]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/memory.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
5 | */ | |
6 | ||
7 | /* | |
8 | * demand-loading started 01.12.91 - seems it is high on the list of | |
9 | * things wanted, and it should be easy to implement. - Linus | |
10 | */ | |
11 | ||
12 | /* | |
13 | * Ok, demand-loading was easy, shared pages a little bit tricker. Shared | |
14 | * pages started 02.12.91, seems to work. - Linus. | |
15 | * | |
16 | * Tested sharing by executing about 30 /bin/sh: under the old kernel it | |
17 | * would have taken more than the 6M I have free, but it worked well as | |
18 | * far as I could see. | |
19 | * | |
20 | * Also corrected some "invalidate()"s - I wasn't doing enough of them. | |
21 | */ | |
22 | ||
23 | /* | |
24 | * Real VM (paging to/from disk) started 18.12.91. Much more work and | |
25 | * thought has to go into this. Oh, well.. | |
26 | * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why. | |
27 | * Found it. Everything seems to work now. | |
28 | * 20.12.91 - Ok, making the swap-device changeable like the root. | |
29 | */ | |
30 | ||
31 | /* | |
32 | * 05.04.94 - Multi-page memory management added for v1.1. | |
33 | * Idea by Alex Bligh (alex@cconcepts.co.uk) | |
34 | * | |
35 | * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG | |
36 | * (Gerhard.Wichert@pdb.siemens.de) | |
37 | * | |
38 | * Aug/Sep 2004 Changed to four level page tables (Andi Kleen) | |
39 | */ | |
40 | ||
41 | #include <linux/kernel_stat.h> | |
42 | #include <linux/mm.h> | |
43 | #include <linux/hugetlb.h> | |
44 | #include <linux/mman.h> | |
45 | #include <linux/swap.h> | |
46 | #include <linux/highmem.h> | |
47 | #include <linux/pagemap.h> | |
9a840895 | 48 | #include <linux/ksm.h> |
1da177e4 | 49 | #include <linux/rmap.h> |
b95f1b31 | 50 | #include <linux/export.h> |
0ff92245 | 51 | #include <linux/delayacct.h> |
1da177e4 | 52 | #include <linux/init.h> |
edc79b2a | 53 | #include <linux/writeback.h> |
8a9f3ccd | 54 | #include <linux/memcontrol.h> |
cddb8a5c | 55 | #include <linux/mmu_notifier.h> |
3dc14741 HD |
56 | #include <linux/kallsyms.h> |
57 | #include <linux/swapops.h> | |
58 | #include <linux/elf.h> | |
5a0e3ad6 | 59 | #include <linux/gfp.h> |
4daae3b4 | 60 | #include <linux/migrate.h> |
2fbc57c5 | 61 | #include <linux/string.h> |
0abdd7a8 | 62 | #include <linux/dma-debug.h> |
1592eef0 | 63 | #include <linux/debugfs.h> |
1da177e4 | 64 | |
6952b61d | 65 | #include <asm/io.h> |
1da177e4 LT |
66 | #include <asm/pgalloc.h> |
67 | #include <asm/uaccess.h> | |
68 | #include <asm/tlb.h> | |
69 | #include <asm/tlbflush.h> | |
70 | #include <asm/pgtable.h> | |
71 | ||
42b77728 JB |
72 | #include "internal.h" |
73 | ||
90572890 PZ |
74 | #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS |
75 | #warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid. | |
75980e97 PZ |
76 | #endif |
77 | ||
d41dee36 | 78 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
79 | /* use the per-pgdat data instead for discontigmem - mbligh */ |
80 | unsigned long max_mapnr; | |
81 | struct page *mem_map; | |
82 | ||
83 | EXPORT_SYMBOL(max_mapnr); | |
84 | EXPORT_SYMBOL(mem_map); | |
85 | #endif | |
86 | ||
1da177e4 LT |
87 | /* |
88 | * A number of key systems in x86 including ioremap() rely on the assumption | |
89 | * that high_memory defines the upper bound on direct map memory, then end | |
90 | * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and | |
91 | * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL | |
92 | * and ZONE_HIGHMEM. | |
93 | */ | |
94 | void * high_memory; | |
1da177e4 | 95 | |
1da177e4 | 96 | EXPORT_SYMBOL(high_memory); |
1da177e4 | 97 | |
32a93233 IM |
98 | /* |
99 | * Randomize the address space (stacks, mmaps, brk, etc.). | |
100 | * | |
101 | * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization, | |
102 | * as ancient (libc5 based) binaries can segfault. ) | |
103 | */ | |
104 | int randomize_va_space __read_mostly = | |
105 | #ifdef CONFIG_COMPAT_BRK | |
106 | 1; | |
107 | #else | |
108 | 2; | |
109 | #endif | |
a62eaf15 AK |
110 | |
111 | static int __init disable_randmaps(char *s) | |
112 | { | |
113 | randomize_va_space = 0; | |
9b41046c | 114 | return 1; |
a62eaf15 AK |
115 | } |
116 | __setup("norandmaps", disable_randmaps); | |
117 | ||
62eede62 | 118 | unsigned long zero_pfn __read_mostly; |
03f6462a | 119 | unsigned long highest_memmap_pfn __read_mostly; |
a13ea5b7 HD |
120 | |
121 | /* | |
122 | * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init() | |
123 | */ | |
124 | static int __init init_zero_pfn(void) | |
125 | { | |
126 | zero_pfn = page_to_pfn(ZERO_PAGE(0)); | |
127 | return 0; | |
128 | } | |
129 | core_initcall(init_zero_pfn); | |
a62eaf15 | 130 | |
d559db08 | 131 | |
34e55232 KH |
132 | #if defined(SPLIT_RSS_COUNTING) |
133 | ||
ea48cf78 | 134 | void sync_mm_rss(struct mm_struct *mm) |
34e55232 KH |
135 | { |
136 | int i; | |
137 | ||
138 | for (i = 0; i < NR_MM_COUNTERS; i++) { | |
05af2e10 DR |
139 | if (current->rss_stat.count[i]) { |
140 | add_mm_counter(mm, i, current->rss_stat.count[i]); | |
141 | current->rss_stat.count[i] = 0; | |
34e55232 KH |
142 | } |
143 | } | |
05af2e10 | 144 | current->rss_stat.events = 0; |
34e55232 KH |
145 | } |
146 | ||
147 | static void add_mm_counter_fast(struct mm_struct *mm, int member, int val) | |
148 | { | |
149 | struct task_struct *task = current; | |
150 | ||
151 | if (likely(task->mm == mm)) | |
152 | task->rss_stat.count[member] += val; | |
153 | else | |
154 | add_mm_counter(mm, member, val); | |
155 | } | |
156 | #define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1) | |
157 | #define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1) | |
158 | ||
159 | /* sync counter once per 64 page faults */ | |
160 | #define TASK_RSS_EVENTS_THRESH (64) | |
161 | static void check_sync_rss_stat(struct task_struct *task) | |
162 | { | |
163 | if (unlikely(task != current)) | |
164 | return; | |
165 | if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH)) | |
ea48cf78 | 166 | sync_mm_rss(task->mm); |
34e55232 | 167 | } |
9547d01b | 168 | #else /* SPLIT_RSS_COUNTING */ |
34e55232 KH |
169 | |
170 | #define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member) | |
171 | #define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member) | |
172 | ||
173 | static void check_sync_rss_stat(struct task_struct *task) | |
174 | { | |
175 | } | |
176 | ||
9547d01b PZ |
177 | #endif /* SPLIT_RSS_COUNTING */ |
178 | ||
179 | #ifdef HAVE_GENERIC_MMU_GATHER | |
180 | ||
181 | static int tlb_next_batch(struct mmu_gather *tlb) | |
182 | { | |
183 | struct mmu_gather_batch *batch; | |
184 | ||
185 | batch = tlb->active; | |
186 | if (batch->next) { | |
187 | tlb->active = batch->next; | |
188 | return 1; | |
189 | } | |
190 | ||
53a59fc6 MH |
191 | if (tlb->batch_count == MAX_GATHER_BATCH_COUNT) |
192 | return 0; | |
193 | ||
9547d01b PZ |
194 | batch = (void *)__get_free_pages(GFP_NOWAIT | __GFP_NOWARN, 0); |
195 | if (!batch) | |
196 | return 0; | |
197 | ||
53a59fc6 | 198 | tlb->batch_count++; |
9547d01b PZ |
199 | batch->next = NULL; |
200 | batch->nr = 0; | |
201 | batch->max = MAX_GATHER_BATCH; | |
202 | ||
203 | tlb->active->next = batch; | |
204 | tlb->active = batch; | |
205 | ||
206 | return 1; | |
207 | } | |
208 | ||
209 | /* tlb_gather_mmu | |
210 | * Called to initialize an (on-stack) mmu_gather structure for page-table | |
211 | * tear-down from @mm. The @fullmm argument is used when @mm is without | |
212 | * users and we're going to destroy the full address space (exit/execve). | |
213 | */ | |
2b047252 | 214 | void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, unsigned long start, unsigned long end) |
9547d01b PZ |
215 | { |
216 | tlb->mm = mm; | |
217 | ||
2b047252 LT |
218 | /* Is it from 0 to ~0? */ |
219 | tlb->fullmm = !(start | (end+1)); | |
1de14c3c | 220 | tlb->need_flush_all = 0; |
2b047252 LT |
221 | tlb->start = start; |
222 | tlb->end = end; | |
9547d01b | 223 | tlb->need_flush = 0; |
9547d01b PZ |
224 | tlb->local.next = NULL; |
225 | tlb->local.nr = 0; | |
226 | tlb->local.max = ARRAY_SIZE(tlb->__pages); | |
227 | tlb->active = &tlb->local; | |
53a59fc6 | 228 | tlb->batch_count = 0; |
9547d01b PZ |
229 | |
230 | #ifdef CONFIG_HAVE_RCU_TABLE_FREE | |
231 | tlb->batch = NULL; | |
232 | #endif | |
233 | } | |
234 | ||
1cf35d47 | 235 | static void tlb_flush_mmu_tlbonly(struct mmu_gather *tlb) |
9547d01b | 236 | { |
9547d01b PZ |
237 | tlb->need_flush = 0; |
238 | tlb_flush(tlb); | |
239 | #ifdef CONFIG_HAVE_RCU_TABLE_FREE | |
240 | tlb_table_flush(tlb); | |
34e55232 | 241 | #endif |
1cf35d47 LT |
242 | } |
243 | ||
244 | static void tlb_flush_mmu_free(struct mmu_gather *tlb) | |
245 | { | |
246 | struct mmu_gather_batch *batch; | |
34e55232 | 247 | |
9547d01b PZ |
248 | for (batch = &tlb->local; batch; batch = batch->next) { |
249 | free_pages_and_swap_cache(batch->pages, batch->nr); | |
250 | batch->nr = 0; | |
251 | } | |
252 | tlb->active = &tlb->local; | |
253 | } | |
254 | ||
1cf35d47 LT |
255 | void tlb_flush_mmu(struct mmu_gather *tlb) |
256 | { | |
257 | if (!tlb->need_flush) | |
258 | return; | |
259 | tlb_flush_mmu_tlbonly(tlb); | |
260 | tlb_flush_mmu_free(tlb); | |
261 | } | |
262 | ||
9547d01b PZ |
263 | /* tlb_finish_mmu |
264 | * Called at the end of the shootdown operation to free up any resources | |
265 | * that were required. | |
266 | */ | |
267 | void tlb_finish_mmu(struct mmu_gather *tlb, unsigned long start, unsigned long end) | |
268 | { | |
269 | struct mmu_gather_batch *batch, *next; | |
270 | ||
271 | tlb_flush_mmu(tlb); | |
272 | ||
273 | /* keep the page table cache within bounds */ | |
274 | check_pgt_cache(); | |
275 | ||
276 | for (batch = tlb->local.next; batch; batch = next) { | |
277 | next = batch->next; | |
278 | free_pages((unsigned long)batch, 0); | |
279 | } | |
280 | tlb->local.next = NULL; | |
281 | } | |
282 | ||
283 | /* __tlb_remove_page | |
284 | * Must perform the equivalent to __free_pte(pte_get_and_clear(ptep)), while | |
285 | * handling the additional races in SMP caused by other CPUs caching valid | |
286 | * mappings in their TLBs. Returns the number of free page slots left. | |
287 | * When out of page slots we must call tlb_flush_mmu(). | |
288 | */ | |
289 | int __tlb_remove_page(struct mmu_gather *tlb, struct page *page) | |
290 | { | |
291 | struct mmu_gather_batch *batch; | |
292 | ||
f21760b1 | 293 | VM_BUG_ON(!tlb->need_flush); |
9547d01b | 294 | |
9547d01b PZ |
295 | batch = tlb->active; |
296 | batch->pages[batch->nr++] = page; | |
297 | if (batch->nr == batch->max) { | |
298 | if (!tlb_next_batch(tlb)) | |
299 | return 0; | |
0b43c3aa | 300 | batch = tlb->active; |
9547d01b | 301 | } |
309381fe | 302 | VM_BUG_ON_PAGE(batch->nr > batch->max, page); |
9547d01b PZ |
303 | |
304 | return batch->max - batch->nr; | |
305 | } | |
306 | ||
307 | #endif /* HAVE_GENERIC_MMU_GATHER */ | |
308 | ||
26723911 PZ |
309 | #ifdef CONFIG_HAVE_RCU_TABLE_FREE |
310 | ||
311 | /* | |
312 | * See the comment near struct mmu_table_batch. | |
313 | */ | |
314 | ||
315 | static void tlb_remove_table_smp_sync(void *arg) | |
316 | { | |
317 | /* Simply deliver the interrupt */ | |
318 | } | |
319 | ||
320 | static void tlb_remove_table_one(void *table) | |
321 | { | |
322 | /* | |
323 | * This isn't an RCU grace period and hence the page-tables cannot be | |
324 | * assumed to be actually RCU-freed. | |
325 | * | |
326 | * It is however sufficient for software page-table walkers that rely on | |
327 | * IRQ disabling. See the comment near struct mmu_table_batch. | |
328 | */ | |
329 | smp_call_function(tlb_remove_table_smp_sync, NULL, 1); | |
330 | __tlb_remove_table(table); | |
331 | } | |
332 | ||
333 | static void tlb_remove_table_rcu(struct rcu_head *head) | |
334 | { | |
335 | struct mmu_table_batch *batch; | |
336 | int i; | |
337 | ||
338 | batch = container_of(head, struct mmu_table_batch, rcu); | |
339 | ||
340 | for (i = 0; i < batch->nr; i++) | |
341 | __tlb_remove_table(batch->tables[i]); | |
342 | ||
343 | free_page((unsigned long)batch); | |
344 | } | |
345 | ||
346 | void tlb_table_flush(struct mmu_gather *tlb) | |
347 | { | |
348 | struct mmu_table_batch **batch = &tlb->batch; | |
349 | ||
350 | if (*batch) { | |
351 | call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu); | |
352 | *batch = NULL; | |
353 | } | |
354 | } | |
355 | ||
356 | void tlb_remove_table(struct mmu_gather *tlb, void *table) | |
357 | { | |
358 | struct mmu_table_batch **batch = &tlb->batch; | |
359 | ||
360 | tlb->need_flush = 1; | |
361 | ||
362 | /* | |
363 | * When there's less then two users of this mm there cannot be a | |
364 | * concurrent page-table walk. | |
365 | */ | |
366 | if (atomic_read(&tlb->mm->mm_users) < 2) { | |
367 | __tlb_remove_table(table); | |
368 | return; | |
369 | } | |
370 | ||
371 | if (*batch == NULL) { | |
372 | *batch = (struct mmu_table_batch *)__get_free_page(GFP_NOWAIT | __GFP_NOWARN); | |
373 | if (*batch == NULL) { | |
374 | tlb_remove_table_one(table); | |
375 | return; | |
376 | } | |
377 | (*batch)->nr = 0; | |
378 | } | |
379 | (*batch)->tables[(*batch)->nr++] = table; | |
380 | if ((*batch)->nr == MAX_TABLE_BATCH) | |
381 | tlb_table_flush(tlb); | |
382 | } | |
383 | ||
9547d01b | 384 | #endif /* CONFIG_HAVE_RCU_TABLE_FREE */ |
26723911 | 385 | |
1da177e4 LT |
386 | /* |
387 | * Note: this doesn't free the actual pages themselves. That | |
388 | * has been handled earlier when unmapping all the memory regions. | |
389 | */ | |
9e1b32ca BH |
390 | static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd, |
391 | unsigned long addr) | |
1da177e4 | 392 | { |
2f569afd | 393 | pgtable_t token = pmd_pgtable(*pmd); |
e0da382c | 394 | pmd_clear(pmd); |
9e1b32ca | 395 | pte_free_tlb(tlb, token, addr); |
e1f56c89 | 396 | atomic_long_dec(&tlb->mm->nr_ptes); |
1da177e4 LT |
397 | } |
398 | ||
e0da382c HD |
399 | static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud, |
400 | unsigned long addr, unsigned long end, | |
401 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
402 | { |
403 | pmd_t *pmd; | |
404 | unsigned long next; | |
e0da382c | 405 | unsigned long start; |
1da177e4 | 406 | |
e0da382c | 407 | start = addr; |
1da177e4 | 408 | pmd = pmd_offset(pud, addr); |
1da177e4 LT |
409 | do { |
410 | next = pmd_addr_end(addr, end); | |
411 | if (pmd_none_or_clear_bad(pmd)) | |
412 | continue; | |
9e1b32ca | 413 | free_pte_range(tlb, pmd, addr); |
1da177e4 LT |
414 | } while (pmd++, addr = next, addr != end); |
415 | ||
e0da382c HD |
416 | start &= PUD_MASK; |
417 | if (start < floor) | |
418 | return; | |
419 | if (ceiling) { | |
420 | ceiling &= PUD_MASK; | |
421 | if (!ceiling) | |
422 | return; | |
1da177e4 | 423 | } |
e0da382c HD |
424 | if (end - 1 > ceiling - 1) |
425 | return; | |
426 | ||
427 | pmd = pmd_offset(pud, start); | |
428 | pud_clear(pud); | |
9e1b32ca | 429 | pmd_free_tlb(tlb, pmd, start); |
1da177e4 LT |
430 | } |
431 | ||
e0da382c HD |
432 | static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd, |
433 | unsigned long addr, unsigned long end, | |
434 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
435 | { |
436 | pud_t *pud; | |
437 | unsigned long next; | |
e0da382c | 438 | unsigned long start; |
1da177e4 | 439 | |
e0da382c | 440 | start = addr; |
1da177e4 | 441 | pud = pud_offset(pgd, addr); |
1da177e4 LT |
442 | do { |
443 | next = pud_addr_end(addr, end); | |
444 | if (pud_none_or_clear_bad(pud)) | |
445 | continue; | |
e0da382c | 446 | free_pmd_range(tlb, pud, addr, next, floor, ceiling); |
1da177e4 LT |
447 | } while (pud++, addr = next, addr != end); |
448 | ||
e0da382c HD |
449 | start &= PGDIR_MASK; |
450 | if (start < floor) | |
451 | return; | |
452 | if (ceiling) { | |
453 | ceiling &= PGDIR_MASK; | |
454 | if (!ceiling) | |
455 | return; | |
1da177e4 | 456 | } |
e0da382c HD |
457 | if (end - 1 > ceiling - 1) |
458 | return; | |
459 | ||
460 | pud = pud_offset(pgd, start); | |
461 | pgd_clear(pgd); | |
9e1b32ca | 462 | pud_free_tlb(tlb, pud, start); |
1da177e4 LT |
463 | } |
464 | ||
465 | /* | |
e0da382c | 466 | * This function frees user-level page tables of a process. |
1da177e4 | 467 | */ |
42b77728 | 468 | void free_pgd_range(struct mmu_gather *tlb, |
e0da382c HD |
469 | unsigned long addr, unsigned long end, |
470 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
471 | { |
472 | pgd_t *pgd; | |
473 | unsigned long next; | |
e0da382c HD |
474 | |
475 | /* | |
476 | * The next few lines have given us lots of grief... | |
477 | * | |
478 | * Why are we testing PMD* at this top level? Because often | |
479 | * there will be no work to do at all, and we'd prefer not to | |
480 | * go all the way down to the bottom just to discover that. | |
481 | * | |
482 | * Why all these "- 1"s? Because 0 represents both the bottom | |
483 | * of the address space and the top of it (using -1 for the | |
484 | * top wouldn't help much: the masks would do the wrong thing). | |
485 | * The rule is that addr 0 and floor 0 refer to the bottom of | |
486 | * the address space, but end 0 and ceiling 0 refer to the top | |
487 | * Comparisons need to use "end - 1" and "ceiling - 1" (though | |
488 | * that end 0 case should be mythical). | |
489 | * | |
490 | * Wherever addr is brought up or ceiling brought down, we must | |
491 | * be careful to reject "the opposite 0" before it confuses the | |
492 | * subsequent tests. But what about where end is brought down | |
493 | * by PMD_SIZE below? no, end can't go down to 0 there. | |
494 | * | |
495 | * Whereas we round start (addr) and ceiling down, by different | |
496 | * masks at different levels, in order to test whether a table | |
497 | * now has no other vmas using it, so can be freed, we don't | |
498 | * bother to round floor or end up - the tests don't need that. | |
499 | */ | |
1da177e4 | 500 | |
e0da382c HD |
501 | addr &= PMD_MASK; |
502 | if (addr < floor) { | |
503 | addr += PMD_SIZE; | |
504 | if (!addr) | |
505 | return; | |
506 | } | |
507 | if (ceiling) { | |
508 | ceiling &= PMD_MASK; | |
509 | if (!ceiling) | |
510 | return; | |
511 | } | |
512 | if (end - 1 > ceiling - 1) | |
513 | end -= PMD_SIZE; | |
514 | if (addr > end - 1) | |
515 | return; | |
516 | ||
42b77728 | 517 | pgd = pgd_offset(tlb->mm, addr); |
1da177e4 LT |
518 | do { |
519 | next = pgd_addr_end(addr, end); | |
520 | if (pgd_none_or_clear_bad(pgd)) | |
521 | continue; | |
42b77728 | 522 | free_pud_range(tlb, pgd, addr, next, floor, ceiling); |
1da177e4 | 523 | } while (pgd++, addr = next, addr != end); |
e0da382c HD |
524 | } |
525 | ||
42b77728 | 526 | void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma, |
3bf5ee95 | 527 | unsigned long floor, unsigned long ceiling) |
e0da382c HD |
528 | { |
529 | while (vma) { | |
530 | struct vm_area_struct *next = vma->vm_next; | |
531 | unsigned long addr = vma->vm_start; | |
532 | ||
8f4f8c16 | 533 | /* |
25d9e2d1 | 534 | * Hide vma from rmap and truncate_pagecache before freeing |
535 | * pgtables | |
8f4f8c16 | 536 | */ |
5beb4930 | 537 | unlink_anon_vmas(vma); |
8f4f8c16 HD |
