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/swapops.h> |
63 | #include <linux/elf.h> | |
5a0e3ad6 | 64 | #include <linux/gfp.h> |
4daae3b4 | 65 | #include <linux/migrate.h> |
2fbc57c5 | 66 | #include <linux/string.h> |
0abdd7a8 | 67 | #include <linux/dma-debug.h> |
1592eef0 | 68 | #include <linux/debugfs.h> |
6b251fc9 | 69 | #include <linux/userfaultfd_k.h> |
bc2466e4 | 70 | #include <linux/dax.h> |
6b31d595 | 71 | #include <linux/oom.h> |
1da177e4 | 72 | |
6952b61d | 73 | #include <asm/io.h> |
33a709b2 | 74 | #include <asm/mmu_context.h> |
1da177e4 | 75 | #include <asm/pgalloc.h> |
7c0f6ba6 | 76 | #include <linux/uaccess.h> |
1da177e4 LT |
77 | #include <asm/tlb.h> |
78 | #include <asm/tlbflush.h> | |
79 | #include <asm/pgtable.h> | |
80 | ||
42b77728 JB |
81 | #include "internal.h" |
82 | ||
af27d940 | 83 | #if defined(LAST_CPUPID_NOT_IN_PAGE_FLAGS) && !defined(CONFIG_COMPILE_TEST) |
90572890 | 84 | #warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid. |
75980e97 PZ |
85 | #endif |
86 | ||
d41dee36 | 87 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
88 | /* use the per-pgdat data instead for discontigmem - mbligh */ |
89 | unsigned long max_mapnr; | |
1da177e4 | 90 | EXPORT_SYMBOL(max_mapnr); |
166f61b9 TH |
91 | |
92 | struct page *mem_map; | |
1da177e4 LT |
93 | EXPORT_SYMBOL(mem_map); |
94 | #endif | |
95 | ||
1da177e4 LT |
96 | /* |
97 | * A number of key systems in x86 including ioremap() rely on the assumption | |
98 | * that high_memory defines the upper bound on direct map memory, then end | |
99 | * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and | |
100 | * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL | |
101 | * and ZONE_HIGHMEM. | |
102 | */ | |
166f61b9 | 103 | void *high_memory; |
1da177e4 | 104 | EXPORT_SYMBOL(high_memory); |
1da177e4 | 105 | |
32a93233 IM |
106 | /* |
107 | * Randomize the address space (stacks, mmaps, brk, etc.). | |
108 | * | |
109 | * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization, | |
110 | * as ancient (libc5 based) binaries can segfault. ) | |
111 | */ | |
112 | int randomize_va_space __read_mostly = | |
113 | #ifdef CONFIG_COMPAT_BRK | |
114 | 1; | |
115 | #else | |
116 | 2; | |
117 | #endif | |
a62eaf15 AK |
118 | |
119 | static int __init disable_randmaps(char *s) | |
120 | { | |
121 | randomize_va_space = 0; | |
9b41046c | 122 | return 1; |
a62eaf15 AK |
123 | } |
124 | __setup("norandmaps", disable_randmaps); | |
125 | ||
62eede62 | 126 | unsigned long zero_pfn __read_mostly; |
0b70068e AB |
127 | EXPORT_SYMBOL(zero_pfn); |
128 | ||
166f61b9 TH |
129 | unsigned long highest_memmap_pfn __read_mostly; |
130 | ||
a13ea5b7 HD |
131 | /* |
132 | * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init() | |
133 | */ | |
134 | static int __init init_zero_pfn(void) | |
135 | { | |
136 | zero_pfn = page_to_pfn(ZERO_PAGE(0)); | |
137 | return 0; | |
138 | } | |
139 | core_initcall(init_zero_pfn); | |
a62eaf15 | 140 | |
d559db08 | 141 | |
34e55232 KH |
142 | #if defined(SPLIT_RSS_COUNTING) |
143 | ||
ea48cf78 | 144 | void sync_mm_rss(struct mm_struct *mm) |
34e55232 KH |
145 | { |
146 | int i; | |
147 | ||
148 | for (i = 0; i < NR_MM_COUNTERS; i++) { | |
05af2e10 DR |
149 | if (current->rss_stat.count[i]) { |
150 | add_mm_counter(mm, i, current->rss_stat.count[i]); | |
151 | current->rss_stat.count[i] = 0; | |
34e55232 KH |
152 | } |
153 | } | |
05af2e10 | 154 | current->rss_stat.events = 0; |
34e55232 KH |
155 | } |
156 | ||
157 | static void add_mm_counter_fast(struct mm_struct *mm, int member, int val) | |
158 | { | |
159 | struct task_struct *task = current; | |
160 | ||
161 | if (likely(task->mm == mm)) | |
162 | task->rss_stat.count[member] += val; | |
163 | else | |
164 | add_mm_counter(mm, member, val); | |
165 | } | |
166 | #define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1) | |
167 | #define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1) | |
168 | ||
169 | /* sync counter once per 64 page faults */ | |
170 | #define TASK_RSS_EVENTS_THRESH (64) | |
171 | static void check_sync_rss_stat(struct task_struct *task) | |
172 | { | |
173 | if (unlikely(task != current)) | |
174 | return; | |
175 | if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH)) | |
ea48cf78 | 176 | sync_mm_rss(task->mm); |
34e55232 | 177 | } |
9547d01b | 178 | #else /* SPLIT_RSS_COUNTING */ |
34e55232 KH |
179 | |
180 | #define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member) | |
181 | #define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member) | |
182 | ||
183 | static void check_sync_rss_stat(struct task_struct *task) | |
184 | { | |
185 | } | |
186 | ||
9547d01b PZ |
187 | #endif /* SPLIT_RSS_COUNTING */ |
188 | ||
189 | #ifdef HAVE_GENERIC_MMU_GATHER | |
190 | ||
ca1d6c7d | 191 | static bool tlb_next_batch(struct mmu_gather *tlb) |
9547d01b PZ |
192 | { |
193 | struct mmu_gather_batch *batch; | |
194 | ||
195 | batch = tlb->active; | |
196 | if (batch->next) { | |
197 | tlb->active = batch->next; | |
ca1d6c7d | 198 | return true; |
9547d01b PZ |
199 | } |
200 | ||
53a59fc6 | 201 | if (tlb->batch_count == MAX_GATHER_BATCH_COUNT) |
ca1d6c7d | 202 | return false; |
53a59fc6 | 203 | |
9547d01b PZ |
204 | batch = (void *)__get_free_pages(GFP_NOWAIT | __GFP_NOWARN, 0); |
205 | if (!batch) | |
ca1d6c7d | 206 | return false; |
9547d01b | 207 | |
53a59fc6 | 208 | tlb->batch_count++; |
9547d01b PZ |
209 | batch->next = NULL; |
210 | batch->nr = 0; | |
211 | batch->max = MAX_GATHER_BATCH; | |
212 | ||
213 | tlb->active->next = batch; | |
214 | tlb->active = batch; | |
215 | ||
ca1d6c7d | 216 | return true; |
9547d01b PZ |
217 | } |
218 | ||
56236a59 MK |
219 | void arch_tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, |
220 | unsigned long start, unsigned long end) | |
9547d01b PZ |
221 | { |
222 | tlb->mm = mm; | |
223 | ||
2b047252 LT |
224 | /* Is it from 0 to ~0? */ |
225 | tlb->fullmm = !(start | (end+1)); | |
1de14c3c | 226 | tlb->need_flush_all = 0; |
9547d01b PZ |
227 | tlb->local.next = NULL; |
228 | tlb->local.nr = 0; | |
229 | tlb->local.max = ARRAY_SIZE(tlb->__pages); | |
230 | tlb->active = &tlb->local; | |
53a59fc6 | 231 | tlb->batch_count = 0; |
9547d01b PZ |
232 | |
233 | #ifdef CONFIG_HAVE_RCU_TABLE_FREE | |
234 | tlb->batch = NULL; | |
235 | #endif | |
e77b0852 | 236 | tlb->page_size = 0; |
fb7332a9 WD |
237 | |
238 | __tlb_reset_range(tlb); | |
9547d01b PZ |
239 | } |
240 | ||
1cf35d47 | 241 | static void tlb_flush_mmu_tlbonly(struct mmu_gather *tlb) |
9547d01b | 242 | { |
721c21c1 WD |
243 | if (!tlb->end) |
244 | return; | |
245 | ||
9547d01b | 246 | tlb_flush(tlb); |
34ee645e | 247 | mmu_notifier_invalidate_range(tlb->mm, tlb->start, tlb->end); |
9547d01b PZ |
248 | #ifdef CONFIG_HAVE_RCU_TABLE_FREE |
249 | tlb_table_flush(tlb); | |
34e55232 | 250 | #endif |
fb7332a9 | 251 | __tlb_reset_range(tlb); |
1cf35d47 LT |
252 | } |
253 | ||
254 | static void tlb_flush_mmu_free(struct mmu_gather *tlb) | |
255 | { | |
256 | struct mmu_gather_batch *batch; | |
34e55232 | 257 | |
721c21c1 | 258 | for (batch = &tlb->local; batch && batch->nr; batch = batch->next) { |
9547d01b PZ |
259 | free_pages_and_swap_cache(batch->pages, batch->nr); |
260 | batch->nr = 0; | |
261 | } | |
262 | tlb->active = &tlb->local; | |
263 | } | |
264 | ||
1cf35d47 LT |
265 | void tlb_flush_mmu(struct mmu_gather *tlb) |
266 | { | |
1cf35d47 LT |
267 | tlb_flush_mmu_tlbonly(tlb); |
268 | tlb_flush_mmu_free(tlb); | |
269 | } | |
270 | ||
9547d01b PZ |
271 | /* tlb_finish_mmu |
272 | * Called at the end of the shootdown operation to free up any resources | |
273 | * that were required. | |
274 | */ | |
56236a59 | 275 | void arch_tlb_finish_mmu(struct mmu_gather *tlb, |
99baac21 | 276 | unsigned long start, unsigned long end, bool force) |
9547d01b PZ |
277 | { |
278 | struct mmu_gather_batch *batch, *next; | |
279 | ||
99baac21 MK |
280 | if (force) |
281 | __tlb_adjust_range(tlb, start, end - start); | |
282 | ||
9547d01b PZ |
283 | tlb_flush_mmu(tlb); |
284 | ||
285 | /* keep the page table cache within bounds */ | |
286 | check_pgt_cache(); | |
287 | ||
288 | for (batch = tlb->local.next; batch; batch = next) { | |
289 | next = batch->next; | |
290 | free_pages((unsigned long)batch, 0); | |
291 | } | |
292 | tlb->local.next = NULL; | |
293 | } | |
294 | ||
295 | /* __tlb_remove_page | |
296 | * Must perform the equivalent to __free_pte(pte_get_and_clear(ptep)), while | |
297 | * handling the additional races in SMP caused by other CPUs caching valid | |
298 | * mappings in their TLBs. Returns the number of free page slots left. | |
299 | * When out of page slots we must call tlb_flush_mmu(). | |
e9d55e15 | 300 | *returns true if the caller should flush. |
9547d01b | 301 | */ |
e77b0852 | 302 | bool __tlb_remove_page_size(struct mmu_gather *tlb, struct page *page, int page_size) |
9547d01b PZ |
303 | { |
304 | struct mmu_gather_batch *batch; | |
305 | ||
fb7332a9 | 306 | VM_BUG_ON(!tlb->end); |
692a68c1 | 307 | VM_WARN_ON(tlb->page_size != page_size); |
e77b0852 | 308 | |
9547d01b | 309 | batch = tlb->active; |
692a68c1 AK |
310 | /* |
311 | * Add the page and check if we are full. If so | |
312 | * force a flush. | |
313 | */ | |
314 | batch->pages[batch->nr++] = page; | |
9547d01b PZ |
315 | if (batch->nr == batch->max) { |
316 | if (!tlb_next_batch(tlb)) | |
e9d55e15 | 317 | return true; |
0b43c3aa | 318 | batch = tlb->active; |
9547d01b | 319 | } |
309381fe | 320 | VM_BUG_ON_PAGE(batch->nr > batch->max, page); |
9547d01b | 321 | |
e9d55e15 | 322 | return false; |
9547d01b PZ |
323 | } |
324 | ||
325 | #endif /* HAVE_GENERIC_MMU_GATHER */ | |
326 | ||
26723911 PZ |
327 | #ifdef CONFIG_HAVE_RCU_TABLE_FREE |
328 | ||
26723911 PZ |
329 | static void tlb_remove_table_smp_sync(void *arg) |
330 | { | |
2ff6ddf1 RR |
331 | struct mm_struct __maybe_unused *mm = arg; |
332 | /* | |
333 | * On most architectures this does nothing. Simply delivering the | |
334 | * interrupt is enough to prevent races with software page table | |
335 | * walking like that done in get_user_pages_fast. | |
336 | * | |
337 | * See the comment near struct mmu_table_batch. | |
338 | */ | |
339 | tlb_flush_remove_tables_local(mm); | |
26723911 PZ |
340 | } |
341 | ||
2ff6ddf1 | 342 | static void tlb_remove_table_one(void *table, struct mmu_gather *tlb) |
26723911 PZ |
343 | { |
344 | /* | |
345 | * This isn't an RCU grace period and hence the page-tables cannot be | |
346 | * assumed to be actually RCU-freed. | |
347 | * | |
348 | * It is however sufficient for software page-table walkers that rely on | |
349 | * IRQ disabling. See the comment near struct mmu_table_batch. | |
350 | */ | |
2ff6ddf1 | 351 | smp_call_function(tlb_remove_table_smp_sync, tlb->mm, 1); |
26723911 PZ |
352 | __tlb_remove_table(table); |
353 | } | |
354 | ||
355 | static void tlb_remove_table_rcu(struct rcu_head *head) | |
356 | { | |
357 | struct mmu_table_batch *batch; | |
358 | int i; | |
359 | ||
360 | batch = container_of(head, struct mmu_table_batch, rcu); | |
361 | ||
362 | for (i = 0; i < batch->nr; i++) | |
363 | __tlb_remove_table(batch->tables[i]); | |
364 | ||
365 | free_page((unsigned long)batch); | |
366 | } | |
367 | ||
368 | void tlb_table_flush(struct mmu_gather *tlb) | |
369 | { | |
370 | struct mmu_table_batch **batch = &tlb->batch; | |
371 | ||
2ff6ddf1 RR |
372 | tlb_flush_remove_tables(tlb->mm); |
373 | ||
26723911 PZ |
374 | if (*batch) { |
375 | call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu); | |
376 | *batch = NULL; | |
377 | } | |
378 | } | |
379 | ||
380 | void tlb_remove_table(struct mmu_gather *tlb, void *table) | |
381 | { | |
382 | struct mmu_table_batch **batch = &tlb->batch; | |
383 | ||
26723911 PZ |
384 | /* |
385 | * When there's less then two users of this mm there cannot be a | |
386 | * concurrent page-table walk. | |
387 | */ | |
388 | if (atomic_read(&tlb->mm->mm_users) < 2) { | |
389 | __tlb_remove_table(table); | |
390 | return; | |
391 | } | |
392 | ||
393 | if (*batch == NULL) { | |
394 | *batch = (struct mmu_table_batch *)__get_free_page(GFP_NOWAIT | __GFP_NOWARN); | |
395 | if (*batch == NULL) { | |
2ff6ddf1 | 396 | tlb_remove_table_one(table, tlb); |
26723911 PZ |
397 | return; |
398 | } | |
399 | (*batch)->nr = 0; | |
400 | } | |
401 | (*batch)->tables[(*batch)->nr++] = table; | |
402 | if ((*batch)->nr == MAX_TABLE_BATCH) | |
403 | tlb_table_flush(tlb); | |
404 | } | |
405 | ||
9547d01b | 406 | #endif /* CONFIG_HAVE_RCU_TABLE_FREE */ |
26723911 | 407 | |
ef549e13 MR |
408 | /** |
409 | * tlb_gather_mmu - initialize an mmu_gather structure for page-table tear-down | |
410 | * @tlb: the mmu_gather structure to initialize | |
411 | * @mm: the mm_struct of the target address space | |
412 | * @start: start of the region that will be removed from the page-table | |
413 | * @end: end of the region that will be removed from the page-table | |
414 | * | |
415 | * Called to initialize an (on-stack) mmu_gather structure for page-table | |
416 | * tear-down from @mm. The @start and @end are set to 0 and -1 | |
417 | * respectively when @mm is without users and we're going to destroy | |
418 | * the full address space (exit/execve). | |
56236a59 MK |
419 | */ |
420 | void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, | |
421 | unsigned long start, unsigned long end) | |
422 | { | |
423 | arch_tlb_gather_mmu(tlb, mm, start, end); | |
99baac21 | 424 | inc_tlb_flush_pending(tlb->mm); |
56236a59 MK |
425 | } |
426 | ||
427 | void tlb_finish_mmu(struct mmu_gather *tlb, | |
428 | unsigned long start, unsigned long end) | |
429 | { | |
99baac21 MK |
430 | /* |
431 | * If there are parallel threads are doing PTE changes on same range | |
432 | * under non-exclusive lock(e.g., mmap_sem read-side) but defer TLB | |
433 | * flush by batching, a thread has stable TLB entry can fail to flush | |
434 | * the TLB by observing pte_none|!pte_dirty, for example so flush TLB | |
435 | * forcefully if we detect parallel PTE batching threads. | |
436 | */ | |
437 | bool force = mm_tlb_flush_nested(tlb->mm); | |
438 | ||
439 | arch_tlb_finish_mmu(tlb, start, end, force); | |
440 | dec_tlb_flush_pending(tlb->mm); | |
56236a59 MK |
441 | } |
442 | ||
1da177e4 LT |
443 | /* |
444 | * Note: this doesn't free the actual pages themselves. That | |
445 | * has been handled earlier when unmapping all the memory regions. | |
446 | */ | |
9e1b32ca BH |
447 | static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd, |
448 | unsigned long addr) | |
1da177e4 | 449 | { |
2f569afd | 450 | pgtable_t token = pmd_pgtable(*pmd); |
e0da382c | 451 | pmd_clear(pmd); |
9e1b32ca | 452 | pte_free_tlb(tlb, token, addr); |
c4812909 | 453 | mm_dec_nr_ptes(tlb->mm); |
1da177e4 LT |
454 | } |
455 | ||
e0da382c HD |
456 | static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud, |
457 | unsigned long addr, unsigned long end, | |
458 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
459 | { |
460 | pmd_t *pmd; | |
461 | unsigned long next; | |
e0da382c | 462 | unsigned long start; |
1da177e4 | 463 | |
e0da382c | 464 | start = addr; |
1da177e4 | 465 | pmd = pmd_offset(pud, addr); |
1da177e4 LT |
466 | do { |
467 | next = pmd_addr_end(addr, end); | |
468 | if (pmd_none_or_clear_bad(pmd)) | |
469 | continue; | |
9e1b32ca | 470 | free_pte_range(tlb, pmd, addr); |
1da177e4 LT |
471 | } while (pmd++, addr = next, addr != end); |
472 | ||
e0da382c HD |
473 | start &= PUD_MASK; |
474 | if (start < floor) | |
475 | return; | |
476 | if (ceiling) { | |
477 | ceiling &= PUD_MASK; | |
478 | if (!ceiling) | |
479 | return; | |
1da177e4 | 480 | } |
e0da382c HD |
481 | if (end - 1 > ceiling - 1) |
482 | return; | |
483 | ||
484 | pmd = pmd_offset(pud, start); | |
485 | pud_clear(pud); | |
9e1b32ca | 486 | pmd_free_tlb(tlb, pmd, start); |
dc6c9a35 | 487 | mm_dec_nr_pmds(tlb->mm); |
1da177e4 LT |
488 | } |
489 | ||
c2febafc | 490 | static inline void free_pud_range(struct mmu_gather *tlb, p4d_t *p4d, |
e0da382c HD |
491 | unsigned long addr, unsigned long end, |
492 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
493 | { |
494 | pud_t *pud; | |
495 | unsigned long next; | |
e0da382c | 496 | unsigned long start; |
1da177e4 | 497 | |
e0da382c | 498 | start = addr; |
c2febafc | 499 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
500 | do { |
501 | next = pud_addr_end(addr, end); | |
502 | if (pud_none_or_clear_bad(pud)) | |
503 | continue; | |
e0da382c | 504 | free_pmd_range(tlb, pud, addr, next, floor, ceiling); |
1da177e4 LT |
505 | } while (pud++, addr = next, addr != end); |
506 | ||
c2febafc KS |
507 | start &= P4D_MASK; |
508 | if (start < floor) | |
509 | return; | |
510 | if (ceiling) { | |
511 | ceiling &= P4D_MASK; | |
512 | if (!ceiling) | |
513 | return; | |
514 | } | |
515 | if (end - 1 > ceiling - 1) | |
516 | return; | |
517 | ||
518 | pud = pud_offset(p4d, start); | |
519 | p4d_clear(p4d); | |
520 | pud_free_tlb(tlb, pud, start); | |
b4e98d9a | 521 | mm_dec_nr_puds(tlb->mm); |
c2febafc KS |
522 | } |
523 | ||
524 | static inline void free_p4d_range(struct mmu_gather *tlb, pgd_t *pgd, | |
525 | unsigned long addr, unsigned long end, | |
526 | unsigned long floor, unsigned long ceiling) | |
527 | { | |
528 | p4d_t *p4d; | |
529 | unsigned long next; | |
530 | unsigned long start; | |
531 | ||
532 | start = addr; | |
533 | p4d = p4d_offset(pgd, addr); | |
534 | do { | |
535 | next = p4d_addr_end(addr, end); | |
536 | if (p4d_none_or_clear_bad(p4d)) | |
537 | continue; | |
538 | free_pud_range(tlb, p4d, addr, next, floor, ceiling); | |
539 | } while (p4d++, addr = next, addr != end); | |
540 | ||
e0da382c HD |
541 | start &= PGDIR_MASK; |
542 | if (start < floor) | |
543 | return; | |
544 | if (ceiling) { | |
545 | ceiling &= PGDIR_MASK; | |
546 | if (!ceiling) | |
547 | return; | |
1da177e4 | 548 | } |
e0da382c HD |
549 | if (end - 1 > ceiling - 1) |
550 | return; | |
551 | ||
c2febafc | 552 | p4d = p4d_offset(pgd, start); |
e0da382c | 553 | pgd_clear(pgd); |
c2febafc | 554 | p4d_free_tlb(tlb, p4d, start); |
1da177e4 LT |
555 | } |
556 | ||
557 | /* | |
e0da382c | 558 | * This function frees user-level page tables of a process. |
1da177e4 | 559 | */ |
42b77728 | 560 | void free_pgd_range(struct mmu_gather *tlb, |
e0da382c HD |
561 | unsigned long addr, unsigned long end, |
562 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
563 | { |
564 | pgd_t *pgd; | |
565 | unsigned long next; | |
e0da382c HD |
566 | |
567 | /* | |
568 | * The next few lines have given us lots of grief... | |
569 | * | |
570 | * Why are we testing PMD* at this top level? Because often | |
571 | * there will be no work to do at all, and we'd prefer not to | |
572 | * go all the way down to the bottom just to discover that. | |
573 | * | |
574 | * Why all these "- 1"s? Because 0 represents both the bottom | |
575 | * of the address space and the top of it (using -1 for the | |
576 | * top wouldn't help much: the masks would do the wrong thing). | |
577 | * The rule is that addr 0 and floor 0 refer to the bottom of | |
578 | * the address space, but end 0 and ceiling 0 refer to the top | |
579 | * Comparisons need to use "end - 1" and "ceiling - 1" (though | |
580 | * that end 0 case should be mythical). | |
581 | * | |
582 | * Wherever addr is brought up or ceiling brought down, we must | |
583 | * be careful to reject "the opposite 0" before it confuses the | |
584 | * subsequent tests. But what about where end is brought down | |
585 | * by PMD_SIZE below? no, end can't go down to 0 there. | |
586 | * | |
587 | * Whereas we round start (addr) and ceiling down, by different | |
588 | * masks at different levels, in order to test whether a table | |
589 | * now has no other vmas using it, so can be freed, we don't | |
590 | * bother to round floor or end up - the tests don't need that. | |
591 | */ | |
1da177e4 | 592 | |
e0da382c HD |
593 | addr &= PMD_MASK; |
594 | if (addr < floor) { | |
595 | addr += PMD_SIZE; | |
596 | if (!addr) | |
597 | return; | |
598 | } | |
599 | if (ceiling) { | |
600 | ceiling &= PMD_MASK; | |
601 | if (!ceiling) | |
602 | return; | |
603 | } | |
604 | if (end - 1 > ceiling - 1) | |
605 | end -= PMD_SIZE; | |
606 | if (addr > end - 1) | |
607 | return; | |
07e32661 AK |
608 | /* |
609 | * We add page table cache pages with PAGE_SIZE, | |
610 | * (see pte_free_tlb()), flush the tlb if we need | |
611 | */ | |
612 | tlb_remove_check_page_size_change(tlb, PAGE_SIZE); | |
42b77728 | 613 | pgd = pgd_offset(tlb->mm, addr); |
1da177e4 LT |
614 | do { |
615 | next = pgd_addr_end(addr, end); | |
616 | if (pgd_none_or_clear_bad(pgd)) | |
617 | continue; | |
c2febafc | 618 | free_p4d_range(tlb, pgd, addr, next, floor, ceiling); |
1da177e4 | 619 | } while (pgd++, addr = next, addr != end); |
e0da382c HD |
620 | } |
621 | ||
42b77728 | 622 | void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma, |
3bf5ee95 | 623 | unsigned long floor, unsigned long ceiling) |
e0da382c HD |
624 | { |
625 | while (vma) { | |
626 | struct vm_area_struct *next = vma->vm_next; | |
627 | unsigned long addr = vma->vm_start; | |
628 | ||
8f4f8c16 | 629 | /* |
25d9e2d1 | 630 | * Hide vma from rmap and truncate_pagecache before freeing |
631 | * pgtables | |
8f4f8c16 | 632 | */ |
5beb4930 | 633 | unlink_anon_vmas(vma); |
8f4f8c16 HD |
634 | unlink_file_vma(vma); |
635 | ||
9da61aef | 636 | if (is_vm_hugetlb_page(vma)) { |
3bf5ee95 | 637 | hugetlb_free_pgd_range(tlb, addr, vma->vm_end, |
166f61b9 | 638 | floor, next ? next->vm_start : ceiling); |
3bf5ee95 HD |
639 | } else { |
640 | /* | |
641 | * Optimization: gather nearby vmas into one call down | |
642 | */ | |
643 | while (next && next->vm_start <= vma->vm_end + PMD_SIZE | |
4866920b | 644 | && !is_vm_hugetlb_page(next)) { |
3bf5ee95 HD |
645 | vma = next; |
646 | next = vma->vm_next; | |
5beb4930 | 647 | unlink_anon_vmas(vma); |
8f4f8c16 | 648 | unlink_file_vma(vma); |
3bf5ee95 HD |
649 | } |
650 | free_pgd_range(tlb, addr, vma->vm_end, | |
166f61b9 | 651 | floor, next ? next->vm_start : ceiling); |
3bf5ee95 | 652 | } |
e0da382c HD |
653 | vma = next; |
654 | } | |
1da177e4 LT |
655 | } |
656 | ||
3ed3a4f0 | 657 | int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address) |
1da177e4 | 658 | { |
c4088ebd | 659 | spinlock_t *ptl; |
2f569afd | 660 | pgtable_t new = pte_alloc_one(mm, address); |
1bb3630e HD |
661 | if (!new) |
662 | return -ENOMEM; | |
663 | ||
362a61ad NP |
664 | /* |
665 | * Ensure all pte setup (eg. pte page lock and page clearing) are | |
666 | * visible before the pte is made visible to other CPUs by being | |
667 | * put into page tables. | |
668 | * | |
669 | * The other side of the story is the pointer chasing in the page | |
670 | * table walking code (when walking the page table without locking; | |
671 | * ie. most of the time). Fortunately, these data accesses consist | |
672 | * of a chain of data-dependent loads, meaning most CPUs (alpha | |
673 | * being the notable exception) will already guarantee loads are | |
674 | * seen in-order. See the alpha page table accessors for the | |
675 | * smp_read_barrier_depends() barriers in page table walking code. | |
676 | */ | |
677 | smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */ | |
678 | ||
c4088ebd | 679 | ptl = pmd_lock(mm, pmd); |
8ac1f832 | 680 | if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ |
c4812909 | 681 | mm_inc_nr_ptes(mm); |
