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)
40 #include <linux/mman.h>
41 #include <linux/swap.h>
42 #include <linux/smp_lock.h>
43 #include <linux/swapctl.h>
44 #include <linux/iobuf.h>
45 #include <asm/uaccess.h>
46 #include <asm/pgalloc.h>
47 #include <linux/highmem.h>
48 #include <linux/pagemap.h>
51 unsigned long max_mapnr
;
52 unsigned long num_physpages
;
54 struct page
*highmem_start_page
;
57 * We special-case the C-O-W ZERO_PAGE, because it's such
58 * a common occurrence (no need to read the page to know
59 * that it's zero - better for the cache and memory subsystem).
61 static inline void copy_cow_page(struct page
* from
, struct page
* to
, unsigned long address
)
63 if (from
== ZERO_PAGE(address
)) {
64 clear_user_highpage(to
, address
);
67 copy_user_highpage(to
, from
, address
);
73 * Note: this doesn't free the actual pages themselves. That
74 * has been handled earlier when unmapping all the memory regions.
76 static inline void free_one_pmd(pmd_t
* dir
)
87 pte
= pte_offset(dir
, 0);
92 static inline void free_one_pgd(pgd_t
* dir
)
104 pmd
= pmd_offset(dir
, 0);
106 for (j
= 0; j
< PTRS_PER_PMD
; j
++)
111 /* Low and high watermarks for page table cache.
112 The system should try to have pgt_water[0] <= cache elements <= pgt_water[1]
114 int pgt_cache_water
[2] = { 25, 50 };
116 /* Returns the number of pages freed */
117 int check_pgt_cache(void)
119 return do_check_pgt_cache(pgt_cache_water
[0], pgt_cache_water
[1]);
124 * This function clears all user-level page tables of a process - this
125 * is needed by execve(), so that old pages aren't in the way.
127 void clear_page_tables(struct mm_struct
*mm
, unsigned long first
, int nr
)
129 pgd_t
* page_dir
= mm
->pgd
;
133 free_one_pgd(page_dir
);
137 /* keep the page table cache within bounds */
141 #define PTE_TABLE_MASK ((PTRS_PER_PTE-1) * sizeof(pte_t))
142 #define PMD_TABLE_MASK ((PTRS_PER_PMD-1) * sizeof(pmd_t))
145 * copy one vm_area from one task to the other. Assumes the page tables
146 * already present in the new task to be cleared in the whole range
147 * covered by this vma.
149 * 08Jan98 Merged into one routine from several inline routines to reduce
150 * variable count and make things faster. -jj
152 int copy_page_range(struct mm_struct
*dst
, struct mm_struct
*src
,
153 struct vm_area_struct
*vma
)
155 pgd_t
* src_pgd
, * dst_pgd
;
156 unsigned long address
= vma
->vm_start
;
157 unsigned long end
= vma
->vm_end
;
158 unsigned long cow
= (vma
->vm_flags
& (VM_SHARED
| VM_MAYWRITE
)) == VM_MAYWRITE
;
160 src_pgd
= pgd_offset(src
, address
)-1;
161 dst_pgd
= pgd_offset(dst
, address
)-1;
164 pmd_t
* src_pmd
, * dst_pmd
;
166 src_pgd
++; dst_pgd
++;
170 if (pgd_none(*src_pgd
))
171 goto skip_copy_pmd_range
;
172 if (pgd_bad(*src_pgd
)) {
175 skip_copy_pmd_range
: address
= (address
+ PGDIR_SIZE
) & PGDIR_MASK
;
176 if (!address
|| (address
>= end
))
180 if (pgd_none(*dst_pgd
)) {
181 if (!pmd_alloc(dst_pgd
, 0))
185 src_pmd
= pmd_offset(src_pgd
, address
);
186 dst_pmd
= pmd_offset(dst_pgd
, address
);
189 pte_t
* src_pte
, * dst_pte
;
193 if (pmd_none(*src_pmd
))
194 goto skip_copy_pte_range
;
195 if (pmd_bad(*src_pmd
)) {
198 skip_copy_pte_range
: address
