unmapped),
 70  * and therefore this tree is called the stable tree.
 71  *
 72  * The stable tree node includes information required for reverse
 73  * mapping from a KSM page to virtual addresses that map this page.
 74  *
 75  * In order to avoid large latencies of the rmap walks on KSM pages,
 76  * KSM maintains two types of nodes in the stable tree:
 77  *
 78  * * the regular nodes that keep the reverse mapping structures in a
 79  *   linked list
 80  * * the "chains" that link nodes ("dups") that represent the same
 81  *   write protected memory content, but each "dup" corresponds to a
 82  *   different KSM page copy of that content
 83  *
 84  * Internally, the regular nodes, "dups" and "chains" are represented
 85  * using the same :c:type:`struct stable_node` structure.
 86  *
 87  * In addition to the stable tree, KSM uses a second data structure called the
 88  * unstable tree: this tree holds pointers to pages which have been found to
 89  * be "unchanged for a period of time".  The unstable tree sorts these pages
 90  * by their contents, but since they are not write-protected, KSM cannot rely
 91  * upon the unstable tree to work correctly - the unstable tree is liable to
 92  * be corrupted as its contents are modified, and so it is called unstable.
 93  *
 94  * KSM solves this problem by several techniques:
 95  *
 96  * 1) The unstable tree is flushed every time KSM completes scanning all
 97  *    memory areas, and then the tree is rebuilt again from the beginning.
 98  * 2) KSM will only insert into the unstable tree, pages whose hash value
 99  *    has not changed since the previous scan of all memory areas.
100  * 3) The unstable tree is a RedBlack Tree - so its balancing is based on the
101  *    colors of the nodes and not on their contents, assuring that even when
102  *    the tree gets "corrupted" it won't get out of balance, so scanning time
103  *    remains the same (also, searching and inserting nodes in an rbtree uses
104  *    the same algorithm, so we have no overhead when we flush and rebuild).
105  * 4) KSM never flushes the stable tree, which means that even if it were to
106  *    take 10 attempts to find a page in the unstable tree, once it is found,
107  *    it is secured in the stable tree.  (When we scan a new page, we first
108  *    compare it against the stable tree, and then against the unstable tree.)
109  *
110  * If the merge_across_nodes tunable is unset, then KSM maintains multiple
111  * stable trees and multiple unstable trees: one of each for each NUMA node.
112  */
113 
114 /**
115  * struct mm_slot - ksm information per mm that is being scanned
116  * @link: link to the mm_slots hash list
117  * @mm_list: link into the mm_slots list, rooted in ksm_mm_head
118  * @rmap_list: head for this mm_slot's singly-linked list of rmap_items
119  * @mm: the mm that this information is valid for
120  */
121 struct mm_slot {
122         struct hlist_node link;
123         struct list_head mm_list;
124         struct rmap_item *rmap_list;
125         struct mm_struct *mm;
126 };
127 
128 /**
129  * struct ksm_scan - cursor for scanning
130  * @mm_slot: the current mm_slot we are scanning
131  * @address: the next address inside that to be scanned
132  * @rmap_list: link to the next rmap to be scanned in the rmap_list
133  * @seqnr: count of completed full scans (needed when removing unstable node)
134  *
135  * There is only the one ksm_scan instance of this cursor structure.
136  */
137 struct ksm_scan {
138         struct mm_slot *mm_slot;
139         unsigned long address;
140         struct rmap_item **rmap_list;
141         unsigned long seqnr;
142 };
143 
144 /**
145  * struct stable_node - node of the stable rbtree
146  * @node: rb node of this ksm page in the stable tree
147  * @head: (overlaying parent) &migrate_nodes indicates temporarily on that list
148  * @hlist_dup: linked into the stable_node->hlist with a stable_node chain
149  * @list: linked into migrate_nodes, pending placement in the proper node tree
150  * @hlist: hlist head of rmap_items using this ksm page
151  * @kpfn: page frame number of this ksm page (perhaps temporarily on wrong nid)
152  * @chain_prune_time: time of the last full garbage collection
153  * @rmap_hlist_len: number of rmap_item entries in hlist or STABLE_NODE_CHAIN
154  * @nid: NUMA node id of stable tree in which linked (may not match kpfn)
155  */
156 struct stable_node {
157         union {
158                 struct rb_node node;    /* when node of stable tree */
159                 struct {                /* when listed for migration */
160                         struct list_head *head;
161                         struct {
162                                 struct hlist_node hlist_dup;
163                                 struct list_head list;
164                         };
165                 };
166         };
167         struct hlist_head hlist;
168         union {
169                 unsigned long kpfn;
170                 unsigned long chain_prune_time;
171         };
172         /*
173          * STABLE_NODE_CHAIN can be any negative number in
174          * rmap_hlist_len negative range, but better not -1 to be able
175          * to reliably detect underflows.
176          */
177 #define STABLE_NODE_CHAIN -1024
178         int rmap_hlist_len;
179 #ifdef CONFIG_NUMA
180         int nid;
181 #endif
182 };
183 
184 /**
185  * struct rmap_item - reverse mapping item for virtual addresses
186  * @rmap_list: next rmap_item in mm_slot's singly-linked rmap_list
187  * @anon_vma: pointer to anon_vma for this mm,address, when in stable tree
188  * @nid: NUMA node id of unstable tree in which linked (may not match page)
189  * @mm: the memory structure this rmap_item is pointing into
190  * @address: the virtual address this rmap_item tracks (+ flags in low bits)
191  * @oldchecksum: previous checksum of the page at that virtual address
192  * @node: rb node of this rmap_item in the unstable tree
193  * @head: pointer to stable_node heading this list in the stable tree
194  * @hlist: link into hlist of rmap_items hanging off that stable_node
195  */
196 struct rmap_item {
197         struct rmap_item *rmap_list;
198         union {
199                 struct anon_vma *anon_vma;      /* when stable */
200 #ifdef CONFIG_NUMA
201                 int nid;                /* when node of unstable tree */
202 #endif
203         };
204         struct mm_struct *mm;
205         unsigned long address;          /* + low bits used for flags below */
206         unsigned int oldchecksum;       /* when unstable */
207         union {
208                 struct rb_node node;    /* when node of unstable tree */
209                 struct {                /* when listed from stable tree */
210                         struct stable_node *head;
211                         struct hlist_node hlist;
212                 };
213         };
214 };
215 
216 #define SEQNR_MASK      0x0ff   /* low bits of unstable tree seqnr */
217 #define UNSTABLE_FLAG   0x100   /* is a node of the unstable tree */
218 #define STABLE_FLAG     0x200   /* is listed from the stable tree */
219 #define KSM_FLAG_MASK   (SEQNR_MASK|UNSTABLE_FLAG|STABLE_FLAG)
220                                 /* to mask all the flags */
221 
222 /* The stable and unstable tree heads */
223 static struct rb_root one_stable_tree[1] = { RB_ROOT };
224 static struct rb_root one_unstable_tree[1] = { RB_ROOT };
225 static struct rb_root *root_stable_tree = one_stable_tree;
226 static struct rb_root *root_unstable_tree = one_unstable_tree;
227 
228 /* Recently migrated nodes of stable tree, pending proper placement */
229 static LIST_HEAD(migrate_nodes);
230 #define STABLE_NODE_DUP_HEAD ((struct list_head *)&migrate_nodes.prev)
231 
232 #define MM_SLOTS_HASH_BITS 10
233 static DEFINE_HASHTABLE(mm_slots_hash, MM_SLOTS_HASH_BITS);
234 
235 static struct mm_slot ksm_mm_head = {
236         .mm_list = LIST_HEAD_INIT(ksm_mm_head.mm_list),
237 };
238 static struct ksm_scan ksm_scan = {
239         .mm_slot = &ksm_mm_head,
240 };
241 
242 static struct kmem_cache *rmap_item_cache;
243 static struct kmem_cache *stable_node_cache;
244 static struct kmem_cache *mm_slot_cache;
245 
246 /* The number of nodes in the stable tree */
247 static unsigned long ksm_pages_shared;
248 
249 /* The number of page slots additionally sharing those nodes */
250 static unsigned long ksm_pages_sharing;
251 
252 /* The number of nodes in the unstable tree */
253 static unsigned long ksm_pages_unshared;
254 
255 /* The number of rmap_items in use: to calculate pages_volatile */
256 static unsigned long ksm_rmap_items;
257 
258 /* The number of stable_node chains */
259 static unsigned long ksm_stable_node_chains;
260 
261 /* The number of stable_node dups linked to the stable_node chains */
262 static unsigned long ksm_stable_node_dups;
263 
264 /* Delay in pruning stale stable_node_dups in the stable_node_chains */
265 static int ksm_stable_node_chains_prune_millisecs = 2000;
266 
267 /* Maximum number of page slots sharing a stable node */
268 static int ksm_max_page_sharing = 256;
269 
270 /* Number of pages ksmd should scan in one batch */
271 static unsigned int ksm_thread_pages_to_scan = 100;
272 
273 /* Milliseconds ksmd should sleep between batches */
274 static unsigned int ksm_thread_sleep_millisecs = 20;
275 
276 /* Checksum of an empty (zeroed) page */
277 static unsigned int zero_checksum __read_mostly;
278 
279 /* Whether to merge empty (zeroed) pages with actual zero pages */
280 static bool ksm_use_zero_pages __read_mostly;
281 
282 #ifdef CONFIG_NUMA
283 /* Zeroed when merging across nodes is not allowed */
284 static unsigned int ksm_merge_across_nodes = 1;
285 static int ksm_nr_node_ids = 1;
286 #else
287 #define ksm_merge_across_nodes  1U
288 #define ksm_nr_node_ids         1
289 #endif
290 
291 #define KSM_RUN_STOP    0
292 #define KSM_RUN_MERGE   1
293 #define KSM_RUN_UNMERGE 2
294 #define KSM_RUN_OFFLINE 4
295 static unsigned long ksm_run = KSM_RUN_STOP;
296 static void wait_while_offlining(void);
297 
298 static DECLARE_WAIT_QUEUE_HEAD(ksm_thread_wait);
299 static DECLARE_WAIT_QUEUE_HEAD(ksm_iter_wait);
300 static DEFINE_MUTEX(ksm_thread_mutex);
301 static DEFINE_SPINLOCK(ksm_mmlist_lock);
302 
303 #define KSM_KMEM_CACHE(__struct, __flags) kmem_cache_create("ksm_"#__struct,\
304                 sizeof(struct __struct), __alignof__(struct __struct),\
305                 (__flags), NULL)
306 
307 static int __init ksm_slab_init(void)
308 {
309         rmap_item_cache = KSM_KMEM_CACHE(rmap_item, 0);
310         if (!rmap_item_cache)
311                 goto out;
312 
313         stable_node_cache = KSM_KMEM_CACHE(stable_node, 0);
314         if (!stable_node_cache)
315                 goto out_free1;
316 
317         mm_slot_cache = KSM_KMEM_CACHE(mm_slot, 0);
318         if (!mm_slot_cache)
319                 goto out_free2;
320 
321         return 0;
322 
323 out_free2:
324         kmem_cache_destroy(stable_node_cache);
325 out_free1:
326         kmem_cache_destroy(rmap_item_cache);
327 out:
328         return -ENOMEM;
329 }
330 
331 static void __init ksm_slab_free(void)
332 {
333         kmem_cache_destroy(mm_slot_cache);
334         kmem_cache_destroy(stable_node_cache);
335         kmem_cache_destroy(rmap_item_cache);
336         mm_slot_cache = NULL;
337 }
338 
339 static __always_inline bool is_stable_node_chain(struct stable_node *chain)
340 {
341         return chain->rmap_hlist_len == STABLE_NODE_CHAIN;
342 }
343 
344 static __always_inline bool is_stable_node_dup(struct stable_node *dup)
345 {
346         return dup->head == STABLE_NODE_DUP_HEAD;
347 }
348 
349 static inline void stable_node_chain_add_dup(struct stable_node *dup,
350                                              struct stable_node *chain)
351 {
352         VM_BUG_ON(is_stable_node_dup(dup));
353         dup->head = STABLE_NODE_DUP_HEAD;
354         VM_BUG_ON(!is_stable_node_chain(chain));
355         hlist_add_head(&dup->hlist_dup, &chain->hlist);
356         ksm_stable_node_dups++;
357 }
358 
359 static inline void __stable_node_dup_del(struct stable_node *dup)
360 {
361         VM_BUG_ON(!is_stable_node_dup(dup));
362         hlist_del(&dup->hlist_dup);
363         ksm_stable_node_dups--;
364 }
365 
366 static inline void stable_node_dup_del(struct stable_node *dup)
367 {
368         VM_BUG_ON(is_stable_node_chain(dup));
369         if (is_stable_node_dup(dup))
370                 __stable_node_dup_del(dup);
371         else
372                 rb_erase(&dup->node, root_stable_tree + NUMA(dup->nid));
373 #ifdef CONFIG_DEBUG_VM
374         dup->head = NULL;
375 #endif
376 }
377 
378 static inline struct rmap_item *alloc_rmap_item(void)
379 {
380         struct rmap_item *rmap_item;
381 
382         rmap_item = kmem_cache_zalloc(rmap_item_cache, GFP_KERNEL |
383                                                 __GFP_NORETRY | __GFP_NOWARN);
384         if (rmap_item)
385                 ksm_rmap_items++;
386         return rmap_item;
387 }
388 
389 static inline void free_rmap_item(struct rmap_item *rmap_item)
390 {
391         ksm_rmap_items--;
392         rmap_item->mm = NULL;   /* debug safety */
393         kmem_cache_free(rmap_item_cache, rmap_item);
394 }
395 
396 static inline struct stable_node *alloc_stable_node(void)
397 {
398         /*
399          * The allocation can take too long with GFP_KERNEL when memory is under
400          * pressure, which may lead to hung task warnings.  Adding __GFP_HIGH
401          * grants access to memory reserves, helping to avoid this problem.
402          */
403         return kmem_cache_alloc(stable_node_cache, GFP_KERNEL | __GFP_HIGH);
404 }
405 
406 static inline void free_stable_node(struct stable_node *stable_node)
407 {
408         VM_BUG_ON(stable_node->rmap_hlist_len &&
409                   !is_stable_node_chain(stable_node));
410         kmem_cache_free(stable_node_cache, stable_node);
411 }
412 
413 static inline struct mm_slot *alloc_mm_slot(void)
414 {
415         if (!mm_slot_cache)     /* initialization failed */
416                 return NULL;
417         return kmem_cache_zalloc(mm_slot_cache, GFP_KERNEL);
418 }
419 
420 static inline void free_mm_slot(struct mm_slot *mm_slot)
421 {
422         kmem_cache_free(mm_slot_cache, mm_slot);
423 }
424 
425 static struct mm_slot *get_mm_slot(struct mm_struct *mm)
426 {
427         struct mm_slot *slot;
428 
429         hash_for_each_possible(mm_slots_hash, slot, link, (unsigned long)mm)
430                 if (slot->mm == mm)
431                         return slot;
432 
433         return NULL;
434 }
435 
436 static void insert_to_mm_slots_hash(struct mm_struct *mm,
437                                     struct mm_slot *mm_slot)
438 {
439         mm_slot->mm = mm;
440         hash_add(mm_slots_hash, &mm_slot->link, (unsigned long)mm);
441 }
442 
443 /*
444  * ksmd, and unmerge_and_remove_all_rmap_items(), must not touch an mm's
445  * page tables after it has passed through ksm_exit() - which, if necessary,
446  * takes mmap_sem briefly to serialize against them.  ksm_exit() does not set
447  * a special flag: they can just back out as soon as mm_users goes to zero.
