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_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
);
285 * Optimization: gather nearby vmas into one call down
287 while (next
&& next
->vm_start
<= vma
->vm_end
+ PMD_SIZE
288 && !is_hugepage_only_range(vma
->vm_mm
, next
->vm_start
,
292 anon_vma_unlink(vma
);
293 unlink_file_vma(vma
);
295 free_pgd_range(tlb
, addr
, vma
->vm_end
,
296 floor
, next
? next
->vm_start
: ceiling
);
302 int __pte_alloc(struct mm_struct
*mm
, pmd_t
*pmd
, unsigned long address
)
304 struct page
*new = pte_alloc_one(mm
, address
);
309 spin_lock(&mm
->page_table_lock
);
310 if (pmd_present(*pmd
)) { /* Another has populated it */
311 pte_lock_deinit(new);
315 inc_page_state(nr_page_table_pages
);
316 pmd_populate(mm
, pmd
, new);
318 spin_unlock(&mm
->page_table_lock
);
322 int __pte_alloc_kernel(pmd_t
*pmd
, unsigned long address
)
324 pte_t
*new = pte_alloc_one_kernel(&init_mm
, address
);
328 spin_lock(&init_mm
.page_table_lock
);
329 if (pmd_present(*pmd
)) /* Another has populated it */
330 pte_free_kernel(new);
332 pmd_populate_kernel(&init_mm
, pmd
, new);
333 spin_unlock(&init_mm
.page_table_lock
);
337 static inline void add_mm_rss(struct mm_struct
*mm
, int file_rss
, int anon_rss
)
340 add_mm_counter(mm
, file_rss
, file_rss
);
342 add_mm_counter(mm
, anon_rss
, anon_rss
);
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.
350 * The calling function must still handle the error.
352 void print_bad_pte(struct vm_area_struct
*vma
, pte_t pte
, unsigned long vaddr
)
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
);
362 static inline int is_cow_mapping(unsigned int flags
)
364 return (flags
& (VM_SHARED
| VM_MAYWRITE
)) == VM_MAYWRITE
;
368 * This function gets the "struct page" associated with a pte.
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".
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
381 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
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
387 struct page
*vm_normal_page(struct vm_area_struct
*vma
, unsigned long addr
, pte_t pte
)
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
)
395 if (!is_cow_mapping(vma
->vm_flags
))
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.
405 * Remove this test eventually!
407 if (unlikely(!pfn_valid(pfn
))) {
408 print_bad_pte(vma
, pte
, addr
);
413 * NOTE! We still have PageReserved() pages in the page
416 * The PAGE_ZERO() pages and various VDSO mappings can
417 * cause them to exist.
419 return pfn_to_page(pfn
);
423 * copy one vm_area from one task to the other. Assumes the page tables
424 * already present in the new task to be cleared in the whole range
425 * covered by this vma.
429 copy_one_pte(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
430 pte_t
*dst_pte
, pte_t
*src_pte
, struct vm_area_struct
*vma
,
431 unsigned long addr
, int *rss
)
433 unsigned long vm_flags
= vma
->vm_flags
;
434 pte_t pte
= *src_pte
;
437 /* pte contains position in swap or file, so copy. */
438 if (unlikely(!pte_present(pte
))) {
439 if (!pte_file(pte
)) {
440 swap_duplicate(pte_to_swp_entry(pte
));
441 /* make sure dst_mm is on swapoff's mmlist. */
442 if (unlikely(list_empty(&dst_mm
->mmlist
))) {
443 spin_lock(&mmlist_lock
);
444 if (list_empty(&dst_mm
->mmlist
))
445 list_add(&dst_mm
->mmlist
,
447 spin_unlock(&mmlist_lock
);
454 * If it's a COW mapping, write protect it both
455 * in the parent and the child
457 if (is_cow_mapping(vm_flags
)) {
458 ptep_set_wrprotect(src_mm
, addr
, src_pte
);
463 * If it's a shared mapping, mark it clean in
466 if (vm_flags
& VM_SHARED
)
467 pte
= pte_mkclean(pte
);
468 pte
= pte_mkold(pte
);
470 page
= vm_normal_page(vma
, addr
, pte
);
474 rss
[!!PageAnon(page
)]++;
478 set_pte_at(dst_mm
, addr
, dst_pte
, pte
);
481 static int copy_pte_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
482 pmd_t
*dst_pmd
, pmd_t
*src_pmd
, struct vm_area_struct
*vma
,
483 unsigned long addr
, unsigned long end
)
485 pte_t
*src_pte
, *dst_pte
;
486 spinlock_t
*src_ptl
, *dst_ptl
;
492 dst_pte
= pte_alloc_map_lock(dst_mm
, dst_pmd
, addr
, &dst_ptl
);
495 src_pte
= pte_offset_map_nested(src_pmd
, addr
);
496 src_ptl
= pte_lockptr(src_mm
, src_pmd
);
501 * We are holding two locks at this point - either of them
502 * could generate latencies in another task on another CPU.
504 if (progress
>= 32) {
506 if (need_resched() ||
507 need_lockbreak(src_ptl
) ||
508 need_lockbreak(dst_ptl
))
511 if (pte_none(*src_pte
)) {
515 copy_one_pte(dst_mm
, src_mm
, dst_pte
, src_pte
, vma
, addr
, rss
);
517 } while (dst_pte
++, src_pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
519 spin_unlock(src_ptl
);
520 pte_unmap_nested(src_pte
- 1);
521 add_mm_rss(dst_mm
, rss
[0], rss
[1]);
522 pte_unmap_unlock(dst_pte
- 1, dst_ptl
);
529 static inline int copy_pmd_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
530 pud_t
*dst_pud
, pud_t
*src_pud
, struct vm_area_struct
*vma
,
531 unsigned long addr
, unsigned long end
)
533 pmd_t
*src_pmd
, *dst_pmd
;
536 dst_pmd
= pmd_alloc(dst_mm
, dst_pud
, addr
);
539 src_pmd
= pmd_offset(src_pud
, addr
);
541 next
= pmd_addr_end(addr
, end
);
542 if (pmd_none_or_clear_bad(src_pmd
))
544 if (copy_pte_range(dst_mm
, src_mm
, dst_pmd
, src_pmd
,
547 } while (dst_pmd
++, src_pmd
++, addr
= next
, addr
!= end
);
551 static inline int copy_pud_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
552 pgd_t
*dst_pgd
, pgd_t
*src_pgd
, struct vm_area_struct
*vma
,
553 unsigned long addr
, unsigned long end
)
555 pud_t
*src_pud
, *dst_pud
;
558 dst_pud
= pud_alloc(dst_mm
, dst_pgd
, addr
);
561 src_pud
= pud_offset(src_pgd
, addr
);
563 next
= pud_addr_end(addr
, end
);
564 if (pud_none_or_clear_bad(src_pud
))
566 if (copy_pmd_range(dst_mm
, src_mm
, dst_pud
, src_pud
,
569 } while (dst_pud
++, src_pud
++, addr
= next
, addr
!= end
);
573 int copy_page_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
574 struct vm_area_struct
*vma
)
576 pgd_t
*src_pgd
, *dst_pgd
;
578 unsigned long addr
= vma
->vm_start
;
579 unsigned long end
= vma
->vm_end
;
582 * Don't copy ptes where a page fault will fill them correctly.
583 * Fork becomes much lighter when there are big shared or private
584 * readonly mappings. The tradeoff is that copy_page_range is more
585 * efficient than faulting.
