]>
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 | 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 LT |
1515 | mmu_notifier_invalidate_range_start(mm, start, end); |
1516 | for ( ; vma && vma->vm_start < end; vma = vma->vm_next) | |
ecf1385d | 1517 | unmap_single_vma(&tlb, vma, start, end, NULL); |
7e027b14 LT |
1518 | mmu_notifier_invalidate_range_end(mm, start, end); |
1519 | tlb_finish_mmu(&tlb, start, end); | |
1da177e4 LT |
1520 | } |
1521 | ||
f5cc4eef AV |
1522 | /** |
1523 | * zap_page_range_single - remove user pages in a given range | |
1524 | * @vma: vm_area_struct holding the applicable pages | |
1525 | * @address: starting address of pages to zap | |
1526 | * @size: number of bytes to zap | |
8a5f14a2 | 1527 | * @details: details of shared cache invalidation |
f5cc4eef AV |
1528 | * |
1529 | * The range must fit into one VMA. | |
1da177e4 | 1530 | */ |
f5cc4eef | 1531 | static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address, |
1da177e4 LT |
1532 | unsigned long size, struct zap_details *details) |
1533 | { | |
1534 | struct mm_struct *mm = vma->vm_mm; | |
d16dfc55 | 1535 | struct mmu_gather tlb; |
1da177e4 | 1536 | unsigned long end = address + size; |
1da177e4 | 1537 | |
1da177e4 | 1538 | lru_add_drain(); |
2b047252 | 1539 | tlb_gather_mmu(&tlb, mm, address, end); |
365e9c87 | 1540 | update_hiwater_rss(mm); |
f5cc4eef | 1541 | mmu_notifier_invalidate_range_start(mm, address, end); |
4f74d2c8 | 1542 | unmap_single_vma(&tlb, vma, address, end, details); |
f5cc4eef | 1543 | mmu_notifier_invalidate_range_end(mm, address, end); |
d16dfc55 | 1544 | tlb_finish_mmu(&tlb, address, end); |
1da177e4 LT |
1545 | } |
1546 | ||
c627f9cc JS |
1547 | /** |
1548 | * zap_vma_ptes - remove ptes mapping the vma | |
1549 | * @vma: vm_area_struct holding ptes to be zapped | |
1550 | * @address: starting address of pages to zap | |
1551 | * @size: number of bytes to zap | |
1552 | * | |
1553 | * This function only unmaps ptes assigned to VM_PFNMAP vmas. | |
1554 | * | |
1555 | * The entire address range must be fully contained within the vma. | |
1556 | * | |
1557 | * Returns 0 if successful. | |
1558 | */ | |
1559 | int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, | |
1560 | unsigned long size) | |
1561 | { | |
1562 | if (address < vma->vm_start || address + size > vma->vm_end || | |
1563 | !(vma->vm_flags & VM_PFNMAP)) | |
1564 | return -1; | |
f5cc4eef | 1565 | zap_page_range_single(vma, address, size, NULL); |
c627f9cc JS |
1566 | return 0; |
1567 | } | |
1568 | EXPORT_SYMBOL_GPL(zap_vma_ptes); | |
1569 | ||
25ca1d6c | 1570 | pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr, |
920c7a5d | 1571 | spinlock_t **ptl) |
c9cfcddf | 1572 | { |
c2febafc KS |
1573 | pgd_t *pgd; |
1574 | p4d_t *p4d; | |
1575 | pud_t *pud; | |
1576 | pmd_t *pmd; | |
1577 | ||
1578 | pgd = pgd_offset(mm, addr); | |
1579 | p4d = p4d_alloc(mm, pgd, addr); | |
1580 | if (!p4d) | |
1581 | return NULL; | |
1582 | pud = pud_alloc(mm, p4d, addr); | |
1583 | if (!pud) | |
1584 | return NULL; | |
1585 | pmd = pmd_alloc(mm, pud, addr); | |
1586 | if (!pmd) | |
1587 | return NULL; | |
1588 | ||
1589 | VM_BUG_ON(pmd_trans_huge(*pmd)); | |
1590 | return pte_alloc_map_lock(mm, pmd, addr, ptl); | |
c9cfcddf LT |
1591 | } |
1592 | ||
238f58d8 LT |
1593 | /* |
1594 | * This is the old fallback for page remapping. | |
1595 | * | |
1596 | * For historical reasons, it only allows reserved pages. Only | |
1597 | * old drivers should use this, and they needed to mark their | |
1598 | * pages reserved for the old functions anyway. | |
1599 | */ | |
423bad60 NP |
1600 | static int insert_page(struct vm_area_struct *vma, unsigned long addr, |
1601 | struct page *page, pgprot_t prot) | |
238f58d8 | 1602 | { |
423bad60 | 1603 | struct mm_struct *mm = vma->vm_mm; |
238f58d8 | 1604 | int retval; |
c9cfcddf | 1605 | pte_t *pte; |
8a9f3ccd BS |
1606 | spinlock_t *ptl; |
1607 | ||
238f58d8 | 1608 | retval = -EINVAL; |
a145dd41 | 1609 | if (PageAnon(page)) |
5b4e655e | 1610 | goto out; |
238f58d8 LT |
1611 | retval = -ENOMEM; |
1612 | flush_dcache_page(page); | |
c9cfcddf | 1613 | pte = get_locked_pte(mm, addr, &ptl); |
238f58d8 | 1614 | if (!pte) |
5b4e655e | 1615 | goto out; |
238f58d8 LT |
1616 | retval = -EBUSY; |
1617 | if (!pte_none(*pte)) | |
1618 | goto out_unlock; | |
1619 | ||
1620 | /* Ok, finally just insert the thing.. */ | |
1621 | get_page(page); | |
eca56ff9 | 1622 | inc_mm_counter_fast(mm, mm_counter_file(page)); |
dd78fedd | 1623 | page_add_file_rmap(page, false); |
238f58d8 LT |
1624 | set_pte_at(mm, addr, pte, mk_pte(page, prot)); |
1625 | ||
1626 | retval = 0; | |
8a9f3ccd BS |
1627 | pte_unmap_unlock(pte, ptl); |
1628 | return retval; | |
238f58d8 LT |
1629 | out_unlock: |
1630 | pte_unmap_unlock(pte, ptl); | |
1631 | out: | |
1632 | return retval; | |
1633 | } | |
1634 | ||
bfa5bf6d REB |
1635 | /** |
1636 | * vm_insert_page - insert single page into user vma | |
1637 | * @vma: user vma to map to | |
1638 | * @addr: target user address of this page | |
1639 | * @page: source kernel page | |
1640 | * | |
a145dd41 LT |
1641 | * This allows drivers to insert individual pages they've allocated |
1642 | * into a user vma. | |
1643 | * | |
1644 | * The page has to be a nice clean _individual_ kernel allocation. | |
1645 | * If you allocate a compound page, you need to have marked it as | |
1646 | * such (__GFP_COMP), or manually just split the page up yourself | |
8dfcc9ba | 1647 | * (see split_page()). |
a145dd41 LT |
1648 | * |
1649 | * NOTE! Traditionally this was done with "remap_pfn_range()" which | |
1650 | * took an arbitrary page protection parameter. This doesn't allow | |
1651 | * that. Your vma protection will have to be set up correctly, which | |
1652 | * means that if you want a shared writable mapping, you'd better | |
1653 | * ask for a shared writable mapping! | |
1654 | * | |
1655 | * The page does not need to be reserved. | |
4b6e1e37 KK |
1656 | * |
1657 | * Usually this function is called from f_op->mmap() handler | |
1658 | * under mm->mmap_sem write-lock, so it can change vma->vm_flags. | |
1659 | * Caller must set VM_MIXEDMAP on vma if it wants to call this | |
1660 | * function from other places, for example from page-fault handler. | |
a145dd41 | 1661 | */ |
423bad60 NP |
1662 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, |
1663 | struct page *page) | |
a145dd41 LT |
1664 | { |
1665 | if (addr < vma->vm_start || addr >= vma->vm_end) | |
1666 | return -EFAULT; | |
1667 | if (!page_count(page)) | |
1668 | return -EINVAL; | |
4b6e1e37 KK |
1669 | if (!(vma->vm_flags & VM_MIXEDMAP)) { |
1670 | BUG_ON(down_read_trylock(&vma->vm_mm->mmap_sem)); | |
1671 | BUG_ON(vma->vm_flags & VM_PFNMAP); | |
1672 | vma->vm_flags |= VM_MIXEDMAP; | |
1673 | } | |
423bad60 | 1674 | return insert_page(vma, addr, page, vma->vm_page_prot); |
a145dd41 | 1675 | } |
e3c3374f | 1676 | EXPORT_SYMBOL(vm_insert_page); |
a145dd41 | 1677 | |
423bad60 | 1678 | static int insert_pfn(struct vm_area_struct *vma, unsigned long addr, |
01c8f1c4 | 1679 | pfn_t pfn, pgprot_t prot) |
423bad60 NP |
1680 | { |
1681 | struct mm_struct *mm = vma->vm_mm; | |
1682 | int retval; | |
1683 | pte_t *pte, entry; | |
1684 | spinlock_t *ptl; | |
1685 | ||
1686 | retval = -ENOMEM; | |
1687 | pte = get_locked_pte(mm, addr, &ptl); | |
1688 | if (!pte) | |
1689 | goto out; | |
1690 | retval = -EBUSY; | |
1691 | if (!pte_none(*pte)) | |
1692 | goto out_unlock; | |
1693 | ||
1694 | /* Ok, finally just insert the thing.. */ | |
01c8f1c4 DW |
1695 | if (pfn_t_devmap(pfn)) |
1696 | entry = pte_mkdevmap(pfn_t_pte(pfn, prot)); | |
1697 | else | |
1698 | entry = pte_mkspecial(pfn_t_pte(pfn, prot)); | |
423bad60 | 1699 | set_pte_at(mm, addr, pte, entry); |
4b3073e1 | 1700 | update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */ |
423bad60 NP |
1701 | |
1702 | retval = 0; | |
1703 | out_unlock: | |
1704 | pte_unmap_unlock(pte, ptl); | |
1705 | out: | |
1706 | return retval; | |
1707 | } | |
1708 | ||
e0dc0d8f NP |
1709 | /** |
1710 | * vm_insert_pfn - insert single pfn into user vma | |
1711 | * @vma: user vma to map to | |
1712 | * @addr: target user address of this page | |
1713 | * @pfn: source kernel pfn | |
1714 | * | |
c462f179 | 1715 | * Similar to vm_insert_page, this allows drivers to insert individual pages |
e0dc0d8f NP |
1716 | * they've allocated into a user vma. Same comments apply. |
1717 | * | |
1718 | * This function should only be called from a vm_ops->fault handler, and | |
1719 | * in that case the handler should return NULL. | |
0d71d10a NP |
1720 | * |
1721 | * vma cannot be a COW mapping. | |
1722 | * | |
1723 | * As this is called only for pages that do not currently exist, we | |
1724 | * do not need to flush old virtual caches or the TLB. | |
e0dc0d8f NP |
1725 | */ |
1726 | int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr, | |
423bad60 | 1727 | unsigned long pfn) |
1745cbc5 AL |
1728 | { |
1729 | return vm_insert_pfn_prot(vma, addr, pfn, vma->vm_page_prot); | |
1730 | } | |
1731 | EXPORT_SYMBOL(vm_insert_pfn); | |
1732 | ||
1733 | /** | |
1734 | * vm_insert_pfn_prot - insert single pfn into user vma with specified pgprot | |
1735 | * @vma: user vma to map to | |
1736 | * @addr: target user address of this page | |
1737 | * @pfn: source kernel pfn | |
1738 | * @pgprot: pgprot flags for the inserted page | |
1739 | * | |
1740 | * This is exactly like vm_insert_pfn, except that it allows drivers to | |
1741 | * to override pgprot on a per-page basis. | |
1742 | * | |
1743 | * This only makes sense for IO mappings, and it makes no sense for | |
1744 | * cow mappings. In general, using multiple vmas is preferable; | |
1745 | * vm_insert_pfn_prot should only be used if using multiple VMAs is | |
1746 | * impractical. | |
1747 | */ | |
1748 | int vm_insert_pfn_prot(struct vm_area_struct *vma, unsigned long addr, | |
1749 | unsigned long pfn, pgprot_t pgprot) | |
e0dc0d8f | 1750 | { |
2ab64037 | 1751 | int ret; |
7e675137 NP |
1752 | /* |
1753 | * Technically, architectures with pte_special can avoid all these | |
1754 | * restrictions (same for remap_pfn_range). However we would like | |
1755 | * consistency in testing and feature parity among all, so we should | |
1756 | * try to keep these invariants in place for everybody. | |
1757 | */ | |
b379d790 JH |
1758 | BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); |
1759 | BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == | |
1760 | (VM_PFNMAP|VM_MIXEDMAP)); | |
1761 | BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); | |
1762 | BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); | |
e0dc0d8f | 1763 | |
423bad60 NP |
1764 | if (addr < vma->vm_start || addr >= vma->vm_end) |
1765 | return -EFAULT; | |
308a047c BP |
1766 | |
1767 | track_pfn_insert(vma, &pgprot, __pfn_to_pfn_t(pfn, PFN_DEV)); | |
2ab64037 | 1768 | |
01c8f1c4 | 1769 | ret = insert_pfn(vma, addr, __pfn_to_pfn_t(pfn, PFN_DEV), pgprot); |
2ab64037 | 1770 | |
2ab64037 | 1771 | return ret; |
423bad60 | 1772 | } |
1745cbc5 | 1773 | EXPORT_SYMBOL(vm_insert_pfn_prot); |
e0dc0d8f | 1774 | |
423bad60 | 1775 | int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr, |
01c8f1c4 | 1776 | pfn_t pfn) |
423bad60 | 1777 | { |
87744ab3 DW |
1778 | pgprot_t pgprot = vma->vm_page_prot; |
1779 | ||
423bad60 | 1780 | BUG_ON(!(vma->vm_flags & VM_MIXEDMAP)); |
e0dc0d8f | 1781 | |
423bad60 NP |
1782 | if (addr < vma->vm_start || addr >= vma->vm_end) |
1783 | return -EFAULT; | |
308a047c BP |
1784 | |
1785 | track_pfn_insert(vma, &pgprot, pfn); | |
e0dc0d8f | 1786 | |
423bad60 NP |
1787 | /* |
1788 | * If we don't have pte special, then we have to use the pfn_valid() | |
1789 | * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* | |
1790 | * refcount the page if pfn_valid is true (hence insert_page rather | |
62eede62 HD |
1791 | * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP |
1792 | * without pte special, it would there be refcounted as a normal page. | |
423bad60 | 1793 | */ |
03fc2da6 | 1794 | if (!HAVE_PTE_SPECIAL && !pfn_t_devmap(pfn) && pfn_t_valid(pfn)) { |
423bad60 NP |
1795 | struct page *page; |
1796 | ||
03fc2da6 DW |
1797 | /* |
1798 | * At this point we are committed to insert_page() | |
1799 | * regardless of whether the caller specified flags that | |
1800 | * result in pfn_t_has_page() == false. | |
1801 | */ | |
1802 | page = pfn_to_page(pfn_t_to_pfn(pfn)); | |
87744ab3 | 1803 | return insert_page(vma, addr, page, pgprot); |
423bad60 | 1804 | } |
87744ab3 | 1805 | return insert_pfn(vma, addr, pfn, pgprot); |
e0dc0d8f | 1806 | } |
423bad60 | 1807 | EXPORT_SYMBOL(vm_insert_mixed); |
e0dc0d8f | 1808 | |
1da177e4 LT |
1809 | /* |
1810 | * maps a range of physical memory into the requested pages. the old | |
1811 | * mappings are removed. any references to nonexistent pages results | |
1812 | * in null mappings (currently treated as "copy-on-access") | |
1813 | */ | |
1814 | static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, | |
1815 | unsigned long addr, unsigned long end, | |
1816 | unsigned long pfn, pgprot_t prot) | |
1817 | { | |
1818 | pte_t *pte; | |
c74df32c | 1819 | spinlock_t *ptl; |
1da177e4 | 1820 | |
c74df32c | 1821 | pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); |
1da177e4 LT |
1822 | if (!pte) |
1823 | return -ENOMEM; | |
6606c3e0 | 1824 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
1825 | do { |
1826 | BUG_ON(!pte_none(*pte)); | |
7e675137 | 1827 | set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); |
1da177e4 LT |
1828 | pfn++; |
1829 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
6606c3e0 | 1830 | arch_leave_lazy_mmu_mode(); |
c74df32c | 1831 | pte_unmap_unlock(pte - 1, ptl); |
1da177e4 LT |
1832 | return 0; |
1833 | } | |
1834 | ||
1835 | static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, | |
1836 | unsigned long addr, unsigned long end, | |
1837 | unsigned long pfn, pgprot_t prot) | |
1838 | { | |
1839 | pmd_t *pmd; | |
1840 | unsigned long next; | |
1841 | ||
1842 | pfn -= addr >> PAGE_SHIFT; | |
1843 | pmd = pmd_alloc(mm, pud, addr); | |
1844 | if (!pmd) | |
1845 | return -ENOMEM; | |
f66055ab | 1846 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
1da177e4 LT |
1847 | do { |
1848 | next = pmd_addr_end(addr, end); | |
1849 | if (remap_pte_range(mm, pmd, addr, next, | |
1850 | pfn + (addr >> PAGE_SHIFT), prot)) | |
1851 | return -ENOMEM; | |
1852 | } while (pmd++, addr = next, addr != end); | |
1853 | return 0; | |
1854 | } | |
1855 | ||
c2febafc | 1856 | static inline int remap_pud_range(struct mm_struct *mm, p4d_t *p4d, |
1da177e4 LT |
1857 | unsigned long addr, unsigned long end, |
1858 | unsigned long pfn, pgprot_t prot) | |
1859 | { | |
1860 | pud_t *pud; | |
1861 | unsigned long next; | |
1862 | ||
1863 | pfn -= addr >> PAGE_SHIFT; | |
c2febafc | 1864 | pud = pud_alloc(mm, p4d, addr); |
1da177e4 LT |
1865 | if (!pud) |
1866 | return -ENOMEM; | |
1867 | do { | |
1868 | next = pud_addr_end(addr, end); | |
1869 | if (remap_pmd_range(mm, pud, addr, next, | |
1870 | pfn + (addr >> PAGE_SHIFT), prot)) | |
1871 | return -ENOMEM; | |
1872 | } while (pud++, addr = next, addr != end); | |
1873 | return 0; | |
1874 | } | |
1875 | ||
c2febafc KS |
1876 | static inline int remap_p4d_range(struct mm_struct *mm, pgd_t *pgd, |
