]>
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. | |
166f61b9 | 33 | * Idea by Alex Bligh (alex@cconcepts.co.uk) |
1da177e4 LT |
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> | |
6e84f315 | 43 | #include <linux/sched/mm.h> |
f7ccbae4 | 44 | #include <linux/sched/coredump.h> |
6a3827d7 | 45 | #include <linux/sched/numa_balancing.h> |
29930025 | 46 | #include <linux/sched/task.h> |
1da177e4 LT |
47 | #include <linux/hugetlb.h> |
48 | #include <linux/mman.h> | |
49 | #include <linux/swap.h> | |
50 | #include <linux/highmem.h> | |
51 | #include <linux/pagemap.h> | |
5042db43 | 52 | #include <linux/memremap.h> |
9a840895 | 53 | #include <linux/ksm.h> |
1da177e4 | 54 | #include <linux/rmap.h> |
b95f1b31 | 55 | #include <linux/export.h> |
0ff92245 | 56 | #include <linux/delayacct.h> |
1da177e4 | 57 | #include <linux/init.h> |
01c8f1c4 | 58 | #include <linux/pfn_t.h> |
edc79b2a | 59 | #include <linux/writeback.h> |
8a9f3ccd | 60 | #include <linux/memcontrol.h> |
cddb8a5c | 61 | #include <linux/mmu_notifier.h> |
3dc14741 HD |
62 | #include <linux/swapops.h> |
63 | #include <linux/elf.h> | |
5a0e3ad6 | 64 | #include <linux/gfp.h> |
4daae3b4 | 65 | #include <linux/migrate.h> |
2fbc57c5 | 66 | #include <linux/string.h> |
0abdd7a8 | 67 | #include <linux/dma-debug.h> |
1592eef0 | 68 | #include <linux/debugfs.h> |
6b251fc9 | 69 | #include <linux/userfaultfd_k.h> |
bc2466e4 | 70 | #include <linux/dax.h> |
6b31d595 | 71 | #include <linux/oom.h> |
98fa15f3 | 72 | #include <linux/numa.h> |
1da177e4 | 73 | |
6952b61d | 74 | #include <asm/io.h> |
33a709b2 | 75 | #include <asm/mmu_context.h> |
1da177e4 | 76 | #include <asm/pgalloc.h> |
7c0f6ba6 | 77 | #include <linux/uaccess.h> |
1da177e4 LT |
78 | #include <asm/tlb.h> |
79 | #include <asm/tlbflush.h> | |
80 | #include <asm/pgtable.h> | |
81 | ||
42b77728 JB |
82 | #include "internal.h" |
83 | ||
af27d940 | 84 | #if defined(LAST_CPUPID_NOT_IN_PAGE_FLAGS) && !defined(CONFIG_COMPILE_TEST) |
90572890 | 85 | #warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid. |
75980e97 PZ |
86 | #endif |
87 | ||
d41dee36 | 88 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
89 | /* use the per-pgdat data instead for discontigmem - mbligh */ |
90 | unsigned long max_mapnr; | |
1da177e4 | 91 | EXPORT_SYMBOL(max_mapnr); |
166f61b9 TH |
92 | |
93 | struct page *mem_map; | |
1da177e4 LT |
94 | EXPORT_SYMBOL(mem_map); |
95 | #endif | |
96 | ||
1da177e4 LT |
97 | /* |
98 | * A number of key systems in x86 including ioremap() rely on the assumption | |
99 | * that high_memory defines the upper bound on direct map memory, then end | |
100 | * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and | |
101 | * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL | |
102 | * and ZONE_HIGHMEM. | |
103 | */ | |
166f61b9 | 104 | void *high_memory; |
1da177e4 | 105 | EXPORT_SYMBOL(high_memory); |
1da177e4 | 106 | |
32a93233 IM |
107 | /* |
108 | * Randomize the address space (stacks, mmaps, brk, etc.). | |
109 | * | |
110 | * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization, | |
111 | * as ancient (libc5 based) binaries can segfault. ) | |
112 | */ | |
113 | int randomize_va_space __read_mostly = | |
114 | #ifdef CONFIG_COMPAT_BRK | |
115 | 1; | |
116 | #else | |
117 | 2; | |
118 | #endif | |
a62eaf15 AK |
119 | |
120 | static int __init disable_randmaps(char *s) | |
121 | { | |
122 | randomize_va_space = 0; | |
9b41046c | 123 | return 1; |
a62eaf15 AK |
124 | } |
125 | __setup("norandmaps", disable_randmaps); | |
126 | ||
62eede62 | 127 | unsigned long zero_pfn __read_mostly; |
0b70068e AB |
128 | EXPORT_SYMBOL(zero_pfn); |
129 | ||
166f61b9 TH |
130 | unsigned long highest_memmap_pfn __read_mostly; |
131 | ||
a13ea5b7 HD |
132 | /* |
133 | * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init() | |
134 | */ | |
135 | static int __init init_zero_pfn(void) | |
136 | { | |
137 | zero_pfn = page_to_pfn(ZERO_PAGE(0)); | |
138 | return 0; | |
139 | } | |
140 | core_initcall(init_zero_pfn); | |
a62eaf15 | 141 | |
d559db08 | 142 | |
34e55232 KH |
143 | #if defined(SPLIT_RSS_COUNTING) |
144 | ||
ea48cf78 | 145 | void sync_mm_rss(struct mm_struct *mm) |
34e55232 KH |
146 | { |
147 | int i; | |
148 | ||
149 | for (i = 0; i < NR_MM_COUNTERS; i++) { | |
05af2e10 DR |
150 | if (current->rss_stat.count[i]) { |
151 | add_mm_counter(mm, i, current->rss_stat.count[i]); | |
152 | current->rss_stat.count[i] = 0; | |
34e55232 KH |
153 | } |
154 | } | |
05af2e10 | 155 | current->rss_stat.events = 0; |
34e55232 KH |
156 | } |
157 | ||
158 | static void add_mm_counter_fast(struct mm_struct *mm, int member, int val) | |
159 | { | |
160 | struct task_struct *task = current; | |
161 | ||
162 | if (likely(task->mm == mm)) | |
163 | task->rss_stat.count[member] += val; | |
164 | else | |
165 | add_mm_counter(mm, member, val); | |
166 | } | |
167 | #define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1) | |
168 | #define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1) | |
169 | ||
170 | /* sync counter once per 64 page faults */ | |
171 | #define TASK_RSS_EVENTS_THRESH (64) | |
172 | static void check_sync_rss_stat(struct task_struct *task) | |
173 | { | |
174 | if (unlikely(task != current)) | |
175 | return; | |
176 | if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH)) | |
ea48cf78 | 177 | sync_mm_rss(task->mm); |
34e55232 | 178 | } |
9547d01b | 179 | #else /* SPLIT_RSS_COUNTING */ |
34e55232 KH |
180 | |
181 | #define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member) | |
182 | #define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member) | |
183 | ||
184 | static void check_sync_rss_stat(struct task_struct *task) | |
185 | { | |
186 | } | |
187 | ||
9547d01b PZ |
188 | #endif /* SPLIT_RSS_COUNTING */ |
189 | ||
1da177e4 LT |
190 | /* |
191 | * Note: this doesn't free the actual pages themselves. That | |
192 | * has been handled earlier when unmapping all the memory regions. | |
193 | */ | |
9e1b32ca BH |
194 | static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd, |
195 | unsigned long addr) | |
1da177e4 | 196 | { |
2f569afd | 197 | pgtable_t token = pmd_pgtable(*pmd); |
e0da382c | 198 | pmd_clear(pmd); |
9e1b32ca | 199 | pte_free_tlb(tlb, token, addr); |
c4812909 | 200 | mm_dec_nr_ptes(tlb->mm); |
1da177e4 LT |
201 | } |
202 | ||
e0da382c HD |
203 | static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud, |
204 | unsigned long addr, unsigned long end, | |
205 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
206 | { |
207 | pmd_t *pmd; | |
208 | unsigned long next; | |
e0da382c | 209 | unsigned long start; |
1da177e4 | 210 | |
e0da382c | 211 | start = addr; |
1da177e4 | 212 | pmd = pmd_offset(pud, addr); |
1da177e4 LT |
213 | do { |
214 | next = pmd_addr_end(addr, end); | |
215 | if (pmd_none_or_clear_bad(pmd)) | |
216 | continue; | |
9e1b32ca | 217 | free_pte_range(tlb, pmd, addr); |
1da177e4 LT |
218 | } while (pmd++, addr = next, addr != end); |
219 | ||
e0da382c HD |
220 | start &= PUD_MASK; |
221 | if (start < floor) | |
222 | return; | |
223 | if (ceiling) { | |
224 | ceiling &= PUD_MASK; | |
225 | if (!ceiling) | |
226 | return; | |
1da177e4 | 227 | } |
e0da382c HD |
228 | if (end - 1 > ceiling - 1) |
229 | return; | |
230 | ||
231 | pmd = pmd_offset(pud, start); | |
232 | pud_clear(pud); | |
9e1b32ca | 233 | pmd_free_tlb(tlb, pmd, start); |
dc6c9a35 | 234 | mm_dec_nr_pmds(tlb->mm); |
1da177e4 LT |
235 | } |
236 | ||
c2febafc | 237 | static inline void free_pud_range(struct mmu_gather *tlb, p4d_t *p4d, |
e0da382c HD |
238 | unsigned long addr, unsigned long end, |
239 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
240 | { |
241 | pud_t *pud; | |
242 | unsigned long next; | |
e0da382c | 243 | unsigned long start; |
1da177e4 | 244 | |
e0da382c | 245 | start = addr; |
c2febafc | 246 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
247 | do { |
248 | next = pud_addr_end(addr, end); | |
249 | if (pud_none_or_clear_bad(pud)) | |
250 | continue; | |
e0da382c | 251 | free_pmd_range(tlb, pud, addr, next, floor, ceiling); |
1da177e4 LT |
252 | } while (pud++, addr = next, addr != end); |
253 | ||
c2febafc KS |
254 | start &= P4D_MASK; |
255 | if (start < floor) | |
256 | return; | |
257 | if (ceiling) { | |
258 | ceiling &= P4D_MASK; | |
259 | if (!ceiling) | |
260 | return; | |
261 | } | |
262 | if (end - 1 > ceiling - 1) | |
263 | return; | |
264 | ||
265 | pud = pud_offset(p4d, start); | |
266 | p4d_clear(p4d); | |
267 | pud_free_tlb(tlb, pud, start); | |
b4e98d9a | 268 | mm_dec_nr_puds(tlb->mm); |
c2febafc KS |
269 | } |
270 | ||
271 | static inline void free_p4d_range(struct mmu_gather *tlb, pgd_t *pgd, | |
272 | unsigned long addr, unsigned long end, | |
273 | unsigned long floor, unsigned long ceiling) | |
274 | { | |
275 | p4d_t *p4d; | |
276 | unsigned long next; | |
277 | unsigned long start; | |
278 | ||
279 | start = addr; | |
280 | p4d = p4d_offset(pgd, addr); | |
281 | do { | |
282 | next = p4d_addr_end(addr, end); | |
283 | if (p4d_none_or_clear_bad(p4d)) | |
284 | continue; | |
285 | free_pud_range(tlb, p4d, addr, next, floor, ceiling); | |
286 | } while (p4d++, addr = next, addr != end); | |
287 | ||
e0da382c HD |
288 | start &= PGDIR_MASK; |
289 | if (start < floor) | |
290 | return; | |
291 | if (ceiling) { | |
292 | ceiling &= PGDIR_MASK; | |
293 | if (!ceiling) | |
294 | return; | |
1da177e4 | 295 | } |
e0da382c HD |
296 | if (end - 1 > ceiling - 1) |
297 | return; | |
298 | ||
c2febafc | 299 | p4d = p4d_offset(pgd, start); |
e0da382c | 300 | pgd_clear(pgd); |
c2febafc | 301 | p4d_free_tlb(tlb, p4d, start); |
1da177e4 LT |
302 | } |
303 | ||
304 | /* | |
e0da382c | 305 | * This function frees user-level page tables of a process. |
1da177e4 | 306 | */ |
42b77728 | 307 | void free_pgd_range(struct mmu_gather *tlb, |
e0da382c HD |
308 | unsigned long addr, unsigned long end, |
309 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
310 | { |
311 | pgd_t *pgd; | |
312 | unsigned long next; | |
e0da382c HD |
313 | |
314 | /* | |
315 | * The next few lines have given us lots of grief... | |
316 | * | |
317 | * Why are we testing PMD* at this top level? Because often | |
318 | * there will be no work to do at all, and we'd prefer not to | |
319 | * go all the way down to the bottom just to discover that. | |
320 | * | |
321 | * Why all these "- 1"s? Because 0 represents both the bottom | |
322 | * of the address space and the top of it (using -1 for the | |
323 | * top wouldn't help much: the masks would do the wrong thing). | |
324 | * The rule is that addr 0 and floor 0 refer to the bottom of | |
325 | * the address space, but end 0 and ceiling 0 refer to the top | |
326 | * Comparisons need to use "end - 1" and "ceiling - 1" (though | |
327 | * that end 0 case should be mythical). | |
328 | * | |
329 | * Wherever addr is brought up or ceiling brought down, we must | |
330 | * be careful to reject "the opposite 0" before it confuses the | |
331 | * subsequent tests. But what about where end is brought down | |
332 | * by PMD_SIZE below? no, end can't go down to 0 there. | |
333 | * | |
334 | * Whereas we round start (addr) and ceiling down, by different | |
335 | * masks at different levels, in order to test whether a table | |
336 | * now has no other vmas using it, so can be freed, we don't | |
337 | * bother to round floor or end up - the tests don't need that. | |
338 | */ | |
1da177e4 | 339 | |
e0da382c HD |
340 | addr &= PMD_MASK; |
341 | if (addr < floor) { | |
342 | addr += PMD_SIZE; | |
343 | if (!addr) | |
344 | return; | |
345 | } | |
346 | if (ceiling) { | |
347 | ceiling &= PMD_MASK; | |
348 | if (!ceiling) | |
349 | return; | |
350 | } | |
351 | if (end - 1 > ceiling - 1) | |
352 | end -= PMD_SIZE; | |
353 | if (addr > end - 1) | |
354 | return; | |
07e32661 AK |
355 | /* |
356 | * We add page table cache pages with PAGE_SIZE, | |
357 | * (see pte_free_tlb()), flush the tlb if we need | |
358 | */ | |
ed6a7935 | 359 | tlb_change_page_size(tlb, PAGE_SIZE); |
42b77728 | 360 | pgd = pgd_offset(tlb->mm, addr); |
1da177e4 LT |
361 | do { |
362 | next = pgd_addr_end(addr, end); | |
363 | if (pgd_none_or_clear_bad(pgd)) | |
364 | continue; | |
c2febafc | 365 | free_p4d_range(tlb, pgd, addr, next, floor, ceiling); |
1da177e4 | 366 | } while (pgd++, addr = next, addr != end); |
e0da382c HD |
367 | } |
368 | ||
42b77728 | 369 | void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma, |
3bf5ee95 | 370 | unsigned long floor, unsigned long ceiling) |
e0da382c HD |
371 | { |
372 | while (vma) { | |
373 | struct vm_area_struct *next = vma->vm_next; | |
374 | unsigned long addr = vma->vm_start; | |
375 | ||
8f4f8c16 | 376 | /* |
25d9e2d1 | 377 | * Hide vma from rmap and truncate_pagecache before freeing |
378 | * pgtables | |
8f4f8c16 | 379 | */ |
5beb4930 | 380 | unlink_anon_vmas(vma); |
8f4f8c16 HD |
381 | unlink_file_vma(vma); |
382 | ||
9da61aef | 383 | if (is_vm_hugetlb_page(vma)) { |
3bf5ee95 | 384 | hugetlb_free_pgd_range(tlb, addr, vma->vm_end, |
166f61b9 | 385 | floor, next ? next->vm_start : ceiling); |
3bf5ee95 HD |
386 | } else { |
387 | /* | |
388 | * Optimization: gather nearby vmas into one call down | |
389 | */ | |
390 | while (next && next->vm_start <= vma->vm_end + PMD_SIZE | |
4866920b | 391 | && !is_vm_hugetlb_page(next)) { |
3bf5ee95 HD |
392 | vma = next; |
393 | next = vma->vm_next; | |
5beb4930 | 394 | unlink_anon_vmas(vma); |
8f4f8c16 | 395 | unlink_file_vma(vma); |
3bf5ee95 HD |
396 | } |
397 | free_pgd_range(tlb, addr, vma->vm_end, | |
166f61b9 | 398 | floor, next ? next->vm_start : ceiling); |
3bf5ee95 | 399 | } |
e0da382c HD |
400 | vma = next; |
401 | } | |
1da177e4 LT |
402 | } |
403 | ||
4cf58924 | 404 | int __pte_alloc(struct mm_struct *mm, pmd_t *pmd) |
1da177e4 | 405 | { |
c4088ebd | 406 | spinlock_t *ptl; |
4cf58924 | 407 | pgtable_t new = pte_alloc_one(mm); |
1bb3630e HD |
408 | if (!new) |
409 | return -ENOMEM; | |
410 | ||
362a61ad NP |
411 | /* |
412 | * Ensure all pte setup (eg. pte page lock and page clearing) are | |
413 | * visible before the pte is made visible to other CPUs by being | |
414 | * put into page tables. | |
415 | * | |
416 | * The other side of the story is the pointer chasing in the page | |
417 | * table walking code (when walking the page table without locking; | |
418 | * ie. most of the time). Fortunately, these data accesses consist | |
419 | * of a chain of data-dependent loads, meaning most CPUs (alpha | |
420 | * being the notable exception) will already guarantee loads are | |
421 | * seen in-order. See the alpha page table accessors for the | |
422 | * smp_read_barrier_depends() barriers in page table walking code. | |
423 | */ | |
424 | smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */ | |
425 | ||
c4088ebd | 426 | ptl = pmd_lock(mm, pmd); |
8ac1f832 | 427 | if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ |
c4812909 | 428 | mm_inc_nr_ptes(mm); |
1da177e4 | 429 | pmd_populate(mm, pmd, new); |
2f569afd | 430 | new = NULL; |
4b471e88 | 431 | } |
c4088ebd | 432 | spin_unlock(ptl); |
2f569afd MS |
433 | if (new) |
434 | pte_free(mm, new); | |
1bb3630e | 435 | return 0; |
1da177e4 LT |
436 | } |
437 | ||
4cf58924 | 438 | int __pte_alloc_kernel(pmd_t *pmd) |
1da177e4 | 439 | { |
4cf58924 | 440 | pte_t *new = pte_alloc_one_kernel(&init_mm); |
1bb3630e HD |
441 | if (!new) |
442 | return -ENOMEM; | |
443 | ||
362a61ad NP |
444 | smp_wmb(); /* See comment in __pte_alloc */ |
445 | ||
1bb3630e | 446 | spin_lock(&init_mm.page_table_lock); |
8ac1f832 | 447 | if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ |
1bb3630e | 448 | pmd_populate_kernel(&init_mm, pmd, new); |
2f569afd | 449 | new = NULL; |
4b471e88 | 450 | } |
1bb3630e | 451 | spin_unlock(&init_mm.page_table_lock); |
2f569afd MS |
452 | if (new) |
453 | pte_free_kernel(&init_mm, new); | |
1bb3630e | 454 | return 0; |
1da177e4 LT |
455 | } |
456 | ||
d559db08 KH |
457 | static inline void init_rss_vec(int *rss) |
458 | { | |
459 | memset(rss, 0, sizeof(int) * NR_MM_COUNTERS); | |
460 | } | |
461 | ||
462 | static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss) | |
ae859762 | 463 | { |
d559db08 KH |
464 | int i; |
465 | ||
34e55232 | 466 | if (current->mm == mm) |
05af2e10 | 467 | sync_mm_rss(mm); |
d559db08 KH |
468 | for (i = 0; i < NR_MM_COUNTERS; i++) |
469 | if (rss[i]) | |
470 | add_mm_counter(mm, i, rss[i]); | |
ae859762 HD |
471 | } |
472 | ||
b5810039 | 473 | /* |
6aab341e LT |
474 | * This function is called to print an error when a bad pte |
475 | * is found. For example, we might have a PFN-mapped pte in | |
476 | * a region that doesn't allow it. | |
b5810039 NP |
477 | * |
478 | * The calling function must still handle the error. | |
479 | */ | |
3dc14741 HD |
480 | static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr, |
481 | pte_t pte, struct page *page) | |
b5810039 | 482 | { |
3dc14741 | 483 | pgd_t *pgd = pgd_offset(vma->vm_mm, addr); |
c2febafc KS |
484 | p4d_t *p4d = p4d_offset(pgd, addr); |
485 | pud_t *pud = pud_offset(p4d, addr); | |
3dc14741 HD |
486 | pmd_t *pmd = pmd_offset(pud, addr); |
487 | struct address_space *mapping; | |
488 | pgoff_t index; | |
d936cf9b HD |
489 | static unsigned long resume; |
490 | static unsigned long nr_shown; | |
491 | static unsigned long nr_unshown; | |
492 | ||
493 | /* | |
494 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
495 | * or allow a steady drip of one report per second. | |
496 | */ | |
497 | if (nr_shown == 60) { | |
498 | if (time_before(jiffies, resume)) { | |
499 | nr_unshown++; | |
500 | return; | |
501 | } | |
502 | if (nr_unshown) { | |
1170532b JP |
503 | pr_alert("BUG: Bad page map: %lu messages suppressed\n", |
504 | nr_unshown); | |
d936cf9b HD |
505 | nr_unshown = 0; |
506 | } | |
507 | nr_shown = 0; | |
508 | } | |
509 | if (nr_shown++ == 0) | |
510 | resume = jiffies + 60 * HZ; | |
3dc14741 HD |
511 | |
512 | mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL; | |
513 | index = linear_page_index(vma, addr); | |
514 | ||
1170532b JP |
515 | pr_alert("BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n", |
516 | current->comm, | |
517 | (long long)pte_val(pte), (long long)pmd_val(*pmd)); | |
718a3821 | 518 | if (page) |
f0b791a3 | 519 | dump_page(page, "bad pte"); |
1170532b JP |
520 | pr_alert("addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n", |
521 | (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index); | |
d75f773c | 522 | pr_alert("file:%pD fault:%ps mmap:%ps readpage:%ps\n", |
2682582a KK |
523 | vma->vm_file, |
524 | vma->vm_ops ? vma->vm_ops->fault : NULL, | |
525 | vma->vm_file ? vma->vm_file->f_op->mmap : NULL, | |
526 | mapping ? mapping->a_ops->readpage : NULL); | |
b5810039 | 527 | dump_stack(); |
373d4d09 | 528 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
b5810039 NP |
529 | } |
530 | ||
ee498ed7 | 531 | /* |
7e675137 | 532 | * vm_normal_page -- This function gets the "struct page" associated with a pte. |
6aab341e | 533 | * |
7e675137 NP |
534 | * "Special" mappings do not wish to be associated with a "struct page" (either |
535 | * it doesn't exist, or it exists but they don't want to touch it). In this | |
536 | * case, NULL is returned here. "Normal" mappings do have a struct page. | |
b379d790 | 537 | * |
7e675137 NP |
538 | * There are 2 broad cases. Firstly, an architecture may define a pte_special() |
539 | * pte bit, in which case this function is trivial. Secondly, an architecture | |
540 | * may not have a spare pte bit, which requires a more complicated scheme, | |
541 | * described below. | |
542 | * | |
543 | * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a | |
544 | * special mapping (even if there are underlying and valid "struct pages"). | |
545 | * COWed pages of a VM_PFNMAP are always normal. | |
6aab341e | 546 | * |
b379d790 JH |
547 | * The way we recognize COWed pages within VM_PFNMAP mappings is through the |
548 | * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit | |
7e675137 NP |
549 | * set, and the vm_pgoff will point to the first PFN mapped: thus every special |
550 | * mapping will always honor the rule | |
6aab341e LT |
551 | * |
552 | * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) | |
553 | * | |
7e675137 NP |
554 | * And for normal mappings this is false. |
555 | * | |
556 | * This restricts such mappings to be a linear translation from virtual address | |
557 | * to pfn. To get around this restriction, we allow arbitrary mappings so long | |
558 | * as the vma is not a COW mapping; in that case, we know that all ptes are | |
559 | * special (because none can have been COWed). | |
b379d790 | 560 | * |
b379d790 | 561 | * |
7e675137 | 562 | * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. |
b379d790 JH |
563 | * |
564 | * VM_MIXEDMAP mappings can likewise contain memory with or without "struct | |
565 | * page" backing, however the difference is that _all_ pages with a struct | |
566 | * page (that is, those where pfn_valid is true) are refcounted and considered | |
567 | * normal pages by the VM. The disadvantage is that pages are refcounted | |
568 | * (which can be slower and simply not an option for some PFNMAP users). The | |
569 | * advantage is that we don't have to follow the strict linearity rule of | |
570 | * PFNMAP mappings in order to support COWable mappings. | |
571 | * | |
ee498ed7 | 572 | */ |
df6ad698 JG |
573 | struct page *_vm_normal_page(struct vm_area_struct *vma, unsigned long addr, |
574 | pte_t pte, bool with_public_device) | |
ee498ed7 | 575 | { |
22b31eec | 576 | unsigned long pfn = pte_pfn(pte); |
7e675137 | 577 | |
00b3a331 | 578 | if (IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL)) { |
b38af472 | 579 | if (likely(!pte_special(pte))) |
22b31eec | 580 | goto check_pfn; |
667a0a06 DV |
581 | if (vma->vm_ops && vma->vm_ops->find_special_page) |
582 | return vma->vm_ops->find_special_page(vma, addr); | |
a13ea5b7 HD |
583 | if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) |
584 | return NULL; | |
df6ad698 JG |
585 | if (is_zero_pfn(pfn)) |
586 | return NULL; | |
587 | ||
588 | /* | |
589 | * Device public pages are special pages (they are ZONE_DEVICE | |
590 | * pages but different from persistent memory). They behave | |
591 | * allmost like normal pages. The difference is that they are | |
592 | * not on the lru and thus should never be involve with any- | |
593 | * thing that involve lru manipulation (mlock, numa balancing, | |
594 | * ...). | |
595 | * | |
596 | * This is why we still want to return NULL for such page from | |
597 | * vm_normal_page() so that we do not have to special case all | |
598 | * call site of vm_normal_page(). | |
599 | */ | |
7d790d2d | 600 | if (likely(pfn <= highest_memmap_pfn)) { |
df6ad698 JG |
601 | struct page *page = pfn_to_page(pfn); |
602 | ||
603 | if (is_device_public_page(page)) { | |
604 | if (with_public_device) | |
605 | return page; | |
606 | return NULL; | |
607 | } | |
608 | } | |
e1fb4a08 DJ |
609 | |
610 | if (pte_devmap(pte)) | |
611 | return NULL; | |
612 | ||
df6ad698 | 613 | print_bad_pte(vma, addr, pte, NULL); |
7e675137 NP |
614 | return NULL; |
615 | } | |
616 | ||
00b3a331 | 617 | /* !CONFIG_ARCH_HAS_PTE_SPECIAL case follows: */ |
7e675137 | 618 | |
b379d790 JH |
619 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { |
620 | if (vma->vm_flags & VM_MIXEDMAP) { | |
621 | if (!pfn_valid(pfn)) | |
622 | return NULL; | |
623 | goto out; | |
624 | } else { | |
7e675137 NP |
625 | unsigned long off; |
