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)
39 #include <linux/kernel_stat.h>
41 #include <linux/hugetlb.h>
42 #include <linux/mman.h>
43 #include <linux/swap.h>
44 #include <linux/highmem.h>
45 #include <linux/pagemap.h>
46 #include <linux/rmap.h>
47 #include <linux/module.h>
48 #include <linux/init.h>
50 #include <asm/pgalloc.h>
51 #include <asm/uaccess.h>
53 #include <asm/tlbflush.h>
54 #include <asm/pgtable.h>
56 #include <linux/swapops.h>
57 #include <linux/elf.h>
59 #ifndef CONFIG_DISCONTIGMEM
60 /* use the per-pgdat data instead for discontigmem - mbligh */
61 unsigned long max_mapnr
;
64 EXPORT_SYMBOL(max_mapnr
);
65 EXPORT_SYMBOL(mem_map
);
68 unsigned long num_physpages
;
70 * A number of key systems in x86 including ioremap() rely on the assumption
71 * that high_memory defines the upper bound on direct map memory, then end
72 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
73 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
77 struct page
*highmem_start_page
;
78 unsigned long vmalloc_earlyreserve
;
80 EXPORT_SYMBOL(num_physpages
);
81 EXPORT_SYMBOL(highmem_start_page
);
82 EXPORT_SYMBOL(high_memory
);
83 EXPORT_SYMBOL(vmalloc_earlyreserve
);
86 * We special-case the C-O-W ZERO_PAGE, because it's such
87 * a common occurrence (no need to read the page to know
88 * that it's zero - better for the cache and memory subsystem).
90 static inline void copy_cow_page(struct page
* from
, struct page
* to
, unsigned long address
)
92 if (from
== ZERO_PAGE(address
)) {
93 clear_user_highpage(to
, address
);
96 copy_user_highpage(to
, from
, address
);
100 * Note: this doesn't free the actual pages themselves. That
101 * has been handled earlier when unmapping all the memory regions.
103 static inline void free_one_pmd(struct mmu_gather
*tlb
, pmd_t
* dir
)
109 if (unlikely(pmd_bad(*dir
))) {
114 page
= pmd_page(*dir
);
116 dec_page_state(nr_page_table_pages
);
118 pte_free_tlb(tlb
, page
);
121 static inline void free_one_pgd(struct mmu_gather
*tlb
, pgd_t
* dir
)
128 if (unlikely(pgd_bad(*dir
))) {
133 pmd
= pmd_offset(dir
, 0);
135 for (j
= 0; j
< PTRS_PER_PMD
; j
++)
136 free_one_pmd(tlb
, pmd
+j
);
137 pmd_free_tlb(tlb
, pmd
);
141 * This function clears all user-level page tables of a process - this
142 * is needed by execve(), so that old pages aren't in the way.
144 * Must be called with pagetable lock held.
146 void clear_page_tables(struct mmu_gather
*tlb
, unsigned long first
, int nr
)
148 pgd_t
* page_dir
= tlb
->mm
->pgd
;
152 free_one_pgd(tlb
, page_dir
);
157 pte_t fastcall
* pte_alloc_map(struct mm_struct
*mm
, pmd_t
*pmd
, unsigned long address
)
159 if (!pmd_present(*pmd
)) {
162 spin_unlock(&mm
->page_table_lock
);
163 new = pte_alloc_one(mm
, address
);
164 spin_lock(&mm
->page_table_lock
);
168 * Because we dropped the lock, we should re-check the
169 * entry, as somebody else could have populated it..
171 if (pmd_present(*pmd
)) {
176 inc_page_state(nr_page_table_pages
);
177 pmd_populate(mm
, pmd
, new);
180 return pte_offset_map(pmd
, address
);
183 pte_t fastcall
* pte_alloc_kernel(struct mm_struct
*mm
, pmd_t
*pmd
, unsigned long address
)
185 if (!pmd_present(*pmd
)) {
188 spin_unlock(&mm
->page_table_lock
);
189 new = pte_alloc_one_kernel(mm
, address
);
190 spin_lock(&mm
->page_table_lock
);
195 * Because we dropped the lock, we should re-check the
196 * entry, as somebody else could have populated it..
198 if (pmd_present(*pmd
)) {
199 pte_free_kernel(new);
202 pmd_populate_kernel(mm
, pmd
, new);
205 return pte_offset_kernel(pmd
, address
);
207 #define PTE_TABLE_MASK ((PTRS_PER_PTE-1) * sizeof(pte_t))
208 #define PMD_TABLE_MASK ((PTRS_PER_PMD-1) * sizeof(pmd_t))
211 * copy one vm_area from one task to the other. Assumes the page tables
212 * already present in the new task to be cleared in the whole range
213 * covered by this vma.
215 * 08Jan98 Merged into one routine from several inline routines to reduce
216 * variable count and make things faster. -jj
218 * dst->page_table_lock is held on entry and exit,
219 * but may be dropped within pmd_alloc() and pte_alloc_map().
