]>
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> | |
9a840895 | 52 | #include <linux/ksm.h> |
1da177e4 | 53 | #include <linux/rmap.h> |
b95f1b31 | 54 | #include <linux/export.h> |
0ff92245 | 55 | #include <linux/delayacct.h> |
1da177e4 | 56 | #include <linux/init.h> |
01c8f1c4 | 57 | #include <linux/pfn_t.h> |
edc79b2a | 58 | #include <linux/writeback.h> |
8a9f3ccd | 59 | #include <linux/memcontrol.h> |
cddb8a5c | 60 | #include <linux/mmu_notifier.h> |
3dc14741 HD |
61 | #include <linux/kallsyms.h> |
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> |
1da177e4 | 72 | |
6952b61d | 73 | #include <asm/io.h> |
33a709b2 | 74 | #include <asm/mmu_context.h> |
1da177e4 | 75 | #include <asm/pgalloc.h> |
7c0f6ba6 | 76 | #include <linux/uaccess.h> |
1da177e4 LT |
77 | #include <asm/tlb.h> |
78 | #include <asm/tlbflush.h> | |
79 | #include <asm/pgtable.h> | |
80 | ||
42b77728 JB |
81 | #include "internal.h" |
82 | ||
90572890 PZ |
83 | #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS |
84 | #warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid. | |
75980e97 PZ |
85 | #endif |
86 | ||
d41dee36 | 87 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
88 | /* use the per-pgdat data instead for discontigmem - mbligh */ |
89 | unsigned long max_mapnr; | |
1da177e4 | 90 | EXPORT_SYMBOL(max_mapnr); |
166f61b9 TH |
91 | |
92 | struct page *mem_map; | |
1da177e4 LT |
93 | EXPORT_SYMBOL(mem_map); |
94 | #endif | |
95 | ||
1da177e4 LT |
96 | /* |
97 | * A number of key systems in x86 including ioremap() rely on the assumption | |
98 | * that high_memory defines the upper bound on direct map memory, then end | |
99 | * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and | |
100 | * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL | |
101 | * and ZONE_HIGHMEM. | |
102 | */ | |
166f61b9 | 103 | void *high_memory; |
1da177e4 | 104 | EXPORT_SYMBOL(high_memory); |
1da177e4 | 105 | |
32a93233 IM |
106 | /* |
107 | * Randomize the address space (stacks, mmaps, brk, etc.). | |
108 | * | |
109 | * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization, | |
110 | * as ancient (libc5 based) binaries can segfault. ) | |
111 | */ | |
112 | int randomize_va_space __read_mostly = | |
113 | #ifdef CONFIG_COMPAT_BRK | |
114 | 1; | |
115 | #else | |
116 | 2; | |
117 | #endif | |
a62eaf15 AK |
118 | |
119 | static int __init disable_randmaps(char *s) | |
120 | { | |
121 | randomize_va_space = 0; | |
9b41046c | 122 | return 1; |
a62eaf15 AK |
123 | } |
124 | __setup("norandmaps", disable_randmaps); | |
125 | ||
62eede62 | 126 | unsigned long zero_pfn __read_mostly; |
0b70068e AB |
127 | EXPORT_SYMBOL(zero_pfn); |
128 | ||
166f61b9 TH |
129 | unsigned long highest_memmap_pfn __read_mostly; |
130 | ||
a13ea5b7 HD |
131 | /* |
132 | * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init() | |
133 | */ | |
134 | static int __init init_zero_pfn(void) | |
135 | { | |
136 | zero_pfn = page_to_pfn(ZERO_PAGE(0)); | |
137 | return 0; | |
138 | } | |
139 | core_initcall(init_zero_pfn); | |
a62eaf15 | 140 | |
d559db08 | 141 | |
34e55232 KH |
142 | #if defined(SPLIT_RSS_COUNTING) |
143 | ||
ea48cf78 | 144 | void sync_mm_rss(struct mm_struct *mm) |
34e55232 KH |
145 | { |
146 | int i; | |
147 | ||
148 | for (i = 0; i < NR_MM_COUNTERS; i++) { | |
05af2e10 DR |
149 | if (current->rss_stat.count[i]) { |
150 | add_mm_counter(mm, i, current->rss_stat.count[i]); | |
151 | current->rss_stat.count[i] = 0; | |
34e55232 KH |
152 | } |
153 | } | |
05af2e10 | 154 | current->rss_stat.events = 0; |
34e55232 KH |
155 | } |
156 | ||
157 | static void add_mm_counter_fast(struct mm_struct *mm, int member, int val) | |
158 | { | |
159 | struct task_struct *task = current; | |
160 | ||
161 | if (likely(task->mm == mm)) | |
162 | task->rss_stat.count[member] += val; | |
163 | else | |
164 | add_mm_counter(mm, member, val); | |
165 | } | |
166 | #define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1) | |
167 | #define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1) | |
168 | ||
169 | /* sync counter once per 64 page faults */ | |
170 | #define TASK_RSS_EVENTS_THRESH (64) | |
171 | static void check_sync_rss_stat(struct task_struct *task) | |
172 | { | |
173 | if (unlikely(task != current)) | |
174 | return; | |
175 | if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH)) | |
ea48cf78 | 176 | sync_mm_rss(task->mm); |
34e55232 | 177 | } |
9547d01b | 178 | #else /* SPLIT_RSS_COUNTING */ |
34e55232 KH |
179 | |
180 | #define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member) | |
181 | #define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member) | |
182 | ||
183 | static void check_sync_rss_stat(struct task_struct *task) | |
184 | { | |
185 | } | |
186 | ||
9547d01b PZ |
187 | #endif /* SPLIT_RSS_COUNTING */ |
188 | ||
189 | #ifdef HAVE_GENERIC_MMU_GATHER | |
190 | ||
ca1d6c7d | 191 | static bool tlb_next_batch(struct mmu_gather *tlb) |
9547d01b PZ |
192 | { |
193 | struct mmu_gather_batch *batch; | |
194 | ||
195 | batch = tlb->active; | |
196 | if (batch->next) { | |
197 | tlb->active = batch->next; | |
ca1d6c7d | 198 | return true; |
9547d01b PZ |
199 | } |
200 | ||
53a59fc6 | 201 | if (tlb->batch_count == MAX_GATHER_BATCH_COUNT) |
ca1d6c7d | 202 | return false; |
53a59fc6 | 203 | |
9547d01b PZ |
204 | batch = (void *)__get_free_pages(GFP_NOWAIT | __GFP_NOWARN, 0); |
205 | if (!batch) | |
ca1d6c7d | 206 | return false; |
9547d01b | 207 | |
53a59fc6 | 208 | tlb->batch_count++; |
9547d01b PZ |
209 | batch->next = NULL; |
210 | batch->nr = 0; | |
211 | batch->max = MAX_GATHER_BATCH; | |
212 | ||
213 | tlb->active->next = batch; | |
214 | tlb->active = batch; | |
215 | ||
ca1d6c7d | 216 | return true; |
9547d01b PZ |
217 | } |
218 | ||
56236a59 MK |
219 | void arch_tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, |
220 | unsigned long start, unsigned long end) | |
9547d01b PZ |
221 | { |
222 | tlb->mm = mm; | |
223 | ||
2b047252 LT |
224 | /* Is it from 0 to ~0? */ |
225 | tlb->fullmm = !(start | (end+1)); | |
1de14c3c | 226 | tlb->need_flush_all = 0; |
9547d01b PZ |
227 | tlb->local.next = NULL; |
228 | tlb->local.nr = 0; | |
229 | tlb->local.max = ARRAY_SIZE(tlb->__pages); | |
230 | tlb->active = &tlb->local; | |
53a59fc6 | 231 | tlb->batch_count = 0; |
9547d01b PZ |
232 | |
233 | #ifdef CONFIG_HAVE_RCU_TABLE_FREE | |
234 | tlb->batch = NULL; | |
235 | #endif | |
e77b0852 | 236 | tlb->page_size = 0; |
fb7332a9 WD |
237 | |
238 | __tlb_reset_range(tlb); | |
9547d01b PZ |
239 | } |
240 | ||
1cf35d47 | 241 | static void tlb_flush_mmu_tlbonly(struct mmu_gather *tlb) |
9547d01b | 242 | { |
721c21c1 WD |
243 | if (!tlb->end) |
244 | return; | |
245 | ||
9547d01b | 246 | tlb_flush(tlb); |
34ee645e | 247 | mmu_notifier_invalidate_range(tlb->mm, tlb->start, tlb->end); |
9547d01b PZ |
248 | #ifdef CONFIG_HAVE_RCU_TABLE_FREE |
249 | tlb_table_flush(tlb); | |
34e55232 | 250 | #endif |
fb7332a9 | 251 | __tlb_reset_range(tlb); |
1cf35d47 LT |
252 | } |
253 | ||
254 | static void tlb_flush_mmu_free(struct mmu_gather *tlb) | |
255 | { | |
256 | struct mmu_gather_batch *batch; | |
34e55232 | 257 | |
721c21c1 | 258 | for (batch = &tlb->local; batch && batch->nr; batch = batch->next) { |
9547d01b PZ |
259 | free_pages_and_swap_cache(batch->pages, batch->nr); |
260 | batch->nr = 0; | |
261 | } | |
262 | tlb->active = &tlb->local; | |
263 | } | |
264 | ||
1cf35d47 LT |
265 | void tlb_flush_mmu(struct mmu_gather *tlb) |
266 | { | |
1cf35d47 LT |
267 | tlb_flush_mmu_tlbonly(tlb); |
268 | tlb_flush_mmu_free(tlb); | |
269 | } | |
270 | ||
9547d01b PZ |
271 | /* tlb_finish_mmu |
272 | * Called at the end of the shootdown operation to free up any resources | |
273 | * that were required. | |
274 | */ | |
56236a59 | 275 | void arch_tlb_finish_mmu(struct mmu_gather *tlb, |
99baac21 | 276 | unsigned long start, unsigned long end, bool force) |
9547d01b PZ |
277 | { |
278 | struct mmu_gather_batch *batch, *next; | |
279 | ||
99baac21 MK |
280 | if (force) |
281 | __tlb_adjust_range(tlb, start, end - start); | |
282 | ||
9547d01b PZ |
283 | tlb_flush_mmu(tlb); |
284 | ||
285 | /* keep the page table cache within bounds */ | |
286 | check_pgt_cache(); | |
287 | ||
288 | for (batch = tlb->local.next; batch; batch = next) { | |
289 | next = batch->next; | |
290 | free_pages((unsigned long)batch, 0); | |
291 | } | |
292 | tlb->local.next = NULL; | |
293 | } | |
294 | ||
295 | /* __tlb_remove_page | |
296 | * Must perform the equivalent to __free_pte(pte_get_and_clear(ptep)), while | |
297 | * handling the additional races in SMP caused by other CPUs caching valid | |
298 | * mappings in their TLBs. Returns the number of free page slots left. | |
299 | * When out of page slots we must call tlb_flush_mmu(). | |
e9d55e15 | 300 | *returns true if the caller should flush. |
9547d01b | 301 | */ |
e77b0852 | 302 | bool __tlb_remove_page_size(struct mmu_gather *tlb, struct page *page, int page_size) |
9547d01b PZ |
303 | { |
304 | struct mmu_gather_batch *batch; | |
305 | ||
fb7332a9 | 306 | VM_BUG_ON(!tlb->end); |
692a68c1 | 307 | VM_WARN_ON(tlb->page_size != page_size); |
e77b0852 | 308 | |
9547d01b | 309 | batch = tlb->active; |
692a68c1 AK |
310 | /* |
311 | * Add the page and check if we are full. If so | |
312 | * force a flush. | |
313 | */ | |
314 | batch->pages[batch->nr++] = page; | |
9547d01b PZ |
315 | if (batch->nr == batch->max) { |
316 | if (!tlb_next_batch(tlb)) | |
e9d55e15 | 317 | return true; |
0b43c3aa | 318 | batch = tlb->active; |
9547d01b | 319 | } |
309381fe | 320 | VM_BUG_ON_PAGE(batch->nr > batch->max, page); |
9547d01b | 321 | |
e9d55e15 | 322 | return false; |
9547d01b PZ |
323 | } |
324 | ||
325 | #endif /* HAVE_GENERIC_MMU_GATHER */ | |
326 | ||
26723911 PZ |
327 | #ifdef CONFIG_HAVE_RCU_TABLE_FREE |
328 | ||
329 | /* | |
330 | * See the comment near struct mmu_table_batch. | |
331 | */ | |
332 | ||
333 | static void tlb_remove_table_smp_sync(void *arg) | |
334 | { | |
335 | /* Simply deliver the interrupt */ | |
336 | } | |
337 | ||
338 | static void tlb_remove_table_one(void *table) | |
339 | { | |
340 | /* | |
341 | * This isn't an RCU grace period and hence the page-tables cannot be | |
342 | * assumed to be actually RCU-freed. | |
343 | * | |
344 | * It is however sufficient for software page-table walkers that rely on | |
345 | * IRQ disabling. See the comment near struct mmu_table_batch. | |
346 | */ | |
347 | smp_call_function(tlb_remove_table_smp_sync, NULL, 1); | |
348 | __tlb_remove_table(table); | |
349 | } | |
350 | ||
351 | static void tlb_remove_table_rcu(struct rcu_head *head) | |
352 | { | |
353 | struct mmu_table_batch *batch; | |
354 | int i; | |
355 | ||
356 | batch = container_of(head, struct mmu_table_batch, rcu); | |
357 | ||
358 | for (i = 0; i < batch->nr; i++) | |
359 | __tlb_remove_table(batch->tables[i]); | |
360 | ||
361 | free_page((unsigned long)batch); | |
362 | } | |
363 | ||
364 | void tlb_table_flush(struct mmu_gather *tlb) | |
365 | { | |
366 | struct mmu_table_batch **batch = &tlb->batch; | |
367 | ||
368 | if (*batch) { | |
369 | call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu); | |
370 | *batch = NULL; | |
371 | } | |
372 | } | |
373 | ||
374 | void tlb_remove_table(struct mmu_gather *tlb, void *table) | |
375 | { | |
376 | struct mmu_table_batch **batch = &tlb->batch; | |
377 | ||
26723911 PZ |
378 | /* |
379 | * When there's less then two users of this mm there cannot be a | |
380 | * concurrent page-table walk. | |
381 | */ | |
382 | if (atomic_read(&tlb->mm->mm_users) < 2) { | |
383 | __tlb_remove_table(table); | |
384 | return; | |
385 | } | |
386 | ||
387 | if (*batch == NULL) { | |
388 | *batch = (struct mmu_table_batch *)__get_free_page(GFP_NOWAIT | __GFP_NOWARN); | |
389 | if (*batch == NULL) { | |
390 | tlb_remove_table_one(table); | |
391 | return; | |
392 | } | |
393 | (*batch)->nr = 0; | |
394 | } | |
395 | (*batch)->tables[(*batch)->nr++] = table; | |
396 | if ((*batch)->nr == MAX_TABLE_BATCH) | |
397 | tlb_table_flush(tlb); | |
398 | } | |
399 | ||
9547d01b | 400 | #endif /* CONFIG_HAVE_RCU_TABLE_FREE */ |
26723911 | 401 | |
56236a59 MK |
402 | /* tlb_gather_mmu |
403 | * Called to initialize an (on-stack) mmu_gather structure for page-table | |
404 | * tear-down from @mm. The @fullmm argument is used when @mm is without | |
405 | * users and we're going to destroy the full address space (exit/execve). | |
406 | */ | |
407 | void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, | |
408 | unsigned long start, unsigned long end) | |
409 | { | |
410 | arch_tlb_gather_mmu(tlb, mm, start, end); | |
99baac21 | 411 | inc_tlb_flush_pending(tlb->mm); |
56236a59 MK |
412 | } |
413 | ||
414 | void tlb_finish_mmu(struct mmu_gather *tlb, | |
415 | unsigned long start, unsigned long end) | |
416 | { | |
99baac21 MK |
417 | /* |
418 | * If there are parallel threads are doing PTE changes on same range | |
419 | * under non-exclusive lock(e.g., mmap_sem read-side) but defer TLB | |
420 | * flush by batching, a thread has stable TLB entry can fail to flush | |
421 | * the TLB by observing pte_none|!pte_dirty, for example so flush TLB | |
422 | * forcefully if we detect parallel PTE batching threads. | |
423 | */ | |
424 | bool force = mm_tlb_flush_nested(tlb->mm); | |
425 | ||
426 | arch_tlb_finish_mmu(tlb, start, end, force); | |
427 | dec_tlb_flush_pending(tlb->mm); | |
56236a59 MK |
428 | } |
429 | ||
1da177e4 LT |
430 | /* |
431 | * Note: this doesn't free the actual pages themselves. That | |
432 | * has been handled earlier when unmapping all the memory regions. | |
433 | */ | |
9e1b32ca BH |
434 | static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd, |
435 | unsigned long addr) | |
1da177e4 | 436 | { |
2f569afd | 437 | pgtable_t token = pmd_pgtable(*pmd); |
e0da382c | 438 | pmd_clear(pmd); |
9e1b32ca | 439 | pte_free_tlb(tlb, token, addr); |
e1f56c89 | 440 | atomic_long_dec(&tlb->mm->nr_ptes); |
1da177e4 LT |
441 | } |
442 | ||
e0da382c HD |
443 | static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud, |
444 | unsigned long addr, unsigned long end, | |
445 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
446 | { |
447 | pmd_t *pmd; | |
448 | unsigned long next; | |
e0da382c | 449 | unsigned long start; |
1da177e4 | 450 | |
e0da382c | 451 | start = addr; |
1da177e4 | 452 | pmd = pmd_offset(pud, addr); |
1da177e4 LT |
453 | do { |
454 | next = pmd_addr_end(addr, end); | |
455 | if (pmd_none_or_clear_bad(pmd)) | |
456 | continue; | |
9e1b32ca | 457 | free_pte_range(tlb, pmd, addr); |
1da177e4 LT |
458 | } while (pmd++, addr = next, addr != end); |
459 | ||
e0da382c HD |
460 | start &= PUD_MASK; |
461 | if (start < floor) | |
462 | return; | |
463 | if (ceiling) { | |
464 | ceiling &= PUD_MASK; | |
465 | if (!ceiling) | |
466 | return; | |
1da177e4 | 467 | } |
e0da382c HD |
468 | if (end - 1 > ceiling - 1) |
469 | return; | |
470 | ||
471 | pmd = pmd_offset(pud, start); | |
472 | pud_clear(pud); | |
9e1b32ca | 473 | pmd_free_tlb(tlb, pmd, start); |
dc6c9a35 | 474 | mm_dec_nr_pmds(tlb->mm); |
1da177e4 LT |
475 | } |
476 | ||
c2febafc | 477 | static inline void free_pud_range(struct mmu_gather *tlb, p4d_t *p4d, |
e0da382c HD |
478 | unsigned long addr, unsigned long end, |
479 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
480 | { |
481 | pud_t *pud; | |
482 | unsigned long next; | |
e0da382c | 483 | unsigned long start; |
1da177e4 | 484 | |
e0da382c | 485 | start = addr; |
c2febafc | 486 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
487 | do { |
488 | next = pud_addr_end(addr, end); | |
489 | if (pud_none_or_clear_bad(pud)) | |
490 | continue; | |
e0da382c | 491 | free_pmd_range(tlb, pud, addr, next, floor, ceiling); |
1da177e4 LT |
492 | } while (pud++, addr = next, addr != end); |
493 | ||
c2febafc KS |
494 | start &= P4D_MASK; |
495 | if (start < floor) | |
496 | return; | |
497 | if (ceiling) { | |
498 | ceiling &= P4D_MASK; | |
499 | if (!ceiling) | |
500 | return; | |
501 | } | |
502 | if (end - 1 > ceiling - 1) | |
503 | return; | |
504 | ||
505 | pud = pud_offset(p4d, start); | |
506 | p4d_clear(p4d); | |
507 | pud_free_tlb(tlb, pud, start); | |
508 | } | |
509 | ||
510 | static inline void free_p4d_range(struct mmu_gather *tlb, pgd_t *pgd, | |
511 | unsigned long addr, unsigned long end, | |
512 | unsigned long floor, unsigned long ceiling) | |
513 | { | |
514 | p4d_t *p4d; | |
515 | unsigned long next; | |
516 | unsigned long start; | |
517 | ||
518 | start = addr; | |
519 | p4d = p4d_offset(pgd, addr); | |
520 | do { | |
521 | next = p4d_addr_end(addr, end); | |
522 | if (p4d_none_or_clear_bad(p4d)) | |
523 | continue; | |
524 | free_pud_range(tlb, p4d, addr, next, floor, ceiling); | |
525 | } while (p4d++, addr = next, addr != end); | |
526 | ||
e0da382c HD |
527 | start &= PGDIR_MASK; |
528 | if (start < floor) | |
529 | return; | |
530 | if (ceiling) { | |
531 | ceiling &= PGDIR_MASK; | |
532 | if (!ceiling) | |
533 | return; | |
1da177e4 | 534 | } |
e0da382c HD |
535 | if (end - 1 > ceiling - 1) |
536 | return; | |
537 | ||
c2febafc | 538 | p4d = p4d_offset(pgd, start); |
e0da382c | 539 | pgd_clear(pgd); |
c2febafc | 540 | p4d_free_tlb(tlb, p4d, start); |
1da177e4 LT |
541 | } |
542 | ||
543 | /* | |
e0da382c | 544 | * This function frees user-level page tables of a process. |
1da177e4 | 545 | */ |
42b77728 | 546 | void free_pgd_range(struct mmu_gather *tlb, |
e0da382c HD |
547 | unsigned long addr, unsigned long end, |
548 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
549 | { |
550 | pgd_t *pgd; | |
551 | unsigned long next; | |
e0da382c HD |
552 | |
553 | /* | |
554 | * The next few lines have given us lots of grief... | |
555 | * | |
556 | * Why are we testing PMD* at this top level? Because often | |
557 | * there will be no work to do at all, and we'd prefer not to | |
558 | * go all the way down to the bottom just to discover that. | |
559 | * | |
560 | * Why all these "- 1"s? Because 0 represents both the bottom | |
561 | * of the address space and the top of it (using -1 for the | |
562 | * top wouldn't help much: the masks would do the wrong thing). | |
563 | * The rule is that addr 0 and floor 0 refer to the bottom of | |
564 | * the address space, but end 0 and ceiling 0 refer to the top | |
565 | * Comparisons need to use "end - 1" and "ceiling - 1" (though | |
566 | * that end 0 case should be mythical). | |
567 | * | |
568 | * Wherever addr is brought up or ceiling brought down, we must | |
569 | * be careful to reject "the opposite 0" before it confuses the | |
570 | * subsequent tests. But what about where end is brought down | |
571 | * by PMD_SIZE below? no, end can't go down to 0 there. | |
572 | * | |
573 | * Whereas we round start (addr) and ceiling down, by different | |
574 | * masks at different levels, in order to test whether a table | |
575 | * now has no other vmas using it, so can be freed, we don't | |
576 | * bother to round floor or end up - the tests don't need that. | |
577 | */ | |
1da177e4 | 578 | |
e0da382c HD |
579 | addr &= PMD_MASK; |
580 | if (addr < floor) { | |
581 | addr += PMD_SIZE; | |
582 | if (!addr) | |
583 | return; | |
584 | } | |
585 | if (ceiling) { | |
586 | ceiling &= PMD_MASK; | |
587 | if (!ceiling) | |
588 | return; | |
589 | } | |
590 | if (end - 1 > ceiling - 1) | |
591 | end -= PMD_SIZE; | |
592 | if (addr > end - 1) | |
593 | return; | |
07e32661 AK |
594 | /* |
595 | * We add page table cache pages with PAGE_SIZE, | |
596 | * (see pte_free_tlb()), flush the tlb if we need | |
597 | */ | |
598 | tlb_remove_check_page_size_change(tlb, PAGE_SIZE); | |
42b77728 | 599 | pgd = pgd_offset(tlb->mm, addr); |
1da177e4 LT |
600 | do { |
601 | next = pgd_addr_end(addr, end); | |
602 | if (pgd_none_or_clear_bad(pgd)) | |
603 | continue; | |
c2febafc | 604 | free_p4d_range(tlb, pgd, addr, next, floor, ceiling); |
1da177e4 | 605 | } while (pgd++, addr = next, addr != end); |
e0da382c HD |
606 | } |
607 | ||
42b77728 | 608 | void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma, |
3bf5ee95 | 609 | unsigned long floor, unsigned long ceiling) |
e0da382c HD |
610 | { |
611 | while (vma) { | |
612 | struct vm_area_struct *next = vma->vm_next; | |
613 | unsigned long addr = vma->vm_start; | |
614 | ||
8f4f8c16 | 615 | /* |
25d9e2d1 NP |
616 | * Hide vma from rmap and truncate_pagecache before freeing |
617 | * pgtables | |
8f4f8c16 | 618 | */ |
5beb4930 | 619 | unlink_anon_vmas(vma); |
8f4f8c16 HD |
620 | unlink_file_vma(vma); |
621 | ||
9da61aef | 622 | if (is_vm_hugetlb_page(vma)) { |
3bf5ee95 | 623 | hugetlb_free_pgd_range(tlb, addr, vma->vm_end, |
166f61b9 | 624 | floor, next ? next->vm_start : ceiling); |
3bf5ee95 HD |
625 | } else { |
626 | /* | |
627 | * Optimization: gather nearby vmas into one call down | |
628 | */ | |
629 | while (next && next->vm_start <= vma->vm_end + PMD_SIZE | |
4866920b | 630 | && !is_vm_hugetlb_page(next)) { |
3bf5ee95 HD |
631 | vma = next; |
632 | next = vma->vm_next; | |
5beb4930 | 633 | unlink_anon_vmas(vma); |
8f4f8c16 | 634 | unlink_file_vma(vma); |
3bf5ee95 HD |
635 | } |
636 | free_pgd_range(tlb, addr, vma->vm_end, | |
166f61b9 | 637 | floor, next ? next->vm_start : ceiling); |
3bf5ee95 | 638 | } |
e0da382c HD |
639 | vma = next; |
640 | } | |
1da177e4 LT |
641 | } |
642 | ||
3ed3a4f0 | 643 | int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address) |
1da177e4 | 644 | { |
c4088ebd | 645 | spinlock_t *ptl; |
2f569afd | 646 | pgtable_t new = pte_alloc_one(mm, address); |
1bb3630e HD |
647 | if (!new) |
648 | return -ENOMEM; | |
649 | ||
362a61ad NP |
650 | /* |
651 | * Ensure all pte setup (eg. pte page lock and page clearing) are | |
652 | * visible before the pte is made visible to other CPUs by being | |
653 | * put into page tables. | |
654 | * | |
655 | * The other side of the story is the pointer chasing in the page | |
656 | * table walking code (when walking the page table without locking; | |
657 | * ie. most of the time). Fortunately, these data accesses consist | |
658 | * of a chain of data-dependent loads, meaning most CPUs (alpha | |
659 | * being the notable exception) will already guarantee loads are | |
660 | * seen in-order. See the alpha page table accessors for the | |
661 | * smp_read_barrier_depends() barriers in page table walking code. | |
662 | */ | |
663 | smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */ | |
664 | ||
c4088ebd | 665 | ptl = pmd_lock(mm, pmd); |
8ac1f832 | 666 | if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ |
e1f56c89 | 667 | atomic_long_inc(&mm->nr_ptes); |
1da177e4 | 668 | pmd_populate(mm, pmd, new); |
2f569afd | 669 | new = NULL; |
4b471e88 | 670 | } |
