4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
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
13 * Ok, demand-loading was easy, shared pages a little bit tricker. Shared
14 * pages started 02.12.91, seems to work. - Linus.
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
20 * Also corrected some "invalidate()"s - I wasn't doing enough of them.
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.
32 * 05.04.94 - Multi-page memory management added for v1.1.
33 * Idea by Alex Bligh (alex@cconcepts.co.uk)
35 * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG
36 * (Gerhard.Wichert@pdb.siemens.de)
38 * Aug/Sep 2004 Changed to four level page tables (Andi Kleen)
41 #include <linux/kernel_stat.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>
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
;
66 EXPORT_SYMBOL(max_mapnr
);
67 EXPORT_SYMBOL(mem_map
);
70 unsigned long num_physpages
;
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
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
)
89 randomize_va_space
= 0;
92 __setup("norandmaps", disable_randmaps
);
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.
101 void pgd_clear_bad(pgd_t
*pgd
)
107 void pud_clear_bad(pud_t
*pud
)
113 void pmd_clear_bad(pmd_t
*pmd
)
120 * Note: this doesn't free the actual pages themselves. That
121 * has been handled earlier when unmapping all the memory regions.
123 static void free_pte_range(struct mmu_gather
*tlb
, pmd_t
*pmd
)
125 struct page
*page
= pmd_page(*pmd
);
127 pte_lock_deinit(page
);
128 pte_free_tlb(tlb
, page
);
129 dec_page_state(nr_page_table_pages
);
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
)
142 pmd
= pmd_offset(pud
, addr
);
144 next
= pmd_addr_end(addr
, end
);
145 if (pmd_none_or_clear_bad(pmd
))
147 free_pte_range(tlb
, pmd
);
148 } while (pmd
++, addr
= next
, addr
!= end
);
158 if (end
- 1 > ceiling
- 1)
161 pmd
= pmd_offset(pud
, start
);
163 pmd_free_tlb(tlb
, pmd
);
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
)
175 pud
= pud_offset(pgd
, addr
);
177 next
= pud_addr_end(addr
, end
);
178 if (pud_none_or_clear_bad(pud
))
180 free_pmd_range(tlb
, pud
, addr
, next
, floor
, ceiling
);
181 } while (pud
++, addr
= next
, addr
!= end
);
187 ceiling
&= PGDIR_MASK
;
191 if (end
- 1 > ceiling
- 1)
194 pud
= pud_offset(pgd
, start
);
196 pud_free_tlb(tlb
, pud
);
200 * This function frees user-level page tables of a process.
202 * Must be called with pagetable lock held.
204 void free_pgd_range(struct mmu_gather
**tlb
,
205 unsigned long addr
, unsigned long end
,
206 unsigned long floor
, unsigned long ceiling
)
213 * The next few lines have given us lots of grief...
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.
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).
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.
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.
249 if (end
- 1 > ceiling
- 1)
255 pgd
= pgd_offset((*tlb
)->mm
, addr
);
257 next
= pgd_addr_end(addr
, end
);
258 if (pgd_none_or_clear_bad(pgd
))
260 free_pud_range(*tlb
, pgd
, addr
, next
, floor
, ceiling
);
261 } while (pgd
++, addr
= next
, addr
!= end
);
264 flush_tlb_pgtables((*tlb
)->mm
, start
, end
);
267 void free_pgtables(struct mmu_gather
**tlb
, struct vm_area_struct
*vma
,
268 unsigned long floor
, unsigned long ceiling
)
271 struct vm_area_struct
*next
= vma
->vm_next
;
272 unsigned long addr
= vma
->vm_start
;
275 * Hide vma from rmap and vmtruncate before freeing pgtables
277 anon_vma_unlink(vma
);
278 unlink_file_vma(vma
);
280 if (is_vm_hugetlb_page(vma
)) {
281 hugetlb_free_pgd_range(tlb
, addr
, vma
->vm_end
,
282 floor
, next
? next
->vm_start
: ceiling
);
285 * Optimization: gather nearby vmas into one call down
287 while (next
&& next
->vm_start
<= vma
->vm_end
+ PMD_SIZE
288 && !is_vm_hugetlb_page(next
)) {
291 anon_vma_unlink(vma
);
292 unlink_file_vma(vma
);
294 free_pgd_range(tlb
, addr
, vma
->vm_end
,
295 floor
, next
? next
->vm_start
: ceiling
);
301 int __pte_alloc(struct mm_struct
*mm
, pmd_t
*pmd
, unsigned long address
)
303 struct page
*new = pte_alloc_one(mm
, address
);
308 spin_lock(&mm
->page_table_lock
);
309 if (pmd_present(*pmd
)) { /* Another has populated it */
310 pte_lock_deinit(new);
314 inc_page_state(nr_page_table_pages
);
315 pmd_populate(mm
, pmd
, new);
317 spin_unlock(&mm
->page_table_lock
);
321 int __pte_alloc_kernel(pmd_t
*pmd
, unsigned long address
)
323 pte_t
*new = pte_alloc_one_kernel(&init_mm
, address
);
327 spin_lock(&init_mm
.page_table_lock
);
328 if (pmd_present(*pmd
)) /* Another has populated it */
329 pte_free_kernel(new);
331 pmd_populate_kernel(&init_mm
, pmd
, new);
332 spin_unlock(&init_mm
.page_table_lock
);
336 static inline void add_mm_rss(struct mm_struct
*mm
, int file_rss
, int anon_rss
)
339 add_mm_counter(mm
, file_rss
, file_rss
);
341 add_mm_counter(mm
, anon_rss
, anon_rss
);
345 * This function is called to print an error when a bad pte
346 * is found. For example, we might have a PFN-mapped pte in
347 * a region that doesn't allow it.
349 * The calling function must still handle the error.
351 void print_bad_pte(struct vm_area_struct
*vma
, pte_t pte
, unsigned long vaddr
)
353 printk(KERN_ERR
"Bad pte = %08llx, process = %s, "
354 "vm_flags = %lx, vaddr = %lx\n",
355 (long long)pte_val(pte
),
356 (vma
->vm_mm
== current
->mm
? current
->comm
: "???"),
357 vma
->vm_flags
, vaddr
);
361 static inline int is_cow_mapping(unsigned int flags
)
363 return (flags
& (VM_SHARED
| VM_MAYWRITE
)) == VM_MAYWRITE
;
367 * This function gets the "struct page" associated with a pte.
369 * NOTE! Some mappings do not have "struct pages". A raw PFN mapping
370 * will have each page table entry just pointing to a raw page frame
371 * number, and as far as the VM layer is concerned, those do not have
372 * pages associated with them - even if the PFN might point to memory
373 * that otherwise is perfectly fine and has a "struct page".
375 * The way we recognize those mappings is through the rules set up
376 * by "remap_pfn_range()": the vma will have the VM_PFNMAP bit set,
377 * and the vm_pgoff will point to the first PFN mapped: thus every
378 * page that is a raw mapping will always honor the rule
380 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
382 * and if that isn't true, the page has been COW'ed (in which case it
383 * _does_ have a "struct page" associated with it even if it is in a
386 struct page
*vm_normal_page(struct vm_area_struct
*vma
, unsigned long addr
, pte_t pte
)
388 unsigned long pfn
= pte_pfn(pte
);
390 if (unlikely(vma
->vm_flags
& VM_PFNMAP
)) {
391 unsigned long off
= (addr
- vma
->vm_start
) >> PAGE_SHIFT
;
392 if (pfn
== vma
->vm_pgoff
+ off
)
394 if (!is_cow_mapping(vma
->vm_flags
))
399 * Add some anal sanity checks for now. Eventually,
400 * we should just do "return pfn_to_page(pfn)", but
401 * in the meantime we check that we get a valid pfn,
402 * and that the resulting page looks ok.
404 if (unlikely(!pfn_valid(pfn
))) {
405 print_bad_pte(vma
, pte
, addr
);
410 * NOTE! We still have PageReserved() pages in the page
413 * The PAGE_ZERO() pages and various VDSO mappings can
414 * cause them to exist.
416 return pfn_to_page(pfn
);
420 * copy one vm_area from one task to the other. Assumes the page tables
421 * already present in the new task to be cleared in the whole range
422 * covered by this vma.