1da177e4 | 682 | pmd_populate(mm, pmd, new); |
2f569afd | 683 | new = NULL; |
4b471e88 | 684 | } |
c4088ebd | 685 | spin_unlock(ptl); |
2f569afd MS |
686 | if (new) |
687 | pte_free(mm, new); | |
1bb3630e | 688 | return 0; |
1da177e4 LT |
689 | } |
690 | ||
1bb3630e | 691 | int __pte_alloc_kernel(pmd_t *pmd, unsigned long address) |
1da177e4 | 692 | { |
1bb3630e HD |
693 | pte_t *new = pte_alloc_one_kernel(&init_mm, address); |
694 | if (!new) | |
695 | return -ENOMEM; | |
696 | ||
362a61ad NP |
697 | smp_wmb(); /* See comment in __pte_alloc */ |
698 | ||
1bb3630e | 699 | spin_lock(&init_mm.page_table_lock); |
8ac1f832 | 700 | if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ |
1bb3630e | 701 | pmd_populate_kernel(&init_mm, pmd, new); |
2f569afd | 702 | new = NULL; |
4b471e88 | 703 | } |
1bb3630e | 704 | spin_unlock(&init_mm.page_table_lock); |
2f569afd MS |
705 | if (new) |
706 | pte_free_kernel(&init_mm, new); | |
1bb3630e | 707 | return 0; |
1da177e4 LT |
708 | } |
709 | ||
d559db08 KH |
710 | static inline void init_rss_vec(int *rss) |
711 | { | |
712 | memset(rss, 0, sizeof(int) * NR_MM_COUNTERS); | |
713 | } | |
714 | ||
715 | static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss) | |
ae859762 | 716 | { |
d559db08 KH |
717 | int i; |
718 | ||
34e55232 | 719 | if (current->mm == mm) |
05af2e10 | 720 | sync_mm_rss(mm); |
d559db08 KH |
721 | for (i = 0; i < NR_MM_COUNTERS; i++) |
722 | if (rss[i]) | |
723 | add_mm_counter(mm, i, rss[i]); | |
ae859762 HD |
724 | } |
725 | ||
b5810039 | 726 | /* |
6aab341e LT |
727 | * This function is called to print an error when a bad pte |
728 | * is found. For example, we might have a PFN-mapped pte in | |
729 | * a region that doesn't allow it. | |
b5810039 NP |
730 | * |
731 | * The calling function must still handle the error. | |
732 | */ | |
3dc14741 HD |
733 | static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr, |
734 | pte_t pte, struct page *page) | |
b5810039 | 735 | { |
3dc14741 | 736 | pgd_t *pgd = pgd_offset(vma->vm_mm, addr); |
c2febafc KS |
737 | p4d_t *p4d = p4d_offset(pgd, addr); |
738 | pud_t *pud = pud_offset(p4d, addr); | |
3dc14741 HD |
739 | pmd_t *pmd = pmd_offset(pud, addr); |
740 | struct address_space *mapping; | |
741 | pgoff_t index; | |
d936cf9b HD |
742 | static unsigned long resume; |
743 | static unsigned long nr_shown; | |
744 | static unsigned long nr_unshown; | |
745 | ||
746 | /* | |
747 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
748 | * or allow a steady drip of one report per second. | |
749 | */ | |
750 | if (nr_shown == 60) { | |
751 | if (time_before(jiffies, resume)) { | |
752 | nr_unshown++; | |
753 | return; | |
754 | } | |
755 | if (nr_unshown) { | |
1170532b JP |
756 | pr_alert("BUG: Bad page map: %lu messages suppressed\n", |
757 | nr_unshown); | |
d936cf9b HD |
758 | nr_unshown = 0; |
759 | } | |
760 | nr_shown = 0; | |
761 | } | |
762 | if (nr_shown++ == 0) | |
763 | resume = jiffies + 60 * HZ; | |
3dc14741 HD |
764 | |
765 | mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL; | |
766 | index = linear_page_index(vma, addr); | |
767 | ||
1170532b JP |
768 | pr_alert("BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n", |
769 | current->comm, | |
770 | (long long)pte_val(pte), (long long)pmd_val(*pmd)); | |
718a3821 | 771 | if (page) |
f0b791a3 | 772 | dump_page(page, "bad pte"); |
1170532b JP |
773 | pr_alert("addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n", |
774 | (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index); | |
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 | */ |
df6ad698 JG |
826 | struct page *_vm_normal_page(struct vm_area_struct *vma, unsigned long addr, |
827 | pte_t pte, bool with_public_device) | |
ee498ed7 | 828 | { |
22b31eec | 829 | unsigned long pfn = pte_pfn(pte); |
7e675137 | 830 | |
00b3a331 | 831 | if (IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL)) { |
b38af472 | 832 | if (likely(!pte_special(pte))) |
22b31eec | 833 | goto check_pfn; |
667a0a06 DV |
834 | if (vma->vm_ops && vma->vm_ops->find_special_page) |
835 | return vma->vm_ops->find_special_page(vma, addr); | |
a13ea5b7 HD |
836 | if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) |
837 | return NULL; | |
df6ad698 JG |
838 | if (is_zero_pfn(pfn)) |
839 | return NULL; | |
840 | ||
841 | /* | |
842 | * Device public pages are special pages (they are ZONE_DEVICE | |
843 | * pages but different from persistent memory). They behave | |
844 | * allmost like normal pages. The difference is that they are | |
845 | * not on the lru and thus should never be involve with any- | |
846 | * thing that involve lru manipulation (mlock, numa balancing, | |
847 | * ...). | |
848 | * | |
849 | * This is why we still want to return NULL for such page from | |
850 | * vm_normal_page() so that we do not have to special case all | |
851 | * call site of vm_normal_page(). | |
852 | */ | |
7d790d2d | 853 | if (likely(pfn <= highest_memmap_pfn)) { |
df6ad698 JG |
854 | struct page *page = pfn_to_page(pfn); |
855 | ||
856 | if (is_device_public_page(page)) { | |
857 | if (with_public_device) | |
858 | return page; | |
859 | return NULL; | |
860 | } | |
861 | } | |
e1fb4a08 DJ |
862 | |
863 | if (pte_devmap(pte)) | |
864 | return NULL; | |
865 | ||
df6ad698 | 866 | print_bad_pte(vma, addr, pte, NULL); |
7e675137 NP |
867 | return NULL; |
868 | } | |
869 | ||
00b3a331 | 870 | /* !CONFIG_ARCH_HAS_PTE_SPECIAL case follows: */ |
7e675137 | 871 | |
b379d790 JH |
872 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { |
873 | if (vma->vm_flags & VM_MIXEDMAP) { | |
874 | if (!pfn_valid(pfn)) | |
875 | return NULL; | |
876 | goto out; | |
877 | } else { | |
7e675137 NP |
878 | unsigned long off; |
879 | off = (addr - vma->vm_start) >> PAGE_SHIFT; | |
b379d790 JH |
880 | if (pfn == vma->vm_pgoff + off) |
881 | return NULL; | |
882 | if (!is_cow_mapping(vma->vm_flags)) | |
883 | return NULL; | |
884 | } | |
6aab341e LT |
885 | } |
886 | ||
b38af472 HD |
887 | if (is_zero_pfn(pfn)) |
888 | return NULL; | |
00b3a331 | 889 | |
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 | |
00b3a331 | 913 | * !CONFIG_ARCH_HAS_PTE_SPECIAL case from vm_normal_page() here. |
28093f9f GS |
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 | ||
e1fb4a08 DJ |
930 | if (pmd_devmap(pmd)) |
931 | return NULL; | |
28093f9f GS |
932 | if (is_zero_pfn(pfn)) |
933 | return NULL; | |
934 | if (unlikely(pfn > highest_memmap_pfn)) | |
935 | return NULL; | |
936 | ||
937 | /* | |
938 | * NOTE! We still have PageReserved() pages in the page tables. | |
939 | * eg. VDSO mappings can cause them to exist. | |
940 | */ | |
941 | out: | |
942 | return pfn_to_page(pfn); | |
943 | } | |
944 | #endif | |
945 | ||
1da177e4 LT |
946 | /* |
947 | * copy one vm_area from one task to the other. Assumes the page tables | |
948 | * already present in the new task to be cleared in the whole range | |
949 | * covered by this vma. | |
1da177e4 LT |
950 | */ |
951 | ||
570a335b | 952 | static inline unsigned long |
1da177e4 | 953 | copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
b5810039 | 954 | pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma, |
8c103762 | 955 | unsigned long addr, int *rss) |
1da177e4 | 956 | { |
b5810039 | 957 | unsigned long vm_flags = vma->vm_flags; |
1da177e4 LT |
958 | pte_t pte = *src_pte; |
959 | struct page *page; | |
1da177e4 LT |
960 | |
961 | /* pte contains position in swap or file, so copy. */ | |
962 | if (unlikely(!pte_present(pte))) { | |
0661a336 KS |
963 | swp_entry_t entry = pte_to_swp_entry(pte); |
964 | ||
965 | if (likely(!non_swap_entry(entry))) { | |
966 | if (swap_duplicate(entry) < 0) | |
967 | return entry.val; | |
968 | ||
969 | /* make sure dst_mm is on swapoff's mmlist. */ | |
970 | if (unlikely(list_empty(&dst_mm->mmlist))) { | |
971 | spin_lock(&mmlist_lock); | |
972 | if (list_empty(&dst_mm->mmlist)) | |
973 | list_add(&dst_mm->mmlist, | |
974 | &src_mm->mmlist); | |
975 | spin_unlock(&mmlist_lock); | |
976 | } | |
977 | rss[MM_SWAPENTS]++; | |
978 | } else if (is_migration_entry(entry)) { | |
979 | page = migration_entry_to_page(entry); | |
980 | ||
eca56ff9 | 981 | rss[mm_counter(page)]++; |
0661a336 KS |
982 | |
983 | if (is_write_migration_entry(entry) && | |
984 | is_cow_mapping(vm_flags)) { | |
985 | /* | |
986 | * COW mappings require pages in both | |
987 | * parent and child to be set to read. | |
988 | */ | |
989 | make_migration_entry_read(&entry); | |
990 | pte = swp_entry_to_pte(entry); | |
991 | if (pte_swp_soft_dirty(*src_pte)) | |
992 | pte = pte_swp_mksoft_dirty(pte); | |
993 | set_pte_at(src_mm, addr, src_pte, pte); | |
0697212a | 994 | } |
5042db43 JG |
995 | } else if (is_device_private_entry(entry)) { |
996 | page = device_private_entry_to_page(entry); | |
997 | ||
998 | /* | |
999 | * Update rss count even for unaddressable pages, as | |
1000 | * they should treated just like normal pages in this | |
1001 | * respect. | |
1002 | * | |
1003 | * We will likely want to have some new rss counters | |
1004 | * for unaddressable pages, at some point. But for now | |
1005 | * keep things as they are. | |
1006 | */ | |
1007 | get_page(page); | |
1008 | rss[mm_counter(page)]++; | |
1009 | page_dup_rmap(page, false); | |
1010 | ||
1011 | /* | |
1012 | * We do not preserve soft-dirty information, because so | |
1013 | * far, checkpoint/restore is the only feature that | |
1014 | * requires that. And checkpoint/restore does not work | |
1015 | * when a device driver is involved (you cannot easily | |
1016 | * save and restore device driver state). | |
1017 | */ | |
1018 | if (is_write_device_private_entry(entry) && | |
1019 | is_cow_mapping(vm_flags)) { | |
1020 | make_device_private_entry_read(&entry); | |
1021 | pte = swp_entry_to_pte(entry); | |
1022 | set_pte_at(src_mm, addr, src_pte, pte); | |
1023 | } | |
1da177e4 | 1024 | } |
ae859762 | 1025 | goto out_set_pte; |
1da177e4 LT |
1026 | } |
1027 | ||
1da177e4 LT |
1028 | /* |
1029 | * If it's a COW mapping, write protect it both | |
1030 | * in the parent and the child | |
1031 | */ | |
67121172 | 1032 | if (is_cow_mapping(vm_flags)) { |
1da177e4 | 1033 | ptep_set_wrprotect(src_mm, addr, src_pte); |
3dc90795 | 1034 | pte = pte_wrprotect(pte); |
1da177e4 LT |
1035 | } |
1036 | ||
1037 | /* | |
1038 | * If it's a shared mapping, mark it clean in | |
1039 | * the child | |
1040 | */ | |
1041 | if (vm_flags & VM_SHARED) | |
1042 | pte = pte_mkclean(pte); | |
1043 | pte = pte_mkold(pte); | |
6aab341e LT |
1044 | |
1045 | page = vm_normal_page(vma, addr, pte); | |
1046 | if (page) { | |
1047 | get_page(page); | |
53f9263b | 1048 | page_dup_rmap(page, false); |
eca56ff9 | 1049 | rss[mm_counter(page)]++; |
df6ad698 JG |
1050 | } else if (pte_devmap(pte)) { |
1051 | page = pte_page(pte); | |
1052 | ||
1053 | /* | |
1054 | * Cache coherent device memory behave like regular page and | |
1055 | * not like persistent memory page. For more informations see | |
1056 | * MEMORY_DEVICE_CACHE_COHERENT in memory_hotplug.h | |
1057 | */ | |
1058 | if (is_device_public_page(page)) { | |
1059 | get_page(page); | |
1060 | page_dup_rmap(page, false); | |
1061 | rss[mm_counter(page)]++; | |
1062 | } | |
6aab341e | 1063 | } |
ae859762 HD |
1064 | |
1065 | out_set_pte: | |
1066 | set_pte_at(dst_mm, addr, dst_pte, pte); | |
570a335b | 1067 | return 0; |
1da177e4 LT |
1068 | } |
1069 | ||
21bda264 | 1070 | static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
71e3aac0 AA |
1071 | pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma, |
1072 | unsigned long addr, unsigned long end) | |
1da177e4 | 1073 | { |
c36987e2 | 1074 | pte_t *orig_src_pte, *orig_dst_pte; |
1da177e4 | 1075 | pte_t *src_pte, *dst_pte; |
c74df32c | 1076 | spinlock_t *src_ptl, *dst_ptl; |
e040f218 | 1077 | int progress = 0; |
d559db08 | 1078 | int rss[NR_MM_COUNTERS]; |
570a335b | 1079 | swp_entry_t entry = (swp_entry_t){0}; |
1da177e4 LT |
1080 | |
1081 | again: | |
d559db08 KH |
1082 | init_rss_vec(rss); |
1083 | ||
c74df32c | 1084 | dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); |
1da177e4 LT |
1085 | if (!dst_pte) |
1086 | return -ENOMEM; | |
ece0e2b6 | 1087 | src_pte = pte_offset_map(src_pmd, addr); |
4c21e2f2 | 1088 | src_ptl = pte_lockptr(src_mm, src_pmd); |
f20dc5f7 | 1089 | spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); |
c36987e2 DN |
1090 | orig_src_pte = src_pte; |
1091 | orig_dst_pte = dst_pte; | |
6606c3e0 | 1092 | arch_enter_lazy_mmu_mode(); |
1da177e4 | 1093 | |
1da177e4 LT |
1094 | do { |
1095 | /* | |
1096 | * We are holding two locks at this point - either of them | |
1097 | * could generate latencies in another task on another CPU. | |
1098 | */ | |
e040f218 HD |
1099 | if (progress >= 32) { |
1100 | progress = 0; | |
1101 | if (need_resched() || | |
95c354fe | 1102 | spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) |
e040f218 HD |
1103 | break; |
1104 | } | |
1da177e4 LT |
1105 | if (pte_none(*src_pte)) { |
1106 | progress++; | |
1107 | continue; | |
1108 | } | |
570a335b HD |
1109 | entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, |
1110 | vma, addr, rss); | |
1111 | if (entry.val) | |
1112 | break; | |
1da177e4 LT |
1113 | progress += 8; |
1114 | } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); | |
1da177e4 | 1115 | |
6606c3e0 | 1116 | arch_leave_lazy_mmu_mode(); |
c74df32c | 1117 | spin_unlock(src_ptl); |
ece0e2b6 | 1118 | pte_unmap(orig_src_pte); |
d559db08 | 1119 | add_mm_rss_vec(dst_mm, rss); |
c36987e2 | 1120 | pte_unmap_unlock(orig_dst_pte, dst_ptl); |
c74df32c | 1121 | cond_resched(); |
570a335b HD |
1122 | |
1123 | if (entry.val) { | |
1124 | if (add_swap_count_continuation(entry, GFP_KERNEL) < 0) | |
1125 | return -ENOMEM; | |
1126 | progress = 0; | |
1127 | } | |
1da177e4 LT |
1128 | if (addr != end) |
1129 | goto again; | |
1130 | return 0; | |
1131 | } | |
1132 | ||
1133 | static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
1134 | pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma, | |
1135 | unsigned long addr, unsigned long end) | |
1136 | { | |
1137 | pmd_t *src_pmd, *dst_pmd; | |
1138 | unsigned long next; | |
1139 | ||
1140 | dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); | |
1141 | if (!dst_pmd) | |
1142 | return -ENOMEM; | |
1143 | src_pmd = pmd_offset(src_pud, addr); | |
1144 | do { | |
1145 | next = pmd_addr_end(addr, end); | |
84c3fc4e ZY |
1146 | if (is_swap_pmd(*src_pmd) || pmd_trans_huge(*src_pmd) |
1147 | || pmd_devmap(*src_pmd)) { | |
71e3aac0 | 1148 | int err; |
a00cc7d9 | 1149 | VM_BUG_ON_VMA(next-addr != HPAGE_PMD_SIZE, vma); |
71e3aac0 AA |
1150 | err = copy_huge_pmd(dst_mm, src_mm, |
1151 | dst_pmd, src_pmd, addr, vma); | |
1152 | if (err == -ENOMEM) | |
1153 | return -ENOMEM; | |
1154 | if (!err) | |
1155 | continue; | |
1156 | /* fall through */ | |
1157 | } | |
1da177e4 LT |
1158 | if (pmd_none_or_clear_bad(src_pmd)) |
1159 | continue; | |
1160 | if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd, | |
1161 | vma, addr, next)) | |
1162 | return -ENOMEM; | |
1163 | } while (dst_pmd++, src_pmd++, addr = next, addr != end); | |
1164 | return 0; | |
1165 | } | |
1166 | ||
1167 | static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
c2febafc | 1168 | p4d_t *dst_p4d, p4d_t *src_p4d, struct vm_area_struct *vma, |
1da177e4 LT |
1169 | unsigned long addr, unsigned long end) |
1170 | { | |
1171 | pud_t *src_pud, *dst_pud; | |
1172 | unsigned long next; | |
1173 | ||
c2febafc | 1174 | dst_pud = pud_alloc(dst_mm, dst_p4d, addr); |
1da177e4 LT |
1175 | if (!dst_pud) |
1176 | return -ENOMEM; | |
c2febafc | 1177 | src_pud = pud_offset(src_p4d, addr); |
1da177e4 LT |
1178 | do { |
1179 | next = pud_addr_end(addr, end); | |
a00cc7d9 MW |
1180 | if (pud_trans_huge(*src_pud) || pud_devmap(*src_pud)) { |
1181 | int err; | |
1182 | ||
1183 | VM_BUG_ON_VMA(next-addr != HPAGE_PUD_SIZE, vma); | |
1184 | err = copy_huge_pud(dst_mm, src_mm, | |
1185 | dst_pud, src_pud, addr, vma); | |
1186 | if (err == -ENOMEM) | |
1187 | return -ENOMEM; | |
1188 | if (!err) | |
1189 | continue; | |
1190 | /* fall through */ | |
1191 | } | |
1da177e4 LT |
1192 | if (pud_none_or_clear_bad(src_pud)) |
1193 | continue; | |
1194 | if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud, | |
1195 | vma, addr, next)) | |
1196 | return -ENOMEM; | |
1197 | } while (dst_pud++, src_pud++, addr = next, addr != end); | |
1198 | return 0; | |
1199 | } | |
1200 | ||
c2febafc KS |
1201 | static inline int copy_p4d_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
1202 | pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma, | |
1203 | unsigned long addr, unsigned long end) | |
1204 | { | |
1205 | p4d_t *src_p4d, *dst_p4d; | |
1206 | unsigned long next; | |
1207 | ||
1208 | dst_p4d = p4d_alloc(dst_mm, dst_pgd, addr); | |
1209 | if (!dst_p4d) | |
1210 | return -ENOMEM; | |
1211 | src_p4d = p4d_offset(src_pgd, addr); | |
1212 | do { | |
1213 | next = p4d_addr_end(addr, end); | |
1214 | if (p4d_none_or_clear_bad(src_p4d)) | |
1215 | continue; | |
1216 | if (copy_pud_range(dst_mm, src_mm, dst_p4d, src_p4d, | |
1217 | vma, addr, next)) | |
1218 | return -ENOMEM; | |
1219 | } while (dst_p4d++, src_p4d++, addr = next, addr != end); | |
1220 | return 0; | |
1221 | } | |
1222 | ||
1da177e4 LT |
1223 | int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
1224 | struct vm_area_struct *vma) | |
1225 | { | |
1226 | pgd_t *src_pgd, *dst_pgd; | |
1227 | unsigned long next; | |
1228 | unsigned long addr = vma->vm_start; | |
1229 | unsigned long end = vma->vm_end; | |
2ec74c3e SG |
1230 | unsigned long mmun_start; /* For mmu_notifiers */ |
1231 | unsigned long mmun_end; /* For mmu_notifiers */ | |
1232 | bool is_cow; | |
cddb8a5c | 1233 | int ret; |
1da177e4 | 1234 | |
d992895b NP |
1235 | /* |
1236 | * Don't copy ptes where a page fault will fill them correctly. | |
1237 | * Fork becomes much lighter when there are big shared or private | |
1238 | * readonly mappings. The tradeoff is that copy_page_range is more | |
1239 | * efficient than faulting. | |
1240 | */ | |
0661a336 KS |
1241 | if (!(vma->vm_flags & (VM_HUGETLB | VM_PFNMAP | VM_MIXEDMAP)) && |
1242 | !vma->anon_vma) | |
1243 | return 0; | |
d992895b | 1244 | |
1da177e4 LT |
1245 | if (is_vm_hugetlb_page(vma)) |
1246 | return copy_hugetlb_page_range(dst_mm, src_mm, vma); | |
1247 | ||
b3b9c293 | 1248 | if (unlikely(vma->vm_flags & VM_PFNMAP)) { |
2ab64037 | 1249 | /* |
1250 | * We do not free on error cases below as remove_vma | |
1251 | * gets called on error from higher level routine | |
1252 | */ | |
5180da41 | 1253 | ret = track_pfn_copy(vma); |
2ab64037 | 1254 | if (ret) |
1255 | return ret; | |
1256 | } | |
1257 | ||
cddb8a5c AA |
1258 | /* |
1259 | * We need to invalidate the secondary MMU mappings only when | |
1260 | * there could be a permission downgrade on the ptes of the | |
1261 | * parent mm. And a permission downgrade will only happen if | |
1262 | * is_cow_mapping() returns true. | |
1263 | */ | |
2ec74c3e SG |
1264 | is_cow = is_cow_mapping(vma->vm_flags); |
1265 | mmun_start = addr; | |
1266 | mmun_end = end; | |
1267 | if (is_cow) | |
1268 | mmu_notifier_invalidate_range_start(src_mm, mmun_start, | |
1269 | mmun_end); | |
cddb8a5c AA |
1270 | |
1271 | ret = 0; | |
1da177e4 LT |
1272 | dst_pgd = pgd_offset(dst_mm, addr); |
1273 | src_pgd = pgd_offset(src_mm, addr); | |
1274 | do { | |
1275 | next = pgd_addr_end(addr, end); | |
1276 | if (pgd_none_or_clear_bad(src_pgd)) | |
1277 | continue; | |
c2febafc | 1278 | if (unlikely(copy_p4d_range(dst_mm, src_mm, dst_pgd, src_pgd, |
cddb8a5c AA |
1279 | vma, addr, next))) { |
1280 | ret = -ENOMEM; | |
1281 | break; | |
1282 | } | |
1da177e4 | 1283 | } while (dst_pgd++, src_pgd++, addr = next, addr != end); |
cddb8a5c | 1284 | |
2ec74c3e SG |
1285 | if (is_cow) |
1286 | mmu_notifier_invalidate_range_end(src_mm, mmun_start, mmun_end); | |
cddb8a5c | 1287 | return ret; |
1da177e4 LT |
1288 | } |
1289 | ||
51c6f666 | 1290 | static unsigned long zap_pte_range(struct mmu_gather *tlb, |
b5810039 | 1291 | struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 | 1292 | unsigned long addr, unsigned long end, |
97a89413 | 1293 | struct zap_details *details) |
1da177e4 | 1294 | { |
b5810039 | 1295 | struct mm_struct *mm = tlb->mm; |
d16dfc55 | 1296 | int force_flush = 0; |
d559db08 | 1297 | int rss[NR_MM_COUNTERS]; |
97a89413 | 1298 | spinlock_t *ptl; |
5f1a1907 | 1299 | pte_t *start_pte; |
97a89413 | 1300 | pte_t *pte; |
8a5f14a2 | 1301 | swp_entry_t entry; |
d559db08 | 1302 | |
07e32661 | 1303 | tlb_remove_check_page_size_change(tlb, PAGE_SIZE); |
d16dfc55 | 1304 | again: |
e303297e | 1305 | init_rss_vec(rss); |
5f1a1907 SR |
1306 | start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
1307 | pte = start_pte; | |
3ea27719 | 1308 | flush_tlb_batched_pending(mm); |
6606c3e0 | 1309 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
1310 | do { |
1311 | pte_t ptent = *pte; | |
166f61b9 | 1312 | if (pte_none(ptent)) |
1da177e4 | 1313 | continue; |
6f5e6b9e | 1314 | |
1da177e4 | 1315 | if (pte_present(ptent)) { |
ee498ed7 | 1316 | struct page *page; |
51c6f666 | 1317 | |
df6ad698 | 1318 | page = _vm_normal_page(vma, addr, ptent, true); |
1da177e4 LT |
1319 | if (unlikely(details) && page) { |
1320 | /* | |
1321 | * unmap_shared_mapping_pages() wants to | |
1322 | * invalidate cache without truncating: | |
1323 | * unmap shared but keep private pages. | |
1324 | */ | |
1325 | if (details->check_mapping && | |
800d8c63 | 1326 | details->check_mapping != page_rmapping(page)) |
1da177e4 | 1327 | continue; |
1da177e4 | 1328 | } |
b5810039 | 1329 | ptent = ptep_get_and_clear_full(mm, addr, pte, |
a600388d | 1330 | tlb->fullmm); |
1da177e4 LT |
1331 | tlb_remove_tlb_entry(tlb, pte, addr); |
1332 | if (unlikely(!page)) | |
1333 | continue; | |
eca56ff9 JM |
1334 | |
1335 | if (!PageAnon(page)) { | |
1cf35d47 LT |
1336 | if (pte_dirty(ptent)) { |
1337 | force_flush = 1; | |
6237bcd9 | 1338 | set_page_dirty(page); |
1cf35d47 | 1339 | } |
4917e5d0 | 1340 | if (pte_young(ptent) && |
64363aad | 1341 | likely(!(vma->vm_flags & VM_SEQ_READ))) |
bf3f3bc5 | 1342 | mark_page_accessed(page); |
6237bcd9 | 1343 | } |
eca56ff9 | 1344 | rss[mm_counter(page)]--; |
d281ee61 | 1345 | page_remove_rmap(page, false); |
3dc14741 HD |
1346 | if (unlikely(page_mapcount(page) < 0)) |
1347 | print_bad_pte(vma, addr, ptent, page); | |
e9d55e15 | 1348 | if (unlikely(__tlb_remove_page(tlb, page))) { |
1cf35d47 | 1349 | force_flush = 1; |
ce9ec37b | 1350 | addr += PAGE_SIZE; |
d16dfc55 | 1351 | break; |
1cf35d47 | 1352 | } |
1da177e4 LT |
1353 | continue; |
1354 | } | |
5042db43 JG |
1355 | |
1356 | entry = pte_to_swp_entry(ptent); | |
1357 | if (non_swap_entry(entry) && is_device_private_entry(entry)) { | |
1358 | struct page *page = device_private_entry_to_page(entry); | |
1359 | ||
1360 | if (unlikely(details && details->check_mapping)) { | |
1361 | /* | |
1362 | * unmap_shared_mapping_pages() wants to | |
1363 | * invalidate cache without truncating: | |
1364 | * unmap shared but keep private pages. | |
1365 | */ | |
1366 | if (details->check_mapping != | |
1367 | page_rmapping(page)) | |
1368 | continue; | |
1369 | } | |
1370 | ||
1371 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); | |
1372 | rss[mm_counter(page)]--; | |
1373 | page_remove_rmap(page, false); | |
1374 | put_page(page); | |