626 | off = (addr - vma->vm_start) >> PAGE_SHIFT; | |
b379d790 JH |
627 | if (pfn == vma->vm_pgoff + off) |
628 | return NULL; | |
629 | if (!is_cow_mapping(vma->vm_flags)) | |
630 | return NULL; | |
631 | } | |
6aab341e LT |
632 | } |
633 | ||
b38af472 HD |
634 | if (is_zero_pfn(pfn)) |
635 | return NULL; | |
00b3a331 | 636 | |
22b31eec HD |
637 | check_pfn: |
638 | if (unlikely(pfn > highest_memmap_pfn)) { | |
639 | print_bad_pte(vma, addr, pte, NULL); | |
640 | return NULL; | |
641 | } | |
6aab341e LT |
642 | |
643 | /* | |
7e675137 | 644 | * NOTE! We still have PageReserved() pages in the page tables. |
7e675137 | 645 | * eg. VDSO mappings can cause them to exist. |
6aab341e | 646 | */ |
b379d790 | 647 | out: |
6aab341e | 648 | return pfn_to_page(pfn); |
ee498ed7 HD |
649 | } |
650 | ||
28093f9f GS |
651 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
652 | struct page *vm_normal_page_pmd(struct vm_area_struct *vma, unsigned long addr, | |
653 | pmd_t pmd) | |
654 | { | |
655 | unsigned long pfn = pmd_pfn(pmd); | |
656 | ||
657 | /* | |
658 | * There is no pmd_special() but there may be special pmds, e.g. | |
659 | * in a direct-access (dax) mapping, so let's just replicate the | |
00b3a331 | 660 | * !CONFIG_ARCH_HAS_PTE_SPECIAL case from vm_normal_page() here. |
28093f9f GS |
661 | */ |
662 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { | |
663 | if (vma->vm_flags & VM_MIXEDMAP) { | |
664 | if (!pfn_valid(pfn)) | |
665 | return NULL; | |
666 | goto out; | |
667 | } else { | |
668 | unsigned long off; | |
669 | off = (addr - vma->vm_start) >> PAGE_SHIFT; | |
670 | if (pfn == vma->vm_pgoff + off) | |
671 | return NULL; | |
672 | if (!is_cow_mapping(vma->vm_flags)) | |
673 | return NULL; | |
674 | } | |
675 | } | |
676 | ||
e1fb4a08 DJ |
677 | if (pmd_devmap(pmd)) |
678 | return NULL; | |
28093f9f GS |
679 | if (is_zero_pfn(pfn)) |
680 | return NULL; | |
681 | if (unlikely(pfn > highest_memmap_pfn)) | |
682 | return NULL; | |
683 | ||
684 | /* | |
685 | * NOTE! We still have PageReserved() pages in the page tables. | |
686 | * eg. VDSO mappings can cause them to exist. | |
687 | */ | |
688 | out: | |
689 | return pfn_to_page(pfn); | |
690 | } | |
691 | #endif | |
692 | ||
1da177e4 LT |
693 | /* |
694 | * copy one vm_area from one task to the other. Assumes the page tables | |
695 | * already present in the new task to be cleared in the whole range | |
696 | * covered by this vma. | |
1da177e4 LT |
697 | */ |
698 | ||
570a335b | 699 | static inline unsigned long |
1da177e4 | 700 | copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
b5810039 | 701 | pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma, |
8c103762 | 702 | unsigned long addr, int *rss) |
1da177e4 | 703 | { |
b5810039 | 704 | unsigned long vm_flags = vma->vm_flags; |
1da177e4 LT |
705 | pte_t pte = *src_pte; |
706 | struct page *page; | |
1da177e4 LT |
707 | |
708 | /* pte contains position in swap or file, so copy. */ | |
709 | if (unlikely(!pte_present(pte))) { | |
0661a336 KS |
710 | swp_entry_t entry = pte_to_swp_entry(pte); |
711 | ||
712 | if (likely(!non_swap_entry(entry))) { | |
713 | if (swap_duplicate(entry) < 0) | |
714 | return entry.val; | |
715 | ||
716 | /* make sure dst_mm is on swapoff's mmlist. */ | |
717 | if (unlikely(list_empty(&dst_mm->mmlist))) { | |
718 | spin_lock(&mmlist_lock); | |
719 | if (list_empty(&dst_mm->mmlist)) | |
720 | list_add(&dst_mm->mmlist, | |
721 | &src_mm->mmlist); | |
722 | spin_unlock(&mmlist_lock); | |
723 | } | |
724 | rss[MM_SWAPENTS]++; | |
725 | } else if (is_migration_entry(entry)) { | |
726 | page = migration_entry_to_page(entry); | |
727 | ||
eca56ff9 | 728 | rss[mm_counter(page)]++; |
0661a336 KS |
729 | |
730 | if (is_write_migration_entry(entry) && | |
731 | is_cow_mapping(vm_flags)) { | |
732 | /* | |
733 | * COW mappings require pages in both | |
734 | * parent and child to be set to read. | |
735 | */ | |
736 | make_migration_entry_read(&entry); | |
737 | pte = swp_entry_to_pte(entry); | |
738 | if (pte_swp_soft_dirty(*src_pte)) | |
739 | pte = pte_swp_mksoft_dirty(pte); | |
740 | set_pte_at(src_mm, addr, src_pte, pte); | |
0697212a | 741 | } |
5042db43 JG |
742 | } else if (is_device_private_entry(entry)) { |
743 | page = device_private_entry_to_page(entry); | |
744 | ||
745 | /* | |
746 | * Update rss count even for unaddressable pages, as | |
747 | * they should treated just like normal pages in this | |
748 | * respect. | |
749 | * | |
750 | * We will likely want to have some new rss counters | |
751 | * for unaddressable pages, at some point. But for now | |
752 | * keep things as they are. | |
753 | */ | |
754 | get_page(page); | |
755 | rss[mm_counter(page)]++; | |
756 | page_dup_rmap(page, false); | |
757 | ||
758 | /* | |
759 | * We do not preserve soft-dirty information, because so | |
760 | * far, checkpoint/restore is the only feature that | |
761 | * requires that. And checkpoint/restore does not work | |
762 | * when a device driver is involved (you cannot easily | |
763 | * save and restore device driver state). | |
764 | */ | |
765 | if (is_write_device_private_entry(entry) && | |
766 | is_cow_mapping(vm_flags)) { | |
767 | make_device_private_entry_read(&entry); | |
768 | pte = swp_entry_to_pte(entry); | |
769 | set_pte_at(src_mm, addr, src_pte, pte); | |
770 | } | |
1da177e4 | 771 | } |
ae859762 | 772 | goto out_set_pte; |
1da177e4 LT |
773 | } |
774 | ||
1da177e4 LT |
775 | /* |
776 | * If it's a COW mapping, write protect it both | |
777 | * in the parent and the child | |
778 | */ | |
1b2de5d0 | 779 | if (is_cow_mapping(vm_flags) && pte_write(pte)) { |
1da177e4 | 780 | ptep_set_wrprotect(src_mm, addr, src_pte); |
3dc90795 | 781 | pte = pte_wrprotect(pte); |
1da177e4 LT |
782 | } |
783 | ||
784 | /* | |
785 | * If it's a shared mapping, mark it clean in | |
786 | * the child | |
787 | */ | |
788 | if (vm_flags & VM_SHARED) | |
789 | pte = pte_mkclean(pte); | |
790 | pte = pte_mkold(pte); | |
6aab341e LT |
791 | |
792 | page = vm_normal_page(vma, addr, pte); | |
793 | if (page) { | |
794 | get_page(page); | |
53f9263b | 795 | page_dup_rmap(page, false); |
eca56ff9 | 796 | rss[mm_counter(page)]++; |
df6ad698 JG |
797 | } else if (pte_devmap(pte)) { |
798 | page = pte_page(pte); | |
799 | ||
800 | /* | |
801 | * Cache coherent device memory behave like regular page and | |
802 | * not like persistent memory page. For more informations see | |
803 | * MEMORY_DEVICE_CACHE_COHERENT in memory_hotplug.h | |
804 | */ | |
805 | if (is_device_public_page(page)) { | |
806 | get_page(page); | |
807 | page_dup_rmap(page, false); | |
808 | rss[mm_counter(page)]++; | |
809 | } | |
6aab341e | 810 | } |
ae859762 HD |
811 | |
812 | out_set_pte: | |
813 | set_pte_at(dst_mm, addr, dst_pte, pte); | |
570a335b | 814 | return 0; |
1da177e4 LT |
815 | } |
816 | ||
21bda264 | 817 | static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
71e3aac0 AA |
818 | pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma, |
819 | unsigned long addr, unsigned long end) | |
1da177e4 | 820 | { |
c36987e2 | 821 | pte_t *orig_src_pte, *orig_dst_pte; |
1da177e4 | 822 | pte_t *src_pte, *dst_pte; |
c74df32c | 823 | spinlock_t *src_ptl, *dst_ptl; |
e040f218 | 824 | int progress = 0; |
d559db08 | 825 | int rss[NR_MM_COUNTERS]; |
570a335b | 826 | swp_entry_t entry = (swp_entry_t){0}; |
1da177e4 LT |
827 | |
828 | again: | |
d559db08 KH |
829 | init_rss_vec(rss); |
830 | ||
c74df32c | 831 | dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); |
1da177e4 LT |
832 | if (!dst_pte) |
833 | return -ENOMEM; | |
ece0e2b6 | 834 | src_pte = pte_offset_map(src_pmd, addr); |
4c21e2f2 | 835 | src_ptl = pte_lockptr(src_mm, src_pmd); |
f20dc5f7 | 836 | spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); |
c36987e2 DN |
837 | orig_src_pte = src_pte; |
838 | orig_dst_pte = dst_pte; | |
6606c3e0 | 839 | arch_enter_lazy_mmu_mode(); |
1da177e4 | 840 | |
1da177e4 LT |
841 | do { |
842 | /* | |
843 | * We are holding two locks at this point - either of them | |
844 | * could generate latencies in another task on another CPU. | |
845 | */ | |
e040f218 HD |
846 | if (progress >= 32) { |
847 | progress = 0; | |
848 | if (need_resched() || | |
95c354fe | 849 | spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) |
e040f218 HD |
850 | break; |
851 | } | |
1da177e4 LT |
852 | if (pte_none(*src_pte)) { |
853 | progress++; | |
854 | continue; | |
855 | } | |
570a335b HD |
856 | entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, |
857 | vma, addr, rss); | |
858 | if (entry.val) | |
859 | break; | |
1da177e4 LT |
860 | progress += 8; |
861 | } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); | |
1da177e4 | 862 | |
6606c3e0 | 863 | arch_leave_lazy_mmu_mode(); |
c74df32c | 864 | spin_unlock(src_ptl); |
ece0e2b6 | 865 | pte_unmap(orig_src_pte); |
d559db08 | 866 | add_mm_rss_vec(dst_mm, rss); |
c36987e2 | 867 | pte_unmap_unlock(orig_dst_pte, dst_ptl); |
c74df32c | 868 | cond_resched(); |
570a335b HD |
869 | |
870 | if (entry.val) { | |
871 | if (add_swap_count_continuation(entry, GFP_KERNEL) < 0) | |
872 | return -ENOMEM; | |
873 | progress = 0; | |
874 | } | |
1da177e4 LT |
875 | if (addr != end) |
876 | goto again; | |
877 | return 0; | |
878 | } | |
879 | ||
880 | static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
881 | pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma, | |
882 | unsigned long addr, unsigned long end) | |
883 | { | |
884 | pmd_t *src_pmd, *dst_pmd; | |
885 | unsigned long next; | |
886 | ||
887 | dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); | |
888 | if (!dst_pmd) | |
889 | return -ENOMEM; | |
890 | src_pmd = pmd_offset(src_pud, addr); | |
891 | do { | |
892 | next = pmd_addr_end(addr, end); | |
84c3fc4e ZY |
893 | if (is_swap_pmd(*src_pmd) || pmd_trans_huge(*src_pmd) |
894 | || pmd_devmap(*src_pmd)) { | |
71e3aac0 | 895 | int err; |
a00cc7d9 | 896 | VM_BUG_ON_VMA(next-addr != HPAGE_PMD_SIZE, vma); |
71e3aac0 AA |
897 | err = copy_huge_pmd(dst_mm, src_mm, |
898 | dst_pmd, src_pmd, addr, vma); | |
899 | if (err == -ENOMEM) | |
900 | return -ENOMEM; | |
901 | if (!err) | |
902 | continue; | |
903 | /* fall through */ | |
904 | } | |
1da177e4 LT |
905 | if (pmd_none_or_clear_bad(src_pmd)) |
906 | continue; | |
907 | if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd, | |
908 | vma, addr, next)) | |
909 | return -ENOMEM; | |
910 | } while (dst_pmd++, src_pmd++, addr = next, addr != end); | |
911 | return 0; | |
912 | } | |
913 | ||
914 | static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
c2febafc | 915 | p4d_t *dst_p4d, p4d_t *src_p4d, struct vm_area_struct *vma, |
1da177e4 LT |
916 | unsigned long addr, unsigned long end) |
917 | { | |
918 | pud_t *src_pud, *dst_pud; | |
919 | unsigned long next; | |
920 | ||
c2febafc | 921 | dst_pud = pud_alloc(dst_mm, dst_p4d, addr); |
1da177e4 LT |
922 | if (!dst_pud) |
923 | return -ENOMEM; | |
c2febafc | 924 | src_pud = pud_offset(src_p4d, addr); |
1da177e4 LT |
925 | do { |
926 | next = pud_addr_end(addr, end); | |
a00cc7d9 MW |
927 | if (pud_trans_huge(*src_pud) || pud_devmap(*src_pud)) { |
928 | int err; | |
929 | ||
930 | VM_BUG_ON_VMA(next-addr != HPAGE_PUD_SIZE, vma); | |
931 | err = copy_huge_pud(dst_mm, src_mm, | |
932 | dst_pud, src_pud, addr, vma); | |
933 | if (err == -ENOMEM) | |
934 | return -ENOMEM; | |
935 | if (!err) | |
936 | continue; | |
937 | /* fall through */ | |
938 | } | |
1da177e4 LT |
939 | if (pud_none_or_clear_bad(src_pud)) |
940 | continue; | |
941 | if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud, | |
942 | vma, addr, next)) | |
943 | return -ENOMEM; | |
944 | } while (dst_pud++, src_pud++, addr = next, addr != end); | |
945 | return 0; | |
946 | } | |
947 | ||
c2febafc KS |
948 | static inline int copy_p4d_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
949 | pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma, | |
950 | unsigned long addr, unsigned long end) | |
951 | { | |
952 | p4d_t *src_p4d, *dst_p4d; | |
953 | unsigned long next; | |
954 | ||
955 | dst_p4d = p4d_alloc(dst_mm, dst_pgd, addr); | |
956 | if (!dst_p4d) | |
957 | return -ENOMEM; | |
958 | src_p4d = p4d_offset(src_pgd, addr); | |
959 | do { | |
960 | next = p4d_addr_end(addr, end); | |
961 | if (p4d_none_or_clear_bad(src_p4d)) | |
962 | continue; | |
963 | if (copy_pud_range(dst_mm, src_mm, dst_p4d, src_p4d, | |
964 | vma, addr, next)) | |
965 | return -ENOMEM; | |
966 | } while (dst_p4d++, src_p4d++, addr = next, addr != end); | |
967 | return 0; | |
968 | } | |
969 | ||
1da177e4 LT |
970 | int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
971 | struct vm_area_struct *vma) | |
972 | { | |
973 | pgd_t *src_pgd, *dst_pgd; | |
974 | unsigned long next; | |
975 | unsigned long addr = vma->vm_start; | |
976 | unsigned long end = vma->vm_end; | |
ac46d4f3 | 977 | struct mmu_notifier_range range; |
2ec74c3e | 978 | bool is_cow; |
cddb8a5c | 979 | int ret; |
1da177e4 | 980 | |
d992895b NP |
981 | /* |
982 | * Don't copy ptes where a page fault will fill them correctly. | |
983 | * Fork becomes much lighter when there are big shared or private | |
984 | * readonly mappings. The tradeoff is that copy_page_range is more | |
985 | * efficient than faulting. | |
986 | */ | |
0661a336 KS |
987 | if (!(vma->vm_flags & (VM_HUGETLB | VM_PFNMAP | VM_MIXEDMAP)) && |
988 | !vma->anon_vma) | |
989 | return 0; | |
d992895b | 990 | |
1da177e4 LT |
991 | if (is_vm_hugetlb_page(vma)) |
992 | return copy_hugetlb_page_range(dst_mm, src_mm, vma); | |
993 | ||
b3b9c293 | 994 | if (unlikely(vma->vm_flags & VM_PFNMAP)) { |
2ab64037 | 995 | /* |
996 | * We do not free on error cases below as remove_vma | |
997 | * gets called on error from higher level routine | |
998 | */ | |
5180da41 | 999 | ret = track_pfn_copy(vma); |
2ab64037 | 1000 | if (ret) |
1001 | return ret; | |
1002 | } | |
1003 | ||
cddb8a5c AA |
1004 | /* |
1005 | * We need to invalidate the secondary MMU mappings only when | |
1006 | * there could be a permission downgrade on the ptes of the | |
1007 | * parent mm. And a permission downgrade will only happen if | |
1008 | * is_cow_mapping() returns true. | |
1009 | */ | |
2ec74c3e | 1010 | is_cow = is_cow_mapping(vma->vm_flags); |
ac46d4f3 JG |
1011 | |
1012 | if (is_cow) { | |
6f4f13e8 JG |
1013 | mmu_notifier_range_init(&range, MMU_NOTIFY_UNMAP, 0, vma, |
1014 | src_mm, addr, end); | |
ac46d4f3 JG |
1015 | mmu_notifier_invalidate_range_start(&range); |
1016 | } | |
cddb8a5c AA |
1017 | |
1018 | ret = 0; | |
1da177e4 LT |
1019 | dst_pgd = pgd_offset(dst_mm, addr); |
1020 | src_pgd = pgd_offset(src_mm, addr); | |
1021 | do { | |
1022 | next = pgd_addr_end(addr, end); | |
1023 | if (pgd_none_or_clear_bad(src_pgd)) | |
1024 | continue; | |
c2febafc | 1025 | if (unlikely(copy_p4d_range(dst_mm, src_mm, dst_pgd, src_pgd, |
cddb8a5c AA |
1026 | vma, addr, next))) { |
1027 | ret = -ENOMEM; | |
1028 | break; | |
1029 | } | |
1da177e4 | 1030 | } while (dst_pgd++, src_pgd++, addr = next, addr != end); |
cddb8a5c | 1031 | |
2ec74c3e | 1032 | if (is_cow) |
ac46d4f3 | 1033 | mmu_notifier_invalidate_range_end(&range); |
cddb8a5c | 1034 | return ret; |
1da177e4 LT |
1035 | } |
1036 | ||
51c6f666 | 1037 | static unsigned long zap_pte_range(struct mmu_gather *tlb, |
b5810039 | 1038 | struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 | 1039 | unsigned long addr, unsigned long end, |
97a89413 | 1040 | struct zap_details *details) |
1da177e4 | 1041 | { |
b5810039 | 1042 | struct mm_struct *mm = tlb->mm; |
d16dfc55 | 1043 | int force_flush = 0; |
d559db08 | 1044 | int rss[NR_MM_COUNTERS]; |
97a89413 | 1045 | spinlock_t *ptl; |
5f1a1907 | 1046 | pte_t *start_pte; |
97a89413 | 1047 | pte_t *pte; |
8a5f14a2 | 1048 | swp_entry_t entry; |
d559db08 | 1049 | |
ed6a7935 | 1050 | tlb_change_page_size(tlb, PAGE_SIZE); |
d16dfc55 | 1051 | again: |
e303297e | 1052 | init_rss_vec(rss); |
5f1a1907 SR |
1053 | start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
1054 | pte = start_pte; | |
3ea27719 | 1055 | flush_tlb_batched_pending(mm); |
6606c3e0 | 1056 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
1057 | do { |
1058 | pte_t ptent = *pte; | |
166f61b9 | 1059 | if (pte_none(ptent)) |
1da177e4 | 1060 | continue; |
6f5e6b9e | 1061 | |
1da177e4 | 1062 | if (pte_present(ptent)) { |
ee498ed7 | 1063 | struct page *page; |
51c6f666 | 1064 | |
df6ad698 | 1065 | page = _vm_normal_page(vma, addr, ptent, true); |
1da177e4 LT |
1066 | if (unlikely(details) && page) { |
1067 | /* | |
1068 | * unmap_shared_mapping_pages() wants to | |
1069 | * invalidate cache without truncating: | |
1070 | * unmap shared but keep private pages. | |
1071 | */ | |
1072 | if (details->check_mapping && | |
800d8c63 | 1073 | details->check_mapping != page_rmapping(page)) |
1da177e4 | 1074 | continue; |
1da177e4 | 1075 | } |
b5810039 | 1076 | ptent = ptep_get_and_clear_full(mm, addr, pte, |
a600388d | 1077 | tlb->fullmm); |
1da177e4 LT |
1078 | tlb_remove_tlb_entry(tlb, pte, addr); |
1079 | if (unlikely(!page)) | |
1080 | continue; | |
eca56ff9 JM |
1081 | |
1082 | if (!PageAnon(page)) { | |
1cf35d47 LT |
1083 | if (pte_dirty(ptent)) { |
1084 | force_flush = 1; | |
6237bcd9 | 1085 | set_page_dirty(page); |
1cf35d47 | 1086 | } |
4917e5d0 | 1087 | if (pte_young(ptent) && |
64363aad | 1088 | likely(!(vma->vm_flags & VM_SEQ_READ))) |
bf3f3bc5 | 1089 | mark_page_accessed(page); |
6237bcd9 | 1090 | } |
eca56ff9 | 1091 | rss[mm_counter(page)]--; |
d281ee61 | 1092 | page_remove_rmap(page, false); |
3dc14741 HD |
1093 | if (unlikely(page_mapcount(page) < 0)) |
1094 | print_bad_pte(vma, addr, ptent, page); | |
e9d55e15 | 1095 | if (unlikely(__tlb_remove_page(tlb, page))) { |
1cf35d47 | 1096 | force_flush = 1; |
ce9ec37b | 1097 | addr += PAGE_SIZE; |
d16dfc55 | 1098 | break; |
1cf35d47 | 1099 | } |
1da177e4 LT |
1100 | continue; |
1101 | } | |
5042db43 JG |
1102 | |
1103 | entry = pte_to_swp_entry(ptent); | |
1104 | if (non_swap_entry(entry) && is_device_private_entry(entry)) { | |
1105 | struct page *page = device_private_entry_to_page(entry); | |
1106 | ||
1107 | if (unlikely(details && details->check_mapping)) { | |
1108 | /* | |
1109 | * unmap_shared_mapping_pages() wants to | |
1110 | * invalidate cache without truncating: | |
1111 | * unmap shared but keep private pages. | |
1112 | */ | |
1113 | if (details->check_mapping != | |
1114 | page_rmapping(page)) | |
1115 | continue; | |
1116 | } | |
1117 | ||
1118 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); | |
1119 | rss[mm_counter(page)]--; | |
1120 | page_remove_rmap(page, false); | |
1121 | put_page(page); | |
1122 | continue; | |
1123 | } | |
1124 | ||
3e8715fd KS |
1125 | /* If details->check_mapping, we leave swap entries. */ |
1126 | if (unlikely(details)) | |
1da177e4 | 1127 | continue; |
b084d435 | 1128 | |
8a5f14a2 KS |
1129 | entry = pte_to_swp_entry(ptent); |
1130 | if (!non_swap_entry(entry)) | |
1131 | rss[MM_SWAPENTS]--; | |
1132 | else if (is_migration_entry(entry)) { | |
1133 | struct page *page; | |
9f9f1acd | 1134 | |
8a5f14a2 | 1135 | page = migration_entry_to_page(entry); |
eca56ff9 | 1136 | rss[mm_counter(page)]--; |
b084d435 | 1137 | } |
8a5f14a2 KS |
1138 | if (unlikely(!free_swap_and_cache(entry))) |
1139 | print_bad_pte(vma, addr, ptent, NULL); | |
9888a1ca | 1140 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); |
97a89413 | 1141 | } while (pte++, addr += PAGE_SIZE, addr != end); |
ae859762 | 1142 | |
d559db08 | 1143 | add_mm_rss_vec(mm, rss); |
6606c3e0 | 1144 | arch_leave_lazy_mmu_mode(); |
51c6f666 | 1145 | |
1cf35d47 | 1146 | /* Do the actual TLB flush before dropping ptl */ |
fb7332a9 | 1147 | if (force_flush) |
1cf35d47 | 1148 | tlb_flush_mmu_tlbonly(tlb); |
1cf35d47 LT |
1149 | pte_unmap_unlock(start_pte, ptl); |
1150 | ||
1151 | /* | |
1152 | * If we forced a TLB flush (either due to running out of | |
1153 | * batch buffers or because we needed to flush dirty TLB | |
1154 | * entries before releasing the ptl), free the batched | |
1155 | * memory too. Restart if we didn't do everything. | |
1156 | */ | |
1157 | if (force_flush) { | |
1158 | force_flush = 0; | |
fa0aafb8 | 1159 | tlb_flush_mmu(tlb); |
2b047252 | 1160 | if (addr != end) |
d16dfc55 PZ |
1161 | goto again; |
1162 | } | |
1163 | ||
51c6f666 | 1164 | return addr; |
1da177e4 LT |
1165 | } |
1166 | ||
51c6f666 | 1167 | static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, |
b5810039 | 1168 | struct vm_area_struct *vma, pud_t *pud, |
1da177e4 | 1169 | unsigned long addr, unsigned long end, |
97a89413 | 1170 | struct zap_details *details) |
1da177e4 LT |
1171 | { |
1172 | pmd_t *pmd; | |
1173 | unsigned long next; | |
1174 | ||
1175 | pmd = pmd_offset(pud, addr); | |
1176 | do { | |
1177 | next = pmd_addr_end(addr, end); | |
84c3fc4e | 1178 | if (is_swap_pmd(*pmd) || pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) { |
53406ed1 | 1179 | if (next - addr != HPAGE_PMD_SIZE) |
fd60775a | 1180 | __split_huge_pmd(vma, pmd, addr, false, NULL); |
53406ed1 | 1181 | else if (zap_huge_pmd(tlb, vma, pmd, addr)) |
1a5a9906 | 1182 | goto next; |
71e3aac0 AA |
1183 | /* fall through */ |
1184 | } | |
1a5a9906 AA |
1185 | /* |
1186 | * Here there can be other concurrent MADV_DONTNEED or | |
1187 | * trans huge page faults running, and if the pmd is | |
1188 | * none or trans huge it can change under us. This is | |
1189 | * because MADV_DONTNEED holds the mmap_sem in read | |
1190 | * mode. | |
1191 | */ | |
1192 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) | |
1193 | goto next; | |
97a89413 | 1194 | next = zap_pte_range(tlb, vma, pmd, addr, next, details); |
1a5a9906 | 1195 | next: |
97a89413 PZ |
1196 | cond_resched(); |
1197 | } while (pmd++, addr = next, addr != end); | |
51c6f666 RH |
1198 | |
1199 | return addr; | |
1da177e4 LT |
1200 | } |
1201 | ||
51c6f666 | 1202 | static inline unsigned long zap_pud_range(struct mmu_gather *tlb, |
c2febafc | 1203 | struct vm_area_struct *vma, p4d_t *p4d, |
1da177e4 | 1204 | unsigned long addr, unsigned long end, |
97a89413 | 1205 | struct zap_details *details) |
1da177e4 LT |
1206 | { |
1207 | pud_t *pud; | |
1208 | unsigned long next; | |
1209 | ||
c2febafc | 1210 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
1211 | do { |
1212 | next = pud_addr_end(addr, end); | |
a00cc7d9 MW |
1213 | if (pud_trans_huge(*pud) || pud_devmap(*pud)) { |
1214 | if (next - addr != HPAGE_PUD_SIZE) { | |
1215 | VM_BUG_ON_VMA(!rwsem_is_locked(&tlb->mm->mmap_sem), vma); | |
1216 | split_huge_pud(vma, pud, addr); | |
1217 | } else if (zap_huge_pud(tlb, vma, pud, addr)) | |
1218 | goto next; | |
1219 | /* fall through */ | |
1220 | } | |
97a89413 | 1221 | if (pud_none_or_clear_bad(pud)) |
1da177e4 | 1222 | continue; |
97a89413 | 1223 | next = zap_pmd_range(tlb, vma, pud, addr, next, details); |
a00cc7d9 MW |
1224 | next: |
1225 | cond_resched(); | |
97a89413 | 1226 | } while (pud++, addr = next, addr != end); |
51c6f666 RH |
1227 | |
1228 | return addr; | |
1da177e4 LT |
1229 | } |
1230 | ||
c2febafc KS |
1231 | static inline unsigned long zap_p4d_range(struct mmu_gather *tlb, |
1232 | struct vm_area_struct *vma, pgd_t *pgd, | |
1233 | unsigned long addr, unsigned long end, | |
1234 | struct zap_details *details) | |
1235 | { | |
1236 | p4d_t *p4d; | |
1237 | unsigned long next; | |
1238 | ||
1239 | p4d = p4d_offset(pgd, addr); | |
1240 | do { | |
1241 | next = p4d_addr_end(addr, end); | |
1242 | if (p4d_none_or_clear_bad(p4d)) | |
1243 | continue; | |