426 copy_one_pte(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
427 pte_t
*dst_pte
, pte_t
*src_pte
, struct vm_area_struct
*vma
,
428 unsigned long addr
, int *rss
)
430 unsigned long vm_flags
= vma
->vm_flags
;
431 pte_t pte
= *src_pte
;
434 /* pte contains position in swap or file, so copy. */
435 if (unlikely(!pte_present(pte
))) {
436 if (!pte_file(pte
)) {
437 swap_duplicate(pte_to_swp_entry(pte
));
438 /* make sure dst_mm is on swapoff's mmlist. */
439 if (unlikely(list_empty(&dst_mm
->mmlist
))) {
440 spin_lock(&mmlist_lock
);
441 if (list_empty(&dst_mm
->mmlist
))
442 list_add(&dst_mm
->mmlist
,
444 spin_unlock(&mmlist_lock
);
451 * If it's a COW mapping, write protect it both
452 * in the parent and the child
454 if (is_cow_mapping(vm_flags
)) {
455 ptep_set_wrprotect(src_mm
, addr
, src_pte
);
460 * If it's a shared mapping, mark it clean in
463 if (vm_flags
& VM_SHARED
)
464 pte
= pte_mkclean(pte
);
465 pte
= pte_mkold(pte
);
467 page
= vm_normal_page(vma
, addr
, pte
);
471 rss
[!!PageAnon(page
)]++;
475 set_pte_at(dst_mm
, addr
, dst_pte
, pte
);
478 static int copy_pte_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
479 pmd_t
*dst_pmd
, pmd_t
*src_pmd
, struct vm_area_struct
*vma
,
480 unsigned long addr
, unsigned long end
)
482 pte_t
*src_pte
, *dst_pte
;
483 spinlock_t
*src_ptl
, *dst_ptl
;
489 dst_pte
= pte_alloc_map_lock(dst_mm
, dst_pmd
, addr
, &dst_ptl
);
492 src_pte
= pte_offset_map_nested(src_pmd
, addr
);
493 src_ptl
= pte_lockptr(src_mm
, src_pmd
);
498 * We are holding two locks at this point - either of them
499 * could generate latencies in another task on another CPU.
501 if (progress
>= 32) {
503 if (need_resched() ||
504 need_lockbreak(src_ptl
) ||
505 need_lockbreak(dst_ptl
))
508 if (pte_none(*src_pte
)) {
512 copy_one_pte(dst_mm
, src_mm
, dst_pte
, src_pte
, vma
, addr
, rss
);
514 } while (dst_pte
++, src_pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
516 spin_unlock(src_ptl
);
517 pte_unmap_nested(src_pte
- 1);
518 add_mm_rss(dst_mm
, rss
[0], rss
[1]);
519 pte_unmap_unlock(dst_pte
- 1, dst_ptl
);
526 static inline int copy_pmd_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
527 pud_t
*dst_pud
, pud_t
*src_pud
, struct vm_area_struct
*vma
,
528 unsigned long addr
, unsigned long end
)
530 pmd_t
*src_pmd
, *dst_pmd
;
533 dst_pmd
= pmd_alloc(dst_mm
, dst_pud
, addr
);
536 src_pmd
= pmd_offset(src_pud
, addr
);
538 next
= pmd_addr_end(addr
, end
);
539 if (pmd_none_or_clear_bad(src_pmd
))
541 if (copy_pte_range(dst_mm
, src_mm
, dst_pmd
, src_pmd
,
544 } while (dst_pmd
++, src_pmd
++, addr
= next
, addr
!= end
);
548 static inline int copy_pud_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
549 pgd_t
*dst_pgd
, pgd_t
*src_pgd
, struct vm_area_struct
*vma
,
550 unsigned long addr
, unsigned long end
)
552 pud_t
*src_pud
, *dst_pud
;
555 dst_pud
= pud_alloc(dst_mm
, dst_pgd
, addr
);
558 src_pud
= pud_offset(src_pgd
, addr
);
560 next
= pud_addr_end(addr
, end
);
561 if (pud_none_or_clear_bad(src_pud
))
563 if (copy_pmd_range(dst_mm
, src_mm
, dst_pud
, src_pud
,
566 } while (dst_pud
++, src_pud
++, addr
= next
, addr
!= end
);
570 int copy_page_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
571 struct vm_area_struct
*vma
)
573 pgd_t
*src_pgd
, *dst_pgd
;
575 unsigned long addr
= vma
->vm_start
;
576 unsigned long end
= vma
->vm_end
;
579 * Don't copy ptes where a page fault will fill them correctly.
580 * Fork becomes much lighter when there are big shared or private
581 * readonly mappings. The tradeoff is that copy_page_range is more
582 * efficient than faulting.
584 if (!(vma
->vm_flags
& (VM_HUGETLB
|VM_NONLINEAR
|VM_PFNMAP
|VM_INSERTPAGE
))) {
589 if (is_vm_hugetlb_page(vma
))
590 return copy_hugetlb_page_range(dst_mm
, src_mm
, vma
);
592 dst_pgd
= pgd_offset(dst_mm
, addr
);
593 src_pgd
= pgd_offset(src_mm
, addr
);
595 next
= pgd_addr_end(addr
, end
);
596 if (pgd_none_or_clear_bad(src_pgd
))
598 if (copy_pud_range(dst_mm
, src_mm
, dst_pgd
, src_pgd
,
601 } while (dst_pgd
++, src_pgd
++, addr
= next
, addr
!= end
);
605 static unsigned long zap_pte_range(struct mmu_gather
*tlb
,
606 struct vm_area_struct
*vma
, pmd_t
*pmd
,
607 unsigned long addr
, unsigned long end
,
608 long *zap_work
, struct zap_details
*details
)
610 struct mm_struct
*mm
= tlb
->mm
;
616 pte
= pte_offset_map_lock(mm
, pmd
, addr
, &ptl
);
619 if (pte_none(ptent
)) {
624 (*zap_work
) -= PAGE_SIZE
;
626 if (pte_present(ptent
)) {
629 page
= vm_normal_page(vma
, addr
, ptent
);
630 if (unlikely(details
) && page
) {
632 * unmap_shared_mapping_pages() wants to
633 * invalidate cache without truncating:
634 * unmap shared but keep private pages.
636 if (details
->check_mapping
&&
637 details
->check_mapping
!= page
->mapping
)
640 * Each page->index must be checked when
641 * invalidating or truncating nonlinear.
643 if (details
->nonlinear_vma
&&
644 (page
->index
< details
->first_index
||
645 page
->index
> details
->last_index
))
648 ptent
= ptep_get_and_clear_full(mm
, addr
, pte
,
650 tlb_remove_tlb_entry(tlb
, pte
, addr
);
653 if (unlikely(details
) && details
->nonlinear_vma
654 && linear_page_index(details
->nonlinear_vma
,
655 addr
) != page
->index
)
656 set_pte_at(mm
, addr
, pte
,
657 pgoff_to_pte(page
->index
));
661 if (pte_dirty(ptent
))
662 set_page_dirty(page
);
663 if (pte_young(ptent
))
664 mark_page_accessed(page
);
667 page_remove_rmap(page
);
668 tlb_remove_page(tlb
, page
);
672 * If details->check_mapping, we leave swap entries;
673 * if details->nonlinear_vma, we leave file entries.
675 if (unlikely(details
))
677 if (!pte_file(ptent
))
678 free_swap_and_cache(pte_to_swp_entry(ptent
));
679 pte_clear_full(mm
, addr
, pte
, tlb
->fullmm
);
680 } while (pte
++, addr
+= PAGE_SIZE
, (addr
!= end
&& *zap_work
> 0));
682 add_mm_rss(mm
, file_rss
, anon_rss
);
683 pte_unmap_unlock(pte
- 1, ptl
);
688 static inline unsigned long zap_pmd_range(struct mmu_gather
*tlb
,
689 struct vm_area_struct
*vma
, pud_t
*pud
,
690 unsigned long addr
, unsigned long end
,
691 long *zap_work
, struct zap_details
*details
)
696 pmd
= pmd_offset(pud
, addr
);
698 next
= pmd_addr_end(addr
, end
);
699 if (pmd_none_or_clear_bad(pmd
)) {
703 next
= zap_pte_range(tlb
, vma
, pmd
, addr
, next
,
705 } while (pmd
++, addr
= next
, (addr
!= end
&& *zap_work
> 0));
710 static inline unsigned long zap_pud_range(struct mmu_gather
*tlb
,
711 struct vm_area_struct
*vma
, pgd_t
*pgd
,
712 unsigned long addr
, unsigned long end
,
713 long *zap_work
, struct zap_details
*details
)
718 pud
= pud_offset(pgd
, addr
);
720 next
= pud_addr_end(addr
, end
);
721 if (pud_none_or_clear_bad(pud
)) {
725 next
= zap_pmd_range(tlb
, vma
, pud
, addr
, next
,
727 } while (pud
++, addr
= next
, (addr
!= end
&& *zap_work
> 0));
732 static unsigned long unmap_page_range(struct mmu_gather
*tlb
,
733 struct vm_area_struct
*vma
,
734 unsigned long addr
, unsigned long end
,
735 long *zap_work
, struct zap_details
*details
)
740 if (details
&& !details
->check_mapping
&& !details
->nonlinear_vma
)
744 tlb_start_vma(tlb
, vma
);
745 pgd
= pgd_offset(vma
->vm_mm
, addr
);
747 next
= pgd_addr_end(addr
, end
);
748 if (pgd_none_or_clear_bad(pgd
)) {
752 next
= zap_pud_range(tlb
, vma
, pgd
, addr
, next
,
754 } while (pgd
++, addr
= next
, (addr
!= end
&& *zap_work
> 0));
755 tlb_end_vma(tlb
, vma
);
760 #ifdef CONFIG_PREEMPT
761 # define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
763 /* No preempt: go for improved straight-line efficiency */
764 # define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
768 * unmap_vmas - unmap a range of memory covered by a list of vma's
769 * @tlbp: address of the caller's struct mmu_gather
770 * @vma: the starting vma
771 * @start_addr: virtual address at which to start unmapping
772 * @end_addr: virtual address at which to end unmapping
773 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
774 * @details: details of nonlinear truncation or shared cache invalidation
776 * Returns the end address of the unmapping (restart addr if interrupted).