448  * ksm_test_exit() is used throughout to make this test for exit: in some
449  * places for correctness, in some places just to avoid unnecessary work.
450  */
451 static inline bool ksm_test_exit(struct mm_struct *mm)
452 {
453         return atomic_read(&mm->mm_users) == 0;
454 }
455 
456 /*
457  * We use break_ksm to break COW on a ksm page: it's a stripped down
458  *
459  *      if (get_user_pages(addr, 1, 1, 1, &page, NULL) == 1)
460  *              put_page(page);
461  *
462  * but taking great care only to touch a ksm page, in a VM_MERGEABLE vma,
463  * in case the application has unmapped and remapped mm,addr meanwhile.
464  * Could a ksm page appear anywhere else?  Actually yes, in a VM_PFNMAP
465  * mmap of /dev/mem or /dev/kmem, where we would not want to touch it.
466  *
467  * FAULT_FLAG/FOLL_REMOTE are because we do this outside the context
468  * of the process that owns 'vma'.  We also do not want to enforce
469  * protection keys here anyway.
470  */
471 static int break_ksm(struct vm_area_struct *vma, unsigned long addr)
472 {
473         struct page *page;
474         vm_fault_t ret = 0;
475 
476         do {
477                 cond_resched();
478                 page = follow_page(vma, addr,
479                                 FOLL_GET | FOLL_MIGRATION | FOLL_REMOTE);
480                 if (IS_ERR_OR_NULL(page))
481                         break;
482                 if (PageKsm(page))
483                         ret = handle_mm_fault(vma, addr,
484                                         FAULT_FLAG_WRITE | FAULT_FLAG_REMOTE);
485                 else
486                         ret = VM_FAULT_WRITE;
487                 put_page(page);
488         } while (!(ret & (VM_FAULT_WRITE | VM_FAULT_SIGBUS | VM_FAULT_SIGSEGV | VM_FAULT_OOM)));
489         /*
490          * We must loop because handle_mm_fault() may back out if there's
491          * any difficulty e.g. if pte accessed bit gets updated concurrently.
492          *
493          * VM_FAULT_WRITE is what we have been hoping for: it indicates that
494          * COW has been broken, even if the vma does not permit VM_WRITE;
495          * but note that a concurrent fault might break PageKsm for us.
496          *
497          * VM_FAULT_SIGBUS could occur if we race with truncation of the
498          * backing file, which also invalidates anonymous pages: that's
499          * okay, that truncation will have unmapped the PageKsm for us.
500          *
501          * VM_FAULT_OOM: at the time of writing (late July 2009), setting
502          * aside mem_cgroup limits, VM_FAULT_OOM would only be set if the
503          * current task has TIF_MEMDIE set, and will be OOM killed on return
504          * to user; and ksmd, having no mm, would never be chosen for that.
505          *
506          * But if the mm is in a limited mem_cgroup, then the fault may fail
507          * with VM_FAULT_OOM even if the current task is not TIF_MEMDIE; and
508          * even ksmd can fail in this way - though it's usually breaking ksm
509          * just to undo a merge it made a moment before, so unlikely to oom.
510          *
511          * That's a pity: we might therefore have more kernel pages allocated
512          * than we're counting as nodes in the stable tree; but ksm_do_scan
513          * will retry to break_cow on each pass, so should recover the page
514          * in due course.  The important thing is to not let VM_MERGEABLE
515          * be cleared while any such pages might remain in the area.
516          */
517         return (ret & VM_FAULT_OOM) ? -ENOMEM : 0;
518 }
519 
520 static struct vm_area_struct *find_mergeable_vma(struct mm_struct *mm,
521                 unsigned long addr)
522 {
523         struct vm_area_struct *vma;
524         if (ksm_test_exit(mm))
525                 return NULL;
526         vma = find_vma(mm, addr);
527         if (!vma || vma->vm_start > addr)
528                 return NULL;
529         if (!(vma->vm_flags & VM_MERGEABLE) || !vma->anon_vma)
530                 return NULL;
531         return vma;
532 }
533 
534 static void break_cow(struct rmap_item *rmap_item)
535 {
536         struct mm_struct *mm = rmap_item->mm;
537         unsigned long addr = rmap_item->address;
538         struct vm_area_struct *vma;
539 
540         /*
541          * It is not an accident that whenever we want to break COW
542          * to undo, we also need to drop a reference to the anon_vma.
543          */
544         put_anon_vma(rmap_item->anon_vma);
545 
546         down_read(&mm->mmap_sem);
547         vma = find_mergeable_vma(mm, addr);
548         if (vma)
549                 break_ksm(vma, addr);
550         up_read(&mm->mmap_sem);
551 }
552 
553 static struct page *get_mergeable_page(struct rmap_item *rmap_item)
554 {
555         struct mm_struct *mm = rmap_item->mm;
556         unsigned long addr = rmap_item->address;
557         struct vm_area_struct *vma;
558         struct page *page;
559 
560         down_read(&mm->mmap_sem);
561         vma = find_mergeable_vma(mm, addr);
562         if (!vma)
563                 goto out;
564 
565         page = follow_page(vma, addr, FOLL_GET);
566         if (IS_ERR_OR_NULL(page))
567                 goto out;
568         if (PageAnon(page)) {
569                 flush_anon_page(vma, page, addr);
570                 flush_dcache_page(page);
571         } else {
572                 put_page(page);
573 out:
574                 page = NULL;
575         }
576         up_read(&mm->mmap_sem);
577         return page;
578 }
579 
580 /*
581  * This helper is used for getting right index into array of tree roots.
582  * When merge_across_nodes knob is set to 1, there are only two rb-trees for
583  * stable and unstable pages from all nodes with roots in index 0. Otherwise,
584  * every node has its own stable and unstable tree.
585  */
586 static inline int get_kpfn_nid(unsigned long kpfn)
587 {
588         return ksm_merge_across_nodes ? 0 : NUMA(pfn_to_nid(kpfn));
589 }
590 
591 static struct stable_node *alloc_stable_node_chain(struct stable_node *dup,
592                                                    struct rb_root *root)
593 {
594         struct stable_node *chain = alloc_stable_node();
595         VM_BUG_ON(is_stable_node_chain(dup));
596         if (likely(chain)) {
597                 INIT_HLIST_HEAD(&chain->hlist);
598                 chain->chain_prune_time = jiffies;
599                 chain->rmap_hlist_len = STABLE_NODE_CHAIN;
600 #if defined (CONFIG_DEBUG_VM) && defined(CONFIG_NUMA)
601                 chain->nid = -1; /* debug */
602 #endif
603                 ksm_stable_node_chains++;
604 
605                 /*
606                  * Put the stable node chain in the first dimension of
607                  * the stable tree and at the same time remove the old
608                  * stable node.
609                  */
610                 rb_replace_node(&dup->node, &chain->node, root);
611 
612                 /*
613                  * Move the old stable node to the second dimension
614                  * queued in the hlist_dup. The invariant is that all
615                  * dup stable_nodes in the chain->hlist point to pages
616                  * that are wrprotected and have the exact same
617                  * content.
618                  */
619                 stable_node_chain_add_dup(dup, chain);
620         }
621         return chain;
622 }
623 
624 static inline void free_stable_node_chain(struct stable_node *chain,
625                                           struct rb_root *root)
626 {
627         rb_erase(&chain->node, root);
628         free_stable_node(chain);
629         ksm_stable_node_chains--;
630 }
631 
632 static void remove_node_from_stable_tree(struct stable_node *stable_node)
633 {
634         struct rmap_item *rmap_item;
635 
636         /* check it's not STABLE_NODE_CHAIN or negative */
637         BUG_ON(stable_node->rmap_hlist_len < 0);
638 
639         hlist_for_each_entry(rmap_item, &stable_node->hlist, hlist) {
640                 if (rmap_item->hlist.next)
641                         ksm_pages_sharing--;
642                 else
643                         ksm_pages_shared--;
644                 VM_BUG_ON(stable_node->rmap_hlist_len <= 0);
645                 stable_node->rmap_hlist_len--;
646                 put_anon_vma(rmap_item->anon_vma);
647                 rmap_item->address &= PAGE_MASK;
648                 cond_resched();
649         }
650 
651         /*
652          * We need the second aligned pointer of the migrate_nodes
653          * list_head to stay clear from the rb_parent_color union
654          * (aligned and different than any node) and also different
655          * from &migrate_nodes. This will verify that future list.h changes
656          * don't break STABLE_NODE_DUP_HEAD. Only recent gcc can handle it.
657          */
658 #if defined(GCC_VERSION) && GCC_VERSION >= 40903
659         BUILD_BUG_ON(STABLE_NODE_DUP_HEAD <= &migrate_nodes);
660         BUILD_BUG_ON(STABLE_NODE_DUP_HEAD >= &migrate_nodes + 1);
661 #endif
662 
663         if (stable_node->head == &migrate_nodes)
664                 list_del(&stable_node->list);
665         else
666                 stable_node_dup_del(stable_node);
667         free_stable_node(stable_node);
668 }
669 
670 /*
671  * get_ksm_page: checks if the page indicated by the stable node
672  * is still its ksm page, despite having held no reference to it.
673  * In which case we can trust the content of the page, and it
674  * returns the gotten page; but if the page has now been zapped,
675  * remove the stale node from the stable tree and return NULL.
676  * But beware, the stable node's page might be being migrated.
677  *
678  * You would expect the stable_node to hold a reference to the ksm page.
679  * But if it increments the page's count, swapping out has to wait for
680  * ksmd to come around again before it can free the page, which may take
681  * seconds or even minutes: much too unresponsive.  So instead we use a
682  * "keyhole reference": access to the ksm page from the stable node peeps
683  * out through its keyhole to see if that page still holds the right key,
684  * pointing back to this stable node.  This relies on freeing a PageAnon
685  * page to reset its page->mapping to NULL, and relies on no other use of
686  * a page to put something that might look like our key in page->mapping.
687  * is on its way to being freed; but it is an anomaly to bear in mind.
688  */
689 static struct page *get_ksm_page(struct stable_node *stable_node, bool lock_it)
690 {
691         struct page *page;
692         void *expected_mapping;
693         unsigned long kpfn;
694 
695         expected_mapping = (void *)((unsigned long)stable_node |
696                                         PAGE_MAPPING_KSM);
697 again:
698         kpfn = READ_ONCE(stable_node->kpfn); /* Address dependency. */
699         page = pfn_to_page(kpfn);
700         if (READ_ONCE(page->mapping) != expected_mapping)
701                 goto stale;
702 
703         /*
704          * We cannot do anything with the page while its refcount is 0.
705          * Usually 0 means free, or tail of a higher-order page: in which
706          * case this node is no longer referenced, and should be freed;
707          * however, it might mean that the page is under page_ref_freeze().
708          * The __remove_mapping() case is easy, again the node is now stale;
709          * but if page is swapcache in migrate_page_move_mapping(), it might
710          * still be our page, in which case it's essential to keep the node.
711          */
712         while (!get_page_unless_zero(page)) {
713                 /*
714                  * Another check for page->mapping != expected_mapping would
715                  * work here too.  We have chosen the !PageSwapCache test to
716                  * optimize the common case, when the page is or is about to
717                  * be freed: PageSwapCache is cleared (under spin_lock_irq)
718                  * in the ref_freeze section of __remove_mapping(); but Anon
719                  * page->mapping reset to NULL later, in free_pages_prepare().
720                  */
721                 if (!PageSwapCache(page))
722                         goto stale;
723                 cpu_relax();
724         }
725 
726         if (READ_ONCE(page->mapping) != expected_mapping) {
727                 put_page(page);
728                 goto stale;
729         }
730 
731         if (lock_it) {
732                 lock_page(page);
733                 if (READ_ONCE(page->mapping) != expected_mapping) {
734                         unlock_page(page);
735                         put_page(page);
736                         goto stale;
737                 }
738         }
739         return page;
740 
741 stale:
742         /*
743          * We come here from above when page->mapping or !PageSwapCache
744          * suggests that the node is stale; but it might be under migration.
745          * We need smp_rmb(), matching the smp_wmb() in ksm_migrate_page(),
746          * before checking whether node->kpfn has been changed.
747          */
748         smp_rmb();
749         if (READ_ONCE(stable_node->kpfn) != kpfn)
750                 goto again;
751         remove_node_from_stable_tree(stable_node);
752         return NULL;
753 }
754 
755 /*
756  * Removing rmap_item from stable or unstable tree.
757  * This function will clean the information from the stable/unstable tree.
758  */
759 static void remove_rmap_item_from_tree(struct rmap_item *rmap_item)
760 {
761         if (rmap_item->address & STABLE_FLAG) {
762                 struct stable_node *stable_node;
763                 struct page *page;
764 
765                 stable_node = rmap_item->head;
766                 page = get_ksm_page(stable_node, true);
767                 if (!page)
768                         goto out;
769 
770                 hlist_del(&rmap_item->hlist);
771                 unlock_page(page);
772                 put_page(page);
773 
774                 if (!hlist_empty(&stable_node->hlist))
775                         ksm_pages_sharing--;
776                 else
777                         ksm_pages_shared--;
778                 VM_BUG_ON(stable_node->rmap_hlist_len <= 0);
779                 stable_node->rmap_hlist_len--;
780 
781                 put_anon_vma(rmap_item->anon_vma);
782                 rmap_item->address &= PAGE_MASK;
783 
784         } else if (rmap_item->address & UNSTABLE_FLAG) {
785                 unsigned char age;
786                 /*
787                  * Usually ksmd can and must skip the rb_erase, because
788                  * root_unstable_tree was already reset to RB_ROOT.
789                  * But be careful when an mm is exiting: do the rb_erase
790                  * if this rmap_item was inserted by this scan, rather
791                  * than left over from before.
792                  */
793                 age = (unsigned char)(ksm_scan.seqnr - rmap_item->address);
794                 BUG_ON(age > 1);
795                 if (!age)
796                         rb_erase(&rmap_item->node,
797                                  root_unstable_tree + NUMA(rmap_item->nid));
798                 ksm_pages_unshared--;
799                 rmap_item->address &= PAGE_MASK;
800         }
801 out:
802         cond_resched();         /* we're called from many long loops */
803 }
804 
805 static void remove_trailing_rmap_items(struct mm_slot *mm_slot,
806                                        struct rmap_item **rmap_list)
807 {
808         while (*rmap_list) {
809                 struct rmap_item *rmap_item = *rmap_list;
810                 *rmap_list = rmap_item->rmap_list;
811                 remove_rmap_item_from_tree(rmap_item);
812                 free_rmap_item(rmap_item);
813         }
814 }
815 
816 /*
817  * Though it's very tempting to unmerge rmap_items from stable tree rather
818  * than check every pte of a given vma, the locking doesn't quite work for
819  * that - an rmap_item is assigned to the stable tree after inserting ksm
820  * page and upping mmap_sem.  Nor does it fit with the way we skip dup'ing
821  * rmap_items from parent to child at fork time (so as not to waste time
822  * if exit comes before the next scan reaches it).