1877 | unsigned long addr, unsigned long end, | |
1878 | unsigned long pfn, pgprot_t prot) | |
1879 | { | |
1880 | p4d_t *p4d; | |
1881 | unsigned long next; | |
1882 | ||
1883 | pfn -= addr >> PAGE_SHIFT; | |
1884 | p4d = p4d_alloc(mm, pgd, addr); | |
1885 | if (!p4d) | |
1886 | return -ENOMEM; | |
1887 | do { | |
1888 | next = p4d_addr_end(addr, end); | |
1889 | if (remap_pud_range(mm, p4d, addr, next, | |
1890 | pfn + (addr >> PAGE_SHIFT), prot)) | |
1891 | return -ENOMEM; | |
1892 | } while (p4d++, addr = next, addr != end); | |
1893 | return 0; | |
1894 | } | |
1895 | ||
bfa5bf6d REB |
1896 | /** |
1897 | * remap_pfn_range - remap kernel memory to userspace | |
1898 | * @vma: user vma to map to | |
1899 | * @addr: target user address to start at | |
1900 | * @pfn: physical address of kernel memory | |
1901 | * @size: size of map area | |
1902 | * @prot: page protection flags for this mapping | |
1903 | * | |
1904 | * Note: this is only safe if the mm semaphore is held when called. | |
1905 | */ | |
1da177e4 LT |
1906 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, |
1907 | unsigned long pfn, unsigned long size, pgprot_t prot) | |
1908 | { | |
1909 | pgd_t *pgd; | |
1910 | unsigned long next; | |
2d15cab8 | 1911 | unsigned long end = addr + PAGE_ALIGN(size); |
1da177e4 | 1912 | struct mm_struct *mm = vma->vm_mm; |
d5957d2f | 1913 | unsigned long remap_pfn = pfn; |
1da177e4 LT |
1914 | int err; |
1915 | ||
1916 | /* | |
1917 | * Physically remapped pages are special. Tell the | |
1918 | * rest of the world about it: | |
1919 | * VM_IO tells people not to look at these pages | |
1920 | * (accesses can have side effects). | |
6aab341e LT |
1921 | * VM_PFNMAP tells the core MM that the base pages are just |
1922 | * raw PFN mappings, and do not have a "struct page" associated | |
1923 | * with them. | |
314e51b9 KK |
1924 | * VM_DONTEXPAND |
1925 | * Disable vma merging and expanding with mremap(). | |
1926 | * VM_DONTDUMP | |
1927 | * Omit vma from core dump, even when VM_IO turned off. | |
fb155c16 LT |
1928 | * |
1929 | * There's a horrible special case to handle copy-on-write | |
1930 | * behaviour that some programs depend on. We mark the "original" | |
1931 | * un-COW'ed pages by matching them up with "vma->vm_pgoff". | |
b3b9c293 | 1932 | * See vm_normal_page() for details. |
1da177e4 | 1933 | */ |
b3b9c293 KK |
1934 | if (is_cow_mapping(vma->vm_flags)) { |
1935 | if (addr != vma->vm_start || end != vma->vm_end) | |
1936 | return -EINVAL; | |
fb155c16 | 1937 | vma->vm_pgoff = pfn; |
b3b9c293 KK |
1938 | } |
1939 | ||
d5957d2f | 1940 | err = track_pfn_remap(vma, &prot, remap_pfn, addr, PAGE_ALIGN(size)); |
b3b9c293 | 1941 | if (err) |
3c8bb73a | 1942 | return -EINVAL; |
fb155c16 | 1943 | |
314e51b9 | 1944 | vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; |
1da177e4 LT |
1945 | |
1946 | BUG_ON(addr >= end); | |
1947 | pfn -= addr >> PAGE_SHIFT; | |
1948 | pgd = pgd_offset(mm, addr); | |
1949 | flush_cache_range(vma, addr, end); | |
1da177e4 LT |
1950 | do { |
1951 | next = pgd_addr_end(addr, end); | |
c2febafc | 1952 | err = remap_p4d_range(mm, pgd, addr, next, |
1da177e4 LT |
1953 | pfn + (addr >> PAGE_SHIFT), prot); |
1954 | if (err) | |
1955 | break; | |
1956 | } while (pgd++, addr = next, addr != end); | |
2ab64037 | 1957 | |
1958 | if (err) | |
d5957d2f | 1959 | untrack_pfn(vma, remap_pfn, PAGE_ALIGN(size)); |
2ab64037 | 1960 | |
1da177e4 LT |
1961 | return err; |
1962 | } | |
1963 | EXPORT_SYMBOL(remap_pfn_range); | |
1964 | ||
b4cbb197 LT |
1965 | /** |
1966 | * vm_iomap_memory - remap memory to userspace | |
1967 | * @vma: user vma to map to | |
1968 | * @start: start of area | |
1969 | * @len: size of area | |
1970 | * | |
1971 | * This is a simplified io_remap_pfn_range() for common driver use. The | |
1972 | * driver just needs to give us the physical memory range to be mapped, | |
1973 | * we'll figure out the rest from the vma information. | |
1974 | * | |
1975 | * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get | |
1976 | * whatever write-combining details or similar. | |
1977 | */ | |
1978 | int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) | |
1979 | { | |
1980 | unsigned long vm_len, pfn, pages; | |
1981 | ||
1982 | /* Check that the physical memory area passed in looks valid */ | |
1983 | if (start + len < start) | |
1984 | return -EINVAL; | |
1985 | /* | |
1986 | * You *really* shouldn't map things that aren't page-aligned, | |
1987 | * but we've historically allowed it because IO memory might | |
1988 | * just have smaller alignment. | |
1989 | */ | |
1990 | len += start & ~PAGE_MASK; | |
1991 | pfn = start >> PAGE_SHIFT; | |
1992 | pages = (len + ~PAGE_MASK) >> PAGE_SHIFT; | |
1993 | if (pfn + pages < pfn) | |
1994 | return -EINVAL; | |
1995 | ||
1996 | /* We start the mapping 'vm_pgoff' pages into the area */ | |
1997 | if (vma->vm_pgoff > pages) | |
1998 | return -EINVAL; | |
1999 | pfn += vma->vm_pgoff; | |
2000 | pages -= vma->vm_pgoff; | |
2001 | ||
2002 | /* Can we fit all of the mapping? */ | |
2003 | vm_len = vma->vm_end - vma->vm_start; | |
2004 | if (vm_len >> PAGE_SHIFT > pages) | |
2005 | return -EINVAL; | |
2006 | ||
2007 | /* Ok, let it rip */ | |
2008 | return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); | |
2009 | } | |
2010 | EXPORT_SYMBOL(vm_iomap_memory); | |
2011 | ||
aee16b3c JF |
2012 | static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, |
2013 | unsigned long addr, unsigned long end, | |
2014 | pte_fn_t fn, void *data) | |
2015 | { | |
2016 | pte_t *pte; | |
2017 | int err; | |
2f569afd | 2018 | pgtable_t token; |
94909914 | 2019 | spinlock_t *uninitialized_var(ptl); |
aee16b3c JF |
2020 | |
2021 | pte = (mm == &init_mm) ? | |
2022 | pte_alloc_kernel(pmd, addr) : | |
2023 | pte_alloc_map_lock(mm, pmd, addr, &ptl); | |
2024 | if (!pte) | |
2025 | return -ENOMEM; | |
2026 | ||
2027 | BUG_ON(pmd_huge(*pmd)); | |
2028 | ||
38e0edb1 JF |
2029 | arch_enter_lazy_mmu_mode(); |
2030 | ||
2f569afd | 2031 | token = pmd_pgtable(*pmd); |
aee16b3c JF |
2032 | |
2033 | do { | |
c36987e2 | 2034 | err = fn(pte++, token, addr, data); |
aee16b3c JF |
2035 | if (err) |
2036 | break; | |
c36987e2 | 2037 | } while (addr += PAGE_SIZE, addr != end); |
aee16b3c | 2038 | |
38e0edb1 JF |
2039 | arch_leave_lazy_mmu_mode(); |
2040 | ||
aee16b3c JF |
2041 | if (mm != &init_mm) |
2042 | pte_unmap_unlock(pte-1, ptl); | |
2043 | return err; | |
2044 | } | |
2045 | ||
2046 | static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, | |
2047 | unsigned long addr, unsigned long end, | |
2048 | pte_fn_t fn, void *data) | |
2049 | { | |
2050 | pmd_t *pmd; | |
2051 | unsigned long next; | |
2052 | int err; | |
2053 | ||
ceb86879 AK |
2054 | BUG_ON(pud_huge(*pud)); |
2055 | ||
aee16b3c JF |
2056 | pmd = pmd_alloc(mm, pud, addr); |
2057 | if (!pmd) | |
2058 | return -ENOMEM; | |
2059 | do { | |
2060 | next = pmd_addr_end(addr, end); | |
2061 | err = apply_to_pte_range(mm, pmd, addr, next, fn, data); | |
2062 | if (err) | |
2063 | break; | |
2064 | } while (pmd++, addr = next, addr != end); | |
2065 | return err; | |
2066 | } | |
2067 | ||
c2febafc | 2068 | static int apply_to_pud_range(struct mm_struct *mm, p4d_t *p4d, |
aee16b3c JF |
2069 | unsigned long addr, unsigned long end, |
2070 | pte_fn_t fn, void *data) | |
2071 | { | |
2072 | pud_t *pud; | |
2073 | unsigned long next; | |
2074 | int err; | |
2075 | ||
c2febafc | 2076 | pud = pud_alloc(mm, p4d, addr); |
aee16b3c JF |
2077 | if (!pud) |
2078 | return -ENOMEM; | |
2079 | do { | |
2080 | next = pud_addr_end(addr, end); | |
2081 | err = apply_to_pmd_range(mm, pud, addr, next, fn, data); | |
2082 | if (err) | |
2083 | break; | |
2084 | } while (pud++, addr = next, addr != end); | |
2085 | return err; | |
2086 | } | |
2087 | ||
c2febafc KS |
2088 | static int apply_to_p4d_range(struct mm_struct *mm, pgd_t *pgd, |
2089 | unsigned long addr, unsigned long end, | |
2090 | pte_fn_t fn, void *data) | |
2091 | { | |
2092 | p4d_t *p4d; | |
2093 | unsigned long next; | |
2094 | int err; | |
2095 | ||
2096 | p4d = p4d_alloc(mm, pgd, addr); | |
2097 | if (!p4d) | |
2098 | return -ENOMEM; | |
2099 | do { | |
2100 | next = p4d_addr_end(addr, end); | |
2101 | err = apply_to_pud_range(mm, p4d, addr, next, fn, data); | |
2102 | if (err) | |
2103 | break; | |
2104 | } while (p4d++, addr = next, addr != end); | |
2105 | return err; | |
2106 | } | |
2107 | ||
aee16b3c JF |
2108 | /* |
2109 | * Scan a region of virtual memory, filling in page tables as necessary | |
2110 | * and calling a provided function on each leaf page table. | |
2111 | */ | |
2112 | int apply_to_page_range(struct mm_struct *mm, unsigned long addr, | |
2113 | unsigned long size, pte_fn_t fn, void *data) | |
2114 | { | |
2115 | pgd_t *pgd; | |
2116 | unsigned long next; | |
57250a5b | 2117 | unsigned long end = addr + size; |
aee16b3c JF |
2118 | int err; |
2119 | ||
9cb65bc3 MP |
2120 | if (WARN_ON(addr >= end)) |
2121 | return -EINVAL; | |
2122 | ||
aee16b3c JF |
2123 | pgd = pgd_offset(mm, addr); |
2124 | do { | |
2125 | next = pgd_addr_end(addr, end); | |
c2febafc | 2126 | err = apply_to_p4d_range(mm, pgd, addr, next, fn, data); |
aee16b3c JF |
2127 | if (err) |
2128 | break; | |
2129 | } while (pgd++, addr = next, addr != end); | |
57250a5b | 2130 | |
aee16b3c JF |
2131 | return err; |
2132 | } | |
2133 | EXPORT_SYMBOL_GPL(apply_to_page_range); | |
2134 | ||
8f4e2101 | 2135 | /* |
9b4bdd2f KS |
2136 | * handle_pte_fault chooses page fault handler according to an entry which was |
2137 | * read non-atomically. Before making any commitment, on those architectures | |
2138 | * or configurations (e.g. i386 with PAE) which might give a mix of unmatched | |
2139 | * parts, do_swap_page must check under lock before unmapping the pte and | |
2140 | * proceeding (but do_wp_page is only called after already making such a check; | |
a335b2e1 | 2141 | * and do_anonymous_page can safely check later on). |
8f4e2101 | 2142 | */ |
4c21e2f2 | 2143 | static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd, |
8f4e2101 HD |
2144 | pte_t *page_table, pte_t orig_pte) |
2145 | { | |
2146 | int same = 1; | |
2147 | #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT) | |
2148 | if (sizeof(pte_t) > sizeof(unsigned long)) { | |
4c21e2f2 HD |
2149 | spinlock_t *ptl = pte_lockptr(mm, pmd); |
2150 | spin_lock(ptl); | |
8f4e2101 | 2151 | same = pte_same(*page_table, orig_pte); |
4c21e2f2 | 2152 | spin_unlock(ptl); |
8f4e2101 HD |
2153 | } |
2154 | #endif | |
2155 | pte_unmap(page_table); | |
2156 | return same; | |
2157 | } | |
2158 | ||
9de455b2 | 2159 | static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma) |
6aab341e | 2160 | { |
0abdd7a8 DW |
2161 | debug_dma_assert_idle(src); |
2162 | ||
6aab341e LT |
2163 | /* |
2164 | * If the source page was a PFN mapping, we don't have | |
2165 | * a "struct page" for it. We do a best-effort copy by | |
2166 | * just copying from the original user address. If that | |
2167 | * fails, we just zero-fill it. Live with it. | |
2168 | */ | |
2169 | if (unlikely(!src)) { | |
9b04c5fe | 2170 | void *kaddr = kmap_atomic(dst); |
5d2a2dbb LT |
2171 | void __user *uaddr = (void __user *)(va & PAGE_MASK); |
2172 | ||
2173 | /* | |
2174 | * This really shouldn't fail, because the page is there | |
2175 | * in the page tables. But it might just be unreadable, | |
2176 | * in which case we just give up and fill the result with | |
2177 | * zeroes. | |
2178 | */ | |
2179 | if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) | |
3ecb01df | 2180 | clear_page(kaddr); |
9b04c5fe | 2181 | kunmap_atomic(kaddr); |
c4ec7b0d | 2182 | flush_dcache_page(dst); |
0ed361de NP |
2183 | } else |
2184 | copy_user_highpage(dst, src, va, vma); | |
6aab341e LT |
2185 | } |
2186 | ||
c20cd45e MH |
2187 | static gfp_t __get_fault_gfp_mask(struct vm_area_struct *vma) |
2188 | { | |
2189 | struct file *vm_file = vma->vm_file; | |
2190 | ||
2191 | if (vm_file) | |
2192 | return mapping_gfp_mask(vm_file->f_mapping) | __GFP_FS | __GFP_IO; | |
2193 | ||
2194 | /* | |
2195 | * Special mappings (e.g. VDSO) do not have any file so fake | |
2196 | * a default GFP_KERNEL for them. | |
2197 | */ | |
2198 | return GFP_KERNEL; | |
2199 | } | |
2200 | ||
fb09a464 KS |
2201 | /* |
2202 | * Notify the address space that the page is about to become writable so that | |
2203 | * it can prohibit this or wait for the page to get into an appropriate state. | |
2204 | * | |
2205 | * We do this without the lock held, so that it can sleep if it needs to. | |
2206 | */ | |
38b8cb7f | 2207 | static int do_page_mkwrite(struct vm_fault *vmf) |
fb09a464 | 2208 | { |
fb09a464 | 2209 | int ret; |
38b8cb7f JK |
2210 | struct page *page = vmf->page; |
2211 | unsigned int old_flags = vmf->flags; | |
fb09a464 | 2212 | |
38b8cb7f | 2213 | vmf->flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; |
fb09a464 | 2214 | |
11bac800 | 2215 | ret = vmf->vma->vm_ops->page_mkwrite(vmf); |
38b8cb7f JK |
2216 | /* Restore original flags so that caller is not surprised */ |
2217 | vmf->flags = old_flags; | |
fb09a464 KS |
2218 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) |
2219 | return ret; | |
2220 | if (unlikely(!(ret & VM_FAULT_LOCKED))) { | |
2221 | lock_page(page); | |
2222 | if (!page->mapping) { | |
2223 | unlock_page(page); | |
2224 | return 0; /* retry */ | |
2225 | } | |
2226 | ret |= VM_FAULT_LOCKED; | |
2227 | } else | |
2228 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
2229 | return ret; | |
2230 | } | |
2231 | ||
97ba0c2b JK |
2232 | /* |
2233 | * Handle dirtying of a page in shared file mapping on a write fault. | |
2234 | * | |
2235 | * The function expects the page to be locked and unlocks it. | |
2236 | */ | |
2237 | static void fault_dirty_shared_page(struct vm_area_struct *vma, | |
2238 | struct page *page) | |
2239 | { | |
2240 | struct address_space *mapping; | |
2241 | bool dirtied; | |
2242 | bool page_mkwrite = vma->vm_ops && vma->vm_ops->page_mkwrite; | |
2243 | ||
2244 | dirtied = set_page_dirty(page); | |
2245 | VM_BUG_ON_PAGE(PageAnon(page), page); | |
2246 | /* | |
2247 | * Take a local copy of the address_space - page.mapping may be zeroed | |
2248 | * by truncate after unlock_page(). The address_space itself remains | |
2249 | * pinned by vma->vm_file's reference. We rely on unlock_page()'s | |
2250 | * release semantics to prevent the compiler from undoing this copying. | |
2251 | */ | |
2252 | mapping = page_rmapping(page); | |
2253 | unlock_page(page); | |
2254 | ||
2255 | if ((dirtied || page_mkwrite) && mapping) { | |
2256 | /* | |
2257 | * Some device drivers do not set page.mapping | |
2258 | * but still dirty their pages | |
2259 | */ | |
2260 | balance_dirty_pages_ratelimited(mapping); | |
2261 | } | |
2262 | ||
2263 | if (!page_mkwrite) | |
2264 | file_update_time(vma->vm_file); | |
2265 | } | |
2266 | ||
4e047f89 SR |
2267 | /* |
2268 | * Handle write page faults for pages that can be reused in the current vma | |
2269 | * | |
2270 | * This can happen either due to the mapping being with the VM_SHARED flag, | |
2271 | * or due to us being the last reference standing to the page. In either | |
2272 | * case, all we need to do here is to mark the page as writable and update | |