221 int copy_page_range(struct mm_struct
*dst
, struct mm_struct
*src
,
222 struct vm_area_struct
*vma
)
224 pgd_t
* src_pgd
, * dst_pgd
;
225 unsigned long address
= vma
->vm_start
;
226 unsigned long end
= vma
->vm_end
;
229 if (is_vm_hugetlb_page(vma
))
230 return copy_hugetlb_page_range(dst
, src
, vma
);
232 cow
= (vma
->vm_flags
& (VM_SHARED
| VM_MAYWRITE
)) == VM_MAYWRITE
;
233 src_pgd
= pgd_offset(src
, address
)-1;
234 dst_pgd
= pgd_offset(dst
, address
)-1;
237 pmd_t
* src_pmd
, * dst_pmd
;
239 src_pgd
++; dst_pgd
++;
243 if (pgd_none(*src_pgd
))
244 goto skip_copy_pmd_range
;
245 if (unlikely(pgd_bad(*src_pgd
))) {
248 skip_copy_pmd_range
: address
= (address
+ PGDIR_SIZE
) & PGDIR_MASK
;
249 if (!address
|| (address
>= end
))
254 src_pmd
= pmd_offset(src_pgd
, address
);
255 dst_pmd
= pmd_alloc(dst
, dst_pgd
, address
);
260 pte_t
* src_pte
, * dst_pte
;
264 if (pmd_none(*src_pmd
))
265 goto skip_copy_pte_range
;
266 if (unlikely(pmd_bad(*src_pmd
))) {
270 address
= (address
+ PMD_SIZE
) & PMD_MASK
;
273 goto cont_copy_pmd_range
;
276 dst_pte
= pte_alloc_map(dst
, dst_pmd
, address
);
279 spin_lock(&src
->page_table_lock
);
280 src_pte
= pte_offset_map_nested(src_pmd
, address
);
282 pte_t pte
= *src_pte
;
289 goto cont_copy_pte_range_noset
;
290 /* pte contains position in swap, so copy. */
291 if (!pte_present(pte
)) {
292 if (!pte_file(pte
)) {
293 swap_duplicate(pte_to_swp_entry(pte
));
294 if (list_empty(&dst
->mmlist
)) {
295 spin_lock(&mmlist_lock
);
296 list_add(&dst
->mmlist
,
298 spin_unlock(&mmlist_lock
);
301 set_pte(dst_pte
, pte
);
302 goto cont_copy_pte_range_noset
;
305 /* the pte points outside of valid memory, the
306 * mapping is assumed to be good, meaningful
307 * and not mapped via rmap - duplicate the
312 page
= pfn_to_page(pfn
);
314 if (!page
|| PageReserved(page
)) {
315 set_pte(dst_pte
, pte
);
316 goto cont_copy_pte_range_noset
;
320 * If it's a COW mapping, write protect it both
321 * in the parent and the child
324 ptep_set_wrprotect(src_pte
);
329 * If it's a shared mapping, mark it clean in
332 if (vma
->vm_flags
& VM_SHARED
)
333 pte
= pte_mkclean(pte
);
334 pte
= pte_mkold(pte
);
337 set_pte(dst_pte
, pte
);
339 cont_copy_pte_range_noset
:
340 address
+= PAGE_SIZE
;
341 if (address
>= end
) {
342 pte_unmap_nested(src_pte
);
348 } while ((unsigned long)src_pte
& PTE_TABLE_MASK
);
349 pte_unmap_nested(src_pte
-1);
350 pte_unmap(dst_pte
-1);
351 spin_unlock(&src
->page_table_lock
);
352 cond_resched_lock(&dst
->page_table_lock
);
356 } while ((unsigned long)src_pmd
& PMD_TABLE_MASK
);
359 spin_unlock(&src
->page_table_lock
);
366 static void zap_pte_range(struct mmu_gather
*tlb
,
367 pmd_t
*pmd
, unsigned long address
,
368 unsigned long size
, struct zap_details
*details
)
370 unsigned long offset
;
375 if (unlikely(pmd_bad(*pmd
))) {
380 ptep
= pte_offset_map(pmd
, address
);
381 offset
= address
& ~PMD_MASK
;
382 if (offset
+ size
> PMD_SIZE
)
383 size
= PMD_SIZE
- offset
;
385 if (details
&& !details
->check_mapping
&& !details
->nonlinear_vma
)
387 for (offset
=0; offset
< size
; ptep
++, offset
+= PAGE_SIZE
) {
391 if (pte_present(pte
)) {
392 struct page
*page
= NULL
;
393 unsigned long pfn
= pte_pfn(pte
);
394 if (pfn_valid(pfn
)) {
395 page
= pfn_to_page(pfn
);
396 if (PageReserved(page
))
399 if (unlikely(details
) && page
) {
401 * unmap_shared_mapping_pages() wants to
402 * invalidate cache without truncating:
403 * unmap shared but keep private pages.
405 if (details
->check_mapping
&&
406 details
->check_mapping
!= page
->mapping
)
409 * Each page->index must be checked when
410 * invalidating or truncating nonlinear.
412 if (details
->nonlinear_vma
&&
413 (page
->index
< details
->first_index
||
414 page
->index
> details
->last_index
))
417 pte
= ptep_get_and_clear(ptep
);
418 tlb_remove_tlb_entry(tlb
, ptep
, address
+offset
);
421 if (unlikely(details
) && details
->nonlinear_vma
422 && linear_page_index(details
->nonlinear_vma
,
423 address
+offset
) != page
->index
)
424 set_pte(ptep
, pgoff_to_pte(page
->index
));
426 set_page_dirty(page
);
427 if (pte_young(pte
) && !PageAnon(page
))
428 mark_page_accessed(page
);
430 page_remove_rmap(page
);
431 tlb_remove_page(tlb
, page
);
435 * If details->check_mapping, we leave swap entries;
436 * if details->nonlinear_vma, we leave file entries.
438 if (unlikely(details
))
441 free_swap_and_cache(pte_to_swp_entry(pte
));
447 static void zap_pmd_range(struct mmu_gather
*tlb
,
448 pgd_t
* dir
, unsigned long address
,
449 unsigned long size
, struct zap_details
*details
)
456 if (unlikely(pgd_bad(*dir
))) {
461 pmd
= pmd_offset(dir
, address
);
462 end
= address
+ size
;
463 if (end
> ((address
+ PGDIR_SIZE
) & PGDIR_MASK
))
464 end
= ((address
+ PGDIR_SIZE
) & PGDIR_MASK
);
466 zap_pte_range(tlb
, pmd
, address
, end
- address
, details
);
467 address
= (address
+ PMD_SIZE
) & PMD_MASK
;
469 } while (address
&& (address
< end
));
472 static void unmap_page_range(struct mmu_gather
*tlb
,
473 struct vm_area_struct
*vma
, unsigned long address
,
474 unsigned long end
, struct zap_details
*details
)
478 BUG_ON(address
>= end
);
479 dir
= pgd_offset(vma
->vm_mm
, address
);
480 tlb_start_vma(tlb
, vma
);
482 zap_pmd_range(tlb
, dir
, address
, end
- address
, details
);
483 address
= (address
+ PGDIR_SIZE
) & PGDIR_MASK
;
485 } while (address
&& (address
< end
));
486 tlb_end_vma(tlb
, vma
);
489 /* Dispose of an entire struct mmu_gather per rescheduling point */
490 #if defined(CONFIG_SMP) && defined(CONFIG_PREEMPT)
491 #define ZAP_BLOCK_SIZE (FREE_PTE_NR * PAGE_SIZE)
494 /* For UP, 256 pages at a time gives nice low latency */
495 #if !defined(CONFIG_SMP) && defined(CONFIG_PREEMPT)
496 #define ZAP_BLOCK_SIZE (256 * PAGE_SIZE)
499 /* No preempt: go for improved straight-line efficiency */
500 #if !defined(CONFIG_PREEMPT)
501 #define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
505 * unmap_vmas - unmap a range of memory covered by a list of vma's
506 * @tlbp: address of the caller's struct mmu_gather
507 * @mm: the controlling mm_struct
508 * @vma: the starting vma
509 * @start_addr: virtual address at which to start unmapping
510 * @end_addr: virtual address at which to end unmapping
511 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
512 * @details: details of nonlinear truncation or shared cache invalidation
514 * Returns the number of vma's which were covered by the unmapping.
