ia64/xen-unstable

view linux-2.6-xen-sparse/mm/memory.c @ 9167:cb5abeaabd1a

[IA64] fix print out in ia64 setup_guest()

Signed-off-by: Isaku Yamahata <yamahata@valinux.co.jp>
author awilliam@xenbuild.aw
date Fri Mar 10 09:19:54 2006 -0700 (2006-03-10)
parents 1c7145a5bb43
children b64ac7e90ac6
line source
1 /*
2 * linux/mm/memory.c
3 *
4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 */
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 */
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 */
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 */
31 /*
32 * 05.04.94 - Multi-page memory management added for v1.1.
33 * Idea by Alex Bligh (alex@cconcepts.co.uk)
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 */
41 #include <linux/kernel_stat.h>
42 #include <linux/mm.h>
43 #include <linux/hugetlb.h>
44 #include <linux/mman.h>
45 #include <linux/swap.h>
46 #include <linux/highmem.h>
47 #include <linux/pagemap.h>
48 #include <linux/rmap.h>
49 #include <linux/module.h>
50 #include <linux/init.h>
52 #include <asm/pgalloc.h>
53 #include <asm/uaccess.h>
54 #include <asm/tlb.h>
55 #include <asm/tlbflush.h>
56 #include <asm/pgtable.h>
58 #include <linux/swapops.h>
59 #include <linux/elf.h>
61 #ifndef CONFIG_NEED_MULTIPLE_NODES
62 /* use the per-pgdat data instead for discontigmem - mbligh */
63 unsigned long max_mapnr;
64 struct page *mem_map;
66 EXPORT_SYMBOL(max_mapnr);
67 EXPORT_SYMBOL(mem_map);
68 #endif
70 unsigned long num_physpages;
71 /*
72 * A number of key systems in x86 including ioremap() rely on the assumption
73 * that high_memory defines the upper bound on direct map memory, then end
74 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
75 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
76 * and ZONE_HIGHMEM.
77 */
78 void * high_memory;
79 unsigned long vmalloc_earlyreserve;
81 EXPORT_SYMBOL(num_physpages);
82 EXPORT_SYMBOL(high_memory);
83 EXPORT_SYMBOL(vmalloc_earlyreserve);
85 int randomize_va_space __read_mostly = 1;
87 static int __init disable_randmaps(char *s)
88 {
89 randomize_va_space = 0;
90 return 0;
91 }
92 __setup("norandmaps", disable_randmaps);
95 /*
96 * If a p?d_bad entry is found while walking page tables, report
97 * the error, before resetting entry to p?d_none. Usually (but
98 * very seldom) called out from the p?d_none_or_clear_bad macros.
99 */
101 void pgd_clear_bad(pgd_t *pgd)
102 {
103 pgd_ERROR(*pgd);
104 pgd_clear(pgd);
105 }
107 void pud_clear_bad(pud_t *pud)
108 {
109 pud_ERROR(*pud);
110 pud_clear(pud);
111 }
113 void pmd_clear_bad(pmd_t *pmd)
114 {
115 pmd_ERROR(*pmd);
116 pmd_clear(pmd);
117 }
119 /*
120 * Note: this doesn't free the actual pages themselves. That
121 * has been handled earlier when unmapping all the memory regions.
122 */
123 static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd)
124 {
125 struct page *page = pmd_page(*pmd);
126 pmd_clear(pmd);
127 pte_lock_deinit(page);
128 pte_free_tlb(tlb, page);
129 dec_page_state(nr_page_table_pages);
130 tlb->mm->nr_ptes--;
131 }
133 static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
134 unsigned long addr, unsigned long end,
135 unsigned long floor, unsigned long ceiling)
136 {
137 pmd_t *pmd;
138 unsigned long next;
139 unsigned long start;
141 start = addr;
142 pmd = pmd_offset(pud, addr);
143 do {
144 next = pmd_addr_end(addr, end);
145 if (pmd_none_or_clear_bad(pmd))
146 continue;
147 free_pte_range(tlb, pmd);
148 } while (pmd++, addr = next, addr != end);
150 start &= PUD_MASK;
151 if (start < floor)
152 return;
153 if (ceiling) {
154 ceiling &= PUD_MASK;
155 if (!ceiling)
156 return;
157 }
158 if (end - 1 > ceiling - 1)
159 return;
161 pmd = pmd_offset(pud, start);
162 pud_clear(pud);
163 pmd_free_tlb(tlb, pmd);
164 }
166 static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
167 unsigned long addr, unsigned long end,
168 unsigned long floor, unsigned long ceiling)
169 {
170 pud_t *pud;
171 unsigned long next;
172 unsigned long start;
174 start = addr;
175 pud = pud_offset(pgd, addr);
176 do {
177 next = pud_addr_end(addr, end);
178 if (pud_none_or_clear_bad(pud))
179 continue;
180 free_pmd_range(tlb, pud, addr, next, floor, ceiling);
181 } while (pud++, addr = next, addr != end);
183 start &= PGDIR_MASK;
184 if (start < floor)
185 return;
186 if (ceiling) {
187 ceiling &= PGDIR_MASK;
188 if (!ceiling)
189 return;
190 }
191 if (end - 1 > ceiling - 1)
192 return;
194 pud = pud_offset(pgd, start);
195 pgd_clear(pgd);
196 pud_free_tlb(tlb, pud);
197 }
199 /*
200 * This function frees user-level page tables of a process.
201 *
202 * Must be called with pagetable lock held.
203 */
204 void free_pgd_range(struct mmu_gather **tlb,
205 unsigned long addr, unsigned long end,
206 unsigned long floor, unsigned long ceiling)
207 {
208 pgd_t *pgd;
209 unsigned long next;
210 unsigned long start;
212 /*
213 * The next few lines have given us lots of grief...
214 *
215 * Why are we testing PMD* at this top level? Because often
216 * there will be no work to do at all, and we'd prefer not to
217 * go all the way down to the bottom just to discover that.
218 *
219 * Why all these "- 1"s? Because 0 represents both the bottom
220 * of the address space and the top of it (using -1 for the
221 * top wouldn't help much: the masks would do the wrong thing).
222 * The rule is that addr 0 and floor 0 refer to the bottom of
223 * the address space, but end 0 and ceiling 0 refer to the top
224 * Comparisons need to use "end - 1" and "ceiling - 1" (though
225 * that end 0 case should be mythical).
226 *
227 * Wherever addr is brought up or ceiling brought down, we must
228 * be careful to reject "the opposite 0" before it confuses the
229 * subsequent tests. But what about where end is brought down
230 * by PMD_SIZE below? no, end can't go down to 0 there.
231 *
232 * Whereas we round start (addr) and ceiling down, by different
233 * masks at different levels, in order to test whether a table
234 * now has no other vmas using it, so can be freed, we don't
235 * bother to round floor or end up - the tests don't need that.
236 */
238 addr &= PMD_MASK;
239 if (addr < floor) {
240 addr += PMD_SIZE;
241 if (!addr)
242 return;
243 }
244 if (ceiling) {
245 ceiling &= PMD_MASK;
246 if (!ceiling)
247 return;
248 }
249 if (end - 1 > ceiling - 1)
250 end -= PMD_SIZE;
251 if (addr > end - 1)
252 return;
254 start = addr;
255 pgd = pgd_offset((*tlb)->mm, addr);
256 do {
257 next = pgd_addr_end(addr, end);
258 if (pgd_none_or_clear_bad(pgd))
259 continue;
260 free_pud_range(*tlb, pgd, addr, next, floor, ceiling);
261 } while (pgd++, addr = next, addr != end);
263 if (!(*tlb)->fullmm)
264 flush_tlb_pgtables((*tlb)->mm, start, end);
265 }
267 void free_pgtables(struct mmu_gather **tlb, struct vm_area_struct *vma,
268 unsigned long floor, unsigned long ceiling)
269 {
270 while (vma) {
271 struct vm_area_struct *next = vma->vm_next;
272 unsigned long addr = vma->vm_start;
274 /*
275 * Hide vma from rmap and vmtruncate before freeing pgtables
276 */
277 anon_vma_unlink(vma);
278 unlink_file_vma(vma);
280 if (is_hugepage_only_range(vma->vm_mm, addr, HPAGE_SIZE)) {
281 hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
282 floor, next? next->vm_start: ceiling);
283 } else {
284 /*
285 * Optimization: gather nearby vmas into one call down
286 */
287 while (next && next->vm_start <= vma->vm_end + PMD_SIZE
288 && !is_hugepage_only_range(vma->vm_mm, next->vm_start,
289 HPAGE_SIZE)) {
290 vma = next;
291 next = vma->vm_next;
292 anon_vma_unlink(vma);
293 unlink_file_vma(vma);
294 }
295 free_pgd_range(tlb, addr, vma->vm_end,
296 floor, next? next->vm_start: ceiling);
297 }
298 vma = next;
299 }
300 }
302 int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
303 {
304 struct page *new = pte_alloc_one(mm, address);
305 if (!new)
306 return -ENOMEM;
308 pte_lock_init(new);
309 spin_lock(&mm->page_table_lock);
310 if (pmd_present(*pmd)) { /* Another has populated it */
311 pte_lock_deinit(new);
312 pte_free(new);
313 } else {
314 mm->nr_ptes++;
315 inc_page_state(nr_page_table_pages);
316 pmd_populate(mm, pmd, new);
317 }
318 spin_unlock(&mm->page_table_lock);
319 return 0;
320 }
322 int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
323 {
324 pte_t *new = pte_alloc_one_kernel(&init_mm, address);
325 if (!new)
326 return -ENOMEM;
328 spin_lock(&init_mm.page_table_lock);
329 if (pmd_present(*pmd)) /* Another has populated it */
330 pte_free_kernel(new);
331 else
332 pmd_populate_kernel(&init_mm, pmd, new);
333 spin_unlock(&init_mm.page_table_lock);
334 return 0;
335 }
337 static inline void add_mm_rss(struct mm_struct *mm, int file_rss, int anon_rss)
338 {
339 if (file_rss)
340 add_mm_counter(mm, file_rss, file_rss);
341 if (anon_rss)
342 add_mm_counter(mm, anon_rss, anon_rss);
343 }
345 /*
346 * This function is called to print an error when a bad pte
347 * is found. For example, we might have a PFN-mapped pte in
348 * a region that doesn't allow it.