538 | unlink_file_vma(vma); |
539 | ||
9da61aef | 540 | if (is_vm_hugetlb_page(vma)) { |
3bf5ee95 | 541 | hugetlb_free_pgd_range(tlb, addr, vma->vm_end, |
e0da382c | 542 | floor, next? next->vm_start: ceiling); |
3bf5ee95 HD |
543 | } else { |
544 | /* | |
545 | * Optimization: gather nearby vmas into one call down | |
546 | */ | |
547 | while (next && next->vm_start <= vma->vm_end + PMD_SIZE | |
4866920b | 548 | && !is_vm_hugetlb_page(next)) { |
3bf5ee95 HD |
549 | vma = next; |
550 | next = vma->vm_next; | |
5beb4930 | 551 | unlink_anon_vmas(vma); |
8f4f8c16 | 552 | unlink_file_vma(vma); |
3bf5ee95 HD |
553 | } |
554 | free_pgd_range(tlb, addr, vma->vm_end, | |
555 | floor, next? next->vm_start: ceiling); | |
556 | } | |
e0da382c HD |
557 | vma = next; |
558 | } | |
1da177e4 LT |
559 | } |
560 | ||
8ac1f832 AA |
561 | int __pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma, |
562 | pmd_t *pmd, unsigned long address) | |
1da177e4 | 563 | { |
c4088ebd | 564 | spinlock_t *ptl; |
2f569afd | 565 | pgtable_t new = pte_alloc_one(mm, address); |
8ac1f832 | 566 | int wait_split_huge_page; |
1bb3630e HD |
567 | if (!new) |
568 | return -ENOMEM; | |
569 | ||
362a61ad NP |
570 | /* |
571 | * Ensure all pte setup (eg. pte page lock and page clearing) are | |
572 | * visible before the pte is made visible to other CPUs by being | |
573 | * put into page tables. | |
574 | * | |
575 | * The other side of the story is the pointer chasing in the page | |
576 | * table walking code (when walking the page table without locking; | |
577 | * ie. most of the time). Fortunately, these data accesses consist | |
578 | * of a chain of data-dependent loads, meaning most CPUs (alpha | |
579 | * being the notable exception) will already guarantee loads are | |
580 | * seen in-order. See the alpha page table accessors for the | |
581 | * smp_read_barrier_depends() barriers in page table walking code. | |
582 | */ | |
583 | smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */ | |
584 | ||
c4088ebd | 585 | ptl = pmd_lock(mm, pmd); |
8ac1f832 AA |
586 | wait_split_huge_page = 0; |
587 | if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ | |
e1f56c89 | 588 | atomic_long_inc(&mm->nr_ptes); |
1da177e4 | 589 | pmd_populate(mm, pmd, new); |
2f569afd | 590 | new = NULL; |
8ac1f832 AA |
591 | } else if (unlikely(pmd_trans_splitting(*pmd))) |
592 | wait_split_huge_page = 1; | |
c4088ebd | 593 | spin_unlock(ptl); |
2f569afd MS |
594 | if (new) |
595 | pte_free(mm, new); | |
8ac1f832 AA |
596 | if (wait_split_huge_page) |
597 | wait_split_huge_page(vma->anon_vma, pmd); | |
1bb3630e | 598 | return 0; |
1da177e4 LT |
599 | } |
600 | ||
1bb3630e | 601 | int __pte_alloc_kernel(pmd_t *pmd, unsigned long address) |
1da177e4 | 602 | { |
1bb3630e HD |
603 | pte_t *new = pte_alloc_one_kernel(&init_mm, address); |
604 | if (!new) | |
605 | return -ENOMEM; | |
606 | ||
362a61ad NP |
607 | smp_wmb(); /* See comment in __pte_alloc */ |
608 | ||
1bb3630e | 609 | spin_lock(&init_mm.page_table_lock); |
8ac1f832 | 610 | if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ |
1bb3630e | 611 | pmd_populate_kernel(&init_mm, pmd, new); |
2f569afd | 612 | new = NULL; |
8ac1f832 AA |
613 | } else |
614 | VM_BUG_ON(pmd_trans_splitting(*pmd)); | |
1bb3630e | 615 | spin_unlock(&init_mm.page_table_lock); |
2f569afd MS |
616 | if (new) |
617 | pte_free_kernel(&init_mm, new); | |
1bb3630e | 618 | return 0; |
1da177e4 LT |
619 | } |
620 | ||
d559db08 KH |
621 | static inline void init_rss_vec(int *rss) |
622 | { | |
623 | memset(rss, 0, sizeof(int) * NR_MM_COUNTERS); | |
624 | } | |
625 | ||
626 | static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss) | |
ae859762 | 627 | { |
d559db08 KH |
628 | int i; |
629 | ||
34e55232 | 630 | if (current->mm == mm) |
05af2e10 | 631 | sync_mm_rss(mm); |
d559db08 KH |
632 | for (i = 0; i < NR_MM_COUNTERS; i++) |
633 | if (rss[i]) | |
634 | add_mm_counter(mm, i, rss[i]); | |
ae859762 HD |
635 | } |
636 | ||
b5810039 | 637 | /* |
6aab341e LT |
638 | * This function is called to print an error when a bad pte |
639 | * is found. For example, we might have a PFN-mapped pte in | |
640 | * a region that doesn't allow it. | |
b5810039 NP |
641 | * |
642 | * The calling function must still handle the error. | |
643 | */ | |
3dc14741 HD |
644 | static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr, |
645 | pte_t pte, struct page *page) | |
b5810039 | 646 | { |
3dc14741 HD |
647 | pgd_t *pgd = pgd_offset(vma->vm_mm, addr); |
648 | pud_t *pud = pud_offset(pgd, addr); | |
649 | pmd_t *pmd = pmd_offset(pud, addr); | |
650 | struct address_space *mapping; | |
651 | pgoff_t index; | |
d936cf9b HD |
652 | static unsigned long resume; |
653 | static unsigned long nr_shown; | |
654 | static unsigned long nr_unshown; | |
655 | ||
656 | /* | |
657 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
658 | * or allow a steady drip of one report per second. | |
659 | */ | |
660 | if (nr_shown == 60) { | |
661 | if (time_before(jiffies, resume)) { | |
662 | nr_unshown++; | |
663 | return; | |
664 | } | |
665 | if (nr_unshown) { | |
1e9e6365 HD |
666 | printk(KERN_ALERT |
667 | "BUG: Bad page map: %lu messages suppressed\n", | |
d936cf9b HD |
668 | nr_unshown); |
669 | nr_unshown = 0; | |
670 | } | |
671 | nr_shown = 0; | |
672 | } | |
673 | if (nr_shown++ == 0) | |
674 | resume = jiffies + 60 * HZ; | |
3dc14741 HD |
675 | |
676 | mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL; | |
677 | index = linear_page_index(vma, addr); | |
678 | ||
1e9e6365 HD |
679 | printk(KERN_ALERT |
680 | "BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n", | |
3dc14741 HD |
681 | current->comm, |
682 | (long long)pte_val(pte), (long long)pmd_val(*pmd)); | |
718a3821 | 683 | if (page) |
f0b791a3 | 684 | dump_page(page, "bad pte"); |
1e9e6365 | 685 | printk(KERN_ALERT |
3dc14741 HD |
686 | "addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n", |
687 | (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index); | |
688 | /* | |
689 | * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y | |
690 | */ | |
691 | if (vma->vm_ops) | |
071361d3 JP |
692 | printk(KERN_ALERT "vma->vm_ops->fault: %pSR\n", |
693 | vma->vm_ops->fault); | |
72c2d531 | 694 | if (vma->vm_file) |
071361d3 JP |
695 | printk(KERN_ALERT "vma->vm_file->f_op->mmap: %pSR\n", |
696 | vma->vm_file->f_op->mmap); | |
b5810039 | 697 | dump_stack(); |
373d4d09 | 698 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
b5810039 NP |
699 | } |
700 | ||
ee498ed7 | 701 | /* |
7e675137 | 702 | * vm_normal_page -- This function gets the "struct page" associated with a pte. |
6aab341e | 703 | * |
7e675137 NP |
704 | * "Special" mappings do not wish to be associated with a "struct page" (either |
705 | * it doesn't exist, or it exists but they don't want to touch it). In this | |
706 | * case, NULL is returned here. "Normal" mappings do have a struct page. | |
b379d790 | 707 | * |
7e675137 NP |
708 | * There are 2 broad cases. Firstly, an architecture may define a pte_special() |
709 | * pte bit, in which case this function is trivial. Secondly, an architecture | |
710 | * may not have a spare pte bit, which requires a more complicated scheme, | |
711 | * described below. | |
712 | * | |
713 | * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a | |
714 | * special mapping (even if there are underlying and valid "struct pages"). | |
715 | * COWed pages of a VM_PFNMAP are always normal. | |
6aab341e | 716 | * |
b379d790 JH |
717 | * The way we recognize COWed pages within VM_PFNMAP mappings is through the |
718 | * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit | |
7e675137 NP |
719 | * set, and the vm_pgoff will point to the first PFN mapped: thus every special |
720 | * mapping will always honor the rule | |
6aab341e LT |
721 | * |
722 | * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) | |
723 | * | |
7e675137 NP |
724 | * And for normal mappings this is false. |
725 | * | |
726 | * This restricts such mappings to be a linear translation from virtual address | |
727 | * to pfn. To get around this restriction, we allow arbitrary mappings so long | |
728 | * as the vma is not a COW mapping; in that case, we know that all ptes are | |
729 | * special (because none can have been COWed). | |
b379d790 | 730 | * |
b379d790 | 731 | * |
7e675137 | 732 | * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. |
b379d790 JH |
733 | * |
734 | * VM_MIXEDMAP mappings can likewise contain memory with or without "struct | |
735 | * page" backing, however the difference is that _all_ pages with a struct | |
736 | * page (that is, those where pfn_valid is true) are refcounted and considered | |
737 | * normal pages by the VM. The disadvantage is that pages are refcounted | |
738 | * (which can be slower and simply not an option for some PFNMAP users). The | |
739 | * advantage is that we don't have to follow the strict linearity rule of | |
740 | * PFNMAP mappings in order to support COWable mappings. | |
741 | * | |
ee498ed7 | 742 | */ |
7e675137 NP |
743 | #ifdef __HAVE_ARCH_PTE_SPECIAL |
744 | # define HAVE_PTE_SPECIAL 1 | |
745 | #else | |
746 | # define HAVE_PTE_SPECIAL 0 | |
747 | #endif | |
748 | struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, | |
749 | pte_t pte) | |
ee498ed7 | 750 | { |
22b31eec | 751 | unsigned long pfn = pte_pfn(pte); |
7e675137 NP |
752 | |
753 | if (HAVE_PTE_SPECIAL) { | |
c46a7c81 | 754 | if (likely(!pte_special(pte) || pte_numa(pte))) |
22b31eec | 755 | goto check_pfn; |
a13ea5b7 HD |
756 | if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) |
757 | return NULL; | |
62eede62 | 758 | if (!is_zero_pfn(pfn)) |
22b31eec | 759 | print_bad_pte(vma, addr, pte, NULL); |
7e675137 NP |
760 | return NULL; |
761 | } | |
762 | ||
763 | /* !HAVE_PTE_SPECIAL case follows: */ | |
764 | ||
b379d790 JH |
765 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { |
766 | if (vma->vm_flags & VM_MIXEDMAP) { | |
767 | if (!pfn_valid(pfn)) | |
768 | return NULL; | |
769 | goto out; | |
770 | } else { | |
7e675137 NP |
771 | unsigned long off; |
772 | off = (addr - vma->vm_start) >> PAGE_SHIFT; | |
b379d790 JH |
773 | if (pfn == vma->vm_pgoff + off) |
774 | return NULL; | |
775 | if (!is_cow_mapping(vma->vm_flags)) | |
776 | return NULL; | |
777 | } | |
6aab341e LT |
778 | } |
779 | ||
22b31eec HD |
780 | check_pfn: |
781 | if (unlikely(pfn > highest_memmap_pfn)) { | |
782 | print_bad_pte(vma, addr, pte, NULL); | |
783 | return NULL; | |
784 | } | |
6aab341e | 785 | |
c46a7c81 MG |
786 | if (is_zero_pfn(pfn)) |
787 | return NULL; | |
788 | ||
6aab341e | 789 | /* |
7e675137 | 790 | * NOTE! We still have PageReserved() pages in the page tables. |
7e675137 | 791 | * eg. VDSO mappings can cause them to exist. |
6aab341e | 792 | */ |
b379d790 | 793 | out: |
6aab341e | 794 | return pfn_to_page(pfn); |
ee498ed7 HD |
795 | } |
796 | ||
1da177e4 LT |
797 | /* |
798 | * copy one vm_area from one task to the other. Assumes the page tables | |
799 | * already present in the new task to be cleared in the whole range | |
800 | * covered by this vma. | |
1da177e4 LT |
801 | */ |
802 | ||
570a335b | 803 | static inline unsigned long |
1da177e4 | 804 | copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
b5810039 | 805 | pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma, |
8c103762 | 806 | unsigned long addr, int *rss) |
1da177e4 | 807 | { |
b5810039 | 808 | unsigned long vm_flags = vma->vm_flags; |
1da177e4 LT |
809 | pte_t pte = *src_pte; |
810 | struct page *page; | |
1da177e4 LT |
811 | |
812 | /* pte contains position in swap or file, so copy. */ | |
813 | if (unlikely(!pte_present(pte))) { | |
814 | if (!pte_file(pte)) { | |
0697212a CL |
815 | swp_entry_t entry = pte_to_swp_entry(pte); |
816 | ||
570a335b HD |
817 | if (swap_duplicate(entry) < 0) |
818 | return entry.val; | |
819 | ||
1da177e4 LT |
820 | /* make sure dst_mm is on swapoff's mmlist. */ |
821 | if (unlikely(list_empty(&dst_mm->mmlist))) { | |
822 | spin_lock(&mmlist_lock); | |
f412ac08 HD |
823 | if (list_empty(&dst_mm->mmlist)) |
824 | list_add(&dst_mm->mmlist, | |
825 | &src_mm->mmlist); | |
1da177e4 LT |
826 | spin_unlock(&mmlist_lock); |
827 | } | |
b084d435 KH |
828 | if (likely(!non_swap_entry(entry))) |
829 | rss[MM_SWAPENTS]++; | |
9f9f1acd KK |
830 | else if (is_migration_entry(entry)) { |
831 | page = migration_entry_to_page(entry); | |
832 | ||
833 | if (PageAnon(page)) | |
834 | rss[MM_ANONPAGES]++; | |
835 | else | |
836 | rss[MM_FILEPAGES]++; | |
837 | ||
838 | if (is_write_migration_entry(entry) && | |
839 | is_cow_mapping(vm_flags)) { | |
840 | /* | |
841 | * COW mappings require pages in both | |
842 | * parent and child to be set to read. | |
843 | */ | |
844 | make_migration_entry_read(&entry); | |
845 | pte = swp_entry_to_pte(entry); | |
c3d16e16 CG |
846 | if (pte_swp_soft_dirty(*src_pte)) |
847 | pte = pte_swp_mksoft_dirty(pte); | |
9f9f1acd KK |
848 | set_pte_at(src_mm, addr, src_pte, pte); |
849 | } | |
0697212a | 850 | } |
1da177e4 | 851 | } |
ae859762 | 852 | goto out_set_pte; |
1da177e4 LT |
853 | } |
854 | ||
1da177e4 LT |
855 | /* |
856 | * If it's a COW mapping, write protect it both | |
857 | * in the parent and the child | |
858 | */ | |
67121172 | 859 | if (is_cow_mapping(vm_flags)) { |
1da177e4 | 860 | ptep_set_wrprotect(src_mm, addr, src_pte); |
3dc90795 | 861 | pte = pte_wrprotect(pte); |
1da177e4 LT |
862 | } |
863 | ||
864 | /* | |
865 | * If it's a shared mapping, mark it clean in | |
866 | * the child | |
867 | */ | |
868 | if (vm_flags & VM_SHARED) | |
869 | pte = pte_mkclean(pte); | |
870 | pte = pte_mkold(pte); | |
6aab341e LT |
871 | |
872 | page = vm_normal_page(vma, addr, pte); | |
873 | if (page) { | |
874 | get_page(page); | |
21333b2b | 875 | page_dup_rmap(page); |
d559db08 KH |
876 | if (PageAnon(page)) |
877 | rss[MM_ANONPAGES]++; | |
878 | else | |
879 | rss[MM_FILEPAGES]++; | |
6aab341e | 880 | } |
ae859762 HD |
881 | |
882 | out_set_pte: | |
883 | set_pte_at(dst_mm, addr, dst_pte, pte); | |
570a335b | 884 | return 0; |
1da177e4 LT |
885 | } |
886 | ||
21bda264 | 887 | static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
71e3aac0 AA |
888 | pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma, |
889 | unsigned long addr, unsigned long end) | |
1da177e4 | 890 | { |
c36987e2 | 891 | pte_t *orig_src_pte, *orig_dst_pte; |
1da177e4 | 892 | pte_t *src_pte, *dst_pte; |
c74df32c | 893 | spinlock_t *src_ptl, *dst_ptl; |
e040f218 | 894 | int progress = 0; |
d559db08 | 895 | int rss[NR_MM_COUNTERS]; |
570a335b | 896 | swp_entry_t entry = (swp_entry_t){0}; |
1da177e4 LT |
897 | |
898 | again: | |
d559db08 KH |
899 | init_rss_vec(rss); |
900 | ||
c74df32c | 901 | dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); |
1da177e4 LT |
902 | if (!dst_pte) |
903 | return -ENOMEM; | |
ece0e2b6 | 904 | src_pte = pte_offset_map(src_pmd, addr); |
4c21e2f2 | 905 | src_ptl = pte_lockptr(src_mm, src_pmd); |
f20dc5f7 | 906 | spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); |
c36987e2 DN |
907 | orig_src_pte = src_pte; |
908 | orig_dst_pte = dst_pte; | |
6606c3e0 | 909 | arch_enter_lazy_mmu_mode(); |
1da177e4 | 910 | |
1da177e4 LT |
911 | do { |
912 | /* | |
913 | * We are holding two locks at this point - either of them | |
914 | * could generate latencies in another task on another CPU. | |
915 | */ | |
e040f218 HD |
916 | if (progress >= 32) { |
917 | progress = 0; | |
918 | if (need_resched() || | |
95c354fe | 919 | spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) |
e040f218 HD |
920 | break; |
921 | } | |
1da177e4 LT |
922 | if (pte_none(*src_pte)) { |
923 | progress++; | |
924 | continue; | |
925 | } | |
570a335b HD |
926 | entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, |
927 | vma, addr, rss); | |
928 | if (entry.val) | |
929 | break; | |
1da177e4 LT |
930 | progress += 8; |
931 | } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); | |
1da177e4 | 932 | |
6606c3e0 | 933 | arch_leave_lazy_mmu_mode(); |
c74df32c | 934 | spin_unlock(src_ptl); |
ece0e2b6 | 935 | pte_unmap(orig_src_pte); |
d559db08 | 936 | add_mm_rss_vec(dst_mm, rss); |
c36987e2 | 937 | pte_unmap_unlock(orig_dst_pte, dst_ptl); |
c74df32c | 938 | cond_resched(); |
570a335b HD |
939 | |
940 | if (entry.val) { | |
941 | if (add_swap_count_continuation(entry, GFP_KERNEL) < 0) | |
942 | return -ENOMEM; | |
943 | progress = 0; | |
944 | } | |
1da177e4 LT |
945 | if (addr != end) |
946 | goto again; | |
947 | return 0; | |
948 | } | |
949 | ||
950 | static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
951 | pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma, | |
952 | unsigned long addr, unsigned long end) | |
953 | { | |
954 | pmd_t *src_pmd, *dst_pmd; | |
955 | unsigned long next; | |
956 | ||
957 | dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); | |
958 | if (!dst_pmd) | |
959 | return -ENOMEM; | |
960 | src_pmd = pmd_offset(src_pud, addr); | |
961 | do { | |
962 | next = pmd_addr_end(addr, end); | |
71e3aac0 AA |
963 | if (pmd_trans_huge(*src_pmd)) { |
964 | int err; | |
14d1a55c | 965 | VM_BUG_ON(next-addr != HPAGE_PMD_SIZE); |
71e3aac0 AA |
966 | err = copy_huge_pmd(dst_mm, src_mm, |
967 | dst_pmd, src_pmd, addr, vma); | |
968 | if (err == -ENOMEM) | |
969 | return -ENOMEM; | |
970 | if (!err) | |
971 | continue; | |
972 | /* fall through */ | |
973 | } | |
1da177e4 LT |
974 | if (pmd_none_or_clear_bad(src_pmd)) |
975 | continue; | |
976 | if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd, | |
977 | vma, addr, next)) | |
978 | return -ENOMEM; | |
979 | } while (dst_pmd++, src_pmd++, addr = next, addr != end); | |
980 | return 0; | |
981 | } | |
982 | ||
983 | static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
984 | pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma, | |
985 | unsigned long addr, unsigned long end) | |
986 | { | |
987 | pud_t *src_pud, *dst_pud; | |
988 | unsigned long next; | |
989 | ||
990 | dst_pud = pud_alloc(dst_mm, dst_pgd, addr); | |
991 | if (!dst_pud) | |
992 | return -ENOMEM; | |
993 | src_pud = pud_offset(src_pgd, addr); | |
994 | do { | |
995 | next = pud_addr_end(addr, end); | |
996 | if (pud_none_or_clear_bad(src_pud)) | |
997 | continue; | |
998 | if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud, | |
999 | vma, addr, next)) | |
1000 | return -ENOMEM; | |
1001 | } while (dst_pud++, src_pud++, addr = next, addr != end); | |
1002 | return 0; | |
1003 | } | |
1004 | ||
1005 | int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
1006 | struct vm_area_struct *vma) | |
1007 | { | |
1008 | pgd_t *src_pgd, *dst_pgd; | |
1009 | unsigned long next; | |
1010 | unsigned long addr = vma->vm_start; | |
1011 | unsigned long end = vma->vm_end; | |
2ec74c3e SG |
1012 | unsigned long mmun_start; /* For mmu_notifiers */ |
1013 | unsigned long mmun_end; /* For mmu_notifiers */ | |
1014 | bool is_cow; | |
cddb8a5c | 1015 | int ret; |
1da177e4 | 1016 | |
d992895b NP |
1017 | /* |
1018 | * Don't copy ptes where a page fault will fill them correctly. | |
1019 | * Fork becomes much lighter when there are big shared or private | |
1020 | * readonly mappings. The tradeoff is that copy_page_range is more | |