= (address
+ PMD_SIZE
) & PMD_MASK
;
201 goto cont_copy_pmd_range
;
203 if (pmd_none(*dst_pmd
)) {
204 if (!pte_alloc(dst_pmd
, 0))
208 src_pte
= pte_offset(src_pmd
, address
);
209 dst_pte
= pte_offset(dst_pmd
, address
);
212 pte_t pte
= *src_pte
;
213 struct page
*ptepage
;
218 goto cont_copy_pte_range_noset
;
219 if (!pte_present(pte
)) {
220 swap_duplicate(pte_to_swp_entry(pte
));
221 goto cont_copy_pte_range
;
223 ptepage
= pte_page(pte
);
224 if ((!VALID_PAGE(ptepage
)) ||
225 PageReserved(ptepage
))
226 goto cont_copy_pte_range
;
228 /* If it's a COW mapping, write protect it both in the parent and the child */
230 ptep_clear_wrprotect(src_pte
);
234 /* If it's a shared mapping, mark it clean in the child */
235 if (vma
->vm_flags
& VM_SHARED
)
236 pte
= pte_mkclean(pte
);
237 pte
= pte_mkold(pte
);
240 cont_copy_pte_range
: set_pte(dst_pte
, pte
);
241 cont_copy_pte_range_noset
: address
+= PAGE_SIZE
;
246 } while ((unsigned long)src_pte
& PTE_TABLE_MASK
);
248 cont_copy_pmd_range
: src_pmd
++;
250 } while ((unsigned long)src_pmd
& PMD_TABLE_MASK
);
260 * Return indicates whether a page was freed so caller can adjust rss
262 static inline int free_pte(pte_t page
)
264 if (pte_present(page
)) {
265 struct page
*ptpage
= pte_page(page
);
266 if ((!VALID_PAGE(ptpage
)) || PageReserved(ptpage
))
269 * free_page() used to be able to clear swap cache
270 * entries. We may now have to do it manually.
273 SetPageDirty(ptpage
);
274 free_page_and_swap_cache(ptpage
);
277 swap_free(pte_to_swp_entry(page
));
281 static inline void forget_pte(pte_t page
)
283 if (!pte_none(page
)) {
284 printk("forget_pte: old mapping existed!\n");
289 static inline int zap_pte_range(struct mm_struct
*mm
, pmd_t
* pmd
, unsigned long address
, unsigned long size
)
301 pte
= pte_offset(pmd
, address
);
302 address
&= ~PMD_MASK
;
303 if (address
+ size
> PMD_SIZE
)
304 size
= PMD_SIZE
- address
;
311 page
= ptep_get_and_clear(pte
);
316 freed
+= free_pte(page
);
321 static inline int zap_pmd_range(struct mm_struct
*mm
, pgd_t
* dir
, unsigned long address
, unsigned long size
)
334 pmd
= pmd_offset(dir
, address
);
335 address
&= ~PGDIR_MASK
;
336 end
= address
+ size
;
337 if (end
> PGDIR_SIZE
)
341 freed
+= zap_pte_range(mm
, pmd
, address
, end
- address
);
342 address
= (address
+ PMD_SIZE
) & PMD_MASK
;
344 } while (address
< end
);
349 * remove user pages in a given range.
351 void zap_page_range(struct mm_struct
*mm
, unsigned long address
, unsigned long size
)
354 unsigned long end
= address
+ size
;
357 dir
= pgd_offset(mm
, address
);
360 * This is a long-lived spinlock. That's fine.
361 * There's no contention, because the page table
362 * lock only protects against kswapd anyway, and
363 * even if kswapd happened to be looking at this
364 * process we _want_ it to get stuck.
368 spin_lock(&mm
->page_table_lock
);
370 freed
+= zap_pmd_range(mm
, dir
, address
, end
- address
);
371 address
= (address
+ PGDIR_SIZE
) & PGDIR_MASK
;
373 } while (address
&& (address
< end
));
374 spin_unlock(&mm
->page_table_lock
);
376 * Update rss for the mm_struct (not necessarily current->mm)
377 * Notice that rss is an unsigned long.
387 * Do a quick page-table lookup for a single page.