c4088ebd | 671 | spin_unlock(ptl); |
2f569afd MS |
672 | if (new) |
673 | pte_free(mm, new); | |
1bb3630e | 674 | return 0; |
1da177e4 LT |
675 | } |
676 | ||
1bb3630e | 677 | int __pte_alloc_kernel(pmd_t *pmd, unsigned long address) |
1da177e4 | 678 | { |
1bb3630e HD |
679 | pte_t *new = pte_alloc_one_kernel(&init_mm, address); |
680 | if (!new) | |
681 | return -ENOMEM; | |
682 | ||
362a61ad NP |
683 | smp_wmb(); /* See comment in __pte_alloc */ |
684 | ||
1bb3630e | 685 | spin_lock(&init_mm.page_table_lock); |
8ac1f832 | 686 | if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ |
1bb3630e | 687 | pmd_populate_kernel(&init_mm, pmd, new); |
2f569afd | 688 | new = NULL; |
4b471e88 | 689 | } |
1bb3630e | 690 | spin_unlock(&init_mm.page_table_lock); |
2f569afd MS |
691 | if (new) |
692 | pte_free_kernel(&init_mm, new); | |
1bb3630e | 693 | return 0; |
1da177e4 LT |
694 | } |
695 | ||
d559db08 KH |
696 | static inline void init_rss_vec(int *rss) |
697 | { | |
698 | memset(rss, 0, sizeof(int) * NR_MM_COUNTERS); | |
699 | } | |
700 | ||
701 | static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss) | |
ae859762 | 702 | { |
d559db08 KH |
703 | int i; |
704 | ||
34e55232 | 705 | if (current->mm == mm) |
05af2e10 | 706 | sync_mm_rss(mm); |
d559db08 KH |
707 | for (i = 0; i < NR_MM_COUNTERS; i++) |
708 | if (rss[i]) | |
709 | add_mm_counter(mm, i, rss[i]); | |
ae859762 HD |
710 | } |
711 | ||
b5810039 | 712 | /* |
6aab341e LT |
713 | * This function is called to print an error when a bad pte |
714 | * is found. For example, we might have a PFN-mapped pte in | |
715 | * a region that doesn't allow it. | |
b5810039 NP |
716 | * |
717 | * The calling function must still handle the error. | |
718 | */ | |
3dc14741 HD |
719 | static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr, |
720 | pte_t pte, struct page *page) | |
b5810039 | 721 | { |
3dc14741 | 722 | pgd_t *pgd = pgd_offset(vma->vm_mm, addr); |
c2febafc KS |
723 | p4d_t *p4d = p4d_offset(pgd, addr); |
724 | pud_t *pud = pud_offset(p4d, addr); | |
3dc14741 HD |
725 | pmd_t *pmd = pmd_offset(pud, addr); |
726 | struct address_space *mapping; | |
727 | pgoff_t index; | |
d936cf9b HD |
728 | static unsigned long resume; |
729 | static unsigned long nr_shown; | |
730 | static unsigned long nr_unshown; | |
731 | ||
732 | /* | |
733 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
734 | * or allow a steady drip of one report per second. | |
735 | */ | |
736 | if (nr_shown == 60) { | |
737 | if (time_before(jiffies, resume)) { | |
738 | nr_unshown++; | |
739 | return; | |
740 | } | |
741 | if (nr_unshown) { | |
1170532b JP |
742 | pr_alert("BUG: Bad page map: %lu messages suppressed\n", |
743 | nr_unshown); | |
d936cf9b HD |
744 | nr_unshown = 0; |
745 | } | |
746 | nr_shown = 0; | |
747 | } | |
748 | if (nr_shown++ == 0) | |
749 | resume = jiffies + 60 * HZ; | |
3dc14741 HD |
750 | |
751 | mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL; | |
752 | index = linear_page_index(vma, addr); | |
753 | ||
1170532b JP |
754 | pr_alert("BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n", |
755 | current->comm, | |
756 | (long long)pte_val(pte), (long long)pmd_val(*pmd)); | |
718a3821 | 757 | if (page) |
f0b791a3 | 758 | dump_page(page, "bad pte"); |
1170532b JP |
759 | pr_alert("addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n", |
760 | (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index); | |
3dc14741 HD |
761 | /* |
762 | * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y | |
763 | */ | |
2682582a KK |
764 | pr_alert("file:%pD fault:%pf mmap:%pf readpage:%pf\n", |
765 | vma->vm_file, | |
766 | vma->vm_ops ? vma->vm_ops->fault : NULL, | |
767 | vma->vm_file ? vma->vm_file->f_op->mmap : NULL, | |
768 | mapping ? mapping->a_ops->readpage : NULL); | |
b5810039 | 769 | dump_stack(); |
373d4d09 | 770 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
b5810039 NP |
771 | } |
772 | ||
ee498ed7 | 773 | /* |
7e675137 | 774 | * vm_normal_page -- This function gets the "struct page" associated with a pte. |
6aab341e | 775 | * |
7e675137 NP |
776 | * "Special" mappings do not wish to be associated with a "struct page" (either |
777 | * it doesn't exist, or it exists but they don't want to touch it). In this | |
778 | * case, NULL is returned here. "Normal" mappings do have a struct page. | |
b379d790 | 779 | * |
7e675137 NP |
780 | * There are 2 broad cases. Firstly, an architecture may define a pte_special() |
781 | * pte bit, in which case this function is trivial. Secondly, an architecture | |
782 | * may not have a spare pte bit, which requires a more complicated scheme, | |
783 | * described below. | |
784 | * | |
785 | * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a | |
786 | * special mapping (even if there are underlying and valid "struct pages"). | |
787 | * COWed pages of a VM_PFNMAP are always normal. | |
6aab341e | 788 | * |
b379d790 JH |
789 | * The way we recognize COWed pages within VM_PFNMAP mappings is through the |
790 | * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit | |
7e675137 NP |
791 | * set, and the vm_pgoff will point to the first PFN mapped: thus every special |
792 | * mapping will always honor the rule | |
6aab341e LT |
793 | * |
794 | * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) | |
795 | * | |
7e675137 NP |
796 | * And for normal mappings this is false. |
797 | * | |
798 | * This restricts such mappings to be a linear translation from virtual address | |
799 | * to pfn. To get around this restriction, we allow arbitrary mappings so long | |
800 | * as the vma is not a COW mapping; in that case, we know that all ptes are | |
801 | * special (because none can have been COWed). | |
b379d790 | 802 | * |
b379d790 | 803 | * |
7e675137 | 804 | * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. |
b379d790 JH |
805 | * |
806 | * VM_MIXEDMAP mappings can likewise contain memory with or without "struct | |
807 | * page" backing, however the difference is that _all_ pages with a struct | |
808 | * page (that is, those where pfn_valid is true) are refcounted and considered | |
809 | * normal pages by the VM. The disadvantage is that pages are refcounted | |
810 | * (which can be slower and simply not an option for some PFNMAP users). The | |
811 | * advantage is that we don't have to follow the strict linearity rule of | |
812 | * PFNMAP mappings in order to support COWable mappings. | |
813 | * | |
ee498ed7 | 814 | */ |
7e675137 NP |
815 | #ifdef __HAVE_ARCH_PTE_SPECIAL |
816 | # define HAVE_PTE_SPECIAL 1 | |
817 | #else | |
818 | # define HAVE_PTE_SPECIAL 0 | |
819 | #endif | |
820 | struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, | |
821 | pte_t pte) | |
ee498ed7 | 822 | { |
22b31eec | 823 | unsigned long pfn = pte_pfn(pte); |
7e675137 NP |
824 | |
825 | if (HAVE_PTE_SPECIAL) { | |
b38af472 | 826 | if (likely(!pte_special(pte))) |
22b31eec | 827 | goto check_pfn; |
667a0a06 DV |
828 | if (vma->vm_ops && vma->vm_ops->find_special_page) |
829 | return vma->vm_ops->find_special_page(vma, addr); | |
a13ea5b7 HD |
830 | if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) |
831 | return NULL; | |
62eede62 | 832 | if (!is_zero_pfn(pfn)) |
22b31eec | 833 | print_bad_pte(vma, addr, pte, NULL); |
7e675137 NP |
834 | return NULL; |
835 | } | |
836 | ||
837 | /* !HAVE_PTE_SPECIAL case follows: */ | |
838 | ||
b379d790 JH |
839 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { |
840 | if (vma->vm_flags & VM_MIXEDMAP) { | |
841 | if (!pfn_valid(pfn)) | |
842 | return NULL; | |
843 | goto out; | |
844 | } else { | |
7e675137 NP |
845 | unsigned long off; |
846 | off = (addr - vma->vm_start) >> PAGE_SHIFT; | |
b379d790 JH |
847 | if (pfn == vma->vm_pgoff + off) |
848 | return NULL; | |
849 | if (!is_cow_mapping(vma->vm_flags)) | |
850 | return NULL; | |
851 | } | |
6aab341e LT |
852 | } |
853 | ||
b38af472 HD |
854 | if (is_zero_pfn(pfn)) |
855 | return NULL; | |
22b31eec HD |
856 | check_pfn: |
857 | if (unlikely(pfn > highest_memmap_pfn)) { | |
858 | print_bad_pte(vma, addr, pte, NULL); | |
859 | return NULL; | |
860 | } | |
6aab341e LT |
861 | |
862 | /* | |
7e675137 | 863 | * NOTE! We still have PageReserved() pages in the page tables. |
7e675137 | 864 | * eg. VDSO mappings can cause them to exist. |
6aab341e | 865 | */ |
b379d790 | 866 | out: |
6aab341e | 867 | return pfn_to_page(pfn); |
ee498ed7 HD |
868 | } |
869 | ||
28093f9f GS |
870 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
871 | struct page *vm_normal_page_pmd(struct vm_area_struct *vma, unsigned long addr, | |
872 | pmd_t pmd) | |
873 | { | |
874 | unsigned long pfn = pmd_pfn(pmd); | |
875 | ||
876 | /* | |
877 | * There is no pmd_special() but there may be special pmds, e.g. | |
878 | * in a direct-access (dax) mapping, so let's just replicate the | |
879 | * !HAVE_PTE_SPECIAL case from vm_normal_page() here. | |
880 | */ | |
881 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { | |
882 | if (vma->vm_flags & VM_MIXEDMAP) { | |
883 | if (!pfn_valid(pfn)) | |
884 | return NULL; | |
885 | goto out; | |
886 | } else { | |
887 | unsigned long off; | |
888 | off = (addr - vma->vm_start) >> PAGE_SHIFT; | |
889 | if (pfn == vma->vm_pgoff + off) | |
890 | return NULL; | |
891 | if (!is_cow_mapping(vma->vm_flags)) | |
892 | return NULL; | |
893 | } | |
894 | } | |
895 | ||
896 | if (is_zero_pfn(pfn)) | |
897 | return NULL; | |
898 | if (unlikely(pfn > highest_memmap_pfn)) | |
899 | return NULL; | |
900 | ||
901 | /* | |
902 | * NOTE! We still have PageReserved() pages in the page tables. | |
903 | * eg. VDSO mappings can cause them to exist. | |
904 | */ | |
905 | out: | |
906 | return pfn_to_page(pfn); | |
907 | } | |
908 | #endif | |
909 | ||
1da177e4 LT |
910 | /* |
911 | * copy one vm_area from one task to the other. Assumes the page tables | |
912 | * already present in the new task to be cleared in the whole range | |
913 | * covered by this vma. | |
1da177e4 LT |
914 | */ |
915 | ||
570a335b | 916 | static inline unsigned long |
1da177e4 | 917 | copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
b5810039 | 918 | pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma, |
8c103762 | 919 | unsigned long addr, int *rss) |
1da177e4 | 920 | { |
b5810039 | 921 | unsigned long vm_flags = vma->vm_flags; |
1da177e4 LT |
922 | pte_t pte = *src_pte; |
923 | struct page *page; | |
1da177e4 LT |
924 | |
925 | /* pte contains position in swap or file, so copy. */ | |
926 | if (unlikely(!pte_present(pte))) { | |
0661a336 KS |
927 | swp_entry_t entry = pte_to_swp_entry(pte); |
928 | ||
929 | if (likely(!non_swap_entry(entry))) { | |
930 | if (swap_duplicate(entry) < 0) | |
931 | return entry.val; | |
932 | ||
933 | /* make sure dst_mm is on swapoff's mmlist. */ | |
934 | if (unlikely(list_empty(&dst_mm->mmlist))) { | |
935 | spin_lock(&mmlist_lock); | |
936 | if (list_empty(&dst_mm->mmlist)) | |
937 | list_add(&dst_mm->mmlist, | |
938 | &src_mm->mmlist); | |
939 | spin_unlock(&mmlist_lock); | |
940 | } | |
941 | rss[MM_SWAPENTS]++; | |
942 | } else if (is_migration_entry(entry)) { | |
943 | page = migration_entry_to_page(entry); | |
944 | ||
eca56ff9 | 945 | rss[mm_counter(page)]++; |
0661a336 KS |
946 | |
947 | if (is_write_migration_entry(entry) && | |
948 | is_cow_mapping(vm_flags)) { | |
949 | /* | |
950 | * COW mappings require pages in both | |
951 | * parent and child to be set to read. | |
952 | */ | |
953 | make_migration_entry_read(&entry); | |
954 | pte = swp_entry_to_pte(entry); | |
955 | if (pte_swp_soft_dirty(*src_pte)) | |
956 | pte = pte_swp_mksoft_dirty(pte); | |
957 | set_pte_at(src_mm, addr, src_pte, pte); | |
0697212a | 958 | } |
1da177e4 | 959 | } |
ae859762 | 960 | goto out_set_pte; |
1da177e4 LT |
961 | } |
962 | ||
1da177e4 LT |
963 | /* |
964 | * If it's a COW mapping, write protect it both | |
965 | * in the parent and the child | |
966 | */ | |
67121172 | 967 | if (is_cow_mapping(vm_flags)) { |
1da177e4 | 968 | ptep_set_wrprotect(src_mm, addr, src_pte); |
3dc90795 | 969 | pte = pte_wrprotect(pte); |
1da177e4 LT |
970 | } |
971 | ||
972 | /* | |
973 | * If it's a shared mapping, mark it clean in | |
974 | * the child | |
975 | */ | |
976 | if (vm_flags & VM_SHARED) | |
977 | pte = pte_mkclean(pte); | |
978 | pte = pte_mkold(pte); | |
6aab341e LT |
979 | |
980 | page = vm_normal_page(vma, addr, pte); | |
981 | if (page) { | |
982 | get_page(page); | |
53f9263b | 983 | page_dup_rmap(page, false); |
eca56ff9 | 984 | rss[mm_counter(page)]++; |
6aab341e | 985 | } |
ae859762 HD |
986 | |
987 | out_set_pte: | |
988 | set_pte_at(dst_mm, addr, dst_pte, pte); | |
570a335b | 989 | return 0; |
1da177e4 LT |
990 | } |
991 | ||
21bda264 | 992 | static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
71e3aac0 AA |
993 | pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma, |
994 | unsigned long addr, unsigned long end) | |
1da177e4 | 995 | { |
c36987e2 | 996 | pte_t *orig_src_pte, *orig_dst_pte; |
1da177e4 | 997 | pte_t *src_pte, *dst_pte; |
c74df32c | 998 | spinlock_t *src_ptl, *dst_ptl; |
e040f218 | 999 | int progress = 0; |
d559db08 | 1000 | int rss[NR_MM_COUNTERS]; |
570a335b | 1001 | swp_entry_t entry = (swp_entry_t){0}; |
1da177e4 LT |
1002 | |
1003 | again: | |
d559db08 KH |
1004 | init_rss_vec(rss); |
1005 | ||
c74df32c | 1006 | dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); |
1da177e4 LT |
1007 | if (!dst_pte) |
1008 | return -ENOMEM; | |
ece0e2b6 | 1009 | src_pte = pte_offset_map(src_pmd, addr); |
4c21e2f2 | 1010 | src_ptl = pte_lockptr(src_mm, src_pmd); |
f20dc5f7 | 1011 | spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); |
c36987e2 DN |
1012 | orig_src_pte = src_pte; |
1013 | orig_dst_pte = dst_pte; | |
6606c3e0 | 1014 | arch_enter_lazy_mmu_mode(); |
1da177e4 | 1015 | |
1da177e4 LT |
1016 | do { |
1017 | /* | |
1018 | * We are holding two locks at this point - either of them | |
1019 | * could generate latencies in another task on another CPU. | |
1020 | */ | |
e040f218 HD |
1021 | if (progress >= 32) { |
1022 | progress = 0; | |
1023 | if (need_resched() || | |
95c354fe | 1024 | spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) |
e040f218 HD |
1025 | break; |
1026 | } | |
1da177e4 LT |
1027 | if (pte_none(*src_pte)) { |
1028 | progress++; | |
1029 | continue; | |
1030 | } | |
570a335b HD |
1031 | entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, |
1032 | vma, addr, rss); | |
1033 | if (entry.val) | |
1034 | break; | |
1da177e4 LT |
1035 | progress += 8; |
1036 | } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); | |
1da177e4 | 1037 | |
6606c3e0 | 1038 | arch_leave_lazy_mmu_mode(); |
c74df32c | 1039 | spin_unlock(src_ptl); |
ece0e2b6 | 1040 | pte_unmap(orig_src_pte); |
d559db08 | 1041 | add_mm_rss_vec(dst_mm, rss); |
c36987e2 | 1042 | pte_unmap_unlock(orig_dst_pte, dst_ptl); |
c74df32c | 1043 | cond_resched(); |
570a335b HD |
1044 | |
1045 | if (entry.val) { | |
1046 | if (add_swap_count_continuation(entry, GFP_KERNEL) < 0) | |
1047 | return -ENOMEM; | |
1048 | progress = 0; | |
1049 | } | |
1da177e4 LT |
1050 | if (addr != end) |
1051 | goto again; | |
1052 | return 0; | |
1053 | } | |
1054 | ||
1055 | static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
1056 | pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma, | |
1057 | unsigned long addr, unsigned long end) | |
1058 | { | |
1059 | pmd_t *src_pmd, *dst_pmd; | |
1060 | unsigned long next; | |
1061 | ||
1062 | dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); | |
1063 | if (!dst_pmd) | |
1064 | return -ENOMEM; | |
1065 | src_pmd = pmd_offset(src_pud, addr); | |
1066 | do { | |
1067 | next = pmd_addr_end(addr, end); | |
5c7fb56e | 1068 | if (pmd_trans_huge(*src_pmd) || pmd_devmap(*src_pmd)) { |
71e3aac0 | 1069 | int err; |
a00cc7d9 | 1070 | VM_BUG_ON_VMA(next-addr != HPAGE_PMD_SIZE, vma); |
71e3aac0 AA |
1071 | err = copy_huge_pmd(dst_mm, src_mm, |
1072 | dst_pmd, src_pmd, addr, vma); | |
1073 | if (err == -ENOMEM) | |
1074 | return -ENOMEM; | |
1075 | if (!err) | |
1076 | continue; | |
1077 | /* fall through */ | |
1078 | } | |
1da177e4 LT |
1079 | if (pmd_none_or_clear_bad(src_pmd)) |
1080 | continue; | |
1081 | if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd, | |
1082 | vma, addr, next)) | |
1083 | return -ENOMEM; | |
1084 | } while (dst_pmd++, src_pmd++, addr = next, addr != end); | |
1085 | return 0; | |
1086 | } | |
1087 | ||
1088 | static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
c2febafc | 1089 | p4d_t *dst_p4d, p4d_t *src_p4d, struct vm_area_struct *vma, |
1da177e4 LT |
1090 | unsigned long addr, unsigned long end) |
1091 | { | |
1092 | pud_t *src_pud, *dst_pud; | |
1093 | unsigned long next; | |
1094 | ||
c2febafc | 1095 | dst_pud = pud_alloc(dst_mm, dst_p4d, addr); |
1da177e4 LT |
1096 | if (!dst_pud) |
1097 | return -ENOMEM; | |
c2febafc | 1098 | src_pud = pud_offset(src_p4d, addr); |
1da177e4 LT |
1099 | do { |
1100 | next = pud_addr_end(addr, end); | |
a00cc7d9 MW |
1101 | if (pud_trans_huge(*src_pud) || pud_devmap(*src_pud)) { |
1102 | int err; | |
1103 | ||
1104 | VM_BUG_ON_VMA(next-addr != HPAGE_PUD_SIZE, vma); | |
1105 | err = copy_huge_pud(dst_mm, src_mm, | |
1106 | dst_pud, src_pud, addr, vma); | |
1107 | if (err == -ENOMEM) | |
1108 | return -ENOMEM; | |
1109 | if (!err) | |
1110 | continue; | |
1111 | /* fall through */ | |
1112 | } | |
1da177e4 LT |
1113 | if (pud_none_or_clear_bad(src_pud)) |
1114 | continue; | |
1115 | if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud, | |
1116 | vma, addr, next)) | |
1117 | return -ENOMEM; | |
1118 | } while (dst_pud++, src_pud++, addr = next, addr != end); | |
1119 | return 0; | |
1120 | } | |
1121 | ||
c2febafc KS |
1122 | static inline int copy_p4d_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
1123 | pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma, | |
1124 | unsigned long addr, unsigned long end) | |
1125 | { | |
1126 | p4d_t *src_p4d, *dst_p4d; | |
1127 | unsigned long next; | |
1128 | ||
1129 | dst_p4d = p4d_alloc(dst_mm, dst_pgd, addr); | |
1130 | if (!dst_p4d) | |
1131 | return -ENOMEM; | |
1132 | src_p4d = p4d_offset(src_pgd, addr); | |
1133 | do { | |
1134 | next = p4d_addr_end(addr, end); | |
1135 | if (p4d_none_or_clear_bad(src_p4d)) | |
1136 | continue; | |
1137 | if (copy_pud_range(dst_mm, src_mm, dst_p4d, src_p4d, | |
1138 | vma, addr, next)) | |
1139 | return -ENOMEM; | |
1140 | } while (dst_p4d++, src_p4d++, addr = next, addr != end); | |
1141 | return 0; | |
1142 | } | |
1143 | ||
1da177e4 LT |
1144 | int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
1145 | struct vm_area_struct *vma) | |
1146 | { | |
1147 | pgd_t *src_pgd, *dst_pgd; | |
1148 | unsigned long next; | |
1149 | unsigned long addr = vma->vm_start; | |
1150 | unsigned long end = vma->vm_end; | |
2ec74c3e SG |
1151 | unsigned long mmun_start; /* For mmu_notifiers */ |
1152 | unsigned long mmun_end; /* For mmu_notifiers */ | |
1153 | bool is_cow; | |
cddb8a5c | 1154 | int ret; |
1da177e4 | 1155 | |
d992895b NP |
1156 | /* |
1157 | * Don't copy ptes where a page fault will fill them correctly. | |
1158 | * Fork becomes much lighter when there are big shared or private | |
1159 | * readonly mappings. The tradeoff is that copy_page_range is more | |
1160 | * efficient than faulting. | |
1161 | */ | |
0661a336 KS |
1162 | if (!(vma->vm_flags & (VM_HUGETLB | VM_PFNMAP | VM_MIXEDMAP)) && |
1163 | !vma->anon_vma) | |
1164 | return 0; | |
d992895b | 1165 | |
1da177e4 LT |
1166 | if (is_vm_hugetlb_page(vma)) |
1167 | return copy_hugetlb_page_range(dst_mm, src_mm, vma); | |
1168 | ||
b3b9c293 | 1169 | if (unlikely(vma->vm_flags & VM_PFNMAP)) { |
2ab64037 | 1170 | /* |
1171 | * We do not free on error cases below as remove_vma | |
1172 | * gets called on error from higher level routine | |
1173 | */ | |
5180da41 | 1174 | ret = track_pfn_copy(vma); |
2ab64037 | 1175 | if (ret) |
1176 | return ret; | |
1177 | } | |
1178 | ||
cddb8a5c AA |
1179 | /* |
1180 | * We need to invalidate the secondary MMU mappings only when | |
1181 | * there could be a permission downgrade on the ptes of the | |
1182 | * parent mm. And a permission downgrade will only happen if | |
1183 | * is_cow_mapping() returns true. | |
1184 | */ | |
2ec74c3e SG |
1185 | is_cow = is_cow_mapping(vma->vm_flags); |
1186 | mmun_start = addr; | |
1187 | mmun_end = end; | |
1188 | if (is_cow) | |
1189 | mmu_notifier_invalidate_range_start(src_mm, mmun_start, | |
1190 | mmun_end); | |
cddb8a5c AA |
1191 | |
1192 | ret = 0; | |
1da177e4 LT |
1193 | dst_pgd = pgd_offset(dst_mm, addr); |
1194 | src_pgd = pgd_offset(src_mm, addr); | |
1195 | do { | |
1196 | next = pgd_addr_end(addr, end); | |
1197 | if (pgd_none_or_clear_bad(src_pgd)) | |
1198 | continue; | |
c2febafc | 1199 | if (unlikely(copy_p4d_range(dst_mm, src_mm, dst_pgd, src_pgd, |
cddb8a5c AA |
1200 | vma, addr, next))) { |
1201 | ret = -ENOMEM; | |
1202 | break; | |
1203 | } | |
1da177e4 | 1204 | } while (dst_pgd++, src_pgd++, addr = next, addr != end); |
cddb8a5c | 1205 | |
2ec74c3e SG |
1206 | if (is_cow) |
1207 | mmu_notifier_invalidate_range_end(src_mm, mmun_start, mmun_end); | |
cddb8a5c | 1208 | return ret; |
1da177e4 LT |
1209 | } |
1210 | ||
51c6f666 | 1211 | static unsigned long zap_pte_range(struct mmu_gather *tlb, |
b5810039 | 1212 | struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 | 1213 | unsigned long addr, unsigned long end, |
97a89413 | 1214 | struct zap_details *details) |
1da177e4 | 1215 | { |
b5810039 | 1216 | struct mm_struct *mm = tlb->mm; |
d16dfc55 | 1217 | int force_flush = 0; |
d559db08 | 1218 | int rss[NR_MM_COUNTERS]; |
97a89413 | 1219 | spinlock_t *ptl; |
5f1a1907 | 1220 | pte_t *start_pte; |
97a89413 | 1221 | pte_t *pte; |
8a5f14a2 | 1222 | swp_entry_t entry; |
d559db08 | 1223 | |
07e32661 | 1224 | tlb_remove_check_page_size_change(tlb, PAGE_SIZE); |
d16dfc55 | 1225 | again: |
e303297e | 1226 | init_rss_vec(rss); |
5f1a1907 SR |
1227 | start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
1228 | pte = start_pte; | |
3ea27719 | 1229 | flush_tlb_batched_pending(mm); |
6606c3e0 | 1230 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
1231 | do { |
1232 | pte_t ptent = *pte; | |
166f61b9 | 1233 | if (pte_none(ptent)) |
1da177e4 | 1234 | continue; |
6f5e6b9e | 1235 | |
1da177e4 | 1236 | if (pte_present(ptent)) { |
ee498ed7 | 1237 | struct page *page; |
51c6f666 | 1238 | |
6aab341e | 1239 | page = vm_normal_page(vma, addr, ptent); |
1da177e4 LT |
1240 | if (unlikely(details) && page) { |
1241 | /* | |
1242 | * unmap_shared_mapping_pages() wants to | |
1243 | * invalidate cache without truncating: | |