587 if (!(vma
->vm_flags
& (VM_HUGETLB
|VM_NONLINEAR
|VM_PFNMAP
|VM_INSERTPAGE
))) {
592 if (is_vm_hugetlb_page(vma
))
593 return copy_hugetlb_page_range(dst_mm
, src_mm
, vma
);
595 dst_pgd
= pgd_offset(dst_mm
, addr
);
596 src_pgd
= pgd_offset(src_mm
, addr
);
598 next
= pgd_addr_end(addr
, end
);
599 if (pgd_none_or_clear_bad(src_pgd
))
601 if (copy_pud_range(dst_mm
, src_mm
, dst_pgd
, src_pgd
,
604 } while (dst_pgd
++, src_pgd
++, addr
= next
, addr
!= end
);
608 static unsigned long zap_pte_range(struct mmu_gather
*tlb
,
609 struct vm_area_struct
*vma
, pmd_t
*pmd
,
610 unsigned long addr
, unsigned long end
,
611 long *zap_work
, struct zap_details
*details
)
613 struct mm_struct
*mm
= tlb
->mm
;
619 pte
= pte_offset_map_lock(mm
, pmd
, addr
, &ptl
);
622 if (pte_none(ptent
)) {
627 (*zap_work
) -= PAGE_SIZE
;
629 if (pte_present(ptent
)) {
632 page
= vm_normal_page(vma
, addr
, ptent
);
633 if (unlikely(details
) && page
) {
635 * unmap_shared_mapping_pages() wants to
636 * invalidate cache without truncating:
637 * unmap shared but keep private pages.
639 if (details
->check_mapping
&&
640 details
->check_mapping
!= page
->mapping
)
643 * Each page->index must be checked when
644 * invalidating or truncating nonlinear.
646 if (details
->nonlinear_vma
&&
647 (page
->index
< details
->first_index
||
648 page
->index
> details
->last_index
))
651 ptent
= ptep_get_and_clear_full(mm
, addr
, pte
,
653 tlb_remove_tlb_entry(tlb
, pte
, addr
);
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
));
664 if (pte_dirty(ptent
))
665 set_page_dirty(page
);
666 if (pte_young(ptent
))
667 mark_page_accessed(page
);
670 page_remove_rmap(page
);
671 tlb_remove_page(tlb
, page
);
675 * If details->check_mapping, we leave swap entries;
676 * if details->nonlinear_vma, we leave file entries.
678 if (unlikely(details
))
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
);
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
)
699 pmd
= pmd_offset(pud
, addr
);
701 next
= pmd_addr_end(addr
, end
);
702 if (pmd_none_or_clear_bad(pmd
)) {
706 next
= zap_pte_range(tlb
, vma
, pmd
, addr
, next
,
708 } while (pmd
++, addr
= next
, (addr
!= end
&& *zap_work
> 0));
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
)
721 pud
= pud_offset(pgd
, addr
);
723 next
= pud_addr_end(addr
, end
);
724 if (pud_none_or_clear_bad(pud
)) {
728 next
= zap_pmd_range(tlb
, vma
, pud
, addr
, next
,
730 } while (pud
++, addr
= next
, (addr
!= end
&& *zap_work
> 0));
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
)
743 if (details
&& !details
->check_mapping
&& !details
->nonlinear_vma
)
747 tlb_start_vma(tlb
, vma
);
748 pgd
= pgd_offset(vma
->vm_mm
, addr
);
750 next
= pgd_addr_end(addr
, end
);
751 if (pgd_none_or_clear_bad(pgd
)) {
755 next
= zap_pud_range(tlb
, vma
, pgd
, addr
, next
,
757 } while (pgd
++, addr
= next
, (addr
!= end
&& *zap_work
> 0));
758 tlb_end_vma(tlb
, vma
);
763 #ifdef CONFIG_PREEMPT
764 # define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
766 /* No preempt: go for improved straight-line efficiency */
767 # define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
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
779 * Returns the end address of the unmapping (restart addr if interrupted).
781 * Unmap all pages in the vma list.
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.
787 * Only addresses between `start' and `end' will be unmapped.
789 * The VMA list must be sorted in ascending virtual address order.
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.
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
)
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
) {
811 start
= max(vma
->vm_start
, start_addr
);
812 if (start
>= vma
->vm_end
)
814 end
= min(vma
->vm_end
, end_addr
);
815 if (end
<= vma
->vm_start
)
818 if (vma
->vm_flags
& VM_ACCOUNT
)
819 *nr_accounted
+= (end
- start
) >> PAGE_SHIFT
;
821 while (start
!= end
) {
822 if (!tlb_start_valid
) {
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
);
833 start
= unmap_page_range(*tlbp
, vma
,
834 start
, end
, &zap_work
, details
);
837 BUG_ON(start
!= end
);
841 tlb_finish_mmu(*tlbp
, tlb_start
, start
);
843 if (need_resched() ||
844 (i_mmap_lock
&& need_lockbreak(i_mmap_lock
))) {
852 *tlbp
= tlb_gather_mmu(vma
->vm_mm
, fullmm
);
854 zap_work
= ZAP_BLOCK_SIZE
;
858 return start
; /* which is now the end (or restart) address */
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
868 unsigned long zap_page_range(struct vm_area_struct
*vma
, unsigned long address
,
869 unsigned long size
, struct zap_details
*details
)
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;
877 tlb
= tlb_gather_mmu(mm
, 0);
878 update_hiwater_rss(mm
);
879 end
= unmap_vmas(&tlb
, vma
, address
, end
, &nr_accounted
, details
);
881 tlb_finish_mmu(tlb
, address
, end
);
886 * Do a quick page-table lookup for a single page.
888 struct page
*follow_page(struct vm_area_struct
*vma
, unsigned long address
,
897 struct mm_struct
*mm
= vma
->vm_mm
;
899 page
= follow_huge_addr(mm
, address
, flags
& FOLL_WRITE
);
901 BUG_ON(flags
& FOLL_GET
);
906 pgd
= pgd_offset(mm
, address
);
907 if (pgd_none(*pgd
) || unlikely(pgd_bad(*pgd
)))
910 pud
= pud_offset(pgd
, address
);
911 if (pud_none(*pud
) || unlikely(pud_bad(*pud
)))
914 pmd
= pmd_offset(pud
, address
);
915 if (pmd_none(*pmd
) || unlikely(pmd_bad(*pmd
)))
918 if (pmd_huge(*pmd
)) {
919 BUG_ON(flags
& FOLL_GET
);
920 page
= follow_huge_pmd(mm
, address
, pmd
, flags
& FOLL_WRITE
);
924 ptep
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
929 if (!pte_present(pte
))
931 if ((flags
& FOLL_WRITE
) && !pte_write(pte
))
933 page
= vm_normal_page(vma
, address
, pte
);
937 if (flags
& FOLL_GET
)
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
);
946 pte_unmap_unlock(ptep
, ptl
);
952 * When core dumping an enormous anonymous area that nobody
953 * has touched so far, we don't want to allocate page tables.
955 if (flags
& FOLL_ANON
) {
956 page
= ZERO_PAGE(address
);
957 if (flags
& FOLL_GET
)
959 BUG_ON(flags
& FOLL_WRITE
);
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
)
969 unsigned int vm_flags
;
972 * Require read or write permissions.
973 * If 'force' is set, we only require the "MAY" flags.
