Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/memory.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
5 | */ | |
6 | ||
7 | /* | |
8 | * demand-loading started 01.12.91 - seems it is high on the list of | |
9 | * things wanted, and it should be easy to implement. - Linus | |
10 | */ | |
11 | ||
12 | /* | |
13 | * Ok, demand-loading was easy, shared pages a little bit tricker. Shared | |
14 | * pages started 02.12.91, seems to work. - Linus. | |
15 | * | |
16 | * Tested sharing by executing about 30 /bin/sh: under the old kernel it | |
17 | * would have taken more than the 6M I have free, but it worked well as | |
18 | * far as I could see. | |
19 | * | |
20 | * Also corrected some "invalidate()"s - I wasn't doing enough of them. | |
21 | */ | |
22 | ||
23 | /* | |
24 | * Real VM (paging to/from disk) started 18.12.91. Much more work and | |
25 | * thought has to go into this. Oh, well.. | |
26 | * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why. | |
27 | * Found it. Everything seems to work now. | |
28 | * 20.12.91 - Ok, making the swap-device changeable like the root. | |
29 | */ | |
30 | ||
31 | /* | |
32 | * 05.04.94 - Multi-page memory management added for v1.1. | |
166f61b9 | 33 | * Idea by Alex Bligh (alex@cconcepts.co.uk) |
1da177e4 LT |
34 | * |
35 | * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG | |
36 | * (Gerhard.Wichert@pdb.siemens.de) | |
37 | * | |
38 | * Aug/Sep 2004 Changed to four level page tables (Andi Kleen) | |
39 | */ | |
40 | ||
41 | #include <linux/kernel_stat.h> | |
42 | #include <linux/mm.h> | |
6e84f315 | 43 | #include <linux/sched/mm.h> |
f7ccbae4 | 44 | #include <linux/sched/coredump.h> |
6a3827d7 | 45 | #include <linux/sched/numa_balancing.h> |
29930025 | 46 | #include <linux/sched/task.h> |
1da177e4 LT |
47 | #include <linux/hugetlb.h> |
48 | #include <linux/mman.h> | |
49 | #include <linux/swap.h> | |
50 | #include <linux/highmem.h> | |
51 | #include <linux/pagemap.h> | |
5042db43 | 52 | #include <linux/memremap.h> |
9a840895 | 53 | #include <linux/ksm.h> |
1da177e4 | 54 | #include <linux/rmap.h> |
b95f1b31 | 55 | #include <linux/export.h> |
0ff92245 | 56 | #include <linux/delayacct.h> |
1da177e4 | 57 | #include <linux/init.h> |
01c8f1c4 | 58 | #include <linux/pfn_t.h> |
edc79b2a | 59 | #include <linux/writeback.h> |
8a9f3ccd | 60 | #include <linux/memcontrol.h> |
cddb8a5c | 61 | #include <linux/mmu_notifier.h> |
3dc14741 HD |
62 | #include <linux/kallsyms.h> |
63 | #include <linux/swapops.h> | |
64 | #include <linux/elf.h> | |
5a0e3ad6 | 65 | #include <linux/gfp.h> |
4daae3b4 | 66 | #include <linux/migrate.h> |
2fbc57c5 | 67 | #include <linux/string.h> |
0abdd7a8 | 68 | #include <linux/dma-debug.h> |
1592eef0 | 69 | #include <linux/debugfs.h> |
6b251fc9 | 70 | #include <linux/userfaultfd_k.h> |
bc2466e4 | 71 | #include <linux/dax.h> |
6b31d595 | 72 | #include <linux/oom.h> |
1da177e4 | 73 | |
6952b61d | 74 | #include <asm/io.h> |
33a709b2 | 75 | #include <asm/mmu_context.h> |
1da177e4 | 76 | #include <asm/pgalloc.h> |
7c0f6ba6 | 77 | #include <linux/uaccess.h> |
1da177e4 LT |
78 | #include <asm/tlb.h> |
79 | #include <asm/tlbflush.h> | |
80 | #include <asm/pgtable.h> | |
81 | ||
42b77728 JB |
82 | #include "internal.h" |
83 | ||
90572890 PZ |
84 | #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS |
85 | #warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid. | |
75980e97 PZ |
86 | #endif |
87 | ||
d41dee36 | 88 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
89 | /* use the per-pgdat data instead for discontigmem - mbligh */ |
90 | unsigned long max_mapnr; | |
1da177e4 | 91 | EXPORT_SYMBOL(max_mapnr); |
166f61b9 TH |
92 | |
93 | struct page *mem_map; | |
1da177e4 LT |
94 | EXPORT_SYMBOL(mem_map); |
95 | #endif | |
96 | ||
1da177e4 LT |
97 | /* |
98 | * A number of key systems in x86 including ioremap() rely on the assumption | |
99 | * that high_memory defines the upper bound on direct map memory, then end | |
100 | * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and | |
101 | * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL | |
102 | * and ZONE_HIGHMEM. | |
103 | */ | |
166f61b9 | 104 | void *high_memory; |
1da177e4 | 105 | EXPORT_SYMBOL(high_memory); |
1da177e4 | 106 | |
32a93233 IM |
107 | /* |
108 | * Randomize the address space (stacks, mmaps, brk, etc.). | |
109 | * | |
110 | * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization, | |
111 | * as ancient (libc5 based) binaries can segfault. ) | |
112 | */ | |
113 | int randomize_va_space __read_mostly = | |
114 | #ifdef CONFIG_COMPAT_BRK | |
115 | 1; | |
116 | #else | |
117 | 2; | |
118 | #endif | |
a62eaf15 AK |
119 | |
120 | static int __init disable_randmaps(char *s) | |
121 | { | |
122 | randomize_va_space = 0; | |
9b41046c | 123 | return 1; |
a62eaf15 AK |
124 | } |
125 | __setup("norandmaps", disable_randmaps); | |
126 | ||
62eede62 | 127 | unsigned long zero_pfn __read_mostly; |
0b70068e AB |
128 | EXPORT_SYMBOL(zero_pfn); |
129 | ||
166f61b9 TH |
130 | unsigned long highest_memmap_pfn __read_mostly; |
131 | ||
a13ea5b7 HD |
132 | /* |
133 | * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init() | |
134 | */ | |
135 | static int __init init_zero_pfn(void) | |
136 | { | |
137 | zero_pfn = page_to_pfn(ZERO_PAGE(0)); | |
138 | return 0; | |
139 | } | |
140 | core_initcall(init_zero_pfn); | |
a62eaf15 | 141 | |
d559db08 | 142 | |
34e55232 KH |
143 | #if defined(SPLIT_RSS_COUNTING) |
144 | ||
ea48cf78 | 145 | void sync_mm_rss(struct mm_struct *mm) |
34e55232 KH |
146 | { |
147 | int i; | |
148 | ||
149 | for (i = 0; i < NR_MM_COUNTERS; i++) { | |
05af2e10 DR |
150 | if (current->rss_stat.count[i]) { |
151 | add_mm_counter(mm, i, current->rss_stat.count[i]); | |
152 | current->rss_stat.count[i] = 0; | |
34e55232 KH |
153 | } |
154 | } | |
05af2e10 | 155 | current->rss_stat.events = 0; |
34e55232 KH |
156 | } |
157 | ||
158 | static void add_mm_counter_fast(struct mm_struct *mm, int member, int val) | |
159 | { | |
160 | struct task_struct *task = current; | |
161 | ||
162 | if (likely(task->mm == mm)) | |
163 | task->rss_stat.count[member] += val; | |
164 | else | |
165 | add_mm_counter(mm, member, val); | |
166 | } | |
167 | #define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1) | |
168 | #define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1) | |
169 | ||
170 | /* sync counter once per 64 page faults */ | |
171 | #define TASK_RSS_EVENTS_THRESH (64) | |
172 | static void check_sync_rss_stat(struct task_struct *task) | |
173 | { | |
174 | if (unlikely(task != current)) | |
175 | return; | |
176 | if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH)) | |
ea48cf78 | 177 | sync_mm_rss(task->mm); |
34e55232 | 178 | } |
9547d01b | 179 | #else /* SPLIT_RSS_COUNTING */ |
34e55232 KH |
180 | |
181 | #define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member) | |
182 | #define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member) | |
183 | ||
184 | static void check_sync_rss_stat(struct task_struct *task) | |
185 | { | |
186 | } | |
187 | ||
9547d01b PZ |
188 | #endif /* SPLIT_RSS_COUNTING */ |
189 | ||
190 | #ifdef HAVE_GENERIC_MMU_GATHER | |
191 | ||
ca1d6c7d | 192 | static bool tlb_next_batch(struct mmu_gather *tlb) |
9547d01b PZ |
193 | { |
194 | struct mmu_gather_batch *batch; | |
195 | ||
196 | batch = tlb->active; | |
197 | if (batch->next) { | |
198 | tlb->active = batch->next; | |
ca1d6c7d | 199 | return true; |
9547d01b PZ |
200 | } |
201 | ||
53a59fc6 | 202 | if (tlb->batch_count == MAX_GATHER_BATCH_COUNT) |
ca1d6c7d | 203 | return false; |
53a59fc6 | 204 | |
9547d01b PZ |
205 | batch = (void *)__get_free_pages(GFP_NOWAIT | __GFP_NOWARN, 0); |
206 | if (!batch) | |
ca1d6c7d | 207 | return false; |
9547d01b | 208 | |
53a59fc6 | 209 | tlb->batch_count++; |
9547d01b PZ |
210 | batch->next = NULL; |
211 | batch->nr = 0; | |
212 | batch->max = MAX_GATHER_BATCH; | |
213 | ||
214 | tlb->active->next = batch; | |
215 | tlb->active = batch; | |
216 | ||
ca1d6c7d | 217 | return true; |
9547d01b PZ |
218 | } |
219 | ||
56236a59 MK |
220 | void arch_tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, |
221 | unsigned long start, unsigned long end) | |
9547d01b PZ |
222 | { |
223 | tlb->mm = mm; | |
224 | ||
2b047252 LT |
225 | /* Is it from 0 to ~0? */ |
226 | tlb->fullmm = !(start | (end+1)); | |
1de14c3c | 227 | tlb->need_flush_all = 0; |
9547d01b PZ |
228 | tlb->local.next = NULL; |
229 | tlb->local.nr = 0; | |
230 | tlb->local.max = ARRAY_SIZE(tlb->__pages); | |
231 | tlb->active = &tlb->local; | |
53a59fc6 | 232 | tlb->batch_count = 0; |
9547d01b PZ |
233 | |
234 | #ifdef CONFIG_HAVE_RCU_TABLE_FREE | |
235 | tlb->batch = NULL; | |
236 | #endif | |
e77b0852 | 237 | tlb->page_size = 0; |
fb7332a9 WD |
238 | |
239 | __tlb_reset_range(tlb); | |
9547d01b PZ |
240 | } |
241 | ||
1cf35d47 | 242 | static void tlb_flush_mmu_tlbonly(struct mmu_gather *tlb) |
9547d01b | 243 | { |
721c21c1 WD |
244 | if (!tlb->end) |
245 | return; | |
246 | ||
9547d01b | 247 | tlb_flush(tlb); |
34ee645e | 248 | mmu_notifier_invalidate_range(tlb->mm, tlb->start, tlb->end); |
9547d01b PZ |
249 | #ifdef CONFIG_HAVE_RCU_TABLE_FREE |
250 | tlb_table_flush(tlb); | |
34e55232 | 251 | #endif |
fb7332a9 | 252 | __tlb_reset_range(tlb); |
1cf35d47 LT |
253 | } |
254 | ||
255 | static void tlb_flush_mmu_free(struct mmu_gather *tlb) | |
256 | { | |
257 | struct mmu_gather_batch *batch; | |
34e55232 | 258 | |
721c21c1 | 259 | for (batch = &tlb->local; batch && batch->nr; batch = batch->next) { |
9547d01b PZ |
260 | free_pages_and_swap_cache(batch->pages, batch->nr); |
261 | batch->nr = 0; | |
262 | } | |
263 | tlb->active = &tlb->local; | |
264 | } | |
265 | ||
1cf35d47 LT |
266 | void tlb_flush_mmu(struct mmu_gather *tlb) |
267 | { | |
1cf35d47 LT |
268 | tlb_flush_mmu_tlbonly(tlb); |
269 | tlb_flush_mmu_free(tlb); | |
270 | } | |
271 | ||
9547d01b PZ |
272 | /* tlb_finish_mmu |
273 | * Called at the end of the shootdown operation to free up any resources | |
274 | * that were required. | |
275 | */ | |
56236a59 | 276 | void arch_tlb_finish_mmu(struct mmu_gather *tlb, |
99baac21 | 277 | unsigned long start, unsigned long end, bool force) |
9547d01b PZ |
278 | { |
279 | struct mmu_gather_batch *batch, *next; | |
280 | ||
99baac21 MK |
281 | if (force) |
282 | __tlb_adjust_range(tlb, start, end - start); | |
283 | ||
9547d01b PZ |
284 | tlb_flush_mmu(tlb); |
285 | ||
286 | /* keep the page table cache within bounds */ | |
287 | check_pgt_cache(); | |
288 | ||
289 | for (batch = tlb->local.next; batch; batch = next) { | |
290 | next = batch->next; | |
291 | free_pages((unsigned long)batch, 0); | |
292 | } | |
293 | tlb->local.next = NULL; | |
294 | } | |
295 | ||
296 | /* __tlb_remove_page | |
297 | * Must perform the equivalent to __free_pte(pte_get_and_clear(ptep)), while | |
298 | * handling the additional races in SMP caused by other CPUs caching valid | |
299 | * mappings in their TLBs. Returns the number of free page slots left. | |
300 | * When out of page slots we must call tlb_flush_mmu(). | |
e9d55e15 | 301 | *returns true if the caller should flush. |
9547d01b | 302 | */ |
e77b0852 | 303 | bool __tlb_remove_page_size(struct mmu_gather *tlb, struct page *page, int page_size) |
9547d01b PZ |
304 | { |
305 | struct mmu_gather_batch *batch; | |
306 | ||
fb7332a9 | 307 | VM_BUG_ON(!tlb->end); |
692a68c1 | 308 | VM_WARN_ON(tlb->page_size != page_size); |
e77b0852 | 309 | |
9547d01b | 310 | batch = tlb->active; |
692a68c1 AK |
311 | /* |
312 | * Add the page and check if we are full. If so | |
313 | * force a flush. | |
314 | */ | |
315 | batch->pages[batch->nr++] = page; | |
9547d01b PZ |
316 | if (batch->nr == batch->max) { |
317 | if (!tlb_next_batch(tlb)) | |
e9d55e15 | 318 | return true; |
0b43c3aa | 319 | batch = tlb->active; |
9547d01b | 320 | } |
309381fe | 321 | VM_BUG_ON_PAGE(batch->nr > batch->max, page); |
9547d01b | 322 | |
e9d55e15 | 323 | return false; |
9547d01b PZ |
324 | } |
325 | ||
326 | #endif /* HAVE_GENERIC_MMU_GATHER */ | |
327 | ||
26723911 PZ |
328 | #ifdef CONFIG_HAVE_RCU_TABLE_FREE |
329 | ||
330 | /* | |
331 | * See the comment near struct mmu_table_batch. | |
332 | */ | |
333 | ||
334 | static void tlb_remove_table_smp_sync(void *arg) | |
335 | { | |
336 | /* Simply deliver the interrupt */ | |
337 | } | |
338 | ||
339 | static void tlb_remove_table_one(void *table) | |
340 | { | |
341 | /* | |
342 | * This isn't an RCU grace period and hence the page-tables cannot be | |
343 | * assumed to be actually RCU-freed. | |
344 | * | |
345 | * It is however sufficient for software page-table walkers that rely on | |
346 | * IRQ disabling. See the comment near struct mmu_table_batch. | |
347 | */ | |
348 | smp_call_function(tlb_remove_table_smp_sync, NULL, 1); | |
349 | __tlb_remove_table(table); | |
350 | } | |
351 | ||
352 | static void tlb_remove_table_rcu(struct rcu_head *head) | |
353 | { | |
354 | struct mmu_table_batch *batch; | |
355 | int i; | |
356 | ||
357 | batch = container_of(head, struct mmu_table_batch, rcu); | |
358 | ||
359 | for (i = 0; i < batch->nr; i++) | |
360 | __tlb_remove_table(batch->tables[i]); | |
361 | ||
362 | free_page((unsigned long)batch); | |
363 | } | |
364 | ||
365 | void tlb_table_flush(struct mmu_gather *tlb) | |
366 | { | |
367 | struct mmu_table_batch **batch = &tlb->batch; | |
368 | ||
369 | if (*batch) { | |
370 | call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu); | |
371 | *batch = NULL; | |
372 | } | |
373 | } | |
374 | ||
375 | void tlb_remove_table(struct mmu_gather *tlb, void *table) | |
376 | { | |
377 | struct mmu_table_batch **batch = &tlb->batch; | |
378 | ||
26723911 PZ |
379 | /* |
380 | * When there's less then two users of this mm there cannot be a | |
381 | * concurrent page-table walk. | |
382 | */ | |
383 | if (atomic_read(&tlb->mm->mm_users) < 2) { | |
384 | __tlb_remove_table(table); | |
385 | return; | |
386 | } | |
387 | ||
388 | if (*batch == NULL) { | |
389 | *batch = (struct mmu_table_batch *)__get_free_page(GFP_NOWAIT | __GFP_NOWARN); | |
390 | if (*batch == NULL) { | |
391 | tlb_remove_table_one(table); | |
392 | return; | |
393 | } | |
394 | (*batch)->nr = 0; | |
395 | } | |
396 | (*batch)->tables[(*batch)->nr++] = table; | |
397 | if ((*batch)->nr == MAX_TABLE_BATCH) | |
398 | tlb_table_flush(tlb); | |
399 | } | |
400 | ||
9547d01b | 401 | #endif /* CONFIG_HAVE_RCU_TABLE_FREE */ |
26723911 | 402 | |
56236a59 MK |
403 | /* tlb_gather_mmu |
404 | * Called to initialize an (on-stack) mmu_gather structure for page-table | |
405 | * tear-down from @mm. The @fullmm argument is used when @mm is without | |
406 | * users and we're going to destroy the full address space (exit/execve). | |
407 | */ | |
408 | void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, | |
409 | unsigned long start, unsigned long end) | |
410 | { | |
411 | arch_tlb_gather_mmu(tlb, mm, start, end); | |
99baac21 | 412 | inc_tlb_flush_pending(tlb->mm); |
56236a59 MK |
413 | } |
414 | ||
415 | void tlb_finish_mmu(struct mmu_gather *tlb, | |
416 | unsigned long start, unsigned long end) | |
417 | { | |
99baac21 MK |
418 | /* |
419 | * If there are parallel threads are doing PTE changes on same range | |
420 | * under non-exclusive lock(e.g., mmap_sem read-side) but defer TLB | |
421 | * flush by batching, a thread has stable TLB entry can fail to flush | |
422 | * the TLB by observing pte_none|!pte_dirty, for example so flush TLB | |
423 | * forcefully if we detect parallel PTE batching threads. | |
424 | */ | |
425 | bool force = mm_tlb_flush_nested(tlb->mm); | |
426 | ||
427 | arch_tlb_finish_mmu(tlb, start, end, force); | |
428 | dec_tlb_flush_pending(tlb->mm); | |
56236a59 MK |
429 | } |
430 | ||
1da177e4 LT |
431 | /* |
432 | * Note: this doesn't free the actual pages themselves. That | |
433 | * has been handled earlier when unmapping all the memory regions. | |
434 | */ | |
9e1b32ca BH |
435 | static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd, |
436 | unsigned long addr) | |
1da177e4 | 437 | { |
2f569afd | 438 | pgtable_t token = pmd_pgtable(*pmd); |
e0da382c | 439 | pmd_clear(pmd); |
9e1b32ca | 440 | pte_free_tlb(tlb, token, addr); |
c4812909 | 441 | mm_dec_nr_ptes(tlb->mm); |
1da177e4 LT |
442 | } |
443 | ||
e0da382c HD |
444 | static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud, |
445 | unsigned long addr, unsigned long end, | |
446 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
447 | { |
448 | pmd_t *pmd; | |
449 | unsigned long next; | |
e0da382c | 450 | unsigned long start; |
1da177e4 | 451 | |
e0da382c | 452 | start = addr; |
1da177e4 | 453 | pmd = pmd_offset(pud, addr); |
1da177e4 LT |
454 | do { |
455 | next = pmd_addr_end(addr, end); | |
456 | if (pmd_none_or_clear_bad(pmd)) | |
457 | continue; | |
9e1b32ca | 458 | free_pte_range(tlb, pmd, addr); |
1da177e4 LT |
459 | } while (pmd++, addr = next, addr != end); |
460 | ||
e0da382c HD |
461 | start &= PUD_MASK; |
462 | if (start < floor) | |
463 | return; | |
464 | if (ceiling) { | |
465 | ceiling &= PUD_MASK; | |
466 | if (!ceiling) | |
467 | return; | |
1da177e4 | 468 | } |
e0da382c HD |
469 | if (end - 1 > ceiling - 1) |
470 | return; | |
471 | ||
472 | pmd = pmd_offset(pud, start); | |
473 | pud_clear(pud); | |
9e1b32ca | 474 | pmd_free_tlb(tlb, pmd, start); |
dc6c9a35 | 475 | mm_dec_nr_pmds(tlb->mm); |
1da177e4 LT |
476 | } |
477 | ||
c2febafc | 478 | static inline void free_pud_range(struct mmu_gather *tlb, p4d_t *p4d, |
e0da382c HD |
479 | unsigned long addr, unsigned long end, |
480 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
481 | { |
482 | pud_t *pud; | |
483 | unsigned long next; | |
e0da382c | 484 | unsigned long start; |
1da177e4 | 485 | |
e0da382c | 486 | start = addr; |
c2febafc | 487 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
488 | do { |
489 | next = pud_addr_end(addr, end); | |
490 | if (pud_none_or_clear_bad(pud)) | |
491 | continue; | |
e0da382c | 492 | free_pmd_range(tlb, pud, addr, next, floor, ceiling); |
1da177e4 LT |
493 | } while (pud++, addr = next, addr != end); |
494 | ||
c2febafc KS |
495 | start &= P4D_MASK; |
496 | if (start < floor) | |
497 | return; | |
498 | if (ceiling) { | |
499 | ceiling &= P4D_MASK; | |
500 | if (!ceiling) | |
501 | return; | |
502 | } | |
503 | if (end - 1 > ceiling - 1) | |
504 | return; | |
505 | ||
506 | pud = pud_offset(p4d, start); | |
507 | p4d_clear(p4d); | |
508 | pud_free_tlb(tlb, pud, start); | |
b4e98d9a | 509 | mm_dec_nr_puds(tlb->mm); |
c2febafc KS |
510 | } |
511 | ||
512 | static inline void free_p4d_range(struct mmu_gather *tlb, pgd_t *pgd, | |
513 | unsigned long addr, unsigned long end, | |
514 | unsigned long floor, unsigned long ceiling) | |
515 | { | |
516 | p4d_t *p4d; | |
517 | unsigned long next; | |
518 | unsigned long start; | |
519 | ||
520 | start = addr; | |
521 | p4d = p4d_offset(pgd, addr); | |
522 | do { | |
523 | next = p4d_addr_end(addr, end); | |
524 | if (p4d_none_or_clear_bad(p4d)) | |
525 | continue; | |
526 | free_pud_range(tlb, p4d, addr, next, floor, ceiling); | |
527 | } while (p4d++, addr = next, addr != end); | |
528 | ||
e0da382c HD |
529 | start &= PGDIR_MASK; |
530 | if (start < floor) | |
531 | return; | |
532 | if (ceiling) { | |
533 | ceiling &= PGDIR_MASK; | |
534 | if (!ceiling) | |
535 | return; | |
1da177e4 | 536 | } |
e0da382c HD |
537 | if (end - 1 > ceiling - 1) |
538 | return; | |
539 | ||
c2febafc | 540 | p4d = p4d_offset(pgd, start); |
e0da382c | 541 | pgd_clear(pgd); |
c2febafc | 542 | p4d_free_tlb(tlb, p4d, start); |
1da177e4 LT |
543 | } |
544 | ||
545 | /* | |
e0da382c | 546 | * This function frees user-level page tables of a process. |
1da177e4 | 547 | */ |
42b77728 | 548 | void free_pgd_range(struct mmu_gather *tlb, |
e0da382c HD |
549 | unsigned long addr, unsigned long end, |
550 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
551 | { |
552 | pgd_t *pgd; | |
553 | unsigned long next; | |
e0da382c HD |
554 | |
555 | /* | |
556 | * The next few lines have given us lots of grief... | |
557 | * | |
558 | * Why are we testing PMD* at this top level? Because often | |
559 | * there will be no work to do at all, and we'd prefer not to | |
560 | * go all the way down to the bottom just to discover that. | |
561 | * | |
562 | * Why all these "- 1"s? Because 0 represents both the bottom | |
563 | * of the address space and the top of it (using -1 for the | |
564 | * top wouldn't help much: the masks would do the wrong thing). | |
565 | * The rule is that addr 0 and floor 0 refer to the bottom of | |
566 | * the address space, but end 0 and ceiling 0 refer to the top | |
567 | * Comparisons need to use "end - 1" and "ceiling - 1" (though | |
568 | * that end 0 case should be mythical). | |
569 | * | |
570 | * Wherever addr is brought up or ceiling brought down, we must | |
571 | * be careful to reject "the opposite 0" before it confuses the | |
572 | * subsequent tests. But what about where end is brought down | |
573 | * by PMD_SIZE below? no, end can't go down to 0 there. | |
574 | * | |
575 | * Whereas we round start (addr) and ceiling down, by different | |
576 | * masks at different levels, in order to test whether a table | |
577 | * now has no other vmas using it, so can be freed, we don't | |
578 | * bother to round floor or end up - the tests don't need that. | |
579 | */ | |
1da177e4 | 580 | |
e0da382c HD |
581 | addr &= PMD_MASK; |
582 | if (addr < floor) { | |
583 | addr += PMD_SIZE; | |
584 | if (!addr) | |
585 | return; | |
586 | } | |
587 | if (ceiling) { | |
588 | ceiling &= PMD_MASK; | |
589 | if (!ceiling) | |
590 | return; | |
591 | } | |
592 | if (end - 1 > ceiling - 1) | |
593 | end -= PMD_SIZE; | |
594 | if (addr > end - 1) | |
595 | return; | |
07e32661 AK |
596 | /* |
597 | * We add page table cache pages with PAGE_SIZE, | |
598 | * (see pte_free_tlb()), flush the tlb if we need | |
599 | */ | |
600 | tlb_remove_check_page_size_change(tlb, PAGE_SIZE); | |
42b77728 | 601 | pgd = pgd_offset(tlb->mm, addr); |
1da177e4 LT |
602 | do { |
603 | next = pgd_addr_end(addr, end); | |
604 | if (pgd_none_or_clear_bad(pgd)) | |
605 | continue; | |
c2febafc | 606 | free_p4d_range(tlb, pgd, addr, next, floor, ceiling); |
1da177e4 | 607 | } while (pgd++, addr = next, addr != end); |
e0da382c HD |
608 | } |
609 | ||
42b77728 | 610 | void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma, |
3bf5ee95 | 611 | unsigned long floor, unsigned long ceiling) |
e0da382c HD |
612 | { |
613 | while (vma) { | |
614 | struct vm_area_struct *next = vma->vm_next; | |
615 | unsigned long addr = vma->vm_start; | |
616 | ||
8f4f8c16 | 617 | /* |
25d9e2d1 | 618 | * Hide vma from rmap and truncate_pagecache before freeing |
619 | * pgtables | |
8f4f8c16 | 620 | */ |
5beb4930 | 621 | unlink_anon_vmas(vma); |
8f4f8c16 HD |
622 | unlink_file_vma(vma); |
623 | ||
9da61aef | 624 | if (is_vm_hugetlb_page(vma)) { |
3bf5ee95 | 625 | hugetlb_free_pgd_range(tlb, addr, vma->vm_end, |
166f61b9 | 626 | floor, next ? next->vm_start : ceiling); |
3bf5ee95 HD |
627 | } else { |
628 | /* | |
629 | * Optimization: gather nearby vmas into one call down | |
630 | */ | |
631 | while (next && next->vm_start <= vma->vm_end + PMD_SIZE | |
4866920b | 632 | && !is_vm_hugetlb_page(next)) { |
3bf5ee95 HD |
633 | vma = next; |
634 | next = vma->vm_next; | |
5beb4930 | 635 | unlink_anon_vmas(vma); |
8f4f8c16 | 636 | unlink_file_vma(vma); |
3bf5ee95 HD |
637 | } |
638 | free_pgd_range(tlb, addr, vma->vm_end, | |
166f61b9 | 639 | floor, next ? next->vm_start : ceiling); |
3bf5ee95 | 640 | } |
e0da382c HD |
641 | vma = next; |
642 | } | |
1da177e4 LT |
643 | } |
644 | ||
3ed3a4f0 | 645 | int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address) |
1da177e4 | 646 | { |
c4088ebd | 647 | spinlock_t *ptl; |