349 *
350 * The calling function must still handle the error.
351 */
352 void print_bad_pte(struct vm_area_struct *vma, pte_t pte, unsigned long vaddr)
353 {
354 printk(KERN_ERR "Bad pte = %08llx, process = %s, "
355 "vm_flags = %lx, vaddr = %lx\n",
356 (long long)pte_val(pte),
357 (vma->vm_mm == current->mm ? current->comm : "???"),
358 vma->vm_flags, vaddr);
359 dump_stack();
360 }
362 static inline int is_cow_mapping(unsigned int flags)
363 {
364 return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
365 }
367 /*
368 * This function gets the "struct page" associated with a pte.
369 *
370 * NOTE! Some mappings do not have "struct pages". A raw PFN mapping
371 * will have each page table entry just pointing to a raw page frame
372 * number, and as far as the VM layer is concerned, those do not have
373 * pages associated with them - even if the PFN might point to memory
374 * that otherwise is perfectly fine and has a "struct page".
375 *
376 * The way we recognize those mappings is through the rules set up
377 * by "remap_pfn_range()": the vma will have the VM_PFNMAP bit set,
378 * and the vm_pgoff will point to the first PFN mapped: thus every
379 * page that is a raw mapping will always honor the rule
380 *
381 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
382 *
383 * and if that isn't true, the page has been COW'ed (in which case it
384 * _does_ have a "struct page" associated with it even if it is in a
385 * VM_PFNMAP range).
386 */
387 struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
388 {
389 unsigned long pfn = pte_pfn(pte);
391 if (vma->vm_flags & VM_PFNMAP) {
392 unsigned long off = (addr - vma->vm_start) >> PAGE_SHIFT;
393 if (pfn == vma->vm_pgoff + off)
394 return NULL;
395 if (!is_cow_mapping(vma->vm_flags))
396 return NULL;
397 }
399 /*
400 * Add some anal sanity checks for now. Eventually,
401 * we should just do "return pfn_to_page(pfn)", but
402 * in the meantime we check that we get a valid pfn,
403 * and that the resulting page looks ok.
404 *
405 * Remove this test eventually!
406 */
407 if (unlikely(!pfn_valid(pfn))) {
408 if (!vma->vm_flags & VM_RESERVED)
409 print_bad_pte(vma, pte, addr);
410 return NULL;
411 }
413 /*
414 * NOTE! We still have PageReserved() pages in the page
415 * tables.
416 *
417 * The PAGE_ZERO() pages and various VDSO mappings can
418 * cause them to exist.
419 */
420 return pfn_to_page(pfn);
421 }
423 /*
424 * copy one vm_area from one task to the other. Assumes the page tables
425 * already present in the new task to be cleared in the whole range
426 * covered by this vma.
427 */
429 static inline void
430 copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
431 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
432 unsigned long addr, int *rss)
433 {
434 unsigned long vm_flags = vma->vm_flags;
435 pte_t pte = *src_pte;
436 struct page *page;
438 /* pte contains position in swap or file, so copy. */
439 if (unlikely(!pte_present(pte))) {
440 if (!pte_file(pte)) {
441 swap_duplicate(pte_to_swp_entry(pte));
442 /* make sure dst_mm is on swapoff's mmlist. */
443 if (unlikely(list_empty(&dst_mm->mmlist))) {
444 spin_lock(&mmlist_lock);
445 if (list_empty(&dst_mm->mmlist))
446 list_add(&dst_mm->mmlist,
447 &src_mm->mmlist);
448 spin_unlock(&mmlist_lock);
449 }
450 }
451 goto out_set_pte;
452 }
454 /*
455 * If it's a COW mapping, write protect it both
456 * in the parent and the child
457 */
458 if (is_cow_mapping(vm_flags)) {
459 ptep_set_wrprotect(src_mm, addr, src_pte);
460 pte = *src_pte;
461 }
463 /*
464 * If it's a shared mapping, mark it clean in
465 * the child
466 */
467 if (vm_flags & VM_SHARED)
468 pte = pte_mkclean(pte);
469 pte = pte_mkold(pte);
471 page = vm_normal_page(vma, addr, pte);
472 if (page) {
473 get_page(page);
474 page_dup_rmap(page);
475 rss[!!PageAnon(page)]++;
476 }
478 out_set_pte:
479 set_pte_at(dst_mm, addr, dst_pte, pte);
480 }
482 static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
483 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
484 unsigned long addr, unsigned long end)
485 {
486 pte_t *src_pte, *dst_pte;
487 spinlock_t *src_ptl, *dst_ptl;
488 int progress = 0;
489 int rss[2];
491 again:
492 rss[1] = rss[0] = 0;
493 dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
494 if (!dst_pte)
495 return -ENOMEM;
496 src_pte = pte_offset_map_nested(src_pmd, addr);
497 src_ptl = pte_lockptr(src_mm, src_pmd);
498 spin_lock(src_ptl);
500 do {
501 /*
502 * We are holding two locks at this point - either of them
503 * could generate latencies in another task on another CPU.
504 */
505 if (progress >= 32) {
506 progress = 0;
507 if (need_resched() ||
508 need_lockbreak(src_ptl) ||
509 need_lockbreak(dst_ptl))
510 break;
511 }
512 if (pte_none(*src_pte)) {
513 progress++;
514 continue;
515 }
516 copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss);
517 progress += 8;
518 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
520 spin_unlock(src_ptl);
521 pte_unmap_nested(src_pte - 1);
522 add_mm_rss(dst_mm, rss[0], rss[1]);
523 pte_unmap_unlock(dst_pte - 1, dst_ptl);
524 cond_resched();
525 if (addr != end)
526 goto again;
527 return 0;
528 }
530 static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
531 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
532 unsigned long addr, unsigned long end)
533 {
534 pmd_t *src_pmd, *dst_pmd;
535 unsigned long next;
537 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
538 if (!dst_pmd)
539 return -ENOMEM;
540 src_pmd = pmd_offset(src_pud, addr);
541 do {
542 next = pmd_addr_end(addr, end);
543 if (pmd_none_or_clear_bad(src_pmd))
544 continue;
545 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
546 vma, addr, next))
547 return -ENOMEM;
548 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
549 return 0;
550 }
552 static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
553 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
554 unsigned long addr, unsigned long end)
555 {
556 pud_t *src_pud, *dst_pud;
557 unsigned long next;
559 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
560 if (!dst_pud)
561 return -ENOMEM;
562 src_pud = pud_offset(src_pgd, addr);
563 do {
564 next = pud_addr_end(addr, end);
565 if (pud_none_or_clear_bad(src_pud))
566 continue;
567 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
568 vma, addr, next))
569 return -ENOMEM;
570 } while (dst_pud++, src_pud++, addr = next, addr != end);
571 return 0;
572 }
574 int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
575 struct vm_area_struct *vma)
576 {
577 pgd_t *src_pgd, *dst_pgd;
578 unsigned long next;
579 unsigned long addr = vma->vm_start;
580 unsigned long end = vma->vm_end;
582 /*
583 * Don't copy ptes where a page fault will fill them correctly.
584 * Fork becomes much lighter when there are big shared or private
585 * readonly mappings. The tradeoff is that copy_page_range is more
586 * efficient than faulting.
587 */
588 if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) {
589 if (!vma->anon_vma)
590 return 0;
591 }
593 if (is_vm_hugetlb_page(vma))
594 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
596 dst_pgd = pgd_offset(dst_mm, addr);
597 src_pgd = pgd_offset(src_mm, addr);
598 do {
599 next = pgd_addr_end(addr, end);
600 if (pgd_none_or_clear_bad(src_pgd))
601 continue;
602 if (copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
603 vma, addr, next))
604 return -ENOMEM;
605 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
606 return 0;
607 }
609 static unsigned long zap_pte_range(struct mmu_gather *tlb,
610 struct vm_area_struct *vma, pmd_t *pmd,
611 unsigned long addr, unsigned long end,
612 long *zap_work, struct zap_details *details)
613 {
614 struct mm_struct *mm = tlb->mm;
615 pte_t *pte;
616 spinlock_t *ptl;
617 int file_rss = 0;
618 int anon_rss = 0;
620 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
621 do {
622 pte_t ptent = *pte;
623 if (pte_none(ptent)) {
624 (*zap_work)--;
625 continue;
626 }
627 if (pte_present(ptent)) {
628 struct page *page;
630 (*zap_work) -= PAGE_SIZE;
632 page = vm_normal_page(vma, addr, ptent);
633 if (unlikely(details) && page) {
634 /*
635 * unmap_shared_mapping_pages() wants to
636 * invalidate cache without truncating:
637 * unmap shared but keep private pages.
638 */
639 if (details->check_mapping &&
640 details->check_mapping != page->mapping)
641 continue;
642 /*
643 * Each page->index must be checked when
644 * invalidating or truncating nonlinear.
645 */
646 if (details->nonlinear_vma &&
647 (page->index < details->first_index ||
648 page->index > details->last_index))
649 continue;
650 }
651 ptent = ptep_get_and_clear_full(mm, addr, pte,
652 tlb->fullmm);
653 tlb_remove_tlb_entry(tlb, pte, addr);
654 if (unlikely(!page))
655 continue;
656 if (unlikely(details) && details->nonlinear_vma
657 && linear_page_index(details->nonlinear_vma,
658 addr) != page->index)
659 set_pte_at(mm, addr, pte,
660 pgoff_to_pte(page->index));
661 if (PageAnon(page))
662 anon_rss--;
663 else {
664 if (pte_dirty(ptent))
665 set_page_dirty(page);
666 if (pte_young(ptent))
667 mark_page_accessed(page);
668 file_rss--;
669 }
670 page_remove_rmap(page);
671 tlb_remove_page(tlb, page);
672 continue;
673 }
674 /*
675 * If details->check_mapping, we leave swap entries;
676 * if details->nonlinear_vma, we leave file entries.