975 vm_flags
= write
? (VM_WRITE
| VM_MAYWRITE
) : (VM_READ
| VM_MAYREAD
);
976 vm_flags
&= force
? (VM_MAYREAD
| VM_MAYWRITE
) : (VM_READ
| VM_WRITE
);
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
);
991 if (write
) /* user gate pages are read-only */
992 return i
? : -EFAULT
;
994 pgd
= pgd_offset_k(pg
);
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
);
1002 return i
? : -EFAULT
;
1003 pte
= pte_offset_map(pmd
, pg
);
1004 if (pte_none(*pte
)) {
1006 return i
? : -EFAULT
;
1009 struct page
*page
= vm_normal_page(gate_vma
, start
, *pte
);
1023 if (!vma
|| (vma
->vm_flags
& (VM_IO
| VM_PFNMAP
))
1024 || !(vm_flags
& vma
->vm_flags
))
1025 return i
? : -EFAULT
;
1027 if (is_vm_hugetlb_page(vma
)) {
1028 i
= follow_hugetlb_page(mm
, vma
, pages
, vmas
,
1033 foll_flags
= FOLL_TOUCH
;
1035 foll_flags
|= FOLL_GET
;
1036 if (!write
&& !(vma
->vm_flags
& VM_LOCKED
) &&
1037 (!vma
->vm_ops
|| !vma
->vm_ops
->nopage
))
1038 foll_flags
|= FOLL_ANON
;
1044 foll_flags
|= FOLL_WRITE
;
1047 while (!(page
= follow_page(vma
, start
, foll_flags
))) {
1049 ret
= __handle_mm_fault(mm
, vma
, start
,
1050 foll_flags
& FOLL_WRITE
);
1052 * The VM_FAULT_WRITE bit tells us that do_wp_page has
1053 * broken COW when necessary, even if maybe_mkwrite
1054 * decided not to set pte_write. We can thus safely do
1055 * subsequent page lookups as if they were reads.
1057 if (ret
& VM_FAULT_WRITE
)
1058 foll_flags
&= ~FOLL_WRITE
;
1060 switch (ret
& ~VM_FAULT_WRITE
) {
1061 case VM_FAULT_MINOR
:
1064 case VM_FAULT_MAJOR
:
1067 case VM_FAULT_SIGBUS
:
1068 return i
? i
: -EFAULT
;
1070 return i
? i
: -ENOMEM
;
1077 flush_dcache_page(page
);
1084 } while (len
&& start
< vma
->vm_end
);
1088 EXPORT_SYMBOL(get_user_pages
);
1090 static int zeromap_pte_range(struct mm_struct
*mm
, pmd_t
*pmd
,
1091 unsigned long addr
, unsigned long end
, pgprot_t prot
)
1096 pte
= pte_alloc_map_lock(mm
, pmd
, addr
, &ptl
);
1100 struct page
*page
= ZERO_PAGE(addr
);
1101 pte_t zero_pte
= pte_wrprotect(mk_pte(page
, prot
));
1102 page_cache_get(page
);
1103 page_add_file_rmap(page
);
1104 inc_mm_counter(mm
, file_rss
);
1105 BUG_ON(!pte_none(*pte
));
1106 set_pte_at(mm
, addr
, pte
, zero_pte
);
1107 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1108 pte_unmap_unlock(pte
- 1, ptl
);
1112 static inline int zeromap_pmd_range(struct mm_struct
*mm
, pud_t
*pud
,
1113 unsigned long addr
, unsigned long end
, pgprot_t prot
)
1118 pmd
= pmd_alloc(mm
, pud
, addr
);
1122 next
= pmd_addr_end(addr
, end
);
1123 if (zeromap_pte_range(mm
, pmd
, addr
, next
, prot
))
1125 } while (pmd
++, addr
= next
, addr
!= end
);
1129 static inline int zeromap_pud_range(struct mm_struct
*mm
, pgd_t
*pgd
,
1130 unsigned long addr
, unsigned long end
, pgprot_t prot
)
1135 pud
= pud_alloc(mm
, pgd
, addr
);
1139 next
= pud_addr_end(addr
, end
);
1140 if (zeromap_pmd_range(mm
, pud
, addr
, next
, prot
))
1142 } while (pud
++, addr
= next
, addr
!= end
);
1146 int zeromap_page_range(struct vm_area_struct
*vma
,
1147 unsigned long addr
, unsigned long size
, pgprot_t prot
)
1151 unsigned long end
= addr
+ size
;
1152 struct mm_struct
*mm
= vma
->vm_mm
;
1155 BUG_ON(addr
>= end
);
1156 pgd
= pgd_offset(mm
, addr
);
1157 flush_cache_range(vma
, addr
, end
);
1159 next
= pgd_addr_end(addr
, end
);
1160 err
= zeromap_pud_range(mm
, pgd
, addr
, next
, prot
);
1163 } while (pgd
++, addr
= next
, addr
!= end
);
1167 pte_t
* fastcall
get_locked_pte(struct mm_struct
*mm
, unsigned long addr
, spinlock_t
**ptl
)
1169 pgd_t
* pgd
= pgd_offset(mm
, addr
);
1170 pud_t
* pud
= pud_alloc(mm
, pgd
, addr
);
1172 pmd_t
* pmd
= pmd_alloc(mm
, pud
, addr
);
1174 return pte_alloc_map_lock(mm
, pmd
, addr
, ptl
);
1180 * This is the old fallback for page remapping.
1182 * For historical reasons, it only allows reserved pages. Only
1183 * old drivers should use this, and they needed to mark their
1184 * pages reserved for the old functions anyway.
1186 static int insert_page(struct mm_struct
*mm
, unsigned long addr
, struct page
*page
, pgprot_t prot
)
1196 flush_dcache_page(page
);
1197 pte
= get_locked_pte(mm
, addr
, &ptl
);
1201 if (!pte_none(*pte
))
1204 /* Ok, finally just insert the thing.. */
1206 inc_mm_counter(mm
, file_rss
);
1207 page_add_file_rmap(page
);
1208 set_pte_at(mm
, addr
, pte
, mk_pte(page
, prot
));
1212 pte_unmap_unlock(pte
, ptl
);
1218 * This allows drivers to insert individual pages they've allocated
1221 * The page has to be a nice clean _individual_ kernel allocation.
1222 * If you allocate a compound page, you need to have marked it as
1223 * such (__GFP_COMP), or manually just split the page up yourself
1224 * (see split_page()).
1226 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1227 * took an arbitrary page protection parameter. This doesn't allow
1228 * that. Your vma protection will have to be set up correctly, which
1229 * means that if you want a shared writable mapping, you'd better
1230 * ask for a shared writable mapping!
1232 * The page does not need to be reserved.