1244 | next = zap_pud_range(tlb, vma, p4d, addr, next, details); | |
1245 | } while (p4d++, addr = next, addr != end); | |
1246 | ||
1247 | return addr; | |
1248 | } | |
1249 | ||
aac45363 | 1250 | void unmap_page_range(struct mmu_gather *tlb, |
038c7aa1 AV |
1251 | struct vm_area_struct *vma, |
1252 | unsigned long addr, unsigned long end, | |
1253 | struct zap_details *details) | |
1da177e4 LT |
1254 | { |
1255 | pgd_t *pgd; | |
1256 | unsigned long next; | |
1257 | ||
1da177e4 LT |
1258 | BUG_ON(addr >= end); |
1259 | tlb_start_vma(tlb, vma); | |
1260 | pgd = pgd_offset(vma->vm_mm, addr); | |
1261 | do { | |
1262 | next = pgd_addr_end(addr, end); | |
97a89413 | 1263 | if (pgd_none_or_clear_bad(pgd)) |
1da177e4 | 1264 | continue; |
c2febafc | 1265 | next = zap_p4d_range(tlb, vma, pgd, addr, next, details); |
97a89413 | 1266 | } while (pgd++, addr = next, addr != end); |
1da177e4 LT |
1267 | tlb_end_vma(tlb, vma); |
1268 | } | |
51c6f666 | 1269 | |
f5cc4eef AV |
1270 | |
1271 | static void unmap_single_vma(struct mmu_gather *tlb, | |
1272 | struct vm_area_struct *vma, unsigned long start_addr, | |
4f74d2c8 | 1273 | unsigned long end_addr, |
f5cc4eef AV |
1274 | struct zap_details *details) |
1275 | { | |
1276 | unsigned long start = max(vma->vm_start, start_addr); | |
1277 | unsigned long end; | |
1278 | ||
1279 | if (start >= vma->vm_end) | |
1280 | return; | |
1281 | end = min(vma->vm_end, end_addr); | |
1282 | if (end <= vma->vm_start) | |
1283 | return; | |
1284 | ||
cbc91f71 SD |
1285 | if (vma->vm_file) |
1286 | uprobe_munmap(vma, start, end); | |
1287 | ||
b3b9c293 | 1288 | if (unlikely(vma->vm_flags & VM_PFNMAP)) |
5180da41 | 1289 | untrack_pfn(vma, 0, 0); |
f5cc4eef AV |
1290 | |
1291 | if (start != end) { | |
1292 | if (unlikely(is_vm_hugetlb_page(vma))) { | |
1293 | /* | |
1294 | * It is undesirable to test vma->vm_file as it | |
1295 | * should be non-null for valid hugetlb area. | |
1296 | * However, vm_file will be NULL in the error | |
7aa6b4ad | 1297 | * cleanup path of mmap_region. When |
f5cc4eef | 1298 | * hugetlbfs ->mmap method fails, |
7aa6b4ad | 1299 | * mmap_region() nullifies vma->vm_file |
f5cc4eef AV |
1300 | * before calling this function to clean up. |
1301 | * Since no pte has actually been setup, it is | |
1302 | * safe to do nothing in this case. | |
1303 | */ | |
24669e58 | 1304 | if (vma->vm_file) { |
83cde9e8 | 1305 | i_mmap_lock_write(vma->vm_file->f_mapping); |
d833352a | 1306 | __unmap_hugepage_range_final(tlb, vma, start, end, NULL); |
83cde9e8 | 1307 | i_mmap_unlock_write(vma->vm_file->f_mapping); |
24669e58 | 1308 | } |
f5cc4eef AV |
1309 | } else |
1310 | unmap_page_range(tlb, vma, start, end, details); | |
1311 | } | |
1da177e4 LT |
1312 | } |
1313 | ||
1da177e4 LT |
1314 | /** |
1315 | * unmap_vmas - unmap a range of memory covered by a list of vma's | |
0164f69d | 1316 | * @tlb: address of the caller's struct mmu_gather |
1da177e4 LT |
1317 | * @vma: the starting vma |
1318 | * @start_addr: virtual address at which to start unmapping | |
1319 | * @end_addr: virtual address at which to end unmapping | |
1da177e4 | 1320 | * |
508034a3 | 1321 | * Unmap all pages in the vma list. |
1da177e4 | 1322 | * |
1da177e4 LT |
1323 | * Only addresses between `start' and `end' will be unmapped. |
1324 | * | |
1325 | * The VMA list must be sorted in ascending virtual address order. | |
1326 | * | |
1327 | * unmap_vmas() assumes that the caller will flush the whole unmapped address | |
1328 | * range after unmap_vmas() returns. So the only responsibility here is to | |
1329 | * ensure that any thus-far unmapped pages are flushed before unmap_vmas() | |
1330 | * drops the lock and schedules. | |
1331 | */ | |
6e8bb019 | 1332 | void unmap_vmas(struct mmu_gather *tlb, |
1da177e4 | 1333 | struct vm_area_struct *vma, unsigned long start_addr, |
4f74d2c8 | 1334 | unsigned long end_addr) |
1da177e4 | 1335 | { |
ac46d4f3 | 1336 | struct mmu_notifier_range range; |
1da177e4 | 1337 | |
6f4f13e8 JG |
1338 | mmu_notifier_range_init(&range, MMU_NOTIFY_UNMAP, 0, vma, vma->vm_mm, |
1339 | start_addr, end_addr); | |
ac46d4f3 | 1340 | mmu_notifier_invalidate_range_start(&range); |
f5cc4eef | 1341 | for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) |
4f74d2c8 | 1342 | unmap_single_vma(tlb, vma, start_addr, end_addr, NULL); |
ac46d4f3 | 1343 | mmu_notifier_invalidate_range_end(&range); |
1da177e4 LT |
1344 | } |
1345 | ||
1346 | /** | |
1347 | * zap_page_range - remove user pages in a given range | |
1348 | * @vma: vm_area_struct holding the applicable pages | |
eb4546bb | 1349 | * @start: starting address of pages to zap |
1da177e4 | 1350 | * @size: number of bytes to zap |
f5cc4eef AV |
1351 | * |
1352 | * Caller must protect the VMA list | |
1da177e4 | 1353 | */ |
7e027b14 | 1354 | void zap_page_range(struct vm_area_struct *vma, unsigned long start, |
ecf1385d | 1355 | unsigned long size) |
1da177e4 | 1356 | { |
ac46d4f3 | 1357 | struct mmu_notifier_range range; |
d16dfc55 | 1358 | struct mmu_gather tlb; |
1da177e4 | 1359 | |
1da177e4 | 1360 | lru_add_drain(); |
6f4f13e8 JG |
1361 | mmu_notifier_range_init(&range, MMU_NOTIFY_UNMAP, 0, vma, vma->vm_mm, |
1362 | start, start + size); | |
ac46d4f3 JG |
1363 | tlb_gather_mmu(&tlb, vma->vm_mm, start, range.end); |
1364 | update_hiwater_rss(vma->vm_mm); | |
1365 | mmu_notifier_invalidate_range_start(&range); | |
1366 | for ( ; vma && vma->vm_start < range.end; vma = vma->vm_next) | |
1367 | unmap_single_vma(&tlb, vma, start, range.end, NULL); | |
1368 | mmu_notifier_invalidate_range_end(&range); | |
1369 | tlb_finish_mmu(&tlb, start, range.end); | |
1da177e4 LT |
1370 | } |
1371 | ||
f5cc4eef AV |
1372 | /** |
1373 | * zap_page_range_single - remove user pages in a given range | |
1374 | * @vma: vm_area_struct holding the applicable pages | |
1375 | * @address: starting address of pages to zap | |
1376 | * @size: number of bytes to zap | |
8a5f14a2 | 1377 | * @details: details of shared cache invalidation |
f5cc4eef AV |
1378 | * |
1379 | * The range must fit into one VMA. | |
1da177e4 | 1380 | */ |
f5cc4eef | 1381 | static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address, |
1da177e4 LT |
1382 | unsigned long size, struct zap_details *details) |
1383 | { | |
ac46d4f3 | 1384 | struct mmu_notifier_range range; |
d16dfc55 | 1385 | struct mmu_gather tlb; |
1da177e4 | 1386 | |
1da177e4 | 1387 | lru_add_drain(); |
6f4f13e8 JG |
1388 | mmu_notifier_range_init(&range, MMU_NOTIFY_UNMAP, 0, vma, vma->vm_mm, |
1389 | address, address + size); | |
ac46d4f3 JG |
1390 | tlb_gather_mmu(&tlb, vma->vm_mm, address, range.end); |
1391 | update_hiwater_rss(vma->vm_mm); | |
1392 | mmu_notifier_invalidate_range_start(&range); | |
1393 | unmap_single_vma(&tlb, vma, address, range.end, details); | |
1394 | mmu_notifier_invalidate_range_end(&range); | |
1395 | tlb_finish_mmu(&tlb, address, range.end); | |
1da177e4 LT |
1396 | } |
1397 | ||
c627f9cc JS |
1398 | /** |
1399 | * zap_vma_ptes - remove ptes mapping the vma | |
1400 | * @vma: vm_area_struct holding ptes to be zapped | |
1401 | * @address: starting address of pages to zap | |
1402 | * @size: number of bytes to zap | |
1403 | * | |
1404 | * This function only unmaps ptes assigned to VM_PFNMAP vmas. | |
1405 | * | |
1406 | * The entire address range must be fully contained within the vma. | |
1407 | * | |
c627f9cc | 1408 | */ |
27d036e3 | 1409 | void zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, |
c627f9cc JS |
1410 | unsigned long size) |
1411 | { | |
1412 | if (address < vma->vm_start || address + size > vma->vm_end || | |
1413 | !(vma->vm_flags & VM_PFNMAP)) | |
27d036e3 LR |
1414 | return; |
1415 | ||
f5cc4eef | 1416 | zap_page_range_single(vma, address, size, NULL); |
c627f9cc JS |
1417 | } |
1418 | EXPORT_SYMBOL_GPL(zap_vma_ptes); | |
1419 | ||
25ca1d6c | 1420 | pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr, |
920c7a5d | 1421 | spinlock_t **ptl) |
c9cfcddf | 1422 | { |
c2febafc KS |
1423 | pgd_t *pgd; |
1424 | p4d_t *p4d; | |
1425 | pud_t *pud; | |
1426 | pmd_t *pmd; | |
1427 | ||
1428 | pgd = pgd_offset(mm, addr); | |
1429 | p4d = p4d_alloc(mm, pgd, addr); | |
1430 | if (!p4d) | |
1431 | return NULL; | |
1432 | pud = pud_alloc(mm, p4d, addr); | |
1433 | if (!pud) | |
1434 | return NULL; | |
1435 | pmd = pmd_alloc(mm, pud, addr); | |
1436 | if (!pmd) | |
1437 | return NULL; | |
1438 | ||
1439 | VM_BUG_ON(pmd_trans_huge(*pmd)); | |
1440 | return pte_alloc_map_lock(mm, pmd, addr, ptl); | |
c9cfcddf LT |
1441 | } |
1442 | ||
238f58d8 LT |
1443 | /* |
1444 | * This is the old fallback for page remapping. | |
1445 | * | |
1446 | * For historical reasons, it only allows reserved pages. Only | |
1447 | * old drivers should use this, and they needed to mark their | |
1448 | * pages reserved for the old functions anyway. | |
1449 | */ | |
423bad60 NP |
1450 | static int insert_page(struct vm_area_struct *vma, unsigned long addr, |
1451 | struct page *page, pgprot_t prot) | |
238f58d8 | 1452 | { |
423bad60 | 1453 | struct mm_struct *mm = vma->vm_mm; |
238f58d8 | 1454 | int retval; |
c9cfcddf | 1455 | pte_t *pte; |
8a9f3ccd BS |
1456 | spinlock_t *ptl; |
1457 | ||
238f58d8 | 1458 | retval = -EINVAL; |
0ee930e6 | 1459 | if (PageAnon(page) || PageSlab(page) || page_has_type(page)) |
5b4e655e | 1460 | goto out; |
238f58d8 LT |
1461 | retval = -ENOMEM; |
1462 | flush_dcache_page(page); | |
c9cfcddf | 1463 | pte = get_locked_pte(mm, addr, &ptl); |
238f58d8 | 1464 | if (!pte) |
5b4e655e | 1465 | goto out; |
238f58d8 LT |
1466 | retval = -EBUSY; |
1467 | if (!pte_none(*pte)) | |
1468 | goto out_unlock; | |
1469 | ||
1470 | /* Ok, finally just insert the thing.. */ | |
1471 | get_page(page); | |
eca56ff9 | 1472 | inc_mm_counter_fast(mm, mm_counter_file(page)); |
dd78fedd | 1473 | page_add_file_rmap(page, false); |
238f58d8 LT |
1474 | set_pte_at(mm, addr, pte, mk_pte(page, prot)); |
1475 | ||
1476 | retval = 0; | |
8a9f3ccd BS |
1477 | pte_unmap_unlock(pte, ptl); |
1478 | return retval; | |
238f58d8 LT |
1479 | out_unlock: |
1480 | pte_unmap_unlock(pte, ptl); | |
1481 | out: | |
1482 | return retval; | |
1483 | } | |
1484 | ||
bfa5bf6d REB |
1485 | /** |
1486 | * vm_insert_page - insert single page into user vma | |
1487 | * @vma: user vma to map to | |
1488 | * @addr: target user address of this page | |
1489 | * @page: source kernel page | |
1490 | * | |
a145dd41 LT |
1491 | * This allows drivers to insert individual pages they've allocated |
1492 | * into a user vma. | |
1493 | * | |
1494 | * The page has to be a nice clean _individual_ kernel allocation. | |
1495 | * If you allocate a compound page, you need to have marked it as | |
1496 | * such (__GFP_COMP), or manually just split the page up yourself | |
8dfcc9ba | 1497 | * (see split_page()). |
a145dd41 LT |
1498 | * |
1499 | * NOTE! Traditionally this was done with "remap_pfn_range()" which | |
1500 | * took an arbitrary page protection parameter. This doesn't allow | |
1501 | * that. Your vma protection will have to be set up correctly, which | |
1502 | * means that if you want a shared writable mapping, you'd better | |
1503 | * ask for a shared writable mapping! | |
1504 | * | |
1505 | * The page does not need to be reserved. | |
4b6e1e37 KK |
1506 | * |
1507 | * Usually this function is called from f_op->mmap() handler | |
1508 | * under mm->mmap_sem write-lock, so it can change vma->vm_flags. | |
1509 | * Caller must set VM_MIXEDMAP on vma if it wants to call this | |
1510 | * function from other places, for example from page-fault handler. | |
a862f68a MR |
1511 | * |
1512 | * Return: %0 on success, negative error code otherwise. | |
a145dd41 | 1513 | */ |
423bad60 NP |
1514 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, |
1515 | struct page *page) | |
a145dd41 LT |
1516 | { |
1517 | if (addr < vma->vm_start || addr >= vma->vm_end) | |
1518 | return -EFAULT; | |
1519 | if (!page_count(page)) | |
1520 | return -EINVAL; | |
4b6e1e37 KK |
1521 | if (!(vma->vm_flags & VM_MIXEDMAP)) { |
1522 | BUG_ON(down_read_trylock(&vma->vm_mm->mmap_sem)); | |
1523 | BUG_ON(vma->vm_flags & VM_PFNMAP); | |
1524 | vma->vm_flags |= VM_MIXEDMAP; | |
1525 | } | |
423bad60 | 1526 | return insert_page(vma, addr, page, vma->vm_page_prot); |
a145dd41 | 1527 | } |
e3c3374f | 1528 | EXPORT_SYMBOL(vm_insert_page); |
a145dd41 | 1529 | |
9b5a8e00 | 1530 | static vm_fault_t insert_pfn(struct vm_area_struct *vma, unsigned long addr, |
b2770da6 | 1531 | pfn_t pfn, pgprot_t prot, bool mkwrite) |
423bad60 NP |
1532 | { |
1533 | struct mm_struct *mm = vma->vm_mm; | |
423bad60 NP |
1534 | pte_t *pte, entry; |
1535 | spinlock_t *ptl; | |
1536 | ||
423bad60 NP |
1537 | pte = get_locked_pte(mm, addr, &ptl); |
1538 | if (!pte) | |
9b5a8e00 | 1539 | return VM_FAULT_OOM; |
b2770da6 RZ |
1540 | if (!pte_none(*pte)) { |
1541 | if (mkwrite) { | |
1542 | /* | |
1543 | * For read faults on private mappings the PFN passed | |
1544 | * in may not match the PFN we have mapped if the | |
1545 | * mapped PFN is a writeable COW page. In the mkwrite | |
1546 | * case we are creating a writable PTE for a shared | |
f2c57d91 JK |
1547 | * mapping and we expect the PFNs to match. If they |
1548 | * don't match, we are likely racing with block | |
1549 | * allocation and mapping invalidation so just skip the | |
1550 | * update. | |
b2770da6 | 1551 | */ |
f2c57d91 JK |
1552 | if (pte_pfn(*pte) != pfn_t_to_pfn(pfn)) { |
1553 | WARN_ON_ONCE(!is_zero_pfn(pte_pfn(*pte))); | |
b2770da6 | 1554 | goto out_unlock; |
f2c57d91 | 1555 | } |
cae85cb8 JK |
1556 | entry = pte_mkyoung(*pte); |
1557 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
1558 | if (ptep_set_access_flags(vma, addr, pte, entry, 1)) | |
1559 | update_mmu_cache(vma, addr, pte); | |
1560 | } | |
1561 | goto out_unlock; | |
b2770da6 | 1562 | } |
423bad60 NP |
1563 | |
1564 | /* Ok, finally just insert the thing.. */ | |
01c8f1c4 DW |
1565 | if (pfn_t_devmap(pfn)) |
1566 | entry = pte_mkdevmap(pfn_t_pte(pfn, prot)); | |
1567 | else | |
1568 | entry = pte_mkspecial(pfn_t_pte(pfn, prot)); | |
b2770da6 | 1569 | |
b2770da6 RZ |
1570 | if (mkwrite) { |
1571 | entry = pte_mkyoung(entry); | |
1572 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
1573 | } | |
1574 | ||
423bad60 | 1575 | set_pte_at(mm, addr, pte, entry); |
4b3073e1 | 1576 | update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */ |
423bad60 | 1577 | |
423bad60 NP |
1578 | out_unlock: |
1579 | pte_unmap_unlock(pte, ptl); | |
9b5a8e00 | 1580 | return VM_FAULT_NOPAGE; |
423bad60 NP |
1581 | } |
1582 | ||
f5e6d1d5 MW |
1583 | /** |
1584 | * vmf_insert_pfn_prot - insert single pfn into user vma with specified pgprot | |
1585 | * @vma: user vma to map to | |
1586 | * @addr: target user address of this page | |
1587 | * @pfn: source kernel pfn | |
1588 | * @pgprot: pgprot flags for the inserted page | |
1589 | * | |
1590 | * This is exactly like vmf_insert_pfn(), except that it allows drivers to | |
1591 | * to override pgprot on a per-page basis. | |
1592 | * | |
1593 | * This only makes sense for IO mappings, and it makes no sense for | |
1594 | * COW mappings. In general, using multiple vmas is preferable; | |
ae2b01f3 | 1595 | * vmf_insert_pfn_prot should only be used if using multiple VMAs is |
f5e6d1d5 MW |
1596 | * impractical. |
1597 | * | |
ae2b01f3 | 1598 | * Context: Process context. May allocate using %GFP_KERNEL. |
f5e6d1d5 MW |
1599 | * Return: vm_fault_t value. |
1600 | */ | |
1601 | vm_fault_t vmf_insert_pfn_prot(struct vm_area_struct *vma, unsigned long addr, | |
1602 | unsigned long pfn, pgprot_t pgprot) | |
1603 | { | |
6d958546 MW |
1604 | /* |
1605 | * Technically, architectures with pte_special can avoid all these | |
1606 | * restrictions (same for remap_pfn_range). However we would like | |
1607 | * consistency in testing and feature parity among all, so we should | |
1608 | * try to keep these invariants in place for everybody. | |
1609 | */ | |
1610 | BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); | |
1611 | BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == | |
1612 | (VM_PFNMAP|VM_MIXEDMAP)); | |
1613 | BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); | |
1614 | BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); | |
1615 | ||
1616 | if (addr < vma->vm_start || addr >= vma->vm_end) | |
1617 | return VM_FAULT_SIGBUS; | |
1618 | ||
1619 | if (!pfn_modify_allowed(pfn, pgprot)) | |
1620 | return VM_FAULT_SIGBUS; | |
1621 | ||
1622 | track_pfn_insert(vma, &pgprot, __pfn_to_pfn_t(pfn, PFN_DEV)); | |
1623 | ||
9b5a8e00 | 1624 | return insert_pfn(vma, addr, __pfn_to_pfn_t(pfn, PFN_DEV), pgprot, |
6d958546 | 1625 | false); |
f5e6d1d5 MW |
1626 | } |
1627 | EXPORT_SYMBOL(vmf_insert_pfn_prot); | |
e0dc0d8f | 1628 | |
ae2b01f3 MW |
1629 | /** |
1630 | * vmf_insert_pfn - insert single pfn into user vma | |
1631 | * @vma: user vma to map to | |
1632 | * @addr: target user address of this page | |
1633 | * @pfn: source kernel pfn | |
1634 | * | |
1635 | * Similar to vm_insert_page, this allows drivers to insert individual pages | |
1636 | * they've allocated into a user vma. Same comments apply. | |
1637 | * | |
1638 | * This function should only be called from a vm_ops->fault handler, and | |
1639 | * in that case the handler should return the result of this function. | |
1640 | * | |
1641 | * vma cannot be a COW mapping. | |
1642 | * | |
1643 | * As this is called only for pages that do not currently exist, we | |
1644 | * do not need to flush old virtual caches or the TLB. | |
1645 | * | |
1646 | * Context: Process context. May allocate using %GFP_KERNEL. | |
1647 | * Return: vm_fault_t value. | |
1648 | */ | |
1649 | vm_fault_t vmf_insert_pfn(struct vm_area_struct *vma, unsigned long addr, | |
1650 | unsigned long pfn) | |
1651 | { | |
1652 | return vmf_insert_pfn_prot(vma, addr, pfn, vma->vm_page_prot); | |
1653 | } | |
1654 | EXPORT_SYMBOL(vmf_insert_pfn); | |
1655 | ||
785a3fab DW |
1656 | static bool vm_mixed_ok(struct vm_area_struct *vma, pfn_t pfn) |
1657 | { | |
1658 | /* these checks mirror the abort conditions in vm_normal_page */ | |
1659 | if (vma->vm_flags & VM_MIXEDMAP) | |
1660 | return true; | |
1661 | if (pfn_t_devmap(pfn)) | |
1662 | return true; | |
1663 | if (pfn_t_special(pfn)) | |
1664 | return true; | |
1665 | if (is_zero_pfn(pfn_t_to_pfn(pfn))) | |
1666 | return true; | |
1667 | return false; | |
1668 | } | |
1669 | ||
79f3aa5b MW |
1670 | static vm_fault_t __vm_insert_mixed(struct vm_area_struct *vma, |
1671 | unsigned long addr, pfn_t pfn, bool mkwrite) | |
423bad60 | 1672 | { |
87744ab3 | 1673 | pgprot_t pgprot = vma->vm_page_prot; |
79f3aa5b | 1674 | int err; |
87744ab3 | 1675 | |
785a3fab | 1676 | BUG_ON(!vm_mixed_ok(vma, pfn)); |
e0dc0d8f | 1677 | |
423bad60 | 1678 | if (addr < vma->vm_start || addr >= vma->vm_end) |
79f3aa5b | 1679 | return VM_FAULT_SIGBUS; |
308a047c BP |
1680 | |
1681 | track_pfn_insert(vma, &pgprot, pfn); | |
e0dc0d8f | 1682 | |
42e4089c | 1683 | if (!pfn_modify_allowed(pfn_t_to_pfn(pfn), pgprot)) |
79f3aa5b | 1684 | return VM_FAULT_SIGBUS; |
42e4089c | 1685 | |
423bad60 NP |
1686 | /* |
1687 | * If we don't have pte special, then we have to use the pfn_valid() | |
1688 | * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* | |
1689 | * refcount the page if pfn_valid is true (hence insert_page rather | |
62eede62 HD |
1690 | * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP |
1691 | * without pte special, it would there be refcounted as a normal page. | |
423bad60 | 1692 | */ |
00b3a331 LD |
1693 | if (!IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL) && |
1694 | !pfn_t_devmap(pfn) && pfn_t_valid(pfn)) { | |
423bad60 NP |
1695 | struct page *page; |
1696 | ||
03fc2da6 DW |
1697 | /* |
1698 | * At this point we are committed to insert_page() | |
1699 | * regardless of whether the caller specified flags that | |
1700 | * result in pfn_t_has_page() == false. | |
1701 | */ | |
1702 | page = pfn_to_page(pfn_t_to_pfn(pfn)); | |
79f3aa5b MW |
1703 | err = insert_page(vma, addr, page, pgprot); |
1704 | } else { | |
9b5a8e00 | 1705 | return insert_pfn(vma, addr, pfn, pgprot, mkwrite); |
423bad60 | 1706 | } |
b2770da6 | 1707 | |
5d747637 MW |
1708 | if (err == -ENOMEM) |
1709 | return VM_FAULT_OOM; | |
1710 | if (err < 0 && err != -EBUSY) | |
1711 | return VM_FAULT_SIGBUS; | |
1712 | ||
1713 | return VM_FAULT_NOPAGE; | |
e0dc0d8f | 1714 | } |
79f3aa5b MW |
1715 | |
1716 | vm_fault_t vmf_insert_mixed(struct vm_area_struct *vma, unsigned long addr, | |
1717 | pfn_t pfn) | |
1718 | { | |
1719 | return __vm_insert_mixed(vma, addr, pfn, false); | |
1720 | } | |
5d747637 | 1721 | EXPORT_SYMBOL(vmf_insert_mixed); |
e0dc0d8f | 1722 | |
ab77dab4 SJ |
1723 | /* |
1724 | * If the insertion of PTE failed because someone else already added a | |
1725 | * different entry in the mean time, we treat that as success as we assume | |
1726 | * the same entry was actually inserted. | |
1727 | */ | |
ab77dab4 SJ |
1728 | vm_fault_t vmf_insert_mixed_mkwrite(struct vm_area_struct *vma, |
1729 | unsigned long addr, pfn_t pfn) | |
b2770da6 | 1730 | { |
79f3aa5b | 1731 | return __vm_insert_mixed(vma, addr, pfn, true); |
b2770da6 | 1732 | } |
ab77dab4 | 1733 | EXPORT_SYMBOL(vmf_insert_mixed_mkwrite); |
b2770da6 | 1734 | |
1da177e4 LT |
1735 | /* |
1736 | * maps a range of physical memory into the requested pages. the old | |
1737 | * mappings are removed. any references to nonexistent pages results | |
1738 | * in null mappings (currently treated as "copy-on-access") | |
1739 | */ | |
1740 | static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, | |
1741 | unsigned long addr, unsigned long end, | |
1742 | unsigned long pfn, pgprot_t prot) | |
1743 | { | |
1744 | pte_t *pte; | |
c74df32c | 1745 | spinlock_t *ptl; |
42e4089c | 1746 | int err = 0; |
1da177e4 | 1747 | |
c74df32c | 1748 | pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); |
1da177e4 LT |
1749 | if (!pte) |
1750 | return -ENOMEM; | |
6606c3e0 | 1751 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
1752 | do { |
1753 | BUG_ON(!pte_none(*pte)); | |
42e4089c AK |
1754 | if (!pfn_modify_allowed(pfn, prot)) { |
1755 | err = -EACCES; | |
1756 | break; | |
1757 | } | |
7e675137 | 1758 | set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); |
1da177e4 LT |
1759 | pfn++; |
1760 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
6606c3e0 | 1761 | arch_leave_lazy_mmu_mode(); |
c74df32c | 1762 | pte_unmap_unlock(pte - 1, ptl); |
42e4089c | 1763 | return err; |
1da177e4 LT |
1764 | } |
1765 | ||
1766 | static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, | |
1767 | unsigned long addr, unsigned long end, | |
1768 | unsigned long pfn, pgprot_t prot) | |
1769 | { | |
1770 | pmd_t *pmd; | |
1771 | unsigned long next; | |
42e4089c | 1772 | int err; |
1da177e4 LT |
1773 | |
1774 | pfn -= addr >> PAGE_SHIFT; | |
1775 | pmd = pmd_alloc(mm, pud, addr); | |
1776 | if (!pmd) | |
1777 | return -ENOMEM; | |
f66055ab | 1778 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
1da177e4 LT |
1779 | do { |
1780 | next = pmd_addr_end(addr, end); | |
42e4089c AK |
1781 | err = remap_pte_range(mm, pmd, addr, next, |
1782 | pfn + (addr >> PAGE_SHIFT), prot); | |
1783 | if (err) | |
1784 | return err; | |
1da177e4 LT |
1785 | } while (pmd++, addr = next, addr != end); |
1786 | return 0; | |
1787 | } | |
1788 | ||
c2febafc | 1789 | static inline int remap_pud_range(struct mm_struct *mm, p4d_t *p4d, |
1da177e4 LT |
1790 | unsigned long addr, unsigned long end, |