677 */
678 if (unlikely(details))
679 continue;
680 if (!pte_file(ptent))
681 free_swap_and_cache(pte_to_swp_entry(ptent));
682 pte_clear_full(mm, addr, pte, tlb->fullmm);
683 } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0));
685 add_mm_rss(mm, file_rss, anon_rss);
686 pte_unmap_unlock(pte - 1, ptl);
688 return addr;
689 }
691 static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
692 struct vm_area_struct *vma, pud_t *pud,
693 unsigned long addr, unsigned long end,
694 long *zap_work, struct zap_details *details)
695 {
696 pmd_t *pmd;
697 unsigned long next;
699 pmd = pmd_offset(pud, addr);
700 do {
701 next = pmd_addr_end(addr, end);
702 if (pmd_none_or_clear_bad(pmd)) {
703 (*zap_work)--;
704 continue;
705 }
706 next = zap_pte_range(tlb, vma, pmd, addr, next,
707 zap_work, details);
708 } while (pmd++, addr = next, (addr != end && *zap_work > 0));
710 return addr;
711 }
713 static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
714 struct vm_area_struct *vma, pgd_t *pgd,
715 unsigned long addr, unsigned long end,
716 long *zap_work, struct zap_details *details)
717 {
718 pud_t *pud;
719 unsigned long next;
721 pud = pud_offset(pgd, addr);
722 do {
723 next = pud_addr_end(addr, end);
724 if (pud_none_or_clear_bad(pud)) {
725 (*zap_work)--;
726 continue;
727 }
728 next = zap_pmd_range(tlb, vma, pud, addr, next,
729 zap_work, details);
730 } while (pud++, addr = next, (addr != end && *zap_work > 0));
732 return addr;
733 }
735 static unsigned long unmap_page_range(struct mmu_gather *tlb,
736 struct vm_area_struct *vma,
737 unsigned long addr, unsigned long end,
738 long *zap_work, struct zap_details *details)
739 {
740 pgd_t *pgd;
741 unsigned long next;
743 if (details && !details->check_mapping && !details->nonlinear_vma)
744 details = NULL;
746 BUG_ON(addr >= end);
747 tlb_start_vma(tlb, vma);
748 pgd = pgd_offset(vma->vm_mm, addr);
749 do {
750 next = pgd_addr_end(addr, end);
751 if (pgd_none_or_clear_bad(pgd)) {
752 (*zap_work)--;
753 continue;
754 }
755 next = zap_pud_range(tlb, vma, pgd, addr, next,
756 zap_work, details);
757 } while (pgd++, addr = next, (addr != end && *zap_work > 0));
758 tlb_end_vma(tlb, vma);
760 return addr;
761 }
763 #ifdef CONFIG_PREEMPT
764 # define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
765 #else
766 /* No preempt: go for improved straight-line efficiency */
767 # define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
768 #endif
770 /**
771 * unmap_vmas - unmap a range of memory covered by a list of vma's
772 * @tlbp: address of the caller's struct mmu_gather
773 * @vma: the starting vma
774 * @start_addr: virtual address at which to start unmapping
775 * @end_addr: virtual address at which to end unmapping
776 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
777 * @details: details of nonlinear truncation or shared cache invalidation
778 *
779 * Returns the end address of the unmapping (restart addr if interrupted).
780 *
781 * Unmap all pages in the vma list.
782 *
783 * We aim to not hold locks for too long (for scheduling latency reasons).
784 * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
785 * return the ending mmu_gather to the caller.
786 *
787 * Only addresses between `start' and `end' will be unmapped.
788 *
789 * The VMA list must be sorted in ascending virtual address order.
790 *
791 * unmap_vmas() assumes that the caller will flush the whole unmapped address
792 * range after unmap_vmas() returns. So the only responsibility here is to
793 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
794 * drops the lock and schedules.
795 */
796 unsigned long unmap_vmas(struct mmu_gather **tlbp,
797 struct vm_area_struct *vma, unsigned long start_addr,
798 unsigned long end_addr, unsigned long *nr_accounted,
799 struct zap_details *details)
800 {
801 long zap_work = ZAP_BLOCK_SIZE;
802 unsigned long tlb_start = 0; /* For tlb_finish_mmu */
803 int tlb_start_valid = 0;
804 unsigned long start = start_addr;
805 spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
806 int fullmm = (*tlbp)->fullmm;
808 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
809 unsigned long end;
811 start = max(vma->vm_start, start_addr);
812 if (start >= vma->vm_end)
813 continue;
814 end = min(vma->vm_end, end_addr);
815 if (end <= vma->vm_start)
816 continue;
818 if (vma->vm_flags & VM_ACCOUNT)
819 *nr_accounted += (end - start) >> PAGE_SHIFT;
821 while (start != end) {
822 if (!tlb_start_valid) {
823 tlb_start = start;
824 tlb_start_valid = 1;
825 }
827 if (unlikely(is_vm_hugetlb_page(vma))) {
828 unmap_hugepage_range(vma, start, end);
829 zap_work -= (end - start) /
830 (HPAGE_SIZE / PAGE_SIZE);
831 start = end;
832 } else
833 start = unmap_page_range(*tlbp, vma,
834 start, end, &zap_work, details);
836 if (zap_work > 0) {
837 BUG_ON(start != end);
838 break;
839 }
841 tlb_finish_mmu(*tlbp, tlb_start, start);
843 if (need_resched() ||
844 (i_mmap_lock && need_lockbreak(i_mmap_lock))) {
845 if (i_mmap_lock) {
846 *tlbp = NULL;
847 goto out;
848 }
849 cond_resched();
850 }
852 *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm);
853 tlb_start_valid = 0;
854 zap_work = ZAP_BLOCK_SIZE;
855 }
856 }
857 out:
858 return start; /* which is now the end (or restart) address */
859 }
861 /**
862 * zap_page_range - remove user pages in a given range
863 * @vma: vm_area_struct holding the applicable pages
864 * @address: starting address of pages to zap
865 * @size: number of bytes to zap
866 * @details: details of nonlinear truncation or shared cache invalidation
867 */
868 unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
869 unsigned long size, struct zap_details *details)
870 {
871 struct mm_struct *mm = vma->vm_mm;
872 struct mmu_gather *tlb;
873 unsigned long end = address + size;
874 unsigned long nr_accounted = 0;
876 lru_add_drain();
877 tlb = tlb_gather_mmu(mm, 0);
878 update_hiwater_rss(mm);
879 end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details);
880 if (tlb)
881 tlb_finish_mmu(tlb, address, end);
882 return end;
883 }
885 /*
886 * Do a quick page-table lookup for a single page.
887 */
888 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
889 unsigned int flags)
890 {
891 pgd_t *pgd;
892 pud_t *pud;
893 pmd_t *pmd;
894 pte_t *ptep, pte;
895 spinlock_t *ptl;
896 struct page *page;
897 struct mm_struct *mm = vma->vm_mm;
899 page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
900 if (!IS_ERR(page)) {
901 BUG_ON(flags & FOLL_GET);
902 goto out;
903 }
905 page = NULL;
906 pgd = pgd_offset(mm, address);
907 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
908 goto no_page_table;
910 pud = pud_offset(pgd, address);
911 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
912 goto no_page_table;
914 pmd = pmd_offset(pud, address);
915 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
916 goto no_page_table;
918 if (pmd_huge(*pmd)) {
919 BUG_ON(flags & FOLL_GET);
920 page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
921 goto out;
922 }
924 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
925 if (!ptep)
926 goto out;
928 pte = *ptep;
929 if (!pte_present(pte))
930 goto unlock;
931 if ((flags & FOLL_WRITE) && !pte_write(pte))
932 goto unlock;
933 page = vm_normal_page(vma, address, pte);
934 if (unlikely(!page))
935 goto unlock;
937 if (flags & FOLL_GET)
938 get_page(page);
939 if (flags & FOLL_TOUCH) {
940 if ((flags & FOLL_WRITE) &&
941 !pte_dirty(pte) && !PageDirty(page))
942 set_page_dirty(page);
943 mark_page_accessed(page);
944 }
945 unlock:
946 pte_unmap_unlock(ptep, ptl);
947 out:
948 return page;
950 no_page_table:
951 /*
952 * When core dumping an enormous anonymous area that nobody
953 * has touched so far, we don't want to allocate page tables.
954 */
955 if (flags & FOLL_ANON) {
956 page = ZERO_PAGE(address);
957 if (flags & FOLL_GET)
958 get_page(page);
959 BUG_ON(flags & FOLL_WRITE);
960 }
961 return page;
962 }
964 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
965 unsigned long start, int len, int write, int force,
966 struct page **pages, struct vm_area_struct **vmas)
967 {
968 int i;
969 unsigned int vm_flags;
971 /*
972 * Require read or write permissions.
973 * If 'force' is set, we only require the "MAY" flags.
974 */
975 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
976 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
977 i = 0;
979 do {
980 struct vm_area_struct *vma;
981 unsigned int foll_flags;
983 vma = find_extend_vma(mm, start);
984 if (!vma && in_gate_area(tsk, start)) {
985 unsigned long pg = start & PAGE_MASK;
986 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
987 pgd_t *pgd;
988 pud_t *pud;
989 pmd_t *pmd;
990 pte_t *pte;
991 if (write) /* user gate pages are read-only */
992 return i ? : -EFAULT;
993 if (pg > TASK_SIZE)
994 pgd = pgd_offset_k(pg);
995 else
996 pgd = pgd_offset_gate(mm, pg);
997 BUG_ON(pgd_none(*pgd));
998 pud = pud_offset(pgd, pg);
999 BUG_ON(pud_none(*pud));
1000 pmd = pmd_offset(pud, pg);
1001 if (pmd_none(*pmd))
1002 return i ? : -EFAULT;
1003 pte = pte_offset_map(pmd, pg);
1004 if (pte_none(*pte)) {
1005 pte_unmap(pte);
1006 return i ? : -EFAULT;
1008 if (pages) {
1009 struct page *page = vm_normal_page(gate_vma, start, *pte);
1010 pages[i] = page;
1011 if (page)
1012 get_page(page);
1014 pte_unmap(pte);
1015 if (vmas)
1016 vmas[i] = gate_vma;
1017 i++;
1018 start += PAGE_SIZE;
1019 len--;
1020 continue;
1023 #ifdef CONFIG_XEN
1024 if (vma && (vma->vm_flags & VM_FOREIGN)) {
1025 struct page **map = vma->vm_private_data;
1026 int offset = (start - vma->vm_start) >> PAGE_SHIFT;
1028 if (map[offset] != NULL) {
1029 if (pages)
1030 pages[i] = map[offset];
1031 if (vmas)
1032 vmas[i] = vma;
1033 i++;
1034 start += PAGE_SIZE;
1035 len--;
1036 continue;
1039 #endif
1040 if (!vma || (vma->vm_flags & (VM_IO | VM_PFNMAP))
1041 || !(vm_flags & vma->vm_flags))
1042 return i ? : -EFAULT;
1044 if (is_vm_hugetlb_page(vma)) {
1045 i = follow_hugetlb_page(mm, vma, pages, vmas,
1046 &start, &len, i);
1047 continue;
1050 foll_flags = FOLL_TOUCH;
1051 if (pages)
1052 foll_flags |= FOLL_GET;
1053 if (!write && !(vma->vm_flags & VM_LOCKED) &&
1054 (!vma->vm_ops || !vma->vm_ops->nopage))
1055 foll_flags |= FOLL_ANON;
1057 do {
1058 struct page *page;
1060 if (write)
1061 foll_flags |= FOLL_WRITE;
1063 cond_resched();
1064 while (!(page = follow_page(vma, start, foll_flags))) {
1065 int ret;
1066 ret = __handle_mm_fault(mm, vma, start,
1067 foll_flags & FOLL_WRITE);
1068 /*
1069 * The VM_FAULT_WRITE bit tells us that do_wp_page has
1070 * broken COW when necessary, even if maybe_mkwrite
1071 * decided not to set pte_write. We can thus safely do
1072 * subsequent page lookups as if they were reads.