1234 int vm_insert_page(struct vm_area_struct
*vma
, unsigned long addr
, struct page
*page
)
1236 if (addr
< vma
->vm_start
|| addr
>= vma
->vm_end
)
1238 if (!page_count(page
))
1240 vma
->vm_flags
|= VM_INSERTPAGE
;
1241 return insert_page(vma
->vm_mm
, addr
, page
, vma
->vm_page_prot
);
1243 EXPORT_SYMBOL(vm_insert_page
);
1246 * maps a range of physical memory into the requested pages. the old
1247 * mappings are removed. any references to nonexistent pages results
1248 * in null mappings (currently treated as "copy-on-access")
1250 static int remap_pte_range(struct mm_struct
*mm
, pmd_t
*pmd
,
1251 unsigned long addr
, unsigned long end
,
1252 unsigned long pfn
, pgprot_t prot
)
1257 pte
= pte_alloc_map_lock(mm
, pmd
, addr
, &ptl
);
1261 BUG_ON(!pte_none(*pte
));
1262 set_pte_at(mm
, addr
, pte
, pfn_pte(pfn
, prot
));
1264 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1265 pte_unmap_unlock(pte
- 1, ptl
);
1269 static inline int remap_pmd_range(struct mm_struct
*mm
, pud_t
*pud
,
1270 unsigned long addr
, unsigned long end
,
1271 unsigned long pfn
, pgprot_t prot
)
1276 pfn
-= addr
>> PAGE_SHIFT
;
1277 pmd
= pmd_alloc(mm
, pud
, addr
);
1281 next
= pmd_addr_end(addr
, end
);
1282 if (remap_pte_range(mm
, pmd
, addr
, next
,
1283 pfn
+ (addr
>> PAGE_SHIFT
), prot
))
1285 } while (pmd
++, addr
= next
, addr
!= end
);
1289 static inline int remap_pud_range(struct mm_struct
*mm
, pgd_t
*pgd
,
1290 unsigned long addr
, unsigned long end
,
1291 unsigned long pfn
, pgprot_t prot
)
1296 pfn
-= addr
>> PAGE_SHIFT
;
1297 pud
= pud_alloc(mm
, pgd
, addr
);
1301 next
= pud_addr_end(addr
, end
);
1302 if (remap_pmd_range(mm
, pud
, addr
, next
,
1303 pfn
+ (addr
>> PAGE_SHIFT
), prot
))
1305 } while (pud
++, addr
= next
, addr
!= end
);
1309 /* Note: this is only safe if the mm semaphore is held when called. */
1310 int remap_pfn_range(struct vm_area_struct
*vma
, unsigned long addr
,
1311 unsigned long pfn
, unsigned long size
, pgprot_t prot
)
1315 unsigned long end
= addr
+ PAGE_ALIGN(size
);
1316 struct mm_struct
*mm
= vma
->vm_mm
;
1320 * Physically remapped pages are special. Tell the
1321 * rest of the world about it:
1322 * VM_IO tells people not to look at these pages
1323 * (accesses can have side effects).
1324 * VM_RESERVED is specified all over the place, because
1325 * in 2.4 it kept swapout's vma scan off this vma; but
1326 * in 2.6 the LRU scan won't even find its pages, so this
1327 * flag means no more than count its pages in reserved_vm,
1328 * and omit it from core dump, even when VM_IO turned off.
1329 * VM_PFNMAP tells the core MM that the base pages are just
1330 * raw PFN mappings, and do not have a "struct page" associated
1333 * There's a horrible special case to handle copy-on-write
1334 * behaviour that some programs depend on. We mark the "original"
1335 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
1337 if (is_cow_mapping(vma
->vm_flags
)) {
1338 if (addr
!= vma
->vm_start
|| end
!= vma
->vm_end
)
1340 vma
->vm_pgoff
= pfn
;
1343 vma
->vm_flags
|= VM_IO
| VM_RESERVED
| VM_PFNMAP
;
1345 BUG_ON(addr
>= end
);
1346 pfn
-= addr
>> PAGE_SHIFT
;
1347 pgd
= pgd_offset(mm
, addr
);
1348 flush_cache_range(vma
, addr
, end
);
1350 next
= pgd_addr_end(addr
, end
);
1351 err
= remap_pud_range(mm
, pgd
, addr
, next
,
1352 pfn
+ (addr
>> PAGE_SHIFT
), prot
);
1355 } while (pgd
++, addr
= next
, addr
!= end
);
1358 EXPORT_SYMBOL(remap_pfn_range
);
1361 * handle_pte_fault chooses page fault handler according to an entry
1362 * which was read non-atomically. Before making any commitment, on
1363 * those architectures or configurations (e.g. i386 with PAE) which
1364 * might give a mix of unmatched parts, do_swap_page and do_file_page
1365 * must check under lock before unmapping the pte and proceeding
1366 * (but do_wp_page is only called after already making such a check;
1367 * and do_anonymous_page and do_no_page can safely check later on).
1369 static inline int pte_unmap_same(struct mm_struct
*mm
, pmd_t
*pmd
,
1370 pte_t
*page_table
, pte_t orig_pte
)
1373 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1374 if (sizeof(pte_t
) > sizeof(unsigned long)) {
1375 spinlock_t
*ptl
= pte_lockptr(mm
, pmd
);
1377 same
= pte_same(*page_table
, orig_pte
);
1381 pte_unmap(page_table
);
1386 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1387 * servicing faults for write access. In the normal case, do always want
1388 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1389 * that do not have writing enabled, when used by access_process_vm.
1391 static inline pte_t
maybe_mkwrite(pte_t pte
, struct vm_area_struct
*vma
)
1393 if (likely(vma
->vm_flags
& VM_WRITE
))
1394 pte
= pte_mkwrite(pte
);
1398 static inline void cow_user_page(struct page
*dst
, struct page
*src
, unsigned long va
)
1401 * If the source page was a PFN mapping, we don't have
1402 * a "struct page" for it. We do a best-effort copy by
1403 * just copying from the original user address. If that
1404 * fails, we just zero-fill it. Live with it.
1406 if (unlikely(!src
)) {
1407 void *kaddr
= kmap_atomic(dst
, KM_USER0
);
1408 void __user
*uaddr
= (void __user
*)(va
& PAGE_MASK
);
1411 * This really shouldn't fail, because the page is there
1412 * in the page tables. But it might just be unreadable,
1413 * in which case we just give up and fill the result with
1416 if (__copy_from_user_inatomic(kaddr
, uaddr
, PAGE_SIZE
))
1417 memset(kaddr
, 0, PAGE_SIZE
);
1418 kunmap_atomic(kaddr
, KM_USER0
);
1422 copy_user_highpage(dst
, src
, va
);
1426 * This routine handles present pages, when users try to write
1427 * to a shared page. It is done by copying the page to a new address
1428 * and decrementing the shared-page counter for the old page.
1430 * Note that this routine assumes that the protection checks have been
1431 * done by the caller (the low-level page fault routine in most cases).
1432 * Thus we can safely just mark it writable once we've done any necessary
1435 * We also mark the page dirty at this point even though the page will
1436 * change only once the write actually happens. This avoids a few races,
1437 * and potentially makes it more efficient.
1439 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1440 * but allow concurrent faults), with pte both mapped and locked.
1441 * We return with mmap_sem still held, but pte unmapped and unlocked.
1443 static int do_wp_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1444 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
1445 spinlock_t
*ptl
, pte_t orig_pte
)
1447 struct page
*old_page
, *new_page
;
1449 int ret
= VM_FAULT_MINOR
;
1451 old_page
= vm_normal_page(vma
, address
, orig_pte
);
1455 if (PageAnon(old_page
) && !TestSetPageLocked(old_page
)) {
1456 int reuse
= can_share_swap_page(old_page
);
1457 unlock_page(old_page
);
1459 flush_cache_page(vma
, address
, pte_pfn(orig_pte
));
1460 entry
= pte_mkyoung(orig_pte
);
1461 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
1462 ptep_set_access_flags(vma
, address
, page_table
, entry
, 1);
1463 update_mmu_cache(vma
, address
, entry
);
1464 lazy_mmu_prot_update(entry
);
1465 ret
|= VM_FAULT_WRITE
;
1471 * Ok, we need to copy. Oh, well..