2f569afd | 648 | pgtable_t new = pte_alloc_one(mm, address); |
1bb3630e HD |
649 | if (!new) |
650 | return -ENOMEM; | |
651 | ||
362a61ad NP |
652 | /* |
653 | * Ensure all pte setup (eg. pte page lock and page clearing) are | |
654 | * visible before the pte is made visible to other CPUs by being | |
655 | * put into page tables. | |
656 | * | |
657 | * The other side of the story is the pointer chasing in the page | |
658 | * table walking code (when walking the page table without locking; | |
659 | * ie. most of the time). Fortunately, these data accesses consist | |
660 | * of a chain of data-dependent loads, meaning most CPUs (alpha | |
661 | * being the notable exception) will already guarantee loads are | |
662 | * seen in-order. See the alpha page table accessors for the | |
663 | * smp_read_barrier_depends() barriers in page table walking code. | |
664 | */ | |
665 | smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */ | |
666 | ||
c4088ebd | 667 | ptl = pmd_lock(mm, pmd); |
8ac1f832 | 668 | if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ |
c4812909 | 669 | mm_inc_nr_ptes(mm); |
1da177e4 | 670 | pmd_populate(mm, pmd, new); |
2f569afd | 671 | new = NULL; |
4b471e88 | 672 | } |
c4088ebd | 673 | spin_unlock(ptl); |
2f569afd MS |
674 | if (new) |
675 | pte_free(mm, new); | |
1bb3630e | 676 | return 0; |
1da177e4 LT |
677 | } |
678 | ||
1bb3630e | 679 | int __pte_alloc_kernel(pmd_t *pmd, unsigned long address) |
1da177e4 | 680 | { |
1bb3630e HD |
681 | pte_t *new = pte_alloc_one_kernel(&init_mm, address); |
682 | if (!new) | |
683 | return -ENOMEM; | |
684 | ||
362a61ad NP |
685 | smp_wmb(); /* See comment in __pte_alloc */ |
686 | ||
1bb3630e | 687 | spin_lock(&init_mm.page_table_lock); |
8ac1f832 | 688 | if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ |
1bb3630e | 689 | pmd_populate_kernel(&init_mm, pmd, new); |
2f569afd | 690 | new = NULL; |
4b471e88 | 691 | } |
1bb3630e | 692 | spin_unlock(&init_mm.page_table_lock); |
2f569afd MS |
693 | if (new) |
694 | pte_free_kernel(&init_mm, new); | |
1bb3630e | 695 | return 0; |
1da177e4 LT |
696 | } |
697 | ||
d559db08 KH |
698 | static inline void init_rss_vec(int *rss) |
699 | { | |
700 | memset(rss, 0, sizeof(int) * NR_MM_COUNTERS); | |
701 | } | |
702 | ||
703 | static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss) | |
ae859762 | 704 | { |
d559db08 KH |
705 | int i; |
706 | ||
34e55232 | 707 | if (current->mm == mm) |
05af2e10 | 708 | sync_mm_rss(mm); |
d559db08 KH |
709 | for (i = 0; i < NR_MM_COUNTERS; i++) |
710 | if (rss[i]) | |
711 | add_mm_counter(mm, i, rss[i]); | |
ae859762 HD |
712 | } |
713 | ||
b5810039 | 714 | /* |
6aab341e LT |
715 | * This function is called to print an error when a bad pte |
716 | * is found. For example, we might have a PFN-mapped pte in | |
717 | * a region that doesn't allow it. | |
b5810039 NP |
718 | * |
719 | * The calling function must still handle the error. | |
720 | */ | |
3dc14741 HD |
721 | static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr, |
722 | pte_t pte, struct page *page) | |
b5810039 | 723 | { |
3dc14741 | 724 | pgd_t *pgd = pgd_offset(vma->vm_mm, addr); |
c2febafc KS |
725 | p4d_t *p4d = p4d_offset(pgd, addr); |
726 | pud_t *pud = pud_offset(p4d, addr); | |
3dc14741 HD |
727 | pmd_t *pmd = pmd_offset(pud, addr); |
728 | struct address_space *mapping; | |
729 | pgoff_t index; | |
d936cf9b HD |
730 | static unsigned long resume; |
731 | static unsigned long nr_shown; | |
732 | static unsigned long nr_unshown; | |
733 | ||
734 | /* | |
735 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
736 | * or allow a steady drip of one report per second. | |
737 | */ | |
738 | if (nr_shown == 60) { | |
739 | if (time_before(jiffies, resume)) { | |
740 | nr_unshown++; | |
741 | return; | |
742 | } | |
743 | if (nr_unshown) { | |
1170532b JP |
744 | pr_alert("BUG: Bad page map: %lu messages suppressed\n", |
745 | nr_unshown); | |
d936cf9b HD |
746 | nr_unshown = 0; |
747 | } | |
748 | nr_shown = 0; | |
749 | } | |
750 | if (nr_shown++ == 0) | |
751 | resume = jiffies + 60 * HZ; | |
3dc14741 HD |
752 | |
753 | mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL; | |
754 | index = linear_page_index(vma, addr); | |
755 | ||
1170532b JP |
756 | pr_alert("BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n", |
757 | current->comm, | |
758 | (long long)pte_val(pte), (long long)pmd_val(*pmd)); | |
718a3821 | 759 | if (page) |
f0b791a3 | 760 | dump_page(page, "bad pte"); |
1170532b JP |
761 | pr_alert("addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n", |
762 | (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index); | |
3dc14741 HD |
763 | /* |
764 | * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y | |
765 | */ | |
2682582a KK |
766 | pr_alert("file:%pD fault:%pf mmap:%pf readpage:%pf\n", |
767 | vma->vm_file, | |
768 | vma->vm_ops ? vma->vm_ops->fault : NULL, | |
769 | vma->vm_file ? vma->vm_file->f_op->mmap : NULL, | |
770 | mapping ? mapping->a_ops->readpage : NULL); | |
b5810039 | 771 | dump_stack(); |
373d4d09 | 772 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
b5810039 NP |
773 | } |
774 | ||
ee498ed7 | 775 | /* |
7e675137 | 776 | * vm_normal_page -- This function gets the "struct page" associated with a pte. |
6aab341e | 777 | * |
7e675137 NP |
778 | * "Special" mappings do not wish to be associated with a "struct page" (either |
779 | * it doesn't exist, or it exists but they don't want to touch it). In this | |
780 | * case, NULL is returned here. "Normal" mappings do have a struct page. | |
b379d790 | 781 | * |
7e675137 NP |
782 | * There are 2 broad cases. Firstly, an architecture may define a pte_special() |
783 | * pte bit, in which case this function is trivial. Secondly, an architecture | |
784 | * may not have a spare pte bit, which requires a more complicated scheme, | |
785 | * described below. | |
786 | * | |
787 | * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a | |
788 | * special mapping (even if there are underlying and valid "struct pages"). | |
789 | * COWed pages of a VM_PFNMAP are always normal. | |
6aab341e | 790 | * |
b379d790 JH |
791 | * The way we recognize COWed pages within VM_PFNMAP mappings is through the |
792 | * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit | |
7e675137 NP |
793 | * set, and the vm_pgoff will point to the first PFN mapped: thus every special |
794 | * mapping will always honor the rule | |
6aab341e LT |
795 | * |
796 | * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) | |
797 | * | |
7e675137 NP |
798 | * And for normal mappings this is false. |
799 | * | |
800 | * This restricts such mappings to be a linear translation from virtual address | |
801 | * to pfn. To get around this restriction, we allow arbitrary mappings so long | |
802 | * as the vma is not a COW mapping; in that case, we know that all ptes are | |
803 | * special (because none can have been COWed). | |
b379d790 | 804 | * |
b379d790 | 805 | * |
7e675137 | 806 | * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. |
b379d790 JH |
807 | * |
808 | * VM_MIXEDMAP mappings can likewise contain memory with or without "struct | |
809 | * page" backing, however the difference is that _all_ pages with a struct | |
810 | * page (that is, those where pfn_valid is true) are refcounted and considered | |
811 | * normal pages by the VM. The disadvantage is that pages are refcounted | |
812 | * (which can be slower and simply not an option for some PFNMAP users). The | |
813 | * advantage is that we don't have to follow the strict linearity rule of | |
814 | * PFNMAP mappings in order to support COWable mappings. | |
815 | * | |
ee498ed7 | 816 | */ |
7e675137 NP |
817 | #ifdef __HAVE_ARCH_PTE_SPECIAL |
818 | # define HAVE_PTE_SPECIAL 1 | |
819 | #else | |
820 | # define HAVE_PTE_SPECIAL 0 | |
821 | #endif | |
df6ad698 JG |
822 | struct page *_vm_normal_page(struct vm_area_struct *vma, unsigned long addr, |
823 | pte_t pte, bool with_public_device) | |
ee498ed7 | 824 | { |
22b31eec | 825 | unsigned long pfn = pte_pfn(pte); |
7e675137 NP |
826 | |
827 | if (HAVE_PTE_SPECIAL) { | |
b38af472 | 828 | if (likely(!pte_special(pte))) |
22b31eec | 829 | goto check_pfn; |
667a0a06 DV |
830 | if (vma->vm_ops && vma->vm_ops->find_special_page) |
831 | return vma->vm_ops->find_special_page(vma, addr); | |
a13ea5b7 HD |
832 | if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) |
833 | return NULL; | |
df6ad698 JG |
834 | if (is_zero_pfn(pfn)) |
835 | return NULL; | |
836 | ||
837 | /* | |
838 | * Device public pages are special pages (they are ZONE_DEVICE | |
839 | * pages but different from persistent memory). They behave | |
840 | * allmost like normal pages. The difference is that they are | |
841 | * not on the lru and thus should never be involve with any- | |
842 | * thing that involve lru manipulation (mlock, numa balancing, | |
843 | * ...). | |
844 | * | |
845 | * This is why we still want to return NULL for such page from | |
846 | * vm_normal_page() so that we do not have to special case all | |
847 | * call site of vm_normal_page(). | |
848 | */ | |
7d790d2d | 849 | if (likely(pfn <= highest_memmap_pfn)) { |
df6ad698 JG |
850 | struct page *page = pfn_to_page(pfn); |
851 | ||
852 | if (is_device_public_page(page)) { | |
853 | if (with_public_device) | |
854 | return page; | |
855 | return NULL; | |
856 | } | |
857 | } | |
858 | print_bad_pte(vma, addr, pte, NULL); | |
7e675137 NP |
859 | return NULL; |
860 | } | |
861 | ||
862 | /* !HAVE_PTE_SPECIAL case follows: */ | |
863 | ||
b379d790 JH |
864 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { |
865 | if (vma->vm_flags & VM_MIXEDMAP) { | |
866 | if (!pfn_valid(pfn)) | |
867 | return NULL; | |
868 | goto out; | |
869 | } else { | |
7e675137 NP |
870 | unsigned long off; |
871 | off = (addr - vma->vm_start) >> PAGE_SHIFT; | |
b379d790 JH |
872 | if (pfn == vma->vm_pgoff + off) |
873 | return NULL; | |
874 | if (!is_cow_mapping(vma->vm_flags)) | |
875 | return NULL; | |
876 | } | |
6aab341e LT |
877 | } |
878 | ||
b38af472 HD |
879 | if (is_zero_pfn(pfn)) |
880 | return NULL; | |
22b31eec HD |
881 | check_pfn: |
882 | if (unlikely(pfn > highest_memmap_pfn)) { | |
883 | print_bad_pte(vma, addr, pte, NULL); | |
884 | return NULL; | |
885 | } | |
6aab341e LT |
886 | |
887 | /* | |
7e675137 | 888 | * NOTE! We still have PageReserved() pages in the page tables. |
7e675137 | 889 | * eg. VDSO mappings can cause them to exist. |
6aab341e | 890 | */ |
b379d790 | 891 | out: |
6aab341e | 892 | return pfn_to_page(pfn); |
ee498ed7 HD |
893 | } |
894 | ||
28093f9f GS |
895 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
896 | struct page *vm_normal_page_pmd(struct vm_area_struct *vma, unsigned long addr, | |
897 | pmd_t pmd) | |
898 | { | |
899 | unsigned long pfn = pmd_pfn(pmd); | |
900 | ||
901 | /* | |
902 | * There is no pmd_special() but there may be special pmds, e.g. | |
903 | * in a direct-access (dax) mapping, so let's just replicate the | |
904 | * !HAVE_PTE_SPECIAL case from vm_normal_page() here. | |
905 | */ | |
906 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { | |
907 | if (vma->vm_flags & VM_MIXEDMAP) { | |
908 | if (!pfn_valid(pfn)) | |
909 | return NULL; | |
910 | goto out; | |
911 | } else { | |
912 | unsigned long off; | |
913 | off = (addr - vma->vm_start) >> PAGE_SHIFT; | |
914 | if (pfn == vma->vm_pgoff + off) | |
915 | return NULL; | |
916 | if (!is_cow_mapping(vma->vm_flags)) | |
917 | return NULL; | |
918 | } | |
919 | } | |
920 | ||
921 | if (is_zero_pfn(pfn)) | |
922 | return NULL; | |
923 | if (unlikely(pfn > highest_memmap_pfn)) | |
924 | return NULL; | |
925 | ||
926 | /* | |
927 | * NOTE! We still have PageReserved() pages in the page tables. | |
928 | * eg. VDSO mappings can cause them to exist. | |
929 | */ | |
930 | out: | |
931 | return pfn_to_page(pfn); | |
932 | } | |
933 | #endif | |
934 | ||
1da177e4 LT |
935 | /* |
936 | * copy one vm_area from one task to the other. Assumes the page tables | |
937 | * already present in the new task to be cleared in the whole range | |
938 | * covered by this vma. | |
1da177e4 LT |
939 | */ |
940 | ||
570a335b | 941 | static inline unsigned long |
1da177e4 | 942 | copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
b5810039 | 943 | pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma, |
8c103762 | 944 | unsigned long addr, int *rss) |
1da177e4 | 945 | { |
b5810039 | 946 | unsigned long vm_flags = vma->vm_flags; |
1da177e4 LT |
947 | pte_t pte = *src_pte; |
948 | struct page *page; | |
1da177e4 LT |
949 | |
950 | /* pte contains position in swap or file, so copy. */ | |
951 | if (unlikely(!pte_present(pte))) { | |
0661a336 KS |
952 | swp_entry_t entry = pte_to_swp_entry(pte); |
953 | ||
954 | if (likely(!non_swap_entry(entry))) { | |
955 | if (swap_duplicate(entry) < 0) | |
956 | return entry.val; | |
957 | ||
958 | /* make sure dst_mm is on swapoff's mmlist. */ | |
959 | if (unlikely(list_empty(&dst_mm->mmlist))) { | |
960 | spin_lock(&mmlist_lock); | |
961 | if (list_empty(&dst_mm->mmlist)) | |
962 | list_add(&dst_mm->mmlist, | |
963 | &src_mm->mmlist); | |
964 | spin_unlock(&mmlist_lock); | |
965 | } | |
966 | rss[MM_SWAPENTS]++; | |
967 | } else if (is_migration_entry(entry)) { | |
968 | page = migration_entry_to_page(entry); | |
969 | ||
eca56ff9 | 970 | rss[mm_counter(page)]++; |
0661a336 KS |
971 | |
972 | if (is_write_migration_entry(entry) && | |
973 | is_cow_mapping(vm_flags)) { | |
974 | /* | |
975 | * COW mappings require pages in both | |
976 | * parent and child to be set to read. | |
977 | */ | |
978 | make_migration_entry_read(&entry); | |
979 | pte = swp_entry_to_pte(entry); | |
980 | if (pte_swp_soft_dirty(*src_pte)) | |
981 | pte = pte_swp_mksoft_dirty(pte); | |
982 | set_pte_at(src_mm, addr, src_pte, pte); | |
0697212a | 983 | } |
5042db43 JG |
984 | } else if (is_device_private_entry(entry)) { |
985 | page = device_private_entry_to_page(entry); | |
986 | ||
987 | /* | |
988 | * Update rss count even for unaddressable pages, as | |
989 | * they should treated just like normal pages in this | |
990 | * respect. | |
991 | * | |
992 | * We will likely want to have some new rss counters | |
993 | * for unaddressable pages, at some point. But for now | |
994 | * keep things as they are. | |
995 | */ | |
996 | get_page(page); | |
997 | rss[mm_counter(page)]++; | |
998 | page_dup_rmap(page, false); | |
999 | ||
1000 | /* | |
1001 | * We do not preserve soft-dirty information, because so | |
1002 | * far, checkpoint/restore is the only feature that | |
1003 | * requires that. And checkpoint/restore does not work | |
1004 | * when a device driver is involved (you cannot easily | |
1005 | * save and restore device driver state). | |
1006 | */ | |
1007 | if (is_write_device_private_entry(entry) && | |
1008 | is_cow_mapping(vm_flags)) { | |
1009 | make_device_private_entry_read(&entry); | |
1010 | pte = swp_entry_to_pte(entry); | |
1011 | set_pte_at(src_mm, addr, src_pte, pte); | |
1012 | } | |
1da177e4 | 1013 | } |
ae859762 | 1014 | goto out_set_pte; |
1da177e4 LT |
1015 | } |
1016 | ||
1da177e4 LT |
1017 | /* |
1018 | * If it's a COW mapping, write protect it both | |
1019 | * in the parent and the child | |
1020 | */ | |
67121172 | 1021 | if (is_cow_mapping(vm_flags)) { |
1da177e4 | 1022 | ptep_set_wrprotect(src_mm, addr, src_pte); |
3dc90795 | 1023 | pte = pte_wrprotect(pte); |
1da177e4 LT |
1024 | } |
1025 | ||
1026 | /* | |
1027 | * If it's a shared mapping, mark it clean in | |
1028 | * the child | |
1029 | */ | |
1030 | if (vm_flags & VM_SHARED) | |
1031 | pte = pte_mkclean(pte); | |
1032 | pte = pte_mkold(pte); | |
6aab341e LT |
1033 | |
1034 | page = vm_normal_page(vma, addr, pte); | |
1035 | if (page) { | |
1036 | get_page(page); | |
53f9263b | 1037 | page_dup_rmap(page, false); |
eca56ff9 | 1038 | rss[mm_counter(page)]++; |
df6ad698 JG |
1039 | } else if (pte_devmap(pte)) { |
1040 | page = pte_page(pte); | |
1041 | ||
1042 | /* | |
1043 | * Cache coherent device memory behave like regular page and | |
1044 | * not like persistent memory page. For more informations see | |
1045 | * MEMORY_DEVICE_CACHE_COHERENT in memory_hotplug.h | |
1046 | */ | |
1047 | if (is_device_public_page(page)) { | |
1048 | get_page(page); | |
1049 | page_dup_rmap(page, false); | |
1050 | rss[mm_counter(page)]++; | |
1051 | } | |
6aab341e | 1052 | } |
ae859762 HD |
1053 | |
1054 | out_set_pte: | |
1055 | set_pte_at(dst_mm, addr, dst_pte, pte); | |
570a335b | 1056 | return 0; |
1da177e4 LT |
1057 | } |
1058 | ||
21bda264 | 1059 | static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
71e3aac0 AA |
1060 | pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma, |
1061 | unsigned long addr, unsigned long end) | |
1da177e4 | 1062 | { |
c36987e2 | 1063 | pte_t *orig_src_pte, *orig_dst_pte; |
1da177e4 | 1064 | pte_t *src_pte, *dst_pte; |
c74df32c | 1065 | spinlock_t *src_ptl, *dst_ptl; |
e040f218 | 1066 | int progress = 0; |
d559db08 | 1067 | int rss[NR_MM_COUNTERS]; |
570a335b | 1068 | swp_entry_t entry = (swp_entry_t){0}; |
1da177e4 LT |
1069 | |
1070 | again: | |
d559db08 KH |
1071 | init_rss_vec(rss); |
1072 | ||
c74df32c | 1073 | dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); |
1da177e4 LT |
1074 | if (!dst_pte) |
1075 | return -ENOMEM; | |
ece0e2b6 | 1076 | src_pte = pte_offset_map(src_pmd, addr); |
4c21e2f2 | 1077 | src_ptl = pte_lockptr(src_mm, src_pmd); |
f20dc5f7 | 1078 | spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); |
c36987e2 DN |
1079 | orig_src_pte = src_pte; |
1080 | orig_dst_pte = dst_pte; | |
6606c3e0 | 1081 | arch_enter_lazy_mmu_mode(); |
1da177e4 | 1082 | |
1da177e4 LT |
1083 | do { |
1084 | /* | |
1085 | * We are holding two locks at this point - either of them | |
1086 | * could generate latencies in another task on another CPU. | |
1087 | */ | |
e040f218 HD |
1088 | if (progress >= 32) { |
1089 | progress = 0; | |
1090 | if (need_resched() || | |
95c354fe | 1091 | spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) |
e040f218 HD |
1092 | break; |
1093 | } | |
1da177e4 LT |
1094 | if (pte_none(*src_pte)) { |
1095 | progress++; | |
1096 | continue; | |
1097 | } | |
570a335b HD |
1098 | entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, |
1099 | vma, addr, rss); | |
1100 | if (entry.val) | |
1101 | break; | |
1da177e4 LT |
1102 | progress += 8; |
1103 | } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); | |
1da177e4 | 1104 | |
6606c3e0 | 1105 | arch_leave_lazy_mmu_mode(); |
c74df32c | 1106 | spin_unlock(src_ptl); |
ece0e2b6 | 1107 | pte_unmap(orig_src_pte); |
d559db08 | 1108 | add_mm_rss_vec(dst_mm, rss); |
c36987e2 | 1109 | pte_unmap_unlock(orig_dst_pte, dst_ptl); |
c74df32c | 1110 | cond_resched(); |
570a335b HD |
1111 | |
1112 | if (entry.val) { | |
1113 | if (add_swap_count_continuation(entry, GFP_KERNEL) < 0) | |
1114 | return -ENOMEM; | |
1115 | progress = 0; | |
1116 | } | |
1da177e4 LT |
1117 | if (addr != end) |
1118 | goto again; | |
1119 | return 0; | |
1120 | } | |
1121 | ||
1122 | static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
1123 | pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma, | |
1124 | unsigned long addr, unsigned long end) | |
1125 | { | |
1126 | pmd_t *src_pmd, *dst_pmd; | |
1127 | unsigned long next; | |
1128 | ||
1129 | dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); | |
1130 | if (!dst_pmd) | |
1131 | return -ENOMEM; | |
1132 | src_pmd = pmd_offset(src_pud, addr); | |
1133 | do { | |
1134 | next = pmd_addr_end(addr, end); | |
84c3fc4e ZY |
1135 | if (is_swap_pmd(*src_pmd) || pmd_trans_huge(*src_pmd) |
1136 | || pmd_devmap(*src_pmd)) { | |
71e3aac0 | 1137 | int err; |
a00cc7d9 | 1138 | VM_BUG_ON_VMA(next-addr != HPAGE_PMD_SIZE, vma); |
71e3aac0 AA |
1139 | err = copy_huge_pmd(dst_mm, src_mm, |
1140 | dst_pmd, src_pmd, addr, vma); | |
1141 | if (err == -ENOMEM) | |
1142 | return -ENOMEM; | |
1143 | if (!err) | |
1144 | continue; | |
1145 | /* fall through */ | |
1146 | } | |
1da177e4 LT |
1147 | if (pmd_none_or_clear_bad(src_pmd)) |
1148 | continue; | |
1149 | if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd, | |
1150 | vma, addr, next)) | |
1151 | return -ENOMEM; | |
1152 | } while (dst_pmd++, src_pmd++, addr = next, addr != end); | |
1153 | return 0; | |
1154 | } | |
1155 | ||
1156 | static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
c2febafc | 1157 | p4d_t *dst_p4d, p4d_t *src_p4d, struct vm_area_struct *vma, |
1da177e4 LT |
1158 | unsigned long addr, unsigned long end) |
1159 | { | |
1160 | pud_t *src_pud, *dst_pud; | |
1161 | unsigned long next; | |
1162 | ||
c2febafc | 1163 | dst_pud = pud_alloc(dst_mm, dst_p4d, addr); |
1da177e4 LT |
1164 | if (!dst_pud) |
1165 | return -ENOMEM; | |
c2febafc | 1166 | src_pud = pud_offset(src_p4d, addr); |
1da177e4 LT |
1167 | do { |
1168 | next = pud_addr_end(addr, end); | |
a00cc7d9 MW |
1169 | if (pud_trans_huge(*src_pud) || pud_devmap(*src_pud)) { |
1170 | int err; | |
1171 | ||
1172 | VM_BUG_ON_VMA(next-addr != HPAGE_PUD_SIZE, vma); | |
1173 | err = copy_huge_pud(dst_mm, src_mm, | |
1174 | dst_pud, src_pud, addr, vma); | |
1175 | if (err == -ENOMEM) | |
1176 | return -ENOMEM; | |
1177 | if (!err) | |
1178 | continue; | |
1179 | /* fall through */ | |
1180 | } | |
1da177e4 LT |
1181 | if (pud_none_or_clear_bad(src_pud)) |
1182 | continue; | |
1183 | if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud, | |
1184 | vma, addr, next)) | |
1185 | return -ENOMEM; | |
1186 | } while (dst_pud++, src_pud++, addr = next, addr != end); | |
1187 | return 0; | |
1188 | } | |
1189 | ||
c2febafc KS |
1190 | static inline int copy_p4d_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
1191 | pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma, | |
1192 | unsigned long addr, unsigned long end) | |
1193 | { | |
1194 | p4d_t *src_p4d, *dst_p4d; | |
1195 | unsigned long next; | |
1196 | ||
1197 | dst_p4d = p4d_alloc(dst_mm, dst_pgd, addr); | |
1198 | if (!dst_p4d) | |
1199 | return -ENOMEM; | |
1200 | src_p4d = p4d_offset(src_pgd, addr); | |
1201 | do { | |
1202 | next = p4d_addr_end(addr, end); | |
1203 | if (p4d_none_or_clear_bad(src_p4d)) | |
1204 | continue; | |
1205 | if (copy_pud_range(dst_mm, src_mm, dst_p4d, src_p4d, | |
1206 | vma, addr, next)) | |
1207 | return -ENOMEM; | |
1208 | } while (dst_p4d++, src_p4d++, addr = next, addr != end); | |
1209 | return 0; | |
1210 | } | |
1211 | ||
1da177e4 LT |
1212 | int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
1213 | struct vm_area_struct *vma) | |
1214 | { | |
1215 | pgd_t *src_pgd, *dst_pgd; | |
1216 | unsigned long next; | |
1217 | unsigned long addr = vma->vm_start; | |
1218 | unsigned long end = vma->vm_end; | |
2ec74c3e SG |
1219 | unsigned long mmun_start; /* For mmu_notifiers */ |
1220 | unsigned long mmun_end; /* For mmu_notifiers */ | |
1221 | bool is_cow; | |
cddb8a5c | 1222 | int ret; |
1da177e4 | 1223 | |
d992895b NP |
1224 | /* |
1225 | * Don't copy ptes where a page fault will fill them correctly. | |
1226 | * Fork becomes much lighter when there are big shared or private | |
1227 | * readonly mappings. The tradeoff is that copy_page_range is more | |
1228 | * efficient than faulting. | |
1229 | */ | |
0661a336 KS |
1230 | if (!(vma->vm_flags & (VM_HUGETLB | VM_PFNMAP | VM_MIXEDMAP)) && |
1231 | !vma->anon_vma) | |
1232 | return 0; | |
d992895b | 1233 | |
1da177e4 LT |
1234 | if (is_vm_hugetlb_page(vma)) |
1235 | return copy_hugetlb_page_range(dst_mm, src_mm, vma); | |
1236 | ||
b3b9c293 | 1237 | if (unlikely(vma->vm_flags & VM_PFNMAP)) { |
2ab64037 | 1238 | /* |
1239 | * We do not free on error cases below as remove_vma | |
1240 | * gets called on error from higher level routine | |
1241 | */ | |
5180da41 | 1242 | ret = track_pfn_copy(vma); |