2273 | * any related book-keeping. | |
2274 | */ | |
997dd98d | 2275 | static inline void wp_page_reuse(struct vm_fault *vmf) |
82b0f8c3 | 2276 | __releases(vmf->ptl) |
4e047f89 | 2277 | { |
82b0f8c3 | 2278 | struct vm_area_struct *vma = vmf->vma; |
a41b70d6 | 2279 | struct page *page = vmf->page; |
4e047f89 SR |
2280 | pte_t entry; |
2281 | /* | |
2282 | * Clear the pages cpupid information as the existing | |
2283 | * information potentially belongs to a now completely | |
2284 | * unrelated process. | |
2285 | */ | |
2286 | if (page) | |
2287 | page_cpupid_xchg_last(page, (1 << LAST_CPUPID_SHIFT) - 1); | |
2288 | ||
2994302b JK |
2289 | flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); |
2290 | entry = pte_mkyoung(vmf->orig_pte); | |
4e047f89 | 2291 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
82b0f8c3 JK |
2292 | if (ptep_set_access_flags(vma, vmf->address, vmf->pte, entry, 1)) |
2293 | update_mmu_cache(vma, vmf->address, vmf->pte); | |
2294 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
4e047f89 SR |
2295 | } |
2296 | ||
2f38ab2c SR |
2297 | /* |
2298 | * Handle the case of a page which we actually need to copy to a new page. | |
2299 | * | |
2300 | * Called with mmap_sem locked and the old page referenced, but | |
2301 | * without the ptl held. | |
2302 | * | |
2303 | * High level logic flow: | |
2304 | * | |
2305 | * - Allocate a page, copy the content of the old page to the new one. | |
2306 | * - Handle book keeping and accounting - cgroups, mmu-notifiers, etc. | |
2307 | * - Take the PTL. If the pte changed, bail out and release the allocated page | |
2308 | * - If the pte is still the way we remember it, update the page table and all | |
2309 | * relevant references. This includes dropping the reference the page-table | |
2310 | * held to the old page, as well as updating the rmap. | |
2311 | * - In any case, unlock the PTL and drop the reference we took to the old page. | |
2312 | */ | |
a41b70d6 | 2313 | static int wp_page_copy(struct vm_fault *vmf) |
2f38ab2c | 2314 | { |
82b0f8c3 | 2315 | struct vm_area_struct *vma = vmf->vma; |
bae473a4 | 2316 | struct mm_struct *mm = vma->vm_mm; |
a41b70d6 | 2317 | struct page *old_page = vmf->page; |
2f38ab2c | 2318 | struct page *new_page = NULL; |
2f38ab2c SR |
2319 | pte_t entry; |
2320 | int page_copied = 0; | |
82b0f8c3 | 2321 | const unsigned long mmun_start = vmf->address & PAGE_MASK; |
bae473a4 | 2322 | const unsigned long mmun_end = mmun_start + PAGE_SIZE; |
2f38ab2c SR |
2323 | struct mem_cgroup *memcg; |
2324 | ||
2325 | if (unlikely(anon_vma_prepare(vma))) | |
2326 | goto oom; | |
2327 | ||
2994302b | 2328 | if (is_zero_pfn(pte_pfn(vmf->orig_pte))) { |
82b0f8c3 JK |
2329 | new_page = alloc_zeroed_user_highpage_movable(vma, |
2330 | vmf->address); | |
2f38ab2c SR |
2331 | if (!new_page) |
2332 | goto oom; | |
2333 | } else { | |
bae473a4 | 2334 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, |
82b0f8c3 | 2335 | vmf->address); |
2f38ab2c SR |
2336 | if (!new_page) |
2337 | goto oom; | |
82b0f8c3 | 2338 | cow_user_page(new_page, old_page, vmf->address, vma); |
2f38ab2c | 2339 | } |
2f38ab2c | 2340 | |
f627c2f5 | 2341 | if (mem_cgroup_try_charge(new_page, mm, GFP_KERNEL, &memcg, false)) |
2f38ab2c SR |
2342 | goto oom_free_new; |
2343 | ||
eb3c24f3 MG |
2344 | __SetPageUptodate(new_page); |
2345 | ||
2f38ab2c SR |
2346 | mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); |
2347 | ||
2348 | /* | |
2349 | * Re-check the pte - we dropped the lock | |
2350 | */ | |
82b0f8c3 | 2351 | vmf->pte = pte_offset_map_lock(mm, vmf->pmd, vmf->address, &vmf->ptl); |
2994302b | 2352 | if (likely(pte_same(*vmf->pte, vmf->orig_pte))) { |
2f38ab2c SR |
2353 | if (old_page) { |
2354 | if (!PageAnon(old_page)) { | |
eca56ff9 JM |
2355 | dec_mm_counter_fast(mm, |
2356 | mm_counter_file(old_page)); | |
2f38ab2c SR |
2357 | inc_mm_counter_fast(mm, MM_ANONPAGES); |
2358 | } | |
2359 | } else { | |
2360 | inc_mm_counter_fast(mm, MM_ANONPAGES); | |
2361 | } | |
2994302b | 2362 | flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); |
2f38ab2c SR |
2363 | entry = mk_pte(new_page, vma->vm_page_prot); |
2364 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
2365 | /* | |
2366 | * Clear the pte entry and flush it first, before updating the | |
2367 | * pte with the new entry. This will avoid a race condition | |
2368 | * seen in the presence of one thread doing SMC and another | |
2369 | * thread doing COW. | |
2370 | */ | |
82b0f8c3 JK |
2371 | ptep_clear_flush_notify(vma, vmf->address, vmf->pte); |
2372 | page_add_new_anon_rmap(new_page, vma, vmf->address, false); | |
f627c2f5 | 2373 | mem_cgroup_commit_charge(new_page, memcg, false, false); |
2f38ab2c SR |
2374 | lru_cache_add_active_or_unevictable(new_page, vma); |
2375 | /* | |
2376 | * We call the notify macro here because, when using secondary | |
2377 | * mmu page tables (such as kvm shadow page tables), we want the | |
2378 | * new page to be mapped directly into the secondary page table. | |
2379 | */ | |
82b0f8c3 JK |
2380 | set_pte_at_notify(mm, vmf->address, vmf->pte, entry); |
2381 | update_mmu_cache(vma, vmf->address, vmf->pte); | |
2f38ab2c SR |
2382 | if (old_page) { |
2383 | /* | |
2384 | * Only after switching the pte to the new page may | |
2385 | * we remove the mapcount here. Otherwise another | |
2386 | * process may come and find the rmap count decremented | |
2387 | * before the pte is switched to the new page, and | |
2388 | * "reuse" the old page writing into it while our pte | |
2389 | * here still points into it and can be read by other | |
2390 | * threads. | |
2391 | * | |
2392 | * The critical issue is to order this | |
2393 | * page_remove_rmap with the ptp_clear_flush above. | |
2394 | * Those stores are ordered by (if nothing else,) | |
2395 | * the barrier present in the atomic_add_negative | |
2396 | * in page_remove_rmap. | |
2397 | * | |
2398 | * Then the TLB flush in ptep_clear_flush ensures that | |
2399 | * no process can access the old page before the | |
2400 | * decremented mapcount is visible. And the old page | |
2401 | * cannot be reused until after the decremented | |
2402 | * mapcount is visible. So transitively, TLBs to | |
2403 | * old page will be flushed before it can be reused. | |
2404 | */ | |
d281ee61 | 2405 | page_remove_rmap(old_page, false); |
2f38ab2c SR |
2406 | } |
2407 | ||
2408 | /* Free the old page.. */ | |
2409 | new_page = old_page; | |
2410 | page_copied = 1; | |
2411 | } else { | |
f627c2f5 | 2412 | mem_cgroup_cancel_charge(new_page, memcg, false); |
2f38ab2c SR |
2413 | } |
2414 | ||
2415 | if (new_page) | |
09cbfeaf | 2416 | put_page(new_page); |
2f38ab2c | 2417 | |
82b0f8c3 | 2418 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
2f38ab2c SR |
2419 | mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); |
2420 | if (old_page) { | |
2421 | /* | |
2422 | * Don't let another task, with possibly unlocked vma, | |
2423 | * keep the mlocked page. | |
2424 | */ | |
2425 | if (page_copied && (vma->vm_flags & VM_LOCKED)) { | |
2426 | lock_page(old_page); /* LRU manipulation */ | |
e90309c9 KS |
2427 | if (PageMlocked(old_page)) |
2428 | munlock_vma_page(old_page); | |
2f38ab2c SR |
2429 | unlock_page(old_page); |
2430 | } | |
09cbfeaf | 2431 | put_page(old_page); |
2f38ab2c SR |
2432 | } |
2433 | return page_copied ? VM_FAULT_WRITE : 0; | |
2434 | oom_free_new: | |
09cbfeaf | 2435 | put_page(new_page); |
2f38ab2c SR |
2436 | oom: |
2437 | if (old_page) | |
09cbfeaf | 2438 | put_page(old_page); |
2f38ab2c SR |
2439 | return VM_FAULT_OOM; |
2440 | } | |
2441 | ||
66a6197c JK |
2442 | /** |
2443 | * finish_mkwrite_fault - finish page fault for a shared mapping, making PTE | |
2444 | * writeable once the page is prepared | |
2445 | * | |
2446 | * @vmf: structure describing the fault | |
2447 | * | |
2448 | * This function handles all that is needed to finish a write page fault in a | |
2449 | * shared mapping due to PTE being read-only once the mapped page is prepared. | |
2450 | * It handles locking of PTE and modifying it. The function returns | |
2451 | * VM_FAULT_WRITE on success, 0 when PTE got changed before we acquired PTE | |
2452 | * lock. | |
2453 | * | |
2454 | * The function expects the page to be locked or other protection against | |
2455 | * concurrent faults / writeback (such as DAX radix tree locks). | |
2456 | */ | |
2457 | int finish_mkwrite_fault(struct vm_fault *vmf) | |
2458 | { | |
2459 | WARN_ON_ONCE(!(vmf->vma->vm_flags & VM_SHARED)); | |
2460 | vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, vmf->address, | |
2461 | &vmf->ptl); | |
2462 | /* | |
2463 | * We might have raced with another page fault while we released the | |
2464 | * pte_offset_map_lock. | |
2465 | */ | |
2466 | if (!pte_same(*vmf->pte, vmf->orig_pte)) { | |
2467 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
a19e2553 | 2468 | return VM_FAULT_NOPAGE; |
66a6197c JK |
2469 | } |
2470 | wp_page_reuse(vmf); | |
a19e2553 | 2471 | return 0; |
66a6197c JK |
2472 | } |
2473 | ||
dd906184 BH |
2474 | /* |
2475 | * Handle write page faults for VM_MIXEDMAP or VM_PFNMAP for a VM_SHARED | |
2476 | * mapping | |
2477 | */ | |
2994302b | 2478 | static int wp_pfn_shared(struct vm_fault *vmf) |
dd906184 | 2479 | { |
82b0f8c3 | 2480 | struct vm_area_struct *vma = vmf->vma; |
bae473a4 | 2481 | |
dd906184 | 2482 | if (vma->vm_ops && vma->vm_ops->pfn_mkwrite) { |
dd906184 BH |
2483 | int ret; |
2484 | ||
82b0f8c3 | 2485 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
fe82221f | 2486 | vmf->flags |= FAULT_FLAG_MKWRITE; |
11bac800 | 2487 | ret = vma->vm_ops->pfn_mkwrite(vmf); |
2f89dc12 | 2488 | if (ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)) |
dd906184 | 2489 | return ret; |
66a6197c | 2490 | return finish_mkwrite_fault(vmf); |
dd906184 | 2491 | } |
997dd98d JK |
2492 | wp_page_reuse(vmf); |
2493 | return VM_FAULT_WRITE; | |
dd906184 BH |
2494 | } |
2495 | ||
a41b70d6 | 2496 | static int wp_page_shared(struct vm_fault *vmf) |
82b0f8c3 | 2497 | __releases(vmf->ptl) |
93e478d4 | 2498 | { |
82b0f8c3 | 2499 | struct vm_area_struct *vma = vmf->vma; |
93e478d4 | 2500 | |
a41b70d6 | 2501 | get_page(vmf->page); |
93e478d4 | 2502 | |
93e478d4 SR |
2503 | if (vma->vm_ops && vma->vm_ops->page_mkwrite) { |
2504 | int tmp; | |
2505 | ||
82b0f8c3 | 2506 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
38b8cb7f | 2507 | tmp = do_page_mkwrite(vmf); |
93e478d4 SR |
2508 | if (unlikely(!tmp || (tmp & |
2509 | (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { | |
a41b70d6 | 2510 | put_page(vmf->page); |
93e478d4 SR |
2511 | return tmp; |
2512 | } | |
66a6197c | 2513 | tmp = finish_mkwrite_fault(vmf); |
a19e2553 | 2514 | if (unlikely(tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { |
a41b70d6 | 2515 | unlock_page(vmf->page); |
a41b70d6 | 2516 | put_page(vmf->page); |
66a6197c | 2517 | return tmp; |
93e478d4 | 2518 | } |
66a6197c JK |
2519 | } else { |
2520 | wp_page_reuse(vmf); | |
997dd98d | 2521 | lock_page(vmf->page); |
93e478d4 | 2522 | } |
997dd98d JK |
2523 | fault_dirty_shared_page(vma, vmf->page); |
2524 | put_page(vmf->page); | |
93e478d4 | 2525 | |
997dd98d | 2526 | return VM_FAULT_WRITE; |
93e478d4 SR |
2527 | } |
2528 | ||
1da177e4 LT |
2529 | /* |
2530 | * This routine handles present pages, when users try to write | |
2531 | * to a shared page. It is done by copying the page to a new address | |
2532 | * and decrementing the shared-page counter for the old page. | |
2533 | * | |
1da177e4 LT |
2534 | * Note that this routine assumes that the protection checks have been |
2535 | * done by the caller (the low-level page fault routine in most cases). | |
2536 | * Thus we can safely just mark it writable once we've done any necessary | |
2537 | * COW. | |
2538 | * | |
2539 | * We also mark the page dirty at this point even though the page will | |
2540 | * change only once the write actually happens. This avoids a few races, | |
2541 | * and potentially makes it more efficient. | |
2542 | * | |
8f4e2101 HD |
2543 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2544 | * but allow concurrent faults), with pte both mapped and locked. | |
2545 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2546 | */ |
2994302b | 2547 | static int do_wp_page(struct vm_fault *vmf) |
82b0f8c3 | 2548 | __releases(vmf->ptl) |
1da177e4 | 2549 | { |
82b0f8c3 | 2550 | struct vm_area_struct *vma = vmf->vma; |
1da177e4 | 2551 | |
a41b70d6 JK |
2552 | vmf->page = vm_normal_page(vma, vmf->address, vmf->orig_pte); |
2553 | if (!vmf->page) { | |
251b97f5 | 2554 | /* |
64e45507 PF |
2555 | * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a |
2556 | * VM_PFNMAP VMA. | |
251b97f5 PZ |
2557 | * |
2558 | * We should not cow pages in a shared writeable mapping. | |
dd906184 | 2559 | * Just mark the pages writable and/or call ops->pfn_mkwrite. |
251b97f5 PZ |
2560 | */ |
2561 | if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) == | |
2562 | (VM_WRITE|VM_SHARED)) | |
2994302b | 2563 | return wp_pfn_shared(vmf); |
2f38ab2c | 2564 | |
82b0f8c3 | 2565 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 2566 | return wp_page_copy(vmf); |
251b97f5 | 2567 | } |
1da177e4 | 2568 | |
d08b3851 | 2569 | /* |
ee6a6457 PZ |
2570 | * Take out anonymous pages first, anonymous shared vmas are |
2571 | * not dirty accountable. | |
d08b3851 | 2572 | */ |
a41b70d6 | 2573 | if (PageAnon(vmf->page) && !PageKsm(vmf->page)) { |
6d0a07ed | 2574 | int total_mapcount; |
a41b70d6 JK |
2575 | if (!trylock_page(vmf->page)) { |
2576 | get_page(vmf->page); | |
82b0f8c3 | 2577 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 2578 | lock_page(vmf->page); |
82b0f8c3 JK |
2579 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
2580 | vmf->address, &vmf->ptl); | |
2994302b | 2581 | if (!pte_same(*vmf->pte, vmf->orig_pte)) { |
a41b70d6 | 2582 | unlock_page(vmf->page); |
82b0f8c3 | 2583 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 2584 | put_page(vmf->page); |
28766805 | 2585 | return 0; |
ab967d86 | 2586 | } |
a41b70d6 | 2587 | put_page(vmf->page); |
ee6a6457 | 2588 | } |
a41b70d6 | 2589 | if (reuse_swap_page(vmf->page, &total_mapcount)) { |
6d0a07ed AA |
2590 | if (total_mapcount == 1) { |
2591 | /* | |
2592 | * The page is all ours. Move it to | |
2593 | * our anon_vma so the rmap code will | |
2594 | * not search our parent or siblings. | |
2595 | * Protected against the rmap code by | |
2596 | * the page lock. | |
2597 | */ | |
a41b70d6 | 2598 | page_move_anon_rmap(vmf->page, vma); |
6d0a07ed | 2599 | } |
a41b70d6 | 2600 | unlock_page(vmf->page); |
997dd98d JK |
2601 | wp_page_reuse(vmf); |
2602 | return VM_FAULT_WRITE; | |
b009c024 | 2603 | } |
a41b70d6 | 2604 | unlock_page(vmf->page); |
ee6a6457 | 2605 | } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) == |
d08b3851 | 2606 | (VM_WRITE|VM_SHARED))) { |
a41b70d6 | 2607 | return wp_page_shared(vmf); |
1da177e4 | 2608 | } |
1da177e4 LT |
2609 | |
2610 | /* | |
2611 | * Ok, we need to copy. Oh, well.. | |
2612 | */ | |
a41b70d6 | 2613 | get_page(vmf->page); |
28766805 | 2614 | |
82b0f8c3 | 2615 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 2616 | return wp_page_copy(vmf); |
1da177e4 LT |
2617 | } |
2618 | ||
97a89413 | 2619 | static void unmap_mapping_range_vma(struct vm_area_struct *vma, |