1021 | * efficient than faulting. | |
1022 | */ | |
4b6e1e37 KK |
1023 | if (!(vma->vm_flags & (VM_HUGETLB | VM_NONLINEAR | |
1024 | VM_PFNMAP | VM_MIXEDMAP))) { | |
d992895b NP |
1025 | if (!vma->anon_vma) |
1026 | return 0; | |
1027 | } | |
1028 | ||
1da177e4 LT |
1029 | if (is_vm_hugetlb_page(vma)) |
1030 | return copy_hugetlb_page_range(dst_mm, src_mm, vma); | |
1031 | ||
b3b9c293 | 1032 | if (unlikely(vma->vm_flags & VM_PFNMAP)) { |
2ab64037 | 1033 | /* |
1034 | * We do not free on error cases below as remove_vma | |
1035 | * gets called on error from higher level routine | |
1036 | */ | |
5180da41 | 1037 | ret = track_pfn_copy(vma); |
2ab64037 | 1038 | if (ret) |
1039 | return ret; | |
1040 | } | |
1041 | ||
cddb8a5c AA |
1042 | /* |
1043 | * We need to invalidate the secondary MMU mappings only when | |
1044 | * there could be a permission downgrade on the ptes of the | |
1045 | * parent mm. And a permission downgrade will only happen if | |
1046 | * is_cow_mapping() returns true. | |
1047 | */ | |
2ec74c3e SG |
1048 | is_cow = is_cow_mapping(vma->vm_flags); |
1049 | mmun_start = addr; | |
1050 | mmun_end = end; | |
1051 | if (is_cow) | |
1052 | mmu_notifier_invalidate_range_start(src_mm, mmun_start, | |
1053 | mmun_end); | |
cddb8a5c AA |
1054 | |
1055 | ret = 0; | |
1da177e4 LT |
1056 | dst_pgd = pgd_offset(dst_mm, addr); |
1057 | src_pgd = pgd_offset(src_mm, addr); | |
1058 | do { | |
1059 | next = pgd_addr_end(addr, end); | |
1060 | if (pgd_none_or_clear_bad(src_pgd)) | |
1061 | continue; | |
cddb8a5c AA |
1062 | if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd, |
1063 | vma, addr, next))) { | |
1064 | ret = -ENOMEM; | |
1065 | break; | |
1066 | } | |
1da177e4 | 1067 | } while (dst_pgd++, src_pgd++, addr = next, addr != end); |
cddb8a5c | 1068 | |
2ec74c3e SG |
1069 | if (is_cow) |
1070 | mmu_notifier_invalidate_range_end(src_mm, mmun_start, mmun_end); | |
cddb8a5c | 1071 | return ret; |
1da177e4 LT |
1072 | } |
1073 | ||
51c6f666 | 1074 | static unsigned long zap_pte_range(struct mmu_gather *tlb, |
b5810039 | 1075 | struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 | 1076 | unsigned long addr, unsigned long end, |
97a89413 | 1077 | struct zap_details *details) |
1da177e4 | 1078 | { |
b5810039 | 1079 | struct mm_struct *mm = tlb->mm; |
d16dfc55 | 1080 | int force_flush = 0; |
d559db08 | 1081 | int rss[NR_MM_COUNTERS]; |
97a89413 | 1082 | spinlock_t *ptl; |
5f1a1907 | 1083 | pte_t *start_pte; |
97a89413 | 1084 | pte_t *pte; |
d559db08 | 1085 | |
d16dfc55 | 1086 | again: |
e303297e | 1087 | init_rss_vec(rss); |
5f1a1907 SR |
1088 | start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
1089 | pte = start_pte; | |
6606c3e0 | 1090 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
1091 | do { |
1092 | pte_t ptent = *pte; | |
51c6f666 | 1093 | if (pte_none(ptent)) { |
1da177e4 | 1094 | continue; |
51c6f666 | 1095 | } |
6f5e6b9e | 1096 | |
1da177e4 | 1097 | if (pte_present(ptent)) { |
ee498ed7 | 1098 | struct page *page; |
51c6f666 | 1099 | |
6aab341e | 1100 | page = vm_normal_page(vma, addr, ptent); |
1da177e4 LT |
1101 | if (unlikely(details) && page) { |
1102 | /* | |
1103 | * unmap_shared_mapping_pages() wants to | |
1104 | * invalidate cache without truncating: | |
1105 | * unmap shared but keep private pages. | |
1106 | */ | |
1107 | if (details->check_mapping && | |
1108 | details->check_mapping != page->mapping) | |
1109 | continue; | |
1110 | /* | |
1111 | * Each page->index must be checked when | |
1112 | * invalidating or truncating nonlinear. | |
1113 | */ | |
1114 | if (details->nonlinear_vma && | |
1115 | (page->index < details->first_index || | |
1116 | page->index > details->last_index)) | |
1117 | continue; | |
1118 | } | |
b5810039 | 1119 | ptent = ptep_get_and_clear_full(mm, addr, pte, |
a600388d | 1120 | tlb->fullmm); |
1da177e4 LT |
1121 | tlb_remove_tlb_entry(tlb, pte, addr); |
1122 | if (unlikely(!page)) | |
1123 | continue; | |
1124 | if (unlikely(details) && details->nonlinear_vma | |
1125 | && linear_page_index(details->nonlinear_vma, | |
41bb3476 CG |
1126 | addr) != page->index) { |
1127 | pte_t ptfile = pgoff_to_pte(page->index); | |
1128 | if (pte_soft_dirty(ptent)) | |
1129 | pte_file_mksoft_dirty(ptfile); | |
1130 | set_pte_at(mm, addr, pte, ptfile); | |
1131 | } | |
1da177e4 | 1132 | if (PageAnon(page)) |
d559db08 | 1133 | rss[MM_ANONPAGES]--; |
6237bcd9 | 1134 | else { |
1cf35d47 LT |
1135 | if (pte_dirty(ptent)) { |
1136 | force_flush = 1; | |
6237bcd9 | 1137 | set_page_dirty(page); |
1cf35d47 | 1138 | } |
4917e5d0 | 1139 | if (pte_young(ptent) && |
64363aad | 1140 | likely(!(vma->vm_flags & VM_SEQ_READ))) |
bf3f3bc5 | 1141 | mark_page_accessed(page); |
d559db08 | 1142 | rss[MM_FILEPAGES]--; |
6237bcd9 | 1143 | } |
edc315fd | 1144 | page_remove_rmap(page); |
3dc14741 HD |
1145 | if (unlikely(page_mapcount(page) < 0)) |
1146 | print_bad_pte(vma, addr, ptent, page); | |
1cf35d47 LT |
1147 | if (unlikely(!__tlb_remove_page(tlb, page))) { |
1148 | force_flush = 1; | |
d16dfc55 | 1149 | break; |
1cf35d47 | 1150 | } |
1da177e4 LT |
1151 | continue; |
1152 | } | |
1153 | /* | |
1154 | * If details->check_mapping, we leave swap entries; | |
1155 | * if details->nonlinear_vma, we leave file entries. | |
1156 | */ | |
1157 | if (unlikely(details)) | |
1158 | continue; | |
2509ef26 HD |
1159 | if (pte_file(ptent)) { |
1160 | if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) | |
1161 | print_bad_pte(vma, addr, ptent, NULL); | |
b084d435 KH |
1162 | } else { |
1163 | swp_entry_t entry = pte_to_swp_entry(ptent); | |
1164 | ||
1165 | if (!non_swap_entry(entry)) | |
1166 | rss[MM_SWAPENTS]--; | |
9f9f1acd KK |
1167 | else if (is_migration_entry(entry)) { |
1168 | struct page *page; | |
1169 | ||
1170 | page = migration_entry_to_page(entry); | |
1171 | ||
1172 | if (PageAnon(page)) | |
1173 | rss[MM_ANONPAGES]--; | |
1174 | else | |
1175 | rss[MM_FILEPAGES]--; | |
1176 | } | |
b084d435 KH |
1177 | if (unlikely(!free_swap_and_cache(entry))) |
1178 | print_bad_pte(vma, addr, ptent, NULL); | |
1179 | } | |
9888a1ca | 1180 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); |
97a89413 | 1181 | } while (pte++, addr += PAGE_SIZE, addr != end); |
ae859762 | 1182 | |
d559db08 | 1183 | add_mm_rss_vec(mm, rss); |
6606c3e0 | 1184 | arch_leave_lazy_mmu_mode(); |
51c6f666 | 1185 | |
1cf35d47 | 1186 | /* Do the actual TLB flush before dropping ptl */ |
d16dfc55 | 1187 | if (force_flush) { |
2b047252 LT |
1188 | unsigned long old_end; |
1189 | ||
2b047252 LT |
1190 | /* |
1191 | * Flush the TLB just for the previous segment, | |
1192 | * then update the range to be the remaining | |
1193 | * TLB range. | |
1194 | */ | |
1195 | old_end = tlb->end; | |
e6c495a9 | 1196 | tlb->end = addr; |
1cf35d47 | 1197 | tlb_flush_mmu_tlbonly(tlb); |
2b047252 LT |
1198 | tlb->start = addr; |
1199 | tlb->end = old_end; | |
1cf35d47 LT |
1200 | } |
1201 | pte_unmap_unlock(start_pte, ptl); | |
1202 | ||
1203 | /* | |
1204 | * If we forced a TLB flush (either due to running out of | |
1205 | * batch buffers or because we needed to flush dirty TLB | |
1206 | * entries before releasing the ptl), free the batched | |
1207 | * memory too. Restart if we didn't do everything. | |
1208 | */ | |
1209 | if (force_flush) { | |
1210 | force_flush = 0; | |
1211 | tlb_flush_mmu_free(tlb); | |
2b047252 LT |
1212 | |
1213 | if (addr != end) | |
d16dfc55 PZ |
1214 | goto again; |
1215 | } | |
1216 | ||
51c6f666 | 1217 | return addr; |
1da177e4 LT |
1218 | } |
1219 | ||
51c6f666 | 1220 | static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, |
b5810039 | 1221 | struct vm_area_struct *vma, pud_t *pud, |
1da177e4 | 1222 | unsigned long addr, unsigned long end, |
97a89413 | 1223 | struct zap_details *details) |
1da177e4 LT |
1224 | { |
1225 | pmd_t *pmd; | |
1226 | unsigned long next; | |
1227 | ||
1228 | pmd = pmd_offset(pud, addr); | |
1229 | do { | |
1230 | next = pmd_addr_end(addr, end); | |
71e3aac0 | 1231 | if (pmd_trans_huge(*pmd)) { |
1a5a9906 | 1232 | if (next - addr != HPAGE_PMD_SIZE) { |
e0897d75 DR |
1233 | #ifdef CONFIG_DEBUG_VM |
1234 | if (!rwsem_is_locked(&tlb->mm->mmap_sem)) { | |
1235 | pr_err("%s: mmap_sem is unlocked! addr=0x%lx end=0x%lx vma->vm_start=0x%lx vma->vm_end=0x%lx\n", | |
1236 | __func__, addr, end, | |
1237 | vma->vm_start, | |
1238 | vma->vm_end); | |
1239 | BUG(); | |
1240 | } | |
1241 | #endif | |
e180377f | 1242 | split_huge_page_pmd(vma, addr, pmd); |
f21760b1 | 1243 | } else if (zap_huge_pmd(tlb, vma, pmd, addr)) |
1a5a9906 | 1244 | goto next; |
71e3aac0 AA |
1245 | /* fall through */ |
1246 | } | |
1a5a9906 AA |
1247 | /* |
1248 | * Here there can be other concurrent MADV_DONTNEED or | |
1249 | * trans huge page faults running, and if the pmd is | |
1250 | * none or trans huge it can change under us. This is | |
1251 | * because MADV_DONTNEED holds the mmap_sem in read | |
1252 | * mode. | |
1253 | */ | |
1254 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) | |
1255 | goto next; | |
97a89413 | 1256 | next = zap_pte_range(tlb, vma, pmd, addr, next, details); |
1a5a9906 | 1257 | next: |
97a89413 PZ |
1258 | cond_resched(); |
1259 | } while (pmd++, addr = next, addr != end); | |
51c6f666 RH |
1260 | |
1261 | return addr; | |
1da177e4 LT |
1262 | } |
1263 | ||
51c6f666 | 1264 | static inline unsigned long zap_pud_range(struct mmu_gather *tlb, |
b5810039 | 1265 | struct vm_area_struct *vma, pgd_t *pgd, |
1da177e4 | 1266 | unsigned long addr, unsigned long end, |
97a89413 | 1267 | struct zap_details *details) |
1da177e4 LT |
1268 | { |
1269 | pud_t *pud; | |
1270 | unsigned long next; | |
1271 | ||
1272 | pud = pud_offset(pgd, addr); | |
1273 | do { | |
1274 | next = pud_addr_end(addr, end); | |
97a89413 | 1275 | if (pud_none_or_clear_bad(pud)) |
1da177e4 | 1276 | continue; |
97a89413 PZ |
1277 | next = zap_pmd_range(tlb, vma, pud, addr, next, details); |
1278 | } while (pud++, addr = next, addr != end); | |
51c6f666 RH |
1279 | |
1280 | return addr; | |
1da177e4 LT |
1281 | } |
1282 | ||
038c7aa1 AV |
1283 | static void unmap_page_range(struct mmu_gather *tlb, |
1284 | struct vm_area_struct *vma, | |
1285 | unsigned long addr, unsigned long end, | |
1286 | struct zap_details *details) | |
1da177e4 LT |
1287 | { |
1288 | pgd_t *pgd; | |
1289 | unsigned long next; | |
1290 | ||
1291 | if (details && !details->check_mapping && !details->nonlinear_vma) | |
1292 | details = NULL; | |
1293 | ||
1294 | BUG_ON(addr >= end); | |
569b846d | 1295 | mem_cgroup_uncharge_start(); |
1da177e4 LT |
1296 | tlb_start_vma(tlb, vma); |
1297 | pgd = pgd_offset(vma->vm_mm, addr); | |
1298 | do { | |
1299 | next = pgd_addr_end(addr, end); | |
97a89413 | 1300 | if (pgd_none_or_clear_bad(pgd)) |
1da177e4 | 1301 | continue; |
97a89413 PZ |
1302 | next = zap_pud_range(tlb, vma, pgd, addr, next, details); |
1303 | } while (pgd++, addr = next, addr != end); | |
1da177e4 | 1304 | tlb_end_vma(tlb, vma); |
569b846d | 1305 | mem_cgroup_uncharge_end(); |
1da177e4 | 1306 | } |
51c6f666 | 1307 | |
f5cc4eef AV |
1308 | |
1309 | static void unmap_single_vma(struct mmu_gather *tlb, | |
1310 | struct vm_area_struct *vma, unsigned long start_addr, | |
4f74d2c8 | 1311 | unsigned long end_addr, |
f5cc4eef AV |
1312 | struct zap_details *details) |
1313 | { | |
1314 | unsigned long start = max(vma->vm_start, start_addr); | |
1315 | unsigned long end; | |
1316 | ||
1317 | if (start >= vma->vm_end) | |
1318 | return; | |
1319 | end = min(vma->vm_end, end_addr); | |
1320 | if (end <= vma->vm_start) | |
1321 | return; | |
1322 | ||
cbc91f71 SD |
1323 | if (vma->vm_file) |
1324 | uprobe_munmap(vma, start, end); | |
1325 | ||
b3b9c293 | 1326 | if (unlikely(vma->vm_flags & VM_PFNMAP)) |
5180da41 | 1327 | untrack_pfn(vma, 0, 0); |
f5cc4eef AV |
1328 | |
1329 | if (start != end) { | |
1330 | if (unlikely(is_vm_hugetlb_page(vma))) { | |
1331 | /* | |
1332 | * It is undesirable to test vma->vm_file as it | |
1333 | * should be non-null for valid hugetlb area. | |
1334 | * However, vm_file will be NULL in the error | |
7aa6b4ad | 1335 | * cleanup path of mmap_region. When |
f5cc4eef | 1336 | * hugetlbfs ->mmap method fails, |
7aa6b4ad | 1337 | * mmap_region() nullifies vma->vm_file |
f5cc4eef AV |
1338 | * before calling this function to clean up. |
1339 | * Since no pte has actually been setup, it is | |
1340 | * safe to do nothing in this case. | |
1341 | */ | |
24669e58 AK |
1342 | if (vma->vm_file) { |
1343 | mutex_lock(&vma->vm_file->f_mapping->i_mmap_mutex); | |
d833352a | 1344 | __unmap_hugepage_range_final(tlb, vma, start, end, NULL); |
24669e58 AK |
1345 | mutex_unlock(&vma->vm_file->f_mapping->i_mmap_mutex); |
1346 | } | |
f5cc4eef AV |
1347 | } else |
1348 | unmap_page_range(tlb, vma, start, end, details); | |
1349 | } | |
1da177e4 LT |
1350 | } |
1351 | ||
1da177e4 LT |
1352 | /** |
1353 | * unmap_vmas - unmap a range of memory covered by a list of vma's | |
0164f69d | 1354 | * @tlb: address of the caller's struct mmu_gather |
1da177e4 LT |
1355 | * @vma: the starting vma |
1356 | * @start_addr: virtual address at which to start unmapping | |
1357 | * @end_addr: virtual address at which to end unmapping | |
1da177e4 | 1358 | * |
508034a3 | 1359 | * Unmap all pages in the vma list. |
1da177e4 | 1360 | * |
1da177e4 LT |
1361 | * Only addresses between `start' and `end' will be unmapped. |
1362 | * | |
1363 | * The VMA list must be sorted in ascending virtual address order. | |
1364 | * | |
1365 | * unmap_vmas() assumes that the caller will flush the whole unmapped address | |
1366 | * range after unmap_vmas() returns. So the only responsibility here is to | |
1367 | * ensure that any thus-far unmapped pages are flushed before unmap_vmas() | |
1368 | * drops the lock and schedules. | |
1369 | */ | |
6e8bb019 | 1370 | void unmap_vmas(struct mmu_gather *tlb, |
1da177e4 | 1371 | struct vm_area_struct *vma, unsigned long start_addr, |
4f74d2c8 | 1372 | unsigned long end_addr) |
1da177e4 | 1373 | { |
cddb8a5c | 1374 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 1375 | |
cddb8a5c | 1376 | mmu_notifier_invalidate_range_start(mm, start_addr, end_addr); |
f5cc4eef | 1377 | for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) |
4f74d2c8 | 1378 | unmap_single_vma(tlb, vma, start_addr, end_addr, NULL); |
cddb8a5c | 1379 | mmu_notifier_invalidate_range_end(mm, start_addr, end_addr); |
1da177e4 LT |
1380 | } |
1381 | ||
1382 | /** | |
1383 | * zap_page_range - remove user pages in a given range | |
1384 | * @vma: vm_area_struct holding the applicable pages | |
eb4546bb | 1385 | * @start: starting address of pages to zap |
1da177e4 LT |
1386 | * @size: number of bytes to zap |
1387 | * @details: details of nonlinear truncation or shared cache invalidation | |
f5cc4eef AV |
1388 | * |
1389 | * Caller must protect the VMA list | |
1da177e4 | 1390 | */ |
7e027b14 | 1391 | void zap_page_range(struct vm_area_struct *vma, unsigned long start, |
1da177e4 LT |
1392 | unsigned long size, struct zap_details *details) |
1393 | { | |
1394 | struct mm_struct *mm = vma->vm_mm; | |
d16dfc55 | 1395 | struct mmu_gather tlb; |
7e027b14 | 1396 | unsigned long end = start + size; |
1da177e4 | 1397 | |
1da177e4 | 1398 | lru_add_drain(); |
2b047252 | 1399 | tlb_gather_mmu(&tlb, mm, start, end); |
365e9c87 | 1400 | update_hiwater_rss(mm); |
7e027b14 LT |
1401 | mmu_notifier_invalidate_range_start(mm, start, end); |
1402 | for ( ; vma && vma->vm_start < end; vma = vma->vm_next) | |
4f74d2c8 | 1403 | unmap_single_vma(&tlb, vma, start, end, details); |
7e027b14 LT |
1404 | mmu_notifier_invalidate_range_end(mm, start, end); |
1405 | tlb_finish_mmu(&tlb, start, end); | |
1da177e4 LT |
1406 | } |
1407 | ||
f5cc4eef AV |
1408 | /** |
1409 | * zap_page_range_single - remove user pages in a given range | |
1410 | * @vma: vm_area_struct holding the applicable pages | |
1411 | * @address: starting address of pages to zap | |
1412 | * @size: number of bytes to zap | |
1413 | * @details: details of nonlinear truncation or shared cache invalidation | |
1414 | * | |
1415 | * The range must fit into one VMA. | |
1da177e4 | 1416 | */ |
f5cc4eef | 1417 | static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address, |
1da177e4 LT |
1418 | unsigned long size, struct zap_details *details) |
1419 | { | |
1420 | struct mm_struct *mm = vma->vm_mm; | |
d16dfc55 | 1421 | struct mmu_gather tlb; |
1da177e4 | 1422 | unsigned long end = address + size; |
1da177e4 | 1423 | |
1da177e4 | 1424 | lru_add_drain(); |
2b047252 | 1425 | tlb_gather_mmu(&tlb, mm, address, end); |
365e9c87 | 1426 | update_hiwater_rss(mm); |
f5cc4eef | 1427 | mmu_notifier_invalidate_range_start(mm, address, end); |
4f74d2c8 | 1428 | unmap_single_vma(&tlb, vma, address, end, details); |
f5cc4eef | 1429 | mmu_notifier_invalidate_range_end(mm, address, end); |
d16dfc55 | 1430 | tlb_finish_mmu(&tlb, address, end); |
1da177e4 LT |
1431 | } |
1432 | ||
c627f9cc JS |
1433 | /** |
1434 | * zap_vma_ptes - remove ptes mapping the vma | |
1435 | * @vma: vm_area_struct holding ptes to be zapped | |
1436 | * @address: starting address of pages to zap | |
1437 | * @size: number of bytes to zap | |
1438 | * | |
1439 | * This function only unmaps ptes assigned to VM_PFNMAP vmas. | |
1440 | * | |
1441 | * The entire address range must be fully contained within the vma. | |
1442 | * | |
1443 | * Returns 0 if successful. | |
1444 | */ | |
1445 | int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, | |
1446 | unsigned long size) | |
1447 | { | |
1448 | if (address < vma->vm_start || address + size > vma->vm_end || | |
1449 | !(vma->vm_flags & VM_PFNMAP)) | |
1450 | return -1; | |
f5cc4eef | 1451 | zap_page_range_single(vma, address, size, NULL); |
c627f9cc JS |
1452 | return 0; |
1453 | } | |
1454 | EXPORT_SYMBOL_GPL(zap_vma_ptes); | |
1455 | ||
25ca1d6c | 1456 | pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr, |
920c7a5d | 1457 | spinlock_t **ptl) |
c9cfcddf LT |
1458 | { |
1459 | pgd_t * pgd = pgd_offset(mm, addr); | |
1460 | pud_t * pud = pud_alloc(mm, pgd, addr); | |
1461 | if (pud) { | |
49c91fb0 | 1462 | pmd_t * pmd = pmd_alloc(mm, pud, addr); |
f66055ab AA |
1463 | if (pmd) { |
1464 | VM_BUG_ON(pmd_trans_huge(*pmd)); | |
c9cfcddf | 1465 | return pte_alloc_map_lock(mm, pmd, addr, ptl); |
f66055ab | 1466 | } |
c9cfcddf LT |
1467 | } |
1468 | return NULL; | |
1469 | } | |
1470 | ||
238f58d8 LT |
1471 | /* |
1472 | * This is the old fallback for page remapping. | |
1473 | * | |
1474 | * For historical reasons, it only allows reserved pages. Only | |
1475 | * old drivers should use this, and they needed to mark their | |
1476 | * pages reserved for the old functions anyway. | |
1477 | */ | |
423bad60 NP |
1478 | static int insert_page(struct vm_area_struct *vma, unsigned long addr, |
1479 | struct page *page, pgprot_t prot) | |
238f58d8 | 1480 | { |
423bad60 | 1481 | struct mm_struct *mm = vma->vm_mm; |
238f58d8 | 1482 | int retval; |
c9cfcddf | 1483 | pte_t *pte; |
8a9f3ccd BS |
1484 | spinlock_t *ptl; |
1485 | ||
238f58d8 | 1486 | retval = -EINVAL; |
a145dd41 | 1487 | if (PageAnon(page)) |
5b4e655e | 1488 | goto out; |