1375 | continue; | |
1376 | } | |
1377 | ||
3e8715fd KS |
1378 | /* If details->check_mapping, we leave swap entries. */ |
1379 | if (unlikely(details)) | |
1da177e4 | 1380 | continue; |
b084d435 | 1381 | |
8a5f14a2 KS |
1382 | entry = pte_to_swp_entry(ptent); |
1383 | if (!non_swap_entry(entry)) | |
1384 | rss[MM_SWAPENTS]--; | |
1385 | else if (is_migration_entry(entry)) { | |
1386 | struct page *page; | |
9f9f1acd | 1387 | |
8a5f14a2 | 1388 | page = migration_entry_to_page(entry); |
eca56ff9 | 1389 | rss[mm_counter(page)]--; |
b084d435 | 1390 | } |
8a5f14a2 KS |
1391 | if (unlikely(!free_swap_and_cache(entry))) |
1392 | print_bad_pte(vma, addr, ptent, NULL); | |
9888a1ca | 1393 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); |
97a89413 | 1394 | } while (pte++, addr += PAGE_SIZE, addr != end); |
ae859762 | 1395 | |
d559db08 | 1396 | add_mm_rss_vec(mm, rss); |
6606c3e0 | 1397 | arch_leave_lazy_mmu_mode(); |
51c6f666 | 1398 | |
1cf35d47 | 1399 | /* Do the actual TLB flush before dropping ptl */ |
fb7332a9 | 1400 | if (force_flush) |
1cf35d47 | 1401 | tlb_flush_mmu_tlbonly(tlb); |
1cf35d47 LT |
1402 | pte_unmap_unlock(start_pte, ptl); |
1403 | ||
1404 | /* | |
1405 | * If we forced a TLB flush (either due to running out of | |
1406 | * batch buffers or because we needed to flush dirty TLB | |
1407 | * entries before releasing the ptl), free the batched | |
1408 | * memory too. Restart if we didn't do everything. | |
1409 | */ | |
1410 | if (force_flush) { | |
1411 | force_flush = 0; | |
1412 | tlb_flush_mmu_free(tlb); | |
2b047252 | 1413 | if (addr != end) |
d16dfc55 PZ |
1414 | goto again; |
1415 | } | |
1416 | ||
51c6f666 | 1417 | return addr; |
1da177e4 LT |
1418 | } |
1419 | ||
51c6f666 | 1420 | static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, |
b5810039 | 1421 | struct vm_area_struct *vma, pud_t *pud, |
1da177e4 | 1422 | unsigned long addr, unsigned long end, |
97a89413 | 1423 | struct zap_details *details) |
1da177e4 LT |
1424 | { |
1425 | pmd_t *pmd; | |
1426 | unsigned long next; | |
1427 | ||
1428 | pmd = pmd_offset(pud, addr); | |
1429 | do { | |
1430 | next = pmd_addr_end(addr, end); | |
84c3fc4e | 1431 | if (is_swap_pmd(*pmd) || pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) { |
53406ed1 | 1432 | if (next - addr != HPAGE_PMD_SIZE) |
fd60775a | 1433 | __split_huge_pmd(vma, pmd, addr, false, NULL); |
53406ed1 | 1434 | else if (zap_huge_pmd(tlb, vma, pmd, addr)) |
1a5a9906 | 1435 | goto next; |
71e3aac0 AA |
1436 | /* fall through */ |
1437 | } | |
1a5a9906 AA |
1438 | /* |
1439 | * Here there can be other concurrent MADV_DONTNEED or | |
1440 | * trans huge page faults running, and if the pmd is | |
1441 | * none or trans huge it can change under us. This is | |
1442 | * because MADV_DONTNEED holds the mmap_sem in read | |
1443 | * mode. | |
1444 | */ | |
1445 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) | |
1446 | goto next; | |
97a89413 | 1447 | next = zap_pte_range(tlb, vma, pmd, addr, next, details); |
1a5a9906 | 1448 | next: |
97a89413 PZ |
1449 | cond_resched(); |
1450 | } while (pmd++, addr = next, addr != end); | |
51c6f666 RH |
1451 | |
1452 | return addr; | |
1da177e4 LT |
1453 | } |
1454 | ||
51c6f666 | 1455 | static inline unsigned long zap_pud_range(struct mmu_gather *tlb, |
c2febafc | 1456 | struct vm_area_struct *vma, p4d_t *p4d, |
1da177e4 | 1457 | unsigned long addr, unsigned long end, |
97a89413 | 1458 | struct zap_details *details) |
1da177e4 LT |
1459 | { |
1460 | pud_t *pud; | |
1461 | unsigned long next; | |
1462 | ||
c2febafc | 1463 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
1464 | do { |
1465 | next = pud_addr_end(addr, end); | |
a00cc7d9 MW |
1466 | if (pud_trans_huge(*pud) || pud_devmap(*pud)) { |
1467 | if (next - addr != HPAGE_PUD_SIZE) { | |
1468 | VM_BUG_ON_VMA(!rwsem_is_locked(&tlb->mm->mmap_sem), vma); | |
1469 | split_huge_pud(vma, pud, addr); | |
1470 | } else if (zap_huge_pud(tlb, vma, pud, addr)) | |
1471 | goto next; | |
1472 | /* fall through */ | |
1473 | } | |
97a89413 | 1474 | if (pud_none_or_clear_bad(pud)) |
1da177e4 | 1475 | continue; |
97a89413 | 1476 | next = zap_pmd_range(tlb, vma, pud, addr, next, details); |
a00cc7d9 MW |
1477 | next: |
1478 | cond_resched(); | |
97a89413 | 1479 | } while (pud++, addr = next, addr != end); |
51c6f666 RH |
1480 | |
1481 | return addr; | |
1da177e4 LT |
1482 | } |
1483 | ||
c2febafc KS |
1484 | static inline unsigned long zap_p4d_range(struct mmu_gather *tlb, |
1485 | struct vm_area_struct *vma, pgd_t *pgd, | |
1486 | unsigned long addr, unsigned long end, | |
1487 | struct zap_details *details) | |
1488 | { | |
1489 | p4d_t *p4d; | |
1490 | unsigned long next; | |
1491 | ||
1492 | p4d = p4d_offset(pgd, addr); | |
1493 | do { | |
1494 | next = p4d_addr_end(addr, end); | |
1495 | if (p4d_none_or_clear_bad(p4d)) | |
1496 | continue; | |
1497 | next = zap_pud_range(tlb, vma, p4d, addr, next, details); | |
1498 | } while (p4d++, addr = next, addr != end); | |
1499 | ||
1500 | return addr; | |
1501 | } | |
1502 | ||
aac45363 | 1503 | void unmap_page_range(struct mmu_gather *tlb, |
038c7aa1 AV |
1504 | struct vm_area_struct *vma, |
1505 | unsigned long addr, unsigned long end, | |
1506 | struct zap_details *details) | |
1da177e4 LT |
1507 | { |
1508 | pgd_t *pgd; | |
1509 | unsigned long next; | |
1510 | ||
1da177e4 LT |
1511 | BUG_ON(addr >= end); |
1512 | tlb_start_vma(tlb, vma); | |
1513 | pgd = pgd_offset(vma->vm_mm, addr); | |
1514 | do { | |
1515 | next = pgd_addr_end(addr, end); | |
97a89413 | 1516 | if (pgd_none_or_clear_bad(pgd)) |
1da177e4 | 1517 | continue; |
c2febafc | 1518 | next = zap_p4d_range(tlb, vma, pgd, addr, next, details); |
97a89413 | 1519 | } while (pgd++, addr = next, addr != end); |
1da177e4 LT |
1520 | tlb_end_vma(tlb, vma); |
1521 | } | |
51c6f666 | 1522 | |
f5cc4eef AV |
1523 | |
1524 | static void unmap_single_vma(struct mmu_gather *tlb, | |
1525 | struct vm_area_struct *vma, unsigned long start_addr, | |
4f74d2c8 | 1526 | unsigned long end_addr, |
f5cc4eef AV |
1527 | struct zap_details *details) |
1528 | { | |
1529 | unsigned long start = max(vma->vm_start, start_addr); | |
1530 | unsigned long end; | |
1531 | ||
1532 | if (start >= vma->vm_end) | |
1533 | return; | |
1534 | end = min(vma->vm_end, end_addr); | |
1535 | if (end <= vma->vm_start) | |
1536 | return; | |
1537 | ||
cbc91f71 SD |
1538 | if (vma->vm_file) |
1539 | uprobe_munmap(vma, start, end); | |
1540 | ||
b3b9c293 | 1541 | if (unlikely(vma->vm_flags & VM_PFNMAP)) |
5180da41 | 1542 | untrack_pfn(vma, 0, 0); |
f5cc4eef AV |
1543 | |
1544 | if (start != end) { | |
1545 | if (unlikely(is_vm_hugetlb_page(vma))) { | |
1546 | /* | |
1547 | * It is undesirable to test vma->vm_file as it | |
1548 | * should be non-null for valid hugetlb area. | |
1549 | * However, vm_file will be NULL in the error | |
7aa6b4ad | 1550 | * cleanup path of mmap_region. When |
f5cc4eef | 1551 | * hugetlbfs ->mmap method fails, |
7aa6b4ad | 1552 | * mmap_region() nullifies vma->vm_file |
f5cc4eef AV |
1553 | * before calling this function to clean up. |
1554 | * Since no pte has actually been setup, it is | |
1555 | * safe to do nothing in this case. | |
1556 | */ | |
24669e58 | 1557 | if (vma->vm_file) { |
83cde9e8 | 1558 | i_mmap_lock_write(vma->vm_file->f_mapping); |
d833352a | 1559 | __unmap_hugepage_range_final(tlb, vma, start, end, NULL); |
83cde9e8 | 1560 | i_mmap_unlock_write(vma->vm_file->f_mapping); |
24669e58 | 1561 | } |
f5cc4eef AV |
1562 | } else |
1563 | unmap_page_range(tlb, vma, start, end, details); | |
1564 | } | |
1da177e4 LT |
1565 | } |
1566 | ||
1da177e4 LT |
1567 | /** |
1568 | * unmap_vmas - unmap a range of memory covered by a list of vma's | |
0164f69d | 1569 | * @tlb: address of the caller's struct mmu_gather |
1da177e4 LT |
1570 | * @vma: the starting vma |
1571 | * @start_addr: virtual address at which to start unmapping | |
1572 | * @end_addr: virtual address at which to end unmapping | |
1da177e4 | 1573 | * |
508034a3 | 1574 | * Unmap all pages in the vma list. |
1da177e4 | 1575 | * |
1da177e4 LT |
1576 | * Only addresses between `start' and `end' will be unmapped. |
1577 | * | |
1578 | * The VMA list must be sorted in ascending virtual address order. | |
1579 | * | |
1580 | * unmap_vmas() assumes that the caller will flush the whole unmapped address | |
1581 | * range after unmap_vmas() returns. So the only responsibility here is to | |
1582 | * ensure that any thus-far unmapped pages are flushed before unmap_vmas() | |
1583 | * drops the lock and schedules. | |
1584 | */ | |
6e8bb019 | 1585 | void unmap_vmas(struct mmu_gather *tlb, |
1da177e4 | 1586 | struct vm_area_struct *vma, unsigned long start_addr, |
4f74d2c8 | 1587 | unsigned long end_addr) |
1da177e4 | 1588 | { |
cddb8a5c | 1589 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 1590 | |
cddb8a5c | 1591 | mmu_notifier_invalidate_range_start(mm, start_addr, end_addr); |
f5cc4eef | 1592 | for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) |
4f74d2c8 | 1593 | unmap_single_vma(tlb, vma, start_addr, end_addr, NULL); |
cddb8a5c | 1594 | mmu_notifier_invalidate_range_end(mm, start_addr, end_addr); |
1da177e4 LT |
1595 | } |
1596 | ||
1597 | /** | |
1598 | * zap_page_range - remove user pages in a given range | |
1599 | * @vma: vm_area_struct holding the applicable pages | |
eb4546bb | 1600 | * @start: starting address of pages to zap |
1da177e4 | 1601 | * @size: number of bytes to zap |
f5cc4eef AV |
1602 | * |
1603 | * Caller must protect the VMA list | |
1da177e4 | 1604 | */ |
7e027b14 | 1605 | void zap_page_range(struct vm_area_struct *vma, unsigned long start, |
ecf1385d | 1606 | unsigned long size) |
1da177e4 LT |
1607 | { |
1608 | struct mm_struct *mm = vma->vm_mm; | |
d16dfc55 | 1609 | struct mmu_gather tlb; |
7e027b14 | 1610 | unsigned long end = start + size; |
1da177e4 | 1611 | |
1da177e4 | 1612 | lru_add_drain(); |
2b047252 | 1613 | tlb_gather_mmu(&tlb, mm, start, end); |
365e9c87 | 1614 | update_hiwater_rss(mm); |
7e027b14 | 1615 | mmu_notifier_invalidate_range_start(mm, start, end); |
4647706e | 1616 | for ( ; vma && vma->vm_start < end; vma = vma->vm_next) { |
ecf1385d | 1617 | unmap_single_vma(&tlb, vma, start, end, NULL); |
4647706e MG |
1618 | |
1619 | /* | |
1620 | * zap_page_range does not specify whether mmap_sem should be | |
1621 | * held for read or write. That allows parallel zap_page_range | |
1622 | * operations to unmap a PTE and defer a flush meaning that | |
1623 | * this call observes pte_none and fails to flush the TLB. | |
1624 | * Rather than adding a complex API, ensure that no stale | |
1625 | * TLB entries exist when this call returns. | |
1626 | */ | |
1627 | flush_tlb_range(vma, start, end); | |
1628 | } | |
1629 | ||
7e027b14 LT |
1630 | mmu_notifier_invalidate_range_end(mm, start, end); |
1631 | tlb_finish_mmu(&tlb, start, end); | |
1da177e4 LT |
1632 | } |
1633 | ||
f5cc4eef AV |
1634 | /** |
1635 | * zap_page_range_single - remove user pages in a given range | |
1636 | * @vma: vm_area_struct holding the applicable pages | |
1637 | * @address: starting address of pages to zap | |
1638 | * @size: number of bytes to zap | |
8a5f14a2 | 1639 | * @details: details of shared cache invalidation |
f5cc4eef AV |
1640 | * |
1641 | * The range must fit into one VMA. | |
1da177e4 | 1642 | */ |
f5cc4eef | 1643 | static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address, |
1da177e4 LT |
1644 | unsigned long size, struct zap_details *details) |
1645 | { | |
1646 | struct mm_struct *mm = vma->vm_mm; | |
d16dfc55 | 1647 | struct mmu_gather tlb; |
1da177e4 | 1648 | unsigned long end = address + size; |
1da177e4 | 1649 | |
1da177e4 | 1650 | lru_add_drain(); |
2b047252 | 1651 | tlb_gather_mmu(&tlb, mm, address, end); |
365e9c87 | 1652 | update_hiwater_rss(mm); |
f5cc4eef | 1653 | mmu_notifier_invalidate_range_start(mm, address, end); |
4f74d2c8 | 1654 | unmap_single_vma(&tlb, vma, address, end, details); |
f5cc4eef | 1655 | mmu_notifier_invalidate_range_end(mm, address, end); |
d16dfc55 | 1656 | tlb_finish_mmu(&tlb, address, end); |
1da177e4 LT |
1657 | } |
1658 | ||
c627f9cc JS |
1659 | /** |
1660 | * zap_vma_ptes - remove ptes mapping the vma | |
1661 | * @vma: vm_area_struct holding ptes to be zapped | |
1662 | * @address: starting address of pages to zap | |
1663 | * @size: number of bytes to zap | |
1664 | * | |
1665 | * This function only unmaps ptes assigned to VM_PFNMAP vmas. | |
1666 | * | |
1667 | * The entire address range must be fully contained within the vma. | |
1668 | * | |
c627f9cc | 1669 | */ |
27d036e3 | 1670 | void zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, |
c627f9cc JS |
1671 | unsigned long size) |
1672 | { | |
1673 | if (address < vma->vm_start || address + size > vma->vm_end || | |
1674 | !(vma->vm_flags & VM_PFNMAP)) | |
27d036e3 LR |
1675 | return; |
1676 | ||
f5cc4eef | 1677 | zap_page_range_single(vma, address, size, NULL); |
c627f9cc JS |
1678 | } |
1679 | EXPORT_SYMBOL_GPL(zap_vma_ptes); | |
1680 | ||
25ca1d6c | 1681 | pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr, |
920c7a5d | 1682 | spinlock_t **ptl) |
c9cfcddf | 1683 | { |
c2febafc KS |
1684 | pgd_t *pgd; |
1685 | p4d_t *p4d; | |
1686 | pud_t *pud; | |
1687 | pmd_t *pmd; | |
1688 | ||
1689 | pgd = pgd_offset(mm, addr); | |
1690 | p4d = p4d_alloc(mm, pgd, addr); | |
1691 | if (!p4d) | |
1692 | return NULL; | |
1693 | pud = pud_alloc(mm, p4d, addr); | |
1694 | if (!pud) | |
1695 | return NULL; | |
1696 | pmd = pmd_alloc(mm, pud, addr); | |
1697 | if (!pmd) | |
1698 | return NULL; | |
1699 | ||
1700 | VM_BUG_ON(pmd_trans_huge(*pmd)); | |
1701 | return pte_alloc_map_lock(mm, pmd, addr, ptl); | |
c9cfcddf LT |
1702 | } |
1703 | ||
238f58d8 LT |
1704 | /* |
1705 | * This is the old fallback for page remapping. | |
1706 | * | |
1707 | * For historical reasons, it only allows reserved pages. Only | |
1708 | * old drivers should use this, and they needed to mark their | |
1709 | * pages reserved for the old functions anyway. | |
1710 | */ | |
423bad60 NP |
1711 | static int insert_page(struct vm_area_struct *vma, unsigned long addr, |
1712 | struct page *page, pgprot_t prot) | |
238f58d8 | 1713 | { |
423bad60 | 1714 | struct mm_struct *mm = vma->vm_mm; |
238f58d8 | 1715 | int retval; |
c9cfcddf | 1716 | pte_t *pte; |
8a9f3ccd BS |
1717 | spinlock_t *ptl; |
1718 | ||
238f58d8 | 1719 | retval = -EINVAL; |
a145dd41 | 1720 | if (PageAnon(page)) |
5b4e655e | 1721 | goto out; |
238f58d8 LT |
1722 | retval = -ENOMEM; |
1723 | flush_dcache_page(page); | |
c9cfcddf | 1724 | pte = get_locked_pte(mm, addr, &ptl); |
238f58d8 | 1725 | if (!pte) |
5b4e655e | 1726 | goto out; |
238f58d8 LT |
1727 | retval = -EBUSY; |
1728 | if (!pte_none(*pte)) | |
1729 | goto out_unlock; | |
1730 | ||
1731 | /* Ok, finally just insert the thing.. */ | |
1732 | get_page(page); | |
eca56ff9 | 1733 | inc_mm_counter_fast(mm, mm_counter_file(page)); |
dd78fedd | 1734 | page_add_file_rmap(page, false); |
238f58d8 LT |
1735 | set_pte_at(mm, addr, pte, mk_pte(page, prot)); |
1736 | ||
1737 | retval = 0; | |
8a9f3ccd BS |
1738 | pte_unmap_unlock(pte, ptl); |
1739 | return retval; | |
238f58d8 LT |
1740 | out_unlock: |
1741 | pte_unmap_unlock(pte, ptl); | |
1742 | out: | |
1743 | return retval; | |
1744 | } | |
1745 | ||
bfa5bf6d REB |
1746 | /** |
1747 | * vm_insert_page - insert single page into user vma | |
1748 | * @vma: user vma to map to | |
1749 | * @addr: target user address of this page | |
1750 | * @page: source kernel page | |
1751 | * | |
a145dd41 LT |
1752 | * This allows drivers to insert individual pages they've allocated |
1753 | * into a user vma. | |
1754 | * | |
1755 | * The page has to be a nice clean _individual_ kernel allocation. | |
1756 | * If you allocate a compound page, you need to have marked it as | |
1757 | * such (__GFP_COMP), or manually just split the page up yourself | |
8dfcc9ba | 1758 | * (see split_page()). |
a145dd41 LT |
1759 | * |
1760 | * NOTE! Traditionally this was done with "remap_pfn_range()" which | |
1761 | * took an arbitrary page protection parameter. This doesn't allow | |
1762 | * that. Your vma protection will have to be set up correctly, which | |
1763 | * means that if you want a shared writable mapping, you'd better | |
1764 | * ask for a shared writable mapping! | |
1765 | * | |
1766 | * The page does not need to be reserved. | |
4b6e1e37 KK |
1767 | * |
1768 | * Usually this function is called from f_op->mmap() handler | |
1769 | * under mm->mmap_sem write-lock, so it can change vma->vm_flags. | |
1770 | * Caller must set VM_MIXEDMAP on vma if it wants to call this | |
1771 | * function from other places, for example from page-fault handler. | |
a145dd41 | 1772 | */ |
423bad60 NP |
1773 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, |
1774 | struct page *page) | |
a145dd41 LT |
1775 | { |
1776 | if (addr < vma->vm_start || addr >= vma->vm_end) | |
1777 | return -EFAULT; | |
1778 | if (!page_count(page)) | |
1779 | return -EINVAL; | |
4b6e1e37 KK |
1780 | if (!(vma->vm_flags & VM_MIXEDMAP)) { |
1781 | BUG_ON(down_read_trylock(&vma->vm_mm->mmap_sem)); | |
1782 | BUG_ON(vma->vm_flags & VM_PFNMAP); | |
1783 | vma->vm_flags |= VM_MIXEDMAP; | |
1784 | } | |
423bad60 | 1785 | return insert_page(vma, addr, page, vma->vm_page_prot); |
a145dd41 | 1786 | } |
e3c3374f | 1787 | EXPORT_SYMBOL(vm_insert_page); |
a145dd41 | 1788 | |
423bad60 | 1789 | static int insert_pfn(struct vm_area_struct *vma, unsigned long addr, |
b2770da6 | 1790 | pfn_t pfn, pgprot_t prot, bool mkwrite) |
423bad60 NP |
1791 | { |
1792 | struct mm_struct *mm = vma->vm_mm; | |
1793 | int retval; | |
1794 | pte_t *pte, entry; | |
1795 | spinlock_t *ptl; | |
1796 | ||
1797 | retval = -ENOMEM; | |
1798 | pte = get_locked_pte(mm, addr, &ptl); | |
1799 | if (!pte) | |
1800 | goto out; | |
1801 | retval = -EBUSY; | |
b2770da6 RZ |
1802 | if (!pte_none(*pte)) { |
1803 | if (mkwrite) { | |
1804 | /* | |
1805 | * For read faults on private mappings the PFN passed | |
1806 | * in may not match the PFN we have mapped if the | |
1807 | * mapped PFN is a writeable COW page. In the mkwrite | |
1808 | * case we are creating a writable PTE for a shared | |
1809 | * mapping and we expect the PFNs to match. | |
1810 | */ | |
1811 | if (WARN_ON_ONCE(pte_pfn(*pte) != pfn_t_to_pfn(pfn))) | |
1812 | goto out_unlock; | |
1813 | entry = *pte; | |
1814 | goto out_mkwrite; | |
1815 | } else | |
1816 | goto out_unlock; | |
1817 | } | |
423bad60 NP |
1818 | |
1819 | /* Ok, finally just insert the thing.. */ | |
01c8f1c4 DW |
1820 | if (pfn_t_devmap(pfn)) |
1821 | entry = pte_mkdevmap(pfn_t_pte(pfn, prot)); | |
1822 | else | |
1823 | entry = pte_mkspecial(pfn_t_pte(pfn, prot)); | |
b2770da6 RZ |
1824 | |
1825 | out_mkwrite: | |
1826 | if (mkwrite) { | |
1827 | entry = pte_mkyoung(entry); | |
1828 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
1829 | } | |
1830 | ||
423bad60 | 1831 | set_pte_at(mm, addr, pte, entry); |
4b3073e1 | 1832 | update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */ |
423bad60 NP |
1833 | |
1834 | retval = 0; | |
1835 | out_unlock: | |
1836 | pte_unmap_unlock(pte, ptl); | |
1837 | out: | |
1838 | return retval; | |
1839 | } | |
1840 | ||
e0dc0d8f NP |
1841 | /** |
1842 | * vm_insert_pfn - insert single pfn into user vma | |
1843 | * @vma: user vma to map to | |
1844 | * @addr: target user address of this page | |
1845 | * @pfn: source kernel pfn | |
1846 | * | |
c462f179 | 1847 | * Similar to vm_insert_page, this allows drivers to insert individual pages |
e0dc0d8f NP |
1848 | * they've allocated into a user vma. Same comments apply. |
1849 | * | |
1850 | * This function should only be called from a vm_ops->fault handler, and | |
1851 | * in that case the handler should return NULL. | |
0d71d10a NP |
1852 | * |
1853 | * vma cannot be a COW mapping. | |
1854 | * | |
1855 | * As this is called only for pages that do not currently exist, we | |
1856 | * do not need to flush old virtual caches or the TLB. | |
e0dc0d8f NP |
1857 | */ |
1858 | int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr, | |
423bad60 | 1859 | unsigned long pfn) |
1745cbc5 AL |
1860 | { |
1861 | return vm_insert_pfn_prot(vma, addr, pfn, vma->vm_page_prot); | |
1862 | } | |
1863 | EXPORT_SYMBOL(vm_insert_pfn); | |
1864 | ||
1865 | /** | |
1866 | * vm_insert_pfn_prot - insert single pfn into user vma with specified pgprot | |
1867 | * @vma: user vma to map to | |
1868 | * @addr: target user address of this page | |
1869 | * @pfn: source kernel pfn | |
1870 | * @pgprot: pgprot flags for the inserted page | |
1871 | * | |
1872 | * This is exactly like vm_insert_pfn, except that it allows drivers to | |
1873 | * to override pgprot on a per-page basis. | |
1874 | * | |
1875 | * This only makes sense for IO mappings, and it makes no sense for | |
1876 | * cow mappings. In general, using multiple vmas is preferable; | |
1877 | * vm_insert_pfn_prot should only be used if using multiple VMAs is | |
1878 | * impractical. | |
1879 | */ | |
1880 | int vm_insert_pfn_prot(struct vm_area_struct *vma, unsigned long addr, | |
1881 | unsigned long pfn, pgprot_t pgprot) | |
e0dc0d8f | 1882 | { |
2ab64037 | 1883 | int ret; |
7e675137 NP |
1884 | /* |
1885 | * Technically, architectures with pte_special can avoid all these | |
1886 | * restrictions (same for remap_pfn_range). However we would like | |
1887 | * consistency in testing and feature parity among all, so we should | |
1888 | * try to keep these invariants in place for everybody. | |
1889 | */ | |
b379d790 JH |
1890 | BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); |
1891 | BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == | |
1892 | (VM_PFNMAP|VM_MIXEDMAP)); | |
1893 | BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); | |
1894 | BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); | |
e0dc0d8f | 1895 | |
423bad60 NP |
1896 | if (addr < vma->vm_start || addr >= vma->vm_end) |
1897 | return -EFAULT; | |
308a047c | 1898 | |
42e4089c AK |
1899 | if (!pfn_modify_allowed(pfn, pgprot)) |
1900 | return -EACCES; | |
1901 | ||
308a047c | 1902 | track_pfn_insert(vma, &pgprot, __pfn_to_pfn_t(pfn, PFN_DEV)); |
2ab64037 | 1903 | |
b2770da6 RZ |
1904 | ret = insert_pfn(vma, addr, __pfn_to_pfn_t(pfn, PFN_DEV), pgprot, |
1905 | false); | |
2ab64037 | 1906 | |
2ab64037 | 1907 | return ret; |
423bad60 | 1908 | } |
1745cbc5 | 1909 | EXPORT_SYMBOL(vm_insert_pfn_prot); |
e0dc0d8f | 1910 | |
785a3fab DW |
1911 | static bool vm_mixed_ok(struct vm_area_struct *vma, pfn_t pfn) |
1912 | { | |
1913 | /* these checks mirror the abort conditions in vm_normal_page */ | |
1914 | if (vma->vm_flags & VM_MIXEDMAP) | |
1915 | return true; | |
1916 | if (pfn_t_devmap(pfn)) | |
1917 | return true; | |
1918 | if (pfn_t_special(pfn)) | |
1919 | return true; | |
1920 | if (is_zero_pfn(pfn_t_to_pfn(pfn))) | |
1921 | return true; | |
1922 | return false; | |
1923 | } | |
1924 | ||
b2770da6 RZ |
1925 | static int __vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr, |
1926 | pfn_t pfn, bool mkwrite) | |
423bad60 | 1927 | { |
87744ab3 DW |
1928 | pgprot_t pgprot = vma->vm_page_prot; |
1929 | ||
785a3fab | 1930 | BUG_ON(!vm_mixed_ok(vma, pfn)); |
e0dc0d8f | 1931 | |
423bad60 NP |
1932 | if (addr < vma->vm_start || addr >= vma->vm_end) |
1933 | return -EFAULT; | |
308a047c BP |
1934 | |
1935 | track_pfn_insert(vma, &pgprot, pfn); | |
