]>
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 | ||
f1521c14 | 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 | |
7607e716 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 | ||
021d5b54 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) { | |
021d5b54 | 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) { | |
021d5b54 | 398 | tlb_table_invalidate(tlb); |
26723911 PZ |
399 | tlb_remove_table_one(table); |
400 | return; | |
401 | } | |
402 | (*batch)->nr = 0; | |
403 | } | |
021d5b54 | 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); |
c4812909 | 450 | mm_dec_nr_ptes(tlb->mm); |
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); | |
b4e98d9a | 518 | mm_dec_nr_puds(tlb->mm); |
c2febafc KS |
519 | } |
520 | ||
521 | static inline void free_p4d_range(struct mmu_gather *tlb, pgd_t *pgd, | |
522 | unsigned long addr, unsigned long end, | |
523 | unsigned long floor, unsigned long ceiling) | |
524 | { | |
525 | p4d_t *p4d; | |
526 | unsigned long next; | |
527 | unsigned long start; | |
528 | ||
529 | start = addr; | |
530 | p4d = p4d_offset(pgd, addr); | |
531 | do { | |
532 | next = p4d_addr_end(addr, end); | |
533 | if (p4d_none_or_clear_bad(p4d)) | |
534 | continue; | |
535 | free_pud_range(tlb, p4d, addr, next, floor, ceiling); | |
536 | } while (p4d++, addr = next, addr != end); | |
537 | ||
e0da382c HD |
538 | start &= PGDIR_MASK; |
539 | if (start < floor) | |
540 | return; | |
541 | if (ceiling) { | |
542 | ceiling &= PGDIR_MASK; | |
543 | if (!ceiling) | |
544 | return; | |
1da177e4 | 545 | } |
e0da382c HD |
546 | if (end - 1 > ceiling - 1) |
547 | return; | |
548 | ||
c2febafc | 549 | p4d = p4d_offset(pgd, start); |
e0da382c | 550 | pgd_clear(pgd); |
c2febafc | 551 | p4d_free_tlb(tlb, p4d, start); |
1da177e4 LT |
552 | } |
553 | ||
554 | /* | |
e0da382c | 555 | * This function frees user-level page tables of a process. |
1da177e4 | 556 | */ |
42b77728 | 557 | void free_pgd_range(struct mmu_gather *tlb, |
e0da382c HD |
558 | unsigned long addr, unsigned long end, |
559 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
560 | { |
561 | pgd_t *pgd; | |
562 | unsigned long next; | |
e0da382c HD |
563 | |
564 | /* | |
565 | * The next few lines have given us lots of grief... | |
566 | * | |
567 | * Why are we testing PMD* at this top level? Because often | |
568 | * there will be no work to do at all, and we'd prefer not to | |
569 | * go all the way down to the bottom just to discover that. | |
570 | * | |
571 | * Why all these "- 1"s? Because 0 represents both the bottom | |
572 | * of the address space and the top of it (using -1 for the | |
573 | * top wouldn't help much: the masks would do the wrong thing). | |
574 | * The rule is that addr 0 and floor 0 refer to the bottom of | |
575 | * the address space, but end 0 and ceiling 0 refer to the top | |
576 | * Comparisons need to use "end - 1" and "ceiling - 1" (though | |
577 | * that end 0 case should be mythical). | |
578 | * | |
579 | * Wherever addr is brought up or ceiling brought down, we must | |
580 | * be careful to reject "the opposite 0" before it confuses the | |
581 | * subsequent tests. But what about where end is brought down | |
582 | * by PMD_SIZE below? no, end can't go down to 0 there. | |
583 | * | |
584 | * Whereas we round start (addr) and ceiling down, by different | |
585 | * masks at different levels, in order to test whether a table | |
586 | * now has no other vmas using it, so can be freed, we don't | |
587 | * bother to round floor or end up - the tests don't need that. | |
588 | */ | |
1da177e4 | 589 | |
e0da382c HD |
590 | addr &= PMD_MASK; |
591 | if (addr < floor) { | |
592 | addr += PMD_SIZE; | |
593 | if (!addr) | |
594 | return; | |
595 | } | |
596 | if (ceiling) { | |
597 | ceiling &= PMD_MASK; | |
598 | if (!ceiling) | |
599 | return; | |
600 | } | |
601 | if (end - 1 > ceiling - 1) | |
602 | end -= PMD_SIZE; | |
603 | if (addr > end - 1) | |
604 | return; | |
07e32661 AK |
605 | /* |
606 | * We add page table cache pages with PAGE_SIZE, | |
607 | * (see pte_free_tlb()), flush the tlb if we need | |
608 | */ | |
609 | tlb_remove_check_page_size_change(tlb, PAGE_SIZE); | |
42b77728 | 610 | pgd = pgd_offset(tlb->mm, addr); |
1da177e4 LT |
611 | do { |
612 | next = pgd_addr_end(addr, end); | |
613 | if (pgd_none_or_clear_bad(pgd)) | |
614 | continue; | |
c2febafc | 615 | free_p4d_range(tlb, pgd, addr, next, floor, ceiling); |
1da177e4 | 616 | } while (pgd++, addr = next, addr != end); |
e0da382c HD |
617 | } |
618 | ||
42b77728 | 619 | void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma, |
3bf5ee95 | 620 | unsigned long floor, unsigned long ceiling) |
e0da382c HD |
621 | { |
622 | while (vma) { | |
623 | struct vm_area_struct *next = vma->vm_next; | |
624 | unsigned long addr = vma->vm_start; | |
625 | ||
8f4f8c16 | 626 | /* |
25d9e2d1 | 627 | * Hide vma from rmap and truncate_pagecache before freeing |
628 | * pgtables | |
8f4f8c16 | 629 | */ |
5beb4930 | 630 | unlink_anon_vmas(vma); |
8f4f8c16 HD |
631 | unlink_file_vma(vma); |
632 | ||
9da61aef | 633 | if (is_vm_hugetlb_page(vma)) { |
3bf5ee95 | 634 | hugetlb_free_pgd_range(tlb, addr, vma->vm_end, |
166f61b9 | 635 | floor, next ? next->vm_start : ceiling); |
3bf5ee95 HD |
636 | } else { |
637 | /* | |
638 | * Optimization: gather nearby vmas into one call down | |
639 | */ | |
640 | while (next && next->vm_start <= vma->vm_end + PMD_SIZE | |
4866920b | 641 | && !is_vm_hugetlb_page(next)) { |
3bf5ee95 HD |
642 | vma = next; |
643 | next = vma->vm_next; | |
5beb4930 | 644 | unlink_anon_vmas(vma); |
8f4f8c16 | 645 | unlink_file_vma(vma); |
3bf5ee95 HD |
646 | } |
647 | free_pgd_range(tlb, addr, vma->vm_end, | |
166f61b9 | 648 | floor, next ? next->vm_start : ceiling); |
3bf5ee95 | 649 | } |
e0da382c HD |
650 | vma = next; |
651 | } | |
1da177e4 LT |
652 | } |
653 | ||
3ed3a4f0 | 654 | int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address) |
1da177e4 | 655 | { |
c4088ebd | 656 | spinlock_t *ptl; |
2f569afd | 657 | pgtable_t new = pte_alloc_one(mm, address); |
1bb3630e HD |
658 | if (!new) |
659 | return -ENOMEM; | |
660 | ||
362a61ad NP |
661 | /* |
662 | * Ensure all pte setup (eg. pte page lock and page clearing) are | |
663 | * visible before the pte is made visible to other CPUs by being | |
664 | * put into page tables. | |
665 | * | |
666 | * The other side of the story is the pointer chasing in the page | |
667 | * table walking code (when walking the page table without locking; | |
668 | * ie. most of the time). Fortunately, these data accesses consist | |
669 | * of a chain of data-dependent loads, meaning most CPUs (alpha | |
670 | * being the notable exception) will already guarantee loads are | |
671 | * seen in-order. See the alpha page table accessors for the | |
672 | * smp_read_barrier_depends() barriers in page table walking code. | |
673 | */ | |
674 | smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */ | |
675 | ||
c4088ebd | 676 | ptl = pmd_lock(mm, pmd); |
8ac1f832 | 677 | if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ |
c4812909 | 678 | mm_inc_nr_ptes(mm); |
1da177e4 | 679 | pmd_populate(mm, pmd, new); |
2f569afd | 680 | new = NULL; |
4b471e88 | 681 | } |
c4088ebd | 682 | spin_unlock(ptl); |
2f569afd MS |
683 | if (new) |
684 | pte_free(mm, new); | |
1bb3630e | 685 | return 0; |
1da177e4 LT |
686 | } |
687 | ||
1bb3630e | 688 | int __pte_alloc_kernel(pmd_t *pmd, unsigned long address) |
1da177e4 | 689 | { |
1bb3630e HD |
690 | pte_t *new = pte_alloc_one_kernel(&init_mm, address); |
691 | if (!new) | |
692 | return -ENOMEM; | |
693 | ||
362a61ad NP |
694 | smp_wmb(); /* See comment in __pte_alloc */ |
695 | ||
1bb3630e | 696 | spin_lock(&init_mm.page_table_lock); |
8ac1f832 | 697 | if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ |
1bb3630e | 698 | pmd_populate_kernel(&init_mm, pmd, new); |
2f569afd | 699 | new = NULL; |
4b471e88 | 700 | } |
1bb3630e | 701 | spin_unlock(&init_mm.page_table_lock); |
2f569afd MS |
702 | if (new) |
703 | pte_free_kernel(&init_mm, new); | |
1bb3630e | 704 | return 0; |
1da177e4 LT |
705 | } |
706 | ||
d559db08 KH |
707 | static inline void init_rss_vec(int *rss) |
708 | { | |
709 | memset(rss, 0, sizeof(int) * NR_MM_COUNTERS); | |
710 | } | |
711 | ||
712 | static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss) | |
ae859762 | 713 | { |
d559db08 KH |
714 | int i; |
715 | ||
34e55232 | 716 | if (current->mm == mm) |
05af2e10 | 717 | sync_mm_rss(mm); |
d559db08 KH |
718 | for (i = 0; i < NR_MM_COUNTERS; i++) |
719 | if (rss[i]) | |
720 | add_mm_counter(mm, i, rss[i]); | |
ae859762 HD |
721 | } |
722 | ||
b5810039 | 723 | /* |
6aab341e LT |
724 | * This function is called to print an error when a bad pte |
725 | * is found. For example, we might have a PFN-mapped pte in | |
726 | * a region that doesn't allow it. | |
b5810039 NP |
727 | * |
728 | * The calling function must still handle the error. | |
729 | */ | |
3dc14741 HD |
730 | static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr, |
731 | pte_t pte, struct page *page) | |
b5810039 | 732 | { |
3dc14741 | 733 | pgd_t *pgd = pgd_offset(vma->vm_mm, addr); |
c2febafc KS |
734 | p4d_t *p4d = p4d_offset(pgd, addr); |
735 | pud_t *pud = pud_offset(p4d, addr); | |
3dc14741 HD |
736 | pmd_t *pmd = pmd_offset(pud, addr); |
737 | struct address_space *mapping; | |
738 | pgoff_t index; | |
d936cf9b HD |
739 | static unsigned long resume; |
740 | static unsigned long nr_shown; | |
741 | static unsigned long nr_unshown; | |
742 | ||
743 | /* | |
744 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
745 | * or allow a steady drip of one report per second. | |
746 | */ | |
747 | if (nr_shown == 60) { | |
748 | if (time_before(jiffies, resume)) { | |
749 | nr_unshown++; | |
750 | return; | |
751 | } | |
752 | if (nr_unshown) { | |
1170532b JP |
753 | pr_alert("BUG: Bad page map: %lu messages suppressed\n", |
754 | nr_unshown); | |
d936cf9b HD |
755 | nr_unshown = 0; |
756 | } | |
757 | nr_shown = 0; | |
758 | } | |
759 | if (nr_shown++ == 0) | |
760 | resume = jiffies + 60 * HZ; | |
3dc14741 HD |
761 | |
762 | mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL; | |
763 | index = linear_page_index(vma, addr); | |
764 | ||
1170532b JP |
765 | pr_alert("BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n", |
766 | current->comm, | |
767 | (long long)pte_val(pte), (long long)pmd_val(*pmd)); | |
718a3821 | 768 | if (page) |
f0b791a3 | 769 | dump_page(page, "bad pte"); |
1170532b JP |
770 | pr_alert("addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n", |
771 | (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index); | |
3dc14741 HD |
772 | /* |
773 | * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y | |
774 | */ | |
2682582a KK |
775 | pr_alert("file:%pD fault:%pf mmap:%pf readpage:%pf\n", |
776 | vma->vm_file, | |
777 | vma->vm_ops ? vma->vm_ops->fault : NULL, | |
778 | vma->vm_file ? vma->vm_file->f_op->mmap : NULL, | |
779 | mapping ? mapping->a_ops->readpage : NULL); | |
b5810039 | 780 | dump_stack(); |
373d4d09 | 781 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
b5810039 NP |
782 | } |
783 | ||
ee498ed7 | 784 | /* |
7e675137 | 785 | * vm_normal_page -- This function gets the "struct page" associated with a pte. |
6aab341e | 786 | * |
7e675137 NP |
787 | * "Special" mappings do not wish to be associated with a "struct page" (either |
788 | * it doesn't exist, or it exists but they don't want to touch it). In this | |
789 | * case, NULL is returned here. "Normal" mappings do have a struct page. | |
b379d790 | 790 | * |
7e675137 NP |
791 | * There are 2 broad cases. Firstly, an architecture may define a pte_special() |
792 | * pte bit, in which case this function is trivial. Secondly, an architecture | |
793 | * may not have a spare pte bit, which requires a more complicated scheme, | |
794 | * described below. | |
795 | * | |
796 | * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a | |
797 | * special mapping (even if there are underlying and valid "struct pages"). | |
798 | * COWed pages of a VM_PFNMAP are always normal. | |
6aab341e | 799 | * |
b379d790 JH |
800 | * The way we recognize COWed pages within VM_PFNMAP mappings is through the |
801 | * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit | |
7e675137 NP |
802 | * set, and the vm_pgoff will point to the first PFN mapped: thus every special |
803 | * mapping will always honor the rule | |
6aab341e LT |
804 | * |
805 | * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) | |
806 | * | |
7e675137 NP |
807 | * And for normal mappings this is false. |
808 | * | |
809 | * This restricts such mappings to be a linear translation from virtual address | |
810 | * to pfn. To get around this restriction, we allow arbitrary mappings so long | |
811 | * as the vma is not a COW mapping; in that case, we know that all ptes are | |
812 | * special (because none can have been COWed). | |
b379d790 | 813 | * |
b379d790 | 814 | * |
7e675137 | 815 | * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. |
b379d790 JH |
816 | * |
817 | * VM_MIXEDMAP mappings can likewise contain memory with or without "struct | |
818 | * page" backing, however the difference is that _all_ pages with a struct | |
819 | * page (that is, those where pfn_valid is true) are refcounted and considered | |
820 | * normal pages by the VM. The disadvantage is that pages are refcounted | |
821 | * (which can be slower and simply not an option for some PFNMAP users). The | |
822 | * advantage is that we don't have to follow the strict linearity rule of | |
823 | * PFNMAP mappings in order to support COWable mappings. | |
824 | * | |
ee498ed7 | 825 | */ |
7e675137 NP |
826 | #ifdef __HAVE_ARCH_PTE_SPECIAL |
827 | # define HAVE_PTE_SPECIAL 1 | |
828 | #else | |
829 | # define HAVE_PTE_SPECIAL 0 | |
830 | #endif | |
df6ad698 JG |
831 | struct page *_vm_normal_page(struct vm_area_struct *vma, unsigned long addr, |
832 | pte_t pte, bool with_public_device) | |
ee498ed7 | 833 | { |
22b31eec | 834 | unsigned long pfn = pte_pfn(pte); |
7e675137 NP |
835 | |
836 | if (HAVE_PTE_SPECIAL) { | |
b38af472 | 837 | if (likely(!pte_special(pte))) |
22b31eec | 838 | goto check_pfn; |
667a0a06 DV |
839 | if (vma->vm_ops && vma->vm_ops->find_special_page) |
840 | return vma->vm_ops->find_special_page(vma, addr); | |
a13ea5b7 HD |
841 | if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) |
842 | return NULL; | |
df6ad698 JG |
843 | if (is_zero_pfn(pfn)) |
844 | return NULL; | |
845 | ||
846 | /* | |
847 | * Device public pages are special pages (they are ZONE_DEVICE | |
848 | * pages but different from persistent memory). They behave | |
849 | * allmost like normal pages. The difference is that they are | |
850 | * not on the lru and thus should never be involve with any- | |
851 | * thing that involve lru manipulation (mlock, numa balancing, | |
852 | * ...). | |
853 | * | |
854 | * This is why we still want to return NULL for such page from | |
855 | * vm_normal_page() so that we do not have to special case all | |
856 | * call site of vm_normal_page(). | |
857 | */ | |
7d790d2d | 858 | if (likely(pfn <= highest_memmap_pfn)) { |
df6ad698 JG |
859 | struct page *page = pfn_to_page(pfn); |
860 | ||
861 | if (is_device_public_page(page)) { | |
862 | if (with_public_device) | |
863 | return page; | |
864 | return NULL; | |
865 | } | |
866 | } | |
867 | print_bad_pte(vma, addr, pte, NULL); | |
7e675137 NP |
868 | return NULL; |
869 | } | |
870 | ||
871 | /* !HAVE_PTE_SPECIAL case follows: */ | |
872 | ||
b379d790 JH |
873 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { |
874 | if (vma->vm_flags & VM_MIXEDMAP) { | |
875 | if (!pfn_valid(pfn)) | |
876 | return NULL; | |
877 | goto out; | |
878 | } else { | |
7e675137 NP |
879 | unsigned long off; |
880 | off = (addr - vma->vm_start) >> PAGE_SHIFT; | |
b379d790 JH |
881 | if (pfn == vma->vm_pgoff + off) |
882 | return NULL; | |
883 | if (!is_cow_mapping(vma->vm_flags)) | |
884 | return NULL; | |
885 | } | |
6aab341e LT |
886 | } |
887 | ||
b38af472 HD |
888 | if (is_zero_pfn(pfn)) |
889 | return NULL; | |
22b31eec HD |
890 | check_pfn: |
891 | if (unlikely(pfn > highest_memmap_pfn)) { | |
892 | print_bad_pte(vma, addr, pte, NULL); | |
893 | return NULL; | |
894 | } | |
6aab341e LT |
895 | |
896 | /* | |
7e675137 | 897 | * NOTE! We still have PageReserved() pages in the page tables. |
7e675137 | 898 | * eg. VDSO mappings can cause them to exist. |
6aab341e | 899 | */ |
b379d790 | 900 | out: |
6aab341e | 901 | return pfn_to_page(pfn); |
ee498ed7 HD |
902 | } |
903 | ||
28093f9f GS |
904 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
905 | struct page *vm_normal_page_pmd(struct vm_area_struct *vma, unsigned long addr, | |
906 | pmd_t pmd) | |
907 | { | |
908 | unsigned long pfn = pmd_pfn(pmd); | |
909 | ||
910 | /* | |
911 | * There is no pmd_special() but there may be special pmds, e.g. | |
912 | * in a direct-access (dax) mapping, so let's just replicate the | |
913 | * !HAVE_PTE_SPECIAL case from vm_normal_page() here. | |
914 | */ | |
915 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { | |
916 | if (vma->vm_flags & VM_MIXEDMAP) { | |
917 | if (!pfn_valid(pfn)) | |
918 | return NULL; | |
919 | goto out; | |
920 | } else { | |
921 | unsigned long off; | |
922 | off = (addr - vma->vm_start) >> PAGE_SHIFT; | |
923 | if (pfn == vma->vm_pgoff + off) | |
924 | return NULL; | |
925 | if (!is_cow_mapping(vma->vm_flags)) | |
926 | return NULL; | |
927 | } | |
928 | } | |
929 | ||
930 | if (is_zero_pfn(pfn)) | |
931 | return NULL; | |
932 | if (unlikely(pfn > highest_memmap_pfn)) | |
933 | return NULL; | |
934 | ||
935 | /* | |
936 | * NOTE! We still have PageReserved() pages in the page tables. | |
937 | * eg. VDSO mappings can cause them to exist. | |
938 | */ | |
939 | out: | |
940 | return pfn_to_page(pfn); | |
941 | } | |
942 | #endif | |
943 | ||
1da177e4 LT |
944 | /* |
945 | * copy one vm_area from one task to the other. Assumes the page tables | |
946 | * already present in the new task to be cleared in the whole range | |
947 | * covered by this vma. | |
1da177e4 LT |
948 | */ |
949 | ||
570a335b | 950 | static inline unsigned long |
1da177e4 | 951 | copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
b5810039 | 952 | pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma, |
8c103762 | 953 | unsigned long addr, int *rss) |
1da177e4 | 954 | { |
b5810039 | 955 | unsigned long vm_flags = vma->vm_flags; |
1da177e4 LT |
956 | pte_t pte = *src_pte; |
957 | struct page *page; | |
1da177e4 LT |
958 | |
959 | /* pte contains position in swap or file, so copy. */ | |
960 | if (unlikely(!pte_present(pte))) { | |
0661a336 KS |
961 | swp_entry_t entry = pte_to_swp_entry(pte); |
962 | ||
963 | if (likely(!non_swap_entry(entry))) { | |
964 | if (swap_duplicate(entry) < 0) | |
965 | return entry.val; | |
966 | ||
967 | /* make sure dst_mm is on swapoff's mmlist. */ | |
968 | if (unlikely(list_empty(&dst_mm->mmlist))) { | |
969 | spin_lock(&mmlist_lock); | |
970 | if (list_empty(&dst_mm->mmlist)) | |
971 | list_add(&dst_mm->mmlist, | |
972 | &src_mm->mmlist); | |
973 | spin_unlock(&mmlist_lock); | |
974 | } | |
975 | rss[MM_SWAPENTS]++; | |
976 | } else if (is_migration_entry(entry)) { | |
977 | page = migration_entry_to_page(entry); | |
978 | ||
eca56ff9 | 979 | rss[mm_counter(page)]++; |
0661a336 KS |
980 | |
981 | if (is_write_migration_entry(entry) && | |
982 | is_cow_mapping(vm_flags)) { | |
983 | /* | |
984 | * COW mappings require pages in both | |
985 | * parent and child to be set to read. | |
986 | */ | |
987 | make_migration_entry_read(&entry); | |
988 | pte = swp_entry_to_pte(entry); | |
989 | if (pte_swp_soft_dirty(*src_pte)) | |
990 | pte = pte_swp_mksoft_dirty(pte); | |
991 | set_pte_at(src_mm, addr, src_pte, pte); | |
0697212a | 992 | } |
5042db43 JG |
993 | } else if (is_device_private_entry(entry)) { |
994 | page = device_private_entry_to_page(entry); | |
995 | ||
996 | /* | |
997 | * Update rss count even for unaddressable pages, as | |
998 | * they should treated just like normal pages in this | |
999 | * respect. | |
1000 | * | |
1001 | * We will likely want to have some new rss counters | |
1002 | * for unaddressable pages, at some point. But for now | |
1003 | * keep things as they are. | |
1004 | */ | |
1005 | get_page(page); | |
1006 | rss[mm_counter(page)]++; | |
1007 | page_dup_rmap(page, false); | |
1008 | ||
1009 | /* | |
1010 | * We do not preserve soft-dirty information, because so | |
1011 | * far, checkpoint/restore is the only feature that | |
1012 | * requires that. And checkpoint/restore does not work | |
1013 | * when a device driver is involved (you cannot easily | |
1014 | * save and restore device driver state). | |
1015 | */ | |
1016 | if (is_write_device_private_entry(entry) && | |
1017 | is_cow_mapping(vm_flags)) { | |
1018 | make_device_private_entry_read(&entry); | |
1019 | pte = swp_entry_to_pte(entry); | |
1020 | set_pte_at(src_mm, addr, src_pte, pte); | |
1021 | } | |
1da177e4 | 1022 | } |
ae859762 | 1023 | goto out_set_pte; |
1da177e4 LT |
1024 | } |
1025 | ||
1da177e4 LT |
1026 | /* |
1027 | * If it's a COW mapping, write protect it both | |
1028 | * in the parent and the child | |
1029 | */ | |
67121172 | 1030 | if (is_cow_mapping(vm_flags)) { |
1da177e4 | 1031 | ptep_set_wrprotect(src_mm, addr, src_pte); |
3dc90795 | 1032 | pte = pte_wrprotect(pte); |
1da177e4 LT |
1033 | } |
1034 | ||
1035 | /* | |
1036 | * If it's a shared mapping, mark it clean in | |
1037 | * the child | |
1038 | */ | |
1039 | if (vm_flags & VM_SHARED) | |
1040 | pte = pte_mkclean(pte); | |
1041 | pte = pte_mkold(pte); | |
6aab341e LT |
1042 | |
1043 | page = vm_normal_page(vma, addr, pte); | |
1044 | if (page) { | |
1045 | get_page(page); | |
53f9263b | 1046 | page_dup_rmap(page, false); |