823  *
824  * Similarly, although we'd like to remove rmap_items (so updating counts
825  * and freeing memory) when unmerging an area, it's easier to leave that
826  * to the next pass of ksmd - consider, for example, how ksmd might be
827  * in cmp_and_merge_page on one of the rmap_items we would be removing.
828  */
829 static int unmerge_ksm_pages(struct vm_area_struct *vma,
830                              unsigned long start, unsigned long end)
831 {
832         unsigned long addr;
833         int err = 0;
834 
835         for (addr = start; addr < end && !err; addr += PAGE_SIZE) {
836                 if (ksm_test_exit(vma->vm_mm))
837                         break;
838                 if (signal_pending(current))
839                         err = -ERESTARTSYS;
840                 else
841                         err = break_ksm(vma, addr);
842         }
843         return err;
844 }
845 
846 static inline struct stable_node *page_stable_node(struct page *page)
847 {
848         return PageKsm(page) ? page_rmapping(page) : NULL;
849 }
850 
851 static inline void set_page_stable_node(struct page *page,
852                                         struct stable_node *stable_node)
853 {
854         page->mapping = (void *)((unsigned long)stable_node | PAGE_MAPPING_KSM);
855 }
856 
857 #ifdef CONFIG_SYSFS
858 /*
859  * Only called through the sysfs control interface:
860  */
861 static int remove_stable_node(struct stable_node *stable_node)
862 {
863         struct page *page;
864         int err;
865 
866         page = get_ksm_page(stable_node, true);
867         if (!page) {
868                 /*
869                  * get_ksm_page did remove_node_from_stable_tree itself.
870                  */
871                 return 0;
872         }
873 
874         if (WARN_ON_ONCE(page_mapped(page))) {
875                 /*
876                  * This should not happen: but if it does, just refuse to let
877                  * merge_across_nodes be switched - there is no need to panic.
878                  */
879                 err = -EBUSY;
880         } else {
881                 /*
882                  * The stable node did not yet appear stale to get_ksm_page(),
883                  * since that allows for an unmapped ksm page to be recognized
884                  * right up until it is freed; but the node is safe to remove.
885                  * This page might be in a pagevec waiting to be freed,
886                  * or it might be PageSwapCache (perhaps under writeback),
887                  * or it might have been removed from swapcache a moment ago.
888                  */
889                 set_page_stable_node(page, NULL);
890                 remove_node_from_stable_tree(stable_node);
891                 err = 0;
892         }
893 
894         unlock_page(page);
895         put_page(page);
896         return err;
897 }
898 
899 static int remove_stable_node_chain(struct stable_node *stable_node,
900                                     struct rb_root *root)
901 {
902         struct stable_node *dup;
903         struct hlist_node *hlist_safe;
904 
905         if (!is_stable_node_chain(stable_node)) {
906                 VM_BUG_ON(is_stable_node_dup(stable_node));
907                 if (remove_stable_node(stable_node))
908                         return true;
909                 else
910                         return false;
911         }
912 
913         hlist_for_each_entry_safe(dup, hlist_safe,
914                                   &stable_node->hlist, hlist_dup) {
915                 VM_BUG_ON(!is_stable_node_dup(dup));
916                 if (remove_stable_node(dup))
917                         return true;
918         }
919         BUG_ON(!hlist_empty(&stable_node->hlist));
920         free_stable_node_chain(stable_node, root);
921         return false;
922 }
923 
924 static int remove_all_stable_nodes(void)
925 {
926         struct stable_node *stable_node, *next;
927         int nid;
928         int err = 0;
929 
930         for (nid = 0; nid < ksm_nr_node_ids; nid++) {
931                 while (root_stable_tree[nid].rb_node) {
932                         stable_node = rb_entry(root_stable_tree[nid].rb_node,
933                                                 struct stable_node, node);
934                         if (remove_stable_node_chain(stable_node,
935                                                      root_stable_tree + nid)) {
936                                 err = -EBUSY;
937                                 break;  /* proceed to next nid */
938                         }
939                         cond_resched();
940                 }
941         }
942         list_for_each_entry_safe(stable_node, next, &migrate_nodes, list) {
943                 if (remove_stable_node(stable_node))
944                         err = -EBUSY;
945                 cond_resched();
946         }
947         return err;
948 }
949 
950 static int unmerge_and_remove_all_rmap_items(void)
951 {
952         struct mm_slot *mm_slot;
953         struct mm_struct *mm;
954         struct vm_area_struct *vma;
955         int err = 0;
956 
957         spin_lock(&ksm_mmlist_lock);
958         ksm_scan.mm_slot = list_entry(ksm_mm_head.mm_list.next,
959                                                 struct mm_slot, mm_list);
960         spin_unlock(&ksm_mmlist_lock);
961 
962         for (mm_slot = ksm_scan.mm_slot;
963                         mm_slot != &ksm_mm_head; mm_slot = ksm_scan.mm_slot) {
964                 mm = mm_slot->mm;
965                 down_read(&mm->mmap_sem);
966                 for (vma = mm->mmap; vma; vma = vma->vm_next) {
967                         if (ksm_test_exit(mm))
968                                 break;
969                         if (!(vma->vm_flags & VM_MERGEABLE) || !vma->anon_vma)
970                                 continue;
971                         err = unmerge_ksm_pages(vma,
972                                                 vma->vm_start, vma->vm_end);
973                         if (err)
974                                 goto error;
975                 }
976 
977                 remove_trailing_rmap_items(mm_slot, &mm_slot->rmap_list);
978                 up_read(&mm->mmap_sem);
979 
980                 spin_lock(&ksm_mmlist_lock);
981                 ksm_scan.mm_slot = list_entry(mm_slot->mm_list.next,
982                                                 struct mm_slot, mm_list);
983                 if (ksm_test_exit(mm)) {
984                         hash_del(&mm_slot->link);
985                         list_del(&mm_slot->mm_list);
986                         spin_unlock(&ksm_mmlist_lock);
987 
988                         free_mm_slot(mm_slot);
989                         clear_bit(MMF_VM_MERGEABLE, &mm->flags);
990                         mmdrop(mm);
991                 } else
992                         spin_unlock(&ksm_mmlist_lock);
993         }
994 
995         /* Clean up stable nodes, but don't worry if some are still busy */
996         remove_all_stable_nodes();
997         ksm_scan.seqnr = 0;
998         return 0;
999 
1000 error:
1001         up_read(&mm->mmap_sem);
1002         spin_lock(&ksm_mmlist_lock);
1003         ksm_scan.mm_slot = &ksm_mm_head;
1004         spin_unlock(&ksm_mmlist_lock);
1005         return err;
1006 }
1007 #endif /* CONFIG_SYSFS */
1008 
1009 static u32 calc_checksum(struct page *page)
1010 {
1011         u32 checksum;
1012         void *addr = kmap_atomic(page);
1013         checksum = xxhash(addr, PAGE_SIZE, 0);
1014         kunmap_atomic(addr);
1015         return checksum;
1016 }
1017 
1018 static int memcmp_pages(struct page *page1, struct page *page2)
1019 {
1020         char *addr1, *addr2;
1021         int ret;
1022 
1023         addr1 = kmap_atomic(page1);
1024         addr2 = kmap_atomic(page2);
1025         ret = memcmp(addr1, addr2, PAGE_SIZE);
1026         kunmap_atomic(addr2);
1027         kunmap_atomic(addr1);
1028         return ret;
1029 }
1030 
1031 static inline int pages_identical(struct page *page1, struct page *page2)
1032 {
1033         return !memcmp_pages(page1, page2);
1034 }
1035 
1036 static int write_protect_page(struct vm_area_struct *vma, struct page *page,
1037                               pte_t *orig_pte)
1038 {
1039         struct mm_struct *mm = vma->vm_mm;
1040         struct page_vma_mapped_walk pvmw = {
1041                 .page = page,
1042                 .vma = vma,
1043         };
1044         int swapped;
1045         int err = -EFAULT;
1046         struct mmu_notifier_range range;
1047 
1048         pvmw.address = page_address_in_vma(page, vma);
1049         if (pvmw.address == -EFAULT)
1050                 goto out;
1051 
1052         BUG_ON(PageTransCompound(page));
1053 
1054         mmu_notifier_range_init(&range, mm, pvmw.address,
1055                                 pvmw.address + PAGE_SIZE);
1056         mmu_notifier_invalidate_range_start(&range);
1057 
1058         if (!page_vma_mapped_walk(&pvmw))
1059                 goto out_mn;
1060         if (WARN_ONCE(!pvmw.pte, "Unexpected PMD mapping?"))
1061                 goto out_unlock;
1062 
1063         if (pte_write(*pvmw.pte) || pte_dirty(*pvmw.pte) ||
1064             (pte_protnone(*pvmw.pte) && pte_savedwrite(*pvmw.pte)) ||
1065                                                 mm_tlb_flush_pending(mm)) {
1066                 pte_t entry;
1067 
1068                 swapped = PageSwapCache(page);
1069                 flush_cache_page(vma, pvmw.address, page_to_pfn(page));
1070                 /*
1071                  * Ok this is tricky, when get_user_pages_fast() run it doesn't
1072                  * take any lock, therefore the check that we are going to make
1073                  * with the pagecount against the mapcount is racey and
1074                  * O_DIRECT can happen right after the check.
1075                  * So we clear the pte and flush the tlb before the check
1076                  * this assure us that no O_DIRECT can happen after the check
1077                  * or in the middle of the check.
1078                  *
1079                  * No need to notify as we are downgrading page table to read
1080                  * only not changing it to point to a new page.
1081                  *
1082                  * See Documentation/vm/mmu_notifier.rst
1083                  */
1084                 entry = ptep_clear_flush(vma, pvmw.address, pvmw.pte);
1085                 /*
1086                  * Check that no O_DIRECT or similar I/O is in progress on the
1087                  * page
1088                  */
1089                 if (page_mapcount(page) + 1 + swapped != page_count(page)) {
1090                         set_pte_at(mm, pvmw.address, pvmw.pte, entry);
1091                         goto out_unlock;
1092                 }
1093                 if (pte_dirty(entry))
1094                         set_page_dirty(page);
1095 
1096                 if (pte_protnone(entry))
1097                         entry = pte_mkclean(pte_clear_savedwrite(entry));
1098                 else
1099                         entry = pte_mkclean(pte_wrprotect(entry));
1100                 set_pte_at_notify(mm, pvmw.address, pvmw.pte, entry);
1101         }
1102         *orig_pte = *pvmw.pte;
1103         err = 0;
1104 
1105 out_unlock:
1106         page_vma_mapped_walk_done(&pvmw);
1107 out_mn:
1108         mmu_notifier_invalidate_range_end(&range);
1109 out:
1110         return err;
1111 }
1112 
1113 /**
1114  * replace_page - replace page in vma by new ksm page
1115  * @vma:      vma that holds the pte pointing to page
1116  * @page:     the page we are replacing by kpage
1117  * @kpage:    the ksm page we replace page by
1118  * @orig_pte: the original value of the pte
1119  *
1120  * Returns 0 on success, -EFAULT on failure.
1121  */
1122 static int replace_page(struct vm_area_struct *vma, struct page *page,
1123                         struct page *kpage, pte_t orig_pte)
1124 {
1125         struct mm_struct *mm = vma->vm_mm;
1126         pmd_t *pmd;
1127         pte_t *ptep;
1128         pte_t newpte;
1129         spinlock_t *ptl;
1130         unsigned long addr;
1131         int err = -EFAULT;
1132         struct mmu_notifier_range range;
1133 
1134         addr = page_address_in_vma(page, vma);
1135         if (addr == -EFAULT)
1136                 goto out;
1137 
1138         pmd = mm_find_pmd(mm, addr);
1139         if (!pmd)
1140                 goto out;
1141 
1142         mmu_notifier_range_init(&range, mm, addr, addr + PAGE_SIZE);
1143         mmu_notifier_invalidate_range_start(&range);
1144 
1145         ptep = pte_offset_map_lock(mm, pmd, addr, &ptl);
1146         if (!pte_same(*ptep, orig_pte)) {
1147                 pte_unmap_unlock(ptep, ptl);
1148                 goto out_mn;
1149         }
1150 
1151         /*
1152          * No need to check ksm_use_zero_pages here: we can only have a
1153          * zero_page here if ksm_use_zero_pages was enabled alreaady.
1154          */
1155         if (!is_zero_pfn(page_to_pfn(kpage))) {
1156                 get_page(kpage);
1157                 page_add_anon_rmap(kpage, vma, addr, false);
1158                 newpte = mk_pte(kpage, vma->vm_page_prot);
1159         } else {
1160                 newpte = pte_mkspecial(pfn_pte(page_to_pfn(kpage),
1161                                                vma->vm_page_prot));
1162                 /*
1163                  * We're replacing an anonymous page with a zero page, which is
1164                  * not anonymous. We need to do proper accounting otherwise we
1165                  * will get wrong values in /proc, and a BUG message in dmesg
1166                  * when tearing down the mm.
1167                  */
1168                 dec_mm_counter(mm, MM_ANONPAGES);
1169         }
1170 
1171         flush_cache_page(vma, addr, pte_pfn(*ptep));
1172         /*
1173          * No need to notify as we are replacing a read only page with another
1174          * read only page with the same content.
1175          *
1176          * See Documentation/vm/mmu_notifier.rst
1177          */
1178         ptep_clear_flush(vma, addr, ptep);
1179         set_pte_at_notify(mm, addr, ptep, newpte);
1180 
1181         page_remove_rmap(page, false);
1182         if (!page_mapped(page))
1183                 try_to_free_swap(page);
1184         put_page(page);
1185 
1186         pte_unmap_unlock(ptep, ptl);
1187         err = 0;
1188 out_mn:
1189         mmu_notifier_invalidate_range_end(&range);
1190 out:
1191         return err;
1192 }
1193 
1194 /*
1195  * try_to_merge_one_page - take two pages and merge them into one
1196  * @vma: the vma that holds the pte pointing to page
1197  * @page: the PageAnon page that we want to replace with kpage
1198  * @kpage: the PageKsm page that we want to map instead of page,
1199  *         or NULL the first time when we want to use page as kpage.
1200  *
1201  * This function returns 0 if the pages were merged, -EFAULT otherwise.