2ab64037 | 1243 | if (ret) |
1244 | return ret; | |
1245 | } | |
1246 | ||
cddb8a5c AA |
1247 | /* |
1248 | * We need to invalidate the secondary MMU mappings only when | |
1249 | * there could be a permission downgrade on the ptes of the | |
1250 | * parent mm. And a permission downgrade will only happen if | |
1251 | * is_cow_mapping() returns true. | |
1252 | */ | |
2ec74c3e SG |
1253 | is_cow = is_cow_mapping(vma->vm_flags); |
1254 | mmun_start = addr; | |
1255 | mmun_end = end; | |
1256 | if (is_cow) | |
1257 | mmu_notifier_invalidate_range_start(src_mm, mmun_start, | |
1258 | mmun_end); | |
cddb8a5c AA |
1259 | |
1260 | ret = 0; | |
1da177e4 LT |
1261 | dst_pgd = pgd_offset(dst_mm, addr); |
1262 | src_pgd = pgd_offset(src_mm, addr); | |
1263 | do { | |
1264 | next = pgd_addr_end(addr, end); | |
1265 | if (pgd_none_or_clear_bad(src_pgd)) | |
1266 | continue; | |
c2febafc | 1267 | if (unlikely(copy_p4d_range(dst_mm, src_mm, dst_pgd, src_pgd, |
cddb8a5c AA |
1268 | vma, addr, next))) { |
1269 | ret = -ENOMEM; | |
1270 | break; | |
1271 | } | |
1da177e4 | 1272 | } while (dst_pgd++, src_pgd++, addr = next, addr != end); |
cddb8a5c | 1273 | |
2ec74c3e SG |
1274 | if (is_cow) |
1275 | mmu_notifier_invalidate_range_end(src_mm, mmun_start, mmun_end); | |
cddb8a5c | 1276 | return ret; |
1da177e4 LT |
1277 | } |
1278 | ||
51c6f666 | 1279 | static unsigned long zap_pte_range(struct mmu_gather *tlb, |
b5810039 | 1280 | struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 | 1281 | unsigned long addr, unsigned long end, |
97a89413 | 1282 | struct zap_details *details) |
1da177e4 | 1283 | { |
b5810039 | 1284 | struct mm_struct *mm = tlb->mm; |
d16dfc55 | 1285 | int force_flush = 0; |
d559db08 | 1286 | int rss[NR_MM_COUNTERS]; |
97a89413 | 1287 | spinlock_t *ptl; |
5f1a1907 | 1288 | pte_t *start_pte; |
97a89413 | 1289 | pte_t *pte; |
8a5f14a2 | 1290 | swp_entry_t entry; |
d559db08 | 1291 | |
07e32661 | 1292 | tlb_remove_check_page_size_change(tlb, PAGE_SIZE); |
d16dfc55 | 1293 | again: |
e303297e | 1294 | init_rss_vec(rss); |
5f1a1907 SR |
1295 | start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
1296 | pte = start_pte; | |
3ea27719 | 1297 | flush_tlb_batched_pending(mm); |
6606c3e0 | 1298 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
1299 | do { |
1300 | pte_t ptent = *pte; | |
166f61b9 | 1301 | if (pte_none(ptent)) |
1da177e4 | 1302 | continue; |
6f5e6b9e | 1303 | |
1da177e4 | 1304 | if (pte_present(ptent)) { |
ee498ed7 | 1305 | struct page *page; |
51c6f666 | 1306 | |
df6ad698 | 1307 | page = _vm_normal_page(vma, addr, ptent, true); |
1da177e4 LT |
1308 | if (unlikely(details) && page) { |
1309 | /* | |
1310 | * unmap_shared_mapping_pages() wants to | |
1311 | * invalidate cache without truncating: | |
1312 | * unmap shared but keep private pages. | |
1313 | */ | |
1314 | if (details->check_mapping && | |
800d8c63 | 1315 | details->check_mapping != page_rmapping(page)) |
1da177e4 | 1316 | continue; |
1da177e4 | 1317 | } |
b5810039 | 1318 | ptent = ptep_get_and_clear_full(mm, addr, pte, |
a600388d | 1319 | tlb->fullmm); |
1da177e4 LT |
1320 | tlb_remove_tlb_entry(tlb, pte, addr); |
1321 | if (unlikely(!page)) | |
1322 | continue; | |
eca56ff9 JM |
1323 | |
1324 | if (!PageAnon(page)) { | |
1cf35d47 LT |
1325 | if (pte_dirty(ptent)) { |
1326 | force_flush = 1; | |
6237bcd9 | 1327 | set_page_dirty(page); |
1cf35d47 | 1328 | } |
4917e5d0 | 1329 | if (pte_young(ptent) && |
64363aad | 1330 | likely(!(vma->vm_flags & VM_SEQ_READ))) |
bf3f3bc5 | 1331 | mark_page_accessed(page); |
6237bcd9 | 1332 | } |
eca56ff9 | 1333 | rss[mm_counter(page)]--; |
d281ee61 | 1334 | page_remove_rmap(page, false); |
3dc14741 HD |
1335 | if (unlikely(page_mapcount(page) < 0)) |
1336 | print_bad_pte(vma, addr, ptent, page); | |
e9d55e15 | 1337 | if (unlikely(__tlb_remove_page(tlb, page))) { |
1cf35d47 | 1338 | force_flush = 1; |
ce9ec37b | 1339 | addr += PAGE_SIZE; |
d16dfc55 | 1340 | break; |
1cf35d47 | 1341 | } |
1da177e4 LT |
1342 | continue; |
1343 | } | |
5042db43 JG |
1344 | |
1345 | entry = pte_to_swp_entry(ptent); | |
1346 | if (non_swap_entry(entry) && is_device_private_entry(entry)) { | |
1347 | struct page *page = device_private_entry_to_page(entry); | |
1348 | ||
1349 | if (unlikely(details && details->check_mapping)) { | |
1350 | /* | |
1351 | * unmap_shared_mapping_pages() wants to | |
1352 | * invalidate cache without truncating: | |
1353 | * unmap shared but keep private pages. | |
1354 | */ | |
1355 | if (details->check_mapping != | |
1356 | page_rmapping(page)) | |
1357 | continue; | |
1358 | } | |
1359 | ||
1360 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); | |
1361 | rss[mm_counter(page)]--; | |
1362 | page_remove_rmap(page, false); | |
1363 | put_page(page); | |
1364 | continue; | |
1365 | } | |
1366 | ||
3e8715fd KS |
1367 | /* If details->check_mapping, we leave swap entries. */ |
1368 | if (unlikely(details)) | |
1da177e4 | 1369 | continue; |
b084d435 | 1370 | |
8a5f14a2 KS |
1371 | entry = pte_to_swp_entry(ptent); |
1372 | if (!non_swap_entry(entry)) | |
1373 | rss[MM_SWAPENTS]--; | |
1374 | else if (is_migration_entry(entry)) { | |
1375 | struct page *page; | |
9f9f1acd | 1376 | |
8a5f14a2 | 1377 | page = migration_entry_to_page(entry); |
eca56ff9 | 1378 | rss[mm_counter(page)]--; |
b084d435 | 1379 | } |
8a5f14a2 KS |
1380 | if (unlikely(!free_swap_and_cache(entry))) |
1381 | print_bad_pte(vma, addr, ptent, NULL); | |
9888a1ca | 1382 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); |
97a89413 | 1383 | } while (pte++, addr += PAGE_SIZE, addr != end); |
ae859762 | 1384 | |
d559db08 | 1385 | add_mm_rss_vec(mm, rss); |
6606c3e0 | 1386 | arch_leave_lazy_mmu_mode(); |
51c6f666 | 1387 | |
1cf35d47 | 1388 | /* Do the actual TLB flush before dropping ptl */ |
fb7332a9 | 1389 | if (force_flush) |
1cf35d47 | 1390 | tlb_flush_mmu_tlbonly(tlb); |
1cf35d47 LT |
1391 | pte_unmap_unlock(start_pte, ptl); |
1392 | ||
1393 | /* | |
1394 | * If we forced a TLB flush (either due to running out of | |
1395 | * batch buffers or because we needed to flush dirty TLB | |
1396 | * entries before releasing the ptl), free the batched | |
1397 | * memory too. Restart if we didn't do everything. | |
1398 | */ | |
1399 | if (force_flush) { | |
1400 | force_flush = 0; | |
1401 | tlb_flush_mmu_free(tlb); | |
2b047252 | 1402 | if (addr != end) |
d16dfc55 PZ |
1403 | goto again; |
1404 | } | |
1405 | ||
51c6f666 | 1406 | return addr; |
1da177e4 LT |
1407 | } |
1408 | ||
51c6f666 | 1409 | static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, |
b5810039 | 1410 | struct vm_area_struct *vma, pud_t *pud, |
1da177e4 | 1411 | unsigned long addr, unsigned long end, |
97a89413 | 1412 | struct zap_details *details) |
1da177e4 LT |
1413 | { |
1414 | pmd_t *pmd; | |
1415 | unsigned long next; | |
1416 | ||
1417 | pmd = pmd_offset(pud, addr); | |
1418 | do { | |
1419 | next = pmd_addr_end(addr, end); | |
84c3fc4e | 1420 | if (is_swap_pmd(*pmd) || pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) { |
1a5a9906 | 1421 | if (next - addr != HPAGE_PMD_SIZE) { |
68428398 HD |
1422 | VM_BUG_ON_VMA(vma_is_anonymous(vma) && |
1423 | !rwsem_is_locked(&tlb->mm->mmap_sem), vma); | |
fd60775a | 1424 | __split_huge_pmd(vma, pmd, addr, false, NULL); |
f21760b1 | 1425 | } else if (zap_huge_pmd(tlb, vma, pmd, addr)) |
1a5a9906 | 1426 | goto next; |
71e3aac0 AA |
1427 | /* fall through */ |
1428 | } | |
1a5a9906 AA |
1429 | /* |
1430 | * Here there can be other concurrent MADV_DONTNEED or | |
1431 | * trans huge page faults running, and if the pmd is | |
1432 | * none or trans huge it can change under us. This is | |
1433 | * because MADV_DONTNEED holds the mmap_sem in read | |
1434 | * mode. | |
1435 | */ | |
1436 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) | |
1437 | goto next; | |
97a89413 | 1438 | next = zap_pte_range(tlb, vma, pmd, addr, next, details); |
1a5a9906 | 1439 | next: |
97a89413 PZ |
1440 | cond_resched(); |
1441 | } while (pmd++, addr = next, addr != end); | |
51c6f666 RH |
1442 | |
1443 | return addr; | |
1da177e4 LT |
1444 | } |
1445 | ||
51c6f666 | 1446 | static inline unsigned long zap_pud_range(struct mmu_gather *tlb, |
c2febafc | 1447 | struct vm_area_struct *vma, p4d_t *p4d, |
1da177e4 | 1448 | unsigned long addr, unsigned long end, |
97a89413 | 1449 | struct zap_details *details) |
1da177e4 LT |
1450 | { |
1451 | pud_t *pud; | |
1452 | unsigned long next; | |
1453 | ||
c2febafc | 1454 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
1455 | do { |
1456 | next = pud_addr_end(addr, end); | |
a00cc7d9 MW |
1457 | if (pud_trans_huge(*pud) || pud_devmap(*pud)) { |
1458 | if (next - addr != HPAGE_PUD_SIZE) { | |
1459 | VM_BUG_ON_VMA(!rwsem_is_locked(&tlb->mm->mmap_sem), vma); | |
1460 | split_huge_pud(vma, pud, addr); | |
1461 | } else if (zap_huge_pud(tlb, vma, pud, addr)) | |
1462 | goto next; | |
1463 | /* fall through */ | |
1464 | } | |
97a89413 | 1465 | if (pud_none_or_clear_bad(pud)) |
1da177e4 | 1466 | continue; |
97a89413 | 1467 | next = zap_pmd_range(tlb, vma, pud, addr, next, details); |
a00cc7d9 MW |
1468 | next: |
1469 | cond_resched(); | |
97a89413 | 1470 | } while (pud++, addr = next, addr != end); |
51c6f666 RH |
1471 | |
1472 | return addr; | |
1da177e4 LT |
1473 | } |
1474 | ||
c2febafc KS |
1475 | static inline unsigned long zap_p4d_range(struct mmu_gather *tlb, |
1476 | struct vm_area_struct *vma, pgd_t *pgd, | |
1477 | unsigned long addr, unsigned long end, | |
1478 | struct zap_details *details) | |
1479 | { | |
1480 | p4d_t *p4d; | |
1481 | unsigned long next; | |
1482 | ||
1483 | p4d = p4d_offset(pgd, addr); | |
1484 | do { | |
1485 | next = p4d_addr_end(addr, end); | |
1486 | if (p4d_none_or_clear_bad(p4d)) | |
1487 | continue; | |
1488 | next = zap_pud_range(tlb, vma, p4d, addr, next, details); | |
1489 | } while (p4d++, addr = next, addr != end); | |
1490 | ||
1491 | return addr; | |
1492 | } | |
1493 | ||
aac45363 | 1494 | void unmap_page_range(struct mmu_gather *tlb, |
038c7aa1 AV |
1495 | struct vm_area_struct *vma, |
1496 | unsigned long addr, unsigned long end, | |
1497 | struct zap_details *details) | |
1da177e4 LT |
1498 | { |
1499 | pgd_t *pgd; | |
1500 | unsigned long next; | |
1501 | ||
1da177e4 LT |
1502 | BUG_ON(addr >= end); |
1503 | tlb_start_vma(tlb, vma); | |
1504 | pgd = pgd_offset(vma->vm_mm, addr); | |
1505 | do { | |
1506 | next = pgd_addr_end(addr, end); | |
97a89413 | 1507 | if (pgd_none_or_clear_bad(pgd)) |
1da177e4 | 1508 | continue; |
c2febafc | 1509 | next = zap_p4d_range(tlb, vma, pgd, addr, next, details); |
97a89413 | 1510 | } while (pgd++, addr = next, addr != end); |
1da177e4 LT |
1511 | tlb_end_vma(tlb, vma); |
1512 | } | |
51c6f666 | 1513 | |
f5cc4eef AV |
1514 | |
1515 | static void unmap_single_vma(struct mmu_gather *tlb, | |
1516 | struct vm_area_struct *vma, unsigned long start_addr, | |
4f74d2c8 | 1517 | unsigned long end_addr, |
f5cc4eef AV |
1518 | struct zap_details *details) |
1519 | { | |
1520 | unsigned long start = max(vma->vm_start, start_addr); | |
1521 | unsigned long end; | |
1522 | ||
1523 | if (start >= vma->vm_end) | |
1524 | return; | |
1525 | end = min(vma->vm_end, end_addr); | |
1526 | if (end <= vma->vm_start) | |
1527 | return; | |
1528 | ||
cbc91f71 SD |
1529 | if (vma->vm_file) |
1530 | uprobe_munmap(vma, start, end); | |
1531 | ||
b3b9c293 | 1532 | if (unlikely(vma->vm_flags & VM_PFNMAP)) |
5180da41 | 1533 | untrack_pfn(vma, 0, 0); |
f5cc4eef AV |
1534 | |
1535 | if (start != end) { | |
1536 | if (unlikely(is_vm_hugetlb_page(vma))) { | |
1537 | /* | |
1538 | * It is undesirable to test vma->vm_file as it | |
1539 | * should be non-null for valid hugetlb area. | |
1540 | * However, vm_file will be NULL in the error | |
7aa6b4ad | 1541 | * cleanup path of mmap_region. When |
f5cc4eef | 1542 | * hugetlbfs ->mmap method fails, |
7aa6b4ad | 1543 | * mmap_region() nullifies vma->vm_file |
f5cc4eef AV |
1544 | * before calling this function to clean up. |
1545 | * Since no pte has actually been setup, it is | |
1546 | * safe to do nothing in this case. | |
1547 | */ | |
24669e58 | 1548 | if (vma->vm_file) { |
83cde9e8 | 1549 | i_mmap_lock_write(vma->vm_file->f_mapping); |
d833352a | 1550 | __unmap_hugepage_range_final(tlb, vma, start, end, NULL); |
83cde9e8 | 1551 | i_mmap_unlock_write(vma->vm_file->f_mapping); |
24669e58 | 1552 | } |
f5cc4eef AV |
1553 | } else |
1554 | unmap_page_range(tlb, vma, start, end, details); | |
1555 | } | |
1da177e4 LT |
1556 | } |
1557 | ||
1da177e4 LT |
1558 | /** |
1559 | * unmap_vmas - unmap a range of memory covered by a list of vma's | |
0164f69d | 1560 | * @tlb: address of the caller's struct mmu_gather |
1da177e4 LT |
1561 | * @vma: the starting vma |
1562 | * @start_addr: virtual address at which to start unmapping | |
1563 | * @end_addr: virtual address at which to end unmapping | |
1da177e4 | 1564 | * |
508034a3 | 1565 | * Unmap all pages in the vma list. |
1da177e4 | 1566 | * |
1da177e4 LT |
1567 | * Only addresses between `start' and `end' will be unmapped. |
1568 | * | |
1569 | * The VMA list must be sorted in ascending virtual address order. | |
1570 | * | |
1571 | * unmap_vmas() assumes that the caller will flush the whole unmapped address | |
1572 | * range after unmap_vmas() returns. So the only responsibility here is to | |
1573 | * ensure that any thus-far unmapped pages are flushed before unmap_vmas() | |
1574 | * drops the lock and schedules. | |
1575 | */ | |
6e8bb019 | 1576 | void unmap_vmas(struct mmu_gather *tlb, |
1da177e4 | 1577 | struct vm_area_struct *vma, unsigned long start_addr, |
4f74d2c8 | 1578 | unsigned long end_addr) |
1da177e4 | 1579 | { |
cddb8a5c | 1580 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 1581 | |
cddb8a5c | 1582 | mmu_notifier_invalidate_range_start(mm, start_addr, end_addr); |
f5cc4eef | 1583 | for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) |
4f74d2c8 | 1584 | unmap_single_vma(tlb, vma, start_addr, end_addr, NULL); |
cddb8a5c | 1585 | mmu_notifier_invalidate_range_end(mm, start_addr, end_addr); |
1da177e4 LT |
1586 | } |
1587 | ||
1588 | /** | |
1589 | * zap_page_range - remove user pages in a given range | |
1590 | * @vma: vm_area_struct holding the applicable pages | |
eb4546bb | 1591 | * @start: starting address of pages to zap |
1da177e4 | 1592 | * @size: number of bytes to zap |
f5cc4eef AV |
1593 | * |
1594 | * Caller must protect the VMA list | |
1da177e4 | 1595 | */ |
7e027b14 | 1596 | void zap_page_range(struct vm_area_struct *vma, unsigned long start, |
ecf1385d | 1597 | unsigned long size) |
1da177e4 LT |
1598 | { |
1599 | struct mm_struct *mm = vma->vm_mm; | |
d16dfc55 | 1600 | struct mmu_gather tlb; |
7e027b14 | 1601 | unsigned long end = start + size; |
1da177e4 | 1602 | |
1da177e4 | 1603 | lru_add_drain(); |
2b047252 | 1604 | tlb_gather_mmu(&tlb, mm, start, end); |
365e9c87 | 1605 | update_hiwater_rss(mm); |
7e027b14 | 1606 | mmu_notifier_invalidate_range_start(mm, start, end); |
4647706e | 1607 | for ( ; vma && vma->vm_start < end; vma = vma->vm_next) { |
ecf1385d | 1608 | unmap_single_vma(&tlb, vma, start, end, NULL); |
4647706e MG |
1609 | |
1610 | /* | |
1611 | * zap_page_range does not specify whether mmap_sem should be | |
1612 | * held for read or write. That allows parallel zap_page_range | |
1613 | * operations to unmap a PTE and defer a flush meaning that | |
1614 | * this call observes pte_none and fails to flush the TLB. | |
1615 | * Rather than adding a complex API, ensure that no stale | |
1616 | * TLB entries exist when this call returns. | |
1617 | */ | |
1618 | flush_tlb_range(vma, start, end); | |
1619 | } | |
1620 | ||
7e027b14 LT |
1621 | mmu_notifier_invalidate_range_end(mm, start, end); |
1622 | tlb_finish_mmu(&tlb, start, end); | |
1da177e4 LT |
1623 | } |
1624 | ||
f5cc4eef AV |
1625 | /** |
1626 | * zap_page_range_single - remove user pages in a given range | |
1627 | * @vma: vm_area_struct holding the applicable pages | |
1628 | * @address: starting address of pages to zap | |
1629 | * @size: number of bytes to zap | |
8a5f14a2 | 1630 | * @details: details of shared cache invalidation |
f5cc4eef AV |
1631 | * |
1632 | * The range must fit into one VMA. | |
1da177e4 | 1633 | */ |
f5cc4eef | 1634 | static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address, |
1da177e4 LT |
1635 | unsigned long size, struct zap_details *details) |
1636 | { | |
1637 | struct mm_struct *mm = vma->vm_mm; | |
d16dfc55 | 1638 | struct mmu_gather tlb; |
1da177e4 | 1639 | unsigned long end = address + size; |
1da177e4 | 1640 | |
1da177e4 | 1641 | lru_add_drain(); |
2b047252 | 1642 | tlb_gather_mmu(&tlb, mm, address, end); |
365e9c87 | 1643 | update_hiwater_rss(mm); |
f5cc4eef | 1644 | mmu_notifier_invalidate_range_start(mm, address, end); |
4f74d2c8 | 1645 | unmap_single_vma(&tlb, vma, address, end, details); |
f5cc4eef | 1646 | mmu_notifier_invalidate_range_end(mm, address, end); |
d16dfc55 | 1647 | tlb_finish_mmu(&tlb, address, end); |
1da177e4 LT |
1648 | } |
1649 | ||
c627f9cc JS |
1650 | /** |
1651 | * zap_vma_ptes - remove ptes mapping the vma | |
1652 | * @vma: vm_area_struct holding ptes to be zapped | |
1653 | * @address: starting address of pages to zap | |
1654 | * @size: number of bytes to zap | |
1655 | * | |
1656 | * This function only unmaps ptes assigned to VM_PFNMAP vmas. | |
1657 | * | |
1658 | * The entire address range must be fully contained within the vma. | |
1659 | * | |
1660 | * Returns 0 if successful. | |
1661 | */ | |
1662 | int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, | |
1663 | unsigned long size) | |
1664 | { | |
1665 | if (address < vma->vm_start || address + size > vma->vm_end || | |
1666 | !(vma->vm_flags & VM_PFNMAP)) | |
1667 | return -1; | |
f5cc4eef | 1668 | zap_page_range_single(vma, address, size, NULL); |
c627f9cc JS |
1669 | return 0; |
1670 | } | |
1671 | EXPORT_SYMBOL_GPL(zap_vma_ptes); | |
1672 | ||
25ca1d6c | 1673 | pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr, |
920c7a5d | 1674 | spinlock_t **ptl) |
c9cfcddf | 1675 | { |
c2febafc KS |
1676 | pgd_t *pgd; |
1677 | p4d_t *p4d; | |
1678 | pud_t *pud; | |
1679 | pmd_t *pmd; | |
1680 | ||
1681 | pgd = pgd_offset(mm, addr); | |
1682 | p4d = p4d_alloc(mm, pgd, addr); | |
1683 | if (!p4d) | |
1684 | return NULL; | |
1685 | pud = pud_alloc(mm, p4d, addr); | |
1686 | if (!pud) | |
1687 | return NULL; | |
1688 | pmd = pmd_alloc(mm, pud, addr); | |
1689 | if (!pmd) | |
1690 | return NULL; | |
1691 | ||
1692 | VM_BUG_ON(pmd_trans_huge(*pmd)); | |
1693 | return pte_alloc_map_lock(mm, pmd, addr, ptl); | |
c9cfcddf LT |
1694 | } |
1695 | ||
238f58d8 LT |
1696 | /* |
1697 | * This is the old fallback for page remapping. | |
1698 | * | |
1699 | * For historical reasons, it only allows reserved pages. Only | |
1700 | * old drivers should use this, and they needed to mark their | |
1701 | * pages reserved for the old functions anyway. | |
1702 | */ | |
423bad60 NP |
1703 | static int insert_page(struct vm_area_struct *vma, unsigned long addr, |
1704 | struct page *page, pgprot_t prot) | |
238f58d8 | 1705 | { |
423bad60 | 1706 | struct mm_struct *mm = vma->vm_mm; |
238f58d8 | 1707 | int retval; |
c9cfcddf | 1708 | pte_t *pte; |
8a9f3ccd BS |
1709 | spinlock_t *ptl; |
1710 | ||
238f58d8 | 1711 | retval = -EINVAL; |
a145dd41 | 1712 | if (PageAnon(page)) |
5b4e655e | 1713 | goto out; |
238f58d8 LT |
1714 | retval = -ENOMEM; |
1715 | flush_dcache_page(page); | |
c9cfcddf | 1716 | pte = get_locked_pte(mm, addr, &ptl); |
238f58d8 | 1717 | if (!pte) |
5b4e655e | 1718 | goto out; |
238f58d8 LT |
1719 | retval = -EBUSY; |
1720 | if (!pte_none(*pte)) | |
1721 | goto out_unlock; | |
1722 | ||
1723 | /* Ok, finally just insert the thing.. */ | |
1724 | get_page(page); | |
eca56ff9 | 1725 | inc_mm_counter_fast(mm, mm_counter_file(page)); |
dd78fedd | 1726 | page_add_file_rmap(page, false); |
238f58d8 LT |
1727 | set_pte_at(mm, addr, pte, mk_pte(page, prot)); |
1728 | ||
1729 | retval = 0; | |
8a9f3ccd BS |
1730 | pte_unmap_unlock(pte, ptl); |
1731 | return retval; | |
238f58d8 LT |
1732 | out_unlock: |
1733 | pte_unmap_unlock(pte, ptl); | |
1734 | out: | |
1735 | return retval; | |
1736 | } | |
1737 | ||
bfa5bf6d REB |
1738 | /** |
1739 | * vm_insert_page - insert single page into user vma | |
1740 | * @vma: user vma to map to | |
1741 | * @addr: target user address of this page | |
1742 | * @page: source kernel page | |
1743 | * | |
a145dd41 LT |
1744 | * This allows drivers to insert individual pages they've allocated |
1745 | * into a user vma. | |
1746 | * | |
1747 | * The page has to be a nice clean _individual_ kernel allocation. | |
1748 | * If you allocate a compound page, you need to have marked it as | |
1749 | * such (__GFP_COMP), or manually just split the page up yourself | |
8dfcc9ba | 1750 | * (see split_page()). |
a145dd41 LT |
1751 | * |
1752 | * NOTE! Traditionally this was done with "remap_pfn_range()" which | |
1753 | * took an arbitrary page protection parameter. This doesn't allow | |
1754 | * that. Your vma protection will have to be set up correctly, which | |
1755 | * means that if you want a shared writable mapping, you'd better | |
1756 | * ask for a shared writable mapping! | |
1757 | * | |
1758 | * The page does not need to be reserved. | |
4b6e1e37 KK |
1759 | * |
1760 | * Usually this function is called from f_op->mmap() handler | |
1761 | * under mm->mmap_sem write-lock, so it can change vma->vm_flags. | |
1762 | * Caller must set VM_MIXEDMAP on vma if it wants to call this | |
1763 | * function from other places, for example from page-fault handler. | |
a145dd41 | 1764 | */ |
423bad60 NP |
1765 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, |
1766 | struct page *page) | |
a145dd41 LT |
1767 | { |
1768 | if (addr < vma->vm_start || addr >= vma->vm_end) | |
1769 | return -EFAULT; | |
1770 | if (!page_count(page)) | |
1771 | return -EINVAL; | |
4b6e1e37 KK |
1772 | if (!(vma->vm_flags & VM_MIXEDMAP)) { |
1773 | BUG_ON(down_read_trylock(&vma->vm_mm->mmap_sem)); | |
1774 | BUG_ON(vma->vm_flags & VM_PFNMAP); | |
1775 | vma->vm_flags |= VM_MIXEDMAP; | |
1776 | } | |
423bad60 | 1777 | return insert_page(vma, addr, page, vma->vm_page_prot); |
a145dd41 | 1778 | } |
e3c3374f | 1779 | EXPORT_SYMBOL(vm_insert_page); |
a145dd41 | 1780 | |
423bad60 | 1781 | static int insert_pfn(struct vm_area_struct *vma, unsigned long addr, |
b2770da6 | 1782 | pfn_t pfn, pgprot_t prot, bool mkwrite) |
423bad60 NP |
1783 | { |
1784 | struct mm_struct *mm = vma->vm_mm; | |
1785 | int retval; | |
1786 | pte_t *pte, entry; | |
1787 | spinlock_t *ptl; | |
1788 | ||
1789 | retval = -ENOMEM; | |
1790 | pte = get_locked_pte(mm, addr, &ptl); | |
1791 | if (!pte) | |
1792 | goto out; | |
1793 | retval = -EBUSY; | |
b2770da6 RZ |
1794 | if (!pte_none(*pte)) { |
1795 | if (mkwrite) { | |
1796 | /* | |
1797 | * For read faults on private mappings the PFN passed | |
1798 | * in may not match the PFN we have mapped if the | |
1799 | * mapped PFN is a writeable COW page. In the mkwrite | |
1800 | * case we are creating a writable PTE for a shared | |
1801 | * mapping and we expect the PFNs to match. | |
1802 | */ | |
1803 | if (WARN_ON_ONCE(pte_pfn(*pte) != pfn_t_to_pfn(pfn))) | |
1804 | goto out_unlock; | |
1805 | entry = *pte; | |
1806 | goto out_mkwrite; | |
1807 | } else | |
1808 | goto out_unlock; | |
1809 | } | |
423bad60 NP |
1810 | |
1811 | /* Ok, finally just insert the thing.. */ | |
01c8f1c4 DW |
1812 | if (pfn_t_devmap(pfn)) |
1813 | entry = pte_mkdevmap(pfn_t_pte(pfn, prot)); | |
1814 | else | |
1815 | entry = pte_mkspecial(pfn_t_pte(pfn, prot)); | |
b2770da6 RZ |
1816 | |
1817 | out_mkwrite: | |
1818 | if (mkwrite) { | |
1819 | entry = pte_mkyoung(entry); | |
1820 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