778 * Unmap all pages in the vma list.
780 * We aim to not hold locks for too long (for scheduling latency reasons).
781 * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
782 * return the ending mmu_gather to the caller.
784 * Only addresses between `start' and `end' will be unmapped.
786 * The VMA list must be sorted in ascending virtual address order.
788 * unmap_vmas() assumes that the caller will flush the whole unmapped address
789 * range after unmap_vmas() returns. So the only responsibility here is to
790 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
791 * drops the lock and schedules.
793 unsigned long unmap_vmas(struct mmu_gather
**tlbp
,
794 struct vm_area_struct
*vma
, unsigned long start_addr
,
795 unsigned long end_addr
, unsigned long *nr_accounted
,
796 struct zap_details
*details
)
798 long zap_work
= ZAP_BLOCK_SIZE
;
799 unsigned long tlb_start
= 0; /* For tlb_finish_mmu */
800 int tlb_start_valid
= 0;
801 unsigned long start
= start_addr
;
802 spinlock_t
*i_mmap_lock
= details
? details
->i_mmap_lock
: NULL
;
803 int fullmm
= (*tlbp
)->fullmm
;
805 for ( ; vma
&& vma
->vm_start
< end_addr
; vma
= vma
->vm_next
) {
808 start
= max(vma
->vm_start
, start_addr
);
809 if (start
>= vma
->vm_end
)
811 end
= min(vma
->vm_end
, end_addr
);
812 if (end
<= vma
->vm_start
)
815 if (vma
->vm_flags
& VM_ACCOUNT
)
816 *nr_accounted
+= (end
- start
) >> PAGE_SHIFT
;
818 while (start
!= end
) {
819 if (!tlb_start_valid
) {
824 if (unlikely(is_vm_hugetlb_page(vma
))) {
825 unmap_hugepage_range(vma
, start
, end
);
826 zap_work
-= (end
- start
) /
827 (HPAGE_SIZE
/ PAGE_SIZE
);
830 start
= unmap_page_range(*tlbp
, vma
,
831 start
, end
, &zap_work
, details
);
834 BUG_ON(start
!= end
);
838 tlb_finish_mmu(*tlbp
, tlb_start
, start
);
840 if (need_resched() ||
841 (i_mmap_lock
&& need_lockbreak(i_mmap_lock
))) {
849 *tlbp
= tlb_gather_mmu(vma
->vm_mm
, fullmm
);
851 zap_work
= ZAP_BLOCK_SIZE
;
855 return start
; /* which is now the end (or restart) address */
859 * zap_page_range - remove user pages in a given range
860 * @vma: vm_area_struct holding the applicable pages
861 * @address: starting address of pages to zap
862 * @size: number of bytes to zap
863 * @details: details of nonlinear truncation or shared cache invalidation
865 unsigned long zap_page_range(struct vm_area_struct
*vma
, unsigned long address
,
866 unsigned long size
, struct zap_details
*details
)
868 struct mm_struct
*mm
= vma
->vm_mm
;
869 struct mmu_gather
*tlb
;
870 unsigned long end
= address
+ size
;
871 unsigned long nr_accounted
= 0;
874 tlb
= tlb_gather_mmu(mm
, 0);
875 update_hiwater_rss(mm
);
876 end
= unmap_vmas(&tlb
, vma
, address
, end
, &nr_accounted
, details
);
878 tlb_finish_mmu(tlb
, address
, end
);
883 * Do a quick page-table lookup for a single page.
885 struct page
*follow_page(struct vm_area_struct
*vma
, unsigned long address
,
894 struct mm_struct
*mm
= vma
->vm_mm
;
896 page
= follow_huge_addr(mm
, address
, flags
& FOLL_WRITE
);
898 BUG_ON(flags
& FOLL_GET
);
903 pgd
= pgd_offset(mm
, address
);
904 if (pgd_none(*pgd
) || unlikely(pgd_bad(*pgd
)))
907 pud
= pud_offset(pgd
, address
);
908 if (pud_none(*pud
) || unlikely(pud_bad(*pud
)))
911 pmd
= pmd_offset(pud
, address
);
912 if (pmd_none(*pmd
) || unlikely(pmd_bad(*pmd
)))
915 if (pmd_huge(*pmd
)) {
916 BUG_ON(flags
& FOLL_GET
);
917 page
= follow_huge_pmd(mm
, address
, pmd
, flags
& FOLL_WRITE
);
921 ptep
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
926 if (!pte_present(pte
))
928 if ((flags
& FOLL_WRITE
) && !pte_write(pte
))
930 page
= vm_normal_page(vma
, address
, pte
);
934 if (flags
& FOLL_GET
)
936 if (flags
& FOLL_TOUCH
) {
937 if ((flags
& FOLL_WRITE
) &&
938 !pte_dirty(pte
) && !PageDirty(page
))
939 set_page_dirty(page
);
940 mark_page_accessed(page
);
943 pte_unmap_unlock(ptep
, ptl
);
949 * When core dumping an enormous anonymous area that nobody
950 * has touched so far, we don't want to allocate page tables.
952 if (flags
& FOLL_ANON
) {
953 page
= ZERO_PAGE(address
);
954 if (flags
& FOLL_GET
)
956 BUG_ON(flags
& FOLL_WRITE
);
961 int get_user_pages(struct task_struct
*tsk
, struct mm_struct
*mm
,
962 unsigned long start
, int len
, int write
, int force
,
963 struct page
**pages
, struct vm_area_struct
**vmas
)
966 unsigned int vm_flags
;
969 * Require read or write permissions.
970 * If 'force' is set, we only require the "MAY" flags.
972 vm_flags
= write
? (VM_WRITE
| VM_MAYWRITE
) : (VM_READ
| VM_MAYREAD
);
973 vm_flags
&= force
? (VM_MAYREAD
| VM_MAYWRITE
) : (VM_READ
| VM_WRITE
);
977 struct vm_area_struct
*vma
;
978 unsigned int foll_flags
;
980 vma
= find_extend_vma(mm
, start
);
981 if (!vma
&& in_gate_area(tsk
, start
)) {
982 unsigned long pg
= start
& PAGE_MASK
;
983 struct vm_area_struct
*gate_vma
= get_gate_vma(tsk
);
988 if (write
) /* user gate pages are read-only */
989 return i
? : -EFAULT
;
991 pgd
= pgd_offset_k(pg
);
993 pgd
= pgd_offset_gate(mm
, pg
);
994 BUG_ON(pgd_none(*pgd
));
995 pud
= pud_offset(pgd
, pg
);
996 BUG_ON(pud_none(*pud
));
997 pmd
= pmd_offset(pud
, pg
);
999 return i
? : -EFAULT
;
1000 pte
= pte_offset_map(pmd
, pg
);
1001 if (pte_none(*pte
)) {
1003 return i
? : -EFAULT
;
1006 struct page
*page
= vm_normal_page(gate_vma
, start
, *pte
);
1020 if (!vma
|| (vma
->vm_flags
& (VM_IO
| VM_PFNMAP
))
1021 || !(vm_flags
& vma
->vm_flags
))
1022 return i
? : -EFAULT
;
1024 if (is_vm_hugetlb_page(vma
)) {
1025 i
= follow_hugetlb_page(mm
, vma
, pages
, vmas
,
1030 foll_flags
= FOLL_TOUCH
;
1032 foll_flags
|= FOLL_GET
;
1033 if (!write
&& !(vma
->vm_flags
& VM_LOCKED
) &&
1034 (!vma
->vm_ops
|| !vma
->vm_ops
->nopage
))
1035 foll_flags
|= FOLL_ANON
;
1041 foll_flags
|= FOLL_WRITE
;
1044 while (!(page
= follow_page(vma
, start
, foll_flags
))) {
1046 ret
= __handle_mm_fault(mm
, vma
, start
,
1047 foll_flags
& FOLL_WRITE
);
1049 * The VM_FAULT_WRITE bit tells us that do_wp_page has
1050 * broken COW when necessary, even if maybe_mkwrite
1051 * decided not to set pte_write. We can thus safely do
1052 * subsequent page lookups as if they were reads.