516 * Unmap all pages in the vma list. Called under page_table_lock.
518 * We aim to not hold page_table_lock for too long (for scheduling latency
519 * reasons). So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
520 * return the ending mmu_gather to the caller.
522 * Only addresses between `start' and `end' will be unmapped.
524 * The VMA list must be sorted in ascending virtual address order.
526 * unmap_vmas() assumes that the caller will flush the whole unmapped address
527 * range after unmap_vmas() returns. So the only responsibility here is to
528 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
529 * drops the lock and schedules.
531 int unmap_vmas(struct mmu_gather
**tlbp
, struct mm_struct
*mm
,
532 struct vm_area_struct
*vma
, unsigned long start_addr
,
533 unsigned long end_addr
, unsigned long *nr_accounted
,
534 struct zap_details
*details
)
536 unsigned long zap_bytes
= ZAP_BLOCK_SIZE
;
537 unsigned long tlb_start
= 0; /* For tlb_finish_mmu */
538 int tlb_start_valid
= 0;
540 int atomic
= details
&& details
->atomic
;
542 for ( ; vma
&& vma
->vm_start
< end_addr
; vma
= vma
->vm_next
) {
546 start
= max(vma
->vm_start
, start_addr
);
547 if (start
>= vma
->vm_end
)
549 end
= min(vma
->vm_end
, end_addr
);
550 if (end
<= vma
->vm_start
)
553 if (vma
->vm_flags
& VM_ACCOUNT
)
554 *nr_accounted
+= (end
- start
) >> PAGE_SHIFT
;
557 while (start
!= end
) {
560 if (!tlb_start_valid
) {
565 if (is_vm_hugetlb_page(vma
)) {
567 unmap_hugepage_range(vma
, start
, end
);
569 block
= min(zap_bytes
, end
- start
);
570 unmap_page_range(*tlbp
, vma
, start
,
571 start
+ block
, details
);
576 if ((long)zap_bytes
> 0)
578 if (!atomic
&& need_resched()) {
579 int fullmm
= tlb_is_full_mm(*tlbp
);
580 tlb_finish_mmu(*tlbp
, tlb_start
, start
);
581 cond_resched_lock(&mm
->page_table_lock
);
582 *tlbp
= tlb_gather_mmu(mm
, fullmm
);
585 zap_bytes
= ZAP_BLOCK_SIZE
;
592 * zap_page_range - remove user pages in a given range
593 * @vma: vm_area_struct holding the applicable pages
594 * @address: starting address of pages to zap
595 * @size: number of bytes to zap
596 * @details: details of nonlinear truncation or shared cache invalidation
598 void zap_page_range(struct vm_area_struct
*vma
, unsigned long address
,
599 unsigned long size
, struct zap_details
*details
)
601 struct mm_struct
*mm
= vma
->vm_mm
;
602 struct mmu_gather
*tlb
;
603 unsigned long end
= address
+ size
;
604 unsigned long nr_accounted
= 0;
606 if (is_vm_hugetlb_page(vma
)) {
607 zap_hugepage_range(vma
, address
, size
);
612 spin_lock(&mm
->page_table_lock
);
613 tlb
= tlb_gather_mmu(mm
, 0);
614 unmap_vmas(&tlb
, mm
, vma
, address
, end
, &nr_accounted
, details
);
615 tlb_finish_mmu(tlb
, address
, end
);
616 spin_unlock(&mm
->page_table_lock
);
620 * Do a quick page-table lookup for a single page.
621 * mm->page_table_lock must be held.
624 follow_page(struct mm_struct
*mm
, unsigned long address
, int write
)
632 page
= follow_huge_addr(mm
, address
, write
);
636 pgd
= pgd_offset(mm
, address
);
637 if (pgd_none(*pgd
) || unlikely(pgd_bad(*pgd
)))
640 pmd
= pmd_offset(pgd
, address
);
644 return follow_huge_pmd(mm
, address
, pmd
, write
);
645 if (unlikely(pmd_bad(*pmd
)))
648 ptep
= pte_offset_map(pmd
, address
);
654 if (pte_present(pte
)) {
655 if (write
&& !pte_write(pte
))
658 if (pfn_valid(pfn
)) {
659 page
= pfn_to_page(pfn
);
660 if (write
&& !pte_dirty(pte
) && !PageDirty(page
))
661 set_page_dirty(page
);
662 mark_page_accessed(page
);
672 * Given a physical address, is there a useful struct page pointing to
673 * it? This may become more complex in the future if we start dealing
674 * with IO-aperture pages for direct-IO.
677 static inline struct page
*get_page_map(struct page
*page
)
679 if (!pfn_valid(page_to_pfn(page
)))
686 untouched_anonymous_page(struct mm_struct
* mm
, struct vm_area_struct
*vma
,
687 unsigned long address
)
692 /* Check if the vma is for an anonymous mapping. */
693 if (vma
->vm_ops
&& vma
->vm_ops
->nopage
)
696 /* Check if page directory entry exists. */
697 pgd
= pgd_offset(mm
, address
);
698 if (pgd_none(*pgd
) || unlikely(pgd_bad(*pgd
)))
701 /* Check if page middle directory entry exists. */
702 pmd
= pmd_offset(pgd
, address
);
703 if (pmd_none(*pmd
) || unlikely(pmd_bad(*pmd
)))
706 /* There is a pte slot for 'address' in 'mm'. */
711 int get_user_pages(struct task_struct
*tsk
, struct mm_struct
*mm
,
712 unsigned long start
, int len
, int write
, int force
,
713 struct page
**pages
, struct vm_area_struct
**vmas
)
719 * Require read or write permissions.
720 * If 'force' is set, we only require the "MAY" flags.
722 flags
= write
? (VM_WRITE
| VM_MAYWRITE
) : (VM_READ
| VM_MAYREAD
);
723 flags
&= force
? (VM_MAYREAD
| VM_MAYWRITE
) : (VM_READ
| VM_WRITE
);
727 struct vm_area_struct
* vma
;
729 vma
= find_extend_vma(mm
, start
);
730 if (!vma
&& in_gate_area(tsk
, start
)) {
731 unsigned long pg
= start
& PAGE_MASK
;
732 struct vm_area_struct
*gate_vma
= get_gate_vma(tsk
);
736 if (write
) /* user gate pages are read-only */
737 return i
? : -EFAULT
;
738 pgd
= pgd_offset_gate(mm
, pg
);
740 return i
? : -EFAULT
;
741 pmd
= pmd_offset(pgd
, pg
);
743 return i
? : -EFAULT
;
744 pte
= pte_offset_map(pmd
, pg
);
746 return i
? : -EFAULT
;
747 if (!pte_present(*pte
)) {
749 return i
? : -EFAULT
;
752 pages
[i
] = pte_page(*pte
);
764 if (!vma
|| (pages
&& (vma
->vm_flags
& VM_IO
))
765 || !(flags
& vma
->vm_flags
))
766 return i
? : -EFAULT
;
768 if (is_vm_hugetlb_page(vma
)) {
769 i
= follow_hugetlb_page(mm
, vma
, pages
, vmas
,
773 spin_lock(&mm
->page_table_lock
);
776 int lookup_write
= write
;
777 while (!(map
= follow_page(mm
, start
, lookup_write
))) {
779 * Shortcut for anonymous pages. We don't want
780 * to force the creation of pages tables for
781 * insanly big anonymously mapped areas that
782 * nobody touched so far. This is important
783 * for doing a core dump for these mappings.