1da177e4 LT |
2620 | unsigned long start_addr, unsigned long end_addr, |
2621 | struct zap_details *details) | |
2622 | { | |
f5cc4eef | 2623 | zap_page_range_single(vma, start_addr, end_addr - start_addr, details); |
1da177e4 LT |
2624 | } |
2625 | ||
6b2dbba8 | 2626 | static inline void unmap_mapping_range_tree(struct rb_root *root, |
1da177e4 LT |
2627 | struct zap_details *details) |
2628 | { | |
2629 | struct vm_area_struct *vma; | |
1da177e4 LT |
2630 | pgoff_t vba, vea, zba, zea; |
2631 | ||
6b2dbba8 | 2632 | vma_interval_tree_foreach(vma, root, |
1da177e4 | 2633 | details->first_index, details->last_index) { |
1da177e4 LT |
2634 | |
2635 | vba = vma->vm_pgoff; | |
d6e93217 | 2636 | vea = vba + vma_pages(vma) - 1; |
1da177e4 LT |
2637 | zba = details->first_index; |
2638 | if (zba < vba) | |
2639 | zba = vba; | |
2640 | zea = details->last_index; | |
2641 | if (zea > vea) | |
2642 | zea = vea; | |
2643 | ||
97a89413 | 2644 | unmap_mapping_range_vma(vma, |
1da177e4 LT |
2645 | ((zba - vba) << PAGE_SHIFT) + vma->vm_start, |
2646 | ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, | |
97a89413 | 2647 | details); |
1da177e4 LT |
2648 | } |
2649 | } | |
2650 | ||
1da177e4 | 2651 | /** |
8a5f14a2 KS |
2652 | * unmap_mapping_range - unmap the portion of all mmaps in the specified |
2653 | * address_space corresponding to the specified page range in the underlying | |
2654 | * file. | |
2655 | * | |
3d41088f | 2656 | * @mapping: the address space containing mmaps to be unmapped. |
1da177e4 LT |
2657 | * @holebegin: byte in first page to unmap, relative to the start of |
2658 | * the underlying file. This will be rounded down to a PAGE_SIZE | |
25d9e2d1 | 2659 | * boundary. Note that this is different from truncate_pagecache(), which |
1da177e4 LT |
2660 | * must keep the partial page. In contrast, we must get rid of |
2661 | * partial pages. | |
2662 | * @holelen: size of prospective hole in bytes. This will be rounded | |
2663 | * up to a PAGE_SIZE boundary. A holelen of zero truncates to the | |
2664 | * end of the file. | |
2665 | * @even_cows: 1 when truncating a file, unmap even private COWed pages; | |
2666 | * but 0 when invalidating pagecache, don't throw away private data. | |
2667 | */ | |
2668 | void unmap_mapping_range(struct address_space *mapping, | |
2669 | loff_t const holebegin, loff_t const holelen, int even_cows) | |
2670 | { | |
aac45363 | 2671 | struct zap_details details = { }; |
1da177e4 LT |
2672 | pgoff_t hba = holebegin >> PAGE_SHIFT; |
2673 | pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
2674 | ||
2675 | /* Check for overflow. */ | |
2676 | if (sizeof(holelen) > sizeof(hlen)) { | |
2677 | long long holeend = | |
2678 | (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
2679 | if (holeend & ~(long long)ULONG_MAX) | |
2680 | hlen = ULONG_MAX - hba + 1; | |
2681 | } | |
2682 | ||
166f61b9 | 2683 | details.check_mapping = even_cows ? NULL : mapping; |
1da177e4 LT |
2684 | details.first_index = hba; |
2685 | details.last_index = hba + hlen - 1; | |
2686 | if (details.last_index < details.first_index) | |
2687 | details.last_index = ULONG_MAX; | |
1da177e4 | 2688 | |
46c043ed | 2689 | i_mmap_lock_write(mapping); |
6b2dbba8 | 2690 | if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap))) |
1da177e4 | 2691 | unmap_mapping_range_tree(&mapping->i_mmap, &details); |
46c043ed | 2692 | i_mmap_unlock_write(mapping); |
1da177e4 LT |
2693 | } |
2694 | EXPORT_SYMBOL(unmap_mapping_range); | |
2695 | ||
1da177e4 | 2696 | /* |
8f4e2101 HD |
2697 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2698 | * but allow concurrent faults), and pte mapped but not yet locked. | |
9a95f3cf PC |
2699 | * We return with pte unmapped and unlocked. |
2700 | * | |
2701 | * We return with the mmap_sem locked or unlocked in the same cases | |
2702 | * as does filemap_fault(). | |
1da177e4 | 2703 | */ |
2994302b | 2704 | int do_swap_page(struct vm_fault *vmf) |
1da177e4 | 2705 | { |
82b0f8c3 | 2706 | struct vm_area_struct *vma = vmf->vma; |
56f31801 | 2707 | struct page *page, *swapcache; |
00501b53 | 2708 | struct mem_cgroup *memcg; |
65500d23 | 2709 | swp_entry_t entry; |
1da177e4 | 2710 | pte_t pte; |
d065bd81 | 2711 | int locked; |
ad8c2ee8 | 2712 | int exclusive = 0; |
83c54070 | 2713 | int ret = 0; |
1da177e4 | 2714 | |
2994302b | 2715 | if (!pte_unmap_same(vma->vm_mm, vmf->pmd, vmf->pte, vmf->orig_pte)) |
8f4e2101 | 2716 | goto out; |
65500d23 | 2717 | |
2994302b | 2718 | entry = pte_to_swp_entry(vmf->orig_pte); |
d1737fdb AK |
2719 | if (unlikely(non_swap_entry(entry))) { |
2720 | if (is_migration_entry(entry)) { | |
82b0f8c3 JK |
2721 | migration_entry_wait(vma->vm_mm, vmf->pmd, |
2722 | vmf->address); | |
d1737fdb AK |
2723 | } else if (is_hwpoison_entry(entry)) { |
2724 | ret = VM_FAULT_HWPOISON; | |
2725 | } else { | |
2994302b | 2726 | print_bad_pte(vma, vmf->address, vmf->orig_pte, NULL); |
d99be1a8 | 2727 | ret = VM_FAULT_SIGBUS; |
d1737fdb | 2728 | } |
0697212a CL |
2729 | goto out; |
2730 | } | |
0ff92245 | 2731 | delayacct_set_flag(DELAYACCT_PF_SWAPIN); |
1da177e4 LT |
2732 | page = lookup_swap_cache(entry); |
2733 | if (!page) { | |
82b0f8c3 JK |
2734 | page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE, vma, |
2735 | vmf->address); | |
1da177e4 LT |
2736 | if (!page) { |
2737 | /* | |
8f4e2101 HD |
2738 | * Back out if somebody else faulted in this pte |
2739 | * while we released the pte lock. | |
1da177e4 | 2740 | */ |
82b0f8c3 JK |
2741 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
2742 | vmf->address, &vmf->ptl); | |
2994302b | 2743 | if (likely(pte_same(*vmf->pte, vmf->orig_pte))) |
1da177e4 | 2744 | ret = VM_FAULT_OOM; |
0ff92245 | 2745 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
65500d23 | 2746 | goto unlock; |
1da177e4 LT |
2747 | } |
2748 | ||
2749 | /* Had to read the page from swap area: Major fault */ | |
2750 | ret = VM_FAULT_MAJOR; | |
f8891e5e | 2751 | count_vm_event(PGMAJFAULT); |
2262185c | 2752 | count_memcg_event_mm(vma->vm_mm, PGMAJFAULT); |
d1737fdb | 2753 | } else if (PageHWPoison(page)) { |
71f72525 WF |
2754 | /* |
2755 | * hwpoisoned dirty swapcache pages are kept for killing | |
2756 | * owner processes (which may be unknown at hwpoison time) | |
2757 | */ | |
d1737fdb AK |
2758 | ret = VM_FAULT_HWPOISON; |
2759 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); | |
56f31801 | 2760 | swapcache = page; |
4779cb31 | 2761 | goto out_release; |
1da177e4 LT |
2762 | } |
2763 | ||
56f31801 | 2764 | swapcache = page; |
82b0f8c3 | 2765 | locked = lock_page_or_retry(page, vma->vm_mm, vmf->flags); |
e709ffd6 | 2766 | |
073e587e | 2767 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
d065bd81 ML |
2768 | if (!locked) { |
2769 | ret |= VM_FAULT_RETRY; | |
2770 | goto out_release; | |
2771 | } | |
073e587e | 2772 | |
4969c119 | 2773 | /* |
31c4a3d3 HD |
2774 | * Make sure try_to_free_swap or reuse_swap_page or swapoff did not |
2775 | * release the swapcache from under us. The page pin, and pte_same | |
2776 | * test below, are not enough to exclude that. Even if it is still | |
2777 | * swapcache, we need to check that the page's swap has not changed. | |
4969c119 | 2778 | */ |
31c4a3d3 | 2779 | if (unlikely(!PageSwapCache(page) || page_private(page) != entry.val)) |
4969c119 AA |
2780 | goto out_page; |
2781 | ||
82b0f8c3 | 2782 | page = ksm_might_need_to_copy(page, vma, vmf->address); |
cbf86cfe HD |
2783 | if (unlikely(!page)) { |
2784 | ret = VM_FAULT_OOM; | |
2785 | page = swapcache; | |
cbf86cfe | 2786 | goto out_page; |
5ad64688 HD |
2787 | } |
2788 | ||
bae473a4 KS |
2789 | if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL, |
2790 | &memcg, false)) { | |
8a9f3ccd | 2791 | ret = VM_FAULT_OOM; |
bc43f75c | 2792 | goto out_page; |
8a9f3ccd BS |
2793 | } |
2794 | ||
1da177e4 | 2795 | /* |
8f4e2101 | 2796 | * Back out if somebody else already faulted in this pte. |
1da177e4 | 2797 | */ |
82b0f8c3 JK |
2798 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, |
2799 | &vmf->ptl); | |
2994302b | 2800 | if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) |
b8107480 | 2801 | goto out_nomap; |
b8107480 KK |
2802 | |
2803 | if (unlikely(!PageUptodate(page))) { | |
2804 | ret = VM_FAULT_SIGBUS; | |
2805 | goto out_nomap; | |
1da177e4 LT |
2806 | } |
2807 | ||
8c7c6e34 KH |
2808 | /* |
2809 | * The page isn't present yet, go ahead with the fault. | |
2810 | * | |
2811 | * Be careful about the sequence of operations here. | |
2812 | * To get its accounting right, reuse_swap_page() must be called | |
2813 | * while the page is counted on swap but not yet in mapcount i.e. | |
2814 | * before page_add_anon_rmap() and swap_free(); try_to_free_swap() | |
2815 | * must be called after the swap_free(), or it will never succeed. | |
8c7c6e34 | 2816 | */ |
1da177e4 | 2817 | |
bae473a4 KS |
2818 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
2819 | dec_mm_counter_fast(vma->vm_mm, MM_SWAPENTS); | |
1da177e4 | 2820 | pte = mk_pte(page, vma->vm_page_prot); |
82b0f8c3 | 2821 | if ((vmf->flags & FAULT_FLAG_WRITE) && reuse_swap_page(page, NULL)) { |
1da177e4 | 2822 | pte = maybe_mkwrite(pte_mkdirty(pte), vma); |
82b0f8c3 | 2823 | vmf->flags &= ~FAULT_FLAG_WRITE; |
9a5b489b | 2824 | ret |= VM_FAULT_WRITE; |
d281ee61 | 2825 | exclusive = RMAP_EXCLUSIVE; |
1da177e4 | 2826 | } |
1da177e4 | 2827 | flush_icache_page(vma, page); |
2994302b | 2828 | if (pte_swp_soft_dirty(vmf->orig_pte)) |
179ef71c | 2829 | pte = pte_mksoft_dirty(pte); |
82b0f8c3 | 2830 | set_pte_at(vma->vm_mm, vmf->address, vmf->pte, pte); |
2994302b | 2831 | vmf->orig_pte = pte; |
00501b53 | 2832 | if (page == swapcache) { |
82b0f8c3 | 2833 | do_page_add_anon_rmap(page, vma, vmf->address, exclusive); |
f627c2f5 | 2834 | mem_cgroup_commit_charge(page, memcg, true, false); |
1a8018fb | 2835 | activate_page(page); |
00501b53 | 2836 | } else { /* ksm created a completely new copy */ |
82b0f8c3 | 2837 | page_add_new_anon_rmap(page, vma, vmf->address, false); |
f627c2f5 | 2838 | mem_cgroup_commit_charge(page, memcg, false, false); |
00501b53 JW |
2839 | lru_cache_add_active_or_unevictable(page, vma); |
2840 | } | |
1da177e4 | 2841 | |
c475a8ab | 2842 | swap_free(entry); |
5ccc5aba VD |
2843 | if (mem_cgroup_swap_full(page) || |
2844 | (vma->vm_flags & VM_LOCKED) || PageMlocked(page)) | |
a2c43eed | 2845 | try_to_free_swap(page); |
c475a8ab | 2846 | unlock_page(page); |
56f31801 | 2847 | if (page != swapcache) { |
4969c119 AA |
2848 | /* |
2849 | * Hold the lock to avoid the swap entry to be reused | |
2850 | * until we take the PT lock for the pte_same() check | |
2851 | * (to avoid false positives from pte_same). For | |
2852 | * further safety release the lock after the swap_free | |
2853 | * so that the swap count won't change under a | |
2854 | * parallel locked swapcache. | |
2855 | */ | |
2856 | unlock_page(swapcache); | |
09cbfeaf | 2857 | put_page(swapcache); |
4969c119 | 2858 | } |
c475a8ab | 2859 | |
82b0f8c3 | 2860 | if (vmf->flags & FAULT_FLAG_WRITE) { |
2994302b | 2861 | ret |= do_wp_page(vmf); |
61469f1d HD |
2862 | if (ret & VM_FAULT_ERROR) |
2863 | ret &= VM_FAULT_ERROR; | |
1da177e4 LT |
2864 | goto out; |
2865 | } | |
2866 | ||
2867 | /* No need to invalidate - it was non-present before */ | |
82b0f8c3 | 2868 | update_mmu_cache(vma, vmf->address, vmf->pte); |
65500d23 | 2869 | unlock: |
82b0f8c3 | 2870 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
1da177e4 LT |
2871 | out: |
2872 | return ret; | |
b8107480 | 2873 | out_nomap: |
f627c2f5 | 2874 | mem_cgroup_cancel_charge(page, memcg, false); |
82b0f8c3 | 2875 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
bc43f75c | 2876 | out_page: |
b8107480 | 2877 | unlock_page(page); |
4779cb31 | 2878 | out_release: |
09cbfeaf | 2879 | put_page(page); |
56f31801 | 2880 | if (page != swapcache) { |
4969c119 | 2881 | unlock_page(swapcache); |
09cbfeaf | 2882 | put_page(swapcache); |
4969c119 | 2883 | } |
65500d23 | 2884 | return ret; |
1da177e4 LT |
2885 | } |
2886 | ||
2887 | /* | |
8f4e2101 HD |
2888 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2889 | * but allow concurrent faults), and pte mapped but not yet locked. | |
2890 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2891 | */ |
82b0f8c3 | 2892 | static int do_anonymous_page(struct vm_fault *vmf) |
1da177e4 | 2893 | { |
82b0f8c3 | 2894 | struct vm_area_struct *vma = vmf->vma; |
00501b53 | 2895 | struct mem_cgroup *memcg; |
8f4e2101 | 2896 | struct page *page; |
6b31d595 | 2897 | int ret = 0; |
1da177e4 | 2898 | pte_t entry; |
1da177e4 | 2899 | |
6b7339f4 KS |
2900 | /* File mapping without ->vm_ops ? */ |
2901 | if (vma->vm_flags & VM_SHARED) | |
2902 | return VM_FAULT_SIGBUS; | |
2903 | ||
7267ec00 KS |
2904 | /* |
2905 | * Use pte_alloc() instead of pte_alloc_map(). We can't run | |
2906 | * pte_offset_map() on pmds where a huge pmd might be created | |
2907 | * from a different thread. | |
2908 | * | |
2909 | * pte_alloc_map() is safe to use under down_write(mmap_sem) or when | |
2910 | * parallel threads are excluded by other means. | |
2911 | * | |
2912 | * Here we only have down_read(mmap_sem). | |
2913 | */ | |
82b0f8c3 | 2914 | if (pte_alloc(vma->vm_mm, vmf->pmd, vmf->address)) |
7267ec00 KS |
2915 | return VM_FAULT_OOM; |
2916 | ||
2917 | /* See the comment in pte_alloc_one_map() */ | |
82b0f8c3 | 2918 | if (unlikely(pmd_trans_unstable(vmf->pmd))) |
7267ec00 KS |
2919 | return 0; |
2920 | ||
11ac5524 | 2921 | /* Use the zero-page for reads */ |
82b0f8c3 | 2922 | if (!(vmf->flags & FAULT_FLAG_WRITE) && |
bae473a4 | 2923 | !mm_forbids_zeropage(vma->vm_mm)) { |
82b0f8c3 | 2924 | entry = pte_mkspecial(pfn_pte(my_zero_pfn(vmf->address), |
62eede62 | 2925 | vma->vm_page_prot)); |
82b0f8c3 JK |
2926 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
2927 | vmf->address, &vmf->ptl); | |
2928 | if (!pte_none(*vmf->pte)) | |
a13ea5b7 | 2929 | goto unlock; |
6b31d595 MH |
2930 | ret = check_stable_address_space(vma->vm_mm); |
2931 | if (ret) | |
2932 | goto unlock; | |
6b251fc9 AA |
2933 | /* Deliver the page fault to userland, check inside PT lock */ |
2934 | if (userfaultfd_missing(vma)) { | |
82b0f8c3 JK |
2935 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
2936 | return handle_userfault(vmf, VM_UFFD_MISSING); | |
6b251fc9 | 2937 | } |
a13ea5b7 HD |
2938 | goto setpte; |
2939 | } | |
2940 | ||
557ed1fa | 2941 | /* Allocate our own private page. */ |
557ed1fa NP |
2942 | if (unlikely(anon_vma_prepare(vma))) |
2943 | goto oom; | |
82b0f8c3 | 2944 | page = alloc_zeroed_user_highpage_movable(vma, vmf->address); |
557ed1fa NP |
2945 | if (!page) |
2946 | goto oom; | |
eb3c24f3 | 2947 | |
bae473a4 | 2948 | if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL, &memcg, false)) |
eb3c24f3 MG |
2949 | goto oom_free_page; |
2950 | ||
52f37629 MK |
2951 | /* |
2952 | * The memory barrier inside __SetPageUptodate makes sure that | |
2953 | * preceeding stores to the page contents become visible before | |
2954 | * the set_pte_at() write. | |
2955 | */ | |
0ed361de | 2956 | __SetPageUptodate(page); |
8f4e2101 | 2957 | |
557ed1fa | 2958 | entry = mk_pte(page, vma->vm_page_prot); |
1ac0cb5d HD |
2959 | if (vma->vm_flags & VM_WRITE) |
2960 | entry = pte_mkwrite(pte_mkdirty(entry)); | |