eca56ff9 | 1047 | rss[mm_counter(page)]++; |
df6ad698 JG |
1048 | } else if (pte_devmap(pte)) { |
1049 | page = pte_page(pte); | |
1050 | ||
1051 | /* | |
1052 | * Cache coherent device memory behave like regular page and | |
1053 | * not like persistent memory page. For more informations see | |
1054 | * MEMORY_DEVICE_CACHE_COHERENT in memory_hotplug.h | |
1055 | */ | |
1056 | if (is_device_public_page(page)) { | |
1057 | get_page(page); | |
1058 | page_dup_rmap(page, false); | |
1059 | rss[mm_counter(page)]++; | |
1060 | } | |
6aab341e | 1061 | } |
ae859762 HD |
1062 | |
1063 | out_set_pte: | |
1064 | set_pte_at(dst_mm, addr, dst_pte, pte); | |
570a335b | 1065 | return 0; |
1da177e4 LT |
1066 | } |
1067 | ||
21bda264 | 1068 | static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
71e3aac0 AA |
1069 | pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma, |
1070 | unsigned long addr, unsigned long end) | |
1da177e4 | 1071 | { |
c36987e2 | 1072 | pte_t *orig_src_pte, *orig_dst_pte; |
1da177e4 | 1073 | pte_t *src_pte, *dst_pte; |
c74df32c | 1074 | spinlock_t *src_ptl, *dst_ptl; |
e040f218 | 1075 | int progress = 0; |
d559db08 | 1076 | int rss[NR_MM_COUNTERS]; |
570a335b | 1077 | swp_entry_t entry = (swp_entry_t){0}; |
1da177e4 LT |
1078 | |
1079 | again: | |
d559db08 KH |
1080 | init_rss_vec(rss); |
1081 | ||
c74df32c | 1082 | dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); |
1da177e4 LT |
1083 | if (!dst_pte) |
1084 | return -ENOMEM; | |
ece0e2b6 | 1085 | src_pte = pte_offset_map(src_pmd, addr); |
4c21e2f2 | 1086 | src_ptl = pte_lockptr(src_mm, src_pmd); |
f20dc5f7 | 1087 | spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); |
c36987e2 DN |
1088 | orig_src_pte = src_pte; |
1089 | orig_dst_pte = dst_pte; | |
6606c3e0 | 1090 | arch_enter_lazy_mmu_mode(); |
1da177e4 | 1091 | |
1da177e4 LT |
1092 | do { |
1093 | /* | |
1094 | * We are holding two locks at this point - either of them | |
1095 | * could generate latencies in another task on another CPU. | |
1096 | */ | |
e040f218 HD |
1097 | if (progress >= 32) { |
1098 | progress = 0; | |
1099 | if (need_resched() || | |
95c354fe | 1100 | spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) |
e040f218 HD |
1101 | break; |
1102 | } | |
1da177e4 LT |
1103 | if (pte_none(*src_pte)) { |
1104 | progress++; | |
1105 | continue; | |
1106 | } | |
570a335b HD |
1107 | entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, |
1108 | vma, addr, rss); | |
1109 | if (entry.val) | |
1110 | break; | |
1da177e4 LT |
1111 | progress += 8; |
1112 | } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); | |
1da177e4 | 1113 | |
6606c3e0 | 1114 | arch_leave_lazy_mmu_mode(); |
c74df32c | 1115 | spin_unlock(src_ptl); |
ece0e2b6 | 1116 | pte_unmap(orig_src_pte); |
d559db08 | 1117 | add_mm_rss_vec(dst_mm, rss); |
c36987e2 | 1118 | pte_unmap_unlock(orig_dst_pte, dst_ptl); |
c74df32c | 1119 | cond_resched(); |
570a335b HD |
1120 | |
1121 | if (entry.val) { | |
1122 | if (add_swap_count_continuation(entry, GFP_KERNEL) < 0) | |
1123 | return -ENOMEM; | |
1124 | progress = 0; | |
1125 | } | |
1da177e4 LT |
1126 | if (addr != end) |
1127 | goto again; | |
1128 | return 0; | |
1129 | } | |
1130 | ||
1131 | static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
1132 | pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma, | |
1133 | unsigned long addr, unsigned long end) | |
1134 | { | |
1135 | pmd_t *src_pmd, *dst_pmd; | |
1136 | unsigned long next; | |
1137 | ||
1138 | dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); | |
1139 | if (!dst_pmd) | |
1140 | return -ENOMEM; | |
1141 | src_pmd = pmd_offset(src_pud, addr); | |
1142 | do { | |
1143 | next = pmd_addr_end(addr, end); | |
84c3fc4e ZY |
1144 | if (is_swap_pmd(*src_pmd) || pmd_trans_huge(*src_pmd) |
1145 | || pmd_devmap(*src_pmd)) { | |
71e3aac0 | 1146 | int err; |
a00cc7d9 | 1147 | VM_BUG_ON_VMA(next-addr != HPAGE_PMD_SIZE, vma); |
71e3aac0 AA |
1148 | err = copy_huge_pmd(dst_mm, src_mm, |
1149 | dst_pmd, src_pmd, addr, vma); | |
1150 | if (err == -ENOMEM) | |
1151 | return -ENOMEM; | |
1152 | if (!err) | |
1153 | continue; | |
1154 | /* fall through */ | |
1155 | } | |
1da177e4 LT |
1156 | if (pmd_none_or_clear_bad(src_pmd)) |
1157 | continue; | |
1158 | if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd, | |
1159 | vma, addr, next)) | |
1160 | return -ENOMEM; | |
1161 | } while (dst_pmd++, src_pmd++, addr = next, addr != end); | |
1162 | return 0; | |
1163 | } | |
1164 | ||
1165 | static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
c2febafc | 1166 | p4d_t *dst_p4d, p4d_t *src_p4d, struct vm_area_struct *vma, |
1da177e4 LT |
1167 | unsigned long addr, unsigned long end) |
1168 | { | |
1169 | pud_t *src_pud, *dst_pud; | |
1170 | unsigned long next; | |
1171 | ||
c2febafc | 1172 | dst_pud = pud_alloc(dst_mm, dst_p4d, addr); |
1da177e4 LT |
1173 | if (!dst_pud) |
1174 | return -ENOMEM; | |
c2febafc | 1175 | src_pud = pud_offset(src_p4d, addr); |
1da177e4 LT |
1176 | do { |
1177 | next = pud_addr_end(addr, end); | |
a00cc7d9 MW |
1178 | if (pud_trans_huge(*src_pud) || pud_devmap(*src_pud)) { |
1179 | int err; | |
1180 | ||
1181 | VM_BUG_ON_VMA(next-addr != HPAGE_PUD_SIZE, vma); | |
1182 | err = copy_huge_pud(dst_mm, src_mm, | |
1183 | dst_pud, src_pud, addr, vma); | |
1184 | if (err == -ENOMEM) | |
1185 | return -ENOMEM; | |
1186 | if (!err) | |
1187 | continue; | |
1188 | /* fall through */ | |
1189 | } | |
1da177e4 LT |
1190 | if (pud_none_or_clear_bad(src_pud)) |
1191 | continue; | |
1192 | if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud, | |
1193 | vma, addr, next)) | |
1194 | return -ENOMEM; | |
1195 | } while (dst_pud++, src_pud++, addr = next, addr != end); | |
1196 | return 0; | |
1197 | } | |
1198 | ||
c2febafc KS |
1199 | static inline int copy_p4d_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
1200 | pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma, | |
1201 | unsigned long addr, unsigned long end) | |
1202 | { | |
1203 | p4d_t *src_p4d, *dst_p4d; | |
1204 | unsigned long next; | |
1205 | ||
1206 | dst_p4d = p4d_alloc(dst_mm, dst_pgd, addr); | |
1207 | if (!dst_p4d) | |
1208 | return -ENOMEM; | |
1209 | src_p4d = p4d_offset(src_pgd, addr); | |
1210 | do { | |
1211 | next = p4d_addr_end(addr, end); | |
1212 | if (p4d_none_or_clear_bad(src_p4d)) | |
1213 | continue; | |
1214 | if (copy_pud_range(dst_mm, src_mm, dst_p4d, src_p4d, | |
1215 | vma, addr, next)) | |
1216 | return -ENOMEM; | |
1217 | } while (dst_p4d++, src_p4d++, addr = next, addr != end); | |
1218 | return 0; | |
1219 | } | |
1220 | ||
1da177e4 LT |
1221 | int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
1222 | struct vm_area_struct *vma) | |
1223 | { | |
1224 | pgd_t *src_pgd, *dst_pgd; | |
1225 | unsigned long next; | |
1226 | unsigned long addr = vma->vm_start; | |
1227 | unsigned long end = vma->vm_end; | |
2ec74c3e SG |
1228 | unsigned long mmun_start; /* For mmu_notifiers */ |
1229 | unsigned long mmun_end; /* For mmu_notifiers */ | |
1230 | bool is_cow; | |
cddb8a5c | 1231 | int ret; |
1da177e4 | 1232 | |
d992895b NP |
1233 | /* |
1234 | * Don't copy ptes where a page fault will fill them correctly. | |
1235 | * Fork becomes much lighter when there are big shared or private | |
1236 | * readonly mappings. The tradeoff is that copy_page_range is more | |
1237 | * efficient than faulting. | |
1238 | */ | |
0661a336 KS |
1239 | if (!(vma->vm_flags & (VM_HUGETLB | VM_PFNMAP | VM_MIXEDMAP)) && |
1240 | !vma->anon_vma) | |
1241 | return 0; | |
d992895b | 1242 | |
1da177e4 LT |
1243 | if (is_vm_hugetlb_page(vma)) |
1244 | return copy_hugetlb_page_range(dst_mm, src_mm, vma); | |
1245 | ||
b3b9c293 | 1246 | if (unlikely(vma->vm_flags & VM_PFNMAP)) { |
2ab64037 | 1247 | /* |
1248 | * We do not free on error cases below as remove_vma | |
1249 | * gets called on error from higher level routine | |
1250 | */ | |
5180da41 | 1251 | ret = track_pfn_copy(vma); |
2ab64037 | 1252 | if (ret) |
1253 | return ret; | |
1254 | } | |
1255 | ||
cddb8a5c AA |
1256 | /* |
1257 | * We need to invalidate the secondary MMU mappings only when | |
1258 | * there could be a permission downgrade on the ptes of the | |
1259 | * parent mm. And a permission downgrade will only happen if | |
1260 | * is_cow_mapping() returns true. | |
1261 | */ | |
2ec74c3e SG |
1262 | is_cow = is_cow_mapping(vma->vm_flags); |
1263 | mmun_start = addr; | |
1264 | mmun_end = end; | |
1265 | if (is_cow) | |
1266 | mmu_notifier_invalidate_range_start(src_mm, mmun_start, | |
1267 | mmun_end); | |
cddb8a5c AA |
1268 | |
1269 | ret = 0; | |
1da177e4 LT |
1270 | dst_pgd = pgd_offset(dst_mm, addr); |
1271 | src_pgd = pgd_offset(src_mm, addr); | |
1272 | do { | |
1273 | next = pgd_addr_end(addr, end); | |
1274 | if (pgd_none_or_clear_bad(src_pgd)) | |
1275 | continue; | |
c2febafc | 1276 | if (unlikely(copy_p4d_range(dst_mm, src_mm, dst_pgd, src_pgd, |
cddb8a5c AA |
1277 | vma, addr, next))) { |
1278 | ret = -ENOMEM; | |
1279 | break; | |
1280 | } | |
1da177e4 | 1281 | } while (dst_pgd++, src_pgd++, addr = next, addr != end); |
cddb8a5c | 1282 | |
2ec74c3e SG |
1283 | if (is_cow) |
1284 | mmu_notifier_invalidate_range_end(src_mm, mmun_start, mmun_end); | |
cddb8a5c | 1285 | return ret; |
1da177e4 LT |
1286 | } |
1287 | ||
51c6f666 | 1288 | static unsigned long zap_pte_range(struct mmu_gather *tlb, |
b5810039 | 1289 | struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 | 1290 | unsigned long addr, unsigned long end, |
97a89413 | 1291 | struct zap_details *details) |
1da177e4 | 1292 | { |
b5810039 | 1293 | struct mm_struct *mm = tlb->mm; |
d16dfc55 | 1294 | int force_flush = 0; |
d559db08 | 1295 | int rss[NR_MM_COUNTERS]; |
97a89413 | 1296 | spinlock_t *ptl; |
5f1a1907 | 1297 | pte_t *start_pte; |
97a89413 | 1298 | pte_t *pte; |
8a5f14a2 | 1299 | swp_entry_t entry; |
d559db08 | 1300 | |
07e32661 | 1301 | tlb_remove_check_page_size_change(tlb, PAGE_SIZE); |
d16dfc55 | 1302 | again: |
e303297e | 1303 | init_rss_vec(rss); |
5f1a1907 SR |
1304 | start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
1305 | pte = start_pte; | |
3ea27719 | 1306 | flush_tlb_batched_pending(mm); |
6606c3e0 | 1307 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
1308 | do { |
1309 | pte_t ptent = *pte; | |
166f61b9 | 1310 | if (pte_none(ptent)) |
1da177e4 | 1311 | continue; |
6f5e6b9e | 1312 | |
1da177e4 | 1313 | if (pte_present(ptent)) { |
ee498ed7 | 1314 | struct page *page; |
51c6f666 | 1315 | |
df6ad698 | 1316 | page = _vm_normal_page(vma, addr, ptent, true); |
1da177e4 LT |
1317 | if (unlikely(details) && page) { |
1318 | /* | |
1319 | * unmap_shared_mapping_pages() wants to | |
1320 | * invalidate cache without truncating: | |
1321 | * unmap shared but keep private pages. | |
1322 | */ | |
1323 | if (details->check_mapping && | |
800d8c63 | 1324 | details->check_mapping != page_rmapping(page)) |
1da177e4 | 1325 | continue; |
1da177e4 | 1326 | } |
b5810039 | 1327 | ptent = ptep_get_and_clear_full(mm, addr, pte, |
a600388d | 1328 | tlb->fullmm); |
1da177e4 LT |
1329 | tlb_remove_tlb_entry(tlb, pte, addr); |
1330 | if (unlikely(!page)) | |
1331 | continue; | |
eca56ff9 JM |
1332 | |
1333 | if (!PageAnon(page)) { | |
1cf35d47 LT |
1334 | if (pte_dirty(ptent)) { |
1335 | force_flush = 1; | |
6237bcd9 | 1336 | set_page_dirty(page); |
1cf35d47 | 1337 | } |
4917e5d0 | 1338 | if (pte_young(ptent) && |
64363aad | 1339 | likely(!(vma->vm_flags & VM_SEQ_READ))) |
bf3f3bc5 | 1340 | mark_page_accessed(page); |
6237bcd9 | 1341 | } |
eca56ff9 | 1342 | rss[mm_counter(page)]--; |
d281ee61 | 1343 | page_remove_rmap(page, false); |
3dc14741 HD |
1344 | if (unlikely(page_mapcount(page) < 0)) |
1345 | print_bad_pte(vma, addr, ptent, page); | |
e9d55e15 | 1346 | if (unlikely(__tlb_remove_page(tlb, page))) { |
1cf35d47 | 1347 | force_flush = 1; |
ce9ec37b | 1348 | addr += PAGE_SIZE; |
d16dfc55 | 1349 | break; |
1cf35d47 | 1350 | } |
1da177e4 LT |
1351 | continue; |
1352 | } | |
5042db43 JG |
1353 | |
1354 | entry = pte_to_swp_entry(ptent); | |
1355 | if (non_swap_entry(entry) && is_device_private_entry(entry)) { | |
1356 | struct page *page = device_private_entry_to_page(entry); | |
1357 | ||
1358 | if (unlikely(details && details->check_mapping)) { | |
1359 | /* | |
1360 | * unmap_shared_mapping_pages() wants to | |
1361 | * invalidate cache without truncating: | |
1362 | * unmap shared but keep private pages. | |
1363 | */ | |
1364 | if (details->check_mapping != | |
1365 | page_rmapping(page)) | |
1366 | continue; | |
1367 | } | |
1368 | ||
1369 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); | |
1370 | rss[mm_counter(page)]--; | |
1371 | page_remove_rmap(page, false); | |
1372 | put_page(page); | |
1373 | continue; | |
1374 | } | |
1375 | ||
3e8715fd KS |
1376 | /* If details->check_mapping, we leave swap entries. */ |
1377 | if (unlikely(details)) | |
1da177e4 | 1378 | continue; |
b084d435 | 1379 | |
8a5f14a2 KS |
1380 | entry = pte_to_swp_entry(ptent); |
1381 | if (!non_swap_entry(entry)) | |
1382 | rss[MM_SWAPENTS]--; | |
1383 | else if (is_migration_entry(entry)) { | |
1384 | struct page *page; | |
9f9f1acd | 1385 | |
8a5f14a2 | 1386 | page = migration_entry_to_page(entry); |
eca56ff9 | 1387 | rss[mm_counter(page)]--; |
b084d435 | 1388 | } |
8a5f14a2 KS |
1389 | if (unlikely(!free_swap_and_cache(entry))) |
1390 | print_bad_pte(vma, addr, ptent, NULL); | |
9888a1ca | 1391 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); |
97a89413 | 1392 | } while (pte++, addr += PAGE_SIZE, addr != end); |
ae859762 | 1393 | |
d559db08 | 1394 | add_mm_rss_vec(mm, rss); |
6606c3e0 | 1395 | arch_leave_lazy_mmu_mode(); |
51c6f666 | 1396 | |
1cf35d47 | 1397 | /* Do the actual TLB flush before dropping ptl */ |
fb7332a9 | 1398 | if (force_flush) |
1cf35d47 | 1399 | tlb_flush_mmu_tlbonly(tlb); |
1cf35d47 LT |
1400 | pte_unmap_unlock(start_pte, ptl); |
1401 | ||
1402 | /* | |
1403 | * If we forced a TLB flush (either due to running out of | |
1404 | * batch buffers or because we needed to flush dirty TLB | |
1405 | * entries before releasing the ptl), free the batched | |
1406 | * memory too. Restart if we didn't do everything. | |
1407 | */ | |
1408 | if (force_flush) { | |
1409 | force_flush = 0; | |
1410 | tlb_flush_mmu_free(tlb); | |
2b047252 | 1411 | if (addr != end) |
d16dfc55 PZ |
1412 | goto again; |
1413 | } | |
1414 | ||
51c6f666 | 1415 | return addr; |
1da177e4 LT |
1416 | } |
1417 | ||
51c6f666 | 1418 | static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, |
b5810039 | 1419 | struct vm_area_struct *vma, pud_t *pud, |
1da177e4 | 1420 | unsigned long addr, unsigned long end, |
97a89413 | 1421 | struct zap_details *details) |
1da177e4 LT |
1422 | { |
1423 | pmd_t *pmd; | |
1424 | unsigned long next; | |
1425 | ||
1426 | pmd = pmd_offset(pud, addr); | |
1427 | do { | |
1428 | next = pmd_addr_end(addr, end); | |
84c3fc4e | 1429 | if (is_swap_pmd(*pmd) || pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) { |
56ccc392 | 1430 | if (next - addr != HPAGE_PMD_SIZE) |
fd60775a | 1431 | __split_huge_pmd(vma, pmd, addr, false, NULL); |
56ccc392 | 1432 | else if (zap_huge_pmd(tlb, vma, pmd, addr)) |
1a5a9906 | 1433 | goto next; |
71e3aac0 AA |
1434 | /* fall through */ |
1435 | } | |
1a5a9906 AA |
1436 | /* |
1437 | * Here there can be other concurrent MADV_DONTNEED or | |
1438 | * trans huge page faults running, and if the pmd is | |
1439 | * none or trans huge it can change under us. This is | |
1440 | * because MADV_DONTNEED holds the mmap_sem in read | |
1441 | * mode. | |
1442 | */ | |
1443 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) | |
1444 | goto next; | |
97a89413 | 1445 | next = zap_pte_range(tlb, vma, pmd, addr, next, details); |
1a5a9906 | 1446 | next: |
97a89413 PZ |
1447 | cond_resched(); |
1448 | } while (pmd++, addr = next, addr != end); | |
51c6f666 RH |
1449 | |
1450 | return addr; | |
1da177e4 LT |
1451 | } |
1452 | ||
51c6f666 | 1453 | static inline unsigned long zap_pud_range(struct mmu_gather *tlb, |
c2febafc | 1454 | struct vm_area_struct *vma, p4d_t *p4d, |
1da177e4 | 1455 | unsigned long addr, unsigned long end, |
97a89413 | 1456 | struct zap_details *details) |
1da177e4 LT |
1457 | { |
1458 | pud_t *pud; | |
1459 | unsigned long next; | |
1460 | ||
c2febafc | 1461 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
1462 | do { |
1463 | next = pud_addr_end(addr, end); | |
a00cc7d9 MW |
1464 | if (pud_trans_huge(*pud) || pud_devmap(*pud)) { |
1465 | if (next - addr != HPAGE_PUD_SIZE) { | |
1466 | VM_BUG_ON_VMA(!rwsem_is_locked(&tlb->mm->mmap_sem), vma); | |
1467 | split_huge_pud(vma, pud, addr); | |
1468 | } else if (zap_huge_pud(tlb, vma, pud, addr)) | |
1469 | goto next; | |
1470 | /* fall through */ | |
1471 | } | |
97a89413 | 1472 | if (pud_none_or_clear_bad(pud)) |
1da177e4 | 1473 | continue; |
97a89413 | 1474 | next = zap_pmd_range(tlb, vma, pud, addr, next, details); |
a00cc7d9 MW |
1475 | next: |
1476 | cond_resched(); | |
97a89413 | 1477 | } while (pud++, addr = next, addr != end); |
51c6f666 RH |
1478 | |
1479 | return addr; | |
1da177e4 LT |
1480 | } |
1481 | ||
c2febafc KS |
1482 | static inline unsigned long zap_p4d_range(struct mmu_gather *tlb, |
1483 | struct vm_area_struct *vma, pgd_t *pgd, | |
1484 | unsigned long addr, unsigned long end, | |
1485 | struct zap_details *details) | |
1486 | { | |
1487 | p4d_t *p4d; | |
1488 | unsigned long next; | |
1489 | ||
1490 | p4d = p4d_offset(pgd, addr); | |
1491 | do { | |
1492 | next = p4d_addr_end(addr, end); | |
1493 | if (p4d_none_or_clear_bad(p4d)) | |
1494 | continue; | |
1495 | next = zap_pud_range(tlb, vma, p4d, addr, next, details); | |
1496 | } while (p4d++, addr = next, addr != end); | |
1497 | ||
1498 | return addr; | |
1499 | } | |
1500 | ||
aac45363 | 1501 | void unmap_page_range(struct mmu_gather *tlb, |
038c7aa1 AV |
1502 | struct vm_area_struct *vma, |
1503 | unsigned long addr, unsigned long end, | |
1504 | struct zap_details *details) | |
1da177e4 LT |
1505 | { |
1506 | pgd_t *pgd; | |
1507 | unsigned long next; | |
1508 | ||
1da177e4 LT |
1509 | BUG_ON(addr >= end); |
1510 | tlb_start_vma(tlb, vma); | |
1511 | pgd = pgd_offset(vma->vm_mm, addr); | |
1512 | do { | |
1513 | next = pgd_addr_end(addr, end); | |
97a89413 | 1514 | if (pgd_none_or_clear_bad(pgd)) |
1da177e4 | 1515 | continue; |
c2febafc | 1516 | next = zap_p4d_range(tlb, vma, pgd, addr, next, details); |
97a89413 | 1517 | } while (pgd++, addr = next, addr != end); |
1da177e4 LT |
1518 | tlb_end_vma(tlb, vma); |
1519 | } | |
51c6f666 | 1520 | |
f5cc4eef AV |
1521 | |
1522 | static void unmap_single_vma(struct mmu_gather *tlb, | |
1523 | struct vm_area_struct *vma, unsigned long start_addr, | |
4f74d2c8 | 1524 | unsigned long end_addr, |
f5cc4eef AV |
1525 | struct zap_details *details) |
1526 | { | |
1527 | unsigned long start = max(vma->vm_start, start_addr); | |
1528 | unsigned long end; | |
1529 | ||
1530 | if (start >= vma->vm_end) | |
1531 | return; | |
1532 | end = min(vma->vm_end, end_addr); | |
1533 | if (end <= vma->vm_start) | |
1534 | return; | |
1535 | ||
cbc91f71 SD |
1536 | if (vma->vm_file) |
1537 | uprobe_munmap(vma, start, end); | |
1538 | ||
b3b9c293 | 1539 | if (unlikely(vma->vm_flags & VM_PFNMAP)) |
5180da41 | 1540 | untrack_pfn(vma, 0, 0); |
f5cc4eef AV |
1541 | |
1542 | if (start != end) { | |
1543 | if (unlikely(is_vm_hugetlb_page(vma))) { | |
1544 | /* | |
1545 | * It is undesirable to test vma->vm_file as it | |
1546 | * should be non-null for valid hugetlb area. | |
1547 | * However, vm_file will be NULL in the error | |
7aa6b4ad | 1548 | * cleanup path of mmap_region. When |
f5cc4eef | 1549 | * hugetlbfs ->mmap method fails, |
7aa6b4ad | 1550 | * mmap_region() nullifies vma->vm_file |
f5cc4eef AV |
1551 | * before calling this function to clean up. |
1552 | * Since no pte has actually been setup, it is | |
1553 | * safe to do nothing in this case. | |
1554 | */ | |
24669e58 | 1555 | if (vma->vm_file) { |
83cde9e8 | 1556 | i_mmap_lock_write(vma->vm_file->f_mapping); |
d833352a | 1557 | __unmap_hugepage_range_final(tlb, vma, start, end, NULL); |
83cde9e8 | 1558 | i_mmap_unlock_write(vma->vm_file->f_mapping); |
24669e58 | 1559 | } |
f5cc4eef AV |
1560 | } else |
1561 | unmap_page_range(tlb, vma, start, end, details); | |
1562 | } | |
1da177e4 LT |
1563 | } |
1564 | ||
1da177e4 LT |
1565 | /** |
1566 | * unmap_vmas - unmap a range of memory covered by a list of vma's | |
0164f69d | 1567 | * @tlb: address of the caller's struct mmu_gather |
1da177e4 LT |
1568 | * @vma: the starting vma |
1569 | * @start_addr: virtual address at which to start unmapping | |
1570 | * @end_addr: virtual address at which to end unmapping | |
1da177e4 | 1571 | * |
508034a3 | 1572 | * Unmap all pages in the vma list. |
1da177e4 | 1573 | * |
1da177e4 LT |
1574 | * Only addresses between `start' and `end' will be unmapped. |
1575 | * | |
1576 | * The VMA list must be sorted in ascending virtual address order. | |
1577 | * | |
1578 | * unmap_vmas() assumes that the caller will flush the whole unmapped address | |
1579 | * range after unmap_vmas() returns. So the only responsibility here is to | |
1580 | * ensure that any thus-far unmapped pages are flushed before unmap_vmas() | |
1581 | * drops the lock and schedules. | |
1582 | */ | |
6e8bb019 | 1583 | void unmap_vmas(struct mmu_gather *tlb, |
1da177e4 | 1584 | struct vm_area_struct *vma, unsigned long start_addr, |
4f74d2c8 | 1585 | unsigned long end_addr) |
1da177e4 | 1586 | { |
cddb8a5c | 1587 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 1588 | |
cddb8a5c | 1589 | mmu_notifier_invalidate_range_start(mm, start_addr, end_addr); |
f5cc4eef | 1590 | for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) |
4f74d2c8 | 1591 | unmap_single_vma(tlb, vma, start_addr, end_addr, NULL); |