786 untouched_anonymous_page(mm
,vma
,start
)) {
787 map
= ZERO_PAGE(start
);
790 spin_unlock(&mm
->page_table_lock
);
791 switch (handle_mm_fault(mm
,vma
,start
,write
)) {
798 case VM_FAULT_SIGBUS
:
799 return i
? i
: -EFAULT
;
801 return i
? i
: -ENOMEM
;
806 * Now that we have performed a write fault
807 * and surely no longer have a shared page we
808 * shouldn't write, we shouldn't ignore an
809 * unwritable page in the page table if
810 * we are forcing write access.
812 lookup_write
= write
&& !force
;
813 spin_lock(&mm
->page_table_lock
);
816 pages
[i
] = get_page_map(map
);
818 spin_unlock(&mm
->page_table_lock
);
820 page_cache_release(pages
[i
]);
824 flush_dcache_page(pages
[i
]);
825 if (!PageReserved(pages
[i
]))
826 page_cache_get(pages
[i
]);
833 } while(len
&& start
< vma
->vm_end
);
834 spin_unlock(&mm
->page_table_lock
);
840 EXPORT_SYMBOL(get_user_pages
);
842 static void zeromap_pte_range(pte_t
* pte
, unsigned long address
,
843 unsigned long size
, pgprot_t prot
)
847 address
&= ~PMD_MASK
;
848 end
= address
+ size
;
852 pte_t zero_pte
= pte_wrprotect(mk_pte(ZERO_PAGE(address
), prot
));
853 BUG_ON(!pte_none(*pte
));
854 set_pte(pte
, zero_pte
);
855 address
+= PAGE_SIZE
;
857 } while (address
&& (address
< end
));
860 static inline int zeromap_pmd_range(struct mm_struct
*mm
, pmd_t
* pmd
, unsigned long address
,
861 unsigned long size
, pgprot_t prot
)
863 unsigned long base
, end
;
865 base
= address
& PGDIR_MASK
;
866 address
&= ~PGDIR_MASK
;
867 end
= address
+ size
;
868 if (end
> PGDIR_SIZE
)
871 pte_t
* pte
= pte_alloc_map(mm
, pmd
, base
+ address
);
874 zeromap_pte_range(pte
, base
+ address
, end
- address
, prot
);
876 address
= (address
+ PMD_SIZE
) & PMD_MASK
;
878 } while (address
&& (address
< end
));
882 int zeromap_page_range(struct vm_area_struct
*vma
, unsigned long address
, unsigned long size
, pgprot_t prot
)
886 unsigned long beg
= address
;
887 unsigned long end
= address
+ size
;
888 struct mm_struct
*mm
= vma
->vm_mm
;
890 dir
= pgd_offset(mm
, address
);
891 flush_cache_range(vma
, beg
, end
);
895 spin_lock(&mm
->page_table_lock
);
897 pmd_t
*pmd
= pmd_alloc(mm
, dir
, address
);
901 error
= zeromap_pmd_range(mm
, pmd
, address
, end
- address
, prot
);
904 address
= (address
+ PGDIR_SIZE
) & PGDIR_MASK
;
906 } while (address
&& (address
< end
));
908 * Why flush? zeromap_pte_range has a BUG_ON for !pte_none()
910 flush_tlb_range(vma
, beg
, end
);
911 spin_unlock(&mm
->page_table_lock
);
916 * maps a range of physical memory into the requested pages. the old
917 * mappings are removed. any references to nonexistent pages results
918 * in null mappings (currently treated as "copy-on-access")
920 static inline void remap_pte_range(pte_t
* pte
, unsigned long address
, unsigned long size
,
921 unsigned long phys_addr
, pgprot_t prot
)
926 address
&= ~PMD_MASK
;
927 end
= address
+ size
;
930 pfn
= phys_addr
>> PAGE_SHIFT
;
932 BUG_ON(!pte_none(*pte
));
933 if (!pfn_valid(pfn
) || PageReserved(pfn_to_page(pfn
)))
934 set_pte(pte
, pfn_pte(pfn
, prot
));
935 address
+= PAGE_SIZE
;
938 } while (address
&& (address
< end
));
941 static inline int remap_pmd_range(struct mm_struct
*mm
, pmd_t
* pmd
, unsigned long address
, unsigned long size
,
942 unsigned long phys_addr
, pgprot_t prot
)
944 unsigned long base
, end
;
946 base
= address
& PGDIR_MASK
;
947 address
&= ~PGDIR_MASK
;
948 end
= address
+ size
;
949 if (end
> PGDIR_SIZE
)
951 phys_addr
-= address
;
953 pte_t
* pte
= pte_alloc_map(mm
, pmd
, base
+ address
);
956 remap_pte_range(pte
, base
+ address
, end
- address
, address
+ phys_addr
, prot
);
958 address
= (address
+ PMD_SIZE
) & PMD_MASK
;
960 } while (address
&& (address
< end
));
964 /* Note: this is only safe if the mm semaphore is held when called. */
965 int remap_page_range(struct vm_area_struct
*vma
, unsigned long from
, unsigned long phys_addr
, unsigned long size
, pgprot_t prot
)
969 unsigned long beg
= from
;
970 unsigned long end
= from
+ size
;
971 struct mm_struct
*mm
= vma
->vm_mm
;
974 dir
= pgd_offset(mm
, from
);
975 flush_cache_range(vma
, beg
, end
);
979 spin_lock(&mm
->page_table_lock
);
981 pmd_t
*pmd
= pmd_alloc(mm
, dir
, from
);
985 error
= remap_pmd_range(mm
, pmd
, from
, end
- from
, phys_addr
+ from
, prot
);
988 from
= (from
+ PGDIR_SIZE
) & PGDIR_MASK
;
990 } while (from
&& (from
< end
));
992 * Why flush? remap_pte_range has a BUG_ON for !pte_none()
994 flush_tlb_range(vma
, beg
, end
);
995 spin_unlock(&mm
->page_table_lock
);
999 EXPORT_SYMBOL(remap_page_range
);
1002 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1003 * servicing faults for write access. In the normal case, do always want
1004 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1005 * that do not have writing enabled, when used by access_process_vm.