1202  */
1203 static int try_to_merge_one_page(struct vm_area_struct *vma,
1204                                  struct page *page, struct page *kpage)
1205 {
1206         pte_t orig_pte = __pte(0);
1207         int err = -EFAULT;
1208 
1209         if (page == kpage)                      /* ksm page forked */
1210                 return 0;
1211 
1212         if (!PageAnon(page))
1213                 goto out;
1214 
1215         /*
1216          * We need the page lock to read a stable PageSwapCache in
1217          * write_protect_page().  We use trylock_page() instead of
1218          * lock_page() because we don't want to wait here - we
1219          * prefer to continue scanning and merging different pages,
1220          * then come back to this page when it is unlocked.
1221          */
1222         if (!trylock_page(page))
1223                 goto out;
1224 
1225         if (PageTransCompound(page)) {
1226                 if (split_huge_page(page))
1227                         goto out_unlock;
1228         }
1229 
1230         /*
1231          * If this anonymous page is mapped only here, its pte may need
1232          * to be write-protected.  If it's mapped elsewhere, all of its
1233          * ptes are necessarily already write-protected.  But in either
1234          * case, we need to lock and check page_count is not raised.
1235          */
1236         if (write_protect_page(vma, page, &orig_pte) == 0) {
1237                 if (!kpage) {
1238                         /*
1239                          * While we hold page lock, upgrade page from
1240                          * PageAnon+anon_vma to PageKsm+NULL stable_node:
1241                          * stable_tree_insert() will update stable_node.
1242                          */
1243                         set_page_stable_node(page, NULL);
1244                         mark_page_accessed(page);
1245                         /*
1246                          * Page reclaim just frees a clean page with no dirty
1247                          * ptes: make sure that the ksm page would be swapped.
1248                          */
1249                         if (!PageDirty(page))
1250                                 SetPageDirty(page);
1251                         err = 0;
1252                 } else if (pages_identical(page, kpage))
1253                         err = replace_page(vma, page, kpage, orig_pte);
1254         }
1255 
1256         if ((vma->vm_flags & VM_LOCKED) && kpage && !err) {
1257                 munlock_vma_page(page);
1258                 if (!PageMlocked(kpage)) {
1259                         unlock_page(page);
1260                         lock_page(kpage);
1261                         mlock_vma_page(kpage);
1262                         page = kpage;           /* for final unlock */
1263                 }
1264         }
1265 
1266 out_unlock:
1267         unlock_page(page);
1268 out:
1269         return err;
1270 }
1271 
1272 /*
1273  * try_to_merge_with_ksm_page - like try_to_merge_two_pages,
1274  * but no new kernel page is allocated: kpage must already be a ksm page.
1275  *
1276  * This function returns 0 if the pages were merged, -EFAULT otherwise.
1277  */
1278 static int try_to_merge_with_ksm_page(struct rmap_item *rmap_item,
1279                                       struct page *page, struct page *kpage)
1280 {
1281         struct mm_struct *mm = rmap_item->mm;
1282         struct vm_area_struct *vma;
1283         int err = -EFAULT;
1284 
1285         down_read(&mm->mmap_sem);
1286         vma = find_mergeable_vma(mm, rmap_item->address);
1287         if (!vma)
1288                 goto out;
1289 
1290         err = try_to_merge_one_page(vma, page, kpage);
1291         if (err)
1292                 goto out;
1293 
1294         /* Unstable nid is in union with stable anon_vma: remove first */
1295         remove_rmap_item_from_tree(rmap_item);
1296 
1297         /* Must get reference to anon_vma while still holding mmap_sem */
1298         rmap_item->anon_vma = vma->anon_vma;
1299         get_anon_vma(vma->anon_vma);
1300 out:
1301         up_read(&mm->mmap_sem);
1302         return err;
1303 }
1304 
1305 /*
1306  * try_to_merge_two_pages - take two identical pages and prepare them
1307  * to be merged into one page.
1308  *
1309  * This function returns the kpage if we successfully merged two identical
1310  * pages into one ksm page, NULL otherwise.
1311  *
1312  * Note that this function upgrades page to ksm page: if one of the pages
1313  * is already a ksm page, try_to_merge_with_ksm_page should be used.
1314  */
1315 static struct page *try_to_merge_two_pages(struct rmap_item *rmap_item,
1316                                            struct page *page,
1317                                            struct rmap_item *tree_rmap_item,
1318                                            struct page *tree_page)
1319 {
1320         int err;
1321 
1322         err = try_to_merge_with_ksm_page(rmap_item, page, NULL);
1323         if (!err) {
1324                 err = try_to_merge_with_ksm_page(tree_rmap_item,
1325                                                         tree_page, page);
1326                 /*
1327                  * If that fails, we have a ksm page with only one pte
1328                  * pointing to it: so break it.
1329                  */
1330                 if (err)
1331                         break_cow(rmap_item);
1332         }
1333         return err ? NULL : page;
1334 }
1335 
1336 static __always_inline
1337 bool __is_page_sharing_candidate(struct stable_node *stable_node, int offset)
1338 {
1339         VM_BUG_ON(stable_node->rmap_hlist_len < 0);
1340         /*
1341          * Check that at least one mapping still exists, otherwise
1342          * there's no much point to merge and share with this
1343          * stable_node, as the underlying tree_page of the other
1344          * sharer is going to be freed soon.
1345          */
1346         return stable_node->rmap_hlist_len &&
1347                 stable_node->rmap_hlist_len + offset < ksm_max_page_sharing;
1348 }
1349 
1350 static __always_inline
1351 bool is_page_sharing_candidate(struct stable_node *stable_node)
1352 {
1353         return __is_page_sharing_candidate(stable_node, 0);
1354 }
1355 
1356 static struct page *stable_node_dup(struct stable_node **_stable_node_dup,
1357                                     struct stable_node **_stable_node,
1358                                     struct rb_root *root,
1359                                     bool prune_stale_stable_nodes)
1360 {
1361         struct stable_node *dup, *found = NULL, *stable_node = *_stable_node;
1362         struct hlist_node *hlist_safe;
1363         struct page *_tree_page, *tree_page = NULL;
1364         int nr = 0;
1365         int found_rmap_hlist_len;
1366 
1367         if (!prune_stale_stable_nodes ||
1368             time_before(jiffies, stable_node->chain_prune_time +
1369                         msecs_to_jiffies(
1370                                 ksm_stable_node_chains_prune_millisecs)))
1371                 prune_stale_stable_nodes = false;
1372         else
1373                 stable_node->chain_prune_time = jiffies;
1374 
1375         hlist_for_each_entry_safe(dup, hlist_safe,
1376                                   &stable_node->hlist, hlist_dup) {
1377                 cond_resched();
1378                 /*
1379                  * We must walk all stable_node_dup to prune the stale
1380                  * stable nodes during lookup.
1381                  *
1382                  * get_ksm_page can drop the nodes from the
1383                  * stable_node->hlist if they point to freed pages
1384                  * (that's why we do a _safe walk). The "dup"
1385                  * stable_node parameter itself will be freed from
1386                  * under us if it returns NULL.
1387                  */
1388                 _tree_page = get_ksm_page(dup, false);
1389                 if (!_tree_page)
1390                         continue;
1391                 nr += 1;
1392                 if (is_page_sharing_candidate(dup)) {
1393                         if (!found ||
1394                             dup->rmap_hlist_len > found_rmap_hlist_len) {
1395                                 if (found)
1396                                         put_page(tree_page);
1397                                 found = dup;
1398                                 found_rmap_hlist_len = found->rmap_hlist_len;
1399                                 tree_page = _tree_page;
1400 
1401                                 /* skip put_page for found dup */
1402                                 if (!prune_stale_stable_nodes)
1403                                         break;
1404                                 continue;
1405                         }
1406                 }
1407                 put_page(_tree_page);
1408         }
1409 
1410         if (found) {
1411                 /*
1412                  * nr is counting all dups in the chain only if
1413                  * prune_stale_stable_nodes is true, otherwise we may
1414                  * break the loop at nr == 1 even if there are
1415                  * multiple entries.
1416                  */
1417                 if (prune_stale_stable_nodes && nr == 1) {
1418                         /*
1419                          * If there's not just one entry it would
1420                          * corrupt memory, better BUG_ON. In KSM
1421                          * context with no lock held it's not even
1422                          * fatal.
1423                          */
1424                         BUG_ON(stable_node->hlist.first->next);
1425 
1426                         /*
1427                          * There's just one entry and it is below the
1428                          * deduplication limit so drop the chain.
1429                          */
1430                         rb_replace_node(&stable_node->node, &found->node,
1431                                         root);
1432                         free_stable_node(stable_node);
1433                         ksm_stable_node_chains--;
1434                         ksm_stable_node_dups--;
1435                         /*
1436                          * NOTE: the caller depends on the stable_node
1437                          * to be equal to stable_node_dup if the chain
1438                          * was collapsed.
1439                          */
1440                         *_stable_node = found;
1441                         /*
1442                          * Just for robustneess as stable_node is
1443                          * otherwise left as a stable pointer, the
1444                          * compiler shall optimize it away at build
1445                          * time.
1446                          */
1447                         stable_node = NULL;
1448                 } else if (stable_node->hlist.first != &found->hlist_dup &&
1449                            __is_page_sharing_candidate(found, 1)) {
1450                         /*
1451                          * If the found stable_node dup can accept one
1452                          * more future merge (in addition to the one
1453                          * that is underway) and is not at the head of
1454                          * the chain, put it there so next search will
1455                          * be quicker in the !prune_stale_stable_nodes
1456                          * case.
1457                          *
1458                          * NOTE: it would be inaccurate to use nr > 1
1459                          * instead of checking the hlist.first pointer
1460                          * directly, because in the
1461                          * prune_stale_stable_nodes case "nr" isn't
1462                          * the position of the found dup in the chain,
1463                          * but the total number of dups in the chain.
1464                          */
1465                         hlist_del(&found->hlist_dup);
1466                         hlist_add_head(&found->hlist_dup,
1467                                        &stable_node->hlist);
1468                 }
1469         }
1470 
1471         *_stable_node_dup = found;
1472         return tree_page;
1473 }
1474 
1475 static struct stable_node *stable_node_dup_any(struct stable_node *stable_node,
1476                                                struct rb_root *root)
1477 {
1478         if (!is_stable_node_chain(stable_node))
1479                 return stable_node;
1480         if (hlist_empty(&stable_node->hlist)) {
1481                 free_stable_node_chain(stable_node, root);
1482                 return NULL;
1483         }
1484         return hlist_entry(stable_node->hlist.first,
1485                            typeof(*stable_node), hlist_dup);
1486 }
1487 
1488 /*
1489  * Like for get_ksm_page, this function can free the *_stable_node and
1490  * *_stable_node_dup if the returned tree_page is NULL.
1491  *
1492  * It can also free and overwrite *_stable_node with the found
1493  * stable_node_dup if the chain is collapsed (in which case
1494  * *_stable_node will be equal to *_stable_node_dup like if the chain
1495  * never existed). It's up to the caller to verify tree_page is not
1496  * NULL before dereferencing *_stable_node or *_stable_node_dup.
1497  *
1498  * *_stable_node_dup is really a second output parameter of this
1499  * function and will be overwritten in all cases, the caller doesn't
1500  * need to initialize it.
1501  */
1502 static struct page *__stable_node_chain(struct stable_node **_stable_node_dup,
1503                                         struct stable_node **_stable_node,
1504                                         struct rb_root *root,
1505                                         bool prune_stale_stable_nodes)
1506 {
1507         struct stable_node *stable_node = *_stable_node;
1508         if (!is_stable_node_chain(stable_node)) {
1509                 if (is_page_sharing_candidate(stable_node)) {
1510                         *_stable_node_dup = stable_node;
1511                         return get_ksm_page(stable_node, false);
1512                 }
1513                 /*
1514                  * _stable_node_dup set to NULL means the stable_node
1515                  * reached the ksm_max_page_sharing limit.
1516                  */
1517                 *_stable_node_dup = NULL;
1518                 return NULL;
1519         }
1520         return stable_node_dup(_stable_node_dup, _stable_node, root,
1521                                prune_stale_stable_nodes);
1522 }
1523 
1524 static __always_inline struct page *chain_prune(struct stable_node **s_n_d,
1525                                                 struct stable_node **s_n,
1526                                                 struct rb_root *root)
1527 {
1528         return __stable_node_chain(s_n_d, s_n, root, true);
1529 }
1530 
1531 static __always_inline struct page *chain(struct stable_node **s_n_d,
1532                                           struct stable_node *s_n,
1533                                           struct rb_root *root)
1534 {
1535         struct stable_node *old_stable_node = s_n;
1536         struct page *tree_page;
1537 
1538         tree_page = __stable_node_chain(s_n_d, &s_n, root, false);
1539         /* not pruning dups so s_n cannot have changed */
1540         VM_BUG_ON(s_n != old_stable_node);
1541         return tree_page;
1542 }
1543 
1544 /*
1545  * stable_tree_search - search for page inside the stable tree
1546  *
1547  * This function checks if there is a page inside the stable tree
1548  * with identical content to the page that we are scanning right now.
1549  *
1550  * This function returns the stable tree node of identical content if found,
1551  * NULL otherwise.
1552  */
1553 static struct page *stable_tree_search(struct page *page)
1554 {
1555         int nid;
1556         struct rb_root *root;
1557         struct rb_node **new;
1558         struct rb_node *parent;
1559         struct stable_node *stable_node, *stable_node_dup, *stable_node_any;
1560         struct stable_node *page_node;
1561 
1562         page_node = page_stable_node(page);
1563         if (page_node && page_node->head != &migrate_nodes) {
1564                 /* ksm page forked */
1565                 get_page(page);
1566                 return page;
1567         }
1568 
1569         nid = get_kpfn_nid(page_to_pfn(page));
1570         root = root_stable_tree + nid;
1571 again:
1572         new = &root->rb_node;
1573         parent = NULL;
1574 
1575         while (*new) {
1576                 struct page *tree_page;
1577                 int ret;
1578 
1579                 cond_resched();
1580                 stable_node = rb_entry(*new, struct stable_node, node);
1581                 stable_node_any = NULL;
1582                 tree_page = chain_prune(&stable_node_dup, &stable_node, root);
1583                 /*
1584                  * NOTE: stable_node may have been freed by
1585                  * chain_prune() if the returned stable_node_dup is
1586                  * not NULL. stable_node_dup may have been inserted in
1587                  * the rbtree instead as a regular stable_node (in
1588                  * order to collapse the stable_node chain if a single
1589                  * stable_node dup was found in it). In such case the
1590                  * stable_node is overwritten by the calleee to point
1591                  * to the stable_node_dup that was collapsed in the
1592                  * stable rbtree and stable_node will be equal to
1593                  * stable_node_dup like if the chain never existed.
1594                  */
1595                 if (!stable_node_dup) {
1596                         /*
1597                          * Either all stable_node dups were full in
1598                          * this stable_node chain, or this chain was
1599                          * empty and should be rb_erased.