1473 page_cache_get(old_page
);
1475 pte_unmap_unlock(page_table
, ptl
);
1477 if (unlikely(anon_vma_prepare(vma
)))
1479 if (old_page
== ZERO_PAGE(address
)) {
1480 new_page
= alloc_zeroed_user_highpage(vma
, address
);
1484 new_page
= alloc_page_vma(GFP_HIGHUSER
, vma
, address
);
1487 cow_user_page(new_page
, old_page
, address
);
1491 * Re-check the pte - we dropped the lock
1493 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
1494 if (likely(pte_same(*page_table
, orig_pte
))) {
1496 page_remove_rmap(old_page
);
1497 if (!PageAnon(old_page
)) {
1498 dec_mm_counter(mm
, file_rss
);
1499 inc_mm_counter(mm
, anon_rss
);
1502 inc_mm_counter(mm
, anon_rss
);
1503 flush_cache_page(vma
, address
, pte_pfn(orig_pte
));
1504 entry
= mk_pte(new_page
, vma
->vm_page_prot
);
1505 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
1506 ptep_establish(vma
, address
, page_table
, entry
);
1507 update_mmu_cache(vma
, address
, entry
);
1508 lazy_mmu_prot_update(entry
);
1509 lru_cache_add_active(new_page
);
1510 page_add_new_anon_rmap(new_page
, vma
, address
);
1512 /* Free the old page.. */
1513 new_page
= old_page
;
1514 ret
|= VM_FAULT_WRITE
;
1517 page_cache_release(new_page
);
1519 page_cache_release(old_page
);
1521 pte_unmap_unlock(page_table
, ptl
);
1525 page_cache_release(old_page
);
1526 return VM_FAULT_OOM
;
1530 * Helper functions for unmap_mapping_range().
1532 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1534 * We have to restart searching the prio_tree whenever we drop the lock,
1535 * since the iterator is only valid while the lock is held, and anyway
1536 * a later vma might be split and reinserted earlier while lock dropped.
1538 * The list of nonlinear vmas could be handled more efficiently, using
1539 * a placeholder, but handle it in the same way until a need is shown.
1540 * It is important to search the prio_tree before nonlinear list: a vma
1541 * may become nonlinear and be shifted from prio_tree to nonlinear list
1542 * while the lock is dropped; but never shifted from list to prio_tree.
1544 * In order to make forward progress despite restarting the search,
1545 * vm_truncate_count is used to mark a vma as now dealt with, so we can
1546 * quickly skip it next time around. Since the prio_tree search only
1547 * shows us those vmas affected by unmapping the range in question, we
1548 * can't efficiently keep all vmas in step with mapping->truncate_count:
1549 * so instead reset them all whenever it wraps back to 0 (then go to 1).
1550 * mapping->truncate_count and vma->vm_truncate_count are protected by
1553 * In order to make forward progress despite repeatedly restarting some
1554 * large vma, note the restart_addr from unmap_vmas when it breaks out:
1555 * and restart from that address when we reach that vma again. It might
1556 * have been split or merged, shrunk or extended, but never shifted: so
1557 * restart_addr remains valid so long as it remains in the vma's range.
1558 * unmap_mapping_range forces truncate_count to leap over page-aligned
1559 * values so we can save vma's restart_addr in its truncate_count field.
1561 #define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
1563 static void reset_vma_truncate_counts(struct address_space
*mapping
)
1565 struct vm_area_struct
*vma
;
1566 struct prio_tree_iter iter
;
1568 vma_prio_tree_foreach(vma
, &iter
, &mapping
->i_mmap
, 0, ULONG_MAX
)
1569 vma
->vm_truncate_count
= 0;
1570 list_for_each_entry(vma
, &mapping
->i_mmap_nonlinear
, shared
.vm_set
.list
)
1571 vma
->vm_truncate_count
= 0;
1574 static int unmap_mapping_range_vma(struct vm_area_struct
*vma
,
1575 unsigned long start_addr
, unsigned long end_addr
,
1576 struct zap_details
*details
)
1578 unsigned long restart_addr
;
1582 restart_addr
= vma
->vm_truncate_count
;
1583 if (is_restart_addr(restart_addr
) && start_addr
< restart_addr
) {
1584 start_addr
= restart_addr
;
1585 if (start_addr
>= end_addr
) {
1586 /* Top of vma has been split off since last time */
1587 vma
->vm_truncate_count
= details
->truncate_count
;
1592 restart_addr
= zap_page_range(vma
, start_addr
,
1593 end_addr
- start_addr
, details
);
1594 need_break
= need_resched() ||
1595 need_lockbreak(details
->i_mmap_lock
);
1597 if (restart_addr
>= end_addr
) {
1598 /* We have now completed this vma: mark it so */
1599 vma
->vm_truncate_count
= details
->truncate_count
;
1603 /* Note restart_addr in vma's truncate_count field */
1604 vma
->vm_truncate_count
= restart_addr
;
1609 spin_unlock(details
->i_mmap_lock
);
1611 spin_lock(details
->i_mmap_lock
);
1615 static inline void unmap_mapping_range_tree(struct prio_tree_root
*root
,
1616 struct zap_details
*details
)
1618 struct vm_area_struct
*vma
;
1619 struct prio_tree_iter iter
;
1620 pgoff_t vba
, vea
, zba
, zea
;
1623 vma_prio_tree_foreach(vma
, &iter
, root
,
1624 details
->first_index
, details
->last_index
) {
1625 /* Skip quickly over those we have already dealt with */
1626 if (vma
->vm_truncate_count
== details
->truncate_count
)
1629 vba
= vma
->vm_pgoff
;
1630 vea
= vba
+ ((vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
) - 1;
1631 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1632 zba
= details
->first_index
;
1635 zea
= details
->last_index
;
1639 if (unmap_mapping_range_vma(vma
,
1640 ((zba
- vba
) << PAGE_SHIFT
) + vma
->vm_start
,
1641 ((zea
- vba
+ 1) << PAGE_SHIFT
) + vma
->vm_start
,
1647 static inline void unmap_mapping_range_list(struct list_head
*head
,
1648 struct zap_details
*details
)
1650 struct vm_area_struct
*vma
;
1653 * In nonlinear VMAs there is no correspondence between virtual address
1654 * offset and file offset. So we must perform an exhaustive search
1655 * across *all* the pages in each nonlinear VMA, not just the pages
1656 * whose virtual address lies outside the file truncation point.
1659 list_for_each_entry(vma
, head
, shared
.vm_set
.list
) {
1660 /* Skip quickly over those we have already dealt with */
1661 if (vma
->vm_truncate_count
== details
->truncate_count
)
1663 details
->nonlinear_vma
= vma
;
1664 if (unmap_mapping_range_vma(vma
, vma
->vm_start
,
1665 vma
->vm_end
, details
) < 0)
1671 * unmap_mapping_range - unmap the portion of all mmaps
1672 * in the specified address_space corresponding to the specified
1673 * page range in the underlying file.
1674 * @mapping: the address space containing mmaps to be unmapped.
1675 * @holebegin: byte in first page to unmap, relative to the start of
1676 * the underlying file. This will be rounded down to a PAGE_SIZE
1677 * boundary. Note that this is different from vmtruncate(), which
1678 * must keep the partial page. In contrast, we must get rid of
1680 * @holelen: size of prospective hole in bytes. This will be rounded
1681 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
1683 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
1684 * but 0 when invalidating pagecache, don't throw away private data.