e0dc0d8f | 1936 | |
42e4089c AK |
1937 | if (!pfn_modify_allowed(pfn_t_to_pfn(pfn), pgprot)) |
1938 | return -EACCES; | |
1939 | ||
423bad60 NP |
1940 | /* |
1941 | * If we don't have pte special, then we have to use the pfn_valid() | |
1942 | * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* | |
1943 | * refcount the page if pfn_valid is true (hence insert_page rather | |
62eede62 HD |
1944 | * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP |
1945 | * without pte special, it would there be refcounted as a normal page. | |
423bad60 | 1946 | */ |
00b3a331 LD |
1947 | if (!IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL) && |
1948 | !pfn_t_devmap(pfn) && pfn_t_valid(pfn)) { | |
423bad60 NP |
1949 | struct page *page; |
1950 | ||
03fc2da6 DW |
1951 | /* |
1952 | * At this point we are committed to insert_page() | |
1953 | * regardless of whether the caller specified flags that | |
1954 | * result in pfn_t_has_page() == false. | |
1955 | */ | |
1956 | page = pfn_to_page(pfn_t_to_pfn(pfn)); | |
87744ab3 | 1957 | return insert_page(vma, addr, page, pgprot); |
423bad60 | 1958 | } |
b2770da6 RZ |
1959 | return insert_pfn(vma, addr, pfn, pgprot, mkwrite); |
1960 | } | |
1961 | ||
1962 | int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr, | |
1963 | pfn_t pfn) | |
1964 | { | |
1965 | return __vm_insert_mixed(vma, addr, pfn, false); | |
1966 | ||
e0dc0d8f | 1967 | } |
423bad60 | 1968 | EXPORT_SYMBOL(vm_insert_mixed); |
e0dc0d8f | 1969 | |
ab77dab4 SJ |
1970 | /* |
1971 | * If the insertion of PTE failed because someone else already added a | |
1972 | * different entry in the mean time, we treat that as success as we assume | |
1973 | * the same entry was actually inserted. | |
1974 | */ | |
1975 | ||
1976 | vm_fault_t vmf_insert_mixed_mkwrite(struct vm_area_struct *vma, | |
1977 | unsigned long addr, pfn_t pfn) | |
b2770da6 | 1978 | { |
ab77dab4 SJ |
1979 | int err; |
1980 | ||
1981 | err = __vm_insert_mixed(vma, addr, pfn, true); | |
1982 | if (err == -ENOMEM) | |
1983 | return VM_FAULT_OOM; | |
1984 | if (err < 0 && err != -EBUSY) | |
1985 | return VM_FAULT_SIGBUS; | |
1986 | return VM_FAULT_NOPAGE; | |
b2770da6 | 1987 | } |
ab77dab4 | 1988 | EXPORT_SYMBOL(vmf_insert_mixed_mkwrite); |
b2770da6 | 1989 | |
1da177e4 LT |
1990 | /* |
1991 | * maps a range of physical memory into the requested pages. the old | |
1992 | * mappings are removed. any references to nonexistent pages results | |
1993 | * in null mappings (currently treated as "copy-on-access") | |
1994 | */ | |
1995 | static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, | |
1996 | unsigned long addr, unsigned long end, | |
1997 | unsigned long pfn, pgprot_t prot) | |
1998 | { | |
1999 | pte_t *pte; | |
c74df32c | 2000 | spinlock_t *ptl; |
42e4089c | 2001 | int err = 0; |
1da177e4 | 2002 | |
c74df32c | 2003 | pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); |
1da177e4 LT |
2004 | if (!pte) |
2005 | return -ENOMEM; | |
6606c3e0 | 2006 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
2007 | do { |
2008 | BUG_ON(!pte_none(*pte)); | |
42e4089c AK |
2009 | if (!pfn_modify_allowed(pfn, prot)) { |
2010 | err = -EACCES; | |
2011 | break; | |
2012 | } | |
7e675137 | 2013 | set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); |
1da177e4 LT |
2014 | pfn++; |
2015 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
6606c3e0 | 2016 | arch_leave_lazy_mmu_mode(); |
c74df32c | 2017 | pte_unmap_unlock(pte - 1, ptl); |
42e4089c | 2018 | return err; |
1da177e4 LT |
2019 | } |
2020 | ||
2021 | static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, | |
2022 | unsigned long addr, unsigned long end, | |
2023 | unsigned long pfn, pgprot_t prot) | |
2024 | { | |
2025 | pmd_t *pmd; | |
2026 | unsigned long next; | |
42e4089c | 2027 | int err; |
1da177e4 LT |
2028 | |
2029 | pfn -= addr >> PAGE_SHIFT; | |
2030 | pmd = pmd_alloc(mm, pud, addr); | |
2031 | if (!pmd) | |
2032 | return -ENOMEM; | |
f66055ab | 2033 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
1da177e4 LT |
2034 | do { |
2035 | next = pmd_addr_end(addr, end); | |
42e4089c AK |
2036 | err = remap_pte_range(mm, pmd, addr, next, |
2037 | pfn + (addr >> PAGE_SHIFT), prot); | |
2038 | if (err) | |
2039 | return err; | |
1da177e4 LT |
2040 | } while (pmd++, addr = next, addr != end); |
2041 | return 0; | |
2042 | } | |
2043 | ||
c2febafc | 2044 | static inline int remap_pud_range(struct mm_struct *mm, p4d_t *p4d, |
1da177e4 LT |
2045 | unsigned long addr, unsigned long end, |
2046 | unsigned long pfn, pgprot_t prot) | |
2047 | { | |
2048 | pud_t *pud; | |
2049 | unsigned long next; | |
42e4089c | 2050 | int err; |
1da177e4 LT |
2051 | |
2052 | pfn -= addr >> PAGE_SHIFT; | |
c2febafc | 2053 | pud = pud_alloc(mm, p4d, addr); |
1da177e4 LT |
2054 | if (!pud) |
2055 | return -ENOMEM; | |
2056 | do { | |
2057 | next = pud_addr_end(addr, end); | |
42e4089c AK |
2058 | err = remap_pmd_range(mm, pud, addr, next, |
2059 | pfn + (addr >> PAGE_SHIFT), prot); | |
2060 | if (err) | |
2061 | return err; | |
1da177e4 LT |
2062 | } while (pud++, addr = next, addr != end); |
2063 | return 0; | |
2064 | } | |
2065 | ||
c2febafc KS |
2066 | static inline int remap_p4d_range(struct mm_struct *mm, pgd_t *pgd, |
2067 | unsigned long addr, unsigned long end, | |
2068 | unsigned long pfn, pgprot_t prot) | |
2069 | { | |
2070 | p4d_t *p4d; | |
2071 | unsigned long next; | |
42e4089c | 2072 | int err; |
c2febafc KS |
2073 | |
2074 | pfn -= addr >> PAGE_SHIFT; | |
2075 | p4d = p4d_alloc(mm, pgd, addr); | |
2076 | if (!p4d) | |
2077 | return -ENOMEM; | |
2078 | do { | |
2079 | next = p4d_addr_end(addr, end); | |
42e4089c AK |
2080 | err = remap_pud_range(mm, p4d, addr, next, |
2081 | pfn + (addr >> PAGE_SHIFT), prot); | |
2082 | if (err) | |
2083 | return err; | |
c2febafc KS |
2084 | } while (p4d++, addr = next, addr != end); |
2085 | return 0; | |
2086 | } | |
2087 | ||
bfa5bf6d REB |
2088 | /** |
2089 | * remap_pfn_range - remap kernel memory to userspace | |
2090 | * @vma: user vma to map to | |
2091 | * @addr: target user address to start at | |
2092 | * @pfn: physical address of kernel memory | |
2093 | * @size: size of map area | |
2094 | * @prot: page protection flags for this mapping | |
2095 | * | |
2096 | * Note: this is only safe if the mm semaphore is held when called. | |
2097 | */ | |
1da177e4 LT |
2098 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, |
2099 | unsigned long pfn, unsigned long size, pgprot_t prot) | |
2100 | { | |
2101 | pgd_t *pgd; | |
2102 | unsigned long next; | |
2d15cab8 | 2103 | unsigned long end = addr + PAGE_ALIGN(size); |
1da177e4 | 2104 | struct mm_struct *mm = vma->vm_mm; |
d5957d2f | 2105 | unsigned long remap_pfn = pfn; |
1da177e4 LT |
2106 | int err; |
2107 | ||
2108 | /* | |
2109 | * Physically remapped pages are special. Tell the | |
2110 | * rest of the world about it: | |
2111 | * VM_IO tells people not to look at these pages | |
2112 | * (accesses can have side effects). | |
6aab341e LT |
2113 | * VM_PFNMAP tells the core MM that the base pages are just |
2114 | * raw PFN mappings, and do not have a "struct page" associated | |
2115 | * with them. | |
314e51b9 KK |
2116 | * VM_DONTEXPAND |
2117 | * Disable vma merging and expanding with mremap(). | |
2118 | * VM_DONTDUMP | |
2119 | * Omit vma from core dump, even when VM_IO turned off. | |
fb155c16 LT |
2120 | * |
2121 | * There's a horrible special case to handle copy-on-write | |
2122 | * behaviour that some programs depend on. We mark the "original" | |
2123 | * un-COW'ed pages by matching them up with "vma->vm_pgoff". | |
b3b9c293 | 2124 | * See vm_normal_page() for details. |
1da177e4 | 2125 | */ |
b3b9c293 KK |
2126 | if (is_cow_mapping(vma->vm_flags)) { |
2127 | if (addr != vma->vm_start || end != vma->vm_end) | |
2128 | return -EINVAL; | |
fb155c16 | 2129 | vma->vm_pgoff = pfn; |
b3b9c293 KK |
2130 | } |
2131 | ||
d5957d2f | 2132 | err = track_pfn_remap(vma, &prot, remap_pfn, addr, PAGE_ALIGN(size)); |
b3b9c293 | 2133 | if (err) |
3c8bb73a | 2134 | return -EINVAL; |
fb155c16 | 2135 | |
314e51b9 | 2136 | vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; |
1da177e4 LT |
2137 | |
2138 | BUG_ON(addr >= end); | |
2139 | pfn -= addr >> PAGE_SHIFT; | |
2140 | pgd = pgd_offset(mm, addr); | |
2141 | flush_cache_range(vma, addr, end); | |
1da177e4 LT |
2142 | do { |
2143 | next = pgd_addr_end(addr, end); | |
c2febafc | 2144 | err = remap_p4d_range(mm, pgd, addr, next, |
1da177e4 LT |
2145 | pfn + (addr >> PAGE_SHIFT), prot); |
2146 | if (err) | |
2147 | break; | |
2148 | } while (pgd++, addr = next, addr != end); | |
2ab64037 | 2149 | |
2150 | if (err) | |
d5957d2f | 2151 | untrack_pfn(vma, remap_pfn, PAGE_ALIGN(size)); |
2ab64037 | 2152 | |
1da177e4 LT |
2153 | return err; |
2154 | } | |
2155 | EXPORT_SYMBOL(remap_pfn_range); | |
2156 | ||
b4cbb197 LT |
2157 | /** |
2158 | * vm_iomap_memory - remap memory to userspace | |
2159 | * @vma: user vma to map to | |
2160 | * @start: start of area | |
2161 | * @len: size of area | |
2162 | * | |
2163 | * This is a simplified io_remap_pfn_range() for common driver use. The | |
2164 | * driver just needs to give us the physical memory range to be mapped, | |
2165 | * we'll figure out the rest from the vma information. | |
2166 | * | |
2167 | * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get | |
2168 | * whatever write-combining details or similar. | |
2169 | */ | |
2170 | int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) | |
2171 | { | |
2172 | unsigned long vm_len, pfn, pages; | |
2173 | ||
2174 | /* Check that the physical memory area passed in looks valid */ | |
2175 | if (start + len < start) | |
2176 | return -EINVAL; | |
2177 | /* | |
2178 | * You *really* shouldn't map things that aren't page-aligned, | |
2179 | * but we've historically allowed it because IO memory might | |
2180 | * just have smaller alignment. | |
2181 | */ | |
2182 | len += start & ~PAGE_MASK; | |
2183 | pfn = start >> PAGE_SHIFT; | |
2184 | pages = (len + ~PAGE_MASK) >> PAGE_SHIFT; | |
2185 | if (pfn + pages < pfn) | |
2186 | return -EINVAL; | |
2187 | ||
2188 | /* We start the mapping 'vm_pgoff' pages into the area */ | |
2189 | if (vma->vm_pgoff > pages) | |
2190 | return -EINVAL; | |
2191 | pfn += vma->vm_pgoff; | |
2192 | pages -= vma->vm_pgoff; | |
2193 | ||
2194 | /* Can we fit all of the mapping? */ | |
2195 | vm_len = vma->vm_end - vma->vm_start; | |
2196 | if (vm_len >> PAGE_SHIFT > pages) | |
2197 | return -EINVAL; | |
2198 | ||
2199 | /* Ok, let it rip */ | |
2200 | return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); | |
2201 | } | |
2202 | EXPORT_SYMBOL(vm_iomap_memory); | |
2203 | ||
aee16b3c JF |
2204 | static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, |
2205 | unsigned long addr, unsigned long end, | |
2206 | pte_fn_t fn, void *data) | |
2207 | { | |
2208 | pte_t *pte; | |
2209 | int err; | |
2f569afd | 2210 | pgtable_t token; |
94909914 | 2211 | spinlock_t *uninitialized_var(ptl); |
aee16b3c JF |
2212 | |
2213 | pte = (mm == &init_mm) ? | |
2214 | pte_alloc_kernel(pmd, addr) : | |
2215 | pte_alloc_map_lock(mm, pmd, addr, &ptl); | |
2216 | if (!pte) | |
2217 | return -ENOMEM; | |
2218 | ||
2219 | BUG_ON(pmd_huge(*pmd)); | |
2220 | ||
38e0edb1 JF |
2221 | arch_enter_lazy_mmu_mode(); |
2222 | ||
2f569afd | 2223 | token = pmd_pgtable(*pmd); |
aee16b3c JF |
2224 | |
2225 | do { | |
c36987e2 | 2226 | err = fn(pte++, token, addr, data); |
aee16b3c JF |
2227 | if (err) |
2228 | break; | |
c36987e2 | 2229 | } while (addr += PAGE_SIZE, addr != end); |
aee16b3c | 2230 | |
38e0edb1 JF |
2231 | arch_leave_lazy_mmu_mode(); |
2232 | ||
aee16b3c JF |
2233 | if (mm != &init_mm) |
2234 | pte_unmap_unlock(pte-1, ptl); | |
2235 | return err; | |
2236 | } | |
2237 | ||
2238 | static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, | |
2239 | unsigned long addr, unsigned long end, | |
2240 | pte_fn_t fn, void *data) | |
2241 | { | |
2242 | pmd_t *pmd; | |
2243 | unsigned long next; | |
2244 | int err; | |
2245 | ||
ceb86879 AK |
2246 | BUG_ON(pud_huge(*pud)); |
2247 | ||
aee16b3c JF |
2248 | pmd = pmd_alloc(mm, pud, addr); |
2249 | if (!pmd) | |
2250 | return -ENOMEM; | |
2251 | do { | |
2252 | next = pmd_addr_end(addr, end); | |
2253 | err = apply_to_pte_range(mm, pmd, addr, next, fn, data); | |
2254 | if (err) | |
2255 | break; | |
2256 | } while (pmd++, addr = next, addr != end); | |
2257 | return err; | |
2258 | } | |
2259 | ||
c2febafc | 2260 | static int apply_to_pud_range(struct mm_struct *mm, p4d_t *p4d, |
aee16b3c JF |
2261 | unsigned long addr, unsigned long end, |
2262 | pte_fn_t fn, void *data) | |
2263 | { | |
2264 | pud_t *pud; | |
2265 | unsigned long next; | |
2266 | int err; | |
2267 | ||
c2febafc | 2268 | pud = pud_alloc(mm, p4d, addr); |
aee16b3c JF |
2269 | if (!pud) |
2270 | return -ENOMEM; | |
2271 | do { | |
2272 | next = pud_addr_end(addr, end); | |
2273 | err = apply_to_pmd_range(mm, pud, addr, next, fn, data); | |
2274 | if (err) | |
2275 | break; | |
2276 | } while (pud++, addr = next, addr != end); | |
2277 | return err; | |
2278 | } | |
2279 | ||
c2febafc KS |
2280 | static int apply_to_p4d_range(struct mm_struct *mm, pgd_t *pgd, |
2281 | unsigned long addr, unsigned long end, | |
2282 | pte_fn_t fn, void *data) | |
2283 | { | |
2284 | p4d_t *p4d; | |
2285 | unsigned long next; | |
2286 | int err; | |
2287 | ||
2288 | p4d = p4d_alloc(mm, pgd, addr); | |
2289 | if (!p4d) | |
2290 | return -ENOMEM; | |
2291 | do { | |
2292 | next = p4d_addr_end(addr, end); | |
2293 | err = apply_to_pud_range(mm, p4d, addr, next, fn, data); | |
2294 | if (err) | |
2295 | break; | |
2296 | } while (p4d++, addr = next, addr != end); | |
2297 | return err; | |
2298 | } | |
2299 | ||
aee16b3c JF |
2300 | /* |
2301 | * Scan a region of virtual memory, filling in page tables as necessary | |
2302 | * and calling a provided function on each leaf page table. | |
2303 | */ | |
2304 | int apply_to_page_range(struct mm_struct *mm, unsigned long addr, | |
2305 | unsigned long size, pte_fn_t fn, void *data) | |
2306 | { | |
2307 | pgd_t *pgd; | |
2308 | unsigned long next; | |
57250a5b | 2309 | unsigned long end = addr + size; |
aee16b3c JF |
2310 | int err; |
2311 | ||
9cb65bc3 MP |
2312 | if (WARN_ON(addr >= end)) |
2313 | return -EINVAL; | |
2314 | ||
aee16b3c JF |
2315 | pgd = pgd_offset(mm, addr); |
2316 | do { | |
2317 | next = pgd_addr_end(addr, end); | |
c2febafc | 2318 | err = apply_to_p4d_range(mm, pgd, addr, next, fn, data); |
aee16b3c JF |
2319 | if (err) |
2320 | break; | |
2321 | } while (pgd++, addr = next, addr != end); | |
57250a5b | 2322 | |
aee16b3c JF |
2323 | return err; |
2324 | } | |
2325 | EXPORT_SYMBOL_GPL(apply_to_page_range); | |
2326 | ||
8f4e2101 | 2327 | /* |
9b4bdd2f KS |
2328 | * handle_pte_fault chooses page fault handler according to an entry which was |
2329 | * read non-atomically. Before making any commitment, on those architectures | |
2330 | * or configurations (e.g. i386 with PAE) which might give a mix of unmatched | |
2331 | * parts, do_swap_page must check under lock before unmapping the pte and | |
2332 | * proceeding (but do_wp_page is only called after already making such a check; | |
a335b2e1 | 2333 | * and do_anonymous_page can safely check later on). |
8f4e2101 | 2334 | */ |
4c21e2f2 | 2335 | static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd, |
8f4e2101 HD |
2336 | pte_t *page_table, pte_t orig_pte) |
2337 | { | |
2338 | int same = 1; | |
2339 | #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT) | |
2340 | if (sizeof(pte_t) > sizeof(unsigned long)) { | |
4c21e2f2 HD |
2341 | spinlock_t *ptl = pte_lockptr(mm, pmd); |
2342 | spin_lock(ptl); | |
8f4e2101 | 2343 | same = pte_same(*page_table, orig_pte); |
4c21e2f2 | 2344 | spin_unlock(ptl); |
8f4e2101 HD |
2345 | } |
2346 | #endif | |
2347 | pte_unmap(page_table); | |
2348 | return same; | |
2349 | } | |
2350 | ||
9de455b2 | 2351 | static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma) |
6aab341e | 2352 | { |
0abdd7a8 DW |
2353 | debug_dma_assert_idle(src); |
2354 | ||
6aab341e LT |
2355 | /* |
2356 | * If the source page was a PFN mapping, we don't have | |
2357 | * a "struct page" for it. We do a best-effort copy by | |
2358 | * just copying from the original user address. If that | |
2359 | * fails, we just zero-fill it. Live with it. | |
2360 | */ | |
2361 | if (unlikely(!src)) { | |
9b04c5fe | 2362 | void *kaddr = kmap_atomic(dst); |
5d2a2dbb LT |
2363 | void __user *uaddr = (void __user *)(va & PAGE_MASK); |
2364 | ||
2365 | /* | |
2366 | * This really shouldn't fail, because the page is there | |
2367 | * in the page tables. But it might just be unreadable, | |
2368 | * in which case we just give up and fill the result with | |
2369 | * zeroes. | |
2370 | */ | |
2371 | if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) | |
3ecb01df | 2372 | clear_page(kaddr); |
9b04c5fe | 2373 | kunmap_atomic(kaddr); |
c4ec7b0d | 2374 | flush_dcache_page(dst); |
0ed361de NP |
2375 | } else |
2376 | copy_user_highpage(dst, src, va, vma); | |
6aab341e LT |
2377 | } |
2378 | ||
c20cd45e MH |
2379 | static gfp_t __get_fault_gfp_mask(struct vm_area_struct *vma) |
2380 | { | |
2381 | struct file *vm_file = vma->vm_file; | |
2382 | ||
2383 | if (vm_file) | |
2384 | return mapping_gfp_mask(vm_file->f_mapping) | __GFP_FS | __GFP_IO; | |
2385 | ||
2386 | /* | |
2387 | * Special mappings (e.g. VDSO) do not have any file so fake | |
2388 | * a default GFP_KERNEL for them. | |
2389 | */ | |
2390 | return GFP_KERNEL; | |
2391 | } | |
2392 | ||
fb09a464 KS |
2393 | /* |
2394 | * Notify the address space that the page is about to become writable so that | |
2395 | * it can prohibit this or wait for the page to get into an appropriate state. | |
2396 | * | |
2397 | * We do this without the lock held, so that it can sleep if it needs to. | |
2398 | */ | |
38b8cb7f | 2399 | static int do_page_mkwrite(struct vm_fault *vmf) |
fb09a464 | 2400 | { |
fb09a464 | 2401 | int ret; |
38b8cb7f JK |
2402 | struct page *page = vmf->page; |
2403 | unsigned int old_flags = vmf->flags; | |
fb09a464 | 2404 | |
38b8cb7f | 2405 | vmf->flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; |
fb09a464 | 2406 | |
11bac800 | 2407 | ret = vmf->vma->vm_ops->page_mkwrite(vmf); |
38b8cb7f JK |
2408 | /* Restore original flags so that caller is not surprised */ |
2409 | vmf->flags = old_flags; | |
fb09a464 KS |
2410 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) |
2411 | return ret; | |
2412 | if (unlikely(!(ret & VM_FAULT_LOCKED))) { | |
2413 | lock_page(page); | |
2414 | if (!page->mapping) { | |
2415 | unlock_page(page); | |
2416 | return 0; /* retry */ | |
2417 | } | |
2418 | ret |= VM_FAULT_LOCKED; | |
2419 | } else | |
2420 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
2421 | return ret; | |
2422 | } | |
2423 | ||
97ba0c2b JK |
2424 | /* |
2425 | * Handle dirtying of a page in shared file mapping on a write fault. | |
2426 | * | |
2427 | * The function expects the page to be locked and unlocks it. | |
2428 | */ | |
2429 | static void fault_dirty_shared_page(struct vm_area_struct *vma, | |
2430 | struct page *page) | |
2431 | { | |
2432 | struct address_space *mapping; | |
2433 | bool dirtied; | |
2434 | bool page_mkwrite = vma->vm_ops && vma->vm_ops->page_mkwrite; | |
2435 | ||
2436 | dirtied = set_page_dirty(page); | |
2437 | VM_BUG_ON_PAGE(PageAnon(page), page); | |
2438 | /* | |
2439 | * Take a local copy of the address_space - page.mapping may be zeroed | |
2440 | * by truncate after unlock_page(). The address_space itself remains | |
2441 | * pinned by vma->vm_file's reference. We rely on unlock_page()'s | |
2442 | * release semantics to prevent the compiler from undoing this copying. | |
2443 | */ | |
2444 | mapping = page_rmapping(page); | |
2445 | unlock_page(page); | |
2446 | ||
2447 | if ((dirtied || page_mkwrite) && mapping) { | |
2448 | /* | |
2449 | * Some device drivers do not set page.mapping | |
2450 | * but still dirty their pages | |
2451 | */ | |
2452 | balance_dirty_pages_ratelimited(mapping); | |
2453 | } | |
2454 | ||
2455 | if (!page_mkwrite) | |
2456 | file_update_time(vma->vm_file); | |
2457 | } | |
2458 | ||
4e047f89 SR |
2459 | /* |
2460 | * Handle write page faults for pages that can be reused in the current vma | |
2461 | * | |
2462 | * This can happen either due to the mapping being with the VM_SHARED flag, | |
2463 | * or due to us being the last reference standing to the page. In either | |
2464 | * case, all we need to do here is to mark the page as writable and update | |
2465 | * any related book-keeping. | |
2466 | */ | |
997dd98d | 2467 | static inline void wp_page_reuse(struct vm_fault *vmf) |
82b0f8c3 | 2468 | __releases(vmf->ptl) |
4e047f89 | 2469 | { |
82b0f8c3 | 2470 | struct vm_area_struct *vma = vmf->vma; |
a41b70d6 | 2471 | struct page *page = vmf->page; |
4e047f89 SR |
2472 | pte_t entry; |
2473 | /* | |
2474 | * Clear the pages cpupid information as the existing | |
2475 | * information potentially belongs to a now completely | |
2476 | * unrelated process. | |
2477 | */ | |
2478 | if (page) | |
2479 | page_cpupid_xchg_last(page, (1 << LAST_CPUPID_SHIFT) - 1); | |
2480 | ||
2994302b JK |
2481 | flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); |
2482 | entry = pte_mkyoung(vmf->orig_pte); | |
4e047f89 | 2483 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
82b0f8c3 JK |
2484 | if (ptep_set_access_flags(vma, vmf->address, vmf->pte, entry, 1)) |
2485 | update_mmu_cache(vma, vmf->address, vmf->pte); | |
2486 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
4e047f89 SR |
2487 | } |
2488 | ||
2f38ab2c SR |
2489 | /* |
2490 | * Handle the case of a page which we actually need to copy to a new page. | |
2491 | * | |
2492 | * Called with mmap_sem locked and the old page referenced, but | |
2493 | * without the ptl held. | |
2494 | * | |
2495 | * High level logic flow: | |
2496 | * | |
2497 | * - Allocate a page, copy the content of the old page to the new one. | |
2498 | * - Handle book keeping and accounting - cgroups, mmu-notifiers, etc. | |
2499 | * - Take the PTL. If the pte changed, bail out and release the allocated page | |
2500 | * - If the pte is still the way we remember it, update the page table and all | |
2501 | * relevant references. This includes dropping the reference the page-table | |
2502 | * held to the old page, as well as updating the rmap. | |
2503 | * - In any case, unlock the PTL and drop the reference we took to the old page. | |
2504 | */ | |
a41b70d6 | 2505 | static int wp_page_copy(struct vm_fault *vmf) |
2f38ab2c | 2506 | { |
82b0f8c3 | 2507 | struct vm_area_struct *vma = vmf->vma; |
bae473a4 | 2508 | struct mm_struct *mm = vma->vm_mm; |
a41b70d6 | 2509 | struct page *old_page = vmf->page; |
2f38ab2c | 2510 | struct page *new_page = NULL; |
2f38ab2c SR |
2511 | pte_t entry; |
2512 | int page_copied = 0; | |
82b0f8c3 | 2513 | const unsigned long mmun_start = vmf->address & PAGE_MASK; |
bae473a4 | 2514 | const unsigned long mmun_end = mmun_start + PAGE_SIZE; |
2f38ab2c SR |
2515 | struct mem_cgroup *memcg; |
2516 | ||
2517 | if (unlikely(anon_vma_prepare(vma))) | |
2518 | goto oom; | |
2519 | ||
2994302b | 2520 | if (is_zero_pfn(pte_pfn(vmf->orig_pte))) { |
82b0f8c3 JK |
2521 | new_page = alloc_zeroed_user_highpage_movable(vma, |
2522 | vmf->address); | |