389 static struct page
* follow_page(unsigned long address
)
394 pgd
= pgd_offset(current
->mm
, address
);
395 pmd
= pmd_offset(pgd
, address
);
397 pte_t
* pte
= pte_offset(pmd
, address
);
398 if (pte
&& pte_present(*pte
))
399 return pte_page(*pte
);
406 * Given a physical address, is there a useful struct page pointing to
407 * it? This may become more complex in the future if we start dealing
408 * with IO-aperture pages in kiobufs.
411 static inline struct page
* get_page_map(struct page
*page
)
413 if (!VALID_PAGE(page
))
419 * Force in an entire range of pages from the current process's user VA,
420 * and pin them in physical memory.
423 #define dprintk(x...)
424 int map_user_kiobuf(int rw
, struct kiobuf
*iobuf
, unsigned long va
, size_t len
)
426 unsigned long ptr
, end
;
428 struct mm_struct
* mm
;
429 struct vm_area_struct
* vma
= 0;
432 int datain
= (rw
== READ
);
434 /* Make sure the iobuf is not already mapped somewhere. */
439 dprintk ("map_user_kiobuf: begin\n");
441 ptr
= va
& PAGE_MASK
;
442 end
= (va
+ len
+ PAGE_SIZE
- 1) & PAGE_MASK
;
443 err
= expand_kiobuf(iobuf
, (end
- ptr
) >> PAGE_SHIFT
);
451 iobuf
->offset
= va
& ~PAGE_MASK
;
457 * First of all, try to fault in all of the necessary pages
460 if (!vma
|| ptr
>= vma
->vm_end
) {
461 vma
= find_vma(current
->mm
, ptr
);
464 if (vma
->vm_start
> ptr
) {
465 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
467 if (expand_stack(vma
, ptr
))
470 if (((datain
) && (!(vma
->vm_flags
& VM_WRITE
))) ||
471 (!(vma
->vm_flags
& VM_READ
))) {
476 if (handle_mm_fault(current
->mm
, vma
, ptr
, datain
) <= 0)
478 spin_lock(&mm
->page_table_lock
);
479 map
= follow_page(ptr
);
481 spin_unlock(&mm
->page_table_lock
);
482 dprintk (KERN_ERR
"Missing page in map_user_kiobuf\n");
485 map
= get_page_map(map
);
487 atomic_inc(&map
->count
);
489 printk (KERN_INFO
"Mapped page missing [%d]\n", i
);
490 spin_unlock(&mm
->page_table_lock
);
491 iobuf
->maplist
[i
] = map
;
492 iobuf
->nr_pages
= ++i
;
498 dprintk ("map_user_kiobuf: end OK\n");
504 dprintk ("map_user_kiobuf: end %d\n", err
);
510 * Unmap all of the pages referenced by a kiobuf. We release the pages,
511 * and unlock them if they were locked.
514 void unmap_kiobuf (struct kiobuf
*iobuf
)
519 for (i
= 0; i
< iobuf
->nr_pages
; i
++) {
520 map
= iobuf
->maplist
[i
];
534 * Lock down all of the pages of a kiovec for IO.
536 * If any page is mapped twice in the kiovec, we return the error -EINVAL.
538 * The optional wait parameter causes the lock call to block until all
539 * pages can be locked if set. If wait==0, the lock operation is
540 * aborted if any locked pages are found and -EAGAIN is returned.
543 int lock_kiovec(int nr
, struct kiobuf
*iovec
[], int wait
)
545 struct kiobuf
*iobuf
;
547 struct page
*page
, **ppage
;
553 for (i
= 0; i
< nr
; i
++) {
560 ppage
= iobuf
->maplist
;
561 for (j
= 0; j
< iobuf
->nr_pages
; ppage
++, j
++) {
566 if (TryLockPage(page
))
576 * We couldn't lock one of the pages. Undo the locking so far,
577 * wait on the page we got to, and try again.
580 unlock_kiovec(nr
, iovec
);
585 * Did the release also unlock the page we got stuck on?
587 if (!PageLocked(page
)) {
589 * If so, we may well have the page mapped twice
590 * in the IO address range. Bad news. Of
591 * course, it _might_ just be a coincidence,
592 * but if it happens more than once, chances
593 * are we have a double-mapped page.