1007 static inline pte_t
maybe_mkwrite(pte_t pte
, struct vm_area_struct
*vma
)
1009 if (likely(vma
->vm_flags
& VM_WRITE
))
1010 pte
= pte_mkwrite(pte
);
1015 * We hold the mm semaphore for reading and vma->vm_mm->page_table_lock
1017 static inline void break_cow(struct vm_area_struct
* vma
, struct page
* new_page
, unsigned long address
,
1022 flush_cache_page(vma
, address
);
1023 entry
= maybe_mkwrite(pte_mkdirty(mk_pte(new_page
, vma
->vm_page_prot
)),
1025 ptep_establish(vma
, address
, page_table
, entry
);
1026 update_mmu_cache(vma
, address
, entry
);
1030 * This routine handles present pages, when users try to write
1031 * to a shared page. It is done by copying the page to a new address
1032 * and decrementing the shared-page counter for the old page.
1034 * Goto-purists beware: the only reason for goto's here is that it results
1035 * in better assembly code.. The "default" path will see no jumps at all.
1037 * Note that this routine assumes that the protection checks have been
1038 * done by the caller (the low-level page fault routine in most cases).
1039 * Thus we can safely just mark it writable once we've done any necessary
1042 * We also mark the page dirty at this point even though the page will
1043 * change only once the write actually happens. This avoids a few races,
1044 * and potentially makes it more efficient.
1046 * We hold the mm semaphore and the page_table_lock on entry and exit
1047 * with the page_table_lock released.
1049 static int do_wp_page(struct mm_struct
*mm
, struct vm_area_struct
* vma
,
1050 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
, pte_t pte
)
1052 struct page
*old_page
, *new_page
;
1053 unsigned long pfn
= pte_pfn(pte
);
1056 if (unlikely(!pfn_valid(pfn
))) {
1058 * This should really halt the system so it can be debugged or
1059 * at least the kernel stops what it's doing before it corrupts
1060 * data, but for the moment just pretend this is OOM.
1062 pte_unmap(page_table
);
1063 printk(KERN_ERR
"do_wp_page: bogus page at address %08lx\n",
1065 spin_unlock(&mm
->page_table_lock
);
1066 return VM_FAULT_OOM
;
1068 old_page
= pfn_to_page(pfn
);
1070 if (!TestSetPageLocked(old_page
)) {
1071 int reuse
= can_share_swap_page(old_page
);
1072 unlock_page(old_page
);
1074 flush_cache_page(vma
, address
);
1075 entry
= maybe_mkwrite(pte_mkyoung(pte_mkdirty(pte
)),
1077 ptep_set_access_flags(vma
, address
, page_table
, entry
, 1);
1078 update_mmu_cache(vma
, address
, entry
);
1079 pte_unmap(page_table
);
1080 spin_unlock(&mm
->page_table_lock
);
1081 return VM_FAULT_MINOR
;
1084 pte_unmap(page_table
);
1087 * Ok, we need to copy. Oh, well..
1089 if (!PageReserved(old_page
))
1090 page_cache_get(old_page
);
1091 spin_unlock(&mm
->page_table_lock
);
1093 if (unlikely(anon_vma_prepare(vma
)))
1095 new_page
= alloc_page_vma(GFP_HIGHUSER
, vma
, address
);
1098 copy_cow_page(old_page
,new_page
,address
);
1101 * Re-check the pte - we dropped the lock
1103 spin_lock(&mm
->page_table_lock
);
1104 page_table
= pte_offset_map(pmd
, address
);
1105 if (likely(pte_same(*page_table
, pte
))) {
1106 if (PageReserved(old_page
))
1109 page_remove_rmap(old_page
);
1110 break_cow(vma
, new_page
, address
, page_table
);
1111 lru_cache_add_active(new_page
);
1112 page_add_anon_rmap(new_page
, vma
, address
);
1114 /* Free the old page.. */
1115 new_page
= old_page
;
1117 pte_unmap(page_table
);
1118 page_cache_release(new_page
);
1119 page_cache_release(old_page
);
1120 spin_unlock(&mm
->page_table_lock
);
1121 return VM_FAULT_MINOR
;
1124 page_cache_release(old_page
);
1125 return VM_FAULT_OOM
;
1129 * Helper function for unmap_mapping_range().
1131 static inline void unmap_mapping_range_list(struct prio_tree_root
*root
,
1132 struct zap_details
*details
)
1134 struct vm_area_struct
*vma
;
1135 struct prio_tree_iter iter
;
1136 pgoff_t vba
, vea
, zba
, zea
;
1138 vma_prio_tree_foreach(vma
, &iter
, root
,
1139 details
->first_index
, details
->last_index
) {
1140 vba
= vma
->vm_pgoff
;
1141 vea
= vba
+ ((vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
) - 1;
1142 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1143 zba
= details
->first_index
;
1146 zea
= details
->last_index
;
1150 ((zba
- vba
) << PAGE_SHIFT
) + vma
->vm_start
,
1151 (zea
- zba
+ 1) << PAGE_SHIFT
, details
);
1156 * unmap_mapping_range - unmap the portion of all mmaps
1157 * in the specified address_space corresponding to the specified
1158 * page range in the underlying file.
1159 * @address_space: the address space containing mmaps to be unmapped.
1160 * @holebegin: byte in first page to unmap, relative to the start of
1161 * the underlying file. This will be rounded down to a PAGE_SIZE
1162 * boundary. Note that this is different from vmtruncate(), which
1163 * must keep the partial page. In contrast, we must get rid of
1165 * @holelen: size of prospective hole in bytes. This will be rounded
1166 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
1168 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
1169 * but 0 when invalidating pagecache, don't throw away private data.