2f38ab2c SR |
2523 | if (!new_page) |
2524 | goto oom; | |
2525 | } else { | |
bae473a4 | 2526 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, |
82b0f8c3 | 2527 | vmf->address); |
2f38ab2c SR |
2528 | if (!new_page) |
2529 | goto oom; | |
82b0f8c3 | 2530 | cow_user_page(new_page, old_page, vmf->address, vma); |
2f38ab2c | 2531 | } |
2f38ab2c | 2532 | |
2cf85583 | 2533 | if (mem_cgroup_try_charge_delay(new_page, mm, GFP_KERNEL, &memcg, false)) |
2f38ab2c SR |
2534 | goto oom_free_new; |
2535 | ||
eb3c24f3 MG |
2536 | __SetPageUptodate(new_page); |
2537 | ||
2f38ab2c SR |
2538 | mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); |
2539 | ||
2540 | /* | |
2541 | * Re-check the pte - we dropped the lock | |
2542 | */ | |
82b0f8c3 | 2543 | vmf->pte = pte_offset_map_lock(mm, vmf->pmd, vmf->address, &vmf->ptl); |
2994302b | 2544 | if (likely(pte_same(*vmf->pte, vmf->orig_pte))) { |
2f38ab2c SR |
2545 | if (old_page) { |
2546 | if (!PageAnon(old_page)) { | |
eca56ff9 JM |
2547 | dec_mm_counter_fast(mm, |
2548 | mm_counter_file(old_page)); | |
2f38ab2c SR |
2549 | inc_mm_counter_fast(mm, MM_ANONPAGES); |
2550 | } | |
2551 | } else { | |
2552 | inc_mm_counter_fast(mm, MM_ANONPAGES); | |
2553 | } | |
2994302b | 2554 | flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); |
2f38ab2c SR |
2555 | entry = mk_pte(new_page, vma->vm_page_prot); |
2556 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
2557 | /* | |
2558 | * Clear the pte entry and flush it first, before updating the | |
2559 | * pte with the new entry. This will avoid a race condition | |
2560 | * seen in the presence of one thread doing SMC and another | |
2561 | * thread doing COW. | |
2562 | */ | |
82b0f8c3 JK |
2563 | ptep_clear_flush_notify(vma, vmf->address, vmf->pte); |
2564 | page_add_new_anon_rmap(new_page, vma, vmf->address, false); | |
f627c2f5 | 2565 | mem_cgroup_commit_charge(new_page, memcg, false, false); |
2f38ab2c SR |
2566 | lru_cache_add_active_or_unevictable(new_page, vma); |
2567 | /* | |
2568 | * We call the notify macro here because, when using secondary | |
2569 | * mmu page tables (such as kvm shadow page tables), we want the | |
2570 | * new page to be mapped directly into the secondary page table. | |
2571 | */ | |
82b0f8c3 JK |
2572 | set_pte_at_notify(mm, vmf->address, vmf->pte, entry); |
2573 | update_mmu_cache(vma, vmf->address, vmf->pte); | |
2f38ab2c SR |
2574 | if (old_page) { |
2575 | /* | |
2576 | * Only after switching the pte to the new page may | |
2577 | * we remove the mapcount here. Otherwise another | |
2578 | * process may come and find the rmap count decremented | |
2579 | * before the pte is switched to the new page, and | |
2580 | * "reuse" the old page writing into it while our pte | |
2581 | * here still points into it and can be read by other | |
2582 | * threads. | |
2583 | * | |
2584 | * The critical issue is to order this | |
2585 | * page_remove_rmap with the ptp_clear_flush above. | |
2586 | * Those stores are ordered by (if nothing else,) | |
2587 | * the barrier present in the atomic_add_negative | |
2588 | * in page_remove_rmap. | |
2589 | * | |
2590 | * Then the TLB flush in ptep_clear_flush ensures that | |
2591 | * no process can access the old page before the | |
2592 | * decremented mapcount is visible. And the old page | |
2593 | * cannot be reused until after the decremented | |
2594 | * mapcount is visible. So transitively, TLBs to | |
2595 | * old page will be flushed before it can be reused. | |
2596 | */ | |
d281ee61 | 2597 | page_remove_rmap(old_page, false); |
2f38ab2c SR |
2598 | } |
2599 | ||
2600 | /* Free the old page.. */ | |
2601 | new_page = old_page; | |
2602 | page_copied = 1; | |
2603 | } else { | |
f627c2f5 | 2604 | mem_cgroup_cancel_charge(new_page, memcg, false); |
2f38ab2c SR |
2605 | } |
2606 | ||
2607 | if (new_page) | |
09cbfeaf | 2608 | put_page(new_page); |
2f38ab2c | 2609 | |
82b0f8c3 | 2610 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
4645b9fe JG |
2611 | /* |
2612 | * No need to double call mmu_notifier->invalidate_range() callback as | |
2613 | * the above ptep_clear_flush_notify() did already call it. | |
2614 | */ | |
2615 | mmu_notifier_invalidate_range_only_end(mm, mmun_start, mmun_end); | |
2f38ab2c SR |
2616 | if (old_page) { |
2617 | /* | |
2618 | * Don't let another task, with possibly unlocked vma, | |
2619 | * keep the mlocked page. | |
2620 | */ | |
2621 | if (page_copied && (vma->vm_flags & VM_LOCKED)) { | |
2622 | lock_page(old_page); /* LRU manipulation */ | |
e90309c9 KS |
2623 | if (PageMlocked(old_page)) |
2624 | munlock_vma_page(old_page); | |
2f38ab2c SR |
2625 | unlock_page(old_page); |
2626 | } | |
09cbfeaf | 2627 | put_page(old_page); |
2f38ab2c SR |
2628 | } |
2629 | return page_copied ? VM_FAULT_WRITE : 0; | |
2630 | oom_free_new: | |
09cbfeaf | 2631 | put_page(new_page); |
2f38ab2c SR |
2632 | oom: |
2633 | if (old_page) | |
09cbfeaf | 2634 | put_page(old_page); |
2f38ab2c SR |
2635 | return VM_FAULT_OOM; |
2636 | } | |
2637 | ||
66a6197c JK |
2638 | /** |
2639 | * finish_mkwrite_fault - finish page fault for a shared mapping, making PTE | |
2640 | * writeable once the page is prepared | |
2641 | * | |
2642 | * @vmf: structure describing the fault | |
2643 | * | |
2644 | * This function handles all that is needed to finish a write page fault in a | |
2645 | * shared mapping due to PTE being read-only once the mapped page is prepared. | |
2646 | * It handles locking of PTE and modifying it. The function returns | |
2647 | * VM_FAULT_WRITE on success, 0 when PTE got changed before we acquired PTE | |
2648 | * lock. | |
2649 | * | |
2650 | * The function expects the page to be locked or other protection against | |
2651 | * concurrent faults / writeback (such as DAX radix tree locks). | |
2652 | */ | |
2653 | int finish_mkwrite_fault(struct vm_fault *vmf) | |
2654 | { | |
2655 | WARN_ON_ONCE(!(vmf->vma->vm_flags & VM_SHARED)); | |
2656 | vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, vmf->address, | |
2657 | &vmf->ptl); | |
2658 | /* | |
2659 | * We might have raced with another page fault while we released the | |
2660 | * pte_offset_map_lock. | |
2661 | */ | |
2662 | if (!pte_same(*vmf->pte, vmf->orig_pte)) { | |
2663 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
a19e2553 | 2664 | return VM_FAULT_NOPAGE; |
66a6197c JK |
2665 | } |
2666 | wp_page_reuse(vmf); | |
a19e2553 | 2667 | return 0; |
66a6197c JK |
2668 | } |
2669 | ||
dd906184 BH |
2670 | /* |
2671 | * Handle write page faults for VM_MIXEDMAP or VM_PFNMAP for a VM_SHARED | |
2672 | * mapping | |
2673 | */ | |
2994302b | 2674 | static int wp_pfn_shared(struct vm_fault *vmf) |
dd906184 | 2675 | { |
82b0f8c3 | 2676 | struct vm_area_struct *vma = vmf->vma; |
bae473a4 | 2677 | |
dd906184 | 2678 | if (vma->vm_ops && vma->vm_ops->pfn_mkwrite) { |
dd906184 BH |
2679 | int ret; |
2680 | ||
82b0f8c3 | 2681 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
fe82221f | 2682 | vmf->flags |= FAULT_FLAG_MKWRITE; |
11bac800 | 2683 | ret = vma->vm_ops->pfn_mkwrite(vmf); |
2f89dc12 | 2684 | if (ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)) |
dd906184 | 2685 | return ret; |
66a6197c | 2686 | return finish_mkwrite_fault(vmf); |
dd906184 | 2687 | } |
997dd98d JK |
2688 | wp_page_reuse(vmf); |
2689 | return VM_FAULT_WRITE; | |
dd906184 BH |
2690 | } |
2691 | ||
a41b70d6 | 2692 | static int wp_page_shared(struct vm_fault *vmf) |
82b0f8c3 | 2693 | __releases(vmf->ptl) |
93e478d4 | 2694 | { |
82b0f8c3 | 2695 | struct vm_area_struct *vma = vmf->vma; |
93e478d4 | 2696 | |
a41b70d6 | 2697 | get_page(vmf->page); |
93e478d4 | 2698 | |
93e478d4 SR |
2699 | if (vma->vm_ops && vma->vm_ops->page_mkwrite) { |
2700 | int tmp; | |
2701 | ||
82b0f8c3 | 2702 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
38b8cb7f | 2703 | tmp = do_page_mkwrite(vmf); |
93e478d4 SR |
2704 | if (unlikely(!tmp || (tmp & |
2705 | (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { | |
a41b70d6 | 2706 | put_page(vmf->page); |
93e478d4 SR |
2707 | return tmp; |
2708 | } | |
66a6197c | 2709 | tmp = finish_mkwrite_fault(vmf); |
a19e2553 | 2710 | if (unlikely(tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { |
a41b70d6 | 2711 | unlock_page(vmf->page); |
a41b70d6 | 2712 | put_page(vmf->page); |
66a6197c | 2713 | return tmp; |
93e478d4 | 2714 | } |
66a6197c JK |
2715 | } else { |
2716 | wp_page_reuse(vmf); | |
997dd98d | 2717 | lock_page(vmf->page); |
93e478d4 | 2718 | } |
997dd98d JK |
2719 | fault_dirty_shared_page(vma, vmf->page); |
2720 | put_page(vmf->page); | |
93e478d4 | 2721 | |
997dd98d | 2722 | return VM_FAULT_WRITE; |
93e478d4 SR |
2723 | } |
2724 | ||
1da177e4 LT |
2725 | /* |
2726 | * This routine handles present pages, when users try to write | |
2727 | * to a shared page. It is done by copying the page to a new address | |
2728 | * and decrementing the shared-page counter for the old page. | |
2729 | * | |
1da177e4 LT |
2730 | * Note that this routine assumes that the protection checks have been |
2731 | * done by the caller (the low-level page fault routine in most cases). | |
2732 | * Thus we can safely just mark it writable once we've done any necessary | |
2733 | * COW. | |
2734 | * | |
2735 | * We also mark the page dirty at this point even though the page will | |
2736 | * change only once the write actually happens. This avoids a few races, | |
2737 | * and potentially makes it more efficient. | |
2738 | * | |
8f4e2101 HD |
2739 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2740 | * but allow concurrent faults), with pte both mapped and locked. | |
2741 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2742 | */ |
2994302b | 2743 | static int do_wp_page(struct vm_fault *vmf) |
82b0f8c3 | 2744 | __releases(vmf->ptl) |
1da177e4 | 2745 | { |
82b0f8c3 | 2746 | struct vm_area_struct *vma = vmf->vma; |
1da177e4 | 2747 | |
a41b70d6 JK |
2748 | vmf->page = vm_normal_page(vma, vmf->address, vmf->orig_pte); |
2749 | if (!vmf->page) { | |
251b97f5 | 2750 | /* |
64e45507 PF |
2751 | * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a |
2752 | * VM_PFNMAP VMA. | |
251b97f5 PZ |
2753 | * |
2754 | * We should not cow pages in a shared writeable mapping. | |
dd906184 | 2755 | * Just mark the pages writable and/or call ops->pfn_mkwrite. |
251b97f5 PZ |
2756 | */ |
2757 | if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) == | |
2758 | (VM_WRITE|VM_SHARED)) | |
2994302b | 2759 | return wp_pfn_shared(vmf); |
2f38ab2c | 2760 | |
82b0f8c3 | 2761 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 2762 | return wp_page_copy(vmf); |
251b97f5 | 2763 | } |
1da177e4 | 2764 | |
d08b3851 | 2765 | /* |
ee6a6457 PZ |
2766 | * Take out anonymous pages first, anonymous shared vmas are |
2767 | * not dirty accountable. | |
d08b3851 | 2768 | */ |
a41b70d6 | 2769 | if (PageAnon(vmf->page) && !PageKsm(vmf->page)) { |
ba3c4ce6 | 2770 | int total_map_swapcount; |
a41b70d6 JK |
2771 | if (!trylock_page(vmf->page)) { |
2772 | get_page(vmf->page); | |
82b0f8c3 | 2773 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 2774 | lock_page(vmf->page); |
82b0f8c3 JK |
2775 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
2776 | vmf->address, &vmf->ptl); | |
2994302b | 2777 | if (!pte_same(*vmf->pte, vmf->orig_pte)) { |
a41b70d6 | 2778 | unlock_page(vmf->page); |
82b0f8c3 | 2779 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 2780 | put_page(vmf->page); |
28766805 | 2781 | return 0; |
ab967d86 | 2782 | } |
a41b70d6 | 2783 | put_page(vmf->page); |
ee6a6457 | 2784 | } |
ba3c4ce6 HY |
2785 | if (reuse_swap_page(vmf->page, &total_map_swapcount)) { |
2786 | if (total_map_swapcount == 1) { | |
6d0a07ed AA |
2787 | /* |
2788 | * The page is all ours. Move it to | |
2789 | * our anon_vma so the rmap code will | |
2790 | * not search our parent or siblings. | |
2791 | * Protected against the rmap code by | |
2792 | * the page lock. | |
2793 | */ | |
a41b70d6 | 2794 | page_move_anon_rmap(vmf->page, vma); |
6d0a07ed | 2795 | } |
a41b70d6 | 2796 | unlock_page(vmf->page); |
997dd98d JK |
2797 | wp_page_reuse(vmf); |
2798 | return VM_FAULT_WRITE; | |
b009c024 | 2799 | } |
a41b70d6 | 2800 | unlock_page(vmf->page); |
ee6a6457 | 2801 | } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) == |
d08b3851 | 2802 | (VM_WRITE|VM_SHARED))) { |
a41b70d6 | 2803 | return wp_page_shared(vmf); |
1da177e4 | 2804 | } |
1da177e4 LT |
2805 | |
2806 | /* | |
2807 | * Ok, we need to copy. Oh, well.. | |
2808 | */ | |
a41b70d6 | 2809 | get_page(vmf->page); |
28766805 | 2810 | |
82b0f8c3 | 2811 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 2812 | return wp_page_copy(vmf); |
1da177e4 LT |
2813 | } |
2814 | ||
97a89413 | 2815 | static void unmap_mapping_range_vma(struct vm_area_struct *vma, |
1da177e4 LT |
2816 | unsigned long start_addr, unsigned long end_addr, |
2817 | struct zap_details *details) | |
2818 | { | |
f5cc4eef | 2819 | zap_page_range_single(vma, start_addr, end_addr - start_addr, details); |
1da177e4 LT |
2820 | } |
2821 | ||
f808c13f | 2822 | static inline void unmap_mapping_range_tree(struct rb_root_cached *root, |
1da177e4 LT |
2823 | struct zap_details *details) |
2824 | { | |
2825 | struct vm_area_struct *vma; | |
1da177e4 LT |
2826 | pgoff_t vba, vea, zba, zea; |
2827 | ||
6b2dbba8 | 2828 | vma_interval_tree_foreach(vma, root, |
1da177e4 | 2829 | details->first_index, details->last_index) { |
1da177e4 LT |
2830 | |
2831 | vba = vma->vm_pgoff; | |
d6e93217 | 2832 | vea = vba + vma_pages(vma) - 1; |
1da177e4 LT |
2833 | zba = details->first_index; |
2834 | if (zba < vba) | |
2835 | zba = vba; | |
2836 | zea = details->last_index; | |
2837 | if (zea > vea) | |
2838 | zea = vea; | |
2839 | ||
97a89413 | 2840 | unmap_mapping_range_vma(vma, |
1da177e4 LT |
2841 | ((zba - vba) << PAGE_SHIFT) + vma->vm_start, |
2842 | ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, | |
97a89413 | 2843 | details); |
1da177e4 LT |
2844 | } |
2845 | } | |
2846 | ||
977fbdcd MW |
2847 | /** |
2848 | * unmap_mapping_pages() - Unmap pages from processes. | |
2849 | * @mapping: The address space containing pages to be unmapped. | |
2850 | * @start: Index of first page to be unmapped. | |
2851 | * @nr: Number of pages to be unmapped. 0 to unmap to end of file. | |
2852 | * @even_cows: Whether to unmap even private COWed pages. | |
2853 | * | |
2854 | * Unmap the pages in this address space from any userspace process which | |
2855 | * has them mmaped. Generally, you want to remove COWed pages as well when | |
2856 | * a file is being truncated, but not when invalidating pages from the page | |
2857 | * cache. | |
2858 | */ | |
2859 | void unmap_mapping_pages(struct address_space *mapping, pgoff_t start, | |
2860 | pgoff_t nr, bool even_cows) | |
2861 | { | |
2862 | struct zap_details details = { }; | |
2863 | ||
2864 | details.check_mapping = even_cows ? NULL : mapping; | |
2865 | details.first_index = start; | |
2866 | details.last_index = start + nr - 1; | |
2867 | if (details.last_index < details.first_index) | |
2868 | details.last_index = ULONG_MAX; | |
2869 | ||
2870 | i_mmap_lock_write(mapping); | |
2871 | if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))) | |
2872 | unmap_mapping_range_tree(&mapping->i_mmap, &details); | |
2873 | i_mmap_unlock_write(mapping); | |
2874 | } | |
2875 | ||
1da177e4 | 2876 | /** |
8a5f14a2 | 2877 | * unmap_mapping_range - unmap the portion of all mmaps in the specified |
977fbdcd | 2878 | * address_space corresponding to the specified byte range in the underlying |
8a5f14a2 KS |
2879 | * file. |
2880 | * | |
3d41088f | 2881 | * @mapping: the address space containing mmaps to be unmapped. |
1da177e4 LT |
2882 | * @holebegin: byte in first page to unmap, relative to the start of |
2883 | * the underlying file. This will be rounded down to a PAGE_SIZE | |
25d9e2d1 | 2884 | * boundary. Note that this is different from truncate_pagecache(), which |
1da177e4 LT |
2885 | * must keep the partial page. In contrast, we must get rid of |
2886 | * partial pages. | |
2887 | * @holelen: size of prospective hole in bytes. This will be rounded | |
2888 | * up to a PAGE_SIZE boundary. A holelen of zero truncates to the | |
2889 | * end of the file. | |
2890 | * @even_cows: 1 when truncating a file, unmap even private COWed pages; | |
2891 | * but 0 when invalidating pagecache, don't throw away private data. | |
2892 | */ | |
2893 | void unmap_mapping_range(struct address_space *mapping, | |
2894 | loff_t const holebegin, loff_t const holelen, int even_cows) | |
2895 | { | |
1da177e4 LT |
2896 | pgoff_t hba = holebegin >> PAGE_SHIFT; |
2897 | pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
2898 | ||
2899 | /* Check for overflow. */ | |
2900 | if (sizeof(holelen) > sizeof(hlen)) { | |
2901 | long long holeend = | |
2902 | (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
2903 | if (holeend & ~(long long)ULONG_MAX) | |
2904 | hlen = ULONG_MAX - hba + 1; | |
2905 | } | |
2906 | ||
977fbdcd | 2907 | unmap_mapping_pages(mapping, hba, hlen, even_cows); |
1da177e4 LT |
2908 | } |
2909 | EXPORT_SYMBOL(unmap_mapping_range); | |
2910 | ||
1da177e4 | 2911 | /* |
8f4e2101 HD |
2912 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2913 | * but allow concurrent faults), and pte mapped but not yet locked. | |
9a95f3cf PC |
2914 | * We return with pte unmapped and unlocked. |
2915 | * | |
2916 | * We return with the mmap_sem locked or unlocked in the same cases | |
2917 | * as does filemap_fault(). | |
1da177e4 | 2918 | */ |
2994302b | 2919 | int do_swap_page(struct vm_fault *vmf) |
1da177e4 | 2920 | { |
82b0f8c3 | 2921 | struct vm_area_struct *vma = vmf->vma; |
eaf649eb | 2922 | struct page *page = NULL, *swapcache; |
00501b53 | 2923 | struct mem_cgroup *memcg; |
65500d23 | 2924 | swp_entry_t entry; |
1da177e4 | 2925 | pte_t pte; |
d065bd81 | 2926 | int locked; |
ad8c2ee8 | 2927 | int exclusive = 0; |
83c54070 | 2928 | int ret = 0; |
1da177e4 | 2929 | |
eaf649eb | 2930 | if (!pte_unmap_same(vma->vm_mm, vmf->pmd, vmf->pte, vmf->orig_pte)) |
8f4e2101 | 2931 | goto out; |
65500d23 | 2932 | |
2994302b | 2933 | entry = pte_to_swp_entry(vmf->orig_pte); |
d1737fdb AK |
2934 | if (unlikely(non_swap_entry(entry))) { |
2935 | if (is_migration_entry(entry)) { | |
82b0f8c3 JK |
2936 | migration_entry_wait(vma->vm_mm, vmf->pmd, |
2937 | vmf->address); | |
5042db43 JG |
2938 | } else if (is_device_private_entry(entry)) { |
2939 | /* | |
2940 | * For un-addressable device memory we call the pgmap | |
2941 | * fault handler callback. The callback must migrate | |
2942 | * the page back to some CPU accessible page. | |
2943 | */ | |
2944 | ret = device_private_entry_fault(vma, vmf->address, entry, | |
2945 | vmf->flags, vmf->pmd); | |
d1737fdb AK |
2946 | } else if (is_hwpoison_entry(entry)) { |
2947 | ret = VM_FAULT_HWPOISON; | |
2948 | } else { | |
2994302b | 2949 | print_bad_pte(vma, vmf->address, vmf->orig_pte, NULL); |
d99be1a8 | 2950 | ret = VM_FAULT_SIGBUS; |
d1737fdb | 2951 | } |
0697212a CL |
2952 | goto out; |
2953 | } | |
0bcac06f MK |
2954 | |
2955 | ||
0ff92245 | 2956 | delayacct_set_flag(DELAYACCT_PF_SWAPIN); |
eaf649eb MK |
2957 | page = lookup_swap_cache(entry, vma, vmf->address); |
2958 | swapcache = page; | |
f8020772 | 2959 | |
1da177e4 | 2960 | if (!page) { |
0bcac06f MK |
2961 | struct swap_info_struct *si = swp_swap_info(entry); |
2962 | ||
aa8d22a1 MK |
2963 | if (si->flags & SWP_SYNCHRONOUS_IO && |
2964 | __swap_count(si, entry) == 1) { | |
0bcac06f | 2965 | /* skip swapcache */ |
e9e9b7ec MK |
2966 | page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, |
2967 | vmf->address); | |
0bcac06f MK |
2968 | if (page) { |
2969 | __SetPageLocked(page); | |
2970 | __SetPageSwapBacked(page); | |
2971 | set_page_private(page, entry.val); | |
2972 | lru_cache_add_anon(page); | |
2973 | swap_readpage(page, true); | |
2974 | } | |
aa8d22a1 | 2975 | } else { |
e9e9b7ec MK |
2976 | page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE, |
2977 | vmf); | |
aa8d22a1 | 2978 | swapcache = page; |
0bcac06f MK |
2979 | } |
2980 | ||
1da177e4 LT |
2981 | if (!page) { |
2982 | /* | |
8f4e2101 HD |
2983 | * Back out if somebody else faulted in this pte |
2984 | * while we released the pte lock. | |
1da177e4 | 2985 | */ |
82b0f8c3 JK |
2986 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
2987 | vmf->address, &vmf->ptl); | |
2994302b | 2988 | if (likely(pte_same(*vmf->pte, vmf->orig_pte))) |
1da177e4 | 2989 | ret = VM_FAULT_OOM; |
0ff92245 | 2990 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
65500d23 | 2991 | goto unlock; |
1da177e4 LT |
2992 | } |
2993 | ||
2994 | /* Had to read the page from swap area: Major fault */ | |
2995 | ret = VM_FAULT_MAJOR; | |
f8891e5e | 2996 | count_vm_event(PGMAJFAULT); |
2262185c | 2997 | count_memcg_event_mm(vma->vm_mm, PGMAJFAULT); |
d1737fdb | 2998 | } else if (PageHWPoison(page)) { |
71f72525 WF |
2999 | /* |
3000 | * hwpoisoned dirty swapcache pages are kept for killing | |
3001 | * owner processes (which may be unknown at hwpoison time) | |
3002 | */ | |
d1737fdb AK |
3003 | ret = VM_FAULT_HWPOISON; |
3004 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); | |
4779cb31 | 3005 | goto out_release; |
1da177e4 LT |
3006 | } |
3007 | ||
82b0f8c3 | 3008 | locked = lock_page_or_retry(page, vma->vm_mm, vmf->flags); |
e709ffd6 | 3009 | |
073e587e | 3010 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
d065bd81 ML |
3011 | if (!locked) { |
3012 | ret |= VM_FAULT_RETRY; | |
3013 | goto out_release; | |
3014 | } | |
073e587e | 3015 | |
4969c119 | 3016 | /* |
31c4a3d3 HD |
3017 | * Make sure try_to_free_swap or reuse_swap_page or swapoff did not |
3018 | * release the swapcache from under us. The page pin, and pte_same | |
3019 | * test below, are not enough to exclude that. Even if it is still | |
3020 | * swapcache, we need to check that the page's swap has not changed. | |
4969c119 | 3021 | */ |
0bcac06f MK |
3022 | if (unlikely((!PageSwapCache(page) || |
3023 | page_private(page) != entry.val)) && swapcache) | |
4969c119 AA |
3024 | goto out_page; |
3025 | ||
82b0f8c3 | 3026 | page = ksm_might_need_to_copy(page, vma, vmf->address); |
cbf86cfe HD |
3027 | if (unlikely(!page)) { |
3028 | ret = VM_FAULT_OOM; | |
3029 | page = swapcache; | |
cbf86cfe | 3030 | goto out_page; |
5ad64688 HD |
3031 | } |
3032 | ||
2cf85583 TH |
3033 | if (mem_cgroup_try_charge_delay(page, vma->vm_mm, GFP_KERNEL, |
3034 | &memcg, false)) { | |
8a9f3ccd | 3035 | ret = VM_FAULT_OOM; |
bc43f75c | 3036 | goto out_page; |
8a9f3ccd BS |
3037 | } |
3038 | ||
1da177e4 | 3039 | /* |
8f4e2101 | 3040 | * Back out if somebody else already faulted in this pte. |
1da177e4 | 3041 | */ |
82b0f8c3 JK |
3042 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, |
3043 | &vmf->ptl); | |
2994302b | 3044 | if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) |
b8107480 | 3045 | goto out_nomap; |
b8107480 KK |
3046 | |
3047 | if (unlikely(!PageUptodate(page))) { | |
3048 | ret = VM_FAULT_SIGBUS; | |
3049 | goto out_nomap; | |
1da177e4 LT |
3050 | } |
3051 | ||
8c7c6e34 KH |
3052 | /* |
3053 | * The page isn't present yet, go ahead with the fault. | |
3054 | * | |
3055 | * Be careful about the sequence of operations here. | |
3056 | * To get its accounting right, reuse_swap_page() must be called | |
3057 | * while the page is counted on swap but not yet in mapcount i.e. | |