1821 | } | |
1822 | ||
423bad60 | 1823 | set_pte_at(mm, addr, pte, entry); |
4b3073e1 | 1824 | update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */ |
423bad60 NP |
1825 | |
1826 | retval = 0; | |
1827 | out_unlock: | |
1828 | pte_unmap_unlock(pte, ptl); | |
1829 | out: | |
1830 | return retval; | |
1831 | } | |
1832 | ||
e0dc0d8f NP |
1833 | /** |
1834 | * vm_insert_pfn - insert single pfn into user vma | |
1835 | * @vma: user vma to map to | |
1836 | * @addr: target user address of this page | |
1837 | * @pfn: source kernel pfn | |
1838 | * | |
c462f179 | 1839 | * Similar to vm_insert_page, this allows drivers to insert individual pages |
e0dc0d8f NP |
1840 | * they've allocated into a user vma. Same comments apply. |
1841 | * | |
1842 | * This function should only be called from a vm_ops->fault handler, and | |
1843 | * in that case the handler should return NULL. | |
0d71d10a NP |
1844 | * |
1845 | * vma cannot be a COW mapping. | |
1846 | * | |
1847 | * As this is called only for pages that do not currently exist, we | |
1848 | * do not need to flush old virtual caches or the TLB. | |
e0dc0d8f NP |
1849 | */ |
1850 | int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr, | |
423bad60 | 1851 | unsigned long pfn) |
1745cbc5 AL |
1852 | { |
1853 | return vm_insert_pfn_prot(vma, addr, pfn, vma->vm_page_prot); | |
1854 | } | |
1855 | EXPORT_SYMBOL(vm_insert_pfn); | |
1856 | ||
1857 | /** | |
1858 | * vm_insert_pfn_prot - insert single pfn into user vma with specified pgprot | |
1859 | * @vma: user vma to map to | |
1860 | * @addr: target user address of this page | |
1861 | * @pfn: source kernel pfn | |
1862 | * @pgprot: pgprot flags for the inserted page | |
1863 | * | |
1864 | * This is exactly like vm_insert_pfn, except that it allows drivers to | |
1865 | * to override pgprot on a per-page basis. | |
1866 | * | |
1867 | * This only makes sense for IO mappings, and it makes no sense for | |
1868 | * cow mappings. In general, using multiple vmas is preferable; | |
1869 | * vm_insert_pfn_prot should only be used if using multiple VMAs is | |
1870 | * impractical. | |
1871 | */ | |
1872 | int vm_insert_pfn_prot(struct vm_area_struct *vma, unsigned long addr, | |
1873 | unsigned long pfn, pgprot_t pgprot) | |
e0dc0d8f | 1874 | { |
2ab64037 | 1875 | int ret; |
7e675137 NP |
1876 | /* |
1877 | * Technically, architectures with pte_special can avoid all these | |
1878 | * restrictions (same for remap_pfn_range). However we would like | |
1879 | * consistency in testing and feature parity among all, so we should | |
1880 | * try to keep these invariants in place for everybody. | |
1881 | */ | |
b379d790 JH |
1882 | BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); |
1883 | BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == | |
1884 | (VM_PFNMAP|VM_MIXEDMAP)); | |
1885 | BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); | |
1886 | BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); | |
e0dc0d8f | 1887 | |
423bad60 NP |
1888 | if (addr < vma->vm_start || addr >= vma->vm_end) |
1889 | return -EFAULT; | |
308a047c BP |
1890 | |
1891 | track_pfn_insert(vma, &pgprot, __pfn_to_pfn_t(pfn, PFN_DEV)); | |
2ab64037 | 1892 | |
b2770da6 RZ |
1893 | ret = insert_pfn(vma, addr, __pfn_to_pfn_t(pfn, PFN_DEV), pgprot, |
1894 | false); | |
2ab64037 | 1895 | |
2ab64037 | 1896 | return ret; |
423bad60 | 1897 | } |
1745cbc5 | 1898 | EXPORT_SYMBOL(vm_insert_pfn_prot); |
e0dc0d8f | 1899 | |
b2770da6 RZ |
1900 | static int __vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr, |
1901 | pfn_t pfn, bool mkwrite) | |
423bad60 | 1902 | { |
87744ab3 DW |
1903 | pgprot_t pgprot = vma->vm_page_prot; |
1904 | ||
423bad60 | 1905 | BUG_ON(!(vma->vm_flags & VM_MIXEDMAP)); |
e0dc0d8f | 1906 | |
423bad60 NP |
1907 | if (addr < vma->vm_start || addr >= vma->vm_end) |
1908 | return -EFAULT; | |
308a047c BP |
1909 | |
1910 | track_pfn_insert(vma, &pgprot, pfn); | |
e0dc0d8f | 1911 | |
423bad60 NP |
1912 | /* |
1913 | * If we don't have pte special, then we have to use the pfn_valid() | |
1914 | * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* | |
1915 | * refcount the page if pfn_valid is true (hence insert_page rather | |
62eede62 HD |
1916 | * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP |
1917 | * without pte special, it would there be refcounted as a normal page. | |
423bad60 | 1918 | */ |
03fc2da6 | 1919 | if (!HAVE_PTE_SPECIAL && !pfn_t_devmap(pfn) && pfn_t_valid(pfn)) { |
423bad60 NP |
1920 | struct page *page; |
1921 | ||
03fc2da6 DW |
1922 | /* |
1923 | * At this point we are committed to insert_page() | |
1924 | * regardless of whether the caller specified flags that | |
1925 | * result in pfn_t_has_page() == false. | |
1926 | */ | |
1927 | page = pfn_to_page(pfn_t_to_pfn(pfn)); | |
87744ab3 | 1928 | return insert_page(vma, addr, page, pgprot); |
423bad60 | 1929 | } |
b2770da6 RZ |
1930 | return insert_pfn(vma, addr, pfn, pgprot, mkwrite); |
1931 | } | |
1932 | ||
1933 | int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr, | |
1934 | pfn_t pfn) | |
1935 | { | |
1936 | return __vm_insert_mixed(vma, addr, pfn, false); | |
1937 | ||
e0dc0d8f | 1938 | } |
423bad60 | 1939 | EXPORT_SYMBOL(vm_insert_mixed); |
e0dc0d8f | 1940 | |
b2770da6 RZ |
1941 | int vm_insert_mixed_mkwrite(struct vm_area_struct *vma, unsigned long addr, |
1942 | pfn_t pfn) | |
1943 | { | |
1944 | return __vm_insert_mixed(vma, addr, pfn, true); | |
1945 | } | |
1946 | EXPORT_SYMBOL(vm_insert_mixed_mkwrite); | |
1947 | ||
1da177e4 LT |
1948 | /* |
1949 | * maps a range of physical memory into the requested pages. the old | |
1950 | * mappings are removed. any references to nonexistent pages results | |
1951 | * in null mappings (currently treated as "copy-on-access") | |
1952 | */ | |
1953 | static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, | |
1954 | unsigned long addr, unsigned long end, | |
1955 | unsigned long pfn, pgprot_t prot) | |
1956 | { | |
1957 | pte_t *pte; | |
c74df32c | 1958 | spinlock_t *ptl; |
1da177e4 | 1959 | |
c74df32c | 1960 | pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); |
1da177e4 LT |
1961 | if (!pte) |
1962 | return -ENOMEM; | |
6606c3e0 | 1963 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
1964 | do { |
1965 | BUG_ON(!pte_none(*pte)); | |
7e675137 | 1966 | set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); |
1da177e4 LT |
1967 | pfn++; |
1968 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
6606c3e0 | 1969 | arch_leave_lazy_mmu_mode(); |
c74df32c | 1970 | pte_unmap_unlock(pte - 1, ptl); |
1da177e4 LT |
1971 | return 0; |
1972 | } | |
1973 | ||
1974 | static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, | |
1975 | unsigned long addr, unsigned long end, | |
1976 | unsigned long pfn, pgprot_t prot) | |
1977 | { | |
1978 | pmd_t *pmd; | |
1979 | unsigned long next; | |
1980 | ||
1981 | pfn -= addr >> PAGE_SHIFT; | |
1982 | pmd = pmd_alloc(mm, pud, addr); | |
1983 | if (!pmd) | |
1984 | return -ENOMEM; | |
f66055ab | 1985 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
1da177e4 LT |
1986 | do { |
1987 | next = pmd_addr_end(addr, end); | |
1988 | if (remap_pte_range(mm, pmd, addr, next, | |
1989 | pfn + (addr >> PAGE_SHIFT), prot)) | |
1990 | return -ENOMEM; | |
1991 | } while (pmd++, addr = next, addr != end); | |
1992 | return 0; | |
1993 | } | |
1994 | ||
c2febafc | 1995 | static inline int remap_pud_range(struct mm_struct *mm, p4d_t *p4d, |
1da177e4 LT |
1996 | unsigned long addr, unsigned long end, |
1997 | unsigned long pfn, pgprot_t prot) | |
1998 | { | |
1999 | pud_t *pud; | |
2000 | unsigned long next; | |
2001 | ||
2002 | pfn -= addr >> PAGE_SHIFT; | |
c2febafc | 2003 | pud = pud_alloc(mm, p4d, addr); |
1da177e4 LT |
2004 | if (!pud) |
2005 | return -ENOMEM; | |
2006 | do { | |
2007 | next = pud_addr_end(addr, end); | |
2008 | if (remap_pmd_range(mm, pud, addr, next, | |
2009 | pfn + (addr >> PAGE_SHIFT), prot)) | |
2010 | return -ENOMEM; | |
2011 | } while (pud++, addr = next, addr != end); | |
2012 | return 0; | |
2013 | } | |
2014 | ||
c2febafc KS |
2015 | static inline int remap_p4d_range(struct mm_struct *mm, pgd_t *pgd, |
2016 | unsigned long addr, unsigned long end, | |
2017 | unsigned long pfn, pgprot_t prot) | |
2018 | { | |
2019 | p4d_t *p4d; | |
2020 | unsigned long next; | |
2021 | ||
2022 | pfn -= addr >> PAGE_SHIFT; | |
2023 | p4d = p4d_alloc(mm, pgd, addr); | |
2024 | if (!p4d) | |
2025 | return -ENOMEM; | |
2026 | do { | |
2027 | next = p4d_addr_end(addr, end); | |
2028 | if (remap_pud_range(mm, p4d, addr, next, | |
2029 | pfn + (addr >> PAGE_SHIFT), prot)) | |
2030 | return -ENOMEM; | |
2031 | } while (p4d++, addr = next, addr != end); | |
2032 | return 0; | |
2033 | } | |
2034 | ||
bfa5bf6d REB |
2035 | /** |
2036 | * remap_pfn_range - remap kernel memory to userspace | |
2037 | * @vma: user vma to map to | |
2038 | * @addr: target user address to start at | |
2039 | * @pfn: physical address of kernel memory | |
2040 | * @size: size of map area | |
2041 | * @prot: page protection flags for this mapping | |
2042 | * | |
2043 | * Note: this is only safe if the mm semaphore is held when called. | |
2044 | */ | |
1da177e4 LT |
2045 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, |
2046 | unsigned long pfn, unsigned long size, pgprot_t prot) | |
2047 | { | |
2048 | pgd_t *pgd; | |
2049 | unsigned long next; | |
2d15cab8 | 2050 | unsigned long end = addr + PAGE_ALIGN(size); |
1da177e4 | 2051 | struct mm_struct *mm = vma->vm_mm; |
d5957d2f | 2052 | unsigned long remap_pfn = pfn; |
1da177e4 LT |
2053 | int err; |
2054 | ||
2055 | /* | |
2056 | * Physically remapped pages are special. Tell the | |
2057 | * rest of the world about it: | |
2058 | * VM_IO tells people not to look at these pages | |
2059 | * (accesses can have side effects). | |
6aab341e LT |
2060 | * VM_PFNMAP tells the core MM that the base pages are just |
2061 | * raw PFN mappings, and do not have a "struct page" associated | |
2062 | * with them. | |
314e51b9 KK |
2063 | * VM_DONTEXPAND |
2064 | * Disable vma merging and expanding with mremap(). | |
2065 | * VM_DONTDUMP | |
2066 | * Omit vma from core dump, even when VM_IO turned off. | |
fb155c16 LT |
2067 | * |
2068 | * There's a horrible special case to handle copy-on-write | |
2069 | * behaviour that some programs depend on. We mark the "original" | |
2070 | * un-COW'ed pages by matching them up with "vma->vm_pgoff". | |
b3b9c293 | 2071 | * See vm_normal_page() for details. |
1da177e4 | 2072 | */ |
b3b9c293 KK |
2073 | if (is_cow_mapping(vma->vm_flags)) { |
2074 | if (addr != vma->vm_start || end != vma->vm_end) | |
2075 | return -EINVAL; | |
fb155c16 | 2076 | vma->vm_pgoff = pfn; |
b3b9c293 KK |
2077 | } |
2078 | ||
d5957d2f | 2079 | err = track_pfn_remap(vma, &prot, remap_pfn, addr, PAGE_ALIGN(size)); |
b3b9c293 | 2080 | if (err) |
3c8bb73a | 2081 | return -EINVAL; |
fb155c16 | 2082 | |
314e51b9 | 2083 | vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; |
1da177e4 LT |
2084 | |
2085 | BUG_ON(addr >= end); | |
2086 | pfn -= addr >> PAGE_SHIFT; | |
2087 | pgd = pgd_offset(mm, addr); | |
2088 | flush_cache_range(vma, addr, end); | |
1da177e4 LT |
2089 | do { |
2090 | next = pgd_addr_end(addr, end); | |
c2febafc | 2091 | err = remap_p4d_range(mm, pgd, addr, next, |
1da177e4 LT |
2092 | pfn + (addr >> PAGE_SHIFT), prot); |
2093 | if (err) | |
2094 | break; | |
2095 | } while (pgd++, addr = next, addr != end); | |
2ab64037 | 2096 | |
2097 | if (err) | |
d5957d2f | 2098 | untrack_pfn(vma, remap_pfn, PAGE_ALIGN(size)); |
2ab64037 | 2099 | |
1da177e4 LT |
2100 | return err; |
2101 | } | |
2102 | EXPORT_SYMBOL(remap_pfn_range); | |
2103 | ||
b4cbb197 LT |
2104 | /** |
2105 | * vm_iomap_memory - remap memory to userspace | |
2106 | * @vma: user vma to map to | |
2107 | * @start: start of area | |
2108 | * @len: size of area | |
2109 | * | |
2110 | * This is a simplified io_remap_pfn_range() for common driver use. The | |
2111 | * driver just needs to give us the physical memory range to be mapped, | |
2112 | * we'll figure out the rest from the vma information. | |
2113 | * | |
2114 | * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get | |
2115 | * whatever write-combining details or similar. | |
2116 | */ | |
2117 | int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) | |
2118 | { | |
2119 | unsigned long vm_len, pfn, pages; | |
2120 | ||
2121 | /* Check that the physical memory area passed in looks valid */ | |
2122 | if (start + len < start) | |
2123 | return -EINVAL; | |
2124 | /* | |
2125 | * You *really* shouldn't map things that aren't page-aligned, | |
2126 | * but we've historically allowed it because IO memory might | |
2127 | * just have smaller alignment. | |
2128 | */ | |
2129 | len += start & ~PAGE_MASK; | |
2130 | pfn = start >> PAGE_SHIFT; | |
2131 | pages = (len + ~PAGE_MASK) >> PAGE_SHIFT; | |
2132 | if (pfn + pages < pfn) | |
2133 | return -EINVAL; | |
2134 | ||
2135 | /* We start the mapping 'vm_pgoff' pages into the area */ | |
2136 | if (vma->vm_pgoff > pages) | |
2137 | return -EINVAL; | |
2138 | pfn += vma->vm_pgoff; | |
2139 | pages -= vma->vm_pgoff; | |
2140 | ||
2141 | /* Can we fit all of the mapping? */ | |
2142 | vm_len = vma->vm_end - vma->vm_start; | |
2143 | if (vm_len >> PAGE_SHIFT > pages) | |
2144 | return -EINVAL; | |
2145 | ||
2146 | /* Ok, let it rip */ | |
2147 | return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); | |
2148 | } | |
2149 | EXPORT_SYMBOL(vm_iomap_memory); | |
2150 | ||
aee16b3c JF |
2151 | static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, |
2152 | unsigned long addr, unsigned long end, | |
2153 | pte_fn_t fn, void *data) | |
2154 | { | |
2155 | pte_t *pte; | |
2156 | int err; | |
2f569afd | 2157 | pgtable_t token; |
94909914 | 2158 | spinlock_t *uninitialized_var(ptl); |
aee16b3c JF |
2159 | |
2160 | pte = (mm == &init_mm) ? | |
2161 | pte_alloc_kernel(pmd, addr) : | |
2162 | pte_alloc_map_lock(mm, pmd, addr, &ptl); | |
2163 | if (!pte) | |
2164 | return -ENOMEM; | |
2165 | ||
2166 | BUG_ON(pmd_huge(*pmd)); | |
2167 | ||
38e0edb1 JF |
2168 | arch_enter_lazy_mmu_mode(); |
2169 | ||
2f569afd | 2170 | token = pmd_pgtable(*pmd); |
aee16b3c JF |
2171 | |
2172 | do { | |
c36987e2 | 2173 | err = fn(pte++, token, addr, data); |
aee16b3c JF |
2174 | if (err) |
2175 | break; | |
c36987e2 | 2176 | } while (addr += PAGE_SIZE, addr != end); |
aee16b3c | 2177 | |
38e0edb1 JF |
2178 | arch_leave_lazy_mmu_mode(); |
2179 | ||
aee16b3c JF |
2180 | if (mm != &init_mm) |
2181 | pte_unmap_unlock(pte-1, ptl); | |
2182 | return err; | |
2183 | } | |
2184 | ||
2185 | static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, | |
2186 | unsigned long addr, unsigned long end, | |
2187 | pte_fn_t fn, void *data) | |
2188 | { | |
2189 | pmd_t *pmd; | |
2190 | unsigned long next; | |
2191 | int err; | |
2192 | ||
ceb86879 AK |
2193 | BUG_ON(pud_huge(*pud)); |
2194 | ||
aee16b3c JF |
2195 | pmd = pmd_alloc(mm, pud, addr); |
2196 | if (!pmd) | |
2197 | return -ENOMEM; | |
2198 | do { | |
2199 | next = pmd_addr_end(addr, end); | |
2200 | err = apply_to_pte_range(mm, pmd, addr, next, fn, data); | |
2201 | if (err) | |
2202 | break; | |
2203 | } while (pmd++, addr = next, addr != end); | |
2204 | return err; | |
2205 | } | |
2206 | ||
c2febafc | 2207 | static int apply_to_pud_range(struct mm_struct *mm, p4d_t *p4d, |
aee16b3c JF |
2208 | unsigned long addr, unsigned long end, |
2209 | pte_fn_t fn, void *data) | |
2210 | { | |
2211 | pud_t *pud; | |
2212 | unsigned long next; | |
2213 | int err; | |
2214 | ||
c2febafc | 2215 | pud = pud_alloc(mm, p4d, addr); |
aee16b3c JF |
2216 | if (!pud) |
2217 | return -ENOMEM; | |
2218 | do { | |
2219 | next = pud_addr_end(addr, end); | |
2220 | err = apply_to_pmd_range(mm, pud, addr, next, fn, data); | |
2221 | if (err) | |
2222 | break; | |
2223 | } while (pud++, addr = next, addr != end); | |
2224 | return err; | |
2225 | } | |
2226 | ||
c2febafc KS |
2227 | static int apply_to_p4d_range(struct mm_struct *mm, pgd_t *pgd, |
2228 | unsigned long addr, unsigned long end, | |
2229 | pte_fn_t fn, void *data) | |
2230 | { | |
2231 | p4d_t *p4d; | |
2232 | unsigned long next; | |
2233 | int err; | |
2234 | ||
2235 | p4d = p4d_alloc(mm, pgd, addr); | |
2236 | if (!p4d) | |
2237 | return -ENOMEM; | |
2238 | do { | |
2239 | next = p4d_addr_end(addr, end); | |
2240 | err = apply_to_pud_range(mm, p4d, addr, next, fn, data); | |
2241 | if (err) | |
2242 | break; | |
2243 | } while (p4d++, addr = next, addr != end); | |
2244 | return err; | |
2245 | } | |
2246 | ||
aee16b3c JF |
2247 | /* |
2248 | * Scan a region of virtual memory, filling in page tables as necessary | |
2249 | * and calling a provided function on each leaf page table. | |
2250 | */ | |
2251 | int apply_to_page_range(struct mm_struct *mm, unsigned long addr, | |
2252 | unsigned long size, pte_fn_t fn, void *data) | |
2253 | { | |
2254 | pgd_t *pgd; | |
2255 | unsigned long next; | |
57250a5b | 2256 | unsigned long end = addr + size; |
aee16b3c JF |
2257 | int err; |
2258 | ||
9cb65bc3 MP |
2259 | if (WARN_ON(addr >= end)) |
2260 | return -EINVAL; | |
2261 | ||
aee16b3c JF |
2262 | pgd = pgd_offset(mm, addr); |
2263 | do { | |
2264 | next = pgd_addr_end(addr, end); | |
c2febafc | 2265 | err = apply_to_p4d_range(mm, pgd, addr, next, fn, data); |
aee16b3c JF |
2266 | if (err) |
2267 | break; | |
2268 | } while (pgd++, addr = next, addr != end); | |
57250a5b | 2269 | |
aee16b3c JF |
2270 | return err; |
2271 | } | |
2272 | EXPORT_SYMBOL_GPL(apply_to_page_range); | |
2273 | ||
8f4e2101 | 2274 | /* |
9b4bdd2f KS |
2275 | * handle_pte_fault chooses page fault handler according to an entry which was |
2276 | * read non-atomically. Before making any commitment, on those architectures | |
2277 | * or configurations (e.g. i386 with PAE) which might give a mix of unmatched | |
2278 | * parts, do_swap_page must check under lock before unmapping the pte and | |
2279 | * proceeding (but do_wp_page is only called after already making such a check; | |
a335b2e1 | 2280 | * and do_anonymous_page can safely check later on). |
8f4e2101 | 2281 | */ |
4c21e2f2 | 2282 | static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd, |
8f4e2101 HD |
2283 | pte_t *page_table, pte_t orig_pte) |
2284 | { | |
2285 | int same = 1; | |
2286 | #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT) | |
2287 | if (sizeof(pte_t) > sizeof(unsigned long)) { | |
4c21e2f2 HD |
2288 | spinlock_t *ptl = pte_lockptr(mm, pmd); |
2289 | spin_lock(ptl); | |
8f4e2101 | 2290 | same = pte_same(*page_table, orig_pte); |
4c21e2f2 | 2291 | spin_unlock(ptl); |
8f4e2101 HD |
2292 | } |
2293 | #endif | |
2294 | pte_unmap(page_table); | |
2295 | return same; | |
2296 | } | |
2297 | ||
9de455b2 | 2298 | static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma) |
6aab341e | 2299 | { |
0abdd7a8 DW |
2300 | debug_dma_assert_idle(src); |
2301 | ||
6aab341e LT |
2302 | /* |
2303 | * If the source page was a PFN mapping, we don't have | |
2304 | * a "struct page" for it. We do a best-effort copy by | |
2305 | * just copying from the original user address. If that | |
2306 | * fails, we just zero-fill it. Live with it. | |
2307 | */ | |
2308 | if (unlikely(!src)) { | |
9b04c5fe | 2309 | void *kaddr = kmap_atomic(dst); |
5d2a2dbb LT |
2310 | void __user *uaddr = (void __user *)(va & PAGE_MASK); |
2311 | ||
2312 | /* | |
2313 | * This really shouldn't fail, because the page is there | |
2314 | * in the page tables. But it might just be unreadable, | |
2315 | * in which case we just give up and fill the result with | |
2316 | * zeroes. | |
2317 | */ | |
2318 | if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) | |
3ecb01df | 2319 | clear_page(kaddr); |
9b04c5fe | 2320 | kunmap_atomic(kaddr); |
c4ec7b0d | 2321 | flush_dcache_page(dst); |
0ed361de NP |
2322 | } else |
2323 | copy_user_highpage(dst, src, va, vma); | |
6aab341e LT |
2324 | } |
2325 | ||
c20cd45e MH |
2326 | static gfp_t __get_fault_gfp_mask(struct vm_area_struct *vma) |
2327 | { | |
2328 | struct file *vm_file = vma->vm_file; | |
2329 | ||
2330 | if (vm_file) | |
2331 | return mapping_gfp_mask(vm_file->f_mapping) | __GFP_FS | __GFP_IO; | |
2332 | ||
2333 | /* | |
2334 | * Special mappings (e.g. VDSO) do not have any file so fake | |
2335 | * a default GFP_KERNEL for them. | |
2336 | */ | |
2337 | return GFP_KERNEL; | |
2338 | } | |
2339 | ||
fb09a464 KS |
2340 | /* |
2341 | * Notify the address space that the page is about to become writable so that | |
2342 | * it can prohibit this or wait for the page to get into an appropriate state. | |
2343 | * | |
2344 | * We do this without the lock held, so that it can sleep if it needs to. | |
2345 | */ | |
38b8cb7f | 2346 | static int do_page_mkwrite(struct vm_fault *vmf) |
fb09a464 | 2347 | { |
fb09a464 | 2348 | int ret; |
38b8cb7f JK |
2349 | struct page *page = vmf->page; |
2350 | unsigned int old_flags = vmf->flags; | |
fb09a464 | 2351 | |
38b8cb7f | 2352 | vmf->flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; |
fb09a464 | 2353 | |
11bac800 | 2354 | ret = vmf->vma->vm_ops->page_mkwrite(vmf); |
38b8cb7f JK |
2355 | /* Restore original flags so that caller is not surprised */ |
2356 | vmf->flags = old_flags; | |
fb09a464 KS |
2357 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) |
2358 | return ret; | |
2359 | if (unlikely(!(ret & VM_FAULT_LOCKED))) { | |
2360 | lock_page(page); | |
2361 | if (!page->mapping) { | |
2362 | unlock_page(page); | |
2363 | return 0; /* retry */ | |
2364 | } | |
2365 | ret |= VM_FAULT_LOCKED; | |
2366 | } else | |
2367 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
2368 | return ret; | |
2369 | } | |
2370 | ||
97ba0c2b JK |
2371 | /* |
2372 | * Handle dirtying of a page in shared file mapping on a write fault. | |
2373 | * | |
2374 | * The function expects the page to be locked and unlocks it. | |
2375 | */ | |
2376 | static void fault_dirty_shared_page(struct vm_area_struct *vma, | |
2377 | struct page *page) | |
2378 | { | |
2379 | struct address_space *mapping; | |
2380 | bool dirtied; | |
2381 | bool page_mkwrite = vma->vm_ops && vma->vm_ops->page_mkwrite; | |
2382 | ||
2383 | dirtied = set_page_dirty(page); | |
2384 | VM_BUG_ON_PAGE(PageAnon(page), page); | |
2385 | /* | |
2386 | * Take a local copy of the address_space - page.mapping may be zeroed | |
2387 | * by truncate after unlock_page(). The address_space itself remains | |
2388 | * pinned by vma->vm_file's reference. We rely on unlock_page()'s | |
2389 | * release semantics to prevent the compiler from undoing this copying. | |
2390 | */ | |
2391 | mapping = page_rmapping(page); | |
2392 | unlock_page(page); | |
2393 | ||
2394 | if ((dirtied || page_mkwrite) && mapping) { | |
2395 | /* | |
2396 | * Some device drivers do not set page.mapping | |
2397 | * but still dirty their pages | |
2398 | */ | |
2399 | balance_dirty_pages_ratelimited(mapping); | |
2400 | } | |
2401 | ||
2402 | if (!page_mkwrite) | |
2403 | file_update_time(vma->vm_file); | |
2404 | } | |
2405 | ||
4e047f89 SR |
2406 | /* |
2407 | * Handle write page faults for pages that can be reused in the current vma | |
2408 | * | |
2409 | * This can happen either due to the mapping being with the VM_SHARED flag, | |
2410 | * or due to us being the last reference standing to the page. In either | |
2411 | * case, all we need to do here is to mark the page as writable and update | |
2412 | * any related book-keeping. | |
2413 | */ | |
997dd98d | 2414 | static inline void wp_page_reuse(struct vm_fault *vmf) |
82b0f8c3 | 2415 | __releases(vmf->ptl) |
4e047f89 | 2416 | { |
82b0f8c3 | 2417 | struct vm_area_struct *vma = vmf->vma; |
a41b70d6 | 2418 | struct page *page = vmf->page; |
4e047f89 SR |
2419 | pte_t entry; |
2420 | /* | |