1244 | * unmap shared but keep private pages. | |
1245 | */ | |
1246 | if (details->check_mapping && | |
800d8c63 | 1247 | details->check_mapping != page_rmapping(page)) |
1da177e4 | 1248 | continue; |
1da177e4 | 1249 | } |
b5810039 | 1250 | ptent = ptep_get_and_clear_full(mm, addr, pte, |
a600388d | 1251 | tlb->fullmm); |
1da177e4 LT |
1252 | tlb_remove_tlb_entry(tlb, pte, addr); |
1253 | if (unlikely(!page)) | |
1254 | continue; | |
eca56ff9 JM |
1255 | |
1256 | if (!PageAnon(page)) { | |
1cf35d47 LT |
1257 | if (pte_dirty(ptent)) { |
1258 | force_flush = 1; | |
6237bcd9 | 1259 | set_page_dirty(page); |
1cf35d47 | 1260 | } |
4917e5d0 | 1261 | if (pte_young(ptent) && |
64363aad | 1262 | likely(!(vma->vm_flags & VM_SEQ_READ))) |
bf3f3bc5 | 1263 | mark_page_accessed(page); |
6237bcd9 | 1264 | } |
eca56ff9 | 1265 | rss[mm_counter(page)]--; |
d281ee61 | 1266 | page_remove_rmap(page, false); |
3dc14741 HD |
1267 | if (unlikely(page_mapcount(page) < 0)) |
1268 | print_bad_pte(vma, addr, ptent, page); | |
e9d55e15 | 1269 | if (unlikely(__tlb_remove_page(tlb, page))) { |
1cf35d47 | 1270 | force_flush = 1; |
ce9ec37b | 1271 | addr += PAGE_SIZE; |
d16dfc55 | 1272 | break; |
1cf35d47 | 1273 | } |
1da177e4 LT |
1274 | continue; |
1275 | } | |
3e8715fd KS |
1276 | /* If details->check_mapping, we leave swap entries. */ |
1277 | if (unlikely(details)) | |
1da177e4 | 1278 | continue; |
b084d435 | 1279 | |
8a5f14a2 KS |
1280 | entry = pte_to_swp_entry(ptent); |
1281 | if (!non_swap_entry(entry)) | |
1282 | rss[MM_SWAPENTS]--; | |
1283 | else if (is_migration_entry(entry)) { | |
1284 | struct page *page; | |
9f9f1acd | 1285 | |
8a5f14a2 | 1286 | page = migration_entry_to_page(entry); |
eca56ff9 | 1287 | rss[mm_counter(page)]--; |
b084d435 | 1288 | } |
8a5f14a2 KS |
1289 | if (unlikely(!free_swap_and_cache(entry))) |
1290 | print_bad_pte(vma, addr, ptent, NULL); | |
9888a1ca | 1291 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); |
97a89413 | 1292 | } while (pte++, addr += PAGE_SIZE, addr != end); |
ae859762 | 1293 | |
d559db08 | 1294 | add_mm_rss_vec(mm, rss); |
6606c3e0 | 1295 | arch_leave_lazy_mmu_mode(); |
51c6f666 | 1296 | |
1cf35d47 | 1297 | /* Do the actual TLB flush before dropping ptl */ |
fb7332a9 | 1298 | if (force_flush) |
1cf35d47 | 1299 | tlb_flush_mmu_tlbonly(tlb); |
1cf35d47 LT |
1300 | pte_unmap_unlock(start_pte, ptl); |
1301 | ||
1302 | /* | |
1303 | * If we forced a TLB flush (either due to running out of | |
1304 | * batch buffers or because we needed to flush dirty TLB | |
1305 | * entries before releasing the ptl), free the batched | |
1306 | * memory too. Restart if we didn't do everything. | |
1307 | */ | |
1308 | if (force_flush) { | |
1309 | force_flush = 0; | |
1310 | tlb_flush_mmu_free(tlb); | |
2b047252 | 1311 | if (addr != end) |
d16dfc55 PZ |
1312 | goto again; |
1313 | } | |
1314 | ||
51c6f666 | 1315 | return addr; |
1da177e4 LT |
1316 | } |
1317 | ||
51c6f666 | 1318 | static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, |
b5810039 | 1319 | struct vm_area_struct *vma, pud_t *pud, |
1da177e4 | 1320 | unsigned long addr, unsigned long end, |
97a89413 | 1321 | struct zap_details *details) |
1da177e4 LT |
1322 | { |
1323 | pmd_t *pmd; | |
1324 | unsigned long next; | |
1325 | ||
1326 | pmd = pmd_offset(pud, addr); | |
1327 | do { | |
1328 | next = pmd_addr_end(addr, end); | |
5c7fb56e | 1329 | if (pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) { |
1a5a9906 | 1330 | if (next - addr != HPAGE_PMD_SIZE) { |
68428398 HD |
1331 | VM_BUG_ON_VMA(vma_is_anonymous(vma) && |
1332 | !rwsem_is_locked(&tlb->mm->mmap_sem), vma); | |
fd60775a | 1333 | __split_huge_pmd(vma, pmd, addr, false, NULL); |
f21760b1 | 1334 | } else if (zap_huge_pmd(tlb, vma, pmd, addr)) |
1a5a9906 | 1335 | goto next; |
71e3aac0 AA |
1336 | /* fall through */ |
1337 | } | |
1a5a9906 AA |
1338 | /* |
1339 | * Here there can be other concurrent MADV_DONTNEED or | |
1340 | * trans huge page faults running, and if the pmd is | |
1341 | * none or trans huge it can change under us. This is | |
1342 | * because MADV_DONTNEED holds the mmap_sem in read | |
1343 | * mode. | |
1344 | */ | |
1345 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) | |
1346 | goto next; | |
97a89413 | 1347 | next = zap_pte_range(tlb, vma, pmd, addr, next, details); |
1a5a9906 | 1348 | next: |
97a89413 PZ |
1349 | cond_resched(); |
1350 | } while (pmd++, addr = next, addr != end); | |
51c6f666 RH |
1351 | |
1352 | return addr; | |
1da177e4 LT |
1353 | } |
1354 | ||
51c6f666 | 1355 | static inline unsigned long zap_pud_range(struct mmu_gather *tlb, |
c2febafc | 1356 | struct vm_area_struct *vma, p4d_t *p4d, |
1da177e4 | 1357 | unsigned long addr, unsigned long end, |
97a89413 | 1358 | struct zap_details *details) |
1da177e4 LT |
1359 | { |
1360 | pud_t *pud; | |
1361 | unsigned long next; | |
1362 | ||
c2febafc | 1363 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
1364 | do { |
1365 | next = pud_addr_end(addr, end); | |
a00cc7d9 MW |
1366 | if (pud_trans_huge(*pud) || pud_devmap(*pud)) { |
1367 | if (next - addr != HPAGE_PUD_SIZE) { | |
1368 | VM_BUG_ON_VMA(!rwsem_is_locked(&tlb->mm->mmap_sem), vma); | |
1369 | split_huge_pud(vma, pud, addr); | |
1370 | } else if (zap_huge_pud(tlb, vma, pud, addr)) | |
1371 | goto next; | |
1372 | /* fall through */ | |
1373 | } | |
97a89413 | 1374 | if (pud_none_or_clear_bad(pud)) |
1da177e4 | 1375 | continue; |
97a89413 | 1376 | next = zap_pmd_range(tlb, vma, pud, addr, next, details); |
a00cc7d9 MW |
1377 | next: |
1378 | cond_resched(); | |
97a89413 | 1379 | } while (pud++, addr = next, addr != end); |
51c6f666 RH |
1380 | |
1381 | return addr; | |
1da177e4 LT |
1382 | } |
1383 | ||
c2febafc KS |
1384 | static inline unsigned long zap_p4d_range(struct mmu_gather *tlb, |
1385 | struct vm_area_struct *vma, pgd_t *pgd, | |
1386 | unsigned long addr, unsigned long end, | |
1387 | struct zap_details *details) | |
1388 | { | |
1389 | p4d_t *p4d; | |
1390 | unsigned long next; | |
1391 | ||
1392 | p4d = p4d_offset(pgd, addr); | |
1393 | do { | |
1394 | next = p4d_addr_end(addr, end); | |
1395 | if (p4d_none_or_clear_bad(p4d)) | |
1396 | continue; | |
1397 | next = zap_pud_range(tlb, vma, p4d, addr, next, details); | |
1398 | } while (p4d++, addr = next, addr != end); | |
1399 | ||
1400 | return addr; | |
1401 | } | |
1402 | ||
aac45363 | 1403 | void unmap_page_range(struct mmu_gather *tlb, |
038c7aa1 AV |
1404 | struct vm_area_struct *vma, |
1405 | unsigned long addr, unsigned long end, | |
1406 | struct zap_details *details) | |
1da177e4 LT |
1407 | { |
1408 | pgd_t *pgd; | |
1409 | unsigned long next; | |
1410 | ||
1da177e4 LT |
1411 | BUG_ON(addr >= end); |
1412 | tlb_start_vma(tlb, vma); | |
1413 | pgd = pgd_offset(vma->vm_mm, addr); | |
1414 | do { | |
1415 | next = pgd_addr_end(addr, end); | |
97a89413 | 1416 | if (pgd_none_or_clear_bad(pgd)) |
1da177e4 | 1417 | continue; |
c2febafc | 1418 | next = zap_p4d_range(tlb, vma, pgd, addr, next, details); |
97a89413 | 1419 | } while (pgd++, addr = next, addr != end); |
1da177e4 LT |
1420 | tlb_end_vma(tlb, vma); |
1421 | } | |
51c6f666 | 1422 | |
f5cc4eef AV |
1423 | |
1424 | static void unmap_single_vma(struct mmu_gather *tlb, | |
1425 | struct vm_area_struct *vma, unsigned long start_addr, | |
4f74d2c8 | 1426 | unsigned long end_addr, |
f5cc4eef AV |
1427 | struct zap_details *details) |
1428 | { | |
1429 | unsigned long start = max(vma->vm_start, start_addr); | |
1430 | unsigned long end; | |
1431 | ||
1432 | if (start >= vma->vm_end) | |
1433 | return; | |
1434 | end = min(vma->vm_end, end_addr); | |
1435 | if (end <= vma->vm_start) | |
1436 | return; | |
1437 | ||
cbc91f71 SD |
1438 | if (vma->vm_file) |
1439 | uprobe_munmap(vma, start, end); | |
1440 | ||
b3b9c293 | 1441 | if (unlikely(vma->vm_flags & VM_PFNMAP)) |
5180da41 | 1442 | untrack_pfn(vma, 0, 0); |
f5cc4eef AV |
1443 | |
1444 | if (start != end) { | |
1445 | if (unlikely(is_vm_hugetlb_page(vma))) { | |
1446 | /* | |
1447 | * It is undesirable to test vma->vm_file as it | |
1448 | * should be non-null for valid hugetlb area. | |
1449 | * However, vm_file will be NULL in the error | |
7aa6b4ad | 1450 | * cleanup path of mmap_region. When |
f5cc4eef | 1451 | * hugetlbfs ->mmap method fails, |
7aa6b4ad | 1452 | * mmap_region() nullifies vma->vm_file |
f5cc4eef AV |
1453 | * before calling this function to clean up. |
1454 | * Since no pte has actually been setup, it is | |
1455 | * safe to do nothing in this case. | |
1456 | */ | |
24669e58 | 1457 | if (vma->vm_file) { |
83cde9e8 | 1458 | i_mmap_lock_write(vma->vm_file->f_mapping); |
d833352a | 1459 | __unmap_hugepage_range_final(tlb, vma, start, end, NULL); |
83cde9e8 | 1460 | i_mmap_unlock_write(vma->vm_file->f_mapping); |
24669e58 | 1461 | } |
f5cc4eef AV |
1462 | } else |
1463 | unmap_page_range(tlb, vma, start, end, details); | |
1464 | } | |
1da177e4 LT |
1465 | } |
1466 | ||
1da177e4 LT |
1467 | /** |
1468 | * unmap_vmas - unmap a range of memory covered by a list of vma's | |
0164f69d | 1469 | * @tlb: address of the caller's struct mmu_gather |
1da177e4 LT |
1470 | * @vma: the starting vma |
1471 | * @start_addr: virtual address at which to start unmapping | |
1472 | * @end_addr: virtual address at which to end unmapping | |
1da177e4 | 1473 | * |
508034a3 | 1474 | * Unmap all pages in the vma list. |
1da177e4 | 1475 | * |
1da177e4 LT |
1476 | * Only addresses between `start' and `end' will be unmapped. |
1477 | * | |
1478 | * The VMA list must be sorted in ascending virtual address order. | |
1479 | * | |
1480 | * unmap_vmas() assumes that the caller will flush the whole unmapped address | |
1481 | * range after unmap_vmas() returns. So the only responsibility here is to | |
1482 | * ensure that any thus-far unmapped pages are flushed before unmap_vmas() | |
1483 | * drops the lock and schedules. | |
1484 | */ | |
6e8bb019 | 1485 | void unmap_vmas(struct mmu_gather *tlb, |
1da177e4 | 1486 | struct vm_area_struct *vma, unsigned long start_addr, |
4f74d2c8 | 1487 | unsigned long end_addr) |
1da177e4 | 1488 | { |
cddb8a5c | 1489 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 1490 | |
cddb8a5c | 1491 | mmu_notifier_invalidate_range_start(mm, start_addr, end_addr); |
f5cc4eef | 1492 | for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) |
4f74d2c8 | 1493 | unmap_single_vma(tlb, vma, start_addr, end_addr, NULL); |
cddb8a5c | 1494 | mmu_notifier_invalidate_range_end(mm, start_addr, end_addr); |
1da177e4 LT |
1495 | } |
1496 | ||
1497 | /** | |
1498 | * zap_page_range - remove user pages in a given range | |
1499 | * @vma: vm_area_struct holding the applicable pages | |
eb4546bb | 1500 | * @start: starting address of pages to zap |
1da177e4 | 1501 | * @size: number of bytes to zap |
f5cc4eef AV |
1502 | * |
1503 | * Caller must protect the VMA list | |
1da177e4 | 1504 | */ |
7e027b14 | 1505 | void zap_page_range(struct vm_area_struct *vma, unsigned long start, |
ecf1385d | 1506 | unsigned long size) |
1da177e4 LT |
1507 | { |
1508 | struct mm_struct *mm = vma->vm_mm; | |
d16dfc55 | 1509 | struct mmu_gather tlb; |
7e027b14 | 1510 | unsigned long end = start + size; |
1da177e4 | 1511 | |
1da177e4 | 1512 | lru_add_drain(); |
2b047252 | 1513 | tlb_gather_mmu(&tlb, mm, start, end); |
365e9c87 | 1514 | update_hiwater_rss(mm); |
7e027b14 | 1515 | mmu_notifier_invalidate_range_start(mm, start, end); |
4647706e | 1516 | for ( ; vma && vma->vm_start < end; vma = vma->vm_next) { |
ecf1385d | 1517 | unmap_single_vma(&tlb, vma, start, end, NULL); |
4647706e MG |
1518 | |
1519 | /* | |
1520 | * zap_page_range does not specify whether mmap_sem should be | |
1521 | * held for read or write. That allows parallel zap_page_range | |
1522 | * operations to unmap a PTE and defer a flush meaning that | |
1523 | * this call observes pte_none and fails to flush the TLB. | |
1524 | * Rather than adding a complex API, ensure that no stale | |
1525 | * TLB entries exist when this call returns. | |
1526 | */ | |
1527 | flush_tlb_range(vma, start, end); | |
1528 | } | |
1529 | ||
7e027b14 LT |
1530 | mmu_notifier_invalidate_range_end(mm, start, end); |
1531 | tlb_finish_mmu(&tlb, start, end); | |
1da177e4 LT |
1532 | } |
1533 | ||
f5cc4eef AV |
1534 | /** |
1535 | * zap_page_range_single - remove user pages in a given range | |
1536 | * @vma: vm_area_struct holding the applicable pages | |
1537 | * @address: starting address of pages to zap | |
1538 | * @size: number of bytes to zap | |
8a5f14a2 | 1539 | * @details: details of shared cache invalidation |
f5cc4eef AV |
1540 | * |
1541 | * The range must fit into one VMA. | |
1da177e4 | 1542 | */ |
f5cc4eef | 1543 | static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address, |
1da177e4 LT |
1544 | unsigned long size, struct zap_details *details) |
1545 | { | |
1546 | struct mm_struct *mm = vma->vm_mm; | |
d16dfc55 | 1547 | struct mmu_gather tlb; |
1da177e4 | 1548 | unsigned long end = address + size; |
1da177e4 | 1549 | |
1da177e4 | 1550 | lru_add_drain(); |
2b047252 | 1551 | tlb_gather_mmu(&tlb, mm, address, end); |
365e9c87 | 1552 | update_hiwater_rss(mm); |
f5cc4eef | 1553 | mmu_notifier_invalidate_range_start(mm, address, end); |
4f74d2c8 | 1554 | unmap_single_vma(&tlb, vma, address, end, details); |
f5cc4eef | 1555 | mmu_notifier_invalidate_range_end(mm, address, end); |
d16dfc55 | 1556 | tlb_finish_mmu(&tlb, address, end); |
1da177e4 LT |
1557 | } |
1558 | ||
c627f9cc JS |
1559 | /** |
1560 | * zap_vma_ptes - remove ptes mapping the vma | |
1561 | * @vma: vm_area_struct holding ptes to be zapped | |
1562 | * @address: starting address of pages to zap | |
1563 | * @size: number of bytes to zap | |
1564 | * | |
1565 | * This function only unmaps ptes assigned to VM_PFNMAP vmas. | |
1566 | * | |
1567 | * The entire address range must be fully contained within the vma. | |
1568 | * | |
1569 | * Returns 0 if successful. | |
1570 | */ | |
1571 | int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, | |
1572 | unsigned long size) | |
1573 | { | |
1574 | if (address < vma->vm_start || address + size > vma->vm_end || | |
1575 | !(vma->vm_flags & VM_PFNMAP)) | |
1576 | return -1; | |
f5cc4eef | 1577 | zap_page_range_single(vma, address, size, NULL); |
c627f9cc JS |
1578 | return 0; |
1579 | } | |
1580 | EXPORT_SYMBOL_GPL(zap_vma_ptes); | |
1581 | ||
25ca1d6c | 1582 | pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr, |
920c7a5d | 1583 | spinlock_t **ptl) |
c9cfcddf | 1584 | { |
c2febafc KS |
1585 | pgd_t *pgd; |
1586 | p4d_t *p4d; | |
1587 | pud_t *pud; | |
1588 | pmd_t *pmd; | |
1589 | ||
1590 | pgd = pgd_offset(mm, addr); | |
1591 | p4d = p4d_alloc(mm, pgd, addr); | |
1592 | if (!p4d) | |
1593 | return NULL; | |
1594 | pud = pud_alloc(mm, p4d, addr); | |
1595 | if (!pud) | |
1596 | return NULL; | |
1597 | pmd = pmd_alloc(mm, pud, addr); | |
1598 | if (!pmd) | |
1599 | return NULL; | |
1600 | ||
1601 | VM_BUG_ON(pmd_trans_huge(*pmd)); | |
1602 | return pte_alloc_map_lock(mm, pmd, addr, ptl); | |
c9cfcddf LT |
1603 | } |
1604 | ||
238f58d8 LT |
1605 | /* |
1606 | * This is the old fallback for page remapping. | |
1607 | * | |
1608 | * For historical reasons, it only allows reserved pages. Only | |
1609 | * old drivers should use this, and they needed to mark their | |
1610 | * pages reserved for the old functions anyway. | |
1611 | */ | |
423bad60 NP |
1612 | static int insert_page(struct vm_area_struct *vma, unsigned long addr, |
1613 | struct page *page, pgprot_t prot) | |
238f58d8 | 1614 | { |
423bad60 | 1615 | struct mm_struct *mm = vma->vm_mm; |
238f58d8 | 1616 | int retval; |
c9cfcddf | 1617 | pte_t *pte; |
8a9f3ccd BS |
1618 | spinlock_t *ptl; |
1619 | ||
238f58d8 | 1620 | retval = -EINVAL; |
a145dd41 | 1621 | if (PageAnon(page)) |
5b4e655e | 1622 | goto out; |
238f58d8 LT |
1623 | retval = -ENOMEM; |
1624 | flush_dcache_page(page); | |
c9cfcddf | 1625 | pte = get_locked_pte(mm, addr, &ptl); |
238f58d8 | 1626 | if (!pte) |
5b4e655e | 1627 | goto out; |
238f58d8 LT |
1628 | retval = -EBUSY; |
1629 | if (!pte_none(*pte)) | |
1630 | goto out_unlock; | |
1631 | ||
1632 | /* Ok, finally just insert the thing.. */ | |
1633 | get_page(page); | |
eca56ff9 | 1634 | inc_mm_counter_fast(mm, mm_counter_file(page)); |
dd78fedd | 1635 | page_add_file_rmap(page, false); |
238f58d8 LT |
1636 | set_pte_at(mm, addr, pte, mk_pte(page, prot)); |
1637 | ||
1638 | retval = 0; | |
8a9f3ccd BS |
1639 | pte_unmap_unlock(pte, ptl); |
1640 | return retval; | |
238f58d8 LT |
1641 | out_unlock: |
1642 | pte_unmap_unlock(pte, ptl); | |
1643 | out: | |
1644 | return retval; | |
1645 | } | |
1646 | ||
bfa5bf6d REB |
1647 | /** |
1648 | * vm_insert_page - insert single page into user vma | |
1649 | * @vma: user vma to map to | |
1650 | * @addr: target user address of this page | |
1651 | * @page: source kernel page | |
1652 | * | |
a145dd41 LT |
1653 | * This allows drivers to insert individual pages they've allocated |
1654 | * into a user vma. | |
1655 | * | |
1656 | * The page has to be a nice clean _individual_ kernel allocation. | |
1657 | * If you allocate a compound page, you need to have marked it as | |
1658 | * such (__GFP_COMP), or manually just split the page up yourself | |
8dfcc9ba | 1659 | * (see split_page()). |
a145dd41 LT |
1660 | * |
1661 | * NOTE! Traditionally this was done with "remap_pfn_range()" which | |
1662 | * took an arbitrary page protection parameter. This doesn't allow | |
1663 | * that. Your vma protection will have to be set up correctly, which | |
1664 | * means that if you want a shared writable mapping, you'd better | |
1665 | * ask for a shared writable mapping! | |
1666 | * | |
1667 | * The page does not need to be reserved. | |
4b6e1e37 KK |
1668 | * |
1669 | * Usually this function is called from f_op->mmap() handler | |
1670 | * under mm->mmap_sem write-lock, so it can change vma->vm_flags. | |
1671 | * Caller must set VM_MIXEDMAP on vma if it wants to call this | |
1672 | * function from other places, for example from page-fault handler. | |
a145dd41 | 1673 | */ |
423bad60 NP |
1674 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, |
1675 | struct page *page) | |
a145dd41 LT |
1676 | { |
1677 | if (addr < vma->vm_start || addr >= vma->vm_end) | |
1678 | return -EFAULT; | |
1679 | if (!page_count(page)) | |
1680 | return -EINVAL; | |
4b6e1e37 KK |
1681 | if (!(vma->vm_flags & VM_MIXEDMAP)) { |
1682 | BUG_ON(down_read_trylock(&vma->vm_mm->mmap_sem)); | |
1683 | BUG_ON(vma->vm_flags & VM_PFNMAP); | |
1684 | vma->vm_flags |= VM_MIXEDMAP; | |
1685 | } | |
423bad60 | 1686 | return insert_page(vma, addr, page, vma->vm_page_prot); |
a145dd41 | 1687 | } |
e3c3374f | 1688 | EXPORT_SYMBOL(vm_insert_page); |
a145dd41 | 1689 | |
423bad60 | 1690 | static int insert_pfn(struct vm_area_struct *vma, unsigned long addr, |
b2770da6 | 1691 | pfn_t pfn, pgprot_t prot, bool mkwrite) |
423bad60 NP |
1692 | { |
1693 | struct mm_struct *mm = vma->vm_mm; | |
1694 | int retval; | |
1695 | pte_t *pte, entry; | |
1696 | spinlock_t *ptl; | |
1697 | ||
1698 | retval = -ENOMEM; | |
1699 | pte = get_locked_pte(mm, addr, &ptl); | |
1700 | if (!pte) | |
1701 | goto out; | |
1702 | retval = -EBUSY; | |
b2770da6 RZ |
1703 | if (!pte_none(*pte)) { |
1704 | if (mkwrite) { | |
1705 | /* | |
1706 | * For read faults on private mappings the PFN passed | |
1707 | * in may not match the PFN we have mapped if the | |
1708 | * mapped PFN is a writeable COW page. In the mkwrite | |
1709 | * case we are creating a writable PTE for a shared | |
1710 | * mapping and we expect the PFNs to match. | |
1711 | */ | |
1712 | if (WARN_ON_ONCE(pte_pfn(*pte) != pfn_t_to_pfn(pfn))) | |
1713 | goto out_unlock; | |
1714 | entry = *pte; | |
1715 | goto out_mkwrite; | |
1716 | } else | |
1717 | goto out_unlock; | |
1718 | } | |
423bad60 NP |
1719 | |
1720 | /* Ok, finally just insert the thing.. */ | |
01c8f1c4 DW |
1721 | if (pfn_t_devmap(pfn)) |
1722 | entry = pte_mkdevmap(pfn_t_pte(pfn, prot)); | |
1723 | else | |
1724 | entry = pte_mkspecial(pfn_t_pte(pfn, prot)); | |
b2770da6 RZ |
1725 | |
1726 | out_mkwrite: | |
1727 | if (mkwrite) { | |
1728 | entry = pte_mkyoung(entry); | |
1729 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
1730 | } | |
1731 | ||
423bad60 | 1732 | set_pte_at(mm, addr, pte, entry); |
4b3073e1 | 1733 | update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */ |
423bad60 NP |
1734 | |
1735 | retval = 0; | |
1736 | out_unlock: | |
1737 | pte_unmap_unlock(pte, ptl); | |
1738 | out: | |
1739 | return retval; | |
1740 | } | |
1741 | ||
e0dc0d8f NP |
1742 | /** |
1743 | * vm_insert_pfn - insert single pfn into user vma | |
1744 | * @vma: user vma to map to | |
1745 | * @addr: target user address of this page | |
1746 | * @pfn: source kernel pfn | |
1747 | * | |
c462f179 | 1748 | * Similar to vm_insert_page, this allows drivers to insert individual pages |
e0dc0d8f NP |
1749 | * they've allocated into a user vma. Same comments apply. |
1750 | * | |
1751 | * This function should only be called from a vm_ops->fault handler, and | |
1752 | * in that case the handler should return NULL. | |
0d71d10a NP |
1753 | * |
1754 | * vma cannot be a COW mapping. | |
1755 | * | |
1756 | * As this is called only for pages that do not currently exist, we | |
1757 | * do not need to flush old virtual caches or the TLB. | |
e0dc0d8f NP |
1758 | */ |
1759 | int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr, | |
423bad60 | 1760 | unsigned long pfn) |
1745cbc5 AL |
1761 | { |
1762 | return vm_insert_pfn_prot(vma, addr, pfn, vma->vm_page_prot); | |
1763 | } | |
1764 | EXPORT_SYMBOL(vm_insert_pfn); | |
1765 | ||
1766 | /** | |
1767 | * vm_insert_pfn_prot - insert single pfn into user vma with specified pgprot | |
1768 | * @vma: user vma to map to | |
1769 | * @addr: target user address of this page | |
1770 | * @pfn: source kernel pfn | |
1771 | * @pgprot: pgprot flags for the inserted page | |
1772 | * | |
1773 | * This is exactly like vm_insert_pfn, except that it allows drivers to | |
1774 | * to override pgprot on a per-page basis. | |
1775 | * | |
1776 | * This only makes sense for IO mappings, and it makes no sense for | |
1777 | * cow mappings. In general, using multiple vmas is preferable; | |
1778 | * vm_insert_pfn_prot should only be used if using multiple VMAs is | |
1779 | * impractical. | |
1780 | */ | |
1781 | int vm_insert_pfn_prot(struct vm_area_struct *vma, unsigned long addr, | |
1782 | unsigned long pfn, pgprot_t pgprot) | |
e0dc0d8f | 1783 | { |
2ab64037 | 1784 | int ret; |
7e675137 NP |
1785 | /* |
1786 | * Technically, architectures with pte_special can avoid all these | |
1787 | * restrictions (same for remap_pfn_range). However we would like | |