238f58d8 LT |
1489 | retval = -ENOMEM; |
1490 | flush_dcache_page(page); | |
c9cfcddf | 1491 | pte = get_locked_pte(mm, addr, &ptl); |
238f58d8 | 1492 | if (!pte) |
5b4e655e | 1493 | goto out; |
238f58d8 LT |
1494 | retval = -EBUSY; |
1495 | if (!pte_none(*pte)) | |
1496 | goto out_unlock; | |
1497 | ||
1498 | /* Ok, finally just insert the thing.. */ | |
1499 | get_page(page); | |
34e55232 | 1500 | inc_mm_counter_fast(mm, MM_FILEPAGES); |
238f58d8 LT |
1501 | page_add_file_rmap(page); |
1502 | set_pte_at(mm, addr, pte, mk_pte(page, prot)); | |
1503 | ||
1504 | retval = 0; | |
8a9f3ccd BS |
1505 | pte_unmap_unlock(pte, ptl); |
1506 | return retval; | |
238f58d8 LT |
1507 | out_unlock: |
1508 | pte_unmap_unlock(pte, ptl); | |
1509 | out: | |
1510 | return retval; | |
1511 | } | |
1512 | ||
bfa5bf6d REB |
1513 | /** |
1514 | * vm_insert_page - insert single page into user vma | |
1515 | * @vma: user vma to map to | |
1516 | * @addr: target user address of this page | |
1517 | * @page: source kernel page | |
1518 | * | |
a145dd41 LT |
1519 | * This allows drivers to insert individual pages they've allocated |
1520 | * into a user vma. | |
1521 | * | |
1522 | * The page has to be a nice clean _individual_ kernel allocation. | |
1523 | * If you allocate a compound page, you need to have marked it as | |
1524 | * such (__GFP_COMP), or manually just split the page up yourself | |
8dfcc9ba | 1525 | * (see split_page()). |
a145dd41 LT |
1526 | * |
1527 | * NOTE! Traditionally this was done with "remap_pfn_range()" which | |
1528 | * took an arbitrary page protection parameter. This doesn't allow | |
1529 | * that. Your vma protection will have to be set up correctly, which | |
1530 | * means that if you want a shared writable mapping, you'd better | |
1531 | * ask for a shared writable mapping! | |
1532 | * | |
1533 | * The page does not need to be reserved. | |
4b6e1e37 KK |
1534 | * |
1535 | * Usually this function is called from f_op->mmap() handler | |
1536 | * under mm->mmap_sem write-lock, so it can change vma->vm_flags. | |
1537 | * Caller must set VM_MIXEDMAP on vma if it wants to call this | |
1538 | * function from other places, for example from page-fault handler. | |
a145dd41 | 1539 | */ |
423bad60 NP |
1540 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, |
1541 | struct page *page) | |
a145dd41 LT |
1542 | { |
1543 | if (addr < vma->vm_start || addr >= vma->vm_end) | |
1544 | return -EFAULT; | |
1545 | if (!page_count(page)) | |
1546 | return -EINVAL; | |
4b6e1e37 KK |
1547 | if (!(vma->vm_flags & VM_MIXEDMAP)) { |
1548 | BUG_ON(down_read_trylock(&vma->vm_mm->mmap_sem)); | |
1549 | BUG_ON(vma->vm_flags & VM_PFNMAP); | |
1550 | vma->vm_flags |= VM_MIXEDMAP; | |
1551 | } | |
423bad60 | 1552 | return insert_page(vma, addr, page, vma->vm_page_prot); |
a145dd41 | 1553 | } |
e3c3374f | 1554 | EXPORT_SYMBOL(vm_insert_page); |
a145dd41 | 1555 | |
423bad60 NP |
1556 | static int insert_pfn(struct vm_area_struct *vma, unsigned long addr, |
1557 | unsigned long pfn, pgprot_t prot) | |
1558 | { | |
1559 | struct mm_struct *mm = vma->vm_mm; | |
1560 | int retval; | |
1561 | pte_t *pte, entry; | |
1562 | spinlock_t *ptl; | |
1563 | ||
1564 | retval = -ENOMEM; | |
1565 | pte = get_locked_pte(mm, addr, &ptl); | |
1566 | if (!pte) | |
1567 | goto out; | |
1568 | retval = -EBUSY; | |
1569 | if (!pte_none(*pte)) | |
1570 | goto out_unlock; | |
1571 | ||
1572 | /* Ok, finally just insert the thing.. */ | |
1573 | entry = pte_mkspecial(pfn_pte(pfn, prot)); | |
1574 | set_pte_at(mm, addr, pte, entry); | |
4b3073e1 | 1575 | update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */ |
423bad60 NP |
1576 | |
1577 | retval = 0; | |
1578 | out_unlock: | |
1579 | pte_unmap_unlock(pte, ptl); | |
1580 | out: | |
1581 | return retval; | |
1582 | } | |
1583 | ||
e0dc0d8f NP |
1584 | /** |
1585 | * vm_insert_pfn - insert single pfn into user vma | |
1586 | * @vma: user vma to map to | |
1587 | * @addr: target user address of this page | |
1588 | * @pfn: source kernel pfn | |
1589 | * | |
c462f179 | 1590 | * Similar to vm_insert_page, this allows drivers to insert individual pages |
e0dc0d8f NP |
1591 | * they've allocated into a user vma. Same comments apply. |
1592 | * | |
1593 | * This function should only be called from a vm_ops->fault handler, and | |
1594 | * in that case the handler should return NULL. | |
0d71d10a NP |
1595 | * |
1596 | * vma cannot be a COW mapping. | |
1597 | * | |
1598 | * As this is called only for pages that do not currently exist, we | |
1599 | * do not need to flush old virtual caches or the TLB. | |
e0dc0d8f NP |
1600 | */ |
1601 | int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr, | |
423bad60 | 1602 | unsigned long pfn) |
e0dc0d8f | 1603 | { |
2ab64037 | 1604 | int ret; |
e4b866ed | 1605 | pgprot_t pgprot = vma->vm_page_prot; |
7e675137 NP |
1606 | /* |
1607 | * Technically, architectures with pte_special can avoid all these | |
1608 | * restrictions (same for remap_pfn_range). However we would like | |
1609 | * consistency in testing and feature parity among all, so we should | |
1610 | * try to keep these invariants in place for everybody. | |
1611 | */ | |
b379d790 JH |
1612 | BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); |
1613 | BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == | |
1614 | (VM_PFNMAP|VM_MIXEDMAP)); | |
1615 | BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); | |
1616 | BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); | |
e0dc0d8f | 1617 | |
423bad60 NP |
1618 | if (addr < vma->vm_start || addr >= vma->vm_end) |
1619 | return -EFAULT; | |
5180da41 | 1620 | if (track_pfn_insert(vma, &pgprot, pfn)) |
2ab64037 | 1621 | return -EINVAL; |
1622 | ||
e4b866ed | 1623 | ret = insert_pfn(vma, addr, pfn, pgprot); |
2ab64037 | 1624 | |
2ab64037 | 1625 | return ret; |
423bad60 NP |
1626 | } |
1627 | EXPORT_SYMBOL(vm_insert_pfn); | |
e0dc0d8f | 1628 | |
423bad60 NP |
1629 | int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr, |
1630 | unsigned long pfn) | |
1631 | { | |
1632 | BUG_ON(!(vma->vm_flags & VM_MIXEDMAP)); | |
e0dc0d8f | 1633 | |
423bad60 NP |
1634 | if (addr < vma->vm_start || addr >= vma->vm_end) |
1635 | return -EFAULT; | |
e0dc0d8f | 1636 | |
423bad60 NP |
1637 | /* |
1638 | * If we don't have pte special, then we have to use the pfn_valid() | |
1639 | * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* | |
1640 | * refcount the page if pfn_valid is true (hence insert_page rather | |
62eede62 HD |
1641 | * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP |
1642 | * without pte special, it would there be refcounted as a normal page. | |
423bad60 NP |
1643 | */ |
1644 | if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) { | |
1645 | struct page *page; | |
1646 | ||
1647 | page = pfn_to_page(pfn); | |
1648 | return insert_page(vma, addr, page, vma->vm_page_prot); | |
1649 | } | |
1650 | return insert_pfn(vma, addr, pfn, vma->vm_page_prot); | |
e0dc0d8f | 1651 | } |
423bad60 | 1652 | EXPORT_SYMBOL(vm_insert_mixed); |
e0dc0d8f | 1653 | |
1da177e4 LT |
1654 | /* |
1655 | * maps a range of physical memory into the requested pages. the old | |
1656 | * mappings are removed. any references to nonexistent pages results | |
1657 | * in null mappings (currently treated as "copy-on-access") | |
1658 | */ | |
1659 | static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, | |
1660 | unsigned long addr, unsigned long end, | |
1661 | unsigned long pfn, pgprot_t prot) | |
1662 | { | |
1663 | pte_t *pte; | |
c74df32c | 1664 | spinlock_t *ptl; |
1da177e4 | 1665 | |
c74df32c | 1666 | pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); |
1da177e4 LT |
1667 | if (!pte) |
1668 | return -ENOMEM; | |
6606c3e0 | 1669 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
1670 | do { |
1671 | BUG_ON(!pte_none(*pte)); | |
7e675137 | 1672 | set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); |
1da177e4 LT |
1673 | pfn++; |
1674 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
6606c3e0 | 1675 | arch_leave_lazy_mmu_mode(); |
c74df32c | 1676 | pte_unmap_unlock(pte - 1, ptl); |
1da177e4 LT |
1677 | return 0; |
1678 | } | |
1679 | ||
1680 | static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, | |
1681 | unsigned long addr, unsigned long end, | |
1682 | unsigned long pfn, pgprot_t prot) | |
1683 | { | |
1684 | pmd_t *pmd; | |
1685 | unsigned long next; | |
1686 | ||
1687 | pfn -= addr >> PAGE_SHIFT; | |
1688 | pmd = pmd_alloc(mm, pud, addr); | |
1689 | if (!pmd) | |
1690 | return -ENOMEM; | |
f66055ab | 1691 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
1da177e4 LT |
1692 | do { |
1693 | next = pmd_addr_end(addr, end); | |
1694 | if (remap_pte_range(mm, pmd, addr, next, | |
1695 | pfn + (addr >> PAGE_SHIFT), prot)) | |
1696 | return -ENOMEM; | |
1697 | } while (pmd++, addr = next, addr != end); | |
1698 | return 0; | |
1699 | } | |
1700 | ||
1701 | static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd, | |
1702 | unsigned long addr, unsigned long end, | |
1703 | unsigned long pfn, pgprot_t prot) | |
1704 | { | |
1705 | pud_t *pud; | |
1706 | unsigned long next; | |
1707 | ||
1708 | pfn -= addr >> PAGE_SHIFT; | |
1709 | pud = pud_alloc(mm, pgd, addr); | |
1710 | if (!pud) | |
1711 | return -ENOMEM; | |
1712 | do { | |
1713 | next = pud_addr_end(addr, end); | |
1714 | if (remap_pmd_range(mm, pud, addr, next, | |
1715 | pfn + (addr >> PAGE_SHIFT), prot)) | |
1716 | return -ENOMEM; | |
1717 | } while (pud++, addr = next, addr != end); | |
1718 | return 0; | |
1719 | } | |
1720 | ||
bfa5bf6d REB |
1721 | /** |
1722 | * remap_pfn_range - remap kernel memory to userspace | |
1723 | * @vma: user vma to map to | |
1724 | * @addr: target user address to start at | |
1725 | * @pfn: physical address of kernel memory | |
1726 | * @size: size of map area | |
1727 | * @prot: page protection flags for this mapping | |
1728 | * | |
1729 | * Note: this is only safe if the mm semaphore is held when called. | |
1730 | */ | |
1da177e4 LT |
1731 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, |
1732 | unsigned long pfn, unsigned long size, pgprot_t prot) | |
1733 | { | |
1734 | pgd_t *pgd; | |
1735 | unsigned long next; | |
2d15cab8 | 1736 | unsigned long end = addr + PAGE_ALIGN(size); |
1da177e4 LT |
1737 | struct mm_struct *mm = vma->vm_mm; |
1738 | int err; | |
1739 | ||
1740 | /* | |
1741 | * Physically remapped pages are special. Tell the | |
1742 | * rest of the world about it: | |
1743 | * VM_IO tells people not to look at these pages | |
1744 | * (accesses can have side effects). | |
6aab341e LT |
1745 | * VM_PFNMAP tells the core MM that the base pages are just |
1746 | * raw PFN mappings, and do not have a "struct page" associated | |
1747 | * with them. | |
314e51b9 KK |
1748 | * VM_DONTEXPAND |
1749 | * Disable vma merging and expanding with mremap(). | |
1750 | * VM_DONTDUMP | |
1751 | * Omit vma from core dump, even when VM_IO turned off. | |
fb155c16 LT |
1752 | * |
1753 | * There's a horrible special case to handle copy-on-write | |
1754 | * behaviour that some programs depend on. We mark the "original" | |
1755 | * un-COW'ed pages by matching them up with "vma->vm_pgoff". | |
b3b9c293 | 1756 | * See vm_normal_page() for details. |
1da177e4 | 1757 | */ |
b3b9c293 KK |
1758 | if (is_cow_mapping(vma->vm_flags)) { |
1759 | if (addr != vma->vm_start || end != vma->vm_end) | |
1760 | return -EINVAL; | |
fb155c16 | 1761 | vma->vm_pgoff = pfn; |
b3b9c293 KK |
1762 | } |
1763 | ||
1764 | err = track_pfn_remap(vma, &prot, pfn, addr, PAGE_ALIGN(size)); | |
1765 | if (err) | |
3c8bb73a | 1766 | return -EINVAL; |
fb155c16 | 1767 | |
314e51b9 | 1768 | vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; |
1da177e4 LT |
1769 | |
1770 | BUG_ON(addr >= end); | |
1771 | pfn -= addr >> PAGE_SHIFT; | |
1772 | pgd = pgd_offset(mm, addr); | |
1773 | flush_cache_range(vma, addr, end); | |
1da177e4 LT |
1774 | do { |
1775 | next = pgd_addr_end(addr, end); | |
1776 | err = remap_pud_range(mm, pgd, addr, next, | |
1777 | pfn + (addr >> PAGE_SHIFT), prot); | |
1778 | if (err) | |
1779 | break; | |
1780 | } while (pgd++, addr = next, addr != end); | |
2ab64037 | 1781 | |
1782 | if (err) | |
5180da41 | 1783 | untrack_pfn(vma, pfn, PAGE_ALIGN(size)); |
2ab64037 | 1784 | |
1da177e4 LT |
1785 | return err; |
1786 | } | |
1787 | EXPORT_SYMBOL(remap_pfn_range); | |
1788 | ||
b4cbb197 LT |
1789 | /** |
1790 | * vm_iomap_memory - remap memory to userspace | |
1791 | * @vma: user vma to map to | |
1792 | * @start: start of area | |
1793 | * @len: size of area | |
1794 | * | |
1795 | * This is a simplified io_remap_pfn_range() for common driver use. The | |
1796 | * driver just needs to give us the physical memory range to be mapped, | |
1797 | * we'll figure out the rest from the vma information. | |
1798 | * | |
1799 | * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get | |
1800 | * whatever write-combining details or similar. | |
1801 | */ | |
1802 | int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) | |
1803 | { | |
1804 | unsigned long vm_len, pfn, pages; | |
1805 | ||
1806 | /* Check that the physical memory area passed in looks valid */ | |
1807 | if (start + len < start) | |
1808 | return -EINVAL; | |
1809 | /* | |
1810 | * You *really* shouldn't map things that aren't page-aligned, | |
1811 | * but we've historically allowed it because IO memory might | |
1812 | * just have smaller alignment. | |
1813 | */ | |
1814 | len += start & ~PAGE_MASK; | |
1815 | pfn = start >> PAGE_SHIFT; | |
1816 | pages = (len + ~PAGE_MASK) >> PAGE_SHIFT; | |
1817 | if (pfn + pages < pfn) | |
1818 | return -EINVAL; | |
1819 | ||
1820 | /* We start the mapping 'vm_pgoff' pages into the area */ | |
1821 | if (vma->vm_pgoff > pages) | |
1822 | return -EINVAL; | |
1823 | pfn += vma->vm_pgoff; | |
1824 | pages -= vma->vm_pgoff; | |
1825 | ||
1826 | /* Can we fit all of the mapping? */ | |
1827 | vm_len = vma->vm_end - vma->vm_start; | |
1828 | if (vm_len >> PAGE_SHIFT > pages) | |
1829 | return -EINVAL; | |
1830 | ||
1831 | /* Ok, let it rip */ | |
1832 | return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); | |
1833 | } | |
1834 | EXPORT_SYMBOL(vm_iomap_memory); | |
1835 | ||
aee16b3c JF |
1836 | static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, |
1837 | unsigned long addr, unsigned long end, | |
1838 | pte_fn_t fn, void *data) | |
1839 | { | |
1840 | pte_t *pte; | |
1841 | int err; | |
2f569afd | 1842 | pgtable_t token; |
94909914 | 1843 | spinlock_t *uninitialized_var(ptl); |
aee16b3c JF |
1844 | |
1845 | pte = (mm == &init_mm) ? | |
1846 | pte_alloc_kernel(pmd, addr) : | |
1847 | pte_alloc_map_lock(mm, pmd, addr, &ptl); | |
1848 | if (!pte) | |
1849 | return -ENOMEM; | |
1850 | ||
1851 | BUG_ON(pmd_huge(*pmd)); | |
1852 | ||
38e0edb1 JF |
1853 | arch_enter_lazy_mmu_mode(); |
1854 | ||
2f569afd | 1855 | token = pmd_pgtable(*pmd); |
aee16b3c JF |
1856 | |
1857 | do { | |
c36987e2 | 1858 | err = fn(pte++, token, addr, data); |
aee16b3c JF |
1859 | if (err) |
1860 | break; | |
c36987e2 | 1861 | } while (addr += PAGE_SIZE, addr != end); |
aee16b3c | 1862 | |
38e0edb1 JF |
1863 | arch_leave_lazy_mmu_mode(); |
1864 | ||
aee16b3c JF |
1865 | if (mm != &init_mm) |
1866 | pte_unmap_unlock(pte-1, ptl); | |
1867 | return err; | |
1868 | } | |
1869 | ||
1870 | static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, | |
1871 | unsigned long addr, unsigned long end, | |
1872 | pte_fn_t fn, void *data) | |
1873 | { | |
1874 | pmd_t *pmd; | |
1875 | unsigned long next; | |
1876 | int err; | |
1877 | ||
ceb86879 AK |
1878 | BUG_ON(pud_huge(*pud)); |
1879 | ||
aee16b3c JF |
1880 | pmd = pmd_alloc(mm, pud, addr); |
1881 | if (!pmd) | |
1882 | return -ENOMEM; | |
1883 | do { | |
1884 | next = pmd_addr_end(addr, end); | |
1885 | err = apply_to_pte_range(mm, pmd, addr, next, fn, data); | |
1886 | if (err) | |
1887 | break; | |
1888 | } while (pmd++, addr = next, addr != end); | |
1889 | return err; | |
1890 | } | |
1891 | ||
1892 | static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd, | |
1893 | unsigned long addr, unsigned long end, | |
1894 | pte_fn_t fn, void *data) | |
1895 | { | |
1896 | pud_t *pud; | |
1897 | unsigned long next; | |
1898 | int err; | |
1899 | ||
1900 | pud = pud_alloc(mm, pgd, addr); | |
1901 | if (!pud) | |
1902 | return -ENOMEM; | |
1903 | do { | |
1904 | next = pud_addr_end(addr, end); | |
1905 | err = apply_to_pmd_range(mm, pud, addr, next, fn, data); | |
1906 | if (err) | |
1907 | break; | |
1908 | } while (pud++, addr = next, addr != end); | |
1909 | return err; | |
1910 | } | |
1911 | ||
1912 | /* | |
1913 | * Scan a region of virtual memory, filling in page tables as necessary | |
1914 | * and calling a provided function on each leaf page table. | |
1915 | */ | |
1916 | int apply_to_page_range(struct mm_struct *mm, unsigned long addr, | |
1917 | unsigned long size, pte_fn_t fn, void *data) | |
1918 | { | |
1919 | pgd_t *pgd; | |
1920 | unsigned long next; | |
57250a5b | 1921 | unsigned long end = addr + size; |
aee16b3c JF |
1922 | int err; |
1923 | ||
1924 | BUG_ON(addr >= end); | |
1925 | pgd = pgd_offset(mm, addr); | |
1926 | do { | |
1927 | next = pgd_addr_end(addr, end); | |
1928 | err = apply_to_pud_range(mm, pgd, addr, next, fn, data); | |
1929 | if (err) | |
1930 | break; | |
1931 | } while (pgd++, addr = next, addr != end); | |
57250a5b | 1932 | |
aee16b3c JF |
1933 | return err; |
1934 | } | |
1935 | EXPORT_SYMBOL_GPL(apply_to_page_range); | |
1936 | ||
8f4e2101 HD |
1937 | /* |
1938 | * handle_pte_fault chooses page fault handler according to an entry | |
1939 | * which was read non-atomically. Before making any commitment, on | |
1940 | * those architectures or configurations (e.g. i386 with PAE) which | |
a335b2e1 | 1941 | * might give a mix of unmatched parts, do_swap_page and do_nonlinear_fault |
8f4e2101 HD |
1942 | * must check under lock before unmapping the pte and proceeding |
1943 | * (but do_wp_page is only called after already making such a check; | |
a335b2e1 | 1944 | * and do_anonymous_page can safely check later on). |
8f4e2101 | 1945 | */ |
4c21e2f2 | 1946 | static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd, |
8f4e2101 HD |
1947 | pte_t *page_table, pte_t orig_pte) |
1948 | { | |
1949 | int same = 1; | |
1950 | #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT) | |
1951 | if (sizeof(pte_t) > sizeof(unsigned long)) { | |
4c21e2f2 HD |
1952 | spinlock_t *ptl = pte_lockptr(mm, pmd); |
1953 | spin_lock(ptl); | |
8f4e2101 | 1954 | same = pte_same(*page_table, orig_pte); |
4c21e2f2 | 1955 | spin_unlock(ptl); |
8f4e2101 HD |
1956 | } |
1957 | #endif | |
1958 | pte_unmap(page_table); | |
1959 | return same; | |
1960 | } | |
1961 | ||
9de455b2 | 1962 | static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma) |
6aab341e | 1963 | { |
0abdd7a8 DW |
1964 | debug_dma_assert_idle(src); |
1965 | ||
6aab341e LT |
1966 | /* |
1967 | * If the source page was a PFN mapping, we don't have | |
1968 | * a "struct page" for it. We do a best-effort copy by | |
1969 | * just copying from the original user address. If that | |