595 if (++doublepage
>= 3)
608 * Unlock all of the pages of a kiovec after IO.
611 int unlock_kiovec(int nr
, struct kiobuf
*iovec
[])
613 struct kiobuf
*iobuf
;
615 struct page
*page
, **ppage
;
617 for (i
= 0; i
< nr
; i
++) {
624 ppage
= iobuf
->maplist
;
625 for (j
= 0; j
< iobuf
->nr_pages
; ppage
++, j
++) {
635 static inline void zeromap_pte_range(pte_t
* pte
, unsigned long address
,
636 unsigned long size
, pgprot_t prot
)
640 address
&= ~PMD_MASK
;
641 end
= address
+ size
;
645 pte_t zero_pte
= pte_wrprotect(mk_pte(ZERO_PAGE(address
), prot
));
646 pte_t oldpage
= ptep_get_and_clear(pte
);
647 set_pte(pte
, zero_pte
);
649 address
+= PAGE_SIZE
;
651 } while (address
&& (address
< end
));
654 static inline int zeromap_pmd_range(pmd_t
* pmd
, unsigned long address
,
655 unsigned long size
, pgprot_t prot
)
659 address
&= ~PGDIR_MASK
;
660 end
= address
+ size
;
661 if (end
> PGDIR_SIZE
)
664 pte_t
* pte
= pte_alloc(pmd
, address
);
667 zeromap_pte_range(pte
, address
, end
- address
, prot
);
668 address
= (address
+ PMD_SIZE
) & PMD_MASK
;
670 } while (address
&& (address
< end
));
674 int zeromap_page_range(unsigned long address
, unsigned long size
, pgprot_t prot
)
678 unsigned long beg
= address
;
679 unsigned long end
= address
+ size
;
681 dir
= pgd_offset(current
->mm
, address
);
682 flush_cache_range(current
->mm
, beg
, end
);
686 pmd_t
*pmd
= pmd_alloc(dir
, address
);
690 error
= zeromap_pmd_range(pmd
, address
, end
- address
, prot
);
693 address
= (address
+ PGDIR_SIZE
) & PGDIR_MASK
;
695 } while (address
&& (address
< end
));
696 flush_tlb_range(current
->mm
, beg
, end
);
701 * maps a range of physical memory into the requested pages. the old
702 * mappings are removed. any references to nonexistent pages results
703 * in null mappings (currently treated as "copy-on-access")
705 static inline void remap_pte_range(pte_t
* pte
, unsigned long address
, unsigned long size
,
706 unsigned long phys_addr
, pgprot_t prot
)
710 address
&= ~PMD_MASK
;
711 end
= address
+ size
;
717 oldpage
= ptep_get_and_clear(pte
);
719 page
= virt_to_page(__va(phys_addr
));
720 if ((!VALID_PAGE(page
)) || PageReserved(page
))
721 set_pte(pte
, mk_pte_phys(phys_addr
, prot
));
723 address
+= PAGE_SIZE
;
724 phys_addr
+= PAGE_SIZE
;
726 } while (address
&& (address
< end
));
729 static inline int remap_pmd_range(pmd_t
* pmd
, unsigned long address
, unsigned long size
,
730 unsigned long phys_addr
, pgprot_t prot
)
734 address
&= ~PGDIR_MASK
;
735 end
= address
+ size
;
736 if (end
> PGDIR_SIZE
)
738 phys_addr
-= address
;
740 pte_t
* pte
= pte_alloc(pmd
, address
);
743 remap_pte_range(pte
, address
, end
- address
, address
+ phys_addr
, prot
);
744 address
= (address
+ PMD_SIZE
) & PMD_MASK
;
746 } while (address
&& (address
< end
));
750 /* Note: this is only safe if the mm semaphore is held when called. */
751 int remap_page_range(unsigned long from
, unsigned long phys_addr
, unsigned long size
, pgprot_t prot
)
755 unsigned long beg
= from
;
756 unsigned long end
= from
+ size
;
759 dir
= pgd_offset(current
->mm
, from
);
760 flush_cache_range(current
->mm
, beg
, end
);
764 pmd_t
*pmd
= pmd_alloc(dir
, from
);
768 error
= remap_pmd_range(pmd
, from
, end
- from
, phys_addr
+ from
, prot
);
771 from
= (from
+ PGDIR_SIZE
) & PGDIR_MASK
;
773 } while (from
&& (from
< end
));
774 flush_tlb_range(current