cddb8a5c | 1592 | mmu_notifier_invalidate_range_end(mm, start_addr, end_addr); |
1da177e4 LT |
1593 | } |
1594 | ||
1595 | /** | |
1596 | * zap_page_range - remove user pages in a given range | |
1597 | * @vma: vm_area_struct holding the applicable pages | |
eb4546bb | 1598 | * @start: starting address of pages to zap |
1da177e4 | 1599 | * @size: number of bytes to zap |
f5cc4eef AV |
1600 | * |
1601 | * Caller must protect the VMA list | |
1da177e4 | 1602 | */ |
7e027b14 | 1603 | void zap_page_range(struct vm_area_struct *vma, unsigned long start, |
ecf1385d | 1604 | unsigned long size) |
1da177e4 LT |
1605 | { |
1606 | struct mm_struct *mm = vma->vm_mm; | |
d16dfc55 | 1607 | struct mmu_gather tlb; |
7e027b14 | 1608 | unsigned long end = start + size; |
1da177e4 | 1609 | |
1da177e4 | 1610 | lru_add_drain(); |
2b047252 | 1611 | tlb_gather_mmu(&tlb, mm, start, end); |
365e9c87 | 1612 | update_hiwater_rss(mm); |
7e027b14 | 1613 | mmu_notifier_invalidate_range_start(mm, start, end); |
4647706e | 1614 | for ( ; vma && vma->vm_start < end; vma = vma->vm_next) { |
ecf1385d | 1615 | unmap_single_vma(&tlb, vma, start, end, NULL); |
4647706e MG |
1616 | |
1617 | /* | |
1618 | * zap_page_range does not specify whether mmap_sem should be | |
1619 | * held for read or write. That allows parallel zap_page_range | |
1620 | * operations to unmap a PTE and defer a flush meaning that | |
1621 | * this call observes pte_none and fails to flush the TLB. | |
1622 | * Rather than adding a complex API, ensure that no stale | |
1623 | * TLB entries exist when this call returns. | |
1624 | */ | |
1625 | flush_tlb_range(vma, start, end); | |
1626 | } | |
1627 | ||
7e027b14 LT |
1628 | mmu_notifier_invalidate_range_end(mm, start, end); |
1629 | tlb_finish_mmu(&tlb, start, end); | |
1da177e4 LT |
1630 | } |
1631 | ||
f5cc4eef AV |
1632 | /** |
1633 | * zap_page_range_single - remove user pages in a given range | |
1634 | * @vma: vm_area_struct holding the applicable pages | |
1635 | * @address: starting address of pages to zap | |
1636 | * @size: number of bytes to zap | |
8a5f14a2 | 1637 | * @details: details of shared cache invalidation |
f5cc4eef AV |
1638 | * |
1639 | * The range must fit into one VMA. | |
1da177e4 | 1640 | */ |
f5cc4eef | 1641 | static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address, |
1da177e4 LT |
1642 | unsigned long size, struct zap_details *details) |
1643 | { | |
1644 | struct mm_struct *mm = vma->vm_mm; | |
d16dfc55 | 1645 | struct mmu_gather tlb; |
1da177e4 | 1646 | unsigned long end = address + size; |
1da177e4 | 1647 | |
1da177e4 | 1648 | lru_add_drain(); |
2b047252 | 1649 | tlb_gather_mmu(&tlb, mm, address, end); |
365e9c87 | 1650 | update_hiwater_rss(mm); |
f5cc4eef | 1651 | mmu_notifier_invalidate_range_start(mm, address, end); |
4f74d2c8 | 1652 | unmap_single_vma(&tlb, vma, address, end, details); |
f5cc4eef | 1653 | mmu_notifier_invalidate_range_end(mm, address, end); |
d16dfc55 | 1654 | tlb_finish_mmu(&tlb, address, end); |
1da177e4 LT |
1655 | } |
1656 | ||
c627f9cc JS |
1657 | /** |
1658 | * zap_vma_ptes - remove ptes mapping the vma | |
1659 | * @vma: vm_area_struct holding ptes to be zapped | |
1660 | * @address: starting address of pages to zap | |
1661 | * @size: number of bytes to zap | |
1662 | * | |
1663 | * This function only unmaps ptes assigned to VM_PFNMAP vmas. | |
1664 | * | |
1665 | * The entire address range must be fully contained within the vma. | |
1666 | * | |
1667 | * Returns 0 if successful. | |
1668 | */ | |
1669 | int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, | |
1670 | unsigned long size) | |
1671 | { | |
1672 | if (address < vma->vm_start || address + size > vma->vm_end || | |
1673 | !(vma->vm_flags & VM_PFNMAP)) | |
1674 | return -1; | |
f5cc4eef | 1675 | zap_page_range_single(vma, address, size, NULL); |
c627f9cc JS |
1676 | return 0; |
1677 | } | |
1678 | EXPORT_SYMBOL_GPL(zap_vma_ptes); | |
1679 | ||
25ca1d6c | 1680 | pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr, |
920c7a5d | 1681 | spinlock_t **ptl) |
c9cfcddf | 1682 | { |
c2febafc KS |
1683 | pgd_t *pgd; |
1684 | p4d_t *p4d; | |
1685 | pud_t *pud; | |
1686 | pmd_t *pmd; | |
1687 | ||
1688 | pgd = pgd_offset(mm, addr); | |
1689 | p4d = p4d_alloc(mm, pgd, addr); | |
1690 | if (!p4d) | |
1691 | return NULL; | |
1692 | pud = pud_alloc(mm, p4d, addr); | |
1693 | if (!pud) | |
1694 | return NULL; | |
1695 | pmd = pmd_alloc(mm, pud, addr); | |
1696 | if (!pmd) | |
1697 | return NULL; | |
1698 | ||
1699 | VM_BUG_ON(pmd_trans_huge(*pmd)); | |
1700 | return pte_alloc_map_lock(mm, pmd, addr, ptl); | |
c9cfcddf LT |
1701 | } |
1702 | ||
238f58d8 LT |
1703 | /* |
1704 | * This is the old fallback for page remapping. | |
1705 | * | |
1706 | * For historical reasons, it only allows reserved pages. Only | |
1707 | * old drivers should use this, and they needed to mark their | |
1708 | * pages reserved for the old functions anyway. | |
1709 | */ | |
423bad60 NP |
1710 | static int insert_page(struct vm_area_struct *vma, unsigned long addr, |
1711 | struct page *page, pgprot_t prot) | |
238f58d8 | 1712 | { |
423bad60 | 1713 | struct mm_struct *mm = vma->vm_mm; |
238f58d8 | 1714 | int retval; |
c9cfcddf | 1715 | pte_t *pte; |
8a9f3ccd BS |
1716 | spinlock_t *ptl; |
1717 | ||
238f58d8 | 1718 | retval = -EINVAL; |
a145dd41 | 1719 | if (PageAnon(page)) |
5b4e655e | 1720 | goto out; |
238f58d8 LT |
1721 | retval = -ENOMEM; |
1722 | flush_dcache_page(page); | |
c9cfcddf | 1723 | pte = get_locked_pte(mm, addr, &ptl); |
238f58d8 | 1724 | if (!pte) |
5b4e655e | 1725 | goto out; |
238f58d8 LT |
1726 | retval = -EBUSY; |
1727 | if (!pte_none(*pte)) | |
1728 | goto out_unlock; | |
1729 | ||
1730 | /* Ok, finally just insert the thing.. */ | |
1731 | get_page(page); | |
eca56ff9 | 1732 | inc_mm_counter_fast(mm, mm_counter_file(page)); |
dd78fedd | 1733 | page_add_file_rmap(page, false); |
238f58d8 LT |
1734 | set_pte_at(mm, addr, pte, mk_pte(page, prot)); |
1735 | ||
1736 | retval = 0; | |
8a9f3ccd BS |
1737 | pte_unmap_unlock(pte, ptl); |
1738 | return retval; | |
238f58d8 LT |
1739 | out_unlock: |
1740 | pte_unmap_unlock(pte, ptl); | |
1741 | out: | |
1742 | return retval; | |
1743 | } | |
1744 | ||
bfa5bf6d REB |
1745 | /** |
1746 | * vm_insert_page - insert single page into user vma | |
1747 | * @vma: user vma to map to | |
1748 | * @addr: target user address of this page | |
1749 | * @page: source kernel page | |
1750 | * | |
a145dd41 LT |
1751 | * This allows drivers to insert individual pages they've allocated |
1752 | * into a user vma. | |
1753 | * | |
1754 | * The page has to be a nice clean _individual_ kernel allocation. | |
1755 | * If you allocate a compound page, you need to have marked it as | |
1756 | * such (__GFP_COMP), or manually just split the page up yourself | |
8dfcc9ba | 1757 | * (see split_page()). |
a145dd41 LT |
1758 | * |
1759 | * NOTE! Traditionally this was done with "remap_pfn_range()" which | |
1760 | * took an arbitrary page protection parameter. This doesn't allow | |
1761 | * that. Your vma protection will have to be set up correctly, which | |
1762 | * means that if you want a shared writable mapping, you'd better | |
1763 | * ask for a shared writable mapping! | |
1764 | * | |
1765 | * The page does not need to be reserved. | |
4b6e1e37 KK |
1766 | * |
1767 | * Usually this function is called from f_op->mmap() handler | |
1768 | * under mm->mmap_sem write-lock, so it can change vma->vm_flags. | |
1769 | * Caller must set VM_MIXEDMAP on vma if it wants to call this | |
1770 | * function from other places, for example from page-fault handler. | |
a145dd41 | 1771 | */ |
423bad60 NP |
1772 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, |
1773 | struct page *page) | |
a145dd41 LT |
1774 | { |
1775 | if (addr < vma->vm_start || addr >= vma->vm_end) | |
1776 | return -EFAULT; | |
1777 | if (!page_count(page)) | |
1778 | return -EINVAL; | |
4b6e1e37 KK |
1779 | if (!(vma->vm_flags & VM_MIXEDMAP)) { |
1780 | BUG_ON(down_read_trylock(&vma->vm_mm->mmap_sem)); | |
1781 | BUG_ON(vma->vm_flags & VM_PFNMAP); | |
1782 | vma->vm_flags |= VM_MIXEDMAP; | |
1783 | } | |
423bad60 | 1784 | return insert_page(vma, addr, page, vma->vm_page_prot); |
a145dd41 | 1785 | } |
e3c3374f | 1786 | EXPORT_SYMBOL(vm_insert_page); |
a145dd41 | 1787 | |
423bad60 | 1788 | static int insert_pfn(struct vm_area_struct *vma, unsigned long addr, |
b2770da6 | 1789 | pfn_t pfn, pgprot_t prot, bool mkwrite) |
423bad60 NP |
1790 | { |
1791 | struct mm_struct *mm = vma->vm_mm; | |
1792 | int retval; | |
1793 | pte_t *pte, entry; | |
1794 | spinlock_t *ptl; | |
1795 | ||
1796 | retval = -ENOMEM; | |
1797 | pte = get_locked_pte(mm, addr, &ptl); | |
1798 | if (!pte) | |
1799 | goto out; | |
1800 | retval = -EBUSY; | |
b2770da6 RZ |
1801 | if (!pte_none(*pte)) { |
1802 | if (mkwrite) { | |
1803 | /* | |
1804 | * For read faults on private mappings the PFN passed | |
1805 | * in may not match the PFN we have mapped if the | |
1806 | * mapped PFN is a writeable COW page. In the mkwrite | |
1807 | * case we are creating a writable PTE for a shared | |
1808 | * mapping and we expect the PFNs to match. | |
1809 | */ | |
1810 | if (WARN_ON_ONCE(pte_pfn(*pte) != pfn_t_to_pfn(pfn))) | |
1811 | goto out_unlock; | |
1812 | entry = *pte; | |
1813 | goto out_mkwrite; | |
1814 | } else | |
1815 | goto out_unlock; | |
1816 | } | |
423bad60 NP |
1817 | |
1818 | /* Ok, finally just insert the thing.. */ | |
01c8f1c4 DW |
1819 | if (pfn_t_devmap(pfn)) |
1820 | entry = pte_mkdevmap(pfn_t_pte(pfn, prot)); | |
1821 | else | |
1822 | entry = pte_mkspecial(pfn_t_pte(pfn, prot)); | |
b2770da6 RZ |
1823 | |
1824 | out_mkwrite: | |
1825 | if (mkwrite) { | |
1826 | entry = pte_mkyoung(entry); | |
1827 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
1828 | } | |
1829 | ||
423bad60 | 1830 | set_pte_at(mm, addr, pte, entry); |
4b3073e1 | 1831 | update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */ |
423bad60 NP |
1832 | |
1833 | retval = 0; | |
1834 | out_unlock: | |
1835 | pte_unmap_unlock(pte, ptl); | |
1836 | out: | |
1837 | return retval; | |
1838 | } | |
1839 | ||
e0dc0d8f NP |
1840 | /** |
1841 | * vm_insert_pfn - insert single pfn into user vma | |
1842 | * @vma: user vma to map to | |
1843 | * @addr: target user address of this page | |
1844 | * @pfn: source kernel pfn | |
1845 | * | |
c462f179 | 1846 | * Similar to vm_insert_page, this allows drivers to insert individual pages |
e0dc0d8f NP |
1847 | * they've allocated into a user vma. Same comments apply. |
1848 | * | |
1849 | * This function should only be called from a vm_ops->fault handler, and | |
1850 | * in that case the handler should return NULL. | |
0d71d10a NP |
1851 | * |
1852 | * vma cannot be a COW mapping. | |
1853 | * | |
1854 | * As this is called only for pages that do not currently exist, we | |
1855 | * do not need to flush old virtual caches or the TLB. | |
e0dc0d8f NP |
1856 | */ |
1857 | int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr, | |
423bad60 | 1858 | unsigned long pfn) |
1745cbc5 AL |
1859 | { |
1860 | return vm_insert_pfn_prot(vma, addr, pfn, vma->vm_page_prot); | |
1861 | } | |
1862 | EXPORT_SYMBOL(vm_insert_pfn); | |
1863 | ||
1864 | /** | |
1865 | * vm_insert_pfn_prot - insert single pfn into user vma with specified pgprot | |
1866 | * @vma: user vma to map to | |
1867 | * @addr: target user address of this page | |
1868 | * @pfn: source kernel pfn | |
1869 | * @pgprot: pgprot flags for the inserted page | |
1870 | * | |
1871 | * This is exactly like vm_insert_pfn, except that it allows drivers to | |
1872 | * to override pgprot on a per-page basis. | |
1873 | * | |
1874 | * This only makes sense for IO mappings, and it makes no sense for | |
1875 | * cow mappings. In general, using multiple vmas is preferable; | |
1876 | * vm_insert_pfn_prot should only be used if using multiple VMAs is | |
1877 | * impractical. | |
1878 | */ | |
1879 | int vm_insert_pfn_prot(struct vm_area_struct *vma, unsigned long addr, | |
1880 | unsigned long pfn, pgprot_t pgprot) | |
e0dc0d8f | 1881 | { |
2ab64037 | 1882 | int ret; |
7e675137 NP |
1883 | /* |
1884 | * Technically, architectures with pte_special can avoid all these | |
1885 | * restrictions (same for remap_pfn_range). However we would like | |
1886 | * consistency in testing and feature parity among all, so we should | |
1887 | * try to keep these invariants in place for everybody. | |
1888 | */ | |
b379d790 JH |
1889 | BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); |
1890 | BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == | |
1891 | (VM_PFNMAP|VM_MIXEDMAP)); | |
1892 | BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); | |
1893 | BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); | |
e0dc0d8f | 1894 | |
423bad60 NP |
1895 | if (addr < vma->vm_start || addr >= vma->vm_end) |
1896 | return -EFAULT; | |
308a047c | 1897 | |
0831b2a2 AK |
1898 | if (!pfn_modify_allowed(pfn, pgprot)) |
1899 | return -EACCES; | |
1900 | ||
308a047c | 1901 | track_pfn_insert(vma, &pgprot, __pfn_to_pfn_t(pfn, PFN_DEV)); |
2ab64037 | 1902 | |
b2770da6 RZ |
1903 | ret = insert_pfn(vma, addr, __pfn_to_pfn_t(pfn, PFN_DEV), pgprot, |
1904 | false); | |
2ab64037 | 1905 | |
2ab64037 | 1906 | return ret; |
423bad60 | 1907 | } |
1745cbc5 | 1908 | EXPORT_SYMBOL(vm_insert_pfn_prot); |
e0dc0d8f | 1909 | |
c11da42b DW |
1910 | static bool vm_mixed_ok(struct vm_area_struct *vma, pfn_t pfn) |
1911 | { | |
1912 | /* these checks mirror the abort conditions in vm_normal_page */ | |
1913 | if (vma->vm_flags & VM_MIXEDMAP) | |
1914 | return true; | |
1915 | if (pfn_t_devmap(pfn)) | |
1916 | return true; | |
1917 | if (pfn_t_special(pfn)) | |
1918 | return true; | |
1919 | if (is_zero_pfn(pfn_t_to_pfn(pfn))) | |
1920 | return true; | |
1921 | return false; | |
1922 | } | |
1923 | ||
b2770da6 RZ |
1924 | static int __vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr, |
1925 | pfn_t pfn, bool mkwrite) | |
423bad60 | 1926 | { |
87744ab3 DW |
1927 | pgprot_t pgprot = vma->vm_page_prot; |
1928 | ||
c11da42b | 1929 | BUG_ON(!vm_mixed_ok(vma, pfn)); |
e0dc0d8f | 1930 | |
423bad60 NP |
1931 | if (addr < vma->vm_start || addr >= vma->vm_end) |
1932 | return -EFAULT; | |
308a047c BP |
1933 | |
1934 | track_pfn_insert(vma, &pgprot, pfn); | |
e0dc0d8f | 1935 | |
0831b2a2 AK |
1936 | if (!pfn_modify_allowed(pfn_t_to_pfn(pfn), pgprot)) |
1937 | return -EACCES; | |
1938 | ||
423bad60 NP |
1939 | /* |
1940 | * If we don't have pte special, then we have to use the pfn_valid() | |
1941 | * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* | |
1942 | * refcount the page if pfn_valid is true (hence insert_page rather | |
62eede62 HD |
1943 | * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP |
1944 | * without pte special, it would there be refcounted as a normal page. | |
423bad60 | 1945 | */ |
03fc2da6 | 1946 | if (!HAVE_PTE_SPECIAL && !pfn_t_devmap(pfn) && pfn_t_valid(pfn)) { |
423bad60 NP |
1947 | struct page *page; |
1948 | ||
03fc2da6 DW |
1949 | /* |
1950 | * At this point we are committed to insert_page() | |
1951 | * regardless of whether the caller specified flags that | |
1952 | * result in pfn_t_has_page() == false. | |
1953 | */ | |
1954 | page = pfn_to_page(pfn_t_to_pfn(pfn)); | |
87744ab3 | 1955 | return insert_page(vma, addr, page, pgprot); |
423bad60 | 1956 | } |
b2770da6 RZ |
1957 | return insert_pfn(vma, addr, pfn, pgprot, mkwrite); |
1958 | } | |
1959 | ||
1960 | int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr, | |
1961 | pfn_t pfn) | |
1962 | { | |
1963 | return __vm_insert_mixed(vma, addr, pfn, false); | |
1964 | ||
e0dc0d8f | 1965 | } |
423bad60 | 1966 | EXPORT_SYMBOL(vm_insert_mixed); |
e0dc0d8f | 1967 | |
b2770da6 RZ |
1968 | int vm_insert_mixed_mkwrite(struct vm_area_struct *vma, unsigned long addr, |
1969 | pfn_t pfn) | |
1970 | { | |
1971 | return __vm_insert_mixed(vma, addr, pfn, true); | |
1972 | } | |
1973 | EXPORT_SYMBOL(vm_insert_mixed_mkwrite); | |
1974 | ||
1da177e4 LT |
1975 | /* |
1976 | * maps a range of physical memory into the requested pages. the old | |
1977 | * mappings are removed. any references to nonexistent pages results | |
1978 | * in null mappings (currently treated as "copy-on-access") | |
1979 | */ | |
1980 | static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, | |
1981 | unsigned long addr, unsigned long end, | |
1982 | unsigned long pfn, pgprot_t prot) | |
1983 | { | |
1984 | pte_t *pte; | |
c74df32c | 1985 | spinlock_t *ptl; |
0831b2a2 | 1986 | int err = 0; |
1da177e4 | 1987 | |
c74df32c | 1988 | pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); |
1da177e4 LT |
1989 | if (!pte) |
1990 | return -ENOMEM; | |
6606c3e0 | 1991 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
1992 | do { |
1993 | BUG_ON(!pte_none(*pte)); | |
0831b2a2 AK |
1994 | if (!pfn_modify_allowed(pfn, prot)) { |
1995 | err = -EACCES; | |
1996 | break; | |
1997 | } | |
7e675137 | 1998 | set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); |
1da177e4 LT |
1999 | pfn++; |
2000 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
6606c3e0 | 2001 | arch_leave_lazy_mmu_mode(); |
c74df32c | 2002 | pte_unmap_unlock(pte - 1, ptl); |
0831b2a2 | 2003 | return err; |
1da177e4 LT |
2004 | } |
2005 | ||
2006 | static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, | |
2007 | unsigned long addr, unsigned long end, | |
2008 | unsigned long pfn, pgprot_t prot) | |
2009 | { | |
2010 | pmd_t *pmd; | |
2011 | unsigned long next; | |
0831b2a2 | 2012 | int err; |
1da177e4 LT |
2013 | |
2014 | pfn -= addr >> PAGE_SHIFT; | |
2015 | pmd = pmd_alloc(mm, pud, addr); | |
2016 | if (!pmd) | |
2017 | return -ENOMEM; | |
f66055ab | 2018 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
1da177e4 LT |
2019 | do { |
2020 | next = pmd_addr_end(addr, end); | |
0831b2a2 AK |
2021 | err = remap_pte_range(mm, pmd, addr, next, |
2022 | pfn + (addr >> PAGE_SHIFT), prot); | |
2023 | if (err) | |
2024 | return err; | |
1da177e4 LT |
2025 | } while (pmd++, addr = next, addr != end); |
2026 | return 0; | |
2027 | } | |
2028 | ||
c2febafc | 2029 | static inline int remap_pud_range(struct mm_struct *mm, p4d_t *p4d, |
1da177e4 LT |
2030 | unsigned long addr, unsigned long end, |
2031 | unsigned long pfn, pgprot_t prot) | |
2032 | { | |
2033 | pud_t *pud; | |
2034 | unsigned long next; | |
0831b2a2 | 2035 | int err; |
1da177e4 LT |
2036 | |
2037 | pfn -= addr >> PAGE_SHIFT; | |
c2febafc | 2038 | pud = pud_alloc(mm, p4d, addr); |
1da177e4 LT |
2039 | if (!pud) |
2040 | return -ENOMEM; | |
2041 | do { | |
2042 | next = pud_addr_end(addr, end); | |
0831b2a2 AK |
2043 | err = remap_pmd_range(mm, pud, addr, next, |
2044 | pfn + (addr >> PAGE_SHIFT), prot); | |
2045 | if (err) | |
2046 | return err; | |
1da177e4 LT |
2047 | } while (pud++, addr = next, addr != end); |
2048 | return 0; | |
2049 | } | |
2050 | ||
c2febafc KS |
2051 | static inline int remap_p4d_range(struct mm_struct *mm, pgd_t *pgd, |
2052 | unsigned long addr, unsigned long end, | |
2053 | unsigned long pfn, pgprot_t prot) | |
2054 | { | |
2055 | p4d_t *p4d; | |
2056 | unsigned long next; | |
0831b2a2 | 2057 | int err; |
c2febafc KS |
2058 | |
2059 | pfn -= addr >> PAGE_SHIFT; | |
2060 | p4d = p4d_alloc(mm, pgd, addr); | |
2061 | if (!p4d) | |
2062 | return -ENOMEM; | |
2063 | do { | |
2064 | next = p4d_addr_end(addr, end); | |
0831b2a2 AK |
2065 | err = remap_pud_range(mm, p4d, addr, next, |
2066 | pfn + (addr >> PAGE_SHIFT), prot); | |
2067 | if (err) | |
2068 | return err; | |
c2febafc KS |
2069 | } while (p4d++, addr = next, addr != end); |
2070 | return 0; | |
2071 | } | |
2072 | ||
bfa5bf6d REB |
2073 | /** |
2074 | * remap_pfn_range - remap kernel memory to userspace | |
2075 | * @vma: user vma to map to | |
2076 | * @addr: target user address to start at | |
2077 | * @pfn: physical address of kernel memory | |
2078 | * @size: size of map area | |
2079 | * @prot: page protection flags for this mapping | |
2080 | * | |
2081 | * Note: this is only safe if the mm semaphore is held when called. | |
2082 | */ | |
1da177e4 LT |
2083 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, |
2084 | unsigned long pfn, unsigned long size, pgprot_t prot) | |
2085 | { | |
2086 | pgd_t *pgd; | |
2087 | unsigned long next; | |
2d15cab8 | 2088 | unsigned long end = addr + PAGE_ALIGN(size); |
1da177e4 | 2089 | struct mm_struct *mm = vma->vm_mm; |
d5957d2f | 2090 | unsigned long remap_pfn = pfn; |
1da177e4 LT |
2091 | int err; |
2092 | ||
2093 | /* | |
2094 | * Physically remapped pages are special. Tell the | |
2095 | * rest of the world about it: | |
2096 | * VM_IO tells people not to look at these pages | |
2097 | * (accesses can have side effects). | |
6aab341e LT |
2098 | * VM_PFNMAP tells the core MM that the base pages are just |
2099 | * raw PFN mappings, and do not have a "struct page" associated | |
2100 | * with them. | |
314e51b9 KK |
2101 | * VM_DONTEXPAND |
2102 | * Disable vma merging and expanding with mremap(). | |
2103 | * VM_DONTDUMP | |
2104 | * Omit vma from core dump, even when VM_IO turned off. | |
fb155c16 LT |
2105 | * |
2106 | * There's a horrible special case to handle copy-on-write | |
2107 | * behaviour that some programs depend on. We mark the "original" | |
2108 | * un-COW'ed pages by matching them up with "vma->vm_pgoff". | |
b3b9c293 | 2109 | * See vm_normal_page() for details. |
1da177e4 | 2110 | */ |
b3b9c293 KK |
2111 | if (is_cow_mapping(vma->vm_flags)) { |
2112 | if (addr != vma->vm_start || end != vma->vm_end) | |
2113 | return -EINVAL; | |
fb155c16 | 2114 | vma->vm_pgoff = pfn; |
b3b9c293 KK |
2115 | } |
2116 | ||
d5957d2f | 2117 | err = track_pfn_remap(vma, &prot, remap_pfn, addr, PAGE_ALIGN(size)); |
b3b9c293 | 2118 | if (err) |
3c8bb73a | 2119 | return -EINVAL; |
fb155c16 | 2120 | |
314e51b9 | 2121 | vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; |