3058 | * before page_add_anon_rmap() and swap_free(); try_to_free_swap() | |
3059 | * must be called after the swap_free(), or it will never succeed. | |
8c7c6e34 | 3060 | */ |
1da177e4 | 3061 | |
bae473a4 KS |
3062 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
3063 | dec_mm_counter_fast(vma->vm_mm, MM_SWAPENTS); | |
1da177e4 | 3064 | pte = mk_pte(page, vma->vm_page_prot); |
82b0f8c3 | 3065 | if ((vmf->flags & FAULT_FLAG_WRITE) && reuse_swap_page(page, NULL)) { |
1da177e4 | 3066 | pte = maybe_mkwrite(pte_mkdirty(pte), vma); |
82b0f8c3 | 3067 | vmf->flags &= ~FAULT_FLAG_WRITE; |
9a5b489b | 3068 | ret |= VM_FAULT_WRITE; |
d281ee61 | 3069 | exclusive = RMAP_EXCLUSIVE; |
1da177e4 | 3070 | } |
1da177e4 | 3071 | flush_icache_page(vma, page); |
2994302b | 3072 | if (pte_swp_soft_dirty(vmf->orig_pte)) |
179ef71c | 3073 | pte = pte_mksoft_dirty(pte); |
82b0f8c3 | 3074 | set_pte_at(vma->vm_mm, vmf->address, vmf->pte, pte); |
ca827d55 | 3075 | arch_do_swap_page(vma->vm_mm, vma, vmf->address, pte, vmf->orig_pte); |
2994302b | 3076 | vmf->orig_pte = pte; |
0bcac06f MK |
3077 | |
3078 | /* ksm created a completely new copy */ | |
3079 | if (unlikely(page != swapcache && swapcache)) { | |
82b0f8c3 | 3080 | page_add_new_anon_rmap(page, vma, vmf->address, false); |
f627c2f5 | 3081 | mem_cgroup_commit_charge(page, memcg, false, false); |
00501b53 | 3082 | lru_cache_add_active_or_unevictable(page, vma); |
0bcac06f MK |
3083 | } else { |
3084 | do_page_add_anon_rmap(page, vma, vmf->address, exclusive); | |
3085 | mem_cgroup_commit_charge(page, memcg, true, false); | |
3086 | activate_page(page); | |
00501b53 | 3087 | } |
1da177e4 | 3088 | |
c475a8ab | 3089 | swap_free(entry); |
5ccc5aba VD |
3090 | if (mem_cgroup_swap_full(page) || |
3091 | (vma->vm_flags & VM_LOCKED) || PageMlocked(page)) | |
a2c43eed | 3092 | try_to_free_swap(page); |
c475a8ab | 3093 | unlock_page(page); |
0bcac06f | 3094 | if (page != swapcache && swapcache) { |
4969c119 AA |
3095 | /* |
3096 | * Hold the lock to avoid the swap entry to be reused | |
3097 | * until we take the PT lock for the pte_same() check | |
3098 | * (to avoid false positives from pte_same). For | |
3099 | * further safety release the lock after the swap_free | |
3100 | * so that the swap count won't change under a | |
3101 | * parallel locked swapcache. | |
3102 | */ | |
3103 | unlock_page(swapcache); | |
09cbfeaf | 3104 | put_page(swapcache); |
4969c119 | 3105 | } |
c475a8ab | 3106 | |
82b0f8c3 | 3107 | if (vmf->flags & FAULT_FLAG_WRITE) { |
2994302b | 3108 | ret |= do_wp_page(vmf); |
61469f1d HD |
3109 | if (ret & VM_FAULT_ERROR) |
3110 | ret &= VM_FAULT_ERROR; | |
1da177e4 LT |
3111 | goto out; |
3112 | } | |
3113 | ||
3114 | /* No need to invalidate - it was non-present before */ | |
82b0f8c3 | 3115 | update_mmu_cache(vma, vmf->address, vmf->pte); |
65500d23 | 3116 | unlock: |
82b0f8c3 | 3117 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
1da177e4 LT |
3118 | out: |
3119 | return ret; | |
b8107480 | 3120 | out_nomap: |
f627c2f5 | 3121 | mem_cgroup_cancel_charge(page, memcg, false); |
82b0f8c3 | 3122 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
bc43f75c | 3123 | out_page: |
b8107480 | 3124 | unlock_page(page); |
4779cb31 | 3125 | out_release: |
09cbfeaf | 3126 | put_page(page); |
0bcac06f | 3127 | if (page != swapcache && swapcache) { |
4969c119 | 3128 | unlock_page(swapcache); |
09cbfeaf | 3129 | put_page(swapcache); |
4969c119 | 3130 | } |
65500d23 | 3131 | return ret; |
1da177e4 LT |
3132 | } |
3133 | ||
3134 | /* | |
8f4e2101 HD |
3135 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
3136 | * but allow concurrent faults), and pte mapped but not yet locked. | |
3137 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 3138 | */ |
82b0f8c3 | 3139 | static int do_anonymous_page(struct vm_fault *vmf) |
1da177e4 | 3140 | { |
82b0f8c3 | 3141 | struct vm_area_struct *vma = vmf->vma; |
00501b53 | 3142 | struct mem_cgroup *memcg; |
8f4e2101 | 3143 | struct page *page; |
6b31d595 | 3144 | int ret = 0; |
1da177e4 | 3145 | pte_t entry; |
1da177e4 | 3146 | |
6b7339f4 KS |
3147 | /* File mapping without ->vm_ops ? */ |
3148 | if (vma->vm_flags & VM_SHARED) | |
3149 | return VM_FAULT_SIGBUS; | |
3150 | ||
7267ec00 KS |
3151 | /* |
3152 | * Use pte_alloc() instead of pte_alloc_map(). We can't run | |
3153 | * pte_offset_map() on pmds where a huge pmd might be created | |
3154 | * from a different thread. | |
3155 | * | |
3156 | * pte_alloc_map() is safe to use under down_write(mmap_sem) or when | |
3157 | * parallel threads are excluded by other means. | |
3158 | * | |
3159 | * Here we only have down_read(mmap_sem). | |
3160 | */ | |
82b0f8c3 | 3161 | if (pte_alloc(vma->vm_mm, vmf->pmd, vmf->address)) |
7267ec00 KS |
3162 | return VM_FAULT_OOM; |
3163 | ||
3164 | /* See the comment in pte_alloc_one_map() */ | |
82b0f8c3 | 3165 | if (unlikely(pmd_trans_unstable(vmf->pmd))) |
7267ec00 KS |
3166 | return 0; |
3167 | ||
11ac5524 | 3168 | /* Use the zero-page for reads */ |
82b0f8c3 | 3169 | if (!(vmf->flags & FAULT_FLAG_WRITE) && |
bae473a4 | 3170 | !mm_forbids_zeropage(vma->vm_mm)) { |
82b0f8c3 | 3171 | entry = pte_mkspecial(pfn_pte(my_zero_pfn(vmf->address), |
62eede62 | 3172 | vma->vm_page_prot)); |
82b0f8c3 JK |
3173 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
3174 | vmf->address, &vmf->ptl); | |
3175 | if (!pte_none(*vmf->pte)) | |
a13ea5b7 | 3176 | goto unlock; |
6b31d595 MH |
3177 | ret = check_stable_address_space(vma->vm_mm); |
3178 | if (ret) | |
3179 | goto unlock; | |
6b251fc9 AA |
3180 | /* Deliver the page fault to userland, check inside PT lock */ |
3181 | if (userfaultfd_missing(vma)) { | |
82b0f8c3 JK |
3182 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
3183 | return handle_userfault(vmf, VM_UFFD_MISSING); | |
6b251fc9 | 3184 | } |
a13ea5b7 HD |
3185 | goto setpte; |
3186 | } | |
3187 | ||
557ed1fa | 3188 | /* Allocate our own private page. */ |
557ed1fa NP |
3189 | if (unlikely(anon_vma_prepare(vma))) |
3190 | goto oom; | |
82b0f8c3 | 3191 | page = alloc_zeroed_user_highpage_movable(vma, vmf->address); |
557ed1fa NP |
3192 | if (!page) |
3193 | goto oom; | |
eb3c24f3 | 3194 | |
2cf85583 TH |
3195 | if (mem_cgroup_try_charge_delay(page, vma->vm_mm, GFP_KERNEL, &memcg, |
3196 | false)) | |
eb3c24f3 MG |
3197 | goto oom_free_page; |
3198 | ||
52f37629 MK |
3199 | /* |
3200 | * The memory barrier inside __SetPageUptodate makes sure that | |
3201 | * preceeding stores to the page contents become visible before | |
3202 | * the set_pte_at() write. | |
3203 | */ | |
0ed361de | 3204 | __SetPageUptodate(page); |
8f4e2101 | 3205 | |
557ed1fa | 3206 | entry = mk_pte(page, vma->vm_page_prot); |
1ac0cb5d HD |
3207 | if (vma->vm_flags & VM_WRITE) |
3208 | entry = pte_mkwrite(pte_mkdirty(entry)); | |
1da177e4 | 3209 | |
82b0f8c3 JK |
3210 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, |
3211 | &vmf->ptl); | |
3212 | if (!pte_none(*vmf->pte)) | |
557ed1fa | 3213 | goto release; |
9ba69294 | 3214 | |
6b31d595 MH |
3215 | ret = check_stable_address_space(vma->vm_mm); |
3216 | if (ret) | |
3217 | goto release; | |
3218 | ||
6b251fc9 AA |
3219 | /* Deliver the page fault to userland, check inside PT lock */ |
3220 | if (userfaultfd_missing(vma)) { | |
82b0f8c3 | 3221 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
f627c2f5 | 3222 | mem_cgroup_cancel_charge(page, memcg, false); |
09cbfeaf | 3223 | put_page(page); |
82b0f8c3 | 3224 | return handle_userfault(vmf, VM_UFFD_MISSING); |
6b251fc9 AA |
3225 | } |
3226 | ||
bae473a4 | 3227 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
82b0f8c3 | 3228 | page_add_new_anon_rmap(page, vma, vmf->address, false); |
f627c2f5 | 3229 | mem_cgroup_commit_charge(page, memcg, false, false); |
00501b53 | 3230 | lru_cache_add_active_or_unevictable(page, vma); |
a13ea5b7 | 3231 | setpte: |
82b0f8c3 | 3232 | set_pte_at(vma->vm_mm, vmf->address, vmf->pte, entry); |
1da177e4 LT |
3233 | |
3234 | /* No need to invalidate - it was non-present before */ | |
82b0f8c3 | 3235 | update_mmu_cache(vma, vmf->address, vmf->pte); |
65500d23 | 3236 | unlock: |
82b0f8c3 | 3237 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
6b31d595 | 3238 | return ret; |
8f4e2101 | 3239 | release: |
f627c2f5 | 3240 | mem_cgroup_cancel_charge(page, memcg, false); |
09cbfeaf | 3241 | put_page(page); |
8f4e2101 | 3242 | goto unlock; |
8a9f3ccd | 3243 | oom_free_page: |
09cbfeaf | 3244 | put_page(page); |
65500d23 | 3245 | oom: |
1da177e4 LT |
3246 | return VM_FAULT_OOM; |
3247 | } | |
3248 | ||
9a95f3cf PC |
3249 | /* |
3250 | * The mmap_sem must have been held on entry, and may have been | |
3251 | * released depending on flags and vma->vm_ops->fault() return value. | |
3252 | * See filemap_fault() and __lock_page_retry(). | |
3253 | */ | |
936ca80d | 3254 | static int __do_fault(struct vm_fault *vmf) |
7eae74af | 3255 | { |
82b0f8c3 | 3256 | struct vm_area_struct *vma = vmf->vma; |
7eae74af KS |
3257 | int ret; |
3258 | ||
11bac800 | 3259 | ret = vma->vm_ops->fault(vmf); |
3917048d | 3260 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY | |
b1aa812b | 3261 | VM_FAULT_DONE_COW))) |
bc2466e4 | 3262 | return ret; |
7eae74af | 3263 | |
667240e0 | 3264 | if (unlikely(PageHWPoison(vmf->page))) { |
7eae74af | 3265 | if (ret & VM_FAULT_LOCKED) |
667240e0 JK |
3266 | unlock_page(vmf->page); |
3267 | put_page(vmf->page); | |
936ca80d | 3268 | vmf->page = NULL; |
7eae74af KS |
3269 | return VM_FAULT_HWPOISON; |
3270 | } | |
3271 | ||
3272 | if (unlikely(!(ret & VM_FAULT_LOCKED))) | |
667240e0 | 3273 | lock_page(vmf->page); |
7eae74af | 3274 | else |
667240e0 | 3275 | VM_BUG_ON_PAGE(!PageLocked(vmf->page), vmf->page); |
7eae74af | 3276 | |
7eae74af KS |
3277 | return ret; |
3278 | } | |
3279 | ||
d0f0931d RZ |
3280 | /* |
3281 | * The ordering of these checks is important for pmds with _PAGE_DEVMAP set. | |
3282 | * If we check pmd_trans_unstable() first we will trip the bad_pmd() check | |
3283 | * inside of pmd_none_or_trans_huge_or_clear_bad(). This will end up correctly | |
3284 | * returning 1 but not before it spams dmesg with the pmd_clear_bad() output. | |
3285 | */ | |
3286 | static int pmd_devmap_trans_unstable(pmd_t *pmd) | |
3287 | { | |
3288 | return pmd_devmap(*pmd) || pmd_trans_unstable(pmd); | |
3289 | } | |
3290 | ||
82b0f8c3 | 3291 | static int pte_alloc_one_map(struct vm_fault *vmf) |
7267ec00 | 3292 | { |
82b0f8c3 | 3293 | struct vm_area_struct *vma = vmf->vma; |
7267ec00 | 3294 | |
82b0f8c3 | 3295 | if (!pmd_none(*vmf->pmd)) |
7267ec00 | 3296 | goto map_pte; |
82b0f8c3 JK |
3297 | if (vmf->prealloc_pte) { |
3298 | vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd); | |
3299 | if (unlikely(!pmd_none(*vmf->pmd))) { | |
3300 | spin_unlock(vmf->ptl); | |
7267ec00 KS |
3301 | goto map_pte; |
3302 | } | |
3303 | ||
c4812909 | 3304 | mm_inc_nr_ptes(vma->vm_mm); |
82b0f8c3 JK |
3305 | pmd_populate(vma->vm_mm, vmf->pmd, vmf->prealloc_pte); |
3306 | spin_unlock(vmf->ptl); | |
7f2b6ce8 | 3307 | vmf->prealloc_pte = NULL; |
82b0f8c3 | 3308 | } else if (unlikely(pte_alloc(vma->vm_mm, vmf->pmd, vmf->address))) { |
7267ec00 KS |
3309 | return VM_FAULT_OOM; |
3310 | } | |
3311 | map_pte: | |
3312 | /* | |
3313 | * If a huge pmd materialized under us just retry later. Use | |
d0f0931d RZ |
3314 | * pmd_trans_unstable() via pmd_devmap_trans_unstable() instead of |
3315 | * pmd_trans_huge() to ensure the pmd didn't become pmd_trans_huge | |
3316 | * under us and then back to pmd_none, as a result of MADV_DONTNEED | |
3317 | * running immediately after a huge pmd fault in a different thread of | |
3318 | * this mm, in turn leading to a misleading pmd_trans_huge() retval. | |
3319 | * All we have to ensure is that it is a regular pmd that we can walk | |
3320 | * with pte_offset_map() and we can do that through an atomic read in | |
3321 | * C, which is what pmd_trans_unstable() provides. | |
7267ec00 | 3322 | */ |
d0f0931d | 3323 | if (pmd_devmap_trans_unstable(vmf->pmd)) |
7267ec00 KS |
3324 | return VM_FAULT_NOPAGE; |
3325 | ||
d0f0931d RZ |
3326 | /* |
3327 | * At this point we know that our vmf->pmd points to a page of ptes | |
3328 | * and it cannot become pmd_none(), pmd_devmap() or pmd_trans_huge() | |
3329 | * for the duration of the fault. If a racing MADV_DONTNEED runs and | |
3330 | * we zap the ptes pointed to by our vmf->pmd, the vmf->ptl will still | |
3331 | * be valid and we will re-check to make sure the vmf->pte isn't | |
3332 | * pte_none() under vmf->ptl protection when we return to | |
3333 | * alloc_set_pte(). | |
3334 | */ | |
82b0f8c3 JK |
3335 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, |
3336 | &vmf->ptl); | |
7267ec00 KS |
3337 | return 0; |
3338 | } | |
3339 | ||
e496cf3d | 3340 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
10102459 KS |
3341 | |
3342 | #define HPAGE_CACHE_INDEX_MASK (HPAGE_PMD_NR - 1) | |
3343 | static inline bool transhuge_vma_suitable(struct vm_area_struct *vma, | |
3344 | unsigned long haddr) | |
3345 | { | |
3346 | if (((vma->vm_start >> PAGE_SHIFT) & HPAGE_CACHE_INDEX_MASK) != | |
3347 | (vma->vm_pgoff & HPAGE_CACHE_INDEX_MASK)) | |
3348 | return false; | |
3349 | if (haddr < vma->vm_start || haddr + HPAGE_PMD_SIZE > vma->vm_end) | |
3350 | return false; | |
3351 | return true; | |
3352 | } | |
3353 | ||
82b0f8c3 | 3354 | static void deposit_prealloc_pte(struct vm_fault *vmf) |
953c66c2 | 3355 | { |
82b0f8c3 | 3356 | struct vm_area_struct *vma = vmf->vma; |
953c66c2 | 3357 | |
82b0f8c3 | 3358 | pgtable_trans_huge_deposit(vma->vm_mm, vmf->pmd, vmf->prealloc_pte); |
953c66c2 AK |
3359 | /* |
3360 | * We are going to consume the prealloc table, | |
3361 | * count that as nr_ptes. | |
3362 | */ | |
c4812909 | 3363 | mm_inc_nr_ptes(vma->vm_mm); |
7f2b6ce8 | 3364 | vmf->prealloc_pte = NULL; |
953c66c2 AK |
3365 | } |
3366 | ||
82b0f8c3 | 3367 | static int do_set_pmd(struct vm_fault *vmf, struct page *page) |
10102459 | 3368 | { |
82b0f8c3 JK |
3369 | struct vm_area_struct *vma = vmf->vma; |
3370 | bool write = vmf->flags & FAULT_FLAG_WRITE; | |
3371 | unsigned long haddr = vmf->address & HPAGE_PMD_MASK; | |
10102459 KS |
3372 | pmd_t entry; |
3373 | int i, ret; | |
3374 | ||
3375 | if (!transhuge_vma_suitable(vma, haddr)) | |
3376 | return VM_FAULT_FALLBACK; | |
3377 | ||
3378 | ret = VM_FAULT_FALLBACK; | |
3379 | page = compound_head(page); | |
3380 | ||
953c66c2 AK |
3381 | /* |
3382 | * Archs like ppc64 need additonal space to store information | |
3383 | * related to pte entry. Use the preallocated table for that. | |
3384 | */ | |
82b0f8c3 JK |
3385 | if (arch_needs_pgtable_deposit() && !vmf->prealloc_pte) { |
3386 | vmf->prealloc_pte = pte_alloc_one(vma->vm_mm, vmf->address); | |
3387 | if (!vmf->prealloc_pte) | |
953c66c2 AK |
3388 | return VM_FAULT_OOM; |
3389 | smp_wmb(); /* See comment in __pte_alloc() */ | |
3390 | } | |
3391 | ||
82b0f8c3 JK |
3392 | vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd); |
3393 | if (unlikely(!pmd_none(*vmf->pmd))) | |
10102459 KS |
3394 | goto out; |
3395 | ||
3396 | for (i = 0; i < HPAGE_PMD_NR; i++) | |
3397 | flush_icache_page(vma, page + i); | |
3398 | ||
3399 | entry = mk_huge_pmd(page, vma->vm_page_prot); | |
3400 | if (write) | |
f55e1014 | 3401 | entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma); |
10102459 | 3402 | |
fadae295 | 3403 | add_mm_counter(vma->vm_mm, mm_counter_file(page), HPAGE_PMD_NR); |
10102459 | 3404 | page_add_file_rmap(page, true); |
953c66c2 AK |
3405 | /* |
3406 | * deposit and withdraw with pmd lock held | |
3407 | */ | |
3408 | if (arch_needs_pgtable_deposit()) | |
82b0f8c3 | 3409 | deposit_prealloc_pte(vmf); |
10102459 | 3410 | |
82b0f8c3 | 3411 | set_pmd_at(vma->vm_mm, haddr, vmf->pmd, entry); |
10102459 | 3412 | |
82b0f8c3 | 3413 | update_mmu_cache_pmd(vma, haddr, vmf->pmd); |
10102459 KS |
3414 | |
3415 | /* fault is handled */ | |
3416 | ret = 0; | |
95ecedcd | 3417 | count_vm_event(THP_FILE_MAPPED); |
10102459 | 3418 | out: |
82b0f8c3 | 3419 | spin_unlock(vmf->ptl); |
10102459 KS |
3420 | return ret; |
3421 | } | |
3422 | #else | |
82b0f8c3 | 3423 | static int do_set_pmd(struct vm_fault *vmf, struct page *page) |
10102459 KS |
3424 | { |
3425 | BUILD_BUG(); | |
3426 | return 0; | |
3427 | } | |
3428 | #endif | |
3429 | ||
8c6e50b0 | 3430 | /** |
7267ec00 KS |
3431 | * alloc_set_pte - setup new PTE entry for given page and add reverse page |
3432 | * mapping. If needed, the fucntion allocates page table or use pre-allocated. | |
8c6e50b0 | 3433 | * |
82b0f8c3 | 3434 | * @vmf: fault environment |
7267ec00 | 3435 | * @memcg: memcg to charge page (only for private mappings) |
8c6e50b0 | 3436 | * @page: page to map |
8c6e50b0 | 3437 | * |
82b0f8c3 JK |
3438 | * Caller must take care of unlocking vmf->ptl, if vmf->pte is non-NULL on |
3439 | * return. | |
8c6e50b0 KS |
3440 | * |
3441 | * Target users are page handler itself and implementations of | |
3442 | * vm_ops->map_pages. | |
3443 | */ | |
82b0f8c3 | 3444 | int alloc_set_pte(struct vm_fault *vmf, struct mem_cgroup *memcg, |
7267ec00 | 3445 | struct page *page) |
3bb97794 | 3446 | { |
82b0f8c3 JK |
3447 | struct vm_area_struct *vma = vmf->vma; |
3448 | bool write = vmf->flags & FAULT_FLAG_WRITE; | |
3bb97794 | 3449 | pte_t entry; |
10102459 KS |
3450 | int ret; |
3451 | ||
82b0f8c3 | 3452 | if (pmd_none(*vmf->pmd) && PageTransCompound(page) && |
e496cf3d | 3453 | IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) { |
10102459 KS |
3454 | /* THP on COW? */ |
3455 | VM_BUG_ON_PAGE(memcg, page); | |
3456 | ||
82b0f8c3 | 3457 | ret = do_set_pmd(vmf, page); |
10102459 | 3458 | if (ret != VM_FAULT_FALLBACK) |
b0b9b3df | 3459 | return ret; |
10102459 | 3460 | } |
3bb97794 | 3461 | |
82b0f8c3 JK |
3462 | if (!vmf->pte) { |
3463 | ret = pte_alloc_one_map(vmf); | |
7267ec00 | 3464 | if (ret) |
b0b9b3df | 3465 | return ret; |
7267ec00 KS |
3466 | } |
3467 | ||
3468 | /* Re-check under ptl */ | |
b0b9b3df HD |
3469 | if (unlikely(!pte_none(*vmf->pte))) |
3470 | return VM_FAULT_NOPAGE; | |
7267ec00 | 3471 | |
3bb97794 KS |
3472 | flush_icache_page(vma, page); |
3473 | entry = mk_pte(page, vma->vm_page_prot); | |
3474 | if (write) | |
3475 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
bae473a4 KS |
3476 | /* copy-on-write page */ |
3477 | if (write && !(vma->vm_flags & VM_SHARED)) { | |
3bb97794 | 3478 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
82b0f8c3 | 3479 | page_add_new_anon_rmap(page, vma, vmf->address, false); |
7267ec00 KS |
3480 | mem_cgroup_commit_charge(page, memcg, false, false); |
3481 | lru_cache_add_active_or_unevictable(page, vma); | |
3bb97794 | 3482 | } else { |
eca56ff9 | 3483 | inc_mm_counter_fast(vma->vm_mm, mm_counter_file(page)); |
dd78fedd | 3484 | page_add_file_rmap(page, false); |
3bb97794 | 3485 | } |
82b0f8c3 | 3486 | set_pte_at(vma->vm_mm, vmf->address, vmf->pte, entry); |
3bb97794 KS |
3487 | |
3488 | /* no need to invalidate: a not-present page won't be cached */ | |
82b0f8c3 | 3489 | update_mmu_cache(vma, vmf->address, vmf->pte); |
7267ec00 | 3490 | |
b0b9b3df | 3491 | return 0; |
3bb97794 KS |
3492 | } |
3493 | ||
9118c0cb JK |
3494 | |
3495 | /** | |
3496 | * finish_fault - finish page fault once we have prepared the page to fault | |
3497 | * | |
3498 | * @vmf: structure describing the fault | |
3499 | * | |
3500 | * This function handles all that is needed to finish a page fault once the | |
3501 | * page to fault in is prepared. It handles locking of PTEs, inserts PTE for | |
3502 | * given page, adds reverse page mapping, handles memcg charges and LRU | |
3503 | * addition. The function returns 0 on success, VM_FAULT_ code in case of | |
3504 | * error. | |
3505 | * | |
3506 | * The function expects the page to be locked and on success it consumes a | |
3507 | * reference of a page being mapped (for the PTE which maps it). | |
3508 | */ | |
3509 | int finish_fault(struct vm_fault *vmf) | |
3510 | { | |
3511 | struct page *page; | |
6b31d595 | 3512 | int ret = 0; |
9118c0cb JK |
3513 | |
3514 | /* Did we COW the page? */ | |
3515 | if ((vmf->flags & FAULT_FLAG_WRITE) && | |
3516 | !(vmf->vma->vm_flags & VM_SHARED)) | |
3517 | page = vmf->cow_page; | |
3518 | else | |
3519 | page = vmf->page; | |
6b31d595 MH |
3520 | |
3521 | /* | |
3522 | * check even for read faults because we might have lost our CoWed | |
3523 | * page | |
3524 | */ | |
3525 | if (!(vmf->vma->vm_flags & VM_SHARED)) | |
3526 | ret = check_stable_address_space(vmf->vma->vm_mm); | |
3527 | if (!ret) | |
3528 | ret = alloc_set_pte(vmf, vmf->memcg, page); | |
9118c0cb JK |
3529 | if (vmf->pte) |
3530 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
3531 | return ret; | |
3532 | } | |
3533 | ||
3a91053a KS |
3534 | static unsigned long fault_around_bytes __read_mostly = |
3535 | rounddown_pow_of_two(65536); | |
a9b0f861 | 3536 | |
a9b0f861 KS |
3537 | #ifdef CONFIG_DEBUG_FS |
3538 | static int fault_around_bytes_get(void *data, u64 *val) | |
1592eef0 | 3539 | { |
a9b0f861 | 3540 | *val = fault_around_bytes; |
1592eef0 KS |
3541 | return 0; |
3542 | } | |
3543 | ||
b4903d6e | 3544 | /* |
da391d64 WK |
3545 | * fault_around_bytes must be rounded down to the nearest page order as it's |
3546 | * what do_fault_around() expects to see. | |
b4903d6e | 3547 | */ |
a9b0f861 | 3548 | static int fault_around_bytes_set(void *data, u64 val) |
1592eef0 | 3549 | { |
a9b0f861 | 3550 | if (val / PAGE_SIZE > PTRS_PER_PTE) |
1592eef0 | 3551 | return -EINVAL; |
b4903d6e AR |
3552 | if (val > PAGE_SIZE) |
3553 | fault_around_bytes = rounddown_pow_of_two(val); | |
3554 | else | |
3555 | fault_around_bytes = PAGE_SIZE; /* rounddown_pow_of_two(0) is undefined */ | |
1592eef0 KS |
3556 | return 0; |
3557 | } | |
0a1345f8 | 3558 | DEFINE_DEBUGFS_ATTRIBUTE(fault_around_bytes_fops, |
a9b0f861 | 3559 | fault_around_bytes_get, fault_around_bytes_set, "%llu\n"); |
1592eef0 KS |
3560 | |
3561 | static int __init fault_around_debugfs(void) | |
3562 | { | |
3563 | void *ret; | |
3564 | ||
0a1345f8 | 3565 | ret = debugfs_create_file_unsafe("fault_around_bytes", 0644, NULL, NULL, |
a9b0f861 | 3566 | &fault_around_bytes_fops); |
1592eef0 | 3567 | if (!ret) |
a9b0f861 | 3568 | pr_warn("Failed to create fault_around_bytes in debugfs"); |
1592eef0 KS |
3569 | return 0; |
3570 | } | |
3571 | late_initcall(fault_around_debugfs); | |
1592eef0 | 3572 | #endif |
8c6e50b0 | 3573 | |
1fdb412b KS |
3574 | /* |
3575 | * do_fault_around() tries to map few pages around the fault address. The hope | |
3576 | * is that the pages will be needed soon and this will lower the number of | |
3577 | * faults to handle. | |
3578 | * | |
3579 | * It uses vm_ops->map_pages() to map the pages, which skips the page if it's | |
3580 | * not ready to be mapped: not up-to-date, locked, etc. | |
3581 | * | |
3582 | * This function is called with the page table lock taken. In the split ptlock | |
3583 | * case the page table lock only protects only those entries which belong to | |
3584 | * the page table corresponding to the fault address. | |
3585 | * | |
3586 | * This function doesn't cross the VMA boundaries, in order to call map_pages() | |
3587 | * only once. | |
3588 | * | |
da391d64 WK |
3589 | * fault_around_bytes defines how many bytes we'll try to map. |
3590 | * do_fault_around() expects it to be set to a power of two less than or equal | |
3591 | * to PTRS_PER_PTE. | |
1fdb412b | 3592 | * |