2421 | * Clear the pages cpupid information as the existing | |
2422 | * information potentially belongs to a now completely | |
2423 | * unrelated process. | |
2424 | */ | |
2425 | if (page) | |
2426 | page_cpupid_xchg_last(page, (1 << LAST_CPUPID_SHIFT) - 1); | |
2427 | ||
2994302b JK |
2428 | flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); |
2429 | entry = pte_mkyoung(vmf->orig_pte); | |
4e047f89 | 2430 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
82b0f8c3 JK |
2431 | if (ptep_set_access_flags(vma, vmf->address, vmf->pte, entry, 1)) |
2432 | update_mmu_cache(vma, vmf->address, vmf->pte); | |
2433 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
4e047f89 SR |
2434 | } |
2435 | ||
2f38ab2c SR |
2436 | /* |
2437 | * Handle the case of a page which we actually need to copy to a new page. | |
2438 | * | |
2439 | * Called with mmap_sem locked and the old page referenced, but | |
2440 | * without the ptl held. | |
2441 | * | |
2442 | * High level logic flow: | |
2443 | * | |
2444 | * - Allocate a page, copy the content of the old page to the new one. | |
2445 | * - Handle book keeping and accounting - cgroups, mmu-notifiers, etc. | |
2446 | * - Take the PTL. If the pte changed, bail out and release the allocated page | |
2447 | * - If the pte is still the way we remember it, update the page table and all | |
2448 | * relevant references. This includes dropping the reference the page-table | |
2449 | * held to the old page, as well as updating the rmap. | |
2450 | * - In any case, unlock the PTL and drop the reference we took to the old page. | |
2451 | */ | |
a41b70d6 | 2452 | static int wp_page_copy(struct vm_fault *vmf) |
2f38ab2c | 2453 | { |
82b0f8c3 | 2454 | struct vm_area_struct *vma = vmf->vma; |
bae473a4 | 2455 | struct mm_struct *mm = vma->vm_mm; |
a41b70d6 | 2456 | struct page *old_page = vmf->page; |
2f38ab2c | 2457 | struct page *new_page = NULL; |
2f38ab2c SR |
2458 | pte_t entry; |
2459 | int page_copied = 0; | |
82b0f8c3 | 2460 | const unsigned long mmun_start = vmf->address & PAGE_MASK; |
bae473a4 | 2461 | const unsigned long mmun_end = mmun_start + PAGE_SIZE; |
2f38ab2c SR |
2462 | struct mem_cgroup *memcg; |
2463 | ||
2464 | if (unlikely(anon_vma_prepare(vma))) | |
2465 | goto oom; | |
2466 | ||
2994302b | 2467 | if (is_zero_pfn(pte_pfn(vmf->orig_pte))) { |
82b0f8c3 JK |
2468 | new_page = alloc_zeroed_user_highpage_movable(vma, |
2469 | vmf->address); | |
2f38ab2c SR |
2470 | if (!new_page) |
2471 | goto oom; | |
2472 | } else { | |
bae473a4 | 2473 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, |
82b0f8c3 | 2474 | vmf->address); |
2f38ab2c SR |
2475 | if (!new_page) |
2476 | goto oom; | |
82b0f8c3 | 2477 | cow_user_page(new_page, old_page, vmf->address, vma); |
2f38ab2c | 2478 | } |
2f38ab2c | 2479 | |
f627c2f5 | 2480 | if (mem_cgroup_try_charge(new_page, mm, GFP_KERNEL, &memcg, false)) |
2f38ab2c SR |
2481 | goto oom_free_new; |
2482 | ||
eb3c24f3 MG |
2483 | __SetPageUptodate(new_page); |
2484 | ||
2f38ab2c SR |
2485 | mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); |
2486 | ||
2487 | /* | |
2488 | * Re-check the pte - we dropped the lock | |
2489 | */ | |
82b0f8c3 | 2490 | vmf->pte = pte_offset_map_lock(mm, vmf->pmd, vmf->address, &vmf->ptl); |
2994302b | 2491 | if (likely(pte_same(*vmf->pte, vmf->orig_pte))) { |
2f38ab2c SR |
2492 | if (old_page) { |
2493 | if (!PageAnon(old_page)) { | |
eca56ff9 JM |
2494 | dec_mm_counter_fast(mm, |
2495 | mm_counter_file(old_page)); | |
2f38ab2c SR |
2496 | inc_mm_counter_fast(mm, MM_ANONPAGES); |
2497 | } | |
2498 | } else { | |
2499 | inc_mm_counter_fast(mm, MM_ANONPAGES); | |
2500 | } | |
2994302b | 2501 | flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); |
2f38ab2c SR |
2502 | entry = mk_pte(new_page, vma->vm_page_prot); |
2503 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
2504 | /* | |
2505 | * Clear the pte entry and flush it first, before updating the | |
2506 | * pte with the new entry. This will avoid a race condition | |
2507 | * seen in the presence of one thread doing SMC and another | |
2508 | * thread doing COW. | |
2509 | */ | |
82b0f8c3 JK |
2510 | ptep_clear_flush_notify(vma, vmf->address, vmf->pte); |
2511 | page_add_new_anon_rmap(new_page, vma, vmf->address, false); | |
f627c2f5 | 2512 | mem_cgroup_commit_charge(new_page, memcg, false, false); |
2f38ab2c SR |
2513 | lru_cache_add_active_or_unevictable(new_page, vma); |
2514 | /* | |
2515 | * We call the notify macro here because, when using secondary | |
2516 | * mmu page tables (such as kvm shadow page tables), we want the | |
2517 | * new page to be mapped directly into the secondary page table. | |
2518 | */ | |
82b0f8c3 JK |
2519 | set_pte_at_notify(mm, vmf->address, vmf->pte, entry); |
2520 | update_mmu_cache(vma, vmf->address, vmf->pte); | |
2f38ab2c SR |
2521 | if (old_page) { |
2522 | /* | |
2523 | * Only after switching the pte to the new page may | |
2524 | * we remove the mapcount here. Otherwise another | |
2525 | * process may come and find the rmap count decremented | |
2526 | * before the pte is switched to the new page, and | |
2527 | * "reuse" the old page writing into it while our pte | |
2528 | * here still points into it and can be read by other | |
2529 | * threads. | |
2530 | * | |
2531 | * The critical issue is to order this | |
2532 | * page_remove_rmap with the ptp_clear_flush above. | |
2533 | * Those stores are ordered by (if nothing else,) | |
2534 | * the barrier present in the atomic_add_negative | |
2535 | * in page_remove_rmap. | |
2536 | * | |
2537 | * Then the TLB flush in ptep_clear_flush ensures that | |
2538 | * no process can access the old page before the | |
2539 | * decremented mapcount is visible. And the old page | |
2540 | * cannot be reused until after the decremented | |
2541 | * mapcount is visible. So transitively, TLBs to | |
2542 | * old page will be flushed before it can be reused. | |
2543 | */ | |
d281ee61 | 2544 | page_remove_rmap(old_page, false); |
2f38ab2c SR |
2545 | } |
2546 | ||
2547 | /* Free the old page.. */ | |
2548 | new_page = old_page; | |
2549 | page_copied = 1; | |
2550 | } else { | |
f627c2f5 | 2551 | mem_cgroup_cancel_charge(new_page, memcg, false); |
2f38ab2c SR |
2552 | } |
2553 | ||
2554 | if (new_page) | |
09cbfeaf | 2555 | put_page(new_page); |
2f38ab2c | 2556 | |
82b0f8c3 | 2557 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
4645b9fe JG |
2558 | /* |
2559 | * No need to double call mmu_notifier->invalidate_range() callback as | |
2560 | * the above ptep_clear_flush_notify() did already call it. | |
2561 | */ | |
2562 | mmu_notifier_invalidate_range_only_end(mm, mmun_start, mmun_end); | |
2f38ab2c SR |
2563 | if (old_page) { |
2564 | /* | |
2565 | * Don't let another task, with possibly unlocked vma, | |
2566 | * keep the mlocked page. | |
2567 | */ | |
2568 | if (page_copied && (vma->vm_flags & VM_LOCKED)) { | |
2569 | lock_page(old_page); /* LRU manipulation */ | |
e90309c9 KS |
2570 | if (PageMlocked(old_page)) |
2571 | munlock_vma_page(old_page); | |
2f38ab2c SR |
2572 | unlock_page(old_page); |
2573 | } | |
09cbfeaf | 2574 | put_page(old_page); |
2f38ab2c SR |
2575 | } |
2576 | return page_copied ? VM_FAULT_WRITE : 0; | |
2577 | oom_free_new: | |
09cbfeaf | 2578 | put_page(new_page); |
2f38ab2c SR |
2579 | oom: |
2580 | if (old_page) | |
09cbfeaf | 2581 | put_page(old_page); |
2f38ab2c SR |
2582 | return VM_FAULT_OOM; |
2583 | } | |
2584 | ||
66a6197c JK |
2585 | /** |
2586 | * finish_mkwrite_fault - finish page fault for a shared mapping, making PTE | |
2587 | * writeable once the page is prepared | |
2588 | * | |
2589 | * @vmf: structure describing the fault | |
2590 | * | |
2591 | * This function handles all that is needed to finish a write page fault in a | |
2592 | * shared mapping due to PTE being read-only once the mapped page is prepared. | |
2593 | * It handles locking of PTE and modifying it. The function returns | |
2594 | * VM_FAULT_WRITE on success, 0 when PTE got changed before we acquired PTE | |
2595 | * lock. | |
2596 | * | |
2597 | * The function expects the page to be locked or other protection against | |
2598 | * concurrent faults / writeback (such as DAX radix tree locks). | |
2599 | */ | |
2600 | int finish_mkwrite_fault(struct vm_fault *vmf) | |
2601 | { | |
2602 | WARN_ON_ONCE(!(vmf->vma->vm_flags & VM_SHARED)); | |
2603 | vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, vmf->address, | |
2604 | &vmf->ptl); | |
2605 | /* | |
2606 | * We might have raced with another page fault while we released the | |
2607 | * pte_offset_map_lock. | |
2608 | */ | |
2609 | if (!pte_same(*vmf->pte, vmf->orig_pte)) { | |
2610 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
a19e2553 | 2611 | return VM_FAULT_NOPAGE; |
66a6197c JK |
2612 | } |
2613 | wp_page_reuse(vmf); | |
a19e2553 | 2614 | return 0; |
66a6197c JK |
2615 | } |
2616 | ||
dd906184 BH |
2617 | /* |
2618 | * Handle write page faults for VM_MIXEDMAP or VM_PFNMAP for a VM_SHARED | |
2619 | * mapping | |
2620 | */ | |
2994302b | 2621 | static int wp_pfn_shared(struct vm_fault *vmf) |
dd906184 | 2622 | { |
82b0f8c3 | 2623 | struct vm_area_struct *vma = vmf->vma; |
bae473a4 | 2624 | |
dd906184 | 2625 | if (vma->vm_ops && vma->vm_ops->pfn_mkwrite) { |
dd906184 BH |
2626 | int ret; |
2627 | ||
82b0f8c3 | 2628 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
fe82221f | 2629 | vmf->flags |= FAULT_FLAG_MKWRITE; |
11bac800 | 2630 | ret = vma->vm_ops->pfn_mkwrite(vmf); |
2f89dc12 | 2631 | if (ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)) |
dd906184 | 2632 | return ret; |
66a6197c | 2633 | return finish_mkwrite_fault(vmf); |
dd906184 | 2634 | } |
997dd98d JK |
2635 | wp_page_reuse(vmf); |
2636 | return VM_FAULT_WRITE; | |
dd906184 BH |
2637 | } |
2638 | ||
a41b70d6 | 2639 | static int wp_page_shared(struct vm_fault *vmf) |
82b0f8c3 | 2640 | __releases(vmf->ptl) |
93e478d4 | 2641 | { |
82b0f8c3 | 2642 | struct vm_area_struct *vma = vmf->vma; |
93e478d4 | 2643 | |
a41b70d6 | 2644 | get_page(vmf->page); |
93e478d4 | 2645 | |
93e478d4 SR |
2646 | if (vma->vm_ops && vma->vm_ops->page_mkwrite) { |
2647 | int tmp; | |
2648 | ||
82b0f8c3 | 2649 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
38b8cb7f | 2650 | tmp = do_page_mkwrite(vmf); |
93e478d4 SR |
2651 | if (unlikely(!tmp || (tmp & |
2652 | (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { | |
a41b70d6 | 2653 | put_page(vmf->page); |
93e478d4 SR |
2654 | return tmp; |
2655 | } | |
66a6197c | 2656 | tmp = finish_mkwrite_fault(vmf); |
a19e2553 | 2657 | if (unlikely(tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { |
a41b70d6 | 2658 | unlock_page(vmf->page); |
a41b70d6 | 2659 | put_page(vmf->page); |
66a6197c | 2660 | return tmp; |
93e478d4 | 2661 | } |
66a6197c JK |
2662 | } else { |
2663 | wp_page_reuse(vmf); | |
997dd98d | 2664 | lock_page(vmf->page); |
93e478d4 | 2665 | } |
997dd98d JK |
2666 | fault_dirty_shared_page(vma, vmf->page); |
2667 | put_page(vmf->page); | |
93e478d4 | 2668 | |
997dd98d | 2669 | return VM_FAULT_WRITE; |
93e478d4 SR |
2670 | } |
2671 | ||
1da177e4 LT |
2672 | /* |
2673 | * This routine handles present pages, when users try to write | |
2674 | * to a shared page. It is done by copying the page to a new address | |
2675 | * and decrementing the shared-page counter for the old page. | |
2676 | * | |
1da177e4 LT |
2677 | * Note that this routine assumes that the protection checks have been |
2678 | * done by the caller (the low-level page fault routine in most cases). | |
2679 | * Thus we can safely just mark it writable once we've done any necessary | |
2680 | * COW. | |
2681 | * | |
2682 | * We also mark the page dirty at this point even though the page will | |
2683 | * change only once the write actually happens. This avoids a few races, | |
2684 | * and potentially makes it more efficient. | |
2685 | * | |
8f4e2101 HD |
2686 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2687 | * but allow concurrent faults), with pte both mapped and locked. | |
2688 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2689 | */ |
2994302b | 2690 | static int do_wp_page(struct vm_fault *vmf) |
82b0f8c3 | 2691 | __releases(vmf->ptl) |
1da177e4 | 2692 | { |
82b0f8c3 | 2693 | struct vm_area_struct *vma = vmf->vma; |
1da177e4 | 2694 | |
a41b70d6 JK |
2695 | vmf->page = vm_normal_page(vma, vmf->address, vmf->orig_pte); |
2696 | if (!vmf->page) { | |
251b97f5 | 2697 | /* |
64e45507 PF |
2698 | * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a |
2699 | * VM_PFNMAP VMA. | |
251b97f5 PZ |
2700 | * |
2701 | * We should not cow pages in a shared writeable mapping. | |
dd906184 | 2702 | * Just mark the pages writable and/or call ops->pfn_mkwrite. |
251b97f5 PZ |
2703 | */ |
2704 | if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) == | |
2705 | (VM_WRITE|VM_SHARED)) | |
2994302b | 2706 | return wp_pfn_shared(vmf); |
2f38ab2c | 2707 | |
82b0f8c3 | 2708 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 2709 | return wp_page_copy(vmf); |
251b97f5 | 2710 | } |
1da177e4 | 2711 | |
d08b3851 | 2712 | /* |
ee6a6457 PZ |
2713 | * Take out anonymous pages first, anonymous shared vmas are |
2714 | * not dirty accountable. | |
d08b3851 | 2715 | */ |
a41b70d6 | 2716 | if (PageAnon(vmf->page) && !PageKsm(vmf->page)) { |
ba3c4ce6 | 2717 | int total_map_swapcount; |
a41b70d6 JK |
2718 | if (!trylock_page(vmf->page)) { |
2719 | get_page(vmf->page); | |
82b0f8c3 | 2720 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 2721 | lock_page(vmf->page); |
82b0f8c3 JK |
2722 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
2723 | vmf->address, &vmf->ptl); | |
2994302b | 2724 | if (!pte_same(*vmf->pte, vmf->orig_pte)) { |
a41b70d6 | 2725 | unlock_page(vmf->page); |
82b0f8c3 | 2726 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 2727 | put_page(vmf->page); |
28766805 | 2728 | return 0; |
ab967d86 | 2729 | } |
a41b70d6 | 2730 | put_page(vmf->page); |
ee6a6457 | 2731 | } |
ba3c4ce6 HY |
2732 | if (reuse_swap_page(vmf->page, &total_map_swapcount)) { |
2733 | if (total_map_swapcount == 1) { | |
6d0a07ed AA |
2734 | /* |
2735 | * The page is all ours. Move it to | |
2736 | * our anon_vma so the rmap code will | |
2737 | * not search our parent or siblings. | |
2738 | * Protected against the rmap code by | |
2739 | * the page lock. | |
2740 | */ | |
a41b70d6 | 2741 | page_move_anon_rmap(vmf->page, vma); |
6d0a07ed | 2742 | } |
a41b70d6 | 2743 | unlock_page(vmf->page); |
997dd98d JK |
2744 | wp_page_reuse(vmf); |
2745 | return VM_FAULT_WRITE; | |
b009c024 | 2746 | } |
a41b70d6 | 2747 | unlock_page(vmf->page); |
ee6a6457 | 2748 | } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) == |
d08b3851 | 2749 | (VM_WRITE|VM_SHARED))) { |
a41b70d6 | 2750 | return wp_page_shared(vmf); |
1da177e4 | 2751 | } |
1da177e4 LT |
2752 | |
2753 | /* | |
2754 | * Ok, we need to copy. Oh, well.. | |
2755 | */ | |
a41b70d6 | 2756 | get_page(vmf->page); |
28766805 | 2757 | |
82b0f8c3 | 2758 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 2759 | return wp_page_copy(vmf); |
1da177e4 LT |
2760 | } |
2761 | ||
97a89413 | 2762 | static void unmap_mapping_range_vma(struct vm_area_struct *vma, |
1da177e4 LT |
2763 | unsigned long start_addr, unsigned long end_addr, |
2764 | struct zap_details *details) | |
2765 | { | |
f5cc4eef | 2766 | zap_page_range_single(vma, start_addr, end_addr - start_addr, details); |
1da177e4 LT |
2767 | } |
2768 | ||
f808c13f | 2769 | static inline void unmap_mapping_range_tree(struct rb_root_cached *root, |
1da177e4 LT |
2770 | struct zap_details *details) |
2771 | { | |
2772 | struct vm_area_struct *vma; | |
1da177e4 LT |
2773 | pgoff_t vba, vea, zba, zea; |
2774 | ||
6b2dbba8 | 2775 | vma_interval_tree_foreach(vma, root, |
1da177e4 | 2776 | details->first_index, details->last_index) { |
1da177e4 LT |
2777 | |
2778 | vba = vma->vm_pgoff; | |
d6e93217 | 2779 | vea = vba + vma_pages(vma) - 1; |
1da177e4 LT |
2780 | zba = details->first_index; |
2781 | if (zba < vba) | |
2782 | zba = vba; | |
2783 | zea = details->last_index; | |
2784 | if (zea > vea) | |
2785 | zea = vea; | |
2786 | ||
97a89413 | 2787 | unmap_mapping_range_vma(vma, |
1da177e4 LT |
2788 | ((zba - vba) << PAGE_SHIFT) + vma->vm_start, |
2789 | ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, | |
97a89413 | 2790 | details); |
1da177e4 LT |
2791 | } |
2792 | } | |
2793 | ||
1da177e4 | 2794 | /** |
8a5f14a2 KS |
2795 | * unmap_mapping_range - unmap the portion of all mmaps in the specified |
2796 | * address_space corresponding to the specified page range in the underlying | |
2797 | * file. | |
2798 | * | |
3d41088f | 2799 | * @mapping: the address space containing mmaps to be unmapped. |
1da177e4 LT |
2800 | * @holebegin: byte in first page to unmap, relative to the start of |
2801 | * the underlying file. This will be rounded down to a PAGE_SIZE | |
25d9e2d1 | 2802 | * boundary. Note that this is different from truncate_pagecache(), which |
1da177e4 LT |
2803 | * must keep the partial page. In contrast, we must get rid of |
2804 | * partial pages. | |
2805 | * @holelen: size of prospective hole in bytes. This will be rounded | |
2806 | * up to a PAGE_SIZE boundary. A holelen of zero truncates to the | |
2807 | * end of the file. | |
2808 | * @even_cows: 1 when truncating a file, unmap even private COWed pages; | |
2809 | * but 0 when invalidating pagecache, don't throw away private data. | |
2810 | */ | |
2811 | void unmap_mapping_range(struct address_space *mapping, | |
2812 | loff_t const holebegin, loff_t const holelen, int even_cows) | |
2813 | { | |
aac45363 | 2814 | struct zap_details details = { }; |
1da177e4 LT |
2815 | pgoff_t hba = holebegin >> PAGE_SHIFT; |
2816 | pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
2817 | ||
2818 | /* Check for overflow. */ | |
2819 | if (sizeof(holelen) > sizeof(hlen)) { | |
2820 | long long holeend = | |
2821 | (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
2822 | if (holeend & ~(long long)ULONG_MAX) | |
2823 | hlen = ULONG_MAX - hba + 1; | |
2824 | } | |
2825 | ||
166f61b9 | 2826 | details.check_mapping = even_cows ? NULL : mapping; |
1da177e4 LT |
2827 | details.first_index = hba; |
2828 | details.last_index = hba + hlen - 1; | |
2829 | if (details.last_index < details.first_index) | |
2830 | details.last_index = ULONG_MAX; | |
1da177e4 | 2831 | |
46c043ed | 2832 | i_mmap_lock_write(mapping); |
f808c13f | 2833 | if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))) |
1da177e4 | 2834 | unmap_mapping_range_tree(&mapping->i_mmap, &details); |
46c043ed | 2835 | i_mmap_unlock_write(mapping); |
1da177e4 LT |
2836 | } |
2837 | EXPORT_SYMBOL(unmap_mapping_range); | |
2838 | ||
1da177e4 | 2839 | /* |
8f4e2101 HD |
2840 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2841 | * but allow concurrent faults), and pte mapped but not yet locked. | |
9a95f3cf PC |
2842 | * We return with pte unmapped and unlocked. |
2843 | * | |
2844 | * We return with the mmap_sem locked or unlocked in the same cases | |
2845 | * as does filemap_fault(). | |
1da177e4 | 2846 | */ |
2994302b | 2847 | int do_swap_page(struct vm_fault *vmf) |
1da177e4 | 2848 | { |
82b0f8c3 | 2849 | struct vm_area_struct *vma = vmf->vma; |
0bcac06f | 2850 | struct page *page = NULL, *swapcache = NULL; |
00501b53 | 2851 | struct mem_cgroup *memcg; |
ec560175 | 2852 | struct vma_swap_readahead swap_ra; |
65500d23 | 2853 | swp_entry_t entry; |
1da177e4 | 2854 | pte_t pte; |
d065bd81 | 2855 | int locked; |
ad8c2ee8 | 2856 | int exclusive = 0; |
83c54070 | 2857 | int ret = 0; |
ec560175 | 2858 | bool vma_readahead = swap_use_vma_readahead(); |
1da177e4 | 2859 | |
ec560175 HY |
2860 | if (vma_readahead) |
2861 | page = swap_readahead_detect(vmf, &swap_ra); | |
2862 | if (!pte_unmap_same(vma->vm_mm, vmf->pmd, vmf->pte, vmf->orig_pte)) { | |
2863 | if (page) | |
2864 | put_page(page); | |
8f4e2101 | 2865 | goto out; |
ec560175 | 2866 | } |
65500d23 | 2867 | |
2994302b | 2868 | entry = pte_to_swp_entry(vmf->orig_pte); |
d1737fdb AK |
2869 | if (unlikely(non_swap_entry(entry))) { |
2870 | if (is_migration_entry(entry)) { | |
82b0f8c3 JK |
2871 | migration_entry_wait(vma->vm_mm, vmf->pmd, |
2872 | vmf->address); | |
5042db43 JG |
2873 | } else if (is_device_private_entry(entry)) { |
2874 | /* | |
2875 | * For un-addressable device memory we call the pgmap | |
2876 | * fault handler callback. The callback must migrate | |
2877 | * the page back to some CPU accessible page. | |
2878 | */ | |
2879 | ret = device_private_entry_fault(vma, vmf->address, entry, | |
2880 | vmf->flags, vmf->pmd); | |
d1737fdb AK |
2881 | } else if (is_hwpoison_entry(entry)) { |
2882 | ret = VM_FAULT_HWPOISON; | |
2883 | } else { | |
2994302b | 2884 | print_bad_pte(vma, vmf->address, vmf->orig_pte, NULL); |
d99be1a8 | 2885 | ret = VM_FAULT_SIGBUS; |
d1737fdb | 2886 | } |
0697212a CL |
2887 | goto out; |
2888 | } | |
0bcac06f MK |
2889 | |
2890 | ||
0ff92245 | 2891 | delayacct_set_flag(DELAYACCT_PF_SWAPIN); |
ec560175 HY |
2892 | if (!page) |
2893 | page = lookup_swap_cache(entry, vma_readahead ? vma : NULL, | |
2894 | vmf->address); | |
1da177e4 | 2895 | if (!page) { |
0bcac06f MK |
2896 | struct swap_info_struct *si = swp_swap_info(entry); |
2897 | ||
aa8d22a1 MK |
2898 | if (si->flags & SWP_SYNCHRONOUS_IO && |
2899 | __swap_count(si, entry) == 1) { | |
0bcac06f MK |
2900 | /* skip swapcache */ |
2901 | page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vmf->address); | |
2902 | if (page) { | |
2903 | __SetPageLocked(page); | |
2904 | __SetPageSwapBacked(page); | |
2905 | set_page_private(page, entry.val); | |
2906 | lru_cache_add_anon(page); | |
2907 | swap_readpage(page, true); | |
2908 | } | |
aa8d22a1 MK |
2909 | } else { |
2910 | if (vma_readahead) | |
2911 | page = do_swap_page_readahead(entry, | |
2912 | GFP_HIGHUSER_MOVABLE, vmf, &swap_ra); | |
2913 | else | |
2914 | page = swapin_readahead(entry, | |
2915 | GFP_HIGHUSER_MOVABLE, vma, vmf->address); | |
2916 | swapcache = page; | |
0bcac06f MK |
2917 | } |
2918 | ||
1da177e4 LT |
2919 | if (!page) { |
2920 | /* | |
8f4e2101 HD |
2921 | * Back out if somebody else faulted in this pte |
2922 | * while we released the pte lock. | |
1da177e4 | 2923 | */ |
82b0f8c3 JK |
2924 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
2925 | vmf->address, &vmf->ptl); | |
2994302b | 2926 | if (likely(pte_same(*vmf->pte, vmf->orig_pte))) |
1da177e4 | 2927 | ret = VM_FAULT_OOM; |
0ff92245 | 2928 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
65500d23 | 2929 | goto unlock; |
1da177e4 LT |
2930 | } |
2931 | ||
2932 | /* Had to read the page from swap area: Major fault */ | |
2933 | ret = VM_FAULT_MAJOR; | |
f8891e5e | 2934 | count_vm_event(PGMAJFAULT); |
2262185c | 2935 | count_memcg_event_mm(vma->vm_mm, PGMAJFAULT); |
d1737fdb | 2936 | } else if (PageHWPoison(page)) { |
71f72525 WF |
2937 | /* |
2938 | * hwpoisoned dirty swapcache pages are kept for killing | |
2939 | * owner processes (which may be unknown at hwpoison time) | |
2940 | */ | |
d1737fdb AK |
2941 | ret = VM_FAULT_HWPOISON; |
2942 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); | |
56f31801 | 2943 | swapcache = page; |
4779cb31 | 2944 | goto out_release; |
1da177e4 LT |
2945 | } |
2946 | ||
82b0f8c3 | 2947 | locked = lock_page_or_retry(page, vma->vm_mm, vmf->flags); |
e709ffd6 | 2948 | |
073e587e | 2949 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
d065bd81 ML |
2950 | if (!locked) { |
2951 | ret |= VM_FAULT_RETRY; | |
2952 | goto out_release; | |
2953 | } | |
073e587e | 2954 | |
4969c119 | 2955 | /* |
31c4a3d3 HD |
2956 | * Make sure try_to_free_swap or reuse_swap_page or swapoff did not |
2957 | * release the swapcache from under us. The page pin, and pte_same | |
2958 | * test below, are not enough to exclude that. Even if it is still | |
2959 | * swapcache, we need to check that the page's swap has not changed. | |
4969c119 | 2960 | */ |
0bcac06f MK |
2961 | if (unlikely((!PageSwapCache(page) || |
2962 | page_private(page) != entry.val)) && swapcache) | |
4969c119 AA |
2963 | goto out_page; |
2964 | ||
82b0f8c3 | 2965 | page = ksm_might_need_to_copy(page, vma, vmf->address); |
cbf86cfe HD |
2966 | if (unlikely(!page)) { |
2967 | ret = VM_FAULT_OOM; | |
2968 | page = swapcache; | |
cbf86cfe | 2969 | goto out_page; |
5ad64688 HD |
2970 | } |
2971 | ||
bae473a4 KS |
2972 | if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL, |
2973 | &memcg, false)) { | |
8a9f3ccd | 2974 | ret = VM_FAULT_OOM; |
bc43f75c | 2975 | goto out_page; |
8a9f3ccd BS |
2976 | } |
2977 | ||
1da177e4 | 2978 | /* |
8f4e2101 | 2979 | * Back out if somebody else already faulted in this pte. |
1da177e4 | 2980 | */ |
82b0f8c3 JK |
2981 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, |
2982 | &vmf->ptl); | |
2994302b | 2983 | if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) |
b8107480 | 2984 | goto out_nomap; |
b8107480 KK |
2985 | |
2986 | if (unlikely(!PageUptodate(page))) { | |
2987 | ret = VM_FAULT_SIGBUS; | |
2988 | goto out_nomap; | |
1da177e4 LT |
2989 | } |
2990 | ||
8c7c6e34 KH |
2991 | /* |
2992 | * The page isn't present yet, go ahead with the fault. | |
2993 | * | |
2994 | * Be careful about the sequence of operations here. | |
2995 | * To get its accounting right, reuse_swap_page() must be called | |
2996 | * while the page is counted on swap but not yet in mapcount i.e. | |
2997 | * before page_add_anon_rmap() and swap_free(); try_to_free_swap() | |
2998 | * must be called after the swap_free(), or it will never succeed. | |
8c7c6e34 | 2999 | */ |
1da177e4 | 3000 | |
bae473a4 KS |
3001 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
3002 | dec_mm_counter_fast(vma->vm_mm, MM_SWAPENTS); | |
1da177e4 | 3003 | pte = mk_pte(page, vma->vm_page_prot); |
82b0f8c3 | 3004 | if ((vmf->flags & FAULT_FLAG_WRITE) && reuse_swap_page(page, NULL)) { |
1da177e4 | 3005 | pte = maybe_mkwrite(pte_mkdirty(pte), vma); |
82b0f8c3 | 3006 | vmf->flags &= ~FAULT_FLAG_WRITE; |
9a5b489b | 3007 | ret |= VM_FAULT_WRITE; |
d281ee61 | 3008 | exclusive = RMAP_EXCLUSIVE; |
1da177e4 | 3009 | } |
1da177e4 | 3010 | flush_icache_page(vma, page); |
2994302b | 3011 | if (pte_swp_soft_dirty(vmf->orig_pte)) |
179ef71c | 3012 | pte = pte_mksoft_dirty(pte); |
82b0f8c3 | 3013 | set_pte_at(vma->vm_mm, vmf->address, vmf->pte, pte); |
2994302b | 3014 | vmf->orig_pte = pte; |
0bcac06f MK |
3015 | |
3016 | /* ksm created a completely new copy */ | |
3017 | if (unlikely(page != swapcache && swapcache)) { | |
82b0f8c3 | 3018 | page_add_new_anon_rmap(page, vma, vmf->address, false); |
f627c2f5 | 3019 | mem_cgroup_commit_charge(page, memcg, false, false); |
00501b53 | 3020 | lru_cache_add_active_or_unevictable(page, vma); |
0bcac06f MK |
3021 | } else { |
3022 | do_page_add_anon_rmap(page, vma, vmf->address, exclusive); | |
3023 | mem_cgroup_commit_charge(page, memcg, true, false); | |
3024 | activate_page(page); | |
00501b53 | 3025 | } |
1da177e4 | 3026 | |
c475a8ab | 3027 | swap_free(entry); |
5ccc5aba VD |
3028 | if (mem_cgroup_swap_full(page) || |
3029 | (vma->vm_flags & VM_LOCKED) || PageMlocked(page)) | |
a2c43eed | 3030 | try_to_free_swap(page); |
c475a8ab | 3031 | unlock_page(page); |
0bcac06f | 3032 | if (page != swapcache && swapcache) { |
4969c119 AA |
3033 | /* |
3034 | * Hold the lock to avoid the swap entry to be reused | |
3035 | * until we take the PT lock for the pte_same() check | |
3036 | * (to avoid false positives from pte_same). For | |
3037 | * further safety release the lock after the swap_free | |
3038 | * so that the swap count won't change under a | |
3039 | * parallel locked swapcache. | |
3040 | */ | |
3041 | unlock_page(swapcache); | |
09cbfeaf | 3042 | put_page(swapcache); |
4969c119 | 3043 | } |
c475a8ab | 3044 | |
82b0f8c3 | 3045 | if (vmf->flags & FAULT_FLAG_WRITE) { |
2994302b | 3046 | ret |= do_wp_page(vmf); |
61469f1d HD |
3047 | if (ret & VM_FAULT_ERROR) |
3048 | ret &= VM_FAULT_ERROR; | |
1da177e4 LT |
3049 | goto out; |
3050 | } | |
3051 | ||
3052 | /* No need to invalidate - it was non-present before */ | |
82b0f8c3 | 3053 | update_mmu_cache(vma, vmf->address, vmf->pte); |
65500d23 | 3054 | unlock: |
82b0f8c3 | 3055 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
1da177e4 LT |
3056 | out: |
3057 | return ret; | |
b8107480 | 3058 | out_nomap: |
f627c2f5 | 3059 | mem_cgroup_cancel_charge(page, memcg, false); |
82b0f8c3 | 3060 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
bc43f75c | 3061 | out_page: |
b8107480 | 3062 | unlock_page(page); |
4779cb31 | 3063 | out_release: |
09cbfeaf | 3064 | put_page(page); |
0bcac06f | 3065 | if (page != swapcache && swapcache) { |
4969c119 | 3066 | unlock_page(swapcache); |
09cbfeaf | 3067 | put_page(swapcache); |
4969c119 | 3068 | } |
65500d23 | 3069 | return ret; |
1da177e4 LT |
3070 | } |
3071 | ||
3072 | /* | |
8f4e2101 HD |
3073 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
3074 | * but allow concurrent faults), and pte mapped but not yet locked. | |
3075 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 3076 | */ |
82b0f8c3 | 3077 | static int do_anonymous_page(struct vm_fault *vmf) |
1da177e4 | 3078 | { |
82b0f8c3 | 3079 | struct vm_area_struct *vma = vmf->vma; |
00501b53 | 3080 | struct mem_cgroup *memcg; |
8f4e2101 | 3081 | struct page *page; |
6b31d595 | 3082 | int ret = 0; |
1da177e4 | 3083 | pte_t entry; |
1da177e4 | 3084 | |
6b7339f4 KS |
3085 | /* File mapping without ->vm_ops ? */ |
3086 | if (vma->vm_flags & VM_SHARED) | |
3087 | return VM_FAULT_SIGBUS; | |
3088 | ||
7267ec00 KS |
3089 | /* |
3090 | * Use pte_alloc() instead of pte_alloc_map(). We can't run | |
3091 | * pte_offset_map() on pmds where a huge pmd might be created | |
3092 | * from a different thread. | |
3093 | * | |
3094 | * pte_alloc_map() is safe to use under down_write(mmap_sem) or when | |
3095 | * parallel threads are excluded by other means. | |
3096 | * | |
3097 | * Here we only have down_read(mmap_sem). | |
3098 | */ | |
82b0f8c3 | 3099 | if (pte_alloc(vma->vm_mm, vmf->pmd, vmf->address)) |
7267ec00 KS |
3100 | return VM_FAULT_OOM; |
3101 | ||
3102 | /* See the comment in pte_alloc_one_map() */ | |
82b0f8c3 | 3103 | if (unlikely(pmd_trans_unstable(vmf->pmd))) |
7267ec00 KS |
3104 | return 0; |
3105 | ||
11ac5524 | 3106 | /* Use the zero-page for reads */ |
82b0f8c3 | 3107 | if (!(vmf->flags & FAULT_FLAG_WRITE) && |
bae473a4 | 3108 | !mm_forbids_zeropage(vma->vm_mm)) { |
82b0f8c3 | 3109 | entry = pte_mkspecial(pfn_pte(my_zero_pfn(vmf->address), |
62eede62 | 3110 | vma->vm_page_prot)); |
82b0f8c3 JK |
3111 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
3112 | vmf->address, &vmf->ptl); | |
3113 | if (!pte_none(*vmf->pte)) | |
a13ea5b7 | 3114 | goto unlock; |
6b31d595 MH |
3115 | ret = check_stable_address_space(vma->vm_mm); |
3116 | if (ret) | |
3117 | goto unlock; | |
6b251fc9 AA |
3118 | /* Deliver the page fault to userland, check inside PT lock */ |
3119 | if (userfaultfd_missing(vma)) { | |
82b0f8c3 JK |
3120 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
3121 | return handle_userfault(vmf, VM_UFFD_MISSING); | |
6b251fc9 | 3122 | } |
a13ea5b7 HD |
3123 | goto setpte; |
3124 | } | |
3125 | ||
557ed1fa | 3126 | /* Allocate our own private page. */ |
557ed1fa NP |
3127 | if (unlikely(anon_vma_prepare(vma))) |
3128 | goto oom; | |
82b0f8c3 | 3129 | page = alloc_zeroed_user_highpage_movable(vma, vmf->address); |
557ed1fa NP |
3130 | if (!page) |
3131 | goto oom; | |
eb3c24f3 | 3132 | |
bae473a4 | 3133 | if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL, &memcg, false)) |
eb3c24f3 MG |
3134 | goto oom_free_page; |
3135 | ||
52f37629 MK |
3136 | /* |
3137 | * The memory barrier inside __SetPageUptodate makes sure that | |
3138 | * preceeding stores to the page contents become visible before | |
3139 | * the set_pte_at() write. | |
3140 | */ | |
0ed361de | 3141 | __SetPageUptodate(page); |
8f4e2101 | 3142 | |
557ed1fa | 3143 | entry = mk_pte(page, vma->vm_page_prot); |
1ac0cb5d HD |
3144 | if (vma->vm_flags & VM_WRITE) |
3145 | entry = pte_mkwrite(pte_mkdirty(entry)); | |
1da177e4 | 3146 | |
82b0f8c3 JK |
3147 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, |
3148 | &vmf->ptl); | |
3149 | if (!pte_none(*vmf->pte)) | |
557ed1fa | 3150 | goto release; |
9ba69294 | 3151 | |
6b31d595 MH |
3152 | ret = check_stable_address_space(vma->vm_mm); |
3153 | if (ret) | |
3154 | goto release; | |
3155 | ||
6b251fc9 AA |
3156 | /* Deliver the page fault to userland, check inside PT lock */ |
3157 | if (userfaultfd_missing(vma)) { | |
82b0f8c3 | 3158 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
f627c2f5 | 3159 | mem_cgroup_cancel_charge(page, memcg, false); |
09cbfeaf | 3160 | put_page(page); |
82b0f8c3 | 3161 | return handle_userfault(vmf, VM_UFFD_MISSING); |
6b251fc9 AA |
3162 | } |
3163 | ||
bae473a4 | 3164 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
82b0f8c3 | 3165 | page_add_new_anon_rmap(page, vma, vmf->address, false); |
f627c2f5 | 3166 | mem_cgroup_commit_charge(page, memcg, false, false); |
00501b53 | 3167 | lru_cache_add_active_or_unevictable(page, vma); |
a13ea5b7 | 3168 | setpte: |
82b0f8c3 | 3169 | set_pte_at(vma->vm_mm, vmf->address, vmf->pte, entry); |
1da177e4 LT |
3170 | |
3171 | /* No need to invalidate - it was non-present before */ | |
82b0f8c3 | 3172 | update_mmu_cache(vma, vmf->address, vmf->pte); |
65500d23 | 3173 | unlock: |
82b0f8c3 | 3174 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
6b31d595 | 3175 | return ret; |
8f4e2101 | 3176 | release: |
f627c2f5 | 3177 | mem_cgroup_cancel_charge(page, memcg, false); |
09cbfeaf | 3178 | put_page(page); |
8f4e2101 | 3179 | goto unlock; |
8a9f3ccd | 3180 | oom_free_page: |
09cbfeaf | 3181 | put_page(page); |
65500d23 | 3182 | oom: |
1da177e4 LT |
3183 | return VM_FAULT_OOM; |
3184 | } | |
3185 | ||
9a95f3cf PC |
3186 | /* |
3187 | * The mmap_sem must have been held on entry, and may have been | |
3188 | * released depending on flags and vma->vm_ops->fault() return value. | |
3189 | * See filemap_fault() and __lock_page_retry(). | |
3190 | */ | |
936ca80d | 3191 | static int __do_fault(struct vm_fault *vmf) |
7eae74af | 3192 | { |
82b0f8c3 | 3193 | struct vm_area_struct *vma = vmf->vma; |
7eae74af KS |
3194 | int ret; |
3195 | ||
11bac800 | 3196 | ret = vma->vm_ops->fault(vmf); |
3917048d | 3197 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY | |
b1aa812b | 3198 | VM_FAULT_DONE_COW))) |
bc2466e4 | 3199 | return ret; |
7eae74af | 3200 | |
667240e0 | 3201 | if (unlikely(PageHWPoison(vmf->page))) { |
7eae74af | 3202 | if (ret & VM_FAULT_LOCKED) |
667240e0 JK |
3203 | unlock_page(vmf->page); |
3204 | put_page(vmf->page); | |
936ca80d | 3205 | vmf->page = NULL; |
7eae74af KS |
3206 | return VM_FAULT_HWPOISON; |
3207 | } | |
3208 | ||
3209 | if (unlikely(!(ret & VM_FAULT_LOCKED))) | |
667240e0 | 3210 | lock_page(vmf->page); |
7eae74af | 3211 | else |
667240e0 | 3212 | VM_BUG_ON_PAGE(!PageLocked(vmf->page), vmf->page); |
7eae74af | 3213 | |
7eae74af KS |
3214 | return ret; |
3215 | } | |
3216 | ||
d0f0931d RZ |
3217 | /* |
3218 | * The ordering of these checks is important for pmds with _PAGE_DEVMAP set. | |
3219 | * If we check pmd_trans_unstable() first we will trip the bad_pmd() check | |
3220 | * inside of pmd_none_or_trans_huge_or_clear_bad(). This will end up correctly | |
3221 | * returning 1 but not before it spams dmesg with the pmd_clear_bad() output. | |
3222 | */ | |
3223 | static int pmd_devmap_trans_unstable(pmd_t *pmd) | |
3224 | { | |
3225 | return pmd_devmap(*pmd) || pmd_trans_unstable(pmd); | |
3226 | } | |
3227 | ||
82b0f8c3 | 3228 | static int pte_alloc_one_map(struct vm_fault *vmf) |
7267ec00 | 3229 | { |
82b0f8c3 | 3230 | struct vm_area_struct *vma = vmf->vma; |
7267ec00 | 3231 | |
82b0f8c3 | 3232 | if (!pmd_none(*vmf->pmd)) |
7267ec00 | 3233 | goto map_pte; |
82b0f8c3 JK |
3234 | if (vmf->prealloc_pte) { |
3235 | vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd); | |
3236 | if (unlikely(!pmd_none(*vmf->pmd))) { | |
3237 | spin_unlock(vmf->ptl); | |
7267ec00 KS |
3238 | goto map_pte; |
3239 | } | |
3240 | ||
c4812909 | 3241 | mm_inc_nr_ptes(vma->vm_mm); |
82b0f8c3 JK |
3242 | pmd_populate(vma->vm_mm, vmf->pmd, vmf->prealloc_pte); |
3243 | spin_unlock(vmf->ptl); | |
7f2b6ce8 | 3244 | vmf->prealloc_pte = NULL; |
82b0f8c3 | 3245 | } else if (unlikely(pte_alloc(vma->vm_mm, vmf->pmd, vmf->address))) { |
7267ec00 KS |
3246 | return VM_FAULT_OOM; |
3247 | } | |
3248 | map_pte: | |
3249 | /* | |
3250 | * If a huge pmd materialized under us just retry later. Use | |
d0f0931d RZ |
3251 | * pmd_trans_unstable() via pmd_devmap_trans_unstable() instead of |
3252 | * pmd_trans_huge() to ensure the pmd didn't become pmd_trans_huge | |
3253 | * under us and then back to pmd_none, as a result of MADV_DONTNEED | |
3254 | * running immediately after a huge pmd fault in a different thread of | |
3255 | * this mm, in turn leading to a misleading pmd_trans_huge() retval. | |
3256 | * All we have to ensure is that it is a regular pmd that we can walk | |
3257 | * with pte_offset_map() and we can do that through an atomic read in | |
3258 | * C, which is what pmd_trans_unstable() provides. | |
7267ec00 | 3259 | */ |
d0f0931d | 3260 | if (pmd_devmap_trans_unstable(vmf->pmd)) |
7267ec00 KS |
3261 | return VM_FAULT_NOPAGE; |
3262 | ||
d0f0931d RZ |
3263 | /* |
3264 | * At this point we know that our vmf->pmd points to a page of ptes | |
3265 | * and it cannot become pmd_none(), pmd_devmap() or pmd_trans_huge() | |
3266 | * for the duration of the fault. If a racing MADV_DONTNEED runs and | |
3267 | * we zap the ptes pointed to by our vmf->pmd, the vmf->ptl will still | |
3268 | * be valid and we will re-check to make sure the vmf->pte isn't | |
3269 | * pte_none() under vmf->ptl protection when we return to | |
3270 | * alloc_set_pte(). | |
3271 | */ | |
82b0f8c3 JK |
3272 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, |
3273 | &vmf->ptl); | |
7267ec00 KS |
3274 | return 0; |
3275 | } | |
3276 | ||
e496cf3d | 3277 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
10102459 KS |
3278 | |
3279 | #define HPAGE_CACHE_INDEX_MASK (HPAGE_PMD_NR - 1) | |
3280 | static inline bool transhuge_vma_suitable(struct vm_area_struct *vma, | |
3281 | unsigned long haddr) | |
3282 | { | |
3283 | if (((vma->vm_start >> PAGE_SHIFT) & HPAGE_CACHE_INDEX_MASK) != | |
3284 | (vma->vm_pgoff & HPAGE_CACHE_INDEX_MASK)) | |
3285 | return false; | |
3286 | if (haddr < vma->vm_start || haddr + HPAGE_PMD_SIZE > vma->vm_end) | |
3287 | return false; | |
3288 | return true; | |
3289 | } | |
3290 | ||
82b0f8c3 | 3291 | static void deposit_prealloc_pte(struct vm_fault *vmf) |
953c66c2 | 3292 | { |
82b0f8c3 | 3293 | struct vm_area_struct *vma = vmf->vma; |
953c66c2 | 3294 | |
82b0f8c3 | 3295 | pgtable_trans_huge_deposit(vma->vm_mm, vmf->pmd, vmf->prealloc_pte); |
953c66c2 AK |
3296 | /* |
3297 | * We are going to consume the prealloc table, | |
3298 | * count that as nr_ptes. | |
3299 | */ | |
c4812909 | 3300 | mm_inc_nr_ptes(vma->vm_mm); |
7f2b6ce8 | 3301 | vmf->prealloc_pte = NULL; |
953c66c2 AK |
3302 | } |
3303 | ||
82b0f8c3 | 3304 | static int do_set_pmd(struct vm_fault *vmf, struct page *page) |
10102459 | 3305 | { |
82b0f8c3 JK |
3306 | struct vm_area_struct *vma = vmf->vma; |
3307 | bool write = vmf->flags & FAULT_FLAG_WRITE; | |
3308 | unsigned long haddr = vmf->address & HPAGE_PMD_MASK; | |
10102459 KS |
3309 | pmd_t entry; |
3310 | int i, ret; | |
3311 | ||
3312 | if (!transhuge_vma_suitable(vma, haddr)) | |
3313 | return VM_FAULT_FALLBACK; | |
3314 | ||
3315 | ret = VM_FAULT_FALLBACK; | |
3316 | page = compound_head(page); | |
3317 | ||
953c66c2 AK |
3318 | /* |
3319 | * Archs like ppc64 need additonal space to store information | |
3320 | * related to pte entry. Use the preallocated table for that. | |
3321 | */ | |
82b0f8c3 JK |
3322 | if (arch_needs_pgtable_deposit() && !vmf->prealloc_pte) { |
3323 | vmf->prealloc_pte = pte_alloc_one(vma->vm_mm, vmf->address); | |
3324 | if (!vmf->prealloc_pte) | |
953c66c2 AK |
3325 | return VM_FAULT_OOM; |
3326 | smp_wmb(); /* See comment in __pte_alloc() */ | |
3327 | } | |
3328 | ||
82b0f8c3 JK |
3329 | vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd); |
3330 | if (unlikely(!pmd_none(*vmf->pmd))) | |
10102459 KS |
3331 | goto out; |
3332 | ||
3333 | for (i = 0; i < HPAGE_PMD_NR; i++) | |
3334 | flush_icache_page(vma, page + i); | |
3335 | ||
3336 | entry = mk_huge_pmd(page, vma->vm_page_prot); | |
3337 | if (write) | |
152e93af | 3338 | entry = maybe_pmd_mkwrite(entry, vma, true); |
10102459 KS |
3339 | |
3340 | add_mm_counter(vma->vm_mm, MM_FILEPAGES, HPAGE_PMD_NR); | |
3341 | page_add_file_rmap(page, true); | |
953c66c2 AK |
3342 | /* |
3343 | * deposit and withdraw with pmd lock held | |
3344 | */ | |
3345 | if (arch_needs_pgtable_deposit()) | |
82b0f8c3 | 3346 | deposit_prealloc_pte(vmf); |
10102459 | 3347 | |
82b0f8c3 | 3348 | set_pmd_at(vma->vm_mm, haddr, vmf->pmd, entry); |
10102459 | 3349 | |
82b0f8c3 | 3350 | update_mmu_cache_pmd(vma, haddr, vmf->pmd); |
10102459 KS |
3351 | |
3352 | /* fault is handled */ | |
3353 | ret = 0; | |
95ecedcd | 3354 | count_vm_event(THP_FILE_MAPPED); |
10102459 | 3355 | out: |
82b0f8c3 | 3356 | spin_unlock(vmf->ptl); |
10102459 KS |
3357 | return ret; |
3358 | } | |
3359 | #else | |
82b0f8c3 | 3360 | static int do_set_pmd(struct vm_fault *vmf, struct page *page) |
10102459 KS |
3361 | { |
3362 | BUILD_BUG(); | |
3363 | return 0; | |
3364 | } | |
3365 | #endif | |
3366 | ||
8c6e50b0 | 3367 | /** |
7267ec00 KS |
3368 | * alloc_set_pte - setup new PTE entry for given page and add reverse page |
3369 | * mapping. If needed, the fucntion allocates page table or use pre-allocated. | |
8c6e50b0 | 3370 | * |
82b0f8c3 | 3371 | * @vmf: fault environment |
7267ec00 | 3372 | * @memcg: memcg to charge page (only for private mappings) |
8c6e50b0 | 3373 | * @page: page to map |
8c6e50b0 | 3374 | * |
82b0f8c3 JK |
3375 | * Caller must take care of unlocking vmf->ptl, if vmf->pte is non-NULL on |
3376 | * return. | |
8c6e50b0 KS |
3377 | * |
3378 | * Target users are page handler itself and implementations of | |
3379 | * vm_ops->map_pages. | |
3380 | */ | |
82b0f8c3 | 3381 | int alloc_set_pte(struct vm_fault *vmf, struct mem_cgroup *memcg, |
7267ec00 | 3382 | struct page *page) |
3bb97794 | 3383 | { |
82b0f8c3 JK |
3384 | struct vm_area_struct *vma = vmf->vma; |
3385 | bool write = vmf->flags & FAULT_FLAG_WRITE; | |
3bb97794 | 3386 | pte_t entry; |
10102459 KS |
3387 | int ret; |
3388 | ||
82b0f8c3 | 3389 | if (pmd_none(*vmf->pmd) && PageTransCompound(page) && |
e496cf3d | 3390 | IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) { |
10102459 KS |
3391 | /* THP on COW? */ |
3392 | VM_BUG_ON_PAGE(memcg, page); | |
3393 | ||
82b0f8c3 | 3394 | ret = do_set_pmd(vmf, page); |
10102459 | 3395 | if (ret != VM_FAULT_FALLBACK) |
b0b9b3df | 3396 | return ret; |
10102459 | 3397 | } |
3bb97794 | 3398 | |
82b0f8c3 JK |
3399 | if (!vmf->pte) { |
3400 | ret = pte_alloc_one_map(vmf); | |
7267ec00 | 3401 | if (ret) |
b0b9b3df | 3402 | return ret; |
7267ec00 KS |
3403 | } |
3404 | ||
3405 | /* Re-check under ptl */ | |
b0b9b3df HD |
3406 | if (unlikely(!pte_none(*vmf->pte))) |
3407 | return VM_FAULT_NOPAGE; | |
7267ec00 | 3408 | |
3bb97794 KS |
3409 | flush_icache_page(vma, page); |
3410 | entry = mk_pte(page, vma->vm_page_prot); | |
3411 | if (write) | |
3412 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
bae473a4 KS |
3413 | /* copy-on-write page */ |
3414 | if (write && !(vma->vm_flags & VM_SHARED)) { | |
3bb97794 | 3415 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
82b0f8c3 | 3416 | page_add_new_anon_rmap(page, vma, vmf->address, false); |
7267ec00 KS |
3417 | mem_cgroup_commit_charge(page, memcg, false, false); |
3418 | lru_cache_add_active_or_unevictable(page, vma); | |
3bb97794 | 3419 | } else { |
eca56ff9 | 3420 | inc_mm_counter_fast(vma->vm_mm, mm_counter_file(page)); |
dd78fedd | 3421 | page_add_file_rmap(page, false); |
3bb97794 | 3422 | } |
82b0f8c3 | 3423 | set_pte_at(vma->vm_mm, vmf->address, vmf->pte, entry); |
3bb97794 KS |
3424 | |
3425 | /* no need to invalidate: a not-present page won't be cached */ | |
82b0f8c3 | 3426 | update_mmu_cache(vma, vmf->address, vmf->pte); |
7267ec00 | 3427 | |
b0b9b3df | 3428 | return 0; |
3bb97794 KS |
3429 | } |
3430 | ||
9118c0cb JK |
3431 | |
3432 | /** | |
3433 | * finish_fault - finish page fault once we have prepared the page to fault | |
3434 | * | |
3435 | * @vmf: structure describing the fault | |
3436 | * | |
3437 | * This function handles all that is needed to finish a page fault once the | |
3438 | * page to fault in is prepared. It handles locking of PTEs, inserts PTE for | |
3439 | * given page, adds reverse page mapping, handles memcg charges and LRU | |
3440 | * addition. The function returns 0 on success, VM_FAULT_ code in case of | |
3441 | * error. | |
3442 | * | |
3443 | * The function expects the page to be locked and on success it consumes a | |
3444 | * reference of a page being mapped (for the PTE which maps it). | |
3445 | */ | |
3446 | int finish_fault(struct vm_fault *vmf) | |
3447 | { | |
3448 | struct page *page; | |
6b31d595 | 3449 | int ret = 0; |
9118c0cb JK |
3450 | |
3451 | /* Did we COW the page? */ | |
3452 | if ((vmf->flags & FAULT_FLAG_WRITE) && | |
3453 | !(vmf->vma->vm_flags & VM_SHARED)) | |
3454 | page = vmf->cow_page; | |
3455 | else | |
3456 | page = vmf->page; | |
6b31d595 MH |
3457 | |
3458 | /* | |
3459 | * check even for read faults because we might have lost our CoWed | |
3460 | * page | |
3461 | */ | |
3462 | if (!(vmf->vma->vm_flags & VM_SHARED)) | |
3463 | ret = check_stable_address_space(vmf->vma->vm_mm); | |
3464 | if (!ret) | |
3465 | ret = alloc_set_pte(vmf, vmf->memcg, page); | |
9118c0cb JK |
3466 | if (vmf->pte) |
3467 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
3468 | return ret; | |
3469 | } | |
3470 | ||
3a91053a KS |
3471 | static unsigned long fault_around_bytes __read_mostly = |
3472 | rounddown_pow_of_two(65536); | |
a9b0f861 | 3473 | |
a9b0f861 KS |
3474 | #ifdef CONFIG_DEBUG_FS |
3475 | static int fault_around_bytes_get(void *data, u64 *val) | |
1592eef0 | 3476 | { |
a9b0f861 | 3477 | *val = fault_around_bytes; |
1592eef0 KS |
3478 | return 0; |
3479 | } | |
3480 | ||
b4903d6e AR |
3481 | /* |
3482 | * fault_around_pages() and fault_around_mask() expects fault_around_bytes | |
3483 | * rounded down to nearest page order. It's what do_fault_around() expects to | |
3484 | * see. | |
3485 | */ | |
a9b0f861 | 3486 | static int fault_around_bytes_set(void *data, u64 val) |
1592eef0 | 3487 | { |
a9b0f861 | 3488 | if (val / PAGE_SIZE > PTRS_PER_PTE) |
1592eef0 | 3489 | return -EINVAL; |
b4903d6e AR |
3490 | if (val > PAGE_SIZE) |
3491 | fault_around_bytes = rounddown_pow_of_two(val); | |
3492 | else | |
3493 | fault_around_bytes = PAGE_SIZE; /* rounddown_pow_of_two(0) is undefined */ | |
1592eef0 KS |
3494 | return 0; |
3495 | } | |
0a1345f8 | 3496 | DEFINE_DEBUGFS_ATTRIBUTE(fault_around_bytes_fops, |
a9b0f861 | 3497 | fault_around_bytes_get, fault_around_bytes_set, "%llu\n"); |
1592eef0 KS |
3498 | |
3499 | static int __init fault_around_debugfs(void) | |
3500 | { | |
3501 | void *ret; | |
3502 | ||
0a1345f8 | 3503 | ret = debugfs_create_file_unsafe("fault_around_bytes", 0644, NULL, NULL, |
a9b0f861 | 3504 | &fault_around_bytes_fops); |
1592eef0 | 3505 | if (!ret) |
a9b0f861 | 3506 | pr_warn("Failed to create fault_around_bytes in debugfs"); |
1592eef0 KS |
3507 | return 0; |
3508 | } | |
3509 | late_initcall(fault_around_debugfs); | |
1592eef0 | 3510 | #endif |
8c6e50b0 | 3511 | |
1fdb412b KS |
3512 | /* |
3513 | * do_fault_around() tries to map few pages around the fault address. The hope | |
3514 | * is that the pages will be needed soon and this will lower the number of | |
3515 | * faults to handle. | |
3516 | * | |
3517 | * It uses vm_ops->map_pages() to map the pages, which skips the page if it's | |
3518 | * not ready to be mapped: not up-to-date, locked, etc. | |
3519 | * | |
3520 | * This function is called with the page table lock taken. In the split ptlock | |
3521 | * case the page table lock only protects only those entries which belong to | |
3522 | * the page table corresponding to the fault address. | |
3523 | * | |
3524 | * This function doesn't cross the VMA boundaries, in order to call map_pages() | |
3525 | * only once. | |
3526 | * | |
3527 | * fault_around_pages() defines how many pages we'll try to map. | |
3528 | * do_fault_around() expects it to return a power of two less than or equal to | |
3529 | * PTRS_PER_PTE. | |
3530 | * | |
3531 | * The virtual address of the area that we map is naturally aligned to the | |
3532 | * fault_around_pages() value (and therefore to page order). This way it's | |
3533 | * easier to guarantee that we don't cross page table boundaries. | |
3534 | */ | |
0721ec8b | 3535 | static int do_fault_around(struct vm_fault *vmf) |
8c6e50b0 | 3536 | { |
82b0f8c3 | 3537 | unsigned long address = vmf->address, nr_pages, mask; |
0721ec8b | 3538 | pgoff_t start_pgoff = vmf->pgoff; |
bae473a4 | 3539 | pgoff_t end_pgoff; |
7267ec00 | 3540 | int off, ret = 0; |
8c6e50b0 | 3541 | |
4db0c3c2 | 3542 | nr_pages = READ_ONCE(fault_around_bytes) >> PAGE_SHIFT; |
aecd6f44 KS |
3543 | mask = ~(nr_pages * PAGE_SIZE - 1) & PAGE_MASK; |
3544 | ||
82b0f8c3 JK |
3545 | vmf->address = max(address & mask, vmf->vma->vm_start); |
3546 | off = ((address - vmf->address) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1); | |
bae473a4 | 3547 | start_pgoff -= off; |
8c6e50b0 KS |
3548 | |
3549 | /* | |
bae473a4 KS |
3550 | * end_pgoff is either end of page table or end of vma |
3551 | * or fault_around_pages() from start_pgoff, depending what is nearest. | |
8c6e50b0 | 3552 | */ |
bae473a4 | 3553 | end_pgoff = start_pgoff - |
82b0f8c3 | 3554 | ((vmf->address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)) + |
8c6e50b0 | 3555 | PTRS_PER_PTE - 1; |
82b0f8c3 | 3556 | end_pgoff = min3(end_pgoff, vma_pages(vmf->vma) + vmf->vma->vm_pgoff - 1, |
bae473a4 | 3557 | start_pgoff + nr_pages - 1); |
8c6e50b0 | 3558 | |
82b0f8c3 JK |
3559 | if (pmd_none(*vmf->pmd)) { |
3560 | vmf->prealloc_pte = pte_alloc_one(vmf->vma->vm_mm, | |
3561 | vmf->address); | |
3562 | if (!vmf->prealloc_pte) | |
c5f88bd2 | 3563 | goto out; |
7267ec00 | 3564 | smp_wmb(); /* See comment in __pte_alloc() */ |
8c6e50b0 KS |
3565 | } |
3566 | ||
82b0f8c3 | 3567 | vmf->vma->vm_ops->map_pages(vmf, start_pgoff, end_pgoff); |
7267ec00 | 3568 | |
7267ec00 | 3569 | /* Huge page is mapped? Page fault is solved */ |
82b0f8c3 | 3570 | if (pmd_trans_huge(*vmf->pmd)) { |
7267ec00 KS |
3571 | ret = VM_FAULT_NOPAGE; |
3572 | goto out; | |
3573 | } | |
3574 | ||
3575 | /* ->map_pages() haven't done anything useful. Cold page cache? */ | |
82b0f8c3 | 3576 | if (!vmf->pte) |
7267ec00 KS |
3577 | goto out; |
3578 | ||
3579 | /* check if the page fault is solved */ | |
82b0f8c3 JK |
3580 | vmf->pte -= (vmf->address >> PAGE_SHIFT) - (address >> PAGE_SHIFT); |
3581 | if (!pte_none(*vmf->pte)) | |
7267ec00 | 3582 | ret = VM_FAULT_NOPAGE; |
82b0f8c3 | 3583 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
bae473a4 | 3584 | out: |
82b0f8c3 JK |
3585 | vmf->address = address; |
3586 | vmf->pte = NULL; | |
7267ec00 | 3587 | return ret; |
8c6e50b0 KS |
3588 | } |
3589 | ||
0721ec8b | 3590 | static int do_read_fault(struct vm_fault *vmf) |
e655fb29 | 3591 | { |
82b0f8c3 | 3592 | struct vm_area_struct *vma = vmf->vma; |
8c6e50b0 KS |
3593 | int ret = 0; |
3594 | ||
3595 | /* | |
3596 | * Let's call ->map_pages() first and use ->fault() as fallback | |
3597 | * if page by the offset is not ready to be mapped (cold cache or | |
3598 | * something). | |
3599 | */ | |
9b4bdd2f | 3600 | if (vma->vm_ops->map_pages && fault_around_bytes >> PAGE_SHIFT > 1) { |
0721ec8b | 3601 | ret = do_fault_around(vmf); |
7267ec00 KS |
3602 | if (ret) |
3603 | return ret; | |
8c6e50b0 | 3604 | } |
e655fb29 | 3605 | |
936ca80d | 3606 | ret = __do_fault(vmf); |
e655fb29 KS |
3607 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
3608 | return ret; | |
3609 | ||
9118c0cb | 3610 | ret |= finish_fault(vmf); |
936ca80d | 3611 | unlock_page(vmf->page); |
7267ec00 | 3612 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
936ca80d | 3613 | put_page(vmf->page); |
e655fb29 KS |
3614 | return ret; |
3615 | } | |
3616 | ||
0721ec8b | 3617 | static int do_cow_fault(struct vm_fault *vmf) |
ec47c3b9 | 3618 | { |
82b0f8c3 | 3619 | struct vm_area_struct *vma = vmf->vma; |
ec47c3b9 KS |
3620 | int ret; |
3621 | ||
3622 | if (unlikely(anon_vma_prepare(vma))) | |
3623 | return VM_FAULT_OOM; | |
3624 | ||
936ca80d JK |
3625 | vmf->cow_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vmf->address); |
3626 | if (!vmf->cow_page) | |
ec47c3b9 KS |
3627 | return VM_FAULT_OOM; |
3628 | ||
936ca80d | 3629 | if (mem_cgroup_try_charge(vmf->cow_page, vma->vm_mm, GFP_KERNEL, |
3917048d | 3630 | &vmf->memcg, false)) { |
936ca80d | 3631 | put_page(vmf->cow_page); |
ec47c3b9 KS |
3632 | return VM_FAULT_OOM; |
3633 | } | |
3634 | ||
936ca80d | 3635 | ret = __do_fault(vmf); |
ec47c3b9 KS |
3636 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
3637 | goto uncharge_out; | |
3917048d JK |
3638 | if (ret & VM_FAULT_DONE_COW) |
3639 | return ret; | |
ec47c3b9 | 3640 | |
b1aa812b | 3641 | copy_user_highpage(vmf->cow_page, vmf->page, vmf->address, vma); |
936ca80d | 3642 | __SetPageUptodate(vmf->cow_page); |
ec47c3b9 | 3643 | |
9118c0cb | 3644 | ret |= finish_fault(vmf); |
b1aa812b JK |
3645 | unlock_page(vmf->page); |
3646 | put_page(vmf->page); | |
7267ec00 KS |
3647 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
3648 | goto uncharge_out; | |
ec47c3b9 KS |
3649 | return ret; |
3650 | uncharge_out: | |
3917048d | 3651 | mem_cgroup_cancel_charge(vmf->cow_page, vmf->memcg, false); |
936ca80d | 3652 | put_page(vmf->cow_page); |
ec47c3b9 KS |
3653 | return ret; |
3654 | } | |
3655 | ||
0721ec8b | 3656 | static int do_shared_fault(struct vm_fault *vmf) |
1da177e4 | 3657 | { |
82b0f8c3 | 3658 | struct vm_area_struct *vma = vmf->vma; |
f0c6d4d2 | 3659 | int ret, tmp; |
1d65f86d | 3660 | |
936ca80d | 3661 | ret = __do_fault(vmf); |
7eae74af | 3662 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
f0c6d4d2 | 3663 | return ret; |
1da177e4 LT |
3664 | |
3665 | /* | |
f0c6d4d2 KS |
3666 | * Check if the backing address space wants to know that the page is |
3667 | * about to become writable | |
1da177e4 | 3668 | */ |
fb09a464 | 3669 | if (vma->vm_ops->page_mkwrite) { |
936ca80d | 3670 | unlock_page(vmf->page); |
38b8cb7f | 3671 | tmp = do_page_mkwrite(vmf); |
fb09a464 KS |
3672 | if (unlikely(!tmp || |
3673 | (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { | |
936ca80d | 3674 | put_page(vmf->page); |
fb09a464 | 3675 | return tmp; |
4294621f | 3676 | } |
fb09a464 KS |
3677 | } |
3678 | ||
9118c0cb | 3679 | ret |= finish_fault(vmf); |
7267ec00 KS |
3680 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | |
3681 | VM_FAULT_RETRY))) { | |
936ca80d JK |
3682 | unlock_page(vmf->page); |
3683 | put_page(vmf->page); | |
f0c6d4d2 | 3684 | return ret; |
1da177e4 | 3685 | } |
b827e496 | 3686 | |
97ba0c2b | 3687 | fault_dirty_shared_page(vma, vmf->page); |
1d65f86d | 3688 | return ret; |
54cb8821 | 3689 | } |
d00806b1 | 3690 | |
9a95f3cf PC |
3691 | /* |
3692 | * We enter with non-exclusive mmap_sem (to exclude vma changes, | |
3693 | * but allow concurrent faults). | |
3694 | * The mmap_sem may have been released depending on flags and our | |
3695 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
3696 | */ | |
82b0f8c3 | 3697 | static int do_fault(struct vm_fault *vmf) |
54cb8821 | 3698 | { |
82b0f8c3 | 3699 | struct vm_area_struct *vma = vmf->vma; |
b0b9b3df | 3700 | int ret; |
54cb8821 | 3701 | |
6b7339f4 KS |
3702 | /* The VMA was not fully populated on mmap() or missing VM_DONTEXPAND */ |
3703 | if (!vma->vm_ops->fault) | |
b0b9b3df HD |
3704 | ret = VM_FAULT_SIGBUS; |
3705 | else if (!(vmf->flags & FAULT_FLAG_WRITE)) | |
3706 | ret = do_read_fault(vmf); | |
3707 | else if (!(vma->vm_flags & VM_SHARED)) | |
3708 | ret = do_cow_fault(vmf); | |
3709 | else | |
3710 | ret = do_shared_fault(vmf); | |
3711 | ||
3712 | /* preallocated pagetable is unused: free it */ | |
3713 | if (vmf->prealloc_pte) { | |
3714 | pte_free(vma->vm_mm, vmf->prealloc_pte); | |
7f2b6ce8 | 3715 | vmf->prealloc_pte = NULL; |
b0b9b3df HD |
3716 | } |
3717 | return ret; | |
54cb8821 NP |
3718 | } |
3719 | ||
b19a9939 | 3720 | static int numa_migrate_prep(struct page *page, struct vm_area_struct *vma, |
04bb2f94 RR |
3721 | unsigned long addr, int page_nid, |
3722 | int *flags) | |
9532fec1 MG |
3723 | { |
3724 | get_page(page); | |
3725 | ||
3726 | count_vm_numa_event(NUMA_HINT_FAULTS); | |
04bb2f94 | 3727 | if (page_nid == numa_node_id()) { |
9532fec1 | 3728 | count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL); |
04bb2f94 RR |
3729 | *flags |= TNF_FAULT_LOCAL; |
3730 | } | |
9532fec1 MG |
3731 | |
3732 | return mpol_misplaced(page, vma, addr); | |
3733 | } | |
3734 | ||
2994302b | 3735 | static int do_numa_page(struct vm_fault *vmf) |
d10e63f2 | 3736 | { |
82b0f8c3 | 3737 | struct vm_area_struct *vma = vmf->vma; |
4daae3b4 | 3738 | struct page *page = NULL; |
8191acbd | 3739 | int page_nid = -1; |
90572890 | 3740 | int last_cpupid; |
cbee9f88 | 3741 | int target_nid; |
b8593bfd | 3742 | bool migrated = false; |
cee216a6 | 3743 | pte_t pte; |
288bc549 | 3744 | bool was_writable = pte_savedwrite(vmf->orig_pte); |
6688cc05 | 3745 | int flags = 0; |
d10e63f2 MG |
3746 | |
3747 | /* | |
166f61b9 TH |
3748 | * The "pte" at this point cannot be used safely without |
3749 | * validation through pte_unmap_same(). It's of NUMA type but | |
3750 | * the pfn may be screwed if the read is non atomic. | |
166f61b9 | 3751 | */ |
82b0f8c3 JK |
3752 | vmf->ptl = pte_lockptr(vma->vm_mm, vmf->pmd); |
3753 | spin_lock(vmf->ptl); | |
cee216a6 | 3754 | if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) { |
82b0f8c3 | 3755 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
4daae3b4 MG |
3756 | goto out; |
3757 | } | |
3758 | ||
cee216a6 AK |
3759 | /* |
3760 | * Make it present again, Depending on how arch implementes non | |
3761 | * accessible ptes, some can allow access by kernel mode. | |
3762 | */ | |
3763 | pte = ptep_modify_prot_start(vma->vm_mm, vmf->address, vmf->pte); | |
4d942466 MG |
3764 | pte = pte_modify(pte, vma->vm_page_prot); |
3765 | pte = pte_mkyoung(pte); | |
b191f9b1 MG |
3766 | if (was_writable) |
3767 | pte = pte_mkwrite(pte); | |
cee216a6 | 3768 | ptep_modify_prot_commit(vma->vm_mm, vmf->address, vmf->pte, pte); |
82b0f8c3 | 3769 | update_mmu_cache(vma, vmf->address, vmf->pte); |
d10e63f2 | 3770 | |
82b0f8c3 | 3771 | page = vm_normal_page(vma, vmf->address, pte); |
d10e63f2 | 3772 | if (!page) { |
82b0f8c3 | 3773 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
d10e63f2 MG |
3774 | return 0; |
3775 | } | |
3776 | ||
e81c4802 KS |
3777 | /* TODO: handle PTE-mapped THP */ |
3778 | if (PageCompound(page)) { | |
82b0f8c3 | 3779 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
e81c4802 KS |
3780 | return 0; |
3781 | } | |
3782 | ||
6688cc05 | 3783 | /* |
bea66fbd MG |
3784 | * Avoid grouping on RO pages in general. RO pages shouldn't hurt as |
3785 | * much anyway since they can be in shared cache state. This misses | |
3786 | * the case where a mapping is writable but the process never writes | |
3787 | * to it but pte_write gets cleared during protection updates and | |
3788 | * pte_dirty has unpredictable behaviour between PTE scan updates, | |
3789 | * background writeback, dirty balancing and application behaviour. | |
6688cc05 | 3790 | */ |
d59dc7bc | 3791 | if (!pte_write(pte)) |
6688cc05 PZ |
3792 | flags |= TNF_NO_GROUP; |
3793 | ||
dabe1d99 RR |
3794 | /* |
3795 | * Flag if the page is shared between multiple address spaces. This | |
3796 | * is later used when determining whether to group tasks together | |
3797 | */ | |
3798 | if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED)) | |
3799 | flags |= TNF_SHARED; | |
3800 | ||
90572890 | 3801 | last_cpupid = page_cpupid_last(page); |
8191acbd | 3802 | page_nid = page_to_nid(page); |
82b0f8c3 | 3803 | target_nid = numa_migrate_prep(page, vma, vmf->address, page_nid, |
bae473a4 | 3804 | &flags); |
82b0f8c3 | 3805 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
4daae3b4 | 3806 | if (target_nid == -1) { |
4daae3b4 MG |
3807 | put_page(page); |
3808 | goto out; | |
3809 | } | |
3810 | ||
3811 | /* Migrate to the requested node */ | |
1bc115d8 | 3812 | migrated = migrate_misplaced_page(page, vma, target_nid); |
6688cc05 | 3813 | if (migrated) { |
8191acbd | 3814 | page_nid = target_nid; |
6688cc05 | 3815 | flags |= TNF_MIGRATED; |
074c2381 MG |
3816 | } else |
3817 | flags |= TNF_MIGRATE_FAIL; | |
4daae3b4 MG |
3818 | |
3819 | out: | |
8191acbd | 3820 | if (page_nid != -1) |
6688cc05 | 3821 | task_numa_fault(last_cpupid, page_nid, 1, flags); |
d10e63f2 MG |
3822 | return 0; |
3823 | } | |
3824 | ||
91a90140 | 3825 | static inline int create_huge_pmd(struct vm_fault *vmf) |
b96375f7 | 3826 | { |
f4200391 | 3827 | if (vma_is_anonymous(vmf->vma)) |
82b0f8c3 | 3828 | return do_huge_pmd_anonymous_page(vmf); |
a2d58167 | 3829 | if (vmf->vma->vm_ops->huge_fault) |
c791ace1 | 3830 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD); |
b96375f7 MW |
3831 | return VM_FAULT_FALLBACK; |
3832 | } | |
3833 | ||
82b0f8c3 | 3834 | static int wp_huge_pmd(struct vm_fault *vmf, pmd_t orig_pmd) |
b96375f7 | 3835 | { |
82b0f8c3 JK |
3836 | if (vma_is_anonymous(vmf->vma)) |
3837 | return do_huge_pmd_wp_page(vmf, orig_pmd); | |
a2d58167 | 3838 | if (vmf->vma->vm_ops->huge_fault) |
c791ace1 | 3839 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD); |
af9e4d5f KS |
3840 | |
3841 | /* COW handled on pte level: split pmd */ | |
82b0f8c3 JK |
3842 | VM_BUG_ON_VMA(vmf->vma->vm_flags & VM_SHARED, vmf->vma); |
3843 | __split_huge_pmd(vmf->vma, vmf->pmd, vmf->address, false, NULL); | |
af9e4d5f | 3844 | |
b96375f7 MW |
3845 | return VM_FAULT_FALLBACK; |
3846 | } | |
3847 | ||
38e08854 LS |
3848 | static inline bool vma_is_accessible(struct vm_area_struct *vma) |
3849 | { | |
3850 | return vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE); | |
3851 | } | |
3852 | ||
a00cc7d9 MW |
3853 | static int create_huge_pud(struct vm_fault *vmf) |
3854 | { | |
3855 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
3856 | /* No support for anonymous transparent PUD pages yet */ | |
3857 | if (vma_is_anonymous(vmf->vma)) | |
3858 | return VM_FAULT_FALLBACK; | |
3859 | if (vmf->vma->vm_ops->huge_fault) | |
c791ace1 | 3860 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD); |
a00cc7d9 MW |
3861 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
3862 | return VM_FAULT_FALLBACK; | |
3863 | } | |
3864 | ||
3865 | static int wp_huge_pud(struct vm_fault *vmf, pud_t orig_pud) | |
3866 | { | |
3867 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
3868 | /* No support for anonymous transparent PUD pages yet */ | |
3869 | if (vma_is_anonymous(vmf->vma)) | |
3870 | return VM_FAULT_FALLBACK; | |
3871 | if (vmf->vma->vm_ops->huge_fault) | |
c791ace1 | 3872 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD); |
a00cc7d9 MW |
3873 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
3874 | return VM_FAULT_FALLBACK; | |
3875 | } | |
3876 | ||
1da177e4 LT |
3877 | /* |
3878 | * These routines also need to handle stuff like marking pages dirty | |
3879 | * and/or accessed for architectures that don't do it in hardware (most | |
3880 | * RISC architectures). The early dirtying is also good on the i386. | |
3881 | * | |
3882 | * There is also a hook called "update_mmu_cache()" that architectures | |
3883 | * with external mmu caches can use to update those (ie the Sparc or | |
3884 | * PowerPC hashed page tables that act as extended TLBs). | |
3885 | * | |
7267ec00 KS |
3886 | * We enter with non-exclusive mmap_sem (to exclude vma changes, but allow |
3887 | * concurrent faults). | |
9a95f3cf | 3888 | * |
7267ec00 KS |
3889 | * The mmap_sem may have been released depending on flags and our return value. |
3890 | * See filemap_fault() and __lock_page_or_retry(). | |
1da177e4 | 3891 | */ |
82b0f8c3 | 3892 | static int handle_pte_fault(struct vm_fault *vmf) |
1da177e4 LT |
3893 | { |
3894 | pte_t entry; | |
3895 | ||
82b0f8c3 | 3896 | if (unlikely(pmd_none(*vmf->pmd))) { |
7267ec00 KS |
3897 | /* |
3898 | * Leave __pte_alloc() until later: because vm_ops->fault may | |
3899 | * want to allocate huge page, and if we expose page table | |
3900 | * for an instant, it will be difficult to retract from | |
3901 | * concurrent faults and from rmap lookups. | |
3902 | */ | |
82b0f8c3 | 3903 | vmf->pte = NULL; |
7267ec00 KS |
3904 | } else { |
3905 | /* See comment in pte_alloc_one_map() */ | |
d0f0931d | 3906 | if (pmd_devmap_trans_unstable(vmf->pmd)) |
7267ec00 KS |
3907 | return 0; |
3908 | /* | |
3909 | * A regular pmd is established and it can't morph into a huge | |
3910 | * pmd from under us anymore at this point because we hold the | |
3911 | * mmap_sem read mode and khugepaged takes it in write mode. | |
3912 | * So now it's safe to run pte_offset_map(). | |
3913 | */ | |
82b0f8c3 | 3914 | vmf->pte = pte_offset_map(vmf->pmd, vmf->address); |
2994302b | 3915 | vmf->orig_pte = *vmf->pte; |
7267ec00 KS |
3916 | |
3917 | /* | |
3918 | * some architectures can have larger ptes than wordsize, | |
3919 | * e.g.ppc44x-defconfig has CONFIG_PTE_64BIT=y and | |
b03a0fe0 PM |
3920 | * CONFIG_32BIT=y, so READ_ONCE cannot guarantee atomic |
3921 | * accesses. The code below just needs a consistent view | |
3922 | * for the ifs and we later double check anyway with the | |
7267ec00 KS |
3923 | * ptl lock held. So here a barrier will do. |
3924 | */ | |
3925 | barrier(); | |
2994302b | 3926 | if (pte_none(vmf->orig_pte)) { |
82b0f8c3 JK |
3927 | pte_unmap(vmf->pte); |
3928 | vmf->pte = NULL; | |
65500d23 | 3929 | } |
1da177e4 LT |
3930 | } |
3931 | ||
82b0f8c3 JK |
3932 | if (!vmf->pte) { |
3933 | if (vma_is_anonymous(vmf->vma)) | |
3934 | return do_anonymous_page(vmf); | |
7267ec00 | 3935 | else |
82b0f8c3 | 3936 | return do_fault(vmf); |
7267ec00 KS |
3937 | } |
3938 | ||
2994302b JK |
3939 | if (!pte_present(vmf->orig_pte)) |
3940 | return do_swap_page(vmf); | |
7267ec00 | 3941 | |
2994302b JK |
3942 | if (pte_protnone(vmf->orig_pte) && vma_is_accessible(vmf->vma)) |
3943 | return do_numa_page(vmf); | |
d10e63f2 | 3944 | |
82b0f8c3 JK |
3945 | vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd); |
3946 | spin_lock(vmf->ptl); | |
2994302b | 3947 | entry = vmf->orig_pte; |
82b0f8c3 | 3948 | if (unlikely(!pte_same(*vmf->pte, entry))) |
8f4e2101 | 3949 | goto unlock; |
82b0f8c3 | 3950 | if (vmf->flags & FAULT_FLAG_WRITE) { |
1da177e4 | 3951 | if (!pte_write(entry)) |
2994302b | 3952 | return do_wp_page(vmf); |
1da177e4 LT |
3953 | entry = pte_mkdirty(entry); |
3954 | } | |
3955 | entry = pte_mkyoung(entry); | |
82b0f8c3 JK |
3956 | if (ptep_set_access_flags(vmf->vma, vmf->address, vmf->pte, entry, |
3957 | vmf->flags & FAULT_FLAG_WRITE)) { | |
3958 | update_mmu_cache(vmf->vma, vmf->address, vmf->pte); | |
1a44e149 AA |
3959 | } else { |
3960 | /* | |
3961 | * This is needed only for protection faults but the arch code | |
3962 | * is not yet telling us if this is a protection fault or not. | |
3963 | * This still avoids useless tlb flushes for .text page faults | |
3964 | * with threads. | |
3965 | */ | |
82b0f8c3 JK |
3966 | if (vmf->flags & FAULT_FLAG_WRITE) |
3967 | flush_tlb_fix_spurious_fault(vmf->vma, vmf->address); | |
1a44e149 | 3968 | } |
8f4e2101 | 3969 | unlock: |
82b0f8c3 | 3970 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
83c54070 | 3971 | return 0; |
1da177e4 LT |
3972 | } |
3973 | ||
3974 | /* | |
3975 | * By the time we get here, we already hold the mm semaphore | |
9a95f3cf PC |
3976 | * |
3977 | * The mmap_sem may have been released depending on flags and our | |
3978 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
1da177e4 | 3979 | */ |
dcddffd4 KS |
3980 | static int __handle_mm_fault(struct vm_area_struct *vma, unsigned long address, |
3981 | unsigned int flags) | |
1da177e4 | 3982 | { |
82b0f8c3 | 3983 | struct vm_fault vmf = { |
bae473a4 | 3984 | .vma = vma, |
1a29d85e | 3985 | .address = address & PAGE_MASK, |
bae473a4 | 3986 | .flags = flags, |
0721ec8b | 3987 | .pgoff = linear_page_index(vma, address), |
667240e0 | 3988 | .gfp_mask = __get_fault_gfp_mask(vma), |
bae473a4 | 3989 | }; |
fde26bed | 3990 | unsigned int dirty = flags & FAULT_FLAG_WRITE; |
dcddffd4 | 3991 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 3992 | pgd_t *pgd; |
c2febafc | 3993 | p4d_t *p4d; |
a2d58167 | 3994 | int ret; |
1da177e4 | 3995 | |
1da177e4 | 3996 | pgd = pgd_offset(mm, address); |
c2febafc KS |
3997 | p4d = p4d_alloc(mm, pgd, address); |
3998 | if (!p4d) | |
3999 | return VM_FAULT_OOM; | |
a00cc7d9 | 4000 | |
c2febafc | 4001 | vmf.pud = pud_alloc(mm, p4d, address); |
a00cc7d9 | 4002 | if (!vmf.pud) |
c74df32c | 4003 | return VM_FAULT_OOM; |
a00cc7d9 | 4004 | if (pud_none(*vmf.pud) && transparent_hugepage_enabled(vma)) { |
a00cc7d9 MW |
4005 | ret = create_huge_pud(&vmf); |
4006 | if (!(ret & VM_FAULT_FALLBACK)) | |
4007 | return ret; | |
4008 | } else { | |
4009 | pud_t orig_pud = *vmf.pud; | |
4010 | ||
4011 | barrier(); | |
4012 | if (pud_trans_huge(orig_pud) || pud_devmap(orig_pud)) { | |
a00cc7d9 | 4013 | |
a00cc7d9 MW |
4014 | /* NUMA case for anonymous PUDs would go here */ |
4015 | ||
4016 | if (dirty && !pud_write(orig_pud)) { | |
4017 | ret = wp_huge_pud(&vmf, orig_pud); | |
4018 | if (!(ret & VM_FAULT_FALLBACK)) | |
4019 | return ret; | |
4020 | } else { | |
4021 | huge_pud_set_accessed(&vmf, orig_pud); | |
4022 | return 0; | |
4023 | } | |
4024 | } | |
4025 | } | |
4026 | ||
4027 | vmf.pmd = pmd_alloc(mm, vmf.pud, address); | |
82b0f8c3 | 4028 | if (!vmf.pmd) |
c74df32c | 4029 | return VM_FAULT_OOM; |
82b0f8c3 | 4030 | if (pmd_none(*vmf.pmd) && transparent_hugepage_enabled(vma)) { |
a2d58167 | 4031 | ret = create_huge_pmd(&vmf); |
c0292554 KS |
4032 | if (!(ret & VM_FAULT_FALLBACK)) |
4033 | return ret; | |
71e3aac0 | 4034 | } else { |
82b0f8c3 | 4035 | pmd_t orig_pmd = *vmf.pmd; |
1f1d06c3 | 4036 | |
71e3aac0 | 4037 | barrier(); |
84c3fc4e ZY |
4038 | if (unlikely(is_swap_pmd(orig_pmd))) { |
4039 | VM_BUG_ON(thp_migration_supported() && | |
4040 | !is_pmd_migration_entry(orig_pmd)); | |
4041 | if (is_pmd_migration_entry(orig_pmd)) | |
4042 | pmd_migration_entry_wait(mm, vmf.pmd); | |
4043 | return 0; | |
4044 | } | |
5c7fb56e | 4045 | if (pmd_trans_huge(orig_pmd) || pmd_devmap(orig_pmd)) { |
38e08854 | 4046 | if (pmd_protnone(orig_pmd) && vma_is_accessible(vma)) |
82b0f8c3 | 4047 | return do_huge_pmd_numa_page(&vmf, orig_pmd); |
d10e63f2 | 4048 | |
fde26bed | 4049 | if (dirty && !pmd_write(orig_pmd)) { |
82b0f8c3 | 4050 | ret = wp_huge_pmd(&vmf, orig_pmd); |
9845cbbd KS |
4051 | if (!(ret & VM_FAULT_FALLBACK)) |
4052 | return ret; | |
a1dd450b | 4053 | } else { |
82b0f8c3 | 4054 | huge_pmd_set_accessed(&vmf, orig_pmd); |
9845cbbd | 4055 | return 0; |
1f1d06c3 | 4056 | } |
71e3aac0 AA |
4057 | } |
4058 | } | |
4059 | ||
82b0f8c3 | 4060 | return handle_pte_fault(&vmf); |
1da177e4 LT |
4061 | } |
4062 | ||
9a95f3cf PC |
4063 | /* |
4064 | * By the time we get here, we already hold the mm semaphore | |
4065 | * | |
4066 | * The mmap_sem may have been released depending on flags and our | |
4067 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
4068 | */ | |
dcddffd4 KS |
4069 | int handle_mm_fault(struct vm_area_struct *vma, unsigned long address, |
4070 | unsigned int flags) | |
519e5247 JW |
4071 | { |
4072 | int ret; | |
4073 | ||
4074 | __set_current_state(TASK_RUNNING); | |
4075 | ||
4076 | count_vm_event(PGFAULT); | |
2262185c | 4077 | count_memcg_event_mm(vma->vm_mm, PGFAULT); |
519e5247 JW |
4078 | |
4079 | /* do counter updates before entering really critical section. */ | |
4080 | check_sync_rss_stat(current); | |
4081 | ||
de0c799b LD |
4082 | if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE, |
4083 | flags & FAULT_FLAG_INSTRUCTION, | |
4084 | flags & FAULT_FLAG_REMOTE)) | |
4085 | return VM_FAULT_SIGSEGV; | |
4086 | ||
519e5247 JW |
4087 | /* |
4088 | * Enable the memcg OOM handling for faults triggered in user | |
4089 | * space. Kernel faults are handled more gracefully. | |
4090 | */ | |
4091 | if (flags & FAULT_FLAG_USER) | |
49426420 | 4092 | mem_cgroup_oom_enable(); |
519e5247 | 4093 | |
bae473a4 KS |
4094 | if (unlikely(is_vm_hugetlb_page(vma))) |
4095 | ret = hugetlb_fault(vma->vm_mm, vma, address, flags); | |
4096 | else | |
4097 | ret = __handle_mm_fault(vma, address, flags); | |
519e5247 | 4098 | |
49426420 JW |
4099 | if (flags & FAULT_FLAG_USER) { |
4100 | mem_cgroup_oom_disable(); | |
166f61b9 TH |
4101 | /* |
4102 | * The task may have entered a memcg OOM situation but | |
4103 | * if the allocation error was handled gracefully (no | |
4104 | * VM_FAULT_OOM), there is no need to kill anything. | |
4105 | * Just clean up the OOM state peacefully. | |
4106 | */ | |
4107 | if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM)) | |
4108 | mem_cgroup_oom_synchronize(false); | |
49426420 | 4109 | } |
3812c8c8 | 4110 | |
519e5247 JW |
4111 | return ret; |
4112 | } | |
e1d6d01a | 4113 | EXPORT_SYMBOL_GPL(handle_mm_fault); |
519e5247 | 4114 | |
90eceff1 KS |
4115 | #ifndef __PAGETABLE_P4D_FOLDED |
4116 | /* | |
4117 | * Allocate p4d page table. | |
4118 | * We've already handled the fast-path in-line. | |
4119 | */ | |
4120 | int __p4d_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) | |
4121 | { | |
4122 | p4d_t *new = p4d_alloc_one(mm, address); | |
4123 | if (!new) | |
4124 | return -ENOMEM; | |
4125 | ||
4126 | smp_wmb(); /* See comment in __pte_alloc */ | |
4127 | ||
4128 | spin_lock(&mm->page_table_lock); | |
4129 | if (pgd_present(*pgd)) /* Another has populated it */ | |
4130 | p4d_free(mm, new); | |
4131 | else | |
4132 | pgd_populate(mm, pgd, new); | |
4133 | spin_unlock(&mm->page_table_lock); | |
4134 | return 0; | |
4135 | } | |
4136 | #endif /* __PAGETABLE_P4D_FOLDED */ | |
4137 | ||
1da177e4 LT |
4138 | #ifndef __PAGETABLE_PUD_FOLDED |
4139 | /* | |
4140 | * Allocate page upper directory. | |
872fec16 | 4141 | * We've already handled the fast-path in-line. |
1da177e4 | 4142 | */ |
c2febafc | 4143 | int __pud_alloc(struct mm_struct *mm, p4d_t *p4d, unsigned long address) |
1da177e4 | 4144 | { |
c74df32c HD |
4145 | pud_t *new = pud_alloc_one(mm, address); |
4146 | if (!new) | |
1bb3630e | 4147 | return -ENOMEM; |
1da177e4 | 4148 | |
362a61ad NP |
4149 | smp_wmb(); /* See comment in __pte_alloc */ |
4150 | ||
872fec16 | 4151 | spin_lock(&mm->page_table_lock); |
c2febafc | 4152 | #ifndef __ARCH_HAS_5LEVEL_HACK |
b4e98d9a KS |
4153 | if (!p4d_present(*p4d)) { |
4154 | mm_inc_nr_puds(mm); | |
c2febafc | 4155 | p4d_populate(mm, p4d, new); |
b4e98d9a | 4156 | } else /* Another has populated it */ |
5e541973 | 4157 | pud_free(mm, new); |
b4e98d9a KS |
4158 | #else |
4159 | if (!pgd_present(*p4d)) { | |
4160 | mm_inc_nr_puds(mm); | |
c2febafc | 4161 | pgd_populate(mm, p4d, new); |
b4e98d9a KS |
4162 | } else /* Another has populated it */ |
4163 | pud_free(mm, new); | |
c2febafc | 4164 | #endif /* __ARCH_HAS_5LEVEL_HACK */ |
c74df32c | 4165 | spin_unlock(&mm->page_table_lock); |
1bb3630e | 4166 | return 0; |
1da177e4 LT |
4167 | } |
4168 | #endif /* __PAGETABLE_PUD_FOLDED */ | |
4169 | ||
4170 | #ifndef __PAGETABLE_PMD_FOLDED | |
4171 | /* | |
4172 | * Allocate page middle directory. | |
872fec16 | 4173 | * We've already handled the fast-path in-line. |
1da177e4 | 4174 | */ |
1bb3630e | 4175 | int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) |
1da177e4 | 4176 | { |
a00cc7d9 | 4177 | spinlock_t *ptl; |
c74df32c HD |
4178 | pmd_t *new = pmd_alloc_one(mm, address); |
4179 | if (!new) | |
1bb3630e | 4180 | return -ENOMEM; |
1da177e4 | 4181 | |
362a61ad NP |
4182 | smp_wmb(); /* See comment in __pte_alloc */ |
4183 | ||
a00cc7d9 | 4184 | ptl = pud_lock(mm, pud); |
1da177e4 | 4185 | #ifndef __ARCH_HAS_4LEVEL_HACK |
dc6c9a35 KS |
4186 | if (!pud_present(*pud)) { |
4187 | mm_inc_nr_pmds(mm); | |
1bb3630e | 4188 | pud_populate(mm, pud, new); |
dc6c9a35 | 4189 | } else /* Another has populated it */ |
5e541973 | 4190 | pmd_free(mm, new); |
dc6c9a35 KS |
4191 | #else |
4192 | if (!pgd_present(*pud)) { | |
4193 | mm_inc_nr_pmds(mm); | |
1bb3630e | 4194 | pgd_populate(mm, pud, new); |
dc6c9a35 KS |
4195 | } else /* Another has populated it */ |
4196 | pmd_free(mm, new); | |
1da177e4 | 4197 | #endif /* __ARCH_HAS_4LEVEL_HACK */ |
a00cc7d9 | 4198 | spin_unlock(ptl); |
1bb3630e | 4199 | return 0; |
e0f39591 | 4200 | } |
1da177e4 LT |
4201 | #endif /* __PAGETABLE_PMD_FOLDED */ |
4202 | ||
09796395 | 4203 | static int __follow_pte_pmd(struct mm_struct *mm, unsigned long address, |
a4d1a885 JG |
4204 | unsigned long *start, unsigned long *end, |
4205 | pte_t **ptepp, pmd_t **pmdpp, spinlock_t **ptlp) | |
f8ad0f49 JW |
4206 | { |
4207 | pgd_t *pgd; | |
c2febafc | 4208 | p4d_t *p4d; |
f8ad0f49 JW |
4209 | pud_t *pud; |
4210 | pmd_t *pmd; | |
4211 | pte_t *ptep; | |
4212 | ||
4213 | pgd = pgd_offset(mm, address); | |
4214 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
4215 | goto out; | |
4216 | ||
c2febafc KS |
4217 | p4d = p4d_offset(pgd, address); |
4218 | if (p4d_none(*p4d) || unlikely(p4d_bad(*p4d))) | |
4219 | goto out; | |
4220 | ||
4221 | pud = pud_offset(p4d, address); | |
f8ad0f49 JW |
4222 | if (pud_none(*pud) || unlikely(pud_bad(*pud))) |
4223 | goto out; | |
4224 | ||
4225 | pmd = pmd_offset(pud, address); | |
f66055ab | 4226 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
f8ad0f49 | 4227 | |
09796395 RZ |
4228 | if (pmd_huge(*pmd)) { |
4229 | if (!pmdpp) | |
4230 | goto out; | |
4231 | ||
a4d1a885 JG |
4232 | if (start && end) { |
4233 | *start = address & PMD_MASK; | |
4234 | *end = *start + PMD_SIZE; | |
4235 | mmu_notifier_invalidate_range_start(mm, *start, *end); | |
4236 | } | |
09796395 RZ |
4237 | *ptlp = pmd_lock(mm, pmd); |
4238 | if (pmd_huge(*pmd)) { | |
4239 | *pmdpp = pmd; | |
4240 | return 0; | |
4241 | } | |
4242 | spin_unlock(*ptlp); | |
a4d1a885 JG |
4243 | if (start && end) |
4244 | mmu_notifier_invalidate_range_end(mm, *start, *end); | |
09796395 RZ |
4245 | } |
4246 | ||
4247 | if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) | |
f8ad0f49 JW |
4248 | goto out; |
4249 | ||
a4d1a885 JG |
4250 | if (start && end) { |
4251 | *start = address & PAGE_MASK; | |
4252 | *end = *start + PAGE_SIZE; | |
4253 | mmu_notifier_invalidate_range_start(mm, *start, *end); | |
4254 | } | |
f8ad0f49 | 4255 | ptep = pte_offset_map_lock(mm, pmd, address, ptlp); |
f8ad0f49 JW |
4256 | if (!pte_present(*ptep)) |
4257 | goto unlock; | |
4258 | *ptepp = ptep; | |
4259 | return 0; | |
4260 | unlock: | |
4261 | pte_unmap_unlock(ptep, *ptlp); | |
a4d1a885 JG |
4262 | if (start && end) |
4263 | mmu_notifier_invalidate_range_end(mm, *start, *end); | |
f8ad0f49 JW |
4264 | out: |
4265 | return -EINVAL; | |
4266 | } | |
4267 | ||
f729c8c9 RZ |
4268 | static inline int follow_pte(struct mm_struct *mm, unsigned long address, |
4269 | pte_t **ptepp, spinlock_t **ptlp) | |
1b36ba81 NK |
4270 | { |
4271 | int res; | |
4272 | ||
4273 | /* (void) is needed to make gcc happy */ | |
4274 | (void) __cond_lock(*ptlp, | |
a4d1a885 JG |
4275 | !(res = __follow_pte_pmd(mm, address, NULL, NULL, |
4276 | ptepp, NULL, ptlp))); | |
09796395 RZ |
4277 | return res; |
4278 | } | |
4279 | ||
4280 | int follow_pte_pmd(struct mm_struct *mm, unsigned long address, | |
a4d1a885 | 4281 | unsigned long *start, unsigned long *end, |
09796395 RZ |
4282 | pte_t **ptepp, pmd_t **pmdpp, spinlock_t **ptlp) |
4283 | { | |
4284 | int res; | |
4285 | ||
4286 | /* (void) is needed to make gcc happy */ | |
4287 | (void) __cond_lock(*ptlp, | |
a4d1a885 JG |
4288 | !(res = __follow_pte_pmd(mm, address, start, end, |
4289 | ptepp, pmdpp, ptlp))); | |
1b36ba81 NK |
4290 | return res; |
4291 | } | |
09796395 | 4292 | EXPORT_SYMBOL(follow_pte_pmd); |
1b36ba81 | 4293 | |
3b6748e2 JW |
4294 | /** |
4295 | * follow_pfn - look up PFN at a user virtual address | |
4296 | * @vma: memory mapping | |
4297 | * @address: user virtual address | |
4298 | * @pfn: location to store found PFN | |
4299 | * | |
4300 | * Only IO mappings and raw PFN mappings are allowed. | |
4301 | * | |
4302 | * Returns zero and the pfn at @pfn on success, -ve otherwise. | |
4303 | */ | |
4304 | int follow_pfn(struct vm_area_struct *vma, unsigned long address, | |
4305 | unsigned long *pfn) | |
4306 | { | |
4307 | int ret = -EINVAL; | |
4308 | spinlock_t *ptl; | |
4309 | pte_t *ptep; | |
4310 | ||
4311 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) | |
4312 | return ret; | |
4313 | ||
4314 | ret = follow_pte(vma->vm_mm, address, &ptep, &ptl); | |
4315 | if (ret) | |
4316 | return ret; | |
4317 | *pfn = pte_pfn(*ptep); | |
4318 | pte_unmap_unlock(ptep, ptl); | |
4319 | return 0; | |
4320 | } | |
4321 | EXPORT_SYMBOL(follow_pfn); | |
4322 | ||
28b2ee20 | 4323 | #ifdef CONFIG_HAVE_IOREMAP_PROT |
d87fe660 | 4324 | int follow_phys(struct vm_area_struct *vma, |
4325 | unsigned long address, unsigned int flags, | |
4326 | unsigned long *prot, resource_size_t *phys) | |
28b2ee20 | 4327 | { |
03668a4d | 4328 | int ret = -EINVAL; |
28b2ee20 RR |
4329 | pte_t *ptep, pte; |
4330 | spinlock_t *ptl; | |
28b2ee20 | 4331 | |
d87fe660 | 4332 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) |
4333 | goto out; | |
28b2ee20 | 4334 | |
03668a4d | 4335 | if (follow_pte(vma->vm_mm, address, &ptep, &ptl)) |
d87fe660 | 4336 | goto out; |
28b2ee20 | 4337 | pte = *ptep; |
03668a4d | 4338 | |
28b2ee20 RR |
4339 | if ((flags & FOLL_WRITE) && !pte_write(pte)) |
4340 | goto unlock; | |
28b2ee20 RR |
4341 | |
4342 | *prot = pgprot_val(pte_pgprot(pte)); | |
03668a4d | 4343 | *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; |
28b2ee20 | 4344 | |
03668a4d | 4345 | ret = 0; |
28b2ee20 RR |
4346 | unlock: |
4347 | pte_unmap_unlock(ptep, ptl); | |
4348 | out: | |
d87fe660 | 4349 | return ret; |
28b2ee20 RR |
4350 | } |
4351 | ||
4352 | int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, | |
4353 | void *buf, int len, int write) | |
4354 | { | |
4355 | resource_size_t phys_addr; | |
4356 | unsigned long prot = 0; | |
2bc7273b | 4357 | void __iomem *maddr; |
28b2ee20 RR |
4358 | int offset = addr & (PAGE_SIZE-1); |
4359 | ||
d87fe660 | 4360 | if (follow_phys(vma, addr, write, &prot, &phys_addr)) |
28b2ee20 RR |
4361 | return -EINVAL; |
4362 | ||
9cb12d7b | 4363 | maddr = ioremap_prot(phys_addr, PAGE_ALIGN(len + offset), prot); |
28b2ee20 RR |
4364 | if (write) |
4365 | memcpy_toio(maddr + offset, buf, len); | |
4366 | else | |
4367 | memcpy_fromio(buf, maddr + offset, len); | |
4368 | iounmap(maddr); | |
4369 | ||
4370 | return len; | |
4371 | } | |
5a73633e | 4372 | EXPORT_SYMBOL_GPL(generic_access_phys); |
28b2ee20 RR |
4373 | #endif |
4374 | ||
0ec76a11 | 4375 | /* |
206cb636 SW |
4376 | * Access another process' address space as given in mm. If non-NULL, use the |
4377 | * given task for page fault accounting. | |
0ec76a11 | 4378 | */ |
84d77d3f | 4379 | int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, |
442486ec | 4380 | unsigned long addr, void *buf, int len, unsigned int gup_flags) |
0ec76a11 | 4381 | { |
0ec76a11 | 4382 | struct vm_area_struct *vma; |
0ec76a11 | 4383 | void *old_buf = buf; |
442486ec | 4384 | int write = gup_flags & FOLL_WRITE; |
0ec76a11 | 4385 | |
0ec76a11 | 4386 | down_read(&mm->mmap_sem); |
183ff22b | 4387 | /* ignore errors, just check how much was successfully transferred */ |
0ec76a11 DH |
4388 | while (len) { |
4389 | int bytes, ret, offset; | |
4390 | void *maddr; | |
28b2ee20 | 4391 | struct page *page = NULL; |
0ec76a11 | 4392 | |
1e987790 | 4393 | ret = get_user_pages_remote(tsk, mm, addr, 1, |
5b56d49f | 4394 | gup_flags, &page, &vma, NULL); |
28b2ee20 | 4395 | if (ret <= 0) { |
dbffcd03 RR |
4396 | #ifndef CONFIG_HAVE_IOREMAP_PROT |
4397 | break; | |
4398 | #else | |
28b2ee20 RR |
4399 | /* |
4400 | * Check if this is a VM_IO | VM_PFNMAP VMA, which | |
4401 | * we can access using slightly different code. | |
4402 | */ | |
28b2ee20 | 4403 | vma = find_vma(mm, addr); |
fe936dfc | 4404 | if (!vma || vma->vm_start > addr) |
28b2ee20 RR |
4405 | break; |
4406 | if (vma->vm_ops && vma->vm_ops->access) | |
4407 | ret = vma->vm_ops->access(vma, addr, buf, | |
4408 | len, write); | |
4409 | if (ret <= 0) | |
28b2ee20 RR |
4410 | break; |
4411 | bytes = ret; | |
dbffcd03 | 4412 | #endif |
0ec76a11 | 4413 | } else { |
28b2ee20 RR |
4414 | bytes = len; |
4415 | offset = addr & (PAGE_SIZE-1); | |
4416 | if (bytes > PAGE_SIZE-offset) | |
4417 | bytes = PAGE_SIZE-offset; | |
4418 | ||
4419 | maddr = kmap(page); | |
4420 | if (write) { | |
4421 | copy_to_user_page(vma, page, addr, | |
4422 | maddr + offset, buf, bytes); | |
4423 | set_page_dirty_lock(page); | |
4424 | } else { | |
4425 | copy_from_user_page(vma, page, addr, | |
4426 | buf, maddr + offset, bytes); | |
4427 | } | |
4428 | kunmap(page); | |
09cbfeaf | 4429 | put_page(page); |
0ec76a11 | 4430 | } |
0ec76a11 DH |
4431 | len -= bytes; |
4432 | buf += bytes; | |
4433 | addr += bytes; | |
4434 | } | |
4435 | up_read(&mm->mmap_sem); | |
0ec76a11 DH |
4436 | |
4437 | return buf - old_buf; | |
4438 | } | |
03252919 | 4439 | |
5ddd36b9 | 4440 | /** |
ae91dbfc | 4441 | * access_remote_vm - access another process' address space |
5ddd36b9 SW |
4442 | * @mm: the mm_struct of the target address space |
4443 | * @addr: start address to access | |
4444 | * @buf: source or destination buffer | |
4445 | * @len: number of bytes to transfer | |
6347e8d5 | 4446 | * @gup_flags: flags modifying lookup behaviour |
5ddd36b9 SW |
4447 | * |
4448 | * The caller must hold a reference on @mm. | |
4449 | */ | |
4450 | int access_remote_vm(struct mm_struct *mm, unsigned long addr, | |
6347e8d5 | 4451 | void *buf, int len, unsigned int gup_flags) |
5ddd36b9 | 4452 | { |
6347e8d5 | 4453 | return __access_remote_vm(NULL, mm, addr, buf, len, gup_flags); |
5ddd36b9 SW |
4454 | } |
4455 | ||
206cb636 SW |
4456 | /* |
4457 | * Access another process' address space. | |
4458 | * Source/target buffer must be kernel space, | |
4459 | * Do not walk the page table directly, use get_user_pages | |
4460 | */ | |
4461 | int access_process_vm(struct task_struct *tsk, unsigned long addr, | |
f307ab6d | 4462 | void *buf, int len, unsigned int gup_flags) |
206cb636 SW |
4463 | { |
4464 | struct mm_struct *mm; | |
4465 | int ret; | |
4466 | ||
4467 | mm = get_task_mm(tsk); | |
4468 | if (!mm) | |
4469 | return 0; | |
4470 | ||
f307ab6d | 4471 | ret = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags); |
442486ec | 4472 | |
206cb636 SW |
4473 | mmput(mm); |
4474 | ||
4475 | return ret; | |
4476 | } | |
fcd35857 | 4477 | EXPORT_SYMBOL_GPL(access_process_vm); |
206cb636 | 4478 | |
03252919 AK |
4479 | /* |
4480 | * Print the name of a VMA. | |
4481 | */ | |
4482 | void print_vma_addr(char *prefix, unsigned long ip) | |
4483 | { | |
4484 | struct mm_struct *mm = current->mm; | |
4485 | struct vm_area_struct *vma; | |
4486 | ||
e8bff74a | 4487 | /* |
0a7f682d | 4488 | * we might be running from an atomic context so we cannot sleep |
e8bff74a | 4489 | */ |
0a7f682d | 4490 | if (!down_read_trylock(&mm->mmap_sem)) |
e8bff74a IM |
4491 | return; |
4492 | ||
03252919 AK |
4493 | vma = find_vma(mm, ip); |
4494 | if (vma && vma->vm_file) { | |
4495 | struct file *f = vma->vm_file; | |
0a7f682d | 4496 | char *buf = (char *)__get_free_page(GFP_NOWAIT); |
03252919 | 4497 | if (buf) { |
2fbc57c5 | 4498 | char *p; |
03252919 | 4499 | |
9bf39ab2 | 4500 | p = file_path(f, buf, PAGE_SIZE); |
03252919 AK |
4501 | if (IS_ERR(p)) |
4502 | p = "?"; | |
2fbc57c5 | 4503 | printk("%s%s[%lx+%lx]", prefix, kbasename(p), |
03252919 AK |
4504 | vma->vm_start, |
4505 | vma->vm_end - vma->vm_start); | |
4506 | free_page((unsigned long)buf); | |
4507 | } | |
4508 | } | |
51a07e50 | 4509 | up_read(&mm->mmap_sem); |
03252919 | 4510 | } |
3ee1afa3 | 4511 | |
662bbcb2 | 4512 | #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP) |
9ec23531 | 4513 | void __might_fault(const char *file, int line) |
3ee1afa3 | 4514 | { |
95156f00 PZ |
4515 | /* |
4516 | * Some code (nfs/sunrpc) uses socket ops on kernel memory while | |
4517 | * holding the mmap_sem, this is safe because kernel memory doesn't | |
4518 | * get paged out, therefore we'll never actually fault, and the | |
4519 | * below annotations will generate false positives. | |
4520 | */ | |
db68ce10 | 4521 | if (uaccess_kernel()) |
95156f00 | 4522 | return; |
9ec23531 | 4523 | if (pagefault_disabled()) |
662bbcb2 | 4524 | return; |
9ec23531 DH |
4525 | __might_sleep(file, line, 0); |
4526 | #if defined(CONFIG_DEBUG_ATOMIC_SLEEP) | |
662bbcb2 | 4527 | if (current->mm) |
3ee1afa3 | 4528 | might_lock_read(¤t->mm->mmap_sem); |
9ec23531 | 4529 | #endif |
3ee1afa3 | 4530 | } |
9ec23531 | 4531 | EXPORT_SYMBOL(__might_fault); |
3ee1afa3 | 4532 | #endif |
47ad8475 AA |
4533 | |
4534 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) | |
4535 | static void clear_gigantic_page(struct page *page, | |
4536 | unsigned long addr, | |
4537 | unsigned int pages_per_huge_page) | |
4538 | { | |
4539 | int i; | |
4540 | struct page *p = page; | |
4541 | ||
4542 | might_sleep(); | |
4543 | for (i = 0; i < pages_per_huge_page; | |
4544 | i++, p = mem_map_next(p, page, i)) { | |
4545 | cond_resched(); | |
4546 | clear_user_highpage(p, addr + i * PAGE_SIZE); | |
4547 | } | |
4548 | } | |
4549 | void clear_huge_page(struct page *page, | |
c79b57e4 | 4550 | unsigned long addr_hint, unsigned int pages_per_huge_page) |
47ad8475 | 4551 | { |
c79b57e4 HY |
4552 | int i, n, base, l; |
4553 | unsigned long addr = addr_hint & | |
4554 | ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); | |
47ad8475 AA |
4555 | |
4556 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { | |
4557 | clear_gigantic_page(page, addr, pages_per_huge_page); | |
4558 | return; | |
4559 | } | |
4560 | ||
c79b57e4 | 4561 | /* Clear sub-page to access last to keep its cache lines hot */ |
47ad8475 | 4562 | might_sleep(); |
c79b57e4 HY |
4563 | n = (addr_hint - addr) / PAGE_SIZE; |
4564 | if (2 * n <= pages_per_huge_page) { | |
4565 | /* If sub-page to access in first half of huge page */ | |
4566 | base = 0; | |
4567 | l = n; | |
4568 | /* Clear sub-pages at the end of huge page */ | |
4569 | for (i = pages_per_huge_page - 1; i >= 2 * n; i--) { | |
4570 | cond_resched(); | |
4571 | clear_user_highpage(page + i, addr + i * PAGE_SIZE); | |
4572 | } | |
4573 | } else { | |
4574 | /* If sub-page to access in second half of huge page */ | |
4575 | base = pages_per_huge_page - 2 * (pages_per_huge_page - n); | |
4576 | l = pages_per_huge_page - n; | |
4577 | /* Clear sub-pages at the begin of huge page */ | |
4578 | for (i = 0; i < base; i++) { | |
4579 | cond_resched(); | |
4580 | clear_user_highpage(page + i, addr + i * PAGE_SIZE); | |
4581 | } | |
4582 | } | |
4583 | /* | |
4584 | * Clear remaining sub-pages in left-right-left-right pattern | |
4585 | * towards the sub-page to access | |
4586 | */ | |
4587 | for (i = 0; i < l; i++) { | |
4588 | int left_idx = base + i; | |
4589 | int right_idx = base + 2 * l - 1 - i; | |
4590 | ||
4591 | cond_resched(); | |
4592 | clear_user_highpage(page + left_idx, | |
4593 | addr + left_idx * PAGE_SIZE); | |
47ad8475 | 4594 | cond_resched(); |
c79b57e4 HY |
4595 | clear_user_highpage(page + right_idx, |
4596 | addr + right_idx * PAGE_SIZE); | |
47ad8475 AA |
4597 | } |
4598 | } | |
4599 | ||
4600 | static void copy_user_gigantic_page(struct page *dst, struct page *src, | |
4601 | unsigned long addr, | |
4602 | struct vm_area_struct *vma, | |
4603 | unsigned int pages_per_huge_page) | |
4604 | { | |
4605 | int i; | |
4606 | struct page *dst_base = dst; | |
4607 | struct page *src_base = src; | |
4608 | ||
4609 | for (i = 0; i < pages_per_huge_page; ) { | |
4610 | cond_resched(); | |
4611 | copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma); | |
4612 | ||
4613 | i++; | |
4614 | dst = mem_map_next(dst, dst_base, i); | |
4615 | src = mem_map_next(src, src_base, i); | |
4616 | } | |
4617 | } | |
4618 | ||
4619 | void copy_user_huge_page(struct page *dst, struct page *src, | |
4620 | unsigned long addr, struct vm_area_struct *vma, | |
4621 | unsigned int pages_per_huge_page) | |
4622 | { | |
4623 | int i; | |
4624 | ||
4625 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { | |
4626 | copy_user_gigantic_page(dst, src, addr, vma, | |
4627 | pages_per_huge_page); | |
4628 | return; | |
4629 | } | |
4630 | ||
4631 | might_sleep(); | |
4632 | for (i = 0; i < pages_per_huge_page; i++) { | |
4633 | cond_resched(); | |
4634 | copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE, vma); | |
4635 | } | |
4636 | } | |
fa4d75c1 MK |
4637 | |
4638 | long copy_huge_page_from_user(struct page *dst_page, | |
4639 | const void __user *usr_src, | |
810a56b9 MK |
4640 | unsigned int pages_per_huge_page, |
4641 | bool allow_pagefault) | |
fa4d75c1 MK |
4642 | { |
4643 | void *src = (void *)usr_src; | |
4644 | void *page_kaddr; | |
4645 | unsigned long i, rc = 0; | |
4646 | unsigned long ret_val = pages_per_huge_page * PAGE_SIZE; | |
4647 | ||
4648 | for (i = 0; i < pages_per_huge_page; i++) { | |
810a56b9 MK |
4649 | if (allow_pagefault) |
4650 | page_kaddr = kmap(dst_page + i); | |
4651 | else | |
4652 | page_kaddr = kmap_atomic(dst_page + i); | |
fa4d75c1 MK |
4653 | rc = copy_from_user(page_kaddr, |
4654 | (const void __user *)(src + i * PAGE_SIZE), | |
4655 | PAGE_SIZE); | |
810a56b9 MK |
4656 | if (allow_pagefault) |
4657 | kunmap(dst_page + i); | |
4658 | else | |
4659 | kunmap_atomic(page_kaddr); | |
fa4d75c1 MK |
4660 | |
4661 | ret_val -= (PAGE_SIZE - rc); | |
4662 | if (rc) | |
4663 | break; | |
4664 | ||
4665 | cond_resched(); | |
4666 | } | |
4667 | return ret_val; | |
4668 | } | |
47ad8475 | 4669 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */ |
49076ec2 | 4670 | |
40b64acd | 4671 | #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS |
b35f1819 KS |
4672 | |
4673 | static struct kmem_cache *page_ptl_cachep; | |
4674 | ||
4675 | void __init ptlock_cache_init(void) | |
4676 | { | |
4677 | page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0, | |
4678 | SLAB_PANIC, NULL); | |
4679 | } | |
4680 | ||
539edb58 | 4681 | bool ptlock_alloc(struct page *page) |
49076ec2 KS |
4682 | { |
4683 | spinlock_t *ptl; | |
4684 | ||
b35f1819 | 4685 | ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL); |
49076ec2 KS |
4686 | if (!ptl) |
4687 | return false; | |
539edb58 | 4688 | page->ptl = ptl; |
49076ec2 KS |
4689 | return true; |
4690 | } | |
4691 | ||
539edb58 | 4692 | void ptlock_free(struct page *page) |
49076ec2 | 4693 | { |
b35f1819 | 4694 | kmem_cache_free(page_ptl_cachep, page->ptl); |
49076ec2 KS |
4695 | } |
4696 | #endif |