1da177e4 | 2961 | |
82b0f8c3 JK |
2962 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, |
2963 | &vmf->ptl); | |
2964 | if (!pte_none(*vmf->pte)) | |
557ed1fa | 2965 | goto release; |
9ba69294 | 2966 | |
6b31d595 MH |
2967 | ret = check_stable_address_space(vma->vm_mm); |
2968 | if (ret) | |
2969 | goto release; | |
2970 | ||
6b251fc9 AA |
2971 | /* Deliver the page fault to userland, check inside PT lock */ |
2972 | if (userfaultfd_missing(vma)) { | |
82b0f8c3 | 2973 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
f627c2f5 | 2974 | mem_cgroup_cancel_charge(page, memcg, false); |
09cbfeaf | 2975 | put_page(page); |
82b0f8c3 | 2976 | return handle_userfault(vmf, VM_UFFD_MISSING); |
6b251fc9 AA |
2977 | } |
2978 | ||
bae473a4 | 2979 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
82b0f8c3 | 2980 | page_add_new_anon_rmap(page, vma, vmf->address, false); |
f627c2f5 | 2981 | mem_cgroup_commit_charge(page, memcg, false, false); |
00501b53 | 2982 | lru_cache_add_active_or_unevictable(page, vma); |
a13ea5b7 | 2983 | setpte: |
82b0f8c3 | 2984 | set_pte_at(vma->vm_mm, vmf->address, vmf->pte, entry); |
1da177e4 LT |
2985 | |
2986 | /* No need to invalidate - it was non-present before */ | |
82b0f8c3 | 2987 | update_mmu_cache(vma, vmf->address, vmf->pte); |
65500d23 | 2988 | unlock: |
82b0f8c3 | 2989 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
6b31d595 | 2990 | return ret; |
8f4e2101 | 2991 | release: |
f627c2f5 | 2992 | mem_cgroup_cancel_charge(page, memcg, false); |
09cbfeaf | 2993 | put_page(page); |
8f4e2101 | 2994 | goto unlock; |
8a9f3ccd | 2995 | oom_free_page: |
09cbfeaf | 2996 | put_page(page); |
65500d23 | 2997 | oom: |
1da177e4 LT |
2998 | return VM_FAULT_OOM; |
2999 | } | |
3000 | ||
9a95f3cf PC |
3001 | /* |
3002 | * The mmap_sem must have been held on entry, and may have been | |
3003 | * released depending on flags and vma->vm_ops->fault() return value. | |
3004 | * See filemap_fault() and __lock_page_retry(). | |
3005 | */ | |
936ca80d | 3006 | static int __do_fault(struct vm_fault *vmf) |
7eae74af | 3007 | { |
82b0f8c3 | 3008 | struct vm_area_struct *vma = vmf->vma; |
7eae74af KS |
3009 | int ret; |
3010 | ||
11bac800 | 3011 | ret = vma->vm_ops->fault(vmf); |
3917048d | 3012 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY | |
b1aa812b | 3013 | VM_FAULT_DONE_COW))) |
bc2466e4 | 3014 | return ret; |
7eae74af | 3015 | |
667240e0 | 3016 | if (unlikely(PageHWPoison(vmf->page))) { |
7eae74af | 3017 | if (ret & VM_FAULT_LOCKED) |
667240e0 JK |
3018 | unlock_page(vmf->page); |
3019 | put_page(vmf->page); | |
936ca80d | 3020 | vmf->page = NULL; |
7eae74af KS |
3021 | return VM_FAULT_HWPOISON; |
3022 | } | |
3023 | ||
3024 | if (unlikely(!(ret & VM_FAULT_LOCKED))) | |
667240e0 | 3025 | lock_page(vmf->page); |
7eae74af | 3026 | else |
667240e0 | 3027 | VM_BUG_ON_PAGE(!PageLocked(vmf->page), vmf->page); |
7eae74af | 3028 | |
7eae74af KS |
3029 | return ret; |
3030 | } | |
3031 | ||
d0f0931d RZ |
3032 | /* |
3033 | * The ordering of these checks is important for pmds with _PAGE_DEVMAP set. | |
3034 | * If we check pmd_trans_unstable() first we will trip the bad_pmd() check | |
3035 | * inside of pmd_none_or_trans_huge_or_clear_bad(). This will end up correctly | |
3036 | * returning 1 but not before it spams dmesg with the pmd_clear_bad() output. | |
3037 | */ | |
3038 | static int pmd_devmap_trans_unstable(pmd_t *pmd) | |
3039 | { | |
3040 | return pmd_devmap(*pmd) || pmd_trans_unstable(pmd); | |
3041 | } | |
3042 | ||
82b0f8c3 | 3043 | static int pte_alloc_one_map(struct vm_fault *vmf) |
7267ec00 | 3044 | { |
82b0f8c3 | 3045 | struct vm_area_struct *vma = vmf->vma; |
7267ec00 | 3046 | |
82b0f8c3 | 3047 | if (!pmd_none(*vmf->pmd)) |
7267ec00 | 3048 | goto map_pte; |
82b0f8c3 JK |
3049 | if (vmf->prealloc_pte) { |
3050 | vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd); | |
3051 | if (unlikely(!pmd_none(*vmf->pmd))) { | |
3052 | spin_unlock(vmf->ptl); | |
7267ec00 KS |
3053 | goto map_pte; |
3054 | } | |
3055 | ||
3056 | atomic_long_inc(&vma->vm_mm->nr_ptes); | |
82b0f8c3 JK |
3057 | pmd_populate(vma->vm_mm, vmf->pmd, vmf->prealloc_pte); |
3058 | spin_unlock(vmf->ptl); | |
7f2b6ce8 | 3059 | vmf->prealloc_pte = NULL; |
82b0f8c3 | 3060 | } else if (unlikely(pte_alloc(vma->vm_mm, vmf->pmd, vmf->address))) { |
7267ec00 KS |
3061 | return VM_FAULT_OOM; |
3062 | } | |
3063 | map_pte: | |
3064 | /* | |
3065 | * If a huge pmd materialized under us just retry later. Use | |
d0f0931d RZ |
3066 | * pmd_trans_unstable() via pmd_devmap_trans_unstable() instead of |
3067 | * pmd_trans_huge() to ensure the pmd didn't become pmd_trans_huge | |
3068 | * under us and then back to pmd_none, as a result of MADV_DONTNEED | |
3069 | * running immediately after a huge pmd fault in a different thread of | |
3070 | * this mm, in turn leading to a misleading pmd_trans_huge() retval. | |
3071 | * All we have to ensure is that it is a regular pmd that we can walk | |
3072 | * with pte_offset_map() and we can do that through an atomic read in | |
3073 | * C, which is what pmd_trans_unstable() provides. | |
7267ec00 | 3074 | */ |
d0f0931d | 3075 | if (pmd_devmap_trans_unstable(vmf->pmd)) |
7267ec00 KS |
3076 | return VM_FAULT_NOPAGE; |
3077 | ||
d0f0931d RZ |
3078 | /* |
3079 | * At this point we know that our vmf->pmd points to a page of ptes | |
3080 | * and it cannot become pmd_none(), pmd_devmap() or pmd_trans_huge() | |
3081 | * for the duration of the fault. If a racing MADV_DONTNEED runs and | |
3082 | * we zap the ptes pointed to by our vmf->pmd, the vmf->ptl will still | |
3083 | * be valid and we will re-check to make sure the vmf->pte isn't | |
3084 | * pte_none() under vmf->ptl protection when we return to | |
3085 | * alloc_set_pte(). | |
3086 | */ | |
82b0f8c3 JK |
3087 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, |
3088 | &vmf->ptl); | |
7267ec00 KS |
3089 | return 0; |
3090 | } | |
3091 | ||
e496cf3d | 3092 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
10102459 KS |
3093 | |
3094 | #define HPAGE_CACHE_INDEX_MASK (HPAGE_PMD_NR - 1) | |
3095 | static inline bool transhuge_vma_suitable(struct vm_area_struct *vma, | |
3096 | unsigned long haddr) | |
3097 | { | |
3098 | if (((vma->vm_start >> PAGE_SHIFT) & HPAGE_CACHE_INDEX_MASK) != | |
3099 | (vma->vm_pgoff & HPAGE_CACHE_INDEX_MASK)) | |
3100 | return false; | |
3101 | if (haddr < vma->vm_start || haddr + HPAGE_PMD_SIZE > vma->vm_end) | |
3102 | return false; | |
3103 | return true; | |
3104 | } | |
3105 | ||
82b0f8c3 | 3106 | static void deposit_prealloc_pte(struct vm_fault *vmf) |
953c66c2 | 3107 | { |
82b0f8c3 | 3108 | struct vm_area_struct *vma = vmf->vma; |
953c66c2 | 3109 | |
82b0f8c3 | 3110 | pgtable_trans_huge_deposit(vma->vm_mm, vmf->pmd, vmf->prealloc_pte); |
953c66c2 AK |
3111 | /* |
3112 | * We are going to consume the prealloc table, | |
3113 | * count that as nr_ptes. | |
3114 | */ | |
3115 | atomic_long_inc(&vma->vm_mm->nr_ptes); | |
7f2b6ce8 | 3116 | vmf->prealloc_pte = NULL; |
953c66c2 AK |
3117 | } |
3118 | ||
82b0f8c3 | 3119 | static int do_set_pmd(struct vm_fault *vmf, struct page *page) |
10102459 | 3120 | { |
82b0f8c3 JK |
3121 | struct vm_area_struct *vma = vmf->vma; |
3122 | bool write = vmf->flags & FAULT_FLAG_WRITE; | |
3123 | unsigned long haddr = vmf->address & HPAGE_PMD_MASK; | |
10102459 KS |
3124 | pmd_t entry; |
3125 | int i, ret; | |
3126 | ||
3127 | if (!transhuge_vma_suitable(vma, haddr)) | |
3128 | return VM_FAULT_FALLBACK; | |
3129 | ||
3130 | ret = VM_FAULT_FALLBACK; | |
3131 | page = compound_head(page); | |
3132 | ||
953c66c2 AK |
3133 | /* |
3134 | * Archs like ppc64 need additonal space to store information | |
3135 | * related to pte entry. Use the preallocated table for that. | |
3136 | */ | |
82b0f8c3 JK |
3137 | if (arch_needs_pgtable_deposit() && !vmf->prealloc_pte) { |
3138 | vmf->prealloc_pte = pte_alloc_one(vma->vm_mm, vmf->address); | |
3139 | if (!vmf->prealloc_pte) | |
953c66c2 AK |
3140 | return VM_FAULT_OOM; |
3141 | smp_wmb(); /* See comment in __pte_alloc() */ | |
3142 | } | |
3143 | ||
82b0f8c3 JK |
3144 | vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd); |
3145 | if (unlikely(!pmd_none(*vmf->pmd))) | |
10102459 KS |
3146 | goto out; |
3147 | ||
3148 | for (i = 0; i < HPAGE_PMD_NR; i++) | |
3149 | flush_icache_page(vma, page + i); | |
3150 | ||
3151 | entry = mk_huge_pmd(page, vma->vm_page_prot); | |
3152 | if (write) | |
3153 | entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma); | |
3154 | ||
3155 | add_mm_counter(vma->vm_mm, MM_FILEPAGES, HPAGE_PMD_NR); | |
3156 | page_add_file_rmap(page, true); | |
953c66c2 AK |
3157 | /* |
3158 | * deposit and withdraw with pmd lock held | |
3159 | */ | |
3160 | if (arch_needs_pgtable_deposit()) | |
82b0f8c3 | 3161 | deposit_prealloc_pte(vmf); |
10102459 | 3162 | |
82b0f8c3 | 3163 | set_pmd_at(vma->vm_mm, haddr, vmf->pmd, entry); |
10102459 | 3164 | |
82b0f8c3 | 3165 | update_mmu_cache_pmd(vma, haddr, vmf->pmd); |
10102459 KS |
3166 | |
3167 | /* fault is handled */ | |
3168 | ret = 0; | |
95ecedcd | 3169 | count_vm_event(THP_FILE_MAPPED); |
10102459 | 3170 | out: |
82b0f8c3 | 3171 | spin_unlock(vmf->ptl); |
10102459 KS |
3172 | return ret; |
3173 | } | |
3174 | #else | |
82b0f8c3 | 3175 | static int do_set_pmd(struct vm_fault *vmf, struct page *page) |
10102459 KS |
3176 | { |
3177 | BUILD_BUG(); | |
3178 | return 0; | |
3179 | } | |
3180 | #endif | |
3181 | ||
8c6e50b0 | 3182 | /** |
7267ec00 KS |
3183 | * alloc_set_pte - setup new PTE entry for given page and add reverse page |
3184 | * mapping. If needed, the fucntion allocates page table or use pre-allocated. | |
8c6e50b0 | 3185 | * |
82b0f8c3 | 3186 | * @vmf: fault environment |
7267ec00 | 3187 | * @memcg: memcg to charge page (only for private mappings) |
8c6e50b0 | 3188 | * @page: page to map |
8c6e50b0 | 3189 | * |
82b0f8c3 JK |
3190 | * Caller must take care of unlocking vmf->ptl, if vmf->pte is non-NULL on |
3191 | * return. | |
8c6e50b0 KS |
3192 | * |
3193 | * Target users are page handler itself and implementations of | |
3194 | * vm_ops->map_pages. | |
3195 | */ | |
82b0f8c3 | 3196 | int alloc_set_pte(struct vm_fault *vmf, struct mem_cgroup *memcg, |
7267ec00 | 3197 | struct page *page) |
3bb97794 | 3198 | { |
82b0f8c3 JK |
3199 | struct vm_area_struct *vma = vmf->vma; |
3200 | bool write = vmf->flags & FAULT_FLAG_WRITE; | |
3bb97794 | 3201 | pte_t entry; |
10102459 KS |
3202 | int ret; |
3203 | ||
82b0f8c3 | 3204 | if (pmd_none(*vmf->pmd) && PageTransCompound(page) && |
e496cf3d | 3205 | IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) { |
10102459 KS |
3206 | /* THP on COW? */ |
3207 | VM_BUG_ON_PAGE(memcg, page); | |
3208 | ||
82b0f8c3 | 3209 | ret = do_set_pmd(vmf, page); |
10102459 | 3210 | if (ret != VM_FAULT_FALLBACK) |
b0b9b3df | 3211 | return ret; |
10102459 | 3212 | } |
3bb97794 | 3213 | |
82b0f8c3 JK |
3214 | if (!vmf->pte) { |
3215 | ret = pte_alloc_one_map(vmf); | |
7267ec00 | 3216 | if (ret) |
b0b9b3df | 3217 | return ret; |
7267ec00 KS |
3218 | } |
3219 | ||
3220 | /* Re-check under ptl */ | |
b0b9b3df HD |
3221 | if (unlikely(!pte_none(*vmf->pte))) |
3222 | return VM_FAULT_NOPAGE; | |
7267ec00 | 3223 | |
3bb97794 KS |
3224 | flush_icache_page(vma, page); |
3225 | entry = mk_pte(page, vma->vm_page_prot); | |
3226 | if (write) | |
3227 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
bae473a4 KS |
3228 | /* copy-on-write page */ |
3229 | if (write && !(vma->vm_flags & VM_SHARED)) { | |
3bb97794 | 3230 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
82b0f8c3 | 3231 | page_add_new_anon_rmap(page, vma, vmf->address, false); |
7267ec00 KS |
3232 | mem_cgroup_commit_charge(page, memcg, false, false); |
3233 | lru_cache_add_active_or_unevictable(page, vma); | |
3bb97794 | 3234 | } else { |
eca56ff9 | 3235 | inc_mm_counter_fast(vma->vm_mm, mm_counter_file(page)); |
dd78fedd | 3236 | page_add_file_rmap(page, false); |
3bb97794 | 3237 | } |
82b0f8c3 | 3238 | set_pte_at(vma->vm_mm, vmf->address, vmf->pte, entry); |
3bb97794 KS |
3239 | |
3240 | /* no need to invalidate: a not-present page won't be cached */ | |
82b0f8c3 | 3241 | update_mmu_cache(vma, vmf->address, vmf->pte); |
7267ec00 | 3242 | |
b0b9b3df | 3243 | return 0; |
3bb97794 KS |
3244 | } |
3245 | ||
9118c0cb JK |
3246 | |
3247 | /** | |
3248 | * finish_fault - finish page fault once we have prepared the page to fault | |
3249 | * | |
3250 | * @vmf: structure describing the fault | |
3251 | * | |
3252 | * This function handles all that is needed to finish a page fault once the | |
3253 | * page to fault in is prepared. It handles locking of PTEs, inserts PTE for | |
3254 | * given page, adds reverse page mapping, handles memcg charges and LRU | |
3255 | * addition. The function returns 0 on success, VM_FAULT_ code in case of | |
3256 | * error. | |
3257 | * | |
3258 | * The function expects the page to be locked and on success it consumes a | |
3259 | * reference of a page being mapped (for the PTE which maps it). | |
3260 | */ | |
3261 | int finish_fault(struct vm_fault *vmf) | |
3262 | { | |
3263 | struct page *page; | |
6b31d595 | 3264 | int ret = 0; |
9118c0cb JK |
3265 | |
3266 | /* Did we COW the page? */ | |
3267 | if ((vmf->flags & FAULT_FLAG_WRITE) && | |
3268 | !(vmf->vma->vm_flags & VM_SHARED)) | |
3269 | page = vmf->cow_page; | |
3270 | else | |
3271 | page = vmf->page; | |
6b31d595 MH |
3272 | |
3273 | /* | |
3274 | * check even for read faults because we might have lost our CoWed | |
3275 | * page | |
3276 | */ | |
3277 | if (!(vmf->vma->vm_flags & VM_SHARED)) | |
3278 | ret = check_stable_address_space(vmf->vma->vm_mm); | |
3279 | if (!ret) | |
3280 | ret = alloc_set_pte(vmf, vmf->memcg, page); | |
9118c0cb JK |
3281 | if (vmf->pte) |
3282 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
3283 | return ret; | |
3284 | } | |
3285 | ||
3a91053a KS |
3286 | static unsigned long fault_around_bytes __read_mostly = |
3287 | rounddown_pow_of_two(65536); | |
a9b0f861 | 3288 | |
a9b0f861 KS |
3289 | #ifdef CONFIG_DEBUG_FS |
3290 | static int fault_around_bytes_get(void *data, u64 *val) | |
1592eef0 | 3291 | { |
a9b0f861 | 3292 | *val = fault_around_bytes; |
1592eef0 KS |
3293 | return 0; |
3294 | } | |
3295 | ||
b4903d6e AR |
3296 | /* |
3297 | * fault_around_pages() and fault_around_mask() expects fault_around_bytes | |
3298 | * rounded down to nearest page order. It's what do_fault_around() expects to | |
3299 | * see. | |
3300 | */ | |
a9b0f861 | 3301 | static int fault_around_bytes_set(void *data, u64 val) |
1592eef0 | 3302 | { |
a9b0f861 | 3303 | if (val / PAGE_SIZE > PTRS_PER_PTE) |
1592eef0 | 3304 | return -EINVAL; |
b4903d6e AR |
3305 | if (val > PAGE_SIZE) |
3306 | fault_around_bytes = rounddown_pow_of_two(val); | |
3307 | else | |
3308 | fault_around_bytes = PAGE_SIZE; /* rounddown_pow_of_two(0) is undefined */ | |
1592eef0 KS |
3309 | return 0; |
3310 | } | |
0a1345f8 | 3311 | DEFINE_DEBUGFS_ATTRIBUTE(fault_around_bytes_fops, |
a9b0f861 | 3312 | fault_around_bytes_get, fault_around_bytes_set, "%llu\n"); |
1592eef0 KS |
3313 | |
3314 | static int __init fault_around_debugfs(void) | |
3315 | { | |
3316 | void *ret; | |
3317 | ||
0a1345f8 | 3318 | ret = debugfs_create_file_unsafe("fault_around_bytes", 0644, NULL, NULL, |