da391d64 WK |
3593 | * The virtual address of the area that we map is naturally aligned to |
3594 | * fault_around_bytes rounded down to the machine page size | |
3595 | * (and therefore to page order). This way it's easier to guarantee | |
3596 | * that we don't cross page table boundaries. | |
1fdb412b | 3597 | */ |
0721ec8b | 3598 | static int do_fault_around(struct vm_fault *vmf) |
8c6e50b0 | 3599 | { |
82b0f8c3 | 3600 | unsigned long address = vmf->address, nr_pages, mask; |
0721ec8b | 3601 | pgoff_t start_pgoff = vmf->pgoff; |
bae473a4 | 3602 | pgoff_t end_pgoff; |
7267ec00 | 3603 | int off, ret = 0; |
8c6e50b0 | 3604 | |
4db0c3c2 | 3605 | nr_pages = READ_ONCE(fault_around_bytes) >> PAGE_SHIFT; |
aecd6f44 KS |
3606 | mask = ~(nr_pages * PAGE_SIZE - 1) & PAGE_MASK; |
3607 | ||
82b0f8c3 JK |
3608 | vmf->address = max(address & mask, vmf->vma->vm_start); |
3609 | off = ((address - vmf->address) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1); | |
bae473a4 | 3610 | start_pgoff -= off; |
8c6e50b0 KS |
3611 | |
3612 | /* | |
da391d64 WK |
3613 | * end_pgoff is either the end of the page table, the end of |
3614 | * the vma or nr_pages from start_pgoff, depending what is nearest. | |
8c6e50b0 | 3615 | */ |
bae473a4 | 3616 | end_pgoff = start_pgoff - |
82b0f8c3 | 3617 | ((vmf->address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)) + |
8c6e50b0 | 3618 | PTRS_PER_PTE - 1; |
82b0f8c3 | 3619 | end_pgoff = min3(end_pgoff, vma_pages(vmf->vma) + vmf->vma->vm_pgoff - 1, |
bae473a4 | 3620 | start_pgoff + nr_pages - 1); |
8c6e50b0 | 3621 | |
82b0f8c3 JK |
3622 | if (pmd_none(*vmf->pmd)) { |
3623 | vmf->prealloc_pte = pte_alloc_one(vmf->vma->vm_mm, | |
3624 | vmf->address); | |
3625 | if (!vmf->prealloc_pte) | |
c5f88bd2 | 3626 | goto out; |
7267ec00 | 3627 | smp_wmb(); /* See comment in __pte_alloc() */ |
8c6e50b0 KS |
3628 | } |
3629 | ||
82b0f8c3 | 3630 | vmf->vma->vm_ops->map_pages(vmf, start_pgoff, end_pgoff); |
7267ec00 | 3631 | |
7267ec00 | 3632 | /* Huge page is mapped? Page fault is solved */ |
82b0f8c3 | 3633 | if (pmd_trans_huge(*vmf->pmd)) { |
7267ec00 KS |
3634 | ret = VM_FAULT_NOPAGE; |
3635 | goto out; | |
3636 | } | |
3637 | ||
3638 | /* ->map_pages() haven't done anything useful. Cold page cache? */ | |
82b0f8c3 | 3639 | if (!vmf->pte) |
7267ec00 KS |
3640 | goto out; |
3641 | ||
3642 | /* check if the page fault is solved */ | |
82b0f8c3 JK |
3643 | vmf->pte -= (vmf->address >> PAGE_SHIFT) - (address >> PAGE_SHIFT); |
3644 | if (!pte_none(*vmf->pte)) | |
7267ec00 | 3645 | ret = VM_FAULT_NOPAGE; |
82b0f8c3 | 3646 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
bae473a4 | 3647 | out: |
82b0f8c3 JK |
3648 | vmf->address = address; |
3649 | vmf->pte = NULL; | |
7267ec00 | 3650 | return ret; |
8c6e50b0 KS |
3651 | } |
3652 | ||
0721ec8b | 3653 | static int do_read_fault(struct vm_fault *vmf) |
e655fb29 | 3654 | { |
82b0f8c3 | 3655 | struct vm_area_struct *vma = vmf->vma; |
8c6e50b0 KS |
3656 | int ret = 0; |
3657 | ||
3658 | /* | |
3659 | * Let's call ->map_pages() first and use ->fault() as fallback | |
3660 | * if page by the offset is not ready to be mapped (cold cache or | |
3661 | * something). | |
3662 | */ | |
9b4bdd2f | 3663 | if (vma->vm_ops->map_pages && fault_around_bytes >> PAGE_SHIFT > 1) { |
0721ec8b | 3664 | ret = do_fault_around(vmf); |
7267ec00 KS |
3665 | if (ret) |
3666 | return ret; | |
8c6e50b0 | 3667 | } |
e655fb29 | 3668 | |
936ca80d | 3669 | ret = __do_fault(vmf); |
e655fb29 KS |
3670 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
3671 | return ret; | |
3672 | ||
9118c0cb | 3673 | ret |= finish_fault(vmf); |
936ca80d | 3674 | unlock_page(vmf->page); |
7267ec00 | 3675 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
936ca80d | 3676 | put_page(vmf->page); |
e655fb29 KS |
3677 | return ret; |
3678 | } | |
3679 | ||
0721ec8b | 3680 | static int do_cow_fault(struct vm_fault *vmf) |
ec47c3b9 | 3681 | { |
82b0f8c3 | 3682 | struct vm_area_struct *vma = vmf->vma; |
ec47c3b9 KS |
3683 | int ret; |
3684 | ||
3685 | if (unlikely(anon_vma_prepare(vma))) | |
3686 | return VM_FAULT_OOM; | |
3687 | ||
936ca80d JK |
3688 | vmf->cow_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vmf->address); |
3689 | if (!vmf->cow_page) | |
ec47c3b9 KS |
3690 | return VM_FAULT_OOM; |
3691 | ||
2cf85583 | 3692 | if (mem_cgroup_try_charge_delay(vmf->cow_page, vma->vm_mm, GFP_KERNEL, |
3917048d | 3693 | &vmf->memcg, false)) { |
936ca80d | 3694 | put_page(vmf->cow_page); |
ec47c3b9 KS |
3695 | return VM_FAULT_OOM; |
3696 | } | |
3697 | ||
936ca80d | 3698 | ret = __do_fault(vmf); |
ec47c3b9 KS |
3699 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
3700 | goto uncharge_out; | |
3917048d JK |
3701 | if (ret & VM_FAULT_DONE_COW) |
3702 | return ret; | |
ec47c3b9 | 3703 | |
b1aa812b | 3704 | copy_user_highpage(vmf->cow_page, vmf->page, vmf->address, vma); |
936ca80d | 3705 | __SetPageUptodate(vmf->cow_page); |
ec47c3b9 | 3706 | |
9118c0cb | 3707 | ret |= finish_fault(vmf); |
b1aa812b JK |
3708 | unlock_page(vmf->page); |
3709 | put_page(vmf->page); | |
7267ec00 KS |
3710 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
3711 | goto uncharge_out; | |
ec47c3b9 KS |
3712 | return ret; |
3713 | uncharge_out: | |
3917048d | 3714 | mem_cgroup_cancel_charge(vmf->cow_page, vmf->memcg, false); |
936ca80d | 3715 | put_page(vmf->cow_page); |
ec47c3b9 KS |
3716 | return ret; |
3717 | } | |
3718 | ||
0721ec8b | 3719 | static int do_shared_fault(struct vm_fault *vmf) |
1da177e4 | 3720 | { |
82b0f8c3 | 3721 | struct vm_area_struct *vma = vmf->vma; |
f0c6d4d2 | 3722 | int ret, tmp; |
1d65f86d | 3723 | |
936ca80d | 3724 | ret = __do_fault(vmf); |
7eae74af | 3725 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
f0c6d4d2 | 3726 | return ret; |
1da177e4 LT |
3727 | |
3728 | /* | |
f0c6d4d2 KS |
3729 | * Check if the backing address space wants to know that the page is |
3730 | * about to become writable | |
1da177e4 | 3731 | */ |
fb09a464 | 3732 | if (vma->vm_ops->page_mkwrite) { |
936ca80d | 3733 | unlock_page(vmf->page); |
38b8cb7f | 3734 | tmp = do_page_mkwrite(vmf); |
fb09a464 KS |
3735 | if (unlikely(!tmp || |
3736 | (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { | |
936ca80d | 3737 | put_page(vmf->page); |
fb09a464 | 3738 | return tmp; |
4294621f | 3739 | } |
fb09a464 KS |
3740 | } |
3741 | ||
9118c0cb | 3742 | ret |= finish_fault(vmf); |
7267ec00 KS |
3743 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | |
3744 | VM_FAULT_RETRY))) { | |
936ca80d JK |
3745 | unlock_page(vmf->page); |
3746 | put_page(vmf->page); | |
f0c6d4d2 | 3747 | return ret; |
1da177e4 | 3748 | } |
b827e496 | 3749 | |
97ba0c2b | 3750 | fault_dirty_shared_page(vma, vmf->page); |
1d65f86d | 3751 | return ret; |
54cb8821 | 3752 | } |
d00806b1 | 3753 | |
9a95f3cf PC |
3754 | /* |
3755 | * We enter with non-exclusive mmap_sem (to exclude vma changes, | |
3756 | * but allow concurrent faults). | |
3757 | * The mmap_sem may have been released depending on flags and our | |
3758 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
3759 | */ | |
82b0f8c3 | 3760 | static int do_fault(struct vm_fault *vmf) |
54cb8821 | 3761 | { |
82b0f8c3 | 3762 | struct vm_area_struct *vma = vmf->vma; |
b0b9b3df | 3763 | int ret; |
54cb8821 | 3764 | |
6b7339f4 KS |
3765 | /* The VMA was not fully populated on mmap() or missing VM_DONTEXPAND */ |
3766 | if (!vma->vm_ops->fault) | |
b0b9b3df HD |
3767 | ret = VM_FAULT_SIGBUS; |
3768 | else if (!(vmf->flags & FAULT_FLAG_WRITE)) | |
3769 | ret = do_read_fault(vmf); | |
3770 | else if (!(vma->vm_flags & VM_SHARED)) | |
3771 | ret = do_cow_fault(vmf); | |
3772 | else | |
3773 | ret = do_shared_fault(vmf); | |
3774 | ||
3775 | /* preallocated pagetable is unused: free it */ | |
3776 | if (vmf->prealloc_pte) { | |
3777 | pte_free(vma->vm_mm, vmf->prealloc_pte); | |
7f2b6ce8 | 3778 | vmf->prealloc_pte = NULL; |
b0b9b3df HD |
3779 | } |
3780 | return ret; | |
54cb8821 NP |
3781 | } |
3782 | ||
b19a9939 | 3783 | static int numa_migrate_prep(struct page *page, struct vm_area_struct *vma, |
04bb2f94 RR |
3784 | unsigned long addr, int page_nid, |
3785 | int *flags) | |
9532fec1 MG |
3786 | { |
3787 | get_page(page); | |
3788 | ||
3789 | count_vm_numa_event(NUMA_HINT_FAULTS); | |
04bb2f94 | 3790 | if (page_nid == numa_node_id()) { |
9532fec1 | 3791 | count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL); |
04bb2f94 RR |
3792 | *flags |= TNF_FAULT_LOCAL; |
3793 | } | |
9532fec1 MG |
3794 | |
3795 | return mpol_misplaced(page, vma, addr); | |
3796 | } | |
3797 | ||
2994302b | 3798 | static int do_numa_page(struct vm_fault *vmf) |
d10e63f2 | 3799 | { |
82b0f8c3 | 3800 | struct vm_area_struct *vma = vmf->vma; |
4daae3b4 | 3801 | struct page *page = NULL; |
8191acbd | 3802 | int page_nid = -1; |
90572890 | 3803 | int last_cpupid; |
cbee9f88 | 3804 | int target_nid; |
b8593bfd | 3805 | bool migrated = false; |
cee216a6 | 3806 | pte_t pte; |
288bc549 | 3807 | bool was_writable = pte_savedwrite(vmf->orig_pte); |
6688cc05 | 3808 | int flags = 0; |
d10e63f2 MG |
3809 | |
3810 | /* | |
166f61b9 TH |
3811 | * The "pte" at this point cannot be used safely without |
3812 | * validation through pte_unmap_same(). It's of NUMA type but | |
3813 | * the pfn may be screwed if the read is non atomic. | |
166f61b9 | 3814 | */ |
82b0f8c3 JK |
3815 | vmf->ptl = pte_lockptr(vma->vm_mm, vmf->pmd); |
3816 | spin_lock(vmf->ptl); | |
cee216a6 | 3817 | if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) { |
82b0f8c3 | 3818 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
4daae3b4 MG |
3819 | goto out; |
3820 | } | |
3821 | ||
cee216a6 AK |
3822 | /* |
3823 | * Make it present again, Depending on how arch implementes non | |
3824 | * accessible ptes, some can allow access by kernel mode. | |
3825 | */ | |
3826 | pte = ptep_modify_prot_start(vma->vm_mm, vmf->address, vmf->pte); | |
4d942466 MG |
3827 | pte = pte_modify(pte, vma->vm_page_prot); |
3828 | pte = pte_mkyoung(pte); | |
b191f9b1 MG |
3829 | if (was_writable) |
3830 | pte = pte_mkwrite(pte); | |
cee216a6 | 3831 | ptep_modify_prot_commit(vma->vm_mm, vmf->address, vmf->pte, pte); |
82b0f8c3 | 3832 | update_mmu_cache(vma, vmf->address, vmf->pte); |
d10e63f2 | 3833 | |
82b0f8c3 | 3834 | page = vm_normal_page(vma, vmf->address, pte); |
d10e63f2 | 3835 | if (!page) { |
82b0f8c3 | 3836 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
d10e63f2 MG |
3837 | return 0; |
3838 | } | |
3839 | ||
e81c4802 KS |
3840 | /* TODO: handle PTE-mapped THP */ |
3841 | if (PageCompound(page)) { | |
82b0f8c3 | 3842 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
e81c4802 KS |
3843 | return 0; |
3844 | } | |
3845 | ||
6688cc05 | 3846 | /* |
bea66fbd MG |
3847 | * Avoid grouping on RO pages in general. RO pages shouldn't hurt as |
3848 | * much anyway since they can be in shared cache state. This misses | |
3849 | * the case where a mapping is writable but the process never writes | |
3850 | * to it but pte_write gets cleared during protection updates and | |
3851 | * pte_dirty has unpredictable behaviour between PTE scan updates, | |
3852 | * background writeback, dirty balancing and application behaviour. | |
6688cc05 | 3853 | */ |
d59dc7bc | 3854 | if (!pte_write(pte)) |
6688cc05 PZ |
3855 | flags |= TNF_NO_GROUP; |
3856 | ||
dabe1d99 RR |
3857 | /* |
3858 | * Flag if the page is shared between multiple address spaces. This | |
3859 | * is later used when determining whether to group tasks together | |
3860 | */ | |
3861 | if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED)) | |
3862 | flags |= TNF_SHARED; | |
3863 | ||
90572890 | 3864 | last_cpupid = page_cpupid_last(page); |
8191acbd | 3865 | page_nid = page_to_nid(page); |
82b0f8c3 | 3866 | target_nid = numa_migrate_prep(page, vma, vmf->address, page_nid, |
bae473a4 | 3867 | &flags); |
82b0f8c3 | 3868 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
4daae3b4 | 3869 | if (target_nid == -1) { |
4daae3b4 MG |
3870 | put_page(page); |
3871 | goto out; | |
3872 | } | |
3873 | ||
3874 | /* Migrate to the requested node */ | |
1bc115d8 | 3875 | migrated = migrate_misplaced_page(page, vma, target_nid); |
6688cc05 | 3876 | if (migrated) { |
8191acbd | 3877 | page_nid = target_nid; |
6688cc05 | 3878 | flags |= TNF_MIGRATED; |
074c2381 MG |
3879 | } else |
3880 | flags |= TNF_MIGRATE_FAIL; | |
4daae3b4 MG |
3881 | |
3882 | out: | |
8191acbd | 3883 | if (page_nid != -1) |
6688cc05 | 3884 | task_numa_fault(last_cpupid, page_nid, 1, flags); |
d10e63f2 MG |
3885 | return 0; |
3886 | } | |
3887 | ||
91a90140 | 3888 | static inline int create_huge_pmd(struct vm_fault *vmf) |
b96375f7 | 3889 | { |
f4200391 | 3890 | if (vma_is_anonymous(vmf->vma)) |
82b0f8c3 | 3891 | return do_huge_pmd_anonymous_page(vmf); |
a2d58167 | 3892 | if (vmf->vma->vm_ops->huge_fault) |
c791ace1 | 3893 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD); |
b96375f7 MW |
3894 | return VM_FAULT_FALLBACK; |
3895 | } | |
3896 | ||
183f24aa GU |
3897 | /* `inline' is required to avoid gcc 4.1.2 build error */ |
3898 | static inline int wp_huge_pmd(struct vm_fault *vmf, pmd_t orig_pmd) | |
b96375f7 | 3899 | { |
82b0f8c3 JK |
3900 | if (vma_is_anonymous(vmf->vma)) |
3901 | return do_huge_pmd_wp_page(vmf, orig_pmd); | |
a2d58167 | 3902 | if (vmf->vma->vm_ops->huge_fault) |
c791ace1 | 3903 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD); |
af9e4d5f KS |
3904 | |
3905 | /* COW handled on pte level: split pmd */ | |
82b0f8c3 JK |
3906 | VM_BUG_ON_VMA(vmf->vma->vm_flags & VM_SHARED, vmf->vma); |
3907 | __split_huge_pmd(vmf->vma, vmf->pmd, vmf->address, false, NULL); | |
af9e4d5f | 3908 | |
b96375f7 MW |
3909 | return VM_FAULT_FALLBACK; |
3910 | } | |
3911 | ||
38e08854 LS |
3912 | static inline bool vma_is_accessible(struct vm_area_struct *vma) |
3913 | { | |
3914 | return vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE); | |
3915 | } | |
3916 | ||
a00cc7d9 MW |
3917 | static int create_huge_pud(struct vm_fault *vmf) |
3918 | { | |
3919 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
3920 | /* No support for anonymous transparent PUD pages yet */ | |
3921 | if (vma_is_anonymous(vmf->vma)) | |
3922 | return VM_FAULT_FALLBACK; | |
3923 | if (vmf->vma->vm_ops->huge_fault) | |
c791ace1 | 3924 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD); |
a00cc7d9 MW |
3925 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
3926 | return VM_FAULT_FALLBACK; | |
3927 | } | |
3928 | ||
3929 | static int wp_huge_pud(struct vm_fault *vmf, pud_t orig_pud) | |
3930 | { | |
3931 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
3932 | /* No support for anonymous transparent PUD pages yet */ | |
3933 | if (vma_is_anonymous(vmf->vma)) | |
3934 | return VM_FAULT_FALLBACK; | |
3935 | if (vmf->vma->vm_ops->huge_fault) | |
c791ace1 | 3936 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD); |
a00cc7d9 MW |
3937 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
3938 | return VM_FAULT_FALLBACK; | |
3939 | } | |
3940 | ||
1da177e4 LT |
3941 | /* |
3942 | * These routines also need to handle stuff like marking pages dirty | |
3943 | * and/or accessed for architectures that don't do it in hardware (most | |
3944 | * RISC architectures). The early dirtying is also good on the i386. | |
3945 | * | |
3946 | * There is also a hook called "update_mmu_cache()" that architectures | |
3947 | * with external mmu caches can use to update those (ie the Sparc or | |
3948 | * PowerPC hashed page tables that act as extended TLBs). | |
3949 | * | |
7267ec00 KS |
3950 | * We enter with non-exclusive mmap_sem (to exclude vma changes, but allow |
3951 | * concurrent faults). | |
9a95f3cf | 3952 | * |
7267ec00 KS |
3953 | * The mmap_sem may have been released depending on flags and our return value. |
3954 | * See filemap_fault() and __lock_page_or_retry(). | |
1da177e4 | 3955 | */ |
82b0f8c3 | 3956 | static int handle_pte_fault(struct vm_fault *vmf) |
1da177e4 LT |
3957 | { |
3958 | pte_t entry; | |
3959 | ||
82b0f8c3 | 3960 | if (unlikely(pmd_none(*vmf->pmd))) { |
7267ec00 KS |
3961 | /* |
3962 | * Leave __pte_alloc() until later: because vm_ops->fault may | |
3963 | * want to allocate huge page, and if we expose page table | |
3964 | * for an instant, it will be difficult to retract from | |
3965 | * concurrent faults and from rmap lookups. | |
3966 | */ | |
82b0f8c3 | 3967 | vmf->pte = NULL; |
7267ec00 KS |
3968 | } else { |
3969 | /* See comment in pte_alloc_one_map() */ | |
d0f0931d | 3970 | if (pmd_devmap_trans_unstable(vmf->pmd)) |
7267ec00 KS |
3971 | return 0; |
3972 | /* | |
3973 | * A regular pmd is established and it can't morph into a huge | |
3974 | * pmd from under us anymore at this point because we hold the | |
3975 | * mmap_sem read mode and khugepaged takes it in write mode. | |
3976 | * So now it's safe to run pte_offset_map(). | |
3977 | */ | |
82b0f8c3 | 3978 | vmf->pte = pte_offset_map(vmf->pmd, vmf->address); |
2994302b | 3979 | vmf->orig_pte = *vmf->pte; |
7267ec00 KS |
3980 | |
3981 | /* | |
3982 | * some architectures can have larger ptes than wordsize, | |
3983 | * e.g.ppc44x-defconfig has CONFIG_PTE_64BIT=y and | |
b03a0fe0 PM |
3984 | * CONFIG_32BIT=y, so READ_ONCE cannot guarantee atomic |
3985 | * accesses. The code below just needs a consistent view | |
3986 | * for the ifs and we later double check anyway with the | |
7267ec00 KS |
3987 | * ptl lock held. So here a barrier will do. |
3988 | */ | |
3989 | barrier(); | |
2994302b | 3990 | if (pte_none(vmf->orig_pte)) { |
82b0f8c3 JK |
3991 | pte_unmap(vmf->pte); |
3992 | vmf->pte = NULL; | |
65500d23 | 3993 | } |
1da177e4 LT |
3994 | } |
3995 | ||
82b0f8c3 JK |
3996 | if (!vmf->pte) { |
3997 | if (vma_is_anonymous(vmf->vma)) | |
3998 | return do_anonymous_page(vmf); | |
7267ec00 | 3999 | else |
82b0f8c3 | 4000 | return do_fault(vmf); |
7267ec00 KS |
4001 | } |
4002 | ||
2994302b JK |
4003 | if (!pte_present(vmf->orig_pte)) |
4004 | return do_swap_page(vmf); | |
7267ec00 | 4005 | |
2994302b JK |
4006 | if (pte_protnone(vmf->orig_pte) && vma_is_accessible(vmf->vma)) |
4007 | return do_numa_page(vmf); | |
d10e63f2 | 4008 | |
82b0f8c3 JK |
4009 | vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd); |
4010 | spin_lock(vmf->ptl); | |
2994302b | 4011 | entry = vmf->orig_pte; |
82b0f8c3 | 4012 | if (unlikely(!pte_same(*vmf->pte, entry))) |
8f4e2101 | 4013 | goto unlock; |
82b0f8c3 | 4014 | if (vmf->flags & FAULT_FLAG_WRITE) { |
f6f37321 | 4015 | if (!pte_write(entry)) |
2994302b | 4016 | return do_wp_page(vmf); |
1da177e4 LT |
4017 | entry = pte_mkdirty(entry); |
4018 | } | |
4019 | entry = pte_mkyoung(entry); | |
82b0f8c3 JK |
4020 | if (ptep_set_access_flags(vmf->vma, vmf->address, vmf->pte, entry, |
4021 | vmf->flags & FAULT_FLAG_WRITE)) { | |
4022 | update_mmu_cache(vmf->vma, vmf->address, vmf->pte); | |
1a44e149 AA |
4023 | } else { |
4024 | /* | |
4025 | * This is needed only for protection faults but the arch code | |
4026 | * is not yet telling us if this is a protection fault or not. | |
4027 | * This still avoids useless tlb flushes for .text page faults | |
4028 | * with threads. | |
4029 | */ | |
82b0f8c3 JK |
4030 | if (vmf->flags & FAULT_FLAG_WRITE) |
4031 | flush_tlb_fix_spurious_fault(vmf->vma, vmf->address); | |
1a44e149 | 4032 | } |
8f4e2101 | 4033 | unlock: |
82b0f8c3 | 4034 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
83c54070 | 4035 | return 0; |
1da177e4 LT |
4036 | } |
4037 | ||
4038 | /* | |
4039 | * By the time we get here, we already hold the mm semaphore | |
9a95f3cf PC |
4040 | * |
4041 | * The mmap_sem may have been released depending on flags and our | |
4042 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
1da177e4 | 4043 | */ |
dcddffd4 KS |
4044 | static int __handle_mm_fault(struct vm_area_struct *vma, unsigned long address, |
4045 | unsigned int flags) | |
1da177e4 | 4046 | { |
82b0f8c3 | 4047 | struct vm_fault vmf = { |
bae473a4 | 4048 | .vma = vma, |
1a29d85e | 4049 | .address = address & PAGE_MASK, |
bae473a4 | 4050 | .flags = flags, |
0721ec8b | 4051 | .pgoff = linear_page_index(vma, address), |
667240e0 | 4052 | .gfp_mask = __get_fault_gfp_mask(vma), |
bae473a4 | 4053 | }; |
fde26bed | 4054 | unsigned int dirty = flags & FAULT_FLAG_WRITE; |
dcddffd4 | 4055 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 4056 | pgd_t *pgd; |
c2febafc | 4057 | p4d_t *p4d; |
a2d58167 | 4058 | int ret; |
1da177e4 | 4059 | |
1da177e4 | 4060 | pgd = pgd_offset(mm, address); |
c2febafc KS |
4061 | p4d = p4d_alloc(mm, pgd, address); |
4062 | if (!p4d) | |
4063 | return VM_FAULT_OOM; | |
a00cc7d9 | 4064 | |
c2febafc | 4065 | vmf.pud = pud_alloc(mm, p4d, address); |
a00cc7d9 | 4066 | if (!vmf.pud) |
c74df32c | 4067 | return VM_FAULT_OOM; |
a00cc7d9 | 4068 | if (pud_none(*vmf.pud) && transparent_hugepage_enabled(vma)) { |
a00cc7d9 MW |
4069 | ret = create_huge_pud(&vmf); |
4070 | if (!(ret & VM_FAULT_FALLBACK)) | |
4071 | return ret; | |
4072 | } else { | |
4073 | pud_t orig_pud = *vmf.pud; | |
4074 | ||
4075 | barrier(); | |
4076 | if (pud_trans_huge(orig_pud) || pud_devmap(orig_pud)) { | |
a00cc7d9 | 4077 | |
a00cc7d9 MW |
4078 | /* NUMA case for anonymous PUDs would go here */ |
4079 | ||
f6f37321 | 4080 | if (dirty && !pud_write(orig_pud)) { |
a00cc7d9 MW |
4081 | ret = wp_huge_pud(&vmf, orig_pud); |
4082 | if (!(ret & VM_FAULT_FALLBACK)) | |
4083 | return ret; | |
4084 | } else { | |
4085 | huge_pud_set_accessed(&vmf, orig_pud); | |
4086 | return 0; | |
4087 | } | |
4088 | } | |
4089 | } | |
4090 | ||
4091 | vmf.pmd = pmd_alloc(mm, vmf.pud, address); | |
82b0f8c3 | 4092 | if (!vmf.pmd) |
c74df32c | 4093 | return VM_FAULT_OOM; |
82b0f8c3 | 4094 | if (pmd_none(*vmf.pmd) && transparent_hugepage_enabled(vma)) { |
a2d58167 | 4095 | ret = create_huge_pmd(&vmf); |
c0292554 KS |
4096 | if (!(ret & VM_FAULT_FALLBACK)) |
4097 | return ret; | |
71e3aac0 | 4098 | } else { |
82b0f8c3 | 4099 | pmd_t orig_pmd = *vmf.pmd; |
1f1d06c3 | 4100 | |
71e3aac0 | 4101 | barrier(); |
84c3fc4e ZY |
4102 | if (unlikely(is_swap_pmd(orig_pmd))) { |
4103 | VM_BUG_ON(thp_migration_supported() && | |
4104 | !is_pmd_migration_entry(orig_pmd)); | |
4105 | if (is_pmd_migration_entry(orig_pmd)) | |
4106 | pmd_migration_entry_wait(mm, vmf.pmd); | |
4107 | return 0; | |
4108 | } | |
5c7fb56e | 4109 | if (pmd_trans_huge(orig_pmd) || pmd_devmap(orig_pmd)) { |
38e08854 | 4110 | if (pmd_protnone(orig_pmd) && vma_is_accessible(vma)) |
82b0f8c3 | 4111 | return do_huge_pmd_numa_page(&vmf, orig_pmd); |
d10e63f2 | 4112 | |
f6f37321 | 4113 | if (dirty && !pmd_write(orig_pmd)) { |
82b0f8c3 | 4114 | ret = wp_huge_pmd(&vmf, orig_pmd); |
9845cbbd KS |
4115 | if (!(ret & VM_FAULT_FALLBACK)) |
4116 | return ret; | |
a1dd450b | 4117 | } else { |
82b0f8c3 | 4118 | huge_pmd_set_accessed(&vmf, orig_pmd); |
9845cbbd | 4119 | return 0; |
1f1d06c3 | 4120 | } |
71e3aac0 AA |
4121 | } |
4122 | } | |
4123 | ||
82b0f8c3 | 4124 | return handle_pte_fault(&vmf); |
1da177e4 LT |
4125 | } |
4126 | ||
9a95f3cf PC |
4127 | /* |
4128 | * By the time we get here, we already hold the mm semaphore | |
4129 | * | |
4130 | * The mmap_sem may have been released depending on flags and our | |
4131 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
4132 | */ | |
dcddffd4 KS |
4133 | int handle_mm_fault(struct vm_area_struct *vma, unsigned long address, |
4134 | unsigned int flags) | |
519e5247 JW |
4135 | { |
4136 | int ret; | |