1da177e4 LT |
2122 | |
2123 | BUG_ON(addr >= end); | |
2124 | pfn -= addr >> PAGE_SHIFT; | |
2125 | pgd = pgd_offset(mm, addr); | |
2126 | flush_cache_range(vma, addr, end); | |
1da177e4 LT |
2127 | do { |
2128 | next = pgd_addr_end(addr, end); | |
c2febafc | 2129 | err = remap_p4d_range(mm, pgd, addr, next, |
1da177e4 LT |
2130 | pfn + (addr >> PAGE_SHIFT), prot); |
2131 | if (err) | |
2132 | break; | |
2133 | } while (pgd++, addr = next, addr != end); | |
2ab64037 | 2134 | |
2135 | if (err) | |
d5957d2f | 2136 | untrack_pfn(vma, remap_pfn, PAGE_ALIGN(size)); |
2ab64037 | 2137 | |
1da177e4 LT |
2138 | return err; |
2139 | } | |
2140 | EXPORT_SYMBOL(remap_pfn_range); | |
2141 | ||
b4cbb197 LT |
2142 | /** |
2143 | * vm_iomap_memory - remap memory to userspace | |
2144 | * @vma: user vma to map to | |
2145 | * @start: start of area | |
2146 | * @len: size of area | |
2147 | * | |
2148 | * This is a simplified io_remap_pfn_range() for common driver use. The | |
2149 | * driver just needs to give us the physical memory range to be mapped, | |
2150 | * we'll figure out the rest from the vma information. | |
2151 | * | |
2152 | * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get | |
2153 | * whatever write-combining details or similar. | |
2154 | */ | |
2155 | int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) | |
2156 | { | |
2157 | unsigned long vm_len, pfn, pages; | |
2158 | ||
2159 | /* Check that the physical memory area passed in looks valid */ | |
2160 | if (start + len < start) | |
2161 | return -EINVAL; | |
2162 | /* | |
2163 | * You *really* shouldn't map things that aren't page-aligned, | |
2164 | * but we've historically allowed it because IO memory might | |
2165 | * just have smaller alignment. | |
2166 | */ | |
2167 | len += start & ~PAGE_MASK; | |
2168 | pfn = start >> PAGE_SHIFT; | |
2169 | pages = (len + ~PAGE_MASK) >> PAGE_SHIFT; | |
2170 | if (pfn + pages < pfn) | |
2171 | return -EINVAL; | |
2172 | ||
2173 | /* We start the mapping 'vm_pgoff' pages into the area */ | |
2174 | if (vma->vm_pgoff > pages) | |
2175 | return -EINVAL; | |
2176 | pfn += vma->vm_pgoff; | |
2177 | pages -= vma->vm_pgoff; | |
2178 | ||
2179 | /* Can we fit all of the mapping? */ | |
2180 | vm_len = vma->vm_end - vma->vm_start; | |
2181 | if (vm_len >> PAGE_SHIFT > pages) | |
2182 | return -EINVAL; | |
2183 | ||
2184 | /* Ok, let it rip */ | |
2185 | return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); | |
2186 | } | |
2187 | EXPORT_SYMBOL(vm_iomap_memory); | |
2188 | ||
aee16b3c JF |
2189 | static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, |
2190 | unsigned long addr, unsigned long end, | |
2191 | pte_fn_t fn, void *data) | |
2192 | { | |
2193 | pte_t *pte; | |
2194 | int err; | |
2f569afd | 2195 | pgtable_t token; |
94909914 | 2196 | spinlock_t *uninitialized_var(ptl); |
aee16b3c JF |
2197 | |
2198 | pte = (mm == &init_mm) ? | |
2199 | pte_alloc_kernel(pmd, addr) : | |
2200 | pte_alloc_map_lock(mm, pmd, addr, &ptl); | |
2201 | if (!pte) | |
2202 | return -ENOMEM; | |
2203 | ||
2204 | BUG_ON(pmd_huge(*pmd)); | |
2205 | ||
38e0edb1 JF |
2206 | arch_enter_lazy_mmu_mode(); |
2207 | ||
2f569afd | 2208 | token = pmd_pgtable(*pmd); |
aee16b3c JF |
2209 | |
2210 | do { | |
c36987e2 | 2211 | err = fn(pte++, token, addr, data); |
aee16b3c JF |
2212 | if (err) |
2213 | break; | |
c36987e2 | 2214 | } while (addr += PAGE_SIZE, addr != end); |
aee16b3c | 2215 | |
38e0edb1 JF |
2216 | arch_leave_lazy_mmu_mode(); |
2217 | ||
aee16b3c JF |
2218 | if (mm != &init_mm) |
2219 | pte_unmap_unlock(pte-1, ptl); | |
2220 | return err; | |
2221 | } | |
2222 | ||
2223 | static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, | |
2224 | unsigned long addr, unsigned long end, | |
2225 | pte_fn_t fn, void *data) | |
2226 | { | |
2227 | pmd_t *pmd; | |
2228 | unsigned long next; | |
2229 | int err; | |
2230 | ||
ceb86879 AK |
2231 | BUG_ON(pud_huge(*pud)); |
2232 | ||
aee16b3c JF |
2233 | pmd = pmd_alloc(mm, pud, addr); |
2234 | if (!pmd) | |
2235 | return -ENOMEM; | |
2236 | do { | |
2237 | next = pmd_addr_end(addr, end); | |
2238 | err = apply_to_pte_range(mm, pmd, addr, next, fn, data); | |
2239 | if (err) | |
2240 | break; | |
2241 | } while (pmd++, addr = next, addr != end); | |
2242 | return err; | |
2243 | } | |
2244 | ||
c2febafc | 2245 | static int apply_to_pud_range(struct mm_struct *mm, p4d_t *p4d, |
aee16b3c JF |
2246 | unsigned long addr, unsigned long end, |
2247 | pte_fn_t fn, void *data) | |
2248 | { | |
2249 | pud_t *pud; | |
2250 | unsigned long next; | |
2251 | int err; | |
2252 | ||
c2febafc | 2253 | pud = pud_alloc(mm, p4d, addr); |
aee16b3c JF |
2254 | if (!pud) |
2255 | return -ENOMEM; | |
2256 | do { | |
2257 | next = pud_addr_end(addr, end); | |
2258 | err = apply_to_pmd_range(mm, pud, addr, next, fn, data); | |
2259 | if (err) | |
2260 | break; | |
2261 | } while (pud++, addr = next, addr != end); | |
2262 | return err; | |
2263 | } | |
2264 | ||
c2febafc KS |
2265 | static int apply_to_p4d_range(struct mm_struct *mm, pgd_t *pgd, |
2266 | unsigned long addr, unsigned long end, | |
2267 | pte_fn_t fn, void *data) | |
2268 | { | |
2269 | p4d_t *p4d; | |
2270 | unsigned long next; | |
2271 | int err; | |
2272 | ||
2273 | p4d = p4d_alloc(mm, pgd, addr); | |
2274 | if (!p4d) | |
2275 | return -ENOMEM; | |
2276 | do { | |
2277 | next = p4d_addr_end(addr, end); | |
2278 | err = apply_to_pud_range(mm, p4d, addr, next, fn, data); | |
2279 | if (err) | |
2280 | break; | |
2281 | } while (p4d++, addr = next, addr != end); | |
2282 | return err; | |
2283 | } | |
2284 | ||
aee16b3c JF |
2285 | /* |
2286 | * Scan a region of virtual memory, filling in page tables as necessary | |
2287 | * and calling a provided function on each leaf page table. | |
2288 | */ | |
2289 | int apply_to_page_range(struct mm_struct *mm, unsigned long addr, | |
2290 | unsigned long size, pte_fn_t fn, void *data) | |
2291 | { | |
2292 | pgd_t *pgd; | |
2293 | unsigned long next; | |
57250a5b | 2294 | unsigned long end = addr + size; |
aee16b3c JF |
2295 | int err; |
2296 | ||
9cb65bc3 MP |
2297 | if (WARN_ON(addr >= end)) |
2298 | return -EINVAL; | |
2299 | ||
aee16b3c JF |
2300 | pgd = pgd_offset(mm, addr); |
2301 | do { | |
2302 | next = pgd_addr_end(addr, end); | |
c2febafc | 2303 | err = apply_to_p4d_range(mm, pgd, addr, next, fn, data); |
aee16b3c JF |
2304 | if (err) |
2305 | break; | |
2306 | } while (pgd++, addr = next, addr != end); | |
57250a5b | 2307 | |
aee16b3c JF |
2308 | return err; |
2309 | } | |
2310 | EXPORT_SYMBOL_GPL(apply_to_page_range); | |
2311 | ||
8f4e2101 | 2312 | /* |
9b4bdd2f KS |
2313 | * handle_pte_fault chooses page fault handler according to an entry which was |
2314 | * read non-atomically. Before making any commitment, on those architectures | |
2315 | * or configurations (e.g. i386 with PAE) which might give a mix of unmatched | |
2316 | * parts, do_swap_page must check under lock before unmapping the pte and | |
2317 | * proceeding (but do_wp_page is only called after already making such a check; | |
a335b2e1 | 2318 | * and do_anonymous_page can safely check later on). |
8f4e2101 | 2319 | */ |
4c21e2f2 | 2320 | static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd, |
8f4e2101 HD |
2321 | pte_t *page_table, pte_t orig_pte) |
2322 | { | |
2323 | int same = 1; | |
2324 | #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT) | |
2325 | if (sizeof(pte_t) > sizeof(unsigned long)) { | |
4c21e2f2 HD |
2326 | spinlock_t *ptl = pte_lockptr(mm, pmd); |
2327 | spin_lock(ptl); | |
8f4e2101 | 2328 | same = pte_same(*page_table, orig_pte); |
4c21e2f2 | 2329 | spin_unlock(ptl); |
8f4e2101 HD |
2330 | } |
2331 | #endif | |
2332 | pte_unmap(page_table); | |
2333 | return same; | |
2334 | } | |
2335 | ||
9de455b2 | 2336 | static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma) |
6aab341e | 2337 | { |
0abdd7a8 DW |
2338 | debug_dma_assert_idle(src); |
2339 | ||
6aab341e LT |
2340 | /* |
2341 | * If the source page was a PFN mapping, we don't have | |
2342 | * a "struct page" for it. We do a best-effort copy by | |
2343 | * just copying from the original user address. If that | |
2344 | * fails, we just zero-fill it. Live with it. | |
2345 | */ | |
2346 | if (unlikely(!src)) { | |
9b04c5fe | 2347 | void *kaddr = kmap_atomic(dst); |
5d2a2dbb LT |
2348 | void __user *uaddr = (void __user *)(va & PAGE_MASK); |
2349 | ||
2350 | /* | |
2351 | * This really shouldn't fail, because the page is there | |
2352 | * in the page tables. But it might just be unreadable, | |
2353 | * in which case we just give up and fill the result with | |
2354 | * zeroes. | |
2355 | */ | |
2356 | if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) | |
3ecb01df | 2357 | clear_page(kaddr); |
9b04c5fe | 2358 | kunmap_atomic(kaddr); |
c4ec7b0d | 2359 | flush_dcache_page(dst); |
0ed361de NP |
2360 | } else |
2361 | copy_user_highpage(dst, src, va, vma); | |
6aab341e LT |
2362 | } |
2363 | ||
c20cd45e MH |
2364 | static gfp_t __get_fault_gfp_mask(struct vm_area_struct *vma) |
2365 | { | |
2366 | struct file *vm_file = vma->vm_file; | |
2367 | ||
2368 | if (vm_file) | |
2369 | return mapping_gfp_mask(vm_file->f_mapping) | __GFP_FS | __GFP_IO; | |
2370 | ||
2371 | /* | |
2372 | * Special mappings (e.g. VDSO) do not have any file so fake | |
2373 | * a default GFP_KERNEL for them. | |
2374 | */ | |
2375 | return GFP_KERNEL; | |
2376 | } | |
2377 | ||
fb09a464 KS |
2378 | /* |
2379 | * Notify the address space that the page is about to become writable so that | |
2380 | * it can prohibit this or wait for the page to get into an appropriate state. | |
2381 | * | |
2382 | * We do this without the lock held, so that it can sleep if it needs to. | |
2383 | */ | |
38b8cb7f | 2384 | static int do_page_mkwrite(struct vm_fault *vmf) |
fb09a464 | 2385 | { |
fb09a464 | 2386 | int ret; |
38b8cb7f JK |
2387 | struct page *page = vmf->page; |
2388 | unsigned int old_flags = vmf->flags; | |
fb09a464 | 2389 | |
38b8cb7f | 2390 | vmf->flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; |
fb09a464 | 2391 | |
11bac800 | 2392 | ret = vmf->vma->vm_ops->page_mkwrite(vmf); |
38b8cb7f JK |
2393 | /* Restore original flags so that caller is not surprised */ |
2394 | vmf->flags = old_flags; | |
fb09a464 KS |
2395 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) |
2396 | return ret; | |
2397 | if (unlikely(!(ret & VM_FAULT_LOCKED))) { | |
2398 | lock_page(page); | |
2399 | if (!page->mapping) { | |
2400 | unlock_page(page); | |
2401 | return 0; /* retry */ | |
2402 | } | |
2403 | ret |= VM_FAULT_LOCKED; | |
2404 | } else | |
2405 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
2406 | return ret; | |
2407 | } | |
2408 | ||
97ba0c2b JK |
2409 | /* |
2410 | * Handle dirtying of a page in shared file mapping on a write fault. | |
2411 | * | |
2412 | * The function expects the page to be locked and unlocks it. | |
2413 | */ | |
2414 | static void fault_dirty_shared_page(struct vm_area_struct *vma, | |
2415 | struct page *page) | |
2416 | { | |
2417 | struct address_space *mapping; | |
2418 | bool dirtied; | |
2419 | bool page_mkwrite = vma->vm_ops && vma->vm_ops->page_mkwrite; | |
2420 | ||
2421 | dirtied = set_page_dirty(page); | |
2422 | VM_BUG_ON_PAGE(PageAnon(page), page); | |
2423 | /* | |
2424 | * Take a local copy of the address_space - page.mapping may be zeroed | |
2425 | * by truncate after unlock_page(). The address_space itself remains | |
2426 | * pinned by vma->vm_file's reference. We rely on unlock_page()'s | |
2427 | * release semantics to prevent the compiler from undoing this copying. | |
2428 | */ | |
2429 | mapping = page_rmapping(page); | |
2430 | unlock_page(page); | |
2431 | ||
2432 | if ((dirtied || page_mkwrite) && mapping) { | |
2433 | /* | |
2434 | * Some device drivers do not set page.mapping | |
2435 | * but still dirty their pages | |
2436 | */ | |
2437 | balance_dirty_pages_ratelimited(mapping); | |
2438 | } | |
2439 | ||
2440 | if (!page_mkwrite) | |
2441 | file_update_time(vma->vm_file); | |
2442 | } | |
2443 | ||
4e047f89 SR |
2444 | /* |
2445 | * Handle write page faults for pages that can be reused in the current vma | |
2446 | * | |
2447 | * This can happen either due to the mapping being with the VM_SHARED flag, | |
2448 | * or due to us being the last reference standing to the page. In either | |
2449 | * case, all we need to do here is to mark the page as writable and update | |
2450 | * any related book-keeping. | |
2451 | */ | |
997dd98d | 2452 | static inline void wp_page_reuse(struct vm_fault *vmf) |
82b0f8c3 | 2453 | __releases(vmf->ptl) |
4e047f89 | 2454 | { |
82b0f8c3 | 2455 | struct vm_area_struct *vma = vmf->vma; |
a41b70d6 | 2456 | struct page *page = vmf->page; |
4e047f89 SR |
2457 | pte_t entry; |
2458 | /* | |
2459 | * Clear the pages cpupid information as the existing | |
2460 | * information potentially belongs to a now completely | |
2461 | * unrelated process. | |
2462 | */ | |
2463 | if (page) | |
2464 | page_cpupid_xchg_last(page, (1 << LAST_CPUPID_SHIFT) - 1); | |
2465 | ||
2994302b JK |
2466 | flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); |
2467 | entry = pte_mkyoung(vmf->orig_pte); | |
4e047f89 | 2468 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
82b0f8c3 JK |
2469 | if (ptep_set_access_flags(vma, vmf->address, vmf->pte, entry, 1)) |
2470 | update_mmu_cache(vma, vmf->address, vmf->pte); | |
2471 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
4e047f89 SR |
2472 | } |
2473 | ||
2f38ab2c SR |
2474 | /* |
2475 | * Handle the case of a page which we actually need to copy to a new page. | |
2476 | * | |
2477 | * Called with mmap_sem locked and the old page referenced, but | |
2478 | * without the ptl held. | |
2479 | * | |
2480 | * High level logic flow: | |
2481 | * | |
2482 | * - Allocate a page, copy the content of the old page to the new one. | |
2483 | * - Handle book keeping and accounting - cgroups, mmu-notifiers, etc. | |
2484 | * - Take the PTL. If the pte changed, bail out and release the allocated page | |
2485 | * - If the pte is still the way we remember it, update the page table and all | |
2486 | * relevant references. This includes dropping the reference the page-table | |
2487 | * held to the old page, as well as updating the rmap. | |
2488 | * - In any case, unlock the PTL and drop the reference we took to the old page. | |
2489 | */ | |
a41b70d6 | 2490 | static int wp_page_copy(struct vm_fault *vmf) |
2f38ab2c | 2491 | { |
82b0f8c3 | 2492 | struct vm_area_struct *vma = vmf->vma; |
bae473a4 | 2493 | struct mm_struct *mm = vma->vm_mm; |
a41b70d6 | 2494 | struct page *old_page = vmf->page; |
2f38ab2c | 2495 | struct page *new_page = NULL; |
2f38ab2c SR |
2496 | pte_t entry; |
2497 | int page_copied = 0; | |
82b0f8c3 | 2498 | const unsigned long mmun_start = vmf->address & PAGE_MASK; |
bae473a4 | 2499 | const unsigned long mmun_end = mmun_start + PAGE_SIZE; |
2f38ab2c SR |
2500 | struct mem_cgroup *memcg; |
2501 | ||
2502 | if (unlikely(anon_vma_prepare(vma))) | |
2503 | goto oom; | |
2504 | ||
2994302b | 2505 | if (is_zero_pfn(pte_pfn(vmf->orig_pte))) { |
82b0f8c3 JK |
2506 | new_page = alloc_zeroed_user_highpage_movable(vma, |
2507 | vmf->address); | |
2f38ab2c SR |
2508 | if (!new_page) |
2509 | goto oom; | |
2510 | } else { | |
bae473a4 | 2511 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, |
82b0f8c3 | 2512 | vmf->address); |
2f38ab2c SR |
2513 | if (!new_page) |
2514 | goto oom; | |
82b0f8c3 | 2515 | cow_user_page(new_page, old_page, vmf->address, vma); |
2f38ab2c | 2516 | } |
2f38ab2c | 2517 | |
f627c2f5 | 2518 | if (mem_cgroup_try_charge(new_page, mm, GFP_KERNEL, &memcg, false)) |
2f38ab2c SR |
2519 | goto oom_free_new; |
2520 | ||
eb3c24f3 MG |
2521 | __SetPageUptodate(new_page); |
2522 | ||
2f38ab2c SR |
2523 | mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); |
2524 | ||
2525 | /* | |
2526 | * Re-check the pte - we dropped the lock | |
2527 | */ | |
82b0f8c3 | 2528 | vmf->pte = pte_offset_map_lock(mm, vmf->pmd, vmf->address, &vmf->ptl); |
2994302b | 2529 | if (likely(pte_same(*vmf->pte, vmf->orig_pte))) { |
2f38ab2c SR |
2530 | if (old_page) { |
2531 | if (!PageAnon(old_page)) { | |
eca56ff9 JM |
2532 | dec_mm_counter_fast(mm, |
2533 | mm_counter_file(old_page)); | |
2f38ab2c SR |
2534 | inc_mm_counter_fast(mm, MM_ANONPAGES); |
2535 | } | |
2536 | } else { | |
2537 | inc_mm_counter_fast(mm, MM_ANONPAGES); | |
2538 | } | |
2994302b | 2539 | flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); |
2f38ab2c SR |
2540 | entry = mk_pte(new_page, vma->vm_page_prot); |
2541 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
2542 | /* | |
2543 | * Clear the pte entry and flush it first, before updating the | |
2544 | * pte with the new entry. This will avoid a race condition | |
2545 | * seen in the presence of one thread doing SMC and another | |
2546 | * thread doing COW. | |
2547 | */ | |
82b0f8c3 JK |
2548 | ptep_clear_flush_notify(vma, vmf->address, vmf->pte); |
2549 | page_add_new_anon_rmap(new_page, vma, vmf->address, false); | |
f627c2f5 | 2550 | mem_cgroup_commit_charge(new_page, memcg, false, false); |
2f38ab2c SR |
2551 | lru_cache_add_active_or_unevictable(new_page, vma); |
2552 | /* | |
2553 | * We call the notify macro here because, when using secondary | |
2554 | * mmu page tables (such as kvm shadow page tables), we want the | |
2555 | * new page to be mapped directly into the secondary page table. | |
2556 | */ | |
82b0f8c3 JK |
2557 | set_pte_at_notify(mm, vmf->address, vmf->pte, entry); |
2558 | update_mmu_cache(vma, vmf->address, vmf->pte); | |
2f38ab2c SR |
2559 | if (old_page) { |
2560 | /* | |
2561 | * Only after switching the pte to the new page may | |
2562 | * we remove the mapcount here. Otherwise another | |
2563 | * process may come and find the rmap count decremented | |
2564 | * before the pte is switched to the new page, and | |
2565 | * "reuse" the old page writing into it while our pte | |
2566 | * here still points into it and can be read by other | |
2567 | * threads. | |
2568 | * | |
2569 | * The critical issue is to order this | |
2570 | * page_remove_rmap with the ptp_clear_flush above. | |
2571 | * Those stores are ordered by (if nothing else,) | |
2572 | * the barrier present in the atomic_add_negative | |
2573 | * in page_remove_rmap. | |
2574 | * | |
2575 | * Then the TLB flush in ptep_clear_flush ensures that | |
2576 | * no process can access the old page before the | |
2577 | * decremented mapcount is visible. And the old page | |
2578 | * cannot be reused until after the decremented | |
2579 | * mapcount is visible. So transitively, TLBs to | |
2580 | * old page will be flushed before it can be reused. | |
2581 | */ | |
d281ee61 | 2582 | page_remove_rmap(old_page, false); |
2f38ab2c SR |
2583 | } |
2584 | ||
2585 | /* Free the old page.. */ | |
2586 | new_page = old_page; | |
2587 | page_copied = 1; | |
2588 | } else { | |
f627c2f5 | 2589 | mem_cgroup_cancel_charge(new_page, memcg, false); |
2f38ab2c SR |
2590 | } |
2591 | ||
2592 | if (new_page) | |
09cbfeaf | 2593 | put_page(new_page); |
2f38ab2c | 2594 | |
82b0f8c3 | 2595 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
4645b9fe JG |
2596 | /* |
2597 | * No need to double call mmu_notifier->invalidate_range() callback as | |
2598 | * the above ptep_clear_flush_notify() did already call it. | |
2599 | */ | |
2600 | mmu_notifier_invalidate_range_only_end(mm, mmun_start, mmun_end); | |
2f38ab2c SR |
2601 | if (old_page) { |
2602 | /* | |
2603 | * Don't let another task, with possibly unlocked vma, | |
2604 | * keep the mlocked page. | |
2605 | */ | |
2606 | if (page_copied && (vma->vm_flags & VM_LOCKED)) { | |
2607 | lock_page(old_page); /* LRU manipulation */ | |
e90309c9 KS |
2608 | if (PageMlocked(old_page)) |
2609 | munlock_vma_page(old_page); | |
2f38ab2c SR |
2610 | unlock_page(old_page); |
2611 | } | |
09cbfeaf | 2612 | put_page(old_page); |
2f38ab2c SR |
2613 | } |
2614 | return page_copied ? VM_FAULT_WRITE : 0; | |
2615 | oom_free_new: | |
09cbfeaf | 2616 | put_page(new_page); |
2f38ab2c SR |
2617 | oom: |
2618 | if (old_page) | |
09cbfeaf | 2619 | put_page(old_page); |
2f38ab2c SR |
2620 | return VM_FAULT_OOM; |
2621 | } | |
2622 | ||
66a6197c JK |
2623 | /** |
2624 | * finish_mkwrite_fault - finish page fault for a shared mapping, making PTE | |
2625 | * writeable once the page is prepared | |
2626 | * | |
2627 | * @vmf: structure describing the fault | |
2628 | * | |
2629 | * This function handles all that is needed to finish a write page fault in a | |
2630 | * shared mapping due to PTE being read-only once the mapped page is prepared. | |
2631 | * It handles locking of PTE and modifying it. The function returns | |
2632 | * VM_FAULT_WRITE on success, 0 when PTE got changed before we acquired PTE | |
2633 | * lock. | |
2634 | * | |
2635 | * The function expects the page to be locked or other protection against | |
2636 | * concurrent faults / writeback (such as DAX radix tree locks). | |
2637 | */ | |
2638 | int finish_mkwrite_fault(struct vm_fault *vmf) | |
2639 | { | |
2640 | WARN_ON_ONCE(!(vmf->vma->vm_flags & VM_SHARED)); | |
2641 | vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, vmf->address, | |
2642 | &vmf->ptl); | |
2643 | /* | |
2644 | * We might have raced with another page fault while we released the | |
2645 | * pte_offset_map_lock. | |
2646 | */ | |
2647 | if (!pte_same(*vmf->pte, vmf->orig_pte)) { | |
2648 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
a19e2553 | 2649 | return VM_FAULT_NOPAGE; |
66a6197c JK |
2650 | } |
2651 | wp_page_reuse(vmf); | |
a19e2553 | 2652 | return 0; |
66a6197c JK |
2653 | } |
2654 | ||
dd906184 BH |
2655 | /* |
2656 | * Handle write page faults for VM_MIXEDMAP or VM_PFNMAP for a VM_SHARED | |
2657 | * mapping | |
2658 | */ | |