1970 | * fails, we just zero-fill it. Live with it. | |
1971 | */ | |
1972 | if (unlikely(!src)) { | |
9b04c5fe | 1973 | void *kaddr = kmap_atomic(dst); |
5d2a2dbb LT |
1974 | void __user *uaddr = (void __user *)(va & PAGE_MASK); |
1975 | ||
1976 | /* | |
1977 | * This really shouldn't fail, because the page is there | |
1978 | * in the page tables. But it might just be unreadable, | |
1979 | * in which case we just give up and fill the result with | |
1980 | * zeroes. | |
1981 | */ | |
1982 | if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) | |
3ecb01df | 1983 | clear_page(kaddr); |
9b04c5fe | 1984 | kunmap_atomic(kaddr); |
c4ec7b0d | 1985 | flush_dcache_page(dst); |
0ed361de NP |
1986 | } else |
1987 | copy_user_highpage(dst, src, va, vma); | |
6aab341e LT |
1988 | } |
1989 | ||
fb09a464 KS |
1990 | /* |
1991 | * Notify the address space that the page is about to become writable so that | |
1992 | * it can prohibit this or wait for the page to get into an appropriate state. | |
1993 | * | |
1994 | * We do this without the lock held, so that it can sleep if it needs to. | |
1995 | */ | |
1996 | static int do_page_mkwrite(struct vm_area_struct *vma, struct page *page, | |
1997 | unsigned long address) | |
1998 | { | |
1999 | struct vm_fault vmf; | |
2000 | int ret; | |
2001 | ||
2002 | vmf.virtual_address = (void __user *)(address & PAGE_MASK); | |
2003 | vmf.pgoff = page->index; | |
2004 | vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; | |
2005 | vmf.page = page; | |
2006 | ||
2007 | ret = vma->vm_ops->page_mkwrite(vma, &vmf); | |
2008 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) | |
2009 | return ret; | |
2010 | if (unlikely(!(ret & VM_FAULT_LOCKED))) { | |
2011 | lock_page(page); | |
2012 | if (!page->mapping) { | |
2013 | unlock_page(page); | |
2014 | return 0; /* retry */ | |
2015 | } | |
2016 | ret |= VM_FAULT_LOCKED; | |
2017 | } else | |
2018 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
2019 | return ret; | |
2020 | } | |
2021 | ||
1da177e4 LT |
2022 | /* |
2023 | * This routine handles present pages, when users try to write | |
2024 | * to a shared page. It is done by copying the page to a new address | |
2025 | * and decrementing the shared-page counter for the old page. | |
2026 | * | |
1da177e4 LT |
2027 | * Note that this routine assumes that the protection checks have been |
2028 | * done by the caller (the low-level page fault routine in most cases). | |
2029 | * Thus we can safely just mark it writable once we've done any necessary | |
2030 | * COW. | |
2031 | * | |
2032 | * We also mark the page dirty at this point even though the page will | |
2033 | * change only once the write actually happens. This avoids a few races, | |
2034 | * and potentially makes it more efficient. | |
2035 | * | |
8f4e2101 HD |
2036 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2037 | * but allow concurrent faults), with pte both mapped and locked. | |
2038 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2039 | */ |
65500d23 HD |
2040 | static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma, |
2041 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
8f4e2101 | 2042 | spinlock_t *ptl, pte_t orig_pte) |
e6219ec8 | 2043 | __releases(ptl) |
1da177e4 | 2044 | { |
2ec74c3e | 2045 | struct page *old_page, *new_page = NULL; |
1da177e4 | 2046 | pte_t entry; |
b009c024 | 2047 | int ret = 0; |
a200ee18 | 2048 | int page_mkwrite = 0; |
d08b3851 | 2049 | struct page *dirty_page = NULL; |
1756954c DR |
2050 | unsigned long mmun_start = 0; /* For mmu_notifiers */ |
2051 | unsigned long mmun_end = 0; /* For mmu_notifiers */ | |
1da177e4 | 2052 | |
6aab341e | 2053 | old_page = vm_normal_page(vma, address, orig_pte); |
251b97f5 PZ |
2054 | if (!old_page) { |
2055 | /* | |
2056 | * VM_MIXEDMAP !pfn_valid() case | |
2057 | * | |
2058 | * We should not cow pages in a shared writeable mapping. | |
2059 | * Just mark the pages writable as we can't do any dirty | |
2060 | * accounting on raw pfn maps. | |
2061 | */ | |
2062 | if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) == | |
2063 | (VM_WRITE|VM_SHARED)) | |
2064 | goto reuse; | |
6aab341e | 2065 | goto gotten; |
251b97f5 | 2066 | } |
1da177e4 | 2067 | |
d08b3851 | 2068 | /* |
ee6a6457 PZ |
2069 | * Take out anonymous pages first, anonymous shared vmas are |
2070 | * not dirty accountable. | |
d08b3851 | 2071 | */ |
9a840895 | 2072 | if (PageAnon(old_page) && !PageKsm(old_page)) { |
ab967d86 HD |
2073 | if (!trylock_page(old_page)) { |
2074 | page_cache_get(old_page); | |
2075 | pte_unmap_unlock(page_table, ptl); | |
2076 | lock_page(old_page); | |
2077 | page_table = pte_offset_map_lock(mm, pmd, address, | |
2078 | &ptl); | |
2079 | if (!pte_same(*page_table, orig_pte)) { | |
2080 | unlock_page(old_page); | |
ab967d86 HD |
2081 | goto unlock; |
2082 | } | |
2083 | page_cache_release(old_page); | |
ee6a6457 | 2084 | } |
b009c024 | 2085 | if (reuse_swap_page(old_page)) { |
c44b6743 RR |
2086 | /* |
2087 | * The page is all ours. Move it to our anon_vma so | |
2088 | * the rmap code will not search our parent or siblings. | |
2089 | * Protected against the rmap code by the page lock. | |
2090 | */ | |
2091 | page_move_anon_rmap(old_page, vma, address); | |
b009c024 ML |
2092 | unlock_page(old_page); |
2093 | goto reuse; | |
2094 | } | |
ab967d86 | 2095 | unlock_page(old_page); |
ee6a6457 | 2096 | } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) == |
d08b3851 | 2097 | (VM_WRITE|VM_SHARED))) { |
ee6a6457 PZ |
2098 | /* |
2099 | * Only catch write-faults on shared writable pages, | |
2100 | * read-only shared pages can get COWed by | |
2101 | * get_user_pages(.write=1, .force=1). | |
2102 | */ | |
9637a5ef | 2103 | if (vma->vm_ops && vma->vm_ops->page_mkwrite) { |
c2ec175c | 2104 | int tmp; |
9637a5ef DH |
2105 | page_cache_get(old_page); |
2106 | pte_unmap_unlock(page_table, ptl); | |
fb09a464 KS |
2107 | tmp = do_page_mkwrite(vma, old_page, address); |
2108 | if (unlikely(!tmp || (tmp & | |
2109 | (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { | |
2110 | page_cache_release(old_page); | |
2111 | return tmp; | |
c2ec175c | 2112 | } |
9637a5ef DH |
2113 | /* |
2114 | * Since we dropped the lock we need to revalidate | |
2115 | * the PTE as someone else may have changed it. If | |
2116 | * they did, we just return, as we can count on the | |
2117 | * MMU to tell us if they didn't also make it writable. | |
2118 | */ | |
2119 | page_table = pte_offset_map_lock(mm, pmd, address, | |
2120 | &ptl); | |
b827e496 NP |
2121 | if (!pte_same(*page_table, orig_pte)) { |
2122 | unlock_page(old_page); | |
9637a5ef | 2123 | goto unlock; |
b827e496 | 2124 | } |
a200ee18 PZ |
2125 | |
2126 | page_mkwrite = 1; | |
1da177e4 | 2127 | } |
d08b3851 PZ |
2128 | dirty_page = old_page; |
2129 | get_page(dirty_page); | |
9637a5ef | 2130 | |
251b97f5 | 2131 | reuse: |
8c8a743c PZ |
2132 | /* |
2133 | * Clear the pages cpupid information as the existing | |
2134 | * information potentially belongs to a now completely | |
2135 | * unrelated process. | |
2136 | */ | |
2137 | if (old_page) | |
2138 | page_cpupid_xchg_last(old_page, (1 << LAST_CPUPID_SHIFT) - 1); | |
2139 | ||
9637a5ef DH |
2140 | flush_cache_page(vma, address, pte_pfn(orig_pte)); |
2141 | entry = pte_mkyoung(orig_pte); | |
2142 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
954ffcb3 | 2143 | if (ptep_set_access_flags(vma, address, page_table, entry,1)) |
4b3073e1 | 2144 | update_mmu_cache(vma, address, page_table); |
72ddc8f7 | 2145 | pte_unmap_unlock(page_table, ptl); |
9637a5ef | 2146 | ret |= VM_FAULT_WRITE; |
72ddc8f7 ML |
2147 | |
2148 | if (!dirty_page) | |
2149 | return ret; | |
2150 | ||
2151 | /* | |
2152 | * Yes, Virginia, this is actually required to prevent a race | |
2153 | * with clear_page_dirty_for_io() from clearing the page dirty | |
2154 | * bit after it clear all dirty ptes, but before a racing | |
2155 | * do_wp_page installs a dirty pte. | |
2156 | * | |
f0c6d4d2 | 2157 | * do_shared_fault is protected similarly. |
72ddc8f7 ML |
2158 | */ |
2159 | if (!page_mkwrite) { | |
2160 | wait_on_page_locked(dirty_page); | |
ed6d7c8e | 2161 | set_page_dirty_balance(dirty_page); |
41c4d25f JK |
2162 | /* file_update_time outside page_lock */ |
2163 | if (vma->vm_file) | |
2164 | file_update_time(vma->vm_file); | |
72ddc8f7 ML |
2165 | } |
2166 | put_page(dirty_page); | |
2167 | if (page_mkwrite) { | |
2168 | struct address_space *mapping = dirty_page->mapping; | |
2169 | ||
2170 | set_page_dirty(dirty_page); | |
2171 | unlock_page(dirty_page); | |
2172 | page_cache_release(dirty_page); | |
2173 | if (mapping) { | |
2174 | /* | |
2175 | * Some device drivers do not set page.mapping | |
2176 | * but still dirty their pages | |
2177 | */ | |
2178 | balance_dirty_pages_ratelimited(mapping); | |
2179 | } | |
2180 | } | |
2181 | ||
72ddc8f7 | 2182 | return ret; |
1da177e4 | 2183 | } |
1da177e4 LT |
2184 | |
2185 | /* | |
2186 | * Ok, we need to copy. Oh, well.. | |
2187 | */ | |
b5810039 | 2188 | page_cache_get(old_page); |
920fc356 | 2189 | gotten: |
8f4e2101 | 2190 | pte_unmap_unlock(page_table, ptl); |
1da177e4 LT |
2191 | |
2192 | if (unlikely(anon_vma_prepare(vma))) | |
65500d23 | 2193 | goto oom; |
a13ea5b7 | 2194 | |
62eede62 | 2195 | if (is_zero_pfn(pte_pfn(orig_pte))) { |
a13ea5b7 HD |
2196 | new_page = alloc_zeroed_user_highpage_movable(vma, address); |
2197 | if (!new_page) | |
2198 | goto oom; | |
2199 | } else { | |
2200 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); | |
2201 | if (!new_page) | |
2202 | goto oom; | |
2203 | cow_user_page(new_page, old_page, address, vma); | |
2204 | } | |
2205 | __SetPageUptodate(new_page); | |
2206 | ||
d715ae08 | 2207 | if (mem_cgroup_charge_anon(new_page, mm, GFP_KERNEL)) |
8a9f3ccd BS |
2208 | goto oom_free_new; |
2209 | ||
6bdb913f | 2210 | mmun_start = address & PAGE_MASK; |
1756954c | 2211 | mmun_end = mmun_start + PAGE_SIZE; |
6bdb913f HE |
2212 | mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); |
2213 | ||
1da177e4 LT |
2214 | /* |
2215 | * Re-check the pte - we dropped the lock | |
2216 | */ | |
8f4e2101 | 2217 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
65500d23 | 2218 | if (likely(pte_same(*page_table, orig_pte))) { |
920fc356 | 2219 | if (old_page) { |
920fc356 | 2220 | if (!PageAnon(old_page)) { |
34e55232 KH |
2221 | dec_mm_counter_fast(mm, MM_FILEPAGES); |
2222 | inc_mm_counter_fast(mm, MM_ANONPAGES); | |
920fc356 HD |
2223 | } |
2224 | } else | |
34e55232 | 2225 | inc_mm_counter_fast(mm, MM_ANONPAGES); |
eca35133 | 2226 | flush_cache_page(vma, address, pte_pfn(orig_pte)); |
65500d23 HD |
2227 | entry = mk_pte(new_page, vma->vm_page_prot); |
2228 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
4ce072f1 SS |
2229 | /* |
2230 | * Clear the pte entry and flush it first, before updating the | |
2231 | * pte with the new entry. This will avoid a race condition | |
2232 | * seen in the presence of one thread doing SMC and another | |
2233 | * thread doing COW. | |
2234 | */ | |
828502d3 | 2235 | ptep_clear_flush(vma, address, page_table); |
9617d95e | 2236 | page_add_new_anon_rmap(new_page, vma, address); |
828502d3 IE |
2237 | /* |
2238 | * We call the notify macro here because, when using secondary | |
2239 | * mmu page tables (such as kvm shadow page tables), we want the | |
2240 | * new page to be mapped directly into the secondary page table. | |
2241 | */ | |
2242 | set_pte_at_notify(mm, address, page_table, entry); | |
4b3073e1 | 2243 | update_mmu_cache(vma, address, page_table); |
945754a1 NP |
2244 | if (old_page) { |
2245 | /* | |
2246 | * Only after switching the pte to the new page may | |
2247 | * we remove the mapcount here. Otherwise another | |
2248 | * process may come and find the rmap count decremented | |
2249 | * before the pte is switched to the new page, and | |
2250 | * "reuse" the old page writing into it while our pte | |
2251 | * here still points into it and can be read by other | |
2252 | * threads. | |
2253 | * | |
2254 | * The critical issue is to order this | |
2255 | * page_remove_rmap with the ptp_clear_flush above. | |
2256 | * Those stores are ordered by (if nothing else,) | |
2257 | * the barrier present in the atomic_add_negative | |
2258 | * in page_remove_rmap. | |
2259 | * | |
2260 | * Then the TLB flush in ptep_clear_flush ensures that | |
2261 | * no process can access the old page before the | |
2262 | * decremented mapcount is visible. And the old page | |
2263 | * cannot be reused until after the decremented | |
2264 | * mapcount is visible. So transitively, TLBs to | |
2265 | * old page will be flushed before it can be reused. | |
2266 | */ | |
edc315fd | 2267 | page_remove_rmap(old_page); |
945754a1 NP |
2268 | } |
2269 | ||
1da177e4 LT |
2270 | /* Free the old page.. */ |
2271 | new_page = old_page; | |
f33ea7f4 | 2272 | ret |= VM_FAULT_WRITE; |
8a9f3ccd BS |
2273 | } else |
2274 | mem_cgroup_uncharge_page(new_page); | |
2275 | ||
6bdb913f HE |
2276 | if (new_page) |
2277 | page_cache_release(new_page); | |
65500d23 | 2278 | unlock: |
8f4e2101 | 2279 | pte_unmap_unlock(page_table, ptl); |
1756954c | 2280 | if (mmun_end > mmun_start) |
6bdb913f | 2281 | mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); |
e15f8c01 ML |
2282 | if (old_page) { |
2283 | /* | |
2284 | * Don't let another task, with possibly unlocked vma, | |
2285 | * keep the mlocked page. | |
2286 | */ | |
2287 | if ((ret & VM_FAULT_WRITE) && (vma->vm_flags & VM_LOCKED)) { | |
2288 | lock_page(old_page); /* LRU manipulation */ | |
2289 | munlock_vma_page(old_page); | |
2290 | unlock_page(old_page); | |
2291 | } | |
2292 | page_cache_release(old_page); | |
2293 | } | |
f33ea7f4 | 2294 | return ret; |
8a9f3ccd | 2295 | oom_free_new: |
6dbf6d3b | 2296 | page_cache_release(new_page); |
65500d23 | 2297 | oom: |
66521d5a | 2298 | if (old_page) |
920fc356 | 2299 | page_cache_release(old_page); |
1da177e4 LT |
2300 | return VM_FAULT_OOM; |
2301 | } | |
2302 | ||
97a89413 | 2303 | static void unmap_mapping_range_vma(struct vm_area_struct *vma, |
1da177e4 LT |
2304 | unsigned long start_addr, unsigned long end_addr, |
2305 | struct zap_details *details) | |
2306 | { | |
f5cc4eef | 2307 | zap_page_range_single(vma, start_addr, end_addr - start_addr, details); |
1da177e4 LT |
2308 | } |
2309 | ||
6b2dbba8 | 2310 | static inline void unmap_mapping_range_tree(struct rb_root *root, |
1da177e4 LT |
2311 | struct zap_details *details) |
2312 | { | |
2313 | struct vm_area_struct *vma; | |
1da177e4 LT |
2314 | pgoff_t vba, vea, zba, zea; |
2315 | ||
6b2dbba8 | 2316 | vma_interval_tree_foreach(vma, root, |
1da177e4 | 2317 | details->first_index, details->last_index) { |
1da177e4 LT |
2318 | |
2319 | vba = vma->vm_pgoff; | |
d6e93217 | 2320 | vea = vba + vma_pages(vma) - 1; |
1da177e4 LT |
2321 | /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */ |
2322 | zba = details->first_index; | |
2323 | if (zba < vba) | |
2324 | zba = vba; | |
2325 | zea = details->last_index; | |
2326 | if (zea > vea) | |
2327 | zea = vea; | |
2328 | ||
97a89413 | 2329 | unmap_mapping_range_vma(vma, |
1da177e4 LT |
2330 | ((zba - vba) << PAGE_SHIFT) + vma->vm_start, |
2331 | ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, | |
97a89413 | 2332 | details); |
1da177e4 LT |
2333 | } |
2334 | } | |
2335 | ||
2336 | static inline void unmap_mapping_range_list(struct list_head *head, | |
2337 | struct zap_details *details) | |
2338 | { | |
2339 | struct vm_area_struct *vma; | |
2340 | ||
2341 | /* | |
2342 | * In nonlinear VMAs there is no correspondence between virtual address | |
2343 | * offset and file offset. So we must perform an exhaustive search | |
2344 | * across *all* the pages in each nonlinear VMA, not just the pages | |
2345 | * whose virtual address lies outside the file truncation point. | |
2346 | */ | |
6b2dbba8 | 2347 | list_for_each_entry(vma, head, shared.nonlinear) { |
1da177e4 | 2348 | details->nonlinear_vma = vma; |
97a89413 | 2349 | unmap_mapping_range_vma(vma, vma->vm_start, vma->vm_end, details); |
1da177e4 LT |
2350 | } |
2351 | } | |
2352 | ||
2353 | /** | |
72fd4a35 | 2354 | * unmap_mapping_range - unmap the portion of all mmaps in the specified address_space corresponding to the specified page range in the underlying file. |
3d41088f | 2355 | * @mapping: the address space containing mmaps to be unmapped. |
1da177e4 LT |
2356 | * @holebegin: byte in first page to unmap, relative to the start of |
2357 | * the underlying file. This will be rounded down to a PAGE_SIZE | |
25d9e2d1 | 2358 | * boundary. Note that this is different from truncate_pagecache(), which |
1da177e4 LT |
2359 | * must keep the partial page. In contrast, we must get rid of |
2360 | * partial pages. | |
2361 | * @holelen: size of prospective hole in bytes. This will be rounded | |
2362 | * up to a PAGE_SIZE boundary. A holelen of zero truncates to the | |
2363 | * end of the file. | |
2364 | * @even_cows: 1 when truncating a file, unmap even private COWed pages; | |
2365 | * but 0 when invalidating pagecache, don't throw away private data. | |
2366 | */ | |
2367 | void unmap_mapping_range(struct address_space *mapping, | |
2368 | loff_t const holebegin, loff_t const holelen, int even_cows) | |
2369 | { | |
2370 | struct zap_details details; | |
2371 | pgoff_t hba = holebegin >> PAGE_SHIFT; | |
2372 | pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
2373 | ||
2374 | /* Check for overflow. */ | |
2375 | if (sizeof(holelen) > sizeof(hlen)) { | |
2376 | long long holeend = | |
2377 | (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
2378 | if (holeend & ~(long long)ULONG_MAX) | |
2379 | hlen = ULONG_MAX - hba + 1; | |
2380 | } | |
2381 | ||
2382 | details.check_mapping = even_cows? NULL: mapping; | |
2383 | details.nonlinear_vma = NULL; | |
2384 | details.first_index = hba; | |
2385 | details.last_index = hba + hlen - 1; | |
2386 | if (details.last_index < details.first_index) | |
2387 | details.last_index = ULONG_MAX; | |
1da177e4 | 2388 | |
1da177e4 | 2389 | |
3d48ae45 | 2390 | mutex_lock(&mapping->i_mmap_mutex); |
6b2dbba8 | 2391 | if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap))) |
1da177e4 LT |
2392 | unmap_mapping_range_tree(&mapping->i_mmap, &details); |
2393 | if (unlikely(!list_empty(&mapping->i_mmap_nonlinear))) | |
2394 | unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details); | |
3d48ae45 | 2395 | mutex_unlock(&mapping->i_mmap_mutex); |
1da177e4 LT |
2396 | } |
2397 | EXPORT_SYMBOL(unmap_mapping_range); | |
2398 | ||
1da177e4 | 2399 | /* |
8f4e2101 HD |
2400 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2401 | * but allow concurrent faults), and pte mapped but not yet locked. | |
9a95f3cf PC |
2402 | * We return with pte unmapped and unlocked. |
2403 | * | |
2404 | * We return with the mmap_sem locked or unlocked in the same cases | |
2405 | * as does filemap_fault(). | |
1da177e4 | 2406 | */ |
65500d23 HD |
2407 | static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma, |
2408 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
30c9f3a9 | 2409 | unsigned int flags, pte_t orig_pte) |
1da177e4 | 2410 | { |
8f4e2101 | 2411 | spinlock_t *ptl; |