1600                          */
1601                         stable_node_any = stable_node_dup_any(stable_node,
1602                                                               root);
1603                         if (!stable_node_any) {
1604                                 /* rb_erase just run */
1605                                 goto again;
1606                         }
1607                         /*
1608                          * Take any of the stable_node dups page of
1609                          * this stable_node chain to let the tree walk
1610                          * continue. All KSM pages belonging to the
1611                          * stable_node dups in a stable_node chain
1612                          * have the same content and they're
1613                          * wrprotected at all times. Any will work
1614                          * fine to continue the walk.
1615                          */
1616                         tree_page = get_ksm_page(stable_node_any, false);
1617                 }
1618                 VM_BUG_ON(!stable_node_dup ^ !!stable_node_any);
1619                 if (!tree_page) {
1620                         /*
1621                          * If we walked over a stale stable_node,
1622                          * get_ksm_page() will call rb_erase() and it
1623                          * may rebalance the tree from under us. So
1624                          * restart the search from scratch. Returning
1625                          * NULL would be safe too, but we'd generate
1626                          * false negative insertions just because some
1627                          * stable_node was stale.
1628                          */
1629                         goto again;
1630                 }
1631 
1632                 ret = memcmp_pages(page, tree_page);
1633                 put_page(tree_page);
1634 
1635                 parent = *new;
1636                 if (ret < 0)
1637                         new = &parent->rb_left;
1638                 else if (ret > 0)
1639                         new = &parent->rb_right;
1640                 else {
1641                         if (page_node) {
1642                                 VM_BUG_ON(page_node->head != &migrate_nodes);
1643                                 /*
1644                                  * Test if the migrated page should be merged
1645                                  * into a stable node dup. If the mapcount is
1646                                  * 1 we can migrate it with another KSM page
1647                                  * without adding it to the chain.
1648                                  */
1649                                 if (page_mapcount(page) > 1)
1650                                         goto chain_append;
1651                         }
1652 
1653                         if (!stable_node_dup) {
1654                                 /*
1655                                  * If the stable_node is a chain and
1656                                  * we got a payload match in memcmp
1657                                  * but we cannot merge the scanned
1658                                  * page in any of the existing
1659                                  * stable_node dups because they're
1660                                  * all full, we need to wait the
1661                                  * scanned page to find itself a match
1662                                  * in the unstable tree to create a
1663                                  * brand new KSM page to add later to
1664                                  * the dups of this stable_node.
1665                                  */
1666                                 return NULL;
1667                         }
1668 
1669                         /*
1670                          * Lock and unlock the stable_node's page (which
1671                          * might already have been migrated) so that page
1672                          * migration is sure to notice its raised count.
1673                          * It would be more elegant to return stable_node
1674                          * than kpage, but that involves more changes.
1675                          */
1676                         tree_page = get_ksm_page(stable_node_dup, true);
1677                         if (unlikely(!tree_page))
1678                                 /*
1679                                  * The tree may have been rebalanced,
1680                                  * so re-evaluate parent and new.
1681                                  */
1682                                 goto again;
1683                         unlock_page(tree_page);
1684 
1685                         if (get_kpfn_nid(stable_node_dup->kpfn) !=
1686                             NUMA(stable_node_dup->nid)) {
1687                                 put_page(tree_page);
1688                                 goto replace;
1689                         }
1690                         return tree_page;
1691                 }
1692         }
1693 
1694         if (!page_node)
1695                 return NULL;
1696 
1697         list_del(&page_node->list);
1698         DO_NUMA(page_node->nid = nid);
1699         rb_link_node(&page_node->node, parent, new);
1700         rb_insert_color(&page_node->node, root);
1701 out:
1702         if (is_page_sharing_candidate(page_node)) {
1703                 get_page(page);
1704                 return page;
1705         } else
1706                 return NULL;
1707 
1708 replace:
1709         /*
1710          * If stable_node was a chain and chain_prune collapsed it,
1711          * stable_node has been updated to be the new regular
1712          * stable_node. A collapse of the chain is indistinguishable
1713          * from the case there was no chain in the stable
1714          * rbtree. Otherwise stable_node is the chain and
1715          * stable_node_dup is the dup to replace.
1716          */
1717         if (stable_node_dup == stable_node) {
1718                 VM_BUG_ON(is_stable_node_chain(stable_node_dup));
1719                 VM_BUG_ON(is_stable_node_dup(stable_node_dup));
1720                 /* there is no chain */
1721                 if (page_node) {
1722                         VM_BUG_ON(page_node->head != &migrate_nodes);
1723                         list_del(&page_node->list);
1724                         DO_NUMA(page_node->nid = nid);
1725                         rb_replace_node(&stable_node_dup->node,
1726                                         &page_node->node,
1727                                         root);
1728                         if (is_page_sharing_candidate(page_node))
1729                                 get_page(page);
1730                         else
1731                                 page = NULL;
1732                 } else {
1733                         rb_erase(&stable_node_dup->node, root);
1734                         page = NULL;
1735                 }
1736         } else {
1737                 VM_BUG_ON(!is_stable_node_chain(stable_node));
1738                 __stable_node_dup_del(stable_node_dup);
1739                 if (page_node) {
1740                         VM_BUG_ON(page_node->head != &migrate_nodes);
1741                         list_del(&page_node->list);
1742                         DO_NUMA(page_node->nid = nid);
1743                         stable_node_chain_add_dup(page_node, stable_node);
1744                         if (is_page_sharing_candidate(page_node))
1745                                 get_page(page);
1746                         else
1747                                 page = NULL;
1748                 } else {
1749                         page = NULL;
1750                 }
1751         }
1752         stable_node_dup->head = &migrate_nodes;
1753         list_add(&stable_node_dup->list, stable_node_dup->head);
1754         return page;
1755 
1756 chain_append:
1757         /* stable_node_dup could be null if it reached the limit */
1758         if (!stable_node_dup)
1759                 stable_node_dup = stable_node_any;
1760         /*
1761          * If stable_node was a chain and chain_prune collapsed it,
1762          * stable_node has been updated to be the new regular
1763          * stable_node. A collapse of the chain is indistinguishable
1764          * from the case there was no chain in the stable
1765          * rbtree. Otherwise stable_node is the chain and
1766          * stable_node_dup is the dup to replace.
1767          */
1768         if (stable_node_dup == stable_node) {
1769                 VM_BUG_ON(is_stable_node_chain(stable_node_dup));
1770                 VM_BUG_ON(is_stable_node_dup(stable_node_dup));
1771                 /* chain is missing so create it */
1772                 stable_node = alloc_stable_node_chain(stable_node_dup,
1773                                                       root);
1774                 if (!stable_node)
1775                         return NULL;
1776         }
1777         /*
1778          * Add this stable_node dup that was
1779          * migrated to the stable_node chain
1780          * of the current nid for this page
1781          * content.
1782          */
1783         VM_BUG_ON(!is_stable_node_chain(stable_node));
1784         VM_BUG_ON(!is_stable_node_dup(stable_node_dup));
1785         VM_BUG_ON(page_node->head != &migrate_nodes);
1786         list_del(&page_node->list);
1787         DO_NUMA(page_node->nid = nid);
1788         stable_node_chain_add_dup(page_node, stable_node);
1789         goto out;
1790 }
1791 
1792 /*
1793  * stable_tree_insert - insert stable tree node pointing to new ksm page
1794  * into the stable tree.
1795  *
1796  * This function returns the stable tree node just allocated on success,
1797  * NULL otherwise.
1798  */
1799 static struct stable_node *stable_tree_insert(struct page *kpage)
1800 {
1801         int nid;
1802         unsigned long kpfn;
1803         struct rb_root *root;
1804         struct rb_node **new;
1805         struct rb_node *parent;
1806         struct stable_node *stable_node, *stable_node_dup, *stable_node_any;
1807         bool need_chain = false;
1808 
1809         kpfn = page_to_pfn(kpage);
1810         nid = get_kpfn_nid(kpfn);
1811         root = root_stable_tree + nid;
1812 again:
1813         parent = NULL;
1814         new = &root->rb_node;
1815 
1816         while (*new) {
1817                 struct page *tree_page;
1818                 int ret;
1819 
1820                 cond_resched();
1821                 stable_node = rb_entry(*new, struct stable_node, node);
1822                 stable_node_any = NULL;
1823                 tree_page = chain(&stable_node_dup, stable_node, root);
1824                 if (!stable_node_dup) {
1825                         /*
1826                          * Either all stable_node dups were full in
1827                          * this stable_node chain, or this chain was
1828                          * empty and should be rb_erased.
1829                          */
1830                         stable_node_any = stable_node_dup_any(stable_node,
1831                                                               root);
1832                         if (!stable_node_any) {
1833                                 /* rb_erase just run */
1834                                 goto again;
1835                         }
1836                         /*
1837                          * Take any of the stable_node dups page of
1838                          * this stable_node chain to let the tree walk
1839                          * continue. All KSM pages belonging to the
1840                          * stable_node dups in a stable_node chain
1841                          * have the same content and they're
1842                          * wrprotected at all times. Any will work
1843                          * fine to continue the walk.
1844                          */
1845                         tree_page = get_ksm_page(stable_node_any, false);
1846                 }
1847                 VM_BUG_ON(!stable_node_dup ^ !!stable_node_any);
1848                 if (!tree_page) {
1849                         /*
1850                          * If we walked over a stale stable_node,
1851                          * get_ksm_page() will call rb_erase() and it
1852                          * may rebalance the tree from under us. So
1853                          * restart the search from scratch. Returning
1854                          * NULL would be safe too, but we'd generate
1855                          * false negative insertions just because some
1856                          * stable_node was stale.
1857                          */
1858                         goto again;
1859                 }
1860 
1861                 ret = memcmp_pages(kpage, tree_page);
1862                 put_page(tree_page);
1863 
1864                 parent = *new;
1865                 if (ret < 0)
1866                         new = &parent->rb_left;
1867                 else if (ret > 0)
1868                         new = &parent->rb_right;
1869                 else {
1870                         need_chain = true;
1871                         break;
1872                 }
1873         }
1874 
1875         stable_node_dup = alloc_stable_node();
1876         if (!stable_node_dup)
1877                 return NULL;
1878 
1879         INIT_HLIST_HEAD(&stable_node_dup->hlist);
1880         stable_node_dup->kpfn = kpfn;
1881         set_page_stable_node(kpage, stable_node_dup);
1882         stable_node_dup->rmap_hlist_len = 0;
1883         DO_NUMA(stable_node_dup->nid = nid);
1884         if (!need_chain) {
1885                 rb_link_node(&stable_node_dup->node, parent, new);
1886                 rb_insert_color(&stable_node_dup->node, root);
1887         } else {
1888                 if (!is_stable_node_chain(stable_node)) {
1889                         struct stable_node *orig = stable_node;
1890                         /* chain is missing so create it */
1891                         stable_node = alloc_stable_node_chain(orig, root);
1892                         if (!stable_node) {
1893                                 free_stable_node(stable_node_dup);
1894                                 return NULL;
1895                         }
1896                 }
1897                 stable_node_chain_add_dup(stable_node_dup, stable_node);
1898         }
1899 
1900         return stable_node_dup;
1901 }
1902 
1903 /*
1904  * unstable_tree_search_insert - search for identical page,
1905  * else insert rmap_item into the unstable tree.
1906  *
1907  * This function searches for a page in the unstable tree identical to the
1908  * page currently being scanned; and if no identical page is found in the
1909  * tree, we insert rmap_item as a new object into the unstable tree.
1910  *
1911  * This function returns pointer to rmap_item found to be identical
1912  * to the currently scanned page, NULL otherwise.
1913  *
1914  * This function does both searching and inserting, because they share
1915  * the same walking algorithm in an rbtree.
1916  */
1917 static
1918 struct rmap_item *unstable_tree_search_insert(struct rmap_item *rmap_item,
1919                                               struct page *page,
1920                                               struct page **tree_pagep)
1921 {
1922         struct rb_node **new;
1923         struct rb_root *root;
1924         struct rb_node *parent = NULL;
1925         int nid;
1926 
1927         nid = get_kpfn_nid(page_to_pfn(page));
1928         root = root_unstable_tree + nid;
1929         new = &root->rb_node;
1930 
1931         while (*new) {
1932                 struct rmap_item *tree_rmap_item;
1933                 struct page *tree_page;
1934                 int ret;
1935 
1936                 cond_resched();
1937                 tree_rmap_item = rb_entry(*new, struct rmap_item, node);
1938                 tree_page = get_mergeable_page(tree_rmap_item);
1939                 if (!tree_page)
1940                         return NULL;
1941 
1942                 /*
1943                  * Don't substitute a ksm page for a forked page.
1944                  */
1945                 if (page == tree_page) {
1946                         put_page(tree_page);
1947                         return NULL;
1948                 }
1949 
1950                 ret = memcmp_pages(page, tree_page);
1951 
1952                 parent = *new;
1953                 if (ret < 0) {
1954                         put_page(tree_page);
1955                         new = &parent->rb_left;
1956                 } else if (ret > 0) {
1957                         put_page(tree_page);
1958                         new = &parent->rb_right;
1959                 } else if (!ksm_merge_across_nodes &&
1960                            page_to_nid(tree_page) != nid) {
1961                         /*
1962                          * If tree_page has been migrated to another NUMA node,
1963                          * it will be flushed out and put in the right unstable
1964                          * tree next time: only merge with it when across_nodes.
1965                          */
1966                         put_page(tree_page);
1967                         return NULL;
1968                 } else {
1969                         *tree_pagep = tree_page;
1970                         return tree_rmap_item;
1971                 }
1972         }
1973 
1974         rmap_item->address |= UNSTABLE_FLAG;
1975         rmap_item->address |= (ksm_scan.seqnr & SEQNR_MASK);
1976         DO_NUMA(rmap_item->nid = nid);
1977         rb_link_node(&rmap_item->node, parent, new);
1978         rb_insert_color(&rmap_item->node, root);
1979 
1980         ksm_pages_unshared++;
1981         return NULL;
1982 }
1983 
1984 /*
1985  * stable_tree_append - add another rmap_item to the linked list of
1986  * rmap_items hanging off a given node of the stable tree, all sharing
1987  * the same ksm page.
1988  */
1989 static void stable_tree_append(struct rmap_item *rmap_item,
1990                                struct stable_node *stable_node,
1991                                bool max_page_sharing_bypass)
1992 {
1993         /*
1994          * rmap won't find this mapping if we don't insert the
1995          * rmap_item in the right stable_node
1996          * duplicate. page_migration could break later if rmap breaks,
1997          * so we can as well crash here. We really need to check for
1998          * rmap_hlist_len == STABLE_NODE_CHAIN, but we can as well check
1999          * for other negative values as an undeflow if detected here
2000          * for the first time (and not when decreasing rmap_hlist_len)
2001          * would be sign of memory corruption in the stable_node.