2994302b | 2659 | static int wp_pfn_shared(struct vm_fault *vmf) |
dd906184 | 2660 | { |
82b0f8c3 | 2661 | struct vm_area_struct *vma = vmf->vma; |
bae473a4 | 2662 | |
dd906184 | 2663 | if (vma->vm_ops && vma->vm_ops->pfn_mkwrite) { |
dd906184 BH |
2664 | int ret; |
2665 | ||
82b0f8c3 | 2666 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
fe82221f | 2667 | vmf->flags |= FAULT_FLAG_MKWRITE; |
11bac800 | 2668 | ret = vma->vm_ops->pfn_mkwrite(vmf); |
2f89dc12 | 2669 | if (ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)) |
dd906184 | 2670 | return ret; |
66a6197c | 2671 | return finish_mkwrite_fault(vmf); |
dd906184 | 2672 | } |
997dd98d JK |
2673 | wp_page_reuse(vmf); |
2674 | return VM_FAULT_WRITE; | |
dd906184 BH |
2675 | } |
2676 | ||
a41b70d6 | 2677 | static int wp_page_shared(struct vm_fault *vmf) |
82b0f8c3 | 2678 | __releases(vmf->ptl) |
93e478d4 | 2679 | { |
82b0f8c3 | 2680 | struct vm_area_struct *vma = vmf->vma; |
93e478d4 | 2681 | |
a41b70d6 | 2682 | get_page(vmf->page); |
93e478d4 | 2683 | |
93e478d4 SR |
2684 | if (vma->vm_ops && vma->vm_ops->page_mkwrite) { |
2685 | int tmp; | |
2686 | ||
82b0f8c3 | 2687 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
38b8cb7f | 2688 | tmp = do_page_mkwrite(vmf); |
93e478d4 SR |
2689 | if (unlikely(!tmp || (tmp & |
2690 | (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { | |
a41b70d6 | 2691 | put_page(vmf->page); |
93e478d4 SR |
2692 | return tmp; |
2693 | } | |
66a6197c | 2694 | tmp = finish_mkwrite_fault(vmf); |
a19e2553 | 2695 | if (unlikely(tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { |
a41b70d6 | 2696 | unlock_page(vmf->page); |
a41b70d6 | 2697 | put_page(vmf->page); |
66a6197c | 2698 | return tmp; |
93e478d4 | 2699 | } |
66a6197c JK |
2700 | } else { |
2701 | wp_page_reuse(vmf); | |
997dd98d | 2702 | lock_page(vmf->page); |
93e478d4 | 2703 | } |
997dd98d JK |
2704 | fault_dirty_shared_page(vma, vmf->page); |
2705 | put_page(vmf->page); | |
93e478d4 | 2706 | |
997dd98d | 2707 | return VM_FAULT_WRITE; |
93e478d4 SR |
2708 | } |
2709 | ||
1da177e4 LT |
2710 | /* |
2711 | * This routine handles present pages, when users try to write | |
2712 | * to a shared page. It is done by copying the page to a new address | |
2713 | * and decrementing the shared-page counter for the old page. | |
2714 | * | |
1da177e4 LT |
2715 | * Note that this routine assumes that the protection checks have been |
2716 | * done by the caller (the low-level page fault routine in most cases). | |
2717 | * Thus we can safely just mark it writable once we've done any necessary | |
2718 | * COW. | |
2719 | * | |
2720 | * We also mark the page dirty at this point even though the page will | |
2721 | * change only once the write actually happens. This avoids a few races, | |
2722 | * and potentially makes it more efficient. | |
2723 | * | |
8f4e2101 HD |
2724 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2725 | * but allow concurrent faults), with pte both mapped and locked. | |
2726 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2727 | */ |
2994302b | 2728 | static int do_wp_page(struct vm_fault *vmf) |
82b0f8c3 | 2729 | __releases(vmf->ptl) |
1da177e4 | 2730 | { |
82b0f8c3 | 2731 | struct vm_area_struct *vma = vmf->vma; |
1da177e4 | 2732 | |
a41b70d6 JK |
2733 | vmf->page = vm_normal_page(vma, vmf->address, vmf->orig_pte); |
2734 | if (!vmf->page) { | |
251b97f5 | 2735 | /* |
64e45507 PF |
2736 | * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a |
2737 | * VM_PFNMAP VMA. | |
251b97f5 PZ |
2738 | * |
2739 | * We should not cow pages in a shared writeable mapping. | |
dd906184 | 2740 | * Just mark the pages writable and/or call ops->pfn_mkwrite. |
251b97f5 PZ |
2741 | */ |
2742 | if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) == | |
2743 | (VM_WRITE|VM_SHARED)) | |
2994302b | 2744 | return wp_pfn_shared(vmf); |
2f38ab2c | 2745 | |
82b0f8c3 | 2746 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 2747 | return wp_page_copy(vmf); |
251b97f5 | 2748 | } |
1da177e4 | 2749 | |
d08b3851 | 2750 | /* |
ee6a6457 PZ |
2751 | * Take out anonymous pages first, anonymous shared vmas are |
2752 | * not dirty accountable. | |
d08b3851 | 2753 | */ |
a41b70d6 | 2754 | if (PageAnon(vmf->page) && !PageKsm(vmf->page)) { |
ba3c4ce6 | 2755 | int total_map_swapcount; |
a41b70d6 JK |
2756 | if (!trylock_page(vmf->page)) { |
2757 | get_page(vmf->page); | |
82b0f8c3 | 2758 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 2759 | lock_page(vmf->page); |
82b0f8c3 JK |
2760 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
2761 | vmf->address, &vmf->ptl); | |
2994302b | 2762 | if (!pte_same(*vmf->pte, vmf->orig_pte)) { |
a41b70d6 | 2763 | unlock_page(vmf->page); |
82b0f8c3 | 2764 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 2765 | put_page(vmf->page); |
28766805 | 2766 | return 0; |
ab967d86 | 2767 | } |
a41b70d6 | 2768 | put_page(vmf->page); |
ee6a6457 | 2769 | } |
ba3c4ce6 HY |
2770 | if (reuse_swap_page(vmf->page, &total_map_swapcount)) { |
2771 | if (total_map_swapcount == 1) { | |
6d0a07ed AA |
2772 | /* |
2773 | * The page is all ours. Move it to | |
2774 | * our anon_vma so the rmap code will | |
2775 | * not search our parent or siblings. | |
2776 | * Protected against the rmap code by | |
2777 | * the page lock. | |
2778 | */ | |
a41b70d6 | 2779 | page_move_anon_rmap(vmf->page, vma); |
6d0a07ed | 2780 | } |
a41b70d6 | 2781 | unlock_page(vmf->page); |
997dd98d JK |
2782 | wp_page_reuse(vmf); |
2783 | return VM_FAULT_WRITE; | |
b009c024 | 2784 | } |
a41b70d6 | 2785 | unlock_page(vmf->page); |
ee6a6457 | 2786 | } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) == |
d08b3851 | 2787 | (VM_WRITE|VM_SHARED))) { |
a41b70d6 | 2788 | return wp_page_shared(vmf); |
1da177e4 | 2789 | } |
1da177e4 LT |
2790 | |
2791 | /* | |
2792 | * Ok, we need to copy. Oh, well.. | |
2793 | */ | |
a41b70d6 | 2794 | get_page(vmf->page); |
28766805 | 2795 | |
82b0f8c3 | 2796 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 2797 | return wp_page_copy(vmf); |
1da177e4 LT |
2798 | } |
2799 | ||
97a89413 | 2800 | static void unmap_mapping_range_vma(struct vm_area_struct *vma, |
1da177e4 LT |
2801 | unsigned long start_addr, unsigned long end_addr, |
2802 | struct zap_details *details) | |
2803 | { | |
f5cc4eef | 2804 | zap_page_range_single(vma, start_addr, end_addr - start_addr, details); |
1da177e4 LT |
2805 | } |
2806 | ||
f808c13f | 2807 | static inline void unmap_mapping_range_tree(struct rb_root_cached *root, |
1da177e4 LT |
2808 | struct zap_details *details) |
2809 | { | |
2810 | struct vm_area_struct *vma; | |
1da177e4 LT |
2811 | pgoff_t vba, vea, zba, zea; |
2812 | ||
6b2dbba8 | 2813 | vma_interval_tree_foreach(vma, root, |
1da177e4 | 2814 | details->first_index, details->last_index) { |
1da177e4 LT |
2815 | |
2816 | vba = vma->vm_pgoff; | |
d6e93217 | 2817 | vea = vba + vma_pages(vma) - 1; |
1da177e4 LT |
2818 | zba = details->first_index; |
2819 | if (zba < vba) | |
2820 | zba = vba; | |
2821 | zea = details->last_index; | |
2822 | if (zea > vea) | |
2823 | zea = vea; | |
2824 | ||
97a89413 | 2825 | unmap_mapping_range_vma(vma, |
1da177e4 LT |
2826 | ((zba - vba) << PAGE_SHIFT) + vma->vm_start, |
2827 | ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, | |
97a89413 | 2828 | details); |
1da177e4 LT |
2829 | } |
2830 | } | |
2831 | ||
1da177e4 | 2832 | /** |
8a5f14a2 KS |
2833 | * unmap_mapping_range - unmap the portion of all mmaps in the specified |
2834 | * address_space corresponding to the specified page range in the underlying | |
2835 | * file. | |
2836 | * | |
3d41088f | 2837 | * @mapping: the address space containing mmaps to be unmapped. |
1da177e4 LT |
2838 | * @holebegin: byte in first page to unmap, relative to the start of |
2839 | * the underlying file. This will be rounded down to a PAGE_SIZE | |
25d9e2d1 | 2840 | * boundary. Note that this is different from truncate_pagecache(), which |
1da177e4 LT |
2841 | * must keep the partial page. In contrast, we must get rid of |
2842 | * partial pages. | |
2843 | * @holelen: size of prospective hole in bytes. This will be rounded | |
2844 | * up to a PAGE_SIZE boundary. A holelen of zero truncates to the | |
2845 | * end of the file. | |
2846 | * @even_cows: 1 when truncating a file, unmap even private COWed pages; | |
2847 | * but 0 when invalidating pagecache, don't throw away private data. | |
2848 | */ | |
2849 | void unmap_mapping_range(struct address_space *mapping, | |
2850 | loff_t const holebegin, loff_t const holelen, int even_cows) | |
2851 | { | |
aac45363 | 2852 | struct zap_details details = { }; |
1da177e4 LT |
2853 | pgoff_t hba = holebegin >> PAGE_SHIFT; |
2854 | pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
2855 | ||
2856 | /* Check for overflow. */ | |
2857 | if (sizeof(holelen) > sizeof(hlen)) { | |
2858 | long long holeend = | |
2859 | (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
2860 | if (holeend & ~(long long)ULONG_MAX) | |
2861 | hlen = ULONG_MAX - hba + 1; | |
2862 | } | |
2863 | ||
166f61b9 | 2864 | details.check_mapping = even_cows ? NULL : mapping; |
1da177e4 LT |
2865 | details.first_index = hba; |
2866 | details.last_index = hba + hlen - 1; | |
2867 | if (details.last_index < details.first_index) | |
2868 | details.last_index = ULONG_MAX; | |
1da177e4 | 2869 | |
46c043ed | 2870 | i_mmap_lock_write(mapping); |
f808c13f | 2871 | if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))) |
1da177e4 | 2872 | unmap_mapping_range_tree(&mapping->i_mmap, &details); |
46c043ed | 2873 | i_mmap_unlock_write(mapping); |
1da177e4 LT |
2874 | } |
2875 | EXPORT_SYMBOL(unmap_mapping_range); | |
2876 | ||
1da177e4 | 2877 | /* |
8f4e2101 HD |
2878 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2879 | * but allow concurrent faults), and pte mapped but not yet locked. | |
9a95f3cf PC |
2880 | * We return with pte unmapped and unlocked. |
2881 | * | |
2882 | * We return with the mmap_sem locked or unlocked in the same cases | |
2883 | * as does filemap_fault(). | |
1da177e4 | 2884 | */ |
2994302b | 2885 | int do_swap_page(struct vm_fault *vmf) |
1da177e4 | 2886 | { |
82b0f8c3 | 2887 | struct vm_area_struct *vma = vmf->vma; |
0bcac06f | 2888 | struct page *page = NULL, *swapcache = NULL; |
00501b53 | 2889 | struct mem_cgroup *memcg; |
ec560175 | 2890 | struct vma_swap_readahead swap_ra; |
65500d23 | 2891 | swp_entry_t entry; |
1da177e4 | 2892 | pte_t pte; |
d065bd81 | 2893 | int locked; |
ad8c2ee8 | 2894 | int exclusive = 0; |
83c54070 | 2895 | int ret = 0; |
ec560175 | 2896 | bool vma_readahead = swap_use_vma_readahead(); |
1da177e4 | 2897 | |
f8020772 | 2898 | if (vma_readahead) { |
ec560175 | 2899 | page = swap_readahead_detect(vmf, &swap_ra); |
f8020772 MK |
2900 | swapcache = page; |
2901 | } | |
2902 | ||
ec560175 HY |
2903 | if (!pte_unmap_same(vma->vm_mm, vmf->pmd, vmf->pte, vmf->orig_pte)) { |
2904 | if (page) | |
2905 | put_page(page); | |
8f4e2101 | 2906 | goto out; |
ec560175 | 2907 | } |
65500d23 | 2908 | |
2994302b | 2909 | entry = pte_to_swp_entry(vmf->orig_pte); |
d1737fdb AK |
2910 | if (unlikely(non_swap_entry(entry))) { |
2911 | if (is_migration_entry(entry)) { | |
82b0f8c3 JK |
2912 | migration_entry_wait(vma->vm_mm, vmf->pmd, |
2913 | vmf->address); | |
5042db43 JG |
2914 | } else if (is_device_private_entry(entry)) { |
2915 | /* | |
2916 | * For un-addressable device memory we call the pgmap | |
2917 | * fault handler callback. The callback must migrate | |
2918 | * the page back to some CPU accessible page. | |
2919 | */ | |
2920 | ret = device_private_entry_fault(vma, vmf->address, entry, | |
2921 | vmf->flags, vmf->pmd); | |
d1737fdb AK |
2922 | } else if (is_hwpoison_entry(entry)) { |
2923 | ret = VM_FAULT_HWPOISON; | |
2924 | } else { | |
2994302b | 2925 | print_bad_pte(vma, vmf->address, vmf->orig_pte, NULL); |
d99be1a8 | 2926 | ret = VM_FAULT_SIGBUS; |
d1737fdb | 2927 | } |
0697212a CL |
2928 | goto out; |
2929 | } | |
0bcac06f MK |
2930 | |
2931 | ||
0ff92245 | 2932 | delayacct_set_flag(DELAYACCT_PF_SWAPIN); |
f8020772 | 2933 | if (!page) { |
ec560175 HY |
2934 | page = lookup_swap_cache(entry, vma_readahead ? vma : NULL, |
2935 | vmf->address); | |
f8020772 MK |
2936 | swapcache = page; |
2937 | } | |
2938 | ||
1da177e4 | 2939 | if (!page) { |
0bcac06f MK |
2940 | struct swap_info_struct *si = swp_swap_info(entry); |
2941 | ||
aa8d22a1 MK |
2942 | if (si->flags & SWP_SYNCHRONOUS_IO && |
2943 | __swap_count(si, entry) == 1) { | |
0bcac06f MK |
2944 | /* skip swapcache */ |
2945 | page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vmf->address); | |
2946 | if (page) { | |
2947 | __SetPageLocked(page); | |
2948 | __SetPageSwapBacked(page); | |
2949 | set_page_private(page, entry.val); | |
2950 | lru_cache_add_anon(page); | |
2951 | swap_readpage(page, true); | |
2952 | } | |
aa8d22a1 MK |
2953 | } else { |
2954 | if (vma_readahead) | |
2955 | page = do_swap_page_readahead(entry, | |
2956 | GFP_HIGHUSER_MOVABLE, vmf, &swap_ra); | |
2957 | else | |
2958 | page = swapin_readahead(entry, | |
2959 | GFP_HIGHUSER_MOVABLE, vma, vmf->address); | |
2960 | swapcache = page; | |
0bcac06f MK |
2961 | } |
2962 | ||
1da177e4 LT |
2963 | if (!page) { |
2964 | /* | |
8f4e2101 HD |
2965 | * Back out if somebody else faulted in this pte |
2966 | * while we released the pte lock. | |
1da177e4 | 2967 | */ |
82b0f8c3 JK |
2968 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
2969 | vmf->address, &vmf->ptl); | |
2994302b | 2970 | if (likely(pte_same(*vmf->pte, vmf->orig_pte))) |
1da177e4 | 2971 | ret = VM_FAULT_OOM; |
0ff92245 | 2972 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
65500d23 | 2973 | goto unlock; |
1da177e4 LT |
2974 | } |
2975 | ||
2976 | /* Had to read the page from swap area: Major fault */ | |
2977 | ret = VM_FAULT_MAJOR; | |
f8891e5e | 2978 | count_vm_event(PGMAJFAULT); |
2262185c | 2979 | count_memcg_event_mm(vma->vm_mm, PGMAJFAULT); |
d1737fdb | 2980 | } else if (PageHWPoison(page)) { |
71f72525 WF |
2981 | /* |
2982 | * hwpoisoned dirty swapcache pages are kept for killing | |
2983 | * owner processes (which may be unknown at hwpoison time) | |
2984 | */ | |
d1737fdb AK |
2985 | ret = VM_FAULT_HWPOISON; |
2986 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); | |
56f31801 | 2987 | swapcache = page; |
4779cb31 | 2988 | goto out_release; |
1da177e4 LT |
2989 | } |
2990 | ||
82b0f8c3 | 2991 | locked = lock_page_or_retry(page, vma->vm_mm, vmf->flags); |
e709ffd6 | 2992 | |
073e587e | 2993 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
d065bd81 ML |
2994 | if (!locked) { |
2995 | ret |= VM_FAULT_RETRY; | |
2996 | goto out_release; | |
2997 | } | |
073e587e | 2998 | |
4969c119 | 2999 | /* |
31c4a3d3 HD |
3000 | * Make sure try_to_free_swap or reuse_swap_page or swapoff did not |
3001 | * release the swapcache from under us. The page pin, and pte_same | |
3002 | * test below, are not enough to exclude that. Even if it is still | |
3003 | * swapcache, we need to check that the page's swap has not changed. | |
4969c119 | 3004 | */ |
0bcac06f MK |
3005 | if (unlikely((!PageSwapCache(page) || |
3006 | page_private(page) != entry.val)) && swapcache) | |
4969c119 AA |
3007 | goto out_page; |
3008 | ||
82b0f8c3 | 3009 | page = ksm_might_need_to_copy(page, vma, vmf->address); |
cbf86cfe HD |
3010 | if (unlikely(!page)) { |
3011 | ret = VM_FAULT_OOM; | |
3012 | page = swapcache; | |
cbf86cfe | 3013 | goto out_page; |
5ad64688 HD |
3014 | } |
3015 | ||
bae473a4 KS |
3016 | if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL, |
3017 | &memcg, false)) { | |
8a9f3ccd | 3018 | ret = VM_FAULT_OOM; |
bc43f75c | 3019 | goto out_page; |
8a9f3ccd BS |
3020 | } |
3021 | ||
1da177e4 | 3022 | /* |
8f4e2101 | 3023 | * Back out if somebody else already faulted in this pte. |
1da177e4 | 3024 | */ |
82b0f8c3 JK |
3025 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, |
3026 | &vmf->ptl); | |
2994302b | 3027 | if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) |
b8107480 | 3028 | goto out_nomap; |
b8107480 KK |
3029 | |
3030 | if (unlikely(!PageUptodate(page))) { | |
3031 | ret = VM_FAULT_SIGBUS; | |
3032 | goto out_nomap; | |
1da177e4 LT |
3033 | } |
3034 | ||
8c7c6e34 KH |
3035 | /* |
3036 | * The page isn't present yet, go ahead with the fault. | |
3037 | * | |
3038 | * Be careful about the sequence of operations here. | |
3039 | * To get its accounting right, reuse_swap_page() must be called | |
3040 | * while the page is counted on swap but not yet in mapcount i.e. | |
3041 | * before page_add_anon_rmap() and swap_free(); try_to_free_swap() | |
3042 | * must be called after the swap_free(), or it will never succeed. | |
8c7c6e34 | 3043 | */ |
1da177e4 | 3044 | |
bae473a4 KS |
3045 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
3046 | dec_mm_counter_fast(vma->vm_mm, MM_SWAPENTS); | |
1da177e4 | 3047 | pte = mk_pte(page, vma->vm_page_prot); |
82b0f8c3 | 3048 | if ((vmf->flags & FAULT_FLAG_WRITE) && reuse_swap_page(page, NULL)) { |
1da177e4 | 3049 | pte = maybe_mkwrite(pte_mkdirty(pte), vma); |
82b0f8c3 | 3050 | vmf->flags &= ~FAULT_FLAG_WRITE; |
9a5b489b | 3051 | ret |= VM_FAULT_WRITE; |
d281ee61 | 3052 | exclusive = RMAP_EXCLUSIVE; |
1da177e4 | 3053 | } |
1da177e4 | 3054 | flush_icache_page(vma, page); |
2994302b | 3055 | if (pte_swp_soft_dirty(vmf->orig_pte)) |
179ef71c | 3056 | pte = pte_mksoft_dirty(pte); |
82b0f8c3 | 3057 | set_pte_at(vma->vm_mm, vmf->address, vmf->pte, pte); |
2994302b | 3058 | vmf->orig_pte = pte; |
0bcac06f MK |
3059 | |
3060 | /* ksm created a completely new copy */ | |
3061 | if (unlikely(page != swapcache && swapcache)) { | |
82b0f8c3 | 3062 | page_add_new_anon_rmap(page, vma, vmf->address, false); |
f627c2f5 | 3063 | mem_cgroup_commit_charge(page, memcg, false, false); |
00501b53 | 3064 | lru_cache_add_active_or_unevictable(page, vma); |
0bcac06f MK |
3065 | } else { |
3066 | do_page_add_anon_rmap(page, vma, vmf->address, exclusive); | |
3067 | mem_cgroup_commit_charge(page, memcg, true, false); | |
3068 | activate_page(page); | |
00501b53 | 3069 | } |
1da177e4 | 3070 | |
c475a8ab | 3071 | swap_free(entry); |
5ccc5aba VD |
3072 | if (mem_cgroup_swap_full(page) || |
3073 | (vma->vm_flags & VM_LOCKED) || PageMlocked(page)) | |
a2c43eed | 3074 | try_to_free_swap(page); |
c475a8ab | 3075 | unlock_page(page); |
0bcac06f | 3076 | if (page != swapcache && swapcache) { |
4969c119 AA |
3077 | /* |
3078 | * Hold the lock to avoid the swap entry to be reused | |
3079 | * until we take the PT lock for the pte_same() check | |
3080 | * (to avoid false positives from pte_same). For | |
3081 | * further safety release the lock after the swap_free | |
3082 | * so that the swap count won't change under a | |
3083 | * parallel locked swapcache. | |
3084 | */ | |
3085 | unlock_page(swapcache); | |
09cbfeaf | 3086 | put_page(swapcache); |
4969c119 | 3087 | } |
c475a8ab | 3088 | |
82b0f8c3 | 3089 | if (vmf->flags & FAULT_FLAG_WRITE) { |
2994302b | 3090 | ret |= do_wp_page(vmf); |
61469f1d HD |
3091 | if (ret & VM_FAULT_ERROR) |
3092 | ret &= VM_FAULT_ERROR; | |
1da177e4 LT |
3093 | goto out; |
3094 | } | |
3095 | ||
3096 | /* No need to invalidate - it was non-present before */ | |
82b0f8c3 | 3097 | update_mmu_cache(vma, vmf->address, vmf->pte); |
65500d23 | 3098 | unlock: |
82b0f8c3 | 3099 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
1da177e4 LT |
3100 | out: |
3101 | return ret; | |
b8107480 | 3102 | out_nomap: |
f627c2f5 | 3103 | mem_cgroup_cancel_charge(page, memcg, false); |
82b0f8c3 | 3104 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
bc43f75c | 3105 | out_page: |
b8107480 | 3106 | unlock_page(page); |
4779cb31 | 3107 | out_release: |
09cbfeaf | 3108 | put_page(page); |
0bcac06f | 3109 | if (page != swapcache && swapcache) { |
4969c119 | 3110 | unlock_page(swapcache); |
09cbfeaf | 3111 | put_page(swapcache); |
4969c119 | 3112 | } |
65500d23 | 3113 | return ret; |
1da177e4 LT |
3114 | } |
3115 | ||
3116 | /* | |
8f4e2101 HD |
3117 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
3118 | * but allow concurrent faults), and pte mapped but not yet locked. | |
3119 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 3120 | */ |
82b0f8c3 | 3121 | static int do_anonymous_page(struct vm_fault *vmf) |
1da177e4 | 3122 | { |
82b0f8c3 | 3123 | struct vm_area_struct *vma = vmf->vma; |
00501b53 | 3124 | struct mem_cgroup *memcg; |
8f4e2101 | 3125 | struct page *page; |
6b31d595 | 3126 | int ret = 0; |
1da177e4 | 3127 | pte_t entry; |
1da177e4 | 3128 | |
6b7339f4 KS |
3129 | /* File mapping without ->vm_ops ? */ |
3130 | if (vma->vm_flags & VM_SHARED) | |
3131 | return VM_FAULT_SIGBUS; | |
3132 | ||
7267ec00 KS |
3133 | /* |
3134 | * Use pte_alloc() instead of pte_alloc_map(). We can't run | |
3135 | * pte_offset_map() on pmds where a huge pmd might be created | |
3136 | * from a different thread. | |
3137 | * | |
3138 | * pte_alloc_map() is safe to use under down_write(mmap_sem) or when | |
3139 | * parallel threads are excluded by other means. | |
3140 | * | |
3141 | * Here we only have down_read(mmap_sem). | |
3142 | */ | |
82b0f8c3 | 3143 | if (pte_alloc(vma->vm_mm, vmf->pmd, vmf->address)) |
7267ec00 KS |
3144 | return VM_FAULT_OOM; |
3145 | ||
3146 | /* See the comment in pte_alloc_one_map() */ | |
82b0f8c3 | 3147 | if (unlikely(pmd_trans_unstable(vmf->pmd))) |
7267ec00 KS |
3148 | return 0; |
3149 | ||
11ac5524 | 3150 | /* Use the zero-page for reads */ |
82b0f8c3 | 3151 | if (!(vmf->flags & FAULT_FLAG_WRITE) && |
bae473a4 | 3152 | !mm_forbids_zeropage(vma->vm_mm)) { |
82b0f8c3 | 3153 | entry = pte_mkspecial(pfn_pte(my_zero_pfn(vmf->address), |
62eede62 | 3154 | vma->vm_page_prot)); |
82b0f8c3 JK |
3155 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
3156 | vmf->address, &vmf->ptl); | |
3157 | if (!pte_none(*vmf->pte)) | |
a13ea5b7 | 3158 | goto unlock; |
6b31d595 MH |