a9b0f861 | 3319 | &fault_around_bytes_fops); |
1592eef0 | 3320 | if (!ret) |
a9b0f861 | 3321 | pr_warn("Failed to create fault_around_bytes in debugfs"); |
1592eef0 KS |
3322 | return 0; |
3323 | } | |
3324 | late_initcall(fault_around_debugfs); | |
1592eef0 | 3325 | #endif |
8c6e50b0 | 3326 | |
1fdb412b KS |
3327 | /* |
3328 | * do_fault_around() tries to map few pages around the fault address. The hope | |
3329 | * is that the pages will be needed soon and this will lower the number of | |
3330 | * faults to handle. | |
3331 | * | |
3332 | * It uses vm_ops->map_pages() to map the pages, which skips the page if it's | |
3333 | * not ready to be mapped: not up-to-date, locked, etc. | |
3334 | * | |
3335 | * This function is called with the page table lock taken. In the split ptlock | |
3336 | * case the page table lock only protects only those entries which belong to | |
3337 | * the page table corresponding to the fault address. | |
3338 | * | |
3339 | * This function doesn't cross the VMA boundaries, in order to call map_pages() | |
3340 | * only once. | |
3341 | * | |
3342 | * fault_around_pages() defines how many pages we'll try to map. | |
3343 | * do_fault_around() expects it to return a power of two less than or equal to | |
3344 | * PTRS_PER_PTE. | |
3345 | * | |
3346 | * The virtual address of the area that we map is naturally aligned to the | |
3347 | * fault_around_pages() value (and therefore to page order). This way it's | |
3348 | * easier to guarantee that we don't cross page table boundaries. | |
3349 | */ | |
0721ec8b | 3350 | static int do_fault_around(struct vm_fault *vmf) |
8c6e50b0 | 3351 | { |
82b0f8c3 | 3352 | unsigned long address = vmf->address, nr_pages, mask; |
0721ec8b | 3353 | pgoff_t start_pgoff = vmf->pgoff; |
bae473a4 | 3354 | pgoff_t end_pgoff; |
7267ec00 | 3355 | int off, ret = 0; |
8c6e50b0 | 3356 | |
4db0c3c2 | 3357 | nr_pages = READ_ONCE(fault_around_bytes) >> PAGE_SHIFT; |
aecd6f44 KS |
3358 | mask = ~(nr_pages * PAGE_SIZE - 1) & PAGE_MASK; |
3359 | ||
82b0f8c3 JK |
3360 | vmf->address = max(address & mask, vmf->vma->vm_start); |
3361 | off = ((address - vmf->address) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1); | |
bae473a4 | 3362 | start_pgoff -= off; |
8c6e50b0 KS |
3363 | |
3364 | /* | |
bae473a4 KS |
3365 | * end_pgoff is either end of page table or end of vma |
3366 | * or fault_around_pages() from start_pgoff, depending what is nearest. | |
8c6e50b0 | 3367 | */ |
bae473a4 | 3368 | end_pgoff = start_pgoff - |
82b0f8c3 | 3369 | ((vmf->address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)) + |
8c6e50b0 | 3370 | PTRS_PER_PTE - 1; |
82b0f8c3 | 3371 | end_pgoff = min3(end_pgoff, vma_pages(vmf->vma) + vmf->vma->vm_pgoff - 1, |
bae473a4 | 3372 | start_pgoff + nr_pages - 1); |
8c6e50b0 | 3373 | |
82b0f8c3 JK |
3374 | if (pmd_none(*vmf->pmd)) { |
3375 | vmf->prealloc_pte = pte_alloc_one(vmf->vma->vm_mm, | |
3376 | vmf->address); | |
3377 | if (!vmf->prealloc_pte) | |
c5f88bd2 | 3378 | goto out; |
7267ec00 | 3379 | smp_wmb(); /* See comment in __pte_alloc() */ |
8c6e50b0 KS |
3380 | } |
3381 | ||
82b0f8c3 | 3382 | vmf->vma->vm_ops->map_pages(vmf, start_pgoff, end_pgoff); |
7267ec00 | 3383 | |
7267ec00 | 3384 | /* Huge page is mapped? Page fault is solved */ |
82b0f8c3 | 3385 | if (pmd_trans_huge(*vmf->pmd)) { |
7267ec00 KS |
3386 | ret = VM_FAULT_NOPAGE; |
3387 | goto out; | |
3388 | } | |
3389 | ||
3390 | /* ->map_pages() haven't done anything useful. Cold page cache? */ | |
82b0f8c3 | 3391 | if (!vmf->pte) |
7267ec00 KS |
3392 | goto out; |
3393 | ||
3394 | /* check if the page fault is solved */ | |
82b0f8c3 JK |
3395 | vmf->pte -= (vmf->address >> PAGE_SHIFT) - (address >> PAGE_SHIFT); |
3396 | if (!pte_none(*vmf->pte)) | |
7267ec00 | 3397 | ret = VM_FAULT_NOPAGE; |
82b0f8c3 | 3398 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
bae473a4 | 3399 | out: |
82b0f8c3 JK |
3400 | vmf->address = address; |
3401 | vmf->pte = NULL; | |
7267ec00 | 3402 | return ret; |
8c6e50b0 KS |
3403 | } |
3404 | ||
0721ec8b | 3405 | static int do_read_fault(struct vm_fault *vmf) |
e655fb29 | 3406 | { |
82b0f8c3 | 3407 | struct vm_area_struct *vma = vmf->vma; |
8c6e50b0 KS |
3408 | int ret = 0; |
3409 | ||
3410 | /* | |
3411 | * Let's call ->map_pages() first and use ->fault() as fallback | |
3412 | * if page by the offset is not ready to be mapped (cold cache or | |
3413 | * something). | |
3414 | */ | |
9b4bdd2f | 3415 | if (vma->vm_ops->map_pages && fault_around_bytes >> PAGE_SHIFT > 1) { |
0721ec8b | 3416 | ret = do_fault_around(vmf); |
7267ec00 KS |
3417 | if (ret) |
3418 | return ret; | |
8c6e50b0 | 3419 | } |
e655fb29 | 3420 | |
936ca80d | 3421 | ret = __do_fault(vmf); |
e655fb29 KS |
3422 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
3423 | return ret; | |
3424 | ||
9118c0cb | 3425 | ret |= finish_fault(vmf); |
936ca80d | 3426 | unlock_page(vmf->page); |
7267ec00 | 3427 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
936ca80d | 3428 | put_page(vmf->page); |
e655fb29 KS |
3429 | return ret; |
3430 | } | |
3431 | ||
0721ec8b | 3432 | static int do_cow_fault(struct vm_fault *vmf) |
ec47c3b9 | 3433 | { |
82b0f8c3 | 3434 | struct vm_area_struct *vma = vmf->vma; |
ec47c3b9 KS |
3435 | int ret; |
3436 | ||
3437 | if (unlikely(anon_vma_prepare(vma))) | |
3438 | return VM_FAULT_OOM; | |
3439 | ||
936ca80d JK |
3440 | vmf->cow_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vmf->address); |
3441 | if (!vmf->cow_page) | |
ec47c3b9 KS |
3442 | return VM_FAULT_OOM; |
3443 | ||
936ca80d | 3444 | if (mem_cgroup_try_charge(vmf->cow_page, vma->vm_mm, GFP_KERNEL, |
3917048d | 3445 | &vmf->memcg, false)) { |
936ca80d | 3446 | put_page(vmf->cow_page); |
ec47c3b9 KS |
3447 | return VM_FAULT_OOM; |
3448 | } | |
3449 | ||
936ca80d | 3450 | ret = __do_fault(vmf); |
ec47c3b9 KS |
3451 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
3452 | goto uncharge_out; | |
3917048d JK |
3453 | if (ret & VM_FAULT_DONE_COW) |
3454 | return ret; | |
ec47c3b9 | 3455 | |
b1aa812b | 3456 | copy_user_highpage(vmf->cow_page, vmf->page, vmf->address, vma); |
936ca80d | 3457 | __SetPageUptodate(vmf->cow_page); |
ec47c3b9 | 3458 | |
9118c0cb | 3459 | ret |= finish_fault(vmf); |
b1aa812b JK |
3460 | unlock_page(vmf->page); |
3461 | put_page(vmf->page); | |
7267ec00 KS |
3462 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
3463 | goto uncharge_out; | |
ec47c3b9 KS |
3464 | return ret; |
3465 | uncharge_out: | |
3917048d | 3466 | mem_cgroup_cancel_charge(vmf->cow_page, vmf->memcg, false); |
936ca80d | 3467 | put_page(vmf->cow_page); |
ec47c3b9 KS |
3468 | return ret; |
3469 | } | |
3470 | ||
0721ec8b | 3471 | static int do_shared_fault(struct vm_fault *vmf) |
1da177e4 | 3472 | { |
82b0f8c3 | 3473 | struct vm_area_struct *vma = vmf->vma; |
f0c6d4d2 | 3474 | int ret, tmp; |
1d65f86d | 3475 | |
936ca80d | 3476 | ret = __do_fault(vmf); |
7eae74af | 3477 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
f0c6d4d2 | 3478 | return ret; |
1da177e4 LT |
3479 | |
3480 | /* | |
f0c6d4d2 KS |
3481 | * Check if the backing address space wants to know that the page is |
3482 | * about to become writable | |
1da177e4 | 3483 | */ |
fb09a464 | 3484 | if (vma->vm_ops->page_mkwrite) { |
936ca80d | 3485 | unlock_page(vmf->page); |
38b8cb7f | 3486 | tmp = do_page_mkwrite(vmf); |
fb09a464 KS |
3487 | if (unlikely(!tmp || |
3488 | (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { | |
936ca80d | 3489 | put_page(vmf->page); |
fb09a464 | 3490 | return tmp; |
4294621f | 3491 | } |
fb09a464 KS |
3492 | } |
3493 | ||
9118c0cb | 3494 | ret |= finish_fault(vmf); |
7267ec00 KS |
3495 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | |
3496 | VM_FAULT_RETRY))) { | |
936ca80d JK |
3497 | unlock_page(vmf->page); |
3498 | put_page(vmf->page); | |
f0c6d4d2 | 3499 | return ret; |
1da177e4 | 3500 | } |
b827e496 | 3501 | |
97ba0c2b | 3502 | fault_dirty_shared_page(vma, vmf->page); |
1d65f86d | 3503 | return ret; |
54cb8821 | 3504 | } |
d00806b1 | 3505 | |
9a95f3cf PC |
3506 | /* |
3507 | * We enter with non-exclusive mmap_sem (to exclude vma changes, | |
3508 | * but allow concurrent faults). | |
3509 | * The mmap_sem may have been released depending on flags and our | |
3510 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
3511 | */ | |
82b0f8c3 | 3512 | static int do_fault(struct vm_fault *vmf) |
54cb8821 | 3513 | { |
82b0f8c3 | 3514 | struct vm_area_struct *vma = vmf->vma; |
b0b9b3df | 3515 | int ret; |
54cb8821 | 3516 | |
6b7339f4 KS |
3517 | /* The VMA was not fully populated on mmap() or missing VM_DONTEXPAND */ |
3518 | if (!vma->vm_ops->fault) | |
b0b9b3df HD |
3519 | ret = VM_FAULT_SIGBUS; |
3520 | else if (!(vmf->flags & FAULT_FLAG_WRITE)) | |
3521 | ret = do_read_fault(vmf); | |
3522 | else if (!(vma->vm_flags & VM_SHARED)) | |
3523 | ret = do_cow_fault(vmf); | |
3524 | else | |
3525 | ret = do_shared_fault(vmf); | |
3526 | ||
3527 | /* preallocated pagetable is unused: free it */ | |
3528 | if (vmf->prealloc_pte) { | |
3529 | pte_free(vma->vm_mm, vmf->prealloc_pte); | |
7f2b6ce8 | 3530 | vmf->prealloc_pte = NULL; |
b0b9b3df HD |
3531 | } |
3532 | return ret; | |
54cb8821 NP |
3533 | } |
3534 | ||
b19a9939 | 3535 | static int numa_migrate_prep(struct page *page, struct vm_area_struct *vma, |
04bb2f94 RR |
3536 | unsigned long addr, int page_nid, |
3537 | int *flags) | |
9532fec1 MG |
3538 | { |
3539 | get_page(page); | |
3540 | ||
3541 | count_vm_numa_event(NUMA_HINT_FAULTS); | |
04bb2f94 | 3542 | if (page_nid == numa_node_id()) { |
9532fec1 | 3543 | count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL); |
04bb2f94 RR |
3544 | *flags |= TNF_FAULT_LOCAL; |
3545 | } | |
9532fec1 MG |
3546 | |
3547 | return mpol_misplaced(page, vma, addr); | |
3548 | } | |
3549 | ||
2994302b | 3550 | static int do_numa_page(struct vm_fault *vmf) |
d10e63f2 | 3551 | { |
82b0f8c3 | 3552 | struct vm_area_struct *vma = vmf->vma; |
4daae3b4 | 3553 | struct page *page = NULL; |
8191acbd | 3554 | int page_nid = -1; |
90572890 | 3555 | int last_cpupid; |
cbee9f88 | 3556 | int target_nid; |
b8593bfd | 3557 | bool migrated = false; |
cee216a6 | 3558 | pte_t pte; |
288bc549 | 3559 | bool was_writable = pte_savedwrite(vmf->orig_pte); |
6688cc05 | 3560 | int flags = 0; |
d10e63f2 MG |
3561 | |
3562 | /* | |
166f61b9 TH |
3563 | * The "pte" at this point cannot be used safely without |
3564 | * validation through pte_unmap_same(). It's of NUMA type but | |
3565 | * the pfn may be screwed if the read is non atomic. | |
166f61b9 | 3566 | */ |
82b0f8c3 JK |
3567 | vmf->ptl = pte_lockptr(vma->vm_mm, vmf->pmd); |
3568 | spin_lock(vmf->ptl); | |
cee216a6 | 3569 | if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) { |
82b0f8c3 | 3570 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
4daae3b4 MG |
3571 | goto out; |
3572 | } | |
3573 | ||
cee216a6 AK |
3574 | /* |
3575 | * Make it present again, Depending on how arch implementes non | |
3576 | * accessible ptes, some can allow access by kernel mode. | |
3577 | */ | |
3578 | pte = ptep_modify_prot_start(vma->vm_mm, vmf->address, vmf->pte); | |
4d942466 MG |
3579 | pte = pte_modify(pte, vma->vm_page_prot); |
3580 | pte = pte_mkyoung(pte); | |
b191f9b1 MG |
3581 | if (was_writable) |
3582 | pte = pte_mkwrite(pte); | |
cee216a6 | 3583 | ptep_modify_prot_commit(vma->vm_mm, vmf->address, vmf->pte, pte); |
82b0f8c3 | 3584 | update_mmu_cache(vma, vmf->address, vmf->pte); |
d10e63f2 | 3585 | |
82b0f8c3 | 3586 | page = vm_normal_page(vma, vmf->address, pte); |
d10e63f2 | 3587 | if (!page) { |
82b0f8c3 | 3588 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
d10e63f2 MG |
3589 | return 0; |
3590 | } | |
3591 | ||
e81c4802 KS |
3592 | /* TODO: handle PTE-mapped THP */ |
3593 | if (PageCompound(page)) { | |
82b0f8c3 | 3594 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
e81c4802 KS |
3595 | return 0; |
3596 | } | |
3597 | ||
6688cc05 | 3598 | /* |
bea66fbd MG |
3599 | * Avoid grouping on RO pages in general. RO pages shouldn't hurt as |
3600 | * much anyway since they can be in shared cache state. This misses | |
3601 | * the case where a mapping is writable but the process never writes | |
3602 | * to it but pte_write gets cleared during protection updates and | |
3603 | * pte_dirty has unpredictable behaviour between PTE scan updates, | |
3604 | * background writeback, dirty balancing and application behaviour. | |
6688cc05 | 3605 | */ |
d59dc7bc | 3606 | if (!pte_write(pte)) |
6688cc05 PZ |
3607 | flags |= TNF_NO_GROUP; |
3608 | ||
dabe1d99 RR |
3609 | /* |
3610 | * Flag if the page is shared between multiple address spaces. This | |
3611 | * is later used when determining whether to group tasks together | |
3612 | */ | |
3613 | if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED)) | |
3614 | flags |= TNF_SHARED; | |
3615 | ||
90572890 | 3616 | last_cpupid = page_cpupid_last(page); |
8191acbd | 3617 | page_nid = page_to_nid(page); |
82b0f8c3 | 3618 | target_nid = numa_migrate_prep(page, vma, vmf->address, page_nid, |
bae473a4 | 3619 | &flags); |
82b0f8c3 | 3620 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
4daae3b4 | 3621 | if (target_nid == -1) { |
4daae3b4 MG |
3622 | put_page(page); |
3623 | goto out; | |
3624 | } | |
3625 | ||
3626 | /* Migrate to the requested node */ | |
1bc115d8 | 3627 | migrated = migrate_misplaced_page(page, vma, target_nid); |
6688cc05 | 3628 | if (migrated) { |
8191acbd | 3629 | page_nid = target_nid; |
6688cc05 | 3630 | flags |= TNF_MIGRATED; |
074c2381 MG |
3631 | } else |
3632 | flags |= TNF_MIGRATE_FAIL; | |
4daae3b4 MG |
3633 | |
3634 | out: | |
8191acbd | 3635 | if (page_nid != -1) |
6688cc05 | 3636 | task_numa_fault(last_cpupid, page_nid, 1, flags); |
d10e63f2 MG |
3637 | return 0; |
3638 | } | |
3639 | ||
91a90140 | 3640 | static inline int create_huge_pmd(struct vm_fault *vmf) |
b96375f7 | 3641 | { |
f4200391 | 3642 | if (vma_is_anonymous(vmf->vma)) |
82b0f8c3 | 3643 | return do_huge_pmd_anonymous_page(vmf); |
a2d58167 | 3644 | if (vmf->vma->vm_ops->huge_fault) |
c791ace1 | 3645 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD); |
b96375f7 MW |
3646 | return VM_FAULT_FALLBACK; |
3647 | } | |
3648 | ||
82b0f8c3 | 3649 | static int wp_huge_pmd(struct vm_fault *vmf, pmd_t orig_pmd) |
b96375f7 | 3650 | { |
82b0f8c3 JK |
3651 | if (vma_is_anonymous(vmf->vma)) |
3652 | return do_huge_pmd_wp_page(vmf, orig_pmd); | |
a2d58167 | 3653 | if (vmf->vma->vm_ops->huge_fault) |
c791ace1 | 3654 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD); |
af9e4d5f KS |
3655 | |
3656 | /* COW handled on pte level: split pmd */ | |
82b0f8c3 JK |
3657 | VM_BUG_ON_VMA(vmf->vma->vm_flags & VM_SHARED, vmf->vma); |
3658 | __split_huge_pmd(vmf->vma, vmf->pmd, vmf->address, false, NULL); | |
af9e4d5f | 3659 | |
b96375f7 MW |
3660 | return VM_FAULT_FALLBACK; |
3661 | } | |
3662 | ||
38e08854 LS |
3663 | static inline bool vma_is_accessible(struct vm_area_struct *vma) |
3664 | { | |
3665 | return vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE); | |
3666 | } | |
3667 | ||
a00cc7d9 MW |
3668 | static int create_huge_pud(struct vm_fault *vmf) |
3669 | { | |
3670 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
3671 | /* No support for anonymous transparent PUD pages yet */ | |
3672 | if (vma_is_anonymous(vmf->vma)) | |