1054 if (ret
& VM_FAULT_WRITE
)
1055 foll_flags
&= ~FOLL_WRITE
;
1057 switch (ret
& ~VM_FAULT_WRITE
) {
1058 case VM_FAULT_MINOR
:
1061 case VM_FAULT_MAJOR
:
1064 case VM_FAULT_SIGBUS
:
1065 return i
? i
: -EFAULT
;
1067 return i
? i
: -ENOMEM
;
1074 flush_dcache_page(page
);
1081 } while (len
&& start
< vma
->vm_end
);
1085 EXPORT_SYMBOL(get_user_pages
);
1087 static int zeromap_pte_range(struct mm_struct
*mm
, pmd_t
*pmd
,
1088 unsigned long addr
, unsigned long end
, pgprot_t prot
)
1093 pte
= pte_alloc_map_lock(mm
, pmd
, addr
, &ptl
);
1097 struct page
*page
= ZERO_PAGE(addr
);
1098 pte_t zero_pte
= pte_wrprotect(mk_pte(page
, prot
));
1099 page_cache_get(page
);
1100 page_add_file_rmap(page
);
1101 inc_mm_counter(mm
, file_rss
);
1102 BUG_ON(!pte_none(*pte
));
1103 set_pte_at(mm
, addr
, pte
, zero_pte
);
1104 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1105 pte_unmap_unlock(pte
- 1, ptl
);
1109 static inline int zeromap_pmd_range(struct mm_struct
*mm
, pud_t
*pud
,
1110 unsigned long addr
, unsigned long end
, pgprot_t prot
)
1115 pmd
= pmd_alloc(mm
, pud
, addr
);
1119 next
= pmd_addr_end(addr
, end
);
1120 if (zeromap_pte_range(mm
, pmd
, addr
, next
, prot
))
1122 } while (pmd
++, addr
= next
, addr
!= end
);
1126 static inline int zeromap_pud_range(struct mm_struct
*mm
, pgd_t
*pgd
,
1127 unsigned long addr
, unsigned long end
, pgprot_t prot
)
1132 pud
= pud_alloc(mm
, pgd
, addr
);
1136 next
= pud_addr_end(addr
, end
);
1137 if (zeromap_pmd_range(mm
, pud
, addr
, next
, prot
))
1139 } while (pud
++, addr
= next
, addr
!= end
);
1143 int zeromap_page_range(struct vm_area_struct
*vma
,
1144 unsigned long addr
, unsigned long size
, pgprot_t prot
)
1148 unsigned long end
= addr
+ size
;
1149 struct mm_struct
*mm
= vma
->vm_mm
;
1152 BUG_ON(addr
>= end
);
1153 pgd
= pgd_offset(mm
, addr
);
1154 flush_cache_range(vma
, addr
, end
);
1156 next
= pgd_addr_end(addr
, end
);
1157 err
= zeromap_pud_range(mm
, pgd
, addr
, next
, prot
);
1160 } while (pgd
++, addr
= next
, addr
!= end
);
1164 pte_t
* fastcall
get_locked_pte(struct mm_struct
*mm
, unsigned long addr
, spinlock_t
**ptl
)
1166 pgd_t
* pgd
= pgd_offset(mm
, addr
);
1167 pud_t
* pud
= pud_alloc(mm
, pgd
, addr
);
1169 pmd_t
* pmd
= pmd_alloc(mm
, pud
, addr
);
1171 return pte_alloc_map_lock(mm
, pmd
, addr
, ptl
);
1177 * This is the old fallback for page remapping.
1179 * For historical reasons, it only allows reserved pages. Only
1180 * old drivers should use this, and they needed to mark their
1181 * pages reserved for the old functions anyway.
1183 static int insert_page(struct mm_struct
*mm
, unsigned long addr
, struct page
*page
, pgprot_t prot
)
1193 flush_dcache_page(page
);
1194 pte
= get_locked_pte(mm
, addr
, &ptl
);
1198 if (!pte_none(*pte
))
1201 /* Ok, finally just insert the thing.. */
1203 inc_mm_counter(mm
, file_rss
);
1204 page_add_file_rmap(page
);
1205 set_pte_at(mm
, addr
, pte
, mk_pte(page
, prot
));
1209 pte_unmap_unlock(pte
, ptl
);
1215 * This allows drivers to insert individual pages they've allocated
1218 * The page has to be a nice clean _individual_ kernel allocation.
1219 * If you allocate a compound page, you need to have marked it as
1220 * such (__GFP_COMP), or manually just split the page up yourself
1221 * (see split_page()).
1223 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1224 * took an arbitrary page protection parameter. This doesn't allow
1225 * that. Your vma protection will have to be set up correctly, which
1226 * means that if you want a shared writable mapping, you'd better
1227 * ask for a shared writable mapping!
1229 * The page does not need to be reserved.
1231 int vm_insert_page(struct vm_area_struct
*vma
, unsigned long addr
, struct page
*page
)
1233 if (addr
< vma
->vm_start
|| addr
>= vma
->vm_end
)
1235 if (!page_count(page
))
1237 vma
->vm_flags
|= VM_INSERTPAGE
;
1238 return insert_page(vma
->vm_mm
, addr
, page
, vma
->vm_page_prot
);
1240 EXPORT_SYMBOL(vm_insert_page
);
1243 * maps a range of physical memory into the requested pages. the old
1244 * mappings are removed. any references to nonexistent pages results
1245 * in null mappings (currently treated as "copy-on-access")
1247 static int remap_pte_range(struct mm_struct
*mm
, pmd_t
*pmd
,
1248 unsigned long addr
, unsigned long end
,
1249 unsigned long pfn
, pgprot_t prot
)
1254 pte
= pte_alloc_map_lock(mm
, pmd
, addr
, &ptl
);
1258 BUG_ON(!pte_none(*pte
));
1259 set_pte_at(mm
, addr
, pte
, pfn_pte(pfn
, prot
));
1261 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1262 pte_unmap_unlock(pte
- 1, ptl
);
1266 static inline int remap_pmd_range(struct mm_struct
*mm
, pud_t
*pud
,
1267 unsigned long addr
, unsigned long end
,
1268 unsigned long pfn
, pgprot_t prot
)
1273 pfn
-= addr
>> PAGE_SHIFT
;
1274 pmd
= pmd_alloc(mm
, pud
, addr
);
1278 next
= pmd_addr_end(addr
, end
);
1279 if (remap_pte_range(mm
, pmd
, addr
, next
,
1280 pfn
+ (addr
>> PAGE_SHIFT
), prot
))
1282 } while (pmd
++, addr
= next
, addr
!= end
);
1286 static inline int remap_pud_range(struct mm_struct
*mm
, pgd_t
*pgd
,
1287 unsigned long addr
, unsigned long end
,
1288 unsigned long pfn
, pgprot_t prot
)
1293 pfn
-= addr
>> PAGE_SHIFT
;
1294 pud
= pud_alloc(mm
, pgd
, addr
);
1298 next
= pud_addr_end(addr
, end
);
1299 if (remap_pmd_range(mm
, pud
, addr
, next
,
1300 pfn
+ (addr
>> PAGE_SHIFT
), prot
))
1302 } while (pud
++, addr
= next
, addr
!= end
);
1306 /* Note: this is only safe if the mm semaphore is held when called. */
1307 int remap_pfn_range(struct vm_area_struct
*vma
, unsigned long addr
,
1308 unsigned long pfn
, unsigned long size
, pgprot_t prot
)
1312 unsigned long end
= addr
+ PAGE_ALIGN(size
);
1313 struct mm_struct
*mm
= vma
->vm_mm
;
1317 * Physically remapped pages are special. Tell the
1318 * rest of the world about it:
1319 * VM_IO tells people not to look at these pages
1320 * (accesses can have side effects).
1321 * VM_RESERVED is specified all over the place, because
1322 * in 2.4 it kept swapout's vma scan off this vma; but
1323 * in 2.6 the LRU scan won't even find its pages, so this
1324 * flag means no more than count its pages in reserved_vm,
1325 * and omit it from core dump, even when VM_IO turned off.