->mm
, beg
, end
);
779 * Establish a new mapping:
780 * - flush the old one
781 * - update the page tables
782 * - inform the TLB about the new one
784 static inline void establish_pte(struct vm_area_struct
* vma
, unsigned long address
, pte_t
*page_table
, pte_t entry
)
786 set_pte(page_table
, entry
);
787 flush_tlb_page(vma
, address
);
788 update_mmu_cache(vma
, address
, entry
);
791 static inline void break_cow(struct vm_area_struct
* vma
, struct page
* old_page
, struct page
* new_page
, unsigned long address
,
794 copy_cow_page(old_page
,new_page
,address
);
795 flush_page_to_ram(new_page
);
796 flush_cache_page(vma
, address
);
797 establish_pte(vma
, address
, page_table
, pte_mkwrite(pte_mkdirty(mk_pte(new_page
, vma
->vm_page_prot
))));
801 * This routine handles present pages, when users try to write
802 * to a shared page. It is done by copying the page to a new address
803 * and decrementing the shared-page counter for the old page.
805 * Goto-purists beware: the only reason for goto's here is that it results
806 * in better assembly code.. The "default" path will see no jumps at all.
808 * Note that this routine assumes that the protection checks have been
809 * done by the caller (the low-level page fault routine in most cases).
810 * Thus we can safely just mark it writable once we've done any necessary
813 * We also mark the page dirty at this point even though the page will
814 * change only once the write actually happens. This avoids a few races,
815 * and potentially makes it more efficient.
817 * We enter with the page table read-lock held, and need to exit without
820 static int do_wp_page(struct mm_struct
*mm
, struct vm_area_struct
* vma
,
821 unsigned long address
, pte_t
*page_table
, pte_t pte
)
823 struct page
*old_page
, *new_page
;
825 old_page
= pte_page(pte
);
826 if (!VALID_PAGE(old_page
))
830 * We can avoid the copy if:
831 * - we're the only user (count == 1)
832 * - the only other user is the swap cache,
833 * and the only swap cache user is itself,
834 * in which case we can just continue to
835 * use the same swap cache (it will be
838 switch (page_count(old_page
)) {
841 * Lock the page so that no one can look it up from
842 * the swap cache, grab a reference and start using it.
843 * Can not do lock_page, holding page_table_lock.
845 if (!PageSwapCache(old_page
) || TryLockPage(old_page
))
847 if (is_page_shared(old_page
)) {
848 UnlockPage(old_page
);
851 UnlockPage(old_page
);
854 flush_cache_page(vma
, address
);
855 establish_pte(vma
, address
, page_table
, pte_mkyoung(pte_mkdirty(pte_mkwrite(pte
))));
856 spin_unlock(&mm
->page_table_lock
);
857 return 1; /* Minor fault */
861 * Ok, we need to copy. Oh, well..
863 spin_unlock(&mm
->page_table_lock
);
864 new_page
= page_cache_alloc();
867 spin_lock(&mm
->page_table_lock
);
870 * Re-check the pte - we dropped the lock
872 if (pte_same(*page_table
, pte
)) {
873 if (PageReserved(old_page
))
875 break_cow(vma
, old_page
, new_page
, address
, page_table
);
877 /* Free the old page.. */
880 spin_unlock(&mm
->page_table_lock
);
881 page_cache_release(new_page
);
882 return 1; /* Minor fault */
885 spin_unlock(&mm
->page_table_lock
);
886 printk("do_wp_page: bogus page at address %08lx (page 0x%lx)\n",address
,(unsigned long)old_page
);
891 * This function zeroes out partial mmap'ed pages at truncation time..