1791 | unsigned long pfn, pgprot_t prot) | |
1792 | { | |
1793 | pud_t *pud; | |
1794 | unsigned long next; | |
42e4089c | 1795 | int err; |
1da177e4 LT |
1796 | |
1797 | pfn -= addr >> PAGE_SHIFT; | |
c2febafc | 1798 | pud = pud_alloc(mm, p4d, addr); |
1da177e4 LT |
1799 | if (!pud) |
1800 | return -ENOMEM; | |
1801 | do { | |
1802 | next = pud_addr_end(addr, end); | |
42e4089c AK |
1803 | err = remap_pmd_range(mm, pud, addr, next, |
1804 | pfn + (addr >> PAGE_SHIFT), prot); | |
1805 | if (err) | |
1806 | return err; | |
1da177e4 LT |
1807 | } while (pud++, addr = next, addr != end); |
1808 | return 0; | |
1809 | } | |
1810 | ||
c2febafc KS |
1811 | static inline int remap_p4d_range(struct mm_struct *mm, pgd_t *pgd, |
1812 | unsigned long addr, unsigned long end, | |
1813 | unsigned long pfn, pgprot_t prot) | |
1814 | { | |
1815 | p4d_t *p4d; | |
1816 | unsigned long next; | |
42e4089c | 1817 | int err; |
c2febafc KS |
1818 | |
1819 | pfn -= addr >> PAGE_SHIFT; | |
1820 | p4d = p4d_alloc(mm, pgd, addr); | |
1821 | if (!p4d) | |
1822 | return -ENOMEM; | |
1823 | do { | |
1824 | next = p4d_addr_end(addr, end); | |
42e4089c AK |
1825 | err = remap_pud_range(mm, p4d, addr, next, |
1826 | pfn + (addr >> PAGE_SHIFT), prot); | |
1827 | if (err) | |
1828 | return err; | |
c2febafc KS |
1829 | } while (p4d++, addr = next, addr != end); |
1830 | return 0; | |
1831 | } | |
1832 | ||
bfa5bf6d REB |
1833 | /** |
1834 | * remap_pfn_range - remap kernel memory to userspace | |
1835 | * @vma: user vma to map to | |
1836 | * @addr: target user address to start at | |
1837 | * @pfn: physical address of kernel memory | |
1838 | * @size: size of map area | |
1839 | * @prot: page protection flags for this mapping | |
1840 | * | |
a862f68a MR |
1841 | * Note: this is only safe if the mm semaphore is held when called. |
1842 | * | |
1843 | * Return: %0 on success, negative error code otherwise. | |
bfa5bf6d | 1844 | */ |
1da177e4 LT |
1845 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, |
1846 | unsigned long pfn, unsigned long size, pgprot_t prot) | |
1847 | { | |
1848 | pgd_t *pgd; | |
1849 | unsigned long next; | |
2d15cab8 | 1850 | unsigned long end = addr + PAGE_ALIGN(size); |
1da177e4 | 1851 | struct mm_struct *mm = vma->vm_mm; |
d5957d2f | 1852 | unsigned long remap_pfn = pfn; |
1da177e4 LT |
1853 | int err; |
1854 | ||
1855 | /* | |
1856 | * Physically remapped pages are special. Tell the | |
1857 | * rest of the world about it: | |
1858 | * VM_IO tells people not to look at these pages | |
1859 | * (accesses can have side effects). | |
6aab341e LT |
1860 | * VM_PFNMAP tells the core MM that the base pages are just |
1861 | * raw PFN mappings, and do not have a "struct page" associated | |
1862 | * with them. | |
314e51b9 KK |
1863 | * VM_DONTEXPAND |
1864 | * Disable vma merging and expanding with mremap(). | |
1865 | * VM_DONTDUMP | |
1866 | * Omit vma from core dump, even when VM_IO turned off. | |
fb155c16 LT |
1867 | * |
1868 | * There's a horrible special case to handle copy-on-write | |
1869 | * behaviour that some programs depend on. We mark the "original" | |
1870 | * un-COW'ed pages by matching them up with "vma->vm_pgoff". | |
b3b9c293 | 1871 | * See vm_normal_page() for details. |
1da177e4 | 1872 | */ |
b3b9c293 KK |
1873 | if (is_cow_mapping(vma->vm_flags)) { |
1874 | if (addr != vma->vm_start || end != vma->vm_end) | |
1875 | return -EINVAL; | |
fb155c16 | 1876 | vma->vm_pgoff = pfn; |
b3b9c293 KK |
1877 | } |
1878 | ||
d5957d2f | 1879 | err = track_pfn_remap(vma, &prot, remap_pfn, addr, PAGE_ALIGN(size)); |
b3b9c293 | 1880 | if (err) |
3c8bb73a | 1881 | return -EINVAL; |
fb155c16 | 1882 | |
314e51b9 | 1883 | vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; |
1da177e4 LT |
1884 | |
1885 | BUG_ON(addr >= end); | |
1886 | pfn -= addr >> PAGE_SHIFT; | |
1887 | pgd = pgd_offset(mm, addr); | |
1888 | flush_cache_range(vma, addr, end); | |
1da177e4 LT |
1889 | do { |
1890 | next = pgd_addr_end(addr, end); | |
c2febafc | 1891 | err = remap_p4d_range(mm, pgd, addr, next, |
1da177e4 LT |
1892 | pfn + (addr >> PAGE_SHIFT), prot); |
1893 | if (err) | |
1894 | break; | |
1895 | } while (pgd++, addr = next, addr != end); | |
2ab64037 | 1896 | |
1897 | if (err) | |
d5957d2f | 1898 | untrack_pfn(vma, remap_pfn, PAGE_ALIGN(size)); |
2ab64037 | 1899 | |
1da177e4 LT |
1900 | return err; |
1901 | } | |
1902 | EXPORT_SYMBOL(remap_pfn_range); | |
1903 | ||
b4cbb197 LT |
1904 | /** |
1905 | * vm_iomap_memory - remap memory to userspace | |
1906 | * @vma: user vma to map to | |
1907 | * @start: start of area | |
1908 | * @len: size of area | |
1909 | * | |
1910 | * This is a simplified io_remap_pfn_range() for common driver use. The | |
1911 | * driver just needs to give us the physical memory range to be mapped, | |
1912 | * we'll figure out the rest from the vma information. | |
1913 | * | |
1914 | * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get | |
1915 | * whatever write-combining details or similar. | |
a862f68a MR |
1916 | * |
1917 | * Return: %0 on success, negative error code otherwise. | |
b4cbb197 LT |
1918 | */ |
1919 | int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) | |
1920 | { | |
1921 | unsigned long vm_len, pfn, pages; | |
1922 | ||
1923 | /* Check that the physical memory area passed in looks valid */ | |
1924 | if (start + len < start) | |
1925 | return -EINVAL; | |
1926 | /* | |
1927 | * You *really* shouldn't map things that aren't page-aligned, | |
1928 | * but we've historically allowed it because IO memory might | |
1929 | * just have smaller alignment. | |
1930 | */ | |
1931 | len += start & ~PAGE_MASK; | |
1932 | pfn = start >> PAGE_SHIFT; | |
1933 | pages = (len + ~PAGE_MASK) >> PAGE_SHIFT; | |
1934 | if (pfn + pages < pfn) | |
1935 | return -EINVAL; | |
1936 | ||
1937 | /* We start the mapping 'vm_pgoff' pages into the area */ | |
1938 | if (vma->vm_pgoff > pages) | |
1939 | return -EINVAL; | |
1940 | pfn += vma->vm_pgoff; | |
1941 | pages -= vma->vm_pgoff; | |
1942 | ||
1943 | /* Can we fit all of the mapping? */ | |
1944 | vm_len = vma->vm_end - vma->vm_start; | |
1945 | if (vm_len >> PAGE_SHIFT > pages) | |
1946 | return -EINVAL; | |
1947 | ||
1948 | /* Ok, let it rip */ | |
1949 | return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); | |
1950 | } | |
1951 | EXPORT_SYMBOL(vm_iomap_memory); | |
1952 | ||
aee16b3c JF |
1953 | static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, |
1954 | unsigned long addr, unsigned long end, | |
1955 | pte_fn_t fn, void *data) | |
1956 | { | |
1957 | pte_t *pte; | |
1958 | int err; | |
2f569afd | 1959 | pgtable_t token; |
94909914 | 1960 | spinlock_t *uninitialized_var(ptl); |
aee16b3c JF |
1961 | |
1962 | pte = (mm == &init_mm) ? | |
1963 | pte_alloc_kernel(pmd, addr) : | |
1964 | pte_alloc_map_lock(mm, pmd, addr, &ptl); | |
1965 | if (!pte) | |
1966 | return -ENOMEM; | |
1967 | ||
1968 | BUG_ON(pmd_huge(*pmd)); | |
1969 | ||
38e0edb1 JF |
1970 | arch_enter_lazy_mmu_mode(); |
1971 | ||
2f569afd | 1972 | token = pmd_pgtable(*pmd); |
aee16b3c JF |
1973 | |
1974 | do { | |
c36987e2 | 1975 | err = fn(pte++, token, addr, data); |
aee16b3c JF |
1976 | if (err) |
1977 | break; | |
c36987e2 | 1978 | } while (addr += PAGE_SIZE, addr != end); |
aee16b3c | 1979 | |
38e0edb1 JF |
1980 | arch_leave_lazy_mmu_mode(); |
1981 | ||
aee16b3c JF |
1982 | if (mm != &init_mm) |
1983 | pte_unmap_unlock(pte-1, ptl); | |
1984 | return err; | |
1985 | } | |
1986 | ||
1987 | static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, | |
1988 | unsigned long addr, unsigned long end, | |
1989 | pte_fn_t fn, void *data) | |
1990 | { | |
1991 | pmd_t *pmd; | |
1992 | unsigned long next; | |
1993 | int err; | |
1994 | ||
ceb86879 AK |
1995 | BUG_ON(pud_huge(*pud)); |
1996 | ||
aee16b3c JF |
1997 | pmd = pmd_alloc(mm, pud, addr); |
1998 | if (!pmd) | |
1999 | return -ENOMEM; | |
2000 | do { | |
2001 | next = pmd_addr_end(addr, end); | |
2002 | err = apply_to_pte_range(mm, pmd, addr, next, fn, data); | |
2003 | if (err) | |
2004 | break; | |
2005 | } while (pmd++, addr = next, addr != end); | |
2006 | return err; | |
2007 | } | |
2008 | ||
c2febafc | 2009 | static int apply_to_pud_range(struct mm_struct *mm, p4d_t *p4d, |
aee16b3c JF |
2010 | unsigned long addr, unsigned long end, |
2011 | pte_fn_t fn, void *data) | |
2012 | { | |
2013 | pud_t *pud; | |
2014 | unsigned long next; | |
2015 | int err; | |
2016 | ||
c2febafc | 2017 | pud = pud_alloc(mm, p4d, addr); |
aee16b3c JF |
2018 | if (!pud) |
2019 | return -ENOMEM; | |
2020 | do { | |
2021 | next = pud_addr_end(addr, end); | |
2022 | err = apply_to_pmd_range(mm, pud, addr, next, fn, data); | |
2023 | if (err) | |
2024 | break; | |
2025 | } while (pud++, addr = next, addr != end); | |
2026 | return err; | |
2027 | } | |
2028 | ||
c2febafc KS |
2029 | static int apply_to_p4d_range(struct mm_struct *mm, pgd_t *pgd, |
2030 | unsigned long addr, unsigned long end, | |
2031 | pte_fn_t fn, void *data) | |
2032 | { | |
2033 | p4d_t *p4d; | |
2034 | unsigned long next; | |
2035 | int err; | |
2036 | ||
2037 | p4d = p4d_alloc(mm, pgd, addr); | |
2038 | if (!p4d) | |
2039 | return -ENOMEM; | |
2040 | do { | |
2041 | next = p4d_addr_end(addr, end); | |
2042 | err = apply_to_pud_range(mm, p4d, addr, next, fn, data); | |
2043 | if (err) | |
2044 | break; | |
2045 | } while (p4d++, addr = next, addr != end); | |
2046 | return err; | |
2047 | } | |
2048 | ||
aee16b3c JF |
2049 | /* |
2050 | * Scan a region of virtual memory, filling in page tables as necessary | |
2051 | * and calling a provided function on each leaf page table. | |
2052 | */ | |
2053 | int apply_to_page_range(struct mm_struct *mm, unsigned long addr, | |
2054 | unsigned long size, pte_fn_t fn, void *data) | |
2055 | { | |
2056 | pgd_t *pgd; | |
2057 | unsigned long next; | |
57250a5b | 2058 | unsigned long end = addr + size; |
aee16b3c JF |
2059 | int err; |
2060 | ||
9cb65bc3 MP |
2061 | if (WARN_ON(addr >= end)) |
2062 | return -EINVAL; | |
2063 | ||
aee16b3c JF |
2064 | pgd = pgd_offset(mm, addr); |
2065 | do { | |
2066 | next = pgd_addr_end(addr, end); | |
c2febafc | 2067 | err = apply_to_p4d_range(mm, pgd, addr, next, fn, data); |
aee16b3c JF |
2068 | if (err) |
2069 | break; | |
2070 | } while (pgd++, addr = next, addr != end); | |
57250a5b | 2071 | |
aee16b3c JF |
2072 | return err; |
2073 | } | |
2074 | EXPORT_SYMBOL_GPL(apply_to_page_range); | |
2075 | ||
8f4e2101 | 2076 | /* |
9b4bdd2f KS |
2077 | * handle_pte_fault chooses page fault handler according to an entry which was |
2078 | * read non-atomically. Before making any commitment, on those architectures | |
2079 | * or configurations (e.g. i386 with PAE) which might give a mix of unmatched | |
2080 | * parts, do_swap_page must check under lock before unmapping the pte and | |
2081 | * proceeding (but do_wp_page is only called after already making such a check; | |
a335b2e1 | 2082 | * and do_anonymous_page can safely check later on). |
8f4e2101 | 2083 | */ |
4c21e2f2 | 2084 | static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd, |
8f4e2101 HD |
2085 | pte_t *page_table, pte_t orig_pte) |
2086 | { | |
2087 | int same = 1; | |
2088 | #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT) | |
2089 | if (sizeof(pte_t) > sizeof(unsigned long)) { | |
4c21e2f2 HD |
2090 | spinlock_t *ptl = pte_lockptr(mm, pmd); |
2091 | spin_lock(ptl); | |
8f4e2101 | 2092 | same = pte_same(*page_table, orig_pte); |
4c21e2f2 | 2093 | spin_unlock(ptl); |
8f4e2101 HD |
2094 | } |
2095 | #endif | |
2096 | pte_unmap(page_table); | |
2097 | return same; | |
2098 | } | |
2099 | ||
9de455b2 | 2100 | static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma) |
6aab341e | 2101 | { |
0abdd7a8 DW |
2102 | debug_dma_assert_idle(src); |
2103 | ||
6aab341e LT |
2104 | /* |
2105 | * If the source page was a PFN mapping, we don't have | |
2106 | * a "struct page" for it. We do a best-effort copy by | |
2107 | * just copying from the original user address. If that | |
2108 | * fails, we just zero-fill it. Live with it. | |
2109 | */ | |
2110 | if (unlikely(!src)) { | |
9b04c5fe | 2111 | void *kaddr = kmap_atomic(dst); |
5d2a2dbb LT |
2112 | void __user *uaddr = (void __user *)(va & PAGE_MASK); |
2113 | ||
2114 | /* | |
2115 | * This really shouldn't fail, because the page is there | |
2116 | * in the page tables. But it might just be unreadable, | |
2117 | * in which case we just give up and fill the result with | |
2118 | * zeroes. | |
2119 | */ | |
2120 | if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) | |
3ecb01df | 2121 | clear_page(kaddr); |
9b04c5fe | 2122 | kunmap_atomic(kaddr); |
c4ec7b0d | 2123 | flush_dcache_page(dst); |
0ed361de NP |
2124 | } else |
2125 | copy_user_highpage(dst, src, va, vma); | |
6aab341e LT |
2126 | } |
2127 | ||
c20cd45e MH |
2128 | static gfp_t __get_fault_gfp_mask(struct vm_area_struct *vma) |
2129 | { | |
2130 | struct file *vm_file = vma->vm_file; | |
2131 | ||
2132 | if (vm_file) | |
2133 | return mapping_gfp_mask(vm_file->f_mapping) | __GFP_FS | __GFP_IO; | |
2134 | ||
2135 | /* | |
2136 | * Special mappings (e.g. VDSO) do not have any file so fake | |
2137 | * a default GFP_KERNEL for them. | |
2138 | */ | |
2139 | return GFP_KERNEL; | |
2140 | } | |
2141 | ||
fb09a464 KS |
2142 | /* |
2143 | * Notify the address space that the page is about to become writable so that | |
2144 | * it can prohibit this or wait for the page to get into an appropriate state. | |
2145 | * | |
2146 | * We do this without the lock held, so that it can sleep if it needs to. | |
2147 | */ | |
2b740303 | 2148 | static vm_fault_t do_page_mkwrite(struct vm_fault *vmf) |
fb09a464 | 2149 | { |
2b740303 | 2150 | vm_fault_t ret; |
38b8cb7f JK |
2151 | struct page *page = vmf->page; |
2152 | unsigned int old_flags = vmf->flags; | |
fb09a464 | 2153 | |
38b8cb7f | 2154 | vmf->flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; |
fb09a464 | 2155 | |
11bac800 | 2156 | ret = vmf->vma->vm_ops->page_mkwrite(vmf); |
38b8cb7f JK |
2157 | /* Restore original flags so that caller is not surprised */ |
2158 | vmf->flags = old_flags; | |
fb09a464 KS |
2159 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) |
2160 | return ret; | |
2161 | if (unlikely(!(ret & VM_FAULT_LOCKED))) { | |
2162 | lock_page(page); | |
2163 | if (!page->mapping) { | |
2164 | unlock_page(page); | |
2165 | return 0; /* retry */ | |
2166 | } | |
2167 | ret |= VM_FAULT_LOCKED; | |
2168 | } else | |
2169 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
2170 | return ret; | |
2171 | } | |
2172 | ||
97ba0c2b JK |
2173 | /* |
2174 | * Handle dirtying of a page in shared file mapping on a write fault. | |
2175 | * | |
2176 | * The function expects the page to be locked and unlocks it. | |
2177 | */ | |
2178 | static void fault_dirty_shared_page(struct vm_area_struct *vma, | |
2179 | struct page *page) | |
2180 | { | |
2181 | struct address_space *mapping; | |
2182 | bool dirtied; | |
2183 | bool page_mkwrite = vma->vm_ops && vma->vm_ops->page_mkwrite; | |
2184 | ||
2185 | dirtied = set_page_dirty(page); | |
2186 | VM_BUG_ON_PAGE(PageAnon(page), page); | |
2187 | /* | |
2188 | * Take a local copy of the address_space - page.mapping may be zeroed | |
2189 | * by truncate after unlock_page(). The address_space itself remains | |
2190 | * pinned by vma->vm_file's reference. We rely on unlock_page()'s | |
2191 | * release semantics to prevent the compiler from undoing this copying. | |
2192 | */ | |
2193 | mapping = page_rmapping(page); | |
2194 | unlock_page(page); | |
2195 | ||
2196 | if ((dirtied || page_mkwrite) && mapping) { | |
2197 | /* | |
2198 | * Some device drivers do not set page.mapping | |
2199 | * but still dirty their pages | |
2200 | */ | |
2201 | balance_dirty_pages_ratelimited(mapping); | |
2202 | } | |
2203 | ||
2204 | if (!page_mkwrite) | |
2205 | file_update_time(vma->vm_file); | |
2206 | } | |
2207 | ||
4e047f89 SR |
2208 | /* |
2209 | * Handle write page faults for pages that can be reused in the current vma | |
2210 | * | |
2211 | * This can happen either due to the mapping being with the VM_SHARED flag, | |
2212 | * or due to us being the last reference standing to the page. In either | |
2213 | * case, all we need to do here is to mark the page as writable and update | |
2214 | * any related book-keeping. | |
2215 | */ | |
997dd98d | 2216 | static inline void wp_page_reuse(struct vm_fault *vmf) |
82b0f8c3 | 2217 | __releases(vmf->ptl) |
4e047f89 | 2218 | { |
82b0f8c3 | 2219 | struct vm_area_struct *vma = vmf->vma; |
a41b70d6 | 2220 | struct page *page = vmf->page; |
4e047f89 SR |
2221 | pte_t entry; |
2222 | /* | |
2223 | * Clear the pages cpupid information as the existing | |
2224 | * information potentially belongs to a now completely | |
2225 | * unrelated process. | |
2226 | */ | |
2227 | if (page) | |
2228 | page_cpupid_xchg_last(page, (1 << LAST_CPUPID_SHIFT) - 1); | |
2229 | ||
2994302b JK |
2230 | flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); |
2231 | entry = pte_mkyoung(vmf->orig_pte); | |
4e047f89 | 2232 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
82b0f8c3 JK |
2233 | if (ptep_set_access_flags(vma, vmf->address, vmf->pte, entry, 1)) |
2234 | update_mmu_cache(vma, vmf->address, vmf->pte); | |
2235 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
4e047f89 SR |
2236 | } |
2237 | ||
2f38ab2c SR |
2238 | /* |
2239 | * Handle the case of a page which we actually need to copy to a new page. | |
2240 | * | |
2241 | * Called with mmap_sem locked and the old page referenced, but | |
2242 | * without the ptl held. | |
2243 | * | |
2244 | * High level logic flow: | |
2245 | * | |
2246 | * - Allocate a page, copy the content of the old page to the new one. | |
2247 | * - Handle book keeping and accounting - cgroups, mmu-notifiers, etc. | |
2248 | * - Take the PTL. If the pte changed, bail out and release the allocated page | |
2249 | * - If the pte is still the way we remember it, update the page table and all | |
2250 | * relevant references. This includes dropping the reference the page-table | |
2251 | * held to the old page, as well as updating the rmap. | |
2252 | * - In any case, unlock the PTL and drop the reference we took to the old page. | |
2253 | */ | |
2b740303 | 2254 | static vm_fault_t wp_page_copy(struct vm_fault *vmf) |
2f38ab2c | 2255 | { |
82b0f8c3 | 2256 | struct vm_area_struct *vma = vmf->vma; |
bae473a4 | 2257 | struct mm_struct *mm = vma->vm_mm; |
a41b70d6 | 2258 | struct page *old_page = vmf->page; |
2f38ab2c | 2259 | struct page *new_page = NULL; |
2f38ab2c SR |
2260 | pte_t entry; |
2261 | int page_copied = 0; | |
2f38ab2c | 2262 | struct mem_cgroup *memcg; |
ac46d4f3 | 2263 | struct mmu_notifier_range range; |
2f38ab2c SR |
2264 | |
2265 | if (unlikely(anon_vma_prepare(vma))) | |
2266 | goto oom; | |
2267 | ||
2994302b | 2268 | if (is_zero_pfn(pte_pfn(vmf->orig_pte))) { |
82b0f8c3 JK |
2269 | new_page = alloc_zeroed_user_highpage_movable(vma, |
2270 | vmf->address); | |
2f38ab2c SR |
2271 | if (!new_page) |
2272 | goto oom; | |
2273 | } else { | |
bae473a4 | 2274 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, |
82b0f8c3 | 2275 | vmf->address); |
2f38ab2c SR |
2276 | if (!new_page) |
2277 | goto oom; | |
82b0f8c3 | 2278 | cow_user_page(new_page, old_page, vmf->address, vma); |
2f38ab2c | 2279 | } |
2f38ab2c | 2280 | |
2cf85583 | 2281 | if (mem_cgroup_try_charge_delay(new_page, mm, GFP_KERNEL, &memcg, false)) |
2f38ab2c SR |
2282 | goto oom_free_new; |
2283 | ||
eb3c24f3 MG |
2284 | __SetPageUptodate(new_page); |
2285 | ||
6f4f13e8 JG |
2286 | mmu_notifier_range_init(&range, MMU_NOTIFY_UNMAP, 0, vma, mm, |
2287 | vmf->address & PAGE_MASK, | |
ac46d4f3 JG |
2288 | (vmf->address & PAGE_MASK) + PAGE_SIZE); |
2289 | mmu_notifier_invalidate_range_start(&range); | |
2f38ab2c SR |
2290 | |
2291 | /* | |
2292 | * Re-check the pte - we dropped the lock | |
2293 | */ | |
82b0f8c3 | 2294 | vmf->pte = pte_offset_map_lock(mm, vmf->pmd, vmf->address, &vmf->ptl); |
2994302b | 2295 | if (likely(pte_same(*vmf->pte, vmf->orig_pte))) { |
2f38ab2c SR |
2296 | if (old_page) { |
2297 | if (!PageAnon(old_page)) { | |
eca56ff9 JM |
2298 | dec_mm_counter_fast(mm, |
2299 | mm_counter_file(old_page)); | |
2f38ab2c SR |
2300 | inc_mm_counter_fast(mm, MM_ANONPAGES); |
2301 | } | |
2302 | } else { | |
2303 | inc_mm_counter_fast(mm, MM_ANONPAGES); | |
2304 | } | |
2994302b | 2305 | flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); |
2f38ab2c SR |
2306 | entry = mk_pte(new_page, vma->vm_page_prot); |
2307 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
2308 | /* | |
2309 | * Clear the pte entry and flush it first, before updating the | |
2310 | * pte with the new entry. This will avoid a race condition | |
2311 | * seen in the presence of one thread doing SMC and another | |
2312 | * thread doing COW. | |
2313 | */ | |
82b0f8c3 JK |
2314 | ptep_clear_flush_notify(vma, vmf->address, vmf->pte); |
2315 | page_add_new_anon_rmap(new_page, vma, vmf->address, false); | |
f627c2f5 | 2316 | mem_cgroup_commit_charge(new_page, memcg, false, false); |
2f38ab2c SR |
2317 | lru_cache_add_active_or_unevictable(new_page, vma); |
2318 | /* | |
2319 | * We call the notify macro here because, when using secondary | |
2320 | * mmu page tables (such as kvm shadow page tables), we want the | |
2321 | * new page to be mapped directly into the secondary page table. | |
2322 | */ | |
82b0f8c3 JK |
2323 | set_pte_at_notify(mm, vmf->address, vmf->pte, entry); |
2324 | update_mmu_cache(vma, vmf->address, vmf->pte); | |
2f38ab2c SR |
2325 | if (old_page) { |
2326 | /* | |
2327 | * Only after switching the pte to the new page may | |
2328 | * we remove the mapcount here. Otherwise another | |
2329 | * process may come and find the rmap count decremented | |
2330 | * before the pte is switched to the new page, and | |
2331 | * "reuse" the old page writing into it while our pte | |
2332 | * here still points into it and can be read by other | |
2333 | * threads. | |
2334 | * | |
2335 | * The critical issue is to order this | |
2336 | * page_remove_rmap with the ptp_clear_flush above. | |
2337 | * Those stores are ordered by (if nothing else,) | |
2338 | * the barrier present in the atomic_add_negative | |
2339 | * in page_remove_rmap. | |
2340 | * | |
2341 | * Then the TLB flush in ptep_clear_flush ensures that | |
2342 | * no process can access the old page before the | |
2343 | * decremented mapcount is visible. And the old page | |
2344 | * cannot be reused until after the decremented | |
2345 | * mapcount is visible. So transitively, TLBs to | |
2346 | * old page will be flushed before it can be reused. | |
2347 | */ | |
d281ee61 | 2348 | page_remove_rmap(old_page, false); |
2f38ab2c SR |
2349 | } |
2350 | ||
2351 | /* Free the old page.. */ | |
2352 | new_page = old_page; | |
2353 | page_copied = 1; | |
2354 | } else { | |
f627c2f5 | 2355 | mem_cgroup_cancel_charge(new_page, memcg, false); |
2f38ab2c SR |
2356 | } |
2357 | ||
2358 | if (new_page) | |
09cbfeaf | 2359 | put_page(new_page); |
2f38ab2c | 2360 | |
82b0f8c3 | 2361 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
4645b9fe JG |
2362 | /* |
2363 | * No need to double call mmu_notifier->invalidate_range() callback as | |
2364 | * the above ptep_clear_flush_notify() did already call it. | |
2365 | */ | |
ac46d4f3 | 2366 | mmu_notifier_invalidate_range_only_end(&range); |
2f38ab2c SR |
2367 | if (old_page) { |
2368 | /* | |
2369 | * Don't let another task, with possibly unlocked vma, | |