56f31801 | 2412 | struct page *page, *swapcache; |
65500d23 | 2413 | swp_entry_t entry; |
1da177e4 | 2414 | pte_t pte; |
d065bd81 | 2415 | int locked; |
56039efa | 2416 | struct mem_cgroup *ptr; |
ad8c2ee8 | 2417 | int exclusive = 0; |
83c54070 | 2418 | int ret = 0; |
1da177e4 | 2419 | |
4c21e2f2 | 2420 | if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) |
8f4e2101 | 2421 | goto out; |
65500d23 HD |
2422 | |
2423 | entry = pte_to_swp_entry(orig_pte); | |
d1737fdb AK |
2424 | if (unlikely(non_swap_entry(entry))) { |
2425 | if (is_migration_entry(entry)) { | |
2426 | migration_entry_wait(mm, pmd, address); | |
2427 | } else if (is_hwpoison_entry(entry)) { | |
2428 | ret = VM_FAULT_HWPOISON; | |
2429 | } else { | |
2430 | print_bad_pte(vma, address, orig_pte, NULL); | |
d99be1a8 | 2431 | ret = VM_FAULT_SIGBUS; |
d1737fdb | 2432 | } |
0697212a CL |
2433 | goto out; |
2434 | } | |
0ff92245 | 2435 | delayacct_set_flag(DELAYACCT_PF_SWAPIN); |
1da177e4 LT |
2436 | page = lookup_swap_cache(entry); |
2437 | if (!page) { | |
02098fea HD |
2438 | page = swapin_readahead(entry, |
2439 | GFP_HIGHUSER_MOVABLE, vma, address); | |
1da177e4 LT |
2440 | if (!page) { |
2441 | /* | |
8f4e2101 HD |
2442 | * Back out if somebody else faulted in this pte |
2443 | * while we released the pte lock. | |
1da177e4 | 2444 | */ |
8f4e2101 | 2445 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
1da177e4 LT |
2446 | if (likely(pte_same(*page_table, orig_pte))) |
2447 | ret = VM_FAULT_OOM; | |
0ff92245 | 2448 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
65500d23 | 2449 | goto unlock; |
1da177e4 LT |
2450 | } |
2451 | ||
2452 | /* Had to read the page from swap area: Major fault */ | |
2453 | ret = VM_FAULT_MAJOR; | |
f8891e5e | 2454 | count_vm_event(PGMAJFAULT); |
456f998e | 2455 | mem_cgroup_count_vm_event(mm, PGMAJFAULT); |
d1737fdb | 2456 | } else if (PageHWPoison(page)) { |
71f72525 WF |
2457 | /* |
2458 | * hwpoisoned dirty swapcache pages are kept for killing | |
2459 | * owner processes (which may be unknown at hwpoison time) | |
2460 | */ | |
d1737fdb AK |
2461 | ret = VM_FAULT_HWPOISON; |
2462 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); | |
56f31801 | 2463 | swapcache = page; |
4779cb31 | 2464 | goto out_release; |
1da177e4 LT |
2465 | } |
2466 | ||
56f31801 | 2467 | swapcache = page; |
d065bd81 | 2468 | locked = lock_page_or_retry(page, mm, flags); |
e709ffd6 | 2469 | |
073e587e | 2470 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
d065bd81 ML |
2471 | if (!locked) { |
2472 | ret |= VM_FAULT_RETRY; | |
2473 | goto out_release; | |
2474 | } | |
073e587e | 2475 | |
4969c119 | 2476 | /* |
31c4a3d3 HD |
2477 | * Make sure try_to_free_swap or reuse_swap_page or swapoff did not |
2478 | * release the swapcache from under us. The page pin, and pte_same | |
2479 | * test below, are not enough to exclude that. Even if it is still | |
2480 | * swapcache, we need to check that the page's swap has not changed. | |
4969c119 | 2481 | */ |
31c4a3d3 | 2482 | if (unlikely(!PageSwapCache(page) || page_private(page) != entry.val)) |
4969c119 AA |
2483 | goto out_page; |
2484 | ||
cbf86cfe HD |
2485 | page = ksm_might_need_to_copy(page, vma, address); |
2486 | if (unlikely(!page)) { | |
2487 | ret = VM_FAULT_OOM; | |
2488 | page = swapcache; | |
cbf86cfe | 2489 | goto out_page; |
5ad64688 HD |
2490 | } |
2491 | ||
2c26fdd7 | 2492 | if (mem_cgroup_try_charge_swapin(mm, page, GFP_KERNEL, &ptr)) { |
8a9f3ccd | 2493 | ret = VM_FAULT_OOM; |
bc43f75c | 2494 | goto out_page; |
8a9f3ccd BS |
2495 | } |
2496 | ||
1da177e4 | 2497 | /* |
8f4e2101 | 2498 | * Back out if somebody else already faulted in this pte. |
1da177e4 | 2499 | */ |
8f4e2101 | 2500 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
9e9bef07 | 2501 | if (unlikely(!pte_same(*page_table, orig_pte))) |
b8107480 | 2502 | goto out_nomap; |
b8107480 KK |
2503 | |
2504 | if (unlikely(!PageUptodate(page))) { | |
2505 | ret = VM_FAULT_SIGBUS; | |
2506 | goto out_nomap; | |
1da177e4 LT |
2507 | } |
2508 | ||
8c7c6e34 KH |
2509 | /* |
2510 | * The page isn't present yet, go ahead with the fault. | |
2511 | * | |
2512 | * Be careful about the sequence of operations here. | |
2513 | * To get its accounting right, reuse_swap_page() must be called | |
2514 | * while the page is counted on swap but not yet in mapcount i.e. | |
2515 | * before page_add_anon_rmap() and swap_free(); try_to_free_swap() | |
2516 | * must be called after the swap_free(), or it will never succeed. | |
03f3c433 KH |
2517 | * Because delete_from_swap_page() may be called by reuse_swap_page(), |
2518 | * mem_cgroup_commit_charge_swapin() may not be able to find swp_entry | |
2519 | * in page->private. In this case, a record in swap_cgroup is silently | |
2520 | * discarded at swap_free(). | |
8c7c6e34 | 2521 | */ |
1da177e4 | 2522 | |
34e55232 | 2523 | inc_mm_counter_fast(mm, MM_ANONPAGES); |
b084d435 | 2524 | dec_mm_counter_fast(mm, MM_SWAPENTS); |
1da177e4 | 2525 | pte = mk_pte(page, vma->vm_page_prot); |
30c9f3a9 | 2526 | if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page)) { |
1da177e4 | 2527 | pte = maybe_mkwrite(pte_mkdirty(pte), vma); |
30c9f3a9 | 2528 | flags &= ~FAULT_FLAG_WRITE; |
9a5b489b | 2529 | ret |= VM_FAULT_WRITE; |
ad8c2ee8 | 2530 | exclusive = 1; |
1da177e4 | 2531 | } |
1da177e4 | 2532 | flush_icache_page(vma, page); |
179ef71c CG |
2533 | if (pte_swp_soft_dirty(orig_pte)) |
2534 | pte = pte_mksoft_dirty(pte); | |
1da177e4 | 2535 | set_pte_at(mm, address, page_table, pte); |
56f31801 | 2536 | if (page == swapcache) |
af34770e | 2537 | do_page_add_anon_rmap(page, vma, address, exclusive); |
56f31801 HD |
2538 | else /* ksm created a completely new copy */ |
2539 | page_add_new_anon_rmap(page, vma, address); | |
03f3c433 KH |
2540 | /* It's better to call commit-charge after rmap is established */ |
2541 | mem_cgroup_commit_charge_swapin(page, ptr); | |
1da177e4 | 2542 | |
c475a8ab | 2543 | swap_free(entry); |
b291f000 | 2544 | if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page)) |
a2c43eed | 2545 | try_to_free_swap(page); |
c475a8ab | 2546 | unlock_page(page); |
56f31801 | 2547 | if (page != swapcache) { |
4969c119 AA |
2548 | /* |
2549 | * Hold the lock to avoid the swap entry to be reused | |
2550 | * until we take the PT lock for the pte_same() check | |
2551 | * (to avoid false positives from pte_same). For | |
2552 | * further safety release the lock after the swap_free | |
2553 | * so that the swap count won't change under a | |
2554 | * parallel locked swapcache. | |
2555 | */ | |
2556 | unlock_page(swapcache); | |
2557 | page_cache_release(swapcache); | |
2558 | } | |
c475a8ab | 2559 | |
30c9f3a9 | 2560 | if (flags & FAULT_FLAG_WRITE) { |
61469f1d HD |
2561 | ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte); |
2562 | if (ret & VM_FAULT_ERROR) | |
2563 | ret &= VM_FAULT_ERROR; | |
1da177e4 LT |
2564 | goto out; |
2565 | } | |
2566 | ||
2567 | /* No need to invalidate - it was non-present before */ | |
4b3073e1 | 2568 | update_mmu_cache(vma, address, page_table); |
65500d23 | 2569 | unlock: |
8f4e2101 | 2570 | pte_unmap_unlock(page_table, ptl); |
1da177e4 LT |
2571 | out: |
2572 | return ret; | |
b8107480 | 2573 | out_nomap: |
7a81b88c | 2574 | mem_cgroup_cancel_charge_swapin(ptr); |
8f4e2101 | 2575 | pte_unmap_unlock(page_table, ptl); |
bc43f75c | 2576 | out_page: |
b8107480 | 2577 | unlock_page(page); |
4779cb31 | 2578 | out_release: |
b8107480 | 2579 | page_cache_release(page); |
56f31801 | 2580 | if (page != swapcache) { |
4969c119 AA |
2581 | unlock_page(swapcache); |
2582 | page_cache_release(swapcache); | |
2583 | } | |
65500d23 | 2584 | return ret; |
1da177e4 LT |
2585 | } |
2586 | ||
320b2b8d | 2587 | /* |
8ca3eb08 TL |
2588 | * This is like a special single-page "expand_{down|up}wards()", |
2589 | * except we must first make sure that 'address{-|+}PAGE_SIZE' | |
320b2b8d | 2590 | * doesn't hit another vma. |
320b2b8d LT |
2591 | */ |
2592 | static inline int check_stack_guard_page(struct vm_area_struct *vma, unsigned long address) | |
2593 | { | |
2594 | address &= PAGE_MASK; | |
2595 | if ((vma->vm_flags & VM_GROWSDOWN) && address == vma->vm_start) { | |
0e8e50e2 LT |
2596 | struct vm_area_struct *prev = vma->vm_prev; |
2597 | ||
2598 | /* | |
2599 | * Is there a mapping abutting this one below? | |
2600 | * | |
2601 | * That's only ok if it's the same stack mapping | |
2602 | * that has gotten split.. | |
2603 | */ | |
2604 | if (prev && prev->vm_end == address) | |
2605 | return prev->vm_flags & VM_GROWSDOWN ? 0 : -ENOMEM; | |
320b2b8d | 2606 | |
d05f3169 | 2607 | expand_downwards(vma, address - PAGE_SIZE); |
320b2b8d | 2608 | } |
8ca3eb08 TL |
2609 | if ((vma->vm_flags & VM_GROWSUP) && address + PAGE_SIZE == vma->vm_end) { |
2610 | struct vm_area_struct *next = vma->vm_next; | |
2611 | ||
2612 | /* As VM_GROWSDOWN but s/below/above/ */ | |
2613 | if (next && next->vm_start == address + PAGE_SIZE) | |
2614 | return next->vm_flags & VM_GROWSUP ? 0 : -ENOMEM; | |
2615 | ||
2616 | expand_upwards(vma, address + PAGE_SIZE); | |
2617 | } | |
320b2b8d LT |
2618 | return 0; |
2619 | } | |
2620 | ||
1da177e4 | 2621 | /* |
8f4e2101 HD |
2622 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2623 | * but allow concurrent faults), and pte mapped but not yet locked. | |
2624 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2625 | */ |
65500d23 HD |
2626 | static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma, |
2627 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
30c9f3a9 | 2628 | unsigned int flags) |
1da177e4 | 2629 | { |
8f4e2101 HD |
2630 | struct page *page; |
2631 | spinlock_t *ptl; | |
1da177e4 | 2632 | pte_t entry; |
1da177e4 | 2633 | |
11ac5524 LT |
2634 | pte_unmap(page_table); |
2635 | ||
2636 | /* Check if we need to add a guard page to the stack */ | |
2637 | if (check_stack_guard_page(vma, address) < 0) | |
320b2b8d LT |
2638 | return VM_FAULT_SIGBUS; |
2639 | ||
11ac5524 | 2640 | /* Use the zero-page for reads */ |
62eede62 HD |
2641 | if (!(flags & FAULT_FLAG_WRITE)) { |
2642 | entry = pte_mkspecial(pfn_pte(my_zero_pfn(address), | |
2643 | vma->vm_page_prot)); | |
11ac5524 | 2644 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
a13ea5b7 HD |
2645 | if (!pte_none(*page_table)) |
2646 | goto unlock; | |
2647 | goto setpte; | |
2648 | } | |
2649 | ||
557ed1fa | 2650 | /* Allocate our own private page. */ |
557ed1fa NP |
2651 | if (unlikely(anon_vma_prepare(vma))) |
2652 | goto oom; | |
2653 | page = alloc_zeroed_user_highpage_movable(vma, address); | |
2654 | if (!page) | |
2655 | goto oom; | |
52f37629 MK |
2656 | /* |
2657 | * The memory barrier inside __SetPageUptodate makes sure that | |
2658 | * preceeding stores to the page contents become visible before | |
2659 | * the set_pte_at() write. | |
2660 | */ | |
0ed361de | 2661 | __SetPageUptodate(page); |
8f4e2101 | 2662 | |
d715ae08 | 2663 | if (mem_cgroup_charge_anon(page, mm, GFP_KERNEL)) |
8a9f3ccd BS |
2664 | goto oom_free_page; |
2665 | ||
557ed1fa | 2666 | entry = mk_pte(page, vma->vm_page_prot); |
1ac0cb5d HD |
2667 | if (vma->vm_flags & VM_WRITE) |
2668 | entry = pte_mkwrite(pte_mkdirty(entry)); | |
1da177e4 | 2669 | |
557ed1fa | 2670 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
1c2fb7a4 | 2671 | if (!pte_none(*page_table)) |
557ed1fa | 2672 | goto release; |
9ba69294 | 2673 | |
34e55232 | 2674 | inc_mm_counter_fast(mm, MM_ANONPAGES); |
557ed1fa | 2675 | page_add_new_anon_rmap(page, vma, address); |
a13ea5b7 | 2676 | setpte: |
65500d23 | 2677 | set_pte_at(mm, address, page_table, entry); |
1da177e4 LT |
2678 | |
2679 | /* No need to invalidate - it was non-present before */ | |
4b3073e1 | 2680 | update_mmu_cache(vma, address, page_table); |
65500d23 | 2681 | unlock: |
8f4e2101 | 2682 | pte_unmap_unlock(page_table, ptl); |
83c54070 | 2683 | return 0; |
8f4e2101 | 2684 | release: |
8a9f3ccd | 2685 | mem_cgroup_uncharge_page(page); |
8f4e2101 HD |
2686 | page_cache_release(page); |
2687 | goto unlock; | |
8a9f3ccd | 2688 | oom_free_page: |
6dbf6d3b | 2689 | page_cache_release(page); |
65500d23 | 2690 | oom: |
1da177e4 LT |
2691 | return VM_FAULT_OOM; |
2692 | } | |
2693 | ||
9a95f3cf PC |
2694 | /* |
2695 | * The mmap_sem must have been held on entry, and may have been | |
2696 | * released depending on flags and vma->vm_ops->fault() return value. | |
2697 | * See filemap_fault() and __lock_page_retry(). | |
2698 | */ | |
7eae74af KS |
2699 | static int __do_fault(struct vm_area_struct *vma, unsigned long address, |
2700 | pgoff_t pgoff, unsigned int flags, struct page **page) | |
2701 | { | |
2702 | struct vm_fault vmf; | |
2703 | int ret; | |
2704 | ||
2705 | vmf.virtual_address = (void __user *)(address & PAGE_MASK); | |
2706 | vmf.pgoff = pgoff; | |
2707 | vmf.flags = flags; | |
2708 | vmf.page = NULL; | |
2709 | ||
2710 | ret = vma->vm_ops->fault(vma, &vmf); | |
2711 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) | |
2712 | return ret; | |
2713 | ||
2714 | if (unlikely(PageHWPoison(vmf.page))) { | |
2715 | if (ret & VM_FAULT_LOCKED) | |
2716 | unlock_page(vmf.page); | |
2717 | page_cache_release(vmf.page); | |
2718 | return VM_FAULT_HWPOISON; | |
2719 | } | |
2720 | ||
2721 | if (unlikely(!(ret & VM_FAULT_LOCKED))) | |
2722 | lock_page(vmf.page); | |
2723 | else | |
2724 | VM_BUG_ON_PAGE(!PageLocked(vmf.page), vmf.page); | |
2725 | ||
2726 | *page = vmf.page; | |
2727 | return ret; | |
2728 | } | |
2729 | ||
8c6e50b0 KS |
2730 | /** |
2731 | * do_set_pte - setup new PTE entry for given page and add reverse page mapping. | |
2732 | * | |
2733 | * @vma: virtual memory area | |
2734 | * @address: user virtual address | |
2735 | * @page: page to map | |
2736 | * @pte: pointer to target page table entry | |
2737 | * @write: true, if new entry is writable | |
2738 | * @anon: true, if it's anonymous page | |
2739 | * | |
2740 | * Caller must hold page table lock relevant for @pte. | |
2741 | * | |
2742 | * Target users are page handler itself and implementations of | |
2743 | * vm_ops->map_pages. | |
2744 | */ | |
2745 | void do_set_pte(struct vm_area_struct *vma, unsigned long address, | |
3bb97794 KS |
2746 | struct page *page, pte_t *pte, bool write, bool anon) |
2747 | { | |
2748 | pte_t entry; | |
2749 | ||
2750 | flush_icache_page(vma, page); | |
2751 | entry = mk_pte(page, vma->vm_page_prot); | |
2752 | if (write) | |
2753 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
2754 | else if (pte_file(*pte) && pte_file_soft_dirty(*pte)) | |
9aed8614 | 2755 | entry = pte_mksoft_dirty(entry); |
3bb97794 KS |
2756 | if (anon) { |
2757 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); | |
2758 | page_add_new_anon_rmap(page, vma, address); | |
2759 | } else { | |
2760 | inc_mm_counter_fast(vma->vm_mm, MM_FILEPAGES); | |
2761 | page_add_file_rmap(page); | |
2762 | } | |
2763 | set_pte_at(vma->vm_mm, address, pte, entry); | |
2764 | ||
2765 | /* no need to invalidate: a not-present page won't be cached */ | |
2766 | update_mmu_cache(vma, address, pte); | |
2767 | } | |
2768 | ||
3a91053a KS |
2769 | static unsigned long fault_around_bytes __read_mostly = |
2770 | rounddown_pow_of_two(65536); | |
a9b0f861 | 2771 | |
a9b0f861 KS |
2772 | #ifdef CONFIG_DEBUG_FS |
2773 | static int fault_around_bytes_get(void *data, u64 *val) | |
1592eef0 | 2774 | { |
a9b0f861 | 2775 | *val = fault_around_bytes; |
1592eef0 KS |
2776 | return 0; |
2777 | } | |
2778 | ||
b4903d6e AR |
2779 | /* |
2780 | * fault_around_pages() and fault_around_mask() expects fault_around_bytes | |
2781 | * rounded down to nearest page order. It's what do_fault_around() expects to | |
2782 | * see. | |
2783 | */ | |
a9b0f861 | 2784 | static int fault_around_bytes_set(void *data, u64 val) |
1592eef0 | 2785 | { |
a9b0f861 | 2786 | if (val / PAGE_SIZE > PTRS_PER_PTE) |
1592eef0 | 2787 | return -EINVAL; |
b4903d6e AR |
2788 | if (val > PAGE_SIZE) |
2789 | fault_around_bytes = rounddown_pow_of_two(val); | |
2790 | else | |
2791 | fault_around_bytes = PAGE_SIZE; /* rounddown_pow_of_two(0) is undefined */ | |
1592eef0 KS |
2792 | return 0; |
2793 | } | |
a9b0f861 KS |
2794 | DEFINE_SIMPLE_ATTRIBUTE(fault_around_bytes_fops, |
2795 | fault_around_bytes_get, fault_around_bytes_set, "%llu\n"); | |
1592eef0 KS |
2796 | |
2797 | static int __init fault_around_debugfs(void) | |
2798 | { | |
2799 | void *ret; | |
2800 | ||
a9b0f861 KS |
2801 | ret = debugfs_create_file("fault_around_bytes", 0644, NULL, NULL, |
2802 | &fault_around_bytes_fops); | |
1592eef0 | 2803 | if (!ret) |
a9b0f861 | 2804 | pr_warn("Failed to create fault_around_bytes in debugfs"); |
1592eef0 KS |
2805 | return 0; |
2806 | } | |
2807 | late_initcall(fault_around_debugfs); | |
1592eef0 | 2808 | #endif |
8c6e50b0 | 2809 | |
1fdb412b KS |
2810 | /* |
2811 | * do_fault_around() tries to map few pages around the fault address. The hope | |
2812 | * is that the pages will be needed soon and this will lower the number of | |
2813 | * faults to handle. | |
2814 | * | |
2815 | * It uses vm_ops->map_pages() to map the pages, which skips the page if it's | |
2816 | * not ready to be mapped: not up-to-date, locked, etc. | |
2817 | * | |
2818 | * This function is called with the page table lock taken. In the split ptlock | |
2819 | * case the page table lock only protects only those entries which belong to | |
2820 | * the page table corresponding to the fault address. | |
2821 | * | |
2822 | * This function doesn't cross the VMA boundaries, in order to call map_pages() | |
2823 | * only once. | |
2824 | * | |
2825 | * fault_around_pages() defines how many pages we'll try to map. | |
2826 | * do_fault_around() expects it to return a power of two less than or equal to | |
2827 | * PTRS_PER_PTE. | |
2828 | * | |
2829 | * The virtual address of the area that we map is naturally aligned to the | |
2830 | * fault_around_pages() value (and therefore to page order). This way it's | |
2831 | * easier to guarantee that we don't cross page table boundaries. | |
2832 | */ | |
8c6e50b0 KS |
2833 | static void do_fault_around(struct vm_area_struct *vma, unsigned long address, |
2834 | pte_t *pte, pgoff_t pgoff, unsigned int flags) | |
2835 | { | |
aecd6f44 | 2836 | unsigned long start_addr, nr_pages, mask; |
8c6e50b0 KS |
2837 | pgoff_t max_pgoff; |
2838 | struct vm_fault vmf; | |
2839 | int off; | |
2840 | ||
aecd6f44 KS |
2841 | nr_pages = ACCESS_ONCE(fault_around_bytes) >> PAGE_SHIFT; |
2842 | mask = ~(nr_pages * PAGE_SIZE - 1) & PAGE_MASK; | |
2843 | ||
2844 | start_addr = max(address & mask, vma->vm_start); | |
8c6e50b0 KS |
2845 | off = ((address - start_addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1); |
2846 | pte -= off; | |
2847 | pgoff -= off; | |
2848 | ||
2849 | /* | |
2850 | * max_pgoff is either end of page table or end of vma | |
850e9c69 | 2851 | * or fault_around_pages() from pgoff, depending what is nearest. |
8c6e50b0 KS |
2852 | */ |
2853 | max_pgoff = pgoff - ((start_addr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)) + | |
2854 | PTRS_PER_PTE - 1; | |
2855 | max_pgoff = min3(max_pgoff, vma_pages(vma) + vma->vm_pgoff - 1, | |
aecd6f44 | 2856 | pgoff + nr_pages - 1); |