1788 | * consistency in testing and feature parity among all, so we should | |
1789 | * try to keep these invariants in place for everybody. | |
1790 | */ | |
b379d790 JH |
1791 | BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); |
1792 | BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == | |
1793 | (VM_PFNMAP|VM_MIXEDMAP)); | |
1794 | BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); | |
1795 | BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); | |
e0dc0d8f | 1796 | |
423bad60 NP |
1797 | if (addr < vma->vm_start || addr >= vma->vm_end) |
1798 | return -EFAULT; | |
308a047c BP |
1799 | |
1800 | track_pfn_insert(vma, &pgprot, __pfn_to_pfn_t(pfn, PFN_DEV)); | |
2ab64037 | 1801 | |
b2770da6 RZ |
1802 | ret = insert_pfn(vma, addr, __pfn_to_pfn_t(pfn, PFN_DEV), pgprot, |
1803 | false); | |
2ab64037 | 1804 | |
2ab64037 | 1805 | return ret; |
423bad60 | 1806 | } |
1745cbc5 | 1807 | EXPORT_SYMBOL(vm_insert_pfn_prot); |
e0dc0d8f | 1808 | |
b2770da6 RZ |
1809 | static int __vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr, |
1810 | pfn_t pfn, bool mkwrite) | |
423bad60 | 1811 | { |
87744ab3 DW |
1812 | pgprot_t pgprot = vma->vm_page_prot; |
1813 | ||
423bad60 | 1814 | BUG_ON(!(vma->vm_flags & VM_MIXEDMAP)); |
e0dc0d8f | 1815 | |
423bad60 NP |
1816 | if (addr < vma->vm_start || addr >= vma->vm_end) |
1817 | return -EFAULT; | |
308a047c BP |
1818 | |
1819 | track_pfn_insert(vma, &pgprot, pfn); | |
e0dc0d8f | 1820 | |
423bad60 NP |
1821 | /* |
1822 | * If we don't have pte special, then we have to use the pfn_valid() | |
1823 | * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* | |
1824 | * refcount the page if pfn_valid is true (hence insert_page rather | |
62eede62 HD |
1825 | * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP |
1826 | * without pte special, it would there be refcounted as a normal page. | |
423bad60 | 1827 | */ |
03fc2da6 | 1828 | if (!HAVE_PTE_SPECIAL && !pfn_t_devmap(pfn) && pfn_t_valid(pfn)) { |
423bad60 NP |
1829 | struct page *page; |
1830 | ||
03fc2da6 DW |
1831 | /* |
1832 | * At this point we are committed to insert_page() | |
1833 | * regardless of whether the caller specified flags that | |
1834 | * result in pfn_t_has_page() == false. | |
1835 | */ | |
1836 | page = pfn_to_page(pfn_t_to_pfn(pfn)); | |
87744ab3 | 1837 | return insert_page(vma, addr, page, pgprot); |
423bad60 | 1838 | } |
b2770da6 RZ |
1839 | return insert_pfn(vma, addr, pfn, pgprot, mkwrite); |
1840 | } | |
1841 | ||
1842 | int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr, | |
1843 | pfn_t pfn) | |
1844 | { | |
1845 | return __vm_insert_mixed(vma, addr, pfn, false); | |
1846 | ||
e0dc0d8f | 1847 | } |
423bad60 | 1848 | EXPORT_SYMBOL(vm_insert_mixed); |
e0dc0d8f | 1849 | |
b2770da6 RZ |
1850 | int vm_insert_mixed_mkwrite(struct vm_area_struct *vma, unsigned long addr, |
1851 | pfn_t pfn) | |
1852 | { | |
1853 | return __vm_insert_mixed(vma, addr, pfn, true); | |
1854 | } | |
1855 | EXPORT_SYMBOL(vm_insert_mixed_mkwrite); | |
1856 | ||
1da177e4 LT |
1857 | /* |
1858 | * maps a range of physical memory into the requested pages. the old | |
1859 | * mappings are removed. any references to nonexistent pages results | |
1860 | * in null mappings (currently treated as "copy-on-access") | |
1861 | */ | |
1862 | static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, | |
1863 | unsigned long addr, unsigned long end, | |
1864 | unsigned long pfn, pgprot_t prot) | |
1865 | { | |
1866 | pte_t *pte; | |
c74df32c | 1867 | spinlock_t *ptl; |
1da177e4 | 1868 | |
c74df32c | 1869 | pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); |
1da177e4 LT |
1870 | if (!pte) |
1871 | return -ENOMEM; | |
6606c3e0 | 1872 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
1873 | do { |
1874 | BUG_ON(!pte_none(*pte)); | |
7e675137 | 1875 | set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); |
1da177e4 LT |
1876 | pfn++; |
1877 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
6606c3e0 | 1878 | arch_leave_lazy_mmu_mode(); |
c74df32c | 1879 | pte_unmap_unlock(pte - 1, ptl); |
1da177e4 LT |
1880 | return 0; |
1881 | } | |
1882 | ||
1883 | static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, | |
1884 | unsigned long addr, unsigned long end, | |
1885 | unsigned long pfn, pgprot_t prot) | |
1886 | { | |
1887 | pmd_t *pmd; | |
1888 | unsigned long next; | |
1889 | ||
1890 | pfn -= addr >> PAGE_SHIFT; | |
1891 | pmd = pmd_alloc(mm, pud, addr); | |
1892 | if (!pmd) | |
1893 | return -ENOMEM; | |
f66055ab | 1894 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
1da177e4 LT |
1895 | do { |
1896 | next = pmd_addr_end(addr, end); | |
1897 | if (remap_pte_range(mm, pmd, addr, next, | |
1898 | pfn + (addr >> PAGE_SHIFT), prot)) | |
1899 | return -ENOMEM; | |
1900 | } while (pmd++, addr = next, addr != end); | |
1901 | return 0; | |
1902 | } | |
1903 | ||
c2febafc | 1904 | static inline int remap_pud_range(struct mm_struct *mm, p4d_t *p4d, |
1da177e4 LT |
1905 | unsigned long addr, unsigned long end, |
1906 | unsigned long pfn, pgprot_t prot) | |
1907 | { | |
1908 | pud_t *pud; | |
1909 | unsigned long next; | |
1910 | ||
1911 | pfn -= addr >> PAGE_SHIFT; | |
c2febafc | 1912 | pud = pud_alloc(mm, p4d, addr); |
1da177e4 LT |
1913 | if (!pud) |
1914 | return -ENOMEM; | |
1915 | do { | |
1916 | next = pud_addr_end(addr, end); | |
1917 | if (remap_pmd_range(mm, pud, addr, next, | |
1918 | pfn + (addr >> PAGE_SHIFT), prot)) | |
1919 | return -ENOMEM; | |
1920 | } while (pud++, addr = next, addr != end); | |
1921 | return 0; | |
1922 | } | |
1923 | ||
c2febafc KS |
1924 | static inline int remap_p4d_range(struct mm_struct *mm, pgd_t *pgd, |
1925 | unsigned long addr, unsigned long end, | |
1926 | unsigned long pfn, pgprot_t prot) | |
1927 | { | |
1928 | p4d_t *p4d; | |
1929 | unsigned long next; | |
1930 | ||
1931 | pfn -= addr >> PAGE_SHIFT; | |
1932 | p4d = p4d_alloc(mm, pgd, addr); | |
1933 | if (!p4d) | |
1934 | return -ENOMEM; | |
1935 | do { | |
1936 | next = p4d_addr_end(addr, end); | |
1937 | if (remap_pud_range(mm, p4d, addr, next, | |
1938 | pfn + (addr >> PAGE_SHIFT), prot)) | |
1939 | return -ENOMEM; | |
1940 | } while (p4d++, addr = next, addr != end); | |
1941 | return 0; | |
1942 | } | |
1943 | ||
bfa5bf6d REB |
1944 | /** |
1945 | * remap_pfn_range - remap kernel memory to userspace | |
1946 | * @vma: user vma to map to | |
1947 | * @addr: target user address to start at | |
1948 | * @pfn: physical address of kernel memory | |
1949 | * @size: size of map area | |
1950 | * @prot: page protection flags for this mapping | |
1951 | * | |
1952 | * Note: this is only safe if the mm semaphore is held when called. | |
1953 | */ | |
1da177e4 LT |
1954 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, |
1955 | unsigned long pfn, unsigned long size, pgprot_t prot) | |
1956 | { | |
1957 | pgd_t *pgd; | |
1958 | unsigned long next; | |
2d15cab8 | 1959 | unsigned long end = addr + PAGE_ALIGN(size); |
1da177e4 | 1960 | struct mm_struct *mm = vma->vm_mm; |
d5957d2f | 1961 | unsigned long remap_pfn = pfn; |
1da177e4 LT |
1962 | int err; |
1963 | ||
1964 | /* | |
1965 | * Physically remapped pages are special. Tell the | |
1966 | * rest of the world about it: | |
1967 | * VM_IO tells people not to look at these pages | |
1968 | * (accesses can have side effects). | |
6aab341e LT |
1969 | * VM_PFNMAP tells the core MM that the base pages are just |
1970 | * raw PFN mappings, and do not have a "struct page" associated | |
1971 | * with them. | |
314e51b9 KK |
1972 | * VM_DONTEXPAND |
1973 | * Disable vma merging and expanding with mremap(). | |
1974 | * VM_DONTDUMP | |
1975 | * Omit vma from core dump, even when VM_IO turned off. | |
fb155c16 LT |
1976 | * |
1977 | * There's a horrible special case to handle copy-on-write | |
1978 | * behaviour that some programs depend on. We mark the "original" | |
1979 | * un-COW'ed pages by matching them up with "vma->vm_pgoff". | |
b3b9c293 | 1980 | * See vm_normal_page() for details. |
1da177e4 | 1981 | */ |
b3b9c293 KK |
1982 | if (is_cow_mapping(vma->vm_flags)) { |
1983 | if (addr != vma->vm_start || end != vma->vm_end) | |
1984 | return -EINVAL; | |
fb155c16 | 1985 | vma->vm_pgoff = pfn; |
b3b9c293 KK |
1986 | } |
1987 | ||
d5957d2f | 1988 | err = track_pfn_remap(vma, &prot, remap_pfn, addr, PAGE_ALIGN(size)); |
b3b9c293 | 1989 | if (err) |
3c8bb73a | 1990 | return -EINVAL; |
fb155c16 | 1991 | |
314e51b9 | 1992 | vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; |
1da177e4 LT |
1993 | |
1994 | BUG_ON(addr >= end); | |
1995 | pfn -= addr >> PAGE_SHIFT; | |
1996 | pgd = pgd_offset(mm, addr); | |
1997 | flush_cache_range(vma, addr, end); | |
1da177e4 LT |
1998 | do { |
1999 | next = pgd_addr_end(addr, end); | |
c2febafc | 2000 | err = remap_p4d_range(mm, pgd, addr, next, |
1da177e4 LT |
2001 | pfn + (addr >> PAGE_SHIFT), prot); |
2002 | if (err) | |
2003 | break; | |
2004 | } while (pgd++, addr = next, addr != end); | |
2ab64037 | 2005 | |
2006 | if (err) | |
d5957d2f | 2007 | untrack_pfn(vma, remap_pfn, PAGE_ALIGN(size)); |
2ab64037 | 2008 | |
1da177e4 LT |
2009 | return err; |
2010 | } | |
2011 | EXPORT_SYMBOL(remap_pfn_range); | |
2012 | ||
b4cbb197 LT |
2013 | /** |
2014 | * vm_iomap_memory - remap memory to userspace | |
2015 | * @vma: user vma to map to | |
2016 | * @start: start of area | |
2017 | * @len: size of area | |
2018 | * | |
2019 | * This is a simplified io_remap_pfn_range() for common driver use. The | |
2020 | * driver just needs to give us the physical memory range to be mapped, | |
2021 | * we'll figure out the rest from the vma information. | |
2022 | * | |
2023 | * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get | |
2024 | * whatever write-combining details or similar. | |
2025 | */ | |
2026 | int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) | |
2027 | { | |
2028 | unsigned long vm_len, pfn, pages; | |
2029 | ||
2030 | /* Check that the physical memory area passed in looks valid */ | |
2031 | if (start + len < start) | |
2032 | return -EINVAL; | |
2033 | /* | |
2034 | * You *really* shouldn't map things that aren't page-aligned, | |
2035 | * but we've historically allowed it because IO memory might | |
2036 | * just have smaller alignment. | |
2037 | */ | |
2038 | len += start & ~PAGE_MASK; | |
2039 | pfn = start >> PAGE_SHIFT; | |
2040 | pages = (len + ~PAGE_MASK) >> PAGE_SHIFT; | |
2041 | if (pfn + pages < pfn) | |
2042 | return -EINVAL; | |
2043 | ||
2044 | /* We start the mapping 'vm_pgoff' pages into the area */ | |
2045 | if (vma->vm_pgoff > pages) | |
2046 | return -EINVAL; | |
2047 | pfn += vma->vm_pgoff; | |
2048 | pages -= vma->vm_pgoff; | |
2049 | ||
2050 | /* Can we fit all of the mapping? */ | |
2051 | vm_len = vma->vm_end - vma->vm_start; | |
2052 | if (vm_len >> PAGE_SHIFT > pages) | |
2053 | return -EINVAL; | |
2054 | ||
2055 | /* Ok, let it rip */ | |
2056 | return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); | |
2057 | } | |
2058 | EXPORT_SYMBOL(vm_iomap_memory); | |
2059 | ||
aee16b3c JF |
2060 | static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, |
2061 | unsigned long addr, unsigned long end, | |
2062 | pte_fn_t fn, void *data) | |
2063 | { | |
2064 | pte_t *pte; | |
2065 | int err; | |
2f569afd | 2066 | pgtable_t token; |
94909914 | 2067 | spinlock_t *uninitialized_var(ptl); |
aee16b3c JF |
2068 | |
2069 | pte = (mm == &init_mm) ? | |
2070 | pte_alloc_kernel(pmd, addr) : | |
2071 | pte_alloc_map_lock(mm, pmd, addr, &ptl); | |
2072 | if (!pte) | |
2073 | return -ENOMEM; | |
2074 | ||
2075 | BUG_ON(pmd_huge(*pmd)); | |
2076 | ||
38e0edb1 JF |
2077 | arch_enter_lazy_mmu_mode(); |
2078 | ||
2f569afd | 2079 | token = pmd_pgtable(*pmd); |
aee16b3c JF |
2080 | |
2081 | do { | |
c36987e2 | 2082 | err = fn(pte++, token, addr, data); |
aee16b3c JF |
2083 | if (err) |
2084 | break; | |
c36987e2 | 2085 | } while (addr += PAGE_SIZE, addr != end); |
aee16b3c | 2086 | |
38e0edb1 JF |
2087 | arch_leave_lazy_mmu_mode(); |
2088 | ||
aee16b3c JF |
2089 | if (mm != &init_mm) |
2090 | pte_unmap_unlock(pte-1, ptl); | |
2091 | return err; | |
2092 | } | |
2093 | ||
2094 | static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, | |
2095 | unsigned long addr, unsigned long end, | |
2096 | pte_fn_t fn, void *data) | |
2097 | { | |
2098 | pmd_t *pmd; | |
2099 | unsigned long next; | |
2100 | int err; | |
2101 | ||
ceb86879 AK |
2102 | BUG_ON(pud_huge(*pud)); |
2103 | ||
aee16b3c JF |
2104 | pmd = pmd_alloc(mm, pud, addr); |
2105 | if (!pmd) | |
2106 | return -ENOMEM; | |
2107 | do { | |
2108 | next = pmd_addr_end(addr, end); | |
2109 | err = apply_to_pte_range(mm, pmd, addr, next, fn, data); | |
2110 | if (err) | |
2111 | break; | |
2112 | } while (pmd++, addr = next, addr != end); | |
2113 | return err; | |
2114 | } | |
2115 | ||
c2febafc | 2116 | static int apply_to_pud_range(struct mm_struct *mm, p4d_t *p4d, |
aee16b3c JF |
2117 | unsigned long addr, unsigned long end, |
2118 | pte_fn_t fn, void *data) | |
2119 | { | |
2120 | pud_t *pud; | |
2121 | unsigned long next; | |
2122 | int err; | |
2123 | ||
c2febafc | 2124 | pud = pud_alloc(mm, p4d, addr); |
aee16b3c JF |
2125 | if (!pud) |
2126 | return -ENOMEM; | |
2127 | do { | |
2128 | next = pud_addr_end(addr, end); | |
2129 | err = apply_to_pmd_range(mm, pud, addr, next, fn, data); | |
2130 | if (err) | |
2131 | break; | |
2132 | } while (pud++, addr = next, addr != end); | |
2133 | return err; | |
2134 | } | |
2135 | ||
c2febafc KS |
2136 | static int apply_to_p4d_range(struct mm_struct *mm, pgd_t *pgd, |
2137 | unsigned long addr, unsigned long end, | |
2138 | pte_fn_t fn, void *data) | |
2139 | { | |
2140 | p4d_t *p4d; | |
2141 | unsigned long next; | |
2142 | int err; | |
2143 | ||
2144 | p4d = p4d_alloc(mm, pgd, addr); | |
2145 | if (!p4d) | |
2146 | return -ENOMEM; | |
2147 | do { | |
2148 | next = p4d_addr_end(addr, end); | |
2149 | err = apply_to_pud_range(mm, p4d, addr, next, fn, data); | |
2150 | if (err) | |
2151 | break; | |
2152 | } while (p4d++, addr = next, addr != end); | |
2153 | return err; | |
2154 | } | |
2155 | ||
aee16b3c JF |
2156 | /* |
2157 | * Scan a region of virtual memory, filling in page tables as necessary | |
2158 | * and calling a provided function on each leaf page table. | |
2159 | */ | |
2160 | int apply_to_page_range(struct mm_struct *mm, unsigned long addr, | |
2161 | unsigned long size, pte_fn_t fn, void *data) | |
2162 | { | |
2163 | pgd_t *pgd; | |
2164 | unsigned long next; | |
57250a5b | 2165 | unsigned long end = addr + size; |
aee16b3c JF |
2166 | int err; |
2167 | ||
9cb65bc3 MP |
2168 | if (WARN_ON(addr >= end)) |
2169 | return -EINVAL; | |
2170 | ||
aee16b3c JF |
2171 | pgd = pgd_offset(mm, addr); |
2172 | do { | |
2173 | next = pgd_addr_end(addr, end); | |
c2febafc | 2174 | err = apply_to_p4d_range(mm, pgd, addr, next, fn, data); |
aee16b3c JF |
2175 | if (err) |
2176 | break; | |
2177 | } while (pgd++, addr = next, addr != end); | |
57250a5b | 2178 | |
aee16b3c JF |
2179 | return err; |
2180 | } | |
2181 | EXPORT_SYMBOL_GPL(apply_to_page_range); | |
2182 | ||
8f4e2101 | 2183 | /* |
9b4bdd2f KS |
2184 | * handle_pte_fault chooses page fault handler according to an entry which was |
2185 | * read non-atomically. Before making any commitment, on those architectures | |
2186 | * or configurations (e.g. i386 with PAE) which might give a mix of unmatched | |
2187 | * parts, do_swap_page must check under lock before unmapping the pte and | |
2188 | * proceeding (but do_wp_page is only called after already making such a check; | |
a335b2e1 | 2189 | * and do_anonymous_page can safely check later on). |
8f4e2101 | 2190 | */ |
4c21e2f2 | 2191 | static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd, |
8f4e2101 HD |
2192 | pte_t *page_table, pte_t orig_pte) |
2193 | { | |
2194 | int same = 1; | |
2195 | #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT) | |
2196 | if (sizeof(pte_t) > sizeof(unsigned long)) { | |
4c21e2f2 HD |
2197 | spinlock_t *ptl = pte_lockptr(mm, pmd); |
2198 | spin_lock(ptl); | |
8f4e2101 | 2199 | same = pte_same(*page_table, orig_pte); |
4c21e2f2 | 2200 | spin_unlock(ptl); |
8f4e2101 HD |
2201 | } |
2202 | #endif | |
2203 | pte_unmap(page_table); | |
2204 | return same; | |
2205 | } | |
2206 | ||
9de455b2 | 2207 | static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma) |
6aab341e | 2208 | { |
0abdd7a8 DW |
2209 | debug_dma_assert_idle(src); |
2210 | ||
6aab341e LT |
2211 | /* |
2212 | * If the source page was a PFN mapping, we don't have | |
2213 | * a "struct page" for it. We do a best-effort copy by | |
2214 | * just copying from the original user address. If that | |
2215 | * fails, we just zero-fill it. Live with it. | |
2216 | */ | |
2217 | if (unlikely(!src)) { | |
9b04c5fe | 2218 | void *kaddr = kmap_atomic(dst); |
5d2a2dbb LT |
2219 | void __user *uaddr = (void __user *)(va & PAGE_MASK); |
2220 | ||
2221 | /* | |
2222 | * This really shouldn't fail, because the page is there | |
2223 | * in the page tables. But it might just be unreadable, | |
2224 | * in which case we just give up and fill the result with | |
2225 | * zeroes. | |
2226 | */ | |
2227 | if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) | |
3ecb01df | 2228 | clear_page(kaddr); |
9b04c5fe | 2229 | kunmap_atomic(kaddr); |
c4ec7b0d | 2230 | flush_dcache_page(dst); |
0ed361de NP |
2231 | } else |
2232 | copy_user_highpage(dst, src, va, vma); | |
6aab341e LT |
2233 | } |
2234 | ||
c20cd45e MH |
2235 | static gfp_t __get_fault_gfp_mask(struct vm_area_struct *vma) |
2236 | { | |
2237 | struct file *vm_file = vma->vm_file; | |
2238 | ||
2239 | if (vm_file) | |
2240 | return mapping_gfp_mask(vm_file->f_mapping) | __GFP_FS | __GFP_IO; | |
2241 | ||
2242 | /* | |
2243 | * Special mappings (e.g. VDSO) do not have any file so fake | |
2244 | * a default GFP_KERNEL for them. | |
2245 | */ | |
2246 | return GFP_KERNEL; | |
2247 | } | |
2248 | ||
fb09a464 KS |
2249 | /* |
2250 | * Notify the address space that the page is about to become writable so that | |
2251 | * it can prohibit this or wait for the page to get into an appropriate state. | |
2252 | * | |
2253 | * We do this without the lock held, so that it can sleep if it needs to. | |
2254 | */ | |
38b8cb7f | 2255 | static int do_page_mkwrite(struct vm_fault *vmf) |
fb09a464 | 2256 | { |
fb09a464 | 2257 | int ret; |
38b8cb7f JK |
2258 | struct page *page = vmf->page; |
2259 | unsigned int old_flags = vmf->flags; | |
fb09a464 | 2260 | |
38b8cb7f | 2261 | vmf->flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; |
fb09a464 | 2262 | |
11bac800 | 2263 | ret = vmf->vma->vm_ops->page_mkwrite(vmf); |
38b8cb7f JK |
2264 | /* Restore original flags so that caller is not surprised */ |
2265 | vmf->flags = old_flags; | |
fb09a464 KS |
2266 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) |
2267 | return ret; | |
2268 | if (unlikely(!(ret & VM_FAULT_LOCKED))) { | |
2269 | lock_page(page); | |
2270 | if (!page->mapping) { | |
2271 | unlock_page(page); | |
2272 | return 0; /* retry */ | |
2273 | } | |
2274 | ret |= VM_FAULT_LOCKED; | |
2275 | } else | |
2276 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
2277 | return ret; | |
2278 | } | |
2279 | ||
97ba0c2b JK |
2280 | /* |
2281 | * Handle dirtying of a page in shared file mapping on a write fault. | |
2282 | * | |
2283 | * The function expects the page to be locked and unlocks it. | |
2284 | */ | |
2285 | static void fault_dirty_shared_page(struct vm_area_struct *vma, | |
2286 | struct page *page) | |
2287 | { | |
2288 | struct address_space *mapping; | |
2289 | bool dirtied; | |
2290 | bool page_mkwrite = vma->vm_ops && vma->vm_ops->page_mkwrite; | |
2291 | ||
2292 | dirtied = set_page_dirty(page); | |
2293 | VM_BUG_ON_PAGE(PageAnon(page), page); | |
2294 | /* | |
2295 | * Take a local copy of the address_space - page.mapping may be zeroed | |
2296 | * by truncate after unlock_page(). The address_space itself remains | |
2297 | * pinned by vma->vm_file's reference. We rely on unlock_page()'s | |
2298 | * release semantics to prevent the compiler from undoing this copying. | |
2299 | */ | |
2300 | mapping = page_rmapping(page); | |
2301 | unlock_page(page); | |
2302 | ||
2303 | if ((dirtied || page_mkwrite) && mapping) { | |
2304 | /* | |
2305 | * Some device drivers do not set page.mapping | |
2306 | * but still dirty their pages | |
2307 | */ | |
2308 | balance_dirty_pages_ratelimited(mapping); | |
2309 | } | |
2310 | ||
2311 | if (!page_mkwrite) | |
2312 | file_update_time(vma->vm_file); | |
2313 | } | |
2314 | ||
4e047f89 SR |
2315 | /* |
2316 | * Handle write page faults for pages that can be reused in the current vma | |
2317 | * | |
2318 | * This can happen either due to the mapping being with the VM_SHARED flag, | |
2319 | * or due to us being the last reference standing to the page. In either | |
2320 | * case, all we need to do here is to mark the page as writable and update | |
2321 | * any related book-keeping. | |
2322 | */ | |
997dd98d | 2323 | static inline void wp_page_reuse(struct vm_fault *vmf) |
82b0f8c3 | 2324 | __releases(vmf->ptl) |
4e047f89 | 2325 | { |
82b0f8c3 | 2326 | struct vm_area_struct *vma = vmf->vma; |
a41b70d6 | 2327 | struct page *page = vmf->page; |
4e047f89 SR |
2328 | pte_t entry; |
2329 | /* | |
2330 | * Clear the pages cpupid information as the existing | |
2331 | * information potentially belongs to a now completely | |
2332 | * unrelated process. | |
2333 | */ | |
2334 | if (page) | |
2335 | page_cpupid_xchg_last(page, (1 << LAST_CPUPID_SHIFT) - 1); | |
2336 | ||
2994302b JK |
2337 | flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); |
2338 | entry = pte_mkyoung(vmf->orig_pte); | |
4e047f89 | 2339 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
82b0f8c3 JK |
2340 | if (ptep_set_access_flags(vma, vmf->address, vmf->pte, entry, 1)) |
2341 | update_mmu_cache(vma, vmf->address, vmf->pte); | |
2342 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
4e047f89 SR |
2343 | } |
2344 | ||
2f38ab2c SR |
2345 | /* |
2346 | * Handle the case of a page which we actually need to copy to a new page. | |
2347 | * | |
2348 | * Called with mmap_sem locked and the old page referenced, but | |
2349 | * without the ptl held. | |
2350 | * | |
2351 | * High level logic flow: | |
2352 | * | |
2353 | * - Allocate a page, copy the content of the old page to the new one. | |
2354 | * - Handle book keeping and accounting - cgroups, mmu-notifiers, etc. | |
2355 | * - Take the PTL. If the pte changed, bail out and release the allocated page | |
2356 | * - If the pte is still the way we remember it, update the page table and all | |