1073 */
1074 if (ret & VM_FAULT_WRITE)
1075 foll_flags &= ~FOLL_WRITE;
1077 switch (ret & ~VM_FAULT_WRITE) {
1078 case VM_FAULT_MINOR:
1079 tsk->min_flt++;
1080 break;
1081 case VM_FAULT_MAJOR:
1082 tsk->maj_flt++;
1083 break;
1084 case VM_FAULT_SIGBUS:
1085 return i ? i : -EFAULT;
1086 case VM_FAULT_OOM:
1087 return i ? i : -ENOMEM;
1088 default:
1089 BUG();
1092 if (pages) {
1093 pages[i] = page;
1094 flush_dcache_page(page);
1096 if (vmas)
1097 vmas[i] = vma;
1098 i++;
1099 start += PAGE_SIZE;
1100 len--;
1101 } while (len && start < vma->vm_end);
1102 } while (len);
1103 return i;
1105 EXPORT_SYMBOL(get_user_pages);
1107 static int zeromap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1108 unsigned long addr, unsigned long end, pgprot_t prot)
1110 pte_t *pte;
1111 spinlock_t *ptl;
1113 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
1114 if (!pte)
1115 return -ENOMEM;
1116 do {
1117 struct page *page = ZERO_PAGE(addr);
1118 pte_t zero_pte = pte_wrprotect(mk_pte(page, prot));
1119 page_cache_get(page);
1120 page_add_file_rmap(page);
1121 inc_mm_counter(mm, file_rss);
1122 BUG_ON(!pte_none(*pte));
1123 set_pte_at(mm, addr, pte, zero_pte);
1124 } while (pte++, addr += PAGE_SIZE, addr != end);
1125 pte_unmap_unlock(pte - 1, ptl);
1126 return 0;
1129 static inline int zeromap_pmd_range(struct mm_struct *mm, pud_t *pud,
1130 unsigned long addr, unsigned long end, pgprot_t prot)
1132 pmd_t *pmd;
1133 unsigned long next;
1135 pmd = pmd_alloc(mm, pud, addr);
1136 if (!pmd)
1137 return -ENOMEM;
1138 do {
1139 next = pmd_addr_end(addr, end);
1140 if (zeromap_pte_range(mm, pmd, addr, next, prot))
1141 return -ENOMEM;
1142 } while (pmd++, addr = next, addr != end);
1143 return 0;
1146 static inline int zeromap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1147 unsigned long addr, unsigned long end, pgprot_t prot)
1149 pud_t *pud;
1150 unsigned long next;
1152 pud = pud_alloc(mm, pgd, addr);
1153 if (!pud)
1154 return -ENOMEM;
1155 do {
1156 next = pud_addr_end(addr, end);
1157 if (zeromap_pmd_range(mm, pud, addr, next, prot))
1158 return -ENOMEM;
1159 } while (pud++, addr = next, addr != end);
1160 return 0;
1163 int zeromap_page_range(struct vm_area_struct *vma,
1164 unsigned long addr, unsigned long size, pgprot_t prot)
1166 pgd_t *pgd;
1167 unsigned long next;
1168 unsigned long end = addr + size;
1169 struct mm_struct *mm = vma->vm_mm;
1170 int err;
1172 BUG_ON(addr >= end);
1173 pgd = pgd_offset(mm, addr);
1174 flush_cache_range(vma, addr, end);
1175 do {
1176 next = pgd_addr_end(addr, end);
1177 err = zeromap_pud_range(mm, pgd, addr, next, prot);
1178 if (err)
1179 break;
1180 } while (pgd++, addr = next, addr != end);
1181 return err;
1184 pte_t * fastcall get_locked_pte(struct mm_struct *mm, unsigned long addr, spinlock_t **ptl)
1186 pgd_t * pgd = pgd_offset(mm, addr);
1187 pud_t * pud = pud_alloc(mm, pgd, addr);
1188 if (pud) {
1189 pmd_t * pmd = pmd_alloc(mm, pud, addr);
1190 if (pmd)
1191 return pte_alloc_map_lock(mm, pmd, addr, ptl);
1193 return NULL;
1196 /*
1197 * This is the old fallback for page remapping.
1199 * For historical reasons, it only allows reserved pages. Only
1200 * old drivers should use this, and they needed to mark their
1201 * pages reserved for the old functions anyway.
1202 */
1203 static int insert_page(struct mm_struct *mm, unsigned long addr, struct page *page, pgprot_t prot)
1205 int retval;
1206 pte_t *pte;
1207 spinlock_t *ptl;
1209 retval = -EINVAL;
1210 if (PageAnon(page))
1211 goto out;
1212 retval = -ENOMEM;
1213 flush_dcache_page(page);
1214 pte = get_locked_pte(mm, addr, &ptl);
1215 if (!pte)
1216 goto out;
1217 retval = -EBUSY;
1218 if (!pte_none(*pte))
1219 goto out_unlock;
1221 /* Ok, finally just insert the thing.. */
1222 get_page(page);
1223 inc_mm_counter(mm, file_rss);
1224 page_add_file_rmap(page);
1225 set_pte_at(mm, addr, pte, mk_pte(page, prot));
1227 retval = 0;
1228 out_unlock:
1229 pte_unmap_unlock(pte, ptl);
1230 out:
1231 return retval;
1234 /*
1235 * This allows drivers to insert individual pages they've allocated
1236 * into a user vma.
1238 * The page has to be a nice clean _individual_ kernel allocation.
1239 * If you allocate a compound page, you need to have marked it as
1240 * such (__GFP_COMP), or manually just split the page up yourself
1241 * (which is mainly an issue of doing "set_page_count(page, 1)" for
1242 * each sub-page, and then freeing them one by one when you free
1243 * them rather than freeing it as a compound page).
1245 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1246 * took an arbitrary page protection parameter. This doesn't allow
1247 * that. Your vma protection will have to be set up correctly, which
1248 * means that if you want a shared writable mapping, you'd better
1249 * ask for a shared writable mapping!
1251 * The page does not need to be reserved.
1252 */
1253 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, struct page *page)
1255 if (addr < vma->vm_start || addr >= vma->vm_end)
1256 return -EFAULT;
1257 if (!page_count(page))
1258 return -EINVAL;
1259 vma->vm_flags |= VM_INSERTPAGE;
1260 return insert_page(vma->vm_mm, addr, page, vma->vm_page_prot);
1262 EXPORT_SYMBOL(vm_insert_page);
1264 /*
1265 * maps a range of physical memory into the requested pages. the old
1266 * mappings are removed. any references to nonexistent pages results
1267 * in null mappings (currently treated as "copy-on-access")
1268 */
1269 static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1270 unsigned long addr, unsigned long end,
1271 unsigned long pfn, pgprot_t prot)
1273 pte_t *pte;
1274 spinlock_t *ptl;
1276 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
1277 if (!pte)
1278 return -ENOMEM;
1279 do {
1280 BUG_ON(!pte_none(*pte));
1281 set_pte_at(mm, addr, pte, pfn_pte(pfn, prot));
1282 pfn++;
1283 } while (pte++, addr += PAGE_SIZE, addr != end);
1284 pte_unmap_unlock(pte - 1, ptl);
1285 return 0;
1288 static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1289 unsigned long addr, unsigned long end,
1290 unsigned long pfn, pgprot_t prot)
1292 pmd_t *pmd;
1293 unsigned long next;
1295 pfn -= addr >> PAGE_SHIFT;
1296 pmd = pmd_alloc(mm, pud, addr);
1297 if (!pmd)
1298 return -ENOMEM;
1299 do {
1300 next = pmd_addr_end(addr, end);
1301 if (remap_pte_range(mm, pmd, addr, next,
1302 pfn + (addr >> PAGE_SHIFT), prot))
1303 return -ENOMEM;
1304 } while (pmd++, addr = next, addr != end);
1305 return 0;
1308 static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1309 unsigned long addr, unsigned long end,
1310 unsigned long pfn, pgprot_t prot)
1312 pud_t *pud;
1313 unsigned long next;
1315 pfn -= addr >> PAGE_SHIFT;
1316 pud = pud_alloc(mm, pgd, addr);
1317 if (!pud)
1318 return -ENOMEM;
1319 do {
1320 next = pud_addr_end(addr, end);
1321 if (remap_pmd_range(mm, pud, addr, next,
1322 pfn + (addr >> PAGE_SHIFT), prot))
1323 return -ENOMEM;
1324 } while (pud++, addr = next, addr != end);
1325 return 0;
1328 /* Note: this is only safe if the mm semaphore is held when called. */
1329 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1330 unsigned long pfn, unsigned long size, pgprot_t prot)
1332 pgd_t *pgd;
1333 unsigned long next;
1334 unsigned long end = addr + PAGE_ALIGN(size);
1335 struct mm_struct *mm = vma->vm_mm;
1336 int err;
1338 /*
1339 * Physically remapped pages are special. Tell the
1340 * rest of the world about it:
1341 * VM_IO tells people not to look at these pages
1342 * (accesses can have side effects).