8c6e50b0 KS |
2857 | |
2858 | /* Check if it makes any sense to call ->map_pages */ | |
2859 | while (!pte_none(*pte)) { | |
2860 | if (++pgoff > max_pgoff) | |
2861 | return; | |
2862 | start_addr += PAGE_SIZE; | |
2863 | if (start_addr >= vma->vm_end) | |
2864 | return; | |
2865 | pte++; | |
2866 | } | |
2867 | ||
2868 | vmf.virtual_address = (void __user *) start_addr; | |
2869 | vmf.pte = pte; | |
2870 | vmf.pgoff = pgoff; | |
2871 | vmf.max_pgoff = max_pgoff; | |
2872 | vmf.flags = flags; | |
2873 | vma->vm_ops->map_pages(vma, &vmf); | |
2874 | } | |
2875 | ||
e655fb29 KS |
2876 | static int do_read_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
2877 | unsigned long address, pmd_t *pmd, | |
2878 | pgoff_t pgoff, unsigned int flags, pte_t orig_pte) | |
2879 | { | |
2880 | struct page *fault_page; | |
2881 | spinlock_t *ptl; | |
3bb97794 | 2882 | pte_t *pte; |
8c6e50b0 KS |
2883 | int ret = 0; |
2884 | ||
2885 | /* | |
2886 | * Let's call ->map_pages() first and use ->fault() as fallback | |
2887 | * if page by the offset is not ready to be mapped (cold cache or | |
2888 | * something). | |
2889 | */ | |
c118678b | 2890 | if (vma->vm_ops->map_pages && !(flags & FAULT_FLAG_NONLINEAR) && |
aecd6f44 | 2891 | fault_around_bytes >> PAGE_SHIFT > 1) { |
8c6e50b0 KS |
2892 | pte = pte_offset_map_lock(mm, pmd, address, &ptl); |
2893 | do_fault_around(vma, address, pte, pgoff, flags); | |
2894 | if (!pte_same(*pte, orig_pte)) | |
2895 | goto unlock_out; | |
2896 | pte_unmap_unlock(pte, ptl); | |
2897 | } | |
e655fb29 KS |
2898 | |
2899 | ret = __do_fault(vma, address, pgoff, flags, &fault_page); | |
2900 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) | |
2901 | return ret; | |
2902 | ||
2903 | pte = pte_offset_map_lock(mm, pmd, address, &ptl); | |
2904 | if (unlikely(!pte_same(*pte, orig_pte))) { | |
2905 | pte_unmap_unlock(pte, ptl); | |
2906 | unlock_page(fault_page); | |
2907 | page_cache_release(fault_page); | |
2908 | return ret; | |
2909 | } | |
3bb97794 | 2910 | do_set_pte(vma, address, fault_page, pte, false, false); |
e655fb29 | 2911 | unlock_page(fault_page); |
8c6e50b0 KS |
2912 | unlock_out: |
2913 | pte_unmap_unlock(pte, ptl); | |
e655fb29 KS |
2914 | return ret; |
2915 | } | |
2916 | ||
ec47c3b9 KS |
2917 | static int do_cow_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
2918 | unsigned long address, pmd_t *pmd, | |
2919 | pgoff_t pgoff, unsigned int flags, pte_t orig_pte) | |
2920 | { | |
2921 | struct page *fault_page, *new_page; | |
2922 | spinlock_t *ptl; | |
3bb97794 | 2923 | pte_t *pte; |
ec47c3b9 KS |
2924 | int ret; |
2925 | ||
2926 | if (unlikely(anon_vma_prepare(vma))) | |
2927 | return VM_FAULT_OOM; | |
2928 | ||
2929 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); | |
2930 | if (!new_page) | |
2931 | return VM_FAULT_OOM; | |
2932 | ||
d715ae08 | 2933 | if (mem_cgroup_charge_anon(new_page, mm, GFP_KERNEL)) { |
ec47c3b9 KS |
2934 | page_cache_release(new_page); |
2935 | return VM_FAULT_OOM; | |
2936 | } | |
2937 | ||
2938 | ret = __do_fault(vma, address, pgoff, flags, &fault_page); | |
2939 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) | |
2940 | goto uncharge_out; | |
2941 | ||
2942 | copy_user_highpage(new_page, fault_page, address, vma); | |
2943 | __SetPageUptodate(new_page); | |
2944 | ||
2945 | pte = pte_offset_map_lock(mm, pmd, address, &ptl); | |
2946 | if (unlikely(!pte_same(*pte, orig_pte))) { | |
2947 | pte_unmap_unlock(pte, ptl); | |
2948 | unlock_page(fault_page); | |
2949 | page_cache_release(fault_page); | |
2950 | goto uncharge_out; | |
2951 | } | |
3bb97794 | 2952 | do_set_pte(vma, address, new_page, pte, true, true); |
ec47c3b9 KS |
2953 | pte_unmap_unlock(pte, ptl); |
2954 | unlock_page(fault_page); | |
2955 | page_cache_release(fault_page); | |
2956 | return ret; | |
2957 | uncharge_out: | |
2958 | mem_cgroup_uncharge_page(new_page); | |
2959 | page_cache_release(new_page); | |
2960 | return ret; | |
2961 | } | |
2962 | ||
f0c6d4d2 | 2963 | static int do_shared_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
16abfa08 | 2964 | unsigned long address, pmd_t *pmd, |
54cb8821 | 2965 | pgoff_t pgoff, unsigned int flags, pte_t orig_pte) |
1da177e4 | 2966 | { |
f0c6d4d2 KS |
2967 | struct page *fault_page; |
2968 | struct address_space *mapping; | |
8f4e2101 | 2969 | spinlock_t *ptl; |
3bb97794 | 2970 | pte_t *pte; |
f0c6d4d2 | 2971 | int dirtied = 0; |
f0c6d4d2 | 2972 | int ret, tmp; |
1d65f86d | 2973 | |
7eae74af KS |
2974 | ret = __do_fault(vma, address, pgoff, flags, &fault_page); |
2975 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) | |
f0c6d4d2 | 2976 | return ret; |
1da177e4 LT |
2977 | |
2978 | /* | |
f0c6d4d2 KS |
2979 | * Check if the backing address space wants to know that the page is |
2980 | * about to become writable | |
1da177e4 | 2981 | */ |
fb09a464 KS |
2982 | if (vma->vm_ops->page_mkwrite) { |
2983 | unlock_page(fault_page); | |
2984 | tmp = do_page_mkwrite(vma, fault_page, address); | |
2985 | if (unlikely(!tmp || | |
2986 | (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { | |
f0c6d4d2 | 2987 | page_cache_release(fault_page); |
fb09a464 | 2988 | return tmp; |
4294621f | 2989 | } |
fb09a464 KS |
2990 | } |
2991 | ||
f0c6d4d2 KS |
2992 | pte = pte_offset_map_lock(mm, pmd, address, &ptl); |
2993 | if (unlikely(!pte_same(*pte, orig_pte))) { | |
2994 | pte_unmap_unlock(pte, ptl); | |
2995 | unlock_page(fault_page); | |
2996 | page_cache_release(fault_page); | |
2997 | return ret; | |
1da177e4 | 2998 | } |
3bb97794 | 2999 | do_set_pte(vma, address, fault_page, pte, true, false); |
f0c6d4d2 | 3000 | pte_unmap_unlock(pte, ptl); |
b827e496 | 3001 | |
f0c6d4d2 KS |
3002 | if (set_page_dirty(fault_page)) |
3003 | dirtied = 1; | |
3004 | mapping = fault_page->mapping; | |
3005 | unlock_page(fault_page); | |
3006 | if ((dirtied || vma->vm_ops->page_mkwrite) && mapping) { | |
3007 | /* | |
3008 | * Some device drivers do not set page.mapping but still | |
3009 | * dirty their pages | |
3010 | */ | |
3011 | balance_dirty_pages_ratelimited(mapping); | |
d08b3851 | 3012 | } |
d00806b1 | 3013 | |
f0c6d4d2 KS |
3014 | /* file_update_time outside page_lock */ |
3015 | if (vma->vm_file && !vma->vm_ops->page_mkwrite) | |
3016 | file_update_time(vma->vm_file); | |
b827e496 | 3017 | |
1d65f86d | 3018 | return ret; |
54cb8821 | 3019 | } |
d00806b1 | 3020 | |
9a95f3cf PC |
3021 | /* |
3022 | * We enter with non-exclusive mmap_sem (to exclude vma changes, | |
3023 | * but allow concurrent faults). | |
3024 | * The mmap_sem may have been released depending on flags and our | |
3025 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
3026 | */ | |
54cb8821 NP |
3027 | static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
3028 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
30c9f3a9 | 3029 | unsigned int flags, pte_t orig_pte) |
54cb8821 NP |
3030 | { |
3031 | pgoff_t pgoff = (((address & PAGE_MASK) | |
0da7e01f | 3032 | - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
54cb8821 | 3033 | |
16abfa08 | 3034 | pte_unmap(page_table); |
e655fb29 KS |
3035 | if (!(flags & FAULT_FLAG_WRITE)) |
3036 | return do_read_fault(mm, vma, address, pmd, pgoff, flags, | |
3037 | orig_pte); | |
ec47c3b9 KS |
3038 | if (!(vma->vm_flags & VM_SHARED)) |
3039 | return do_cow_fault(mm, vma, address, pmd, pgoff, flags, | |
3040 | orig_pte); | |
f0c6d4d2 | 3041 | return do_shared_fault(mm, vma, address, pmd, pgoff, flags, orig_pte); |
54cb8821 NP |
3042 | } |
3043 | ||
1da177e4 LT |
3044 | /* |
3045 | * Fault of a previously existing named mapping. Repopulate the pte | |
3046 | * from the encoded file_pte if possible. This enables swappable | |
3047 | * nonlinear vmas. | |
8f4e2101 HD |
3048 | * |
3049 | * We enter with non-exclusive mmap_sem (to exclude vma changes, | |
3050 | * but allow concurrent faults), and pte mapped but not yet locked. | |
9a95f3cf PC |
3051 | * We return with pte unmapped and unlocked. |
3052 | * The mmap_sem may have been released depending on flags and our | |
3053 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
1da177e4 | 3054 | */ |
d0217ac0 | 3055 | static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
65500d23 | 3056 | unsigned long address, pte_t *page_table, pmd_t *pmd, |
30c9f3a9 | 3057 | unsigned int flags, pte_t orig_pte) |
1da177e4 | 3058 | { |
65500d23 | 3059 | pgoff_t pgoff; |
1da177e4 | 3060 | |
30c9f3a9 LT |
3061 | flags |= FAULT_FLAG_NONLINEAR; |
3062 | ||
4c21e2f2 | 3063 | if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) |
83c54070 | 3064 | return 0; |
1da177e4 | 3065 | |
2509ef26 | 3066 | if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) { |
65500d23 HD |
3067 | /* |
3068 | * Page table corrupted: show pte and kill process. | |
3069 | */ | |
3dc14741 | 3070 | print_bad_pte(vma, address, orig_pte, NULL); |
d99be1a8 | 3071 | return VM_FAULT_SIGBUS; |
65500d23 | 3072 | } |
65500d23 HD |
3073 | |
3074 | pgoff = pte_to_pgoff(orig_pte); | |
e655fb29 KS |
3075 | if (!(flags & FAULT_FLAG_WRITE)) |
3076 | return do_read_fault(mm, vma, address, pmd, pgoff, flags, | |
3077 | orig_pte); | |
ec47c3b9 KS |
3078 | if (!(vma->vm_flags & VM_SHARED)) |
3079 | return do_cow_fault(mm, vma, address, pmd, pgoff, flags, | |
3080 | orig_pte); | |
f0c6d4d2 | 3081 | return do_shared_fault(mm, vma, address, pmd, pgoff, flags, orig_pte); |
1da177e4 LT |
3082 | } |
3083 | ||
b19a9939 | 3084 | static int numa_migrate_prep(struct page *page, struct vm_area_struct *vma, |
04bb2f94 RR |
3085 | unsigned long addr, int page_nid, |
3086 | int *flags) | |
9532fec1 MG |
3087 | { |
3088 | get_page(page); | |
3089 | ||
3090 | count_vm_numa_event(NUMA_HINT_FAULTS); | |
04bb2f94 | 3091 | if (page_nid == numa_node_id()) { |
9532fec1 | 3092 | count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL); |
04bb2f94 RR |
3093 | *flags |= TNF_FAULT_LOCAL; |
3094 | } | |
9532fec1 MG |
3095 | |
3096 | return mpol_misplaced(page, vma, addr); | |
3097 | } | |
3098 | ||
b19a9939 | 3099 | static int do_numa_page(struct mm_struct *mm, struct vm_area_struct *vma, |
d10e63f2 MG |
3100 | unsigned long addr, pte_t pte, pte_t *ptep, pmd_t *pmd) |
3101 | { | |
4daae3b4 | 3102 | struct page *page = NULL; |
d10e63f2 | 3103 | spinlock_t *ptl; |
8191acbd | 3104 | int page_nid = -1; |
90572890 | 3105 | int last_cpupid; |
cbee9f88 | 3106 | int target_nid; |
b8593bfd | 3107 | bool migrated = false; |
6688cc05 | 3108 | int flags = 0; |
d10e63f2 MG |
3109 | |
3110 | /* | |
3111 | * The "pte" at this point cannot be used safely without | |
3112 | * validation through pte_unmap_same(). It's of NUMA type but | |
3113 | * the pfn may be screwed if the read is non atomic. | |
3114 | * | |
3115 | * ptep_modify_prot_start is not called as this is clearing | |
3116 | * the _PAGE_NUMA bit and it is not really expected that there | |
3117 | * would be concurrent hardware modifications to the PTE. | |
3118 | */ | |
3119 | ptl = pte_lockptr(mm, pmd); | |
3120 | spin_lock(ptl); | |
4daae3b4 MG |
3121 | if (unlikely(!pte_same(*ptep, pte))) { |
3122 | pte_unmap_unlock(ptep, ptl); | |
3123 | goto out; | |
3124 | } | |
3125 | ||
d10e63f2 MG |
3126 | pte = pte_mknonnuma(pte); |
3127 | set_pte_at(mm, addr, ptep, pte); | |
3128 | update_mmu_cache(vma, addr, ptep); | |
3129 | ||
3130 | page = vm_normal_page(vma, addr, pte); | |
3131 | if (!page) { | |
3132 | pte_unmap_unlock(ptep, ptl); | |
3133 | return 0; | |
3134 | } | |
a1a46184 | 3135 | BUG_ON(is_zero_pfn(page_to_pfn(page))); |
d10e63f2 | 3136 | |
6688cc05 PZ |
3137 | /* |
3138 | * Avoid grouping on DSO/COW pages in specific and RO pages | |
3139 | * in general, RO pages shouldn't hurt as much anyway since | |
3140 | * they can be in shared cache state. | |
3141 | */ | |
3142 | if (!pte_write(pte)) | |
3143 | flags |= TNF_NO_GROUP; | |
3144 | ||
dabe1d99 RR |
3145 | /* |
3146 | * Flag if the page is shared between multiple address spaces. This | |
3147 | * is later used when determining whether to group tasks together | |
3148 | */ | |
3149 | if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED)) | |
3150 | flags |= TNF_SHARED; | |
3151 | ||
90572890 | 3152 | last_cpupid = page_cpupid_last(page); |
8191acbd | 3153 | page_nid = page_to_nid(page); |
04bb2f94 | 3154 | target_nid = numa_migrate_prep(page, vma, addr, page_nid, &flags); |
d10e63f2 | 3155 | pte_unmap_unlock(ptep, ptl); |
4daae3b4 | 3156 | if (target_nid == -1) { |
4daae3b4 MG |
3157 | put_page(page); |
3158 | goto out; | |
3159 | } | |
3160 | ||
3161 | /* Migrate to the requested node */ | |
1bc115d8 | 3162 | migrated = migrate_misplaced_page(page, vma, target_nid); |
6688cc05 | 3163 | if (migrated) { |
8191acbd | 3164 | page_nid = target_nid; |
6688cc05 PZ |
3165 | flags |= TNF_MIGRATED; |
3166 | } | |
4daae3b4 MG |
3167 | |
3168 | out: | |
8191acbd | 3169 | if (page_nid != -1) |
6688cc05 | 3170 | task_numa_fault(last_cpupid, page_nid, 1, flags); |
d10e63f2 MG |
3171 | return 0; |
3172 | } | |
3173 | ||
1da177e4 LT |
3174 | /* |
3175 | * These routines also need to handle stuff like marking pages dirty | |
3176 | * and/or accessed for architectures that don't do it in hardware (most | |
3177 | * RISC architectures). The early dirtying is also good on the i386. | |
3178 | * | |
3179 | * There is also a hook called "update_mmu_cache()" that architectures | |
3180 | * with external mmu caches can use to update those (ie the Sparc or | |
3181 | * PowerPC hashed page tables that act as extended TLBs). | |
3182 | * | |
c74df32c HD |
3183 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
3184 | * but allow concurrent faults), and pte mapped but not yet locked. | |
9a95f3cf PC |
3185 | * We return with pte unmapped and unlocked. |
3186 | * | |
3187 | * The mmap_sem may have been released depending on flags and our | |
3188 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
1da177e4 | 3189 | */ |
c0292554 | 3190 | static int handle_pte_fault(struct mm_struct *mm, |
71e3aac0 AA |
3191 | struct vm_area_struct *vma, unsigned long address, |
3192 | pte_t *pte, pmd_t *pmd, unsigned int flags) | |
1da177e4 LT |
3193 | { |
3194 | pte_t entry; | |
8f4e2101 | 3195 | spinlock_t *ptl; |
1da177e4 | 3196 | |
c0d73261 | 3197 | entry = ACCESS_ONCE(*pte); |
1da177e4 | 3198 | if (!pte_present(entry)) { |
65500d23 | 3199 | if (pte_none(entry)) { |
f4b81804 | 3200 | if (vma->vm_ops) { |
3c18ddd1 | 3201 | if (likely(vma->vm_ops->fault)) |
54cb8821 | 3202 | return do_linear_fault(mm, vma, address, |
30c9f3a9 | 3203 | pte, pmd, flags, entry); |
f4b81804 JS |
3204 | } |
3205 | return do_anonymous_page(mm, vma, address, | |
30c9f3a9 | 3206 | pte, pmd, flags); |
65500d23 | 3207 | } |
1da177e4 | 3208 | if (pte_file(entry)) |
d0217ac0 | 3209 | return do_nonlinear_fault(mm, vma, address, |
30c9f3a9 | 3210 | pte, pmd, flags, entry); |
65500d23 | 3211 | return do_swap_page(mm, vma, address, |
30c9f3a9 | 3212 | pte, pmd, flags, entry); |
1da177e4 LT |
3213 | } |
3214 | ||
d10e63f2 MG |
3215 | if (pte_numa(entry)) |
3216 | return do_numa_page(mm, vma, address, entry, pte, pmd); | |
3217 | ||
4c21e2f2 | 3218 | ptl = pte_lockptr(mm, pmd); |
8f4e2101 HD |
3219 | spin_lock(ptl); |
3220 | if (unlikely(!pte_same(*pte, entry))) | |
3221 | goto unlock; | |
30c9f3a9 | 3222 | if (flags & FAULT_FLAG_WRITE) { |
1da177e4 | 3223 | if (!pte_write(entry)) |
8f4e2101 HD |
3224 | return do_wp_page(mm, vma, address, |
3225 | pte, pmd, ptl, entry); | |
1da177e4 LT |
3226 | entry = pte_mkdirty(entry); |
3227 | } | |
3228 | entry = pte_mkyoung(entry); | |
30c9f3a9 | 3229 | if (ptep_set_access_flags(vma, address, pte, entry, flags & FAULT_FLAG_WRITE)) { |
4b3073e1 | 3230 | update_mmu_cache(vma, address, pte); |
1a44e149 AA |
3231 | } else { |
3232 | /* | |
3233 | * This is needed only for protection faults but the arch code | |
3234 | * is not yet telling us if this is a protection fault or not. | |
3235 | * This still avoids useless tlb flushes for .text page faults | |
3236 | * with threads. | |
3237 | */ | |
30c9f3a9 | 3238 | if (flags & FAULT_FLAG_WRITE) |
61c77326 | 3239 | flush_tlb_fix_spurious_fault(vma, address); |
1a44e149 | 3240 | } |
8f4e2101 HD |
3241 | unlock: |
3242 | pte_unmap_unlock(pte, ptl); | |
83c54070 | 3243 | return 0; |
1da177e4 LT |
3244 | } |
3245 | ||
3246 | /* | |
3247 | * By the time we get here, we already hold the mm semaphore | |
9a95f3cf PC |
3248 | * |
3249 | * The mmap_sem may have been released depending on flags and our | |
3250 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
1da177e4 | 3251 | */ |
519e5247 JW |
3252 | static int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
3253 | unsigned long address, unsigned int flags) | |
1da177e4 LT |
3254 | { |
3255 | pgd_t *pgd; | |
3256 | pud_t *pud; | |
3257 | pmd_t *pmd; | |
3258 | pte_t *pte; | |
3259 | ||
ac9b9c66 | 3260 | if (unlikely(is_vm_hugetlb_page(vma))) |
30c9f3a9 | 3261 | return hugetlb_fault(mm, vma, address, flags); |
1da177e4 | 3262 | |
1da177e4 | 3263 | pgd = pgd_offset(mm, address); |
1da177e4 LT |
3264 | pud = pud_alloc(mm, pgd, address); |
3265 | if (!pud) | |
c74df32c | 3266 | return VM_FAULT_OOM; |
1da177e4 LT |
3267 | pmd = pmd_alloc(mm, pud, address); |
3268 | if (!pmd) | |
c74df32c | 3269 | return VM_FAULT_OOM; |
71e3aac0 | 3270 | if (pmd_none(*pmd) && transparent_hugepage_enabled(vma)) { |
c0292554 | 3271 | int ret = VM_FAULT_FALLBACK; |
71e3aac0 | 3272 | if (!vma->vm_ops) |
c0292554 KS |
3273 | ret = do_huge_pmd_anonymous_page(mm, vma, address, |
3274 | pmd, flags); | |
3275 | if (!(ret & VM_FAULT_FALLBACK)) | |
3276 | return ret; | |
71e3aac0 AA |
3277 | } else { |
3278 | pmd_t orig_pmd = *pmd; | |
1f1d06c3 DR |
3279 | int ret; |
3280 | ||
71e3aac0 AA |
3281 | barrier(); |
3282 | if (pmd_trans_huge(orig_pmd)) { | |
a1dd450b WD |
3283 | unsigned int dirty = flags & FAULT_FLAG_WRITE; |
3284 | ||
e53289c0 LT |
3285 | /* |
3286 | * If the pmd is splitting, return and retry the | |
3287 | * the fault. Alternative: wait until the split | |
3288 | * is done, and goto retry. | |
3289 | */ | |
3290 | if (pmd_trans_splitting(orig_pmd)) | |
3291 | return 0; | |
3292 | ||
3d59eebc | 3293 | if (pmd_numa(orig_pmd)) |
4daae3b4 | 3294 | return do_huge_pmd_numa_page(mm, vma, address, |
d10e63f2 MG |
3295 | orig_pmd, pmd); |
3296 | ||
3d59eebc | 3297 | if (dirty && !pmd_write(orig_pmd)) { |
1f1d06c3 DR |
3298 | ret = do_huge_pmd_wp_page(mm, vma, address, pmd, |
3299 | orig_pmd); | |
9845cbbd KS |
3300 | if (!(ret & VM_FAULT_FALLBACK)) |
3301 | return ret; | |
a1dd450b WD |
3302 | } else { |
3303 | huge_pmd_set_accessed(mm, vma, address, pmd, | |
3304 | orig_pmd, dirty); | |
9845cbbd | 3305 | return 0; |
1f1d06c3 | 3306 | } |
71e3aac0 AA |
3307 | } |
3308 | } | |
3309 | ||
3310 | /* | |
3311 | * Use __pte_alloc instead of pte_alloc_map, because we can't | |
3312 | * run pte_offset_map on the pmd, if an huge pmd could | |
3313 | * materialize from under us from a different thread. | |
3314 | */ | |
4fd01770 MG |
3315 | if (unlikely(pmd_none(*pmd)) && |
3316 | unlikely(__pte_alloc(mm, vma, pmd, address))) | |
c74df32c | 3317 | return VM_FAULT_OOM; |
71e3aac0 AA |
3318 | /* if an huge pmd materialized from under us just retry later */ |