3159 | ret = check_stable_address_space(vma->vm_mm); |
3160 | if (ret) | |
3161 | goto unlock; | |
6b251fc9 AA |
3162 | /* Deliver the page fault to userland, check inside PT lock */ |
3163 | if (userfaultfd_missing(vma)) { | |
82b0f8c3 JK |
3164 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
3165 | return handle_userfault(vmf, VM_UFFD_MISSING); | |
6b251fc9 | 3166 | } |
a13ea5b7 HD |
3167 | goto setpte; |
3168 | } | |
3169 | ||
557ed1fa | 3170 | /* Allocate our own private page. */ |
557ed1fa NP |
3171 | if (unlikely(anon_vma_prepare(vma))) |
3172 | goto oom; | |
82b0f8c3 | 3173 | page = alloc_zeroed_user_highpage_movable(vma, vmf->address); |
557ed1fa NP |
3174 | if (!page) |
3175 | goto oom; | |
eb3c24f3 | 3176 | |
bae473a4 | 3177 | if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL, &memcg, false)) |
eb3c24f3 MG |
3178 | goto oom_free_page; |
3179 | ||
52f37629 MK |
3180 | /* |
3181 | * The memory barrier inside __SetPageUptodate makes sure that | |
3182 | * preceeding stores to the page contents become visible before | |
3183 | * the set_pte_at() write. | |
3184 | */ | |
0ed361de | 3185 | __SetPageUptodate(page); |
8f4e2101 | 3186 | |
557ed1fa | 3187 | entry = mk_pte(page, vma->vm_page_prot); |
1ac0cb5d HD |
3188 | if (vma->vm_flags & VM_WRITE) |
3189 | entry = pte_mkwrite(pte_mkdirty(entry)); | |
1da177e4 | 3190 | |
82b0f8c3 JK |
3191 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, |
3192 | &vmf->ptl); | |
3193 | if (!pte_none(*vmf->pte)) | |
557ed1fa | 3194 | goto release; |
9ba69294 | 3195 | |
6b31d595 MH |
3196 | ret = check_stable_address_space(vma->vm_mm); |
3197 | if (ret) | |
3198 | goto release; | |
3199 | ||
6b251fc9 AA |
3200 | /* Deliver the page fault to userland, check inside PT lock */ |
3201 | if (userfaultfd_missing(vma)) { | |
82b0f8c3 | 3202 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
f627c2f5 | 3203 | mem_cgroup_cancel_charge(page, memcg, false); |
09cbfeaf | 3204 | put_page(page); |
82b0f8c3 | 3205 | return handle_userfault(vmf, VM_UFFD_MISSING); |
6b251fc9 AA |
3206 | } |
3207 | ||
bae473a4 | 3208 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
82b0f8c3 | 3209 | page_add_new_anon_rmap(page, vma, vmf->address, false); |
f627c2f5 | 3210 | mem_cgroup_commit_charge(page, memcg, false, false); |
00501b53 | 3211 | lru_cache_add_active_or_unevictable(page, vma); |
a13ea5b7 | 3212 | setpte: |
82b0f8c3 | 3213 | set_pte_at(vma->vm_mm, vmf->address, vmf->pte, entry); |
1da177e4 LT |
3214 | |
3215 | /* No need to invalidate - it was non-present before */ | |
82b0f8c3 | 3216 | update_mmu_cache(vma, vmf->address, vmf->pte); |
65500d23 | 3217 | unlock: |
82b0f8c3 | 3218 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
6b31d595 | 3219 | return ret; |
8f4e2101 | 3220 | release: |
f627c2f5 | 3221 | mem_cgroup_cancel_charge(page, memcg, false); |
09cbfeaf | 3222 | put_page(page); |
8f4e2101 | 3223 | goto unlock; |
8a9f3ccd | 3224 | oom_free_page: |
09cbfeaf | 3225 | put_page(page); |
65500d23 | 3226 | oom: |
1da177e4 LT |
3227 | return VM_FAULT_OOM; |
3228 | } | |
3229 | ||
9a95f3cf PC |
3230 | /* |
3231 | * The mmap_sem must have been held on entry, and may have been | |
3232 | * released depending on flags and vma->vm_ops->fault() return value. | |
3233 | * See filemap_fault() and __lock_page_retry(). | |
3234 | */ | |
936ca80d | 3235 | static int __do_fault(struct vm_fault *vmf) |
7eae74af | 3236 | { |
82b0f8c3 | 3237 | struct vm_area_struct *vma = vmf->vma; |
7eae74af KS |
3238 | int ret; |
3239 | ||
11bac800 | 3240 | ret = vma->vm_ops->fault(vmf); |
3917048d | 3241 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY | |
b1aa812b | 3242 | VM_FAULT_DONE_COW))) |
bc2466e4 | 3243 | return ret; |
7eae74af | 3244 | |
667240e0 | 3245 | if (unlikely(PageHWPoison(vmf->page))) { |
7eae74af | 3246 | if (ret & VM_FAULT_LOCKED) |
667240e0 JK |
3247 | unlock_page(vmf->page); |
3248 | put_page(vmf->page); | |
936ca80d | 3249 | vmf->page = NULL; |
7eae74af KS |
3250 | return VM_FAULT_HWPOISON; |
3251 | } | |
3252 | ||
3253 | if (unlikely(!(ret & VM_FAULT_LOCKED))) | |
667240e0 | 3254 | lock_page(vmf->page); |
7eae74af | 3255 | else |
667240e0 | 3256 | VM_BUG_ON_PAGE(!PageLocked(vmf->page), vmf->page); |
7eae74af | 3257 | |
7eae74af KS |
3258 | return ret; |
3259 | } | |
3260 | ||
d0f0931d RZ |
3261 | /* |
3262 | * The ordering of these checks is important for pmds with _PAGE_DEVMAP set. | |
3263 | * If we check pmd_trans_unstable() first we will trip the bad_pmd() check | |
3264 | * inside of pmd_none_or_trans_huge_or_clear_bad(). This will end up correctly | |
3265 | * returning 1 but not before it spams dmesg with the pmd_clear_bad() output. | |
3266 | */ | |
3267 | static int pmd_devmap_trans_unstable(pmd_t *pmd) | |
3268 | { | |
3269 | return pmd_devmap(*pmd) || pmd_trans_unstable(pmd); | |
3270 | } | |
3271 | ||
82b0f8c3 | 3272 | static int pte_alloc_one_map(struct vm_fault *vmf) |
7267ec00 | 3273 | { |
82b0f8c3 | 3274 | struct vm_area_struct *vma = vmf->vma; |
7267ec00 | 3275 | |
82b0f8c3 | 3276 | if (!pmd_none(*vmf->pmd)) |
7267ec00 | 3277 | goto map_pte; |
82b0f8c3 JK |
3278 | if (vmf->prealloc_pte) { |
3279 | vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd); | |
3280 | if (unlikely(!pmd_none(*vmf->pmd))) { | |
3281 | spin_unlock(vmf->ptl); | |
7267ec00 KS |
3282 | goto map_pte; |
3283 | } | |
3284 | ||
c4812909 | 3285 | mm_inc_nr_ptes(vma->vm_mm); |
82b0f8c3 JK |
3286 | pmd_populate(vma->vm_mm, vmf->pmd, vmf->prealloc_pte); |
3287 | spin_unlock(vmf->ptl); | |
7f2b6ce8 | 3288 | vmf->prealloc_pte = NULL; |
82b0f8c3 | 3289 | } else if (unlikely(pte_alloc(vma->vm_mm, vmf->pmd, vmf->address))) { |
7267ec00 KS |
3290 | return VM_FAULT_OOM; |
3291 | } | |
3292 | map_pte: | |
3293 | /* | |
3294 | * If a huge pmd materialized under us just retry later. Use | |
d0f0931d RZ |
3295 | * pmd_trans_unstable() via pmd_devmap_trans_unstable() instead of |
3296 | * pmd_trans_huge() to ensure the pmd didn't become pmd_trans_huge | |
3297 | * under us and then back to pmd_none, as a result of MADV_DONTNEED | |
3298 | * running immediately after a huge pmd fault in a different thread of | |
3299 | * this mm, in turn leading to a misleading pmd_trans_huge() retval. | |
3300 | * All we have to ensure is that it is a regular pmd that we can walk | |
3301 | * with pte_offset_map() and we can do that through an atomic read in | |
3302 | * C, which is what pmd_trans_unstable() provides. | |
7267ec00 | 3303 | */ |
d0f0931d | 3304 | if (pmd_devmap_trans_unstable(vmf->pmd)) |
7267ec00 KS |
3305 | return VM_FAULT_NOPAGE; |
3306 | ||
d0f0931d RZ |
3307 | /* |
3308 | * At this point we know that our vmf->pmd points to a page of ptes | |
3309 | * and it cannot become pmd_none(), pmd_devmap() or pmd_trans_huge() | |
3310 | * for the duration of the fault. If a racing MADV_DONTNEED runs and | |
3311 | * we zap the ptes pointed to by our vmf->pmd, the vmf->ptl will still | |
3312 | * be valid and we will re-check to make sure the vmf->pte isn't | |
3313 | * pte_none() under vmf->ptl protection when we return to | |
3314 | * alloc_set_pte(). | |
3315 | */ | |
82b0f8c3 JK |
3316 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, |
3317 | &vmf->ptl); | |
7267ec00 KS |
3318 | return 0; |
3319 | } | |
3320 | ||
e496cf3d | 3321 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
10102459 KS |
3322 | |
3323 | #define HPAGE_CACHE_INDEX_MASK (HPAGE_PMD_NR - 1) | |
3324 | static inline bool transhuge_vma_suitable(struct vm_area_struct *vma, | |
3325 | unsigned long haddr) | |
3326 | { | |
3327 | if (((vma->vm_start >> PAGE_SHIFT) & HPAGE_CACHE_INDEX_MASK) != | |
3328 | (vma->vm_pgoff & HPAGE_CACHE_INDEX_MASK)) | |
3329 | return false; | |
3330 | if (haddr < vma->vm_start || haddr + HPAGE_PMD_SIZE > vma->vm_end) | |
3331 | return false; | |
3332 | return true; | |
3333 | } | |
3334 | ||
82b0f8c3 | 3335 | static void deposit_prealloc_pte(struct vm_fault *vmf) |
953c66c2 | 3336 | { |
82b0f8c3 | 3337 | struct vm_area_struct *vma = vmf->vma; |
953c66c2 | 3338 | |
82b0f8c3 | 3339 | pgtable_trans_huge_deposit(vma->vm_mm, vmf->pmd, vmf->prealloc_pte); |
953c66c2 AK |
3340 | /* |
3341 | * We are going to consume the prealloc table, | |
3342 | * count that as nr_ptes. | |
3343 | */ | |
c4812909 | 3344 | mm_inc_nr_ptes(vma->vm_mm); |
7f2b6ce8 | 3345 | vmf->prealloc_pte = NULL; |
953c66c2 AK |
3346 | } |
3347 | ||
82b0f8c3 | 3348 | static int do_set_pmd(struct vm_fault *vmf, struct page *page) |
10102459 | 3349 | { |
82b0f8c3 JK |
3350 | struct vm_area_struct *vma = vmf->vma; |
3351 | bool write = vmf->flags & FAULT_FLAG_WRITE; | |
3352 | unsigned long haddr = vmf->address & HPAGE_PMD_MASK; | |
10102459 KS |
3353 | pmd_t entry; |
3354 | int i, ret; | |
3355 | ||
3356 | if (!transhuge_vma_suitable(vma, haddr)) | |
3357 | return VM_FAULT_FALLBACK; | |
3358 | ||
3359 | ret = VM_FAULT_FALLBACK; | |
3360 | page = compound_head(page); | |
3361 | ||
953c66c2 AK |
3362 | /* |
3363 | * Archs like ppc64 need additonal space to store information | |
3364 | * related to pte entry. Use the preallocated table for that. | |
3365 | */ | |
82b0f8c3 JK |
3366 | if (arch_needs_pgtable_deposit() && !vmf->prealloc_pte) { |
3367 | vmf->prealloc_pte = pte_alloc_one(vma->vm_mm, vmf->address); | |
3368 | if (!vmf->prealloc_pte) | |
953c66c2 AK |
3369 | return VM_FAULT_OOM; |
3370 | smp_wmb(); /* See comment in __pte_alloc() */ | |
3371 | } | |
3372 | ||
82b0f8c3 JK |
3373 | vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd); |
3374 | if (unlikely(!pmd_none(*vmf->pmd))) | |
10102459 KS |
3375 | goto out; |
3376 | ||
3377 | for (i = 0; i < HPAGE_PMD_NR; i++) | |
3378 | flush_icache_page(vma, page + i); | |
3379 | ||
3380 | entry = mk_huge_pmd(page, vma->vm_page_prot); | |
3381 | if (write) | |
f55e1014 | 3382 | entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma); |
10102459 KS |
3383 | |
3384 | add_mm_counter(vma->vm_mm, MM_FILEPAGES, HPAGE_PMD_NR); | |
3385 | page_add_file_rmap(page, true); | |
953c66c2 AK |
3386 | /* |
3387 | * deposit and withdraw with pmd lock held | |
3388 | */ | |
3389 | if (arch_needs_pgtable_deposit()) | |
82b0f8c3 | 3390 | deposit_prealloc_pte(vmf); |
10102459 | 3391 | |
82b0f8c3 | 3392 | set_pmd_at(vma->vm_mm, haddr, vmf->pmd, entry); |
10102459 | 3393 | |
82b0f8c3 | 3394 | update_mmu_cache_pmd(vma, haddr, vmf->pmd); |
10102459 KS |
3395 | |
3396 | /* fault is handled */ | |
3397 | ret = 0; | |
95ecedcd | 3398 | count_vm_event(THP_FILE_MAPPED); |
10102459 | 3399 | out: |
82b0f8c3 | 3400 | spin_unlock(vmf->ptl); |
10102459 KS |
3401 | return ret; |
3402 | } | |
3403 | #else | |
82b0f8c3 | 3404 | static int do_set_pmd(struct vm_fault *vmf, struct page *page) |
10102459 KS |
3405 | { |
3406 | BUILD_BUG(); | |
3407 | return 0; | |
3408 | } | |
3409 | #endif | |
3410 | ||
8c6e50b0 | 3411 | /** |
7267ec00 KS |
3412 | * alloc_set_pte - setup new PTE entry for given page and add reverse page |
3413 | * mapping. If needed, the fucntion allocates page table or use pre-allocated. | |
8c6e50b0 | 3414 | * |
82b0f8c3 | 3415 | * @vmf: fault environment |
7267ec00 | 3416 | * @memcg: memcg to charge page (only for private mappings) |
8c6e50b0 | 3417 | * @page: page to map |
8c6e50b0 | 3418 | * |
82b0f8c3 JK |
3419 | * Caller must take care of unlocking vmf->ptl, if vmf->pte is non-NULL on |
3420 | * return. | |
8c6e50b0 KS |
3421 | * |
3422 | * Target users are page handler itself and implementations of | |
3423 | * vm_ops->map_pages. | |
3424 | */ | |
82b0f8c3 | 3425 | int alloc_set_pte(struct vm_fault *vmf, struct mem_cgroup *memcg, |
7267ec00 | 3426 | struct page *page) |
3bb97794 | 3427 | { |
82b0f8c3 JK |
3428 | struct vm_area_struct *vma = vmf->vma; |
3429 | bool write = vmf->flags & FAULT_FLAG_WRITE; | |
3bb97794 | 3430 | pte_t entry; |
10102459 KS |
3431 | int ret; |
3432 | ||
82b0f8c3 | 3433 | if (pmd_none(*vmf->pmd) && PageTransCompound(page) && |
e496cf3d | 3434 | IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) { |
10102459 KS |
3435 | /* THP on COW? */ |
3436 | VM_BUG_ON_PAGE(memcg, page); | |
3437 | ||
82b0f8c3 | 3438 | ret = do_set_pmd(vmf, page); |
10102459 | 3439 | if (ret != VM_FAULT_FALLBACK) |
b0b9b3df | 3440 | return ret; |
10102459 | 3441 | } |
3bb97794 | 3442 | |
82b0f8c3 JK |
3443 | if (!vmf->pte) { |
3444 | ret = pte_alloc_one_map(vmf); | |
7267ec00 | 3445 | if (ret) |
b0b9b3df | 3446 | return ret; |
7267ec00 KS |
3447 | } |
3448 | ||
3449 | /* Re-check under ptl */ | |
b0b9b3df HD |
3450 | if (unlikely(!pte_none(*vmf->pte))) |
3451 | return VM_FAULT_NOPAGE; | |
7267ec00 | 3452 | |
3bb97794 KS |
3453 | flush_icache_page(vma, page); |
3454 | entry = mk_pte(page, vma->vm_page_prot); | |
3455 | if (write) | |
3456 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
bae473a4 KS |
3457 | /* copy-on-write page */ |
3458 | if (write && !(vma->vm_flags & VM_SHARED)) { | |
3bb97794 | 3459 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
82b0f8c3 | 3460 | page_add_new_anon_rmap(page, vma, vmf->address, false); |
7267ec00 KS |
3461 | mem_cgroup_commit_charge(page, memcg, false, false); |
3462 | lru_cache_add_active_or_unevictable(page, vma); | |
3bb97794 | 3463 | } else { |
eca56ff9 | 3464 | inc_mm_counter_fast(vma->vm_mm, mm_counter_file(page)); |
dd78fedd | 3465 | page_add_file_rmap(page, false); |
3bb97794 | 3466 | } |
82b0f8c3 | 3467 | set_pte_at(vma->vm_mm, vmf->address, vmf->pte, entry); |
3bb97794 KS |
3468 | |
3469 | /* no need to invalidate: a not-present page won't be cached */ | |
82b0f8c3 | 3470 | update_mmu_cache(vma, vmf->address, vmf->pte); |
7267ec00 | 3471 | |
b0b9b3df | 3472 | return 0; |
3bb97794 KS |
3473 | } |
3474 | ||
9118c0cb JK |
3475 | |
3476 | /** | |
3477 | * finish_fault - finish page fault once we have prepared the page to fault | |
3478 | * | |
3479 | * @vmf: structure describing the fault | |
3480 | * | |
3481 | * This function handles all that is needed to finish a page fault once the | |
3482 | * page to fault in is prepared. It handles locking of PTEs, inserts PTE for | |
3483 | * given page, adds reverse page mapping, handles memcg charges and LRU | |
3484 | * addition. The function returns 0 on success, VM_FAULT_ code in case of | |
3485 | * error. | |
3486 | * | |
3487 | * The function expects the page to be locked and on success it consumes a | |
3488 | * reference of a page being mapped (for the PTE which maps it). | |
3489 | */ | |
3490 | int finish_fault(struct vm_fault *vmf) | |
3491 | { | |
3492 | struct page *page; | |
6b31d595 | 3493 | int ret = 0; |
9118c0cb JK |
3494 | |
3495 | /* Did we COW the page? */ | |
3496 | if ((vmf->flags & FAULT_FLAG_WRITE) && | |
3497 | !(vmf->vma->vm_flags & VM_SHARED)) | |
3498 | page = vmf->cow_page; | |
3499 | else | |
3500 | page = vmf->page; | |
6b31d595 MH |
3501 | |
3502 | /* | |
3503 | * check even for read faults because we might have lost our CoWed | |
3504 | * page | |
3505 | */ | |
3506 | if (!(vmf->vma->vm_flags & VM_SHARED)) | |
3507 | ret = check_stable_address_space(vmf->vma->vm_mm); | |
3508 | if (!ret) | |
3509 | ret = alloc_set_pte(vmf, vmf->memcg, page); | |
9118c0cb JK |
3510 | if (vmf->pte) |
3511 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
3512 | return ret; | |
3513 | } | |
3514 | ||
3a91053a KS |
3515 | static unsigned long fault_around_bytes __read_mostly = |
3516 | rounddown_pow_of_two(65536); | |
a9b0f861 | 3517 | |
a9b0f861 KS |
3518 | #ifdef CONFIG_DEBUG_FS |
3519 | static int fault_around_bytes_get(void *data, u64 *val) | |
1592eef0 | 3520 | { |
a9b0f861 | 3521 | *val = fault_around_bytes; |
1592eef0 KS |
3522 | return 0; |
3523 | } | |
3524 | ||
b4903d6e AR |
3525 | /* |
3526 | * fault_around_pages() and fault_around_mask() expects fault_around_bytes | |
3527 | * rounded down to nearest page order. It's what do_fault_around() expects to | |
3528 | * see. | |
3529 | */ | |
a9b0f861 | 3530 | static int fault_around_bytes_set(void *data, u64 val) |
1592eef0 | 3531 | { |
a9b0f861 | 3532 | if (val / PAGE_SIZE > PTRS_PER_PTE) |
1592eef0 | 3533 | return -EINVAL; |
b4903d6e AR |
3534 | if (val > PAGE_SIZE) |
3535 | fault_around_bytes = rounddown_pow_of_two(val); | |
3536 | else | |
3537 | fault_around_bytes = PAGE_SIZE; /* rounddown_pow_of_two(0) is undefined */ | |
1592eef0 KS |
3538 | return 0; |
3539 | } | |
0a1345f8 | 3540 | DEFINE_DEBUGFS_ATTRIBUTE(fault_around_bytes_fops, |
a9b0f861 | 3541 | fault_around_bytes_get, fault_around_bytes_set, "%llu\n"); |
1592eef0 KS |
3542 | |
3543 | static int __init fault_around_debugfs(void) | |
3544 | { | |
3545 | void *ret; | |
3546 | ||
0a1345f8 | 3547 | ret = debugfs_create_file_unsafe("fault_around_bytes", 0644, NULL, NULL, |
a9b0f861 | 3548 | &fault_around_bytes_fops); |
1592eef0 | 3549 | if (!ret) |
a9b0f861 | 3550 | pr_warn("Failed to create fault_around_bytes in debugfs"); |
1592eef0 KS |
3551 | return 0; |
3552 | } | |
3553 | late_initcall(fault_around_debugfs); | |
1592eef0 | 3554 | #endif |
8c6e50b0 | 3555 | |
1fdb412b KS |
3556 | /* |
3557 | * do_fault_around() tries to map few pages around the fault address. The hope | |
3558 | * is that the pages will be needed soon and this will lower the number of | |
3559 | * faults to handle. | |
3560 | * | |
3561 | * It uses vm_ops->map_pages() to map the pages, which skips the page if it's | |
3562 | * not ready to be mapped: not up-to-date, locked, etc. | |
3563 | * | |
3564 | * This function is called with the page table lock taken. In the split ptlock | |
3565 | * case the page table lock only protects only those entries which belong to | |
3566 | * the page table corresponding to the fault address. | |
3567 | * | |
3568 | * This function doesn't cross the VMA boundaries, in order to call map_pages() | |
3569 | * only once. | |
3570 | * | |
3571 | * fault_around_pages() defines how many pages we'll try to map. | |
3572 | * do_fault_around() expects it to return a power of two less than or equal to | |
3573 | * PTRS_PER_PTE. | |
3574 | * | |
3575 | * The virtual address of the area that we map is naturally aligned to the | |
3576 | * fault_around_pages() value (and therefore to page order). This way it's | |
3577 | * easier to guarantee that we don't cross page table boundaries. | |
3578 | */ | |
0721ec8b | 3579 | static int do_fault_around(struct vm_fault *vmf) |
8c6e50b0 | 3580 | { |
82b0f8c3 | 3581 | unsigned long address = vmf->address, nr_pages, mask; |
0721ec8b | 3582 | pgoff_t start_pgoff = vmf->pgoff; |
bae473a4 | 3583 | pgoff_t end_pgoff; |
7267ec00 | 3584 | int off, ret = 0; |
8c6e50b0 | 3585 | |
4db0c3c2 | 3586 | nr_pages = READ_ONCE(fault_around_bytes) >> PAGE_SHIFT; |
aecd6f44 KS |
3587 | mask = ~(nr_pages * PAGE_SIZE - 1) & PAGE_MASK; |
3588 | ||
82b0f8c3 JK |
3589 | vmf->address = max(address & mask, vmf->vma->vm_start); |
3590 | off = ((address - vmf->address) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1); | |
bae473a4 | 3591 | start_pgoff -= off; |
8c6e50b0 KS |
3592 | |
3593 | /* | |
bae473a4 KS |
3594 | * end_pgoff is either end of page table or end of vma |
3595 | * or fault_around_pages() from start_pgoff, depending what is nearest. | |
8c6e50b0 | 3596 | */ |
bae473a4 | 3597 | end_pgoff = start_pgoff - |
82b0f8c3 | 3598 | ((vmf->address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)) + |
8c6e50b0 | 3599 | PTRS_PER_PTE - 1; |
82b0f8c3 | 3600 | end_pgoff = min3(end_pgoff, vma_pages(vmf->vma) + vmf->vma->vm_pgoff - 1, |
bae473a4 | 3601 | start_pgoff + nr_pages - 1); |
8c6e50b0 | 3602 | |
82b0f8c3 JK |
3603 | if (pmd_none(*vmf->pmd)) { |
3604 | vmf->prealloc_pte = pte_alloc_one(vmf->vma->vm_mm, | |
3605 | vmf->address); | |
3606 | if (!vmf->prealloc_pte) | |
c5f88bd2 | 3607 | goto out; |
7267ec00 | 3608 | smp_wmb(); /* See comment in __pte_alloc() */ |
8c6e50b0 KS |
3609 | } |
3610 | ||
82b0f8c3 | 3611 | vmf->vma->vm_ops->map_pages(vmf, start_pgoff, end_pgoff); |
7267ec00 | 3612 | |
7267ec00 | 3613 | /* Huge page is mapped? Page fault is solved */ |
82b0f8c3 | 3614 | if (pmd_trans_huge(*vmf->pmd)) { |
7267ec00 KS |
3615 | ret = VM_FAULT_NOPAGE; |
3616 | goto out; | |
3617 | } | |
3618 | ||
3619 | /* ->map_pages() haven't done anything useful. Cold page cache? */ | |
82b0f8c3 | 3620 | if (!vmf->pte) |
7267ec00 KS |
3621 | goto out; |
3622 | ||
3623 | /* check if the page fault is solved */ | |
82b0f8c3 JK |
3624 | vmf->pte -= (vmf->address >> PAGE_SHIFT) - (address >> PAGE_SHIFT); |
3625 | if (!pte_none(*vmf->pte)) | |
7267ec00 | 3626 | ret = VM_FAULT_NOPAGE; |
82b0f8c3 | 3627 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
bae473a4 | 3628 | out: |
82b0f8c3 JK |
3629 | vmf->address = address; |
3630 | vmf->pte = NULL; | |
7267ec00 | 3631 | return ret; |
8c6e50b0 KS |
3632 | } |
3633 | ||
0721ec8b | 3634 | static int do_read_fault(struct vm_fault *vmf) |
e655fb29 | 3635 | { |
82b0f8c3 | 3636 | struct vm_area_struct *vma = vmf->vma; |
8c6e50b0 KS |
3637 | int ret = 0; |
3638 | ||
3639 | /* | |
3640 | * Let's call ->map_pages() first and use ->fault() as fallback | |
3641 | * if page by the offset is not ready to be mapped (cold cache or | |
3642 | * something). | |
3643 | */ | |
9b4bdd2f | 3644 | if (vma->vm_ops->map_pages && fault_around_bytes >> PAGE_SHIFT > 1) { |
0721ec8b | 3645 | ret = do_fault_around(vmf); |
7267ec00 KS |
3646 | if (ret) |
3647 | return ret; | |
8c6e50b0 | 3648 | } |
e655fb29 | 3649 | |
936ca80d | 3650 | ret = __do_fault(vmf); |
e655fb29 KS |
3651 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
3652 | return ret; | |
3653 | ||
9118c0cb | 3654 | ret |= finish_fault(vmf); |
936ca80d | 3655 | unlock_page(vmf->page); |
7267ec00 | 3656 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
936ca80d | 3657 | put_page(vmf->page); |
e655fb29 KS |
3658 | return ret; |
3659 | } | |
3660 | ||
0721ec8b | 3661 | static int do_cow_fault(struct vm_fault *vmf) |
ec47c3b9 | 3662 | { |
82b0f8c3 | 3663 | struct vm_area_struct *vma = vmf->vma; |