2002          */
2003         BUG_ON(stable_node->rmap_hlist_len < 0);
2004 
2005         stable_node->rmap_hlist_len++;
2006         if (!max_page_sharing_bypass)
2007                 /* possibly non fatal but unexpected overflow, only warn */
2008                 WARN_ON_ONCE(stable_node->rmap_hlist_len >
2009                              ksm_max_page_sharing);
2010 
2011         rmap_item->head = stable_node;
2012         rmap_item->address |= STABLE_FLAG;
2013         hlist_add_head(&rmap_item->hlist, &stable_node->hlist);
2014 
2015         if (rmap_item->hlist.next)
2016                 ksm_pages_sharing++;
2017         else
2018                 ksm_pages_shared++;
2019 }
2020 
2021 /*
2022  * cmp_and_merge_page - first see if page can be merged into the stable tree;
2023  * if not, compare checksum to previous and if it's the same, see if page can
2024  * be inserted into the unstable tree, or merged with a page already there and
2025  * both transferred to the stable tree.
2026  *
2027  * @page: the page that we are searching identical page to.
2028  * @rmap_item: the reverse mapping into the virtual address of this page
2029  */
2030 static void cmp_and_merge_page(struct page *page, struct rmap_item *rmap_item)
2031 {
2032         struct mm_struct *mm = rmap_item->mm;
2033         struct rmap_item *tree_rmap_item;
2034         struct page *tree_page = NULL;
2035         struct stable_node *stable_node;
2036         struct page *kpage;
2037         unsigned int checksum;
2038         int err;
2039         bool max_page_sharing_bypass = false;
2040 
2041         stable_node = page_stable_node(page);
2042         if (stable_node) {
2043                 if (stable_node->head != &migrate_nodes &&
2044                     get_kpfn_nid(READ_ONCE(stable_node->kpfn)) !=
2045                     NUMA(stable_node->nid)) {
2046                         stable_node_dup_del(stable_node);
2047                         stable_node->head = &migrate_nodes;
2048                         list_add(&stable_node->list, stable_node->head);
2049                 }
2050                 if (stable_node->head != &migrate_nodes &&
2051                     rmap_item->head == stable_node)
2052                         return;
2053                 /*
2054                  * If it's a KSM fork, allow it to go over the sharing limit
2055                  * without warnings.
2056                  */
2057                 if (!is_page_sharing_candidate(stable_node))
2058                         max_page_sharing_bypass = true;
2059         }
2060 
2061         /* We first start with searching the page inside the stable tree */
2062         kpage = stable_tree_search(page);
2063         if (kpage == page && rmap_item->head == stable_node) {
2064                 put_page(kpage);
2065                 return;
2066         }
2067 
2068         remove_rmap_item_from_tree(rmap_item);
2069 
2070         if (kpage) {
2071                 err = try_to_merge_with_ksm_page(rmap_item, page, kpage);
2072                 if (!err) {
2073                         /*
2074                          * The page was successfully merged:
2075                          * add its rmap_item to the stable tree.
2076                          */
2077                         lock_page(kpage);
2078                         stable_tree_append(rmap_item, page_stable_node(kpage),
2079                                            max_page_sharing_bypass);
2080                         unlock_page(kpage);
2081                 }
2082                 put_page(kpage);
2083                 return;
2084         }
2085 
2086         /*
2087          * If the hash value of the page has changed from the last time
2088          * we calculated it, this page is changing frequently: therefore we
2089          * don't want to insert it in the unstable tree, and we don't want
2090          * to waste our time searching for something identical to it there.
2091          */
2092         checksum = calc_checksum(page);
2093         if (rmap_item->oldchecksum != checksum) {
2094                 rmap_item->oldchecksum = checksum;
2095                 return;
2096         }
2097 
2098         /*
2099          * Same checksum as an empty page. We attempt to merge it with the
2100          * appropriate zero page if the user enabled this via sysfs.
2101          */
2102         if (ksm_use_zero_pages && (checksum == zero_checksum)) {
2103                 struct vm_area_struct *vma;
2104 
2105                 down_read(&mm->mmap_sem);
2106                 vma = find_mergeable_vma(mm, rmap_item->address);
2107                 err = try_to_merge_one_page(vma, page,
2108                                             ZERO_PAGE(rmap_item->address));
2109                 up_read(&mm->mmap_sem);
2110                 /*
2111                  * In case of failure, the page was not really empty, so we
2112                  * need to continue. Otherwise we're done.
2113                  */
2114                 if (!err)
2115                         return;
2116         }
2117         tree_rmap_item =
2118                 unstable_tree_search_insert(rmap_item, page, &tree_page);
2119         if (tree_rmap_item) {
2120                 bool split;
2121 
2122                 kpage = try_to_merge_two_pages(rmap_item, page,
2123                                                 tree_rmap_item, tree_page);
2124                 /*
2125                  * If both pages we tried to merge belong to the same compound
2126                  * page, then we actually ended up increasing the reference
2127                  * count of the same compound page twice, and split_huge_page
2128                  * failed.
2129                  * Here we set a flag if that happened, and we use it later to
2130                  * try split_huge_page again. Since we call put_page right
2131                  * afterwards, the reference count will be correct and
2132                  * split_huge_page should succeed.
2133                  */
2134                 split = PageTransCompound(page)
2135                         && compound_head(page) == compound_head(tree_page);
2136                 put_page(tree_page);
2137                 if (kpage) {
2138                         /*
2139                          * The pages were successfully merged: insert new
2140                          * node in the stable tree and add both rmap_items.
2141                          */
2142                         lock_page(kpage);
2143                         stable_node = stable_tree_insert(kpage);
2144                         if (stable_node) {
2145                                 stable_tree_append(tree_rmap_item, stable_node,
2146                                                    false);
2147                                 stable_tree_append(rmap_item, stable_node,
2148                                                    false);
2149                         }
2150                         unlock_page(kpage);
2151 
2152                         /*
2153                          * If we fail to insert the page into the stable tree,
2154                          * we will have 2 virtual addresses that are pointing
2155                          * to a ksm page left outside the stable tree,
2156                          * in which case we need to break_cow on both.
2157                          */
2158                         if (!stable_node) {
2159                                 break_cow(tree_rmap_item);
2160                                 break_cow(rmap_item);
2161                         }
2162                 } else if (split) {
2163                         /*
2164                          * We are here if we tried to merge two pages and
2165                          * failed because they both belonged to the same
2166                          * compound page. We will split the page now, but no
2167                          * merging will take place.
2168                          * We do not want to add the cost of a full lock; if
2169                          * the page is locked, it is better to skip it and
2170                          * perhaps try again later.
2171                          */
2172                         if (!trylock_page(page))
2173                                 return;
2174                         split_huge_page(page);
2175                         unlock_page(page);
2176                 }
2177         }
2178 }
2179 
2180 static struct rmap_item *get_next_rmap_item(struct mm_slot *mm_slot,
2181                                             struct rmap_item **rmap_list,
2182                                             unsigned long addr)
2183 {
2184         struct rmap_item *rmap_item;
2185 
2186         while (*rmap_list) {
2187                 rmap_item = *rmap_list;
2188                 if ((rmap_item->address & PAGE_MASK) == addr)
2189                         return rmap_item;
2190                 if (rmap_item->address > addr)
2191                         break;
2192                 *rmap_list = rmap_item->rmap_list;
2193                 remove_rmap_item_from_tree(rmap_item);
2194                 free_rmap_item(rmap_item);
2195         }
2196 
2197         rmap_item = alloc_rmap_item();
2198         if (rmap_item) {
2199                 /* It has already been zeroed */
2200                 rmap_item->mm = mm_slot->mm;
2201                 rmap_item->address = addr;
2202                 rmap_item->rmap_list = *rmap_list;
2203                 *rmap_list = rmap_item;
2204         }
2205         return rmap_item;
2206 }
2207 
2208 static struct rmap_item *scan_get_next_rmap_item(struct page **page)
2209 {
2210         struct mm_struct *mm;
2211         struct mm_slot *slot;
2212         struct vm_area_struct *vma;
2213         struct rmap_item *rmap_item;
2214         int nid;
2215 
2216         if (list_empty(&ksm_mm_head.mm_list))
2217                 return NULL;
2218 
2219         slot = ksm_scan.mm_slot;
2220         if (slot == &ksm_mm_head) {
2221                 /*
2222                  * A number of pages can hang around indefinitely on per-cpu
2223                  * pagevecs, raised page count preventing write_protect_page
2224                  * from merging them.  Though it doesn't really matter much,
2225                  * it is puzzling to see some stuck in pages_volatile until
2226                  * other activity jostles them out, and they also prevented
2227                  * LTP's KSM test from succeeding deterministically; so drain
2228                  * them here (here rather than on entry to ksm_do_scan(),
2229                  * so we don't IPI too often when pages_to_scan is set low).
2230                  */
2231                 lru_add_drain_all();
2232 
2233                 /*
2234                  * Whereas stale stable_nodes on the stable_tree itself
2235                  * get pruned in the regular course of stable_tree_search(),
2236                  * those moved out to the migrate_nodes list can accumulate:
2237                  * so prune them once before each full scan.
2238                  */
2239                 if (!ksm_merge_across_nodes) {
2240                         struct stable_node *stable_node, *next;
2241                         struct page *page;
2242 
2243                         list_for_each_entry_safe(stable_node, next,
2244                                                  &migrate_nodes, list) {
2245                                 page = get_ksm_page(stable_node, false);
2246                                 if (page)
2247                                         put_page(page);
2248                                 cond_resched();
2249                         }
2250                 }
2251 
2252                 for (nid = 0; nid < ksm_nr_node_ids; nid++)
2253                         root_unstable_tree[nid] = RB_ROOT;
2254 
2255                 spin_lock(&ksm_mmlist_lock);
2256                 slot = list_entry(slot->mm_list.next, struct mm_slot, mm_list);
2257                 ksm_scan.mm_slot = slot;
2258                 spin_unlock(&ksm_mmlist_lock);
2259                 /*
2260                  * Although we tested list_empty() above, a racing __ksm_exit
2261                  * of the last mm on the list may have removed it since then.
2262                  */
2263                 if (slot == &ksm_mm_head)
2264                         return NULL;
2265 next_mm:
2266                 ksm_scan.address = 0;
2267                 ksm_scan.rmap_list = &slot->rmap_list;
2268         }
2269 
2270         mm = slot->mm;
2271         down_read(&mm->mmap_sem);
2272         if (ksm_test_exit(mm))
2273                 vma = NULL;
2274         else
2275                 vma = find_vma(mm, ksm_scan.address);
2276 
2277         for (; vma; vma = vma->vm_next) {
2278                 if (!(vma->vm_flags & VM_MERGEABLE))
2279                         continue;
2280                 if (ksm_scan.address < vma->vm_start)
2281                         ksm_scan.address = vma->vm_start;
2282                 if (!vma->anon_vma)
2283                         ksm_scan.address = vma->vm_end;
2284 
2285                 while (ksm_scan.address < vma->vm_end) {
2286                         if (ksm_test_exit(mm))
2287                                 break;
2288                         *page = follow_page(vma, ksm_scan.address, FOLL_GET);
2289                         if (IS_ERR_OR_NULL(*page)) {
2290                                 ksm_scan.address += PAGE_SIZE;
2291                                 cond_resched();
2292                                 continue;
2293                         }
2294                         if (PageAnon(*page)) {
2295                                 flush_anon_page(vma, *page, ksm_scan.address);
2296                                 flush_dcache_page(*page);
2297                                 rmap_item = get_next_rmap_item(slot,
2298                                         ksm_scan.rmap_list, ksm_scan.address);
2299                                 if (rmap_item) {
2300                                         ksm_scan.rmap_list =
2301                                                         &rmap_item->rmap_list;
2302                                         ksm_scan.address += PAGE_SIZE;
2303                                 } else
2304                                         put_page(*page);
2305                                 up_read(&mm->mmap_sem);
2306                                 return rmap_item;
2307                         }
2308                         put_page(*page);
2309                         ksm_scan.address += PAGE_SIZE;
2310                         cond_resched();
2311                 }
2312         }
2313 
2314         if (ksm_test_exit(mm)) {
2315                 ksm_scan.address = 0;
2316                 ksm_scan.rmap_list = &slot->rmap_list;
2317         }
2318         /*
2319          * Nuke all the rmap_items that are above this current rmap:
2320          * because there were no VM_MERGEABLE vmas with such addresses.
2321          */
2322         remove_trailing_rmap_items(slot, ksm_scan.rmap_list);
2323 
2324         spin_lock(&ksm_mmlist_lock);
2325         ksm_scan.mm_slot = list_entry(slot->mm_list.next,
2326                                                 struct mm_slot, mm_list);
2327         if (ksm_scan.address == 0) {
2328                 /*
2329                  * We've completed a full scan of all vmas, holding mmap_sem
2330                  * throughout, and found no VM_MERGEABLE: so do the same as
2331                  * __ksm_exit does to remove this mm from all our lists now.
2332                  * This applies either when cleaning up after __ksm_exit
2333                  * (but beware: we can reach here even before __ksm_exit),
2334                  * or when all VM_MERGEABLE areas have been unmapped (and
2335                  * mmap_sem then protects against race with MADV_MERGEABLE).
2336                  */
2337                 hash_del(&slot->link);
2338                 list_del(&slot->mm_list);
2339                 spin_unlock(&ksm_mmlist_lock);
2340 
2341                 free_mm_slot(slot);
2342                 clear_bit(MMF_VM_MERGEABLE, &mm->flags);
2343                 up_read(&mm->mmap_sem);
2344                 mmdrop(mm);
2345         } else {
2346                 up_read(&mm->mmap_sem);
2347                 /*
2348                  * up_read(&mm->mmap_sem) first because after
2349                  * spin_unlock(&ksm_mmlist_lock) run, the "mm" may
2350                  * already have been freed under us by __ksm_exit()
2351                  * because the "mm_slot" is still hashed and
2352                  * ksm_scan.mm_slot doesn't point to it anymore.
2353                  */
2354                 spin_unlock(&ksm_mmlist_lock);
2355         }
2356 
2357         /* Repeat until we've completed scanning the whole list */
2358         slot = ksm_scan.mm_slot;
2359         if (slot != &ksm_mm_head)
2360                 goto next_mm;
2361 
2362         ksm_scan.seqnr++;
2363         return NULL;
2364 }
2365 
2366 /**
2367  * ksm_do_scan  - the ksm scanner main worker function.
2368  * @scan_npages:  number of pages we want to scan before we return.