1326 * VM_PFNMAP tells the core MM that the base pages are just
1327 * raw PFN mappings, and do not have a "struct page" associated
1330 * There's a horrible special case to handle copy-on-write
1331 * behaviour that some programs depend on. We mark the "original"
1332 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
1334 if (is_cow_mapping(vma
->vm_flags
)) {
1335 if (addr
!= vma
->vm_start
|| end
!= vma
->vm_end
)
1337 vma
->vm_pgoff
= pfn
;
1340 vma
->vm_flags
|= VM_IO
| VM_RESERVED
| VM_PFNMAP
;
1342 BUG_ON(addr
>= end
);
1343 pfn
-= addr
>> PAGE_SHIFT
;
1344 pgd
= pgd_offset(mm
, addr
);
1345 flush_cache_range(vma
, addr
, end
);
1347 next
= pgd_addr_end(addr
, end
);
1348 err
= remap_pud_range(mm
, pgd
, addr
, next
,
1349 pfn
+ (addr
>> PAGE_SHIFT
), prot
);
1352 } while (pgd
++, addr
= next
, addr
!= end
);
1355 EXPORT_SYMBOL(remap_pfn_range
);
1358 * handle_pte_fault chooses page fault handler according to an entry
1359 * which was read non-atomically. Before making any commitment, on
1360 * those architectures or configurations (e.g. i386 with PAE) which
1361 * might give a mix of unmatched parts, do_swap_page and do_file_page
1362 * must check under lock before unmapping the pte and proceeding
1363 * (but do_wp_page is only called after already making such a check;
1364 * and do_anonymous_page and do_no_page can safely check later on).
1366 static inline int pte_unmap_same(struct mm_struct
*mm
, pmd_t
*pmd
,
1367 pte_t
*page_table
, pte_t orig_pte
)
1370 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1371 if (sizeof(pte_t
) > sizeof(unsigned long)) {
1372 spinlock_t
*ptl
= pte_lockptr(mm
, pmd
);
1374 same
= pte_same(*page_table
, orig_pte
);
1378 pte_unmap(page_table
);
1383 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1384 * servicing faults for write access. In the normal case, do always want
1385 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1386 * that do not have writing enabled, when used by access_process_vm.
1388 static inline pte_t
maybe_mkwrite(pte_t pte
, struct vm_area_struct
*vma
)
1390 if (likely(vma
->vm_flags
& VM_WRITE
))
1391 pte
= pte_mkwrite(pte
);
1395 static inline void cow_user_page(struct page
*dst
, struct page
*src
, unsigned long va
)
1398 * If the source page was a PFN mapping, we don't have
1399 * a "struct page" for it. We do a best-effort copy by
1400 * just copying from the original user address. If that
1401 * fails, we just zero-fill it. Live with it.
1403 if (unlikely(!src
)) {
1404 void *kaddr
= kmap_atomic(dst
, KM_USER0
);
1405 void __user
*uaddr
= (void __user
*)(va
& PAGE_MASK
);
1408 * This really shouldn't fail, because the page is there
1409 * in the page tables. But it might just be unreadable,
1410 * in which case we just give up and fill the result with
1413 if (__copy_from_user_inatomic(kaddr
, uaddr
, PAGE_SIZE
))
1414 memset(kaddr
, 0, PAGE_SIZE
);
1415 kunmap_atomic(kaddr
, KM_USER0
);
1419 copy_user_highpage(dst
, src
, va
);
1423 * This routine handles present pages, when users try to write
1424 * to a shared page. It is done by copying the page to a new address
1425 * and decrementing the shared-page counter for the old page.
1427 * Note that this routine assumes that the protection checks have been
1428 * done by the caller (the low-level page fault routine in most cases).
1429 * Thus we can safely just mark it writable once we've done any necessary
1432 * We also mark the page dirty at this point even though the page will
1433 * change only once the write actually happens. This avoids a few races,
1434 * and potentially makes it more efficient.
1436 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1437 * but allow concurrent faults), with pte both mapped and locked.
1438 * We return with mmap_sem still held, but pte unmapped and unlocked.
1440 static int do_wp_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1441 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
1442 spinlock_t
*ptl
, pte_t orig_pte
)
1444 struct page
*old_page
, *new_page
;
1446 int ret
= VM_FAULT_MINOR
;
1448 old_page
= vm_normal_page(vma
, address
, orig_pte
);
1452 if (PageAnon(old_page
) && !TestSetPageLocked(old_page
)) {
1453 int reuse
= can_share_swap_page(old_page
);
1454 unlock_page(old_page
);
1456 flush_cache_page(vma
, address
, pte_pfn(orig_pte
));
1457 entry
= pte_mkyoung(orig_pte
);
1458 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
1459 ptep_set_access_flags(vma
, address
, page_table
, entry
, 1);
1460 update_mmu_cache(vma
, address
, entry
);
1461 lazy_mmu_prot_update(entry
);
1462 ret
|= VM_FAULT_WRITE
;
1468 * Ok, we need to copy. Oh, well..
1470 page_cache_get(old_page
);
1472 pte_unmap_unlock(page_table
, ptl
);
1474 if (unlikely(anon_vma_prepare(vma
)))
1476 if (old_page
== ZERO_PAGE(address
)) {
1477 new_page
= alloc_zeroed_user_highpage(vma
, address
);
1481 new_page
= alloc_page_vma(GFP_HIGHUSER
, vma
, address
);
1484 cow_user_page(new_page
, old_page
, address
);
1488 * Re-check the pte - we dropped the lock
1490 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
1491 if (likely(pte_same(*page_table
, orig_pte
))) {
1493 page_remove_rmap(old_page
);
1494 if (!PageAnon(old_page
)) {
1495 dec_mm_counter(mm
, file_rss
);
1496 inc_mm_counter(mm
, anon_rss
);
1499 inc_mm_counter(mm
, anon_rss
);
1500 flush_cache_page(vma
, address
, pte_pfn(orig_pte
));
1501 entry
= mk_pte(new_page
, vma
->vm_page_prot
);
1502 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
1503 ptep_establish(vma
, address
, page_table
, entry
);
1504 update_mmu_cache(vma
, address
, entry
);
1505 lazy_mmu_prot_update(entry
);
1506 lru_cache_add_active(new_page
);
1507 page_add_new_anon_rmap(new_page
, vma
, address
);
1509 /* Free the old page.. */
1510 new_page
= old_page
;
1511 ret
|= VM_FAULT_WRITE
;
1514 page_cache_release(new_page
);
1516 page_cache_release(old_page
);
1518 pte_unmap_unlock(page_table
, ptl
);
1522 page_cache_release(old_page
);
1523 return VM_FAULT_OOM
;
1527 * Helper functions for unmap_mapping_range().
1529 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1531 * We have to restart searching the prio_tree whenever we drop the lock,
1532 * since the iterator is only valid while the lock is held, and anyway
1533 * a later vma might be split and reinserted earlier while lock dropped.
1535 * The list of nonlinear vmas could be handled more efficiently, using
1536 * a placeholder, but handle it in the same way until a need is shown.
1537 * It is important to search the prio_tree before nonlinear list: a vma
1538 * may become nonlinear and be shifted from prio_tree to nonlinear list
1539 * while the lock is dropped; but never shifted from list to prio_tree.
1541 * In order to make forward progress despite restarting the search,
1542 * vm_truncate_count is used to mark a vma as now dealt with, so we can
1543 * quickly skip it next time around. Since the prio_tree search only
1544 * shows us those vmas affected by unmapping the range in question, we
1545 * can't efficiently keep all vmas in step with mapping->truncate_count:
1546 * so instead reset them all whenever it wraps back to 0 (then go to 1).
1547 * mapping->truncate_count and vma->vm_truncate_count are protected by
1550 * In order to make forward progress despite repeatedly restarting some
1551 * large vma, note the restart_addr from unmap_vmas when it breaks out:
1552 * and restart from that address when we reach that vma again. It might
1553 * have been split or merged, shrunk or extended, but never shifted: so
1554 * restart_addr remains valid so long as it remains in the vma's range.
1555 * unmap_mapping_range forces truncate_count to leap over page-aligned
1556 * values so we can save vma's restart_addr in its truncate_count field.