893 static void partial_clear(struct vm_area_struct
*vma
, unsigned long address
)
899 pte_t
*page_table
, pte
;
901 page_dir
= pgd_offset(vma
->vm_mm
, address
);
902 if (pgd_none(*page_dir
))
904 if (pgd_bad(*page_dir
)) {
905 pgd_ERROR(*page_dir
);
909 page_middle
= pmd_offset(page_dir
, address
);
910 if (pmd_none(*page_middle
))
912 if (pmd_bad(*page_middle
)) {
913 pmd_ERROR(*page_middle
);
914 pmd_clear(page_middle
);
917 page_table
= pte_offset(page_middle
, address
);
919 if (!pte_present(pte
))
921 flush_cache_page(vma
, address
);
922 page
= pte_page(pte
);
923 if ((!VALID_PAGE(page
)) || PageReserved(page
))
925 offset
= address
& ~PAGE_MASK
;
926 memclear_highpage_flush(page
, offset
, PAGE_SIZE
- offset
);
929 static void vmtruncate_list(struct vm_area_struct
*mpnt
,
930 unsigned long pgoff
, unsigned long partial
)
933 struct mm_struct
*mm
= mpnt
->vm_mm
;
934 unsigned long start
= mpnt
->vm_start
;
935 unsigned long end
= mpnt
->vm_end
;
936 unsigned long len
= end
- start
;
939 /* mapping wholly truncated? */
940 if (mpnt
->vm_pgoff
>= pgoff
) {
941 flush_cache_range(mm
, start
, end
);
942 zap_page_range(mm
, start
, len
);
943 flush_tlb_range(mm
, start
, end
);
947 /* mapping wholly unaffected? */
948 len
= len
>> PAGE_SHIFT
;
949 diff
= pgoff
- mpnt
->vm_pgoff
;
953 /* Ok, partially affected.. */
954 start
+= diff
<< PAGE_SHIFT
;
955 len
= (len
- diff
) << PAGE_SHIFT
;
956 if (start
& ~PAGE_MASK
) {
957 partial_clear(mpnt
, start
);
958 start
= (start
+ ~PAGE_MASK
) & PAGE_MASK
;
960 flush_cache_range(mm
, start
, end
);
961 zap_page_range(mm
, start
, len
);
962 flush_tlb_range(mm
, start
, end
);
963 } while ((mpnt
= mpnt
->vm_next_share
) != NULL
);
968 * Handle all mappings that got truncated by a "truncate()"
971 * NOTE! We have to be ready to update the memory sharing
972 * between the file and the memory map for a potential last
973 * incomplete page. Ugly, but necessary.
975 void vmtruncate(struct inode
* inode
, loff_t offset
)
977 unsigned long partial
, pgoff
;
978 struct address_space
*mapping
= inode
->i_mapping
;
981 if (inode
->i_size
< offset
)
983 inode
->i_size
= offset
;
984 truncate_inode_pages(mapping
, offset
);
985 spin_lock(&mapping
->i_shared_lock
);
986 if (!mapping
->i_mmap
&& !mapping
->i_mmap_shared
)
989 pgoff
= (offset
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
990 partial
= (unsigned long)offset
& (PAGE_CACHE_SIZE
- 1);
992 if (mapping
->i_mmap
!= NULL
)
993 vmtruncate_list(mapping
->i_mmap
, pgoff
, partial
);
994 if (mapping
->i_mmap_shared
!= NULL
)
995 vmtruncate_list(mapping
->i_mmap_shared
, pgoff
, partial
);
998 spin_unlock(&mapping
->i_shared_lock
);
999 /* this should go into ->truncate */
1000 inode
->i_size
= offset
;
1001 if (inode
->i_op
&& inode
->i_op
->truncate
)
1002 inode
->i_op
->truncate(inode
);
1006 limit
= current
->rlim
[RLIMIT_FSIZE
].rlim_cur
;
1007 if (limit
!= RLIM_INFINITY
) {
1008 if (inode
->i_size
>= limit
) {
1009 send_sig(SIGXFSZ
, current
, 0);
1012 if (offset
> limit
) {
1013 send_sig(SIGXFSZ
, current
, 0);
1017 inode
->i_size
= offset
;
1018 if (inode
->i_op
&& inode
->i_op
->truncate
)
1019 inode
->i_op
->truncate(inode
);
1027 * Primitive swap readahead code. We simply read an aligned block of
1028 * (1 << page_cluster) entries in the swap area. This method is chosen
1029 * because it doesn't cost us any seek time. We also make sure to queue
1030 * the 'original' request together with the readahead ones...