2370 | * keep the mlocked page. | |
2371 | */ | |
2372 | if (page_copied && (vma->vm_flags & VM_LOCKED)) { | |
2373 | lock_page(old_page); /* LRU manipulation */ | |
e90309c9 KS |
2374 | if (PageMlocked(old_page)) |
2375 | munlock_vma_page(old_page); | |
2f38ab2c SR |
2376 | unlock_page(old_page); |
2377 | } | |
09cbfeaf | 2378 | put_page(old_page); |
2f38ab2c SR |
2379 | } |
2380 | return page_copied ? VM_FAULT_WRITE : 0; | |
2381 | oom_free_new: | |
09cbfeaf | 2382 | put_page(new_page); |
2f38ab2c SR |
2383 | oom: |
2384 | if (old_page) | |
09cbfeaf | 2385 | put_page(old_page); |
2f38ab2c SR |
2386 | return VM_FAULT_OOM; |
2387 | } | |
2388 | ||
66a6197c JK |
2389 | /** |
2390 | * finish_mkwrite_fault - finish page fault for a shared mapping, making PTE | |
2391 | * writeable once the page is prepared | |
2392 | * | |
2393 | * @vmf: structure describing the fault | |
2394 | * | |
2395 | * This function handles all that is needed to finish a write page fault in a | |
2396 | * shared mapping due to PTE being read-only once the mapped page is prepared. | |
a862f68a | 2397 | * It handles locking of PTE and modifying it. |
66a6197c JK |
2398 | * |
2399 | * The function expects the page to be locked or other protection against | |
2400 | * concurrent faults / writeback (such as DAX radix tree locks). | |
a862f68a MR |
2401 | * |
2402 | * Return: %VM_FAULT_WRITE on success, %0 when PTE got changed before | |
2403 | * we acquired PTE lock. | |
66a6197c | 2404 | */ |
2b740303 | 2405 | vm_fault_t finish_mkwrite_fault(struct vm_fault *vmf) |
66a6197c JK |
2406 | { |
2407 | WARN_ON_ONCE(!(vmf->vma->vm_flags & VM_SHARED)); | |
2408 | vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, vmf->address, | |
2409 | &vmf->ptl); | |
2410 | /* | |
2411 | * We might have raced with another page fault while we released the | |
2412 | * pte_offset_map_lock. | |
2413 | */ | |
2414 | if (!pte_same(*vmf->pte, vmf->orig_pte)) { | |
2415 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
a19e2553 | 2416 | return VM_FAULT_NOPAGE; |
66a6197c JK |
2417 | } |
2418 | wp_page_reuse(vmf); | |
a19e2553 | 2419 | return 0; |
66a6197c JK |
2420 | } |
2421 | ||
dd906184 BH |
2422 | /* |
2423 | * Handle write page faults for VM_MIXEDMAP or VM_PFNMAP for a VM_SHARED | |
2424 | * mapping | |
2425 | */ | |
2b740303 | 2426 | static vm_fault_t wp_pfn_shared(struct vm_fault *vmf) |
dd906184 | 2427 | { |
82b0f8c3 | 2428 | struct vm_area_struct *vma = vmf->vma; |
bae473a4 | 2429 | |
dd906184 | 2430 | if (vma->vm_ops && vma->vm_ops->pfn_mkwrite) { |
2b740303 | 2431 | vm_fault_t ret; |
dd906184 | 2432 | |
82b0f8c3 | 2433 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
fe82221f | 2434 | vmf->flags |= FAULT_FLAG_MKWRITE; |
11bac800 | 2435 | ret = vma->vm_ops->pfn_mkwrite(vmf); |
2f89dc12 | 2436 | if (ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)) |
dd906184 | 2437 | return ret; |
66a6197c | 2438 | return finish_mkwrite_fault(vmf); |
dd906184 | 2439 | } |
997dd98d JK |
2440 | wp_page_reuse(vmf); |
2441 | return VM_FAULT_WRITE; | |
dd906184 BH |
2442 | } |
2443 | ||
2b740303 | 2444 | static vm_fault_t wp_page_shared(struct vm_fault *vmf) |
82b0f8c3 | 2445 | __releases(vmf->ptl) |
93e478d4 | 2446 | { |
82b0f8c3 | 2447 | struct vm_area_struct *vma = vmf->vma; |
93e478d4 | 2448 | |
a41b70d6 | 2449 | get_page(vmf->page); |
93e478d4 | 2450 | |
93e478d4 | 2451 | if (vma->vm_ops && vma->vm_ops->page_mkwrite) { |
2b740303 | 2452 | vm_fault_t tmp; |
93e478d4 | 2453 | |
82b0f8c3 | 2454 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
38b8cb7f | 2455 | tmp = do_page_mkwrite(vmf); |
93e478d4 SR |
2456 | if (unlikely(!tmp || (tmp & |
2457 | (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { | |
a41b70d6 | 2458 | put_page(vmf->page); |
93e478d4 SR |
2459 | return tmp; |
2460 | } | |
66a6197c | 2461 | tmp = finish_mkwrite_fault(vmf); |
a19e2553 | 2462 | if (unlikely(tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { |
a41b70d6 | 2463 | unlock_page(vmf->page); |
a41b70d6 | 2464 | put_page(vmf->page); |
66a6197c | 2465 | return tmp; |
93e478d4 | 2466 | } |
66a6197c JK |
2467 | } else { |
2468 | wp_page_reuse(vmf); | |
997dd98d | 2469 | lock_page(vmf->page); |
93e478d4 | 2470 | } |
997dd98d JK |
2471 | fault_dirty_shared_page(vma, vmf->page); |
2472 | put_page(vmf->page); | |
93e478d4 | 2473 | |
997dd98d | 2474 | return VM_FAULT_WRITE; |
93e478d4 SR |
2475 | } |
2476 | ||
1da177e4 LT |
2477 | /* |
2478 | * This routine handles present pages, when users try to write | |
2479 | * to a shared page. It is done by copying the page to a new address | |
2480 | * and decrementing the shared-page counter for the old page. | |
2481 | * | |
1da177e4 LT |
2482 | * Note that this routine assumes that the protection checks have been |
2483 | * done by the caller (the low-level page fault routine in most cases). | |
2484 | * Thus we can safely just mark it writable once we've done any necessary | |
2485 | * COW. | |
2486 | * | |
2487 | * We also mark the page dirty at this point even though the page will | |
2488 | * change only once the write actually happens. This avoids a few races, | |
2489 | * and potentially makes it more efficient. | |
2490 | * | |
8f4e2101 HD |
2491 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2492 | * but allow concurrent faults), with pte both mapped and locked. | |
2493 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2494 | */ |
2b740303 | 2495 | static vm_fault_t do_wp_page(struct vm_fault *vmf) |
82b0f8c3 | 2496 | __releases(vmf->ptl) |
1da177e4 | 2497 | { |
82b0f8c3 | 2498 | struct vm_area_struct *vma = vmf->vma; |
1da177e4 | 2499 | |
a41b70d6 JK |
2500 | vmf->page = vm_normal_page(vma, vmf->address, vmf->orig_pte); |
2501 | if (!vmf->page) { | |
251b97f5 | 2502 | /* |
64e45507 PF |
2503 | * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a |
2504 | * VM_PFNMAP VMA. | |
251b97f5 PZ |
2505 | * |
2506 | * We should not cow pages in a shared writeable mapping. | |
dd906184 | 2507 | * Just mark the pages writable and/or call ops->pfn_mkwrite. |
251b97f5 PZ |
2508 | */ |
2509 | if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) == | |
2510 | (VM_WRITE|VM_SHARED)) | |
2994302b | 2511 | return wp_pfn_shared(vmf); |
2f38ab2c | 2512 | |
82b0f8c3 | 2513 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 2514 | return wp_page_copy(vmf); |
251b97f5 | 2515 | } |
1da177e4 | 2516 | |
d08b3851 | 2517 | /* |
ee6a6457 PZ |
2518 | * Take out anonymous pages first, anonymous shared vmas are |
2519 | * not dirty accountable. | |
d08b3851 | 2520 | */ |
52d1e606 | 2521 | if (PageAnon(vmf->page)) { |
ba3c4ce6 | 2522 | int total_map_swapcount; |
52d1e606 KT |
2523 | if (PageKsm(vmf->page) && (PageSwapCache(vmf->page) || |
2524 | page_count(vmf->page) != 1)) | |
2525 | goto copy; | |
a41b70d6 JK |
2526 | if (!trylock_page(vmf->page)) { |
2527 | get_page(vmf->page); | |
82b0f8c3 | 2528 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 2529 | lock_page(vmf->page); |
82b0f8c3 JK |
2530 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
2531 | vmf->address, &vmf->ptl); | |
2994302b | 2532 | if (!pte_same(*vmf->pte, vmf->orig_pte)) { |
a41b70d6 | 2533 | unlock_page(vmf->page); |
82b0f8c3 | 2534 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 2535 | put_page(vmf->page); |
28766805 | 2536 | return 0; |
ab967d86 | 2537 | } |
a41b70d6 | 2538 | put_page(vmf->page); |
ee6a6457 | 2539 | } |
52d1e606 KT |
2540 | if (PageKsm(vmf->page)) { |
2541 | bool reused = reuse_ksm_page(vmf->page, vmf->vma, | |
2542 | vmf->address); | |
2543 | unlock_page(vmf->page); | |
2544 | if (!reused) | |
2545 | goto copy; | |
2546 | wp_page_reuse(vmf); | |
2547 | return VM_FAULT_WRITE; | |
2548 | } | |
ba3c4ce6 HY |
2549 | if (reuse_swap_page(vmf->page, &total_map_swapcount)) { |
2550 | if (total_map_swapcount == 1) { | |
6d0a07ed AA |
2551 | /* |
2552 | * The page is all ours. Move it to | |
2553 | * our anon_vma so the rmap code will | |
2554 | * not search our parent or siblings. | |
2555 | * Protected against the rmap code by | |
2556 | * the page lock. | |
2557 | */ | |
a41b70d6 | 2558 | page_move_anon_rmap(vmf->page, vma); |
6d0a07ed | 2559 | } |
a41b70d6 | 2560 | unlock_page(vmf->page); |
997dd98d JK |
2561 | wp_page_reuse(vmf); |
2562 | return VM_FAULT_WRITE; | |
b009c024 | 2563 | } |
a41b70d6 | 2564 | unlock_page(vmf->page); |
ee6a6457 | 2565 | } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) == |
d08b3851 | 2566 | (VM_WRITE|VM_SHARED))) { |
a41b70d6 | 2567 | return wp_page_shared(vmf); |
1da177e4 | 2568 | } |
52d1e606 | 2569 | copy: |
1da177e4 LT |
2570 | /* |
2571 | * Ok, we need to copy. Oh, well.. | |
2572 | */ | |
a41b70d6 | 2573 | get_page(vmf->page); |
28766805 | 2574 | |
82b0f8c3 | 2575 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 2576 | return wp_page_copy(vmf); |
1da177e4 LT |
2577 | } |
2578 | ||
97a89413 | 2579 | static void unmap_mapping_range_vma(struct vm_area_struct *vma, |
1da177e4 LT |
2580 | unsigned long start_addr, unsigned long end_addr, |
2581 | struct zap_details *details) | |
2582 | { | |
f5cc4eef | 2583 | zap_page_range_single(vma, start_addr, end_addr - start_addr, details); |
1da177e4 LT |
2584 | } |
2585 | ||
f808c13f | 2586 | static inline void unmap_mapping_range_tree(struct rb_root_cached *root, |
1da177e4 LT |
2587 | struct zap_details *details) |
2588 | { | |
2589 | struct vm_area_struct *vma; | |
1da177e4 LT |
2590 | pgoff_t vba, vea, zba, zea; |
2591 | ||
6b2dbba8 | 2592 | vma_interval_tree_foreach(vma, root, |
1da177e4 | 2593 | details->first_index, details->last_index) { |
1da177e4 LT |
2594 | |
2595 | vba = vma->vm_pgoff; | |
d6e93217 | 2596 | vea = vba + vma_pages(vma) - 1; |
1da177e4 LT |
2597 | zba = details->first_index; |
2598 | if (zba < vba) | |
2599 | zba = vba; | |
2600 | zea = details->last_index; | |
2601 | if (zea > vea) | |
2602 | zea = vea; | |
2603 | ||
97a89413 | 2604 | unmap_mapping_range_vma(vma, |
1da177e4 LT |
2605 | ((zba - vba) << PAGE_SHIFT) + vma->vm_start, |
2606 | ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, | |
97a89413 | 2607 | details); |
1da177e4 LT |
2608 | } |
2609 | } | |
2610 | ||
977fbdcd MW |
2611 | /** |
2612 | * unmap_mapping_pages() - Unmap pages from processes. | |
2613 | * @mapping: The address space containing pages to be unmapped. | |
2614 | * @start: Index of first page to be unmapped. | |
2615 | * @nr: Number of pages to be unmapped. 0 to unmap to end of file. | |
2616 | * @even_cows: Whether to unmap even private COWed pages. | |
2617 | * | |
2618 | * Unmap the pages in this address space from any userspace process which | |
2619 | * has them mmaped. Generally, you want to remove COWed pages as well when | |
2620 | * a file is being truncated, but not when invalidating pages from the page | |
2621 | * cache. | |
2622 | */ | |
2623 | void unmap_mapping_pages(struct address_space *mapping, pgoff_t start, | |
2624 | pgoff_t nr, bool even_cows) | |
2625 | { | |
2626 | struct zap_details details = { }; | |
2627 | ||
2628 | details.check_mapping = even_cows ? NULL : mapping; | |
2629 | details.first_index = start; | |
2630 | details.last_index = start + nr - 1; | |
2631 | if (details.last_index < details.first_index) | |
2632 | details.last_index = ULONG_MAX; | |
2633 | ||
2634 | i_mmap_lock_write(mapping); | |
2635 | if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))) | |
2636 | unmap_mapping_range_tree(&mapping->i_mmap, &details); | |
2637 | i_mmap_unlock_write(mapping); | |
2638 | } | |
2639 | ||
1da177e4 | 2640 | /** |
8a5f14a2 | 2641 | * unmap_mapping_range - unmap the portion of all mmaps in the specified |
977fbdcd | 2642 | * address_space corresponding to the specified byte range in the underlying |
8a5f14a2 KS |
2643 | * file. |
2644 | * | |
3d41088f | 2645 | * @mapping: the address space containing mmaps to be unmapped. |
1da177e4 LT |
2646 | * @holebegin: byte in first page to unmap, relative to the start of |
2647 | * the underlying file. This will be rounded down to a PAGE_SIZE | |
25d9e2d1 | 2648 | * boundary. Note that this is different from truncate_pagecache(), which |
1da177e4 LT |
2649 | * must keep the partial page. In contrast, we must get rid of |
2650 | * partial pages. | |
2651 | * @holelen: size of prospective hole in bytes. This will be rounded | |
2652 | * up to a PAGE_SIZE boundary. A holelen of zero truncates to the | |
2653 | * end of the file. | |
2654 | * @even_cows: 1 when truncating a file, unmap even private COWed pages; | |
2655 | * but 0 when invalidating pagecache, don't throw away private data. | |
2656 | */ | |
2657 | void unmap_mapping_range(struct address_space *mapping, | |
2658 | loff_t const holebegin, loff_t const holelen, int even_cows) | |
2659 | { | |
1da177e4 LT |
2660 | pgoff_t hba = holebegin >> PAGE_SHIFT; |
2661 | pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
2662 | ||
2663 | /* Check for overflow. */ | |
2664 | if (sizeof(holelen) > sizeof(hlen)) { | |
2665 | long long holeend = | |
2666 | (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
2667 | if (holeend & ~(long long)ULONG_MAX) | |
2668 | hlen = ULONG_MAX - hba + 1; | |
2669 | } | |
2670 | ||
977fbdcd | 2671 | unmap_mapping_pages(mapping, hba, hlen, even_cows); |
1da177e4 LT |
2672 | } |
2673 | EXPORT_SYMBOL(unmap_mapping_range); | |
2674 | ||
1da177e4 | 2675 | /* |
8f4e2101 HD |
2676 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2677 | * but allow concurrent faults), and pte mapped but not yet locked. | |
9a95f3cf PC |
2678 | * We return with pte unmapped and unlocked. |
2679 | * | |
2680 | * We return with the mmap_sem locked or unlocked in the same cases | |
2681 | * as does filemap_fault(). | |
1da177e4 | 2682 | */ |
2b740303 | 2683 | vm_fault_t do_swap_page(struct vm_fault *vmf) |
1da177e4 | 2684 | { |
82b0f8c3 | 2685 | struct vm_area_struct *vma = vmf->vma; |
eaf649eb | 2686 | struct page *page = NULL, *swapcache; |
00501b53 | 2687 | struct mem_cgroup *memcg; |
65500d23 | 2688 | swp_entry_t entry; |
1da177e4 | 2689 | pte_t pte; |
d065bd81 | 2690 | int locked; |
ad8c2ee8 | 2691 | int exclusive = 0; |
2b740303 | 2692 | vm_fault_t ret = 0; |
1da177e4 | 2693 | |
eaf649eb | 2694 | if (!pte_unmap_same(vma->vm_mm, vmf->pmd, vmf->pte, vmf->orig_pte)) |
8f4e2101 | 2695 | goto out; |
65500d23 | 2696 | |
2994302b | 2697 | entry = pte_to_swp_entry(vmf->orig_pte); |
d1737fdb AK |
2698 | if (unlikely(non_swap_entry(entry))) { |
2699 | if (is_migration_entry(entry)) { | |
82b0f8c3 JK |
2700 | migration_entry_wait(vma->vm_mm, vmf->pmd, |
2701 | vmf->address); | |
5042db43 JG |
2702 | } else if (is_device_private_entry(entry)) { |
2703 | /* | |
2704 | * For un-addressable device memory we call the pgmap | |
2705 | * fault handler callback. The callback must migrate | |
2706 | * the page back to some CPU accessible page. | |
2707 | */ | |
2708 | ret = device_private_entry_fault(vma, vmf->address, entry, | |
2709 | vmf->flags, vmf->pmd); | |
d1737fdb AK |
2710 | } else if (is_hwpoison_entry(entry)) { |
2711 | ret = VM_FAULT_HWPOISON; | |
2712 | } else { | |
2994302b | 2713 | print_bad_pte(vma, vmf->address, vmf->orig_pte, NULL); |
d99be1a8 | 2714 | ret = VM_FAULT_SIGBUS; |
d1737fdb | 2715 | } |
0697212a CL |
2716 | goto out; |
2717 | } | |
0bcac06f MK |
2718 | |
2719 | ||
0ff92245 | 2720 | delayacct_set_flag(DELAYACCT_PF_SWAPIN); |
eaf649eb MK |
2721 | page = lookup_swap_cache(entry, vma, vmf->address); |
2722 | swapcache = page; | |
f8020772 | 2723 | |
1da177e4 | 2724 | if (!page) { |
0bcac06f MK |
2725 | struct swap_info_struct *si = swp_swap_info(entry); |
2726 | ||
aa8d22a1 MK |
2727 | if (si->flags & SWP_SYNCHRONOUS_IO && |
2728 | __swap_count(si, entry) == 1) { | |
0bcac06f | 2729 | /* skip swapcache */ |
e9e9b7ec MK |
2730 | page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, |
2731 | vmf->address); | |
0bcac06f MK |
2732 | if (page) { |
2733 | __SetPageLocked(page); | |
2734 | __SetPageSwapBacked(page); | |
2735 | set_page_private(page, entry.val); | |
2736 | lru_cache_add_anon(page); | |
2737 | swap_readpage(page, true); | |
2738 | } | |
aa8d22a1 | 2739 | } else { |
e9e9b7ec MK |
2740 | page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE, |
2741 | vmf); | |
aa8d22a1 | 2742 | swapcache = page; |
0bcac06f MK |
2743 | } |
2744 | ||
1da177e4 LT |
2745 | if (!page) { |
2746 | /* | |
8f4e2101 HD |
2747 | * Back out if somebody else faulted in this pte |
2748 | * while we released the pte lock. | |
1da177e4 | 2749 | */ |
82b0f8c3 JK |
2750 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
2751 | vmf->address, &vmf->ptl); | |
2994302b | 2752 | if (likely(pte_same(*vmf->pte, vmf->orig_pte))) |
1da177e4 | 2753 | ret = VM_FAULT_OOM; |
0ff92245 | 2754 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
65500d23 | 2755 | goto unlock; |
1da177e4 LT |
2756 | } |
2757 | ||
2758 | /* Had to read the page from swap area: Major fault */ | |
2759 | ret = VM_FAULT_MAJOR; | |
f8891e5e | 2760 | count_vm_event(PGMAJFAULT); |
2262185c | 2761 | count_memcg_event_mm(vma->vm_mm, PGMAJFAULT); |
d1737fdb | 2762 | } else if (PageHWPoison(page)) { |
71f72525 WF |
2763 | /* |
2764 | * hwpoisoned dirty swapcache pages are kept for killing | |
2765 | * owner processes (which may be unknown at hwpoison time) | |
2766 | */ | |
d1737fdb AK |
2767 | ret = VM_FAULT_HWPOISON; |
2768 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); | |
4779cb31 | 2769 | goto out_release; |
1da177e4 LT |
2770 | } |
2771 | ||
82b0f8c3 | 2772 | locked = lock_page_or_retry(page, vma->vm_mm, vmf->flags); |
e709ffd6 | 2773 | |
073e587e | 2774 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
d065bd81 ML |
2775 | if (!locked) { |
2776 | ret |= VM_FAULT_RETRY; | |
2777 | goto out_release; | |
2778 | } | |
073e587e | 2779 | |
4969c119 | 2780 | /* |
31c4a3d3 HD |
2781 | * Make sure try_to_free_swap or reuse_swap_page or swapoff did not |
2782 | * release the swapcache from under us. The page pin, and pte_same | |
2783 | * test below, are not enough to exclude that. Even if it is still | |
2784 | * swapcache, we need to check that the page's swap has not changed. | |
4969c119 | 2785 | */ |
0bcac06f MK |
2786 | if (unlikely((!PageSwapCache(page) || |
2787 | page_private(page) != entry.val)) && swapcache) | |
4969c119 AA |
2788 | goto out_page; |
2789 | ||
82b0f8c3 | 2790 | page = ksm_might_need_to_copy(page, vma, vmf->address); |
cbf86cfe HD |
2791 | if (unlikely(!page)) { |
2792 | ret = VM_FAULT_OOM; | |
2793 | page = swapcache; | |
cbf86cfe | 2794 | goto out_page; |
5ad64688 HD |
2795 | } |
2796 | ||
2cf85583 TH |
2797 | if (mem_cgroup_try_charge_delay(page, vma->vm_mm, GFP_KERNEL, |
2798 | &memcg, false)) { | |
8a9f3ccd | 2799 | ret = VM_FAULT_OOM; |
bc43f75c | 2800 | goto out_page; |
8a9f3ccd BS |
2801 | } |
2802 | ||
1da177e4 | 2803 | /* |
8f4e2101 | 2804 | * Back out if somebody else already faulted in this pte. |
1da177e4 | 2805 | */ |
82b0f8c3 JK |
2806 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, |
2807 | &vmf->ptl); | |
2994302b | 2808 | if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) |
b8107480 | 2809 | goto out_nomap; |
b8107480 KK |
2810 | |
2811 | if (unlikely(!PageUptodate(page))) { | |
2812 | ret = VM_FAULT_SIGBUS; | |
2813 | goto out_nomap; | |
1da177e4 LT |
2814 | } |
2815 | ||
8c7c6e34 KH |
2816 | /* |
2817 | * The page isn't present yet, go ahead with the fault. | |
2818 | * | |
2819 | * Be careful about the sequence of operations here. | |
2820 | * To get its accounting right, reuse_swap_page() must be called | |
2821 | * while the page is counted on swap but not yet in mapcount i.e. | |
2822 | * before page_add_anon_rmap() and swap_free(); try_to_free_swap() | |
2823 | * must be called after the swap_free(), or it will never succeed. | |
8c7c6e34 | 2824 | */ |
1da177e4 | 2825 | |
bae473a4 KS |
2826 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
2827 | dec_mm_counter_fast(vma->vm_mm, MM_SWAPENTS); | |
1da177e4 | 2828 | pte = mk_pte(page, vma->vm_page_prot); |
82b0f8c3 | 2829 | if ((vmf->flags & FAULT_FLAG_WRITE) && reuse_swap_page(page, NULL)) { |
1da177e4 | 2830 | pte = maybe_mkwrite(pte_mkdirty(pte), vma); |
82b0f8c3 | 2831 | vmf->flags &= ~FAULT_FLAG_WRITE; |
9a5b489b | 2832 | ret |= VM_FAULT_WRITE; |
d281ee61 | 2833 | exclusive = RMAP_EXCLUSIVE; |
1da177e4 | 2834 | } |
1da177e4 | 2835 | flush_icache_page(vma, page); |
2994302b | 2836 | if (pte_swp_soft_dirty(vmf->orig_pte)) |
179ef71c | 2837 | pte = pte_mksoft_dirty(pte); |
82b0f8c3 | 2838 | set_pte_at(vma->vm_mm, vmf->address, vmf->pte, pte); |
ca827d55 | 2839 | arch_do_swap_page(vma->vm_mm, vma, vmf->address, pte, vmf->orig_pte); |
2994302b | 2840 | vmf->orig_pte = pte; |
0bcac06f MK |
2841 | |
2842 | /* ksm created a completely new copy */ | |
2843 | if (unlikely(page != swapcache && swapcache)) { | |
82b0f8c3 | 2844 | page_add_new_anon_rmap(page, vma, vmf->address, false); |
f627c2f5 | 2845 | mem_cgroup_commit_charge(page, memcg, false, false); |
00501b53 | 2846 | lru_cache_add_active_or_unevictable(page, vma); |
0bcac06f MK |
2847 | } else { |
2848 | do_page_add_anon_rmap(page, vma, vmf->address, exclusive); | |
2849 | mem_cgroup_commit_charge(page, memcg, true, false); | |
2850 | activate_page(page); | |
00501b53 | 2851 | } |
1da177e4 | 2852 | |
c475a8ab | 2853 | swap_free(entry); |
5ccc5aba VD |
2854 | if (mem_cgroup_swap_full(page) || |
2855 | (vma->vm_flags & VM_LOCKED) || PageMlocked(page)) | |
a2c43eed | 2856 | try_to_free_swap(page); |
c475a8ab | 2857 | unlock_page(page); |
0bcac06f | 2858 | if (page != swapcache && swapcache) { |
4969c119 AA |
2859 | /* |
2860 | * Hold the lock to avoid the swap entry to be reused | |
2861 | * until we take the PT lock for the pte_same() check | |
2862 | * (to avoid false positives from pte_same). For | |
2863 | * further safety release the lock after the swap_free | |
2864 | * so that the swap count won't change under a | |
2865 | * parallel locked swapcache. | |
2866 | */ | |
2867 | unlock_page(swapcache); | |
09cbfeaf | 2868 | put_page(swapcache); |
4969c119 | 2869 | } |
c475a8ab | 2870 | |
82b0f8c3 | 2871 | if (vmf->flags & FAULT_FLAG_WRITE) { |
2994302b | 2872 | ret |= do_wp_page(vmf); |
61469f1d HD |
2873 | if (ret & VM_FAULT_ERROR) |
2874 | ret &= VM_FAULT_ERROR; | |
1da177e4 LT |
2875 | goto out; |
2876 | } | |
2877 | ||
2878 | /* No need to invalidate - it was non-present before */ | |
82b0f8c3 | 2879 | update_mmu_cache(vma, vmf->address, vmf->pte); |
65500d23 | 2880 | unlock: |
82b0f8c3 | 2881 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
1da177e4 LT |
2882 | out: |
2883 | return ret; | |
b8107480 | 2884 | out_nomap: |
f627c2f5 | 2885 | mem_cgroup_cancel_charge(page, memcg, false); |
82b0f8c3 | 2886 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
bc43f75c | 2887 | out_page: |
b8107480 | 2888 | unlock_page(page); |
4779cb31 | 2889 | out_release: |
09cbfeaf | 2890 | put_page(page); |
0bcac06f | 2891 | if (page != swapcache && swapcache) { |
4969c119 | 2892 | unlock_page(swapcache); |
09cbfeaf | 2893 | put_page(swapcache); |
4969c119 | 2894 | } |
65500d23 | 2895 | return ret; |
1da177e4 LT |
2896 | } |
2897 | ||
2898 | /* | |
8f4e2101 HD |
2899 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2900 | * but allow concurrent faults), and pte mapped but not yet locked. | |
2901 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2902 | */ |
2b740303 | 2903 | static vm_fault_t do_anonymous_page(struct vm_fault *vmf) |
1da177e4 | 2904 | { |
82b0f8c3 | 2905 | struct vm_area_struct *vma = vmf->vma; |