ec47c3b9 KS |
3664 | int ret; |
3665 | ||
3666 | if (unlikely(anon_vma_prepare(vma))) | |
3667 | return VM_FAULT_OOM; | |
3668 | ||
936ca80d JK |
3669 | vmf->cow_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vmf->address); |
3670 | if (!vmf->cow_page) | |
ec47c3b9 KS |
3671 | return VM_FAULT_OOM; |
3672 | ||
936ca80d | 3673 | if (mem_cgroup_try_charge(vmf->cow_page, vma->vm_mm, GFP_KERNEL, |
3917048d | 3674 | &vmf->memcg, false)) { |
936ca80d | 3675 | put_page(vmf->cow_page); |
ec47c3b9 KS |
3676 | return VM_FAULT_OOM; |
3677 | } | |
3678 | ||
936ca80d | 3679 | ret = __do_fault(vmf); |
ec47c3b9 KS |
3680 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
3681 | goto uncharge_out; | |
3917048d JK |
3682 | if (ret & VM_FAULT_DONE_COW) |
3683 | return ret; | |
ec47c3b9 | 3684 | |
b1aa812b | 3685 | copy_user_highpage(vmf->cow_page, vmf->page, vmf->address, vma); |
936ca80d | 3686 | __SetPageUptodate(vmf->cow_page); |
ec47c3b9 | 3687 | |
9118c0cb | 3688 | ret |= finish_fault(vmf); |
b1aa812b JK |
3689 | unlock_page(vmf->page); |
3690 | put_page(vmf->page); | |
7267ec00 KS |
3691 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
3692 | goto uncharge_out; | |
ec47c3b9 KS |
3693 | return ret; |
3694 | uncharge_out: | |
3917048d | 3695 | mem_cgroup_cancel_charge(vmf->cow_page, vmf->memcg, false); |
936ca80d | 3696 | put_page(vmf->cow_page); |
ec47c3b9 KS |
3697 | return ret; |
3698 | } | |
3699 | ||
0721ec8b | 3700 | static int do_shared_fault(struct vm_fault *vmf) |
1da177e4 | 3701 | { |
82b0f8c3 | 3702 | struct vm_area_struct *vma = vmf->vma; |
f0c6d4d2 | 3703 | int ret, tmp; |
1d65f86d | 3704 | |
936ca80d | 3705 | ret = __do_fault(vmf); |
7eae74af | 3706 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
f0c6d4d2 | 3707 | return ret; |
1da177e4 LT |
3708 | |
3709 | /* | |
f0c6d4d2 KS |
3710 | * Check if the backing address space wants to know that the page is |
3711 | * about to become writable | |
1da177e4 | 3712 | */ |
fb09a464 | 3713 | if (vma->vm_ops->page_mkwrite) { |
936ca80d | 3714 | unlock_page(vmf->page); |
38b8cb7f | 3715 | tmp = do_page_mkwrite(vmf); |
fb09a464 KS |
3716 | if (unlikely(!tmp || |
3717 | (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { | |
936ca80d | 3718 | put_page(vmf->page); |
fb09a464 | 3719 | return tmp; |
4294621f | 3720 | } |
fb09a464 KS |
3721 | } |
3722 | ||
9118c0cb | 3723 | ret |= finish_fault(vmf); |
7267ec00 KS |
3724 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | |
3725 | VM_FAULT_RETRY))) { | |
936ca80d JK |
3726 | unlock_page(vmf->page); |
3727 | put_page(vmf->page); | |
f0c6d4d2 | 3728 | return ret; |
1da177e4 | 3729 | } |
b827e496 | 3730 | |
97ba0c2b | 3731 | fault_dirty_shared_page(vma, vmf->page); |
1d65f86d | 3732 | return ret; |
54cb8821 | 3733 | } |
d00806b1 | 3734 | |
9a95f3cf PC |
3735 | /* |
3736 | * We enter with non-exclusive mmap_sem (to exclude vma changes, | |
3737 | * but allow concurrent faults). | |
3738 | * The mmap_sem may have been released depending on flags and our | |
3739 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
3740 | */ | |
82b0f8c3 | 3741 | static int do_fault(struct vm_fault *vmf) |
54cb8821 | 3742 | { |
82b0f8c3 | 3743 | struct vm_area_struct *vma = vmf->vma; |
b0b9b3df | 3744 | int ret; |
54cb8821 | 3745 | |
f8c85d07 AK |
3746 | /* |
3747 | * The VMA was not fully populated on mmap() or missing VM_DONTEXPAND | |
3748 | */ | |
3749 | if (!vma->vm_ops->fault) { | |
3750 | /* | |
3751 | * If we find a migration pmd entry or a none pmd entry, which | |
3752 | * should never happen, return SIGBUS | |
3753 | */ | |
3754 | if (unlikely(!pmd_present(*vmf->pmd))) | |
3755 | ret = VM_FAULT_SIGBUS; | |
3756 | else { | |
3757 | vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, | |
3758 | vmf->pmd, | |
3759 | vmf->address, | |
3760 | &vmf->ptl); | |
3761 | /* | |
3762 | * Make sure this is not a temporary clearing of pte | |
3763 | * by holding ptl and checking again. A R/M/W update | |
3764 | * of pte involves: take ptl, clearing the pte so that | |
3765 | * we don't have concurrent modification by hardware | |
3766 | * followed by an update. | |
3767 | */ | |
3768 | if (unlikely(pte_none(*vmf->pte))) | |
3769 | ret = VM_FAULT_SIGBUS; | |
3770 | else | |
3771 | ret = VM_FAULT_NOPAGE; | |
3772 | ||
3773 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
3774 | } | |
3775 | } else if (!(vmf->flags & FAULT_FLAG_WRITE)) | |
b0b9b3df HD |
3776 | ret = do_read_fault(vmf); |
3777 | else if (!(vma->vm_flags & VM_SHARED)) | |
3778 | ret = do_cow_fault(vmf); | |
3779 | else | |
3780 | ret = do_shared_fault(vmf); | |
3781 | ||
3782 | /* preallocated pagetable is unused: free it */ | |
3783 | if (vmf->prealloc_pte) { | |
3784 | pte_free(vma->vm_mm, vmf->prealloc_pte); | |
7f2b6ce8 | 3785 | vmf->prealloc_pte = NULL; |
b0b9b3df HD |
3786 | } |
3787 | return ret; | |
54cb8821 NP |
3788 | } |
3789 | ||
b19a9939 | 3790 | static int numa_migrate_prep(struct page *page, struct vm_area_struct *vma, |
04bb2f94 RR |
3791 | unsigned long addr, int page_nid, |
3792 | int *flags) | |
9532fec1 MG |
3793 | { |
3794 | get_page(page); | |
3795 | ||
3796 | count_vm_numa_event(NUMA_HINT_FAULTS); | |
04bb2f94 | 3797 | if (page_nid == numa_node_id()) { |
9532fec1 | 3798 | count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL); |
04bb2f94 RR |
3799 | *flags |= TNF_FAULT_LOCAL; |
3800 | } | |
9532fec1 MG |
3801 | |
3802 | return mpol_misplaced(page, vma, addr); | |
3803 | } | |
3804 | ||
2994302b | 3805 | static int do_numa_page(struct vm_fault *vmf) |
d10e63f2 | 3806 | { |
82b0f8c3 | 3807 | struct vm_area_struct *vma = vmf->vma; |
4daae3b4 | 3808 | struct page *page = NULL; |
8191acbd | 3809 | int page_nid = -1; |
90572890 | 3810 | int last_cpupid; |
cbee9f88 | 3811 | int target_nid; |
b8593bfd | 3812 | bool migrated = false; |
cee216a6 | 3813 | pte_t pte; |
288bc549 | 3814 | bool was_writable = pte_savedwrite(vmf->orig_pte); |
6688cc05 | 3815 | int flags = 0; |
d10e63f2 MG |
3816 | |
3817 | /* | |
166f61b9 TH |
3818 | * The "pte" at this point cannot be used safely without |
3819 | * validation through pte_unmap_same(). It's of NUMA type but | |
3820 | * the pfn may be screwed if the read is non atomic. | |
166f61b9 | 3821 | */ |
82b0f8c3 JK |
3822 | vmf->ptl = pte_lockptr(vma->vm_mm, vmf->pmd); |
3823 | spin_lock(vmf->ptl); | |
cee216a6 | 3824 | if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) { |
82b0f8c3 | 3825 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
4daae3b4 MG |
3826 | goto out; |
3827 | } | |
3828 | ||
cee216a6 AK |
3829 | /* |
3830 | * Make it present again, Depending on how arch implementes non | |
3831 | * accessible ptes, some can allow access by kernel mode. | |
3832 | */ | |
3833 | pte = ptep_modify_prot_start(vma->vm_mm, vmf->address, vmf->pte); | |
4d942466 MG |
3834 | pte = pte_modify(pte, vma->vm_page_prot); |
3835 | pte = pte_mkyoung(pte); | |
b191f9b1 MG |
3836 | if (was_writable) |
3837 | pte = pte_mkwrite(pte); | |
cee216a6 | 3838 | ptep_modify_prot_commit(vma->vm_mm, vmf->address, vmf->pte, pte); |
82b0f8c3 | 3839 | update_mmu_cache(vma, vmf->address, vmf->pte); |
d10e63f2 | 3840 | |
82b0f8c3 | 3841 | page = vm_normal_page(vma, vmf->address, pte); |
d10e63f2 | 3842 | if (!page) { |
82b0f8c3 | 3843 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
d10e63f2 MG |
3844 | return 0; |
3845 | } | |
3846 | ||
e81c4802 KS |
3847 | /* TODO: handle PTE-mapped THP */ |
3848 | if (PageCompound(page)) { | |
82b0f8c3 | 3849 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
e81c4802 KS |
3850 | return 0; |
3851 | } | |
3852 | ||
6688cc05 | 3853 | /* |
bea66fbd MG |
3854 | * Avoid grouping on RO pages in general. RO pages shouldn't hurt as |
3855 | * much anyway since they can be in shared cache state. This misses | |
3856 | * the case where a mapping is writable but the process never writes | |
3857 | * to it but pte_write gets cleared during protection updates and | |
3858 | * pte_dirty has unpredictable behaviour between PTE scan updates, | |
3859 | * background writeback, dirty balancing and application behaviour. | |
6688cc05 | 3860 | */ |
d59dc7bc | 3861 | if (!pte_write(pte)) |
6688cc05 PZ |
3862 | flags |= TNF_NO_GROUP; |
3863 | ||
dabe1d99 RR |
3864 | /* |
3865 | * Flag if the page is shared between multiple address spaces. This | |
3866 | * is later used when determining whether to group tasks together | |
3867 | */ | |
3868 | if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED)) | |
3869 | flags |= TNF_SHARED; | |
3870 | ||
90572890 | 3871 | last_cpupid = page_cpupid_last(page); |
8191acbd | 3872 | page_nid = page_to_nid(page); |
82b0f8c3 | 3873 | target_nid = numa_migrate_prep(page, vma, vmf->address, page_nid, |
bae473a4 | 3874 | &flags); |
82b0f8c3 | 3875 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
4daae3b4 | 3876 | if (target_nid == -1) { |
4daae3b4 MG |
3877 | put_page(page); |
3878 | goto out; | |
3879 | } | |
3880 | ||
3881 | /* Migrate to the requested node */ | |
1bc115d8 | 3882 | migrated = migrate_misplaced_page(page, vma, target_nid); |
6688cc05 | 3883 | if (migrated) { |
8191acbd | 3884 | page_nid = target_nid; |
6688cc05 | 3885 | flags |= TNF_MIGRATED; |
074c2381 MG |
3886 | } else |
3887 | flags |= TNF_MIGRATE_FAIL; | |
4daae3b4 MG |
3888 | |
3889 | out: | |
8191acbd | 3890 | if (page_nid != -1) |
6688cc05 | 3891 | task_numa_fault(last_cpupid, page_nid, 1, flags); |
d10e63f2 MG |
3892 | return 0; |
3893 | } | |
3894 | ||
91a90140 | 3895 | static inline int create_huge_pmd(struct vm_fault *vmf) |
b96375f7 | 3896 | { |
f4200391 | 3897 | if (vma_is_anonymous(vmf->vma)) |
82b0f8c3 | 3898 | return do_huge_pmd_anonymous_page(vmf); |
a2d58167 | 3899 | if (vmf->vma->vm_ops->huge_fault) |
c791ace1 | 3900 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD); |
b96375f7 MW |
3901 | return VM_FAULT_FALLBACK; |
3902 | } | |
3903 | ||
183f24aa GU |
3904 | /* `inline' is required to avoid gcc 4.1.2 build error */ |
3905 | static inline int wp_huge_pmd(struct vm_fault *vmf, pmd_t orig_pmd) | |
b96375f7 | 3906 | { |
82b0f8c3 JK |
3907 | if (vma_is_anonymous(vmf->vma)) |
3908 | return do_huge_pmd_wp_page(vmf, orig_pmd); | |
a2d58167 | 3909 | if (vmf->vma->vm_ops->huge_fault) |
c791ace1 | 3910 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD); |
af9e4d5f KS |
3911 | |
3912 | /* COW handled on pte level: split pmd */ | |
82b0f8c3 JK |
3913 | VM_BUG_ON_VMA(vmf->vma->vm_flags & VM_SHARED, vmf->vma); |
3914 | __split_huge_pmd(vmf->vma, vmf->pmd, vmf->address, false, NULL); | |
af9e4d5f | 3915 | |
b96375f7 MW |
3916 | return VM_FAULT_FALLBACK; |
3917 | } | |
3918 | ||
38e08854 LS |
3919 | static inline bool vma_is_accessible(struct vm_area_struct *vma) |
3920 | { | |
3921 | return vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE); | |
3922 | } | |
3923 | ||
a00cc7d9 MW |
3924 | static int create_huge_pud(struct vm_fault *vmf) |
3925 | { | |
3926 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
3927 | /* No support for anonymous transparent PUD pages yet */ | |
3928 | if (vma_is_anonymous(vmf->vma)) | |
3929 | return VM_FAULT_FALLBACK; | |
3930 | if (vmf->vma->vm_ops->huge_fault) | |
c791ace1 | 3931 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD); |
a00cc7d9 MW |
3932 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
3933 | return VM_FAULT_FALLBACK; | |
3934 | } | |
3935 | ||
3936 | static int wp_huge_pud(struct vm_fault *vmf, pud_t orig_pud) | |
3937 | { | |
3938 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
3939 | /* No support for anonymous transparent PUD pages yet */ | |
3940 | if (vma_is_anonymous(vmf->vma)) | |
3941 | return VM_FAULT_FALLBACK; | |
3942 | if (vmf->vma->vm_ops->huge_fault) | |
c791ace1 | 3943 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD); |
a00cc7d9 MW |
3944 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
3945 | return VM_FAULT_FALLBACK; | |
3946 | } | |
3947 | ||
1da177e4 LT |
3948 | /* |
3949 | * These routines also need to handle stuff like marking pages dirty | |
3950 | * and/or accessed for architectures that don't do it in hardware (most | |
3951 | * RISC architectures). The early dirtying is also good on the i386. | |
3952 | * | |
3953 | * There is also a hook called "update_mmu_cache()" that architectures | |
3954 | * with external mmu caches can use to update those (ie the Sparc or | |
3955 | * PowerPC hashed page tables that act as extended TLBs). | |
3956 | * | |
7267ec00 KS |
3957 | * We enter with non-exclusive mmap_sem (to exclude vma changes, but allow |
3958 | * concurrent faults). | |
9a95f3cf | 3959 | * |
7267ec00 KS |
3960 | * The mmap_sem may have been released depending on flags and our return value. |
3961 | * See filemap_fault() and __lock_page_or_retry(). | |
1da177e4 | 3962 | */ |
82b0f8c3 | 3963 | static int handle_pte_fault(struct vm_fault *vmf) |
1da177e4 LT |
3964 | { |
3965 | pte_t entry; | |
3966 | ||
82b0f8c3 | 3967 | if (unlikely(pmd_none(*vmf->pmd))) { |
7267ec00 KS |
3968 | /* |
3969 | * Leave __pte_alloc() until later: because vm_ops->fault may | |
3970 | * want to allocate huge page, and if we expose page table | |
3971 | * for an instant, it will be difficult to retract from | |
3972 | * concurrent faults and from rmap lookups. | |
3973 | */ | |
82b0f8c3 | 3974 | vmf->pte = NULL; |
7267ec00 KS |
3975 | } else { |
3976 | /* See comment in pte_alloc_one_map() */ | |
d0f0931d | 3977 | if (pmd_devmap_trans_unstable(vmf->pmd)) |
7267ec00 KS |
3978 | return 0; |
3979 | /* | |
3980 | * A regular pmd is established and it can't morph into a huge | |
3981 | * pmd from under us anymore at this point because we hold the | |
3982 | * mmap_sem read mode and khugepaged takes it in write mode. | |
3983 | * So now it's safe to run pte_offset_map(). | |
3984 | */ | |
82b0f8c3 | 3985 | vmf->pte = pte_offset_map(vmf->pmd, vmf->address); |
2994302b | 3986 | vmf->orig_pte = *vmf->pte; |
7267ec00 KS |
3987 | |
3988 | /* | |
3989 | * some architectures can have larger ptes than wordsize, | |
3990 | * e.g.ppc44x-defconfig has CONFIG_PTE_64BIT=y and | |
b03a0fe0 PM |
3991 | * CONFIG_32BIT=y, so READ_ONCE cannot guarantee atomic |
3992 | * accesses. The code below just needs a consistent view | |
3993 | * for the ifs and we later double check anyway with the | |
7267ec00 KS |
3994 | * ptl lock held. So here a barrier will do. |
3995 | */ | |
3996 | barrier(); | |
2994302b | 3997 | if (pte_none(vmf->orig_pte)) { |
82b0f8c3 JK |
3998 | pte_unmap(vmf->pte); |
3999 | vmf->pte = NULL; | |
65500d23 | 4000 | } |
1da177e4 LT |
4001 | } |
4002 | ||
82b0f8c3 JK |
4003 | if (!vmf->pte) { |
4004 | if (vma_is_anonymous(vmf->vma)) | |
4005 | return do_anonymous_page(vmf); | |
7267ec00 | 4006 | else |
82b0f8c3 | 4007 | return do_fault(vmf); |
7267ec00 KS |
4008 | } |
4009 | ||
2994302b JK |
4010 | if (!pte_present(vmf->orig_pte)) |
4011 | return do_swap_page(vmf); | |
7267ec00 | 4012 | |
2994302b JK |
4013 | if (pte_protnone(vmf->orig_pte) && vma_is_accessible(vmf->vma)) |
4014 | return do_numa_page(vmf); | |
d10e63f2 | 4015 | |
82b0f8c3 JK |
4016 | vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd); |
4017 | spin_lock(vmf->ptl); | |
2994302b | 4018 | entry = vmf->orig_pte; |
82b0f8c3 | 4019 | if (unlikely(!pte_same(*vmf->pte, entry))) |
8f4e2101 | 4020 | goto unlock; |
82b0f8c3 | 4021 | if (vmf->flags & FAULT_FLAG_WRITE) { |
f6f37321 | 4022 | if (!pte_write(entry)) |
2994302b | 4023 | return do_wp_page(vmf); |
1da177e4 LT |
4024 | entry = pte_mkdirty(entry); |
4025 | } | |
4026 | entry = pte_mkyoung(entry); | |
82b0f8c3 JK |
4027 | if (ptep_set_access_flags(vmf->vma, vmf->address, vmf->pte, entry, |
4028 | vmf->flags & FAULT_FLAG_WRITE)) { | |
4029 | update_mmu_cache(vmf->vma, vmf->address, vmf->pte); | |
1a44e149 AA |
4030 | } else { |
4031 | /* | |
4032 | * This is needed only for protection faults but the arch code | |
4033 | * is not yet telling us if this is a protection fault or not. | |
4034 | * This still avoids useless tlb flushes for .text page faults | |
4035 | * with threads. | |
4036 | */ | |
82b0f8c3 JK |
4037 | if (vmf->flags & FAULT_FLAG_WRITE) |
4038 | flush_tlb_fix_spurious_fault(vmf->vma, vmf->address); | |
1a44e149 | 4039 | } |
8f4e2101 | 4040 | unlock: |
82b0f8c3 | 4041 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
83c54070 | 4042 | return 0; |
1da177e4 LT |
4043 | } |
4044 | ||
4045 | /* | |
4046 | * By the time we get here, we already hold the mm semaphore | |
9a95f3cf PC |
4047 | * |
4048 | * The mmap_sem may have been released depending on flags and our | |
4049 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
1da177e4 | 4050 | */ |
dcddffd4 KS |
4051 | static int __handle_mm_fault(struct vm_area_struct *vma, unsigned long address, |
4052 | unsigned int flags) | |
1da177e4 | 4053 | { |
82b0f8c3 | 4054 | struct vm_fault vmf = { |
bae473a4 | 4055 | .vma = vma, |
1a29d85e | 4056 | .address = address & PAGE_MASK, |
bae473a4 | 4057 | .flags = flags, |
0721ec8b | 4058 | .pgoff = linear_page_index(vma, address), |
667240e0 | 4059 | .gfp_mask = __get_fault_gfp_mask(vma), |
bae473a4 | 4060 | }; |
fde26bed | 4061 | unsigned int dirty = flags & FAULT_FLAG_WRITE; |
dcddffd4 | 4062 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 4063 | pgd_t *pgd; |
c2febafc | 4064 | p4d_t *p4d; |
a2d58167 | 4065 | int ret; |
1da177e4 | 4066 | |
1da177e4 | 4067 | pgd = pgd_offset(mm, address); |
c2febafc KS |
4068 | p4d = p4d_alloc(mm, pgd, address); |
4069 | if (!p4d) | |
4070 | return VM_FAULT_OOM; | |
a00cc7d9 | 4071 | |
c2febafc | 4072 | vmf.pud = pud_alloc(mm, p4d, address); |
a00cc7d9 | 4073 | if (!vmf.pud) |
c74df32c | 4074 | return VM_FAULT_OOM; |
a00cc7d9 | 4075 | if (pud_none(*vmf.pud) && transparent_hugepage_enabled(vma)) { |
a00cc7d9 MW |
4076 | ret = create_huge_pud(&vmf); |
4077 | if (!(ret & VM_FAULT_FALLBACK)) | |
4078 | return ret; | |
4079 | } else { | |
4080 | pud_t orig_pud = *vmf.pud; | |
4081 | ||
4082 | barrier(); | |
4083 | if (pud_trans_huge(orig_pud) || pud_devmap(orig_pud)) { | |
a00cc7d9 | 4084 | |
a00cc7d9 MW |
4085 | /* NUMA case for anonymous PUDs would go here */ |
4086 | ||
f6f37321 | 4087 | if (dirty && !pud_write(orig_pud)) { |
a00cc7d9 MW |
4088 | ret = wp_huge_pud(&vmf, orig_pud); |
4089 | if (!(ret & VM_FAULT_FALLBACK)) | |
4090 | return ret; | |
4091 | } else { | |
4092 | huge_pud_set_accessed(&vmf, orig_pud); | |
4093 | return 0; | |
4094 | } | |
4095 | } | |
4096 | } | |
4097 | ||
4098 | vmf.pmd = pmd_alloc(mm, vmf.pud, address); | |
82b0f8c3 | 4099 | if (!vmf.pmd) |
c74df32c | 4100 | return VM_FAULT_OOM; |
82b0f8c3 | 4101 | if (pmd_none(*vmf.pmd) && transparent_hugepage_enabled(vma)) { |
a2d58167 | 4102 | ret = create_huge_pmd(&vmf); |
c0292554 KS |
4103 | if (!(ret & VM_FAULT_FALLBACK)) |
4104 | return ret; | |
71e3aac0 | 4105 | } else { |
82b0f8c3 | 4106 | pmd_t orig_pmd = *vmf.pmd; |
1f1d06c3 | 4107 | |
71e3aac0 | 4108 | barrier(); |
84c3fc4e ZY |
4109 | if (unlikely(is_swap_pmd(orig_pmd))) { |
4110 | VM_BUG_ON(thp_migration_supported() && | |
4111 | !is_pmd_migration_entry(orig_pmd)); | |
4112 | if (is_pmd_migration_entry(orig_pmd)) | |
4113 | pmd_migration_entry_wait(mm, vmf.pmd); | |
4114 | return 0; | |
4115 | } | |
5c7fb56e | 4116 | if (pmd_trans_huge(orig_pmd) || pmd_devmap(orig_pmd)) { |
38e08854 | 4117 | if (pmd_protnone(orig_pmd) && vma_is_accessible(vma)) |
82b0f8c3 | 4118 | return do_huge_pmd_numa_page(&vmf, orig_pmd); |
d10e63f2 | 4119 | |
f6f37321 | 4120 | if (dirty && !pmd_write(orig_pmd)) { |
82b0f8c3 | 4121 | ret = wp_huge_pmd(&vmf, orig_pmd); |
9845cbbd KS |
4122 | if (!(ret & VM_FAULT_FALLBACK)) |
4123 | return ret; | |
a1dd450b | 4124 | } else { |
82b0f8c3 | 4125 | huge_pmd_set_accessed(&vmf, orig_pmd); |
9845cbbd | 4126 | return 0; |
1f1d06c3 | 4127 | } |
71e3aac0 AA |
4128 | } |
4129 | } | |
4130 | ||
82b0f8c3 | 4131 | return handle_pte_fault(&vmf); |
1da177e4 LT |
4132 | } |
4133 | ||
9a95f3cf PC |
4134 | /* |
4135 | * By the time we get here, we already hold the mm semaphore | |
4136 | * | |
4137 | * The mmap_sem may have been released depending on flags and our | |
4138 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
4139 | */ | |
dcddffd4 KS |
4140 | int handle_mm_fault(struct vm_area_struct *vma, unsigned long address, |
4141 | unsigned int flags) | |
519e5247 JW |
4142 | { |
4143 | int ret; | |
4144 | ||
4145 | __set_current_state(TASK_RUNNING); | |
4146 | ||
4147 | count_vm_event(PGFAULT); | |
2262185c | 4148 | count_memcg_event_mm(vma->vm_mm, PGFAULT); |
519e5247 JW |
4149 | |
4150 | /* do counter updates before entering really critical section. */ | |
4151 | check_sync_rss_stat(current); | |
4152 | ||
de0c799b LD |
4153 | if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE, |
4154 | flags & FAULT_FLAG_INSTRUCTION, | |
4155 | flags & FAULT_FLAG_REMOTE)) | |
4156 | return VM_FAULT_SIGSEGV; | |
4157 | ||
519e5247 JW |
4158 | /* |
4159 | * Enable the memcg OOM handling for faults triggered in user | |
4160 | * space. Kernel faults are handled more gracefully. | |
4161 | */ | |
4162 | if (flags & FAULT_FLAG_USER) | |
49426420 | 4163 | mem_cgroup_oom_enable(); |
519e5247 | 4164 | |
bae473a4 KS |
4165 | if (unlikely(is_vm_hugetlb_page(vma))) |
4166 | ret = hugetlb_fault(vma->vm_mm, vma, address, flags); | |
4167 | else | |
4168 | ret = __handle_mm_fault(vma, address, flags); | |
519e5247 | 4169 | |
49426420 JW |
4170 | if (flags & FAULT_FLAG_USER) { |
4171 | mem_cgroup_oom_disable(); | |
166f61b9 TH |
4172 | /* |
4173 | * The task may have entered a memcg OOM situation but | |
4174 | * if the allocation error was handled gracefully (no | |
4175 | * VM_FAULT_OOM), there is no need to kill anything. | |
4176 | * Just clean up the OOM state peacefully. | |
4177 | */ | |
4178 | if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM)) | |
4179 | mem_cgroup_oom_synchronize(false); | |