1171 void unmap_mapping_range(struct address_space
*mapping
,
1172 loff_t
const holebegin
, loff_t
const holelen
, int even_cows
)
1174 struct zap_details details
;
1175 pgoff_t hba
= holebegin
>> PAGE_SHIFT
;
1176 pgoff_t hlen
= (holelen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1178 /* Check for overflow. */
1179 if (sizeof(holelen
) > sizeof(hlen
)) {
1181 (holebegin
+ holelen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1182 if (holeend
& ~(long long)ULONG_MAX
)
1183 hlen
= ULONG_MAX
- hba
+ 1;
1186 details
.check_mapping
= even_cows
? NULL
: mapping
;
1187 details
.nonlinear_vma
= NULL
;
1188 details
.first_index
= hba
;
1189 details
.last_index
= hba
+ hlen
- 1;
1190 details
.atomic
= 1; /* A spinlock is held */
1191 if (details
.last_index
< details
.first_index
)
1192 details
.last_index
= ULONG_MAX
;
1194 spin_lock(&mapping
->i_mmap_lock
);
1195 /* Protect against page fault */
1196 atomic_inc(&mapping
->truncate_count
);
1198 if (unlikely(!prio_tree_empty(&mapping
->i_mmap
)))
1199 unmap_mapping_range_list(&mapping
->i_mmap
, &details
);
1202 * In nonlinear VMAs there is no correspondence between virtual address
1203 * offset and file offset. So we must perform an exhaustive search
1204 * across *all* the pages in each nonlinear VMA, not just the pages
1205 * whose virtual address lies outside the file truncation point.
1207 if (unlikely(!list_empty(&mapping
->i_mmap_nonlinear
))) {
1208 struct vm_area_struct
*vma
;
1209 list_for_each_entry(vma
, &mapping
->i_mmap_nonlinear
,
1210 shared
.vm_set
.list
) {
1211 details
.nonlinear_vma
= vma
;
1212 zap_page_range(vma
, vma
->vm_start
,
1213 vma
->vm_end
- vma
->vm_start
, &details
);
1216 spin_unlock(&mapping
->i_mmap_lock
);
1218 EXPORT_SYMBOL(unmap_mapping_range
);
1221 * Handle all mappings that got truncated by a "truncate()"
1224 * NOTE! We have to be ready to update the memory sharing
1225 * between the file and the memory map for a potential last
1226 * incomplete page. Ugly, but necessary.
1228 int vmtruncate(struct inode
* inode
, loff_t offset
)
1230 struct address_space
*mapping
= inode
->i_mapping
;
1231 unsigned long limit
;
1233 if (inode
->i_size
< offset
)
1236 * truncation of in-use swapfiles is disallowed - it would cause
1237 * subsequent swapout to scribble on the now-freed blocks.
1239 if (IS_SWAPFILE(inode
))
1241 i_size_write(inode
, offset
);
1242 unmap_mapping_range(mapping
, offset
+ PAGE_SIZE
- 1, 0, 1);
1243 truncate_inode_pages(mapping
, offset
);
1247 limit
= current
->signal
->rlim
[RLIMIT_FSIZE
].rlim_cur
;
1248 if (limit
!= RLIM_INFINITY
&& offset
> limit
)
1250 if (offset
> inode
->i_sb
->s_maxbytes
)
1252 i_size_write(inode
, offset
);
1255 if (inode
->i_op
&& inode
->i_op
->truncate
)
1256 inode
->i_op
->truncate(inode
);
1259 send_sig(SIGXFSZ
, current
, 0);
1266 EXPORT_SYMBOL(vmtruncate
);
1269 * Primitive swap readahead code. We simply read an aligned block of
1270 * (1 << page_cluster) entries in the swap area. This method is chosen
1271 * because it doesn't cost us any seek time. We also make sure to queue
1272 * the 'original' request together with the readahead ones...
1274 * This has been extended to use the NUMA policies from the mm triggering
1277 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
1279 void swapin_readahead(swp_entry_t entry
, unsigned long addr
,struct vm_area_struct
*vma
)
1282 struct vm_area_struct
*next_vma
= vma
? vma
->vm_next
: NULL
;
1285 struct page
*new_page
;
1286 unsigned long offset
;
1289 * Get the number of handles we should do readahead io to.
1291 num
= valid_swaphandles(entry
, &offset
);
1292 for (i
= 0; i
< num
; offset
++, i
++) {
1293 /* Ok, do the async read-ahead now */
1294 new_page
= read_swap_cache_async(swp_entry(swp_type(entry
),
1295 offset
), vma
, addr
);
1298 page_cache_release(new_page
);
1301 * Find the next applicable VMA for the NUMA policy.
1307 if (addr
>= vma
->vm_end
) {
1309 next_vma
= vma
? vma
->vm_next
: NULL
;
1311 if (vma
&& addr
< vma
->vm_start
)
1314 if (next_vma
&& addr
>= next_vma
->vm_start
) {
1316 next_vma
= vma
->vm_next
;
1321 lru_add_drain(); /* Push any new pages onto the LRU now */
1325 * We hold the mm semaphore and the page_table_lock on entry and
1326 * should release the pagetable lock on exit..
1328 static int do_swap_page(struct mm_struct
* mm
,
1329 struct vm_area_struct
* vma
, unsigned long address
,
1330 pte_t
*page_table
, pmd_t
*pmd
, pte_t orig_pte
, int write_access
)
1333 swp_entry_t entry
= pte_to_swp_entry(orig_pte
);
1335 int ret
= VM_FAULT_MINOR
;
1337 pte_unmap(page_table
);
1338 spin_unlock(&mm
->page_table_lock
);
1339 page
= lookup_swap_cache(entry
);
1341 swapin_readahead(entry
, address
, vma
);
1342 page
= read_swap_cache_async(entry
, vma
, address
);
1345 * Back out if somebody else faulted in this pte while
1346 * we released the page table lock.
1348 spin_lock(&mm
->page_table_lock
);
1349 page_table
= pte_offset_map(pmd
, address
);
1350 if (likely(pte_same(*page_table
, orig_pte
)))
1353 ret
= VM_FAULT_MINOR
;
1354 pte_unmap(page_table
);
1355 spin_unlock(&mm
->page_table_lock
);
1359 /* Had to read the page from swap area: Major fault */
1360 ret
= VM_FAULT_MAJOR
;
1361 inc_page_state(pgmajfault
);
1365 mark_page_accessed(page
);
1369 * Back out if somebody else faulted in this pte while we
1370 * released the page table lock.
1372 spin_lock(&mm
->page_table_lock
);
1373 page_table
= pte_offset_map(pmd
, address
);
1374 if (unlikely(!pte_same(*page_table
, orig_pte
))) {
1375 pte_unmap(page_table
);
1376 spin_unlock(&mm
->page_table_lock
);
1378 page_cache_release(page
);
1379 ret
= VM_FAULT_MINOR
;
1383 /* The page isn't present yet, go ahead with the fault. */
1387 remove_exclusive_swap_page(page
);
1390 pte
= mk_pte(page
, vma
->vm_page_prot
);
1391 if (write_access
&& can_share_swap_page(page
)) {
1392 pte
= maybe_mkwrite(pte_mkdirty(pte
), vma
);
1397 flush_icache_page(vma
, page
);
1398 set_pte(page_table
, pte
);
1399 page_add_anon_rmap(page
, vma
, address
);
1402 if (do_wp_page(mm
, vma
, address
,
1403 page_table
, pmd
, pte
) == VM_FAULT_OOM
)
1408 /* No need to invalidate - it was non-present before */
1409 update_mmu_cache(vma
, address
, pte
);
1410 pte_unmap(page_table
);
1411 spin_unlock(&mm
->page_table_lock
);
1417 * We are called with the MM semaphore and page_table_lock
1418 * spinlock held to protect against concurrent faults in
1419 * multithreaded programs.