3673 | return VM_FAULT_FALLBACK; | |
3674 | if (vmf->vma->vm_ops->huge_fault) | |
c791ace1 | 3675 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD); |
a00cc7d9 MW |
3676 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
3677 | return VM_FAULT_FALLBACK; | |
3678 | } | |
3679 | ||
3680 | static int wp_huge_pud(struct vm_fault *vmf, pud_t orig_pud) | |
3681 | { | |
3682 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
3683 | /* No support for anonymous transparent PUD pages yet */ | |
3684 | if (vma_is_anonymous(vmf->vma)) | |
3685 | return VM_FAULT_FALLBACK; | |
3686 | if (vmf->vma->vm_ops->huge_fault) | |
c791ace1 | 3687 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD); |
a00cc7d9 MW |
3688 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
3689 | return VM_FAULT_FALLBACK; | |
3690 | } | |
3691 | ||
1da177e4 LT |
3692 | /* |
3693 | * These routines also need to handle stuff like marking pages dirty | |
3694 | * and/or accessed for architectures that don't do it in hardware (most | |
3695 | * RISC architectures). The early dirtying is also good on the i386. | |
3696 | * | |
3697 | * There is also a hook called "update_mmu_cache()" that architectures | |
3698 | * with external mmu caches can use to update those (ie the Sparc or | |
3699 | * PowerPC hashed page tables that act as extended TLBs). | |
3700 | * | |
7267ec00 KS |
3701 | * We enter with non-exclusive mmap_sem (to exclude vma changes, but allow |
3702 | * concurrent faults). | |
9a95f3cf | 3703 | * |
7267ec00 KS |
3704 | * The mmap_sem may have been released depending on flags and our return value. |
3705 | * See filemap_fault() and __lock_page_or_retry(). | |
1da177e4 | 3706 | */ |
82b0f8c3 | 3707 | static int handle_pte_fault(struct vm_fault *vmf) |
1da177e4 LT |
3708 | { |
3709 | pte_t entry; | |
3710 | ||
82b0f8c3 | 3711 | if (unlikely(pmd_none(*vmf->pmd))) { |
7267ec00 KS |
3712 | /* |
3713 | * Leave __pte_alloc() until later: because vm_ops->fault may | |
3714 | * want to allocate huge page, and if we expose page table | |
3715 | * for an instant, it will be difficult to retract from | |
3716 | * concurrent faults and from rmap lookups. | |
3717 | */ | |
82b0f8c3 | 3718 | vmf->pte = NULL; |
7267ec00 KS |
3719 | } else { |
3720 | /* See comment in pte_alloc_one_map() */ | |
d0f0931d | 3721 | if (pmd_devmap_trans_unstable(vmf->pmd)) |
7267ec00 KS |
3722 | return 0; |
3723 | /* | |
3724 | * A regular pmd is established and it can't morph into a huge | |
3725 | * pmd from under us anymore at this point because we hold the | |
3726 | * mmap_sem read mode and khugepaged takes it in write mode. | |
3727 | * So now it's safe to run pte_offset_map(). | |
3728 | */ | |
82b0f8c3 | 3729 | vmf->pte = pte_offset_map(vmf->pmd, vmf->address); |
2994302b | 3730 | vmf->orig_pte = *vmf->pte; |
7267ec00 KS |
3731 | |
3732 | /* | |
3733 | * some architectures can have larger ptes than wordsize, | |
3734 | * e.g.ppc44x-defconfig has CONFIG_PTE_64BIT=y and | |
3735 | * CONFIG_32BIT=y, so READ_ONCE or ACCESS_ONCE cannot guarantee | |
3736 | * atomic accesses. The code below just needs a consistent | |
3737 | * view for the ifs and we later double check anyway with the | |
3738 | * ptl lock held. So here a barrier will do. | |
3739 | */ | |
3740 | barrier(); | |
2994302b | 3741 | if (pte_none(vmf->orig_pte)) { |
82b0f8c3 JK |
3742 | pte_unmap(vmf->pte); |
3743 | vmf->pte = NULL; | |
65500d23 | 3744 | } |
1da177e4 LT |
3745 | } |
3746 | ||
82b0f8c3 JK |
3747 | if (!vmf->pte) { |
3748 | if (vma_is_anonymous(vmf->vma)) | |
3749 | return do_anonymous_page(vmf); | |
7267ec00 | 3750 | else |
82b0f8c3 | 3751 | return do_fault(vmf); |
7267ec00 KS |
3752 | } |
3753 | ||
2994302b JK |
3754 | if (!pte_present(vmf->orig_pte)) |
3755 | return do_swap_page(vmf); | |
7267ec00 | 3756 | |
2994302b JK |
3757 | if (pte_protnone(vmf->orig_pte) && vma_is_accessible(vmf->vma)) |
3758 | return do_numa_page(vmf); | |
d10e63f2 | 3759 | |
82b0f8c3 JK |
3760 | vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd); |
3761 | spin_lock(vmf->ptl); | |
2994302b | 3762 | entry = vmf->orig_pte; |
82b0f8c3 | 3763 | if (unlikely(!pte_same(*vmf->pte, entry))) |
8f4e2101 | 3764 | goto unlock; |
82b0f8c3 | 3765 | if (vmf->flags & FAULT_FLAG_WRITE) { |
1da177e4 | 3766 | if (!pte_write(entry)) |
2994302b | 3767 | return do_wp_page(vmf); |
1da177e4 LT |
3768 | entry = pte_mkdirty(entry); |
3769 | } | |
3770 | entry = pte_mkyoung(entry); | |
82b0f8c3 JK |
3771 | if (ptep_set_access_flags(vmf->vma, vmf->address, vmf->pte, entry, |
3772 | vmf->flags & FAULT_FLAG_WRITE)) { | |
3773 | update_mmu_cache(vmf->vma, vmf->address, vmf->pte); | |
1a44e149 AA |
3774 | } else { |
3775 | /* | |
3776 | * This is needed only for protection faults but the arch code | |
3777 | * is not yet telling us if this is a protection fault or not. | |
3778 | * This still avoids useless tlb flushes for .text page faults | |
3779 | * with threads. | |
3780 | */ | |
82b0f8c3 JK |
3781 | if (vmf->flags & FAULT_FLAG_WRITE) |
3782 | flush_tlb_fix_spurious_fault(vmf->vma, vmf->address); | |
1a44e149 | 3783 | } |
8f4e2101 | 3784 | unlock: |
82b0f8c3 | 3785 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
83c54070 | 3786 | return 0; |
1da177e4 LT |
3787 | } |
3788 | ||
3789 | /* | |
3790 | * By the time we get here, we already hold the mm semaphore | |
9a95f3cf PC |
3791 | * |
3792 | * The mmap_sem may have been released depending on flags and our | |
3793 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
1da177e4 | 3794 | */ |
dcddffd4 KS |
3795 | static int __handle_mm_fault(struct vm_area_struct *vma, unsigned long address, |
3796 | unsigned int flags) | |
1da177e4 | 3797 | { |
82b0f8c3 | 3798 | struct vm_fault vmf = { |
bae473a4 | 3799 | .vma = vma, |
1a29d85e | 3800 | .address = address & PAGE_MASK, |
bae473a4 | 3801 | .flags = flags, |
0721ec8b | 3802 | .pgoff = linear_page_index(vma, address), |
667240e0 | 3803 | .gfp_mask = __get_fault_gfp_mask(vma), |
bae473a4 | 3804 | }; |
dcddffd4 | 3805 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 3806 | pgd_t *pgd; |
c2febafc | 3807 | p4d_t *p4d; |
a2d58167 | 3808 | int ret; |
1da177e4 | 3809 | |
1da177e4 | 3810 | pgd = pgd_offset(mm, address); |
c2febafc KS |
3811 | p4d = p4d_alloc(mm, pgd, address); |
3812 | if (!p4d) | |
3813 | return VM_FAULT_OOM; | |
a00cc7d9 | 3814 | |
c2febafc | 3815 | vmf.pud = pud_alloc(mm, p4d, address); |
a00cc7d9 | 3816 | if (!vmf.pud) |
c74df32c | 3817 | return VM_FAULT_OOM; |
a00cc7d9 | 3818 | if (pud_none(*vmf.pud) && transparent_hugepage_enabled(vma)) { |
a00cc7d9 MW |
3819 | ret = create_huge_pud(&vmf); |
3820 | if (!(ret & VM_FAULT_FALLBACK)) | |
3821 | return ret; | |
3822 | } else { | |
3823 | pud_t orig_pud = *vmf.pud; | |
3824 | ||
3825 | barrier(); | |
3826 | if (pud_trans_huge(orig_pud) || pud_devmap(orig_pud)) { | |
3827 | unsigned int dirty = flags & FAULT_FLAG_WRITE; | |
3828 | ||
a00cc7d9 MW |
3829 | /* NUMA case for anonymous PUDs would go here */ |
3830 | ||
3831 | if (dirty && !pud_write(orig_pud)) { | |
3832 | ret = wp_huge_pud(&vmf, orig_pud); | |
3833 | if (!(ret & VM_FAULT_FALLBACK)) | |
3834 | return ret; | |
3835 | } else { | |
3836 | huge_pud_set_accessed(&vmf, orig_pud); | |
3837 | return 0; | |
3838 | } | |
3839 | } | |
3840 | } | |
3841 | ||
3842 | vmf.pmd = pmd_alloc(mm, vmf.pud, address); | |
82b0f8c3 | 3843 | if (!vmf.pmd) |
c74df32c | 3844 | return VM_FAULT_OOM; |
82b0f8c3 | 3845 | if (pmd_none(*vmf.pmd) && transparent_hugepage_enabled(vma)) { |
a2d58167 | 3846 | ret = create_huge_pmd(&vmf); |
c0292554 KS |
3847 | if (!(ret & VM_FAULT_FALLBACK)) |
3848 | return ret; | |
71e3aac0 | 3849 | } else { |
82b0f8c3 | 3850 | pmd_t orig_pmd = *vmf.pmd; |
1f1d06c3 | 3851 | |
71e3aac0 | 3852 | barrier(); |
5c7fb56e | 3853 | if (pmd_trans_huge(orig_pmd) || pmd_devmap(orig_pmd)) { |
38e08854 | 3854 | if (pmd_protnone(orig_pmd) && vma_is_accessible(vma)) |
82b0f8c3 | 3855 | return do_huge_pmd_numa_page(&vmf, orig_pmd); |
d10e63f2 | 3856 | |
82b0f8c3 | 3857 | if ((vmf.flags & FAULT_FLAG_WRITE) && |
bae473a4 | 3858 | !pmd_write(orig_pmd)) { |
82b0f8c3 | 3859 | ret = wp_huge_pmd(&vmf, orig_pmd); |
9845cbbd KS |
3860 | if (!(ret & VM_FAULT_FALLBACK)) |
3861 | return ret; | |
a1dd450b | 3862 | } else { |
82b0f8c3 | 3863 | huge_pmd_set_accessed(&vmf, orig_pmd); |
9845cbbd | 3864 | return 0; |
1f1d06c3 | 3865 | } |
71e3aac0 AA |
3866 | } |
3867 | } | |
3868 | ||
82b0f8c3 | 3869 | return handle_pte_fault(&vmf); |
1da177e4 LT |
3870 | } |
3871 | ||
9a95f3cf PC |
3872 | /* |
3873 | * By the time we get here, we already hold the mm semaphore | |
3874 | * | |
3875 | * The mmap_sem may have been released depending on flags and our | |
3876 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
3877 | */ | |
dcddffd4 KS |
3878 | int handle_mm_fault(struct vm_area_struct *vma, unsigned long address, |
3879 | unsigned int flags) | |
519e5247 JW |
3880 | { |
3881 | int ret; | |
3882 | ||
3883 | __set_current_state(TASK_RUNNING); | |
3884 | ||
3885 | count_vm_event(PGFAULT); | |
2262185c | 3886 | count_memcg_event_mm(vma->vm_mm, PGFAULT); |
519e5247 JW |
3887 | |
3888 | /* do counter updates before entering really critical section. */ | |
3889 | check_sync_rss_stat(current); | |
3890 | ||
3891 | /* | |
3892 | * Enable the memcg OOM handling for faults triggered in user | |
3893 | * space. Kernel faults are handled more gracefully. | |
3894 | */ | |
3895 | if (flags & FAULT_FLAG_USER) | |
49426420 | 3896 | mem_cgroup_oom_enable(); |
519e5247 | 3897 | |
bae473a4 KS |
3898 | if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE, |
3899 | flags & FAULT_FLAG_INSTRUCTION, | |
3900 | flags & FAULT_FLAG_REMOTE)) | |
3901 | return VM_FAULT_SIGSEGV; | |
3902 | ||
3903 | if (unlikely(is_vm_hugetlb_page(vma))) | |
3904 | ret = hugetlb_fault(vma->vm_mm, vma, address, flags); | |
3905 | else | |
3906 | ret = __handle_mm_fault(vma, address, flags); | |
519e5247 | 3907 | |
49426420 JW |
3908 | if (flags & FAULT_FLAG_USER) { |
3909 | mem_cgroup_oom_disable(); | |
166f61b9 TH |
3910 | /* |
3911 | * The task may have entered a memcg OOM situation but | |
3912 | * if the allocation error was handled gracefully (no | |
3913 | * VM_FAULT_OOM), there is no need to kill anything. | |
3914 | * Just clean up the OOM state peacefully. | |
3915 | */ | |
3916 | if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM)) | |
3917 | mem_cgroup_oom_synchronize(false); | |
49426420 | 3918 | } |
3812c8c8 | 3919 | |
519e5247 JW |
3920 | return ret; |
3921 | } | |
e1d6d01a | 3922 | EXPORT_SYMBOL_GPL(handle_mm_fault); |
519e5247 | 3923 | |
90eceff1 KS |
3924 | #ifndef __PAGETABLE_P4D_FOLDED |
3925 | /* | |
3926 | * Allocate p4d page table. | |
3927 | * We've already handled the fast-path in-line. | |
3928 | */ | |
3929 | int __p4d_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) | |
3930 | { | |
3931 | p4d_t *new = p4d_alloc_one(mm, address); | |
3932 | if (!new) | |
3933 | return -ENOMEM; | |
3934 | ||
3935 | smp_wmb(); /* See comment in __pte_alloc */ | |
3936 | ||
3937 | spin_lock(&mm->page_table_lock); | |
3938 | if (pgd_present(*pgd)) /* Another has populated it */ | |
3939 | p4d_free(mm, new); | |
3940 | else | |
3941 | pgd_populate(mm, pgd, new); | |
3942 | spin_unlock(&mm->page_table_lock); | |
3943 | return 0; | |
3944 | } | |
3945 | #endif /* __PAGETABLE_P4D_FOLDED */ | |
3946 | ||
1da177e4 LT |
3947 | #ifndef __PAGETABLE_PUD_FOLDED |
3948 | /* | |
3949 | * Allocate page upper directory. | |
872fec16 | 3950 | * We've already handled the fast-path in-line. |
1da177e4 | 3951 | */ |
c2febafc | 3952 | int __pud_alloc(struct mm_struct *mm, p4d_t *p4d, unsigned long address) |
1da177e4 | 3953 | { |
c74df32c HD |
3954 | pud_t *new = pud_alloc_one(mm, address); |
3955 | if (!new) | |
1bb3630e | 3956 | return -ENOMEM; |
1da177e4 | 3957 | |
362a61ad NP |
3958 | smp_wmb(); /* See comment in __pte_alloc */ |
3959 | ||
872fec16 | 3960 | spin_lock(&mm->page_table_lock); |
c2febafc KS |
3961 | #ifndef __ARCH_HAS_5LEVEL_HACK |
3962 | if (p4d_present(*p4d)) /* Another has populated it */ | |
5e541973 | 3963 | pud_free(mm, new); |
1bb3630e | 3964 | else |
c2febafc KS |
3965 | p4d_populate(mm, p4d, new); |
3966 | #else | |
3967 | if (pgd_present(*p4d)) /* Another has populated it */ | |
5e541973 | 3968 | pud_free(mm, new); |
1bb3630e | 3969 | else |
c2febafc KS |
3970 | pgd_populate(mm, p4d, new); |
3971 | #endif /* __ARCH_HAS_5LEVEL_HACK */ | |
c74df32c | 3972 | spin_unlock(&mm->page_table_lock); |
1bb3630e | 3973 | return 0; |
1da177e4 LT |
3974 | } |
3975 | #endif /* __PAGETABLE_PUD_FOLDED */ | |
3976 | ||
3977 | #ifndef __PAGETABLE_PMD_FOLDED | |
3978 | /* | |
3979 | * Allocate page middle directory. | |
872fec16 | 3980 | * We've already handled the fast-path in-line. |
1da177e4 | 3981 | */ |
1bb3630e | 3982 | int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) |
1da177e4 | 3983 | { |
a00cc7d9 | 3984 | spinlock_t *ptl; |
c74df32c HD |
3985 | pmd_t *new = pmd_alloc_one(mm, address); |
3986 | if (!new) | |
1bb3630e | 3987 | return -ENOMEM; |
1da177e4 | 3988 | |
362a61ad NP |
3989 | smp_wmb(); /* See comment in __pte_alloc */ |
3990 | ||
a00cc7d9 | 3991 | ptl = pud_lock(mm, pud); |
1da177e4 | 3992 | #ifndef __ARCH_HAS_4LEVEL_HACK |
dc6c9a35 KS |
3993 | if (!pud_present(*pud)) { |
3994 | mm_inc_nr_pmds(mm); | |
1bb3630e | 3995 | pud_populate(mm, pud, new); |
dc6c9a35 | 3996 | } else /* Another has populated it */ |
5e541973 | 3997 | pmd_free(mm, new); |
dc6c9a35 KS |
3998 | #else |
3999 | if (!pgd_present(*pud)) { | |
4000 | mm_inc_nr_pmds(mm); | |
1bb3630e | 4001 | pgd_populate(mm, pud, new); |
dc6c9a35 KS |
4002 | } else /* Another has populated it */ |
4003 | pmd_free(mm, new); | |
1da177e4 | 4004 | #endif /* __ARCH_HAS_4LEVEL_HACK */ |
a00cc7d9 | 4005 | spin_unlock(ptl); |
1bb3630e | 4006 | return 0; |
e0f39591 | 4007 | } |
1da177e4 LT |
4008 | #endif /* __PAGETABLE_PMD_FOLDED */ |
4009 | ||
09796395 | 4010 | static int __follow_pte_pmd(struct mm_struct *mm, unsigned long address, |
a4d1a885 JG |
4011 | unsigned long *start, unsigned long *end, |
4012 | pte_t **ptepp, pmd_t **pmdpp, spinlock_t **ptlp) | |
f8ad0f49 JW |
4013 | { |
4014 | pgd_t *pgd; | |
c2febafc | 4015 | p4d_t *p4d; |
f8ad0f49 JW |
4016 | pud_t *pud; |
4017 | pmd_t *pmd; | |
4018 | pte_t *ptep; | |
4019 | ||
4020 | pgd = pgd_offset(mm, address); | |
4021 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
4022 | goto out; | |
4023 | ||
c2febafc KS |
4024 | p4d = p4d_offset(pgd, address); |
4025 | if (p4d_none(*p4d) || unlikely(p4d_bad(*p4d))) | |
4026 | goto out; | |
4027 | ||
4028 | pud = pud_offset(p4d, address); | |
f8ad0f49 JW |
4029 | if (pud_none(*pud) || unlikely(pud_bad(*pud))) |
4030 | goto out; | |
4031 | ||
4032 | pmd = pmd_offset(pud, address); | |
f66055ab | 4033 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
f8ad0f49 | 4034 | |
09796395 RZ |
4035 | if (pmd_huge(*pmd)) { |
4036 | if (!pmdpp) | |
4037 | goto out; | |
4038 | ||
a4d1a885 JG |
4039 | if (start && end) { |
4040 | *start = address & PMD_MASK; | |
4041 | *end = *start + PMD_SIZE; | |
4042 | mmu_notifier_invalidate_range_start(mm, *start, *end); | |
4043 | } | |
09796395 RZ |
4044 | *ptlp = pmd_lock(mm, pmd); |
4045 | if (pmd_huge(*pmd)) { | |
4046 | *pmdpp = pmd; | |
4047 | return 0; | |
4048 | } | |