49426420 | 4180 | } |
3812c8c8 | 4181 | |
519e5247 JW |
4182 | return ret; |
4183 | } | |
e1d6d01a | 4184 | EXPORT_SYMBOL_GPL(handle_mm_fault); |
519e5247 | 4185 | |
90eceff1 KS |
4186 | #ifndef __PAGETABLE_P4D_FOLDED |
4187 | /* | |
4188 | * Allocate p4d page table. | |
4189 | * We've already handled the fast-path in-line. | |
4190 | */ | |
4191 | int __p4d_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) | |
4192 | { | |
4193 | p4d_t *new = p4d_alloc_one(mm, address); | |
4194 | if (!new) | |
4195 | return -ENOMEM; | |
4196 | ||
4197 | smp_wmb(); /* See comment in __pte_alloc */ | |
4198 | ||
4199 | spin_lock(&mm->page_table_lock); | |
4200 | if (pgd_present(*pgd)) /* Another has populated it */ | |
4201 | p4d_free(mm, new); | |
4202 | else | |
4203 | pgd_populate(mm, pgd, new); | |
4204 | spin_unlock(&mm->page_table_lock); | |
4205 | return 0; | |
4206 | } | |
4207 | #endif /* __PAGETABLE_P4D_FOLDED */ | |
4208 | ||
1da177e4 LT |
4209 | #ifndef __PAGETABLE_PUD_FOLDED |
4210 | /* | |
4211 | * Allocate page upper directory. | |
872fec16 | 4212 | * We've already handled the fast-path in-line. |
1da177e4 | 4213 | */ |
c2febafc | 4214 | int __pud_alloc(struct mm_struct *mm, p4d_t *p4d, unsigned long address) |
1da177e4 | 4215 | { |
c74df32c HD |
4216 | pud_t *new = pud_alloc_one(mm, address); |
4217 | if (!new) | |
1bb3630e | 4218 | return -ENOMEM; |
1da177e4 | 4219 | |
362a61ad NP |
4220 | smp_wmb(); /* See comment in __pte_alloc */ |
4221 | ||
872fec16 | 4222 | spin_lock(&mm->page_table_lock); |
c2febafc | 4223 | #ifndef __ARCH_HAS_5LEVEL_HACK |
b4e98d9a KS |
4224 | if (!p4d_present(*p4d)) { |
4225 | mm_inc_nr_puds(mm); | |
c2febafc | 4226 | p4d_populate(mm, p4d, new); |
b4e98d9a | 4227 | } else /* Another has populated it */ |
5e541973 | 4228 | pud_free(mm, new); |
b4e98d9a KS |
4229 | #else |
4230 | if (!pgd_present(*p4d)) { | |
4231 | mm_inc_nr_puds(mm); | |
c2febafc | 4232 | pgd_populate(mm, p4d, new); |
b4e98d9a KS |
4233 | } else /* Another has populated it */ |
4234 | pud_free(mm, new); | |
c2febafc | 4235 | #endif /* __ARCH_HAS_5LEVEL_HACK */ |
c74df32c | 4236 | spin_unlock(&mm->page_table_lock); |
1bb3630e | 4237 | return 0; |
1da177e4 LT |
4238 | } |
4239 | #endif /* __PAGETABLE_PUD_FOLDED */ | |
4240 | ||
4241 | #ifndef __PAGETABLE_PMD_FOLDED | |
4242 | /* | |
4243 | * Allocate page middle directory. | |
872fec16 | 4244 | * We've already handled the fast-path in-line. |
1da177e4 | 4245 | */ |
1bb3630e | 4246 | int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) |
1da177e4 | 4247 | { |
a00cc7d9 | 4248 | spinlock_t *ptl; |
c74df32c HD |
4249 | pmd_t *new = pmd_alloc_one(mm, address); |
4250 | if (!new) | |
1bb3630e | 4251 | return -ENOMEM; |
1da177e4 | 4252 | |
362a61ad NP |
4253 | smp_wmb(); /* See comment in __pte_alloc */ |
4254 | ||
a00cc7d9 | 4255 | ptl = pud_lock(mm, pud); |
1da177e4 | 4256 | #ifndef __ARCH_HAS_4LEVEL_HACK |
dc6c9a35 KS |
4257 | if (!pud_present(*pud)) { |
4258 | mm_inc_nr_pmds(mm); | |
1bb3630e | 4259 | pud_populate(mm, pud, new); |
dc6c9a35 | 4260 | } else /* Another has populated it */ |
5e541973 | 4261 | pmd_free(mm, new); |
dc6c9a35 KS |
4262 | #else |
4263 | if (!pgd_present(*pud)) { | |
4264 | mm_inc_nr_pmds(mm); | |
1bb3630e | 4265 | pgd_populate(mm, pud, new); |
dc6c9a35 KS |
4266 | } else /* Another has populated it */ |
4267 | pmd_free(mm, new); | |
1da177e4 | 4268 | #endif /* __ARCH_HAS_4LEVEL_HACK */ |
a00cc7d9 | 4269 | spin_unlock(ptl); |
1bb3630e | 4270 | return 0; |
e0f39591 | 4271 | } |
1da177e4 LT |
4272 | #endif /* __PAGETABLE_PMD_FOLDED */ |
4273 | ||
09796395 | 4274 | static int __follow_pte_pmd(struct mm_struct *mm, unsigned long address, |
a4d1a885 JG |
4275 | unsigned long *start, unsigned long *end, |
4276 | pte_t **ptepp, pmd_t **pmdpp, spinlock_t **ptlp) | |
f8ad0f49 JW |
4277 | { |
4278 | pgd_t *pgd; | |
c2febafc | 4279 | p4d_t *p4d; |
f8ad0f49 JW |
4280 | pud_t *pud; |
4281 | pmd_t *pmd; | |
4282 | pte_t *ptep; | |
4283 | ||
4284 | pgd = pgd_offset(mm, address); | |
4285 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
4286 | goto out; | |
4287 | ||
c2febafc KS |
4288 | p4d = p4d_offset(pgd, address); |
4289 | if (p4d_none(*p4d) || unlikely(p4d_bad(*p4d))) | |
4290 | goto out; | |
4291 | ||
4292 | pud = pud_offset(p4d, address); | |
f8ad0f49 JW |
4293 | if (pud_none(*pud) || unlikely(pud_bad(*pud))) |
4294 | goto out; | |
4295 | ||
4296 | pmd = pmd_offset(pud, address); | |
f66055ab | 4297 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
f8ad0f49 | 4298 | |
09796395 RZ |
4299 | if (pmd_huge(*pmd)) { |
4300 | if (!pmdpp) | |
4301 | goto out; | |
4302 | ||
a4d1a885 JG |
4303 | if (start && end) { |
4304 | *start = address & PMD_MASK; | |
4305 | *end = *start + PMD_SIZE; | |
4306 | mmu_notifier_invalidate_range_start(mm, *start, *end); | |
4307 | } | |
09796395 RZ |
4308 | *ptlp = pmd_lock(mm, pmd); |
4309 | if (pmd_huge(*pmd)) { | |
4310 | *pmdpp = pmd; | |
4311 | return 0; | |
4312 | } | |
4313 | spin_unlock(*ptlp); | |
a4d1a885 JG |
4314 | if (start && end) |
4315 | mmu_notifier_invalidate_range_end(mm, *start, *end); | |
09796395 RZ |
4316 | } |
4317 | ||
4318 | if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) | |
f8ad0f49 JW |
4319 | goto out; |
4320 | ||
a4d1a885 JG |
4321 | if (start && end) { |
4322 | *start = address & PAGE_MASK; | |
4323 | *end = *start + PAGE_SIZE; | |
4324 | mmu_notifier_invalidate_range_start(mm, *start, *end); | |
4325 | } | |
f8ad0f49 | 4326 | ptep = pte_offset_map_lock(mm, pmd, address, ptlp); |
f8ad0f49 JW |
4327 | if (!pte_present(*ptep)) |
4328 | goto unlock; | |
4329 | *ptepp = ptep; | |
4330 | return 0; | |
4331 | unlock: | |
4332 | pte_unmap_unlock(ptep, *ptlp); | |
a4d1a885 JG |
4333 | if (start && end) |
4334 | mmu_notifier_invalidate_range_end(mm, *start, *end); | |
f8ad0f49 JW |
4335 | out: |
4336 | return -EINVAL; | |
4337 | } | |
4338 | ||
f729c8c9 RZ |
4339 | static inline int follow_pte(struct mm_struct *mm, unsigned long address, |
4340 | pte_t **ptepp, spinlock_t **ptlp) | |
1b36ba81 NK |
4341 | { |
4342 | int res; | |
4343 | ||
4344 | /* (void) is needed to make gcc happy */ | |
4345 | (void) __cond_lock(*ptlp, | |
a4d1a885 JG |
4346 | !(res = __follow_pte_pmd(mm, address, NULL, NULL, |
4347 | ptepp, NULL, ptlp))); | |
09796395 RZ |
4348 | return res; |
4349 | } | |
4350 | ||
4351 | int follow_pte_pmd(struct mm_struct *mm, unsigned long address, | |
a4d1a885 | 4352 | unsigned long *start, unsigned long *end, |
09796395 RZ |
4353 | pte_t **ptepp, pmd_t **pmdpp, spinlock_t **ptlp) |
4354 | { | |
4355 | int res; | |
4356 | ||
4357 | /* (void) is needed to make gcc happy */ | |
4358 | (void) __cond_lock(*ptlp, | |
a4d1a885 JG |
4359 | !(res = __follow_pte_pmd(mm, address, start, end, |
4360 | ptepp, pmdpp, ptlp))); | |
1b36ba81 NK |
4361 | return res; |
4362 | } | |
09796395 | 4363 | EXPORT_SYMBOL(follow_pte_pmd); |
1b36ba81 | 4364 | |
3b6748e2 JW |
4365 | /** |
4366 | * follow_pfn - look up PFN at a user virtual address | |
4367 | * @vma: memory mapping | |
4368 | * @address: user virtual address | |
4369 | * @pfn: location to store found PFN | |
4370 | * | |
4371 | * Only IO mappings and raw PFN mappings are allowed. | |
4372 | * | |
4373 | * Returns zero and the pfn at @pfn on success, -ve otherwise. | |
4374 | */ | |
4375 | int follow_pfn(struct vm_area_struct *vma, unsigned long address, | |
4376 | unsigned long *pfn) | |
4377 | { | |
4378 | int ret = -EINVAL; | |
4379 | spinlock_t *ptl; | |
4380 | pte_t *ptep; | |
4381 | ||
4382 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) | |
4383 | return ret; | |
4384 | ||
4385 | ret = follow_pte(vma->vm_mm, address, &ptep, &ptl); | |
4386 | if (ret) | |
4387 | return ret; | |
4388 | *pfn = pte_pfn(*ptep); | |
4389 | pte_unmap_unlock(ptep, ptl); | |
4390 | return 0; | |
4391 | } | |
4392 | EXPORT_SYMBOL(follow_pfn); | |
4393 | ||
28b2ee20 | 4394 | #ifdef CONFIG_HAVE_IOREMAP_PROT |
d87fe660 | 4395 | int follow_phys(struct vm_area_struct *vma, |
4396 | unsigned long address, unsigned int flags, | |
4397 | unsigned long *prot, resource_size_t *phys) | |
28b2ee20 | 4398 | { |
03668a4d | 4399 | int ret = -EINVAL; |
28b2ee20 RR |
4400 | pte_t *ptep, pte; |
4401 | spinlock_t *ptl; | |
28b2ee20 | 4402 | |
d87fe660 | 4403 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) |
4404 | goto out; | |
28b2ee20 | 4405 | |
03668a4d | 4406 | if (follow_pte(vma->vm_mm, address, &ptep, &ptl)) |
d87fe660 | 4407 | goto out; |
28b2ee20 | 4408 | pte = *ptep; |
03668a4d | 4409 | |
f6f37321 | 4410 | if ((flags & FOLL_WRITE) && !pte_write(pte)) |
28b2ee20 | 4411 | goto unlock; |
28b2ee20 RR |
4412 | |
4413 | *prot = pgprot_val(pte_pgprot(pte)); | |
03668a4d | 4414 | *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; |
28b2ee20 | 4415 | |
03668a4d | 4416 | ret = 0; |
28b2ee20 RR |
4417 | unlock: |
4418 | pte_unmap_unlock(ptep, ptl); | |
4419 | out: | |
d87fe660 | 4420 | return ret; |
28b2ee20 RR |
4421 | } |
4422 | ||
4423 | int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, | |
4424 | void *buf, int len, int write) | |
4425 | { | |
4426 | resource_size_t phys_addr; | |
4427 | unsigned long prot = 0; | |
2bc7273b | 4428 | void __iomem *maddr; |
28b2ee20 RR |
4429 | int offset = addr & (PAGE_SIZE-1); |
4430 | ||
d87fe660 | 4431 | if (follow_phys(vma, addr, write, &prot, &phys_addr)) |
28b2ee20 RR |
4432 | return -EINVAL; |
4433 | ||
9cb12d7b | 4434 | maddr = ioremap_prot(phys_addr, PAGE_ALIGN(len + offset), prot); |
7b90b5c2 | 4435 | if (!maddr) |
4436 | return -ENOMEM; | |
4437 | ||
28b2ee20 RR |
4438 | if (write) |
4439 | memcpy_toio(maddr + offset, buf, len); | |
4440 | else | |
4441 | memcpy_fromio(buf, maddr + offset, len); | |
4442 | iounmap(maddr); | |
4443 | ||
4444 | return len; | |
4445 | } | |
5a73633e | 4446 | EXPORT_SYMBOL_GPL(generic_access_phys); |
28b2ee20 RR |
4447 | #endif |
4448 | ||
0ec76a11 | 4449 | /* |
206cb636 SW |
4450 | * Access another process' address space as given in mm. If non-NULL, use the |
4451 | * given task for page fault accounting. | |
0ec76a11 | 4452 | */ |
84d77d3f | 4453 | int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, |
442486ec | 4454 | unsigned long addr, void *buf, int len, unsigned int gup_flags) |
0ec76a11 | 4455 | { |
0ec76a11 | 4456 | struct vm_area_struct *vma; |
0ec76a11 | 4457 | void *old_buf = buf; |
442486ec | 4458 | int write = gup_flags & FOLL_WRITE; |
0ec76a11 | 4459 | |
0ec76a11 | 4460 | down_read(&mm->mmap_sem); |
183ff22b | 4461 | /* ignore errors, just check how much was successfully transferred */ |
0ec76a11 DH |
4462 | while (len) { |
4463 | int bytes, ret, offset; | |
4464 | void *maddr; | |
28b2ee20 | 4465 | struct page *page = NULL; |
0ec76a11 | 4466 | |
1e987790 | 4467 | ret = get_user_pages_remote(tsk, mm, addr, 1, |
5b56d49f | 4468 | gup_flags, &page, &vma, NULL); |
28b2ee20 | 4469 | if (ret <= 0) { |
dbffcd03 RR |
4470 | #ifndef CONFIG_HAVE_IOREMAP_PROT |
4471 | break; | |
4472 | #else | |
28b2ee20 RR |
4473 | /* |
4474 | * Check if this is a VM_IO | VM_PFNMAP VMA, which | |
4475 | * we can access using slightly different code. | |
4476 | */ | |
28b2ee20 | 4477 | vma = find_vma(mm, addr); |
fe936dfc | 4478 | if (!vma || vma->vm_start > addr) |
28b2ee20 RR |
4479 | break; |
4480 | if (vma->vm_ops && vma->vm_ops->access) | |
4481 | ret = vma->vm_ops->access(vma, addr, buf, | |
4482 | len, write); | |
4483 | if (ret <= 0) | |
28b2ee20 RR |
4484 | break; |
4485 | bytes = ret; | |
dbffcd03 | 4486 | #endif |
0ec76a11 | 4487 | } else { |
28b2ee20 RR |
4488 | bytes = len; |
4489 | offset = addr & (PAGE_SIZE-1); | |
4490 | if (bytes > PAGE_SIZE-offset) | |
4491 | bytes = PAGE_SIZE-offset; | |
4492 | ||
4493 | maddr = kmap(page); | |
4494 | if (write) { | |
4495 | copy_to_user_page(vma, page, addr, | |
4496 | maddr + offset, buf, bytes); | |
4497 | set_page_dirty_lock(page); | |
4498 | } else { | |
4499 | copy_from_user_page(vma, page, addr, | |
4500 | buf, maddr + offset, bytes); | |
4501 | } | |
4502 | kunmap(page); | |
09cbfeaf | 4503 | put_page(page); |
0ec76a11 | 4504 | } |
0ec76a11 DH |
4505 | len -= bytes; |
4506 | buf += bytes; | |
4507 | addr += bytes; | |
4508 | } | |
4509 | up_read(&mm->mmap_sem); | |
0ec76a11 DH |
4510 | |
4511 | return buf - old_buf; | |
4512 | } | |
03252919 | 4513 | |
5ddd36b9 | 4514 | /** |
ae91dbfc | 4515 | * access_remote_vm - access another process' address space |
5ddd36b9 SW |
4516 | * @mm: the mm_struct of the target address space |
4517 | * @addr: start address to access | |
4518 | * @buf: source or destination buffer | |
4519 | * @len: number of bytes to transfer | |
6347e8d5 | 4520 | * @gup_flags: flags modifying lookup behaviour |
5ddd36b9 SW |
4521 | * |
4522 | * The caller must hold a reference on @mm. | |
4523 | */ | |
4524 | int access_remote_vm(struct mm_struct *mm, unsigned long addr, | |
6347e8d5 | 4525 | void *buf, int len, unsigned int gup_flags) |
5ddd36b9 | 4526 | { |
6347e8d5 | 4527 | return __access_remote_vm(NULL, mm, addr, buf, len, gup_flags); |
5ddd36b9 SW |
4528 | } |
4529 | ||
206cb636 SW |
4530 | /* |
4531 | * Access another process' address space. | |
4532 | * Source/target buffer must be kernel space, | |
4533 | * Do not walk the page table directly, use get_user_pages | |
4534 | */ | |
4535 | int access_process_vm(struct task_struct *tsk, unsigned long addr, | |
f307ab6d | 4536 | void *buf, int len, unsigned int gup_flags) |
206cb636 SW |
4537 | { |
4538 | struct mm_struct *mm; | |
4539 | int ret; | |
4540 | ||
4541 | mm = get_task_mm(tsk); | |
4542 | if (!mm) | |
4543 | return 0; | |
4544 | ||
f307ab6d | 4545 | ret = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags); |
442486ec | 4546 | |
206cb636 SW |
4547 | mmput(mm); |
4548 | ||
4549 | return ret; | |
4550 | } | |
fcd35857 | 4551 | EXPORT_SYMBOL_GPL(access_process_vm); |
206cb636 | 4552 | |
03252919 AK |
4553 | /* |
4554 | * Print the name of a VMA. | |
4555 | */ | |
4556 | void print_vma_addr(char *prefix, unsigned long ip) | |
4557 | { | |
4558 | struct mm_struct *mm = current->mm; | |
4559 | struct vm_area_struct *vma; | |
4560 | ||
e8bff74a | 4561 | /* |
0a7f682d | 4562 | * we might be running from an atomic context so we cannot sleep |
e8bff74a | 4563 | */ |
0a7f682d | 4564 | if (!down_read_trylock(&mm->mmap_sem)) |
e8bff74a IM |
4565 | return; |
4566 | ||
03252919 AK |
4567 | vma = find_vma(mm, ip); |
4568 | if (vma && vma->vm_file) { | |
4569 | struct file *f = vma->vm_file; | |
0a7f682d | 4570 | char *buf = (char *)__get_free_page(GFP_NOWAIT); |
03252919 | 4571 | if (buf) { |
2fbc57c5 | 4572 | char *p; |
03252919 | 4573 | |
9bf39ab2 | 4574 | p = file_path(f, buf, PAGE_SIZE); |
03252919 AK |
4575 | if (IS_ERR(p)) |
4576 | p = "?"; | |
2fbc57c5 | 4577 | printk("%s%s[%lx+%lx]", prefix, kbasename(p), |
03252919 AK |
4578 | vma->vm_start, |
4579 | vma->vm_end - vma->vm_start); | |
4580 | free_page((unsigned long)buf); | |
4581 | } | |
4582 | } | |
51a07e50 | 4583 | up_read(&mm->mmap_sem); |
03252919 | 4584 | } |
3ee1afa3 | 4585 | |
662bbcb2 | 4586 | #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP) |
9ec23531 | 4587 | void __might_fault(const char *file, int line) |
3ee1afa3 | 4588 | { |
95156f00 PZ |
4589 | /* |
4590 | * Some code (nfs/sunrpc) uses socket ops on kernel memory while | |
4591 | * holding the mmap_sem, this is safe because kernel memory doesn't | |
4592 | * get paged out, therefore we'll never actually fault, and the | |
4593 | * below annotations will generate false positives. | |
4594 | */ | |
db68ce10 | 4595 | if (uaccess_kernel()) |
95156f00 | 4596 | return; |
9ec23531 | 4597 | if (pagefault_disabled()) |
662bbcb2 | 4598 | return; |
9ec23531 DH |
4599 | __might_sleep(file, line, 0); |
4600 | #if defined(CONFIG_DEBUG_ATOMIC_SLEEP) | |
662bbcb2 | 4601 | if (current->mm) |
3ee1afa3 | 4602 | might_lock_read(¤t->mm->mmap_sem); |
9ec23531 | 4603 | #endif |
3ee1afa3 | 4604 | } |
9ec23531 | 4605 | EXPORT_SYMBOL(__might_fault); |
3ee1afa3 | 4606 | #endif |
47ad8475 AA |
4607 | |
4608 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) | |
4609 | static void clear_gigantic_page(struct page *page, | |
4610 | unsigned long addr, | |
4611 | unsigned int pages_per_huge_page) | |
4612 | { | |
4613 | int i; | |
4614 | struct page *p = page; | |
4615 | ||
4616 | might_sleep(); | |
4617 | for (i = 0; i < pages_per_huge_page; | |
4618 | i++, p = mem_map_next(p, page, i)) { | |
4619 | cond_resched(); | |
4620 | clear_user_highpage(p, addr + i * PAGE_SIZE); | |
4621 | } | |
4622 | } | |
4623 | void clear_huge_page(struct page *page, | |
c79b57e4 | 4624 | unsigned long addr_hint, unsigned int pages_per_huge_page) |
47ad8475 | 4625 | { |
c79b57e4 HY |
4626 | int i, n, base, l; |
4627 | unsigned long addr = addr_hint & | |
4628 | ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); | |
47ad8475 AA |
4629 | |
4630 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { | |
4631 | clear_gigantic_page(page, addr, pages_per_huge_page); | |
4632 | return; | |
4633 | } | |
4634 | ||
c79b57e4 | 4635 | /* Clear sub-page to access last to keep its cache lines hot */ |
47ad8475 | 4636 | might_sleep(); |
c79b57e4 HY |
4637 | n = (addr_hint - addr) / PAGE_SIZE; |
4638 | if (2 * n <= pages_per_huge_page) { | |
4639 | /* If sub-page to access in first half of huge page */ | |
4640 | base = 0; | |
4641 | l = n; | |
4642 | /* Clear sub-pages at the end of huge page */ | |
4643 | for (i = pages_per_huge_page - 1; i >= 2 * n; i--) { | |
4644 | cond_resched(); | |
4645 | clear_user_highpage(page + i, addr + i * PAGE_SIZE); | |
4646 | } | |
4647 | } else { | |
4648 | /* If sub-page to access in second half of huge page */ | |
4649 | base = pages_per_huge_page - 2 * (pages_per_huge_page - n); | |
4650 | l = pages_per_huge_page - n; | |
4651 | /* Clear sub-pages at the begin of huge page */ | |
4652 | for (i = 0; i < base; i++) { | |
4653 | cond_resched(); | |
4654 | clear_user_highpage(page + i, addr + i * PAGE_SIZE); | |
4655 | } | |
4656 | } | |
4657 | /* | |
4658 | * Clear remaining sub-pages in left-right-left-right pattern | |
4659 | * towards the sub-page to access | |
4660 | */ | |
4661 | for (i = 0; i < l; i++) { | |
4662 | int left_idx = base + i; | |
4663 | int right_idx = base + 2 * l - 1 - i; | |
4664 | ||
4665 | cond_resched(); | |
4666 | clear_user_highpage(page + left_idx, | |
4667 | addr + left_idx * PAGE_SIZE); | |
47ad8475 | 4668 | cond_resched(); |
c79b57e4 HY |
4669 | clear_user_highpage(page + right_idx, |
4670 | addr + right_idx * PAGE_SIZE); | |
47ad8475 AA |
4671 | } |
4672 | } | |
4673 | ||
4674 | static void copy_user_gigantic_page(struct page *dst, struct page *src, | |
4675 | unsigned long addr, | |
4676 | struct vm_area_struct *vma, | |
4677 | unsigned int pages_per_huge_page) | |
4678 | { | |
4679 | int i; | |
4680 | struct page *dst_base = dst; | |
4681 | struct page *src_base = src; | |
4682 | ||
4683 | for (i = 0; i < pages_per_huge_page; ) { | |
4684 | cond_resched(); | |
4685 | copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma); | |
4686 | ||
4687 | i++; | |
4688 | dst = mem_map_next(dst, dst_base, i); | |
4689 | src = mem_map_next(src, src_base, i); | |
4690 | } | |
4691 | } | |
4692 | ||
4693 | void copy_user_huge_page(struct page *dst, struct page *src, | |
4694 | unsigned long addr, struct vm_area_struct *vma, | |
4695 | unsigned int pages_per_huge_page) | |
4696 | { | |
4697 | int i; | |
4698 | ||
4699 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { | |
4700 | copy_user_gigantic_page(dst, src, addr, vma, | |
4701 | pages_per_huge_page); | |
4702 | return; | |
4703 | } | |
4704 | ||
4705 | might_sleep(); | |
4706 | for (i = 0; i < pages_per_huge_page; i++) { | |
4707 | cond_resched(); | |
4708 | copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE, vma); | |
4709 | } | |
4710 | } | |
fa4d75c1 MK |
4711 | |
4712 | long copy_huge_page_from_user(struct page *dst_page, | |
4713 | const void __user *usr_src, | |
810a56b9 MK |
4714 | unsigned int pages_per_huge_page, |
4715 | bool allow_pagefault) | |
fa4d75c1 MK |
4716 | { |
4717 | void *src = (void *)usr_src; | |
4718 | void *page_kaddr; | |
4719 | unsigned long i, rc = 0; | |
4720 | unsigned long ret_val = pages_per_huge_page * PAGE_SIZE; | |
4721 | ||
4722 | for (i = 0; i < pages_per_huge_page; i++) { | |
810a56b9 MK |
4723 | if (allow_pagefault) |
4724 | page_kaddr = kmap(dst_page + i); | |
4725 | else | |
4726 | page_kaddr = kmap_atomic(dst_page + i); | |
fa4d75c1 MK |
4727 | rc = copy_from_user(page_kaddr, |
4728 | (const void __user *)(src + i * PAGE_SIZE), | |
4729 | PAGE_SIZE); | |
810a56b9 MK |
4730 | if (allow_pagefault) |
4731 | kunmap(dst_page + i); | |
4732 | else | |
4733 | kunmap_atomic(page_kaddr); | |
fa4d75c1 MK |
4734 | |
4735 | ret_val -= (PAGE_SIZE - rc); | |
4736 | if (rc) | |
4737 | break; | |
4738 | ||
4739 | cond_resched(); | |
4740 | } | |
4741 | return ret_val; | |
4742 | } | |
47ad8475 | 4743 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */ |
49076ec2 | 4744 | |
40b64acd | 4745 | #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS |
b35f1819 KS |
4746 | |
4747 | static struct kmem_cache *page_ptl_cachep; | |
4748 | ||
4749 | void __init ptlock_cache_init(void) | |
4750 | { | |
4751 | page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0, | |
4752 | SLAB_PANIC, NULL); | |
4753 | } | |
4754 | ||
539edb58 | 4755 | bool ptlock_alloc(struct page *page) |
49076ec2 KS |
4756 | { |
4757 | spinlock_t *ptl; | |
4758 | ||
b35f1819 | 4759 | ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL); |
49076ec2 KS |
4760 | if (!ptl) |
4761 | return false; | |
539edb58 | 4762 | page->ptl = ptl; |
49076ec2 KS |
4763 | return true; |
4764 | } | |
4765 | ||
539edb58 | 4766 | void ptlock_free(struct page *page) |
49076ec2 | 4767 | { |
b35f1819 | 4768 | kmem_cache_free(page_ptl_cachep, page->ptl); |
49076ec2 KS |
4769 | } |
4770 | #endif |