1032 void swapin_readahead(swp_entry_t entry
)
1035 struct page
*new_page
;
1036 unsigned long offset
;
1039 * Get the number of handles we should do readahead io to. Also,
1040 * grab temporary references on them, releasing them as io completes.
1042 num
= valid_swaphandles(entry
, &offset
);
1043 for (i
= 0; i
< num
; offset
++, i
++) {
1044 /* Don't block on I/O for read-ahead */
1045 if (atomic_read(&nr_async_pages
) >= pager_daemon
.swap_cluster
1046 * (1 << page_cluster
)) {
1048 swap_free(SWP_ENTRY(SWP_TYPE(entry
), offset
++));
1051 /* Ok, do the async read-ahead now */
1052 new_page
= read_swap_cache_async(SWP_ENTRY(SWP_TYPE(entry
), offset
), 0);
1053 if (new_page
!= NULL
)
1054 page_cache_release(new_page
);
1055 swap_free(SWP_ENTRY(SWP_TYPE(entry
), offset
));
1060 static int do_swap_page(struct mm_struct
* mm
,
1061 struct vm_area_struct
* vma
, unsigned long address
,
1062 pte_t
* page_table
, swp_entry_t entry
, int write_access
)
1064 struct page
*page
= lookup_swap_cache(entry
);
1069 swapin_readahead(entry
);
1070 page
= read_swap_cache(entry
);
1075 flush_page_to_ram(page
);
1076 flush_icache_page(vma
, page
);
1081 pte
= mk_pte(page
, vma
->vm_page_prot
);
1084 * Freeze the "shared"ness of the page, ie page_count + swap_count.
1085 * Must lock page before transferring our swap count to already
1086 * obtained page count.
1090 if (write_access
&& !is_page_shared(page
))
1091 pte
= pte_mkwrite(pte_mkdirty(pte
));
1094 set_pte(page_table
, pte
);
1095 /* No need to invalidate - it was non-present before */
1096 update_mmu_cache(vma
, address
, pte
);
1097 return 1; /* Minor fault */
1101 * This only needs the MM semaphore
1103 static int do_anonymous_page(struct mm_struct
* mm
, struct vm_area_struct
* vma
, pte_t
*page_table
, int write_access
, unsigned long addr
)
1105 struct page
*page
= NULL
;
1106 pte_t entry
= pte_wrprotect(mk_pte(ZERO_PAGE(addr
), vma
->vm_page_prot
));
1108 page
= alloc_page(GFP_HIGHUSER
);
1111 clear_user_highpage(page
, addr
);
1112 entry
= pte_mkwrite(pte_mkdirty(mk_pte(page
, vma
->vm_page_prot
)));
1114 flush_page_to_ram(page
);
1116 set_pte(page_table
, entry
);
1117 /* No need to invalidate - it was non-present before */
1118 update_mmu_cache(vma
, addr
, entry
);
1119 return 1; /* Minor fault */
1123 * do_no_page() tries to create a new page mapping. It aggressively
1124 * tries to share with existing pages, but makes a separate copy if
1125 * the "write_access" parameter is true in order to avoid the next
1128 * As this is called only for pages that do not currently exist, we
1129 * do not need to flush old virtual caches or the TLB.
1131 * This is called with the MM semaphore held.
1133 static int do_no_page(struct mm_struct
* mm
, struct vm_area_struct
* vma
,
1134 unsigned long address
, int write_access
, pte_t
*page_table
)
1136 struct page
* new_page
;
1139 if (!vma
->vm_ops
|| !vma
->vm_ops
->nopage
)
1140 return do_anonymous_page(mm
, vma
, page_table
, write_access
, address
);
1143 * The third argument is "no_share", which tells the low-level code
1144 * to copy, not share the page even if sharing is possible. It's
1145 * essentially an early COW detection.