1422 do_anonymous_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1423 pte_t
*page_table
, pmd_t
*pmd
, int write_access
,
1427 struct page
* page
= ZERO_PAGE(addr
);
1429 /* Read-only mapping of ZERO_PAGE. */
1430 entry
= pte_wrprotect(mk_pte(ZERO_PAGE(addr
), vma
->vm_page_prot
));
1432 /* ..except if it's a write access */
1434 /* Allocate our own private page. */
1435 pte_unmap(page_table
);
1436 spin_unlock(&mm
->page_table_lock
);
1438 if (unlikely(anon_vma_prepare(vma
)))
1440 page
= alloc_page_vma(GFP_HIGHUSER
, vma
, addr
);
1443 clear_user_highpage(page
, addr
);
1445 spin_lock(&mm
->page_table_lock
);
1446 page_table
= pte_offset_map(pmd
, addr
);
1448 if (!pte_none(*page_table
)) {
1449 pte_unmap(page_table
);
1450 page_cache_release(page
);
1451 spin_unlock(&mm
->page_table_lock
);
1455 entry
= maybe_mkwrite(pte_mkdirty(mk_pte(page
,
1456 vma
->vm_page_prot
)),
1458 lru_cache_add_active(page
);
1459 mark_page_accessed(page
);
1460 page_add_anon_rmap(page
, vma
, addr
);
1463 set_pte(page_table
, entry
);
1464 pte_unmap(page_table
);
1466 /* No need to invalidate - it was non-present before */
1467 update_mmu_cache(vma
, addr
, entry
);
1468 spin_unlock(&mm
->page_table_lock
);
1470 return VM_FAULT_MINOR
;
1472 return VM_FAULT_OOM
;
1476 * do_no_page() tries to create a new page mapping. It aggressively
1477 * tries to share with existing pages, but makes a separate copy if
1478 * the "write_access" parameter is true in order to avoid the next
1481 * As this is called only for pages that do not currently exist, we
1482 * do not need to flush old virtual caches or the TLB.
1484 * This is called with the MM semaphore held and the page table
1485 * spinlock held. Exit with the spinlock released.
1488 do_no_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1489 unsigned long address
, int write_access
, pte_t
*page_table
, pmd_t
*pmd
)
1491 struct page
* new_page
;
1492 struct address_space
*mapping
= NULL
;
1495 int ret
= VM_FAULT_MINOR
;
1498 if (!vma
->vm_ops
|| !vma
->vm_ops
->nopage
)
1499 return do_anonymous_page(mm
, vma
, page_table
,
1500 pmd
, write_access
, address
);
1501 pte_unmap(page_table
);
1502 spin_unlock(&mm
->page_table_lock
);
1505 mapping
= vma
->vm_file
->f_mapping
;
1506 sequence
= atomic_read(&mapping
->truncate_count
);
1508 smp_rmb(); /* Prevent CPU from reordering lock-free ->nopage() */
1510 new_page
= vma
->vm_ops
->nopage(vma
, address
& PAGE_MASK
, &ret
);
1512 /* no page was available -- either SIGBUS or OOM */
1513 if (new_page
== NOPAGE_SIGBUS
)
1514 return VM_FAULT_SIGBUS
;
1515 if (new_page
== NOPAGE_OOM
)
1516 return VM_FAULT_OOM
;
1519 * Should we do an early C-O-W break?
1521 if (write_access
&& !(vma
->vm_flags
& VM_SHARED
)) {
1524 if (unlikely(anon_vma_prepare(vma
)))
1526 page
= alloc_page_vma(GFP_HIGHUSER
, vma
, address
);
1529 copy_user_highpage(page
, new_page
, address
);
1530 page_cache_release(new_page
);
1535 spin_lock(&mm
->page_table_lock
);
1537 * For a file-backed vma, someone could have truncated or otherwise
1538 * invalidated this page. If unmap_mapping_range got called,
1539 * retry getting the page.
1542 (unlikely(sequence
!= atomic_read(&mapping
->truncate_count
)))) {
1543 sequence
= atomic_read(&mapping
->truncate_count
);
1544 spin_unlock(&mm
->page_table_lock
);
1545 page_cache_release(new_page
);
1548 page_table
= pte_offset_map(pmd
, address
);
1551 * This silly early PAGE_DIRTY setting removes a race
1552 * due to the bad i386 page protection. But it's valid
1553 * for other architectures too.
1555 * Note that if write_access is true, we either now have
1556 * an exclusive copy of the page, or this is a shared mapping,
1557 * so we can make it writable and dirty to avoid having to
1558 * handle that later.
1560 /* Only go through if we didn't race with anybody else... */
1561 if (pte_none(*page_table
)) {
1562 if (!PageReserved(new_page
))
1564 flush_icache_page(vma
, new_page
);
1565 entry
= mk_pte(new_page
, vma
->vm_page_prot
);
1567 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
1568 set_pte(page_table
, entry
);
1570 lru_cache_add_active(new_page
);
1571 page_add_anon_rmap(new_page
, vma
, address
);
1573 page_add_file_rmap(new_page
);
1574 pte_unmap(page_table
);
1576 /* One of our sibling threads was faster, back out. */
1577 pte_unmap(page_table
);
1578 page_cache_release(new_page
);
1579 spin_unlock(&mm
->page_table_lock
);
1583 /* no need to invalidate: a not-present page shouldn't be cached */
1584 update_mmu_cache(vma
, address
, entry
);
1585 spin_unlock(&mm
->page_table_lock
);
1589 page_cache_release(new_page
);
1595 * Fault of a previously existing named mapping. Repopulate the pte
1596 * from the encoded file_pte if possible. This enables swappable
1599 static int do_file_page(struct mm_struct
* mm
, struct vm_area_struct
* vma
,
1600 unsigned long address
, int write_access
, pte_t
*pte
, pmd_t
*pmd
)
1602 unsigned long pgoff
;
1605 BUG_ON(!vma
->vm_ops
|| !vma
->vm_ops
->nopage
);
1607 * Fall back to the linear mapping if the fs does not support
1610 if (!vma
->vm_ops
|| !vma
->vm_ops
->populate
||
1611 (write_access
&& !(vma
->vm_flags
& VM_SHARED
))) {
1613 return do_no_page(mm
, vma
, address
, write_access
, pte
, pmd
);
1616 pgoff
= pte_to_pgoff(*pte
);
1619 spin_unlock(&mm
->page_table_lock
);
1621 err
= vma
->vm_ops
->populate(vma
, address
& PAGE_MASK
, PAGE_SIZE
, vma
->vm_page_prot
, pgoff
, 0);
1623 return VM_FAULT_OOM
;
1625 return VM_FAULT_SIGBUS
;
1626 return VM_FAULT_MAJOR
;
1630 * These routines also need to handle stuff like marking pages dirty
1631 * and/or accessed for architectures that don't do it in hardware (most
1632 * RISC architectures). The early dirtying is also good on the i386.