1686 void unmap_mapping_range(struct address_space
*mapping
,
1687 loff_t
const holebegin
, loff_t
const holelen
, int even_cows
)
1689 struct zap_details details
;
1690 pgoff_t hba
= holebegin
>> PAGE_SHIFT
;
1691 pgoff_t hlen
= (holelen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1693 /* Check for overflow. */
1694 if (sizeof(holelen
) > sizeof(hlen
)) {
1696 (holebegin
+ holelen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1697 if (holeend
& ~(long long)ULONG_MAX
)
1698 hlen
= ULONG_MAX
- hba
+ 1;
1701 details
.check_mapping
= even_cows
? NULL
: mapping
;
1702 details
.nonlinear_vma
= NULL
;
1703 details
.first_index
= hba
;
1704 details
.last_index
= hba
+ hlen
- 1;
1705 if (details
.last_index
< details
.first_index
)
1706 details
.last_index
= ULONG_MAX
;
1707 details
.i_mmap_lock
= &mapping
->i_mmap_lock
;
1709 spin_lock(&mapping
->i_mmap_lock
);
1711 /* serialize i_size write against truncate_count write */
1713 /* Protect against page faults, and endless unmapping loops */
1714 mapping
->truncate_count
++;
1716 * For archs where spin_lock has inclusive semantics like ia64
1717 * this smp_mb() will prevent to read pagetable contents
1718 * before the truncate_count increment is visible to
1722 if (unlikely(is_restart_addr(mapping
->truncate_count
))) {
1723 if (mapping
->truncate_count
== 0)
1724 reset_vma_truncate_counts(mapping
);
1725 mapping
->truncate_count
++;
1727 details
.truncate_count
= mapping
->truncate_count
;
1729 if (unlikely(!prio_tree_empty(&mapping
->i_mmap
)))
1730 unmap_mapping_range_tree(&mapping
->i_mmap
, &details
);
1731 if (unlikely(!list_empty(&mapping
->i_mmap_nonlinear
)))
1732 unmap_mapping_range_list(&mapping
->i_mmap_nonlinear
, &details
);
1733 spin_unlock(&mapping
->i_mmap_lock
);
1735 EXPORT_SYMBOL(unmap_mapping_range
);
1738 * Handle all mappings that got truncated by a "truncate()"
1741 * NOTE! We have to be ready to update the memory sharing
1742 * between the file and the memory map for a potential last
1743 * incomplete page. Ugly, but necessary.
1745 int vmtruncate(struct inode
* inode
, loff_t offset
)
1747 struct address_space
*mapping
= inode
->i_mapping
;
1748 unsigned long limit
;
1750 if (inode
->i_size
< offset
)
1753 * truncation of in-use swapfiles is disallowed - it would cause
1754 * subsequent swapout to scribble on the now-freed blocks.
1756 if (IS_SWAPFILE(inode
))
1758 i_size_write(inode
, offset
);
1759 unmap_mapping_range(mapping
, offset
+ PAGE_SIZE
- 1, 0, 1);
1760 truncate_inode_pages(mapping
, offset
);
1764 limit
= current
->signal
->rlim
[RLIMIT_FSIZE
].rlim_cur
;
1765 if (limit
!= RLIM_INFINITY
&& offset
> limit
)
1767 if (offset
> inode
->i_sb
->s_maxbytes
)
1769 i_size_write(inode
, offset
);
1772 if (inode
->i_op
&& inode
->i_op
->truncate
)
1773 inode
->i_op
->truncate(inode
);
1776 send_sig(SIGXFSZ
, current
, 0);
1782 EXPORT_SYMBOL(vmtruncate
);
1784 int vmtruncate_range(struct inode
*inode
, loff_t offset
, loff_t end
)
1786 struct address_space
*mapping
= inode
->i_mapping
;
1789 * If the underlying filesystem is not going to provide
1790 * a way to truncate a range of blocks (punch a hole) -
1791 * we should return failure right now.
1793 if (!inode
->i_op
|| !inode
->i_op
->truncate_range
)
1796 mutex_lock(&inode
->i_mutex
);
1797 down_write(&inode
->i_alloc_sem
);
1798 unmap_mapping_range(mapping
, offset
, (end
- offset
), 1);
1799 truncate_inode_pages_range(mapping
, offset
, end
);
1800 inode
->i_op
->truncate_range(inode
, offset
, end
);
1801 up_write(&inode
->i_alloc_sem
);
1802 mutex_unlock(&inode
->i_mutex
);
1806 EXPORT_SYMBOL(vmtruncate_range
);
1809 * Primitive swap readahead code. We simply read an aligned block of
1810 * (1 << page_cluster) entries in the swap area. This method is chosen
1811 * because it doesn't cost us any seek time. We also make sure to queue
1812 * the 'original' request together with the readahead ones...
1814 * This has been extended to use the NUMA policies from the mm triggering
1817 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
1819 void swapin_readahead(swp_entry_t entry
, unsigned long addr
,struct vm_area_struct
*vma
)
1822 struct vm_area_struct
*next_vma
= vma
? vma
->vm_next
: NULL
;
1825 struct page
*new_page
;
1826 unsigned long offset
;
1829 * Get the number of handles we should do readahead io to.
1831 num
= valid_swaphandles(entry
, &offset
);
1832 for (i
= 0; i
< num
; offset
++, i
++) {
1833 /* Ok, do the async read-ahead now */
1834 new_page
= read_swap_cache_async(swp_entry(swp_type(entry
),
1835 offset
), vma
, addr
);
1838 page_cache_release(new_page
);
1841 * Find the next applicable VMA for the NUMA policy.
1847 if (addr
>= vma
->vm_end
) {
1849 next_vma
= vma
? vma
->vm_next
: NULL
;
1851 if (vma
&& addr
< vma
->vm_start
)
1854 if (next_vma
&& addr
>= next_vma
->vm_start
) {
1856 next_vma
= vma
->vm_next
;
1861 lru_add_drain(); /* Push any new pages onto the LRU now */
1865 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1866 * but allow concurrent faults), and pte mapped but not yet locked.
1867 * We return with mmap_sem still held, but pte unmapped and unlocked.
1869 static int do_swap_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1870 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
1871 int write_access
, pte_t orig_pte
)
1877 int ret
= VM_FAULT_MINOR
;
1879 if (!pte_unmap_same(mm
, pmd
, page_table
, orig_pte
))
1882 entry
= pte_to_swp_entry(orig_pte
);
1884 page
= lookup_swap_cache(entry
);
1886 swapin_readahead(entry
, address
, vma
);
1887 page
= read_swap_cache_async(entry
, vma
, address
);
1890 * Back out if somebody else faulted in this pte
1891 * while we released the pte lock.
1893 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
1894 if (likely(pte_same(*page_table
, orig_pte
)))
1899 /* Had to read the page from swap area: Major fault */
1900 ret
= VM_FAULT_MAJOR
;
1901 inc_page_state(pgmajfault
);
1905 mark_page_accessed(page
);
1907 if (!PageSwapCache(page
)) {
1908 /* Page migration has occured */
1910 page_cache_release(page
);
1915 * Back out if somebody else already faulted in this pte.
1917 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
1918 if (unlikely(!pte_same(*page_table
, orig_pte
)))
1921 if (unlikely(!PageUptodate(page
))) {
1922 ret
= VM_FAULT_SIGBUS
;
1926 /* The page isn't present yet, go ahead with the fault. */
1928 inc_mm_counter(mm
, anon_rss
);
1929 pte
= mk_pte(page
, vma
->vm_page_prot
);
1930 if (write_access
&& can_share_swap_page(page
)) {
1931 pte
= maybe_mkwrite(pte_mkdirty(pte
), vma
);
1935 flush_icache_page(vma
, page
);
1936 set_pte_at(mm
, address
, page_table
, pte
);
1937 page_add_anon_rmap(page
, vma
, address
);
1941 remove_exclusive_swap_page(page
);
1945 if (do_wp_page(mm
, vma
, address
,
1946 page_table
, pmd
, ptl
, pte
) == VM_FAULT_OOM
)
1951 /* No need to invalidate - it was non-present before */
1952 update_mmu_cache(vma
, address
, pte
);
1953 lazy_mmu_prot_update(pte
);
1955 pte_unmap_unlock(page_table
, ptl
);
1959 pte_unmap_unlock(page_table
, ptl
);
1961 page_cache_release(page
);
1966 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1967 * but allow concurrent faults), and pte mapped but not yet locked.