1558 #define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
1560 static void reset_vma_truncate_counts(struct address_space
*mapping
)
1562 struct vm_area_struct
*vma
;
1563 struct prio_tree_iter iter
;
1565 vma_prio_tree_foreach(vma
, &iter
, &mapping
->i_mmap
, 0, ULONG_MAX
)
1566 vma
->vm_truncate_count
= 0;
1567 list_for_each_entry(vma
, &mapping
->i_mmap_nonlinear
, shared
.vm_set
.list
)
1568 vma
->vm_truncate_count
= 0;
1571 static int unmap_mapping_range_vma(struct vm_area_struct
*vma
,
1572 unsigned long start_addr
, unsigned long end_addr
,
1573 struct zap_details
*details
)
1575 unsigned long restart_addr
;
1579 restart_addr
= vma
->vm_truncate_count
;
1580 if (is_restart_addr(restart_addr
) && start_addr
< restart_addr
) {
1581 start_addr
= restart_addr
;
1582 if (start_addr
>= end_addr
) {
1583 /* Top of vma has been split off since last time */
1584 vma
->vm_truncate_count
= details
->truncate_count
;
1589 restart_addr
= zap_page_range(vma
, start_addr
,
1590 end_addr
- start_addr
, details
);
1591 need_break
= need_resched() ||
1592 need_lockbreak(details
->i_mmap_lock
);
1594 if (restart_addr
>= end_addr
) {
1595 /* We have now completed this vma: mark it so */
1596 vma
->vm_truncate_count
= details
->truncate_count
;
1600 /* Note restart_addr in vma's truncate_count field */
1601 vma
->vm_truncate_count
= restart_addr
;
1606 spin_unlock(details
->i_mmap_lock
);
1608 spin_lock(details
->i_mmap_lock
);
1612 static inline void unmap_mapping_range_tree(struct prio_tree_root
*root
,
1613 struct zap_details
*details
)
1615 struct vm_area_struct
*vma
;
1616 struct prio_tree_iter iter
;
1617 pgoff_t vba
, vea
, zba
, zea
;
1620 vma_prio_tree_foreach(vma
, &iter
, root
,
1621 details
->first_index
, details
->last_index
) {
1622 /* Skip quickly over those we have already dealt with */
1623 if (vma
->vm_truncate_count
== details
->truncate_count
)
1626 vba
= vma
->vm_pgoff
;
1627 vea
= vba
+ ((vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
) - 1;
1628 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1629 zba
= details
->first_index
;
1632 zea
= details
->last_index
;
1636 if (unmap_mapping_range_vma(vma
,
1637 ((zba
- vba
) << PAGE_SHIFT
) + vma
->vm_start
,
1638 ((zea
- vba
+ 1) << PAGE_SHIFT
) + vma
->vm_start
,
1644 static inline void unmap_mapping_range_list(struct list_head
*head
,
1645 struct zap_details
*details
)
1647 struct vm_area_struct
*vma
;
1650 * In nonlinear VMAs there is no correspondence between virtual address
1651 * offset and file offset. So we must perform an exhaustive search
1652 * across *all* the pages in each nonlinear VMA, not just the pages
1653 * whose virtual address lies outside the file truncation point.
1656 list_for_each_entry(vma
, head
, shared
.vm_set
.list
) {
1657 /* Skip quickly over those we have already dealt with */
1658 if (vma
->vm_truncate_count
== details
->truncate_count
)
1660 details
->nonlinear_vma
= vma
;
1661 if (unmap_mapping_range_vma(vma
, vma
->vm_start
,
1662 vma
->vm_end
, details
) < 0)
1668 * unmap_mapping_range - unmap the portion of all mmaps
1669 * in the specified address_space corresponding to the specified
1670 * page range in the underlying file.
1671 * @mapping: the address space containing mmaps to be unmapped.
1672 * @holebegin: byte in first page to unmap, relative to the start of
1673 * the underlying file. This will be rounded down to a PAGE_SIZE
1674 * boundary. Note that this is different from vmtruncate(), which
1675 * must keep the partial page. In contrast, we must get rid of
1677 * @holelen: size of prospective hole in bytes. This will be rounded
1678 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
1680 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
1681 * but 0 when invalidating pagecache, don't throw away private data.
1683 void unmap_mapping_range(struct address_space
*mapping
,
1684 loff_t
const holebegin
, loff_t
const holelen
, int even_cows
)
1686 struct zap_details details
;
1687 pgoff_t hba
= holebegin
>> PAGE_SHIFT
;
1688 pgoff_t hlen
= (holelen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1690 /* Check for overflow. */
1691 if (sizeof(holelen
) > sizeof(hlen
)) {
1693 (holebegin
+ holelen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1694 if (holeend
& ~(long long)ULONG_MAX
)
1695 hlen
= ULONG_MAX
- hba
+ 1;
1698 details
.check_mapping
= even_cows
? NULL
: mapping
;
1699 details
.nonlinear_vma
= NULL
;
1700 details
.first_index
= hba
;
1701 details
.last_index
= hba
+ hlen
- 1;
1702 if (details
.last_index
< details
.first_index
)
1703 details
.last_index
= ULONG_MAX
;
1704 details
.i_mmap_lock
= &mapping
->i_mmap_lock
;
1706 spin_lock(&mapping
->i_mmap_lock
);
1708 /* serialize i_size write against truncate_count write */
1710 /* Protect against page faults, and endless unmapping loops */
1711 mapping
->truncate_count
++;
1713 * For archs where spin_lock has inclusive semantics like ia64
1714 * this smp_mb() will prevent to read pagetable contents
1715 * before the truncate_count increment is visible to
1719 if (unlikely(is_restart_addr(mapping
->truncate_count
))) {
1720 if (mapping
->truncate_count
== 0)
1721 reset_vma_truncate_counts(mapping
);
1722 mapping
->truncate_count
++;
1724 details
.truncate_count
= mapping
->truncate_count
;
1726 if (unlikely(!prio_tree_empty(&mapping
->i_mmap
)))
1727 unmap_mapping_range_tree(&mapping
->i_mmap
, &details
);
1728 if (unlikely(!list_empty(&mapping
->i_mmap_nonlinear
)))
1729 unmap_mapping_range_list(&mapping
->i_mmap_nonlinear
, &details
);
1730 spin_unlock(&mapping
->i_mmap_lock
);
1732 EXPORT_SYMBOL(unmap_mapping_range
);
1735 * Handle all mappings that got truncated by a "truncate()"
1738 * NOTE! We have to be ready to update the memory sharing
1739 * between the file and the memory map for a potential last
1740 * incomplete page. Ugly, but necessary.
1742 int vmtruncate(struct inode
* inode
, loff_t offset
)
1744 struct address_space
*mapping
= inode
->i_mapping
;
1745 unsigned long limit
;
1747 if (inode
->i_size
< offset
)
1750 * truncation of in-use swapfiles is disallowed - it would cause
1751 * subsequent swapout to scribble on the now-freed blocks.
1753 if (IS_SWAPFILE(inode
))
1755 i_size_write(inode
, offset
);
1756 unmap_mapping_range(mapping
, offset
+ PAGE_SIZE
- 1, 0, 1);
1757 truncate_inode_pages(mapping
, offset
);
1761 limit
= current
->signal
->rlim
[RLIMIT_FSIZE
].rlim_cur
;
1762 if (limit
!= RLIM_INFINITY
&& offset
> limit
)
1764 if (offset
> inode
->i_sb
->s_maxbytes
)
1766 i_size_write(inode
, offset
);
1769 if (inode
->i_op
&& inode
->i_op
->truncate
)
1770 inode
->i_op
->truncate(inode
);
1773 send_sig(SIGXFSZ
, current
, 0);
1779 EXPORT_SYMBOL(vmtruncate
);
1781 int vmtruncate_range(struct inode
*inode
, loff_t offset
, loff_t end
)
1783 struct address_space
*mapping
= inode
->i_mapping
;
1786 * If the underlying filesystem is not going to provide
1787 * a way to truncate a range of blocks (punch a hole) -
1788 * we should return failure right now.
1790 if (!inode
->i_op
|| !inode
->i_op
->truncate_range
)
1793 mutex_lock(&inode
->i_mutex
);
1794 down_write(&inode
->i_alloc_sem
);
1795 unmap_mapping_range(mapping
, offset
, (end
- offset
), 1);
1796 truncate_inode_pages_range(mapping
, offset
, end
);
1797 inode
->i_op
->truncate_range(inode
, offset
, end
);
1798 up_write(&inode
->i_alloc_sem
);
1799 mutex_unlock(&inode
->i_mutex
);
1803 EXPORT_SYMBOL(vmtruncate_range
);
1806 * Primitive swap readahead code. We simply read an aligned block of
1807 * (1 << page_cluster) entries in the swap area. This method is chosen
1808 * because it doesn't cost us any seek time. We also make sure to queue
1809 * the 'original' request together with the readahead ones...
1811 * This has been extended to use the NUMA policies from the mm triggering
1814 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
1816 void swapin_readahead(swp_entry_t entry
, unsigned long addr
,struct vm_area_struct
*vma
)
1819 struct vm_area_struct
*next_vma
= vma
? vma
->vm_next
: NULL
;
1822 struct page
*new_page
;
1823 unsigned long offset
;
1826 * Get the number of handles we should do readahead io to.
1828 num
= valid_swaphandles(entry
, &offset
);
1829 for (i
= 0; i
< num
; offset
++, i
++) {
1830 /* Ok, do the async read-ahead now */
1831 new_page
= read_swap_cache_async(swp_entry(swp_type(entry
),
1832 offset
), vma
, addr
);
1835 page_cache_release(new_page
);
1838 * Find the next applicable VMA for the NUMA policy.