2369  */
2370 static void ksm_do_scan(unsigned int scan_npages)
2371 {
2372         struct rmap_item *rmap_item;
2373         struct page *uninitialized_var(page);
2374 
2375         while (scan_npages-- && likely(!freezing(current))) {
2376                 cond_resched();
2377                 rmap_item = scan_get_next_rmap_item(&page);
2378                 if (!rmap_item)
2379                         return;
2380                 cmp_and_merge_page(page, rmap_item);
2381                 put_page(page);
2382         }
2383 }
2384 
2385 static int ksmd_should_run(void)
2386 {
2387         return (ksm_run & KSM_RUN_MERGE) && !list_empty(&ksm_mm_head.mm_list);
2388 }
2389 
2390 static int ksm_scan_thread(void *nothing)
2391 {
2392         unsigned int sleep_ms;
2393 
2394         set_freezable();
2395         set_user_nice(current, 5);
2396 
2397         while (!kthread_should_stop()) {
2398                 mutex_lock(&ksm_thread_mutex);
2399                 wait_while_offlining();
2400                 if (ksmd_should_run())
2401                         ksm_do_scan(ksm_thread_pages_to_scan);
2402                 mutex_unlock(&ksm_thread_mutex);
2403 
2404                 try_to_freeze();
2405 
2406                 if (ksmd_should_run()) {
2407                         sleep_ms = READ_ONCE(ksm_thread_sleep_millisecs);
2408                         wait_event_interruptible_timeout(ksm_iter_wait,
2409                                 sleep_ms != READ_ONCE(ksm_thread_sleep_millisecs),
2410                                 msecs_to_jiffies(sleep_ms));
2411                 } else {
2412                         wait_event_freezable(ksm_thread_wait,
2413                                 ksmd_should_run() || kthread_should_stop());
2414                 }
2415         }
2416         return 0;
2417 }
2418 
2419 int ksm_madvise(struct vm_area_struct *vma, unsigned long start,
2420                 unsigned long end, int advice, unsigned long *vm_flags)
2421 {
2422         struct mm_struct *mm = vma->vm_mm;
2423         int err;
2424 
2425         switch (advice) {
2426         case MADV_MERGEABLE:
2427                 /*
2428                  * Be somewhat over-protective for now!
2429                  */
2430                 if (*vm_flags & (VM_MERGEABLE | VM_SHARED  | VM_MAYSHARE   |
2431                                  VM_PFNMAP    | VM_IO      | VM_DONTEXPAND |
2432                                  VM_HUGETLB | VM_MIXEDMAP))
2433                         return 0;               /* just ignore the advice */
2434 
2435                 if (vma_is_dax(vma))
2436                         return 0;
2437 
2438 #ifdef VM_SAO
2439                 if (*vm_flags & VM_SAO)
2440                         return 0;
2441 #endif
2442 #ifdef VM_SPARC_ADI
2443                 if (*vm_flags & VM_SPARC_ADI)
2444                         return 0;
2445 #endif
2446 
2447                 if (!test_bit(MMF_VM_MERGEABLE, &mm->flags)) {
2448                         err = __ksm_enter(mm);
2449                         if (err)
2450                                 return err;
2451                 }
2452 
2453                 *vm_flags |= VM_MERGEABLE;
2454                 break;
2455 
2456         case MADV_UNMERGEABLE:
2457                 if (!(*vm_flags & VM_MERGEABLE))
2458                         return 0;               /* just ignore the advice */
2459 
2460                 if (vma->anon_vma) {
2461                         err = unmerge_ksm_pages(vma, start, end);
2462                         if (err)
2463                                 return err;
2464                 }
2465 
2466                 *vm_flags &= ~VM_MERGEABLE;
2467                 break;
2468         }
2469 
2470         return 0;
2471 }
2472 
2473 int __ksm_enter(struct mm_struct *mm)
2474 {
2475         struct mm_slot *mm_slot;
2476         int needs_wakeup;
2477 
2478         mm_slot = alloc_mm_slot();
2479         if (!mm_slot)
2480                 return -ENOMEM;
2481 
2482         /* Check ksm_run too?  Would need tighter locking */
2483         needs_wakeup = list_empty(&ksm_mm_head.mm_list);
2484 
2485         spin_lock(&ksm_mmlist_lock);
2486         insert_to_mm_slots_hash(mm, mm_slot);
2487         /*
2488          * When KSM_RUN_MERGE (or KSM_RUN_STOP),
2489          * insert just behind the scanning cursor, to let the area settle
2490          * down a little; when fork is followed by immediate exec, we don't
2491          * want ksmd to waste time setting up and tearing down an rmap_list.
2492          *
2493          * But when KSM_RUN_UNMERGE, it's important to insert ahead of its
2494          * scanning cursor, otherwise KSM pages in newly forked mms will be
2495          * missed: then we might as well insert at the end of the list.
2496          */
2497         if (ksm_run & KSM_RUN_UNMERGE)
2498                 list_add_tail(&mm_slot->mm_list, &ksm_mm_head.mm_list);
2499         else
2500                 list_add_tail(&mm_slot->mm_list, &ksm_scan.mm_slot->mm_list);
2501         spin_unlock(&ksm_mmlist_lock);
2502 
2503         set_bit(MMF_VM_MERGEABLE, &mm->flags);
2504         mmgrab(mm);
2505 
2506         if (needs_wakeup)
2507                 wake_up_interruptible(&ksm_thread_wait);
2508 
2509         return 0;
2510 }
2511 
2512 void __ksm_exit(struct mm_struct *mm)
2513 {
2514         struct mm_slot *mm_slot;
2515         int easy_to_free = 0;
2516 
2517         /*
2518          * This process is exiting: if it's straightforward (as is the
2519          * case when ksmd was never running), free mm_slot immediately.
2520          * But if it's at the cursor or has rmap_items linked to it, use
2521          * mmap_sem to synchronize with any break_cows before pagetables
2522          * are freed, and leave the mm_slot on the list for ksmd to free.
2523          * Beware: ksm may already have noticed it exiting and freed the slot.
2524          */
2525 
2526         spin_lock(&ksm_mmlist_lock);
2527         mm_slot = get_mm_slot(mm);
2528         if (mm_slot && ksm_scan.mm_slot != mm_slot) {
2529                 if (!mm_slot->rmap_list) {
2530                         hash_del(&mm_slot->link);
2531                         list_del(&mm_slot->mm_list);
2532                         easy_to_free = 1;
2533                 } else {
2534                         list_move(&mm_slot->mm_list,
2535                                   &ksm_scan.mm_slot->mm_list);
2536                 }
2537         }
2538         spin_unlock(&ksm_mmlist_lock);
2539 
2540         if (easy_to_free) {
2541                 free_mm_slot(mm_slot);
2542                 clear_bit(MMF_VM_MERGEABLE, &mm->flags);
2543                 mmdrop(mm);
2544         } else if (mm_slot) {
2545                 down_write(&mm->mmap_sem);
2546                 up_write(&mm->mmap_sem);
2547         }
2548 }
2549 
2550 struct page *ksm_might_need_to_copy(struct page *page,
2551                         struct vm_area_struct *vma, unsigned long address)
2552 {
2553         struct anon_vma *anon_vma = page_anon_vma(page);
2554         struct page *new_page;
2555 
2556         if (PageKsm(page)) {
2557                 if (page_stable_node(page) &&
2558                     !(ksm_run & KSM_RUN_UNMERGE))
2559                         return page;    /* no need to copy it */
2560         } else if (!anon_vma) {
2561                 return page;            /* no need to copy it */
2562         } else if (anon_vma->root == vma->anon_vma->root &&
2563                  page->index == linear_page_index(vma, address)) {
2564                 return page;            /* still no need to copy it */
2565         }
2566         if (!PageUptodate(page))
2567                 return page;            /* let do_swap_page report the error */
2568 
2569         new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
2570         if (new_page) {
2571                 copy_user_highpage(new_page, page, address, vma);
2572 
2573                 SetPageDirty(new_page);
2574                 __SetPageUptodate(new_page);
2575                 __SetPageLocked(new_page);
2576         }
2577 
2578         return new_page;
2579 }
2580 
2581 void rmap_walk_ksm(struct page *page, struct rmap_walk_control *rwc)
2582 {
2583         struct stable_node *stable_node;
2584         struct rmap_item *rmap_item;
2585         int search_new_forks = 0;
2586 
2587         VM_BUG_ON_PAGE(!PageKsm(page), page);
2588 
2589         /*
2590          * Rely on the page lock to protect against concurrent modifications
2591          * to that page's node of the stable tree.
2592          */
2593         VM_BUG_ON_PAGE(!PageLocked(page), page);
2594 
2595         stable_node = page_stable_node(page);
2596         if (!stable_node)
2597                 return;
2598 again:
2599         hlist_for_each_entry(rmap_item, &stable_node->hlist, hlist) {
2600                 struct anon_vma *anon_vma = rmap_item->anon_vma;
2601                 struct anon_vma_chain *vmac;
2602                 struct vm_area_struct *vma;
2603 
2604                 cond_resched();
2605                 anon_vma_lock_read(anon_vma);
2606                 anon_vma_interval_tree_foreach(vmac, &anon_vma->rb_root,
2607                                                0, ULONG_MAX) {
2608                         unsigned long addr;
2609 
2610                         cond_resched();
2611                         vma = vmac->vma;
2612 
2613                         /* Ignore the stable/unstable/sqnr flags */
2614                         addr = rmap_item->address & ~KSM_FLAG_MASK;
2615 
2616                         if (addr < vma->vm_start || addr >= vma->vm_end)
2617                                 continue;
2618                         /*
2619                          * Initially we examine only the vma which covers this
2620                          * rmap_item; but later, if there is still work to do,
2621                          * we examine covering vmas in other mms: in case they
2622                          * were forked from the original since ksmd passed.
2623                          */
2624                         if ((rmap_item->mm == vma->vm_mm) == search_new_forks)
2625                                 continue;
2626 
2627                         if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg))
2628                                 continue;
2629 
2630                         if (!rwc->rmap_one(page, vma, addr, rwc->arg)) {
2631                                 anon_vma_unlock_read(anon_vma);
2632                                 return;
2633                         }
2634                         if (rwc->done && rwc->done(page)) {
2635                                 anon_vma_unlock_read(anon_vma);
2636                                 return;
2637                         }
2638                 }
2639                 anon_vma_unlock_read(anon_vma);
2640         }
2641         if (!search_new_forks++)
2642                 goto again;
2643 }
2644 
2645 #ifdef CONFIG_MIGRATION
2646 void ksm_migrate_page(struct page *newpage, struct page *oldpage)
2647 {
2648         struct stable_node *stable_node;
2649 
2650         VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage);
2651         VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
2652         VM_BUG_ON_PAGE(newpage->mapping != oldpage->mapping, newpage);
2653 
2654         stable_node = page_stable_node(newpage);
2655         if (stable_node) {
2656                 VM_BUG_ON_PAGE(stable_node->kpfn != page_to_pfn(oldpage), oldpage);
2657                 stable_node->kpfn = page_to_pfn(newpage);
2658                 /*
2659                  * newpage->mapping was set in advance; now we need smp_wmb()
2660                  * to make sure that the new stable_node->kpfn is visible
2661                  * to get_ksm_page() before it can see that oldpage->mapping
2662                  * has gone stale (or that PageSwapCache has been cleared).
2663                  */
2664                 smp_wmb();
2665                 set_page_stable_node(oldpage, NULL);
2666         }
2667 }
2668 #endif /* CONFIG_MIGRATION */
2669 
2670 #ifdef CONFIG_MEMORY_HOTREMOVE
2671 static void wait_while_offlining(void)
2672 {
2673         while (ksm_run & KSM_RUN_OFFLINE) {
2674                 mutex_unlock(&ksm_thread_mutex);
2675                 wait_on_bit(&ksm_run, ilog2(KSM_RUN_OFFLINE),
2676                             TASK_UNINTERRUPTIBLE);
2677                 mutex_lock(&ksm_thread_mutex);
2678         }
2679 }
2680 
2681 static bool stable_node_dup_remove_range(struct stable_node *stable_node,
2682                                          unsigned long start_pfn,
2683                                          unsigned long end_pfn)
2684 {
2685         if (stable_node->kpfn >= start_pfn &&
2686             stable_node->kpfn < end_pfn) {
2687                 /*
2688                  * Don't get_ksm_page, page has already gone:
2689                  * which is why we keep kpfn instead of page*
2690                  */
2691                 remove_node_from_stable_tree(stable_node);
2692                 return true;
2693         }
2694         return false;
2695 }
2696 
2697 static bool stable_node_chain_remove_range(struct stable_node *stable_node,
2698                                            unsigned long start_pfn,
2699                                            unsigned long end_pfn,
2700                                            struct rb_root *root)
2701 {
2702         struct stable_node *dup;
2703         struct hlist_node *hlist_safe;
2704 
2705         if (!is_stable_node_chain(stable_node)) {
2706                 VM_BUG_ON(is_stable_node_dup(stable_node));
2707                 return stable_node_dup_remove_range(stable_node, start_pfn,
2708                                                     end_pfn);
2709         }
2710 
2711         hlist_for_each_entry_safe(dup, hlist_safe,
2712                                   &stable_node->hlist, hlist_dup) {
2713                 VM_BUG_ON(!is_stable_node_dup(dup));
2714                 stable_node_dup_remove_range(dup, start_pfn, end_pfn);
2715         }
2716         if (hlist_empty(&stable_node->hlist)) {
2717                 free_stable_node_chain(stable_node, root);
2718                 return true; /* notify caller that tree was rebalanced */
2719         } else
2720                 return false;
2721 }
2722 
2723 static void ksm_check_stable_tree(unsigned long start_pfn,
2724                                   unsigned long end_pfn)
2725 {
2726         struct stable_node *stable_node, *next;
2727         struct rb_node *node;
2728         int nid;
2729 
2730         for (nid = 0; nid < ksm_nr_node_ids; nid++) {
2731                 node = rb_first(root_stable_tree + nid);
2732                 while (node) {
2733                         stable_node = rb_entry(node, struct stable_node, node);
2734                         if (stable_node_chain_remove_range(stable_node,
2735                                                            start_pfn, end_pfn,
2736                                                            root_stable_tree +
2737                                                            nid))
2738                                 node = rb_first(root_stable_tree + nid);
2739                         else
2740                                 node = rb_next(node);
2741                         cond_resched();
2742                 }
2743         }
2744         list_for_each_entry_safe(stable_node, next, &migrate_nodes, list) {
2745                 if (stable_node->kpfn >= start_pfn &&
2746                     stable_node->kpfn < end_pfn)
2747                         remove_node_from_stable_tree(stable_node);
2748                 cond_resched();
2749         }
2750 }
2751 
2752 static int ksm_memory_callback(struct notifier_block *self,
2753                                unsigned long action, void *arg)
2754 {
2755         struct memory_notify *mn = arg;
2756 
2757         switch (action) {
2758         case MEM_GOING_OFFLINE:
2759                 /*
2760                  * Prevent ksm_do_scan(), unmerge_and_remove_all_rmap_items()
2761                  * and remove_all_stable_nodes() while memory is going offline:
2762                  * it is unsafe for them to touch the stable tree at this time.
2763                  * But unmerge_ksm_pages(), rmap lookups and other entry points
2764                  * which do not need the ksm_thread_mutex are all safe.
2765                  */
2766                 mutex_lock(&ksm_thread_mutex);
2767                 ksm_run |= KSM_RUN_OFFLINE;
2768                 mutex_unlock(&ksm_thread_mutex);
2769                 break;
2770 
2771         case MEM_OFFLINE:
2772                 /*
2773                  * Most of the work is done by page migration; but there might
2774                  * be a few stable_nodes left over, still pointing to struct
2775                  * pages which have been offlined: prune those from the tree,
2776                  * otherwise get_ksm_page() might later try to access a
2777                  * non-existent struct page.