2357 | * relevant references. This includes dropping the reference the page-table | |
2358 | * held to the old page, as well as updating the rmap. | |
2359 | * - In any case, unlock the PTL and drop the reference we took to the old page. | |
2360 | */ | |
a41b70d6 | 2361 | static int wp_page_copy(struct vm_fault *vmf) |
2f38ab2c | 2362 | { |
82b0f8c3 | 2363 | struct vm_area_struct *vma = vmf->vma; |
bae473a4 | 2364 | struct mm_struct *mm = vma->vm_mm; |
a41b70d6 | 2365 | struct page *old_page = vmf->page; |
2f38ab2c | 2366 | struct page *new_page = NULL; |
2f38ab2c SR |
2367 | pte_t entry; |
2368 | int page_copied = 0; | |
82b0f8c3 | 2369 | const unsigned long mmun_start = vmf->address & PAGE_MASK; |
bae473a4 | 2370 | const unsigned long mmun_end = mmun_start + PAGE_SIZE; |
2f38ab2c SR |
2371 | struct mem_cgroup *memcg; |
2372 | ||
2373 | if (unlikely(anon_vma_prepare(vma))) | |
2374 | goto oom; | |
2375 | ||
2994302b | 2376 | if (is_zero_pfn(pte_pfn(vmf->orig_pte))) { |
82b0f8c3 JK |
2377 | new_page = alloc_zeroed_user_highpage_movable(vma, |
2378 | vmf->address); | |
2f38ab2c SR |
2379 | if (!new_page) |
2380 | goto oom; | |
2381 | } else { | |
bae473a4 | 2382 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, |
82b0f8c3 | 2383 | vmf->address); |
2f38ab2c SR |
2384 | if (!new_page) |
2385 | goto oom; | |
82b0f8c3 | 2386 | cow_user_page(new_page, old_page, vmf->address, vma); |
2f38ab2c | 2387 | } |
2f38ab2c | 2388 | |
f627c2f5 | 2389 | if (mem_cgroup_try_charge(new_page, mm, GFP_KERNEL, &memcg, false)) |
2f38ab2c SR |
2390 | goto oom_free_new; |
2391 | ||
eb3c24f3 MG |
2392 | __SetPageUptodate(new_page); |
2393 | ||
2f38ab2c SR |
2394 | mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); |
2395 | ||
2396 | /* | |
2397 | * Re-check the pte - we dropped the lock | |
2398 | */ | |
82b0f8c3 | 2399 | vmf->pte = pte_offset_map_lock(mm, vmf->pmd, vmf->address, &vmf->ptl); |
2994302b | 2400 | if (likely(pte_same(*vmf->pte, vmf->orig_pte))) { |
2f38ab2c SR |
2401 | if (old_page) { |
2402 | if (!PageAnon(old_page)) { | |
eca56ff9 JM |
2403 | dec_mm_counter_fast(mm, |
2404 | mm_counter_file(old_page)); | |
2f38ab2c SR |
2405 | inc_mm_counter_fast(mm, MM_ANONPAGES); |
2406 | } | |
2407 | } else { | |
2408 | inc_mm_counter_fast(mm, MM_ANONPAGES); | |
2409 | } | |
2994302b | 2410 | flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); |
2f38ab2c SR |
2411 | entry = mk_pte(new_page, vma->vm_page_prot); |
2412 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
2413 | /* | |
2414 | * Clear the pte entry and flush it first, before updating the | |
2415 | * pte with the new entry. This will avoid a race condition | |
2416 | * seen in the presence of one thread doing SMC and another | |
2417 | * thread doing COW. | |
2418 | */ | |
82b0f8c3 JK |
2419 | ptep_clear_flush_notify(vma, vmf->address, vmf->pte); |
2420 | page_add_new_anon_rmap(new_page, vma, vmf->address, false); | |
f627c2f5 | 2421 | mem_cgroup_commit_charge(new_page, memcg, false, false); |
2f38ab2c SR |
2422 | lru_cache_add_active_or_unevictable(new_page, vma); |
2423 | /* | |
2424 | * We call the notify macro here because, when using secondary | |
2425 | * mmu page tables (such as kvm shadow page tables), we want the | |
2426 | * new page to be mapped directly into the secondary page table. | |
2427 | */ | |
82b0f8c3 JK |
2428 | set_pte_at_notify(mm, vmf->address, vmf->pte, entry); |
2429 | update_mmu_cache(vma, vmf->address, vmf->pte); | |
2f38ab2c SR |
2430 | if (old_page) { |
2431 | /* | |
2432 | * Only after switching the pte to the new page may | |
2433 | * we remove the mapcount here. Otherwise another | |
2434 | * process may come and find the rmap count decremented | |
2435 | * before the pte is switched to the new page, and | |
2436 | * "reuse" the old page writing into it while our pte | |
2437 | * here still points into it and can be read by other | |
2438 | * threads. | |
2439 | * | |
2440 | * The critical issue is to order this | |
2441 | * page_remove_rmap with the ptp_clear_flush above. | |
2442 | * Those stores are ordered by (if nothing else,) | |
2443 | * the barrier present in the atomic_add_negative | |
2444 | * in page_remove_rmap. | |
2445 | * | |
2446 | * Then the TLB flush in ptep_clear_flush ensures that | |
2447 | * no process can access the old page before the | |
2448 | * decremented mapcount is visible. And the old page | |
2449 | * cannot be reused until after the decremented | |
2450 | * mapcount is visible. So transitively, TLBs to | |
2451 | * old page will be flushed before it can be reused. | |
2452 | */ | |
d281ee61 | 2453 | page_remove_rmap(old_page, false); |
2f38ab2c SR |
2454 | } |
2455 | ||
2456 | /* Free the old page.. */ | |
2457 | new_page = old_page; | |
2458 | page_copied = 1; | |
2459 | } else { | |
f627c2f5 | 2460 | mem_cgroup_cancel_charge(new_page, memcg, false); |
2f38ab2c SR |
2461 | } |
2462 | ||
2463 | if (new_page) | |
09cbfeaf | 2464 | put_page(new_page); |
2f38ab2c | 2465 | |
82b0f8c3 | 2466 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
2f38ab2c SR |
2467 | mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); |
2468 | if (old_page) { | |
2469 | /* | |
2470 | * Don't let another task, with possibly unlocked vma, | |
2471 | * keep the mlocked page. | |
2472 | */ | |
2473 | if (page_copied && (vma->vm_flags & VM_LOCKED)) { | |
2474 | lock_page(old_page); /* LRU manipulation */ | |
e90309c9 KS |
2475 | if (PageMlocked(old_page)) |
2476 | munlock_vma_page(old_page); | |
2f38ab2c SR |
2477 | unlock_page(old_page); |
2478 | } | |
09cbfeaf | 2479 | put_page(old_page); |
2f38ab2c SR |
2480 | } |
2481 | return page_copied ? VM_FAULT_WRITE : 0; | |
2482 | oom_free_new: | |
09cbfeaf | 2483 | put_page(new_page); |
2f38ab2c SR |
2484 | oom: |
2485 | if (old_page) | |
09cbfeaf | 2486 | put_page(old_page); |
2f38ab2c SR |
2487 | return VM_FAULT_OOM; |
2488 | } | |
2489 | ||
66a6197c JK |
2490 | /** |
2491 | * finish_mkwrite_fault - finish page fault for a shared mapping, making PTE | |
2492 | * writeable once the page is prepared | |
2493 | * | |
2494 | * @vmf: structure describing the fault | |
2495 | * | |
2496 | * This function handles all that is needed to finish a write page fault in a | |
2497 | * shared mapping due to PTE being read-only once the mapped page is prepared. | |
2498 | * It handles locking of PTE and modifying it. The function returns | |
2499 | * VM_FAULT_WRITE on success, 0 when PTE got changed before we acquired PTE | |
2500 | * lock. | |
2501 | * | |
2502 | * The function expects the page to be locked or other protection against | |
2503 | * concurrent faults / writeback (such as DAX radix tree locks). | |
2504 | */ | |
2505 | int finish_mkwrite_fault(struct vm_fault *vmf) | |
2506 | { | |
2507 | WARN_ON_ONCE(!(vmf->vma->vm_flags & VM_SHARED)); | |
2508 | vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, vmf->address, | |
2509 | &vmf->ptl); | |
2510 | /* | |
2511 | * We might have raced with another page fault while we released the | |
2512 | * pte_offset_map_lock. | |
2513 | */ | |
2514 | if (!pte_same(*vmf->pte, vmf->orig_pte)) { | |
2515 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
a19e2553 | 2516 | return VM_FAULT_NOPAGE; |
66a6197c JK |
2517 | } |
2518 | wp_page_reuse(vmf); | |
a19e2553 | 2519 | return 0; |
66a6197c JK |
2520 | } |
2521 | ||
dd906184 BH |
2522 | /* |
2523 | * Handle write page faults for VM_MIXEDMAP or VM_PFNMAP for a VM_SHARED | |
2524 | * mapping | |
2525 | */ | |
2994302b | 2526 | static int wp_pfn_shared(struct vm_fault *vmf) |
dd906184 | 2527 | { |
82b0f8c3 | 2528 | struct vm_area_struct *vma = vmf->vma; |
bae473a4 | 2529 | |
dd906184 | 2530 | if (vma->vm_ops && vma->vm_ops->pfn_mkwrite) { |
dd906184 BH |
2531 | int ret; |
2532 | ||
82b0f8c3 | 2533 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
fe82221f | 2534 | vmf->flags |= FAULT_FLAG_MKWRITE; |
11bac800 | 2535 | ret = vma->vm_ops->pfn_mkwrite(vmf); |
2f89dc12 | 2536 | if (ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)) |
dd906184 | 2537 | return ret; |
66a6197c | 2538 | return finish_mkwrite_fault(vmf); |
dd906184 | 2539 | } |
997dd98d JK |
2540 | wp_page_reuse(vmf); |
2541 | return VM_FAULT_WRITE; | |
dd906184 BH |
2542 | } |
2543 | ||
a41b70d6 | 2544 | static int wp_page_shared(struct vm_fault *vmf) |
82b0f8c3 | 2545 | __releases(vmf->ptl) |
93e478d4 | 2546 | { |
82b0f8c3 | 2547 | struct vm_area_struct *vma = vmf->vma; |
93e478d4 | 2548 | |
a41b70d6 | 2549 | get_page(vmf->page); |
93e478d4 | 2550 | |
93e478d4 SR |
2551 | if (vma->vm_ops && vma->vm_ops->page_mkwrite) { |
2552 | int tmp; | |
2553 | ||
82b0f8c3 | 2554 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
38b8cb7f | 2555 | tmp = do_page_mkwrite(vmf); |
93e478d4 SR |
2556 | if (unlikely(!tmp || (tmp & |
2557 | (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { | |
a41b70d6 | 2558 | put_page(vmf->page); |
93e478d4 SR |
2559 | return tmp; |
2560 | } | |
66a6197c | 2561 | tmp = finish_mkwrite_fault(vmf); |
a19e2553 | 2562 | if (unlikely(tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { |
a41b70d6 | 2563 | unlock_page(vmf->page); |
a41b70d6 | 2564 | put_page(vmf->page); |
66a6197c | 2565 | return tmp; |
93e478d4 | 2566 | } |
66a6197c JK |
2567 | } else { |
2568 | wp_page_reuse(vmf); | |
997dd98d | 2569 | lock_page(vmf->page); |
93e478d4 | 2570 | } |
997dd98d JK |
2571 | fault_dirty_shared_page(vma, vmf->page); |
2572 | put_page(vmf->page); | |
93e478d4 | 2573 | |
997dd98d | 2574 | return VM_FAULT_WRITE; |
93e478d4 SR |
2575 | } |
2576 | ||
1da177e4 LT |
2577 | /* |
2578 | * This routine handles present pages, when users try to write | |
2579 | * to a shared page. It is done by copying the page to a new address | |
2580 | * and decrementing the shared-page counter for the old page. | |
2581 | * | |
1da177e4 LT |
2582 | * Note that this routine assumes that the protection checks have been |
2583 | * done by the caller (the low-level page fault routine in most cases). | |
2584 | * Thus we can safely just mark it writable once we've done any necessary | |
2585 | * COW. | |
2586 | * | |
2587 | * We also mark the page dirty at this point even though the page will | |
2588 | * change only once the write actually happens. This avoids a few races, | |
2589 | * and potentially makes it more efficient. | |
2590 | * | |
8f4e2101 HD |
2591 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2592 | * but allow concurrent faults), with pte both mapped and locked. | |
2593 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2594 | */ |
2994302b | 2595 | static int do_wp_page(struct vm_fault *vmf) |
82b0f8c3 | 2596 | __releases(vmf->ptl) |
1da177e4 | 2597 | { |
82b0f8c3 | 2598 | struct vm_area_struct *vma = vmf->vma; |
1da177e4 | 2599 | |
a41b70d6 JK |
2600 | vmf->page = vm_normal_page(vma, vmf->address, vmf->orig_pte); |
2601 | if (!vmf->page) { | |
251b97f5 | 2602 | /* |
64e45507 PF |
2603 | * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a |
2604 | * VM_PFNMAP VMA. | |
251b97f5 PZ |
2605 | * |
2606 | * We should not cow pages in a shared writeable mapping. | |
dd906184 | 2607 | * Just mark the pages writable and/or call ops->pfn_mkwrite. |
251b97f5 PZ |
2608 | */ |
2609 | if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) == | |
2610 | (VM_WRITE|VM_SHARED)) | |
2994302b | 2611 | return wp_pfn_shared(vmf); |
2f38ab2c | 2612 | |
82b0f8c3 | 2613 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 2614 | return wp_page_copy(vmf); |
251b97f5 | 2615 | } |
1da177e4 | 2616 | |
d08b3851 | 2617 | /* |
ee6a6457 PZ |
2618 | * Take out anonymous pages first, anonymous shared vmas are |
2619 | * not dirty accountable. | |
d08b3851 | 2620 | */ |
a41b70d6 | 2621 | if (PageAnon(vmf->page) && !PageKsm(vmf->page)) { |
ba3c4ce6 | 2622 | int total_map_swapcount; |
a41b70d6 JK |
2623 | if (!trylock_page(vmf->page)) { |
2624 | get_page(vmf->page); | |
82b0f8c3 | 2625 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 2626 | lock_page(vmf->page); |
82b0f8c3 JK |
2627 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
2628 | vmf->address, &vmf->ptl); | |
2994302b | 2629 | if (!pte_same(*vmf->pte, vmf->orig_pte)) { |
a41b70d6 | 2630 | unlock_page(vmf->page); |
82b0f8c3 | 2631 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 2632 | put_page(vmf->page); |
28766805 | 2633 | return 0; |
ab967d86 | 2634 | } |
a41b70d6 | 2635 | put_page(vmf->page); |
ee6a6457 | 2636 | } |
ba3c4ce6 HY |
2637 | if (reuse_swap_page(vmf->page, &total_map_swapcount)) { |
2638 | if (total_map_swapcount == 1) { | |
6d0a07ed AA |
2639 | /* |
2640 | * The page is all ours. Move it to | |
2641 | * our anon_vma so the rmap code will | |
2642 | * not search our parent or siblings. | |
2643 | * Protected against the rmap code by | |
2644 | * the page lock. | |
2645 | */ | |
a41b70d6 | 2646 | page_move_anon_rmap(vmf->page, vma); |
6d0a07ed | 2647 | } |
a41b70d6 | 2648 | unlock_page(vmf->page); |
997dd98d JK |
2649 | wp_page_reuse(vmf); |
2650 | return VM_FAULT_WRITE; | |
b009c024 | 2651 | } |
a41b70d6 | 2652 | unlock_page(vmf->page); |
ee6a6457 | 2653 | } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) == |
d08b3851 | 2654 | (VM_WRITE|VM_SHARED))) { |
a41b70d6 | 2655 | return wp_page_shared(vmf); |
1da177e4 | 2656 | } |
1da177e4 LT |
2657 | |
2658 | /* | |
2659 | * Ok, we need to copy. Oh, well.. | |
2660 | */ | |
a41b70d6 | 2661 | get_page(vmf->page); |
28766805 | 2662 | |
82b0f8c3 | 2663 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 2664 | return wp_page_copy(vmf); |
1da177e4 LT |
2665 | } |
2666 | ||
97a89413 | 2667 | static void unmap_mapping_range_vma(struct vm_area_struct *vma, |
1da177e4 LT |
2668 | unsigned long start_addr, unsigned long end_addr, |
2669 | struct zap_details *details) | |
2670 | { | |
f5cc4eef | 2671 | zap_page_range_single(vma, start_addr, end_addr - start_addr, details); |
1da177e4 LT |
2672 | } |
2673 | ||
6b2dbba8 | 2674 | static inline void unmap_mapping_range_tree(struct rb_root *root, |
1da177e4 LT |
2675 | struct zap_details *details) |
2676 | { | |
2677 | struct vm_area_struct *vma; | |
1da177e4 LT |
2678 | pgoff_t vba, vea, zba, zea; |
2679 | ||
6b2dbba8 | 2680 | vma_interval_tree_foreach(vma, root, |
1da177e4 | 2681 | details->first_index, details->last_index) { |
1da177e4 LT |
2682 | |
2683 | vba = vma->vm_pgoff; | |
d6e93217 | 2684 | vea = vba + vma_pages(vma) - 1; |
1da177e4 LT |
2685 | zba = details->first_index; |
2686 | if (zba < vba) | |
2687 | zba = vba; | |
2688 | zea = details->last_index; | |
2689 | if (zea > vea) | |
2690 | zea = vea; | |
2691 | ||
97a89413 | 2692 | unmap_mapping_range_vma(vma, |
1da177e4 LT |
2693 | ((zba - vba) << PAGE_SHIFT) + vma->vm_start, |
2694 | ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, | |
97a89413 | 2695 | details); |
1da177e4 LT |
2696 | } |
2697 | } | |
2698 | ||
1da177e4 | 2699 | /** |
8a5f14a2 KS |
2700 | * unmap_mapping_range - unmap the portion of all mmaps in the specified |
2701 | * address_space corresponding to the specified page range in the underlying | |
2702 | * file. | |
2703 | * | |
3d41088f | 2704 | * @mapping: the address space containing mmaps to be unmapped. |
1da177e4 LT |
2705 | * @holebegin: byte in first page to unmap, relative to the start of |
2706 | * the underlying file. This will be rounded down to a PAGE_SIZE | |
25d9e2d1 | 2707 | * boundary. Note that this is different from truncate_pagecache(), which |
1da177e4 LT |
2708 | * must keep the partial page. In contrast, we must get rid of |
2709 | * partial pages. | |
2710 | * @holelen: size of prospective hole in bytes. This will be rounded | |
2711 | * up to a PAGE_SIZE boundary. A holelen of zero truncates to the | |
2712 | * end of the file. | |
2713 | * @even_cows: 1 when truncating a file, unmap even private COWed pages; | |
2714 | * but 0 when invalidating pagecache, don't throw away private data. | |
2715 | */ | |
2716 | void unmap_mapping_range(struct address_space *mapping, | |
2717 | loff_t const holebegin, loff_t const holelen, int even_cows) | |
2718 | { | |
aac45363 | 2719 | struct zap_details details = { }; |
1da177e4 LT |
2720 | pgoff_t hba = holebegin >> PAGE_SHIFT; |
2721 | pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
2722 | ||
2723 | /* Check for overflow. */ | |
2724 | if (sizeof(holelen) > sizeof(hlen)) { | |
2725 | long long holeend = | |
2726 | (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
2727 | if (holeend & ~(long long)ULONG_MAX) | |
2728 | hlen = ULONG_MAX - hba + 1; | |
2729 | } | |
2730 | ||
166f61b9 | 2731 | details.check_mapping = even_cows ? NULL : mapping; |
1da177e4 LT |
2732 | details.first_index = hba; |
2733 | details.last_index = hba + hlen - 1; | |
2734 | if (details.last_index < details.first_index) | |
2735 | details.last_index = ULONG_MAX; | |
1da177e4 | 2736 | |
46c043ed | 2737 | i_mmap_lock_write(mapping); |
6b2dbba8 | 2738 | if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap))) |
1da177e4 | 2739 | unmap_mapping_range_tree(&mapping->i_mmap, &details); |
46c043ed | 2740 | i_mmap_unlock_write(mapping); |
1da177e4 LT |
2741 | } |
2742 | EXPORT_SYMBOL(unmap_mapping_range); | |
2743 | ||
1da177e4 | 2744 | /* |
8f4e2101 HD |
2745 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2746 | * but allow concurrent faults), and pte mapped but not yet locked. | |
9a95f3cf PC |
2747 | * We return with pte unmapped and unlocked. |
2748 | * | |
2749 | * We return with the mmap_sem locked or unlocked in the same cases | |
2750 | * as does filemap_fault(). | |
1da177e4 | 2751 | */ |
2994302b | 2752 | int do_swap_page(struct vm_fault *vmf) |
1da177e4 | 2753 | { |
82b0f8c3 | 2754 | struct vm_area_struct *vma = vmf->vma; |
ec560175 | 2755 | struct page *page = NULL, *swapcache; |
00501b53 | 2756 | struct mem_cgroup *memcg; |
ec560175 | 2757 | struct vma_swap_readahead swap_ra; |
65500d23 | 2758 | swp_entry_t entry; |
1da177e4 | 2759 | pte_t pte; |
d065bd81 | 2760 | int locked; |
ad8c2ee8 | 2761 | int exclusive = 0; |
83c54070 | 2762 | int ret = 0; |
ec560175 | 2763 | bool vma_readahead = swap_use_vma_readahead(); |
1da177e4 | 2764 | |
ec560175 HY |
2765 | if (vma_readahead) |
2766 | page = swap_readahead_detect(vmf, &swap_ra); | |
2767 | if (!pte_unmap_same(vma->vm_mm, vmf->pmd, vmf->pte, vmf->orig_pte)) { | |
2768 | if (page) | |
2769 | put_page(page); | |
8f4e2101 | 2770 | goto out; |
ec560175 | 2771 | } |
65500d23 | 2772 | |
2994302b | 2773 | entry = pte_to_swp_entry(vmf->orig_pte); |
d1737fdb AK |
2774 | if (unlikely(non_swap_entry(entry))) { |
2775 | if (is_migration_entry(entry)) { | |
82b0f8c3 JK |
2776 | migration_entry_wait(vma->vm_mm, vmf->pmd, |
2777 | vmf->address); | |
d1737fdb AK |
2778 | } else if (is_hwpoison_entry(entry)) { |
2779 | ret = VM_FAULT_HWPOISON; | |
2780 | } else { | |
2994302b | 2781 | print_bad_pte(vma, vmf->address, vmf->orig_pte, NULL); |
d99be1a8 | 2782 | ret = VM_FAULT_SIGBUS; |
d1737fdb | 2783 | } |
0697212a CL |
2784 | goto out; |
2785 | } | |
0ff92245 | 2786 | delayacct_set_flag(DELAYACCT_PF_SWAPIN); |
ec560175 HY |
2787 | if (!page) |
2788 | page = lookup_swap_cache(entry, vma_readahead ? vma : NULL, | |
2789 | vmf->address); | |
1da177e4 | 2790 | if (!page) { |
ec560175 HY |
2791 | if (vma_readahead) |
2792 | page = do_swap_page_readahead(entry, | |
2793 | GFP_HIGHUSER_MOVABLE, vmf, &swap_ra); | |
2794 | else | |
2795 | page = swapin_readahead(entry, | |
2796 | GFP_HIGHUSER_MOVABLE, vma, vmf->address); | |
1da177e4 LT |
2797 | if (!page) { |
2798 | /* | |
8f4e2101 HD |
2799 | * Back out if somebody else faulted in this pte |
2800 | * while we released the pte lock. | |
1da177e4 | 2801 | */ |
82b0f8c3 JK |
2802 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
2803 | vmf->address, &vmf->ptl); | |
2994302b | 2804 | if (likely(pte_same(*vmf->pte, vmf->orig_pte))) |
1da177e4 | 2805 | ret = VM_FAULT_OOM; |
0ff92245 | 2806 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
65500d23 | 2807 | goto unlock; |
1da177e4 LT |
2808 | } |
2809 | ||
2810 | /* Had to read the page from swap area: Major fault */ | |
2811 | ret = VM_FAULT_MAJOR; | |
f8891e5e | 2812 | count_vm_event(PGMAJFAULT); |
2262185c | 2813 | count_memcg_event_mm(vma->vm_mm, PGMAJFAULT); |
d1737fdb | 2814 | } else if (PageHWPoison(page)) { |
71f72525 WF |
2815 | /* |
2816 | * hwpoisoned dirty swapcache pages are kept for killing | |
2817 | * owner processes (which may be unknown at hwpoison time) | |
2818 | */ | |
d1737fdb AK |
2819 | ret = VM_FAULT_HWPOISON; |
2820 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); | |
56f31801 | 2821 | swapcache = page; |
4779cb31 | 2822 | goto out_release; |
1da177e4 LT |
2823 | } |
2824 | ||
56f31801 | 2825 | swapcache = page; |
82b0f8c3 | 2826 | locked = lock_page_or_retry(page, vma->vm_mm, vmf->flags); |
e709ffd6 | 2827 | |
073e587e | 2828 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
d065bd81 ML |
2829 | if (!locked) { |
2830 | ret |= VM_FAULT_RETRY; | |
2831 | goto out_release; | |
2832 | } | |
073e587e | 2833 | |
4969c119 | 2834 | /* |
31c4a3d3 HD |
2835 | * Make sure try_to_free_swap or reuse_swap_page or swapoff did not |
2836 | * release the swapcache from under us. The page pin, and pte_same | |
2837 | * test below, are not enough to exclude that. Even if it is still | |
2838 | * swapcache, we need to check that the page's swap has not changed. | |
4969c119 | 2839 | */ |
31c4a3d3 | 2840 | if (unlikely(!PageSwapCache(page) || page_private(page) != entry.val)) |
4969c119 AA |
2841 | goto out_page; |
2842 | ||
82b0f8c3 | 2843 | page = ksm_might_need_to_copy(page, vma, vmf->address); |
cbf86cfe HD |
2844 | if (unlikely(!page)) { |
2845 | ret = VM_FAULT_OOM; | |
2846 | page = swapcache; | |
cbf86cfe | 2847 | goto out_page; |
5ad64688 HD |
2848 | } |
2849 | ||
bae473a4 KS |
2850 | if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL, |
2851 | &memcg, false)) { | |
8a9f3ccd | 2852 | ret = VM_FAULT_OOM; |
bc43f75c | 2853 | goto out_page; |
8a9f3ccd BS |
2854 | } |
2855 | ||
1da177e4 | 2856 | /* |
8f4e2101 | 2857 | * Back out if somebody else already faulted in this pte. |
1da177e4 | 2858 | */ |
82b0f8c3 JK |
2859 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, |
2860 | &vmf->ptl); | |
2994302b | 2861 | if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) |
b8107480 | 2862 | goto out_nomap; |
b8107480 KK |
2863 | |
2864 | if (unlikely(!PageUptodate(page))) { | |
2865 | ret = VM_FAULT_SIGBUS; | |
2866 | goto out_nomap; | |
1da177e4 LT |
2867 | } |
2868 | ||
8c7c6e34 KH |
2869 | /* |
2870 | * The page isn't present yet, go ahead with the fault. | |
2871 | * | |
2872 | * Be careful about the sequence of operations here. | |