1844 if (addr
>= vma
->vm_end
) {
1846 next_vma
= vma
? vma
->vm_next
: NULL
;
1848 if (vma
&& addr
< vma
->vm_start
)
1851 if (next_vma
&& addr
>= next_vma
->vm_start
) {
1853 next_vma
= vma
->vm_next
;
1858 lru_add_drain(); /* Push any new pages onto the LRU now */
1862 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1863 * but allow concurrent faults), and pte mapped but not yet locked.
1864 * We return with mmap_sem still held, but pte unmapped and unlocked.
1866 static int do_swap_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1867 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
1868 int write_access
, pte_t orig_pte
)
1874 int ret
= VM_FAULT_MINOR
;
1876 if (!pte_unmap_same(mm
, pmd
, page_table
, orig_pte
))
1879 entry
= pte_to_swp_entry(orig_pte
);
1881 page
= lookup_swap_cache(entry
);
1883 swapin_readahead(entry
, address
, vma
);
1884 page
= read_swap_cache_async(entry
, vma
, address
);
1887 * Back out if somebody else faulted in this pte
1888 * while we released the pte lock.
1890 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
1891 if (likely(pte_same(*page_table
, orig_pte
)))
1896 /* Had to read the page from swap area: Major fault */
1897 ret
= VM_FAULT_MAJOR
;
1898 inc_page_state(pgmajfault
);
1902 mark_page_accessed(page
);
1904 if (!PageSwapCache(page
)) {
1905 /* Page migration has occured */
1907 page_cache_release(page
);
1912 * Back out if somebody else already faulted in this pte.
1914 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
1915 if (unlikely(!pte_same(*page_table
, orig_pte
)))
1918 if (unlikely(!PageUptodate(page
))) {
1919 ret
= VM_FAULT_SIGBUS
;
1923 /* The page isn't present yet, go ahead with the fault. */
1925 inc_mm_counter(mm
, anon_rss
);
1926 pte
= mk_pte(page
, vma
->vm_page_prot
);
1927 if (write_access
&& can_share_swap_page(page
)) {
1928 pte
= maybe_mkwrite(pte_mkdirty(pte
), vma
);
1932 flush_icache_page(vma
, page
);
1933 set_pte_at(mm
, address
, page_table
, pte
);
1934 page_add_anon_rmap(page
, vma
, address
);
1938 remove_exclusive_swap_page(page
);
1942 if (do_wp_page(mm
, vma
, address
,
1943 page_table
, pmd
, ptl
, pte
) == VM_FAULT_OOM
)
1948 /* No need to invalidate - it was non-present before */
1949 update_mmu_cache(vma
, address
, pte
);
1950 lazy_mmu_prot_update(pte
);
1952 pte_unmap_unlock(page_table
, ptl
);
1956 pte_unmap_unlock(page_table
, ptl
);
1958 page_cache_release(page
);
1963 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1964 * but allow concurrent faults), and pte mapped but not yet locked.
1965 * We return with mmap_sem still held, but pte unmapped and unlocked.
1967 static int do_anonymous_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1968 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
1976 /* Allocate our own private page. */
1977 pte_unmap(page_table
);
1979 if (unlikely(anon_vma_prepare(vma
)))
1981 page
= alloc_zeroed_user_highpage(vma
, address
);
1985 entry
= mk_pte(page
, vma
->vm_page_prot
);
1986 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
1988 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
1989 if (!pte_none(*page_table
))
1991 inc_mm_counter(mm
, anon_rss
);
1992 lru_cache_add_active(page
);
1993 page_add_new_anon_rmap(page
, vma
, address
);
1995 /* Map the ZERO_PAGE - vm_page_prot is readonly */
1996 page
= ZERO_PAGE(address
);
1997 page_cache_get(page
);
1998 entry
= mk_pte(page
, vma
->vm_page_prot
);
2000 ptl
= pte_lockptr(mm
, pmd
);
2002 if (!pte_none(*page_table
))
2004 inc_mm_counter(mm
, file_rss
);
2005 page_add_file_rmap(page
);
2008 set_pte_at(mm
, address
, page_table
, entry
);
2010 /* No need to invalidate - it was non-present before */
2011 update_mmu_cache(vma
, address
, entry
);
2012 lazy_mmu_prot_update(entry
);
2014 pte_unmap_unlock(page_table
, ptl
);
2015 return VM_FAULT_MINOR
;
2017 page_cache_release(page
);
2020 return VM_FAULT_OOM
;
2024 * do_no_page() tries to create a new page mapping. It aggressively
2025 * tries to share with existing pages, but makes a separate copy if
2026 * the "write_access" parameter is true in order to avoid the next
2029 * As this is called only for pages that do not currently exist, we
2030 * do not need to flush old virtual caches or the TLB.
2032 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2033 * but allow concurrent faults), and pte mapped but not yet locked.
2034 * We return with mmap_sem still held, but pte unmapped and unlocked.
2036 static int do_no_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2037 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
2041 struct page
*new_page
;
2042 struct address_space
*mapping
= NULL
;
2044 unsigned int sequence
= 0;
2045 int ret
= VM_FAULT_MINOR
;
2048 pte_unmap(page_table
);
2049 BUG_ON(vma
->vm_flags
& VM_PFNMAP
);
2052 mapping
= vma
->vm_file
->f_mapping
;
2053 sequence
= mapping
->truncate_count
;
2054 smp_rmb(); /* serializes i_size against truncate_count */
2057 new_page
= vma
->vm_ops
->nopage(vma
, address
& PAGE_MASK
, &ret
);
2059 * No smp_rmb is needed here as long as there's a full
2060 * spin_lock/unlock sequence inside the ->nopage callback
2061 * (for the pagecache lookup) that acts as an implicit
2062 * smp_mb() and prevents the i_size read to happen
2063 * after the next truncate_count read.
2066 /* no page was available -- either SIGBUS or OOM */
2067 if (new_page
== NOPAGE_SIGBUS
)
2068 return VM_FAULT_SIGBUS
;
2069 if (new_page
== NOPAGE_OOM
)
2070 return VM_FAULT_OOM
;
2073 * Should we do an early C-O-W break?
2075 if (write_access
&& !(vma
->vm_flags
& VM_SHARED
)) {
2078 if (unlikely(anon_vma_prepare(vma
)))
2080 page
= alloc_page_vma(GFP_HIGHUSER
, vma
, address
);
2083 copy_user_highpage(page
, new_page
, address
);
2084 page_cache_release(new_page
);
2089 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
2091 * For a file-backed vma, someone could have truncated or otherwise
2092 * invalidated this page. If unmap_mapping_range got called,
2093 * retry getting the page.
2095 if (mapping
&& unlikely(sequence
!= mapping
->truncate_count
)) {
2096 pte_unmap_unlock(page_table
, ptl
);
2097 page_cache_release(new_page
);
2099 sequence
= mapping
->truncate_count
;
2105 * This silly early PAGE_DIRTY setting removes a race
2106 * due to the bad i386 page protection. But it's valid
2107 * for other architectures too.
2109 * Note that if write_access is true, we either now have
2110 * an exclusive copy of the page, or this is a shared mapping,
2111 * so we can make it writable and dirty to avoid having to
2112 * handle that later.
2114 /* Only go through if we didn't race with anybody else... */
2115 if (pte_none(*page_table
)) {
2116 flush_icache_page(vma
, new_page
);
2117 entry
= mk_pte(new_page
, vma
->vm_page_prot
);
2119 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
2120 set_pte_at(mm
, address
, page_table
, entry
);
2122 inc_mm_counter(mm
, anon_rss
);
2123 lru_cache_add_active(new_page
);
2124 page_add_new_anon_rmap(new_page
, vma
, address
);
2126 inc_mm_counter(mm
, file_rss
);
2127 page_add_file_rmap(new_page
);
2130 /* One of our sibling threads was faster, back out. */
2131 page_cache_release(new_page
);
2135 /* no need to invalidate: a not-present page shouldn't be cached */
2136 update_mmu_cache(vma
, address
, entry
);
2137 lazy_mmu_prot_update(entry
);
2139 pte_unmap_unlock(page_table
, ptl
);
2142 page_cache_release(new_page
);
2143 return VM_FAULT_OOM
;
2147 * Fault of a previously existing named mapping. Repopulate the pte
2148 * from the encoded file_pte if possible. This enables swappable
2151 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2152 * but allow concurrent faults), and pte mapped but not yet locked.
2153 * We return with mmap_sem still held, but pte unmapped and unlocked.
2155 static int do_file_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2156 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
2157 int write_access
, pte_t orig_pte
)
2162 if (!pte_unmap_same(mm
, pmd
, page_table
, orig_pte
))
2163 return VM_FAULT_MINOR
;
2165 if (unlikely(!(vma
->vm_flags
& VM_NONLINEAR
))) {
2167 * Page table corrupted: show pte and kill process.