1147 new_page
= vma
->vm_ops
->nopage(vma
, address
& PAGE_MASK
, (vma
->vm_flags
& VM_SHARED
)?0:write_access
);
1148 if (new_page
== NULL
) /* no page was available -- SIGBUS */
1150 if (new_page
== NOPAGE_OOM
)
1154 * This silly early PAGE_DIRTY setting removes a race
1155 * due to the bad i386 page protection. But it's valid
1156 * for other architectures too.
1158 * Note that if write_access is true, we either now have
1159 * an exclusive copy of the page, or this is a shared mapping,
1160 * so we can make it writable and dirty to avoid having to
1161 * handle that later.
1163 flush_page_to_ram(new_page
);
1164 flush_icache_page(vma
, new_page
);
1165 entry
= mk_pte(new_page
, vma
->vm_page_prot
);
1167 entry
= pte_mkwrite(pte_mkdirty(entry
));
1168 } else if (page_count(new_page
) > 1 &&
1169 !(vma
->vm_flags
& VM_SHARED
))
1170 entry
= pte_wrprotect(entry
);
1171 set_pte(page_table
, entry
);
1172 /* no need to invalidate: a not-present page shouldn't be cached */
1173 update_mmu_cache(vma
, address
, entry
);
1174 return 2; /* Major fault */
1178 * These routines also need to handle stuff like marking pages dirty
1179 * and/or accessed for architectures that don't do it in hardware (most
1180 * RISC architectures). The early dirtying is also good on the i386.
1182 * There is also a hook called "update_mmu_cache()" that architectures
1183 * with external mmu caches can use to update those (ie the Sparc or
1184 * PowerPC hashed page tables that act as extended TLBs).
1186 * Note the "page_table_lock". It is to protect against kswapd removing
1187 * pages from under us. Note that kswapd only ever _removes_ pages, never
1188 * adds them. As such, once we have noticed that the page is not present,
1189 * we can drop the lock early.
1191 * The adding of pages is protected by the MM semaphore (which we hold),
1192 * so we don't need to worry about a page being suddenly been added into
1195 static inline int handle_pte_fault(struct mm_struct
*mm
,
1196 struct vm_area_struct
* vma
, unsigned long address
,
1197 int write_access
, pte_t
* pte
)
1202 * We need the page table lock to synchronize with kswapd
1203 * and the SMP-safe atomic PTE updates.
1205 spin_lock(&mm
->page_table_lock
);
1207 if (!pte_present(entry
)) {
1209 * If it truly wasn't present, we know that kswapd
1210 * and the PTE updates will not touch it later. So
1213 spin_unlock(&mm
->page_table_lock
);
1214 if (pte_none(entry
))
1215 return do_no_page(mm
, vma
, address
, write_access
, pte
);
1216 return do_swap_page(mm
, vma
, address
, pte
, pte_to_swp_entry(entry
), write_access
);
1220 if (!pte_write(entry
))
1221 return do_wp_page(mm
, vma
, address
, pte
, entry
);
1223 entry
= pte_mkdirty(entry
);
1225 entry
= pte_mkyoung(entry
);
1226 establish_pte(vma
, address
, pte
, entry
);
1227 spin_unlock(&mm
->page_table_lock
);
1232 * By the time we get here, we already hold the mm semaphore
1234 int handle_mm_fault(struct mm_struct
*mm
, struct vm_area_struct
* vma
,
1235 unsigned long address
, int write_access
)
1241 pgd
= pgd_offset(mm
, address
);
1242 pmd
= pmd_alloc(pgd
, address
);
1245 pte_t
* pte
= pte_alloc(pmd
, address
);
1247 ret
= handle_pte_fault(mm
, vma
, address
, write_access
, pte
);
1253 * Simplistic page force-in..
1255 int make_pages_present(unsigned long addr
, unsigned long end
)
1258 struct mm_struct
*mm
= current
->mm
;
1259 struct vm_area_struct
* vma
;
1261 vma
= find_vma(mm
, addr
);
1262 write
= (vma
->vm_flags
& VM_WRITE
) != 0;
1266 if (handle_mm_fault(mm
, vma
, addr
, write
) < 0)
1269 } while (addr
< end
);