1343 * VM_RESERVED is specified all over the place, because
1344 * in 2.4 it kept swapout's vma scan off this vma; but
1345 * in 2.6 the LRU scan won't even find its pages, so this
1346 * flag means no more than count its pages in reserved_vm,
1347 * and omit it from core dump, even when VM_IO turned off.
1348 * VM_PFNMAP tells the core MM that the base pages are just
1349 * raw PFN mappings, and do not have a "struct page" associated
1350 * with them.
1352 * There's a horrible special case to handle copy-on-write
1353 * behaviour that some programs depend on. We mark the "original"
1354 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
1355 */
1356 if (is_cow_mapping(vma->vm_flags)) {
1357 if (addr != vma->vm_start || end != vma->vm_end)
1358 return -EINVAL;
1359 vma->vm_pgoff = pfn;
1362 vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
1364 BUG_ON(addr >= end);
1365 pfn -= addr >> PAGE_SHIFT;
1366 pgd = pgd_offset(mm, addr);
1367 flush_cache_range(vma, addr, end);
1368 do {
1369 next = pgd_addr_end(addr, end);
1370 err = remap_pud_range(mm, pgd, addr, next,
1371 pfn + (addr >> PAGE_SHIFT), prot);
1372 if (err)
1373 break;
1374 } while (pgd++, addr = next, addr != end);
1375 return err;
1377 EXPORT_SYMBOL(remap_pfn_range);
1379 #ifdef CONFIG_XEN
1380 static inline int generic_pte_range(struct mm_struct *mm, pmd_t *pmd,
1381 unsigned long addr, unsigned long end,
1382 pte_fn_t fn, void *data)
1384 pte_t *pte;
1385 int err;
1386 struct page *pte_page;
1388 pte = (mm == &init_mm) ?
1389 pte_alloc_kernel(pmd, addr) :
1390 pte_alloc_map(mm, pmd, addr);
1391 if (!pte)
1392 return -ENOMEM;
1394 pte_page = pmd_page(*pmd);
1396 do {
1397 err = fn(pte, pte_page, addr, data);
1398 if (err)
1399 break;
1400 } while (pte++, addr += PAGE_SIZE, addr != end);
1402 if (mm != &init_mm)
1403 pte_unmap(pte-1);
1404 return err;
1407 static inline int generic_pmd_range(struct mm_struct *mm, pud_t *pud,
1408 unsigned long addr, unsigned long end,
1409 pte_fn_t fn, void *data)
1411 pmd_t *pmd;
1412 unsigned long next;
1413 int err;
1415 pmd = pmd_alloc(mm, pud, addr);
1416 if (!pmd)
1417 return -ENOMEM;
1418 do {
1419 next = pmd_addr_end(addr, end);
1420 err = generic_pte_range(mm, pmd, addr, next, fn, data);
1421 if (err)
1422 break;
1423 } while (pmd++, addr = next, addr != end);
1424 return err;
1427 static inline int generic_pud_range(struct mm_struct *mm, pgd_t *pgd,
1428 unsigned long addr, unsigned long end,
1429 pte_fn_t fn, void *data)
1431 pud_t *pud;
1432 unsigned long next;
1433 int err;
1435 pud = pud_alloc(mm, pgd, addr);
1436 if (!pud)
1437 return -ENOMEM;
1438 do {
1439 next = pud_addr_end(addr, end);
1440 err = generic_pmd_range(mm, pud, addr, next, fn, data);
1441 if (err)
1442 break;
1443 } while (pud++, addr = next, addr != end);
1444 return err;
1447 /*
1448 * Scan a region of virtual memory, filling in page tables as necessary
1449 * and calling a provided function on each leaf page table.
1450 */
1451 int generic_page_range(struct mm_struct *mm, unsigned long addr,
1452 unsigned long size, pte_fn_t fn, void *data)
1454 pgd_t *pgd;
1455 unsigned long next;
1456 unsigned long end = addr + size;
1457 int err;
1459 BUG_ON(addr >= end);
1460 pgd = pgd_offset(mm, addr);
1461 do {
1462 next = pgd_addr_end(addr, end);
1463 err = generic_pud_range(mm, pgd, addr, next, fn, data);
1464 if (err)
1465 break;
1466 } while (pgd++, addr = next, addr != end);
1467 return err;
1469 #endif
1471 /*
1472 * handle_pte_fault chooses page fault handler according to an entry
1473 * which was read non-atomically. Before making any commitment, on
1474 * those architectures or configurations (e.g. i386 with PAE) which
1475 * might give a mix of unmatched parts, do_swap_page and do_file_page
1476 * must check under lock before unmapping the pte and proceeding
1477 * (but do_wp_page is only called after already making such a check;
1478 * and do_anonymous_page and do_no_page can safely check later on).
1479 */
1480 static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
1481 pte_t *page_table, pte_t orig_pte)
1483 int same = 1;
1484 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1485 if (sizeof(pte_t) > sizeof(unsigned long)) {
1486 spinlock_t *ptl = pte_lockptr(mm, pmd);
1487 spin_lock(ptl);
1488 same = pte_same(*page_table, orig_pte);
1489 spin_unlock(ptl);
1491 #endif
1492 pte_unmap(page_table);
1493 return same;
1496 /*
1497 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1498 * servicing faults for write access. In the normal case, do always want
1499 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1500 * that do not have writing enabled, when used by access_process_vm.
1501 */
1502 static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1504 if (likely(vma->vm_flags & VM_WRITE))
1505 pte = pte_mkwrite(pte);
1506 return pte;
1509 static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va)
1511 /*
1512 * If the source page was a PFN mapping, we don't have
1513 * a "struct page" for it. We do a best-effort copy by
1514 * just copying from the original user address. If that
1515 * fails, we just zero-fill it. Live with it.
1516 */
1517 if (unlikely(!src)) {
1518 void *kaddr = kmap_atomic(dst, KM_USER0);
1519 void __user *uaddr = (void __user *)(va & PAGE_MASK);
1521 /*
1522 * This really shouldn't fail, because the page is there
1523 * in the page tables. But it might just be unreadable,
1524 * in which case we just give up and fill the result with
1525 * zeroes.
1526 */
1527 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
1528 memset(kaddr, 0, PAGE_SIZE);
1529 kunmap_atomic(kaddr, KM_USER0);
1530 return;
1533 copy_user_highpage(dst, src, va);
1536 /*
1537 * This routine handles present pages, when users try to write
1538 * to a shared page. It is done by copying the page to a new address
1539 * and decrementing the shared-page counter for the old page.
1541 * Note that this routine assumes that the protection checks have been
1542 * done by the caller (the low-level page fault routine in most cases).
1543 * Thus we can safely just mark it writable once we've done any necessary
1544 * COW.
1546 * We also mark the page dirty at this point even though the page will
1547 * change only once the write actually happens. This avoids a few races,
1548 * and potentially makes it more efficient.
1550 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1551 * but allow concurrent faults), with pte both mapped and locked.
1552 * We return with mmap_sem still held, but pte unmapped and unlocked.
1553 */
1554 static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
1555 unsigned long address, pte_t *page_table, pmd_t *pmd,
1556 spinlock_t *ptl, pte_t orig_pte)
1558 struct page *old_page, *new_page;
1559 pte_t entry;
1560 int ret = VM_FAULT_MINOR;
1562 old_page = vm_normal_page(vma, address, orig_pte);
1563 if (!old_page)
1564 goto gotten;
1566 if (PageAnon(old_page) && !TestSetPageLocked(old_page)) {
1567 int reuse = can_share_swap_page(old_page);
1568 unlock_page(old_page);
1569 if (reuse) {
1570 flush_cache_page(vma, address, pte_pfn(orig_pte));
1571 entry = pte_mkyoung(orig_pte);
1572 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1573 ptep_set_access_flags(vma, address, page_table, entry, 1);
1574 update_mmu_cache(vma, address, entry);
1575 lazy_mmu_prot_update(entry);
1576 ret |= VM_FAULT_WRITE;
1577 goto unlock;
1581 /*
1582 * Ok, we need to copy. Oh, well..
1583 */
1584 page_cache_get(old_page);
1585 gotten:
1586 pte_unmap_unlock(page_table, ptl);
1588 if (unlikely(anon_vma_prepare(vma)))
1589 goto oom;
1590 if (old_page == ZERO_PAGE(address)) {
1591 new_page = alloc_zeroed_user_highpage(vma, address);
1592 if (!new_page)
1593 goto oom;
1594 } else {
1595 new_page = alloc_page_vma(GFP_HIGHUSER, vma, address);
1596 if (!new_page)
1597 goto oom;
1598 cow_user_page(new_page, old_page, address);
1601 /*
1602 * Re-check the pte - we dropped the lock
1603 */
1604 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
1605 if (likely(pte_same(*page_table, orig_pte))) {
1606 if (old_page) {
1607 page_remove_rmap(old_page);
1608 if (!PageAnon(old_page)) {
1609 dec_mm_counter(mm, file_rss);
1610 inc_mm_counter(mm, anon_rss);
1612 } else
1613 inc_mm_counter(mm, anon_rss);
1614 flush_cache_page(vma, address, pte_pfn(orig_pte));
1615 entry = mk_pte(new_page, vma->vm_page_prot);
1616 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1617 ptep_establish(vma, address, page_table, entry);
1618 update_mmu_cache(vma, address, entry);
1619 lazy_mmu_prot_update(entry);
1620 lru_cache_add_active(new_page);
1621 page_add_new_anon_rmap(new_page, vma, address);
1623 /* Free the old page.. */
1624 new_page = old_page;
1625 ret |= VM_FAULT_WRITE;
1627 if (new_page)
1628 page_cache_release(new_page);
1629 if (old_page)
1630 page_cache_release(old_page);
1631 unlock:
1632 pte_unmap_unlock(page_table, ptl);
1633 return ret;
1634 oom:
1635 if (old_page)
1636 page_cache_release(old_page);
1637 return VM_FAULT_OOM;
1640 /*
1641 * Helper functions for unmap_mapping_range().