00501b53 | 2906 | struct mem_cgroup *memcg; |
8f4e2101 | 2907 | struct page *page; |
2b740303 | 2908 | vm_fault_t ret = 0; |
1da177e4 | 2909 | pte_t entry; |
1da177e4 | 2910 | |
6b7339f4 KS |
2911 | /* File mapping without ->vm_ops ? */ |
2912 | if (vma->vm_flags & VM_SHARED) | |
2913 | return VM_FAULT_SIGBUS; | |
2914 | ||
7267ec00 KS |
2915 | /* |
2916 | * Use pte_alloc() instead of pte_alloc_map(). We can't run | |
2917 | * pte_offset_map() on pmds where a huge pmd might be created | |
2918 | * from a different thread. | |
2919 | * | |
2920 | * pte_alloc_map() is safe to use under down_write(mmap_sem) or when | |
2921 | * parallel threads are excluded by other means. | |
2922 | * | |
2923 | * Here we only have down_read(mmap_sem). | |
2924 | */ | |
4cf58924 | 2925 | if (pte_alloc(vma->vm_mm, vmf->pmd)) |
7267ec00 KS |
2926 | return VM_FAULT_OOM; |
2927 | ||
2928 | /* See the comment in pte_alloc_one_map() */ | |
82b0f8c3 | 2929 | if (unlikely(pmd_trans_unstable(vmf->pmd))) |
7267ec00 KS |
2930 | return 0; |
2931 | ||
11ac5524 | 2932 | /* Use the zero-page for reads */ |
82b0f8c3 | 2933 | if (!(vmf->flags & FAULT_FLAG_WRITE) && |
bae473a4 | 2934 | !mm_forbids_zeropage(vma->vm_mm)) { |
82b0f8c3 | 2935 | entry = pte_mkspecial(pfn_pte(my_zero_pfn(vmf->address), |
62eede62 | 2936 | vma->vm_page_prot)); |
82b0f8c3 JK |
2937 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
2938 | vmf->address, &vmf->ptl); | |
2939 | if (!pte_none(*vmf->pte)) | |
a13ea5b7 | 2940 | goto unlock; |
6b31d595 MH |
2941 | ret = check_stable_address_space(vma->vm_mm); |
2942 | if (ret) | |
2943 | goto unlock; | |
6b251fc9 AA |
2944 | /* Deliver the page fault to userland, check inside PT lock */ |
2945 | if (userfaultfd_missing(vma)) { | |
82b0f8c3 JK |
2946 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
2947 | return handle_userfault(vmf, VM_UFFD_MISSING); | |
6b251fc9 | 2948 | } |
a13ea5b7 HD |
2949 | goto setpte; |
2950 | } | |
2951 | ||
557ed1fa | 2952 | /* Allocate our own private page. */ |
557ed1fa NP |
2953 | if (unlikely(anon_vma_prepare(vma))) |
2954 | goto oom; | |
82b0f8c3 | 2955 | page = alloc_zeroed_user_highpage_movable(vma, vmf->address); |
557ed1fa NP |
2956 | if (!page) |
2957 | goto oom; | |
eb3c24f3 | 2958 | |
2cf85583 TH |
2959 | if (mem_cgroup_try_charge_delay(page, vma->vm_mm, GFP_KERNEL, &memcg, |
2960 | false)) | |
eb3c24f3 MG |
2961 | goto oom_free_page; |
2962 | ||
52f37629 MK |
2963 | /* |
2964 | * The memory barrier inside __SetPageUptodate makes sure that | |
2965 | * preceeding stores to the page contents become visible before | |
2966 | * the set_pte_at() write. | |
2967 | */ | |
0ed361de | 2968 | __SetPageUptodate(page); |
8f4e2101 | 2969 | |
557ed1fa | 2970 | entry = mk_pte(page, vma->vm_page_prot); |
1ac0cb5d HD |
2971 | if (vma->vm_flags & VM_WRITE) |
2972 | entry = pte_mkwrite(pte_mkdirty(entry)); | |
1da177e4 | 2973 | |
82b0f8c3 JK |
2974 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, |
2975 | &vmf->ptl); | |
2976 | if (!pte_none(*vmf->pte)) | |
557ed1fa | 2977 | goto release; |
9ba69294 | 2978 | |
6b31d595 MH |
2979 | ret = check_stable_address_space(vma->vm_mm); |
2980 | if (ret) | |
2981 | goto release; | |
2982 | ||
6b251fc9 AA |
2983 | /* Deliver the page fault to userland, check inside PT lock */ |
2984 | if (userfaultfd_missing(vma)) { | |
82b0f8c3 | 2985 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
f627c2f5 | 2986 | mem_cgroup_cancel_charge(page, memcg, false); |
09cbfeaf | 2987 | put_page(page); |
82b0f8c3 | 2988 | return handle_userfault(vmf, VM_UFFD_MISSING); |
6b251fc9 AA |
2989 | } |
2990 | ||
bae473a4 | 2991 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
82b0f8c3 | 2992 | page_add_new_anon_rmap(page, vma, vmf->address, false); |
f627c2f5 | 2993 | mem_cgroup_commit_charge(page, memcg, false, false); |
00501b53 | 2994 | lru_cache_add_active_or_unevictable(page, vma); |
a13ea5b7 | 2995 | setpte: |
82b0f8c3 | 2996 | set_pte_at(vma->vm_mm, vmf->address, vmf->pte, entry); |
1da177e4 LT |
2997 | |
2998 | /* No need to invalidate - it was non-present before */ | |
82b0f8c3 | 2999 | update_mmu_cache(vma, vmf->address, vmf->pte); |
65500d23 | 3000 | unlock: |
82b0f8c3 | 3001 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
6b31d595 | 3002 | return ret; |
8f4e2101 | 3003 | release: |
f627c2f5 | 3004 | mem_cgroup_cancel_charge(page, memcg, false); |
09cbfeaf | 3005 | put_page(page); |
8f4e2101 | 3006 | goto unlock; |
8a9f3ccd | 3007 | oom_free_page: |
09cbfeaf | 3008 | put_page(page); |
65500d23 | 3009 | oom: |
1da177e4 LT |
3010 | return VM_FAULT_OOM; |
3011 | } | |
3012 | ||
9a95f3cf PC |
3013 | /* |
3014 | * The mmap_sem must have been held on entry, and may have been | |
3015 | * released depending on flags and vma->vm_ops->fault() return value. | |
3016 | * See filemap_fault() and __lock_page_retry(). | |
3017 | */ | |
2b740303 | 3018 | static vm_fault_t __do_fault(struct vm_fault *vmf) |
7eae74af | 3019 | { |
82b0f8c3 | 3020 | struct vm_area_struct *vma = vmf->vma; |
2b740303 | 3021 | vm_fault_t ret; |
7eae74af | 3022 | |
63f3655f MH |
3023 | /* |
3024 | * Preallocate pte before we take page_lock because this might lead to | |
3025 | * deadlocks for memcg reclaim which waits for pages under writeback: | |
3026 | * lock_page(A) | |
3027 | * SetPageWriteback(A) | |
3028 | * unlock_page(A) | |
3029 | * lock_page(B) | |
3030 | * lock_page(B) | |
3031 | * pte_alloc_pne | |
3032 | * shrink_page_list | |
3033 | * wait_on_page_writeback(A) | |
3034 | * SetPageWriteback(B) | |
3035 | * unlock_page(B) | |
3036 | * # flush A, B to clear the writeback | |
3037 | */ | |
3038 | if (pmd_none(*vmf->pmd) && !vmf->prealloc_pte) { | |
3039 | vmf->prealloc_pte = pte_alloc_one(vmf->vma->vm_mm); | |
3040 | if (!vmf->prealloc_pte) | |
3041 | return VM_FAULT_OOM; | |
3042 | smp_wmb(); /* See comment in __pte_alloc() */ | |
3043 | } | |
3044 | ||
11bac800 | 3045 | ret = vma->vm_ops->fault(vmf); |
3917048d | 3046 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY | |
b1aa812b | 3047 | VM_FAULT_DONE_COW))) |
bc2466e4 | 3048 | return ret; |
7eae74af | 3049 | |
667240e0 | 3050 | if (unlikely(PageHWPoison(vmf->page))) { |
7eae74af | 3051 | if (ret & VM_FAULT_LOCKED) |
667240e0 JK |
3052 | unlock_page(vmf->page); |
3053 | put_page(vmf->page); | |
936ca80d | 3054 | vmf->page = NULL; |
7eae74af KS |
3055 | return VM_FAULT_HWPOISON; |
3056 | } | |
3057 | ||
3058 | if (unlikely(!(ret & VM_FAULT_LOCKED))) | |
667240e0 | 3059 | lock_page(vmf->page); |
7eae74af | 3060 | else |
667240e0 | 3061 | VM_BUG_ON_PAGE(!PageLocked(vmf->page), vmf->page); |
7eae74af | 3062 | |
7eae74af KS |
3063 | return ret; |
3064 | } | |
3065 | ||
d0f0931d RZ |
3066 | /* |
3067 | * The ordering of these checks is important for pmds with _PAGE_DEVMAP set. | |
3068 | * If we check pmd_trans_unstable() first we will trip the bad_pmd() check | |
3069 | * inside of pmd_none_or_trans_huge_or_clear_bad(). This will end up correctly | |
3070 | * returning 1 but not before it spams dmesg with the pmd_clear_bad() output. | |
3071 | */ | |
3072 | static int pmd_devmap_trans_unstable(pmd_t *pmd) | |
3073 | { | |
3074 | return pmd_devmap(*pmd) || pmd_trans_unstable(pmd); | |
3075 | } | |
3076 | ||
2b740303 | 3077 | static vm_fault_t pte_alloc_one_map(struct vm_fault *vmf) |
7267ec00 | 3078 | { |
82b0f8c3 | 3079 | struct vm_area_struct *vma = vmf->vma; |
7267ec00 | 3080 | |
82b0f8c3 | 3081 | if (!pmd_none(*vmf->pmd)) |
7267ec00 | 3082 | goto map_pte; |
82b0f8c3 JK |
3083 | if (vmf->prealloc_pte) { |
3084 | vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd); | |
3085 | if (unlikely(!pmd_none(*vmf->pmd))) { | |
3086 | spin_unlock(vmf->ptl); | |
7267ec00 KS |
3087 | goto map_pte; |
3088 | } | |
3089 | ||
c4812909 | 3090 | mm_inc_nr_ptes(vma->vm_mm); |
82b0f8c3 JK |
3091 | pmd_populate(vma->vm_mm, vmf->pmd, vmf->prealloc_pte); |
3092 | spin_unlock(vmf->ptl); | |
7f2b6ce8 | 3093 | vmf->prealloc_pte = NULL; |
4cf58924 | 3094 | } else if (unlikely(pte_alloc(vma->vm_mm, vmf->pmd))) { |
7267ec00 KS |
3095 | return VM_FAULT_OOM; |
3096 | } | |
3097 | map_pte: | |
3098 | /* | |
3099 | * If a huge pmd materialized under us just retry later. Use | |
d0f0931d RZ |
3100 | * pmd_trans_unstable() via pmd_devmap_trans_unstable() instead of |
3101 | * pmd_trans_huge() to ensure the pmd didn't become pmd_trans_huge | |
3102 | * under us and then back to pmd_none, as a result of MADV_DONTNEED | |
3103 | * running immediately after a huge pmd fault in a different thread of | |
3104 | * this mm, in turn leading to a misleading pmd_trans_huge() retval. | |
3105 | * All we have to ensure is that it is a regular pmd that we can walk | |
3106 | * with pte_offset_map() and we can do that through an atomic read in | |
3107 | * C, which is what pmd_trans_unstable() provides. | |
7267ec00 | 3108 | */ |
d0f0931d | 3109 | if (pmd_devmap_trans_unstable(vmf->pmd)) |
7267ec00 KS |
3110 | return VM_FAULT_NOPAGE; |
3111 | ||
d0f0931d RZ |
3112 | /* |
3113 | * At this point we know that our vmf->pmd points to a page of ptes | |
3114 | * and it cannot become pmd_none(), pmd_devmap() or pmd_trans_huge() | |
3115 | * for the duration of the fault. If a racing MADV_DONTNEED runs and | |
3116 | * we zap the ptes pointed to by our vmf->pmd, the vmf->ptl will still | |
3117 | * be valid and we will re-check to make sure the vmf->pte isn't | |
3118 | * pte_none() under vmf->ptl protection when we return to | |
3119 | * alloc_set_pte(). | |
3120 | */ | |
82b0f8c3 JK |
3121 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, |
3122 | &vmf->ptl); | |
7267ec00 KS |
3123 | return 0; |
3124 | } | |
3125 | ||
e496cf3d | 3126 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
10102459 KS |
3127 | |
3128 | #define HPAGE_CACHE_INDEX_MASK (HPAGE_PMD_NR - 1) | |
3129 | static inline bool transhuge_vma_suitable(struct vm_area_struct *vma, | |
3130 | unsigned long haddr) | |
3131 | { | |
3132 | if (((vma->vm_start >> PAGE_SHIFT) & HPAGE_CACHE_INDEX_MASK) != | |
3133 | (vma->vm_pgoff & HPAGE_CACHE_INDEX_MASK)) | |
3134 | return false; | |
3135 | if (haddr < vma->vm_start || haddr + HPAGE_PMD_SIZE > vma->vm_end) | |
3136 | return false; | |
3137 | return true; | |
3138 | } | |
3139 | ||
82b0f8c3 | 3140 | static void deposit_prealloc_pte(struct vm_fault *vmf) |
953c66c2 | 3141 | { |
82b0f8c3 | 3142 | struct vm_area_struct *vma = vmf->vma; |
953c66c2 | 3143 | |
82b0f8c3 | 3144 | pgtable_trans_huge_deposit(vma->vm_mm, vmf->pmd, vmf->prealloc_pte); |
953c66c2 AK |
3145 | /* |
3146 | * We are going to consume the prealloc table, | |
3147 | * count that as nr_ptes. | |
3148 | */ | |
c4812909 | 3149 | mm_inc_nr_ptes(vma->vm_mm); |
7f2b6ce8 | 3150 | vmf->prealloc_pte = NULL; |
953c66c2 AK |
3151 | } |
3152 | ||
2b740303 | 3153 | static vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page) |
10102459 | 3154 | { |
82b0f8c3 JK |
3155 | struct vm_area_struct *vma = vmf->vma; |
3156 | bool write = vmf->flags & FAULT_FLAG_WRITE; | |
3157 | unsigned long haddr = vmf->address & HPAGE_PMD_MASK; | |
10102459 | 3158 | pmd_t entry; |
2b740303 SJ |
3159 | int i; |
3160 | vm_fault_t ret; | |
10102459 KS |
3161 | |
3162 | if (!transhuge_vma_suitable(vma, haddr)) | |
3163 | return VM_FAULT_FALLBACK; | |
3164 | ||
3165 | ret = VM_FAULT_FALLBACK; | |
3166 | page = compound_head(page); | |
3167 | ||
953c66c2 AK |
3168 | /* |
3169 | * Archs like ppc64 need additonal space to store information | |
3170 | * related to pte entry. Use the preallocated table for that. | |
3171 | */ | |
82b0f8c3 | 3172 | if (arch_needs_pgtable_deposit() && !vmf->prealloc_pte) { |
4cf58924 | 3173 | vmf->prealloc_pte = pte_alloc_one(vma->vm_mm); |
82b0f8c3 | 3174 | if (!vmf->prealloc_pte) |
953c66c2 AK |
3175 | return VM_FAULT_OOM; |
3176 | smp_wmb(); /* See comment in __pte_alloc() */ | |
3177 | } | |
3178 | ||
82b0f8c3 JK |
3179 | vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd); |
3180 | if (unlikely(!pmd_none(*vmf->pmd))) | |
10102459 KS |
3181 | goto out; |
3182 | ||
3183 | for (i = 0; i < HPAGE_PMD_NR; i++) | |
3184 | flush_icache_page(vma, page + i); | |
3185 | ||
3186 | entry = mk_huge_pmd(page, vma->vm_page_prot); | |
3187 | if (write) | |
f55e1014 | 3188 | entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma); |
10102459 | 3189 | |
fadae295 | 3190 | add_mm_counter(vma->vm_mm, mm_counter_file(page), HPAGE_PMD_NR); |
10102459 | 3191 | page_add_file_rmap(page, true); |
953c66c2 AK |
3192 | /* |
3193 | * deposit and withdraw with pmd lock held | |
3194 | */ | |
3195 | if (arch_needs_pgtable_deposit()) | |
82b0f8c3 | 3196 | deposit_prealloc_pte(vmf); |
10102459 | 3197 | |
82b0f8c3 | 3198 | set_pmd_at(vma->vm_mm, haddr, vmf->pmd, entry); |
10102459 | 3199 | |
82b0f8c3 | 3200 | update_mmu_cache_pmd(vma, haddr, vmf->pmd); |
10102459 KS |
3201 | |
3202 | /* fault is handled */ | |
3203 | ret = 0; | |
95ecedcd | 3204 | count_vm_event(THP_FILE_MAPPED); |
10102459 | 3205 | out: |
82b0f8c3 | 3206 | spin_unlock(vmf->ptl); |
10102459 KS |
3207 | return ret; |
3208 | } | |
3209 | #else | |
2b740303 | 3210 | static vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page) |
10102459 KS |
3211 | { |
3212 | BUILD_BUG(); | |
3213 | return 0; | |
3214 | } | |
3215 | #endif | |
3216 | ||
8c6e50b0 | 3217 | /** |
7267ec00 KS |
3218 | * alloc_set_pte - setup new PTE entry for given page and add reverse page |
3219 | * mapping. If needed, the fucntion allocates page table or use pre-allocated. | |
8c6e50b0 | 3220 | * |
82b0f8c3 | 3221 | * @vmf: fault environment |
7267ec00 | 3222 | * @memcg: memcg to charge page (only for private mappings) |
8c6e50b0 | 3223 | * @page: page to map |
8c6e50b0 | 3224 | * |
82b0f8c3 JK |
3225 | * Caller must take care of unlocking vmf->ptl, if vmf->pte is non-NULL on |
3226 | * return. | |
8c6e50b0 KS |
3227 | * |
3228 | * Target users are page handler itself and implementations of | |
3229 | * vm_ops->map_pages. | |
a862f68a MR |
3230 | * |
3231 | * Return: %0 on success, %VM_FAULT_ code in case of error. | |
8c6e50b0 | 3232 | */ |
2b740303 | 3233 | vm_fault_t alloc_set_pte(struct vm_fault *vmf, struct mem_cgroup *memcg, |
7267ec00 | 3234 | struct page *page) |
3bb97794 | 3235 | { |
82b0f8c3 JK |
3236 | struct vm_area_struct *vma = vmf->vma; |
3237 | bool write = vmf->flags & FAULT_FLAG_WRITE; | |
3bb97794 | 3238 | pte_t entry; |
2b740303 | 3239 | vm_fault_t ret; |
10102459 | 3240 | |
82b0f8c3 | 3241 | if (pmd_none(*vmf->pmd) && PageTransCompound(page) && |
e496cf3d | 3242 | IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) { |
10102459 KS |
3243 | /* THP on COW? */ |
3244 | VM_BUG_ON_PAGE(memcg, page); | |
3245 | ||
82b0f8c3 | 3246 | ret = do_set_pmd(vmf, page); |
10102459 | 3247 | if (ret != VM_FAULT_FALLBACK) |
b0b9b3df | 3248 | return ret; |
10102459 | 3249 | } |
3bb97794 | 3250 | |
82b0f8c3 JK |
3251 | if (!vmf->pte) { |
3252 | ret = pte_alloc_one_map(vmf); | |
7267ec00 | 3253 | if (ret) |
b0b9b3df | 3254 | return ret; |
7267ec00 KS |
3255 | } |
3256 | ||
3257 | /* Re-check under ptl */ | |
b0b9b3df HD |
3258 | if (unlikely(!pte_none(*vmf->pte))) |
3259 | return VM_FAULT_NOPAGE; | |
7267ec00 | 3260 | |
3bb97794 KS |
3261 | flush_icache_page(vma, page); |
3262 | entry = mk_pte(page, vma->vm_page_prot); | |
3263 | if (write) | |
3264 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
bae473a4 KS |
3265 | /* copy-on-write page */ |
3266 | if (write && !(vma->vm_flags & VM_SHARED)) { | |
3bb97794 | 3267 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
82b0f8c3 | 3268 | page_add_new_anon_rmap(page, vma, vmf->address, false); |
7267ec00 KS |
3269 | mem_cgroup_commit_charge(page, memcg, false, false); |
3270 | lru_cache_add_active_or_unevictable(page, vma); | |
3bb97794 | 3271 | } else { |
eca56ff9 | 3272 | inc_mm_counter_fast(vma->vm_mm, mm_counter_file(page)); |
dd78fedd | 3273 | page_add_file_rmap(page, false); |
3bb97794 | 3274 | } |
82b0f8c3 | 3275 | set_pte_at(vma->vm_mm, vmf->address, vmf->pte, entry); |
3bb97794 KS |
3276 | |
3277 | /* no need to invalidate: a not-present page won't be cached */ | |
82b0f8c3 | 3278 | update_mmu_cache(vma, vmf->address, vmf->pte); |
7267ec00 | 3279 | |
b0b9b3df | 3280 | return 0; |
3bb97794 KS |
3281 | } |
3282 | ||
9118c0cb JK |
3283 | |
3284 | /** | |
3285 | * finish_fault - finish page fault once we have prepared the page to fault | |
3286 | * | |
3287 | * @vmf: structure describing the fault | |
3288 | * | |
3289 | * This function handles all that is needed to finish a page fault once the | |
3290 | * page to fault in is prepared. It handles locking of PTEs, inserts PTE for | |
3291 | * given page, adds reverse page mapping, handles memcg charges and LRU | |
a862f68a | 3292 | * addition. |
9118c0cb JK |
3293 | * |
3294 | * The function expects the page to be locked and on success it consumes a | |
3295 | * reference of a page being mapped (for the PTE which maps it). | |
a862f68a MR |
3296 | * |
3297 | * Return: %0 on success, %VM_FAULT_ code in case of error. | |
9118c0cb | 3298 | */ |
2b740303 | 3299 | vm_fault_t finish_fault(struct vm_fault *vmf) |
9118c0cb JK |
3300 | { |
3301 | struct page *page; | |
2b740303 | 3302 | vm_fault_t ret = 0; |
9118c0cb JK |
3303 | |
3304 | /* Did we COW the page? */ | |
3305 | if ((vmf->flags & FAULT_FLAG_WRITE) && | |
3306 | !(vmf->vma->vm_flags & VM_SHARED)) | |
3307 | page = vmf->cow_page; | |
3308 | else | |
3309 | page = vmf->page; | |
6b31d595 MH |
3310 | |
3311 | /* | |
3312 | * check even for read faults because we might have lost our CoWed | |
3313 | * page | |
3314 | */ | |
3315 | if (!(vmf->vma->vm_flags & VM_SHARED)) | |
3316 | ret = check_stable_address_space(vmf->vma->vm_mm); | |
3317 | if (!ret) | |
3318 | ret = alloc_set_pte(vmf, vmf->memcg, page); | |
9118c0cb JK |
3319 | if (vmf->pte) |
3320 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
3321 | return ret; | |
3322 | } | |
3323 | ||
3a91053a KS |
3324 | static unsigned long fault_around_bytes __read_mostly = |
3325 | rounddown_pow_of_two(65536); | |
a9b0f861 | 3326 | |
a9b0f861 KS |
3327 | #ifdef CONFIG_DEBUG_FS |
3328 | static int fault_around_bytes_get(void *data, u64 *val) | |
1592eef0 | 3329 | { |
a9b0f861 | 3330 | *val = fault_around_bytes; |
1592eef0 KS |
3331 | return 0; |
3332 | } | |
3333 | ||
b4903d6e | 3334 | /* |
da391d64 WK |
3335 | * fault_around_bytes must be rounded down to the nearest page order as it's |
3336 | * what do_fault_around() expects to see. | |
b4903d6e | 3337 | */ |
a9b0f861 | 3338 | static int fault_around_bytes_set(void *data, u64 val) |
1592eef0 | 3339 | { |
a9b0f861 | 3340 | if (val / PAGE_SIZE > PTRS_PER_PTE) |
1592eef0 | 3341 | return -EINVAL; |
b4903d6e AR |
3342 | if (val > PAGE_SIZE) |
3343 | fault_around_bytes = rounddown_pow_of_two(val); | |
3344 | else | |
3345 | fault_around_bytes = PAGE_SIZE; /* rounddown_pow_of_two(0) is undefined */ | |
1592eef0 KS |
3346 | return 0; |
3347 | } | |
0a1345f8 | 3348 | DEFINE_DEBUGFS_ATTRIBUTE(fault_around_bytes_fops, |
a9b0f861 | 3349 | fault_around_bytes_get, fault_around_bytes_set, "%llu\n"); |
1592eef0 KS |
3350 | |
3351 | static int __init fault_around_debugfs(void) | |
3352 | { | |
d9f7979c GKH |
3353 | debugfs_create_file_unsafe("fault_around_bytes", 0644, NULL, NULL, |
3354 | &fault_around_bytes_fops); | |
1592eef0 KS |
3355 | return 0; |
3356 | } | |
3357 | late_initcall(fault_around_debugfs); | |
1592eef0 | 3358 | #endif |
8c6e50b0 | 3359 | |
1fdb412b KS |
3360 | /* |
3361 | * do_fault_around() tries to map few pages around the fault address. The hope | |
3362 | * is that the pages will be needed soon and this will lower the number of | |
3363 | * faults to handle. | |
3364 | * | |
3365 | * It uses vm_ops->map_pages() to map the pages, which skips the page if it's | |
3366 | * not ready to be mapped: not up-to-date, locked, etc. | |
3367 | * | |
3368 | * This function is called with the page table lock taken. In the split ptlock | |
3369 | * case the page table lock only protects only those entries which belong to | |
3370 | * the page table corresponding to the fault address. | |
3371 | * | |
3372 | * This function doesn't cross the VMA boundaries, in order to call map_pages() | |
3373 | * only once. | |
3374 | * | |
da391d64 WK |
3375 | * fault_around_bytes defines how many bytes we'll try to map. |
3376 | * do_fault_around() expects it to be set to a power of two less than or equal | |
3377 | * to PTRS_PER_PTE. | |
1fdb412b | 3378 | * |
da391d64 WK |
3379 | * The virtual address of the area that we map is naturally aligned to |
3380 | * fault_around_bytes rounded down to the machine page size | |
3381 | * (and therefore to page order). This way it's easier to guarantee | |
3382 | * that we don't cross page table boundaries. | |
1fdb412b | 3383 | */ |
2b740303 | 3384 | static vm_fault_t do_fault_around(struct vm_fault *vmf) |
8c6e50b0 | 3385 | { |
82b0f8c3 | 3386 | unsigned long address = vmf->address, nr_pages, mask; |
0721ec8b | 3387 | pgoff_t start_pgoff = vmf->pgoff; |
bae473a4 | 3388 | pgoff_t end_pgoff; |
2b740303 SJ |
3389 | int off; |
3390 | vm_fault_t ret = 0; | |
8c6e50b0 | 3391 | |
4db0c3c2 | 3392 | nr_pages = READ_ONCE(fault_around_bytes) >> PAGE_SHIFT; |
aecd6f44 KS |
3393 | mask = ~(nr_pages * PAGE_SIZE - 1) & PAGE_MASK; |
3394 | ||
82b0f8c3 JK |
3395 | vmf->address = max(address & mask, vmf->vma->vm_start); |
3396 | off = ((address - vmf->address) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1); | |
bae473a4 | 3397 | start_pgoff -= off; |
8c6e50b0 KS |
3398 | |
3399 | /* | |
da391d64 WK |
3400 | * end_pgoff is either the end of the page table, the end of |
3401 | * the vma or nr_pages from start_pgoff, depending what is nearest. | |
8c6e50b0 | 3402 | */ |
bae473a4 | 3403 | end_pgoff = start_pgoff - |
82b0f8c3 | 3404 | ((vmf->address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)) + |
8c6e50b0 | 3405 | PTRS_PER_PTE - 1; |
82b0f8c3 | 3406 | end_pgoff = min3(end_pgoff, vma_pages(vmf->vma) + vmf->vma->vm_pgoff - 1, |
bae473a4 | 3407 | start_pgoff + nr_pages - 1); |
8c6e50b0 | 3408 | |
82b0f8c3 | 3409 | if (pmd_none(*vmf->pmd)) { |
4cf58924 | 3410 | vmf->prealloc_pte = pte_alloc_one(vmf->vma->vm_mm); |
82b0f8c3 | 3411 | if (!vmf->prealloc_pte) |
c5f88bd2 | 3412 | goto out; |
7267ec00 | 3413 | smp_wmb(); /* See comment in __pte_alloc() */ |
8c6e50b0 KS |
3414 | } |
3415 | ||
82b0f8c3 | 3416 | vmf->vma->vm_ops->map_pages(vmf, start_pgoff, end_pgoff); |
7267ec00 | 3417 | |
7267ec00 | 3418 | /* Huge page is mapped? Page fault is solved */ |
82b0f8c3 | 3419 | if (pmd_trans_huge(*vmf->pmd)) { |
7267ec00 KS |
3420 | ret = VM_FAULT_NOPAGE; |
3421 | goto out; | |
3422 | } | |
3423 | ||
3424 | /* ->map_pages() haven't done anything useful. Cold page cache? */ | |
82b0f8c3 | 3425 | if (!vmf->pte) |
7267ec00 KS |
3426 | goto out; |
3427 | ||
3428 | /* check if the page fault is solved */ | |
82b0f8c3 JK |
3429 | vmf->pte -= (vmf->address >> PAGE_SHIFT) - (address >> PAGE_SHIFT); |
3430 | if (!pte_none(*vmf->pte)) | |
7267ec00 | 3431 | ret = VM_FAULT_NOPAGE; |
82b0f8c3 | 3432 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
bae473a4 | 3433 | out: |
82b0f8c3 JK |
3434 | vmf->address = address; |
3435 | vmf->pte = NULL; | |
7267ec00 | 3436 | return ret; |
8c6e50b0 KS |
3437 | } |
3438 | ||
2b740303 | 3439 | static vm_fault_t do_read_fault(struct vm_fault *vmf) |
e655fb29 | 3440 | { |
82b0f8c3 | 3441 | struct vm_area_struct *vma = vmf->vma; |
2b740303 | 3442 | vm_fault_t ret = 0; |
8c6e50b0 KS |
3443 | |
3444 | /* | |