1634 * There is also a hook called "update_mmu_cache()" that architectures
1635 * with external mmu caches can use to update those (ie the Sparc or
1636 * PowerPC hashed page tables that act as extended TLBs).
1638 * Note the "page_table_lock". It is to protect against kswapd removing
1639 * pages from under us. Note that kswapd only ever _removes_ pages, never
1640 * adds them. As such, once we have noticed that the page is not present,
1641 * we can drop the lock early.
1643 * The adding of pages is protected by the MM semaphore (which we hold),
1644 * so we don't need to worry about a page being suddenly been added into
1647 * We enter with the pagetable spinlock held, we are supposed to
1648 * release it when done.
1650 static inline int handle_pte_fault(struct mm_struct
*mm
,
1651 struct vm_area_struct
* vma
, unsigned long address
,
1652 int write_access
, pte_t
*pte
, pmd_t
*pmd
)
1657 if (!pte_present(entry
)) {
1659 * If it truly wasn't present, we know that kswapd
1660 * and the PTE updates will not touch it later. So
1663 if (pte_none(entry
))
1664 return do_no_page(mm
, vma
, address
, write_access
, pte
, pmd
);
1665 if (pte_file(entry
))
1666 return do_file_page(mm
, vma
, address
, write_access
, pte
, pmd
);
1667 return do_swap_page(mm
, vma
, address
, pte
, pmd
, entry
, write_access
);
1671 if (!pte_write(entry
))
1672 return do_wp_page(mm
, vma
, address
, pte
, pmd
, entry
);
1674 entry
= pte_mkdirty(entry
);
1676 entry
= pte_mkyoung(entry
);
1677 ptep_set_access_flags(vma
, address
, pte
, entry
, write_access
);
1678 update_mmu_cache(vma
, address
, entry
);
1680 spin_unlock(&mm
->page_table_lock
);
1681 return VM_FAULT_MINOR
;
1685 * By the time we get here, we already hold the mm semaphore
1687 int handle_mm_fault(struct mm_struct
*mm
, struct vm_area_struct
* vma
,
1688 unsigned long address
, int write_access
)
1693 __set_current_state(TASK_RUNNING
);
1694 pgd
= pgd_offset(mm
, address
);
1696 inc_page_state(pgfault
);
1698 if (is_vm_hugetlb_page(vma
))
1699 return VM_FAULT_SIGBUS
; /* mapping truncation does this. */
1702 * We need the page table lock to synchronize with kswapd
1703 * and the SMP-safe atomic PTE updates.
1705 spin_lock(&mm
->page_table_lock
);
1706 pmd
= pmd_alloc(mm
, pgd
, address
);
1709 pte_t
* pte
= pte_alloc_map(mm
, pmd
, address
);
1711 return handle_pte_fault(mm
, vma
, address
, write_access
, pte
, pmd
);
1713 spin_unlock(&mm
->page_table_lock
);
1714 return VM_FAULT_OOM
;
1718 * Allocate page middle directory.
1720 * We've already handled the fast-path in-line, and we own the
1723 * On a two-level page table, this ends up actually being entirely
1726 pmd_t fastcall
*__pmd_alloc(struct mm_struct
*mm
, pgd_t
*pgd
, unsigned long address
)
1730 spin_unlock(&mm
->page_table_lock
);
1731 new = pmd_alloc_one(mm
, address
);
1732 spin_lock(&mm
->page_table_lock
);
1737 * Because we dropped the lock, we should re-check the
1738 * entry, as somebody else could have populated it..
1740 if (pgd_present(*pgd
)) {
1744 pgd_populate(mm
, pgd
, new);
1746 return pmd_offset(pgd
, address
);
1749 int make_pages_present(unsigned long addr
, unsigned long end
)
1751 int ret
, len
, write
;
1752 struct vm_area_struct
* vma
;
1754 vma
= find_vma(current
->mm
, addr
);
1757 write
= (vma
->vm_flags
& VM_WRITE
) != 0;
1760 if (end
> vma
->vm_end
)
1762 len
= (end
+PAGE_SIZE
-1)/PAGE_SIZE
-addr
/PAGE_SIZE
;
1763 ret
= get_user_pages(current
, current
->mm
, addr
,
1764 len
, write
, 0, NULL
, NULL
);
1767 return ret
== len
? 0 : -1;
1771 * Map a vmalloc()-space virtual address to the physical page.
1773 struct page
* vmalloc_to_page(void * vmalloc_addr
)
1775 unsigned long addr
= (unsigned long) vmalloc_addr
;
1776 struct page
*page
= NULL
;
1777 pgd_t
*pgd
= pgd_offset_k(addr
);
1781 if (!pgd_none(*pgd
)) {
1782 pmd
= pmd_offset(pgd
, addr
);
1783 if (!pmd_none(*pmd
)) {
1785 ptep
= pte_offset_map(pmd
, addr
);
1787 if (pte_present(pte
))
1788 page
= pte_page(pte
);
1796 EXPORT_SYMBOL(vmalloc_to_page
);
1798 #if !defined(CONFIG_ARCH_GATE_AREA)
1800 #if defined(AT_SYSINFO_EHDR)
1801 struct vm_area_struct gate_vma
;
1803 static int __init
gate_vma_init(void)
1805 gate_vma
.vm_mm
= NULL
;
1806 gate_vma
.vm_start
= FIXADDR_USER_START
;
1807 gate_vma
.vm_end
= FIXADDR_USER_END
;
1808 gate_vma
.vm_page_prot
= PAGE_READONLY
;
1809 gate_vma
.vm_flags
= 0;
1812 __initcall(gate_vma_init
);
1815 struct vm_area_struct
*get_gate_vma(struct task_struct
*tsk
)
1817 #ifdef AT_SYSINFO_EHDR
1824 int in_gate_area(struct task_struct
*task
, unsigned long addr
)
1826 #ifdef AT_SYSINFO_EHDR
1827 if ((addr
>= FIXADDR_USER_START
) && (addr
< FIXADDR_USER_END
))