2873 | * To get its accounting right, reuse_swap_page() must be called | |
2874 | * while the page is counted on swap but not yet in mapcount i.e. | |
2875 | * before page_add_anon_rmap() and swap_free(); try_to_free_swap() | |
2876 | * must be called after the swap_free(), or it will never succeed. | |
8c7c6e34 | 2877 | */ |
1da177e4 | 2878 | |
bae473a4 KS |
2879 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
2880 | dec_mm_counter_fast(vma->vm_mm, MM_SWAPENTS); | |
1da177e4 | 2881 | pte = mk_pte(page, vma->vm_page_prot); |
82b0f8c3 | 2882 | if ((vmf->flags & FAULT_FLAG_WRITE) && reuse_swap_page(page, NULL)) { |
1da177e4 | 2883 | pte = maybe_mkwrite(pte_mkdirty(pte), vma); |
82b0f8c3 | 2884 | vmf->flags &= ~FAULT_FLAG_WRITE; |
9a5b489b | 2885 | ret |= VM_FAULT_WRITE; |
d281ee61 | 2886 | exclusive = RMAP_EXCLUSIVE; |
1da177e4 | 2887 | } |
1da177e4 | 2888 | flush_icache_page(vma, page); |
2994302b | 2889 | if (pte_swp_soft_dirty(vmf->orig_pte)) |
179ef71c | 2890 | pte = pte_mksoft_dirty(pte); |
82b0f8c3 | 2891 | set_pte_at(vma->vm_mm, vmf->address, vmf->pte, pte); |
2994302b | 2892 | vmf->orig_pte = pte; |
00501b53 | 2893 | if (page == swapcache) { |
82b0f8c3 | 2894 | do_page_add_anon_rmap(page, vma, vmf->address, exclusive); |
f627c2f5 | 2895 | mem_cgroup_commit_charge(page, memcg, true, false); |
1a8018fb | 2896 | activate_page(page); |
00501b53 | 2897 | } else { /* ksm created a completely new copy */ |
82b0f8c3 | 2898 | page_add_new_anon_rmap(page, vma, vmf->address, false); |
f627c2f5 | 2899 | mem_cgroup_commit_charge(page, memcg, false, false); |
00501b53 JW |
2900 | lru_cache_add_active_or_unevictable(page, vma); |
2901 | } | |
1da177e4 | 2902 | |
c475a8ab | 2903 | swap_free(entry); |
5ccc5aba VD |
2904 | if (mem_cgroup_swap_full(page) || |
2905 | (vma->vm_flags & VM_LOCKED) || PageMlocked(page)) | |
a2c43eed | 2906 | try_to_free_swap(page); |
c475a8ab | 2907 | unlock_page(page); |
56f31801 | 2908 | if (page != swapcache) { |
4969c119 AA |
2909 | /* |
2910 | * Hold the lock to avoid the swap entry to be reused | |
2911 | * until we take the PT lock for the pte_same() check | |
2912 | * (to avoid false positives from pte_same). For | |
2913 | * further safety release the lock after the swap_free | |
2914 | * so that the swap count won't change under a | |
2915 | * parallel locked swapcache. | |
2916 | */ | |
2917 | unlock_page(swapcache); | |
09cbfeaf | 2918 | put_page(swapcache); |
4969c119 | 2919 | } |
c475a8ab | 2920 | |
82b0f8c3 | 2921 | if (vmf->flags & FAULT_FLAG_WRITE) { |
2994302b | 2922 | ret |= do_wp_page(vmf); |
61469f1d HD |
2923 | if (ret & VM_FAULT_ERROR) |
2924 | ret &= VM_FAULT_ERROR; | |
1da177e4 LT |
2925 | goto out; |
2926 | } | |
2927 | ||
2928 | /* No need to invalidate - it was non-present before */ | |
82b0f8c3 | 2929 | update_mmu_cache(vma, vmf->address, vmf->pte); |
65500d23 | 2930 | unlock: |
82b0f8c3 | 2931 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
1da177e4 LT |
2932 | out: |
2933 | return ret; | |
b8107480 | 2934 | out_nomap: |
f627c2f5 | 2935 | mem_cgroup_cancel_charge(page, memcg, false); |
82b0f8c3 | 2936 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
bc43f75c | 2937 | out_page: |
b8107480 | 2938 | unlock_page(page); |
4779cb31 | 2939 | out_release: |
09cbfeaf | 2940 | put_page(page); |
56f31801 | 2941 | if (page != swapcache) { |
4969c119 | 2942 | unlock_page(swapcache); |
09cbfeaf | 2943 | put_page(swapcache); |
4969c119 | 2944 | } |
65500d23 | 2945 | return ret; |
1da177e4 LT |
2946 | } |
2947 | ||
2948 | /* | |
8f4e2101 HD |
2949 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2950 | * but allow concurrent faults), and pte mapped but not yet locked. | |
2951 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2952 | */ |
82b0f8c3 | 2953 | static int do_anonymous_page(struct vm_fault *vmf) |
1da177e4 | 2954 | { |
82b0f8c3 | 2955 | struct vm_area_struct *vma = vmf->vma; |
00501b53 | 2956 | struct mem_cgroup *memcg; |
8f4e2101 | 2957 | struct page *page; |
6b31d595 | 2958 | int ret = 0; |
1da177e4 | 2959 | pte_t entry; |
1da177e4 | 2960 | |
6b7339f4 KS |
2961 | /* File mapping without ->vm_ops ? */ |
2962 | if (vma->vm_flags & VM_SHARED) | |
2963 | return VM_FAULT_SIGBUS; | |
2964 | ||
7267ec00 KS |
2965 | /* |
2966 | * Use pte_alloc() instead of pte_alloc_map(). We can't run | |
2967 | * pte_offset_map() on pmds where a huge pmd might be created | |
2968 | * from a different thread. | |
2969 | * | |
2970 | * pte_alloc_map() is safe to use under down_write(mmap_sem) or when | |
2971 | * parallel threads are excluded by other means. | |
2972 | * | |
2973 | * Here we only have down_read(mmap_sem). | |
2974 | */ | |
82b0f8c3 | 2975 | if (pte_alloc(vma->vm_mm, vmf->pmd, vmf->address)) |
7267ec00 KS |
2976 | return VM_FAULT_OOM; |
2977 | ||
2978 | /* See the comment in pte_alloc_one_map() */ | |
82b0f8c3 | 2979 | if (unlikely(pmd_trans_unstable(vmf->pmd))) |
7267ec00 KS |
2980 | return 0; |
2981 | ||
11ac5524 | 2982 | /* Use the zero-page for reads */ |
82b0f8c3 | 2983 | if (!(vmf->flags & FAULT_FLAG_WRITE) && |
bae473a4 | 2984 | !mm_forbids_zeropage(vma->vm_mm)) { |
82b0f8c3 | 2985 | entry = pte_mkspecial(pfn_pte(my_zero_pfn(vmf->address), |
62eede62 | 2986 | vma->vm_page_prot)); |
82b0f8c3 JK |
2987 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
2988 | vmf->address, &vmf->ptl); | |
2989 | if (!pte_none(*vmf->pte)) | |
a13ea5b7 | 2990 | goto unlock; |
6b31d595 MH |
2991 | ret = check_stable_address_space(vma->vm_mm); |
2992 | if (ret) | |
2993 | goto unlock; | |
6b251fc9 AA |
2994 | /* Deliver the page fault to userland, check inside PT lock */ |
2995 | if (userfaultfd_missing(vma)) { | |
82b0f8c3 JK |
2996 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
2997 | return handle_userfault(vmf, VM_UFFD_MISSING); | |
6b251fc9 | 2998 | } |
a13ea5b7 HD |
2999 | goto setpte; |
3000 | } | |
3001 | ||
557ed1fa | 3002 | /* Allocate our own private page. */ |
557ed1fa NP |
3003 | if (unlikely(anon_vma_prepare(vma))) |
3004 | goto oom; | |
82b0f8c3 | 3005 | page = alloc_zeroed_user_highpage_movable(vma, vmf->address); |
557ed1fa NP |
3006 | if (!page) |
3007 | goto oom; | |
eb3c24f3 | 3008 | |
bae473a4 | 3009 | if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL, &memcg, false)) |
eb3c24f3 MG |
3010 | goto oom_free_page; |
3011 | ||
52f37629 MK |
3012 | /* |
3013 | * The memory barrier inside __SetPageUptodate makes sure that | |
3014 | * preceeding stores to the page contents become visible before | |
3015 | * the set_pte_at() write. | |
3016 | */ | |
0ed361de | 3017 | __SetPageUptodate(page); |
8f4e2101 | 3018 | |
557ed1fa | 3019 | entry = mk_pte(page, vma->vm_page_prot); |
1ac0cb5d HD |
3020 | if (vma->vm_flags & VM_WRITE) |
3021 | entry = pte_mkwrite(pte_mkdirty(entry)); | |
1da177e4 | 3022 | |
82b0f8c3 JK |
3023 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, |
3024 | &vmf->ptl); | |
3025 | if (!pte_none(*vmf->pte)) | |
557ed1fa | 3026 | goto release; |
9ba69294 | 3027 | |
6b31d595 MH |
3028 | ret = check_stable_address_space(vma->vm_mm); |
3029 | if (ret) | |
3030 | goto release; | |
3031 | ||
6b251fc9 AA |
3032 | /* Deliver the page fault to userland, check inside PT lock */ |
3033 | if (userfaultfd_missing(vma)) { | |
82b0f8c3 | 3034 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
f627c2f5 | 3035 | mem_cgroup_cancel_charge(page, memcg, false); |
09cbfeaf | 3036 | put_page(page); |
82b0f8c3 | 3037 | return handle_userfault(vmf, VM_UFFD_MISSING); |
6b251fc9 AA |
3038 | } |
3039 | ||
bae473a4 | 3040 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
82b0f8c3 | 3041 | page_add_new_anon_rmap(page, vma, vmf->address, false); |
f627c2f5 | 3042 | mem_cgroup_commit_charge(page, memcg, false, false); |
00501b53 | 3043 | lru_cache_add_active_or_unevictable(page, vma); |
a13ea5b7 | 3044 | setpte: |
82b0f8c3 | 3045 | set_pte_at(vma->vm_mm, vmf->address, vmf->pte, entry); |
1da177e4 LT |
3046 | |
3047 | /* No need to invalidate - it was non-present before */ | |
82b0f8c3 | 3048 | update_mmu_cache(vma, vmf->address, vmf->pte); |
65500d23 | 3049 | unlock: |
82b0f8c3 | 3050 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
6b31d595 | 3051 | return ret; |
8f4e2101 | 3052 | release: |
f627c2f5 | 3053 | mem_cgroup_cancel_charge(page, memcg, false); |
09cbfeaf | 3054 | put_page(page); |
8f4e2101 | 3055 | goto unlock; |
8a9f3ccd | 3056 | oom_free_page: |
09cbfeaf | 3057 | put_page(page); |
65500d23 | 3058 | oom: |
1da177e4 LT |
3059 | return VM_FAULT_OOM; |
3060 | } | |
3061 | ||
9a95f3cf PC |
3062 | /* |
3063 | * The mmap_sem must have been held on entry, and may have been | |
3064 | * released depending on flags and vma->vm_ops->fault() return value. | |
3065 | * See filemap_fault() and __lock_page_retry(). | |
3066 | */ | |
936ca80d | 3067 | static int __do_fault(struct vm_fault *vmf) |
7eae74af | 3068 | { |
82b0f8c3 | 3069 | struct vm_area_struct *vma = vmf->vma; |
7eae74af KS |
3070 | int ret; |
3071 | ||
11bac800 | 3072 | ret = vma->vm_ops->fault(vmf); |
3917048d | 3073 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY | |
b1aa812b | 3074 | VM_FAULT_DONE_COW))) |
bc2466e4 | 3075 | return ret; |
7eae74af | 3076 | |
667240e0 | 3077 | if (unlikely(PageHWPoison(vmf->page))) { |
7eae74af | 3078 | if (ret & VM_FAULT_LOCKED) |
667240e0 JK |
3079 | unlock_page(vmf->page); |
3080 | put_page(vmf->page); | |
936ca80d | 3081 | vmf->page = NULL; |
7eae74af KS |
3082 | return VM_FAULT_HWPOISON; |
3083 | } | |
3084 | ||
3085 | if (unlikely(!(ret & VM_FAULT_LOCKED))) | |
667240e0 | 3086 | lock_page(vmf->page); |
7eae74af | 3087 | else |
667240e0 | 3088 | VM_BUG_ON_PAGE(!PageLocked(vmf->page), vmf->page); |
7eae74af | 3089 | |
7eae74af KS |
3090 | return ret; |
3091 | } | |
3092 | ||
d0f0931d RZ |
3093 | /* |
3094 | * The ordering of these checks is important for pmds with _PAGE_DEVMAP set. | |
3095 | * If we check pmd_trans_unstable() first we will trip the bad_pmd() check | |
3096 | * inside of pmd_none_or_trans_huge_or_clear_bad(). This will end up correctly | |
3097 | * returning 1 but not before it spams dmesg with the pmd_clear_bad() output. | |
3098 | */ | |
3099 | static int pmd_devmap_trans_unstable(pmd_t *pmd) | |
3100 | { | |
3101 | return pmd_devmap(*pmd) || pmd_trans_unstable(pmd); | |
3102 | } | |
3103 | ||
82b0f8c3 | 3104 | static int pte_alloc_one_map(struct vm_fault *vmf) |
7267ec00 | 3105 | { |
82b0f8c3 | 3106 | struct vm_area_struct *vma = vmf->vma; |
7267ec00 | 3107 | |
82b0f8c3 | 3108 | if (!pmd_none(*vmf->pmd)) |
7267ec00 | 3109 | goto map_pte; |
82b0f8c3 JK |
3110 | if (vmf->prealloc_pte) { |
3111 | vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd); | |
3112 | if (unlikely(!pmd_none(*vmf->pmd))) { | |
3113 | spin_unlock(vmf->ptl); | |
7267ec00 KS |
3114 | goto map_pte; |
3115 | } | |
3116 | ||
3117 | atomic_long_inc(&vma->vm_mm->nr_ptes); | |
82b0f8c3 JK |
3118 | pmd_populate(vma->vm_mm, vmf->pmd, vmf->prealloc_pte); |
3119 | spin_unlock(vmf->ptl); | |
7f2b6ce8 | 3120 | vmf->prealloc_pte = NULL; |
82b0f8c3 | 3121 | } else if (unlikely(pte_alloc(vma->vm_mm, vmf->pmd, vmf->address))) { |
7267ec00 KS |
3122 | return VM_FAULT_OOM; |
3123 | } | |
3124 | map_pte: | |
3125 | /* | |
3126 | * If a huge pmd materialized under us just retry later. Use | |
d0f0931d RZ |
3127 | * pmd_trans_unstable() via pmd_devmap_trans_unstable() instead of |
3128 | * pmd_trans_huge() to ensure the pmd didn't become pmd_trans_huge | |
3129 | * under us and then back to pmd_none, as a result of MADV_DONTNEED | |
3130 | * running immediately after a huge pmd fault in a different thread of | |
3131 | * this mm, in turn leading to a misleading pmd_trans_huge() retval. | |
3132 | * All we have to ensure is that it is a regular pmd that we can walk | |
3133 | * with pte_offset_map() and we can do that through an atomic read in | |
3134 | * C, which is what pmd_trans_unstable() provides. | |
7267ec00 | 3135 | */ |
d0f0931d | 3136 | if (pmd_devmap_trans_unstable(vmf->pmd)) |
7267ec00 KS |
3137 | return VM_FAULT_NOPAGE; |
3138 | ||
d0f0931d RZ |
3139 | /* |
3140 | * At this point we know that our vmf->pmd points to a page of ptes | |
3141 | * and it cannot become pmd_none(), pmd_devmap() or pmd_trans_huge() | |
3142 | * for the duration of the fault. If a racing MADV_DONTNEED runs and | |
3143 | * we zap the ptes pointed to by our vmf->pmd, the vmf->ptl will still | |
3144 | * be valid and we will re-check to make sure the vmf->pte isn't | |
3145 | * pte_none() under vmf->ptl protection when we return to | |
3146 | * alloc_set_pte(). | |
3147 | */ | |
82b0f8c3 JK |
3148 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, |
3149 | &vmf->ptl); | |
7267ec00 KS |
3150 | return 0; |
3151 | } | |
3152 | ||
e496cf3d | 3153 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
10102459 KS |
3154 | |
3155 | #define HPAGE_CACHE_INDEX_MASK (HPAGE_PMD_NR - 1) | |
3156 | static inline bool transhuge_vma_suitable(struct vm_area_struct *vma, | |
3157 | unsigned long haddr) | |
3158 | { | |
3159 | if (((vma->vm_start >> PAGE_SHIFT) & HPAGE_CACHE_INDEX_MASK) != | |
3160 | (vma->vm_pgoff & HPAGE_CACHE_INDEX_MASK)) | |
3161 | return false; | |
3162 | if (haddr < vma->vm_start || haddr + HPAGE_PMD_SIZE > vma->vm_end) | |
3163 | return false; | |
3164 | return true; | |
3165 | } | |
3166 | ||
82b0f8c3 | 3167 | static void deposit_prealloc_pte(struct vm_fault *vmf) |
953c66c2 | 3168 | { |
82b0f8c3 | 3169 | struct vm_area_struct *vma = vmf->vma; |
953c66c2 | 3170 | |
82b0f8c3 | 3171 | pgtable_trans_huge_deposit(vma->vm_mm, vmf->pmd, vmf->prealloc_pte); |
953c66c2 AK |
3172 | /* |
3173 | * We are going to consume the prealloc table, | |
3174 | * count that as nr_ptes. | |
3175 | */ | |
3176 | atomic_long_inc(&vma->vm_mm->nr_ptes); | |
7f2b6ce8 | 3177 | vmf->prealloc_pte = NULL; |
953c66c2 AK |
3178 | } |
3179 | ||
82b0f8c3 | 3180 | static int do_set_pmd(struct vm_fault *vmf, struct page *page) |
10102459 | 3181 | { |
82b0f8c3 JK |
3182 | struct vm_area_struct *vma = vmf->vma; |
3183 | bool write = vmf->flags & FAULT_FLAG_WRITE; | |
3184 | unsigned long haddr = vmf->address & HPAGE_PMD_MASK; | |
10102459 KS |
3185 | pmd_t entry; |
3186 | int i, ret; | |
3187 | ||
3188 | if (!transhuge_vma_suitable(vma, haddr)) | |
3189 | return VM_FAULT_FALLBACK; | |
3190 | ||
3191 | ret = VM_FAULT_FALLBACK; | |
3192 | page = compound_head(page); | |
3193 | ||
953c66c2 AK |
3194 | /* |
3195 | * Archs like ppc64 need additonal space to store information | |
3196 | * related to pte entry. Use the preallocated table for that. | |
3197 | */ | |
82b0f8c3 JK |
3198 | if (arch_needs_pgtable_deposit() && !vmf->prealloc_pte) { |
3199 | vmf->prealloc_pte = pte_alloc_one(vma->vm_mm, vmf->address); | |
3200 | if (!vmf->prealloc_pte) | |
953c66c2 AK |
3201 | return VM_FAULT_OOM; |
3202 | smp_wmb(); /* See comment in __pte_alloc() */ | |
3203 | } | |
3204 | ||
82b0f8c3 JK |
3205 | vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd); |
3206 | if (unlikely(!pmd_none(*vmf->pmd))) | |
10102459 KS |
3207 | goto out; |
3208 | ||
3209 | for (i = 0; i < HPAGE_PMD_NR; i++) | |
3210 | flush_icache_page(vma, page + i); | |
3211 | ||
3212 | entry = mk_huge_pmd(page, vma->vm_page_prot); | |
3213 | if (write) | |
3214 | entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma); | |
3215 | ||
3216 | add_mm_counter(vma->vm_mm, MM_FILEPAGES, HPAGE_PMD_NR); | |
3217 | page_add_file_rmap(page, true); | |
953c66c2 AK |
3218 | /* |
3219 | * deposit and withdraw with pmd lock held | |
3220 | */ | |
3221 | if (arch_needs_pgtable_deposit()) | |
82b0f8c3 | 3222 | deposit_prealloc_pte(vmf); |
10102459 | 3223 | |
82b0f8c3 | 3224 | set_pmd_at(vma->vm_mm, haddr, vmf->pmd, entry); |
10102459 | 3225 | |
82b0f8c3 | 3226 | update_mmu_cache_pmd(vma, haddr, vmf->pmd); |
10102459 KS |
3227 | |
3228 | /* fault is handled */ | |
3229 | ret = 0; | |
95ecedcd | 3230 | count_vm_event(THP_FILE_MAPPED); |
10102459 | 3231 | out: |
82b0f8c3 | 3232 | spin_unlock(vmf->ptl); |
10102459 KS |
3233 | return ret; |
3234 | } | |
3235 | #else | |
82b0f8c3 | 3236 | static int do_set_pmd(struct vm_fault *vmf, struct page *page) |
10102459 KS |
3237 | { |
3238 | BUILD_BUG(); | |
3239 | return 0; | |
3240 | } | |
3241 | #endif | |
3242 | ||
8c6e50b0 | 3243 | /** |
7267ec00 KS |
3244 | * alloc_set_pte - setup new PTE entry for given page and add reverse page |
3245 | * mapping. If needed, the fucntion allocates page table or use pre-allocated. | |
8c6e50b0 | 3246 | * |
82b0f8c3 | 3247 | * @vmf: fault environment |
7267ec00 | 3248 | * @memcg: memcg to charge page (only for private mappings) |
8c6e50b0 | 3249 | * @page: page to map |
8c6e50b0 | 3250 | * |
82b0f8c3 JK |
3251 | * Caller must take care of unlocking vmf->ptl, if vmf->pte is non-NULL on |
3252 | * return. | |
8c6e50b0 KS |
3253 | * |
3254 | * Target users are page handler itself and implementations of | |
3255 | * vm_ops->map_pages. | |
3256 | */ | |
82b0f8c3 | 3257 | int alloc_set_pte(struct vm_fault *vmf, struct mem_cgroup *memcg, |
7267ec00 | 3258 | struct page *page) |
3bb97794 | 3259 | { |
82b0f8c3 JK |
3260 | struct vm_area_struct *vma = vmf->vma; |
3261 | bool write = vmf->flags & FAULT_FLAG_WRITE; | |
3bb97794 | 3262 | pte_t entry; |
10102459 KS |
3263 | int ret; |
3264 | ||
82b0f8c3 | 3265 | if (pmd_none(*vmf->pmd) && PageTransCompound(page) && |
e496cf3d | 3266 | IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) { |
10102459 KS |
3267 | /* THP on COW? */ |
3268 | VM_BUG_ON_PAGE(memcg, page); | |
3269 | ||
82b0f8c3 | 3270 | ret = do_set_pmd(vmf, page); |
10102459 | 3271 | if (ret != VM_FAULT_FALLBACK) |
b0b9b3df | 3272 | return ret; |
10102459 | 3273 | } |
3bb97794 | 3274 | |
82b0f8c3 JK |
3275 | if (!vmf->pte) { |
3276 | ret = pte_alloc_one_map(vmf); | |
7267ec00 | 3277 | if (ret) |
b0b9b3df | 3278 | return ret; |
7267ec00 KS |
3279 | } |
3280 | ||
3281 | /* Re-check under ptl */ | |
b0b9b3df HD |
3282 | if (unlikely(!pte_none(*vmf->pte))) |
3283 | return VM_FAULT_NOPAGE; | |
7267ec00 | 3284 | |
3bb97794 KS |
3285 | flush_icache_page(vma, page); |
3286 | entry = mk_pte(page, vma->vm_page_prot); | |
3287 | if (write) | |
3288 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
bae473a4 KS |
3289 | /* copy-on-write page */ |
3290 | if (write && !(vma->vm_flags & VM_SHARED)) { | |
3bb97794 | 3291 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
82b0f8c3 | 3292 | page_add_new_anon_rmap(page, vma, vmf->address, false); |
7267ec00 KS |
3293 | mem_cgroup_commit_charge(page, memcg, false, false); |
3294 | lru_cache_add_active_or_unevictable(page, vma); | |
3bb97794 | 3295 | } else { |
eca56ff9 | 3296 | inc_mm_counter_fast(vma->vm_mm, mm_counter_file(page)); |
dd78fedd | 3297 | page_add_file_rmap(page, false); |
3bb97794 | 3298 | } |
82b0f8c3 | 3299 | set_pte_at(vma->vm_mm, vmf->address, vmf->pte, entry); |
3bb97794 KS |
3300 | |
3301 | /* no need to invalidate: a not-present page won't be cached */ | |
82b0f8c3 | 3302 | update_mmu_cache(vma, vmf->address, vmf->pte); |
7267ec00 | 3303 | |
b0b9b3df | 3304 | return 0; |
3bb97794 KS |
3305 | } |
3306 | ||
9118c0cb JK |
3307 | |
3308 | /** | |
3309 | * finish_fault - finish page fault once we have prepared the page to fault | |
3310 | * | |
3311 | * @vmf: structure describing the fault | |
3312 | * | |
3313 | * This function handles all that is needed to finish a page fault once the | |
3314 | * page to fault in is prepared. It handles locking of PTEs, inserts PTE for | |
3315 | * given page, adds reverse page mapping, handles memcg charges and LRU | |
3316 | * addition. The function returns 0 on success, VM_FAULT_ code in case of | |
3317 | * error. | |
3318 | * | |
3319 | * The function expects the page to be locked and on success it consumes a | |
3320 | * reference of a page being mapped (for the PTE which maps it). | |
3321 | */ | |
3322 | int finish_fault(struct vm_fault *vmf) | |
3323 | { | |
3324 | struct page *page; | |
6b31d595 | 3325 | int ret = 0; |
9118c0cb JK |
3326 | |
3327 | /* Did we COW the page? */ | |
3328 | if ((vmf->flags & FAULT_FLAG_WRITE) && | |
3329 | !(vmf->vma->vm_flags & VM_SHARED)) | |
3330 | page = vmf->cow_page; | |
3331 | else | |
3332 | page = vmf->page; | |
6b31d595 MH |
3333 | |
3334 | /* | |
3335 | * check even for read faults because we might have lost our CoWed | |
3336 | * page | |
3337 | */ | |
3338 | if (!(vmf->vma->vm_flags & VM_SHARED)) | |
3339 | ret = check_stable_address_space(vmf->vma->vm_mm); | |
3340 | if (!ret) | |
3341 | ret = alloc_set_pte(vmf, vmf->memcg, page); | |
9118c0cb JK |
3342 | if (vmf->pte) |
3343 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
3344 | return ret; | |
3345 | } | |
3346 | ||
3a91053a KS |
3347 | static unsigned long fault_around_bytes __read_mostly = |
3348 | rounddown_pow_of_two(65536); | |
a9b0f861 | 3349 | |
a9b0f861 KS |
3350 | #ifdef CONFIG_DEBUG_FS |
3351 | static int fault_around_bytes_get(void *data, u64 *val) | |
1592eef0 | 3352 | { |
a9b0f861 | 3353 | *val = fault_around_bytes; |
1592eef0 KS |
3354 | return 0; |
3355 | } | |
3356 | ||
b4903d6e AR |
3357 | /* |
3358 | * fault_around_pages() and fault_around_mask() expects fault_around_bytes | |
3359 | * rounded down to nearest page order. It's what do_fault_around() expects to | |
3360 | * see. | |
3361 | */ | |
a9b0f861 | 3362 | static int fault_around_bytes_set(void *data, u64 val) |
1592eef0 | 3363 | { |
a9b0f861 | 3364 | if (val / PAGE_SIZE > PTRS_PER_PTE) |
1592eef0 | 3365 | return -EINVAL; |
b4903d6e AR |
3366 | if (val > PAGE_SIZE) |
3367 | fault_around_bytes = rounddown_pow_of_two(val); | |
3368 | else | |
3369 | fault_around_bytes = PAGE_SIZE; /* rounddown_pow_of_two(0) is undefined */ | |
1592eef0 KS |
3370 | return 0; |
3371 | } | |
0a1345f8 | 3372 | DEFINE_DEBUGFS_ATTRIBUTE(fault_around_bytes_fops, |
a9b0f861 | 3373 | fault_around_bytes_get, fault_around_bytes_set, "%llu\n"); |
1592eef0 KS |
3374 | |
3375 | static int __init fault_around_debugfs(void) | |
3376 | { | |
3377 | void *ret; | |
3378 | ||
0a1345f8 | 3379 | ret = debugfs_create_file_unsafe("fault_around_bytes", 0644, NULL, NULL, |
a9b0f861 | 3380 | &fault_around_bytes_fops); |
1592eef0 | 3381 | if (!ret) |
a9b0f861 | 3382 | pr_warn("Failed to create fault_around_bytes in debugfs"); |
1592eef0 KS |
3383 | return 0; |
3384 | } | |
3385 | late_initcall(fault_around_debugfs); | |
1592eef0 | 3386 | #endif |
8c6e50b0 | 3387 | |
1fdb412b KS |
3388 | /* |
3389 | * do_fault_around() tries to map few pages around the fault address. The hope | |
3390 | * is that the pages will be needed soon and this will lower the number of | |
3391 | * faults to handle. | |
3392 | * | |
3393 | * It uses vm_ops->map_pages() to map the pages, which skips the page if it's | |