3445 | * Let's call ->map_pages() first and use ->fault() as fallback | |
3446 | * if page by the offset is not ready to be mapped (cold cache or | |
3447 | * something). | |
3448 | */ | |
9b4bdd2f | 3449 | if (vma->vm_ops->map_pages && fault_around_bytes >> PAGE_SHIFT > 1) { |
0721ec8b | 3450 | ret = do_fault_around(vmf); |
7267ec00 KS |
3451 | if (ret) |
3452 | return ret; | |
8c6e50b0 | 3453 | } |
e655fb29 | 3454 | |
936ca80d | 3455 | ret = __do_fault(vmf); |
e655fb29 KS |
3456 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
3457 | return ret; | |
3458 | ||
9118c0cb | 3459 | ret |= finish_fault(vmf); |
936ca80d | 3460 | unlock_page(vmf->page); |
7267ec00 | 3461 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
936ca80d | 3462 | put_page(vmf->page); |
e655fb29 KS |
3463 | return ret; |
3464 | } | |
3465 | ||
2b740303 | 3466 | static vm_fault_t do_cow_fault(struct vm_fault *vmf) |
ec47c3b9 | 3467 | { |
82b0f8c3 | 3468 | struct vm_area_struct *vma = vmf->vma; |
2b740303 | 3469 | vm_fault_t ret; |
ec47c3b9 KS |
3470 | |
3471 | if (unlikely(anon_vma_prepare(vma))) | |
3472 | return VM_FAULT_OOM; | |
3473 | ||
936ca80d JK |
3474 | vmf->cow_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vmf->address); |
3475 | if (!vmf->cow_page) | |
ec47c3b9 KS |
3476 | return VM_FAULT_OOM; |
3477 | ||
2cf85583 | 3478 | if (mem_cgroup_try_charge_delay(vmf->cow_page, vma->vm_mm, GFP_KERNEL, |
3917048d | 3479 | &vmf->memcg, false)) { |
936ca80d | 3480 | put_page(vmf->cow_page); |
ec47c3b9 KS |
3481 | return VM_FAULT_OOM; |
3482 | } | |
3483 | ||
936ca80d | 3484 | ret = __do_fault(vmf); |
ec47c3b9 KS |
3485 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
3486 | goto uncharge_out; | |
3917048d JK |
3487 | if (ret & VM_FAULT_DONE_COW) |
3488 | return ret; | |
ec47c3b9 | 3489 | |
b1aa812b | 3490 | copy_user_highpage(vmf->cow_page, vmf->page, vmf->address, vma); |
936ca80d | 3491 | __SetPageUptodate(vmf->cow_page); |
ec47c3b9 | 3492 | |
9118c0cb | 3493 | ret |= finish_fault(vmf); |
b1aa812b JK |
3494 | unlock_page(vmf->page); |
3495 | put_page(vmf->page); | |
7267ec00 KS |
3496 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
3497 | goto uncharge_out; | |
ec47c3b9 KS |
3498 | return ret; |
3499 | uncharge_out: | |
3917048d | 3500 | mem_cgroup_cancel_charge(vmf->cow_page, vmf->memcg, false); |
936ca80d | 3501 | put_page(vmf->cow_page); |
ec47c3b9 KS |
3502 | return ret; |
3503 | } | |
3504 | ||
2b740303 | 3505 | static vm_fault_t do_shared_fault(struct vm_fault *vmf) |
1da177e4 | 3506 | { |
82b0f8c3 | 3507 | struct vm_area_struct *vma = vmf->vma; |
2b740303 | 3508 | vm_fault_t ret, tmp; |
1d65f86d | 3509 | |
936ca80d | 3510 | ret = __do_fault(vmf); |
7eae74af | 3511 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
f0c6d4d2 | 3512 | return ret; |
1da177e4 LT |
3513 | |
3514 | /* | |
f0c6d4d2 KS |
3515 | * Check if the backing address space wants to know that the page is |
3516 | * about to become writable | |
1da177e4 | 3517 | */ |
fb09a464 | 3518 | if (vma->vm_ops->page_mkwrite) { |
936ca80d | 3519 | unlock_page(vmf->page); |
38b8cb7f | 3520 | tmp = do_page_mkwrite(vmf); |
fb09a464 KS |
3521 | if (unlikely(!tmp || |
3522 | (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { | |
936ca80d | 3523 | put_page(vmf->page); |
fb09a464 | 3524 | return tmp; |
4294621f | 3525 | } |
fb09a464 KS |
3526 | } |
3527 | ||
9118c0cb | 3528 | ret |= finish_fault(vmf); |
7267ec00 KS |
3529 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | |
3530 | VM_FAULT_RETRY))) { | |
936ca80d JK |
3531 | unlock_page(vmf->page); |
3532 | put_page(vmf->page); | |
f0c6d4d2 | 3533 | return ret; |
1da177e4 | 3534 | } |
b827e496 | 3535 | |
97ba0c2b | 3536 | fault_dirty_shared_page(vma, vmf->page); |
1d65f86d | 3537 | return ret; |
54cb8821 | 3538 | } |
d00806b1 | 3539 | |
9a95f3cf PC |
3540 | /* |
3541 | * We enter with non-exclusive mmap_sem (to exclude vma changes, | |
3542 | * but allow concurrent faults). | |
3543 | * The mmap_sem may have been released depending on flags and our | |
3544 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
fc8efd2d JS |
3545 | * If mmap_sem is released, vma may become invalid (for example |
3546 | * by other thread calling munmap()). | |
9a95f3cf | 3547 | */ |
2b740303 | 3548 | static vm_fault_t do_fault(struct vm_fault *vmf) |
54cb8821 | 3549 | { |
82b0f8c3 | 3550 | struct vm_area_struct *vma = vmf->vma; |
fc8efd2d | 3551 | struct mm_struct *vm_mm = vma->vm_mm; |
2b740303 | 3552 | vm_fault_t ret; |
54cb8821 | 3553 | |
ff09d7ec AK |
3554 | /* |
3555 | * The VMA was not fully populated on mmap() or missing VM_DONTEXPAND | |
3556 | */ | |
3557 | if (!vma->vm_ops->fault) { | |
3558 | /* | |
3559 | * If we find a migration pmd entry or a none pmd entry, which | |
3560 | * should never happen, return SIGBUS | |
3561 | */ | |
3562 | if (unlikely(!pmd_present(*vmf->pmd))) | |
3563 | ret = VM_FAULT_SIGBUS; | |
3564 | else { | |
3565 | vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, | |
3566 | vmf->pmd, | |
3567 | vmf->address, | |
3568 | &vmf->ptl); | |
3569 | /* | |
3570 | * Make sure this is not a temporary clearing of pte | |
3571 | * by holding ptl and checking again. A R/M/W update | |
3572 | * of pte involves: take ptl, clearing the pte so that | |
3573 | * we don't have concurrent modification by hardware | |
3574 | * followed by an update. | |
3575 | */ | |
3576 | if (unlikely(pte_none(*vmf->pte))) | |
3577 | ret = VM_FAULT_SIGBUS; | |
3578 | else | |
3579 | ret = VM_FAULT_NOPAGE; | |
3580 | ||
3581 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
3582 | } | |
3583 | } else if (!(vmf->flags & FAULT_FLAG_WRITE)) | |
b0b9b3df HD |
3584 | ret = do_read_fault(vmf); |
3585 | else if (!(vma->vm_flags & VM_SHARED)) | |
3586 | ret = do_cow_fault(vmf); | |
3587 | else | |
3588 | ret = do_shared_fault(vmf); | |
3589 | ||
3590 | /* preallocated pagetable is unused: free it */ | |
3591 | if (vmf->prealloc_pte) { | |
fc8efd2d | 3592 | pte_free(vm_mm, vmf->prealloc_pte); |
7f2b6ce8 | 3593 | vmf->prealloc_pte = NULL; |
b0b9b3df HD |
3594 | } |
3595 | return ret; | |
54cb8821 NP |
3596 | } |
3597 | ||
b19a9939 | 3598 | static int numa_migrate_prep(struct page *page, struct vm_area_struct *vma, |
04bb2f94 RR |
3599 | unsigned long addr, int page_nid, |
3600 | int *flags) | |
9532fec1 MG |
3601 | { |
3602 | get_page(page); | |
3603 | ||
3604 | count_vm_numa_event(NUMA_HINT_FAULTS); | |
04bb2f94 | 3605 | if (page_nid == numa_node_id()) { |
9532fec1 | 3606 | count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL); |
04bb2f94 RR |
3607 | *flags |= TNF_FAULT_LOCAL; |
3608 | } | |
9532fec1 MG |
3609 | |
3610 | return mpol_misplaced(page, vma, addr); | |
3611 | } | |
3612 | ||
2b740303 | 3613 | static vm_fault_t do_numa_page(struct vm_fault *vmf) |
d10e63f2 | 3614 | { |
82b0f8c3 | 3615 | struct vm_area_struct *vma = vmf->vma; |
4daae3b4 | 3616 | struct page *page = NULL; |
98fa15f3 | 3617 | int page_nid = NUMA_NO_NODE; |
90572890 | 3618 | int last_cpupid; |
cbee9f88 | 3619 | int target_nid; |
b8593bfd | 3620 | bool migrated = false; |
04a86453 | 3621 | pte_t pte, old_pte; |
288bc549 | 3622 | bool was_writable = pte_savedwrite(vmf->orig_pte); |
6688cc05 | 3623 | int flags = 0; |
d10e63f2 MG |
3624 | |
3625 | /* | |
166f61b9 TH |
3626 | * The "pte" at this point cannot be used safely without |
3627 | * validation through pte_unmap_same(). It's of NUMA type but | |
3628 | * the pfn may be screwed if the read is non atomic. | |
166f61b9 | 3629 | */ |
82b0f8c3 JK |
3630 | vmf->ptl = pte_lockptr(vma->vm_mm, vmf->pmd); |
3631 | spin_lock(vmf->ptl); | |
cee216a6 | 3632 | if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) { |
82b0f8c3 | 3633 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
4daae3b4 MG |
3634 | goto out; |
3635 | } | |
3636 | ||
cee216a6 AK |
3637 | /* |
3638 | * Make it present again, Depending on how arch implementes non | |
3639 | * accessible ptes, some can allow access by kernel mode. | |
3640 | */ | |
04a86453 AK |
3641 | old_pte = ptep_modify_prot_start(vma, vmf->address, vmf->pte); |
3642 | pte = pte_modify(old_pte, vma->vm_page_prot); | |
4d942466 | 3643 | pte = pte_mkyoung(pte); |
b191f9b1 MG |
3644 | if (was_writable) |
3645 | pte = pte_mkwrite(pte); | |
04a86453 | 3646 | ptep_modify_prot_commit(vma, vmf->address, vmf->pte, old_pte, pte); |
82b0f8c3 | 3647 | update_mmu_cache(vma, vmf->address, vmf->pte); |
d10e63f2 | 3648 | |
82b0f8c3 | 3649 | page = vm_normal_page(vma, vmf->address, pte); |
d10e63f2 | 3650 | if (!page) { |
82b0f8c3 | 3651 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
d10e63f2 MG |
3652 | return 0; |
3653 | } | |
3654 | ||
e81c4802 KS |
3655 | /* TODO: handle PTE-mapped THP */ |
3656 | if (PageCompound(page)) { | |
82b0f8c3 | 3657 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
e81c4802 KS |
3658 | return 0; |
3659 | } | |
3660 | ||
6688cc05 | 3661 | /* |
bea66fbd MG |
3662 | * Avoid grouping on RO pages in general. RO pages shouldn't hurt as |
3663 | * much anyway since they can be in shared cache state. This misses | |
3664 | * the case where a mapping is writable but the process never writes | |
3665 | * to it but pte_write gets cleared during protection updates and | |
3666 | * pte_dirty has unpredictable behaviour between PTE scan updates, | |
3667 | * background writeback, dirty balancing and application behaviour. | |
6688cc05 | 3668 | */ |
d59dc7bc | 3669 | if (!pte_write(pte)) |
6688cc05 PZ |
3670 | flags |= TNF_NO_GROUP; |
3671 | ||
dabe1d99 RR |
3672 | /* |
3673 | * Flag if the page is shared between multiple address spaces. This | |
3674 | * is later used when determining whether to group tasks together | |
3675 | */ | |
3676 | if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED)) | |
3677 | flags |= TNF_SHARED; | |
3678 | ||
90572890 | 3679 | last_cpupid = page_cpupid_last(page); |
8191acbd | 3680 | page_nid = page_to_nid(page); |
82b0f8c3 | 3681 | target_nid = numa_migrate_prep(page, vma, vmf->address, page_nid, |
bae473a4 | 3682 | &flags); |
82b0f8c3 | 3683 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
98fa15f3 | 3684 | if (target_nid == NUMA_NO_NODE) { |
4daae3b4 MG |
3685 | put_page(page); |
3686 | goto out; | |
3687 | } | |
3688 | ||
3689 | /* Migrate to the requested node */ | |
1bc115d8 | 3690 | migrated = migrate_misplaced_page(page, vma, target_nid); |
6688cc05 | 3691 | if (migrated) { |
8191acbd | 3692 | page_nid = target_nid; |
6688cc05 | 3693 | flags |= TNF_MIGRATED; |
074c2381 MG |
3694 | } else |
3695 | flags |= TNF_MIGRATE_FAIL; | |
4daae3b4 MG |
3696 | |
3697 | out: | |
98fa15f3 | 3698 | if (page_nid != NUMA_NO_NODE) |
6688cc05 | 3699 | task_numa_fault(last_cpupid, page_nid, 1, flags); |
d10e63f2 MG |
3700 | return 0; |
3701 | } | |
3702 | ||
2b740303 | 3703 | static inline vm_fault_t create_huge_pmd(struct vm_fault *vmf) |
b96375f7 | 3704 | { |
f4200391 | 3705 | if (vma_is_anonymous(vmf->vma)) |
82b0f8c3 | 3706 | return do_huge_pmd_anonymous_page(vmf); |
a2d58167 | 3707 | if (vmf->vma->vm_ops->huge_fault) |
c791ace1 | 3708 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD); |
b96375f7 MW |
3709 | return VM_FAULT_FALLBACK; |
3710 | } | |
3711 | ||
183f24aa | 3712 | /* `inline' is required to avoid gcc 4.1.2 build error */ |
2b740303 | 3713 | static inline vm_fault_t wp_huge_pmd(struct vm_fault *vmf, pmd_t orig_pmd) |
b96375f7 | 3714 | { |
82b0f8c3 JK |
3715 | if (vma_is_anonymous(vmf->vma)) |
3716 | return do_huge_pmd_wp_page(vmf, orig_pmd); | |
a2d58167 | 3717 | if (vmf->vma->vm_ops->huge_fault) |
c791ace1 | 3718 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD); |
af9e4d5f KS |
3719 | |
3720 | /* COW handled on pte level: split pmd */ | |
82b0f8c3 JK |
3721 | VM_BUG_ON_VMA(vmf->vma->vm_flags & VM_SHARED, vmf->vma); |
3722 | __split_huge_pmd(vmf->vma, vmf->pmd, vmf->address, false, NULL); | |
af9e4d5f | 3723 | |
b96375f7 MW |
3724 | return VM_FAULT_FALLBACK; |
3725 | } | |
3726 | ||
38e08854 LS |
3727 | static inline bool vma_is_accessible(struct vm_area_struct *vma) |
3728 | { | |
3729 | return vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE); | |
3730 | } | |
3731 | ||
2b740303 | 3732 | static vm_fault_t create_huge_pud(struct vm_fault *vmf) |
a00cc7d9 MW |
3733 | { |
3734 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
3735 | /* No support for anonymous transparent PUD pages yet */ | |
3736 | if (vma_is_anonymous(vmf->vma)) | |
3737 | return VM_FAULT_FALLBACK; | |
3738 | if (vmf->vma->vm_ops->huge_fault) | |
c791ace1 | 3739 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD); |
a00cc7d9 MW |
3740 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
3741 | return VM_FAULT_FALLBACK; | |
3742 | } | |
3743 | ||
2b740303 | 3744 | static vm_fault_t wp_huge_pud(struct vm_fault *vmf, pud_t orig_pud) |
a00cc7d9 MW |
3745 | { |
3746 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
3747 | /* No support for anonymous transparent PUD pages yet */ | |
3748 | if (vma_is_anonymous(vmf->vma)) | |
3749 | return VM_FAULT_FALLBACK; | |
3750 | if (vmf->vma->vm_ops->huge_fault) | |
c791ace1 | 3751 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD); |
a00cc7d9 MW |
3752 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
3753 | return VM_FAULT_FALLBACK; | |
3754 | } | |
3755 | ||
1da177e4 LT |
3756 | /* |
3757 | * These routines also need to handle stuff like marking pages dirty | |
3758 | * and/or accessed for architectures that don't do it in hardware (most | |
3759 | * RISC architectures). The early dirtying is also good on the i386. | |
3760 | * | |
3761 | * There is also a hook called "update_mmu_cache()" that architectures | |
3762 | * with external mmu caches can use to update those (ie the Sparc or | |
3763 | * PowerPC hashed page tables that act as extended TLBs). | |
3764 | * | |
7267ec00 KS |
3765 | * We enter with non-exclusive mmap_sem (to exclude vma changes, but allow |
3766 | * concurrent faults). | |
9a95f3cf | 3767 | * |
7267ec00 KS |
3768 | * The mmap_sem may have been released depending on flags and our return value. |
3769 | * See filemap_fault() and __lock_page_or_retry(). | |
1da177e4 | 3770 | */ |
2b740303 | 3771 | static vm_fault_t handle_pte_fault(struct vm_fault *vmf) |
1da177e4 LT |
3772 | { |
3773 | pte_t entry; | |
3774 | ||
82b0f8c3 | 3775 | if (unlikely(pmd_none(*vmf->pmd))) { |
7267ec00 KS |
3776 | /* |
3777 | * Leave __pte_alloc() until later: because vm_ops->fault may | |
3778 | * want to allocate huge page, and if we expose page table | |
3779 | * for an instant, it will be difficult to retract from | |
3780 | * concurrent faults and from rmap lookups. | |
3781 | */ | |
82b0f8c3 | 3782 | vmf->pte = NULL; |
7267ec00 KS |
3783 | } else { |
3784 | /* See comment in pte_alloc_one_map() */ | |
d0f0931d | 3785 | if (pmd_devmap_trans_unstable(vmf->pmd)) |
7267ec00 KS |
3786 | return 0; |
3787 | /* | |
3788 | * A regular pmd is established and it can't morph into a huge | |
3789 | * pmd from under us anymore at this point because we hold the | |
3790 | * mmap_sem read mode and khugepaged takes it in write mode. | |
3791 | * So now it's safe to run pte_offset_map(). | |
3792 | */ | |
82b0f8c3 | 3793 | vmf->pte = pte_offset_map(vmf->pmd, vmf->address); |
2994302b | 3794 | vmf->orig_pte = *vmf->pte; |
7267ec00 KS |
3795 | |
3796 | /* | |
3797 | * some architectures can have larger ptes than wordsize, | |
3798 | * e.g.ppc44x-defconfig has CONFIG_PTE_64BIT=y and | |
b03a0fe0 PM |
3799 | * CONFIG_32BIT=y, so READ_ONCE cannot guarantee atomic |
3800 | * accesses. The code below just needs a consistent view | |
3801 | * for the ifs and we later double check anyway with the | |
7267ec00 KS |
3802 | * ptl lock held. So here a barrier will do. |
3803 | */ | |
3804 | barrier(); | |
2994302b | 3805 | if (pte_none(vmf->orig_pte)) { |
82b0f8c3 JK |
3806 | pte_unmap(vmf->pte); |
3807 | vmf->pte = NULL; | |
65500d23 | 3808 | } |
1da177e4 LT |
3809 | } |
3810 | ||
82b0f8c3 JK |
3811 | if (!vmf->pte) { |
3812 | if (vma_is_anonymous(vmf->vma)) | |
3813 | return do_anonymous_page(vmf); | |
7267ec00 | 3814 | else |
82b0f8c3 | 3815 | return do_fault(vmf); |
7267ec00 KS |
3816 | } |
3817 | ||
2994302b JK |
3818 | if (!pte_present(vmf->orig_pte)) |
3819 | return do_swap_page(vmf); | |
7267ec00 | 3820 | |
2994302b JK |
3821 | if (pte_protnone(vmf->orig_pte) && vma_is_accessible(vmf->vma)) |
3822 | return do_numa_page(vmf); | |
d10e63f2 | 3823 | |
82b0f8c3 JK |
3824 | vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd); |
3825 | spin_lock(vmf->ptl); | |
2994302b | 3826 | entry = vmf->orig_pte; |
82b0f8c3 | 3827 | if (unlikely(!pte_same(*vmf->pte, entry))) |
8f4e2101 | 3828 | goto unlock; |
82b0f8c3 | 3829 | if (vmf->flags & FAULT_FLAG_WRITE) { |
f6f37321 | 3830 | if (!pte_write(entry)) |
2994302b | 3831 | return do_wp_page(vmf); |
1da177e4 LT |
3832 | entry = pte_mkdirty(entry); |
3833 | } | |
3834 | entry = pte_mkyoung(entry); | |
82b0f8c3 JK |
3835 | if (ptep_set_access_flags(vmf->vma, vmf->address, vmf->pte, entry, |
3836 | vmf->flags & FAULT_FLAG_WRITE)) { | |
3837 | update_mmu_cache(vmf->vma, vmf->address, vmf->pte); | |
1a44e149 AA |
3838 | } else { |
3839 | /* | |
3840 | * This is needed only for protection faults but the arch code | |
3841 | * is not yet telling us if this is a protection fault or not. | |
3842 | * This still avoids useless tlb flushes for .text page faults | |
3843 | * with threads. | |
3844 | */ | |
82b0f8c3 JK |
3845 | if (vmf->flags & FAULT_FLAG_WRITE) |
3846 | flush_tlb_fix_spurious_fault(vmf->vma, vmf->address); | |
1a44e149 | 3847 | } |
8f4e2101 | 3848 | unlock: |
82b0f8c3 | 3849 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
83c54070 | 3850 | return 0; |
1da177e4 LT |
3851 | } |
3852 | ||
3853 | /* | |
3854 | * By the time we get here, we already hold the mm semaphore | |
9a95f3cf PC |
3855 | * |
3856 | * The mmap_sem may have been released depending on flags and our | |
3857 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
1da177e4 | 3858 | */ |
2b740303 SJ |
3859 | static vm_fault_t __handle_mm_fault(struct vm_area_struct *vma, |
3860 | unsigned long address, unsigned int flags) | |
1da177e4 | 3861 | { |
82b0f8c3 | 3862 | struct vm_fault vmf = { |
bae473a4 | 3863 | .vma = vma, |
1a29d85e | 3864 | .address = address & PAGE_MASK, |
bae473a4 | 3865 | .flags = flags, |
0721ec8b | 3866 | .pgoff = linear_page_index(vma, address), |
667240e0 | 3867 | .gfp_mask = __get_fault_gfp_mask(vma), |
bae473a4 | 3868 | }; |
fde26bed | 3869 | unsigned int dirty = flags & FAULT_FLAG_WRITE; |
dcddffd4 | 3870 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 3871 | pgd_t *pgd; |
c2febafc | 3872 | p4d_t *p4d; |
2b740303 | 3873 | vm_fault_t ret; |
1da177e4 | 3874 | |
1da177e4 | 3875 | pgd = pgd_offset(mm, address); |
c2febafc KS |
3876 | p4d = p4d_alloc(mm, pgd, address); |
3877 | if (!p4d) | |
3878 | return VM_FAULT_OOM; | |
a00cc7d9 | 3879 | |
c2febafc | 3880 | vmf.pud = pud_alloc(mm, p4d, address); |
a00cc7d9 | 3881 | if (!vmf.pud) |
c74df32c | 3882 | return VM_FAULT_OOM; |
7635d9cb | 3883 | if (pud_none(*vmf.pud) && __transparent_hugepage_enabled(vma)) { |
a00cc7d9 MW |
3884 | ret = create_huge_pud(&vmf); |
3885 | if (!(ret & VM_FAULT_FALLBACK)) | |
3886 | return ret; | |
3887 | } else { | |
3888 | pud_t orig_pud = *vmf.pud; | |
3889 | ||
3890 | barrier(); | |
3891 | if (pud_trans_huge(orig_pud) || pud_devmap(orig_pud)) { | |
a00cc7d9 | 3892 | |
a00cc7d9 MW |
3893 | /* NUMA case for anonymous PUDs would go here */ |
3894 | ||
f6f37321 | 3895 | if (dirty && !pud_write(orig_pud)) { |
a00cc7d9 MW |
3896 | ret = wp_huge_pud(&vmf, orig_pud); |
3897 | if (!(ret & VM_FAULT_FALLBACK)) | |
3898 | return ret; | |
3899 | } else { | |
3900 | huge_pud_set_accessed(&vmf, orig_pud); | |
3901 | return 0; | |
3902 | } | |
3903 | } | |
3904 | } | |
3905 | ||
3906 | vmf.pmd = pmd_alloc(mm, vmf.pud, address); | |
82b0f8c3 | 3907 | if (!vmf.pmd) |
c74df32c | 3908 | return VM_FAULT_OOM; |
7635d9cb | 3909 | if (pmd_none(*vmf.pmd) && __transparent_hugepage_enabled(vma)) { |
a2d58167 | 3910 | ret = create_huge_pmd(&vmf); |
c0292554 KS |
3911 | if (!(ret & VM_FAULT_FALLBACK)) |
3912 | return ret; | |
71e3aac0 | 3913 | } else { |
82b0f8c3 | 3914 | pmd_t orig_pmd = *vmf.pmd; |
1f1d06c3 | 3915 | |
71e3aac0 | 3916 | barrier(); |
84c3fc4e ZY |
3917 | if (unlikely(is_swap_pmd(orig_pmd))) { |
3918 | VM_BUG_ON(thp_migration_supported() && | |
3919 | !is_pmd_migration_entry(orig_pmd)); | |
3920 | if (is_pmd_migration_entry(orig_pmd)) | |
3921 | pmd_migration_entry_wait(mm, vmf.pmd); | |
3922 | return 0; | |
3923 | } | |
5c7fb56e | 3924 | if (pmd_trans_huge(orig_pmd) || pmd_devmap(orig_pmd)) { |
38e08854 | 3925 | if (pmd_protnone(orig_pmd) && vma_is_accessible(vma)) |
82b0f8c3 | 3926 | return do_huge_pmd_numa_page(&vmf, orig_pmd); |
d10e63f2 | 3927 | |
f6f37321 | 3928 | if (dirty && !pmd_write(orig_pmd)) { |
82b0f8c3 | 3929 | ret = wp_huge_pmd(&vmf, orig_pmd); |
9845cbbd KS |
3930 | if (!(ret & VM_FAULT_FALLBACK)) |
3931 | return ret; | |
a1dd450b | 3932 | } else { |
82b0f8c3 | 3933 | huge_pmd_set_accessed(&vmf, orig_pmd); |
9845cbbd | 3934 | return 0; |
1f1d06c3 | 3935 | } |
71e3aac0 AA |
3936 | } |
3937 | } | |
3938 | ||
82b0f8c3 | 3939 | return handle_pte_fault(&vmf); |
1da177e4 LT |
3940 | } |
3941 | ||
9a95f3cf PC |
3942 | /* |
3943 | * By the time we get here, we already hold the mm semaphore | |
3944 | * | |
3945 | * The mmap_sem may have been released depending on flags and our | |
3946 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
3947 | */ | |
2b740303 | 3948 | vm_fault_t handle_mm_fault(struct vm_area_struct *vma, unsigned long address, |
dcddffd4 | 3949 | unsigned int flags) |
519e5247 | 3950 | { |
2b740303 | 3951 | vm_fault_t ret; |
519e5247 JW |
3952 | |
3953 | __set_current_state(TASK_RUNNING); | |
3954 | ||
3955 | count_vm_event(PGFAULT); | |
2262185c | 3956 | count_memcg_event_mm(vma->vm_mm, PGFAULT); |
519e5247 JW |
3957 | |
3958 | /* do counter updates before entering really critical section. */ | |
3959 | check_sync_rss_stat(current); | |
3960 | ||
de0c799b LD |
3961 | if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE, |
3962 | flags & FAULT_FLAG_INSTRUCTION, | |
3963 | flags & FAULT_FLAG_REMOTE)) | |
3964 | return VM_FAULT_SIGSEGV; | |
3965 | ||
519e5247 JW |
3966 | /* |
3967 | * Enable the memcg OOM handling for faults triggered in user | |
3968 | * space. Kernel faults are handled more gracefully. | |
3969 | */ | |
3970 | if (flags & FAULT_FLAG_USER) | |
29ef680a | 3971 | mem_cgroup_enter_user_fault(); |
519e5247 | 3972 | |
bae473a4 KS |
3973 | if (unlikely(is_vm_hugetlb_page(vma))) |
3974 | ret = hugetlb_fault(vma->vm_mm, vma, address, flags); | |
3975 | else | |
3976 | ret = __handle_mm_fault(vma, address, flags); | |
519e5247 | 3977 | |
49426420 | 3978 | if (flags & FAULT_FLAG_USER) { |
29ef680a | 3979 | mem_cgroup_exit_user_fault(); |
166f61b9 TH |
3980 | /* |
3981 | * The task may have entered a memcg OOM situation but | |
3982 | * if the allocation error was handled gracefully (no | |
3983 | * VM_FAULT_OOM), there is no need to kill anything. | |
3984 | * Just clean up the OOM state peacefully. | |
3985 | */ | |
3986 | if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM)) | |
3987 | mem_cgroup_oom_synchronize(false); | |
49426420 | 3988 | } |
3812c8c8 | 3989 | |
519e5247 JW |
3990 | return ret; |
3991 | } | |
e1d6d01a | 3992 | EXPORT_SYMBOL_GPL(handle_mm_fault); |
519e5247 | 3993 | |
90eceff1 KS |
3994 | #ifndef __PAGETABLE_P4D_FOLDED |
3995 | /* | |