1968 * We return with mmap_sem still held, but pte unmapped and unlocked.
1970 static int do_anonymous_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1971 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
1979 /* Allocate our own private page. */
1980 pte_unmap(page_table
);
1982 if (unlikely(anon_vma_prepare(vma
)))
1984 page
= alloc_zeroed_user_highpage(vma
, address
);
1988 entry
= mk_pte(page
, vma
->vm_page_prot
);
1989 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
1991 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
1992 if (!pte_none(*page_table
))
1994 inc_mm_counter(mm
, anon_rss
);
1995 lru_cache_add_active(page
);
1996 page_add_new_anon_rmap(page
, vma
, address
);
1998 /* Map the ZERO_PAGE - vm_page_prot is readonly */
1999 page
= ZERO_PAGE(address
);
2000 page_cache_get(page
);
2001 entry
= mk_pte(page
, vma
->vm_page_prot
);
2003 ptl
= pte_lockptr(mm
, pmd
);
2005 if (!pte_none(*page_table
))
2007 inc_mm_counter(mm
, file_rss
);
2008 page_add_file_rmap(page
);
2011 set_pte_at(mm
, address
, page_table
, entry
);
2013 /* No need to invalidate - it was non-present before */
2014 update_mmu_cache(vma
, address
, entry
);
2015 lazy_mmu_prot_update(entry
);
2017 pte_unmap_unlock(page_table
, ptl
);
2018 return VM_FAULT_MINOR
;
2020 page_cache_release(page
);
2023 return VM_FAULT_OOM
;
2027 * do_no_page() tries to create a new page mapping. It aggressively
2028 * tries to share with existing pages, but makes a separate copy if
2029 * the "write_access" parameter is true in order to avoid the next
2032 * As this is called only for pages that do not currently exist, we
2033 * do not need to flush old virtual caches or the TLB.
2035 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2036 * but allow concurrent faults), and pte mapped but not yet locked.
2037 * We return with mmap_sem still held, but pte unmapped and unlocked.
2039 static int do_no_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2040 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
2044 struct page
*new_page
;
2045 struct address_space
*mapping
= NULL
;
2047 unsigned int sequence
= 0;
2048 int ret
= VM_FAULT_MINOR
;
2051 pte_unmap(page_table
);
2052 BUG_ON(vma
->vm_flags
& VM_PFNMAP
);
2055 mapping
= vma
->vm_file
->f_mapping
;
2056 sequence
= mapping
->truncate_count
;
2057 smp_rmb(); /* serializes i_size against truncate_count */
2060 new_page
= vma
->vm_ops
->nopage(vma
, address
& PAGE_MASK
, &ret
);
2062 * No smp_rmb is needed here as long as there's a full
2063 * spin_lock/unlock sequence inside the ->nopage callback
2064 * (for the pagecache lookup) that acts as an implicit
2065 * smp_mb() and prevents the i_size read to happen
2066 * after the next truncate_count read.
2069 /* no page was available -- either SIGBUS or OOM */
2070 if (new_page
== NOPAGE_SIGBUS
)
2071 return VM_FAULT_SIGBUS
;
2072 if (new_page
== NOPAGE_OOM
)
2073 return VM_FAULT_OOM
;
2076 * Should we do an early C-O-W break?
2078 if (write_access
&& !(vma
->vm_flags
& VM_SHARED
)) {
2081 if (unlikely(anon_vma_prepare(vma
)))
2083 page
= alloc_page_vma(GFP_HIGHUSER
, vma
, address
);
2086 copy_user_highpage(page
, new_page
, address
);
2087 page_cache_release(new_page
);
2092 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
2094 * For a file-backed vma, someone could have truncated or otherwise
2095 * invalidated this page. If unmap_mapping_range got called,
2096 * retry getting the page.
2098 if (mapping
&& unlikely(sequence
!= mapping
->truncate_count
)) {
2099 pte_unmap_unlock(page_table
, ptl
);
2100 page_cache_release(new_page
);
2102 sequence
= mapping
->truncate_count
;
2108 * This silly early PAGE_DIRTY setting removes a race
2109 * due to the bad i386 page protection. But it's valid
2110 * for other architectures too.
2112 * Note that if write_access is true, we either now have
2113 * an exclusive copy of the page, or this is a shared mapping,
2114 * so we can make it writable and dirty to avoid having to
2115 * handle that later.
2117 /* Only go through if we didn't race with anybody else... */
2118 if (pte_none(*page_table
)) {
2119 flush_icache_page(vma
, new_page
);
2120 entry
= mk_pte(new_page
, vma
->vm_page_prot
);
2122 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
2123 set_pte_at(mm
, address
, page_table
, entry
);
2125 inc_mm_counter(mm
, anon_rss
);
2126 lru_cache_add_active(new_page
);
2127 page_add_new_anon_rmap(new_page
, vma
, address
);
2129 inc_mm_counter(mm
, file_rss
);
2130 page_add_file_rmap(new_page
);
2133 /* One of our sibling threads was faster, back out. */
2134 page_cache_release(new_page
);
2138 /* no need to invalidate: a not-present page shouldn't be cached */
2139 update_mmu_cache(vma
, address
, entry
);
2140 lazy_mmu_prot_update(entry
);
2142 pte_unmap_unlock(page_table
, ptl
);
2145 page_cache_release(new_page
);
2146 return VM_FAULT_OOM
;
2150 * Fault of a previously existing named mapping. Repopulate the pte
2151 * from the encoded file_pte if possible. This enables swappable
2154 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2155 * but allow concurrent faults), and pte mapped but not yet locked.
2156 * We return with mmap_sem still held, but pte unmapped and unlocked.
2158 static int do_file_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2159 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
2160 int write_access
, pte_t orig_pte
)
2165 if (!pte_unmap_same(mm
, pmd
, page_table
, orig_pte
))
2166 return VM_FAULT_MINOR
;
2168 if (unlikely(!(vma
->vm_flags
& VM_NONLINEAR
))) {
2170 * Page table corrupted: show pte and kill process.
2172 print_bad_pte(vma
, orig_pte
, address
);
2173 return VM_FAULT_OOM
;
2175 /* We can then assume vm->vm_ops && vma->vm_ops->populate */
2177 pgoff
= pte_to_pgoff(orig_pte
);
2178 err
= vma
->vm_ops
->populate(vma
, address
& PAGE_MASK
, PAGE_SIZE
,
2179 vma
->vm_page_prot
, pgoff
, 0);
2181 return VM_FAULT_OOM
;
2183 return VM_FAULT_SIGBUS
;
2184 return VM_FAULT_MAJOR
;
2188 * These routines also need to handle stuff like marking pages dirty
2189 * and/or accessed for architectures that don't do it in hardware (most
2190 * RISC architectures). The early dirtying is also good on the i386.
2192 * There is also a hook called "update_mmu_cache()" that architectures
2193 * with external mmu caches can use to update those (ie the Sparc or
2194 * PowerPC hashed page tables that act as extended TLBs).
2196 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2197 * but allow concurrent faults), and pte mapped but not yet locked.
2198 * We return with mmap_sem still held, but pte unmapped and unlocked.