3394 | * not ready to be mapped: not up-to-date, locked, etc. | |
3395 | * | |
3396 | * This function is called with the page table lock taken. In the split ptlock | |
3397 | * case the page table lock only protects only those entries which belong to | |
3398 | * the page table corresponding to the fault address. | |
3399 | * | |
3400 | * This function doesn't cross the VMA boundaries, in order to call map_pages() | |
3401 | * only once. | |
3402 | * | |
3403 | * fault_around_pages() defines how many pages we'll try to map. | |
3404 | * do_fault_around() expects it to return a power of two less than or equal to | |
3405 | * PTRS_PER_PTE. | |
3406 | * | |
3407 | * The virtual address of the area that we map is naturally aligned to the | |
3408 | * fault_around_pages() value (and therefore to page order). This way it's | |
3409 | * easier to guarantee that we don't cross page table boundaries. | |
3410 | */ | |
0721ec8b | 3411 | static int do_fault_around(struct vm_fault *vmf) |
8c6e50b0 | 3412 | { |
82b0f8c3 | 3413 | unsigned long address = vmf->address, nr_pages, mask; |
0721ec8b | 3414 | pgoff_t start_pgoff = vmf->pgoff; |
bae473a4 | 3415 | pgoff_t end_pgoff; |
7267ec00 | 3416 | int off, ret = 0; |
8c6e50b0 | 3417 | |
4db0c3c2 | 3418 | nr_pages = READ_ONCE(fault_around_bytes) >> PAGE_SHIFT; |
aecd6f44 KS |
3419 | mask = ~(nr_pages * PAGE_SIZE - 1) & PAGE_MASK; |
3420 | ||
82b0f8c3 JK |
3421 | vmf->address = max(address & mask, vmf->vma->vm_start); |
3422 | off = ((address - vmf->address) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1); | |
bae473a4 | 3423 | start_pgoff -= off; |
8c6e50b0 KS |
3424 | |
3425 | /* | |
bae473a4 KS |
3426 | * end_pgoff is either end of page table or end of vma |
3427 | * or fault_around_pages() from start_pgoff, depending what is nearest. | |
8c6e50b0 | 3428 | */ |
bae473a4 | 3429 | end_pgoff = start_pgoff - |
82b0f8c3 | 3430 | ((vmf->address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)) + |
8c6e50b0 | 3431 | PTRS_PER_PTE - 1; |
82b0f8c3 | 3432 | end_pgoff = min3(end_pgoff, vma_pages(vmf->vma) + vmf->vma->vm_pgoff - 1, |
bae473a4 | 3433 | start_pgoff + nr_pages - 1); |
8c6e50b0 | 3434 | |
82b0f8c3 JK |
3435 | if (pmd_none(*vmf->pmd)) { |
3436 | vmf->prealloc_pte = pte_alloc_one(vmf->vma->vm_mm, | |
3437 | vmf->address); | |
3438 | if (!vmf->prealloc_pte) | |
c5f88bd2 | 3439 | goto out; |
7267ec00 | 3440 | smp_wmb(); /* See comment in __pte_alloc() */ |
8c6e50b0 KS |
3441 | } |
3442 | ||
82b0f8c3 | 3443 | vmf->vma->vm_ops->map_pages(vmf, start_pgoff, end_pgoff); |
7267ec00 | 3444 | |
7267ec00 | 3445 | /* Huge page is mapped? Page fault is solved */ |
82b0f8c3 | 3446 | if (pmd_trans_huge(*vmf->pmd)) { |
7267ec00 KS |
3447 | ret = VM_FAULT_NOPAGE; |
3448 | goto out; | |
3449 | } | |
3450 | ||
3451 | /* ->map_pages() haven't done anything useful. Cold page cache? */ | |
82b0f8c3 | 3452 | if (!vmf->pte) |
7267ec00 KS |
3453 | goto out; |
3454 | ||
3455 | /* check if the page fault is solved */ | |
82b0f8c3 JK |
3456 | vmf->pte -= (vmf->address >> PAGE_SHIFT) - (address >> PAGE_SHIFT); |
3457 | if (!pte_none(*vmf->pte)) | |
7267ec00 | 3458 | ret = VM_FAULT_NOPAGE; |
82b0f8c3 | 3459 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
bae473a4 | 3460 | out: |
82b0f8c3 JK |
3461 | vmf->address = address; |
3462 | vmf->pte = NULL; | |
7267ec00 | 3463 | return ret; |
8c6e50b0 KS |
3464 | } |
3465 | ||
0721ec8b | 3466 | static int do_read_fault(struct vm_fault *vmf) |
e655fb29 | 3467 | { |
82b0f8c3 | 3468 | struct vm_area_struct *vma = vmf->vma; |
8c6e50b0 KS |
3469 | int ret = 0; |
3470 | ||
3471 | /* | |
3472 | * Let's call ->map_pages() first and use ->fault() as fallback | |
3473 | * if page by the offset is not ready to be mapped (cold cache or | |
3474 | * something). | |
3475 | */ | |
9b4bdd2f | 3476 | if (vma->vm_ops->map_pages && fault_around_bytes >> PAGE_SHIFT > 1) { |
0721ec8b | 3477 | ret = do_fault_around(vmf); |
7267ec00 KS |
3478 | if (ret) |
3479 | return ret; | |
8c6e50b0 | 3480 | } |
e655fb29 | 3481 | |
936ca80d | 3482 | ret = __do_fault(vmf); |
e655fb29 KS |
3483 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
3484 | return ret; | |
3485 | ||
9118c0cb | 3486 | ret |= finish_fault(vmf); |
936ca80d | 3487 | unlock_page(vmf->page); |
7267ec00 | 3488 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
936ca80d | 3489 | put_page(vmf->page); |
e655fb29 KS |
3490 | return ret; |
3491 | } | |
3492 | ||
0721ec8b | 3493 | static int do_cow_fault(struct vm_fault *vmf) |
ec47c3b9 | 3494 | { |
82b0f8c3 | 3495 | struct vm_area_struct *vma = vmf->vma; |
ec47c3b9 KS |
3496 | int ret; |
3497 | ||
3498 | if (unlikely(anon_vma_prepare(vma))) | |
3499 | return VM_FAULT_OOM; | |
3500 | ||
936ca80d JK |
3501 | vmf->cow_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vmf->address); |
3502 | if (!vmf->cow_page) | |
ec47c3b9 KS |
3503 | return VM_FAULT_OOM; |
3504 | ||
936ca80d | 3505 | if (mem_cgroup_try_charge(vmf->cow_page, vma->vm_mm, GFP_KERNEL, |
3917048d | 3506 | &vmf->memcg, false)) { |
936ca80d | 3507 | put_page(vmf->cow_page); |
ec47c3b9 KS |
3508 | return VM_FAULT_OOM; |
3509 | } | |
3510 | ||
936ca80d | 3511 | ret = __do_fault(vmf); |
ec47c3b9 KS |
3512 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
3513 | goto uncharge_out; | |
3917048d JK |
3514 | if (ret & VM_FAULT_DONE_COW) |
3515 | return ret; | |
ec47c3b9 | 3516 | |
b1aa812b | 3517 | copy_user_highpage(vmf->cow_page, vmf->page, vmf->address, vma); |
936ca80d | 3518 | __SetPageUptodate(vmf->cow_page); |
ec47c3b9 | 3519 | |
9118c0cb | 3520 | ret |= finish_fault(vmf); |
b1aa812b JK |
3521 | unlock_page(vmf->page); |
3522 | put_page(vmf->page); | |
7267ec00 KS |
3523 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
3524 | goto uncharge_out; | |
ec47c3b9 KS |
3525 | return ret; |
3526 | uncharge_out: | |
3917048d | 3527 | mem_cgroup_cancel_charge(vmf->cow_page, vmf->memcg, false); |
936ca80d | 3528 | put_page(vmf->cow_page); |
ec47c3b9 KS |
3529 | return ret; |
3530 | } | |
3531 | ||
0721ec8b | 3532 | static int do_shared_fault(struct vm_fault *vmf) |
1da177e4 | 3533 | { |
82b0f8c3 | 3534 | struct vm_area_struct *vma = vmf->vma; |
f0c6d4d2 | 3535 | int ret, tmp; |
1d65f86d | 3536 | |
936ca80d | 3537 | ret = __do_fault(vmf); |
7eae74af | 3538 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
f0c6d4d2 | 3539 | return ret; |
1da177e4 LT |
3540 | |
3541 | /* | |
f0c6d4d2 KS |
3542 | * Check if the backing address space wants to know that the page is |
3543 | * about to become writable | |
1da177e4 | 3544 | */ |
fb09a464 | 3545 | if (vma->vm_ops->page_mkwrite) { |
936ca80d | 3546 | unlock_page(vmf->page); |
38b8cb7f | 3547 | tmp = do_page_mkwrite(vmf); |
fb09a464 KS |
3548 | if (unlikely(!tmp || |
3549 | (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { | |
936ca80d | 3550 | put_page(vmf->page); |
fb09a464 | 3551 | return tmp; |
4294621f | 3552 | } |
fb09a464 KS |
3553 | } |
3554 | ||
9118c0cb | 3555 | ret |= finish_fault(vmf); |
7267ec00 KS |
3556 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | |
3557 | VM_FAULT_RETRY))) { | |
936ca80d JK |
3558 | unlock_page(vmf->page); |
3559 | put_page(vmf->page); | |
f0c6d4d2 | 3560 | return ret; |
1da177e4 | 3561 | } |
b827e496 | 3562 | |
97ba0c2b | 3563 | fault_dirty_shared_page(vma, vmf->page); |
1d65f86d | 3564 | return ret; |
54cb8821 | 3565 | } |
d00806b1 | 3566 | |
9a95f3cf PC |
3567 | /* |
3568 | * We enter with non-exclusive mmap_sem (to exclude vma changes, | |
3569 | * but allow concurrent faults). | |
3570 | * The mmap_sem may have been released depending on flags and our | |
3571 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
3572 | */ | |
82b0f8c3 | 3573 | static int do_fault(struct vm_fault *vmf) |
54cb8821 | 3574 | { |
82b0f8c3 | 3575 | struct vm_area_struct *vma = vmf->vma; |
b0b9b3df | 3576 | int ret; |
54cb8821 | 3577 | |
6b7339f4 KS |
3578 | /* The VMA was not fully populated on mmap() or missing VM_DONTEXPAND */ |
3579 | if (!vma->vm_ops->fault) | |
b0b9b3df HD |
3580 | ret = VM_FAULT_SIGBUS; |
3581 | else if (!(vmf->flags & FAULT_FLAG_WRITE)) | |
3582 | ret = do_read_fault(vmf); | |
3583 | else if (!(vma->vm_flags & VM_SHARED)) | |
3584 | ret = do_cow_fault(vmf); | |
3585 | else | |
3586 | ret = do_shared_fault(vmf); | |
3587 | ||
3588 | /* preallocated pagetable is unused: free it */ | |
3589 | if (vmf->prealloc_pte) { | |
3590 | pte_free(vma->vm_mm, vmf->prealloc_pte); | |
7f2b6ce8 | 3591 | vmf->prealloc_pte = NULL; |
b0b9b3df HD |
3592 | } |
3593 | return ret; | |
54cb8821 NP |
3594 | } |
3595 | ||
b19a9939 | 3596 | static int numa_migrate_prep(struct page *page, struct vm_area_struct *vma, |
04bb2f94 RR |
3597 | unsigned long addr, int page_nid, |
3598 | int *flags) | |
9532fec1 MG |
3599 | { |
3600 | get_page(page); | |
3601 | ||
3602 | count_vm_numa_event(NUMA_HINT_FAULTS); | |
04bb2f94 | 3603 | if (page_nid == numa_node_id()) { |
9532fec1 | 3604 | count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL); |
04bb2f94 RR |
3605 | *flags |= TNF_FAULT_LOCAL; |
3606 | } | |
9532fec1 MG |
3607 | |
3608 | return mpol_misplaced(page, vma, addr); | |
3609 | } | |
3610 | ||
2994302b | 3611 | static int do_numa_page(struct vm_fault *vmf) |
d10e63f2 | 3612 | { |
82b0f8c3 | 3613 | struct vm_area_struct *vma = vmf->vma; |
4daae3b4 | 3614 | struct page *page = NULL; |
8191acbd | 3615 | int page_nid = -1; |
90572890 | 3616 | int last_cpupid; |
cbee9f88 | 3617 | int target_nid; |
b8593bfd | 3618 | bool migrated = false; |
cee216a6 | 3619 | pte_t pte; |
288bc549 | 3620 | bool was_writable = pte_savedwrite(vmf->orig_pte); |
6688cc05 | 3621 | int flags = 0; |
d10e63f2 MG |
3622 | |
3623 | /* | |
166f61b9 TH |
3624 | * The "pte" at this point cannot be used safely without |
3625 | * validation through pte_unmap_same(). It's of NUMA type but | |
3626 | * the pfn may be screwed if the read is non atomic. | |
166f61b9 | 3627 | */ |
82b0f8c3 JK |
3628 | vmf->ptl = pte_lockptr(vma->vm_mm, vmf->pmd); |
3629 | spin_lock(vmf->ptl); | |
cee216a6 | 3630 | if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) { |
82b0f8c3 | 3631 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
4daae3b4 MG |
3632 | goto out; |
3633 | } | |
3634 | ||
cee216a6 AK |
3635 | /* |
3636 | * Make it present again, Depending on how arch implementes non | |
3637 | * accessible ptes, some can allow access by kernel mode. | |
3638 | */ | |
3639 | pte = ptep_modify_prot_start(vma->vm_mm, vmf->address, vmf->pte); | |
4d942466 MG |
3640 | pte = pte_modify(pte, vma->vm_page_prot); |
3641 | pte = pte_mkyoung(pte); | |
b191f9b1 MG |
3642 | if (was_writable) |
3643 | pte = pte_mkwrite(pte); | |
cee216a6 | 3644 | ptep_modify_prot_commit(vma->vm_mm, vmf->address, vmf->pte, pte); |
82b0f8c3 | 3645 | update_mmu_cache(vma, vmf->address, vmf->pte); |
d10e63f2 | 3646 | |
82b0f8c3 | 3647 | page = vm_normal_page(vma, vmf->address, pte); |
d10e63f2 | 3648 | if (!page) { |
82b0f8c3 | 3649 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
d10e63f2 MG |
3650 | return 0; |
3651 | } | |
3652 | ||
e81c4802 KS |
3653 | /* TODO: handle PTE-mapped THP */ |
3654 | if (PageCompound(page)) { | |
82b0f8c3 | 3655 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
e81c4802 KS |
3656 | return 0; |
3657 | } | |
3658 | ||
6688cc05 | 3659 | /* |
bea66fbd MG |
3660 | * Avoid grouping on RO pages in general. RO pages shouldn't hurt as |
3661 | * much anyway since they can be in shared cache state. This misses | |
3662 | * the case where a mapping is writable but the process never writes | |
3663 | * to it but pte_write gets cleared during protection updates and | |
3664 | * pte_dirty has unpredictable behaviour between PTE scan updates, | |
3665 | * background writeback, dirty balancing and application behaviour. | |
6688cc05 | 3666 | */ |
d59dc7bc | 3667 | if (!pte_write(pte)) |
6688cc05 PZ |
3668 | flags |= TNF_NO_GROUP; |
3669 | ||
dabe1d99 RR |
3670 | /* |
3671 | * Flag if the page is shared between multiple address spaces. This | |
3672 | * is later used when determining whether to group tasks together | |
3673 | */ | |
3674 | if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED)) | |
3675 | flags |= TNF_SHARED; | |
3676 | ||
90572890 | 3677 | last_cpupid = page_cpupid_last(page); |
8191acbd | 3678 | page_nid = page_to_nid(page); |
82b0f8c3 | 3679 | target_nid = numa_migrate_prep(page, vma, vmf->address, page_nid, |
bae473a4 | 3680 | &flags); |
82b0f8c3 | 3681 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
4daae3b4 | 3682 | if (target_nid == -1) { |
4daae3b4 MG |
3683 | put_page(page); |
3684 | goto out; | |
3685 | } | |
3686 | ||
3687 | /* Migrate to the requested node */ | |
1bc115d8 | 3688 | migrated = migrate_misplaced_page(page, vma, target_nid); |
6688cc05 | 3689 | if (migrated) { |
8191acbd | 3690 | page_nid = target_nid; |
6688cc05 | 3691 | flags |= TNF_MIGRATED; |
074c2381 MG |
3692 | } else |
3693 | flags |= TNF_MIGRATE_FAIL; | |
4daae3b4 MG |
3694 | |
3695 | out: | |
8191acbd | 3696 | if (page_nid != -1) |
6688cc05 | 3697 | task_numa_fault(last_cpupid, page_nid, 1, flags); |
d10e63f2 MG |
3698 | return 0; |
3699 | } | |
3700 | ||
91a90140 | 3701 | static inline int create_huge_pmd(struct vm_fault *vmf) |
b96375f7 | 3702 | { |
f4200391 | 3703 | if (vma_is_anonymous(vmf->vma)) |
82b0f8c3 | 3704 | return do_huge_pmd_anonymous_page(vmf); |
a2d58167 | 3705 | if (vmf->vma->vm_ops->huge_fault) |
c791ace1 | 3706 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD); |
b96375f7 MW |
3707 | return VM_FAULT_FALLBACK; |
3708 | } | |
3709 | ||
82b0f8c3 | 3710 | static int wp_huge_pmd(struct vm_fault *vmf, pmd_t orig_pmd) |
b96375f7 | 3711 | { |
82b0f8c3 JK |
3712 | if (vma_is_anonymous(vmf->vma)) |
3713 | return do_huge_pmd_wp_page(vmf, orig_pmd); | |
a2d58167 | 3714 | if (vmf->vma->vm_ops->huge_fault) |
c791ace1 | 3715 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD); |
af9e4d5f KS |
3716 | |
3717 | /* COW handled on pte level: split pmd */ | |
82b0f8c3 JK |
3718 | VM_BUG_ON_VMA(vmf->vma->vm_flags & VM_SHARED, vmf->vma); |
3719 | __split_huge_pmd(vmf->vma, vmf->pmd, vmf->address, false, NULL); | |
af9e4d5f | 3720 | |
b96375f7 MW |
3721 | return VM_FAULT_FALLBACK; |
3722 | } | |
3723 | ||
38e08854 LS |
3724 | static inline bool vma_is_accessible(struct vm_area_struct *vma) |
3725 | { | |
3726 | return vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE); | |
3727 | } | |
3728 | ||
a00cc7d9 MW |
3729 | static int create_huge_pud(struct vm_fault *vmf) |
3730 | { | |
3731 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
3732 | /* No support for anonymous transparent PUD pages yet */ | |
3733 | if (vma_is_anonymous(vmf->vma)) | |
3734 | return VM_FAULT_FALLBACK; | |
3735 | if (vmf->vma->vm_ops->huge_fault) | |
c791ace1 | 3736 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD); |
a00cc7d9 MW |
3737 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
3738 | return VM_FAULT_FALLBACK; | |
3739 | } | |
3740 | ||
3741 | static int wp_huge_pud(struct vm_fault *vmf, pud_t orig_pud) | |
3742 | { | |
3743 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
3744 | /* No support for anonymous transparent PUD pages yet */ | |
3745 | if (vma_is_anonymous(vmf->vma)) | |
3746 | return VM_FAULT_FALLBACK; | |
3747 | if (vmf->vma->vm_ops->huge_fault) | |
c791ace1 | 3748 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD); |
a00cc7d9 MW |
3749 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
3750 | return VM_FAULT_FALLBACK; | |
3751 | } | |
3752 | ||
1da177e4 LT |
3753 | /* |
3754 | * These routines also need to handle stuff like marking pages dirty | |
3755 | * and/or accessed for architectures that don't do it in hardware (most | |
3756 | * RISC architectures). The early dirtying is also good on the i386. | |
3757 | * | |
3758 | * There is also a hook called "update_mmu_cache()" that architectures | |
3759 | * with external mmu caches can use to update those (ie the Sparc or | |
3760 | * PowerPC hashed page tables that act as extended TLBs). | |
3761 | * | |
7267ec00 KS |
3762 | * We enter with non-exclusive mmap_sem (to exclude vma changes, but allow |
3763 | * concurrent faults). | |
9a95f3cf | 3764 | * |
7267ec00 KS |
3765 | * The mmap_sem may have been released depending on flags and our return value. |
3766 | * See filemap_fault() and __lock_page_or_retry(). | |
1da177e4 | 3767 | */ |
82b0f8c3 | 3768 | static int handle_pte_fault(struct vm_fault *vmf) |
1da177e4 LT |
3769 | { |
3770 | pte_t entry; | |
3771 | ||
82b0f8c3 | 3772 | if (unlikely(pmd_none(*vmf->pmd))) { |
7267ec00 KS |
3773 | /* |
3774 | * Leave __pte_alloc() until later: because vm_ops->fault may | |
3775 | * want to allocate huge page, and if we expose page table | |
3776 | * for an instant, it will be difficult to retract from | |
3777 | * concurrent faults and from rmap lookups. | |
3778 | */ | |
82b0f8c3 | 3779 | vmf->pte = NULL; |
7267ec00 KS |
3780 | } else { |
3781 | /* See comment in pte_alloc_one_map() */ | |
d0f0931d | 3782 | if (pmd_devmap_trans_unstable(vmf->pmd)) |
7267ec00 KS |
3783 | return 0; |
3784 | /* | |
3785 | * A regular pmd is established and it can't morph into a huge | |
3786 | * pmd from under us anymore at this point because we hold the | |
3787 | * mmap_sem read mode and khugepaged takes it in write mode. | |
3788 | * So now it's safe to run pte_offset_map(). | |
3789 | */ | |
82b0f8c3 | 3790 | vmf->pte = pte_offset_map(vmf->pmd, vmf->address); |
2994302b | 3791 | vmf->orig_pte = *vmf->pte; |
7267ec00 KS |
3792 | |
3793 | /* | |
3794 | * some architectures can have larger ptes than wordsize, | |
3795 | * e.g.ppc44x-defconfig has CONFIG_PTE_64BIT=y and | |
3796 | * CONFIG_32BIT=y, so READ_ONCE or ACCESS_ONCE cannot guarantee | |
3797 | * atomic accesses. The code below just needs a consistent | |
3798 | * view for the ifs and we later double check anyway with the | |
3799 | * ptl lock held. So here a barrier will do. | |
3800 | */ | |
3801 | barrier(); | |
2994302b | 3802 | if (pte_none(vmf->orig_pte)) { |
82b0f8c3 JK |
3803 | pte_unmap(vmf->pte); |
3804 | vmf->pte = NULL; | |
65500d23 | 3805 | } |
1da177e4 LT |
3806 | } |
3807 | ||
82b0f8c3 JK |
3808 | if (!vmf->pte) { |
3809 | if (vma_is_anonymous(vmf->vma)) | |
3810 | return do_anonymous_page(vmf); | |
7267ec00 | 3811 | else |
82b0f8c3 | 3812 | return do_fault(vmf); |
7267ec00 KS |
3813 | } |
3814 | ||
2994302b JK |
3815 | if (!pte_present(vmf->orig_pte)) |
3816 | return do_swap_page(vmf); | |
7267ec00 | 3817 | |
2994302b JK |
3818 | if (pte_protnone(vmf->orig_pte) && vma_is_accessible(vmf->vma)) |
3819 | return do_numa_page(vmf); | |
d10e63f2 | 3820 | |
82b0f8c3 JK |
3821 | vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd); |
3822 | spin_lock(vmf->ptl); | |
2994302b | 3823 | entry = vmf->orig_pte; |
82b0f8c3 | 3824 | if (unlikely(!pte_same(*vmf->pte, entry))) |
8f4e2101 | 3825 | goto unlock; |
82b0f8c3 | 3826 | if (vmf->flags & FAULT_FLAG_WRITE) { |
1da177e4 | 3827 | if (!pte_write(entry)) |
2994302b | 3828 | return do_wp_page(vmf); |
1da177e4 LT |
3829 | entry = pte_mkdirty(entry); |
3830 | } | |
3831 | entry = pte_mkyoung(entry); | |
82b0f8c3 JK |
3832 | if (ptep_set_access_flags(vmf->vma, vmf->address, vmf->pte, entry, |
3833 | vmf->flags & FAULT_FLAG_WRITE)) { | |
3834 | update_mmu_cache(vmf->vma, vmf->address, vmf->pte); | |
1a44e149 AA |
3835 | } else { |
3836 | /* | |
3837 | * This is needed only for protection faults but the arch code | |
3838 | * is not yet telling us if this is a protection fault or not. | |
3839 | * This still avoids useless tlb flushes for .text page faults | |
3840 | * with threads. | |
3841 | */ | |
82b0f8c3 JK |
3842 | if (vmf->flags & FAULT_FLAG_WRITE) |
3843 | flush_tlb_fix_spurious_fault(vmf->vma, vmf->address); | |
1a44e149 | 3844 | } |
8f4e2101 | 3845 | unlock: |
82b0f8c3 | 3846 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
83c54070 | 3847 | return 0; |
1da177e4 LT |
3848 | } |
3849 | ||
3850 | /* | |
3851 | * By the time we get here, we already hold the mm semaphore | |
9a95f3cf PC |
3852 | * |
3853 | * The mmap_sem may have been released depending on flags and our | |
3854 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
1da177e4 | 3855 | */ |
dcddffd4 KS |
3856 | static int __handle_mm_fault(struct vm_area_struct *vma, unsigned long address, |
3857 | unsigned int flags) | |
1da177e4 | 3858 | { |
82b0f8c3 | 3859 | struct vm_fault vmf = { |
bae473a4 | 3860 | .vma = vma, |
1a29d85e | 3861 | .address = address & PAGE_MASK, |
bae473a4 | 3862 | .flags = flags, |
0721ec8b | 3863 | .pgoff = linear_page_index(vma, address), |
667240e0 | 3864 | .gfp_mask = __get_fault_gfp_mask(vma), |
bae473a4 | 3865 | }; |
dcddffd4 | 3866 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 3867 | pgd_t *pgd; |
c2febafc | 3868 | p4d_t *p4d; |
a2d58167 | 3869 | int ret; |
1da177e4 | 3870 | |
1da177e4 | 3871 | pgd = pgd_offset(mm, address); |
c2febafc KS |
3872 | p4d = p4d_alloc(mm, pgd, address); |
3873 | if (!p4d) | |
3874 | return VM_FAULT_OOM; | |
a00cc7d9 | 3875 | |
c2febafc | 3876 | vmf.pud = pud_alloc(mm, p4d, address); |
a00cc7d9 | 3877 | if (!vmf.pud) |
c74df32c | 3878 | return VM_FAULT_OOM; |
a00cc7d9 | 3879 | if (pud_none(*vmf.pud) && transparent_hugepage_enabled(vma)) { |
a00cc7d9 MW |
3880 | ret = create_huge_pud(&vmf); |
3881 | if (!(ret & VM_FAULT_FALLBACK)) | |
3882 | return ret; | |
3883 | } else { | |
3884 | pud_t orig_pud = *vmf.pud; | |
3885 | ||
3886 | barrier(); | |
3887 | if (pud_trans_huge(orig_pud) || pud_devmap(orig_pud)) { | |
3888 | unsigned int dirty = flags & FAULT_FLAG_WRITE; | |
3889 | ||
a00cc7d9 MW |
3890 | /* NUMA case for anonymous PUDs would go here */ |
3891 | ||
3892 | if (dirty && !pud_write(orig_pud)) { | |
3893 | ret = wp_huge_pud(&vmf, orig_pud); | |
3894 | if (!(ret & VM_FAULT_FALLBACK)) | |
3895 | return ret; | |
3896 | } else { | |
3897 | huge_pud_set_accessed(&vmf, orig_pud); | |
3898 | return 0; | |
3899 | } | |
3900 | } | |
3901 | } | |
3902 | ||
3903 | vmf.pmd = pmd_alloc(mm, vmf.pud, address); | |
82b0f8c3 | 3904 | if (!vmf.pmd) |
c74df32c | 3905 | return VM_FAULT_OOM; |
82b0f8c3 | 3906 | if (pmd_none(*vmf.pmd) && transparent_hugepage_enabled(vma)) { |
a2d58167 | 3907 | ret = create_huge_pmd(&vmf); |
c0292554 KS |
3908 | if (!(ret & VM_FAULT_FALLBACK)) |
3909 | return ret; | |
71e3aac0 | 3910 | } else { |
82b0f8c3 | 3911 | pmd_t orig_pmd = *vmf.pmd; |
1f1d06c3 | 3912 | |
71e3aac0 | 3913 | barrier(); |
5c7fb56e | 3914 | if (pmd_trans_huge(orig_pmd) || pmd_devmap(orig_pmd)) { |
38e08854 | 3915 | if (pmd_protnone(orig_pmd) && vma_is_accessible(vma)) |
82b0f8c3 | 3916 | return do_huge_pmd_numa_page(&vmf, orig_pmd); |