1643 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1645 * We have to restart searching the prio_tree whenever we drop the lock,
1646 * since the iterator is only valid while the lock is held, and anyway
1647 * a later vma might be split and reinserted earlier while lock dropped.
1649 * The list of nonlinear vmas could be handled more efficiently, using
1650 * a placeholder, but handle it in the same way until a need is shown.
1651 * It is important to search the prio_tree before nonlinear list: a vma
1652 * may become nonlinear and be shifted from prio_tree to nonlinear list
1653 * while the lock is dropped; but never shifted from list to prio_tree.
1655 * In order to make forward progress despite restarting the search,
1656 * vm_truncate_count is used to mark a vma as now dealt with, so we can
1657 * quickly skip it next time around. Since the prio_tree search only
1658 * shows us those vmas affected by unmapping the range in question, we
1659 * can't efficiently keep all vmas in step with mapping->truncate_count:
1660 * so instead reset them all whenever it wraps back to 0 (then go to 1).
1661 * mapping->truncate_count and vma->vm_truncate_count are protected by
1662 * i_mmap_lock.
1664 * In order to make forward progress despite repeatedly restarting some
1665 * large vma, note the restart_addr from unmap_vmas when it breaks out:
1666 * and restart from that address when we reach that vma again. It might
1667 * have been split or merged, shrunk or extended, but never shifted: so
1668 * restart_addr remains valid so long as it remains in the vma's range.
1669 * unmap_mapping_range forces truncate_count to leap over page-aligned
1670 * values so we can save vma's restart_addr in its truncate_count field.
1671 */
1672 #define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
1674 static void reset_vma_truncate_counts(struct address_space *mapping)
1676 struct vm_area_struct *vma;
1677 struct prio_tree_iter iter;
1679 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
1680 vma->vm_truncate_count = 0;
1681 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
1682 vma->vm_truncate_count = 0;
1685 static int unmap_mapping_range_vma(struct vm_area_struct *vma,
1686 unsigned long start_addr, unsigned long end_addr,
1687 struct zap_details *details)
1689 unsigned long restart_addr;
1690 int need_break;
1692 again:
1693 restart_addr = vma->vm_truncate_count;
1694 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
1695 start_addr = restart_addr;
1696 if (start_addr >= end_addr) {
1697 /* Top of vma has been split off since last time */
1698 vma->vm_truncate_count = details->truncate_count;
1699 return 0;
1703 restart_addr = zap_page_range(vma, start_addr,
1704 end_addr - start_addr, details);
1705 need_break = need_resched() ||
1706 need_lockbreak(details->i_mmap_lock);
1708 if (restart_addr >= end_addr) {
1709 /* We have now completed this vma: mark it so */
1710 vma->vm_truncate_count = details->truncate_count;
1711 if (!need_break)
1712 return 0;
1713 } else {
1714 /* Note restart_addr in vma's truncate_count field */
1715 vma->vm_truncate_count = restart_addr;
1716 if (!need_break)
1717 goto again;
1720 spin_unlock(details->i_mmap_lock);
1721 cond_resched();
1722 spin_lock(details->i_mmap_lock);
1723 return -EINTR;
1726 static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
1727 struct zap_details *details)
1729 struct vm_area_struct *vma;
1730 struct prio_tree_iter iter;
1731 pgoff_t vba, vea, zba, zea;
1733 restart:
1734 vma_prio_tree_foreach(vma, &iter, root,
1735 details->first_index, details->last_index) {
1736 /* Skip quickly over those we have already dealt with */
1737 if (vma->vm_truncate_count == details->truncate_count)
1738 continue;
1740 vba = vma->vm_pgoff;
1741 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
1742 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1743 zba = details->first_index;
1744 if (zba < vba)
1745 zba = vba;
1746 zea = details->last_index;
1747 if (zea > vea)
1748 zea = vea;
1750 if (unmap_mapping_range_vma(vma,
1751 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
1752 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
1753 details) < 0)
1754 goto restart;
1758 static inline void unmap_mapping_range_list(struct list_head *head,
1759 struct zap_details *details)
1761 struct vm_area_struct *vma;
1763 /*
1764 * In nonlinear VMAs there is no correspondence between virtual address
1765 * offset and file offset. So we must perform an exhaustive search
1766 * across *all* the pages in each nonlinear VMA, not just the pages
1767 * whose virtual address lies outside the file truncation point.
1768 */
1769 restart:
1770 list_for_each_entry(vma, head, shared.vm_set.list) {
1771 /* Skip quickly over those we have already dealt with */
1772 if (vma->vm_truncate_count == details->truncate_count)
1773 continue;
1774 details->nonlinear_vma = vma;
1775 if (unmap_mapping_range_vma(vma, vma->vm_start,
1776 vma->vm_end, details) < 0)
1777 goto restart;
1781 /**
1782 * unmap_mapping_range - unmap the portion of all mmaps
1783 * in the specified address_space corresponding to the specified
1784 * page range in the underlying file.
1785 * @mapping: the address space containing mmaps to be unmapped.
1786 * @holebegin: byte in first page to unmap, relative to the start of
1787 * the underlying file. This will be rounded down to a PAGE_SIZE
1788 * boundary. Note that this is different from vmtruncate(), which
1789 * must keep the partial page. In contrast, we must get rid of
1790 * partial pages.
1791 * @holelen: size of prospective hole in bytes. This will be rounded
1792 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
1793 * end of the file.
1794 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
1795 * but 0 when invalidating pagecache, don't throw away private data.
1796 */
1797 void unmap_mapping_range(struct address_space *mapping,
1798 loff_t const holebegin, loff_t const holelen, int even_cows)
1800 struct zap_details details;
1801 pgoff_t hba = holebegin >> PAGE_SHIFT;
1802 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1804 /* Check for overflow. */
1805 if (sizeof(holelen) > sizeof(hlen)) {
1806 long long holeend =
1807 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1808 if (holeend & ~(long long)ULONG_MAX)
1809 hlen = ULONG_MAX - hba + 1;
1812 details.check_mapping = even_cows? NULL: mapping;
1813 details.nonlinear_vma = NULL;
1814 details.first_index = hba;
1815 details.last_index = hba + hlen - 1;
1816 if (details.last_index < details.first_index)
1817 details.last_index = ULONG_MAX;
1818 details.i_mmap_lock = &mapping->i_mmap_lock;
1820 spin_lock(&mapping->i_mmap_lock);
1822 /* serialize i_size write against truncate_count write */
1823 smp_wmb();
1824 /* Protect against page faults, and endless unmapping loops */
1825 mapping->truncate_count++;
1826 /*
1827 * For archs where spin_lock has inclusive semantics like ia64
1828 * this smp_mb() will prevent to read pagetable contents
1829 * before the truncate_count increment is visible to
1830 * other cpus.
1831 */
1832 smp_mb();
1833 if (unlikely(is_restart_addr(mapping->truncate_count))) {
1834 if (mapping->truncate_count == 0)
1835 reset_vma_truncate_counts(mapping);
1836 mapping->truncate_count++;
1838 details.truncate_count = mapping->truncate_count;
1840 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
1841 unmap_mapping_range_tree(&mapping->i_mmap, &details);
1842 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
1843 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
1844 spin_unlock(&mapping->i_mmap_lock);
1846 EXPORT_SYMBOL(unmap_mapping_range);
1848 /*
1849 * Handle all mappings that got truncated by a "truncate()"
1850 * system call.
1852 * NOTE! We have to be ready to update the memory sharing
1853 * between the file and the memory map for a potential last
1854 * incomplete page. Ugly, but necessary.
1855 */
1856 int vmtruncate(struct inode * inode, loff_t offset)
1858 struct address_space *mapping = inode->i_mapping;
1859 unsigned long limit;
1861 if (inode->i_size < offset)
1862 goto do_expand;
1863 /*
1864 * truncation of in-use swapfiles is disallowed - it would cause
1865 * subsequent swapout to scribble on the now-freed blocks.
1866 */
1867 if (IS_SWAPFILE(inode))
1868 goto out_busy;
1869 i_size_write(inode, offset);
1870 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
1871 truncate_inode_pages(mapping, offset);
1872 goto out_truncate;
1874 do_expand:
1875 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1876 if (limit != RLIM_INFINITY && offset > limit)
1877 goto out_sig;
1878 if (offset > inode->i_sb->s_maxbytes)
1879 goto out_big;
1880 i_size_write(inode, offset);
1882 out_truncate:
1883 if (inode->i_op && inode->i_op->truncate)
1884 inode->i_op->truncate(inode);
1885 return 0;
1886 out_sig:
1887 send_sig(SIGXFSZ, current, 0);
1888 out_big:
1889 return -EFBIG;
1890 out_busy:
1891 return -ETXTBSY;
1893 EXPORT_SYMBOL(vmtruncate);
1895 int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
1897 struct address_space *mapping = inode->i_mapping;
1899 /*
1900 * If the underlying filesystem is not going to provide
1901 * a way to truncate a range of blocks (punch a hole) -
1902 * we should return failure right now.
1903 */
1904 if (!inode->i_op || !inode->i_op->truncate_range)
1905 return -ENOSYS;
1907 mutex_lock(&inode->i_mutex);
1908 down_write(&inode->i_alloc_sem);
1909 unmap_mapping_range(mapping, offset, (end - offset), 1);
1910 truncate_inode_pages_range(mapping, offset, end);
1911 inode->i_op->truncate_range(inode, offset, end);
1912 up_write(&inode->i_alloc_sem);
1913 mutex_unlock(&inode->i_mutex);
1915 return 0;
1917 EXPORT_SYMBOL(vmtruncate_range);
1919 /*
1920 * Primitive swap readahead code. We simply read an aligned block of
1921 * (1 << page_cluster) entries in the swap area. This method is chosen
1922 * because it doesn't cost us any seek time. We also make sure to queue
1923 * the 'original' request together with the readahead ones...
1925 * This has been extended to use the NUMA policies from the mm triggering
1926 * the readahead.