3319 | if (unlikely(pmd_trans_huge(*pmd))) | |
3320 | return 0; | |
3321 | /* | |
3322 | * A regular pmd is established and it can't morph into a huge pmd | |
3323 | * from under us anymore at this point because we hold the mmap_sem | |
3324 | * read mode and khugepaged takes it in write mode. So now it's | |
3325 | * safe to run pte_offset_map(). | |
3326 | */ | |
3327 | pte = pte_offset_map(pmd, address); | |
1da177e4 | 3328 | |
30c9f3a9 | 3329 | return handle_pte_fault(mm, vma, address, pte, pmd, flags); |
1da177e4 LT |
3330 | } |
3331 | ||
9a95f3cf PC |
3332 | /* |
3333 | * By the time we get here, we already hold the mm semaphore | |
3334 | * | |
3335 | * The mmap_sem may have been released depending on flags and our | |
3336 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
3337 | */ | |
519e5247 JW |
3338 | int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
3339 | unsigned long address, unsigned int flags) | |
3340 | { | |
3341 | int ret; | |
3342 | ||
3343 | __set_current_state(TASK_RUNNING); | |
3344 | ||
3345 | count_vm_event(PGFAULT); | |
3346 | mem_cgroup_count_vm_event(mm, PGFAULT); | |
3347 | ||
3348 | /* do counter updates before entering really critical section. */ | |
3349 | check_sync_rss_stat(current); | |
3350 | ||
3351 | /* | |
3352 | * Enable the memcg OOM handling for faults triggered in user | |
3353 | * space. Kernel faults are handled more gracefully. | |
3354 | */ | |
3355 | if (flags & FAULT_FLAG_USER) | |
49426420 | 3356 | mem_cgroup_oom_enable(); |
519e5247 JW |
3357 | |
3358 | ret = __handle_mm_fault(mm, vma, address, flags); | |
3359 | ||
49426420 JW |
3360 | if (flags & FAULT_FLAG_USER) { |
3361 | mem_cgroup_oom_disable(); | |
3362 | /* | |
3363 | * The task may have entered a memcg OOM situation but | |
3364 | * if the allocation error was handled gracefully (no | |
3365 | * VM_FAULT_OOM), there is no need to kill anything. | |
3366 | * Just clean up the OOM state peacefully. | |
3367 | */ | |
3368 | if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM)) | |
3369 | mem_cgroup_oom_synchronize(false); | |
3370 | } | |
3812c8c8 | 3371 | |
519e5247 JW |
3372 | return ret; |
3373 | } | |
3374 | ||
1da177e4 LT |
3375 | #ifndef __PAGETABLE_PUD_FOLDED |
3376 | /* | |
3377 | * Allocate page upper directory. | |
872fec16 | 3378 | * We've already handled the fast-path in-line. |
1da177e4 | 3379 | */ |
1bb3630e | 3380 | int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) |
1da177e4 | 3381 | { |
c74df32c HD |
3382 | pud_t *new = pud_alloc_one(mm, address); |
3383 | if (!new) | |
1bb3630e | 3384 | return -ENOMEM; |
1da177e4 | 3385 | |
362a61ad NP |
3386 | smp_wmb(); /* See comment in __pte_alloc */ |
3387 | ||
872fec16 | 3388 | spin_lock(&mm->page_table_lock); |
1bb3630e | 3389 | if (pgd_present(*pgd)) /* Another has populated it */ |
5e541973 | 3390 | pud_free(mm, new); |
1bb3630e HD |
3391 | else |
3392 | pgd_populate(mm, pgd, new); | |
c74df32c | 3393 | spin_unlock(&mm->page_table_lock); |
1bb3630e | 3394 | return 0; |
1da177e4 LT |
3395 | } |
3396 | #endif /* __PAGETABLE_PUD_FOLDED */ | |
3397 | ||
3398 | #ifndef __PAGETABLE_PMD_FOLDED | |
3399 | /* | |
3400 | * Allocate page middle directory. | |
872fec16 | 3401 | * We've already handled the fast-path in-line. |
1da177e4 | 3402 | */ |
1bb3630e | 3403 | int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) |
1da177e4 | 3404 | { |
c74df32c HD |
3405 | pmd_t *new = pmd_alloc_one(mm, address); |
3406 | if (!new) | |
1bb3630e | 3407 | return -ENOMEM; |
1da177e4 | 3408 | |
362a61ad NP |
3409 | smp_wmb(); /* See comment in __pte_alloc */ |
3410 | ||
872fec16 | 3411 | spin_lock(&mm->page_table_lock); |
1da177e4 | 3412 | #ifndef __ARCH_HAS_4LEVEL_HACK |
1bb3630e | 3413 | if (pud_present(*pud)) /* Another has populated it */ |
5e541973 | 3414 | pmd_free(mm, new); |
1bb3630e HD |
3415 | else |
3416 | pud_populate(mm, pud, new); | |
1da177e4 | 3417 | #else |
1bb3630e | 3418 | if (pgd_present(*pud)) /* Another has populated it */ |
5e541973 | 3419 | pmd_free(mm, new); |
1bb3630e HD |
3420 | else |
3421 | pgd_populate(mm, pud, new); | |
1da177e4 | 3422 | #endif /* __ARCH_HAS_4LEVEL_HACK */ |
c74df32c | 3423 | spin_unlock(&mm->page_table_lock); |
1bb3630e | 3424 | return 0; |
e0f39591 | 3425 | } |
1da177e4 LT |
3426 | #endif /* __PAGETABLE_PMD_FOLDED */ |
3427 | ||
1da177e4 LT |
3428 | #if !defined(__HAVE_ARCH_GATE_AREA) |
3429 | ||
3430 | #if defined(AT_SYSINFO_EHDR) | |
5ce7852c | 3431 | static struct vm_area_struct gate_vma; |
1da177e4 LT |
3432 | |
3433 | static int __init gate_vma_init(void) | |
3434 | { | |
3435 | gate_vma.vm_mm = NULL; | |
3436 | gate_vma.vm_start = FIXADDR_USER_START; | |
3437 | gate_vma.vm_end = FIXADDR_USER_END; | |
b6558c4a RM |
3438 | gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC; |
3439 | gate_vma.vm_page_prot = __P101; | |
909af768 | 3440 | |
1da177e4 LT |
3441 | return 0; |
3442 | } | |
3443 | __initcall(gate_vma_init); | |
3444 | #endif | |
3445 | ||
31db58b3 | 3446 | struct vm_area_struct *get_gate_vma(struct mm_struct *mm) |
1da177e4 LT |
3447 | { |
3448 | #ifdef AT_SYSINFO_EHDR | |
3449 | return &gate_vma; | |
3450 | #else | |
3451 | return NULL; | |
3452 | #endif | |
3453 | } | |
3454 | ||
cae5d390 | 3455 | int in_gate_area_no_mm(unsigned long addr) |
1da177e4 LT |
3456 | { |
3457 | #ifdef AT_SYSINFO_EHDR | |
3458 | if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END)) | |
3459 | return 1; | |
3460 | #endif | |
3461 | return 0; | |
3462 | } | |
3463 | ||
3464 | #endif /* __HAVE_ARCH_GATE_AREA */ | |
0ec76a11 | 3465 | |
1b36ba81 | 3466 | static int __follow_pte(struct mm_struct *mm, unsigned long address, |
f8ad0f49 JW |
3467 | pte_t **ptepp, spinlock_t **ptlp) |
3468 | { | |
3469 | pgd_t *pgd; | |
3470 | pud_t *pud; | |
3471 | pmd_t *pmd; | |
3472 | pte_t *ptep; | |
3473 | ||
3474 | pgd = pgd_offset(mm, address); | |
3475 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
3476 | goto out; | |
3477 | ||
3478 | pud = pud_offset(pgd, address); | |
3479 | if (pud_none(*pud) || unlikely(pud_bad(*pud))) | |
3480 | goto out; | |
3481 | ||
3482 | pmd = pmd_offset(pud, address); | |
f66055ab | 3483 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
f8ad0f49 JW |
3484 | if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) |
3485 | goto out; | |
3486 | ||
3487 | /* We cannot handle huge page PFN maps. Luckily they don't exist. */ | |
3488 | if (pmd_huge(*pmd)) | |
3489 | goto out; | |
3490 | ||
3491 | ptep = pte_offset_map_lock(mm, pmd, address, ptlp); | |
3492 | if (!ptep) | |
3493 | goto out; | |
3494 | if (!pte_present(*ptep)) | |
3495 | goto unlock; | |
3496 | *ptepp = ptep; | |
3497 | return 0; | |
3498 | unlock: | |
3499 | pte_unmap_unlock(ptep, *ptlp); | |
3500 | out: | |
3501 | return -EINVAL; | |
3502 | } | |
3503 | ||
1b36ba81 NK |
3504 | static inline int follow_pte(struct mm_struct *mm, unsigned long address, |
3505 | pte_t **ptepp, spinlock_t **ptlp) | |
3506 | { | |
3507 | int res; | |
3508 | ||
3509 | /* (void) is needed to make gcc happy */ | |
3510 | (void) __cond_lock(*ptlp, | |
3511 | !(res = __follow_pte(mm, address, ptepp, ptlp))); | |
3512 | return res; | |
3513 | } | |
3514 | ||
3b6748e2 JW |
3515 | /** |
3516 | * follow_pfn - look up PFN at a user virtual address | |
3517 | * @vma: memory mapping | |
3518 | * @address: user virtual address | |
3519 | * @pfn: location to store found PFN | |
3520 | * | |
3521 | * Only IO mappings and raw PFN mappings are allowed. | |
3522 | * | |
3523 | * Returns zero and the pfn at @pfn on success, -ve otherwise. | |
3524 | */ | |
3525 | int follow_pfn(struct vm_area_struct *vma, unsigned long address, | |
3526 | unsigned long *pfn) | |
3527 | { | |
3528 | int ret = -EINVAL; | |
3529 | spinlock_t *ptl; | |
3530 | pte_t *ptep; | |
3531 | ||
3532 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) | |
3533 | return ret; | |
3534 | ||
3535 | ret = follow_pte(vma->vm_mm, address, &ptep, &ptl); | |
3536 | if (ret) | |
3537 | return ret; | |
3538 | *pfn = pte_pfn(*ptep); | |
3539 | pte_unmap_unlock(ptep, ptl); | |
3540 | return 0; | |
3541 | } | |
3542 | EXPORT_SYMBOL(follow_pfn); | |
3543 | ||
28b2ee20 | 3544 | #ifdef CONFIG_HAVE_IOREMAP_PROT |
d87fe660 | 3545 | int follow_phys(struct vm_area_struct *vma, |
3546 | unsigned long address, unsigned int flags, | |
3547 | unsigned long *prot, resource_size_t *phys) | |
28b2ee20 | 3548 | { |
03668a4d | 3549 | int ret = -EINVAL; |
28b2ee20 RR |
3550 | pte_t *ptep, pte; |
3551 | spinlock_t *ptl; | |
28b2ee20 | 3552 | |
d87fe660 | 3553 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) |
3554 | goto out; | |
28b2ee20 | 3555 | |
03668a4d | 3556 | if (follow_pte(vma->vm_mm, address, &ptep, &ptl)) |
d87fe660 | 3557 | goto out; |
28b2ee20 | 3558 | pte = *ptep; |
03668a4d | 3559 | |
28b2ee20 RR |
3560 | if ((flags & FOLL_WRITE) && !pte_write(pte)) |
3561 | goto unlock; | |
28b2ee20 RR |
3562 | |
3563 | *prot = pgprot_val(pte_pgprot(pte)); | |
03668a4d | 3564 | *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; |
28b2ee20 | 3565 | |
03668a4d | 3566 | ret = 0; |
28b2ee20 RR |
3567 | unlock: |
3568 | pte_unmap_unlock(ptep, ptl); | |
3569 | out: | |
d87fe660 | 3570 | return ret; |
28b2ee20 RR |
3571 | } |
3572 | ||
3573 | int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, | |
3574 | void *buf, int len, int write) | |
3575 | { | |
3576 | resource_size_t phys_addr; | |
3577 | unsigned long prot = 0; | |
2bc7273b | 3578 | void __iomem *maddr; |
28b2ee20 RR |
3579 | int offset = addr & (PAGE_SIZE-1); |
3580 | ||
d87fe660 | 3581 | if (follow_phys(vma, addr, write, &prot, &phys_addr)) |
28b2ee20 RR |
3582 | return -EINVAL; |
3583 | ||
3584 | maddr = ioremap_prot(phys_addr, PAGE_SIZE, prot); | |
3585 | if (write) | |
3586 | memcpy_toio(maddr + offset, buf, len); | |
3587 | else | |
3588 | memcpy_fromio(buf, maddr + offset, len); | |
3589 | iounmap(maddr); | |
3590 | ||
3591 | return len; | |
3592 | } | |
5a73633e | 3593 | EXPORT_SYMBOL_GPL(generic_access_phys); |
28b2ee20 RR |
3594 | #endif |
3595 | ||
0ec76a11 | 3596 | /* |
206cb636 SW |
3597 | * Access another process' address space as given in mm. If non-NULL, use the |
3598 | * given task for page fault accounting. | |
0ec76a11 | 3599 | */ |
206cb636 SW |
3600 | static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, |
3601 | unsigned long addr, void *buf, int len, int write) | |
0ec76a11 | 3602 | { |
0ec76a11 | 3603 | struct vm_area_struct *vma; |
0ec76a11 DH |
3604 | void *old_buf = buf; |
3605 | ||
0ec76a11 | 3606 | down_read(&mm->mmap_sem); |
183ff22b | 3607 | /* ignore errors, just check how much was successfully transferred */ |
0ec76a11 DH |
3608 | while (len) { |
3609 | int bytes, ret, offset; | |
3610 | void *maddr; | |
28b2ee20 | 3611 | struct page *page = NULL; |
0ec76a11 DH |
3612 | |
3613 | ret = get_user_pages(tsk, mm, addr, 1, | |
3614 | write, 1, &page, &vma); | |
28b2ee20 | 3615 | if (ret <= 0) { |
dbffcd03 RR |
3616 | #ifndef CONFIG_HAVE_IOREMAP_PROT |
3617 | break; | |
3618 | #else | |
28b2ee20 RR |
3619 | /* |
3620 | * Check if this is a VM_IO | VM_PFNMAP VMA, which | |
3621 | * we can access using slightly different code. | |
3622 | */ | |
28b2ee20 | 3623 | vma = find_vma(mm, addr); |
fe936dfc | 3624 | if (!vma || vma->vm_start > addr) |
28b2ee20 RR |
3625 | break; |
3626 | if (vma->vm_ops && vma->vm_ops->access) | |
3627 | ret = vma->vm_ops->access(vma, addr, buf, | |
3628 | len, write); | |
3629 | if (ret <= 0) | |
28b2ee20 RR |
3630 | break; |
3631 | bytes = ret; | |
dbffcd03 | 3632 | #endif |
0ec76a11 | 3633 | } else { |
28b2ee20 RR |
3634 | bytes = len; |
3635 | offset = addr & (PAGE_SIZE-1); | |
3636 | if (bytes > PAGE_SIZE-offset) | |
3637 | bytes = PAGE_SIZE-offset; | |
3638 | ||
3639 | maddr = kmap(page); | |
3640 | if (write) { | |
3641 | copy_to_user_page(vma, page, addr, | |
3642 | maddr + offset, buf, bytes); | |
3643 | set_page_dirty_lock(page); | |
3644 | } else { | |
3645 | copy_from_user_page(vma, page, addr, | |
3646 | buf, maddr + offset, bytes); | |
3647 | } | |
3648 | kunmap(page); | |
3649 | page_cache_release(page); | |
0ec76a11 | 3650 | } |
0ec76a11 DH |
3651 | len -= bytes; |
3652 | buf += bytes; | |
3653 | addr += bytes; | |
3654 | } | |
3655 | up_read(&mm->mmap_sem); | |
0ec76a11 DH |
3656 | |
3657 | return buf - old_buf; | |
3658 | } | |
03252919 | 3659 | |
5ddd36b9 | 3660 | /** |
ae91dbfc | 3661 | * access_remote_vm - access another process' address space |
5ddd36b9 SW |
3662 | * @mm: the mm_struct of the target address space |
3663 | * @addr: start address to access | |
3664 | * @buf: source or destination buffer | |
3665 | * @len: number of bytes to transfer | |
3666 | * @write: whether the access is a write | |
3667 | * | |
3668 | * The caller must hold a reference on @mm. | |
3669 | */ | |
3670 | int access_remote_vm(struct mm_struct *mm, unsigned long addr, | |
3671 | void *buf, int len, int write) | |
3672 | { | |
3673 | return __access_remote_vm(NULL, mm, addr, buf, len, write); | |
3674 | } | |
3675 | ||
206cb636 SW |
3676 | /* |
3677 | * Access another process' address space. | |
3678 | * Source/target buffer must be kernel space, | |
3679 | * Do not walk the page table directly, use get_user_pages | |
3680 | */ | |
3681 | int access_process_vm(struct task_struct *tsk, unsigned long addr, | |
3682 | void *buf, int len, int write) | |
3683 | { | |
3684 | struct mm_struct *mm; | |
3685 | int ret; | |
3686 | ||
3687 | mm = get_task_mm(tsk); | |
3688 | if (!mm) | |
3689 | return 0; | |
3690 | ||
3691 | ret = __access_remote_vm(tsk, mm, addr, buf, len, write); | |
3692 | mmput(mm); | |
3693 | ||
3694 | return ret; | |
3695 | } | |
3696 | ||
03252919 AK |
3697 | /* |
3698 | * Print the name of a VMA. | |
3699 | */ | |
3700 | void print_vma_addr(char *prefix, unsigned long ip) | |
3701 | { | |
3702 | struct mm_struct *mm = current->mm; | |
3703 | struct vm_area_struct *vma; | |
3704 | ||
e8bff74a IM |
3705 | /* |
3706 | * Do not print if we are in atomic | |
3707 | * contexts (in exception stacks, etc.): | |
3708 | */ | |
3709 | if (preempt_count()) | |
3710 | return; | |
3711 | ||
03252919 AK |
3712 | down_read(&mm->mmap_sem); |
3713 | vma = find_vma(mm, ip); | |
3714 | if (vma && vma->vm_file) { | |
3715 | struct file *f = vma->vm_file; | |
3716 | char *buf = (char *)__get_free_page(GFP_KERNEL); | |
3717 | if (buf) { | |
2fbc57c5 | 3718 | char *p; |
03252919 | 3719 | |
cf28b486 | 3720 | p = d_path(&f->f_path, buf, PAGE_SIZE); |
03252919 AK |
3721 | if (IS_ERR(p)) |
3722 | p = "?"; | |
2fbc57c5 | 3723 | printk("%s%s[%lx+%lx]", prefix, kbasename(p), |
03252919 AK |
3724 | vma->vm_start, |
3725 | vma->vm_end - vma->vm_start); | |
3726 | free_page((unsigned long)buf); | |
3727 | } | |
3728 | } | |
51a07e50 | 3729 | up_read(&mm->mmap_sem); |
03252919 | 3730 | } |
3ee1afa3 | 3731 | |
662bbcb2 | 3732 | #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP) |
3ee1afa3 NP |
3733 | void might_fault(void) |
3734 | { | |
95156f00 PZ |
3735 | /* |
3736 | * Some code (nfs/sunrpc) uses socket ops on kernel memory while | |
3737 | * holding the mmap_sem, this is safe because kernel memory doesn't | |
3738 | * get paged out, therefore we'll never actually fault, and the | |
3739 | * below annotations will generate false positives. | |
3740 | */ | |
3741 | if (segment_eq(get_fs(), KERNEL_DS)) | |
3742 | return; | |
3743 | ||
3ee1afa3 NP |
3744 | /* |
3745 | * it would be nicer only to annotate paths which are not under | |
3746 | * pagefault_disable, however that requires a larger audit and | |
3747 | * providing helpers like get_user_atomic. | |
3748 | */ | |
662bbcb2 MT |
3749 | if (in_atomic()) |
3750 | return; | |
3751 | ||
3752 | __might_sleep(__FILE__, __LINE__, 0); | |
3753 | ||
3754 | if (current->mm) | |
3ee1afa3 NP |
3755 | might_lock_read(¤t->mm->mmap_sem); |
3756 | } | |
3757 | EXPORT_SYMBOL(might_fault); | |
3758 | #endif | |
47ad8475 AA |
3759 | |
3760 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) | |
3761 | static void clear_gigantic_page(struct page *page, | |
3762 | unsigned long addr, | |
3763 | unsigned int pages_per_huge_page) | |
3764 | { | |
3765 | int i; | |
3766 | struct page *p = page; | |
3767 | ||
3768 | might_sleep(); | |
3769 | for (i = 0; i < pages_per_huge_page; | |
3770 | i++, p = mem_map_next(p, page, i)) { | |
3771 | cond_resched(); | |
3772 | clear_user_highpage(p, addr + i * PAGE_SIZE); | |
3773 | } | |
3774 | } | |
3775 | void clear_huge_page(struct page *page, | |
3776 | unsigned long addr, unsigned int pages_per_huge_page) | |
3777 | { | |
3778 | int i; | |
3779 | ||
3780 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { | |
3781 | clear_gigantic_page(page, addr, pages_per_huge_page); | |
3782 | return; | |
3783 | } | |
3784 | ||
3785 | might_sleep(); | |
3786 | for (i = 0; i < pages_per_huge_page; i++) { | |
3787 | cond_resched(); | |
3788 | clear_user_highpage(page + i, addr + i * PAGE_SIZE); | |
3789 | } | |
3790 | } | |
3791 | ||
3792 | static void copy_user_gigantic_page(struct page *dst, struct page *src, | |
3793 | unsigned long addr, | |
3794 | struct vm_area_struct *vma, | |
3795 | unsigned int pages_per_huge_page) | |
3796 | { | |
3797 | int i; | |
3798 | struct page *dst_base = dst; | |
3799 | struct page *src_base = src; | |
3800 | ||
3801 | for (i = 0; i < pages_per_huge_page; ) { | |
3802 | cond_resched(); | |
3803 | copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma); | |
3804 | ||
3805 | i++; | |
3806 | dst = mem_map_next(dst, dst_base, i); | |
3807 | src = mem_map_next(src, src_base, i); | |
3808 | } | |
3809 | } | |
3810 | ||
3811 | void copy_user_huge_page(struct page *dst, struct page *src, | |
3812 | unsigned long addr, struct vm_area_struct *vma, | |
3813 | unsigned int pages_per_huge_page) | |
3814 | { | |
3815 | int i; | |
3816 | ||
3817 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { | |
3818 | copy_user_gigantic_page(dst, src, addr, vma, | |
3819 | pages_per_huge_page); | |
3820 | return; | |
3821 | } | |
3822 | ||
3823 | might_sleep(); | |
3824 | for (i = 0; i < pages_per_huge_page; i++) { | |
3825 | cond_resched(); | |
3826 | copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE, vma); | |
3827 | } | |
3828 | } | |
3829 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */ | |
49076ec2 | 3830 | |
40b64acd | 3831 | #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS |
b35f1819 KS |
3832 | |
3833 | static struct kmem_cache *page_ptl_cachep; | |
3834 | ||
3835 | void __init ptlock_cache_init(void) | |
3836 | { | |
3837 | page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0, | |
3838 | SLAB_PANIC, NULL); | |
3839 | } | |
3840 | ||
539edb58 | 3841 | bool ptlock_alloc(struct page *page) |
49076ec2 KS |
3842 | { |
3843 | spinlock_t *ptl; | |
3844 | ||
b35f1819 | 3845 | ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL); |
49076ec2 KS |
3846 | if (!ptl) |
3847 | return false; | |
539edb58 | 3848 | page->ptl = ptl; |
49076ec2 KS |
3849 | return true; |
3850 | } | |
3851 | ||
539edb58 | 3852 | void ptlock_free(struct page *page) |
49076ec2 | 3853 | { |
b35f1819 | 3854 | kmem_cache_free(page_ptl_cachep, page->ptl); |
49076ec2 KS |
3855 | } |
3856 | #endif |