d10e63f2 | 3917 | |
82b0f8c3 | 3918 | if ((vmf.flags & FAULT_FLAG_WRITE) && |
bae473a4 | 3919 | !pmd_write(orig_pmd)) { |
82b0f8c3 | 3920 | ret = wp_huge_pmd(&vmf, orig_pmd); |
9845cbbd KS |
3921 | if (!(ret & VM_FAULT_FALLBACK)) |
3922 | return ret; | |
a1dd450b | 3923 | } else { |
82b0f8c3 | 3924 | huge_pmd_set_accessed(&vmf, orig_pmd); |
9845cbbd | 3925 | return 0; |
1f1d06c3 | 3926 | } |
71e3aac0 AA |
3927 | } |
3928 | } | |
3929 | ||
82b0f8c3 | 3930 | return handle_pte_fault(&vmf); |
1da177e4 LT |
3931 | } |
3932 | ||
9a95f3cf PC |
3933 | /* |
3934 | * By the time we get here, we already hold the mm semaphore | |
3935 | * | |
3936 | * The mmap_sem may have been released depending on flags and our | |
3937 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
3938 | */ | |
dcddffd4 KS |
3939 | int handle_mm_fault(struct vm_area_struct *vma, unsigned long address, |
3940 | unsigned int flags) | |
519e5247 JW |
3941 | { |
3942 | int ret; | |
3943 | ||
3944 | __set_current_state(TASK_RUNNING); | |
3945 | ||
3946 | count_vm_event(PGFAULT); | |
2262185c | 3947 | count_memcg_event_mm(vma->vm_mm, PGFAULT); |
519e5247 JW |
3948 | |
3949 | /* do counter updates before entering really critical section. */ | |
3950 | check_sync_rss_stat(current); | |
3951 | ||
3952 | /* | |
3953 | * Enable the memcg OOM handling for faults triggered in user | |
3954 | * space. Kernel faults are handled more gracefully. | |
3955 | */ | |
3956 | if (flags & FAULT_FLAG_USER) | |
49426420 | 3957 | mem_cgroup_oom_enable(); |
519e5247 | 3958 | |
bae473a4 KS |
3959 | if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE, |
3960 | flags & FAULT_FLAG_INSTRUCTION, | |
3961 | flags & FAULT_FLAG_REMOTE)) | |
3962 | return VM_FAULT_SIGSEGV; | |
3963 | ||
3964 | if (unlikely(is_vm_hugetlb_page(vma))) | |
3965 | ret = hugetlb_fault(vma->vm_mm, vma, address, flags); | |
3966 | else | |
3967 | ret = __handle_mm_fault(vma, address, flags); | |
519e5247 | 3968 | |
49426420 JW |
3969 | if (flags & FAULT_FLAG_USER) { |
3970 | mem_cgroup_oom_disable(); | |
166f61b9 TH |
3971 | /* |
3972 | * The task may have entered a memcg OOM situation but | |
3973 | * if the allocation error was handled gracefully (no | |
3974 | * VM_FAULT_OOM), there is no need to kill anything. | |
3975 | * Just clean up the OOM state peacefully. | |
3976 | */ | |
3977 | if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM)) | |
3978 | mem_cgroup_oom_synchronize(false); | |
49426420 | 3979 | } |
3812c8c8 | 3980 | |
519e5247 JW |
3981 | return ret; |
3982 | } | |
e1d6d01a | 3983 | EXPORT_SYMBOL_GPL(handle_mm_fault); |
519e5247 | 3984 | |
90eceff1 KS |
3985 | #ifndef __PAGETABLE_P4D_FOLDED |
3986 | /* | |
3987 | * Allocate p4d page table. | |
3988 | * We've already handled the fast-path in-line. | |
3989 | */ | |
3990 | int __p4d_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) | |
3991 | { | |
3992 | p4d_t *new = p4d_alloc_one(mm, address); | |
3993 | if (!new) | |
3994 | return -ENOMEM; | |
3995 | ||
3996 | smp_wmb(); /* See comment in __pte_alloc */ | |
3997 | ||
3998 | spin_lock(&mm->page_table_lock); | |
3999 | if (pgd_present(*pgd)) /* Another has populated it */ | |
4000 | p4d_free(mm, new); | |
4001 | else | |
4002 | pgd_populate(mm, pgd, new); | |
4003 | spin_unlock(&mm->page_table_lock); | |
4004 | return 0; | |
4005 | } | |
4006 | #endif /* __PAGETABLE_P4D_FOLDED */ | |
4007 | ||
1da177e4 LT |
4008 | #ifndef __PAGETABLE_PUD_FOLDED |
4009 | /* | |
4010 | * Allocate page upper directory. | |
872fec16 | 4011 | * We've already handled the fast-path in-line. |
1da177e4 | 4012 | */ |
c2febafc | 4013 | int __pud_alloc(struct mm_struct *mm, p4d_t *p4d, unsigned long address) |
1da177e4 | 4014 | { |
c74df32c HD |
4015 | pud_t *new = pud_alloc_one(mm, address); |
4016 | if (!new) | |
1bb3630e | 4017 | return -ENOMEM; |
1da177e4 | 4018 | |
362a61ad NP |
4019 | smp_wmb(); /* See comment in __pte_alloc */ |
4020 | ||
872fec16 | 4021 | spin_lock(&mm->page_table_lock); |
c2febafc KS |
4022 | #ifndef __ARCH_HAS_5LEVEL_HACK |
4023 | if (p4d_present(*p4d)) /* Another has populated it */ | |
5e541973 | 4024 | pud_free(mm, new); |
1bb3630e | 4025 | else |
c2febafc KS |
4026 | p4d_populate(mm, p4d, new); |
4027 | #else | |
4028 | if (pgd_present(*p4d)) /* Another has populated it */ | |
5e541973 | 4029 | pud_free(mm, new); |
1bb3630e | 4030 | else |
c2febafc KS |
4031 | pgd_populate(mm, p4d, new); |
4032 | #endif /* __ARCH_HAS_5LEVEL_HACK */ | |
c74df32c | 4033 | spin_unlock(&mm->page_table_lock); |
1bb3630e | 4034 | return 0; |
1da177e4 LT |
4035 | } |
4036 | #endif /* __PAGETABLE_PUD_FOLDED */ | |
4037 | ||
4038 | #ifndef __PAGETABLE_PMD_FOLDED | |
4039 | /* | |
4040 | * Allocate page middle directory. | |
872fec16 | 4041 | * We've already handled the fast-path in-line. |
1da177e4 | 4042 | */ |
1bb3630e | 4043 | int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) |
1da177e4 | 4044 | { |
a00cc7d9 | 4045 | spinlock_t *ptl; |
c74df32c HD |
4046 | pmd_t *new = pmd_alloc_one(mm, address); |
4047 | if (!new) | |
1bb3630e | 4048 | return -ENOMEM; |
1da177e4 | 4049 | |
362a61ad NP |
4050 | smp_wmb(); /* See comment in __pte_alloc */ |
4051 | ||
a00cc7d9 | 4052 | ptl = pud_lock(mm, pud); |
1da177e4 | 4053 | #ifndef __ARCH_HAS_4LEVEL_HACK |
dc6c9a35 KS |
4054 | if (!pud_present(*pud)) { |
4055 | mm_inc_nr_pmds(mm); | |
1bb3630e | 4056 | pud_populate(mm, pud, new); |
dc6c9a35 | 4057 | } else /* Another has populated it */ |
5e541973 | 4058 | pmd_free(mm, new); |
dc6c9a35 KS |
4059 | #else |
4060 | if (!pgd_present(*pud)) { | |
4061 | mm_inc_nr_pmds(mm); | |
1bb3630e | 4062 | pgd_populate(mm, pud, new); |
dc6c9a35 KS |
4063 | } else /* Another has populated it */ |
4064 | pmd_free(mm, new); | |
1da177e4 | 4065 | #endif /* __ARCH_HAS_4LEVEL_HACK */ |
a00cc7d9 | 4066 | spin_unlock(ptl); |
1bb3630e | 4067 | return 0; |
e0f39591 | 4068 | } |
1da177e4 LT |
4069 | #endif /* __PAGETABLE_PMD_FOLDED */ |
4070 | ||
09796395 | 4071 | static int __follow_pte_pmd(struct mm_struct *mm, unsigned long address, |
a4d1a885 JG |
4072 | unsigned long *start, unsigned long *end, |
4073 | pte_t **ptepp, pmd_t **pmdpp, spinlock_t **ptlp) | |
f8ad0f49 JW |
4074 | { |
4075 | pgd_t *pgd; | |
c2febafc | 4076 | p4d_t *p4d; |
f8ad0f49 JW |
4077 | pud_t *pud; |
4078 | pmd_t *pmd; | |
4079 | pte_t *ptep; | |
4080 | ||
4081 | pgd = pgd_offset(mm, address); | |
4082 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
4083 | goto out; | |
4084 | ||
c2febafc KS |
4085 | p4d = p4d_offset(pgd, address); |
4086 | if (p4d_none(*p4d) || unlikely(p4d_bad(*p4d))) | |
4087 | goto out; | |
4088 | ||
4089 | pud = pud_offset(p4d, address); | |
f8ad0f49 JW |
4090 | if (pud_none(*pud) || unlikely(pud_bad(*pud))) |
4091 | goto out; | |
4092 | ||
4093 | pmd = pmd_offset(pud, address); | |
f66055ab | 4094 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
f8ad0f49 | 4095 | |
09796395 RZ |
4096 | if (pmd_huge(*pmd)) { |
4097 | if (!pmdpp) | |
4098 | goto out; | |
4099 | ||
a4d1a885 JG |
4100 | if (start && end) { |
4101 | *start = address & PMD_MASK; | |
4102 | *end = *start + PMD_SIZE; | |
4103 | mmu_notifier_invalidate_range_start(mm, *start, *end); | |
4104 | } | |
09796395 RZ |
4105 | *ptlp = pmd_lock(mm, pmd); |
4106 | if (pmd_huge(*pmd)) { | |
4107 | *pmdpp = pmd; | |
4108 | return 0; | |
4109 | } | |
4110 | spin_unlock(*ptlp); | |
a4d1a885 JG |
4111 | if (start && end) |
4112 | mmu_notifier_invalidate_range_end(mm, *start, *end); | |
09796395 RZ |
4113 | } |
4114 | ||
4115 | if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) | |
f8ad0f49 JW |
4116 | goto out; |
4117 | ||
a4d1a885 JG |
4118 | if (start && end) { |
4119 | *start = address & PAGE_MASK; | |
4120 | *end = *start + PAGE_SIZE; | |
4121 | mmu_notifier_invalidate_range_start(mm, *start, *end); | |
4122 | } | |
f8ad0f49 | 4123 | ptep = pte_offset_map_lock(mm, pmd, address, ptlp); |
f8ad0f49 JW |
4124 | if (!pte_present(*ptep)) |
4125 | goto unlock; | |
4126 | *ptepp = ptep; | |
4127 | return 0; | |
4128 | unlock: | |
4129 | pte_unmap_unlock(ptep, *ptlp); | |
a4d1a885 JG |
4130 | if (start && end) |
4131 | mmu_notifier_invalidate_range_end(mm, *start, *end); | |
f8ad0f49 JW |
4132 | out: |
4133 | return -EINVAL; | |
4134 | } | |
4135 | ||
f729c8c9 RZ |
4136 | static inline int follow_pte(struct mm_struct *mm, unsigned long address, |
4137 | pte_t **ptepp, spinlock_t **ptlp) | |
1b36ba81 NK |
4138 | { |
4139 | int res; | |
4140 | ||
4141 | /* (void) is needed to make gcc happy */ | |
4142 | (void) __cond_lock(*ptlp, | |
a4d1a885 JG |
4143 | !(res = __follow_pte_pmd(mm, address, NULL, NULL, |
4144 | ptepp, NULL, ptlp))); | |
09796395 RZ |
4145 | return res; |
4146 | } | |
4147 | ||
4148 | int follow_pte_pmd(struct mm_struct *mm, unsigned long address, | |
a4d1a885 | 4149 | unsigned long *start, unsigned long *end, |
09796395 RZ |
4150 | pte_t **ptepp, pmd_t **pmdpp, spinlock_t **ptlp) |
4151 | { | |
4152 | int res; | |
4153 | ||
4154 | /* (void) is needed to make gcc happy */ | |
4155 | (void) __cond_lock(*ptlp, | |
a4d1a885 JG |
4156 | !(res = __follow_pte_pmd(mm, address, start, end, |
4157 | ptepp, pmdpp, ptlp))); | |
1b36ba81 NK |
4158 | return res; |
4159 | } | |
09796395 | 4160 | EXPORT_SYMBOL(follow_pte_pmd); |
1b36ba81 | 4161 | |
3b6748e2 JW |
4162 | /** |
4163 | * follow_pfn - look up PFN at a user virtual address | |
4164 | * @vma: memory mapping | |
4165 | * @address: user virtual address | |
4166 | * @pfn: location to store found PFN | |
4167 | * | |
4168 | * Only IO mappings and raw PFN mappings are allowed. | |
4169 | * | |
4170 | * Returns zero and the pfn at @pfn on success, -ve otherwise. | |
4171 | */ | |
4172 | int follow_pfn(struct vm_area_struct *vma, unsigned long address, | |
4173 | unsigned long *pfn) | |
4174 | { | |
4175 | int ret = -EINVAL; | |
4176 | spinlock_t *ptl; | |
4177 | pte_t *ptep; | |
4178 | ||
4179 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) | |
4180 | return ret; | |
4181 | ||
4182 | ret = follow_pte(vma->vm_mm, address, &ptep, &ptl); | |
4183 | if (ret) | |
4184 | return ret; | |
4185 | *pfn = pte_pfn(*ptep); | |
4186 | pte_unmap_unlock(ptep, ptl); | |
4187 | return 0; | |
4188 | } | |
4189 | EXPORT_SYMBOL(follow_pfn); | |
4190 | ||
28b2ee20 | 4191 | #ifdef CONFIG_HAVE_IOREMAP_PROT |
d87fe660 | 4192 | int follow_phys(struct vm_area_struct *vma, |
4193 | unsigned long address, unsigned int flags, | |
4194 | unsigned long *prot, resource_size_t *phys) | |
28b2ee20 | 4195 | { |
03668a4d | 4196 | int ret = -EINVAL; |
28b2ee20 RR |
4197 | pte_t *ptep, pte; |
4198 | spinlock_t *ptl; | |
28b2ee20 | 4199 | |
d87fe660 | 4200 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) |
4201 | goto out; | |
28b2ee20 | 4202 | |
03668a4d | 4203 | if (follow_pte(vma->vm_mm, address, &ptep, &ptl)) |
d87fe660 | 4204 | goto out; |
28b2ee20 | 4205 | pte = *ptep; |
03668a4d | 4206 | |
28b2ee20 RR |
4207 | if ((flags & FOLL_WRITE) && !pte_write(pte)) |
4208 | goto unlock; | |
28b2ee20 RR |
4209 | |
4210 | *prot = pgprot_val(pte_pgprot(pte)); | |
03668a4d | 4211 | *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; |
28b2ee20 | 4212 | |
03668a4d | 4213 | ret = 0; |
28b2ee20 RR |
4214 | unlock: |
4215 | pte_unmap_unlock(ptep, ptl); | |
4216 | out: | |
d87fe660 | 4217 | return ret; |
28b2ee20 RR |
4218 | } |
4219 | ||
4220 | int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, | |
4221 | void *buf, int len, int write) | |
4222 | { | |
4223 | resource_size_t phys_addr; | |
4224 | unsigned long prot = 0; | |
2bc7273b | 4225 | void __iomem *maddr; |
28b2ee20 RR |
4226 | int offset = addr & (PAGE_SIZE-1); |
4227 | ||
d87fe660 | 4228 | if (follow_phys(vma, addr, write, &prot, &phys_addr)) |
28b2ee20 RR |
4229 | return -EINVAL; |
4230 | ||
9cb12d7b | 4231 | maddr = ioremap_prot(phys_addr, PAGE_ALIGN(len + offset), prot); |
28b2ee20 RR |
4232 | if (write) |
4233 | memcpy_toio(maddr + offset, buf, len); | |
4234 | else | |
4235 | memcpy_fromio(buf, maddr + offset, len); | |
4236 | iounmap(maddr); | |
4237 | ||
4238 | return len; | |
4239 | } | |
5a73633e | 4240 | EXPORT_SYMBOL_GPL(generic_access_phys); |
28b2ee20 RR |
4241 | #endif |
4242 | ||
0ec76a11 | 4243 | /* |
206cb636 SW |
4244 | * Access another process' address space as given in mm. If non-NULL, use the |
4245 | * given task for page fault accounting. | |
0ec76a11 | 4246 | */ |
84d77d3f | 4247 | int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, |
442486ec | 4248 | unsigned long addr, void *buf, int len, unsigned int gup_flags) |
0ec76a11 | 4249 | { |
0ec76a11 | 4250 | struct vm_area_struct *vma; |
0ec76a11 | 4251 | void *old_buf = buf; |
442486ec | 4252 | int write = gup_flags & FOLL_WRITE; |
0ec76a11 | 4253 | |
0ec76a11 | 4254 | down_read(&mm->mmap_sem); |
183ff22b | 4255 | /* ignore errors, just check how much was successfully transferred */ |
0ec76a11 DH |
4256 | while (len) { |
4257 | int bytes, ret, offset; | |
4258 | void *maddr; | |
28b2ee20 | 4259 | struct page *page = NULL; |
0ec76a11 | 4260 | |
1e987790 | 4261 | ret = get_user_pages_remote(tsk, mm, addr, 1, |
5b56d49f | 4262 | gup_flags, &page, &vma, NULL); |
28b2ee20 | 4263 | if (ret <= 0) { |
dbffcd03 RR |
4264 | #ifndef CONFIG_HAVE_IOREMAP_PROT |
4265 | break; | |
4266 | #else | |
28b2ee20 RR |
4267 | /* |
4268 | * Check if this is a VM_IO | VM_PFNMAP VMA, which | |
4269 | * we can access using slightly different code. | |
4270 | */ | |
28b2ee20 | 4271 | vma = find_vma(mm, addr); |
fe936dfc | 4272 | if (!vma || vma->vm_start > addr) |
28b2ee20 RR |
4273 | break; |
4274 | if (vma->vm_ops && vma->vm_ops->access) | |
4275 | ret = vma->vm_ops->access(vma, addr, buf, | |
4276 | len, write); | |
4277 | if (ret <= 0) | |
28b2ee20 RR |
4278 | break; |
4279 | bytes = ret; | |
dbffcd03 | 4280 | #endif |
0ec76a11 | 4281 | } else { |
28b2ee20 RR |
4282 | bytes = len; |
4283 | offset = addr & (PAGE_SIZE-1); | |
4284 | if (bytes > PAGE_SIZE-offset) | |
4285 | bytes = PAGE_SIZE-offset; | |
4286 | ||
4287 | maddr = kmap(page); | |
4288 | if (write) { | |
4289 | copy_to_user_page(vma, page, addr, | |
4290 | maddr + offset, buf, bytes); | |
4291 | set_page_dirty_lock(page); | |
4292 | } else { | |
4293 | copy_from_user_page(vma, page, addr, | |
4294 | buf, maddr + offset, bytes); | |
4295 | } | |
4296 | kunmap(page); | |
09cbfeaf | 4297 | put_page(page); |
0ec76a11 | 4298 | } |
0ec76a11 DH |
4299 | len -= bytes; |
4300 | buf += bytes; | |
4301 | addr += bytes; | |
4302 | } | |
4303 | up_read(&mm->mmap_sem); | |
0ec76a11 DH |
4304 | |
4305 | return buf - old_buf; | |
4306 | } | |
03252919 | 4307 | |
5ddd36b9 | 4308 | /** |
ae91dbfc | 4309 | * access_remote_vm - access another process' address space |
5ddd36b9 SW |
4310 | * @mm: the mm_struct of the target address space |
4311 | * @addr: start address to access | |
4312 | * @buf: source or destination buffer | |
4313 | * @len: number of bytes to transfer | |
6347e8d5 | 4314 | * @gup_flags: flags modifying lookup behaviour |
5ddd36b9 SW |
4315 | * |
4316 | * The caller must hold a reference on @mm. | |
4317 | */ | |
4318 | int access_remote_vm(struct mm_struct *mm, unsigned long addr, | |
6347e8d5 | 4319 | void *buf, int len, unsigned int gup_flags) |
5ddd36b9 | 4320 | { |
6347e8d5 | 4321 | return __access_remote_vm(NULL, mm, addr, buf, len, gup_flags); |
5ddd36b9 SW |
4322 | } |
4323 | ||
206cb636 SW |
4324 | /* |
4325 | * Access another process' address space. | |
4326 | * Source/target buffer must be kernel space, | |
4327 | * Do not walk the page table directly, use get_user_pages | |
4328 | */ | |
4329 | int access_process_vm(struct task_struct *tsk, unsigned long addr, | |
f307ab6d | 4330 | void *buf, int len, unsigned int gup_flags) |
206cb636 SW |
4331 | { |
4332 | struct mm_struct *mm; | |
4333 | int ret; | |
4334 | ||
4335 | mm = get_task_mm(tsk); | |
4336 | if (!mm) | |
4337 | return 0; | |
4338 | ||
f307ab6d | 4339 | ret = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags); |
442486ec | 4340 | |
206cb636 SW |
4341 | mmput(mm); |
4342 | ||
4343 | return ret; | |
4344 | } | |
fcd35857 | 4345 | EXPORT_SYMBOL_GPL(access_process_vm); |
206cb636 | 4346 | |
03252919 AK |
4347 | /* |
4348 | * Print the name of a VMA. | |
4349 | */ | |
4350 | void print_vma_addr(char *prefix, unsigned long ip) | |
4351 | { | |
4352 | struct mm_struct *mm = current->mm; | |
4353 | struct vm_area_struct *vma; | |
4354 | ||
e8bff74a IM |
4355 | /* |
4356 | * Do not print if we are in atomic | |
4357 | * contexts (in exception stacks, etc.): | |
4358 | */ | |
4359 | if (preempt_count()) | |
4360 | return; | |
4361 | ||
03252919 AK |
4362 | down_read(&mm->mmap_sem); |
4363 | vma = find_vma(mm, ip); | |
4364 | if (vma && vma->vm_file) { | |
4365 | struct file *f = vma->vm_file; | |
4366 | char *buf = (char *)__get_free_page(GFP_KERNEL); | |
4367 | if (buf) { | |
2fbc57c5 | 4368 | char *p; |
03252919 | 4369 | |
9bf39ab2 | 4370 | p = file_path(f, buf, PAGE_SIZE); |
03252919 AK |
4371 | if (IS_ERR(p)) |
4372 | p = "?"; | |
2fbc57c5 | 4373 | printk("%s%s[%lx+%lx]", prefix, kbasename(p), |
03252919 AK |
4374 | vma->vm_start, |
4375 | vma->vm_end - vma->vm_start); | |
4376 | free_page((unsigned long)buf); | |
4377 | } | |
4378 | } | |
51a07e50 | 4379 | up_read(&mm->mmap_sem); |
03252919 | 4380 | } |
3ee1afa3 | 4381 | |
662bbcb2 | 4382 | #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP) |
9ec23531 | 4383 | void __might_fault(const char *file, int line) |
3ee1afa3 | 4384 | { |
95156f00 PZ |
4385 | /* |
4386 | * Some code (nfs/sunrpc) uses socket ops on kernel memory while | |
4387 | * holding the mmap_sem, this is safe because kernel memory doesn't | |
4388 | * get paged out, therefore we'll never actually fault, and the | |
4389 | * below annotations will generate false positives. | |
4390 | */ | |
db68ce10 | 4391 | if (uaccess_kernel()) |
95156f00 | 4392 | return; |
9ec23531 | 4393 | if (pagefault_disabled()) |
662bbcb2 | 4394 | return; |
9ec23531 DH |
4395 | __might_sleep(file, line, 0); |
4396 | #if defined(CONFIG_DEBUG_ATOMIC_SLEEP) | |
662bbcb2 | 4397 | if (current->mm) |
3ee1afa3 | 4398 | might_lock_read(¤t->mm->mmap_sem); |
9ec23531 | 4399 | #endif |
3ee1afa3 | 4400 | } |
9ec23531 | 4401 | EXPORT_SYMBOL(__might_fault); |
3ee1afa3 | 4402 | #endif |
47ad8475 AA |
4403 | |
4404 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) | |
4405 | static void clear_gigantic_page(struct page *page, | |
4406 | unsigned long addr, | |
4407 | unsigned int pages_per_huge_page) | |
4408 | { | |
4409 | int i; | |
4410 | struct page *p = page; | |
4411 | ||
4412 | might_sleep(); | |
4413 | for (i = 0; i < pages_per_huge_page; | |
4414 | i++, p = mem_map_next(p, page, i)) { | |
4415 | cond_resched(); | |
4416 | clear_user_highpage(p, addr + i * PAGE_SIZE); | |
4417 | } | |
4418 | } | |
4419 | void clear_huge_page(struct page *page, | |
4420 | unsigned long addr, unsigned int pages_per_huge_page) | |
4421 | { | |
4422 | int i; | |
4423 | ||
4424 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { | |
4425 | clear_gigantic_page(page, addr, pages_per_huge_page); | |
4426 | return; | |
4427 | } | |
4428 | ||
4429 | might_sleep(); | |
4430 | for (i = 0; i < pages_per_huge_page; i++) { | |
4431 | cond_resched(); | |
4432 | clear_user_highpage(page + i, addr + i * PAGE_SIZE); | |
4433 | } | |
4434 | } | |
4435 | ||
4436 | static void copy_user_gigantic_page(struct page *dst, struct page *src, | |
4437 | unsigned long addr, | |
4438 | struct vm_area_struct *vma, | |
4439 | unsigned int pages_per_huge_page) | |
4440 | { | |
4441 | int i; | |
4442 | struct page *dst_base = dst; | |
4443 | struct page *src_base = src; | |
4444 | ||
4445 | for (i = 0; i < pages_per_huge_page; ) { | |
4446 | cond_resched(); | |
4447 | copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma); | |
4448 | ||
4449 | i++; | |
4450 | dst = mem_map_next(dst, dst_base, i); | |
4451 | src = mem_map_next(src, src_base, i); | |
4452 | } | |
4453 | } | |
4454 | ||
4455 | void copy_user_huge_page(struct page *dst, struct page *src, | |
4456 | unsigned long addr, struct vm_area_struct *vma, | |
4457 | unsigned int pages_per_huge_page) | |
4458 | { | |
4459 | int i; | |
4460 | ||
4461 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { | |
4462 | copy_user_gigantic_page(dst, src, addr, vma, | |
4463 | pages_per_huge_page); | |
4464 | return; | |
4465 | } | |
4466 | ||
4467 | might_sleep(); | |
4468 | for (i = 0; i < pages_per_huge_page; i++) { | |
4469 | cond_resched(); | |
4470 | copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE, vma); | |
4471 | } | |
4472 | } | |
fa4d75c1 MK |
4473 | |
4474 | long copy_huge_page_from_user(struct page *dst_page, | |
4475 | const void __user *usr_src, | |
810a56b9 MK |
4476 | unsigned int pages_per_huge_page, |
4477 | bool allow_pagefault) | |
fa4d75c1 MK |
4478 | { |
4479 | void *src = (void *)usr_src; | |
4480 | void *page_kaddr; | |
4481 | unsigned long i, rc = 0; | |
4482 | unsigned long ret_val = pages_per_huge_page * PAGE_SIZE; | |
4483 | ||
4484 | for (i = 0; i < pages_per_huge_page; i++) { | |
810a56b9 MK |
4485 | if (allow_pagefault) |
4486 | page_kaddr = kmap(dst_page + i); | |
4487 | else | |
4488 | page_kaddr = kmap_atomic(dst_page + i); | |
fa4d75c1 MK |
4489 | rc = copy_from_user(page_kaddr, |
4490 | (const void __user *)(src + i * PAGE_SIZE), | |
4491 | PAGE_SIZE); | |
810a56b9 MK |
4492 | if (allow_pagefault) |
4493 | kunmap(dst_page + i); | |
4494 | else | |
4495 | kunmap_atomic(page_kaddr); | |
fa4d75c1 MK |
4496 | |
4497 | ret_val -= (PAGE_SIZE - rc); | |
4498 | if (rc) | |
4499 | break; | |
4500 | ||
4501 | cond_resched(); | |
4502 | } | |
4503 | return ret_val; | |
4504 | } | |
47ad8475 | 4505 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */ |
49076ec2 | 4506 | |
40b64acd | 4507 | #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS |
b35f1819 KS |
4508 | |
4509 | static struct kmem_cache *page_ptl_cachep; | |
4510 | ||
4511 | void __init ptlock_cache_init(void) | |
4512 | { | |
4513 | page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0, | |
4514 | SLAB_PANIC, NULL); | |
4515 | } | |
4516 | ||
539edb58 | 4517 | bool ptlock_alloc(struct page *page) |
49076ec2 KS |
4518 | { |
4519 | spinlock_t *ptl; | |
4520 | ||
b35f1819 | 4521 | ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL); |
49076ec2 KS |
4522 | if (!ptl) |
4523 | return false; | |
539edb58 | 4524 | page->ptl = ptl; |
49076ec2 KS |
4525 | return true; |
4526 | } | |
4527 | ||
539edb58 | 4528 | void ptlock_free(struct page *page) |
49076ec2 | 4529 | { |
b35f1819 | 4530 | kmem_cache_free(page_ptl_cachep, page->ptl); |
49076ec2 KS |
4531 | } |
4532 | #endif |