2778                  */
2779                 ksm_check_stable_tree(mn->start_pfn,
2780                                       mn->start_pfn + mn->nr_pages);
2781                 /* fallthrough */
2782 
2783         case MEM_CANCEL_OFFLINE:
2784                 mutex_lock(&ksm_thread_mutex);
2785                 ksm_run &= ~KSM_RUN_OFFLINE;
2786                 mutex_unlock(&ksm_thread_mutex);
2787 
2788                 smp_mb();       /* wake_up_bit advises this */
2789                 wake_up_bit(&ksm_run, ilog2(KSM_RUN_OFFLINE));
2790                 break;
2791         }
2792         return NOTIFY_OK;
2793 }
2794 #else
2795 static void wait_while_offlining(void)
2796 {
2797 }
2798 #endif /* CONFIG_MEMORY_HOTREMOVE */
2799 
2800 #ifdef CONFIG_SYSFS
2801 /*
2802  * This all compiles without CONFIG_SYSFS, but is a waste of space.
2803  */
2804 
2805 #define KSM_ATTR_RO(_name) \
2806         static struct kobj_attribute _name##_attr = __ATTR_RO(_name)
2807 #define KSM_ATTR(_name) \
2808         static struct kobj_attribute _name##_attr = \
2809                 __ATTR(_name, 0644, _name##_show, _name##_store)
2810 
2811 static ssize_t sleep_millisecs_show(struct kobject *kobj,
2812                                     struct kobj_attribute *attr, char *buf)
2813 {
2814         return sprintf(buf, "%u\n", ksm_thread_sleep_millisecs);
2815 }
2816 
2817 static ssize_t sleep_millisecs_store(struct kobject *kobj,
2818                                      struct kobj_attribute *attr,
2819                                      const char *buf, size_t count)
2820 {
2821         unsigned long msecs;
2822         int err;
2823 
2824         err = kstrtoul(buf, 10, &msecs);
2825         if (err || msecs > UINT_MAX)
2826                 return -EINVAL;
2827 
2828         ksm_thread_sleep_millisecs = msecs;
2829         wake_up_interruptible(&ksm_iter_wait);
2830 
2831         return count;
2832 }
2833 KSM_ATTR(sleep_millisecs);
2834 
2835 static ssize_t pages_to_scan_show(struct kobject *kobj,
2836                                   struct kobj_attribute *attr, char *buf)
2837 {
2838         return sprintf(buf, "%u\n", ksm_thread_pages_to_scan);
2839 }
2840 
2841 static ssize_t pages_to_scan_store(struct kobject *kobj,
2842                                    struct kobj_attribute *attr,
2843                                    const char *buf, size_t count)
2844 {
2845         int err;
2846         unsigned long nr_pages;
2847 
2848         err = kstrtoul(buf, 10, &nr_pages);
2849         if (err || nr_pages > UINT_MAX)
2850                 return -EINVAL;
2851 
2852         ksm_thread_pages_to_scan = nr_pages;
2853 
2854         return count;
2855 }
2856 KSM_ATTR(pages_to_scan);
2857 
2858 static ssize_t run_show(struct kobject *kobj, struct kobj_attribute *attr,
2859                         char *buf)
2860 {
2861         return sprintf(buf, "%lu\n", ksm_run);
2862 }
2863 
2864 static ssize_t run_store(struct kobject *kobj, struct kobj_attribute *attr,
2865                          const char *buf, size_t count)
2866 {
2867         int err;
2868         unsigned long flags;
2869 
2870         err = kstrtoul(buf, 10, &flags);
2871         if (err || flags > UINT_MAX)
2872                 return -EINVAL;
2873         if (flags > KSM_RUN_UNMERGE)
2874                 return -EINVAL;
2875 
2876         /*
2877          * KSM_RUN_MERGE sets ksmd running, and 0 stops it running.
2878          * KSM_RUN_UNMERGE stops it running and unmerges all rmap_items,
2879          * breaking COW to free the pages_shared (but leaves mm_slots
2880          * on the list for when ksmd may be set running again).
2881          */
2882 
2883         mutex_lock(&ksm_thread_mutex);
2884         wait_while_offlining();
2885         if (ksm_run != flags) {
2886                 ksm_run = flags;
2887                 if (flags & KSM_RUN_UNMERGE) {
2888                         set_current_oom_origin();
2889                         err = unmerge_and_remove_all_rmap_items();
2890                         clear_current_oom_origin();
2891                         if (err) {
2892                                 ksm_run = KSM_RUN_STOP;
2893                                 count = err;
2894                         }
2895                 }
2896         }
2897         mutex_unlock(&ksm_thread_mutex);
2898 
2899         if (flags & KSM_RUN_MERGE)
2900                 wake_up_interruptible(&ksm_thread_wait);
2901 
2902         return count;
2903 }
2904 KSM_ATTR(run);
2905 
2906 #ifdef CONFIG_NUMA
2907 static ssize_t merge_across_nodes_show(struct kobject *kobj,
2908                                 struct kobj_attribute *attr, char *buf)
2909 {
2910         return sprintf(buf, "%u\n", ksm_merge_across_nodes);
2911 }
2912 
2913 static ssize_t merge_across_nodes_store(struct kobject *kobj,
2914                                    struct kobj_attribute *attr,
2915                                    const char *buf, size_t count)
2916 {
2917         int err;
2918         unsigned long knob;
2919 
2920         err = kstrtoul(buf, 10, &knob);
2921         if (err)
2922                 return err;
2923         if (knob > 1)
2924                 return -EINVAL;
2925 
2926         mutex_lock(&ksm_thread_mutex);
2927         wait_while_offlining();
2928         if (ksm_merge_across_nodes != knob) {
2929                 if (ksm_pages_shared || remove_all_stable_nodes())
2930                         err = -EBUSY;
2931                 else if (root_stable_tree == one_stable_tree) {
2932                         struct rb_root *buf;
2933                         /*
2934                          * This is the first time that we switch away from the
2935                          * default of merging across nodes: must now allocate
2936                          * a buffer to hold as many roots as may be needed.
2937                          * Allocate stable and unstable together:
2938                          * MAXSMP NODES_SHIFT 10 will use 16kB.
2939                          */
2940                         buf = kcalloc(nr_node_ids + nr_node_ids, sizeof(*buf),
2941                                       GFP_KERNEL);
2942                         /* Let us assume that RB_ROOT is NULL is zero */
2943                         if (!buf)
2944                                 err = -ENOMEM;
2945                         else {
2946                                 root_stable_tree = buf;
2947                                 root_unstable_tree = buf + nr_node_ids;
2948                                 /* Stable tree is empty but not the unstable */
2949                                 root_unstable_tree[0] = one_unstable_tree[0];
2950                         }
2951                 }
2952                 if (!err) {
2953                         ksm_merge_across_nodes = knob;
2954                         ksm_nr_node_ids = knob ? 1 : nr_node_ids;
2955                 }
2956         }
2957         mutex_unlock(&ksm_thread_mutex);
2958 
2959         return err ? err : count;
2960 }
2961 KSM_ATTR(merge_across_nodes);
2962 #endif
2963 
2964 static ssize_t use_zero_pages_show(struct kobject *kobj,
2965                                 struct kobj_attribute *attr, char *buf)
2966 {
2967         return sprintf(buf, "%u\n", ksm_use_zero_pages);
2968 }
2969 static ssize_t use_zero_pages_store(struct kobject *kobj,
2970                                    struct kobj_attribute *attr,
2971                                    const char *buf, size_t count)
2972 {
2973         int err;
2974         bool value;
2975 
2976         err = kstrtobool(buf, &value);
2977         if (err)
2978                 return -EINVAL;
2979 
2980         ksm_use_zero_pages = value;
2981 
2982         return count;
2983 }
2984 KSM_ATTR(use_zero_pages);
2985 
2986 static ssize_t max_page_sharing_show(struct kobject *kobj,
2987                                      struct kobj_attribute *attr, char *buf)
2988 {
2989         return sprintf(buf, "%u\n", ksm_max_page_sharing);
2990 }
2991 
2992 static ssize_t max_page_sharing_store(struct kobject *kobj,
2993                                       struct kobj_attribute *attr,
2994                                       const char *buf, size_t count)
2995 {
2996         int err;
2997         int knob;
2998 
2999         err = kstrtoint(buf, 10, &knob);
3000         if (err)
3001                 return err;
3002         /*
3003          * When a KSM page is created it is shared by 2 mappings. This
3004          * being a signed comparison, it implicitly verifies it's not
3005          * negative.
3006          */
3007         if (knob < 2)
3008                 return -EINVAL;
3009 
3010         if (READ_ONCE(ksm_max_page_sharing) == knob)
3011                 return count;
3012 
3013         mutex_lock(&ksm_thread_mutex);
3014         wait_while_offlining();
3015         if (ksm_max_page_sharing != knob) {
3016                 if (ksm_pages_shared || remove_all_stable_nodes())
3017                         err = -EBUSY;
3018                 else
3019                         ksm_max_page_sharing = knob;
3020         }
3021         mutex_unlock(&ksm_thread_mutex);
3022 
3023         return err ? err : count;
3024 }
3025 KSM_ATTR(max_page_sharing);
3026 
3027 static ssize_t pages_shared_show(struct kobject *kobj,
3028                                  struct kobj_attribute *attr, char *buf)
3029 {
3030         return sprintf(buf, "%lu\n", ksm_pages_shared);
3031 }
3032 KSM_ATTR_RO(pages_shared);
3033 
3034 static ssize_t pages_sharing_show(struct kobject *kobj,
3035                                   struct kobj_attribute *attr, char *buf)
3036 {
3037         return sprintf(buf, "%lu\n", ksm_pages_sharing);
3038 }
3039 KSM_ATTR_RO(pages_sharing);
3040 
3041 static ssize_t pages_unshared_show(struct kobject *kobj,
3042                                    struct kobj_attribute *attr, char *buf)
3043 {
3044         return sprintf(buf, "%lu\n", ksm_pages_unshared);
3045 }
3046 KSM_ATTR_RO(pages_unshared);
3047 
3048 static ssize_t pages_volatile_show(struct kobject *kobj,
3049                                    struct kobj_attribute *attr, char *buf)
3050 {
3051         long ksm_pages_volatile;
3052 
3053         ksm_pages_volatile = ksm_rmap_items - ksm_pages_shared
3054                                 - ksm_pages_sharing - ksm_pages_unshared;
3055         /*
3056          * It was not worth any locking to calculate that statistic,
3057          * but it might therefore sometimes be negative: conceal that.
3058          */
3059         if (ksm_pages_volatile < 0)
3060                 ksm_pages_volatile = 0;
3061         return sprintf(buf, "%ld\n", ksm_pages_volatile);
3062 }
3063 KSM_ATTR_RO(pages_volatile);
3064 
3065 static ssize_t stable_node_dups_show(struct kobject *kobj,
3066                                      struct kobj_attribute *attr, char *buf)
3067 {
3068         return sprintf(buf, "%lu\n", ksm_stable_node_dups);
3069 }
3070 KSM_ATTR_RO(stable_node_dups);
3071 
3072 static ssize_t stable_node_chains_show(struct kobject *kobj,
3073                                        struct kobj_attribute *attr, char *buf)
3074 {
3075         return sprintf(buf, "%lu\n", ksm_stable_node_chains);
3076 }
3077 KSM_ATTR_RO(stable_node_chains);
3078 
3079 static ssize_t
3080 stable_node_chains_prune_millisecs_show(struct kobject *kobj,
3081                                         struct kobj_attribute *attr,
3082                                         char *buf)
3083 {
3084         return sprintf(buf, "%u\n", ksm_stable_node_chains_prune_millisecs);
3085 }
3086 
3087 static ssize_t
3088 stable_node_chains_prune_millisecs_store(struct kobject *kobj,
3089                                          struct kobj_attribute *attr,
3090                                          const char *buf, size_t count)
3091 {
3092         unsigned long msecs;
3093         int err;
3094 
3095         err = kstrtoul(buf, 10, &msecs);
3096         if (err || msecs > UINT_MAX)
3097                 return -EINVAL;
3098 
3099         ksm_stable_node_chains_prune_millisecs = msecs;
3100 
3101         return count;
3102 }
3103 KSM_ATTR(stable_node_chains_prune_millisecs);
3104 
3105 static ssize_t full_scans_show(struct kobject *kobj,
3106                                struct kobj_attribute *attr, char *buf)
3107 {
3108         return sprintf(buf, "%lu\n", ksm_scan.seqnr);
3109 }
3110 KSM_ATTR_RO(full_scans);
3111 
3112 static struct attribute *ksm_attrs[] = {
3113         &sleep_millisecs_attr.attr,
3114         &pages_to_scan_attr.attr,
3115         &run_attr.attr,
3116         &pages_shared_attr.attr,
3117         &pages_sharing_attr.attr,
3118         &pages_unshared_attr.attr,
3119         &pages_volatile_attr.attr,
3120         &full_scans_attr.attr,
3121 #ifdef CONFIG_NUMA
3122         &merge_across_nodes_attr.attr,
3123 #endif
3124         &max_page_sharing_attr.attr,
3125         &stable_node_chains_attr.attr,
3126         &stable_node_dups_attr.attr,
3127         &stable_node_chains_prune_millisecs_attr.attr,
3128         &use_zero_pages_attr.attr,
3129         NULL,
3130 };
3131 
3132 static const struct attribute_group ksm_attr_group = {
3133         .attrs = ksm_attrs,
3134         .name = "ksm",
3135 };
3136 #endif /* CONFIG_SYSFS */
3137 
3138 static int __init ksm_init(void)
3139 {
3140         struct task_struct *ksm_thread;
3141         int err;
3142 
3143         /* The correct value depends on page size and endianness */
3144         zero_checksum = calc_checksum(ZERO_PAGE(0));
3145         /* Default to false for backwards compatibility */
3146         ksm_use_zero_pages = false;
3147 
3148         err = ksm_slab_init();
3149         if (err)
3150                 goto out;
3151 
3152         ksm_thread = kthread_run(ksm_scan_thread, NULL, "ksmd");
3153         if (IS_ERR(ksm_thread)) {
3154                 pr_err("ksm: creating kthread failed\n");
3155                 err = PTR_ERR(ksm_thread);
3156                 goto out_free;
3157         }
3158 
3159 #ifdef CONFIG_SYSFS
3160         err = sysfs_create_group(mm_kobj, &ksm_attr_group);
3161         if (err) {
3162                 pr_err("ksm: register sysfs failed\n");
3163                 kthread_stop(ksm_thread);
3164                 goto out_free;
3165         }
3166 #else
3167         ksm_run = KSM_RUN_MERGE;        /* no way for user to start it */
3168 
3169 #endif /* CONFIG_SYSFS */
3170 
3171 #ifdef CONFIG_MEMORY_HOTREMOVE
3172         /* There is no significance to this priority 100 */
3173         hotplug_memory_notifier(ksm_memory_callback, 100);
3174 #endif
3175         return 0;
3176 
3177 out_free:
3178         ksm_slab_free();
3179 out:
3180         return err;
3181 }
3182 subsys_initcall(ksm_init);
3183 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp