]>
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 | ||
568c61fa | 84 | #if defined(LAST_CPUPID_NOT_IN_PAGE_FLAGS) && !defined(CONFIG_COMPILE_TEST) |
90572890 | 85 | #warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid. |
75980e97 PZ |
86 | #endif |
87 | ||
d41dee36 | 88 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
89 | /* use the per-pgdat data instead for discontigmem - mbligh */ |
90 | unsigned long max_mapnr; | |
1da177e4 | 91 | EXPORT_SYMBOL(max_mapnr); |
166f61b9 TH |
92 | |
93 | struct page *mem_map; | |
1da177e4 LT |
94 | EXPORT_SYMBOL(mem_map); |
95 | #endif | |
96 | ||
1da177e4 LT |
97 | /* |
98 | * A number of key systems in x86 including ioremap() rely on the assumption | |
99 | * that high_memory defines the upper bound on direct map memory, then end | |
100 | * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and | |
101 | * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL | |
102 | * and ZONE_HIGHMEM. | |
103 | */ | |
166f61b9 | 104 | void *high_memory; |
1da177e4 | 105 | EXPORT_SYMBOL(high_memory); |
1da177e4 | 106 | |
32a93233 IM |
107 | /* |
108 | * Randomize the address space (stacks, mmaps, brk, etc.). | |
109 | * | |
110 | * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization, | |
111 | * as ancient (libc5 based) binaries can segfault. ) | |
112 | */ | |
113 | int randomize_va_space __read_mostly = | |
114 | #ifdef CONFIG_COMPAT_BRK | |
115 | 1; | |
116 | #else | |
117 | 2; | |
118 | #endif | |
a62eaf15 AK |
119 | |
120 | static int __init disable_randmaps(char *s) | |
121 | { | |
122 | randomize_va_space = 0; | |
9b41046c | 123 | return 1; |
a62eaf15 AK |
124 | } |
125 | __setup("norandmaps", disable_randmaps); | |
126 | ||
62eede62 | 127 | unsigned long zero_pfn __read_mostly; |
0b70068e AB |
128 | EXPORT_SYMBOL(zero_pfn); |
129 | ||
166f61b9 TH |
130 | unsigned long highest_memmap_pfn __read_mostly; |
131 | ||
a13ea5b7 HD |
132 | /* |
133 | * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init() | |
134 | */ | |
135 | static int __init init_zero_pfn(void) | |
136 | { | |
137 | zero_pfn = page_to_pfn(ZERO_PAGE(0)); | |
138 | return 0; | |
139 | } | |
140 | core_initcall(init_zero_pfn); | |
a62eaf15 | 141 | |
d559db08 | 142 | |
34e55232 KH |
143 | #if defined(SPLIT_RSS_COUNTING) |
144 | ||
ea48cf78 | 145 | void sync_mm_rss(struct mm_struct *mm) |
34e55232 KH |
146 | { |
147 | int i; | |
148 | ||
149 | for (i = 0; i < NR_MM_COUNTERS; i++) { | |
05af2e10 DR |
150 | if (current->rss_stat.count[i]) { |
151 | add_mm_counter(mm, i, current->rss_stat.count[i]); | |
152 | current->rss_stat.count[i] = 0; | |
34e55232 KH |
153 | } |
154 | } | |
05af2e10 | 155 | current->rss_stat.events = 0; |
34e55232 KH |
156 | } |
157 | ||
158 | static void add_mm_counter_fast(struct mm_struct *mm, int member, int val) | |
159 | { | |
160 | struct task_struct *task = current; | |
161 | ||
162 | if (likely(task->mm == mm)) | |
163 | task->rss_stat.count[member] += val; | |
164 | else | |
165 | add_mm_counter(mm, member, val); | |
166 | } | |
167 | #define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1) | |
168 | #define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1) | |
169 | ||
170 | /* sync counter once per 64 page faults */ | |
171 | #define TASK_RSS_EVENTS_THRESH (64) | |
172 | static void check_sync_rss_stat(struct task_struct *task) | |
173 | { | |
174 | if (unlikely(task != current)) | |
175 | return; | |
176 | if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH)) | |
ea48cf78 | 177 | sync_mm_rss(task->mm); |
34e55232 | 178 | } |
9547d01b | 179 | #else /* SPLIT_RSS_COUNTING */ |
34e55232 KH |
180 | |
181 | #define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member) | |
182 | #define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member) | |
183 | ||
184 | static void check_sync_rss_stat(struct task_struct *task) | |
185 | { | |
186 | } | |
187 | ||
9547d01b PZ |
188 | #endif /* SPLIT_RSS_COUNTING */ |
189 | ||
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); |
fb7332a9 | 249 | __tlb_reset_range(tlb); |
1cf35d47 LT |
250 | } |
251 | ||
252 | static void tlb_flush_mmu_free(struct mmu_gather *tlb) | |
253 | { | |
254 | struct mmu_gather_batch *batch; | |
34e55232 | 255 | |
3e099461 NP |
256 | #ifdef CONFIG_HAVE_RCU_TABLE_FREE |
257 | tlb_table_flush(tlb); | |
258 | #endif | |
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 | ||
e9afa7c1 PZ |
334 | /* |
335 | * If we want tlb_remove_table() to imply TLB invalidates. | |
336 | */ | |
337 | static inline void tlb_table_invalidate(struct mmu_gather *tlb) | |
338 | { | |
339 | #ifdef CONFIG_HAVE_RCU_TABLE_INVALIDATE | |
340 | /* | |
341 | * Invalidate page-table caches used by hardware walkers. Then we still | |
342 | * need to RCU-sched wait while freeing the pages because software | |
343 | * walkers can still be in-flight. | |
344 | */ | |
345 | tlb_flush_mmu_tlbonly(tlb); | |
346 | #endif | |
347 | } | |
348 | ||
26723911 PZ |
349 | static void tlb_remove_table_smp_sync(void *arg) |
350 | { | |
351 | /* Simply deliver the interrupt */ | |
352 | } | |
353 | ||
354 | static void tlb_remove_table_one(void *table) | |
355 | { | |
356 | /* | |
357 | * This isn't an RCU grace period and hence the page-tables cannot be | |
358 | * assumed to be actually RCU-freed. | |
359 | * | |
360 | * It is however sufficient for software page-table walkers that rely on | |
361 | * IRQ disabling. See the comment near struct mmu_table_batch. | |
362 | */ | |
363 | smp_call_function(tlb_remove_table_smp_sync, NULL, 1); | |
364 | __tlb_remove_table(table); | |
365 | } | |
366 | ||
367 | static void tlb_remove_table_rcu(struct rcu_head *head) | |
368 | { | |
369 | struct mmu_table_batch *batch; | |
370 | int i; | |
371 | ||
372 | batch = container_of(head, struct mmu_table_batch, rcu); | |
373 | ||
374 | for (i = 0; i < batch->nr; i++) | |
375 | __tlb_remove_table(batch->tables[i]); | |
376 | ||
377 | free_page((unsigned long)batch); | |
378 | } | |
379 | ||
380 | void tlb_table_flush(struct mmu_gather *tlb) | |
381 | { | |
382 | struct mmu_table_batch **batch = &tlb->batch; | |
383 | ||
384 | if (*batch) { | |
e9afa7c1 | 385 | tlb_table_invalidate(tlb); |
26723911 PZ |
386 | call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu); |
387 | *batch = NULL; | |
388 | } | |
389 | } | |
390 | ||
391 | void tlb_remove_table(struct mmu_gather *tlb, void *table) | |
392 | { | |
393 | struct mmu_table_batch **batch = &tlb->batch; | |
394 | ||
26723911 PZ |
395 | if (*batch == NULL) { |
396 | *batch = (struct mmu_table_batch *)__get_free_page(GFP_NOWAIT | __GFP_NOWARN); | |
397 | if (*batch == NULL) { | |
e9afa7c1 | 398 | tlb_table_invalidate(tlb); |
26723911 PZ |
399 | tlb_remove_table_one(table); |
400 | return; | |
401 | } | |
402 | (*batch)->nr = 0; | |
403 | } | |
e9afa7c1 | 404 | |
26723911 PZ |
405 | (*batch)->tables[(*batch)->nr++] = table; |
406 | if ((*batch)->nr == MAX_TABLE_BATCH) | |
407 | tlb_table_flush(tlb); | |
408 | } | |
409 | ||
9547d01b | 410 | #endif /* CONFIG_HAVE_RCU_TABLE_FREE */ |
26723911 | 411 | |
56236a59 MK |
412 | /* tlb_gather_mmu |
413 | * Called to initialize an (on-stack) mmu_gather structure for page-table | |
414 | * tear-down from @mm. The @fullmm argument is used when @mm is without | |
415 | * users and we're going to destroy the full address space (exit/execve). | |
416 | */ | |
417 | void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, | |
418 | unsigned long start, unsigned long end) | |
419 | { | |
420 | arch_tlb_gather_mmu(tlb, mm, start, end); | |
99baac21 | 421 | inc_tlb_flush_pending(tlb->mm); |
56236a59 MK |
422 | } |
423 | ||
424 | void tlb_finish_mmu(struct mmu_gather *tlb, | |
425 | unsigned long start, unsigned long end) | |
426 | { | |
99baac21 MK |
427 | /* |
428 | * If there are parallel threads are doing PTE changes on same range | |
429 | * under non-exclusive lock(e.g., mmap_sem read-side) but defer TLB | |
430 | * flush by batching, a thread has stable TLB entry can fail to flush | |
431 | * the TLB by observing pte_none|!pte_dirty, for example so flush TLB | |
432 | * forcefully if we detect parallel PTE batching threads. | |
433 | */ | |
434 | bool force = mm_tlb_flush_nested(tlb->mm); | |
435 | ||
436 | arch_tlb_finish_mmu(tlb, start, end, force); | |
437 | dec_tlb_flush_pending(tlb->mm); | |
56236a59 MK |
438 | } |
439 | ||
1da177e4 LT |
440 | /* |
441 | * Note: this doesn't free the actual pages themselves. That | |
442 | * has been handled earlier when unmapping all the memory regions. | |
443 | */ | |
9e1b32ca BH |
444 | static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd, |
445 | unsigned long addr) | |
1da177e4 | 446 | { |
2f569afd | 447 | pgtable_t token = pmd_pgtable(*pmd); |
e0da382c | 448 | pmd_clear(pmd); |
9e1b32ca | 449 | pte_free_tlb(tlb, token, addr); |
e1f56c89 | 450 | atomic_long_dec(&tlb->mm->nr_ptes); |
1da177e4 LT |
451 | } |
452 | ||
e0da382c HD |
453 | static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud, |
454 | unsigned long addr, unsigned long end, | |
455 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
456 | { |
457 | pmd_t *pmd; | |
458 | unsigned long next; | |
e0da382c | 459 | unsigned long start; |
1da177e4 | 460 | |
e0da382c | 461 | start = addr; |
1da177e4 | 462 | pmd = pmd_offset(pud, addr); |
1da177e4 LT |
463 | do { |
464 | next = pmd_addr_end(addr, end); | |
465 | if (pmd_none_or_clear_bad(pmd)) | |
466 | continue; | |
9e1b32ca | 467 | free_pte_range(tlb, pmd, addr); |
1da177e4 LT |
468 | } while (pmd++, addr = next, addr != end); |
469 | ||
e0da382c HD |
470 | start &= PUD_MASK; |
471 | if (start < floor) | |
472 | return; | |
473 | if (ceiling) { | |
474 | ceiling &= PUD_MASK; | |
475 | if (!ceiling) | |
476 | return; | |
1da177e4 | 477 | } |
e0da382c HD |
478 | if (end - 1 > ceiling - 1) |
479 | return; | |
480 | ||
481 | pmd = pmd_offset(pud, start); | |
482 | pud_clear(pud); | |
9e1b32ca | 483 | pmd_free_tlb(tlb, pmd, start); |
dc6c9a35 | 484 | mm_dec_nr_pmds(tlb->mm); |
1da177e4 LT |
485 | } |
486 | ||
c2febafc | 487 | static inline void free_pud_range(struct mmu_gather *tlb, p4d_t *p4d, |
e0da382c HD |
488 | unsigned long addr, unsigned long end, |
489 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
490 | { |
491 | pud_t *pud; | |
492 | unsigned long next; | |
e0da382c | 493 | unsigned long start; |
1da177e4 | 494 | |
e0da382c | 495 | start = addr; |
c2febafc | 496 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
497 | do { |
498 | next = pud_addr_end(addr, end); | |
499 | if (pud_none_or_clear_bad(pud)) | |
500 | continue; | |
e0da382c | 501 | free_pmd_range(tlb, pud, addr, next, floor, ceiling); |
1da177e4 LT |
502 | } while (pud++, addr = next, addr != end); |
503 | ||
c2febafc KS |
504 | start &= P4D_MASK; |
505 | if (start < floor) | |
506 | return; | |
507 | if (ceiling) { | |
508 | ceiling &= P4D_MASK; | |
509 | if (!ceiling) | |
510 | return; | |
511 | } | |
512 | if (end - 1 > ceiling - 1) | |
513 | return; | |
514 | ||
515 | pud = pud_offset(p4d, start); | |
516 | p4d_clear(p4d); | |
517 | pud_free_tlb(tlb, pud, start); | |
518 | } | |
519 | ||
520 | static inline void free_p4d_range(struct mmu_gather *tlb, pgd_t *pgd, | |
521 | unsigned long addr, unsigned long end, | |
522 | unsigned long floor, unsigned long ceiling) | |
523 | { | |
524 | p4d_t *p4d; | |
525 | unsigned long next; | |
526 | unsigned long start; | |
527 | ||
528 | start = addr; | |
529 | p4d = p4d_offset(pgd, addr); | |
530 | do { | |
531 | next = p4d_addr_end(addr, end); | |
532 | if (p4d_none_or_clear_bad(p4d)) | |
533 | continue; | |
534 | free_pud_range(tlb, p4d, addr, next, floor, ceiling); | |
535 | } while (p4d++, addr = next, addr != end); | |
536 | ||
e0da382c HD |
537 | start &= PGDIR_MASK; |
538 | if (start < floor) | |
539 | return; | |
540 | if (ceiling) { | |
541 | ceiling &= PGDIR_MASK; | |
542 | if (!ceiling) | |
543 | return; | |
1da177e4 | 544 | } |
e0da382c HD |
545 | if (end - 1 > ceiling - 1) |
546 | return; | |
547 | ||
c2febafc | 548 | p4d = p4d_offset(pgd, start); |
e0da382c | 549 | pgd_clear(pgd); |
c2febafc | 550 | p4d_free_tlb(tlb, p4d, start); |
1da177e4 LT |
551 | } |
552 | ||
553 | /* | |
e0da382c | 554 | * This function frees user-level page tables of a process. |
1da177e4 | 555 | */ |
42b77728 | 556 | void free_pgd_range(struct mmu_gather *tlb, |
e0da382c HD |
557 | unsigned long addr, unsigned long end, |
558 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
559 | { |
560 | pgd_t *pgd; | |
561 | unsigned long next; | |
e0da382c HD |
562 | |
563 | /* | |
564 | * The next few lines have given us lots of grief... | |
565 | * | |
566 | * Why are we testing PMD* at this top level? Because often | |
567 | * there will be no work to do at all, and we'd prefer not to | |
568 | * go all the way down to the bottom just to discover that. | |
569 | * | |
570 | * Why all these "- 1"s? Because 0 represents both the bottom | |
571 | * of the address space and the top of it (using -1 for the | |
572 | * top wouldn't help much: the masks would do the wrong thing). | |
573 | * The rule is that addr 0 and floor 0 refer to the bottom of | |
574 | * the address space, but end 0 and ceiling 0 refer to the top | |
575 | * Comparisons need to use "end - 1" and "ceiling - 1" (though | |
576 | * that end 0 case should be mythical). | |
577 | * | |
578 | * Wherever addr is brought up or ceiling brought down, we must | |
579 | * be careful to reject "the opposite 0" before it confuses the | |
580 | * subsequent tests. But what about where end is brought down | |
581 | * by PMD_SIZE below? no, end can't go down to 0 there. | |
582 | * | |
583 | * Whereas we round start (addr) and ceiling down, by different | |
584 | * masks at different levels, in order to test whether a table | |
585 | * now has no other vmas using it, so can be freed, we don't | |
586 | * bother to round floor or end up - the tests don't need that. | |
587 | */ | |
1da177e4 | 588 | |
e0da382c HD |
589 | addr &= PMD_MASK; |
590 | if (addr < floor) { | |
591 | addr += PMD_SIZE; | |
592 | if (!addr) | |
593 | return; | |
594 | } | |
595 | if (ceiling) { | |
596 | ceiling &= PMD_MASK; | |
597 | if (!ceiling) | |
598 | return; | |
599 | } | |
600 | if (end - 1 > ceiling - 1) | |
601 | end -= PMD_SIZE; | |
602 | if (addr > end - 1) | |
603 | return; | |
07e32661 AK |
604 | /* |
605 | * We add page table cache pages with PAGE_SIZE, | |
606 | * (see pte_free_tlb()), flush the tlb if we need | |
607 | */ | |
608 | tlb_remove_check_page_size_change(tlb, PAGE_SIZE); | |
42b77728 | 609 | pgd = pgd_offset(tlb->mm, addr); |
1da177e4 LT |
610 | do { |
611 | next = pgd_addr_end(addr, end); | |
612 | if (pgd_none_or_clear_bad(pgd)) | |
613 | continue; | |
c2febafc | 614 | free_p4d_range(tlb, pgd, addr, next, floor, ceiling); |
1da177e4 | 615 | } while (pgd++, addr = next, addr != end); |
e0da382c HD |
616 | } |
617 | ||
42b77728 | 618 | void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma, |
3bf5ee95 | 619 | unsigned long floor, unsigned long ceiling) |
e0da382c HD |
620 | { |
621 | while (vma) { | |
622 | struct vm_area_struct *next = vma->vm_next; | |
623 | unsigned long addr = vma->vm_start; | |
624 | ||
8f4f8c16 | 625 | /* |
25d9e2d1 NP |
626 | * Hide vma from rmap and truncate_pagecache before freeing |
627 | * pgtables | |
8f4f8c16 | 628 | */ |
5beb4930 | 629 | unlink_anon_vmas(vma); |
8f4f8c16 HD |
630 | unlink_file_vma(vma); |
631 | ||
9da61aef | 632 | if (is_vm_hugetlb_page(vma)) { |
3bf5ee95 | 633 | hugetlb_free_pgd_range(tlb, addr, vma->vm_end, |
166f61b9 | 634 | floor, next ? next->vm_start : ceiling); |
3bf5ee95 HD |
635 | } else { |
636 | /* | |
637 | * Optimization: gather nearby vmas into one call down | |
638 | */ | |
639 | while (next && next->vm_start <= vma->vm_end + PMD_SIZE | |
4866920b | 640 | && !is_vm_hugetlb_page(next)) { |
3bf5ee95 HD |
641 | vma = next; |
642 | next = vma->vm_next; | |
5beb4930 | 643 | unlink_anon_vmas(vma); |
8f4f8c16 | 644 | unlink_file_vma(vma); |
3bf5ee95 HD |
645 | } |
646 | free_pgd_range(tlb, addr, vma->vm_end, | |
166f61b9 | 647 | floor, next ? next->vm_start : ceiling); |
3bf5ee95 | 648 | } |
e0da382c HD |
649 | vma = next; |
650 | } | |
1da177e4 LT |
651 | } |
652 | ||
3ed3a4f0 | 653 | int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address) |
1da177e4 | 654 | { |
c4088ebd | 655 | spinlock_t *ptl; |
2f569afd | 656 | pgtable_t new = pte_alloc_one(mm, address); |
1bb3630e HD |
657 | if (!new) |
658 | return -ENOMEM; | |
659 | ||
362a61ad NP |
660 | /* |
661 | * Ensure all pte setup (eg. pte page lock and page clearing) are | |
662 | * visible before the pte is made visible to other CPUs by being | |
663 | * put into page tables. | |
664 | * | |
665 | * The other side of the story is the pointer chasing in the page | |
666 | * table walking code (when walking the page table without locking; | |
667 | * ie. most of the time). Fortunately, these data accesses consist | |
668 | * of a chain of data-dependent loads, meaning most CPUs (alpha | |
669 | * being the notable exception) will already guarantee loads are | |
670 | * seen in-order. See the alpha page table accessors for the | |
671 | * smp_read_barrier_depends() barriers in page table walking code. | |
672 | */ | |
673 | smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */ | |
674 | ||
c4088ebd | 675 | ptl = pmd_lock(mm, pmd); |
8ac1f832 | 676 | if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ |
e1f56c89 | 677 | atomic_long_inc(&mm->nr_ptes); |
1da177e4 | 678 | pmd_populate(mm, pmd, new); |
2f569afd | 679 | new = NULL; |
4b471e88 | 680 | } |
c4088ebd | 681 | spin_unlock(ptl); |
2f569afd MS |
682 | if (new) |
683 | pte_free(mm, new); | |
1bb3630e | 684 | return 0; |
1da177e4 LT |
685 | } |
686 | ||
1bb3630e | 687 | int __pte_alloc_kernel(pmd_t *pmd, unsigned long address) |
1da177e4 | 688 | { |
1bb3630e HD |
689 | pte_t *new = pte_alloc_one_kernel(&init_mm, address); |
690 | if (!new) | |
691 | return -ENOMEM; | |
692 | ||
362a61ad NP |
693 | smp_wmb(); /* See comment in __pte_alloc */ |
694 | ||
1bb3630e | 695 | spin_lock(&init_mm.page_table_lock); |
8ac1f832 | 696 | if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ |
1bb3630e | 697 | pmd_populate_kernel(&init_mm, pmd, new); |
2f569afd | 698 | new = NULL; |
4b471e88 | 699 | } |
1bb3630e | 700 | spin_unlock(&init_mm.page_table_lock); |
2f569afd MS |
701 | if (new) |
702 | pte_free_kernel(&init_mm, new); | |
1bb3630e | 703 | return 0; |
1da177e4 LT |
704 | } |
705 | ||
d559db08 KH |
706 | static inline void init_rss_vec(int *rss) |
707 | { | |
708 | memset(rss, 0, sizeof(int) * NR_MM_COUNTERS); | |
709 | } | |
710 | ||
711 | static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss) | |
ae859762 | 712 | { |
d559db08 KH |
713 | int i; |
714 | ||
34e55232 | 715 | if (current->mm == mm) |
05af2e10 | 716 | sync_mm_rss(mm); |
d559db08 KH |
717 | for (i = 0; i < NR_MM_COUNTERS; i++) |
718 | if (rss[i]) | |
719 | add_mm_counter(mm, i, rss[i]); | |
ae859762 HD |
720 | } |
721 | ||
b5810039 | 722 | /* |
6aab341e LT |
723 | * This function is called to print an error when a bad pte |
724 | * is found. For example, we might have a PFN-mapped pte in | |
725 | * a region that doesn't allow it. | |
b5810039 NP |
726 | * |
727 | * The calling function must still handle the error. | |
728 | */ | |
3dc14741 HD |
729 | static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr, |
730 | pte_t pte, struct page *page) | |
b5810039 | 731 | { |
3dc14741 | 732 | pgd_t *pgd = pgd_offset(vma->vm_mm, addr); |
c2febafc KS |
733 | p4d_t *p4d = p4d_offset(pgd, addr); |
734 | pud_t *pud = pud_offset(p4d, addr); | |
3dc14741 HD |
735 | pmd_t *pmd = pmd_offset(pud, addr); |
736 | struct address_space *mapping; | |
737 | pgoff_t index; | |
d936cf9b HD |
738 | static unsigned long resume; |
739 | static unsigned long nr_shown; | |
740 | static unsigned long nr_unshown; | |
741 | ||
742 | /* | |
743 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
744 | * or allow a steady drip of one report per second. | |
745 | */ | |
746 | if (nr_shown == 60) { | |
747 | if (time_before(jiffies, resume)) { | |
748 | nr_unshown++; | |
749 | return; | |
750 | } | |
751 | if (nr_unshown) { | |
1170532b JP |
752 | pr_alert("BUG: Bad page map: %lu messages suppressed\n", |
753 | nr_unshown); | |
d936cf9b HD |
754 | nr_unshown = 0; |
755 | } | |
756 | nr_shown = 0; | |
757 | } | |
758 | if (nr_shown++ == 0) | |
759 | resume = jiffies + 60 * HZ; | |
3dc14741 HD |
760 | |
761 | mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL; | |
762 | index = linear_page_index(vma, addr); | |
763 | ||
1170532b JP |
764 | pr_alert("BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n", |
765 | current->comm, | |
766 | (long long)pte_val(pte), (long long)pmd_val(*pmd)); | |
718a3821 | 767 | if (page) |
f0b791a3 | 768 | dump_page(page, "bad pte"); |
1170532b JP |
769 | pr_alert("addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n", |
770 | (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index); | |
3dc14741 HD |
771 | /* |
772 | * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y | |
773 | */ | |
2682582a KK |
774 | pr_alert("file:%pD fault:%pf mmap:%pf readpage:%pf\n", |
775 | vma->vm_file, | |
776 | vma->vm_ops ? vma->vm_ops->fault : NULL, | |
777 | vma->vm_file ? vma->vm_file->f_op->mmap : NULL, | |
778 | mapping ? mapping->a_ops->readpage : NULL); | |
b5810039 | 779 | dump_stack(); |
373d4d09 | 780 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
b5810039 NP |
781 | } |
782 | ||
ee498ed7 | 783 | /* |
7e675137 | 784 | * vm_normal_page -- This function gets the "struct page" associated with a pte. |
6aab341e | 785 | * |
7e675137 NP |
786 | * "Special" mappings do not wish to be associated with a "struct page" (either |
787 | * it doesn't exist, or it exists but they don't want to touch it). In this | |
788 | * case, NULL is returned here. "Normal" mappings do have a struct page. | |
b379d790 | 789 | * |
7e675137 NP |
790 | * There are 2 broad cases. Firstly, an architecture may define a pte_special() |
791 | * pte bit, in which case this function is trivial. Secondly, an architecture | |
792 | * may not have a spare pte bit, which requires a more complicated scheme, | |
793 | * described below. | |
794 | * | |
795 | * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a | |
796 | * special mapping (even if there are underlying and valid "struct pages"). | |
797 | * COWed pages of a VM_PFNMAP are always normal. | |
6aab341e | 798 | * |
b379d790 JH |
799 | * The way we recognize COWed pages within VM_PFNMAP mappings is through the |
800 | * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit | |
7e675137 NP |
801 | * set, and the vm_pgoff will point to the first PFN mapped: thus every special |
802 | * mapping will always honor the rule | |
6aab341e LT |
803 | * |
804 | * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) | |
805 | * | |
7e675137 NP |
806 | * And for normal mappings this is false. |
807 | * | |
808 | * This restricts such mappings to be a linear translation from virtual address | |
809 | * to pfn. To get around this restriction, we allow arbitrary mappings so long | |
810 | * as the vma is not a COW mapping; in that case, we know that all ptes are | |
811 | * special (because none can have been COWed). | |
b379d790 | 812 | * |
b379d790 | 813 | * |
7e675137 | 814 | * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. |
b379d790 JH |
815 | * |
816 | * VM_MIXEDMAP mappings can likewise contain memory with or without "struct | |
817 | * page" backing, however the difference is that _all_ pages with a struct | |
818 | * page (that is, those where pfn_valid is true) are refcounted and considered | |
819 | * normal pages by the VM. The disadvantage is that pages are refcounted | |
820 | * (which can be slower and simply not an option for some PFNMAP users). The | |
821 | * advantage is that we don't have to follow the strict linearity rule of | |
822 | * PFNMAP mappings in order to support COWable mappings. | |
823 | * | |
ee498ed7 | 824 | */ |
7e675137 NP |
825 | #ifdef __HAVE_ARCH_PTE_SPECIAL |
826 | # define HAVE_PTE_SPECIAL 1 | |
827 | #else | |
828 | # define HAVE_PTE_SPECIAL 0 | |
829 | #endif | |
df6ad698 JG |
830 | struct page *_vm_normal_page(struct vm_area_struct *vma, unsigned long addr, |
831 | pte_t pte, bool with_public_device) | |
ee498ed7 | 832 | { |
22b31eec | 833 | unsigned long pfn = pte_pfn(pte); |
7e675137 NP |
834 | |
835 | if (HAVE_PTE_SPECIAL) { | |
b38af472 | 836 | if (likely(!pte_special(pte))) |
22b31eec | 837 | goto check_pfn; |
667a0a06 DV |
838 | if (vma->vm_ops && vma->vm_ops->find_special_page) |
839 | return vma->vm_ops->find_special_page(vma, addr); | |
a13ea5b7 HD |
840 | if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) |
841 | return NULL; | |
df6ad698 JG |
842 | if (is_zero_pfn(pfn)) |
843 | return NULL; | |
844 | ||
845 | /* | |
846 | * Device public pages are special pages (they are ZONE_DEVICE | |
847 | * pages but different from persistent memory). They behave | |
848 | * allmost like normal pages. The difference is that they are | |
849 | * not on the lru and thus should never be involve with any- | |
850 | * thing that involve lru manipulation (mlock, numa balancing, | |
851 | * ...). | |
852 | * | |
853 | * This is why we still want to return NULL for such page from | |
854 | * vm_normal_page() so that we do not have to special case all | |
855 | * call site of vm_normal_page(). | |
856 | */ | |
7d790d2d | 857 | if (likely(pfn <= highest_memmap_pfn)) { |
df6ad698 JG |
858 | struct page *page = pfn_to_page(pfn); |
859 | ||
860 | if (is_device_public_page(page)) { | |
861 | if (with_public_device) | |
862 | return page; | |
863 | return NULL; | |
864 | } | |
865 | } | |
866 | print_bad_pte(vma, addr, pte, NULL); | |
7e675137 NP |
867 | return NULL; |
868 | } | |
869 | ||
870 | /* !HAVE_PTE_SPECIAL case follows: */ | |
871 | ||
b379d790 JH |
872 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { |
873 | if (vma->vm_flags & VM_MIXEDMAP) { | |
874 | if (!pfn_valid(pfn)) | |
875 | return NULL; | |
876 | goto out; | |
877 | } else { | |
7e675137 NP |
878 | unsigned long off; |
879 | off = (addr - vma->vm_start) >> PAGE_SHIFT; | |
b379d790 JH |
880 | if (pfn == vma->vm_pgoff + off) |
881 | return NULL; | |
882 | if (!is_cow_mapping(vma->vm_flags)) | |
883 | return NULL; | |
884 | } | |
6aab341e LT |
885 | } |
886 | ||
b38af472 HD |
887 | if (is_zero_pfn(pfn)) |
888 | return NULL; | |
22b31eec HD |
889 | check_pfn: |
890 | if (unlikely(pfn > highest_memmap_pfn)) { | |
891 | print_bad_pte(vma, addr, pte, NULL); | |
892 | return NULL; | |
893 | } | |
6aab341e LT |
894 | |
895 | /* | |
7e675137 | 896 | * NOTE! We still have PageReserved() pages in the page tables. |
7e675137 | 897 | * eg. VDSO mappings can cause them to exist. |
6aab341e | 898 | */ |
b379d790 | 899 | out: |
6aab341e | 900 | return pfn_to_page(pfn); |
ee498ed7 HD |
901 | } |
902 | ||
28093f9f GS |
903 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
904 | struct page *vm_normal_page_pmd(struct vm_area_struct *vma, unsigned long addr, | |
905 | pmd_t pmd) | |
906 | { | |
907 | unsigned long pfn = pmd_pfn(pmd); | |
908 | ||
909 | /* | |
910 | * There is no pmd_special() but there may be special pmds, e.g. | |
911 | * in a direct-access (dax) mapping, so let's just replicate the | |
912 | * !HAVE_PTE_SPECIAL case from vm_normal_page() here. | |
913 | */ | |
914 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { | |
915 | if (vma->vm_flags & VM_MIXEDMAP) { | |
916 | if (!pfn_valid(pfn)) | |
917 | return NULL; | |
918 | goto out; | |
919 | } else { | |
920 | unsigned long off; | |
921 | off = (addr - vma->vm_start) >> PAGE_SHIFT; | |
922 | if (pfn == vma->vm_pgoff + off) | |
923 | return NULL; | |
924 | if (!is_cow_mapping(vma->vm_flags)) | |
925 | return NULL; | |
926 | } | |
927 | } | |
928 | ||
929 | if (is_zero_pfn(pfn)) | |
930 | return NULL; | |
931 | if (unlikely(pfn > highest_memmap_pfn)) | |
932 | return NULL; | |
933 | ||
934 | /* | |
935 | * NOTE! We still have PageReserved() pages in the page tables. | |
936 | * eg. VDSO mappings can cause them to exist. | |
937 | */ | |
938 | out: | |
939 | return pfn_to_page(pfn); | |
940 | } | |
941 | #endif | |
942 | ||
1da177e4 LT |
943 | /* |
944 | * copy one vm_area from one task to the other. Assumes the page tables | |
945 | * already present in the new task to be cleared in the whole range | |
946 | * covered by this vma. | |
1da177e4 LT |
947 | */ |
948 | ||
570a335b | 949 | static inline unsigned long |
1da177e4 | 950 | copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
b5810039 | 951 | pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma, |
8c103762 | 952 | unsigned long addr, int *rss) |
1da177e4 | 953 | { |
b5810039 | 954 | unsigned long vm_flags = vma->vm_flags; |
1da177e4 LT |
955 | pte_t pte = *src_pte; |
956 | struct page *page; | |
1da177e4 LT |
957 | |
958 | /* pte contains position in swap or file, so copy. */ | |
959 | if (unlikely(!pte_present(pte))) { | |
0661a336 KS |
960 | swp_entry_t entry = pte_to_swp_entry(pte); |
961 | ||
962 | if (likely(!non_swap_entry(entry))) { | |
963 | if (swap_duplicate(entry) < 0) | |
964 | return entry.val; | |
965 | ||
966 | /* make sure dst_mm is on swapoff's mmlist. */ | |
967 | if (unlikely(list_empty(&dst_mm->mmlist))) { | |
968 | spin_lock(&mmlist_lock); | |
969 | if (list_empty(&dst_mm->mmlist)) | |
970 | list_add(&dst_mm->mmlist, | |
971 | &src_mm->mmlist); | |
972 | spin_unlock(&mmlist_lock); | |
973 | } | |
974 | rss[MM_SWAPENTS]++; | |
975 | } else if (is_migration_entry(entry)) { | |
976 | page = migration_entry_to_page(entry); | |
977 | ||
eca56ff9 | 978 | rss[mm_counter(page)]++; |
0661a336 KS |
979 | |
980 | if (is_write_migration_entry(entry) && | |
981 | is_cow_mapping(vm_flags)) { | |
982 | /* | |
983 | * COW mappings require pages in both | |
984 | * parent and child to be set to read. | |
985 | */ | |
986 | make_migration_entry_read(&entry); | |
987 | pte = swp_entry_to_pte(entry); | |
988 | if (pte_swp_soft_dirty(*src_pte)) | |
989 | pte = pte_swp_mksoft_dirty(pte); | |
990 | set_pte_at(src_mm, addr, src_pte, pte); | |
0697212a | 991 | } |
5042db43 JG |
992 | } else if (is_device_private_entry(entry)) { |
993 | page = device_private_entry_to_page(entry); | |
994 | ||
995 | /* | |
996 | * Update rss count even for unaddressable pages, as | |
997 | * they should treated just like normal pages in this | |
998 | * respect. | |
999 | * | |
1000 | * We will likely want to have some new rss counters | |
1001 | * for unaddressable pages, at some point. But for now | |
1002 | * keep things as they are. | |
1003 | */ | |
1004 | get_page(page); | |
1005 | rss[mm_counter(page)]++; | |
1006 | page_dup_rmap(page, false); | |
1007 | ||
1008 | /* | |
1009 | * We do not preserve soft-dirty information, because so | |
1010 | * far, checkpoint/restore is the only feature that | |
1011 | * requires that. And checkpoint/restore does not work | |
1012 | * when a device driver is involved (you cannot easily | |
1013 | * save and restore device driver state). | |
1014 | */ | |
1015 | if (is_write_device_private_entry(entry) && | |
1016 | is_cow_mapping(vm_flags)) { | |
1017 | make_device_private_entry_read(&entry); | |
1018 | pte = swp_entry_to_pte(entry); | |
1019 | set_pte_at(src_mm, addr, src_pte, pte); | |
1020 | } | |
1da177e4 | 1021 | } |
ae859762 | 1022 | goto out_set_pte; |
1da177e4 LT |
1023 | } |
1024 | ||
1da177e4 LT |
1025 | /* |
1026 | * If it's a COW mapping, write protect it both | |
1027 | * in the parent and the child | |
1028 | */ | |
67121172 | 1029 | if (is_cow_mapping(vm_flags)) { |
1da177e4 | 1030 | ptep_set_wrprotect(src_mm, addr, src_pte); |
3dc90795 | 1031 | pte = pte_wrprotect(pte); |
1da177e4 LT |
1032 | } |
1033 | ||
1034 | /* | |
1035 | * If it's a shared mapping, mark it clean in | |
1036 | * the child | |
1037 | */ | |
1038 | if (vm_flags & VM_SHARED) | |
1039 | pte = pte_mkclean(pte); | |
1040 | pte = pte_mkold(pte); | |
6aab341e LT |
1041 | |
1042 | page = vm_normal_page(vma, addr, pte); | |
1043 | if (page) { | |
1044 | get_page(page); | |
53f9263b | 1045 | page_dup_rmap(page, false); |
eca56ff9 | 1046 | rss[mm_counter(page)]++; |
df6ad698 JG |
1047 | } else if (pte_devmap(pte)) { |
1048 | page = pte_page(pte); | |
1049 | ||
1050 | /* | |
1051 | * Cache coherent device memory behave like regular page and | |
1052 | * not like persistent memory page. For more informations see | |
1053 | * MEMORY_DEVICE_CACHE_COHERENT in memory_hotplug.h | |
1054 | */ | |
1055 | if (is_device_public_page(page)) { | |
1056 | get_page(page); | |
1057 | page_dup_rmap(page, false); | |
1058 | rss[mm_counter(page)]++; | |
1059 | } | |
6aab341e | 1060 | } |
ae859762 HD |
1061 | |
1062 | out_set_pte: | |
1063 | set_pte_at(dst_mm, addr, dst_pte, pte); | |
570a335b | 1064 | return 0; |
1da177e4 LT |
1065 | } |
1066 | ||
21bda264 | 1067 | static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
71e3aac0 AA |
1068 | pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma, |
1069 | unsigned long addr, unsigned long end) | |
1da177e4 | 1070 | { |
c36987e2 | 1071 | pte_t *orig_src_pte, *orig_dst_pte; |
1da177e4 | 1072 | pte_t *src_pte, *dst_pte; |
c74df32c | 1073 | spinlock_t *src_ptl, *dst_ptl; |
e040f218 | 1074 | int progress = 0; |
d559db08 | 1075 | int rss[NR_MM_COUNTERS]; |
570a335b | 1076 | swp_entry_t entry = (swp_entry_t){0}; |
1da177e4 LT |
1077 | |
1078 | again: | |
d559db08 KH |
1079 | init_rss_vec(rss); |
1080 | ||
c74df32c | 1081 | dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); |
1da177e4 LT |
1082 | if (!dst_pte) |
1083 | return -ENOMEM; | |
ece0e2b6 | 1084 | src_pte = pte_offset_map(src_pmd, addr); |
4c21e2f2 | 1085 | src_ptl = pte_lockptr(src_mm, src_pmd); |
f20dc5f7 | 1086 | spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); |
c36987e2 DN |
1087 | orig_src_pte = src_pte; |
1088 | orig_dst_pte = dst_pte; | |
6606c3e0 | 1089 | arch_enter_lazy_mmu_mode(); |
1da177e4 | 1090 | |
1da177e4 LT |
1091 | do { |
1092 | /* | |
1093 | * We are holding two locks at this point - either of them | |
1094 | * could generate latencies in another task on another CPU. | |
1095 | */ | |
e040f218 HD |
1096 | if (progress >= 32) { |
1097 | progress = 0; | |
1098 | if (need_resched() || | |
95c354fe | 1099 | spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) |
e040f218 HD |
1100 | break; |
1101 | } | |
1da177e4 LT |
1102 | if (pte_none(*src_pte)) { |
1103 | progress++; | |
1104 | continue; | |
1105 | } | |
570a335b HD |
1106 | entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, |
1107 | vma, addr, rss); | |
1108 | if (entry.val) | |
1109 | break; | |
1da177e4 LT |
1110 | progress += 8; |
1111 | } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); | |
1da177e4 | 1112 | |
6606c3e0 | 1113 | arch_leave_lazy_mmu_mode(); |
c74df32c | 1114 | spin_unlock(src_ptl); |
ece0e2b6 | 1115 | pte_unmap(orig_src_pte); |
d559db08 | 1116 | add_mm_rss_vec(dst_mm, rss); |
c36987e2 | 1117 | pte_unmap_unlock(orig_dst_pte, dst_ptl); |
c74df32c | 1118 | cond_resched(); |
570a335b HD |
1119 | |
1120 | if (entry.val) { | |
1121 | if (add_swap_count_continuation(entry, GFP_KERNEL) < 0) | |
1122 | return -ENOMEM; | |
1123 | progress = 0; | |
1124 | } | |
1da177e4 LT |
1125 | if (addr != end) |
1126 | goto again; | |
1127 | return 0; | |
1128 | } | |
1129 | ||
1130 | static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
1131 | pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma, | |
1132 | unsigned long addr, unsigned long end) | |
1133 | { | |
1134 | pmd_t *src_pmd, *dst_pmd; | |
1135 | unsigned long next; | |
1136 | ||
1137 | dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); | |
1138 | if (!dst_pmd) | |
1139 | return -ENOMEM; | |
1140 | src_pmd = pmd_offset(src_pud, addr); | |
1141 | do { | |
1142 | next = pmd_addr_end(addr, end); | |
84c3fc4e ZY |
1143 | if (is_swap_pmd(*src_pmd) || pmd_trans_huge(*src_pmd) |
1144 | || pmd_devmap(*src_pmd)) { | |
71e3aac0 | 1145 | int err; |
a00cc7d9 | 1146 | VM_BUG_ON_VMA(next-addr != HPAGE_PMD_SIZE, vma); |
71e3aac0 AA |
1147 | err = copy_huge_pmd(dst_mm, src_mm, |
1148 | dst_pmd, src_pmd, addr, vma); | |
1149 | if (err == -ENOMEM) | |
1150 | return -ENOMEM; | |
1151 | if (!err) | |
1152 | continue; | |
1153 | /* fall through */ | |
1154 | } | |
1da177e4 LT |
1155 | if (pmd_none_or_clear_bad(src_pmd)) |
1156 | continue; | |
1157 | if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd, | |
1158 | vma, addr, next)) | |
1159 | return -ENOMEM; | |
1160 | } while (dst_pmd++, src_pmd++, addr = next, addr != end); | |
1161 | return 0; | |
1162 | } | |
1163 | ||
1164 | static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
c2febafc | 1165 | p4d_t *dst_p4d, p4d_t *src_p4d, struct vm_area_struct *vma, |
1da177e4 LT |
1166 | unsigned long addr, unsigned long end) |
1167 | { | |
1168 | pud_t *src_pud, *dst_pud; | |
1169 | unsigned long next; | |
1170 | ||
c2febafc | 1171 | dst_pud = pud_alloc(dst_mm, dst_p4d, addr); |
1da177e4 LT |
1172 | if (!dst_pud) |
1173 | return -ENOMEM; | |
c2febafc | 1174 | src_pud = pud_offset(src_p4d, addr); |
1da177e4 LT |
1175 | do { |
1176 | next = pud_addr_end(addr, end); | |
a00cc7d9 MW |
1177 | if (pud_trans_huge(*src_pud) || pud_devmap(*src_pud)) { |
1178 | int err; | |
1179 | ||
1180 | VM_BUG_ON_VMA(next-addr != HPAGE_PUD_SIZE, vma); | |
1181 | err = copy_huge_pud(dst_mm, src_mm, | |
1182 | dst_pud, src_pud, addr, vma); | |
1183 | if (err == -ENOMEM) | |
1184 | return -ENOMEM; | |
1185 | if (!err) | |
1186 | continue; | |
1187 | /* fall through */ | |
1188 | } | |
1da177e4 LT |
1189 | if (pud_none_or_clear_bad(src_pud)) |
1190 | continue; | |
1191 | if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud, | |
1192 | vma, addr, next)) | |
1193 | return -ENOMEM; | |
1194 | } while (dst_pud++, src_pud++, addr = next, addr != end); | |
1195 | return 0; | |
1196 | } | |
1197 | ||
c2febafc KS |
1198 | static inline int copy_p4d_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
1199 | pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma, | |
1200 | unsigned long addr, unsigned long end) | |
1201 | { | |
1202 | p4d_t *src_p4d, *dst_p4d; | |
1203 | unsigned long next; | |
1204 | ||
1205 | dst_p4d = p4d_alloc(dst_mm, dst_pgd, addr); | |
1206 | if (!dst_p4d) | |
1207 | return -ENOMEM; | |
1208 | src_p4d = p4d_offset(src_pgd, addr); | |
1209 | do { | |
1210 | next = p4d_addr_end(addr, end); | |
1211 | if (p4d_none_or_clear_bad(src_p4d)) | |
1212 | continue; | |
1213 | if (copy_pud_range(dst_mm, src_mm, dst_p4d, src_p4d, | |
1214 | vma, addr, next)) | |
1215 | return -ENOMEM; | |
1216 | } while (dst_p4d++, src_p4d++, addr = next, addr != end); | |
1217 | return 0; | |
1218 | } | |
1219 | ||
1da177e4 LT |
1220 | int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
1221 | struct vm_area_struct *vma) | |
1222 | { | |
1223 | pgd_t *src_pgd, *dst_pgd; | |
1224 | unsigned long next; | |
1225 | unsigned long addr = vma->vm_start; | |
1226 | unsigned long end = vma->vm_end; | |
2ec74c3e SG |
1227 | unsigned long mmun_start; /* For mmu_notifiers */ |
1228 | unsigned long mmun_end; /* For mmu_notifiers */ | |
1229 | bool is_cow; | |
cddb8a5c | 1230 | int ret; |
1da177e4 | 1231 | |
d992895b NP |
1232 | /* |
1233 | * Don't copy ptes where a page fault will fill them correctly. | |
1234 | * Fork becomes much lighter when there are big shared or private | |
1235 | * readonly mappings. The tradeoff is that copy_page_range is more | |
1236 | * efficient than faulting. | |
1237 | */ | |
0661a336 KS |
1238 | if (!(vma->vm_flags & (VM_HUGETLB | VM_PFNMAP | VM_MIXEDMAP)) && |
1239 | !vma->anon_vma) | |
1240 | return 0; | |
d992895b | 1241 | |
1da177e4 LT |
1242 | if (is_vm_hugetlb_page(vma)) |
1243 | return copy_hugetlb_page_range(dst_mm, src_mm, vma); | |
1244 | ||
b3b9c293 | 1245 | if (unlikely(vma->vm_flags & VM_PFNMAP)) { |
2ab64037 | 1246 | /* |
1247 | * We do not free on error cases below as remove_vma | |
1248 | * gets called on error from higher level routine | |
1249 | */ | |
5180da41 | 1250 | ret = track_pfn_copy(vma); |
2ab64037 | 1251 | if (ret) |
1252 | return ret; | |
1253 | } | |
1254 | ||
cddb8a5c AA |
1255 | /* |
1256 | * We need to invalidate the secondary MMU mappings only when | |
1257 | * there could be a permission downgrade on the ptes of the | |
1258 | * parent mm. And a permission downgrade will only happen if | |
1259 | * is_cow_mapping() returns true. | |
1260 | */ | |
2ec74c3e SG |
1261 | is_cow = is_cow_mapping(vma->vm_flags); |
1262 | mmun_start = addr; | |
1263 | mmun_end = end; | |
1264 | if (is_cow) | |
1265 | mmu_notifier_invalidate_range_start(src_mm, mmun_start, | |
1266 | mmun_end); | |
cddb8a5c AA |
1267 | |
1268 | ret = 0; | |
1da177e4 LT |
1269 | dst_pgd = pgd_offset(dst_mm, addr); |
1270 | src_pgd = pgd_offset(src_mm, addr); | |
1271 | do { | |
1272 | next = pgd_addr_end(addr, end); | |
1273 | if (pgd_none_or_clear_bad(src_pgd)) | |
1274 | continue; | |
c2febafc | 1275 | if (unlikely(copy_p4d_range(dst_mm, src_mm, dst_pgd, src_pgd, |
cddb8a5c AA |
1276 | vma, addr, next))) { |
1277 | ret = -ENOMEM; | |
1278 | break; | |
1279 | } | |
1da177e4 | 1280 | } while (dst_pgd++, src_pgd++, addr = next, addr != end); |
cddb8a5c | 1281 | |
2ec74c3e SG |
1282 | if (is_cow) |
1283 | mmu_notifier_invalidate_range_end(src_mm, mmun_start, mmun_end); | |
cddb8a5c | 1284 | return ret; |
1da177e4 LT |
1285 | } |
1286 | ||
51c6f666 | 1287 | static unsigned long zap_pte_range(struct mmu_gather *tlb, |
b5810039 | 1288 | struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 | 1289 | unsigned long addr, unsigned long end, |
97a89413 | 1290 | struct zap_details *details) |
1da177e4 | 1291 | { |
b5810039 | 1292 | struct mm_struct *mm = tlb->mm; |
d16dfc55 | 1293 | int force_flush = 0; |
d559db08 | 1294 | int rss[NR_MM_COUNTERS]; |
97a89413 | 1295 | spinlock_t *ptl; |
5f1a1907 | 1296 | pte_t *start_pte; |
97a89413 | 1297 | pte_t *pte; |
8a5f14a2 | 1298 | swp_entry_t entry; |
d559db08 | 1299 | |
07e32661 | 1300 | tlb_remove_check_page_size_change(tlb, PAGE_SIZE); |
d16dfc55 | 1301 | again: |
e303297e | 1302 | init_rss_vec(rss); |
5f1a1907 SR |
1303 | start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
1304 | pte = start_pte; | |
3ea27719 | 1305 | flush_tlb_batched_pending(mm); |
6606c3e0 | 1306 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
1307 | do { |
1308 | pte_t ptent = *pte; | |
166f61b9 | 1309 | if (pte_none(ptent)) |
1da177e4 | 1310 | continue; |
6f5e6b9e | 1311 | |
1da177e4 | 1312 | if (pte_present(ptent)) { |
ee498ed7 | 1313 | struct page *page; |
51c6f666 | 1314 | |
df6ad698 | 1315 | page = _vm_normal_page(vma, addr, ptent, true); |
1da177e4 LT |
1316 | if (unlikely(details) && page) { |
1317 | /* | |
1318 | * unmap_shared_mapping_pages() wants to | |
1319 | * invalidate cache without truncating: | |
1320 | * unmap shared but keep private pages. | |
1321 | */ | |
1322 | if (details->check_mapping && | |
800d8c63 | 1323 | details->check_mapping != page_rmapping(page)) |
1da177e4 | 1324 | continue; |
1da177e4 | 1325 | } |
b5810039 | 1326 | ptent = ptep_get_and_clear_full(mm, addr, pte, |
a600388d | 1327 | tlb->fullmm); |
1da177e4 LT |
1328 | tlb_remove_tlb_entry(tlb, pte, addr); |
1329 | if (unlikely(!page)) | |
1330 | continue; | |
eca56ff9 JM |
1331 | |
1332 | if (!PageAnon(page)) { | |
1cf35d47 LT |
1333 | if (pte_dirty(ptent)) { |
1334 | force_flush = 1; | |
6237bcd9 | 1335 | set_page_dirty(page); |
1cf35d47 | 1336 | } |
4917e5d0 | 1337 | if (pte_young(ptent) && |
64363aad | 1338 | likely(!(vma->vm_flags & VM_SEQ_READ))) |
bf3f3bc5 | 1339 | mark_page_accessed(page); |
6237bcd9 | 1340 | } |
eca56ff9 | 1341 | rss[mm_counter(page)]--; |
d281ee61 | 1342 | page_remove_rmap(page, false); |
3dc14741 HD |
1343 | if (unlikely(page_mapcount(page) < 0)) |
1344 | print_bad_pte(vma, addr, ptent, page); | |
e9d55e15 | 1345 | if (unlikely(__tlb_remove_page(tlb, page))) { |
1cf35d47 | 1346 | force_flush = 1; |
ce9ec37b | 1347 | addr += PAGE_SIZE; |
d16dfc55 | 1348 | break; |
1cf35d47 | 1349 | } |
1da177e4 LT |
1350 | continue; |
1351 | } | |
5042db43 JG |
1352 | |
1353 | entry = pte_to_swp_entry(ptent); | |
1354 | if (non_swap_entry(entry) && is_device_private_entry(entry)) { | |
1355 | struct page *page = device_private_entry_to_page(entry); | |
1356 | ||
1357 | if (unlikely(details && details->check_mapping)) { | |
1358 | /* | |
1359 | * unmap_shared_mapping_pages() wants to | |
1360 | * invalidate cache without truncating: | |
1361 | * unmap shared but keep private pages. | |
1362 | */ | |
1363 | if (details->check_mapping != | |
1364 | page_rmapping(page)) | |
1365 | continue; | |
1366 | } | |
1367 | ||
1368 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); | |
1369 | rss[mm_counter(page)]--; | |
1370 | page_remove_rmap(page, false); | |
1371 | put_page(page); | |
1372 | continue; | |
1373 | } | |
1374 | ||
3e8715fd KS |
1375 | /* If details->check_mapping, we leave swap entries. */ |
1376 | if (unlikely(details)) | |
1da177e4 | 1377 | continue; |
b084d435 | 1378 | |
8a5f14a2 KS |
1379 | entry = pte_to_swp_entry(ptent); |
1380 | if (!non_swap_entry(entry)) | |
1381 | rss[MM_SWAPENTS]--; | |
1382 | else if (is_migration_entry(entry)) { | |
1383 | struct page *page; | |
9f9f1acd | 1384 | |
8a5f14a2 | 1385 | page = migration_entry_to_page(entry); |
eca56ff9 | 1386 | rss[mm_counter(page)]--; |
b084d435 | 1387 | } |
8a5f14a2 KS |
1388 | if (unlikely(!free_swap_and_cache(entry))) |
1389 | print_bad_pte(vma, addr, ptent, NULL); | |
9888a1ca | 1390 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); |
97a89413 | 1391 | } while (pte++, addr += PAGE_SIZE, addr != end); |
ae859762 | 1392 | |
d559db08 | 1393 | add_mm_rss_vec(mm, rss); |
6606c3e0 | 1394 | arch_leave_lazy_mmu_mode(); |
51c6f666 | 1395 | |
1cf35d47 | 1396 | /* Do the actual TLB flush before dropping ptl */ |
fb7332a9 | 1397 | if (force_flush) |
1cf35d47 | 1398 | tlb_flush_mmu_tlbonly(tlb); |
1cf35d47 LT |
1399 | pte_unmap_unlock(start_pte, ptl); |
1400 | ||
1401 | /* | |
1402 | * If we forced a TLB flush (either due to running out of | |
1403 | * batch buffers or because we needed to flush dirty TLB | |
1404 | * entries before releasing the ptl), free the batched | |
1405 | * memory too. Restart if we didn't do everything. | |
1406 | */ | |
1407 | if (force_flush) { | |
1408 | force_flush = 0; | |
1409 | tlb_flush_mmu_free(tlb); | |
2b047252 | 1410 | if (addr != end) |
d16dfc55 PZ |
1411 | goto again; |
1412 | } | |
1413 | ||
51c6f666 | 1414 | return addr; |
1da177e4 LT |
1415 | } |
1416 | ||
51c6f666 | 1417 | static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, |
b5810039 | 1418 | struct vm_area_struct *vma, pud_t *pud, |
1da177e4 | 1419 | unsigned long addr, unsigned long end, |
97a89413 | 1420 | struct zap_details *details) |
1da177e4 LT |
1421 | { |
1422 | pmd_t *pmd; | |
1423 | unsigned long next; | |
1424 | ||
1425 | pmd = pmd_offset(pud, addr); | |
1426 | do { | |
1427 | next = pmd_addr_end(addr, end); | |
84c3fc4e | 1428 | if (is_swap_pmd(*pmd) || pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) { |
249778d9 | 1429 | if (next - addr != HPAGE_PMD_SIZE) |
fd60775a | 1430 | __split_huge_pmd(vma, pmd, addr, false, NULL); |
249778d9 | 1431 | else if (zap_huge_pmd(tlb, vma, pmd, addr)) |
1a5a9906 | 1432 | goto next; |
71e3aac0 AA |
1433 | /* fall through */ |
1434 | } | |
1a5a9906 AA |
1435 | /* |
1436 | * Here there can be other concurrent MADV_DONTNEED or | |
1437 | * trans huge page faults running, and if the pmd is | |
1438 | * none or trans huge it can change under us. This is | |
1439 | * because MADV_DONTNEED holds the mmap_sem in read | |
1440 | * mode. | |
1441 | */ | |
1442 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) | |
1443 | goto next; | |
97a89413 | 1444 | next = zap_pte_range(tlb, vma, pmd, addr, next, details); |
1a5a9906 | 1445 | next: |
97a89413 PZ |
1446 | cond_resched(); |
1447 | } while (pmd++, addr = next, addr != end); | |
51c6f666 RH |
1448 | |
1449 | return addr; | |
1da177e4 LT |
1450 | } |
1451 | ||
51c6f666 | 1452 | static inline unsigned long zap_pud_range(struct mmu_gather *tlb, |
c2febafc | 1453 | struct vm_area_struct *vma, p4d_t *p4d, |
1da177e4 | 1454 | unsigned long addr, unsigned long end, |
97a89413 | 1455 | struct zap_details *details) |
1da177e4 LT |
1456 | { |
1457 | pud_t *pud; | |
1458 | unsigned long next; | |
1459 | ||
c2febafc | 1460 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
1461 | do { |
1462 | next = pud_addr_end(addr, end); | |
a00cc7d9 MW |
1463 | if (pud_trans_huge(*pud) || pud_devmap(*pud)) { |
1464 | if (next - addr != HPAGE_PUD_SIZE) { | |
1465 | VM_BUG_ON_VMA(!rwsem_is_locked(&tlb->mm->mmap_sem), vma); | |
1466 | split_huge_pud(vma, pud, addr); | |
1467 | } else if (zap_huge_pud(tlb, vma, pud, addr)) | |
1468 | goto next; | |
1469 | /* fall through */ | |
1470 | } | |
97a89413 | 1471 | if (pud_none_or_clear_bad(pud)) |
1da177e4 | 1472 | continue; |
97a89413 | 1473 | next = zap_pmd_range(tlb, vma, pud, addr, next, details); |
a00cc7d9 MW |
1474 | next: |
1475 | cond_resched(); | |
97a89413 | 1476 | } while (pud++, addr = next, addr != end); |
51c6f666 RH |
1477 | |
1478 | return addr; | |
1da177e4 LT |
1479 | } |
1480 | ||
c2febafc KS |
1481 | static inline unsigned long zap_p4d_range(struct mmu_gather *tlb, |
1482 | struct vm_area_struct *vma, pgd_t *pgd, | |
1483 | unsigned long addr, unsigned long end, | |
1484 | struct zap_details *details) | |
1485 | { | |
1486 | p4d_t *p4d; | |
1487 | unsigned long next; | |
1488 | ||
1489 | p4d = p4d_offset(pgd, addr); | |
1490 | do { | |
1491 | next = p4d_addr_end(addr, end); | |
1492 | if (p4d_none_or_clear_bad(p4d)) | |
1493 | continue; | |
1494 | next = zap_pud_range(tlb, vma, p4d, addr, next, details); | |
1495 | } while (p4d++, addr = next, addr != end); | |
1496 | ||
1497 | return addr; | |
1498 | } | |
1499 | ||
aac45363 | 1500 | void unmap_page_range(struct mmu_gather *tlb, |
038c7aa1 AV |
1501 | struct vm_area_struct *vma, |
1502 | unsigned long addr, unsigned long end, | |
1503 | struct zap_details *details) | |
1da177e4 LT |
1504 | { |
1505 | pgd_t *pgd; | |
1506 | unsigned long next; | |
1507 | ||
1da177e4 LT |
1508 | BUG_ON(addr >= end); |
1509 | tlb_start_vma(tlb, vma); | |
1510 | pgd = pgd_offset(vma->vm_mm, addr); | |
1511 | do { | |
1512 | next = pgd_addr_end(addr, end); | |
97a89413 | 1513 | if (pgd_none_or_clear_bad(pgd)) |
1da177e4 | 1514 | continue; |
c2febafc | 1515 | next = zap_p4d_range(tlb, vma, pgd, addr, next, details); |
97a89413 | 1516 | } while (pgd++, addr = next, addr != end); |
1da177e4 LT |
1517 | tlb_end_vma(tlb, vma); |
1518 | } | |
51c6f666 | 1519 | |
f5cc4eef AV |
1520 | |
1521 | static void unmap_single_vma(struct mmu_gather *tlb, | |
1522 | struct vm_area_struct *vma, unsigned long start_addr, | |
4f74d2c8 | 1523 | unsigned long end_addr, |
f5cc4eef AV |
1524 | struct zap_details *details) |
1525 | { | |
1526 | unsigned long start = max(vma->vm_start, start_addr); | |
1527 | unsigned long end; | |
1528 | ||
1529 | if (start >= vma->vm_end) | |
1530 | return; | |
1531 | end = min(vma->vm_end, end_addr); | |
1532 | if (end <= vma->vm_start) | |
1533 | return; | |
1534 | ||
cbc91f71 SD |
1535 | if (vma->vm_file) |
1536 | uprobe_munmap(vma, start, end); | |
1537 | ||
b3b9c293 | 1538 | if (unlikely(vma->vm_flags & VM_PFNMAP)) |
5180da41 | 1539 | untrack_pfn(vma, 0, 0); |
f5cc4eef AV |
1540 | |
1541 | if (start != end) { | |
1542 | if (unlikely(is_vm_hugetlb_page(vma))) { | |
1543 | /* | |
1544 | * It is undesirable to test vma->vm_file as it | |
1545 | * should be non-null for valid hugetlb area. | |
1546 | * However, vm_file will be NULL in the error | |
7aa6b4ad | 1547 | * cleanup path of mmap_region. When |
f5cc4eef | 1548 | * hugetlbfs ->mmap method fails, |
7aa6b4ad | 1549 | * mmap_region() nullifies vma->vm_file |
f5cc4eef AV |
1550 | * before calling this function to clean up. |
1551 | * Since no pte has actually been setup, it is | |
1552 | * safe to do nothing in this case. | |
1553 | */ | |
24669e58 | 1554 | if (vma->vm_file) { |
83cde9e8 | 1555 | i_mmap_lock_write(vma->vm_file->f_mapping); |
d833352a | 1556 | __unmap_hugepage_range_final(tlb, vma, start, end, NULL); |
83cde9e8 | 1557 | i_mmap_unlock_write(vma->vm_file->f_mapping); |
24669e58 | 1558 | } |
f5cc4eef AV |
1559 | } else |
1560 | unmap_page_range(tlb, vma, start, end, details); | |
1561 | } | |
1da177e4 LT |
1562 | } |
1563 | ||
1da177e4 LT |
1564 | /** |
1565 | * unmap_vmas - unmap a range of memory covered by a list of vma's | |
0164f69d | 1566 | * @tlb: address of the caller's struct mmu_gather |
1da177e4 LT |
1567 | * @vma: the starting vma |
1568 | * @start_addr: virtual address at which to start unmapping | |
1569 | * @end_addr: virtual address at which to end unmapping | |
1da177e4 | 1570 | * |
508034a3 | 1571 | * Unmap all pages in the vma list. |
1da177e4 | 1572 | * |
1da177e4 LT |
1573 | * Only addresses between `start' and `end' will be unmapped. |
1574 | * | |
1575 | * The VMA list must be sorted in ascending virtual address order. | |
1576 | * | |
1577 | * unmap_vmas() assumes that the caller will flush the whole unmapped address | |
1578 | * range after unmap_vmas() returns. So the only responsibility here is to | |
1579 | * ensure that any thus-far unmapped pages are flushed before unmap_vmas() | |
1580 | * drops the lock and schedules. | |
1581 | */ | |
6e8bb019 | 1582 | void unmap_vmas(struct mmu_gather *tlb, |
1da177e4 | 1583 | struct vm_area_struct *vma, unsigned long start_addr, |
4f74d2c8 | 1584 | unsigned long end_addr) |
1da177e4 | 1585 | { |
cddb8a5c | 1586 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 1587 | |
cddb8a5c | 1588 | mmu_notifier_invalidate_range_start(mm, start_addr, end_addr); |
f5cc4eef | 1589 | for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) |
4f74d2c8 | 1590 | unmap_single_vma(tlb, vma, start_addr, end_addr, NULL); |
cddb8a5c | 1591 | mmu_notifier_invalidate_range_end(mm, start_addr, end_addr); |
1da177e4 LT |
1592 | } |
1593 | ||
1594 | /** | |
1595 | * zap_page_range - remove user pages in a given range | |
1596 | * @vma: vm_area_struct holding the applicable pages | |
eb4546bb | 1597 | * @start: starting address of pages to zap |
1da177e4 | 1598 | * @size: number of bytes to zap |
f5cc4eef AV |
1599 | * |
1600 | * Caller must protect the VMA list | |
1da177e4 | 1601 | */ |
7e027b14 | 1602 | void zap_page_range(struct vm_area_struct *vma, unsigned long start, |
ecf1385d | 1603 | unsigned long size) |
1da177e4 LT |
1604 | { |
1605 | struct mm_struct *mm = vma->vm_mm; | |
d16dfc55 | 1606 | struct mmu_gather tlb; |
7e027b14 | 1607 | unsigned long end = start + size; |
1da177e4 | 1608 | |
1da177e4 | 1609 | lru_add_drain(); |
2b047252 | 1610 | tlb_gather_mmu(&tlb, mm, start, end); |
365e9c87 | 1611 | update_hiwater_rss(mm); |
7e027b14 | 1612 | mmu_notifier_invalidate_range_start(mm, start, end); |
4647706e | 1613 | for ( ; vma && vma->vm_start < end; vma = vma->vm_next) { |
ecf1385d | 1614 | unmap_single_vma(&tlb, vma, start, end, NULL); |
4647706e MG |
1615 | |
1616 | /* | |
1617 | * zap_page_range does not specify whether mmap_sem should be | |
1618 | * held for read or write. That allows parallel zap_page_range | |
1619 | * operations to unmap a PTE and defer a flush meaning that | |
1620 | * this call observes pte_none and fails to flush the TLB. | |
1621 | * Rather than adding a complex API, ensure that no stale | |
1622 | * TLB entries exist when this call returns. | |
1623 | */ | |
1624 | flush_tlb_range(vma, start, end); | |
1625 | } | |
1626 | ||
7e027b14 LT |
1627 | mmu_notifier_invalidate_range_end(mm, start, end); |
1628 | tlb_finish_mmu(&tlb, start, end); | |
1da177e4 LT |
1629 | } |
1630 | ||
f5cc4eef AV |
1631 | /** |
1632 | * zap_page_range_single - remove user pages in a given range | |
1633 | * @vma: vm_area_struct holding the applicable pages | |
1634 | * @address: starting address of pages to zap | |
1635 | * @size: number of bytes to zap | |
8a5f14a2 | 1636 | * @details: details of shared cache invalidation |
f5cc4eef AV |
1637 | * |
1638 | * The range must fit into one VMA. | |
1da177e4 | 1639 | */ |
f5cc4eef | 1640 | static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address, |
1da177e4 LT |
1641 | unsigned long size, struct zap_details *details) |
1642 | { | |
1643 | struct mm_struct *mm = vma->vm_mm; | |
d16dfc55 | 1644 | struct mmu_gather tlb; |
1da177e4 | 1645 | unsigned long end = address + size; |
1da177e4 | 1646 | |
1da177e4 | 1647 | lru_add_drain(); |
2b047252 | 1648 | tlb_gather_mmu(&tlb, mm, address, end); |
365e9c87 | 1649 | update_hiwater_rss(mm); |
f5cc4eef | 1650 | mmu_notifier_invalidate_range_start(mm, address, end); |
4f74d2c8 | 1651 | unmap_single_vma(&tlb, vma, address, end, details); |
f5cc4eef | 1652 | mmu_notifier_invalidate_range_end(mm, address, end); |
d16dfc55 | 1653 | tlb_finish_mmu(&tlb, address, end); |
1da177e4 LT |
1654 | } |
1655 | ||
c627f9cc JS |
1656 | /** |
1657 | * zap_vma_ptes - remove ptes mapping the vma | |
1658 | * @vma: vm_area_struct holding ptes to be zapped | |
1659 | * @address: starting address of pages to zap | |
1660 | * @size: number of bytes to zap | |
1661 | * | |
1662 | * This function only unmaps ptes assigned to VM_PFNMAP vmas. | |
1663 | * | |
1664 | * The entire address range must be fully contained within the vma. | |
1665 | * | |
1666 | * Returns 0 if successful. | |
1667 | */ | |
1668 | int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, | |
1669 | unsigned long size) | |
1670 | { | |
1671 | if (address < vma->vm_start || address + size > vma->vm_end || | |
1672 | !(vma->vm_flags & VM_PFNMAP)) | |
1673 | return -1; | |
f5cc4eef | 1674 | zap_page_range_single(vma, address, size, NULL); |
c627f9cc JS |
1675 | return 0; |
1676 | } | |
1677 | EXPORT_SYMBOL_GPL(zap_vma_ptes); | |
1678 | ||
25ca1d6c | 1679 | pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr, |
920c7a5d | 1680 | spinlock_t **ptl) |
c9cfcddf | 1681 | { |
c2febafc KS |
1682 | pgd_t *pgd; |
1683 | p4d_t *p4d; | |
1684 | pud_t *pud; | |
1685 | pmd_t *pmd; | |
1686 | ||
1687 | pgd = pgd_offset(mm, addr); | |
1688 | p4d = p4d_alloc(mm, pgd, addr); | |
1689 | if (!p4d) | |
1690 | return NULL; | |
1691 | pud = pud_alloc(mm, p4d, addr); | |
1692 | if (!pud) | |
1693 | return NULL; | |
1694 | pmd = pmd_alloc(mm, pud, addr); | |
1695 | if (!pmd) | |
1696 | return NULL; | |
1697 | ||
1698 | VM_BUG_ON(pmd_trans_huge(*pmd)); | |
1699 | return pte_alloc_map_lock(mm, pmd, addr, ptl); | |
c9cfcddf LT |
1700 | } |
1701 | ||
238f58d8 LT |
1702 | /* |
1703 | * This is the old fallback for page remapping. | |
1704 | * | |
1705 | * For historical reasons, it only allows reserved pages. Only | |
1706 | * old drivers should use this, and they needed to mark their | |
1707 | * pages reserved for the old functions anyway. | |
1708 | */ | |
423bad60 NP |
1709 | static int insert_page(struct vm_area_struct *vma, unsigned long addr, |
1710 | struct page *page, pgprot_t prot) | |
238f58d8 | 1711 | { |
423bad60 | 1712 | struct mm_struct *mm = vma->vm_mm; |
238f58d8 | 1713 | int retval; |
c9cfcddf | 1714 | pte_t *pte; |
8a9f3ccd BS |
1715 | spinlock_t *ptl; |
1716 | ||
238f58d8 | 1717 | retval = -EINVAL; |
a145dd41 | 1718 | if (PageAnon(page)) |
5b4e655e | 1719 | goto out; |
238f58d8 LT |
1720 | retval = -ENOMEM; |
1721 | flush_dcache_page(page); | |
c9cfcddf | 1722 | pte = get_locked_pte(mm, addr, &ptl); |
238f58d8 | 1723 | if (!pte) |
5b4e655e | 1724 | goto out; |
238f58d8 LT |
1725 | retval = -EBUSY; |
1726 | if (!pte_none(*pte)) | |
1727 | goto out_unlock; | |
1728 | ||
1729 | /* Ok, finally just insert the thing.. */ | |
1730 | get_page(page); | |
eca56ff9 | 1731 | inc_mm_counter_fast(mm, mm_counter_file(page)); |
dd78fedd | 1732 | page_add_file_rmap(page, false); |
238f58d8 LT |
1733 | set_pte_at(mm, addr, pte, mk_pte(page, prot)); |
1734 | ||
1735 | retval = 0; | |
8a9f3ccd BS |
1736 | pte_unmap_unlock(pte, ptl); |
1737 | return retval; | |
238f58d8 LT |
1738 | out_unlock: |
1739 | pte_unmap_unlock(pte, ptl); | |
1740 | out: | |
1741 | return retval; | |
1742 | } | |
1743 | ||
bfa5bf6d REB |
1744 | /** |
1745 | * vm_insert_page - insert single page into user vma | |
1746 | * @vma: user vma to map to | |
1747 | * @addr: target user address of this page | |
1748 | * @page: source kernel page | |
1749 | * | |
a145dd41 LT |
1750 | * This allows drivers to insert individual pages they've allocated |
1751 | * into a user vma. | |
1752 | * | |
1753 | * The page has to be a nice clean _individual_ kernel allocation. | |
1754 | * If you allocate a compound page, you need to have marked it as | |
1755 | * such (__GFP_COMP), or manually just split the page up yourself | |
8dfcc9ba | 1756 | * (see split_page()). |
a145dd41 LT |
1757 | * |
1758 | * NOTE! Traditionally this was done with "remap_pfn_range()" which | |
1759 | * took an arbitrary page protection parameter. This doesn't allow | |
1760 | * that. Your vma protection will have to be set up correctly, which | |
1761 | * means that if you want a shared writable mapping, you'd better | |
1762 | * ask for a shared writable mapping! | |
1763 | * | |
1764 | * The page does not need to be reserved. | |
4b6e1e37 KK |
1765 | * |
1766 | * Usually this function is called from f_op->mmap() handler | |
1767 | * under mm->mmap_sem write-lock, so it can change vma->vm_flags. | |
1768 | * Caller must set VM_MIXEDMAP on vma if it wants to call this | |
1769 | * function from other places, for example from page-fault handler. | |
a145dd41 | 1770 | */ |
423bad60 NP |
1771 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, |
1772 | struct page *page) | |
a145dd41 LT |
1773 | { |
1774 | if (addr < vma->vm_start || addr >= vma->vm_end) | |
1775 | return -EFAULT; | |
1776 | if (!page_count(page)) | |
1777 | return -EINVAL; | |
4b6e1e37 KK |
1778 | if (!(vma->vm_flags & VM_MIXEDMAP)) { |
1779 | BUG_ON(down_read_trylock(&vma->vm_mm->mmap_sem)); | |
1780 | BUG_ON(vma->vm_flags & VM_PFNMAP); | |
1781 | vma->vm_flags |= VM_MIXEDMAP; | |
1782 | } | |
423bad60 | 1783 | return insert_page(vma, addr, page, vma->vm_page_prot); |
a145dd41 | 1784 | } |
e3c3374f | 1785 | EXPORT_SYMBOL(vm_insert_page); |
a145dd41 | 1786 | |
423bad60 | 1787 | static int insert_pfn(struct vm_area_struct *vma, unsigned long addr, |
b2770da6 | 1788 | pfn_t pfn, pgprot_t prot, bool mkwrite) |
423bad60 NP |
1789 | { |
1790 | struct mm_struct *mm = vma->vm_mm; | |
1791 | int retval; | |
1792 | pte_t *pte, entry; | |
1793 | spinlock_t *ptl; | |
1794 | ||
1795 | retval = -ENOMEM; | |
1796 | pte = get_locked_pte(mm, addr, &ptl); | |
1797 | if (!pte) | |
1798 | goto out; | |
1799 | retval = -EBUSY; | |
b2770da6 RZ |
1800 | if (!pte_none(*pte)) { |
1801 | if (mkwrite) { | |
1802 | /* | |
1803 | * For read faults on private mappings the PFN passed | |
1804 | * in may not match the PFN we have mapped if the | |
1805 | * mapped PFN is a writeable COW page. In the mkwrite | |
1806 | * case we are creating a writable PTE for a shared | |
1807 | * mapping and we expect the PFNs to match. | |
1808 | */ | |
1809 | if (WARN_ON_ONCE(pte_pfn(*pte) != pfn_t_to_pfn(pfn))) | |
1810 | goto out_unlock; | |
1811 | entry = *pte; | |
1812 | goto out_mkwrite; | |
1813 | } else | |
1814 | goto out_unlock; | |
1815 | } | |
423bad60 NP |
1816 | |
1817 | /* Ok, finally just insert the thing.. */ | |
01c8f1c4 DW |
1818 | if (pfn_t_devmap(pfn)) |
1819 | entry = pte_mkdevmap(pfn_t_pte(pfn, prot)); | |
1820 | else | |
1821 | entry = pte_mkspecial(pfn_t_pte(pfn, prot)); | |
b2770da6 RZ |
1822 | |
1823 | out_mkwrite: | |
1824 | if (mkwrite) { | |
1825 | entry = pte_mkyoung(entry); | |
1826 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
1827 | } | |
1828 | ||
423bad60 | 1829 | set_pte_at(mm, addr, pte, entry); |
4b3073e1 | 1830 | update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */ |
423bad60 NP |
1831 | |
1832 | retval = 0; | |
1833 | out_unlock: | |
1834 | pte_unmap_unlock(pte, ptl); | |
1835 | out: | |
1836 | return retval; | |
1837 | } | |
1838 | ||
e0dc0d8f NP |
1839 | /** |
1840 | * vm_insert_pfn - insert single pfn into user vma | |
1841 | * @vma: user vma to map to | |
1842 | * @addr: target user address of this page | |
1843 | * @pfn: source kernel pfn | |
1844 | * | |
c462f179 | 1845 | * Similar to vm_insert_page, this allows drivers to insert individual pages |
e0dc0d8f NP |
1846 | * they've allocated into a user vma. Same comments apply. |
1847 | * | |
1848 | * This function should only be called from a vm_ops->fault handler, and | |
1849 | * in that case the handler should return NULL. | |
0d71d10a NP |
1850 | * |
1851 | * vma cannot be a COW mapping. | |
1852 | * | |
1853 | * As this is called only for pages that do not currently exist, we | |
1854 | * do not need to flush old virtual caches or the TLB. | |
e0dc0d8f NP |
1855 | */ |
1856 | int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr, | |
423bad60 | 1857 | unsigned long pfn) |
1745cbc5 AL |
1858 | { |
1859 | return vm_insert_pfn_prot(vma, addr, pfn, vma->vm_page_prot); | |
1860 | } | |
1861 | EXPORT_SYMBOL(vm_insert_pfn); | |
1862 | ||
1863 | /** | |
1864 | * vm_insert_pfn_prot - insert single pfn into user vma with specified pgprot | |
1865 | * @vma: user vma to map to | |
1866 | * @addr: target user address of this page | |
1867 | * @pfn: source kernel pfn | |
1868 | * @pgprot: pgprot flags for the inserted page | |
1869 | * | |
1870 | * This is exactly like vm_insert_pfn, except that it allows drivers to | |
1871 | * to override pgprot on a per-page basis. | |
1872 | * | |
1873 | * This only makes sense for IO mappings, and it makes no sense for | |
1874 | * cow mappings. In general, using multiple vmas is preferable; | |
1875 | * vm_insert_pfn_prot should only be used if using multiple VMAs is | |
1876 | * impractical. | |
1877 | */ | |
1878 | int vm_insert_pfn_prot(struct vm_area_struct *vma, unsigned long addr, | |
1879 | unsigned long pfn, pgprot_t pgprot) | |
e0dc0d8f | 1880 | { |
2ab64037 | 1881 | int ret; |
7e675137 NP |
1882 | /* |
1883 | * Technically, architectures with pte_special can avoid all these | |
1884 | * restrictions (same for remap_pfn_range). However we would like | |
1885 | * consistency in testing and feature parity among all, so we should | |
1886 | * try to keep these invariants in place for everybody. | |
1887 | */ | |
b379d790 JH |
1888 | BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); |
1889 | BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == | |
1890 | (VM_PFNMAP|VM_MIXEDMAP)); | |
1891 | BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); | |
1892 | BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); | |
e0dc0d8f | 1893 | |
423bad60 NP |
1894 | if (addr < vma->vm_start || addr >= vma->vm_end) |
1895 | return -EFAULT; | |
308a047c | 1896 | |
a4116334 AK |
1897 | if (!pfn_modify_allowed(pfn, pgprot)) |
1898 | return -EACCES; | |
1899 | ||
308a047c | 1900 | track_pfn_insert(vma, &pgprot, __pfn_to_pfn_t(pfn, PFN_DEV)); |
2ab64037 | 1901 | |
b2770da6 RZ |
1902 | ret = insert_pfn(vma, addr, __pfn_to_pfn_t(pfn, PFN_DEV), pgprot, |
1903 | false); | |
2ab64037 | 1904 | |
2ab64037 | 1905 | return ret; |
423bad60 | 1906 | } |
1745cbc5 | 1907 | EXPORT_SYMBOL(vm_insert_pfn_prot); |
e0dc0d8f | 1908 | |
b2770da6 RZ |
1909 | static int __vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr, |
1910 | pfn_t pfn, bool mkwrite) | |
423bad60 | 1911 | { |
87744ab3 DW |
1912 | pgprot_t pgprot = vma->vm_page_prot; |
1913 | ||
423bad60 | 1914 | BUG_ON(!(vma->vm_flags & VM_MIXEDMAP)); |
e0dc0d8f | 1915 | |
423bad60 NP |
1916 | if (addr < vma->vm_start || addr >= vma->vm_end) |
1917 | return -EFAULT; | |
308a047c BP |
1918 | |
1919 | track_pfn_insert(vma, &pgprot, pfn); | |
e0dc0d8f | 1920 | |
a4116334 AK |
1921 | if (!pfn_modify_allowed(pfn_t_to_pfn(pfn), pgprot)) |
1922 | return -EACCES; | |
1923 | ||
423bad60 NP |
1924 | /* |
1925 | * If we don't have pte special, then we have to use the pfn_valid() | |
1926 | * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* | |
1927 | * refcount the page if pfn_valid is true (hence insert_page rather | |
62eede62 HD |
1928 | * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP |
1929 | * without pte special, it would there be refcounted as a normal page. | |
423bad60 | 1930 | */ |
03fc2da6 | 1931 | if (!HAVE_PTE_SPECIAL && !pfn_t_devmap(pfn) && pfn_t_valid(pfn)) { |
423bad60 NP |
1932 | struct page *page; |
1933 | ||
03fc2da6 DW |
1934 | /* |
1935 | * At this point we are committed to insert_page() | |
1936 | * regardless of whether the caller specified flags that | |
1937 | * result in pfn_t_has_page() == false. | |
1938 | */ | |
1939 | page = pfn_to_page(pfn_t_to_pfn(pfn)); | |
87744ab3 | 1940 | return insert_page(vma, addr, page, pgprot); |
423bad60 | 1941 | } |
b2770da6 RZ |
1942 | return insert_pfn(vma, addr, pfn, pgprot, mkwrite); |
1943 | } | |
1944 | ||
1945 | int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr, | |
1946 | pfn_t pfn) | |
1947 | { | |
1948 | return __vm_insert_mixed(vma, addr, pfn, false); | |
1949 | ||
e0dc0d8f | 1950 | } |
423bad60 | 1951 | EXPORT_SYMBOL(vm_insert_mixed); |
e0dc0d8f | 1952 | |
b2770da6 RZ |
1953 | int vm_insert_mixed_mkwrite(struct vm_area_struct *vma, unsigned long addr, |
1954 | pfn_t pfn) | |
1955 | { | |
1956 | return __vm_insert_mixed(vma, addr, pfn, true); | |
1957 | } | |
1958 | EXPORT_SYMBOL(vm_insert_mixed_mkwrite); | |
1959 | ||
1da177e4 LT |
1960 | /* |
1961 | * maps a range of physical memory into the requested pages. the old | |
1962 | * mappings are removed. any references to nonexistent pages results | |
1963 | * in null mappings (currently treated as "copy-on-access") | |
1964 | */ | |
1965 | static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, | |
1966 | unsigned long addr, unsigned long end, | |
1967 | unsigned long pfn, pgprot_t prot) | |
1968 | { | |
1969 | pte_t *pte; | |
c74df32c | 1970 | spinlock_t *ptl; |
a4116334 | 1971 | int err = 0; |
1da177e4 | 1972 | |
c74df32c | 1973 | pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); |
1da177e4 LT |
1974 | if (!pte) |
1975 | return -ENOMEM; | |
6606c3e0 | 1976 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
1977 | do { |
1978 | BUG_ON(!pte_none(*pte)); | |
a4116334 AK |
1979 | if (!pfn_modify_allowed(pfn, prot)) { |
1980 | err = -EACCES; | |
1981 | break; | |
1982 | } | |
7e675137 | 1983 | set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); |
1da177e4 LT |
1984 | pfn++; |
1985 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
6606c3e0 | 1986 | arch_leave_lazy_mmu_mode(); |
c74df32c | 1987 | pte_unmap_unlock(pte - 1, ptl); |
a4116334 | 1988 | return err; |
1da177e4 LT |
1989 | } |
1990 | ||
1991 | static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, | |
1992 | unsigned long addr, unsigned long end, | |
1993 | unsigned long pfn, pgprot_t prot) | |
1994 | { | |
1995 | pmd_t *pmd; | |
1996 | unsigned long next; | |
a4116334 | 1997 | int err; |
1da177e4 LT |
1998 | |
1999 | pfn -= addr >> PAGE_SHIFT; | |
2000 | pmd = pmd_alloc(mm, pud, addr); | |
2001 | if (!pmd) | |
2002 | return -ENOMEM; | |
f66055ab | 2003 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
1da177e4 LT |
2004 | do { |
2005 | next = pmd_addr_end(addr, end); | |
a4116334 AK |
2006 | err = remap_pte_range(mm, pmd, addr, next, |
2007 | pfn + (addr >> PAGE_SHIFT), prot); | |
2008 | if (err) | |
2009 | return err; | |
1da177e4 LT |
2010 | } while (pmd++, addr = next, addr != end); |
2011 | return 0; | |
2012 | } | |
2013 | ||
c2febafc | 2014 | static inline int remap_pud_range(struct mm_struct *mm, p4d_t *p4d, |
1da177e4 LT |
2015 | unsigned long addr, unsigned long end, |
2016 | unsigned long pfn, pgprot_t prot) | |
2017 | { | |
2018 | pud_t *pud; | |
2019 | unsigned long next; | |
a4116334 | 2020 | int err; |
1da177e4 LT |
2021 | |
2022 | pfn -= addr >> PAGE_SHIFT; | |
c2febafc | 2023 | pud = pud_alloc(mm, p4d, addr); |
1da177e4 LT |
2024 | if (!pud) |
2025 | return -ENOMEM; | |
2026 | do { | |
2027 | next = pud_addr_end(addr, end); | |
a4116334 AK |
2028 | err = remap_pmd_range(mm, pud, addr, next, |
2029 | pfn + (addr >> PAGE_SHIFT), prot); | |
2030 | if (err) | |
2031 | return err; | |
1da177e4 LT |
2032 | } while (pud++, addr = next, addr != end); |
2033 | return 0; | |
2034 | } | |
2035 | ||
c2febafc KS |
2036 | static inline int remap_p4d_range(struct mm_struct *mm, pgd_t *pgd, |
2037 | unsigned long addr, unsigned long end, | |
2038 | unsigned long pfn, pgprot_t prot) | |
2039 | { | |
2040 | p4d_t *p4d; | |
2041 | unsigned long next; | |
a4116334 | 2042 | int err; |
c2febafc KS |
2043 | |
2044 | pfn -= addr >> PAGE_SHIFT; | |
2045 | p4d = p4d_alloc(mm, pgd, addr); | |
2046 | if (!p4d) | |
2047 | return -ENOMEM; | |
2048 | do { | |
2049 | next = p4d_addr_end(addr, end); | |
a4116334 AK |
2050 | err = remap_pud_range(mm, p4d, addr, next, |
2051 | pfn + (addr >> PAGE_SHIFT), prot); | |
2052 | if (err) | |
2053 | return err; | |
c2febafc KS |
2054 | } while (p4d++, addr = next, addr != end); |
2055 | return 0; | |
2056 | } | |
2057 | ||
bfa5bf6d REB |
2058 | /** |
2059 | * remap_pfn_range - remap kernel memory to userspace | |
2060 | * @vma: user vma to map to | |
2061 | * @addr: target user address to start at | |
2062 | * @pfn: physical address of kernel memory | |
2063 | * @size: size of map area | |
2064 | * @prot: page protection flags for this mapping | |
2065 | * | |
2066 | * Note: this is only safe if the mm semaphore is held when called. | |
2067 | */ | |
1da177e4 LT |
2068 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, |
2069 | unsigned long pfn, unsigned long size, pgprot_t prot) | |
2070 | { | |
2071 | pgd_t *pgd; | |
2072 | unsigned long next; | |
2d15cab8 | 2073 | unsigned long end = addr + PAGE_ALIGN(size); |
1da177e4 | 2074 | struct mm_struct *mm = vma->vm_mm; |
d5957d2f | 2075 | unsigned long remap_pfn = pfn; |
1da177e4 LT |
2076 | int err; |
2077 | ||
2078 | /* | |
2079 | * Physically remapped pages are special. Tell the | |
2080 | * rest of the world about it: | |
2081 | * VM_IO tells people not to look at these pages | |
2082 | * (accesses can have side effects). | |
6aab341e LT |
2083 | * VM_PFNMAP tells the core MM that the base pages are just |
2084 | * raw PFN mappings, and do not have a "struct page" associated | |
2085 | * with them. | |
314e51b9 KK |
2086 | * VM_DONTEXPAND |
2087 | * Disable vma merging and expanding with mremap(). | |
2088 | * VM_DONTDUMP | |
2089 | * Omit vma from core dump, even when VM_IO turned off. | |
fb155c16 LT |
2090 | * |
2091 | * There's a horrible special case to handle copy-on-write | |
2092 | * behaviour that some programs depend on. We mark the "original" | |
2093 | * un-COW'ed pages by matching them up with "vma->vm_pgoff". | |
b3b9c293 | 2094 | * See vm_normal_page() for details. |
1da177e4 | 2095 | */ |
b3b9c293 KK |
2096 | if (is_cow_mapping(vma->vm_flags)) { |
2097 | if (addr != vma->vm_start || end != vma->vm_end) | |
2098 | return -EINVAL; | |
fb155c16 | 2099 | vma->vm_pgoff = pfn; |
b3b9c293 KK |
2100 | } |
2101 | ||
d5957d2f | 2102 | err = track_pfn_remap(vma, &prot, remap_pfn, addr, PAGE_ALIGN(size)); |
b3b9c293 | 2103 | if (err) |
3c8bb73a | 2104 | return -EINVAL; |
fb155c16 | 2105 | |
314e51b9 | 2106 | vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; |
1da177e4 LT |
2107 | |
2108 | BUG_ON(addr >= end); | |
2109 | pfn -= addr >> PAGE_SHIFT; | |
2110 | pgd = pgd_offset(mm, addr); | |
2111 | flush_cache_range(vma, addr, end); | |
1da177e4 LT |
2112 | do { |
2113 | next = pgd_addr_end(addr, end); | |
c2febafc | 2114 | err = remap_p4d_range(mm, pgd, addr, next, |
1da177e4 LT |
2115 | pfn + (addr >> PAGE_SHIFT), prot); |
2116 | if (err) | |
2117 | break; | |
2118 | } while (pgd++, addr = next, addr != end); | |
2ab64037 | 2119 | |
2120 | if (err) | |
d5957d2f | 2121 | untrack_pfn(vma, remap_pfn, PAGE_ALIGN(size)); |
2ab64037 | 2122 | |
1da177e4 LT |
2123 | return err; |
2124 | } | |
2125 | EXPORT_SYMBOL(remap_pfn_range); | |
2126 | ||
b4cbb197 LT |
2127 | /** |
2128 | * vm_iomap_memory - remap memory to userspace | |
2129 | * @vma: user vma to map to | |
2130 | * @start: start of area | |
2131 | * @len: size of area | |
2132 | * | |
2133 | * This is a simplified io_remap_pfn_range() for common driver use. The | |
2134 | * driver just needs to give us the physical memory range to be mapped, | |
2135 | * we'll figure out the rest from the vma information. | |
2136 | * | |
2137 | * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get | |
2138 | * whatever write-combining details or similar. | |
2139 | */ | |
2140 | int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) | |
2141 | { | |
2142 | unsigned long vm_len, pfn, pages; | |
2143 | ||
2144 | /* Check that the physical memory area passed in looks valid */ | |
2145 | if (start + len < start) | |
2146 | return -EINVAL; | |
2147 | /* | |
2148 | * You *really* shouldn't map things that aren't page-aligned, | |
2149 | * but we've historically allowed it because IO memory might | |
2150 | * just have smaller alignment. | |
2151 | */ | |
2152 | len += start & ~PAGE_MASK; | |
2153 | pfn = start >> PAGE_SHIFT; | |
2154 | pages = (len + ~PAGE_MASK) >> PAGE_SHIFT; | |
2155 | if (pfn + pages < pfn) | |
2156 | return -EINVAL; | |
2157 | ||
2158 | /* We start the mapping 'vm_pgoff' pages into the area */ | |
2159 | if (vma->vm_pgoff > pages) | |
2160 | return -EINVAL; | |
2161 | pfn += vma->vm_pgoff; | |
2162 | pages -= vma->vm_pgoff; | |
2163 | ||
2164 | /* Can we fit all of the mapping? */ | |
2165 | vm_len = vma->vm_end - vma->vm_start; | |
2166 | if (vm_len >> PAGE_SHIFT > pages) | |
2167 | return -EINVAL; | |
2168 | ||
2169 | /* Ok, let it rip */ | |
2170 | return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); | |
2171 | } | |
2172 | EXPORT_SYMBOL(vm_iomap_memory); | |
2173 | ||
aee16b3c JF |
2174 | static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, |
2175 | unsigned long addr, unsigned long end, | |
2176 | pte_fn_t fn, void *data) | |
2177 | { | |
2178 | pte_t *pte; | |
2179 | int err; | |
2f569afd | 2180 | pgtable_t token; |
94909914 | 2181 | spinlock_t *uninitialized_var(ptl); |
aee16b3c JF |
2182 | |
2183 | pte = (mm == &init_mm) ? | |
2184 | pte_alloc_kernel(pmd, addr) : | |
2185 | pte_alloc_map_lock(mm, pmd, addr, &ptl); | |
2186 | if (!pte) | |
2187 | return -ENOMEM; | |
2188 | ||
2189 | BUG_ON(pmd_huge(*pmd)); | |
2190 | ||
38e0edb1 JF |
2191 | arch_enter_lazy_mmu_mode(); |
2192 | ||
2f569afd | 2193 | token = pmd_pgtable(*pmd); |
aee16b3c JF |
2194 | |
2195 | do { | |
c36987e2 | 2196 | err = fn(pte++, token, addr, data); |
aee16b3c JF |
2197 | if (err) |
2198 | break; | |
c36987e2 | 2199 | } while (addr += PAGE_SIZE, addr != end); |
aee16b3c | 2200 | |
38e0edb1 JF |
2201 | arch_leave_lazy_mmu_mode(); |
2202 | ||
aee16b3c JF |
2203 | if (mm != &init_mm) |
2204 | pte_unmap_unlock(pte-1, ptl); | |
2205 | return err; | |
2206 | } | |
2207 | ||
2208 | static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, | |
2209 | unsigned long addr, unsigned long end, | |
2210 | pte_fn_t fn, void *data) | |
2211 | { | |
2212 | pmd_t *pmd; | |
2213 | unsigned long next; | |
2214 | int err; | |
2215 | ||
ceb86879 AK |
2216 | BUG_ON(pud_huge(*pud)); |
2217 | ||
aee16b3c JF |
2218 | pmd = pmd_alloc(mm, pud, addr); |
2219 | if (!pmd) | |
2220 | return -ENOMEM; | |
2221 | do { | |
2222 | next = pmd_addr_end(addr, end); | |
2223 | err = apply_to_pte_range(mm, pmd, addr, next, fn, data); | |
2224 | if (err) | |
2225 | break; | |
2226 | } while (pmd++, addr = next, addr != end); | |
2227 | return err; | |
2228 | } | |
2229 | ||
c2febafc | 2230 | static int apply_to_pud_range(struct mm_struct *mm, p4d_t *p4d, |
aee16b3c JF |
2231 | unsigned long addr, unsigned long end, |
2232 | pte_fn_t fn, void *data) | |
2233 | { | |
2234 | pud_t *pud; | |
2235 | unsigned long next; | |
2236 | int err; | |
2237 | ||
c2febafc | 2238 | pud = pud_alloc(mm, p4d, addr); |
aee16b3c JF |
2239 | if (!pud) |
2240 | return -ENOMEM; | |
2241 | do { | |
2242 | next = pud_addr_end(addr, end); | |
2243 | err = apply_to_pmd_range(mm, pud, addr, next, fn, data); | |
2244 | if (err) | |
2245 | break; | |
2246 | } while (pud++, addr = next, addr != end); | |
2247 | return err; | |
2248 | } | |
2249 | ||
c2febafc KS |
2250 | static int apply_to_p4d_range(struct mm_struct *mm, pgd_t *pgd, |
2251 | unsigned long addr, unsigned long end, | |
2252 | pte_fn_t fn, void *data) | |
2253 | { | |
2254 | p4d_t *p4d; | |
2255 | unsigned long next; | |
2256 | int err; | |
2257 | ||
2258 | p4d = p4d_alloc(mm, pgd, addr); | |
2259 | if (!p4d) | |
2260 | return -ENOMEM; | |
2261 | do { | |
2262 | next = p4d_addr_end(addr, end); | |
2263 | err = apply_to_pud_range(mm, p4d, addr, next, fn, data); | |
2264 | if (err) | |
2265 | break; | |
2266 | } while (p4d++, addr = next, addr != end); | |
2267 | return err; | |
2268 | } | |
2269 | ||
aee16b3c JF |
2270 | /* |
2271 | * Scan a region of virtual memory, filling in page tables as necessary | |
2272 | * and calling a provided function on each leaf page table. | |
2273 | */ | |
2274 | int apply_to_page_range(struct mm_struct *mm, unsigned long addr, | |
2275 | unsigned long size, pte_fn_t fn, void *data) | |
2276 | { | |
2277 | pgd_t *pgd; | |
2278 | unsigned long next; | |
57250a5b | 2279 | unsigned long end = addr + size; |
aee16b3c JF |
2280 | int err; |
2281 | ||
9cb65bc3 MP |
2282 | if (WARN_ON(addr >= end)) |
2283 | return -EINVAL; | |
2284 | ||
aee16b3c JF |
2285 | pgd = pgd_offset(mm, addr); |
2286 | do { | |
2287 | next = pgd_addr_end(addr, end); | |
c2febafc | 2288 | err = apply_to_p4d_range(mm, pgd, addr, next, fn, data); |
aee16b3c JF |
2289 | if (err) |
2290 | break; | |
2291 | } while (pgd++, addr = next, addr != end); | |
57250a5b | 2292 | |
aee16b3c JF |
2293 | return err; |
2294 | } | |
2295 | EXPORT_SYMBOL_GPL(apply_to_page_range); | |
2296 | ||
8f4e2101 | 2297 | /* |
9b4bdd2f KS |
2298 | * handle_pte_fault chooses page fault handler according to an entry which was |
2299 | * read non-atomically. Before making any commitment, on those architectures | |
2300 | * or configurations (e.g. i386 with PAE) which might give a mix of unmatched | |
2301 | * parts, do_swap_page must check under lock before unmapping the pte and | |
2302 | * proceeding (but do_wp_page is only called after already making such a check; | |
a335b2e1 | 2303 | * and do_anonymous_page can safely check later on). |
8f4e2101 | 2304 | */ |
4c21e2f2 | 2305 | static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd, |
8f4e2101 HD |
2306 | pte_t *page_table, pte_t orig_pte) |
2307 | { | |
2308 | int same = 1; | |
2309 | #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT) | |
2310 | if (sizeof(pte_t) > sizeof(unsigned long)) { | |
4c21e2f2 HD |
2311 | spinlock_t *ptl = pte_lockptr(mm, pmd); |
2312 | spin_lock(ptl); | |
8f4e2101 | 2313 | same = pte_same(*page_table, orig_pte); |
4c21e2f2 | 2314 | spin_unlock(ptl); |
8f4e2101 HD |
2315 | } |
2316 | #endif | |
2317 | pte_unmap(page_table); | |
2318 | return same; | |
2319 | } | |
2320 | ||
9de455b2 | 2321 | static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma) |
6aab341e | 2322 | { |
0abdd7a8 DW |
2323 | debug_dma_assert_idle(src); |
2324 | ||
6aab341e LT |
2325 | /* |
2326 | * If the source page was a PFN mapping, we don't have | |
2327 | * a "struct page" for it. We do a best-effort copy by | |
2328 | * just copying from the original user address. If that | |
2329 | * fails, we just zero-fill it. Live with it. | |
2330 | */ | |
2331 | if (unlikely(!src)) { | |
9b04c5fe | 2332 | void *kaddr = kmap_atomic(dst); |
5d2a2dbb LT |
2333 | void __user *uaddr = (void __user *)(va & PAGE_MASK); |
2334 | ||
2335 | /* | |
2336 | * This really shouldn't fail, because the page is there | |
2337 | * in the page tables. But it might just be unreadable, | |
2338 | * in which case we just give up and fill the result with | |
2339 | * zeroes. | |
2340 | */ | |
2341 | if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) | |
3ecb01df | 2342 | clear_page(kaddr); |
9b04c5fe | 2343 | kunmap_atomic(kaddr); |
c4ec7b0d | 2344 | flush_dcache_page(dst); |
0ed361de NP |
2345 | } else |
2346 | copy_user_highpage(dst, src, va, vma); | |
6aab341e LT |
2347 | } |
2348 | ||
c20cd45e MH |
2349 | static gfp_t __get_fault_gfp_mask(struct vm_area_struct *vma) |
2350 | { | |
2351 | struct file *vm_file = vma->vm_file; | |
2352 | ||
2353 | if (vm_file) | |
2354 | return mapping_gfp_mask(vm_file->f_mapping) | __GFP_FS | __GFP_IO; | |
2355 | ||
2356 | /* | |
2357 | * Special mappings (e.g. VDSO) do not have any file so fake | |
2358 | * a default GFP_KERNEL for them. | |
2359 | */ | |
2360 | return GFP_KERNEL; | |
2361 | } | |
2362 | ||
fb09a464 KS |
2363 | /* |
2364 | * Notify the address space that the page is about to become writable so that | |
2365 | * it can prohibit this or wait for the page to get into an appropriate state. | |
2366 | * | |
2367 | * We do this without the lock held, so that it can sleep if it needs to. | |
2368 | */ | |
38b8cb7f | 2369 | static int do_page_mkwrite(struct vm_fault *vmf) |
fb09a464 | 2370 | { |
fb09a464 | 2371 | int ret; |
38b8cb7f JK |
2372 | struct page *page = vmf->page; |
2373 | unsigned int old_flags = vmf->flags; | |
fb09a464 | 2374 | |
38b8cb7f | 2375 | vmf->flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; |
fb09a464 | 2376 | |
11bac800 | 2377 | ret = vmf->vma->vm_ops->page_mkwrite(vmf); |
38b8cb7f JK |
2378 | /* Restore original flags so that caller is not surprised */ |
2379 | vmf->flags = old_flags; | |
fb09a464 KS |
2380 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) |
2381 | return ret; | |
2382 | if (unlikely(!(ret & VM_FAULT_LOCKED))) { | |
2383 | lock_page(page); | |
2384 | if (!page->mapping) { | |
2385 | unlock_page(page); | |
2386 | return 0; /* retry */ | |
2387 | } | |
2388 | ret |= VM_FAULT_LOCKED; | |
2389 | } else | |
2390 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
2391 | return ret; | |
2392 | } | |
2393 | ||
97ba0c2b JK |
2394 | /* |
2395 | * Handle dirtying of a page in shared file mapping on a write fault. | |
2396 | * | |
2397 | * The function expects the page to be locked and unlocks it. | |
2398 | */ | |
2399 | static void fault_dirty_shared_page(struct vm_area_struct *vma, | |
2400 | struct page *page) | |
2401 | { | |
2402 | struct address_space *mapping; | |
2403 | bool dirtied; | |
2404 | bool page_mkwrite = vma->vm_ops && vma->vm_ops->page_mkwrite; | |
2405 | ||
2406 | dirtied = set_page_dirty(page); | |
2407 | VM_BUG_ON_PAGE(PageAnon(page), page); | |
2408 | /* | |
2409 | * Take a local copy of the address_space - page.mapping may be zeroed | |
2410 | * by truncate after unlock_page(). The address_space itself remains | |
2411 | * pinned by vma->vm_file's reference. We rely on unlock_page()'s | |
2412 | * release semantics to prevent the compiler from undoing this copying. | |
2413 | */ | |
2414 | mapping = page_rmapping(page); | |
2415 | unlock_page(page); | |
2416 | ||
2417 | if ((dirtied || page_mkwrite) && mapping) { | |
2418 | /* | |
2419 | * Some device drivers do not set page.mapping | |
2420 | * but still dirty their pages | |
2421 | */ | |
2422 | balance_dirty_pages_ratelimited(mapping); | |
2423 | } | |
2424 | ||
2425 | if (!page_mkwrite) | |
2426 | file_update_time(vma->vm_file); | |
2427 | } | |
2428 | ||
4e047f89 SR |
2429 | /* |
2430 | * Handle write page faults for pages that can be reused in the current vma | |
2431 | * | |
2432 | * This can happen either due to the mapping being with the VM_SHARED flag, | |
2433 | * or due to us being the last reference standing to the page. In either | |
2434 | * case, all we need to do here is to mark the page as writable and update | |
2435 | * any related book-keeping. | |
2436 | */ | |
997dd98d | 2437 | static inline void wp_page_reuse(struct vm_fault *vmf) |
82b0f8c3 | 2438 | __releases(vmf->ptl) |
4e047f89 | 2439 | { |
82b0f8c3 | 2440 | struct vm_area_struct *vma = vmf->vma; |
a41b70d6 | 2441 | struct page *page = vmf->page; |
4e047f89 SR |
2442 | pte_t entry; |
2443 | /* | |
2444 | * Clear the pages cpupid information as the existing | |
2445 | * information potentially belongs to a now completely | |
2446 | * unrelated process. | |
2447 | */ | |
2448 | if (page) | |
2449 | page_cpupid_xchg_last(page, (1 << LAST_CPUPID_SHIFT) - 1); | |
2450 | ||
2994302b JK |
2451 | flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); |
2452 | entry = pte_mkyoung(vmf->orig_pte); | |
4e047f89 | 2453 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
82b0f8c3 JK |
2454 | if (ptep_set_access_flags(vma, vmf->address, vmf->pte, entry, 1)) |
2455 | update_mmu_cache(vma, vmf->address, vmf->pte); | |
2456 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
4e047f89 SR |
2457 | } |
2458 | ||
2f38ab2c SR |
2459 | /* |
2460 | * Handle the case of a page which we actually need to copy to a new page. | |
2461 | * | |
2462 | * Called with mmap_sem locked and the old page referenced, but | |
2463 | * without the ptl held. | |
2464 | * | |
2465 | * High level logic flow: | |
2466 | * | |
2467 | * - Allocate a page, copy the content of the old page to the new one. | |
2468 | * - Handle book keeping and accounting - cgroups, mmu-notifiers, etc. | |
2469 | * - Take the PTL. If the pte changed, bail out and release the allocated page | |
2470 | * - If the pte is still the way we remember it, update the page table and all | |
2471 | * relevant references. This includes dropping the reference the page-table | |
2472 | * held to the old page, as well as updating the rmap. | |
2473 | * - In any case, unlock the PTL and drop the reference we took to the old page. | |
2474 | */ | |
a41b70d6 | 2475 | static int wp_page_copy(struct vm_fault *vmf) |
2f38ab2c | 2476 | { |
82b0f8c3 | 2477 | struct vm_area_struct *vma = vmf->vma; |
bae473a4 | 2478 | struct mm_struct *mm = vma->vm_mm; |
a41b70d6 | 2479 | struct page *old_page = vmf->page; |
2f38ab2c | 2480 | struct page *new_page = NULL; |
2f38ab2c SR |
2481 | pte_t entry; |
2482 | int page_copied = 0; | |
82b0f8c3 | 2483 | const unsigned long mmun_start = vmf->address & PAGE_MASK; |
bae473a4 | 2484 | const unsigned long mmun_end = mmun_start + PAGE_SIZE; |
2f38ab2c SR |
2485 | struct mem_cgroup *memcg; |
2486 | ||
2487 | if (unlikely(anon_vma_prepare(vma))) | |
2488 | goto oom; | |
2489 | ||
2994302b | 2490 | if (is_zero_pfn(pte_pfn(vmf->orig_pte))) { |
82b0f8c3 JK |
2491 | new_page = alloc_zeroed_user_highpage_movable(vma, |
2492 | vmf->address); | |
2f38ab2c SR |
2493 | if (!new_page) |
2494 | goto oom; | |
2495 | } else { | |
bae473a4 | 2496 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, |
82b0f8c3 | 2497 | vmf->address); |
2f38ab2c SR |
2498 | if (!new_page) |
2499 | goto oom; | |
82b0f8c3 | 2500 | cow_user_page(new_page, old_page, vmf->address, vma); |
2f38ab2c | 2501 | } |
2f38ab2c | 2502 | |
f627c2f5 | 2503 | if (mem_cgroup_try_charge(new_page, mm, GFP_KERNEL, &memcg, false)) |
2f38ab2c SR |
2504 | goto oom_free_new; |
2505 | ||
eb3c24f3 MG |
2506 | __SetPageUptodate(new_page); |
2507 | ||
2f38ab2c SR |
2508 | mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); |
2509 | ||
2510 | /* | |
2511 | * Re-check the pte - we dropped the lock | |
2512 | */ | |
82b0f8c3 | 2513 | vmf->pte = pte_offset_map_lock(mm, vmf->pmd, vmf->address, &vmf->ptl); |
2994302b | 2514 | if (likely(pte_same(*vmf->pte, vmf->orig_pte))) { |
2f38ab2c SR |
2515 | if (old_page) { |
2516 | if (!PageAnon(old_page)) { | |
eca56ff9 JM |
2517 | dec_mm_counter_fast(mm, |
2518 | mm_counter_file(old_page)); | |
2f38ab2c SR |
2519 | inc_mm_counter_fast(mm, MM_ANONPAGES); |
2520 | } | |
2521 | } else { | |
2522 | inc_mm_counter_fast(mm, MM_ANONPAGES); | |
2523 | } | |
2994302b | 2524 | flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); |
2f38ab2c SR |
2525 | entry = mk_pte(new_page, vma->vm_page_prot); |
2526 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
2527 | /* | |
2528 | * Clear the pte entry and flush it first, before updating the | |
2529 | * pte with the new entry. This will avoid a race condition | |
2530 | * seen in the presence of one thread doing SMC and another | |
2531 | * thread doing COW. | |
2532 | */ | |
82b0f8c3 JK |
2533 | ptep_clear_flush_notify(vma, vmf->address, vmf->pte); |
2534 | page_add_new_anon_rmap(new_page, vma, vmf->address, false); | |
f627c2f5 | 2535 | mem_cgroup_commit_charge(new_page, memcg, false, false); |
2f38ab2c SR |
2536 | lru_cache_add_active_or_unevictable(new_page, vma); |
2537 | /* | |
2538 | * We call the notify macro here because, when using secondary | |
2539 | * mmu page tables (such as kvm shadow page tables), we want the | |
2540 | * new page to be mapped directly into the secondary page table. | |
2541 | */ | |
82b0f8c3 JK |
2542 | set_pte_at_notify(mm, vmf->address, vmf->pte, entry); |
2543 | update_mmu_cache(vma, vmf->address, vmf->pte); | |
2f38ab2c SR |
2544 | if (old_page) { |
2545 | /* | |
2546 | * Only after switching the pte to the new page may | |
2547 | * we remove the mapcount here. Otherwise another | |
2548 | * process may come and find the rmap count decremented | |
2549 | * before the pte is switched to the new page, and | |
2550 | * "reuse" the old page writing into it while our pte | |
2551 | * here still points into it and can be read by other | |
2552 | * threads. | |
2553 | * | |
2554 | * The critical issue is to order this | |
2555 | * page_remove_rmap with the ptp_clear_flush above. | |
2556 | * Those stores are ordered by (if nothing else,) | |
2557 | * the barrier present in the atomic_add_negative | |
2558 | * in page_remove_rmap. | |
2559 | * | |
2560 | * Then the TLB flush in ptep_clear_flush ensures that | |
2561 | * no process can access the old page before the | |
2562 | * decremented mapcount is visible. And the old page | |
2563 | * cannot be reused until after the decremented | |
2564 | * mapcount is visible. So transitively, TLBs to | |
2565 | * old page will be flushed before it can be reused. | |
2566 | */ | |
d281ee61 | 2567 | page_remove_rmap(old_page, false); |
2f38ab2c SR |
2568 | } |
2569 | ||
2570 | /* Free the old page.. */ | |
2571 | new_page = old_page; | |
2572 | page_copied = 1; | |
2573 | } else { | |
f627c2f5 | 2574 | mem_cgroup_cancel_charge(new_page, memcg, false); |
2f38ab2c SR |
2575 | } |
2576 | ||
2577 | if (new_page) | |
09cbfeaf | 2578 | put_page(new_page); |
2f38ab2c | 2579 | |
82b0f8c3 | 2580 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
2f38ab2c SR |
2581 | mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); |
2582 | if (old_page) { | |
2583 | /* | |
2584 | * Don't let another task, with possibly unlocked vma, | |
2585 | * keep the mlocked page. | |
2586 | */ | |
2587 | if (page_copied && (vma->vm_flags & VM_LOCKED)) { | |
2588 | lock_page(old_page); /* LRU manipulation */ | |
e90309c9 KS |
2589 | if (PageMlocked(old_page)) |
2590 | munlock_vma_page(old_page); | |
2f38ab2c SR |
2591 | unlock_page(old_page); |
2592 | } | |
09cbfeaf | 2593 | put_page(old_page); |
2f38ab2c SR |
2594 | } |
2595 | return page_copied ? VM_FAULT_WRITE : 0; | |
2596 | oom_free_new: | |
09cbfeaf | 2597 | put_page(new_page); |
2f38ab2c SR |
2598 | oom: |
2599 | if (old_page) | |
09cbfeaf | 2600 | put_page(old_page); |
2f38ab2c SR |
2601 | return VM_FAULT_OOM; |
2602 | } | |
2603 | ||
66a6197c JK |
2604 | /** |
2605 | * finish_mkwrite_fault - finish page fault for a shared mapping, making PTE | |
2606 | * writeable once the page is prepared | |
2607 | * | |
2608 | * @vmf: structure describing the fault | |
2609 | * | |
2610 | * This function handles all that is needed to finish a write page fault in a | |
2611 | * shared mapping due to PTE being read-only once the mapped page is prepared. | |
2612 | * It handles locking of PTE and modifying it. The function returns | |
2613 | * VM_FAULT_WRITE on success, 0 when PTE got changed before we acquired PTE | |
2614 | * lock. | |
2615 | * | |
2616 | * The function expects the page to be locked or other protection against | |
2617 | * concurrent faults / writeback (such as DAX radix tree locks). | |
2618 | */ | |
2619 | int finish_mkwrite_fault(struct vm_fault *vmf) | |
2620 | { | |
2621 | WARN_ON_ONCE(!(vmf->vma->vm_flags & VM_SHARED)); | |
2622 | vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, vmf->address, | |
2623 | &vmf->ptl); | |
2624 | /* | |
2625 | * We might have raced with another page fault while we released the | |
2626 | * pte_offset_map_lock. | |
2627 | */ | |
2628 | if (!pte_same(*vmf->pte, vmf->orig_pte)) { | |
2629 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
a19e2553 | 2630 | return VM_FAULT_NOPAGE; |
66a6197c JK |
2631 | } |
2632 | wp_page_reuse(vmf); | |
a19e2553 | 2633 | return 0; |
66a6197c JK |
2634 | } |
2635 | ||
dd906184 BH |
2636 | /* |
2637 | * Handle write page faults for VM_MIXEDMAP or VM_PFNMAP for a VM_SHARED | |
2638 | * mapping | |
2639 | */ | |
2994302b | 2640 | static int wp_pfn_shared(struct vm_fault *vmf) |
dd906184 | 2641 | { |
82b0f8c3 | 2642 | struct vm_area_struct *vma = vmf->vma; |
bae473a4 | 2643 | |
dd906184 | 2644 | if (vma->vm_ops && vma->vm_ops->pfn_mkwrite) { |
dd906184 BH |
2645 | int ret; |
2646 | ||
82b0f8c3 | 2647 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
fe82221f | 2648 | vmf->flags |= FAULT_FLAG_MKWRITE; |
11bac800 | 2649 | ret = vma->vm_ops->pfn_mkwrite(vmf); |
2f89dc12 | 2650 | if (ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)) |
dd906184 | 2651 | return ret; |
66a6197c | 2652 | return finish_mkwrite_fault(vmf); |
dd906184 | 2653 | } |
997dd98d JK |
2654 | wp_page_reuse(vmf); |
2655 | return VM_FAULT_WRITE; | |
dd906184 BH |
2656 | } |
2657 | ||
a41b70d6 | 2658 | static int wp_page_shared(struct vm_fault *vmf) |
82b0f8c3 | 2659 | __releases(vmf->ptl) |
93e478d4 | 2660 | { |
82b0f8c3 | 2661 | struct vm_area_struct *vma = vmf->vma; |
93e478d4 | 2662 | |
a41b70d6 | 2663 | get_page(vmf->page); |
93e478d4 | 2664 | |
93e478d4 SR |
2665 | if (vma->vm_ops && vma->vm_ops->page_mkwrite) { |
2666 | int tmp; | |
2667 | ||
82b0f8c3 | 2668 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
38b8cb7f | 2669 | tmp = do_page_mkwrite(vmf); |
93e478d4 SR |
2670 | if (unlikely(!tmp || (tmp & |
2671 | (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { | |
a41b70d6 | 2672 | put_page(vmf->page); |
93e478d4 SR |
2673 | return tmp; |
2674 | } | |
66a6197c | 2675 | tmp = finish_mkwrite_fault(vmf); |
a19e2553 | 2676 | if (unlikely(tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { |
a41b70d6 | 2677 | unlock_page(vmf->page); |
a41b70d6 | 2678 | put_page(vmf->page); |
66a6197c | 2679 | return tmp; |
93e478d4 | 2680 | } |
66a6197c JK |
2681 | } else { |
2682 | wp_page_reuse(vmf); | |
997dd98d | 2683 | lock_page(vmf->page); |
93e478d4 | 2684 | } |
997dd98d JK |
2685 | fault_dirty_shared_page(vma, vmf->page); |
2686 | put_page(vmf->page); | |
93e478d4 | 2687 | |
997dd98d | 2688 | return VM_FAULT_WRITE; |
93e478d4 SR |
2689 | } |
2690 | ||
1da177e4 LT |
2691 | /* |
2692 | * This routine handles present pages, when users try to write | |
2693 | * to a shared page. It is done by copying the page to a new address | |
2694 | * and decrementing the shared-page counter for the old page. | |
2695 | * | |
1da177e4 LT |
2696 | * Note that this routine assumes that the protection checks have been |
2697 | * done by the caller (the low-level page fault routine in most cases). | |
2698 | * Thus we can safely just mark it writable once we've done any necessary | |
2699 | * COW. | |
2700 | * | |
2701 | * We also mark the page dirty at this point even though the page will | |
2702 | * change only once the write actually happens. This avoids a few races, | |
2703 | * and potentially makes it more efficient. | |
2704 | * | |
8f4e2101 HD |
2705 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2706 | * but allow concurrent faults), with pte both mapped and locked. | |
2707 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2708 | */ |
2994302b | 2709 | static int do_wp_page(struct vm_fault *vmf) |
82b0f8c3 | 2710 | __releases(vmf->ptl) |
1da177e4 | 2711 | { |
82b0f8c3 | 2712 | struct vm_area_struct *vma = vmf->vma; |
1da177e4 | 2713 | |
a41b70d6 JK |
2714 | vmf->page = vm_normal_page(vma, vmf->address, vmf->orig_pte); |
2715 | if (!vmf->page) { | |
251b97f5 | 2716 | /* |
64e45507 PF |
2717 | * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a |
2718 | * VM_PFNMAP VMA. | |
251b97f5 PZ |
2719 | * |
2720 | * We should not cow pages in a shared writeable mapping. | |
dd906184 | 2721 | * Just mark the pages writable and/or call ops->pfn_mkwrite. |
251b97f5 PZ |
2722 | */ |
2723 | if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) == | |
2724 | (VM_WRITE|VM_SHARED)) | |
2994302b | 2725 | return wp_pfn_shared(vmf); |
2f38ab2c | 2726 | |
82b0f8c3 | 2727 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 2728 | return wp_page_copy(vmf); |
251b97f5 | 2729 | } |
1da177e4 | 2730 | |
d08b3851 | 2731 | /* |
ee6a6457 PZ |
2732 | * Take out anonymous pages first, anonymous shared vmas are |
2733 | * not dirty accountable. | |
d08b3851 | 2734 | */ |
a41b70d6 | 2735 | if (PageAnon(vmf->page) && !PageKsm(vmf->page)) { |
ba3c4ce6 | 2736 | int total_map_swapcount; |
a41b70d6 JK |
2737 | if (!trylock_page(vmf->page)) { |
2738 | get_page(vmf->page); | |
82b0f8c3 | 2739 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 2740 | lock_page(vmf->page); |
82b0f8c3 JK |
2741 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
2742 | vmf->address, &vmf->ptl); | |
2994302b | 2743 | if (!pte_same(*vmf->pte, vmf->orig_pte)) { |
a41b70d6 | 2744 | unlock_page(vmf->page); |
82b0f8c3 | 2745 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 2746 | put_page(vmf->page); |
28766805 | 2747 | return 0; |
ab967d86 | 2748 | } |
a41b70d6 | 2749 | put_page(vmf->page); |
ee6a6457 | 2750 | } |
ba3c4ce6 HY |
2751 | if (reuse_swap_page(vmf->page, &total_map_swapcount)) { |
2752 | if (total_map_swapcount == 1) { | |
6d0a07ed AA |
2753 | /* |
2754 | * The page is all ours. Move it to | |
2755 | * our anon_vma so the rmap code will | |
2756 | * not search our parent or siblings. | |
2757 | * Protected against the rmap code by | |
2758 | * the page lock. | |
2759 | */ | |
a41b70d6 | 2760 | page_move_anon_rmap(vmf->page, vma); |
6d0a07ed | 2761 | } |
a41b70d6 | 2762 | unlock_page(vmf->page); |
997dd98d JK |
2763 | wp_page_reuse(vmf); |
2764 | return VM_FAULT_WRITE; | |
b009c024 | 2765 | } |
a41b70d6 | 2766 | unlock_page(vmf->page); |
ee6a6457 | 2767 | } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) == |
d08b3851 | 2768 | (VM_WRITE|VM_SHARED))) { |
a41b70d6 | 2769 | return wp_page_shared(vmf); |
1da177e4 | 2770 | } |
1da177e4 LT |
2771 | |
2772 | /* | |
2773 | * Ok, we need to copy. Oh, well.. | |
2774 | */ | |
a41b70d6 | 2775 | get_page(vmf->page); |
28766805 | 2776 | |
82b0f8c3 | 2777 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 2778 | return wp_page_copy(vmf); |
1da177e4 LT |
2779 | } |
2780 | ||
97a89413 | 2781 | static void unmap_mapping_range_vma(struct vm_area_struct *vma, |
1da177e4 LT |
2782 | unsigned long start_addr, unsigned long end_addr, |
2783 | struct zap_details *details) | |
2784 | { | |
f5cc4eef | 2785 | zap_page_range_single(vma, start_addr, end_addr - start_addr, details); |
1da177e4 LT |
2786 | } |
2787 | ||
f808c13f | 2788 | static inline void unmap_mapping_range_tree(struct rb_root_cached *root, |
1da177e4 LT |
2789 | struct zap_details *details) |
2790 | { | |
2791 | struct vm_area_struct *vma; | |
1da177e4 LT |
2792 | pgoff_t vba, vea, zba, zea; |
2793 | ||
6b2dbba8 | 2794 | vma_interval_tree_foreach(vma, root, |
1da177e4 | 2795 | details->first_index, details->last_index) { |
1da177e4 LT |
2796 | |
2797 | vba = vma->vm_pgoff; | |
d6e93217 | 2798 | vea = vba + vma_pages(vma) - 1; |
1da177e4 LT |
2799 | zba = details->first_index; |
2800 | if (zba < vba) | |
2801 | zba = vba; | |
2802 | zea = details->last_index; | |
2803 | if (zea > vea) | |
2804 | zea = vea; | |
2805 | ||
97a89413 | 2806 | unmap_mapping_range_vma(vma, |
1da177e4 LT |
2807 | ((zba - vba) << PAGE_SHIFT) + vma->vm_start, |
2808 | ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, | |
97a89413 | 2809 | details); |
1da177e4 LT |
2810 | } |
2811 | } | |
2812 | ||
1da177e4 | 2813 | /** |
8a5f14a2 KS |
2814 | * unmap_mapping_range - unmap the portion of all mmaps in the specified |
2815 | * address_space corresponding to the specified page range in the underlying | |
2816 | * file. | |
2817 | * | |
3d41088f | 2818 | * @mapping: the address space containing mmaps to be unmapped. |
1da177e4 LT |
2819 | * @holebegin: byte in first page to unmap, relative to the start of |
2820 | * the underlying file. This will be rounded down to a PAGE_SIZE | |
25d9e2d1 | 2821 | * boundary. Note that this is different from truncate_pagecache(), which |
1da177e4 LT |
2822 | * must keep the partial page. In contrast, we must get rid of |
2823 | * partial pages. | |
2824 | * @holelen: size of prospective hole in bytes. This will be rounded | |
2825 | * up to a PAGE_SIZE boundary. A holelen of zero truncates to the | |
2826 | * end of the file. | |
2827 | * @even_cows: 1 when truncating a file, unmap even private COWed pages; | |
2828 | * but 0 when invalidating pagecache, don't throw away private data. | |
2829 | */ | |
2830 | void unmap_mapping_range(struct address_space *mapping, | |
2831 | loff_t const holebegin, loff_t const holelen, int even_cows) | |
2832 | { | |
aac45363 | 2833 | struct zap_details details = { }; |
1da177e4 LT |
2834 | pgoff_t hba = holebegin >> PAGE_SHIFT; |
2835 | pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
2836 | ||
2837 | /* Check for overflow. */ | |
2838 | if (sizeof(holelen) > sizeof(hlen)) { | |
2839 | long long holeend = | |
2840 | (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
2841 | if (holeend & ~(long long)ULONG_MAX) | |
2842 | hlen = ULONG_MAX - hba + 1; | |
2843 | } | |
2844 | ||
166f61b9 | 2845 | details.check_mapping = even_cows ? NULL : mapping; |
1da177e4 LT |
2846 | details.first_index = hba; |
2847 | details.last_index = hba + hlen - 1; | |
2848 | if (details.last_index < details.first_index) | |
2849 | details.last_index = ULONG_MAX; | |
1da177e4 | 2850 | |
46c043ed | 2851 | i_mmap_lock_write(mapping); |
f808c13f | 2852 | if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))) |
1da177e4 | 2853 | unmap_mapping_range_tree(&mapping->i_mmap, &details); |
46c043ed | 2854 | i_mmap_unlock_write(mapping); |
1da177e4 LT |
2855 | } |
2856 | EXPORT_SYMBOL(unmap_mapping_range); | |
2857 | ||
1da177e4 | 2858 | /* |
8f4e2101 HD |
2859 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2860 | * but allow concurrent faults), and pte mapped but not yet locked. | |
9a95f3cf PC |
2861 | * We return with pte unmapped and unlocked. |
2862 | * | |
2863 | * We return with the mmap_sem locked or unlocked in the same cases | |
2864 | * as does filemap_fault(). | |
1da177e4 | 2865 | */ |
2994302b | 2866 | int do_swap_page(struct vm_fault *vmf) |
1da177e4 | 2867 | { |
82b0f8c3 | 2868 | struct vm_area_struct *vma = vmf->vma; |
ec560175 | 2869 | struct page *page = NULL, *swapcache; |
00501b53 | 2870 | struct mem_cgroup *memcg; |
ec560175 | 2871 | struct vma_swap_readahead swap_ra; |
65500d23 | 2872 | swp_entry_t entry; |
1da177e4 | 2873 | pte_t pte; |
d065bd81 | 2874 | int locked; |
ad8c2ee8 | 2875 | int exclusive = 0; |
83c54070 | 2876 | int ret = 0; |
ec560175 | 2877 | bool vma_readahead = swap_use_vma_readahead(); |
1da177e4 | 2878 | |
ec560175 HY |
2879 | if (vma_readahead) |
2880 | page = swap_readahead_detect(vmf, &swap_ra); | |
2881 | if (!pte_unmap_same(vma->vm_mm, vmf->pmd, vmf->pte, vmf->orig_pte)) { | |
2882 | if (page) | |
2883 | put_page(page); | |
8f4e2101 | 2884 | goto out; |
ec560175 | 2885 | } |
65500d23 | 2886 | |
2994302b | 2887 | entry = pte_to_swp_entry(vmf->orig_pte); |
d1737fdb AK |
2888 | if (unlikely(non_swap_entry(entry))) { |
2889 | if (is_migration_entry(entry)) { | |
82b0f8c3 JK |
2890 | migration_entry_wait(vma->vm_mm, vmf->pmd, |
2891 | vmf->address); | |
5042db43 JG |
2892 | } else if (is_device_private_entry(entry)) { |
2893 | /* | |
2894 | * For un-addressable device memory we call the pgmap | |
2895 | * fault handler callback. The callback must migrate | |
2896 | * the page back to some CPU accessible page. | |
2897 | */ | |
2898 | ret = device_private_entry_fault(vma, vmf->address, entry, | |
2899 | vmf->flags, vmf->pmd); | |
d1737fdb AK |
2900 | } else if (is_hwpoison_entry(entry)) { |
2901 | ret = VM_FAULT_HWPOISON; | |
2902 | } else { | |
2994302b | 2903 | print_bad_pte(vma, vmf->address, vmf->orig_pte, NULL); |
d99be1a8 | 2904 | ret = VM_FAULT_SIGBUS; |
d1737fdb | 2905 | } |
0697212a CL |
2906 | goto out; |
2907 | } | |
0ff92245 | 2908 | delayacct_set_flag(DELAYACCT_PF_SWAPIN); |
ec560175 HY |
2909 | if (!page) |
2910 | page = lookup_swap_cache(entry, vma_readahead ? vma : NULL, | |
2911 | vmf->address); | |
1da177e4 | 2912 | if (!page) { |
ec560175 HY |
2913 | if (vma_readahead) |
2914 | page = do_swap_page_readahead(entry, | |
2915 | GFP_HIGHUSER_MOVABLE, vmf, &swap_ra); | |
2916 | else | |
2917 | page = swapin_readahead(entry, | |
2918 | GFP_HIGHUSER_MOVABLE, vma, vmf->address); | |
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 | ||
56f31801 | 2947 | swapcache = page; |
82b0f8c3 | 2948 | locked = lock_page_or_retry(page, vma->vm_mm, vmf->flags); |
e709ffd6 | 2949 | |
073e587e | 2950 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
d065bd81 ML |
2951 | if (!locked) { |
2952 | ret |= VM_FAULT_RETRY; | |
2953 | goto out_release; | |
2954 | } | |
073e587e | 2955 | |
4969c119 | 2956 | /* |
31c4a3d3 HD |
2957 | * Make sure try_to_free_swap or reuse_swap_page or swapoff did not |
2958 | * release the swapcache from under us. The page pin, and pte_same | |
2959 | * test below, are not enough to exclude that. Even if it is still | |
2960 | * swapcache, we need to check that the page's swap has not changed. | |
4969c119 | 2961 | */ |
31c4a3d3 | 2962 | if (unlikely(!PageSwapCache(page) || page_private(page) != entry.val)) |
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; |
00501b53 | 3015 | if (page == swapcache) { |
82b0f8c3 | 3016 | do_page_add_anon_rmap(page, vma, vmf->address, exclusive); |
f627c2f5 | 3017 | mem_cgroup_commit_charge(page, memcg, true, false); |
1a8018fb | 3018 | activate_page(page); |
00501b53 | 3019 | } else { /* ksm created a completely new copy */ |
82b0f8c3 | 3020 | page_add_new_anon_rmap(page, vma, vmf->address, false); |
f627c2f5 | 3021 | mem_cgroup_commit_charge(page, memcg, false, false); |
00501b53 JW |
3022 | lru_cache_add_active_or_unevictable(page, vma); |
3023 | } | |
1da177e4 | 3024 | |
c475a8ab | 3025 | swap_free(entry); |
5ccc5aba VD |
3026 | if (mem_cgroup_swap_full(page) || |
3027 | (vma->vm_flags & VM_LOCKED) || PageMlocked(page)) | |
a2c43eed | 3028 | try_to_free_swap(page); |
c475a8ab | 3029 | unlock_page(page); |
56f31801 | 3030 | if (page != swapcache) { |
4969c119 AA |
3031 | /* |
3032 | * Hold the lock to avoid the swap entry to be reused | |
3033 | * until we take the PT lock for the pte_same() check | |
3034 | * (to avoid false positives from pte_same). For | |
3035 | * further safety release the lock after the swap_free | |
3036 | * so that the swap count won't change under a | |
3037 | * parallel locked swapcache. | |
3038 | */ | |
3039 | unlock_page(swapcache); | |
09cbfeaf | 3040 | put_page(swapcache); |
4969c119 | 3041 | } |
c475a8ab | 3042 | |
82b0f8c3 | 3043 | if (vmf->flags & FAULT_FLAG_WRITE) { |
2994302b | 3044 | ret |= do_wp_page(vmf); |
61469f1d HD |
3045 | if (ret & VM_FAULT_ERROR) |
3046 | ret &= VM_FAULT_ERROR; | |
1da177e4 LT |
3047 | goto out; |
3048 | } | |
3049 | ||
3050 | /* No need to invalidate - it was non-present before */ | |
82b0f8c3 | 3051 | update_mmu_cache(vma, vmf->address, vmf->pte); |
65500d23 | 3052 | unlock: |
82b0f8c3 | 3053 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
1da177e4 LT |
3054 | out: |
3055 | return ret; | |
b8107480 | 3056 | out_nomap: |
f627c2f5 | 3057 | mem_cgroup_cancel_charge(page, memcg, false); |
82b0f8c3 | 3058 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
bc43f75c | 3059 | out_page: |
b8107480 | 3060 | unlock_page(page); |
4779cb31 | 3061 | out_release: |
09cbfeaf | 3062 | put_page(page); |
56f31801 | 3063 | if (page != swapcache) { |
4969c119 | 3064 | unlock_page(swapcache); |
09cbfeaf | 3065 | put_page(swapcache); |
4969c119 | 3066 | } |
65500d23 | 3067 | return ret; |
1da177e4 LT |
3068 | } |
3069 | ||
3070 | /* | |
8f4e2101 HD |
3071 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
3072 | * but allow concurrent faults), and pte mapped but not yet locked. | |
3073 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 3074 | */ |
82b0f8c3 | 3075 | static int do_anonymous_page(struct vm_fault *vmf) |
1da177e4 | 3076 | { |
82b0f8c3 | 3077 | struct vm_area_struct *vma = vmf->vma; |
00501b53 | 3078 | struct mem_cgroup *memcg; |
8f4e2101 | 3079 | struct page *page; |
6b31d595 | 3080 | int ret = 0; |
1da177e4 | 3081 | pte_t entry; |
1da177e4 | 3082 | |
6b7339f4 KS |
3083 | /* File mapping without ->vm_ops ? */ |
3084 | if (vma->vm_flags & VM_SHARED) | |
3085 | return VM_FAULT_SIGBUS; | |
3086 | ||
7267ec00 KS |
3087 | /* |
3088 | * Use pte_alloc() instead of pte_alloc_map(). We can't run | |
3089 | * pte_offset_map() on pmds where a huge pmd might be created | |
3090 | * from a different thread. | |
3091 | * | |
3092 | * pte_alloc_map() is safe to use under down_write(mmap_sem) or when | |
3093 | * parallel threads are excluded by other means. | |
3094 | * | |
3095 | * Here we only have down_read(mmap_sem). | |
3096 | */ | |
82b0f8c3 | 3097 | if (pte_alloc(vma->vm_mm, vmf->pmd, vmf->address)) |
7267ec00 KS |
3098 | return VM_FAULT_OOM; |
3099 | ||
3100 | /* See the comment in pte_alloc_one_map() */ | |
82b0f8c3 | 3101 | if (unlikely(pmd_trans_unstable(vmf->pmd))) |
7267ec00 KS |
3102 | return 0; |
3103 | ||
11ac5524 | 3104 | /* Use the zero-page for reads */ |
82b0f8c3 | 3105 | if (!(vmf->flags & FAULT_FLAG_WRITE) && |
bae473a4 | 3106 | !mm_forbids_zeropage(vma->vm_mm)) { |
82b0f8c3 | 3107 | entry = pte_mkspecial(pfn_pte(my_zero_pfn(vmf->address), |
62eede62 | 3108 | vma->vm_page_prot)); |
82b0f8c3 JK |
3109 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
3110 | vmf->address, &vmf->ptl); | |
3111 | if (!pte_none(*vmf->pte)) | |
a13ea5b7 | 3112 | goto unlock; |
6b31d595 MH |
3113 | ret = check_stable_address_space(vma->vm_mm); |
3114 | if (ret) | |
3115 | goto unlock; | |
6b251fc9 AA |
3116 | /* Deliver the page fault to userland, check inside PT lock */ |
3117 | if (userfaultfd_missing(vma)) { | |
82b0f8c3 JK |
3118 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
3119 | return handle_userfault(vmf, VM_UFFD_MISSING); | |
6b251fc9 | 3120 | } |
a13ea5b7 HD |
3121 | goto setpte; |
3122 | } | |
3123 | ||
557ed1fa | 3124 | /* Allocate our own private page. */ |
557ed1fa NP |
3125 | if (unlikely(anon_vma_prepare(vma))) |
3126 | goto oom; | |
82b0f8c3 | 3127 | page = alloc_zeroed_user_highpage_movable(vma, vmf->address); |
557ed1fa NP |
3128 | if (!page) |
3129 | goto oom; | |
eb3c24f3 | 3130 | |
bae473a4 | 3131 | if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL, &memcg, false)) |
eb3c24f3 MG |
3132 | goto oom_free_page; |
3133 | ||
52f37629 MK |
3134 | /* |
3135 | * The memory barrier inside __SetPageUptodate makes sure that | |
3136 | * preceeding stores to the page contents become visible before | |
3137 | * the set_pte_at() write. | |
3138 | */ | |
0ed361de | 3139 | __SetPageUptodate(page); |
8f4e2101 | 3140 | |
557ed1fa | 3141 | entry = mk_pte(page, vma->vm_page_prot); |
1ac0cb5d HD |
3142 | if (vma->vm_flags & VM_WRITE) |
3143 | entry = pte_mkwrite(pte_mkdirty(entry)); | |
1da177e4 | 3144 | |
82b0f8c3 JK |
3145 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, |
3146 | &vmf->ptl); | |
3147 | if (!pte_none(*vmf->pte)) | |
557ed1fa | 3148 | goto release; |
9ba69294 | 3149 | |
6b31d595 MH |
3150 | ret = check_stable_address_space(vma->vm_mm); |
3151 | if (ret) | |
3152 | goto release; | |
3153 | ||
6b251fc9 AA |
3154 | /* Deliver the page fault to userland, check inside PT lock */ |
3155 | if (userfaultfd_missing(vma)) { | |
82b0f8c3 | 3156 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
f627c2f5 | 3157 | mem_cgroup_cancel_charge(page, memcg, false); |
09cbfeaf | 3158 | put_page(page); |
82b0f8c3 | 3159 | return handle_userfault(vmf, VM_UFFD_MISSING); |
6b251fc9 AA |
3160 | } |
3161 | ||
bae473a4 | 3162 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
82b0f8c3 | 3163 | page_add_new_anon_rmap(page, vma, vmf->address, false); |
f627c2f5 | 3164 | mem_cgroup_commit_charge(page, memcg, false, false); |
00501b53 | 3165 | lru_cache_add_active_or_unevictable(page, vma); |
a13ea5b7 | 3166 | setpte: |
82b0f8c3 | 3167 | set_pte_at(vma->vm_mm, vmf->address, vmf->pte, entry); |
1da177e4 LT |
3168 | |
3169 | /* No need to invalidate - it was non-present before */ | |
82b0f8c3 | 3170 | update_mmu_cache(vma, vmf->address, vmf->pte); |
65500d23 | 3171 | unlock: |
82b0f8c3 | 3172 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
6b31d595 | 3173 | return ret; |
8f4e2101 | 3174 | release: |
f627c2f5 | 3175 | mem_cgroup_cancel_charge(page, memcg, false); |
09cbfeaf | 3176 | put_page(page); |
8f4e2101 | 3177 | goto unlock; |
8a9f3ccd | 3178 | oom_free_page: |
09cbfeaf | 3179 | put_page(page); |
65500d23 | 3180 | oom: |
1da177e4 LT |
3181 | return VM_FAULT_OOM; |
3182 | } | |
3183 | ||
9a95f3cf PC |
3184 | /* |
3185 | * The mmap_sem must have been held on entry, and may have been | |
3186 | * released depending on flags and vma->vm_ops->fault() return value. | |
3187 | * See filemap_fault() and __lock_page_retry(). | |
3188 | */ | |
936ca80d | 3189 | static int __do_fault(struct vm_fault *vmf) |
7eae74af | 3190 | { |
82b0f8c3 | 3191 | struct vm_area_struct *vma = vmf->vma; |
7eae74af KS |
3192 | int ret; |
3193 | ||
11bac800 | 3194 | ret = vma->vm_ops->fault(vmf); |
3917048d | 3195 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY | |
b1aa812b | 3196 | VM_FAULT_DONE_COW))) |
bc2466e4 | 3197 | return ret; |
7eae74af | 3198 | |
667240e0 | 3199 | if (unlikely(PageHWPoison(vmf->page))) { |
7eae74af | 3200 | if (ret & VM_FAULT_LOCKED) |
667240e0 JK |
3201 | unlock_page(vmf->page); |
3202 | put_page(vmf->page); | |
936ca80d | 3203 | vmf->page = NULL; |
7eae74af KS |
3204 | return VM_FAULT_HWPOISON; |
3205 | } | |
3206 | ||
3207 | if (unlikely(!(ret & VM_FAULT_LOCKED))) | |
667240e0 | 3208 | lock_page(vmf->page); |
7eae74af | 3209 | else |
667240e0 | 3210 | VM_BUG_ON_PAGE(!PageLocked(vmf->page), vmf->page); |
7eae74af | 3211 | |
7eae74af KS |
3212 | return ret; |
3213 | } | |
3214 | ||
d0f0931d RZ |
3215 | /* |
3216 | * The ordering of these checks is important for pmds with _PAGE_DEVMAP set. | |
3217 | * If we check pmd_trans_unstable() first we will trip the bad_pmd() check | |
3218 | * inside of pmd_none_or_trans_huge_or_clear_bad(). This will end up correctly | |
3219 | * returning 1 but not before it spams dmesg with the pmd_clear_bad() output. | |
3220 | */ | |
3221 | static int pmd_devmap_trans_unstable(pmd_t *pmd) | |
3222 | { | |
3223 | return pmd_devmap(*pmd) || pmd_trans_unstable(pmd); | |
3224 | } | |
3225 | ||
82b0f8c3 | 3226 | static int pte_alloc_one_map(struct vm_fault *vmf) |
7267ec00 | 3227 | { |
82b0f8c3 | 3228 | struct vm_area_struct *vma = vmf->vma; |
7267ec00 | 3229 | |
82b0f8c3 | 3230 | if (!pmd_none(*vmf->pmd)) |
7267ec00 | 3231 | goto map_pte; |
82b0f8c3 JK |
3232 | if (vmf->prealloc_pte) { |
3233 | vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd); | |
3234 | if (unlikely(!pmd_none(*vmf->pmd))) { | |
3235 | spin_unlock(vmf->ptl); | |
7267ec00 KS |
3236 | goto map_pte; |
3237 | } | |
3238 | ||
3239 | atomic_long_inc(&vma->vm_mm->nr_ptes); | |
82b0f8c3 JK |
3240 | pmd_populate(vma->vm_mm, vmf->pmd, vmf->prealloc_pte); |
3241 | spin_unlock(vmf->ptl); | |
7f2b6ce8 | 3242 | vmf->prealloc_pte = NULL; |
82b0f8c3 | 3243 | } else if (unlikely(pte_alloc(vma->vm_mm, vmf->pmd, vmf->address))) { |
7267ec00 KS |
3244 | return VM_FAULT_OOM; |
3245 | } | |
3246 | map_pte: | |
3247 | /* | |
3248 | * If a huge pmd materialized under us just retry later. Use | |
d0f0931d RZ |
3249 | * pmd_trans_unstable() via pmd_devmap_trans_unstable() instead of |
3250 | * pmd_trans_huge() to ensure the pmd didn't become pmd_trans_huge | |
3251 | * under us and then back to pmd_none, as a result of MADV_DONTNEED | |
3252 | * running immediately after a huge pmd fault in a different thread of | |
3253 | * this mm, in turn leading to a misleading pmd_trans_huge() retval. | |
3254 | * All we have to ensure is that it is a regular pmd that we can walk | |
3255 | * with pte_offset_map() and we can do that through an atomic read in | |
3256 | * C, which is what pmd_trans_unstable() provides. | |
7267ec00 | 3257 | */ |
d0f0931d | 3258 | if (pmd_devmap_trans_unstable(vmf->pmd)) |
7267ec00 KS |
3259 | return VM_FAULT_NOPAGE; |
3260 | ||
d0f0931d RZ |
3261 | /* |
3262 | * At this point we know that our vmf->pmd points to a page of ptes | |
3263 | * and it cannot become pmd_none(), pmd_devmap() or pmd_trans_huge() | |
3264 | * for the duration of the fault. If a racing MADV_DONTNEED runs and | |
3265 | * we zap the ptes pointed to by our vmf->pmd, the vmf->ptl will still | |
3266 | * be valid and we will re-check to make sure the vmf->pte isn't | |
3267 | * pte_none() under vmf->ptl protection when we return to | |
3268 | * alloc_set_pte(). | |
3269 | */ | |
82b0f8c3 JK |
3270 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, |
3271 | &vmf->ptl); | |
7267ec00 KS |
3272 | return 0; |
3273 | } | |
3274 | ||
e496cf3d | 3275 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
10102459 KS |
3276 | |
3277 | #define HPAGE_CACHE_INDEX_MASK (HPAGE_PMD_NR - 1) | |
3278 | static inline bool transhuge_vma_suitable(struct vm_area_struct *vma, | |
3279 | unsigned long haddr) | |
3280 | { | |
3281 | if (((vma->vm_start >> PAGE_SHIFT) & HPAGE_CACHE_INDEX_MASK) != | |
3282 | (vma->vm_pgoff & HPAGE_CACHE_INDEX_MASK)) | |
3283 | return false; | |
3284 | if (haddr < vma->vm_start || haddr + HPAGE_PMD_SIZE > vma->vm_end) | |
3285 | return false; | |
3286 | return true; | |
3287 | } | |
3288 | ||
82b0f8c3 | 3289 | static void deposit_prealloc_pte(struct vm_fault *vmf) |
953c66c2 | 3290 | { |
82b0f8c3 | 3291 | struct vm_area_struct *vma = vmf->vma; |
953c66c2 | 3292 | |
82b0f8c3 | 3293 | pgtable_trans_huge_deposit(vma->vm_mm, vmf->pmd, vmf->prealloc_pte); |
953c66c2 AK |
3294 | /* |
3295 | * We are going to consume the prealloc table, | |
3296 | * count that as nr_ptes. | |
3297 | */ | |
3298 | atomic_long_inc(&vma->vm_mm->nr_ptes); | |
7f2b6ce8 | 3299 | vmf->prealloc_pte = NULL; |
953c66c2 AK |
3300 | } |
3301 | ||
82b0f8c3 | 3302 | static int do_set_pmd(struct vm_fault *vmf, struct page *page) |
10102459 | 3303 | { |
82b0f8c3 JK |
3304 | struct vm_area_struct *vma = vmf->vma; |
3305 | bool write = vmf->flags & FAULT_FLAG_WRITE; | |
3306 | unsigned long haddr = vmf->address & HPAGE_PMD_MASK; | |
10102459 KS |
3307 | pmd_t entry; |
3308 | int i, ret; | |
3309 | ||
3310 | if (!transhuge_vma_suitable(vma, haddr)) | |
3311 | return VM_FAULT_FALLBACK; | |
3312 | ||
3313 | ret = VM_FAULT_FALLBACK; | |
3314 | page = compound_head(page); | |
3315 | ||
953c66c2 AK |
3316 | /* |
3317 | * Archs like ppc64 need additonal space to store information | |
3318 | * related to pte entry. Use the preallocated table for that. | |
3319 | */ | |
82b0f8c3 JK |
3320 | if (arch_needs_pgtable_deposit() && !vmf->prealloc_pte) { |
3321 | vmf->prealloc_pte = pte_alloc_one(vma->vm_mm, vmf->address); | |
3322 | if (!vmf->prealloc_pte) | |
953c66c2 AK |
3323 | return VM_FAULT_OOM; |
3324 | smp_wmb(); /* See comment in __pte_alloc() */ | |
3325 | } | |
3326 | ||
82b0f8c3 JK |
3327 | vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd); |
3328 | if (unlikely(!pmd_none(*vmf->pmd))) | |
10102459 KS |
3329 | goto out; |
3330 | ||
3331 | for (i = 0; i < HPAGE_PMD_NR; i++) | |
3332 | flush_icache_page(vma, page + i); | |
3333 | ||
3334 | entry = mk_huge_pmd(page, vma->vm_page_prot); | |
3335 | if (write) | |
3336 | entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma); | |
3337 | ||
3338 | add_mm_counter(vma->vm_mm, MM_FILEPAGES, HPAGE_PMD_NR); | |
3339 | page_add_file_rmap(page, true); | |
953c66c2 AK |
3340 | /* |
3341 | * deposit and withdraw with pmd lock held | |
3342 | */ | |
3343 | if (arch_needs_pgtable_deposit()) | |
82b0f8c3 | 3344 | deposit_prealloc_pte(vmf); |
10102459 | 3345 | |
82b0f8c3 | 3346 | set_pmd_at(vma->vm_mm, haddr, vmf->pmd, entry); |
10102459 | 3347 | |
82b0f8c3 | 3348 | update_mmu_cache_pmd(vma, haddr, vmf->pmd); |
10102459 KS |
3349 | |
3350 | /* fault is handled */ | |
3351 | ret = 0; | |
95ecedcd | 3352 | count_vm_event(THP_FILE_MAPPED); |
10102459 | 3353 | out: |
82b0f8c3 | 3354 | spin_unlock(vmf->ptl); |
10102459 KS |
3355 | return ret; |
3356 | } | |
3357 | #else | |
82b0f8c3 | 3358 | static int do_set_pmd(struct vm_fault *vmf, struct page *page) |
10102459 KS |
3359 | { |
3360 | BUILD_BUG(); | |
3361 | return 0; | |
3362 | } | |
3363 | #endif | |
3364 | ||
8c6e50b0 | 3365 | /** |
7267ec00 KS |
3366 | * alloc_set_pte - setup new PTE entry for given page and add reverse page |
3367 | * mapping. If needed, the fucntion allocates page table or use pre-allocated. | |
8c6e50b0 | 3368 | * |
82b0f8c3 | 3369 | * @vmf: fault environment |
7267ec00 | 3370 | * @memcg: memcg to charge page (only for private mappings) |
8c6e50b0 | 3371 | * @page: page to map |
8c6e50b0 | 3372 | * |
82b0f8c3 JK |
3373 | * Caller must take care of unlocking vmf->ptl, if vmf->pte is non-NULL on |
3374 | * return. | |
8c6e50b0 KS |
3375 | * |
3376 | * Target users are page handler itself and implementations of | |
3377 | * vm_ops->map_pages. | |
3378 | */ | |
82b0f8c3 | 3379 | int alloc_set_pte(struct vm_fault *vmf, struct mem_cgroup *memcg, |
7267ec00 | 3380 | struct page *page) |
3bb97794 | 3381 | { |
82b0f8c3 JK |
3382 | struct vm_area_struct *vma = vmf->vma; |
3383 | bool write = vmf->flags & FAULT_FLAG_WRITE; | |
3bb97794 | 3384 | pte_t entry; |
10102459 KS |
3385 | int ret; |
3386 | ||
82b0f8c3 | 3387 | if (pmd_none(*vmf->pmd) && PageTransCompound(page) && |
e496cf3d | 3388 | IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) { |
10102459 KS |
3389 | /* THP on COW? */ |
3390 | VM_BUG_ON_PAGE(memcg, page); | |
3391 | ||
82b0f8c3 | 3392 | ret = do_set_pmd(vmf, page); |
10102459 | 3393 | if (ret != VM_FAULT_FALLBACK) |
b0b9b3df | 3394 | return ret; |
10102459 | 3395 | } |
3bb97794 | 3396 | |
82b0f8c3 JK |
3397 | if (!vmf->pte) { |
3398 | ret = pte_alloc_one_map(vmf); | |
7267ec00 | 3399 | if (ret) |
b0b9b3df | 3400 | return ret; |
7267ec00 KS |
3401 | } |
3402 | ||
3403 | /* Re-check under ptl */ | |
b0b9b3df HD |
3404 | if (unlikely(!pte_none(*vmf->pte))) |
3405 | return VM_FAULT_NOPAGE; | |
7267ec00 | 3406 | |
3bb97794 KS |
3407 | flush_icache_page(vma, page); |
3408 | entry = mk_pte(page, vma->vm_page_prot); | |
3409 | if (write) | |
3410 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
bae473a4 KS |
3411 | /* copy-on-write page */ |
3412 | if (write && !(vma->vm_flags & VM_SHARED)) { | |
3bb97794 | 3413 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
82b0f8c3 | 3414 | page_add_new_anon_rmap(page, vma, vmf->address, false); |
7267ec00 KS |
3415 | mem_cgroup_commit_charge(page, memcg, false, false); |
3416 | lru_cache_add_active_or_unevictable(page, vma); | |
3bb97794 | 3417 | } else { |
eca56ff9 | 3418 | inc_mm_counter_fast(vma->vm_mm, mm_counter_file(page)); |
dd78fedd | 3419 | page_add_file_rmap(page, false); |
3bb97794 | 3420 | } |
82b0f8c3 | 3421 | set_pte_at(vma->vm_mm, vmf->address, vmf->pte, entry); |
3bb97794 KS |
3422 | |
3423 | /* no need to invalidate: a not-present page won't be cached */ | |
82b0f8c3 | 3424 | update_mmu_cache(vma, vmf->address, vmf->pte); |
7267ec00 | 3425 | |
b0b9b3df | 3426 | return 0; |
3bb97794 KS |
3427 | } |
3428 | ||
9118c0cb JK |
3429 | |
3430 | /** | |
3431 | * finish_fault - finish page fault once we have prepared the page to fault | |
3432 | * | |
3433 | * @vmf: structure describing the fault | |
3434 | * | |
3435 | * This function handles all that is needed to finish a page fault once the | |
3436 | * page to fault in is prepared. It handles locking of PTEs, inserts PTE for | |
3437 | * given page, adds reverse page mapping, handles memcg charges and LRU | |
3438 | * addition. The function returns 0 on success, VM_FAULT_ code in case of | |
3439 | * error. | |
3440 | * | |
3441 | * The function expects the page to be locked and on success it consumes a | |
3442 | * reference of a page being mapped (for the PTE which maps it). | |
3443 | */ | |
3444 | int finish_fault(struct vm_fault *vmf) | |
3445 | { | |
3446 | struct page *page; | |
6b31d595 | 3447 | int ret = 0; |
9118c0cb JK |
3448 | |
3449 | /* Did we COW the page? */ | |
3450 | if ((vmf->flags & FAULT_FLAG_WRITE) && | |
3451 | !(vmf->vma->vm_flags & VM_SHARED)) | |
3452 | page = vmf->cow_page; | |
3453 | else | |
3454 | page = vmf->page; | |
6b31d595 MH |
3455 | |
3456 | /* | |
3457 | * check even for read faults because we might have lost our CoWed | |
3458 | * page | |
3459 | */ | |
3460 | if (!(vmf->vma->vm_flags & VM_SHARED)) | |
3461 | ret = check_stable_address_space(vmf->vma->vm_mm); | |
3462 | if (!ret) | |
3463 | ret = alloc_set_pte(vmf, vmf->memcg, page); | |
9118c0cb JK |
3464 | if (vmf->pte) |
3465 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
3466 | return ret; | |
3467 | } | |
3468 | ||
3a91053a KS |
3469 | static unsigned long fault_around_bytes __read_mostly = |
3470 | rounddown_pow_of_two(65536); | |
a9b0f861 | 3471 | |
a9b0f861 KS |
3472 | #ifdef CONFIG_DEBUG_FS |
3473 | static int fault_around_bytes_get(void *data, u64 *val) | |
1592eef0 | 3474 | { |
a9b0f861 | 3475 | *val = fault_around_bytes; |
1592eef0 KS |
3476 | return 0; |
3477 | } | |
3478 | ||
b4903d6e AR |
3479 | /* |
3480 | * fault_around_pages() and fault_around_mask() expects fault_around_bytes | |
3481 | * rounded down to nearest page order. It's what do_fault_around() expects to | |
3482 | * see. | |
3483 | */ | |
a9b0f861 | 3484 | static int fault_around_bytes_set(void *data, u64 val) |
1592eef0 | 3485 | { |
a9b0f861 | 3486 | if (val / PAGE_SIZE > PTRS_PER_PTE) |
1592eef0 | 3487 | return -EINVAL; |
b4903d6e AR |
3488 | if (val > PAGE_SIZE) |
3489 | fault_around_bytes = rounddown_pow_of_two(val); | |
3490 | else | |
3491 | fault_around_bytes = PAGE_SIZE; /* rounddown_pow_of_two(0) is undefined */ | |
1592eef0 KS |
3492 | return 0; |
3493 | } | |
0a1345f8 | 3494 | DEFINE_DEBUGFS_ATTRIBUTE(fault_around_bytes_fops, |
a9b0f861 | 3495 | fault_around_bytes_get, fault_around_bytes_set, "%llu\n"); |
1592eef0 KS |
3496 | |
3497 | static int __init fault_around_debugfs(void) | |
3498 | { | |
3499 | void *ret; | |
3500 | ||
0a1345f8 | 3501 | ret = debugfs_create_file_unsafe("fault_around_bytes", 0644, NULL, NULL, |
a9b0f861 | 3502 | &fault_around_bytes_fops); |
1592eef0 | 3503 | if (!ret) |
a9b0f861 | 3504 | pr_warn("Failed to create fault_around_bytes in debugfs"); |
1592eef0 KS |
3505 | return 0; |
3506 | } | |
3507 | late_initcall(fault_around_debugfs); | |
1592eef0 | 3508 | #endif |
8c6e50b0 | 3509 | |
1fdb412b KS |
3510 | /* |
3511 | * do_fault_around() tries to map few pages around the fault address. The hope | |
3512 | * is that the pages will be needed soon and this will lower the number of | |
3513 | * faults to handle. | |
3514 | * | |
3515 | * It uses vm_ops->map_pages() to map the pages, which skips the page if it's | |
3516 | * not ready to be mapped: not up-to-date, locked, etc. | |
3517 | * | |
3518 | * This function is called with the page table lock taken. In the split ptlock | |
3519 | * case the page table lock only protects only those entries which belong to | |
3520 | * the page table corresponding to the fault address. | |
3521 | * | |
3522 | * This function doesn't cross the VMA boundaries, in order to call map_pages() | |
3523 | * only once. | |
3524 | * | |
3525 | * fault_around_pages() defines how many pages we'll try to map. | |
3526 | * do_fault_around() expects it to return a power of two less than or equal to | |
3527 | * PTRS_PER_PTE. | |
3528 | * | |
3529 | * The virtual address of the area that we map is naturally aligned to the | |
3530 | * fault_around_pages() value (and therefore to page order). This way it's | |
3531 | * easier to guarantee that we don't cross page table boundaries. | |
3532 | */ | |
0721ec8b | 3533 | static int do_fault_around(struct vm_fault *vmf) |
8c6e50b0 | 3534 | { |
82b0f8c3 | 3535 | unsigned long address = vmf->address, nr_pages, mask; |
0721ec8b | 3536 | pgoff_t start_pgoff = vmf->pgoff; |
bae473a4 | 3537 | pgoff_t end_pgoff; |
7267ec00 | 3538 | int off, ret = 0; |
8c6e50b0 | 3539 | |
4db0c3c2 | 3540 | nr_pages = READ_ONCE(fault_around_bytes) >> PAGE_SHIFT; |
aecd6f44 KS |
3541 | mask = ~(nr_pages * PAGE_SIZE - 1) & PAGE_MASK; |
3542 | ||
82b0f8c3 JK |
3543 | vmf->address = max(address & mask, vmf->vma->vm_start); |
3544 | off = ((address - vmf->address) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1); | |
bae473a4 | 3545 | start_pgoff -= off; |
8c6e50b0 KS |
3546 | |
3547 | /* | |
bae473a4 KS |
3548 | * end_pgoff is either end of page table or end of vma |
3549 | * or fault_around_pages() from start_pgoff, depending what is nearest. | |
8c6e50b0 | 3550 | */ |
bae473a4 | 3551 | end_pgoff = start_pgoff - |
82b0f8c3 | 3552 | ((vmf->address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)) + |
8c6e50b0 | 3553 | PTRS_PER_PTE - 1; |
82b0f8c3 | 3554 | end_pgoff = min3(end_pgoff, vma_pages(vmf->vma) + vmf->vma->vm_pgoff - 1, |
bae473a4 | 3555 | start_pgoff + nr_pages - 1); |
8c6e50b0 | 3556 | |
82b0f8c3 JK |
3557 | if (pmd_none(*vmf->pmd)) { |
3558 | vmf->prealloc_pte = pte_alloc_one(vmf->vma->vm_mm, | |
3559 | vmf->address); | |
3560 | if (!vmf->prealloc_pte) | |
c5f88bd2 | 3561 | goto out; |
7267ec00 | 3562 | smp_wmb(); /* See comment in __pte_alloc() */ |
8c6e50b0 KS |
3563 | } |
3564 | ||
82b0f8c3 | 3565 | vmf->vma->vm_ops->map_pages(vmf, start_pgoff, end_pgoff); |
7267ec00 | 3566 | |
7267ec00 | 3567 | /* Huge page is mapped? Page fault is solved */ |
82b0f8c3 | 3568 | if (pmd_trans_huge(*vmf->pmd)) { |
7267ec00 KS |
3569 | ret = VM_FAULT_NOPAGE; |
3570 | goto out; | |
3571 | } | |
3572 | ||
3573 | /* ->map_pages() haven't done anything useful. Cold page cache? */ | |
82b0f8c3 | 3574 | if (!vmf->pte) |
7267ec00 KS |
3575 | goto out; |
3576 | ||
3577 | /* check if the page fault is solved */ | |
82b0f8c3 JK |
3578 | vmf->pte -= (vmf->address >> PAGE_SHIFT) - (address >> PAGE_SHIFT); |
3579 | if (!pte_none(*vmf->pte)) | |
7267ec00 | 3580 | ret = VM_FAULT_NOPAGE; |
82b0f8c3 | 3581 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
bae473a4 | 3582 | out: |
82b0f8c3 JK |
3583 | vmf->address = address; |
3584 | vmf->pte = NULL; | |
7267ec00 | 3585 | return ret; |
8c6e50b0 KS |
3586 | } |
3587 | ||
0721ec8b | 3588 | static int do_read_fault(struct vm_fault *vmf) |
e655fb29 | 3589 | { |
82b0f8c3 | 3590 | struct vm_area_struct *vma = vmf->vma; |
8c6e50b0 KS |
3591 | int ret = 0; |
3592 | ||
3593 | /* | |
3594 | * Let's call ->map_pages() first and use ->fault() as fallback | |
3595 | * if page by the offset is not ready to be mapped (cold cache or | |
3596 | * something). | |
3597 | */ | |
9b4bdd2f | 3598 | if (vma->vm_ops->map_pages && fault_around_bytes >> PAGE_SHIFT > 1) { |
0721ec8b | 3599 | ret = do_fault_around(vmf); |
7267ec00 KS |
3600 | if (ret) |
3601 | return ret; | |
8c6e50b0 | 3602 | } |
e655fb29 | 3603 | |
936ca80d | 3604 | ret = __do_fault(vmf); |
e655fb29 KS |
3605 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
3606 | return ret; | |
3607 | ||
9118c0cb | 3608 | ret |= finish_fault(vmf); |
936ca80d | 3609 | unlock_page(vmf->page); |
7267ec00 | 3610 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
936ca80d | 3611 | put_page(vmf->page); |
e655fb29 KS |
3612 | return ret; |
3613 | } | |
3614 | ||
0721ec8b | 3615 | static int do_cow_fault(struct vm_fault *vmf) |
ec47c3b9 | 3616 | { |
82b0f8c3 | 3617 | struct vm_area_struct *vma = vmf->vma; |
ec47c3b9 KS |
3618 | int ret; |
3619 | ||
3620 | if (unlikely(anon_vma_prepare(vma))) | |
3621 | return VM_FAULT_OOM; | |
3622 | ||
936ca80d JK |
3623 | vmf->cow_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vmf->address); |
3624 | if (!vmf->cow_page) | |
ec47c3b9 KS |
3625 | return VM_FAULT_OOM; |
3626 | ||
936ca80d | 3627 | if (mem_cgroup_try_charge(vmf->cow_page, vma->vm_mm, GFP_KERNEL, |
3917048d | 3628 | &vmf->memcg, false)) { |
936ca80d | 3629 | put_page(vmf->cow_page); |
ec47c3b9 KS |
3630 | return VM_FAULT_OOM; |
3631 | } | |
3632 | ||
936ca80d | 3633 | ret = __do_fault(vmf); |
ec47c3b9 KS |
3634 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
3635 | goto uncharge_out; | |
3917048d JK |
3636 | if (ret & VM_FAULT_DONE_COW) |
3637 | return ret; | |
ec47c3b9 | 3638 | |
b1aa812b | 3639 | copy_user_highpage(vmf->cow_page, vmf->page, vmf->address, vma); |
936ca80d | 3640 | __SetPageUptodate(vmf->cow_page); |
ec47c3b9 | 3641 | |
9118c0cb | 3642 | ret |= finish_fault(vmf); |
b1aa812b JK |
3643 | unlock_page(vmf->page); |
3644 | put_page(vmf->page); | |
7267ec00 KS |
3645 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
3646 | goto uncharge_out; | |
ec47c3b9 KS |
3647 | return ret; |
3648 | uncharge_out: | |
3917048d | 3649 | mem_cgroup_cancel_charge(vmf->cow_page, vmf->memcg, false); |
936ca80d | 3650 | put_page(vmf->cow_page); |
ec47c3b9 KS |
3651 | return ret; |
3652 | } | |
3653 | ||
0721ec8b | 3654 | static int do_shared_fault(struct vm_fault *vmf) |
1da177e4 | 3655 | { |
82b0f8c3 | 3656 | struct vm_area_struct *vma = vmf->vma; |
f0c6d4d2 | 3657 | int ret, tmp; |
1d65f86d | 3658 | |
936ca80d | 3659 | ret = __do_fault(vmf); |
7eae74af | 3660 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
f0c6d4d2 | 3661 | return ret; |
1da177e4 LT |
3662 | |
3663 | /* | |
f0c6d4d2 KS |
3664 | * Check if the backing address space wants to know that the page is |
3665 | * about to become writable | |
1da177e4 | 3666 | */ |
fb09a464 | 3667 | if (vma->vm_ops->page_mkwrite) { |
936ca80d | 3668 | unlock_page(vmf->page); |
38b8cb7f | 3669 | tmp = do_page_mkwrite(vmf); |
fb09a464 KS |
3670 | if (unlikely(!tmp || |
3671 | (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { | |
936ca80d | 3672 | put_page(vmf->page); |
fb09a464 | 3673 | return tmp; |
4294621f | 3674 | } |
fb09a464 KS |
3675 | } |
3676 | ||
9118c0cb | 3677 | ret |= finish_fault(vmf); |
7267ec00 KS |
3678 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | |
3679 | VM_FAULT_RETRY))) { | |
936ca80d JK |
3680 | unlock_page(vmf->page); |
3681 | put_page(vmf->page); | |
f0c6d4d2 | 3682 | return ret; |
1da177e4 | 3683 | } |
b827e496 | 3684 | |
97ba0c2b | 3685 | fault_dirty_shared_page(vma, vmf->page); |
1d65f86d | 3686 | return ret; |
54cb8821 | 3687 | } |
d00806b1 | 3688 | |
9a95f3cf PC |
3689 | /* |
3690 | * We enter with non-exclusive mmap_sem (to exclude vma changes, | |
3691 | * but allow concurrent faults). | |
3692 | * The mmap_sem may have been released depending on flags and our | |
3693 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
3694 | */ | |
82b0f8c3 | 3695 | static int do_fault(struct vm_fault *vmf) |
54cb8821 | 3696 | { |
82b0f8c3 | 3697 | struct vm_area_struct *vma = vmf->vma; |
b0b9b3df | 3698 | int ret; |
54cb8821 | 3699 | |
6b43a997 AK |
3700 | /* |
3701 | * The VMA was not fully populated on mmap() or missing VM_DONTEXPAND | |
3702 | */ | |
3703 | if (!vma->vm_ops->fault) { | |
3704 | /* | |
3705 | * If we find a migration pmd entry or a none pmd entry, which | |
3706 | * should never happen, return SIGBUS | |
3707 | */ | |
3708 | if (unlikely(!pmd_present(*vmf->pmd))) | |
3709 | ret = VM_FAULT_SIGBUS; | |
3710 | else { | |
3711 | vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, | |
3712 | vmf->pmd, | |
3713 | vmf->address, | |
3714 | &vmf->ptl); | |
3715 | /* | |
3716 | * Make sure this is not a temporary clearing of pte | |
3717 | * by holding ptl and checking again. A R/M/W update | |
3718 | * of pte involves: take ptl, clearing the pte so that | |
3719 | * we don't have concurrent modification by hardware | |
3720 | * followed by an update. | |
3721 | */ | |
3722 | if (unlikely(pte_none(*vmf->pte))) | |
3723 | ret = VM_FAULT_SIGBUS; | |
3724 | else | |
3725 | ret = VM_FAULT_NOPAGE; | |
3726 | ||
3727 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
3728 | } | |
3729 | } else if (!(vmf->flags & FAULT_FLAG_WRITE)) | |
b0b9b3df HD |
3730 | ret = do_read_fault(vmf); |
3731 | else if (!(vma->vm_flags & VM_SHARED)) | |
3732 | ret = do_cow_fault(vmf); | |
3733 | else | |
3734 | ret = do_shared_fault(vmf); | |
3735 | ||
3736 | /* preallocated pagetable is unused: free it */ | |
3737 | if (vmf->prealloc_pte) { | |
3738 | pte_free(vma->vm_mm, vmf->prealloc_pte); | |
7f2b6ce8 | 3739 | vmf->prealloc_pte = NULL; |
b0b9b3df HD |
3740 | } |
3741 | return ret; | |
54cb8821 NP |
3742 | } |
3743 | ||
b19a9939 | 3744 | static int numa_migrate_prep(struct page *page, struct vm_area_struct *vma, |
04bb2f94 RR |
3745 | unsigned long addr, int page_nid, |
3746 | int *flags) | |
9532fec1 MG |
3747 | { |
3748 | get_page(page); | |
3749 | ||
3750 | count_vm_numa_event(NUMA_HINT_FAULTS); | |
04bb2f94 | 3751 | if (page_nid == numa_node_id()) { |
9532fec1 | 3752 | count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL); |
04bb2f94 RR |
3753 | *flags |= TNF_FAULT_LOCAL; |
3754 | } | |
9532fec1 MG |
3755 | |
3756 | return mpol_misplaced(page, vma, addr); | |
3757 | } | |
3758 | ||
2994302b | 3759 | static int do_numa_page(struct vm_fault *vmf) |
d10e63f2 | 3760 | { |
82b0f8c3 | 3761 | struct vm_area_struct *vma = vmf->vma; |
4daae3b4 | 3762 | struct page *page = NULL; |
8191acbd | 3763 | int page_nid = -1; |
90572890 | 3764 | int last_cpupid; |
cbee9f88 | 3765 | int target_nid; |
b8593bfd | 3766 | bool migrated = false; |
cee216a6 | 3767 | pte_t pte; |
288bc549 | 3768 | bool was_writable = pte_savedwrite(vmf->orig_pte); |
6688cc05 | 3769 | int flags = 0; |
d10e63f2 MG |
3770 | |
3771 | /* | |
166f61b9 TH |
3772 | * The "pte" at this point cannot be used safely without |
3773 | * validation through pte_unmap_same(). It's of NUMA type but | |
3774 | * the pfn may be screwed if the read is non atomic. | |
166f61b9 | 3775 | */ |
82b0f8c3 JK |
3776 | vmf->ptl = pte_lockptr(vma->vm_mm, vmf->pmd); |
3777 | spin_lock(vmf->ptl); | |
cee216a6 | 3778 | if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) { |
82b0f8c3 | 3779 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
4daae3b4 MG |
3780 | goto out; |
3781 | } | |
3782 | ||
cee216a6 AK |
3783 | /* |
3784 | * Make it present again, Depending on how arch implementes non | |
3785 | * accessible ptes, some can allow access by kernel mode. | |
3786 | */ | |
3787 | pte = ptep_modify_prot_start(vma->vm_mm, vmf->address, vmf->pte); | |
4d942466 MG |
3788 | pte = pte_modify(pte, vma->vm_page_prot); |
3789 | pte = pte_mkyoung(pte); | |
b191f9b1 MG |
3790 | if (was_writable) |
3791 | pte = pte_mkwrite(pte); | |
cee216a6 | 3792 | ptep_modify_prot_commit(vma->vm_mm, vmf->address, vmf->pte, pte); |
82b0f8c3 | 3793 | update_mmu_cache(vma, vmf->address, vmf->pte); |
d10e63f2 | 3794 | |
82b0f8c3 | 3795 | page = vm_normal_page(vma, vmf->address, pte); |
d10e63f2 | 3796 | if (!page) { |
82b0f8c3 | 3797 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
d10e63f2 MG |
3798 | return 0; |
3799 | } | |
3800 | ||
e81c4802 KS |
3801 | /* TODO: handle PTE-mapped THP */ |
3802 | if (PageCompound(page)) { | |
82b0f8c3 | 3803 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
e81c4802 KS |
3804 | return 0; |
3805 | } | |
3806 | ||
6688cc05 | 3807 | /* |
bea66fbd MG |
3808 | * Avoid grouping on RO pages in general. RO pages shouldn't hurt as |
3809 | * much anyway since they can be in shared cache state. This misses | |
3810 | * the case where a mapping is writable but the process never writes | |
3811 | * to it but pte_write gets cleared during protection updates and | |
3812 | * pte_dirty has unpredictable behaviour between PTE scan updates, | |
3813 | * background writeback, dirty balancing and application behaviour. | |
6688cc05 | 3814 | */ |
d59dc7bc | 3815 | if (!pte_write(pte)) |
6688cc05 PZ |
3816 | flags |= TNF_NO_GROUP; |
3817 | ||
dabe1d99 RR |
3818 | /* |
3819 | * Flag if the page is shared between multiple address spaces. This | |
3820 | * is later used when determining whether to group tasks together | |
3821 | */ | |
3822 | if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED)) | |
3823 | flags |= TNF_SHARED; | |
3824 | ||
90572890 | 3825 | last_cpupid = page_cpupid_last(page); |
8191acbd | 3826 | page_nid = page_to_nid(page); |
82b0f8c3 | 3827 | target_nid = numa_migrate_prep(page, vma, vmf->address, page_nid, |
bae473a4 | 3828 | &flags); |
82b0f8c3 | 3829 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
4daae3b4 | 3830 | if (target_nid == -1) { |
4daae3b4 MG |
3831 | put_page(page); |
3832 | goto out; | |
3833 | } | |
3834 | ||
3835 | /* Migrate to the requested node */ | |
1bc115d8 | 3836 | migrated = migrate_misplaced_page(page, vma, target_nid); |
6688cc05 | 3837 | if (migrated) { |
8191acbd | 3838 | page_nid = target_nid; |
6688cc05 | 3839 | flags |= TNF_MIGRATED; |
074c2381 MG |
3840 | } else |
3841 | flags |= TNF_MIGRATE_FAIL; | |
4daae3b4 MG |
3842 | |
3843 | out: | |
8191acbd | 3844 | if (page_nid != -1) |
6688cc05 | 3845 | task_numa_fault(last_cpupid, page_nid, 1, flags); |
d10e63f2 MG |
3846 | return 0; |
3847 | } | |
3848 | ||
91a90140 | 3849 | static inline int create_huge_pmd(struct vm_fault *vmf) |
b96375f7 | 3850 | { |
f4200391 | 3851 | if (vma_is_anonymous(vmf->vma)) |
82b0f8c3 | 3852 | return do_huge_pmd_anonymous_page(vmf); |
a2d58167 | 3853 | if (vmf->vma->vm_ops->huge_fault) |
c791ace1 | 3854 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD); |
b96375f7 MW |
3855 | return VM_FAULT_FALLBACK; |
3856 | } | |
3857 | ||
82b0f8c3 | 3858 | static int wp_huge_pmd(struct vm_fault *vmf, pmd_t orig_pmd) |
b96375f7 | 3859 | { |
82b0f8c3 JK |
3860 | if (vma_is_anonymous(vmf->vma)) |
3861 | return do_huge_pmd_wp_page(vmf, orig_pmd); | |
a2d58167 | 3862 | if (vmf->vma->vm_ops->huge_fault) |
c791ace1 | 3863 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD); |
af9e4d5f KS |
3864 | |
3865 | /* COW handled on pte level: split pmd */ | |
82b0f8c3 JK |
3866 | VM_BUG_ON_VMA(vmf->vma->vm_flags & VM_SHARED, vmf->vma); |
3867 | __split_huge_pmd(vmf->vma, vmf->pmd, vmf->address, false, NULL); | |
af9e4d5f | 3868 | |
b96375f7 MW |
3869 | return VM_FAULT_FALLBACK; |
3870 | } | |
3871 | ||
38e08854 LS |
3872 | static inline bool vma_is_accessible(struct vm_area_struct *vma) |
3873 | { | |
3874 | return vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE); | |
3875 | } | |
3876 | ||
a00cc7d9 MW |
3877 | static int create_huge_pud(struct vm_fault *vmf) |
3878 | { | |
3879 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
3880 | /* No support for anonymous transparent PUD pages yet */ | |
3881 | if (vma_is_anonymous(vmf->vma)) | |
3882 | return VM_FAULT_FALLBACK; | |
3883 | if (vmf->vma->vm_ops->huge_fault) | |
c791ace1 | 3884 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD); |
a00cc7d9 MW |
3885 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
3886 | return VM_FAULT_FALLBACK; | |
3887 | } | |
3888 | ||
3889 | static int wp_huge_pud(struct vm_fault *vmf, pud_t orig_pud) | |
3890 | { | |
3891 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
3892 | /* No support for anonymous transparent PUD pages yet */ | |
3893 | if (vma_is_anonymous(vmf->vma)) | |
3894 | return VM_FAULT_FALLBACK; | |
3895 | if (vmf->vma->vm_ops->huge_fault) | |
c791ace1 | 3896 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD); |
a00cc7d9 MW |
3897 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
3898 | return VM_FAULT_FALLBACK; | |
3899 | } | |
3900 | ||
1da177e4 LT |
3901 | /* |
3902 | * These routines also need to handle stuff like marking pages dirty | |
3903 | * and/or accessed for architectures that don't do it in hardware (most | |
3904 | * RISC architectures). The early dirtying is also good on the i386. | |
3905 | * | |
3906 | * There is also a hook called "update_mmu_cache()" that architectures | |
3907 | * with external mmu caches can use to update those (ie the Sparc or | |
3908 | * PowerPC hashed page tables that act as extended TLBs). | |
3909 | * | |
7267ec00 KS |
3910 | * We enter with non-exclusive mmap_sem (to exclude vma changes, but allow |
3911 | * concurrent faults). | |
9a95f3cf | 3912 | * |
7267ec00 KS |
3913 | * The mmap_sem may have been released depending on flags and our return value. |
3914 | * See filemap_fault() and __lock_page_or_retry(). | |
1da177e4 | 3915 | */ |
82b0f8c3 | 3916 | static int handle_pte_fault(struct vm_fault *vmf) |
1da177e4 LT |
3917 | { |
3918 | pte_t entry; | |
3919 | ||
82b0f8c3 | 3920 | if (unlikely(pmd_none(*vmf->pmd))) { |
7267ec00 KS |
3921 | /* |
3922 | * Leave __pte_alloc() until later: because vm_ops->fault may | |
3923 | * want to allocate huge page, and if we expose page table | |
3924 | * for an instant, it will be difficult to retract from | |
3925 | * concurrent faults and from rmap lookups. | |
3926 | */ | |
82b0f8c3 | 3927 | vmf->pte = NULL; |
7267ec00 KS |
3928 | } else { |
3929 | /* See comment in pte_alloc_one_map() */ | |
d0f0931d | 3930 | if (pmd_devmap_trans_unstable(vmf->pmd)) |
7267ec00 KS |
3931 | return 0; |
3932 | /* | |
3933 | * A regular pmd is established and it can't morph into a huge | |
3934 | * pmd from under us anymore at this point because we hold the | |
3935 | * mmap_sem read mode and khugepaged takes it in write mode. | |
3936 | * So now it's safe to run pte_offset_map(). | |
3937 | */ | |
82b0f8c3 | 3938 | vmf->pte = pte_offset_map(vmf->pmd, vmf->address); |
2994302b | 3939 | vmf->orig_pte = *vmf->pte; |
7267ec00 KS |
3940 | |
3941 | /* | |
3942 | * some architectures can have larger ptes than wordsize, | |
3943 | * e.g.ppc44x-defconfig has CONFIG_PTE_64BIT=y and | |
3944 | * CONFIG_32BIT=y, so READ_ONCE or ACCESS_ONCE cannot guarantee | |
3945 | * atomic accesses. The code below just needs a consistent | |
3946 | * view for the ifs and we later double check anyway with the | |
3947 | * ptl lock held. So here a barrier will do. | |
3948 | */ | |
3949 | barrier(); | |
2994302b | 3950 | if (pte_none(vmf->orig_pte)) { |
82b0f8c3 JK |
3951 | pte_unmap(vmf->pte); |
3952 | vmf->pte = NULL; | |
65500d23 | 3953 | } |
1da177e4 LT |
3954 | } |
3955 | ||
82b0f8c3 JK |
3956 | if (!vmf->pte) { |
3957 | if (vma_is_anonymous(vmf->vma)) | |
3958 | return do_anonymous_page(vmf); | |
7267ec00 | 3959 | else |
82b0f8c3 | 3960 | return do_fault(vmf); |
7267ec00 KS |
3961 | } |
3962 | ||
2994302b JK |
3963 | if (!pte_present(vmf->orig_pte)) |
3964 | return do_swap_page(vmf); | |
7267ec00 | 3965 | |
2994302b JK |
3966 | if (pte_protnone(vmf->orig_pte) && vma_is_accessible(vmf->vma)) |
3967 | return do_numa_page(vmf); | |
d10e63f2 | 3968 | |
82b0f8c3 JK |
3969 | vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd); |
3970 | spin_lock(vmf->ptl); | |
2994302b | 3971 | entry = vmf->orig_pte; |
82b0f8c3 | 3972 | if (unlikely(!pte_same(*vmf->pte, entry))) |
8f4e2101 | 3973 | goto unlock; |
82b0f8c3 | 3974 | if (vmf->flags & FAULT_FLAG_WRITE) { |
1da177e4 | 3975 | if (!pte_write(entry)) |
2994302b | 3976 | return do_wp_page(vmf); |
1da177e4 LT |
3977 | entry = pte_mkdirty(entry); |
3978 | } | |
3979 | entry = pte_mkyoung(entry); | |
82b0f8c3 JK |
3980 | if (ptep_set_access_flags(vmf->vma, vmf->address, vmf->pte, entry, |
3981 | vmf->flags & FAULT_FLAG_WRITE)) { | |
3982 | update_mmu_cache(vmf->vma, vmf->address, vmf->pte); | |
1a44e149 AA |
3983 | } else { |
3984 | /* | |
3985 | * This is needed only for protection faults but the arch code | |
3986 | * is not yet telling us if this is a protection fault or not. | |
3987 | * This still avoids useless tlb flushes for .text page faults | |
3988 | * with threads. | |
3989 | */ | |
82b0f8c3 JK |
3990 | if (vmf->flags & FAULT_FLAG_WRITE) |
3991 | flush_tlb_fix_spurious_fault(vmf->vma, vmf->address); | |
1a44e149 | 3992 | } |
8f4e2101 | 3993 | unlock: |
82b0f8c3 | 3994 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
83c54070 | 3995 | return 0; |
1da177e4 LT |
3996 | } |
3997 | ||
3998 | /* | |
3999 | * By the time we get here, we already hold the mm semaphore | |
9a95f3cf PC |
4000 | * |
4001 | * The mmap_sem may have been released depending on flags and our | |
4002 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
1da177e4 | 4003 | */ |
dcddffd4 KS |
4004 | static int __handle_mm_fault(struct vm_area_struct *vma, unsigned long address, |
4005 | unsigned int flags) | |
1da177e4 | 4006 | { |
82b0f8c3 | 4007 | struct vm_fault vmf = { |
bae473a4 | 4008 | .vma = vma, |
1a29d85e | 4009 | .address = address & PAGE_MASK, |
bae473a4 | 4010 | .flags = flags, |
0721ec8b | 4011 | .pgoff = linear_page_index(vma, address), |
667240e0 | 4012 | .gfp_mask = __get_fault_gfp_mask(vma), |
bae473a4 | 4013 | }; |
fde26bed | 4014 | unsigned int dirty = flags & FAULT_FLAG_WRITE; |
dcddffd4 | 4015 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 4016 | pgd_t *pgd; |
c2febafc | 4017 | p4d_t *p4d; |
a2d58167 | 4018 | int ret; |
1da177e4 | 4019 | |
1da177e4 | 4020 | pgd = pgd_offset(mm, address); |
c2febafc KS |
4021 | p4d = p4d_alloc(mm, pgd, address); |
4022 | if (!p4d) | |
4023 | return VM_FAULT_OOM; | |
a00cc7d9 | 4024 | |
c2febafc | 4025 | vmf.pud = pud_alloc(mm, p4d, address); |
a00cc7d9 | 4026 | if (!vmf.pud) |
c74df32c | 4027 | return VM_FAULT_OOM; |
a00cc7d9 | 4028 | if (pud_none(*vmf.pud) && transparent_hugepage_enabled(vma)) { |
a00cc7d9 MW |
4029 | ret = create_huge_pud(&vmf); |
4030 | if (!(ret & VM_FAULT_FALLBACK)) | |
4031 | return ret; | |
4032 | } else { | |
4033 | pud_t orig_pud = *vmf.pud; | |
4034 | ||
4035 | barrier(); | |
4036 | if (pud_trans_huge(orig_pud) || pud_devmap(orig_pud)) { | |
a00cc7d9 | 4037 | |
a00cc7d9 MW |
4038 | /* NUMA case for anonymous PUDs would go here */ |
4039 | ||
4040 | if (dirty && !pud_write(orig_pud)) { | |
4041 | ret = wp_huge_pud(&vmf, orig_pud); | |
4042 | if (!(ret & VM_FAULT_FALLBACK)) | |
4043 | return ret; | |
4044 | } else { | |
4045 | huge_pud_set_accessed(&vmf, orig_pud); | |
4046 | return 0; | |
4047 | } | |
4048 | } | |
4049 | } | |
4050 | ||
4051 | vmf.pmd = pmd_alloc(mm, vmf.pud, address); | |
82b0f8c3 | 4052 | if (!vmf.pmd) |
c74df32c | 4053 | return VM_FAULT_OOM; |
82b0f8c3 | 4054 | if (pmd_none(*vmf.pmd) && transparent_hugepage_enabled(vma)) { |
a2d58167 | 4055 | ret = create_huge_pmd(&vmf); |
c0292554 KS |
4056 | if (!(ret & VM_FAULT_FALLBACK)) |
4057 | return ret; | |
71e3aac0 | 4058 | } else { |
82b0f8c3 | 4059 | pmd_t orig_pmd = *vmf.pmd; |
1f1d06c3 | 4060 | |
71e3aac0 | 4061 | barrier(); |
84c3fc4e ZY |
4062 | if (unlikely(is_swap_pmd(orig_pmd))) { |
4063 | VM_BUG_ON(thp_migration_supported() && | |
4064 | !is_pmd_migration_entry(orig_pmd)); | |
4065 | if (is_pmd_migration_entry(orig_pmd)) | |
4066 | pmd_migration_entry_wait(mm, vmf.pmd); | |
4067 | return 0; | |
4068 | } | |
5c7fb56e | 4069 | if (pmd_trans_huge(orig_pmd) || pmd_devmap(orig_pmd)) { |
38e08854 | 4070 | if (pmd_protnone(orig_pmd) && vma_is_accessible(vma)) |
82b0f8c3 | 4071 | return do_huge_pmd_numa_page(&vmf, orig_pmd); |
d10e63f2 | 4072 | |
fde26bed | 4073 | if (dirty && !pmd_write(orig_pmd)) { |
82b0f8c3 | 4074 | ret = wp_huge_pmd(&vmf, orig_pmd); |
9845cbbd KS |
4075 | if (!(ret & VM_FAULT_FALLBACK)) |
4076 | return ret; | |
a1dd450b | 4077 | } else { |
82b0f8c3 | 4078 | huge_pmd_set_accessed(&vmf, orig_pmd); |
9845cbbd | 4079 | return 0; |
1f1d06c3 | 4080 | } |
71e3aac0 AA |
4081 | } |
4082 | } | |
4083 | ||
82b0f8c3 | 4084 | return handle_pte_fault(&vmf); |
1da177e4 LT |
4085 | } |
4086 | ||
9a95f3cf PC |
4087 | /* |
4088 | * By the time we get here, we already hold the mm semaphore | |
4089 | * | |
4090 | * The mmap_sem may have been released depending on flags and our | |
4091 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
4092 | */ | |
dcddffd4 KS |
4093 | int handle_mm_fault(struct vm_area_struct *vma, unsigned long address, |
4094 | unsigned int flags) | |
519e5247 JW |
4095 | { |
4096 | int ret; | |
4097 | ||
4098 | __set_current_state(TASK_RUNNING); | |
4099 | ||
4100 | count_vm_event(PGFAULT); | |
2262185c | 4101 | count_memcg_event_mm(vma->vm_mm, PGFAULT); |
519e5247 JW |
4102 | |
4103 | /* do counter updates before entering really critical section. */ | |
4104 | check_sync_rss_stat(current); | |
4105 | ||
de0c799b LD |
4106 | if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE, |
4107 | flags & FAULT_FLAG_INSTRUCTION, | |
4108 | flags & FAULT_FLAG_REMOTE)) | |
4109 | return VM_FAULT_SIGSEGV; | |
4110 | ||
519e5247 JW |
4111 | /* |
4112 | * Enable the memcg OOM handling for faults triggered in user | |
4113 | * space. Kernel faults are handled more gracefully. | |
4114 | */ | |
4115 | if (flags & FAULT_FLAG_USER) | |
49426420 | 4116 | mem_cgroup_oom_enable(); |