1928 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
1929 */
1930 void swapin_readahead(swp_entry_t entry, unsigned long addr,struct vm_area_struct *vma)
1932 #ifdef CONFIG_NUMA
1933 struct vm_area_struct *next_vma = vma ? vma->vm_next : NULL;
1934 #endif
1935 int i, num;
1936 struct page *new_page;
1937 unsigned long offset;
1939 /*
1940 * Get the number of handles we should do readahead io to.
1941 */
1942 num = valid_swaphandles(entry, &offset);
1943 for (i = 0; i < num; offset++, i++) {
1944 /* Ok, do the async read-ahead now */
1945 new_page = read_swap_cache_async(swp_entry(swp_type(entry),
1946 offset), vma, addr);
1947 if (!new_page)
1948 break;
1949 page_cache_release(new_page);
1950 #ifdef CONFIG_NUMA
1951 /*
1952 * Find the next applicable VMA for the NUMA policy.
1953 */
1954 addr += PAGE_SIZE;
1955 if (addr == 0)
1956 vma = NULL;
1957 if (vma) {
1958 if (addr >= vma->vm_end) {
1959 vma = next_vma;
1960 next_vma = vma ? vma->vm_next : NULL;
1962 if (vma && addr < vma->vm_start)
1963 vma = NULL;
1964 } else {
1965 if (next_vma && addr >= next_vma->vm_start) {
1966 vma = next_vma;
1967 next_vma = vma->vm_next;
1970 #endif
1972 lru_add_drain(); /* Push any new pages onto the LRU now */
1975 /*
1976 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1977 * but allow concurrent faults), and pte mapped but not yet locked.
1978 * We return with mmap_sem still held, but pte unmapped and unlocked.
1979 */
1980 static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
1981 unsigned long address, pte_t *page_table, pmd_t *pmd,
1982 int write_access, pte_t orig_pte)
1984 spinlock_t *ptl;
1985 struct page *page;
1986 swp_entry_t entry;
1987 pte_t pte;
1988 int ret = VM_FAULT_MINOR;
1990 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
1991 goto out;
1993 entry = pte_to_swp_entry(orig_pte);
1994 again:
1995 page = lookup_swap_cache(entry);
1996 if (!page) {
1997 swapin_readahead(entry, address, vma);
1998 page = read_swap_cache_async(entry, vma, address);
1999 if (!page) {
2000 /*
2001 * Back out if somebody else faulted in this pte
2002 * while we released the pte lock.
2003 */
2004 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2005 if (likely(pte_same(*page_table, orig_pte)))
2006 ret = VM_FAULT_OOM;
2007 goto unlock;
2010 /* Had to read the page from swap area: Major fault */
2011 ret = VM_FAULT_MAJOR;
2012 inc_page_state(pgmajfault);
2013 grab_swap_token();
2016 mark_page_accessed(page);
2017 lock_page(page);
2018 if (!PageSwapCache(page)) {
2019 /* Page migration has occured */
2020 unlock_page(page);
2021 page_cache_release(page);
2022 goto again;
2025 /*
2026 * Back out if somebody else already faulted in this pte.
2027 */
2028 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2029 if (unlikely(!pte_same(*page_table, orig_pte)))
2030 goto out_nomap;
2032 if (unlikely(!PageUptodate(page))) {
2033 ret = VM_FAULT_SIGBUS;
2034 goto out_nomap;
2037 /* The page isn't present yet, go ahead with the fault. */
2039 inc_mm_counter(mm, anon_rss);
2040 pte = mk_pte(page, vma->vm_page_prot);
2041 if (write_access && can_share_swap_page(page)) {
2042 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
2043 write_access = 0;
2046 flush_icache_page(vma, page);
2047 set_pte_at(mm, address, page_table, pte);
2048 page_add_anon_rmap(page, vma, address);
2050 swap_free(entry);
2051 if (vm_swap_full())
2052 remove_exclusive_swap_page(page);
2053 unlock_page(page);
2055 if (write_access) {
2056 if (do_wp_page(mm, vma, address,
2057 page_table, pmd, ptl, pte) == VM_FAULT_OOM)
2058 ret = VM_FAULT_OOM;
2059 goto out;
2062 /* No need to invalidate - it was non-present before */
2063 update_mmu_cache(vma, address, pte);
2064 lazy_mmu_prot_update(pte);
2065 unlock:
2066 pte_unmap_unlock(page_table, ptl);
2067 out:
2068 return ret;
2069 out_nomap:
2070 pte_unmap_unlock(page_table, ptl);
2071 unlock_page(page);
2072 page_cache_release(page);
2073 return ret;
2076 /*
2077 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2078 * but allow concurrent faults), and pte mapped but not yet locked.
2079 * We return with mmap_sem still held, but pte unmapped and unlocked.
2080 */
2081 static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2082 unsigned long address, pte_t *page_table, pmd_t *pmd,
2083 int write_access)
2085 struct page *page;
2086 spinlock_t *ptl;
2087 pte_t entry;
2089 if (write_access) {
2090 /* Allocate our own private page. */
2091 pte_unmap(page_table);
2093 if (unlikely(anon_vma_prepare(vma)))
2094 goto oom;
2095 page = alloc_zeroed_user_highpage(vma, address);
2096 if (!page)
2097 goto oom;
2099 entry = mk_pte(page, vma->vm_page_prot);
2100 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2102 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2103 if (!pte_none(*page_table))
2104 goto release;
2105 inc_mm_counter(mm, anon_rss);
2106 lru_cache_add_active(page);
2107 page_add_new_anon_rmap(page, vma, address);
2108 } else {
2109 /* Map the ZERO_PAGE - vm_page_prot is readonly */
2110 page = ZERO_PAGE(address);
2111 page_cache_get(page);
2112 entry = mk_pte(page, vma->vm_page_prot);
2114 ptl = pte_lockptr(mm, pmd);
2115 spin_lock(ptl);
2116 if (!pte_none(*page_table))
2117 goto release;
2118 inc_mm_counter(mm, file_rss);
2119 page_add_file_rmap(page);
2122 set_pte_at(mm, address, page_table, entry);
2124 /* No need to invalidate - it was non-present before */
2125 update_mmu_cache(vma, address, entry);
2126 lazy_mmu_prot_update(entry);
2127 unlock:
2128 pte_unmap_unlock(page_table, ptl);
2129 return VM_FAULT_MINOR;
2130 release:
2131 page_cache_release(page);
2132 goto unlock;
2133 oom:
2134 return VM_FAULT_OOM;
2137 /*
2138 * do_no_page() tries to create a new page mapping. It aggressively
2139 * tries to share with existing pages, but makes a separate copy if
2140 * the "write_access" parameter is true in order to avoid the next
2141 * page fault.
2143 * As this is called only for pages that do not currently exist, we
2144 * do not need to flush old virtual caches or the TLB.
2146 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2147 * but allow concurrent faults), and pte mapped but not yet locked.
2148 * We return with mmap_sem still held, but pte unmapped and unlocked.
2149 */
2150 static int do_no_page(struct mm_struct *mm, struct vm_area_struct *vma,
2151 unsigned long address, pte_t *page_table, pmd_t *pmd,
2152 int write_access)
2154 spinlock_t *ptl;
2155 struct page *new_page;
2156 struct address_space *mapping = NULL;
2157 pte_t entry;
2158 unsigned int sequence = 0;
2159 int ret = VM_FAULT_MINOR;
2160 int anon = 0;
2162 pte_unmap(page_table);
2163 BUG_ON(vma->vm_flags & VM_PFNMAP);
2165 if (vma->vm_file) {
2166 mapping = vma->vm_file->f_mapping;
2167 sequence = mapping->truncate_count;
2168 smp_rmb(); /* serializes i_size against truncate_count */
2170 retry:
2171 new_page = vma->vm_ops->nopage(vma, address & PAGE_MASK, &ret);
2172 /*
2173 * No smp_rmb is needed here as long as there's a full
2174 * spin_lock/unlock sequence inside the ->nopage callback
2175 * (for the pagecache lookup) that acts as an implicit
2176 * smp_mb() and prevents the i_size read to happen
2177 * after the next truncate_count read.
2178 */
2180 /* no page was available -- either SIGBUS or OOM */
2181 if (new_page == NOPAGE_SIGBUS)
2182 return VM_FAULT_SIGBUS;
2183 if (new_page == NOPAGE_OOM)
2184 return VM_FAULT_OOM;
2186 /*
2187 * Should we do an early C-O-W break?
2188 */
2189 if (write_access && !(vma->vm_flags & VM_SHARED)) {
2190 struct page *page;
2192 if (unlikely(anon_vma_prepare(vma)))
2193 goto oom;
2194 page = alloc_page_vma(GFP_HIGHUSER, vma, address);
2195 if (!page)
2196 goto oom;
2197 copy_user_highpage(page, new_page, address);
2198 page_cache_release(new_page);
2199 new_page = page;
2200 anon = 1;
2203 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2204 /*
2205 * For a file-backed vma, someone could have truncated or otherwise
2206 * invalidated this page. If unmap_mapping_range got called,
2207 * retry getting the page.
2208 */
2209 if (mapping && unlikely(sequence != mapping->truncate_count)) {
2210 pte_unmap_unlock(page_table, ptl);
2211 page_cache_release(new_page);
2212 cond_resched();
2213 sequence = mapping->truncate_count;
2214 smp_rmb();
2215 goto retry;
2218 /*
2219 * This silly early PAGE_DIRTY setting removes a race
2220 * due to the bad i386 page protection. But it's valid
2221 * for other architectures too.
2223 * Note that if write_access is true, we either now have
2224 * an exclusive copy of the page, or this is a shared mapping,
2225 * so we can make it writable and dirty to avoid having to
2226 * handle that later.