4049 | spin_unlock(*ptlp); | |
a4d1a885 JG |
4050 | if (start && end) |
4051 | mmu_notifier_invalidate_range_end(mm, *start, *end); | |
09796395 RZ |
4052 | } |
4053 | ||
4054 | if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) | |
f8ad0f49 JW |
4055 | goto out; |
4056 | ||
a4d1a885 JG |
4057 | if (start && end) { |
4058 | *start = address & PAGE_MASK; | |
4059 | *end = *start + PAGE_SIZE; | |
4060 | mmu_notifier_invalidate_range_start(mm, *start, *end); | |
4061 | } | |
f8ad0f49 | 4062 | ptep = pte_offset_map_lock(mm, pmd, address, ptlp); |
f8ad0f49 JW |
4063 | if (!pte_present(*ptep)) |
4064 | goto unlock; | |
4065 | *ptepp = ptep; | |
4066 | return 0; | |
4067 | unlock: | |
4068 | pte_unmap_unlock(ptep, *ptlp); | |
a4d1a885 JG |
4069 | if (start && end) |
4070 | mmu_notifier_invalidate_range_end(mm, *start, *end); | |
f8ad0f49 JW |
4071 | out: |
4072 | return -EINVAL; | |
4073 | } | |
4074 | ||
f729c8c9 RZ |
4075 | static inline int follow_pte(struct mm_struct *mm, unsigned long address, |
4076 | pte_t **ptepp, spinlock_t **ptlp) | |
1b36ba81 NK |
4077 | { |
4078 | int res; | |
4079 | ||
4080 | /* (void) is needed to make gcc happy */ | |
4081 | (void) __cond_lock(*ptlp, | |
a4d1a885 JG |
4082 | !(res = __follow_pte_pmd(mm, address, NULL, NULL, |
4083 | ptepp, NULL, ptlp))); | |
09796395 RZ |
4084 | return res; |
4085 | } | |
4086 | ||
4087 | int follow_pte_pmd(struct mm_struct *mm, unsigned long address, | |
a4d1a885 | 4088 | unsigned long *start, unsigned long *end, |
09796395 RZ |
4089 | pte_t **ptepp, pmd_t **pmdpp, spinlock_t **ptlp) |
4090 | { | |
4091 | int res; | |
4092 | ||
4093 | /* (void) is needed to make gcc happy */ | |
4094 | (void) __cond_lock(*ptlp, | |
a4d1a885 JG |
4095 | !(res = __follow_pte_pmd(mm, address, start, end, |
4096 | ptepp, pmdpp, ptlp))); | |
1b36ba81 NK |
4097 | return res; |
4098 | } | |
09796395 | 4099 | EXPORT_SYMBOL(follow_pte_pmd); |
1b36ba81 | 4100 | |
3b6748e2 JW |
4101 | /** |
4102 | * follow_pfn - look up PFN at a user virtual address | |
4103 | * @vma: memory mapping | |
4104 | * @address: user virtual address | |
4105 | * @pfn: location to store found PFN | |
4106 | * | |
4107 | * Only IO mappings and raw PFN mappings are allowed. | |
4108 | * | |
4109 | * Returns zero and the pfn at @pfn on success, -ve otherwise. | |
4110 | */ | |
4111 | int follow_pfn(struct vm_area_struct *vma, unsigned long address, | |
4112 | unsigned long *pfn) | |
4113 | { | |
4114 | int ret = -EINVAL; | |
4115 | spinlock_t *ptl; | |
4116 | pte_t *ptep; | |
4117 | ||
4118 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) | |
4119 | return ret; | |
4120 | ||
4121 | ret = follow_pte(vma->vm_mm, address, &ptep, &ptl); | |
4122 | if (ret) | |
4123 | return ret; | |
4124 | *pfn = pte_pfn(*ptep); | |
4125 | pte_unmap_unlock(ptep, ptl); | |
4126 | return 0; | |
4127 | } | |
4128 | EXPORT_SYMBOL(follow_pfn); | |
4129 | ||
28b2ee20 | 4130 | #ifdef CONFIG_HAVE_IOREMAP_PROT |
d87fe660 | 4131 | int follow_phys(struct vm_area_struct *vma, |
4132 | unsigned long address, unsigned int flags, | |
4133 | unsigned long *prot, resource_size_t *phys) | |
28b2ee20 | 4134 | { |
03668a4d | 4135 | int ret = -EINVAL; |
28b2ee20 RR |
4136 | pte_t *ptep, pte; |
4137 | spinlock_t *ptl; | |
28b2ee20 | 4138 | |
d87fe660 | 4139 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) |
4140 | goto out; | |
28b2ee20 | 4141 | |
03668a4d | 4142 | if (follow_pte(vma->vm_mm, address, &ptep, &ptl)) |
d87fe660 | 4143 | goto out; |
28b2ee20 | 4144 | pte = *ptep; |
03668a4d | 4145 | |
28b2ee20 RR |
4146 | if ((flags & FOLL_WRITE) && !pte_write(pte)) |
4147 | goto unlock; | |
28b2ee20 RR |
4148 | |
4149 | *prot = pgprot_val(pte_pgprot(pte)); | |
03668a4d | 4150 | *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; |
28b2ee20 | 4151 | |
03668a4d | 4152 | ret = 0; |
28b2ee20 RR |
4153 | unlock: |
4154 | pte_unmap_unlock(ptep, ptl); | |
4155 | out: | |
d87fe660 | 4156 | return ret; |
28b2ee20 RR |
4157 | } |
4158 | ||
4159 | int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, | |
4160 | void *buf, int len, int write) | |
4161 | { | |
4162 | resource_size_t phys_addr; | |
4163 | unsigned long prot = 0; | |
2bc7273b | 4164 | void __iomem *maddr; |
28b2ee20 RR |
4165 | int offset = addr & (PAGE_SIZE-1); |
4166 | ||
d87fe660 | 4167 | if (follow_phys(vma, addr, write, &prot, &phys_addr)) |
28b2ee20 RR |
4168 | return -EINVAL; |
4169 | ||
9cb12d7b | 4170 | maddr = ioremap_prot(phys_addr, PAGE_ALIGN(len + offset), prot); |
28b2ee20 RR |
4171 | if (write) |
4172 | memcpy_toio(maddr + offset, buf, len); | |
4173 | else | |
4174 | memcpy_fromio(buf, maddr + offset, len); | |
4175 | iounmap(maddr); | |
4176 | ||
4177 | return len; | |
4178 | } | |
5a73633e | 4179 | EXPORT_SYMBOL_GPL(generic_access_phys); |
28b2ee20 RR |
4180 | #endif |
4181 | ||
0ec76a11 | 4182 | /* |
206cb636 SW |
4183 | * Access another process' address space as given in mm. If non-NULL, use the |
4184 | * given task for page fault accounting. | |
0ec76a11 | 4185 | */ |
84d77d3f | 4186 | int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, |
442486ec | 4187 | unsigned long addr, void *buf, int len, unsigned int gup_flags) |
0ec76a11 | 4188 | { |
0ec76a11 | 4189 | struct vm_area_struct *vma; |
0ec76a11 | 4190 | void *old_buf = buf; |
442486ec | 4191 | int write = gup_flags & FOLL_WRITE; |
0ec76a11 | 4192 | |
0ec76a11 | 4193 | down_read(&mm->mmap_sem); |
183ff22b | 4194 | /* ignore errors, just check how much was successfully transferred */ |
0ec76a11 DH |
4195 | while (len) { |
4196 | int bytes, ret, offset; | |
4197 | void *maddr; | |
28b2ee20 | 4198 | struct page *page = NULL; |
0ec76a11 | 4199 | |
1e987790 | 4200 | ret = get_user_pages_remote(tsk, mm, addr, 1, |
5b56d49f | 4201 | gup_flags, &page, &vma, NULL); |
28b2ee20 | 4202 | if (ret <= 0) { |
dbffcd03 RR |
4203 | #ifndef CONFIG_HAVE_IOREMAP_PROT |
4204 | break; | |
4205 | #else | |
28b2ee20 RR |
4206 | /* |
4207 | * Check if this is a VM_IO | VM_PFNMAP VMA, which | |
4208 | * we can access using slightly different code. | |
4209 | */ | |
28b2ee20 | 4210 | vma = find_vma(mm, addr); |
fe936dfc | 4211 | if (!vma || vma->vm_start > addr) |
28b2ee20 RR |
4212 | break; |
4213 | if (vma->vm_ops && vma->vm_ops->access) | |
4214 | ret = vma->vm_ops->access(vma, addr, buf, | |
4215 | len, write); | |
4216 | if (ret <= 0) | |
28b2ee20 RR |
4217 | break; |
4218 | bytes = ret; | |
dbffcd03 | 4219 | #endif |
0ec76a11 | 4220 | } else { |
28b2ee20 RR |
4221 | bytes = len; |
4222 | offset = addr & (PAGE_SIZE-1); | |
4223 | if (bytes > PAGE_SIZE-offset) | |
4224 | bytes = PAGE_SIZE-offset; | |
4225 | ||
4226 | maddr = kmap(page); | |
4227 | if (write) { | |
4228 | copy_to_user_page(vma, page, addr, | |
4229 | maddr + offset, buf, bytes); | |
4230 | set_page_dirty_lock(page); | |
4231 | } else { | |
4232 | copy_from_user_page(vma, page, addr, | |
4233 | buf, maddr + offset, bytes); | |
4234 | } | |
4235 | kunmap(page); | |
09cbfeaf | 4236 | put_page(page); |
0ec76a11 | 4237 | } |
0ec76a11 DH |
4238 | len -= bytes; |
4239 | buf += bytes; | |
4240 | addr += bytes; | |
4241 | } | |
4242 | up_read(&mm->mmap_sem); | |
0ec76a11 DH |
4243 | |
4244 | return buf - old_buf; | |
4245 | } | |
03252919 | 4246 | |
5ddd36b9 | 4247 | /** |
ae91dbfc | 4248 | * access_remote_vm - access another process' address space |
5ddd36b9 SW |
4249 | * @mm: the mm_struct of the target address space |
4250 | * @addr: start address to access | |
4251 | * @buf: source or destination buffer | |
4252 | * @len: number of bytes to transfer | |
6347e8d5 | 4253 | * @gup_flags: flags modifying lookup behaviour |
5ddd36b9 SW |
4254 | * |
4255 | * The caller must hold a reference on @mm. | |
4256 | */ | |
4257 | int access_remote_vm(struct mm_struct *mm, unsigned long addr, | |
6347e8d5 | 4258 | void *buf, int len, unsigned int gup_flags) |
5ddd36b9 | 4259 | { |
6347e8d5 | 4260 | return __access_remote_vm(NULL, mm, addr, buf, len, gup_flags); |
5ddd36b9 SW |
4261 | } |
4262 | ||
206cb636 SW |
4263 | /* |
4264 | * Access another process' address space. | |
4265 | * Source/target buffer must be kernel space, | |
4266 | * Do not walk the page table directly, use get_user_pages | |
4267 | */ | |
4268 | int access_process_vm(struct task_struct *tsk, unsigned long addr, | |
f307ab6d | 4269 | void *buf, int len, unsigned int gup_flags) |
206cb636 SW |
4270 | { |
4271 | struct mm_struct *mm; | |
4272 | int ret; | |
4273 | ||
4274 | mm = get_task_mm(tsk); | |
4275 | if (!mm) | |
4276 | return 0; | |
4277 | ||
f307ab6d | 4278 | ret = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags); |
442486ec | 4279 | |
206cb636 SW |
4280 | mmput(mm); |
4281 | ||
4282 | return ret; | |
4283 | } | |
fcd35857 | 4284 | EXPORT_SYMBOL_GPL(access_process_vm); |
206cb636 | 4285 | |
03252919 AK |
4286 | /* |
4287 | * Print the name of a VMA. | |
4288 | */ | |
4289 | void print_vma_addr(char *prefix, unsigned long ip) | |
4290 | { | |
4291 | struct mm_struct *mm = current->mm; | |
4292 | struct vm_area_struct *vma; | |
4293 | ||
e8bff74a IM |
4294 | /* |
4295 | * Do not print if we are in atomic | |
4296 | * contexts (in exception stacks, etc.): | |
4297 | */ | |
4298 | if (preempt_count()) | |
4299 | return; | |
4300 | ||
03252919 AK |
4301 | down_read(&mm->mmap_sem); |
4302 | vma = find_vma(mm, ip); | |
4303 | if (vma && vma->vm_file) { | |
4304 | struct file *f = vma->vm_file; | |
4305 | char *buf = (char *)__get_free_page(GFP_KERNEL); | |
4306 | if (buf) { | |
2fbc57c5 | 4307 | char *p; |
03252919 | 4308 | |
9bf39ab2 | 4309 | p = file_path(f, buf, PAGE_SIZE); |
03252919 AK |
4310 | if (IS_ERR(p)) |
4311 | p = "?"; | |
2fbc57c5 | 4312 | printk("%s%s[%lx+%lx]", prefix, kbasename(p), |
03252919 AK |
4313 | vma->vm_start, |
4314 | vma->vm_end - vma->vm_start); | |
4315 | free_page((unsigned long)buf); | |
4316 | } | |
4317 | } | |
51a07e50 | 4318 | up_read(&mm->mmap_sem); |
03252919 | 4319 | } |
3ee1afa3 | 4320 | |
662bbcb2 | 4321 | #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP) |
9ec23531 | 4322 | void __might_fault(const char *file, int line) |
3ee1afa3 | 4323 | { |
95156f00 PZ |
4324 | /* |
4325 | * Some code (nfs/sunrpc) uses socket ops on kernel memory while | |
4326 | * holding the mmap_sem, this is safe because kernel memory doesn't | |
4327 | * get paged out, therefore we'll never actually fault, and the | |
4328 | * below annotations will generate false positives. | |
4329 | */ | |
db68ce10 | 4330 | if (uaccess_kernel()) |
95156f00 | 4331 | return; |
9ec23531 | 4332 | if (pagefault_disabled()) |
662bbcb2 | 4333 | return; |
9ec23531 DH |
4334 | __might_sleep(file, line, 0); |
4335 | #if defined(CONFIG_DEBUG_ATOMIC_SLEEP) | |
662bbcb2 | 4336 | if (current->mm) |
3ee1afa3 | 4337 | might_lock_read(¤t->mm->mmap_sem); |
9ec23531 | 4338 | #endif |
3ee1afa3 | 4339 | } |
9ec23531 | 4340 | EXPORT_SYMBOL(__might_fault); |
3ee1afa3 | 4341 | #endif |
47ad8475 AA |
4342 | |
4343 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) | |
4344 | static void clear_gigantic_page(struct page *page, | |
4345 | unsigned long addr, | |
4346 | unsigned int pages_per_huge_page) | |
4347 | { | |
4348 | int i; | |
4349 | struct page *p = page; | |
4350 | ||
4351 | might_sleep(); | |
4352 | for (i = 0; i < pages_per_huge_page; | |
4353 | i++, p = mem_map_next(p, page, i)) { | |
4354 | cond_resched(); | |
4355 | clear_user_highpage(p, addr + i * PAGE_SIZE); | |
4356 | } | |
4357 | } | |
4358 | void clear_huge_page(struct page *page, | |
4359 | unsigned long addr, unsigned int pages_per_huge_page) | |
4360 | { | |
4361 | int i; | |
4362 | ||
4363 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { | |
4364 | clear_gigantic_page(page, addr, pages_per_huge_page); | |
4365 | return; | |
4366 | } | |
4367 | ||
4368 | might_sleep(); | |
4369 | for (i = 0; i < pages_per_huge_page; i++) { | |
4370 | cond_resched(); | |
4371 | clear_user_highpage(page + i, addr + i * PAGE_SIZE); | |
4372 | } | |
4373 | } | |
4374 | ||
4375 | static void copy_user_gigantic_page(struct page *dst, struct page *src, | |
4376 | unsigned long addr, | |
4377 | struct vm_area_struct *vma, | |
4378 | unsigned int pages_per_huge_page) | |
4379 | { | |
4380 | int i; | |
4381 | struct page *dst_base = dst; | |
4382 | struct page *src_base = src; | |
4383 | ||
4384 | for (i = 0; i < pages_per_huge_page; ) { | |
4385 | cond_resched(); | |
4386 | copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma); | |
4387 | ||
4388 | i++; | |
4389 | dst = mem_map_next(dst, dst_base, i); | |
4390 | src = mem_map_next(src, src_base, i); | |
4391 | } | |
4392 | } | |
4393 | ||
4394 | void copy_user_huge_page(struct page *dst, struct page *src, | |
4395 | unsigned long addr, struct vm_area_struct *vma, | |
4396 | unsigned int pages_per_huge_page) | |
4397 | { | |
4398 | int i; | |
4399 | ||
4400 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { | |
4401 | copy_user_gigantic_page(dst, src, addr, vma, | |
4402 | pages_per_huge_page); | |
4403 | return; | |
4404 | } | |
4405 | ||
4406 | might_sleep(); | |
4407 | for (i = 0; i < pages_per_huge_page; i++) { | |
4408 | cond_resched(); | |
4409 | copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE, vma); | |
4410 | } | |
4411 | } | |
fa4d75c1 MK |
4412 | |
4413 | long copy_huge_page_from_user(struct page *dst_page, | |
4414 | const void __user *usr_src, | |
810a56b9 MK |
4415 | unsigned int pages_per_huge_page, |
4416 | bool allow_pagefault) | |
fa4d75c1 MK |
4417 | { |
4418 | void *src = (void *)usr_src; | |
4419 | void *page_kaddr; | |
4420 | unsigned long i, rc = 0; | |
4421 | unsigned long ret_val = pages_per_huge_page * PAGE_SIZE; | |
4422 | ||
4423 | for (i = 0; i < pages_per_huge_page; i++) { | |
810a56b9 MK |
4424 | if (allow_pagefault) |
4425 | page_kaddr = kmap(dst_page + i); | |
4426 | else | |
4427 | page_kaddr = kmap_atomic(dst_page + i); | |
fa4d75c1 MK |
4428 | rc = copy_from_user(page_kaddr, |
4429 | (const void __user *)(src + i * PAGE_SIZE), | |
4430 | PAGE_SIZE); | |
810a56b9 MK |
4431 | if (allow_pagefault) |
4432 | kunmap(dst_page + i); | |
4433 | else | |
4434 | kunmap_atomic(page_kaddr); | |
fa4d75c1 MK |
4435 | |
4436 | ret_val -= (PAGE_SIZE - rc); | |
4437 | if (rc) | |
4438 | break; | |
4439 | ||
4440 | cond_resched(); | |
4441 | } | |
4442 | return ret_val; | |
4443 | } | |
47ad8475 | 4444 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */ |
49076ec2 | 4445 | |
40b64acd | 4446 | #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS |
b35f1819 KS |
4447 | |
4448 | static struct kmem_cache *page_ptl_cachep; | |
4449 | ||
4450 | void __init ptlock_cache_init(void) | |
4451 | { | |
4452 | page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0, | |
4453 | SLAB_PANIC, NULL); | |
4454 | } | |
4455 | ||
539edb58 | 4456 | bool ptlock_alloc(struct page *page) |
49076ec2 KS |
4457 | { |
4458 | spinlock_t *ptl; | |
4459 | ||
b35f1819 | 4460 | ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL); |
49076ec2 KS |
4461 | if (!ptl) |
4462 | return false; | |
539edb58 | 4463 | page->ptl = ptl; |
49076ec2 KS |
4464 | return true; |
4465 | } | |
4466 | ||
539edb58 | 4467 | void ptlock_free(struct page *page) |
49076ec2 | 4468 | { |
b35f1819 | 4469 | kmem_cache_free(page_ptl_cachep, page->ptl); |
49076ec2 KS |
4470 | } |
4471 | #endif |