2200 static inline int handle_pte_fault(struct mm_struct
*mm
,
2201 struct vm_area_struct
*vma
, unsigned long address
,
2202 pte_t
*pte
, pmd_t
*pmd
, int write_access
)
2208 old_entry
= entry
= *pte
;
2209 if (!pte_present(entry
)) {
2210 if (pte_none(entry
)) {
2211 if (!vma
->vm_ops
|| !vma
->vm_ops
->nopage
)
2212 return do_anonymous_page(mm
, vma
, address
,
2213 pte
, pmd
, write_access
);
2214 return do_no_page(mm
, vma
, address
,
2215 pte
, pmd
, write_access
);
2217 if (pte_file(entry
))
2218 return do_file_page(mm
, vma
, address
,
2219 pte
, pmd
, write_access
, entry
);
2220 return do_swap_page(mm
, vma
, address
,
2221 pte
, pmd
, write_access
, entry
);
2224 ptl
= pte_lockptr(mm
, pmd
);
2226 if (unlikely(!pte_same(*pte
, entry
)))
2229 if (!pte_write(entry
))
2230 return do_wp_page(mm
, vma
, address
,
2231 pte
, pmd
, ptl
, entry
);
2232 entry
= pte_mkdirty(entry
);
2234 entry
= pte_mkyoung(entry
);
2235 if (!pte_same(old_entry
, entry
)) {
2236 ptep_set_access_flags(vma
, address
, pte
, entry
, write_access
);
2237 update_mmu_cache(vma
, address
, entry
);
2238 lazy_mmu_prot_update(entry
);
2241 * This is needed only for protection faults but the arch code
2242 * is not yet telling us if this is a protection fault or not.
2243 * This still avoids useless tlb flushes for .text page faults
2247 flush_tlb_page(vma
, address
);
2250 pte_unmap_unlock(pte
, ptl
);
2251 return VM_FAULT_MINOR
;
2255 * By the time we get here, we already hold the mm semaphore
2257 int __handle_mm_fault(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2258 unsigned long address
, int write_access
)
2265 __set_current_state(TASK_RUNNING
);
2267 inc_page_state(pgfault
);
2269 if (unlikely(is_vm_hugetlb_page(vma
)))
2270 return hugetlb_fault(mm
, vma
, address
, write_access
);
2272 pgd
= pgd_offset(mm
, address
);
2273 pud
= pud_alloc(mm
, pgd
, address
);
2275 return VM_FAULT_OOM
;
2276 pmd
= pmd_alloc(mm
, pud
, address
);
2278 return VM_FAULT_OOM
;
2279 pte
= pte_alloc_map(mm
, pmd
, address
);
2281 return VM_FAULT_OOM
;
2283 return handle_pte_fault(mm
, vma
, address
, pte
, pmd
, write_access
);
2286 EXPORT_SYMBOL_GPL(__handle_mm_fault
);
2288 #ifndef __PAGETABLE_PUD_FOLDED
2290 * Allocate page upper directory.
2291 * We've already handled the fast-path in-line.
2293 int __pud_alloc(struct mm_struct
*mm
, pgd_t
*pgd
, unsigned long address
)
2295 pud_t
*new = pud_alloc_one(mm
, address
);
2299 spin_lock(&mm
->page_table_lock
);
2300 if (pgd_present(*pgd
)) /* Another has populated it */
2303 pgd_populate(mm
, pgd
, new);
2304 spin_unlock(&mm
->page_table_lock
);
2308 /* Workaround for gcc 2.96 */
2309 int __pud_alloc(struct mm_struct
*mm
, pgd_t
*pgd
, unsigned long address
)
2313 #endif /* __PAGETABLE_PUD_FOLDED */
2315 #ifndef __PAGETABLE_PMD_FOLDED
2317 * Allocate page middle directory.
2318 * We've already handled the fast-path in-line.
2320 int __pmd_alloc(struct mm_struct
*mm
, pud_t
*pud
, unsigned long address
)
2322 pmd_t
*new = pmd_alloc_one(mm
, address
);
2326 spin_lock(&mm
->page_table_lock
);
2327 #ifndef __ARCH_HAS_4LEVEL_HACK
2328 if (pud_present(*pud
)) /* Another has populated it */
2331 pud_populate(mm
, pud
, new);
2333 if (pgd_present(*pud
)) /* Another has populated it */
2336 pgd_populate(mm
, pud
, new);
2337 #endif /* __ARCH_HAS_4LEVEL_HACK */
2338 spin_unlock(&mm
->page_table_lock
);
2342 /* Workaround for gcc 2.96 */
2343 int __pmd_alloc(struct mm_struct
*mm
, pud_t
*pud
, unsigned long address
)
2347 #endif /* __PAGETABLE_PMD_FOLDED */
2349 int make_pages_present(unsigned long addr
, unsigned long end
)
2351 int ret
, len
, write
;
2352 struct vm_area_struct
* vma
;
2354 vma
= find_vma(current
->mm
, addr
);
2357 write
= (vma
->vm_flags
& VM_WRITE
) != 0;
2360 if (end
> vma
->vm_end
)
2362 len
= (end
+PAGE_SIZE
-1)/PAGE_SIZE
-addr
/PAGE_SIZE
;
2363 ret
= get_user_pages(current
, current
->mm
, addr
,
2364 len
, write
, 0, NULL
, NULL
);
2367 return ret
== len
? 0 : -1;
2371 * Map a vmalloc()-space virtual address to the physical page.
2373 struct page
* vmalloc_to_page(void * vmalloc_addr
)
2375 unsigned long addr
= (unsigned long) vmalloc_addr
;
2376 struct page
*page
= NULL
;
2377 pgd_t
*pgd
= pgd_offset_k(addr
);
2382 if (!pgd_none(*pgd
)) {
2383 pud
= pud_offset(pgd
, addr
);
2384 if (!pud_none(*pud
)) {
2385 pmd
= pmd_offset(pud
, addr
);
2386 if (!pmd_none(*pmd
)) {
2387 ptep
= pte_offset_map(pmd
, addr
);
2389 if (pte_present(pte
))
2390 page
= pte_page(pte
);
2398 EXPORT_SYMBOL(vmalloc_to_page
);
2401 * Map a vmalloc()-space virtual address to the physical page frame number.
2403 unsigned long vmalloc_to_pfn(void * vmalloc_addr
)
2405 return page_to_pfn(vmalloc_to_page(vmalloc_addr
));
2408 EXPORT_SYMBOL(vmalloc_to_pfn
);
2410 #if !defined(__HAVE_ARCH_GATE_AREA)
2412 #if defined(AT_SYSINFO_EHDR)
2413 static struct vm_area_struct gate_vma
;
2415 static int __init
gate_vma_init(void)
2417 gate_vma
.vm_mm
= NULL
;
2418 gate_vma
.vm_start
= FIXADDR_USER_START
;
2419 gate_vma
.vm_end
= FIXADDR_USER_END
;
2420 gate_vma
.vm_page_prot
= PAGE_READONLY
;
2421 gate_vma
.vm_flags
= 0;
2424 __initcall(gate_vma_init
);
2427 struct vm_area_struct
*get_gate_vma(struct task_struct
*tsk
)
2429 #ifdef AT_SYSINFO_EHDR
2436 int in_gate_area_no_task(unsigned long addr
)
2438 #ifdef AT_SYSINFO_EHDR
2439 if ((addr
>= FIXADDR_USER_START
) && (addr
< FIXADDR_USER_END
))
2445 #endif /* __HAVE_ARCH_GATE_AREA */