2227 */
2228 /* Only go through if we didn't race with anybody else... */
2229 if (pte_none(*page_table)) {
2230 flush_icache_page(vma, new_page);
2231 entry = mk_pte(new_page, vma->vm_page_prot);
2232 if (write_access)
2233 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2234 set_pte_at(mm, address, page_table, entry);
2235 if (anon) {
2236 inc_mm_counter(mm, anon_rss);
2237 lru_cache_add_active(new_page);
2238 page_add_new_anon_rmap(new_page, vma, address);
2239 } else {
2240 inc_mm_counter(mm, file_rss);
2241 page_add_file_rmap(new_page);
2243 } else {
2244 /* One of our sibling threads was faster, back out. */
2245 page_cache_release(new_page);
2246 goto unlock;
2249 /* no need to invalidate: a not-present page shouldn't be cached */
2250 update_mmu_cache(vma, address, entry);
2251 lazy_mmu_prot_update(entry);
2252 unlock:
2253 pte_unmap_unlock(page_table, ptl);
2254 return ret;
2255 oom:
2256 page_cache_release(new_page);
2257 return VM_FAULT_OOM;
2260 /*
2261 * Fault of a previously existing named mapping. Repopulate the pte
2262 * from the encoded file_pte if possible. This enables swappable
2263 * nonlinear vmas.
2265 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2266 * but allow concurrent faults), and pte mapped but not yet locked.
2267 * We return with mmap_sem still held, but pte unmapped and unlocked.
2268 */
2269 static int do_file_page(struct mm_struct *mm, struct vm_area_struct *vma,
2270 unsigned long address, pte_t *page_table, pmd_t *pmd,
2271 int write_access, pte_t orig_pte)
2273 pgoff_t pgoff;
2274 int err;
2276 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
2277 return VM_FAULT_MINOR;
2279 if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) {
2280 /*
2281 * Page table corrupted: show pte and kill process.
2282 */
2283 print_bad_pte(vma, orig_pte, address);
2284 return VM_FAULT_OOM;
2286 /* We can then assume vm->vm_ops && vma->vm_ops->populate */
2288 pgoff = pte_to_pgoff(orig_pte);
2289 err = vma->vm_ops->populate(vma, address & PAGE_MASK, PAGE_SIZE,
2290 vma->vm_page_prot, pgoff, 0);
2291 if (err == -ENOMEM)
2292 return VM_FAULT_OOM;
2293 if (err)
2294 return VM_FAULT_SIGBUS;
2295 return VM_FAULT_MAJOR;
2298 /*
2299 * These routines also need to handle stuff like marking pages dirty
2300 * and/or accessed for architectures that don't do it in hardware (most
2301 * RISC architectures). The early dirtying is also good on the i386.
2303 * There is also a hook called "update_mmu_cache()" that architectures
2304 * with external mmu caches can use to update those (ie the Sparc or
2305 * PowerPC hashed page tables that act as extended TLBs).
2307 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2308 * but allow concurrent faults), and pte mapped but not yet locked.
2309 * We return with mmap_sem still held, but pte unmapped and unlocked.
2310 */
2311 static inline int handle_pte_fault(struct mm_struct *mm,
2312 struct vm_area_struct *vma, unsigned long address,
2313 pte_t *pte, pmd_t *pmd, int write_access)
2315 pte_t entry;
2316 pte_t old_entry;
2317 spinlock_t *ptl;
2319 old_entry = entry = *pte;
2320 if (!pte_present(entry)) {
2321 if (pte_none(entry)) {
2322 if (!vma->vm_ops || !vma->vm_ops->nopage)
2323 return do_anonymous_page(mm, vma, address,
2324 pte, pmd, write_access);
2325 return do_no_page(mm, vma, address,
2326 pte, pmd, write_access);
2328 if (pte_file(entry))
2329 return do_file_page(mm, vma, address,
2330 pte, pmd, write_access, entry);
2331 return do_swap_page(mm, vma, address,
2332 pte, pmd, write_access, entry);
2335 ptl = pte_lockptr(mm, pmd);
2336 spin_lock(ptl);
2337 if (unlikely(!pte_same(*pte, entry)))
2338 goto unlock;
2339 if (write_access) {
2340 if (!pte_write(entry))
2341 return do_wp_page(mm, vma, address,
2342 pte, pmd, ptl, entry);
2343 entry = pte_mkdirty(entry);
2345 entry = pte_mkyoung(entry);
2346 if (!pte_same(old_entry, entry)) {
2347 ptep_set_access_flags(vma, address, pte, entry, write_access);
2348 update_mmu_cache(vma, address, entry);
2349 lazy_mmu_prot_update(entry);
2350 } else {
2351 /*
2352 * This is needed only for protection faults but the arch code
2353 * is not yet telling us if this is a protection fault or not.
2354 * This still avoids useless tlb flushes for .text page faults
2355 * with threads.
2356 */
2357 if (write_access)
2358 flush_tlb_page(vma, address);
2360 unlock:
2361 pte_unmap_unlock(pte, ptl);
2362 return VM_FAULT_MINOR;
2365 /*
2366 * By the time we get here, we already hold the mm semaphore
2367 */
2368 int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2369 unsigned long address, int write_access)
2371 pgd_t *pgd;
2372 pud_t *pud;
2373 pmd_t *pmd;
2374 pte_t *pte;
2376 __set_current_state(TASK_RUNNING);
2378 inc_page_state(pgfault);
2380 if (unlikely(is_vm_hugetlb_page(vma)))
2381 return hugetlb_fault(mm, vma, address, write_access);
2383 pgd = pgd_offset(mm, address);
2384 pud = pud_alloc(mm, pgd, address);
2385 if (!pud)
2386 return VM_FAULT_OOM;
2387 pmd = pmd_alloc(mm, pud, address);
2388 if (!pmd)
2389 return VM_FAULT_OOM;
2390 pte = pte_alloc_map(mm, pmd, address);
2391 if (!pte)
2392 return VM_FAULT_OOM;
2394 return handle_pte_fault(mm, vma, address, pte, pmd, write_access);
2397 EXPORT_SYMBOL_GPL(__handle_mm_fault);
2399 #ifndef __PAGETABLE_PUD_FOLDED
2400 /*
2401 * Allocate page upper directory.
2402 * We've already handled the fast-path in-line.
2403 */
2404 int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
2406 pud_t *new = pud_alloc_one(mm, address);
2407 if (!new)
2408 return -ENOMEM;
2410 spin_lock(&mm->page_table_lock);
2411 if (pgd_present(*pgd)) /* Another has populated it */
2412 pud_free(new);
2413 else
2414 pgd_populate(mm, pgd, new);
2415 spin_unlock(&mm->page_table_lock);
2416 return 0;
2418 #else
2419 /* Workaround for gcc 2.96 */
2420 int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
2422 return 0;
2424 #endif /* __PAGETABLE_PUD_FOLDED */
2426 #ifndef __PAGETABLE_PMD_FOLDED
2427 /*
2428 * Allocate page middle directory.
2429 * We've already handled the fast-path in-line.
2430 */
2431 int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
2433 pmd_t *new = pmd_alloc_one(mm, address);
2434 if (!new)
2435 return -ENOMEM;
2437 spin_lock(&mm->page_table_lock);
2438 #ifndef __ARCH_HAS_4LEVEL_HACK
2439 if (pud_present(*pud)) /* Another has populated it */
2440 pmd_free(new);
2441 else
2442 pud_populate(mm, pud, new);
2443 #else
2444 if (pgd_present(*pud)) /* Another has populated it */
2445 pmd_free(new);
2446 else
2447 pgd_populate(mm, pud, new);
2448 #endif /* __ARCH_HAS_4LEVEL_HACK */
2449 spin_unlock(&mm->page_table_lock);
2450 return 0;
2452 #else
2453 /* Workaround for gcc 2.96 */
2454 int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
2456 return 0;
2458 #endif /* __PAGETABLE_PMD_FOLDED */
2460 int make_pages_present(unsigned long addr, unsigned long end)
2462 int ret, len, write;
2463 struct vm_area_struct * vma;
2465 vma = find_vma(current->mm, addr);
2466 if (!vma)
2467 return -1;
2468 write = (vma->vm_flags & VM_WRITE) != 0;
2469 if (addr >= end)
2470 BUG();
2471 if (end > vma->vm_end)
2472 BUG();
2473 len = (end+PAGE_SIZE-1)/PAGE_SIZE-addr/PAGE_SIZE;
2474 ret = get_user_pages(current, current->mm, addr,
2475 len, write, 0, NULL, NULL);
2476 if (ret < 0)
2477 return ret;
2478 return ret == len ? 0 : -1;
2481 /*
2482 * Map a vmalloc()-space virtual address to the physical page.
2483 */
2484 struct page * vmalloc_to_page(void * vmalloc_addr)
2486 unsigned long addr = (unsigned long) vmalloc_addr;
2487 struct page *page = NULL;
2488 pgd_t *pgd = pgd_offset_k(addr);
2489 pud_t *pud;
2490 pmd_t *pmd;
2491 pte_t *ptep, pte;
2493 if (!pgd_none(*pgd)) {
2494 pud = pud_offset(pgd, addr);
2495 if (!pud_none(*pud)) {
2496 pmd = pmd_offset(pud, addr);
2497 if (!pmd_none(*pmd)) {
2498 ptep = pte_offset_map(pmd, addr);
2499 pte = *ptep;
2500 if (pte_present(pte))
2501 page = pte_page(pte);
2502 pte_unmap(ptep);
2506 return page;
2509 EXPORT_SYMBOL(vmalloc_to_page);
2511 /*
2512 * Map a vmalloc()-space virtual address to the physical page frame number.
2513 */
2514 unsigned long vmalloc_to_pfn(void * vmalloc_addr)
2516 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
2519 EXPORT_SYMBOL(vmalloc_to_pfn);
2521 #if !defined(__HAVE_ARCH_GATE_AREA)
2523 #if defined(AT_SYSINFO_EHDR)
2524 static struct vm_area_struct gate_vma;
2526 static int __init gate_vma_init(void)
2528 gate_vma.vm_mm = NULL;
2529 gate_vma.vm_start = FIXADDR_USER_START;
2530 gate_vma.vm_end = FIXADDR_USER_END;
2531 gate_vma.vm_page_prot = PAGE_READONLY;
2532 gate_vma.vm_flags = 0;
2533 return 0;
2535 __initcall(gate_vma_init);
2536 #endif
2538 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
2540 #ifdef AT_SYSINFO_EHDR
2541 return &gate_vma;
2542 #else
2543 return NULL;
2544 #endif
2547 int in_gate_area_no_task(unsigned long addr)
2549 #ifdef AT_SYSINFO_EHDR
2550 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
2551 return 1;
2552 #endif
2553 return 0;
2556 #endif /* __HAVE_ARCH_GATE_AREA */