3996 | * Allocate p4d page table. | |
3997 | * We've already handled the fast-path in-line. | |
3998 | */ | |
3999 | int __p4d_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) | |
4000 | { | |
4001 | p4d_t *new = p4d_alloc_one(mm, address); | |
4002 | if (!new) | |
4003 | return -ENOMEM; | |
4004 | ||
4005 | smp_wmb(); /* See comment in __pte_alloc */ | |
4006 | ||
4007 | spin_lock(&mm->page_table_lock); | |
4008 | if (pgd_present(*pgd)) /* Another has populated it */ | |
4009 | p4d_free(mm, new); | |
4010 | else | |
4011 | pgd_populate(mm, pgd, new); | |
4012 | spin_unlock(&mm->page_table_lock); | |
4013 | return 0; | |
4014 | } | |
4015 | #endif /* __PAGETABLE_P4D_FOLDED */ | |
4016 | ||
1da177e4 LT |
4017 | #ifndef __PAGETABLE_PUD_FOLDED |
4018 | /* | |
4019 | * Allocate page upper directory. | |
872fec16 | 4020 | * We've already handled the fast-path in-line. |
1da177e4 | 4021 | */ |
c2febafc | 4022 | int __pud_alloc(struct mm_struct *mm, p4d_t *p4d, unsigned long address) |
1da177e4 | 4023 | { |
c74df32c HD |
4024 | pud_t *new = pud_alloc_one(mm, address); |
4025 | if (!new) | |
1bb3630e | 4026 | return -ENOMEM; |
1da177e4 | 4027 | |
362a61ad NP |
4028 | smp_wmb(); /* See comment in __pte_alloc */ |
4029 | ||
872fec16 | 4030 | spin_lock(&mm->page_table_lock); |
c2febafc | 4031 | #ifndef __ARCH_HAS_5LEVEL_HACK |
b4e98d9a KS |
4032 | if (!p4d_present(*p4d)) { |
4033 | mm_inc_nr_puds(mm); | |
c2febafc | 4034 | p4d_populate(mm, p4d, new); |
b4e98d9a | 4035 | } else /* Another has populated it */ |
5e541973 | 4036 | pud_free(mm, new); |
b4e98d9a KS |
4037 | #else |
4038 | if (!pgd_present(*p4d)) { | |
4039 | mm_inc_nr_puds(mm); | |
c2febafc | 4040 | pgd_populate(mm, p4d, new); |
b4e98d9a KS |
4041 | } else /* Another has populated it */ |
4042 | pud_free(mm, new); | |
c2febafc | 4043 | #endif /* __ARCH_HAS_5LEVEL_HACK */ |
c74df32c | 4044 | spin_unlock(&mm->page_table_lock); |
1bb3630e | 4045 | return 0; |
1da177e4 LT |
4046 | } |
4047 | #endif /* __PAGETABLE_PUD_FOLDED */ | |
4048 | ||
4049 | #ifndef __PAGETABLE_PMD_FOLDED | |
4050 | /* | |
4051 | * Allocate page middle directory. | |
872fec16 | 4052 | * We've already handled the fast-path in-line. |
1da177e4 | 4053 | */ |
1bb3630e | 4054 | int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) |
1da177e4 | 4055 | { |
a00cc7d9 | 4056 | spinlock_t *ptl; |
c74df32c HD |
4057 | pmd_t *new = pmd_alloc_one(mm, address); |
4058 | if (!new) | |
1bb3630e | 4059 | return -ENOMEM; |
1da177e4 | 4060 | |
362a61ad NP |
4061 | smp_wmb(); /* See comment in __pte_alloc */ |
4062 | ||
a00cc7d9 | 4063 | ptl = pud_lock(mm, pud); |
1da177e4 | 4064 | #ifndef __ARCH_HAS_4LEVEL_HACK |
dc6c9a35 KS |
4065 | if (!pud_present(*pud)) { |
4066 | mm_inc_nr_pmds(mm); | |
1bb3630e | 4067 | pud_populate(mm, pud, new); |
dc6c9a35 | 4068 | } else /* Another has populated it */ |
5e541973 | 4069 | pmd_free(mm, new); |
dc6c9a35 KS |
4070 | #else |
4071 | if (!pgd_present(*pud)) { | |
4072 | mm_inc_nr_pmds(mm); | |
1bb3630e | 4073 | pgd_populate(mm, pud, new); |
dc6c9a35 KS |
4074 | } else /* Another has populated it */ |
4075 | pmd_free(mm, new); | |
1da177e4 | 4076 | #endif /* __ARCH_HAS_4LEVEL_HACK */ |
a00cc7d9 | 4077 | spin_unlock(ptl); |
1bb3630e | 4078 | return 0; |
e0f39591 | 4079 | } |
1da177e4 LT |
4080 | #endif /* __PAGETABLE_PMD_FOLDED */ |
4081 | ||
09796395 | 4082 | static int __follow_pte_pmd(struct mm_struct *mm, unsigned long address, |
ac46d4f3 | 4083 | struct mmu_notifier_range *range, |
a4d1a885 | 4084 | pte_t **ptepp, pmd_t **pmdpp, spinlock_t **ptlp) |
f8ad0f49 JW |
4085 | { |
4086 | pgd_t *pgd; | |
c2febafc | 4087 | p4d_t *p4d; |
f8ad0f49 JW |
4088 | pud_t *pud; |
4089 | pmd_t *pmd; | |
4090 | pte_t *ptep; | |
4091 | ||
4092 | pgd = pgd_offset(mm, address); | |
4093 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
4094 | goto out; | |
4095 | ||
c2febafc KS |
4096 | p4d = p4d_offset(pgd, address); |
4097 | if (p4d_none(*p4d) || unlikely(p4d_bad(*p4d))) | |
4098 | goto out; | |
4099 | ||
4100 | pud = pud_offset(p4d, address); | |
f8ad0f49 JW |
4101 | if (pud_none(*pud) || unlikely(pud_bad(*pud))) |
4102 | goto out; | |
4103 | ||
4104 | pmd = pmd_offset(pud, address); | |
f66055ab | 4105 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
f8ad0f49 | 4106 | |
09796395 RZ |
4107 | if (pmd_huge(*pmd)) { |
4108 | if (!pmdpp) | |
4109 | goto out; | |
4110 | ||
ac46d4f3 | 4111 | if (range) { |
6f4f13e8 JG |
4112 | mmu_notifier_range_init(range, MMU_NOTIFY_UNMAP, 0, |
4113 | NULL, mm, address & PMD_MASK, | |
4114 | (address & PMD_MASK) + PMD_SIZE); | |
ac46d4f3 | 4115 | mmu_notifier_invalidate_range_start(range); |
a4d1a885 | 4116 | } |
09796395 RZ |
4117 | *ptlp = pmd_lock(mm, pmd); |
4118 | if (pmd_huge(*pmd)) { | |
4119 | *pmdpp = pmd; | |
4120 | return 0; | |
4121 | } | |
4122 | spin_unlock(*ptlp); | |
ac46d4f3 JG |
4123 | if (range) |
4124 | mmu_notifier_invalidate_range_end(range); | |
09796395 RZ |
4125 | } |
4126 | ||
4127 | if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) | |
f8ad0f49 JW |
4128 | goto out; |
4129 | ||
ac46d4f3 | 4130 | if (range) { |
6f4f13e8 JG |
4131 | mmu_notifier_range_init(range, MMU_NOTIFY_UNMAP, 0, NULL, mm, |
4132 | address & PAGE_MASK, | |
4133 | (address & PAGE_MASK) + PAGE_SIZE); | |
ac46d4f3 | 4134 | mmu_notifier_invalidate_range_start(range); |
a4d1a885 | 4135 | } |
f8ad0f49 | 4136 | ptep = pte_offset_map_lock(mm, pmd, address, ptlp); |
f8ad0f49 JW |
4137 | if (!pte_present(*ptep)) |
4138 | goto unlock; | |
4139 | *ptepp = ptep; | |
4140 | return 0; | |
4141 | unlock: | |
4142 | pte_unmap_unlock(ptep, *ptlp); | |
ac46d4f3 JG |
4143 | if (range) |
4144 | mmu_notifier_invalidate_range_end(range); | |
f8ad0f49 JW |
4145 | out: |
4146 | return -EINVAL; | |
4147 | } | |
4148 | ||
f729c8c9 RZ |
4149 | static inline int follow_pte(struct mm_struct *mm, unsigned long address, |
4150 | pte_t **ptepp, spinlock_t **ptlp) | |
1b36ba81 NK |
4151 | { |
4152 | int res; | |
4153 | ||
4154 | /* (void) is needed to make gcc happy */ | |
4155 | (void) __cond_lock(*ptlp, | |
ac46d4f3 | 4156 | !(res = __follow_pte_pmd(mm, address, NULL, |
a4d1a885 | 4157 | ptepp, NULL, ptlp))); |
09796395 RZ |
4158 | return res; |
4159 | } | |
4160 | ||
4161 | int follow_pte_pmd(struct mm_struct *mm, unsigned long address, | |
ac46d4f3 JG |
4162 | struct mmu_notifier_range *range, |
4163 | pte_t **ptepp, pmd_t **pmdpp, spinlock_t **ptlp) | |
09796395 RZ |
4164 | { |
4165 | int res; | |
4166 | ||
4167 | /* (void) is needed to make gcc happy */ | |
4168 | (void) __cond_lock(*ptlp, | |
ac46d4f3 | 4169 | !(res = __follow_pte_pmd(mm, address, range, |
a4d1a885 | 4170 | ptepp, pmdpp, ptlp))); |
1b36ba81 NK |
4171 | return res; |
4172 | } | |
09796395 | 4173 | EXPORT_SYMBOL(follow_pte_pmd); |
1b36ba81 | 4174 | |
3b6748e2 JW |
4175 | /** |
4176 | * follow_pfn - look up PFN at a user virtual address | |
4177 | * @vma: memory mapping | |
4178 | * @address: user virtual address | |
4179 | * @pfn: location to store found PFN | |
4180 | * | |
4181 | * Only IO mappings and raw PFN mappings are allowed. | |
4182 | * | |
a862f68a | 4183 | * Return: zero and the pfn at @pfn on success, -ve otherwise. |
3b6748e2 JW |
4184 | */ |
4185 | int follow_pfn(struct vm_area_struct *vma, unsigned long address, | |
4186 | unsigned long *pfn) | |
4187 | { | |
4188 | int ret = -EINVAL; | |
4189 | spinlock_t *ptl; | |
4190 | pte_t *ptep; | |
4191 | ||
4192 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) | |
4193 | return ret; | |
4194 | ||
4195 | ret = follow_pte(vma->vm_mm, address, &ptep, &ptl); | |
4196 | if (ret) | |
4197 | return ret; | |
4198 | *pfn = pte_pfn(*ptep); | |
4199 | pte_unmap_unlock(ptep, ptl); | |
4200 | return 0; | |
4201 | } | |
4202 | EXPORT_SYMBOL(follow_pfn); | |
4203 | ||
28b2ee20 | 4204 | #ifdef CONFIG_HAVE_IOREMAP_PROT |
d87fe660 | 4205 | int follow_phys(struct vm_area_struct *vma, |
4206 | unsigned long address, unsigned int flags, | |
4207 | unsigned long *prot, resource_size_t *phys) | |
28b2ee20 | 4208 | { |
03668a4d | 4209 | int ret = -EINVAL; |
28b2ee20 RR |
4210 | pte_t *ptep, pte; |
4211 | spinlock_t *ptl; | |
28b2ee20 | 4212 | |
d87fe660 | 4213 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) |
4214 | goto out; | |
28b2ee20 | 4215 | |
03668a4d | 4216 | if (follow_pte(vma->vm_mm, address, &ptep, &ptl)) |
d87fe660 | 4217 | goto out; |
28b2ee20 | 4218 | pte = *ptep; |
03668a4d | 4219 | |
f6f37321 | 4220 | if ((flags & FOLL_WRITE) && !pte_write(pte)) |
28b2ee20 | 4221 | goto unlock; |
28b2ee20 RR |
4222 | |
4223 | *prot = pgprot_val(pte_pgprot(pte)); | |
03668a4d | 4224 | *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; |
28b2ee20 | 4225 | |
03668a4d | 4226 | ret = 0; |
28b2ee20 RR |
4227 | unlock: |
4228 | pte_unmap_unlock(ptep, ptl); | |
4229 | out: | |
d87fe660 | 4230 | return ret; |
28b2ee20 RR |
4231 | } |
4232 | ||
4233 | int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, | |
4234 | void *buf, int len, int write) | |
4235 | { | |
4236 | resource_size_t phys_addr; | |
4237 | unsigned long prot = 0; | |
2bc7273b | 4238 | void __iomem *maddr; |
28b2ee20 RR |
4239 | int offset = addr & (PAGE_SIZE-1); |
4240 | ||
d87fe660 | 4241 | if (follow_phys(vma, addr, write, &prot, &phys_addr)) |
28b2ee20 RR |
4242 | return -EINVAL; |
4243 | ||
9cb12d7b | 4244 | maddr = ioremap_prot(phys_addr, PAGE_ALIGN(len + offset), prot); |
24eee1e4 | 4245 | if (!maddr) |
4246 | return -ENOMEM; | |
4247 | ||
28b2ee20 RR |
4248 | if (write) |
4249 | memcpy_toio(maddr + offset, buf, len); | |
4250 | else | |
4251 | memcpy_fromio(buf, maddr + offset, len); | |
4252 | iounmap(maddr); | |
4253 | ||
4254 | return len; | |
4255 | } | |
5a73633e | 4256 | EXPORT_SYMBOL_GPL(generic_access_phys); |
28b2ee20 RR |
4257 | #endif |
4258 | ||
0ec76a11 | 4259 | /* |
206cb636 SW |
4260 | * Access another process' address space as given in mm. If non-NULL, use the |
4261 | * given task for page fault accounting. | |
0ec76a11 | 4262 | */ |
84d77d3f | 4263 | int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, |
442486ec | 4264 | unsigned long addr, void *buf, int len, unsigned int gup_flags) |
0ec76a11 | 4265 | { |
0ec76a11 | 4266 | struct vm_area_struct *vma; |
0ec76a11 | 4267 | void *old_buf = buf; |
442486ec | 4268 | int write = gup_flags & FOLL_WRITE; |
0ec76a11 | 4269 | |
0ec76a11 | 4270 | down_read(&mm->mmap_sem); |
183ff22b | 4271 | /* ignore errors, just check how much was successfully transferred */ |
0ec76a11 DH |
4272 | while (len) { |
4273 | int bytes, ret, offset; | |
4274 | void *maddr; | |
28b2ee20 | 4275 | struct page *page = NULL; |
0ec76a11 | 4276 | |
1e987790 | 4277 | ret = get_user_pages_remote(tsk, mm, addr, 1, |
5b56d49f | 4278 | gup_flags, &page, &vma, NULL); |
28b2ee20 | 4279 | if (ret <= 0) { |
dbffcd03 RR |
4280 | #ifndef CONFIG_HAVE_IOREMAP_PROT |
4281 | break; | |
4282 | #else | |
28b2ee20 RR |
4283 | /* |
4284 | * Check if this is a VM_IO | VM_PFNMAP VMA, which | |
4285 | * we can access using slightly different code. | |
4286 | */ | |
28b2ee20 | 4287 | vma = find_vma(mm, addr); |
fe936dfc | 4288 | if (!vma || vma->vm_start > addr) |
28b2ee20 RR |
4289 | break; |
4290 | if (vma->vm_ops && vma->vm_ops->access) | |
4291 | ret = vma->vm_ops->access(vma, addr, buf, | |
4292 | len, write); | |
4293 | if (ret <= 0) | |
28b2ee20 RR |
4294 | break; |
4295 | bytes = ret; | |
dbffcd03 | 4296 | #endif |
0ec76a11 | 4297 | } else { |
28b2ee20 RR |
4298 | bytes = len; |
4299 | offset = addr & (PAGE_SIZE-1); | |
4300 | if (bytes > PAGE_SIZE-offset) | |
4301 | bytes = PAGE_SIZE-offset; | |
4302 | ||
4303 | maddr = kmap(page); | |
4304 | if (write) { | |
4305 | copy_to_user_page(vma, page, addr, | |
4306 | maddr + offset, buf, bytes); | |
4307 | set_page_dirty_lock(page); | |
4308 | } else { | |
4309 | copy_from_user_page(vma, page, addr, | |
4310 | buf, maddr + offset, bytes); | |
4311 | } | |
4312 | kunmap(page); | |
09cbfeaf | 4313 | put_page(page); |
0ec76a11 | 4314 | } |
0ec76a11 DH |
4315 | len -= bytes; |
4316 | buf += bytes; | |
4317 | addr += bytes; | |
4318 | } | |
4319 | up_read(&mm->mmap_sem); | |
0ec76a11 DH |
4320 | |
4321 | return buf - old_buf; | |
4322 | } | |
03252919 | 4323 | |
5ddd36b9 | 4324 | /** |
ae91dbfc | 4325 | * access_remote_vm - access another process' address space |
5ddd36b9 SW |
4326 | * @mm: the mm_struct of the target address space |
4327 | * @addr: start address to access | |
4328 | * @buf: source or destination buffer | |
4329 | * @len: number of bytes to transfer | |
6347e8d5 | 4330 | * @gup_flags: flags modifying lookup behaviour |
5ddd36b9 SW |
4331 | * |
4332 | * The caller must hold a reference on @mm. | |
a862f68a MR |
4333 | * |
4334 | * Return: number of bytes copied from source to destination. | |
5ddd36b9 SW |
4335 | */ |
4336 | int access_remote_vm(struct mm_struct *mm, unsigned long addr, | |
6347e8d5 | 4337 | void *buf, int len, unsigned int gup_flags) |
5ddd36b9 | 4338 | { |
6347e8d5 | 4339 | return __access_remote_vm(NULL, mm, addr, buf, len, gup_flags); |
5ddd36b9 SW |
4340 | } |
4341 | ||
206cb636 SW |
4342 | /* |
4343 | * Access another process' address space. | |
4344 | * Source/target buffer must be kernel space, | |
4345 | * Do not walk the page table directly, use get_user_pages | |
4346 | */ | |
4347 | int access_process_vm(struct task_struct *tsk, unsigned long addr, | |
f307ab6d | 4348 | void *buf, int len, unsigned int gup_flags) |
206cb636 SW |
4349 | { |
4350 | struct mm_struct *mm; | |
4351 | int ret; | |
4352 | ||
4353 | mm = get_task_mm(tsk); | |
4354 | if (!mm) | |
4355 | return 0; | |
4356 | ||
f307ab6d | 4357 | ret = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags); |
442486ec | 4358 | |
206cb636 SW |
4359 | mmput(mm); |
4360 | ||
4361 | return ret; | |
4362 | } | |
fcd35857 | 4363 | EXPORT_SYMBOL_GPL(access_process_vm); |
206cb636 | 4364 | |
03252919 AK |
4365 | /* |
4366 | * Print the name of a VMA. | |
4367 | */ | |
4368 | void print_vma_addr(char *prefix, unsigned long ip) | |
4369 | { | |
4370 | struct mm_struct *mm = current->mm; | |
4371 | struct vm_area_struct *vma; | |
4372 | ||
e8bff74a | 4373 | /* |
0a7f682d | 4374 | * we might be running from an atomic context so we cannot sleep |
e8bff74a | 4375 | */ |
0a7f682d | 4376 | if (!down_read_trylock(&mm->mmap_sem)) |
e8bff74a IM |
4377 | return; |
4378 | ||
03252919 AK |
4379 | vma = find_vma(mm, ip); |
4380 | if (vma && vma->vm_file) { | |
4381 | struct file *f = vma->vm_file; | |
0a7f682d | 4382 | char *buf = (char *)__get_free_page(GFP_NOWAIT); |
03252919 | 4383 | if (buf) { |
2fbc57c5 | 4384 | char *p; |
03252919 | 4385 | |
9bf39ab2 | 4386 | p = file_path(f, buf, PAGE_SIZE); |
03252919 AK |
4387 | if (IS_ERR(p)) |
4388 | p = "?"; | |
2fbc57c5 | 4389 | printk("%s%s[%lx+%lx]", prefix, kbasename(p), |
03252919 AK |
4390 | vma->vm_start, |
4391 | vma->vm_end - vma->vm_start); | |
4392 | free_page((unsigned long)buf); | |
4393 | } | |
4394 | } | |
51a07e50 | 4395 | up_read(&mm->mmap_sem); |
03252919 | 4396 | } |
3ee1afa3 | 4397 | |
662bbcb2 | 4398 | #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP) |
9ec23531 | 4399 | void __might_fault(const char *file, int line) |
3ee1afa3 | 4400 | { |
95156f00 PZ |
4401 | /* |
4402 | * Some code (nfs/sunrpc) uses socket ops on kernel memory while | |
4403 | * holding the mmap_sem, this is safe because kernel memory doesn't | |
4404 | * get paged out, therefore we'll never actually fault, and the | |
4405 | * below annotations will generate false positives. | |
4406 | */ | |
db68ce10 | 4407 | if (uaccess_kernel()) |
95156f00 | 4408 | return; |
9ec23531 | 4409 | if (pagefault_disabled()) |
662bbcb2 | 4410 | return; |
9ec23531 DH |
4411 | __might_sleep(file, line, 0); |
4412 | #if defined(CONFIG_DEBUG_ATOMIC_SLEEP) | |
662bbcb2 | 4413 | if (current->mm) |
3ee1afa3 | 4414 | might_lock_read(¤t->mm->mmap_sem); |
9ec23531 | 4415 | #endif |
3ee1afa3 | 4416 | } |
9ec23531 | 4417 | EXPORT_SYMBOL(__might_fault); |
3ee1afa3 | 4418 | #endif |
47ad8475 AA |
4419 | |
4420 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) | |
c6ddfb6c HY |
4421 | /* |
4422 | * Process all subpages of the specified huge page with the specified | |
4423 | * operation. The target subpage will be processed last to keep its | |
4424 | * cache lines hot. | |
4425 | */ | |
4426 | static inline void process_huge_page( | |
4427 | unsigned long addr_hint, unsigned int pages_per_huge_page, | |
4428 | void (*process_subpage)(unsigned long addr, int idx, void *arg), | |
4429 | void *arg) | |
47ad8475 | 4430 | { |
c79b57e4 HY |
4431 | int i, n, base, l; |
4432 | unsigned long addr = addr_hint & | |
4433 | ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); | |
47ad8475 | 4434 | |
c6ddfb6c | 4435 | /* Process target subpage last to keep its cache lines hot */ |
47ad8475 | 4436 | might_sleep(); |
c79b57e4 HY |
4437 | n = (addr_hint - addr) / PAGE_SIZE; |
4438 | if (2 * n <= pages_per_huge_page) { | |
c6ddfb6c | 4439 | /* If target subpage in first half of huge page */ |
c79b57e4 HY |
4440 | base = 0; |
4441 | l = n; | |
c6ddfb6c | 4442 | /* Process subpages at the end of huge page */ |
c79b57e4 HY |
4443 | for (i = pages_per_huge_page - 1; i >= 2 * n; i--) { |
4444 | cond_resched(); | |
c6ddfb6c | 4445 | process_subpage(addr + i * PAGE_SIZE, i, arg); |
c79b57e4 HY |
4446 | } |
4447 | } else { | |
c6ddfb6c | 4448 | /* If target subpage in second half of huge page */ |
c79b57e4 HY |
4449 | base = pages_per_huge_page - 2 * (pages_per_huge_page - n); |
4450 | l = pages_per_huge_page - n; | |
c6ddfb6c | 4451 | /* Process subpages at the begin of huge page */ |
c79b57e4 HY |
4452 | for (i = 0; i < base; i++) { |
4453 | cond_resched(); | |
c6ddfb6c | 4454 | process_subpage(addr + i * PAGE_SIZE, i, arg); |
c79b57e4 HY |
4455 | } |
4456 | } | |
4457 | /* | |
c6ddfb6c HY |
4458 | * Process remaining subpages in left-right-left-right pattern |
4459 | * towards the target subpage | |
c79b57e4 HY |
4460 | */ |
4461 | for (i = 0; i < l; i++) { | |
4462 | int left_idx = base + i; | |
4463 | int right_idx = base + 2 * l - 1 - i; | |
4464 | ||
4465 | cond_resched(); | |
c6ddfb6c | 4466 | process_subpage(addr + left_idx * PAGE_SIZE, left_idx, arg); |
47ad8475 | 4467 | cond_resched(); |
c6ddfb6c | 4468 | process_subpage(addr + right_idx * PAGE_SIZE, right_idx, arg); |
47ad8475 AA |
4469 | } |
4470 | } | |
4471 | ||
c6ddfb6c HY |
4472 | static void clear_gigantic_page(struct page *page, |
4473 | unsigned long addr, | |
4474 | unsigned int pages_per_huge_page) | |
4475 | { | |
4476 | int i; | |
4477 | struct page *p = page; | |
4478 | ||
4479 | might_sleep(); | |
4480 | for (i = 0; i < pages_per_huge_page; | |
4481 | i++, p = mem_map_next(p, page, i)) { | |
4482 | cond_resched(); | |
4483 | clear_user_highpage(p, addr + i * PAGE_SIZE); | |
4484 | } | |
4485 | } | |
4486 | ||
4487 | static void clear_subpage(unsigned long addr, int idx, void *arg) | |
4488 | { | |
4489 | struct page *page = arg; | |
4490 | ||
4491 | clear_user_highpage(page + idx, addr); | |
4492 | } | |
4493 | ||
4494 | void clear_huge_page(struct page *page, | |
4495 | unsigned long addr_hint, unsigned int pages_per_huge_page) | |
4496 | { | |
4497 | unsigned long addr = addr_hint & | |
4498 | ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); | |
4499 | ||
4500 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { | |
4501 | clear_gigantic_page(page, addr, pages_per_huge_page); | |
4502 | return; | |
4503 | } | |
4504 | ||
4505 | process_huge_page(addr_hint, pages_per_huge_page, clear_subpage, page); | |
4506 | } | |
4507 | ||
47ad8475 AA |
4508 | static void copy_user_gigantic_page(struct page *dst, struct page *src, |
4509 | unsigned long addr, | |
4510 | struct vm_area_struct *vma, | |
4511 | unsigned int pages_per_huge_page) | |
4512 | { | |
4513 | int i; | |
4514 | struct page *dst_base = dst; | |
4515 | struct page *src_base = src; | |
4516 | ||
4517 | for (i = 0; i < pages_per_huge_page; ) { | |
4518 | cond_resched(); | |
4519 | copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma); | |
4520 | ||
4521 | i++; | |
4522 | dst = mem_map_next(dst, dst_base, i); | |
4523 | src = mem_map_next(src, src_base, i); | |
4524 | } | |
4525 | } | |
4526 | ||
c9f4cd71 HY |
4527 | struct copy_subpage_arg { |
4528 | struct page *dst; | |
4529 | struct page *src; | |
4530 | struct vm_area_struct *vma; | |
4531 | }; | |
4532 | ||
4533 | static void copy_subpage(unsigned long addr, int idx, void *arg) | |
4534 | { | |
4535 | struct copy_subpage_arg *copy_arg = arg; | |
4536 | ||
4537 | copy_user_highpage(copy_arg->dst + idx, copy_arg->src + idx, | |
4538 | addr, copy_arg->vma); | |
4539 | } | |
4540 | ||
47ad8475 | 4541 | void copy_user_huge_page(struct page *dst, struct page *src, |
c9f4cd71 | 4542 | unsigned long addr_hint, struct vm_area_struct *vma, |
47ad8475 AA |
4543 | unsigned int pages_per_huge_page) |
4544 | { | |
c9f4cd71 HY |
4545 | unsigned long addr = addr_hint & |
4546 | ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); | |
4547 | struct copy_subpage_arg arg = { | |
4548 | .dst = dst, | |
4549 | .src = src, | |
4550 | .vma = vma, | |
4551 | }; | |
47ad8475 AA |
4552 | |
4553 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { | |
4554 | copy_user_gigantic_page(dst, src, addr, vma, | |
4555 | pages_per_huge_page); | |
4556 | return; | |
4557 | } | |
4558 | ||
c9f4cd71 | 4559 | process_huge_page(addr_hint, pages_per_huge_page, copy_subpage, &arg); |
47ad8475 | 4560 | } |
fa4d75c1 MK |
4561 | |
4562 | long copy_huge_page_from_user(struct page *dst_page, | |
4563 | const void __user *usr_src, | |
810a56b9 MK |
4564 | unsigned int pages_per_huge_page, |
4565 | bool allow_pagefault) | |
fa4d75c1 MK |
4566 | { |
4567 | void *src = (void *)usr_src; | |
4568 | void *page_kaddr; | |
4569 | unsigned long i, rc = 0; | |
4570 | unsigned long ret_val = pages_per_huge_page * PAGE_SIZE; | |
4571 | ||
4572 | for (i = 0; i < pages_per_huge_page; i++) { | |
810a56b9 MK |
4573 | if (allow_pagefault) |
4574 | page_kaddr = kmap(dst_page + i); | |
4575 | else | |
4576 | page_kaddr = kmap_atomic(dst_page + i); | |
fa4d75c1 MK |
4577 | rc = copy_from_user(page_kaddr, |
4578 | (const void __user *)(src + i * PAGE_SIZE), | |
4579 | PAGE_SIZE); | |
810a56b9 MK |
4580 | if (allow_pagefault) |
4581 | kunmap(dst_page + i); | |
4582 | else | |
4583 | kunmap_atomic(page_kaddr); | |
fa4d75c1 MK |
4584 | |
4585 | ret_val -= (PAGE_SIZE - rc); | |
4586 | if (rc) | |
4587 | break; | |
4588 | ||
4589 | cond_resched(); | |
4590 | } | |
4591 | return ret_val; | |
4592 | } | |
47ad8475 | 4593 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */ |
49076ec2 | 4594 | |
40b64acd | 4595 | #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS |
b35f1819 KS |
4596 | |
4597 | static struct kmem_cache *page_ptl_cachep; | |
4598 | ||
4599 | void __init ptlock_cache_init(void) | |
4600 | { | |
4601 | page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0, | |
4602 | SLAB_PANIC, NULL); | |
4603 | } | |
4604 | ||
539edb58 | 4605 | bool ptlock_alloc(struct page *page) |
49076ec2 KS |
4606 | { |
4607 | spinlock_t *ptl; | |
4608 | ||
b35f1819 | 4609 | ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL); |
49076ec2 KS |
4610 | if (!ptl) |
4611 | return false; | |
539edb58 | 4612 | page->ptl = ptl; |
49076ec2 KS |
4613 | return true; |
4614 | } | |
4615 | ||
539edb58 | 4616 | void ptlock_free(struct page *page) |
49076ec2 | 4617 | { |
b35f1819 | 4618 | kmem_cache_free(page_ptl_cachep, page->ptl); |
49076ec2 KS |
4619 | } |
4620 | #endif |