519e5247 | 4117 | |
bae473a4 KS |
4118 | if (unlikely(is_vm_hugetlb_page(vma))) |
4119 | ret = hugetlb_fault(vma->vm_mm, vma, address, flags); | |
4120 | else | |
4121 | ret = __handle_mm_fault(vma, address, flags); | |
519e5247 | 4122 | |
49426420 JW |
4123 | if (flags & FAULT_FLAG_USER) { |
4124 | mem_cgroup_oom_disable(); | |
166f61b9 TH |
4125 | /* |
4126 | * The task may have entered a memcg OOM situation but | |
4127 | * if the allocation error was handled gracefully (no | |
4128 | * VM_FAULT_OOM), there is no need to kill anything. | |
4129 | * Just clean up the OOM state peacefully. | |
4130 | */ | |
4131 | if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM)) | |
4132 | mem_cgroup_oom_synchronize(false); | |
49426420 | 4133 | } |
3812c8c8 | 4134 | |
519e5247 JW |
4135 | return ret; |
4136 | } | |
e1d6d01a | 4137 | EXPORT_SYMBOL_GPL(handle_mm_fault); |
519e5247 | 4138 | |
90eceff1 KS |
4139 | #ifndef __PAGETABLE_P4D_FOLDED |
4140 | /* | |
4141 | * Allocate p4d page table. | |
4142 | * We've already handled the fast-path in-line. | |
4143 | */ | |
4144 | int __p4d_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) | |
4145 | { | |
4146 | p4d_t *new = p4d_alloc_one(mm, address); | |
4147 | if (!new) | |
4148 | return -ENOMEM; | |
4149 | ||
4150 | smp_wmb(); /* See comment in __pte_alloc */ | |
4151 | ||
4152 | spin_lock(&mm->page_table_lock); | |
4153 | if (pgd_present(*pgd)) /* Another has populated it */ | |
4154 | p4d_free(mm, new); | |
4155 | else | |
4156 | pgd_populate(mm, pgd, new); | |
4157 | spin_unlock(&mm->page_table_lock); | |
4158 | return 0; | |
4159 | } | |
4160 | #endif /* __PAGETABLE_P4D_FOLDED */ | |
4161 | ||
1da177e4 LT |
4162 | #ifndef __PAGETABLE_PUD_FOLDED |
4163 | /* | |
4164 | * Allocate page upper directory. | |
872fec16 | 4165 | * We've already handled the fast-path in-line. |
1da177e4 | 4166 | */ |
c2febafc | 4167 | int __pud_alloc(struct mm_struct *mm, p4d_t *p4d, unsigned long address) |
1da177e4 | 4168 | { |
c74df32c HD |
4169 | pud_t *new = pud_alloc_one(mm, address); |
4170 | if (!new) | |
1bb3630e | 4171 | return -ENOMEM; |
1da177e4 | 4172 | |
362a61ad NP |
4173 | smp_wmb(); /* See comment in __pte_alloc */ |
4174 | ||
872fec16 | 4175 | spin_lock(&mm->page_table_lock); |
c2febafc KS |
4176 | #ifndef __ARCH_HAS_5LEVEL_HACK |
4177 | if (p4d_present(*p4d)) /* Another has populated it */ | |
5e541973 | 4178 | pud_free(mm, new); |
1bb3630e | 4179 | else |
c2febafc KS |
4180 | p4d_populate(mm, p4d, new); |
4181 | #else | |
4182 | if (pgd_present(*p4d)) /* Another has populated it */ | |
5e541973 | 4183 | pud_free(mm, new); |
1bb3630e | 4184 | else |
c2febafc KS |
4185 | pgd_populate(mm, p4d, new); |
4186 | #endif /* __ARCH_HAS_5LEVEL_HACK */ | |
c74df32c | 4187 | spin_unlock(&mm->page_table_lock); |
1bb3630e | 4188 | return 0; |
1da177e4 LT |
4189 | } |
4190 | #endif /* __PAGETABLE_PUD_FOLDED */ | |
4191 | ||
4192 | #ifndef __PAGETABLE_PMD_FOLDED | |
4193 | /* | |
4194 | * Allocate page middle directory. | |
872fec16 | 4195 | * We've already handled the fast-path in-line. |
1da177e4 | 4196 | */ |
1bb3630e | 4197 | int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) |
1da177e4 | 4198 | { |
a00cc7d9 | 4199 | spinlock_t *ptl; |
c74df32c HD |
4200 | pmd_t *new = pmd_alloc_one(mm, address); |
4201 | if (!new) | |
1bb3630e | 4202 | return -ENOMEM; |
1da177e4 | 4203 | |
362a61ad NP |
4204 | smp_wmb(); /* See comment in __pte_alloc */ |
4205 | ||
a00cc7d9 | 4206 | ptl = pud_lock(mm, pud); |
1da177e4 | 4207 | #ifndef __ARCH_HAS_4LEVEL_HACK |
dc6c9a35 KS |
4208 | if (!pud_present(*pud)) { |
4209 | mm_inc_nr_pmds(mm); | |
1bb3630e | 4210 | pud_populate(mm, pud, new); |
dc6c9a35 | 4211 | } else /* Another has populated it */ |
5e541973 | 4212 | pmd_free(mm, new); |
dc6c9a35 KS |
4213 | #else |
4214 | if (!pgd_present(*pud)) { | |
4215 | mm_inc_nr_pmds(mm); | |
1bb3630e | 4216 | pgd_populate(mm, pud, new); |
dc6c9a35 KS |
4217 | } else /* Another has populated it */ |
4218 | pmd_free(mm, new); | |
1da177e4 | 4219 | #endif /* __ARCH_HAS_4LEVEL_HACK */ |
a00cc7d9 | 4220 | spin_unlock(ptl); |
1bb3630e | 4221 | return 0; |
e0f39591 | 4222 | } |
1da177e4 LT |
4223 | #endif /* __PAGETABLE_PMD_FOLDED */ |
4224 | ||
09796395 | 4225 | static int __follow_pte_pmd(struct mm_struct *mm, unsigned long address, |
a4d1a885 JG |
4226 | unsigned long *start, unsigned long *end, |
4227 | pte_t **ptepp, pmd_t **pmdpp, spinlock_t **ptlp) | |
f8ad0f49 JW |
4228 | { |
4229 | pgd_t *pgd; | |
c2febafc | 4230 | p4d_t *p4d; |
f8ad0f49 JW |
4231 | pud_t *pud; |
4232 | pmd_t *pmd; | |
4233 | pte_t *ptep; | |
4234 | ||
4235 | pgd = pgd_offset(mm, address); | |
4236 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
4237 | goto out; | |
4238 | ||
c2febafc KS |
4239 | p4d = p4d_offset(pgd, address); |
4240 | if (p4d_none(*p4d) || unlikely(p4d_bad(*p4d))) | |
4241 | goto out; | |
4242 | ||
4243 | pud = pud_offset(p4d, address); | |
f8ad0f49 JW |
4244 | if (pud_none(*pud) || unlikely(pud_bad(*pud))) |
4245 | goto out; | |
4246 | ||
4247 | pmd = pmd_offset(pud, address); | |
f66055ab | 4248 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
f8ad0f49 | 4249 | |
09796395 RZ |
4250 | if (pmd_huge(*pmd)) { |
4251 | if (!pmdpp) | |
4252 | goto out; | |
4253 | ||
a4d1a885 JG |
4254 | if (start && end) { |
4255 | *start = address & PMD_MASK; | |
4256 | *end = *start + PMD_SIZE; | |
4257 | mmu_notifier_invalidate_range_start(mm, *start, *end); | |
4258 | } | |
09796395 RZ |
4259 | *ptlp = pmd_lock(mm, pmd); |
4260 | if (pmd_huge(*pmd)) { | |
4261 | *pmdpp = pmd; | |
4262 | return 0; | |
4263 | } | |
4264 | spin_unlock(*ptlp); | |
a4d1a885 JG |
4265 | if (start && end) |
4266 | mmu_notifier_invalidate_range_end(mm, *start, *end); | |
09796395 RZ |
4267 | } |
4268 | ||
4269 | if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) | |
f8ad0f49 JW |
4270 | goto out; |
4271 | ||
a4d1a885 JG |
4272 | if (start && end) { |
4273 | *start = address & PAGE_MASK; | |
4274 | *end = *start + PAGE_SIZE; | |
4275 | mmu_notifier_invalidate_range_start(mm, *start, *end); | |
4276 | } | |
f8ad0f49 | 4277 | ptep = pte_offset_map_lock(mm, pmd, address, ptlp); |
f8ad0f49 JW |
4278 | if (!pte_present(*ptep)) |
4279 | goto unlock; | |
4280 | *ptepp = ptep; | |
4281 | return 0; | |
4282 | unlock: | |
4283 | pte_unmap_unlock(ptep, *ptlp); | |
a4d1a885 JG |
4284 | if (start && end) |
4285 | mmu_notifier_invalidate_range_end(mm, *start, *end); | |
f8ad0f49 JW |
4286 | out: |
4287 | return -EINVAL; | |
4288 | } | |
4289 | ||
f729c8c9 RZ |
4290 | static inline int follow_pte(struct mm_struct *mm, unsigned long address, |
4291 | pte_t **ptepp, spinlock_t **ptlp) | |
1b36ba81 NK |
4292 | { |
4293 | int res; | |
4294 | ||
4295 | /* (void) is needed to make gcc happy */ | |
4296 | (void) __cond_lock(*ptlp, | |
a4d1a885 JG |
4297 | !(res = __follow_pte_pmd(mm, address, NULL, NULL, |
4298 | ptepp, NULL, ptlp))); | |
09796395 RZ |
4299 | return res; |
4300 | } | |
4301 | ||
4302 | int follow_pte_pmd(struct mm_struct *mm, unsigned long address, | |
a4d1a885 | 4303 | unsigned long *start, unsigned long *end, |
09796395 RZ |
4304 | pte_t **ptepp, pmd_t **pmdpp, spinlock_t **ptlp) |
4305 | { | |
4306 | int res; | |
4307 | ||
4308 | /* (void) is needed to make gcc happy */ | |
4309 | (void) __cond_lock(*ptlp, | |
a4d1a885 JG |
4310 | !(res = __follow_pte_pmd(mm, address, start, end, |
4311 | ptepp, pmdpp, ptlp))); | |
1b36ba81 NK |
4312 | return res; |
4313 | } | |
09796395 | 4314 | EXPORT_SYMBOL(follow_pte_pmd); |
1b36ba81 | 4315 | |
3b6748e2 JW |
4316 | /** |
4317 | * follow_pfn - look up PFN at a user virtual address | |
4318 | * @vma: memory mapping | |
4319 | * @address: user virtual address | |
4320 | * @pfn: location to store found PFN | |
4321 | * | |
4322 | * Only IO mappings and raw PFN mappings are allowed. | |
4323 | * | |
4324 | * Returns zero and the pfn at @pfn on success, -ve otherwise. | |
4325 | */ | |
4326 | int follow_pfn(struct vm_area_struct *vma, unsigned long address, | |
4327 | unsigned long *pfn) | |
4328 | { | |
4329 | int ret = -EINVAL; | |
4330 | spinlock_t *ptl; | |
4331 | pte_t *ptep; | |
4332 | ||
4333 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) | |
4334 | return ret; | |
4335 | ||
4336 | ret = follow_pte(vma->vm_mm, address, &ptep, &ptl); | |
4337 | if (ret) | |
4338 | return ret; | |
4339 | *pfn = pte_pfn(*ptep); | |
4340 | pte_unmap_unlock(ptep, ptl); | |
4341 | return 0; | |
4342 | } | |
4343 | EXPORT_SYMBOL(follow_pfn); | |
4344 | ||
28b2ee20 | 4345 | #ifdef CONFIG_HAVE_IOREMAP_PROT |
d87fe660 | 4346 | int follow_phys(struct vm_area_struct *vma, |
4347 | unsigned long address, unsigned int flags, | |
4348 | unsigned long *prot, resource_size_t *phys) | |
28b2ee20 | 4349 | { |
03668a4d | 4350 | int ret = -EINVAL; |
28b2ee20 RR |
4351 | pte_t *ptep, pte; |
4352 | spinlock_t *ptl; | |
28b2ee20 | 4353 | |
d87fe660 | 4354 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) |
4355 | goto out; | |
28b2ee20 | 4356 | |
03668a4d | 4357 | if (follow_pte(vma->vm_mm, address, &ptep, &ptl)) |
d87fe660 | 4358 | goto out; |
28b2ee20 | 4359 | pte = *ptep; |
03668a4d | 4360 | |
28b2ee20 RR |
4361 | if ((flags & FOLL_WRITE) && !pte_write(pte)) |
4362 | goto unlock; | |
28b2ee20 RR |
4363 | |
4364 | *prot = pgprot_val(pte_pgprot(pte)); | |
03668a4d | 4365 | *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; |
28b2ee20 | 4366 | |
03668a4d | 4367 | ret = 0; |
28b2ee20 RR |
4368 | unlock: |
4369 | pte_unmap_unlock(ptep, ptl); | |
4370 | out: | |
d87fe660 | 4371 | return ret; |
28b2ee20 RR |
4372 | } |
4373 | ||
4374 | int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, | |
4375 | void *buf, int len, int write) | |
4376 | { | |
4377 | resource_size_t phys_addr; | |
4378 | unsigned long prot = 0; | |
2bc7273b | 4379 | void __iomem *maddr; |
28b2ee20 RR |
4380 | int offset = addr & (PAGE_SIZE-1); |
4381 | ||
d87fe660 | 4382 | if (follow_phys(vma, addr, write, &prot, &phys_addr)) |
28b2ee20 RR |
4383 | return -EINVAL; |
4384 | ||
9cb12d7b | 4385 | maddr = ioremap_prot(phys_addr, PAGE_ALIGN(len + offset), prot); |
cf7ab2ab | 4386 | if (!maddr) |
4387 | return -ENOMEM; | |
4388 | ||
28b2ee20 RR |
4389 | if (write) |
4390 | memcpy_toio(maddr + offset, buf, len); | |
4391 | else | |
4392 | memcpy_fromio(buf, maddr + offset, len); | |
4393 | iounmap(maddr); | |
4394 | ||
4395 | return len; | |
4396 | } | |
5a73633e | 4397 | EXPORT_SYMBOL_GPL(generic_access_phys); |
28b2ee20 RR |
4398 | #endif |
4399 | ||
0ec76a11 | 4400 | /* |
206cb636 SW |
4401 | * Access another process' address space as given in mm. If non-NULL, use the |
4402 | * given task for page fault accounting. | |
0ec76a11 | 4403 | */ |
84d77d3f | 4404 | int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, |
442486ec | 4405 | unsigned long addr, void *buf, int len, unsigned int gup_flags) |
0ec76a11 | 4406 | { |
0ec76a11 | 4407 | struct vm_area_struct *vma; |
0ec76a11 | 4408 | void *old_buf = buf; |
442486ec | 4409 | int write = gup_flags & FOLL_WRITE; |
0ec76a11 | 4410 | |
0ec76a11 | 4411 | down_read(&mm->mmap_sem); |
183ff22b | 4412 | /* ignore errors, just check how much was successfully transferred */ |
0ec76a11 DH |
4413 | while (len) { |
4414 | int bytes, ret, offset; | |
4415 | void *maddr; | |
28b2ee20 | 4416 | struct page *page = NULL; |
0ec76a11 | 4417 | |
1e987790 | 4418 | ret = get_user_pages_remote(tsk, mm, addr, 1, |
5b56d49f | 4419 | gup_flags, &page, &vma, NULL); |
28b2ee20 | 4420 | if (ret <= 0) { |
dbffcd03 RR |
4421 | #ifndef CONFIG_HAVE_IOREMAP_PROT |
4422 | break; | |
4423 | #else | |
28b2ee20 RR |
4424 | /* |
4425 | * Check if this is a VM_IO | VM_PFNMAP VMA, which | |
4426 | * we can access using slightly different code. | |
4427 | */ | |
28b2ee20 | 4428 | vma = find_vma(mm, addr); |
fe936dfc | 4429 | if (!vma || vma->vm_start > addr) |
28b2ee20 RR |
4430 | break; |
4431 | if (vma->vm_ops && vma->vm_ops->access) | |
4432 | ret = vma->vm_ops->access(vma, addr, buf, | |
4433 | len, write); | |
4434 | if (ret <= 0) | |
28b2ee20 RR |
4435 | break; |
4436 | bytes = ret; | |
dbffcd03 | 4437 | #endif |
0ec76a11 | 4438 | } else { |
28b2ee20 RR |
4439 | bytes = len; |
4440 | offset = addr & (PAGE_SIZE-1); | |
4441 | if (bytes > PAGE_SIZE-offset) | |
4442 | bytes = PAGE_SIZE-offset; | |
4443 | ||
4444 | maddr = kmap(page); | |
4445 | if (write) { | |
4446 | copy_to_user_page(vma, page, addr, | |
4447 | maddr + offset, buf, bytes); | |
4448 | set_page_dirty_lock(page); | |
4449 | } else { | |
4450 | copy_from_user_page(vma, page, addr, | |
4451 | buf, maddr + offset, bytes); | |
4452 | } | |
4453 | kunmap(page); | |
09cbfeaf | 4454 | put_page(page); |
0ec76a11 | 4455 | } |
0ec76a11 DH |
4456 | len -= bytes; |
4457 | buf += bytes; | |
4458 | addr += bytes; | |
4459 | } | |
4460 | up_read(&mm->mmap_sem); | |
0ec76a11 DH |
4461 | |
4462 | return buf - old_buf; | |
4463 | } | |
03252919 | 4464 | |
5ddd36b9 | 4465 | /** |
ae91dbfc | 4466 | * access_remote_vm - access another process' address space |
5ddd36b9 SW |
4467 | * @mm: the mm_struct of the target address space |
4468 | * @addr: start address to access | |
4469 | * @buf: source or destination buffer | |
4470 | * @len: number of bytes to transfer | |
6347e8d5 | 4471 | * @gup_flags: flags modifying lookup behaviour |
5ddd36b9 SW |
4472 | * |
4473 | * The caller must hold a reference on @mm. | |
4474 | */ | |
4475 | int access_remote_vm(struct mm_struct *mm, unsigned long addr, | |
6347e8d5 | 4476 | void *buf, int len, unsigned int gup_flags) |
5ddd36b9 | 4477 | { |
6347e8d5 | 4478 | return __access_remote_vm(NULL, mm, addr, buf, len, gup_flags); |
5ddd36b9 SW |
4479 | } |
4480 | ||
206cb636 SW |
4481 | /* |
4482 | * Access another process' address space. | |
4483 | * Source/target buffer must be kernel space, | |
4484 | * Do not walk the page table directly, use get_user_pages | |
4485 | */ | |
4486 | int access_process_vm(struct task_struct *tsk, unsigned long addr, | |
f307ab6d | 4487 | void *buf, int len, unsigned int gup_flags) |
206cb636 SW |
4488 | { |
4489 | struct mm_struct *mm; | |
4490 | int ret; | |
4491 | ||
4492 | mm = get_task_mm(tsk); | |
4493 | if (!mm) | |
4494 | return 0; | |
4495 | ||
f307ab6d | 4496 | ret = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags); |
442486ec | 4497 | |
206cb636 SW |
4498 | mmput(mm); |
4499 | ||
4500 | return ret; | |
4501 | } | |
fcd35857 | 4502 | EXPORT_SYMBOL_GPL(access_process_vm); |
206cb636 | 4503 | |
03252919 AK |
4504 | /* |
4505 | * Print the name of a VMA. | |
4506 | */ | |
4507 | void print_vma_addr(char *prefix, unsigned long ip) | |
4508 | { | |
4509 | struct mm_struct *mm = current->mm; | |
4510 | struct vm_area_struct *vma; | |
4511 | ||
e8bff74a IM |
4512 | /* |
4513 | * Do not print if we are in atomic | |
4514 | * contexts (in exception stacks, etc.): | |
4515 | */ | |
4516 | if (preempt_count()) | |
4517 | return; | |
4518 | ||
03252919 AK |
4519 | down_read(&mm->mmap_sem); |
4520 | vma = find_vma(mm, ip); | |
4521 | if (vma && vma->vm_file) { | |
4522 | struct file *f = vma->vm_file; | |
4523 | char *buf = (char *)__get_free_page(GFP_KERNEL); | |
4524 | if (buf) { | |
2fbc57c5 | 4525 | char *p; |
03252919 | 4526 | |
9bf39ab2 | 4527 | p = file_path(f, buf, PAGE_SIZE); |
03252919 AK |
4528 | if (IS_ERR(p)) |
4529 | p = "?"; | |
2fbc57c5 | 4530 | printk("%s%s[%lx+%lx]", prefix, kbasename(p), |
03252919 AK |
4531 | vma->vm_start, |
4532 | vma->vm_end - vma->vm_start); | |
4533 | free_page((unsigned long)buf); | |
4534 | } | |
4535 | } | |
51a07e50 | 4536 | up_read(&mm->mmap_sem); |
03252919 | 4537 | } |
3ee1afa3 | 4538 | |
662bbcb2 | 4539 | #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP) |
9ec23531 | 4540 | void __might_fault(const char *file, int line) |
3ee1afa3 | 4541 | { |
95156f00 PZ |
4542 | /* |
4543 | * Some code (nfs/sunrpc) uses socket ops on kernel memory while | |
4544 | * holding the mmap_sem, this is safe because kernel memory doesn't | |
4545 | * get paged out, therefore we'll never actually fault, and the | |
4546 | * below annotations will generate false positives. | |
4547 | */ | |
db68ce10 | 4548 | if (uaccess_kernel()) |
95156f00 | 4549 | return; |
9ec23531 | 4550 | if (pagefault_disabled()) |
662bbcb2 | 4551 | return; |
9ec23531 DH |
4552 | __might_sleep(file, line, 0); |
4553 | #if defined(CONFIG_DEBUG_ATOMIC_SLEEP) | |
662bbcb2 | 4554 | if (current->mm) |
3ee1afa3 | 4555 | might_lock_read(¤t->mm->mmap_sem); |
9ec23531 | 4556 | #endif |
3ee1afa3 | 4557 | } |
9ec23531 | 4558 | EXPORT_SYMBOL(__might_fault); |
3ee1afa3 | 4559 | #endif |
47ad8475 AA |
4560 | |
4561 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) | |
4562 | static void clear_gigantic_page(struct page *page, | |
4563 | unsigned long addr, | |
4564 | unsigned int pages_per_huge_page) | |
4565 | { | |
4566 | int i; | |
4567 | struct page *p = page; | |
4568 | ||
4569 | might_sleep(); | |
4570 | for (i = 0; i < pages_per_huge_page; | |
4571 | i++, p = mem_map_next(p, page, i)) { | |
4572 | cond_resched(); | |
4573 | clear_user_highpage(p, addr + i * PAGE_SIZE); | |
4574 | } | |
4575 | } | |
4576 | void clear_huge_page(struct page *page, | |
c79b57e4 | 4577 | unsigned long addr_hint, unsigned int pages_per_huge_page) |
47ad8475 | 4578 | { |
c79b57e4 HY |
4579 | int i, n, base, l; |
4580 | unsigned long addr = addr_hint & | |
4581 | ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); | |
47ad8475 AA |
4582 | |
4583 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { | |
4584 | clear_gigantic_page(page, addr, pages_per_huge_page); | |
4585 | return; | |
4586 | } | |
4587 | ||
c79b57e4 | 4588 | /* Clear sub-page to access last to keep its cache lines hot */ |
47ad8475 | 4589 | might_sleep(); |
c79b57e4 HY |
4590 | n = (addr_hint - addr) / PAGE_SIZE; |
4591 | if (2 * n <= pages_per_huge_page) { | |
4592 | /* If sub-page to access in first half of huge page */ | |
4593 | base = 0; | |
4594 | l = n; | |
4595 | /* Clear sub-pages at the end of huge page */ | |
4596 | for (i = pages_per_huge_page - 1; i >= 2 * n; i--) { | |
4597 | cond_resched(); | |
4598 | clear_user_highpage(page + i, addr + i * PAGE_SIZE); | |
4599 | } | |
4600 | } else { | |
4601 | /* If sub-page to access in second half of huge page */ | |
4602 | base = pages_per_huge_page - 2 * (pages_per_huge_page - n); | |
4603 | l = pages_per_huge_page - n; | |
4604 | /* Clear sub-pages at the begin of huge page */ | |
4605 | for (i = 0; i < base; i++) { | |
4606 | cond_resched(); | |
4607 | clear_user_highpage(page + i, addr + i * PAGE_SIZE); | |
4608 | } | |
4609 | } | |
4610 | /* | |
4611 | * Clear remaining sub-pages in left-right-left-right pattern | |
4612 | * towards the sub-page to access | |
4613 | */ | |
4614 | for (i = 0; i < l; i++) { | |
4615 | int left_idx = base + i; | |
4616 | int right_idx = base + 2 * l - 1 - i; | |
4617 | ||
4618 | cond_resched(); | |
4619 | clear_user_highpage(page + left_idx, | |
4620 | addr + left_idx * PAGE_SIZE); | |
47ad8475 | 4621 | cond_resched(); |
c79b57e4 HY |
4622 | clear_user_highpage(page + right_idx, |
4623 | addr + right_idx * PAGE_SIZE); | |
47ad8475 AA |
4624 | } |
4625 | } | |
4626 | ||
4627 | static void copy_user_gigantic_page(struct page *dst, struct page *src, | |
4628 | unsigned long addr, | |
4629 | struct vm_area_struct *vma, | |
4630 | unsigned int pages_per_huge_page) | |
4631 | { | |
4632 | int i; | |
4633 | struct page *dst_base = dst; | |
4634 | struct page *src_base = src; | |
4635 | ||
4636 | for (i = 0; i < pages_per_huge_page; ) { | |
4637 | cond_resched(); | |
4638 | copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma); | |
4639 | ||
4640 | i++; | |
4641 | dst = mem_map_next(dst, dst_base, i); | |
4642 | src = mem_map_next(src, src_base, i); | |
4643 | } | |
4644 | } | |
4645 | ||
4646 | void copy_user_huge_page(struct page *dst, struct page *src, | |
4647 | unsigned long addr, struct vm_area_struct *vma, | |
4648 | unsigned int pages_per_huge_page) | |
4649 | { | |
4650 | int i; | |
4651 | ||
4652 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { | |
4653 | copy_user_gigantic_page(dst, src, addr, vma, | |
4654 | pages_per_huge_page); | |
4655 | return; | |
4656 | } | |
4657 | ||
4658 | might_sleep(); | |
4659 | for (i = 0; i < pages_per_huge_page; i++) { | |
4660 | cond_resched(); | |
4661 | copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE, vma); | |
4662 | } | |
4663 | } | |
fa4d75c1 MK |
4664 | |
4665 | long copy_huge_page_from_user(struct page *dst_page, | |
4666 | const void __user *usr_src, | |
810a56b9 MK |
4667 | unsigned int pages_per_huge_page, |
4668 | bool allow_pagefault) | |
fa4d75c1 MK |
4669 | { |
4670 | void *src = (void *)usr_src; | |
4671 | void *page_kaddr; | |
4672 | unsigned long i, rc = 0; | |
4673 | unsigned long ret_val = pages_per_huge_page * PAGE_SIZE; | |
4674 | ||
4675 | for (i = 0; i < pages_per_huge_page; i++) { | |
810a56b9 MK |
4676 | if (allow_pagefault) |
4677 | page_kaddr = kmap(dst_page + i); | |
4678 | else | |
4679 | page_kaddr = kmap_atomic(dst_page + i); | |
fa4d75c1 MK |
4680 | rc = copy_from_user(page_kaddr, |
4681 | (const void __user *)(src + i * PAGE_SIZE), | |
4682 | PAGE_SIZE); | |
810a56b9 MK |
4683 | if (allow_pagefault) |
4684 | kunmap(dst_page + i); | |
4685 | else | |
4686 | kunmap_atomic(page_kaddr); | |
fa4d75c1 MK |
4687 | |
4688 | ret_val -= (PAGE_SIZE - rc); | |
4689 | if (rc) | |
4690 | break; | |
4691 | ||
4692 | cond_resched(); | |
4693 | } | |
4694 | return ret_val; | |
4695 | } | |
47ad8475 | 4696 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */ |
49076ec2 | 4697 | |
40b64acd | 4698 | #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS |
b35f1819 KS |
4699 | |
4700 | static struct kmem_cache *page_ptl_cachep; | |
4701 | ||
4702 | void __init ptlock_cache_init(void) | |
4703 | { | |
4704 | page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0, | |
4705 | SLAB_PANIC, NULL); | |
4706 | } | |
4707 | ||
539edb58 | 4708 | bool ptlock_alloc(struct page *page) |
49076ec2 KS |
4709 | { |
4710 | spinlock_t *ptl; | |
4711 | ||
b35f1819 | 4712 | ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL); |
49076ec2 KS |
4713 | if (!ptl) |
4714 | return false; | |
539edb58 | 4715 | page->ptl = ptl; |
49076ec2 KS |
4716 | return true; |
4717 | } | |
4718 | ||
539edb58 | 4719 | void ptlock_free(struct page *page) |
49076ec2 | 4720 | { |
b35f1819 | 4721 | kmem_cache_free(page_ptl_cachep, page->ptl); |
49076ec2 KS |
4722 | } |
4723 | #endif |