4137 | ||
4138 | __set_current_state(TASK_RUNNING); | |
4139 | ||
4140 | count_vm_event(PGFAULT); | |
2262185c | 4141 | count_memcg_event_mm(vma->vm_mm, PGFAULT); |
519e5247 JW |
4142 | |
4143 | /* do counter updates before entering really critical section. */ | |
4144 | check_sync_rss_stat(current); | |
4145 | ||
de0c799b LD |
4146 | if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE, |
4147 | flags & FAULT_FLAG_INSTRUCTION, | |
4148 | flags & FAULT_FLAG_REMOTE)) | |
4149 | return VM_FAULT_SIGSEGV; | |
4150 | ||
519e5247 JW |
4151 | /* |
4152 | * Enable the memcg OOM handling for faults triggered in user | |
4153 | * space. Kernel faults are handled more gracefully. | |
4154 | */ | |
4155 | if (flags & FAULT_FLAG_USER) | |
49426420 | 4156 | mem_cgroup_oom_enable(); |
519e5247 | 4157 | |
bae473a4 KS |
4158 | if (unlikely(is_vm_hugetlb_page(vma))) |
4159 | ret = hugetlb_fault(vma->vm_mm, vma, address, flags); | |
4160 | else | |
4161 | ret = __handle_mm_fault(vma, address, flags); | |
519e5247 | 4162 | |
49426420 JW |
4163 | if (flags & FAULT_FLAG_USER) { |
4164 | mem_cgroup_oom_disable(); | |
166f61b9 TH |
4165 | /* |
4166 | * The task may have entered a memcg OOM situation but | |
4167 | * if the allocation error was handled gracefully (no | |
4168 | * VM_FAULT_OOM), there is no need to kill anything. | |
4169 | * Just clean up the OOM state peacefully. | |
4170 | */ | |
4171 | if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM)) | |
4172 | mem_cgroup_oom_synchronize(false); | |
49426420 | 4173 | } |
3812c8c8 | 4174 | |
519e5247 JW |
4175 | return ret; |
4176 | } | |
e1d6d01a | 4177 | EXPORT_SYMBOL_GPL(handle_mm_fault); |
519e5247 | 4178 | |
90eceff1 KS |
4179 | #ifndef __PAGETABLE_P4D_FOLDED |
4180 | /* | |
4181 | * Allocate p4d page table. | |
4182 | * We've already handled the fast-path in-line. | |
4183 | */ | |
4184 | int __p4d_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) | |
4185 | { | |
4186 | p4d_t *new = p4d_alloc_one(mm, address); | |
4187 | if (!new) | |
4188 | return -ENOMEM; | |
4189 | ||
4190 | smp_wmb(); /* See comment in __pte_alloc */ | |
4191 | ||
4192 | spin_lock(&mm->page_table_lock); | |
4193 | if (pgd_present(*pgd)) /* Another has populated it */ | |
4194 | p4d_free(mm, new); | |
4195 | else | |
4196 | pgd_populate(mm, pgd, new); | |
4197 | spin_unlock(&mm->page_table_lock); | |
4198 | return 0; | |
4199 | } | |
4200 | #endif /* __PAGETABLE_P4D_FOLDED */ | |
4201 | ||
1da177e4 LT |
4202 | #ifndef __PAGETABLE_PUD_FOLDED |
4203 | /* | |
4204 | * Allocate page upper directory. | |
872fec16 | 4205 | * We've already handled the fast-path in-line. |
1da177e4 | 4206 | */ |
c2febafc | 4207 | int __pud_alloc(struct mm_struct *mm, p4d_t *p4d, unsigned long address) |
1da177e4 | 4208 | { |
c74df32c HD |
4209 | pud_t *new = pud_alloc_one(mm, address); |
4210 | if (!new) | |
1bb3630e | 4211 | return -ENOMEM; |
1da177e4 | 4212 | |
362a61ad NP |
4213 | smp_wmb(); /* See comment in __pte_alloc */ |
4214 | ||
872fec16 | 4215 | spin_lock(&mm->page_table_lock); |
c2febafc | 4216 | #ifndef __ARCH_HAS_5LEVEL_HACK |
b4e98d9a KS |
4217 | if (!p4d_present(*p4d)) { |
4218 | mm_inc_nr_puds(mm); | |
c2febafc | 4219 | p4d_populate(mm, p4d, new); |
b4e98d9a | 4220 | } else /* Another has populated it */ |
5e541973 | 4221 | pud_free(mm, new); |
b4e98d9a KS |
4222 | #else |
4223 | if (!pgd_present(*p4d)) { | |
4224 | mm_inc_nr_puds(mm); | |
c2febafc | 4225 | pgd_populate(mm, p4d, new); |
b4e98d9a KS |
4226 | } else /* Another has populated it */ |
4227 | pud_free(mm, new); | |
c2febafc | 4228 | #endif /* __ARCH_HAS_5LEVEL_HACK */ |
c74df32c | 4229 | spin_unlock(&mm->page_table_lock); |
1bb3630e | 4230 | return 0; |
1da177e4 LT |
4231 | } |
4232 | #endif /* __PAGETABLE_PUD_FOLDED */ | |
4233 | ||
4234 | #ifndef __PAGETABLE_PMD_FOLDED | |
4235 | /* | |
4236 | * Allocate page middle directory. | |
872fec16 | 4237 | * We've already handled the fast-path in-line. |
1da177e4 | 4238 | */ |
1bb3630e | 4239 | int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) |
1da177e4 | 4240 | { |
a00cc7d9 | 4241 | spinlock_t *ptl; |
c74df32c HD |
4242 | pmd_t *new = pmd_alloc_one(mm, address); |
4243 | if (!new) | |
1bb3630e | 4244 | return -ENOMEM; |
1da177e4 | 4245 | |
362a61ad NP |
4246 | smp_wmb(); /* See comment in __pte_alloc */ |
4247 | ||
a00cc7d9 | 4248 | ptl = pud_lock(mm, pud); |
1da177e4 | 4249 | #ifndef __ARCH_HAS_4LEVEL_HACK |
dc6c9a35 KS |
4250 | if (!pud_present(*pud)) { |
4251 | mm_inc_nr_pmds(mm); | |
1bb3630e | 4252 | pud_populate(mm, pud, new); |
dc6c9a35 | 4253 | } else /* Another has populated it */ |
5e541973 | 4254 | pmd_free(mm, new); |
dc6c9a35 KS |
4255 | #else |
4256 | if (!pgd_present(*pud)) { | |
4257 | mm_inc_nr_pmds(mm); | |
1bb3630e | 4258 | pgd_populate(mm, pud, new); |
dc6c9a35 KS |
4259 | } else /* Another has populated it */ |
4260 | pmd_free(mm, new); | |
1da177e4 | 4261 | #endif /* __ARCH_HAS_4LEVEL_HACK */ |
a00cc7d9 | 4262 | spin_unlock(ptl); |
1bb3630e | 4263 | return 0; |
e0f39591 | 4264 | } |
1da177e4 LT |
4265 | #endif /* __PAGETABLE_PMD_FOLDED */ |
4266 | ||
09796395 | 4267 | static int __follow_pte_pmd(struct mm_struct *mm, unsigned long address, |
a4d1a885 JG |
4268 | unsigned long *start, unsigned long *end, |
4269 | pte_t **ptepp, pmd_t **pmdpp, spinlock_t **ptlp) | |
f8ad0f49 JW |
4270 | { |
4271 | pgd_t *pgd; | |
c2febafc | 4272 | p4d_t *p4d; |
f8ad0f49 JW |
4273 | pud_t *pud; |
4274 | pmd_t *pmd; | |
4275 | pte_t *ptep; | |
4276 | ||
4277 | pgd = pgd_offset(mm, address); | |
4278 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
4279 | goto out; | |
4280 | ||
c2febafc KS |
4281 | p4d = p4d_offset(pgd, address); |
4282 | if (p4d_none(*p4d) || unlikely(p4d_bad(*p4d))) | |
4283 | goto out; | |
4284 | ||
4285 | pud = pud_offset(p4d, address); | |
f8ad0f49 JW |
4286 | if (pud_none(*pud) || unlikely(pud_bad(*pud))) |
4287 | goto out; | |
4288 | ||
4289 | pmd = pmd_offset(pud, address); | |
f66055ab | 4290 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
f8ad0f49 | 4291 | |
09796395 RZ |
4292 | if (pmd_huge(*pmd)) { |
4293 | if (!pmdpp) | |
4294 | goto out; | |
4295 | ||
a4d1a885 JG |
4296 | if (start && end) { |
4297 | *start = address & PMD_MASK; | |
4298 | *end = *start + PMD_SIZE; | |
4299 | mmu_notifier_invalidate_range_start(mm, *start, *end); | |
4300 | } | |
09796395 RZ |
4301 | *ptlp = pmd_lock(mm, pmd); |
4302 | if (pmd_huge(*pmd)) { | |
4303 | *pmdpp = pmd; | |
4304 | return 0; | |
4305 | } | |
4306 | spin_unlock(*ptlp); | |
a4d1a885 JG |
4307 | if (start && end) |
4308 | mmu_notifier_invalidate_range_end(mm, *start, *end); | |
09796395 RZ |
4309 | } |
4310 | ||
4311 | if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) | |
f8ad0f49 JW |
4312 | goto out; |
4313 | ||
a4d1a885 JG |
4314 | if (start && end) { |
4315 | *start = address & PAGE_MASK; | |
4316 | *end = *start + PAGE_SIZE; | |
4317 | mmu_notifier_invalidate_range_start(mm, *start, *end); | |
4318 | } | |
f8ad0f49 | 4319 | ptep = pte_offset_map_lock(mm, pmd, address, ptlp); |
f8ad0f49 JW |
4320 | if (!pte_present(*ptep)) |
4321 | goto unlock; | |
4322 | *ptepp = ptep; | |
4323 | return 0; | |
4324 | unlock: | |
4325 | pte_unmap_unlock(ptep, *ptlp); | |
a4d1a885 JG |
4326 | if (start && end) |
4327 | mmu_notifier_invalidate_range_end(mm, *start, *end); | |
f8ad0f49 JW |
4328 | out: |
4329 | return -EINVAL; | |
4330 | } | |
4331 | ||
f729c8c9 RZ |
4332 | static inline int follow_pte(struct mm_struct *mm, unsigned long address, |
4333 | pte_t **ptepp, spinlock_t **ptlp) | |
1b36ba81 NK |
4334 | { |
4335 | int res; | |
4336 | ||
4337 | /* (void) is needed to make gcc happy */ | |
4338 | (void) __cond_lock(*ptlp, | |
a4d1a885 JG |
4339 | !(res = __follow_pte_pmd(mm, address, NULL, NULL, |
4340 | ptepp, NULL, ptlp))); | |
09796395 RZ |
4341 | return res; |
4342 | } | |
4343 | ||
4344 | int follow_pte_pmd(struct mm_struct *mm, unsigned long address, | |
a4d1a885 | 4345 | unsigned long *start, unsigned long *end, |
09796395 RZ |
4346 | pte_t **ptepp, pmd_t **pmdpp, spinlock_t **ptlp) |
4347 | { | |
4348 | int res; | |
4349 | ||
4350 | /* (void) is needed to make gcc happy */ | |
4351 | (void) __cond_lock(*ptlp, | |
a4d1a885 JG |
4352 | !(res = __follow_pte_pmd(mm, address, start, end, |
4353 | ptepp, pmdpp, ptlp))); | |
1b36ba81 NK |
4354 | return res; |
4355 | } | |
09796395 | 4356 | EXPORT_SYMBOL(follow_pte_pmd); |
1b36ba81 | 4357 | |
3b6748e2 JW |
4358 | /** |
4359 | * follow_pfn - look up PFN at a user virtual address | |
4360 | * @vma: memory mapping | |
4361 | * @address: user virtual address | |
4362 | * @pfn: location to store found PFN | |
4363 | * | |
4364 | * Only IO mappings and raw PFN mappings are allowed. | |
4365 | * | |
4366 | * Returns zero and the pfn at @pfn on success, -ve otherwise. | |
4367 | */ | |
4368 | int follow_pfn(struct vm_area_struct *vma, unsigned long address, | |
4369 | unsigned long *pfn) | |
4370 | { | |
4371 | int ret = -EINVAL; | |
4372 | spinlock_t *ptl; | |
4373 | pte_t *ptep; | |
4374 | ||
4375 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) | |
4376 | return ret; | |
4377 | ||
4378 | ret = follow_pte(vma->vm_mm, address, &ptep, &ptl); | |
4379 | if (ret) | |
4380 | return ret; | |
4381 | *pfn = pte_pfn(*ptep); | |
4382 | pte_unmap_unlock(ptep, ptl); | |
4383 | return 0; | |
4384 | } | |
4385 | EXPORT_SYMBOL(follow_pfn); | |
4386 | ||
28b2ee20 | 4387 | #ifdef CONFIG_HAVE_IOREMAP_PROT |
d87fe660 | 4388 | int follow_phys(struct vm_area_struct *vma, |
4389 | unsigned long address, unsigned int flags, | |
4390 | unsigned long *prot, resource_size_t *phys) | |
28b2ee20 | 4391 | { |
03668a4d | 4392 | int ret = -EINVAL; |
28b2ee20 RR |
4393 | pte_t *ptep, pte; |
4394 | spinlock_t *ptl; | |
28b2ee20 | 4395 | |
d87fe660 | 4396 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) |
4397 | goto out; | |
28b2ee20 | 4398 | |
03668a4d | 4399 | if (follow_pte(vma->vm_mm, address, &ptep, &ptl)) |
d87fe660 | 4400 | goto out; |
28b2ee20 | 4401 | pte = *ptep; |
03668a4d | 4402 | |
f6f37321 | 4403 | if ((flags & FOLL_WRITE) && !pte_write(pte)) |
28b2ee20 | 4404 | goto unlock; |
28b2ee20 RR |
4405 | |
4406 | *prot = pgprot_val(pte_pgprot(pte)); | |
03668a4d | 4407 | *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; |
28b2ee20 | 4408 | |
03668a4d | 4409 | ret = 0; |
28b2ee20 RR |
4410 | unlock: |
4411 | pte_unmap_unlock(ptep, ptl); | |
4412 | out: | |
d87fe660 | 4413 | return ret; |
28b2ee20 RR |
4414 | } |
4415 | ||
4416 | int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, | |
4417 | void *buf, int len, int write) | |
4418 | { | |
4419 | resource_size_t phys_addr; | |
4420 | unsigned long prot = 0; | |
2bc7273b | 4421 | void __iomem *maddr; |
28b2ee20 RR |
4422 | int offset = addr & (PAGE_SIZE-1); |
4423 | ||
d87fe660 | 4424 | if (follow_phys(vma, addr, write, &prot, &phys_addr)) |
28b2ee20 RR |
4425 | return -EINVAL; |
4426 | ||
9cb12d7b | 4427 | maddr = ioremap_prot(phys_addr, PAGE_ALIGN(len + offset), prot); |
24eee1e4 | 4428 | if (!maddr) |
4429 | return -ENOMEM; | |
4430 | ||
28b2ee20 RR |
4431 | if (write) |
4432 | memcpy_toio(maddr + offset, buf, len); | |
4433 | else | |
4434 | memcpy_fromio(buf, maddr + offset, len); | |
4435 | iounmap(maddr); | |
4436 | ||
4437 | return len; | |
4438 | } | |
5a73633e | 4439 | EXPORT_SYMBOL_GPL(generic_access_phys); |
28b2ee20 RR |
4440 | #endif |
4441 | ||
0ec76a11 | 4442 | /* |
206cb636 SW |
4443 | * Access another process' address space as given in mm. If non-NULL, use the |
4444 | * given task for page fault accounting. | |
0ec76a11 | 4445 | */ |
84d77d3f | 4446 | int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, |
442486ec | 4447 | unsigned long addr, void *buf, int len, unsigned int gup_flags) |
0ec76a11 | 4448 | { |
0ec76a11 | 4449 | struct vm_area_struct *vma; |
0ec76a11 | 4450 | void *old_buf = buf; |
442486ec | 4451 | int write = gup_flags & FOLL_WRITE; |
0ec76a11 | 4452 | |
0ec76a11 | 4453 | down_read(&mm->mmap_sem); |
183ff22b | 4454 | /* ignore errors, just check how much was successfully transferred */ |
0ec76a11 DH |
4455 | while (len) { |
4456 | int bytes, ret, offset; | |
4457 | void *maddr; | |
28b2ee20 | 4458 | struct page *page = NULL; |
0ec76a11 | 4459 | |
1e987790 | 4460 | ret = get_user_pages_remote(tsk, mm, addr, 1, |
5b56d49f | 4461 | gup_flags, &page, &vma, NULL); |
28b2ee20 | 4462 | if (ret <= 0) { |
dbffcd03 RR |
4463 | #ifndef CONFIG_HAVE_IOREMAP_PROT |
4464 | break; | |
4465 | #else | |
28b2ee20 RR |
4466 | /* |
4467 | * Check if this is a VM_IO | VM_PFNMAP VMA, which | |
4468 | * we can access using slightly different code. | |
4469 | */ | |
28b2ee20 | 4470 | vma = find_vma(mm, addr); |
fe936dfc | 4471 | if (!vma || vma->vm_start > addr) |
28b2ee20 RR |
4472 | break; |
4473 | if (vma->vm_ops && vma->vm_ops->access) | |
4474 | ret = vma->vm_ops->access(vma, addr, buf, | |
4475 | len, write); | |
4476 | if (ret <= 0) | |
28b2ee20 RR |
4477 | break; |
4478 | bytes = ret; | |
dbffcd03 | 4479 | #endif |
0ec76a11 | 4480 | } else { |
28b2ee20 RR |
4481 | bytes = len; |
4482 | offset = addr & (PAGE_SIZE-1); | |
4483 | if (bytes > PAGE_SIZE-offset) | |
4484 | bytes = PAGE_SIZE-offset; | |
4485 | ||
4486 | maddr = kmap(page); | |
4487 | if (write) { | |
4488 | copy_to_user_page(vma, page, addr, | |
4489 | maddr + offset, buf, bytes); | |
4490 | set_page_dirty_lock(page); | |
4491 | } else { | |
4492 | copy_from_user_page(vma, page, addr, | |
4493 | buf, maddr + offset, bytes); | |
4494 | } | |
4495 | kunmap(page); | |
09cbfeaf | 4496 | put_page(page); |
0ec76a11 | 4497 | } |
0ec76a11 DH |
4498 | len -= bytes; |
4499 | buf += bytes; | |
4500 | addr += bytes; | |
4501 | } | |
4502 | up_read(&mm->mmap_sem); | |
0ec76a11 DH |
4503 | |
4504 | return buf - old_buf; | |
4505 | } | |
03252919 | 4506 | |
5ddd36b9 | 4507 | /** |
ae91dbfc | 4508 | * access_remote_vm - access another process' address space |
5ddd36b9 SW |
4509 | * @mm: the mm_struct of the target address space |
4510 | * @addr: start address to access | |
4511 | * @buf: source or destination buffer | |
4512 | * @len: number of bytes to transfer | |
6347e8d5 | 4513 | * @gup_flags: flags modifying lookup behaviour |
5ddd36b9 SW |
4514 | * |
4515 | * The caller must hold a reference on @mm. | |
4516 | */ | |
4517 | int access_remote_vm(struct mm_struct *mm, unsigned long addr, | |
6347e8d5 | 4518 | void *buf, int len, unsigned int gup_flags) |
5ddd36b9 | 4519 | { |
6347e8d5 | 4520 | return __access_remote_vm(NULL, mm, addr, buf, len, gup_flags); |
5ddd36b9 SW |
4521 | } |
4522 | ||
206cb636 SW |
4523 | /* |
4524 | * Access another process' address space. | |
4525 | * Source/target buffer must be kernel space, | |
4526 | * Do not walk the page table directly, use get_user_pages | |
4527 | */ | |
4528 | int access_process_vm(struct task_struct *tsk, unsigned long addr, | |
f307ab6d | 4529 | void *buf, int len, unsigned int gup_flags) |
206cb636 SW |
4530 | { |
4531 | struct mm_struct *mm; | |
4532 | int ret; | |
4533 | ||
4534 | mm = get_task_mm(tsk); | |
4535 | if (!mm) | |
4536 | return 0; | |
4537 | ||
f307ab6d | 4538 | ret = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags); |
442486ec | 4539 | |
206cb636 SW |
4540 | mmput(mm); |
4541 | ||
4542 | return ret; | |
4543 | } | |
fcd35857 | 4544 | EXPORT_SYMBOL_GPL(access_process_vm); |
206cb636 | 4545 | |
03252919 AK |
4546 | /* |
4547 | * Print the name of a VMA. | |
4548 | */ | |
4549 | void print_vma_addr(char *prefix, unsigned long ip) | |
4550 | { | |
4551 | struct mm_struct *mm = current->mm; | |
4552 | struct vm_area_struct *vma; | |
4553 | ||
e8bff74a | 4554 | /* |
0a7f682d | 4555 | * we might be running from an atomic context so we cannot sleep |
e8bff74a | 4556 | */ |
0a7f682d | 4557 | if (!down_read_trylock(&mm->mmap_sem)) |
e8bff74a IM |
4558 | return; |
4559 | ||
03252919 AK |
4560 | vma = find_vma(mm, ip); |
4561 | if (vma && vma->vm_file) { | |
4562 | struct file *f = vma->vm_file; | |
0a7f682d | 4563 | char *buf = (char *)__get_free_page(GFP_NOWAIT); |
03252919 | 4564 | if (buf) { |
2fbc57c5 | 4565 | char *p; |
03252919 | 4566 | |
9bf39ab2 | 4567 | p = file_path(f, buf, PAGE_SIZE); |
03252919 AK |
4568 | if (IS_ERR(p)) |
4569 | p = "?"; | |
2fbc57c5 | 4570 | printk("%s%s[%lx+%lx]", prefix, kbasename(p), |
03252919 AK |
4571 | vma->vm_start, |
4572 | vma->vm_end - vma->vm_start); | |
4573 | free_page((unsigned long)buf); | |
4574 | } | |
4575 | } | |
51a07e50 | 4576 | up_read(&mm->mmap_sem); |
03252919 | 4577 | } |
3ee1afa3 | 4578 | |
662bbcb2 | 4579 | #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP) |
9ec23531 | 4580 | void __might_fault(const char *file, int line) |
3ee1afa3 | 4581 | { |
95156f00 PZ |
4582 | /* |
4583 | * Some code (nfs/sunrpc) uses socket ops on kernel memory while | |
4584 | * holding the mmap_sem, this is safe because kernel memory doesn't | |
4585 | * get paged out, therefore we'll never actually fault, and the | |
4586 | * below annotations will generate false positives. | |
4587 | */ | |
db68ce10 | 4588 | if (uaccess_kernel()) |
95156f00 | 4589 | return; |
9ec23531 | 4590 | if (pagefault_disabled()) |
662bbcb2 | 4591 | return; |
9ec23531 DH |
4592 | __might_sleep(file, line, 0); |
4593 | #if defined(CONFIG_DEBUG_ATOMIC_SLEEP) | |
662bbcb2 | 4594 | if (current->mm) |
3ee1afa3 | 4595 | might_lock_read(¤t->mm->mmap_sem); |
9ec23531 | 4596 | #endif |
3ee1afa3 | 4597 | } |
9ec23531 | 4598 | EXPORT_SYMBOL(__might_fault); |
3ee1afa3 | 4599 | #endif |
47ad8475 AA |
4600 | |
4601 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) | |
4602 | static void clear_gigantic_page(struct page *page, | |
4603 | unsigned long addr, | |
4604 | unsigned int pages_per_huge_page) | |
4605 | { | |
4606 | int i; | |
4607 | struct page *p = page; | |
4608 | ||
4609 | might_sleep(); | |
4610 | for (i = 0; i < pages_per_huge_page; | |
4611 | i++, p = mem_map_next(p, page, i)) { | |
4612 | cond_resched(); | |
4613 | clear_user_highpage(p, addr + i * PAGE_SIZE); | |
4614 | } | |
4615 | } | |
4616 | void clear_huge_page(struct page *page, | |
c79b57e4 | 4617 | unsigned long addr_hint, unsigned int pages_per_huge_page) |
47ad8475 | 4618 | { |
c79b57e4 HY |
4619 | int i, n, base, l; |
4620 | unsigned long addr = addr_hint & | |
4621 | ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1); | |
47ad8475 AA |
4622 | |
4623 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { | |
4624 | clear_gigantic_page(page, addr, pages_per_huge_page); | |
4625 | return; | |
4626 | } | |
4627 | ||
c79b57e4 | 4628 | /* Clear sub-page to access last to keep its cache lines hot */ |
47ad8475 | 4629 | might_sleep(); |
c79b57e4 HY |
4630 | n = (addr_hint - addr) / PAGE_SIZE; |
4631 | if (2 * n <= pages_per_huge_page) { | |
4632 | /* If sub-page to access in first half of huge page */ | |
4633 | base = 0; | |
4634 | l = n; | |
4635 | /* Clear sub-pages at the end of huge page */ | |
4636 | for (i = pages_per_huge_page - 1; i >= 2 * n; i--) { | |
4637 | cond_resched(); | |
4638 | clear_user_highpage(page + i, addr + i * PAGE_SIZE); | |
4639 | } | |
4640 | } else { | |
4641 | /* If sub-page to access in second half of huge page */ | |
4642 | base = pages_per_huge_page - 2 * (pages_per_huge_page - n); | |
4643 | l = pages_per_huge_page - n; | |
4644 | /* Clear sub-pages at the begin of huge page */ | |
4645 | for (i = 0; i < base; i++) { | |
4646 | cond_resched(); | |
4647 | clear_user_highpage(page + i, addr + i * PAGE_SIZE); | |
4648 | } | |
4649 | } | |
4650 | /* | |
4651 | * Clear remaining sub-pages in left-right-left-right pattern | |
4652 | * towards the sub-page to access | |
4653 | */ | |
4654 | for (i = 0; i < l; i++) { | |
4655 | int left_idx = base + i; | |
4656 | int right_idx = base + 2 * l - 1 - i; | |
4657 | ||
4658 | cond_resched(); | |
4659 | clear_user_highpage(page + left_idx, | |
4660 | addr + left_idx * PAGE_SIZE); | |
47ad8475 | 4661 | cond_resched(); |
c79b57e4 HY |
4662 | clear_user_highpage(page + right_idx, |
4663 | addr + right_idx * PAGE_SIZE); | |
47ad8475 AA |
4664 | } |
4665 | } | |
4666 | ||
4667 | static void copy_user_gigantic_page(struct page *dst, struct page *src, | |
4668 | unsigned long addr, | |
4669 | struct vm_area_struct *vma, | |
4670 | unsigned int pages_per_huge_page) | |
4671 | { | |
4672 | int i; | |
4673 | struct page *dst_base = dst; | |
4674 | struct page *src_base = src; | |
4675 | ||
4676 | for (i = 0; i < pages_per_huge_page; ) { | |
4677 | cond_resched(); | |
4678 | copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma); | |
4679 | ||
4680 | i++; | |
4681 | dst = mem_map_next(dst, dst_base, i); | |
4682 | src = mem_map_next(src, src_base, i); | |
4683 | } | |
4684 | } | |
4685 | ||
4686 | void copy_user_huge_page(struct page *dst, struct page *src, | |
4687 | unsigned long addr, struct vm_area_struct *vma, | |
4688 | unsigned int pages_per_huge_page) | |
4689 | { | |
4690 | int i; | |
4691 | ||
4692 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { | |
4693 | copy_user_gigantic_page(dst, src, addr, vma, | |
4694 | pages_per_huge_page); | |
4695 | return; | |
4696 | } | |
4697 | ||
4698 | might_sleep(); | |
4699 | for (i = 0; i < pages_per_huge_page; i++) { | |
4700 | cond_resched(); | |
4701 | copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE, vma); | |
4702 | } | |
4703 | } | |
fa4d75c1 MK |
4704 | |
4705 | long copy_huge_page_from_user(struct page *dst_page, | |
4706 | const void __user *usr_src, | |
810a56b9 MK |
4707 | unsigned int pages_per_huge_page, |
4708 | bool allow_pagefault) | |
fa4d75c1 MK |
4709 | { |
4710 | void *src = (void *)usr_src; | |
4711 | void *page_kaddr; | |
4712 | unsigned long i, rc = 0; | |
4713 | unsigned long ret_val = pages_per_huge_page * PAGE_SIZE; | |
4714 | ||
4715 | for (i = 0; i < pages_per_huge_page; i++) { | |
810a56b9 MK |
4716 | if (allow_pagefault) |
4717 | page_kaddr = kmap(dst_page + i); | |
4718 | else | |
4719 | page_kaddr = kmap_atomic(dst_page + i); | |
fa4d75c1 MK |
4720 | rc = copy_from_user(page_kaddr, |
4721 | (const void __user *)(src + i * PAGE_SIZE), | |
4722 | PAGE_SIZE); | |
810a56b9 MK |
4723 | if (allow_pagefault) |
4724 | kunmap(dst_page + i); | |
4725 | else | |
4726 | kunmap_atomic(page_kaddr); | |
fa4d75c1 MK |
4727 | |
4728 | ret_val -= (PAGE_SIZE - rc); | |
4729 | if (rc) | |
4730 | break; | |
4731 | ||
4732 | cond_resched(); | |
4733 | } | |
4734 | return ret_val; | |
4735 | } | |
47ad8475 | 4736 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */ |
49076ec2 | 4737 | |
40b64acd | 4738 | #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS |
b35f1819 KS |
4739 | |
4740 | static struct kmem_cache *page_ptl_cachep; | |
4741 | ||
4742 | void __init ptlock_cache_init(void) | |
4743 | { | |
4744 | page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0, | |
4745 | SLAB_PANIC, NULL); | |
4746 | } | |
4747 | ||
539edb58 | 4748 | bool ptlock_alloc(struct page *page) |
49076ec2 KS |
4749 | { |
4750 | spinlock_t *ptl; | |
4751 | ||
b35f1819 | 4752 | ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL); |
49076ec2 KS |
4753 | if (!ptl) |
4754 | return false; | |
539edb58 | 4755 | page->ptl = ptl; |
49076ec2 KS |
4756 | return true; |
4757 | } | |
4758 | ||
539edb58 | 4759 | void ptlock_free(struct page *page) |
49076ec2 | 4760 | { |
b35f1819 | 4761 | kmem_cache_free(page_ptl_cachep, page->ptl); |
49076ec2 KS |
4762 | } |
4763 | #endif |