2169 print_bad_pte(vma
, orig_pte
, address
);
2170 return VM_FAULT_OOM
;
2172 /* We can then assume vm->vm_ops && vma->vm_ops->populate */
2174 pgoff
= pte_to_pgoff(orig_pte
);
2175 err
= vma
->vm_ops
->populate(vma
, address
& PAGE_MASK
, PAGE_SIZE
,
2176 vma
->vm_page_prot
, pgoff
, 0);
2178 return VM_FAULT_OOM
;
2180 return VM_FAULT_SIGBUS
;
2181 return VM_FAULT_MAJOR
;
2185 * These routines also need to handle stuff like marking pages dirty
2186 * and/or accessed for architectures that don't do it in hardware (most
2187 * RISC architectures). The early dirtying is also good on the i386.
2189 * There is also a hook called "update_mmu_cache()" that architectures
2190 * with external mmu caches can use to update those (ie the Sparc or
2191 * PowerPC hashed page tables that act as extended TLBs).
2193 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2194 * but allow concurrent faults), and pte mapped but not yet locked.
2195 * We return with mmap_sem still held, but pte unmapped and unlocked.
2197 static inline int handle_pte_fault(struct mm_struct
*mm
,
2198 struct vm_area_struct
*vma
, unsigned long address
,
2199 pte_t
*pte
, pmd_t
*pmd
, int write_access
)
2205 old_entry
= entry
= *pte
;
2206 if (!pte_present(entry
)) {
2207 if (pte_none(entry
)) {
2208 if (!vma
->vm_ops
|| !vma
->vm_ops
->nopage
)
2209 return do_anonymous_page(mm
, vma
, address
,
2210 pte
, pmd
, write_access
);
2211 return do_no_page(mm
, vma
, address
,
2212 pte
, pmd
, write_access
);
2214 if (pte_file(entry
))
2215 return do_file_page(mm
, vma
, address
,
2216 pte
, pmd
, write_access
, entry
);
2217 return do_swap_page(mm
, vma
, address
,
2218 pte
, pmd
, write_access
, entry
);
2221 ptl
= pte_lockptr(mm
, pmd
);
2223 if (unlikely(!pte_same(*pte
, entry
)))
2226 if (!pte_write(entry
))
2227 return do_wp_page(mm
, vma
, address
,
2228 pte
, pmd
, ptl
, entry
);
2229 entry
= pte_mkdirty(entry
);
2231 entry
= pte_mkyoung(entry
);
2232 if (!pte_same(old_entry
, entry
)) {
2233 ptep_set_access_flags(vma
, address
, pte
, entry
, write_access
);
2234 update_mmu_cache(vma
, address
, entry
);
2235 lazy_mmu_prot_update(entry
);
2238 * This is needed only for protection faults but the arch code
2239 * is not yet telling us if this is a protection fault or not.
2240 * This still avoids useless tlb flushes for .text page faults
2244 flush_tlb_page(vma
, address
);
2247 pte_unmap_unlock(pte
, ptl
);
2248 return VM_FAULT_MINOR
;
2252 * By the time we get here, we already hold the mm semaphore
2254 int __handle_mm_fault(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2255 unsigned long address
, int write_access
)
2262 __set_current_state(TASK_RUNNING
);
2264 inc_page_state(pgfault
);
2266 if (unlikely(is_vm_hugetlb_page(vma
)))
2267 return hugetlb_fault(mm
, vma
, address
, write_access
);
2269 pgd
= pgd_offset(mm
, address
);
2270 pud
= pud_alloc(mm
, pgd
, address
);
2272 return VM_FAULT_OOM
;
2273 pmd
= pmd_alloc(mm
, pud
, address
);
2275 return VM_FAULT_OOM
;
2276 pte
= pte_alloc_map(mm
, pmd
, address
);
2278 return VM_FAULT_OOM
;
2280 return handle_pte_fault(mm
, vma
, address
, pte
, pmd
, write_access
);
2283 EXPORT_SYMBOL_GPL(__handle_mm_fault
);
2285 #ifndef __PAGETABLE_PUD_FOLDED
2287 * Allocate page upper directory.
2288 * We've already handled the fast-path in-line.
2290 int __pud_alloc(struct mm_struct
*mm
, pgd_t
*pgd
, unsigned long address
)
2292 pud_t
*new = pud_alloc_one(mm
, address
);
2296 spin_lock(&mm
->page_table_lock
);
2297 if (pgd_present(*pgd
)) /* Another has populated it */
2300 pgd_populate(mm
, pgd
, new);
2301 spin_unlock(&mm
->page_table_lock
);
2305 /* Workaround for gcc 2.96 */
2306 int __pud_alloc(struct mm_struct
*mm
, pgd_t
*pgd
, unsigned long address
)
2310 #endif /* __PAGETABLE_PUD_FOLDED */
2312 #ifndef __PAGETABLE_PMD_FOLDED
2314 * Allocate page middle directory.
2315 * We've already handled the fast-path in-line.
2317 int __pmd_alloc(struct mm_struct
*mm
, pud_t
*pud
, unsigned long address
)
2319 pmd_t
*new = pmd_alloc_one(mm
, address
);
2323 spin_lock(&mm
->page_table_lock
);
2324 #ifndef __ARCH_HAS_4LEVEL_HACK
2325 if (pud_present(*pud
)) /* Another has populated it */
2328 pud_populate(mm
, pud
, new);
2330 if (pgd_present(*pud
)) /* Another has populated it */
2333 pgd_populate(mm
, pud
, new);
2334 #endif /* __ARCH_HAS_4LEVEL_HACK */
2335 spin_unlock(&mm
->page_table_lock
);
2339 /* Workaround for gcc 2.96 */
2340 int __pmd_alloc(struct mm_struct
*mm
, pud_t
*pud
, unsigned long address
)
2344 #endif /* __PAGETABLE_PMD_FOLDED */
2346 int make_pages_present(unsigned long addr
, unsigned long end
)
2348 int ret
, len
, write
;
2349 struct vm_area_struct
* vma
;
2351 vma
= find_vma(current
->mm
, addr
);
2354 write
= (vma
->vm_flags
& VM_WRITE
) != 0;
2357 if (end
> vma
->vm_end
)
2359 len
= (end
+PAGE_SIZE
-1)/PAGE_SIZE
-addr
/PAGE_SIZE
;
2360 ret
= get_user_pages(current
, current
->mm
, addr
,
2361 len
, write
, 0, NULL
, NULL
);
2364 return ret
== len
? 0 : -1;
2368 * Map a vmalloc()-space virtual address to the physical page.
2370 struct page
* vmalloc_to_page(void * vmalloc_addr
)
2372 unsigned long addr
= (unsigned long) vmalloc_addr
;
2373 struct page
*page
= NULL
;
2374 pgd_t
*pgd
= pgd_offset_k(addr
);
2379 if (!pgd_none(*pgd
)) {
2380 pud
= pud_offset(pgd
, addr
);
2381 if (!pud_none(*pud
)) {
2382 pmd
= pmd_offset(pud
, addr
);
2383 if (!pmd_none(*pmd
)) {
2384 ptep
= pte_offset_map(pmd
, addr
);
2386 if (pte_present(pte
))
2387 page
= pte_page(pte
);
2395 EXPORT_SYMBOL(vmalloc_to_page
);
2398 * Map a vmalloc()-space virtual address to the physical page frame number.
2400 unsigned long vmalloc_to_pfn(void * vmalloc_addr
)
2402 return page_to_pfn(vmalloc_to_page(vmalloc_addr
));
2405 EXPORT_SYMBOL(vmalloc_to_pfn
);
2407 #if !defined(__HAVE_ARCH_GATE_AREA)
2409 #if defined(AT_SYSINFO_EHDR)
2410 static struct vm_area_struct gate_vma
;
2412 static int __init
gate_vma_init(void)
2414 gate_vma
.vm_mm
= NULL
;
2415 gate_vma
.vm_start
= FIXADDR_USER_START
;
2416 gate_vma
.vm_end
= FIXADDR_USER_END
;
2417 gate_vma
.vm_page_prot
= PAGE_READONLY
;
2418 gate_vma
.vm_flags
= 0;
2421 __initcall(gate_vma_init
);
2424 struct vm_area_struct
*get_gate_vma(struct task_struct
*tsk
)
2426 #ifdef AT_SYSINFO_EHDR
2433 int in_gate_area_no_task(unsigned long addr
)
2435 #ifdef AT_SYSINFO_EHDR
2436 if ((addr
>= FIXADDR_USER_START
) && (addr
< FIXADDR_USER_END
))
2442 #endif /* __HAVE_ARCH_GATE_AREA */